User’s Guide and Developer’s Manual¶
Introduction¶
QMCPACK is an open-source, high-performance electronic structure code that implements numerous Quantum Monte Carlo (QMC) algorithms. Its main applications are electronic structure calculations of molecular, periodic 2D, and periodic 3D solid-state systems. Real-space variational Monte Carlo (VMC), diffusion Monte Carlo (DMC), and a number of other advanced QMC algorithms are implemented. A full set of orbital-space auxiliary-field QMC (AFQMC) methods is also implemented. By directly solving the Schrodinger equation, QMC methods offer greater accuracy than methods such as density functional theory but at a trade-off of much greater computational expense. Distinct from many other correlated many-body methods, QMC methods are readily applicable to both isolated molecular systems and to bulk (periodic) systems including metals and insulators. The few systematic errors in these methods are increasingly testable allowing for greater confidence in predictions and convergence to e.g. chemically accurate results in some cases.
QMCPACK is written in C++ and is designed with the modularity afforded by object-oriented programming. High parallel and computational efficiencies are achievable on the largest supercomputers. Because of the modular architecture, the addition of new wavefunctions, algorithms, and observables is relatively straightforward. For parallelization, QMCPACK uses a fully hybrid (OpenMP,CUDA)/MPI approach to optimize memory usage and to take advantage of the growing number of cores per SMP node or graphical processing units (GPUs) and accelerators. Finally, QMCPACK uses standard file formats for input and output in XML and HDF5 to facilitate data exchange.
This manual currently serves as an introduction to the essential features of QMCPACK and as a guide to installing and running it. Over time this manual will be expanded to include a fuller introduction to QMC methods in general and to include more of the specialized features in QMCPACK.
Besides studying this manual we recommend reading a recent review of QMCPACK developments [KAB+20] as well as the QMCPACK citation paper [KBB+18].
Quickstart and a first QMCPACK calculation¶
In case you are keen to get started, this section describes how to quickly build and run a first QMCPACK calculation on a standard UNIX or Linux-like system. The build system usually works without much fuss on these systems. If C++, MPI, BLAS/LAPACK, FFTW, HDF5, and CMake are already installed, QMCPACK can be built and run within five minutes. For supercomputers, cross-compilation systems, and some computer clusters, the build system might require hints on the locations of libraries and which versions to use, typical of any code; see Obtaining, installing, and validating QMCPACK. Installation instructions for common workstations and supercomputers includes complete examples of installations for common workstations and supercomputers that you can reuse.
To build QMCPACK:
Download the latest QMCPACK distribution from http://www.qmcpack.org.
Untar the archive (e.g.,
tar xvf qmcpack_v1.3.tar.gz).Check the instructions in the README file.
Run CMake in a suitable build directory to configure QMCPACK for your system:
cd qmcpack/build; cmake ..If CMake is unable to find all needed libraries, see Obtaining, installing, and validating QMCPACK for instructions and specific build instructions for common systems.
Build QMCPACK:
makeormake -j 16; use the latter for a faster parallel build on a system using, for example, 16 processes.The QMCPACK executable is
bin/qmcpack.
QMCPACK is distributed with examples illustrating different capabilities. Most of the examples are designed to run quickly with modest resources. We’ll run a short diffusion Monte Carlo calculation of a water molecule:
Go to the appropriate example directory:
cd ../examples/molecules/H2O.- (Optional) Put the QMCPACK binary on your path:
export PATH=\$PATH:location-of-qmcpack/build/bin Run QMCPACK:
../../../build/bin/qmcpack simple-H2O.xmlorqmcpack simple-H2O.xmlif you followed the step above.The run will output to the screen and generate a number of files:
$ls H2O* H2O.HF.wfs.xml H2O.s001.scalar.dat H2O.s002.cont.xml H2O.s002.qmc.xml H2O.s002.stat.h5 H2O.s001.qmc.xml H2O.s001.stat.h5 H2O.s002.dmc.dat H2O.s002.scalar.dat
Partially summarized results are in the standard text files with the suffixes scalar.dat and dmc.dat. They are viewable with any standard editor.
If you have Python and matplotlib installed, you can use the analysis utility to produce statistics and plots of the data. See Analyzing QMCPACK data for information on analyzing QMCPACK data.
export PATH=$PATH:location-of-qmcpack/nexus/bin
export PYTHONPATH=$PYTHONPATH:location-of-qmcpack/nexus/library
qmca H2O.s002.scalar.dat # For statistical analysis of the DMC data
qmca -t -q e H2O.s002.scalar.dat # Graphical plot of DMC energy
The last command will produce a graph as per Fig. 1. This shows the average energy of the DMC walkers at each timestep. In a real simulation we would have to check equilibration, convergence with walker population, time step, etc.
Congratulations, you have completed a DMC calculation with QMCPACK!
Trace of walker energies produced by the qmca tool for a simple water molecule example.¶
Authors and History¶
Development of QMCPACK was started in the late 2000s by Jeongnim Kim while in the group of Professor David Ceperley at the University of Illinois at Urbana-Champaign, with later contributions being made at Oak Ridge National Laboratory (ORNL). Over the years, many others have contributed, including students and researchers in the groups of Professor David Ceperley and Professor Richard M. Martin, and increasingly staff and postdocs at Lawrence Livermore National Laboratory, Sandia National Laboratories, Argonne National Laboratory, and ORNL.
Additional developers, contributors, and advisors include Anouar Benali, Mark A. Berrill, David M. Ceperley, Simone Chiesa, Raymond C. III Clay, Bryan Clark, Kris T. Delaney, Kenneth P. Esler, Paul R. C. Kent, Jaron T. Krogel, Ying Wai Li, Ye Luo, Jeremy McMinis, Miguel A. Morales, William D. Parker, Nichols A. Romero, Luke Shulenburger, Norman M. Tubman, and Jordan E. Vincent. See the authors of [KAB+20] and [KBB+18].
If you should be added to these lists, please let us know.
Development of QMCPACK has been supported financially by several grants, including the following:
“Center for Predictive Simulation of Functional Materials”, supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division, as part of the Computational Materials Sciences Program.
The Exascale Computing Project (17-SC-20-SC), a joint project of the U.S. Department of Energy’s Office of Science and National Nuclear Security Administration, responsible for delivering a capable exascale ecosystem, including software, applications, and hardware technology, to support the nation’s exascale computing imperative.
“Network for ab initio many-body methods: development, education and training” supported through the Predictive Theory and Modeling for Materials and Chemical Science program by the U.S. Department of Energy Office of Science, Basic Energy Sciences.
“QMC Endstation,” supported by Accelerating Delivery of Petascale Computing Environment at the DOE Leadership Computing Facility at ORNL.
PetaApps, supported by the US National Science Foundation.
Materials Computation Center (MCC), supported by the US National Science Foundation.
Support and Contacting the Developers¶
Questions about installing, applying, or extending QMCPACK can be posted on the QMCPACK Google group at https://groups.google.com/forum/#!forum/qmcpack. You may also email any of the developers, but we recommend checking the group first. Particular attention is given to any problem reports. Technical questions can also be posted on the QMCPACK GitHub repository https://github.com/QMCPACK/qmcpack/issues.
Performance¶
QMCPACK implements modern Monte Carlo (MC) algorithms, is highly parallel, and is written using very efficient code for high per-CPU or on-node performance. In particular, the code is highly vectorizable, giving high performance on modern central processing units (CPUs) and GPUs. We believe QMCPACK delivers performance either comparable to or better than other QMC codes when similar calculations are run, particularly for the most common QMC methods and for large systems. If you find a calculation where this is not the case, or you simply find performance slower than expected, please post on the Google group or contact one of the developers. These reports are valuable. If your calculation is sufficiently mainstream we will optimize QMCPACK to improve the performance.
Open Source License¶
QMCPACK is distributed under the University of Illinois at Urbana-Champaign/National Center for Supercomputing Applications (UIUC/NCSA) Open Source License.
University of Illinois/NCSA Open Source License
Copyright (c) 2003, University of Illinois Board of Trustees.
All rights reserved.
Developed by:
Jeongnim Kim
Condensed Matter Physics,
National Center for Supercomputing Applications, University of Illinois
Materials computation Center, University of Illinois
http://www.mcc.uiuc.edu/qmc/
Permission is hereby granted, free of charge, to any person obtaining a
copy of this software and associated documentation files (the
``Software''), to deal with the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimers.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimers in
the documentation and/or other materials provided with the
distribution.
* Neither the names of the NCSA, the MCC, the University of Illinois,
nor the names of its contributors may be used to endorse or promote
products derived from this Software without specific prior written
permission.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
THE CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
OTHER DEALINGS WITH THE SOFTWARE.
Copyright is generally believed to remain with the authors of the individual sections of code. See the various notations in the source code as well as the code history.
Contributing to QMCPACK¶
QMCPACK is fully open source, and we welcome contributions. If you are planning a development, early discussions are encouraged. Please post on the QMCPACK Google group, on the QMCPACK GitHub repository, or contact one of the developers. We can tell you whether anyone else is working on a similar feature or whether any related work has been done in the past. Credit for your contribution can be obtained, for example, through citation of a paper or by becoming one of the authors on the next version of the standard QMCPACK reference citation.
See Development Guide for details about developing for QMCPACK, including instructions on how to work with GitHub, the style guide, and examples about the code architecture.
Contributions are made under the same license as QMCPACK, the UIUC/NCSA open source license. If this is problematic, please discuss with a developer.
Please note the following guidelines for contributions:
Additions should be fully synchronized with the latest release version and the latest develop branch on GitHub. Merging of code developed on older versions is error prone.
Code should be cleanly formatted, commented, portable, and accessible to other programmers. That is, if you need to use any clever tricks, add a comment to note this, why the trick is needed, how it works, etc. Although we appreciate high performance, ease of maintenance and accessibility are also considerations.
Comment your code. You are not only writing it for the compiler for also for other humans! (We know this is a repeat of the previous point, but it is important enough to repeat.)
Write a brief description of the method, algorithms, and inputs and outputs suitable for inclusion in this manual.
Develop tests that exercise the functionality that can be used for validation and for examples. Where is it practical to write them, we prefer unit tests and fully deterministic tests ahead of stochastic tests. Stochastic tests naturally fail on occasion, which is a property that does not scale to hundreds of tests. We can help with this and tests integration into the test system.
QMCPACK Roadmap¶
A general outline of the QMCPACK roadmap is given in the following sections. Suggestions for improvements are welcome, particularly those that would facilitate new scientific applications. For example, if an interface to a particular quantum chemical or density functional code would help, this would be given strong consideration.
Code¶
We will continue to improve the accessibility and usability of QMCPACK through combinations of more convenient input parameters, improved workflow, integration with more quantum chemical and density functional codes, and a wider range of examples. Suggestions are very welcome, both from new users of QMC and from those experienced with other QMC codes.
A main development focus is the creation of a single performance portable version of the code. All features will consequently be available on all platforms, including accelerators (GPUs) from NVIDIA, AMD, and Intel. The internal design is being updated for greater simplicity. Overall we expect this to increase performance and improve the overall consistency and robustness of the code. It will also enable us to remove legacy implementations.
Documentation and examples¶
This manual describes the core features of QMCPACK that are required for routine research calculations and standard QMC workflows, i.e., the VMC and DMC methods, auxiliary field QMC, how to obtain and optimize trial wavefunctions, and simple observables. This covers at least 95% of use cases.
Because of its history as an academically developed research code, QMCPACK contains a variety of additional QMC methods, trial wavefunction forms, potentials, etc., that, although far from critical, might be very useful for specialized calculations or particular material or chemical systems. These “secret features” (every code has these) are not actually secret but simply lack descriptions, example inputs, and tests. You are encouraged to browse and read the source code to find them. New descriptions will be added over time but can also be prioritized and added on request (e.g., if a specialized Jastrow factor would help or a historical Jastrow form is needed for benchmarking).
- KAB+20(1,2)
P. R. C. Kent, Abdulgani Annaberdiyev, Anouar Benali, M. Chandler Bennett, Edgar Josué Landinez Borda, Peter Doak, Hongxia Hao, Kenneth D. Jordan, Jaron T. Krogel, Ilkka Kylänpää, Joonho Lee, Ye Luo, Fionn D. Malone, Cody A. Melton, Lubos Mitas, Miguel A. Morales, Eric Neuscamman, Fernando A. Reboredo, Brenda Rubenstein, Kayahan Saritas, Shiv Upadhyay, Guangming Wang, Shuai Zhang, and Luning Zhao. QMCPACK: Advances in the development, efficiency, and application of auxiliary field and real-space variational and diffusion quantum Monte Carlo. Journal of Chemical Physics, 152(17):174105, May 2020. doi:10.1063/5.0004860.
- KBB+18(1,2)
Jeongnim Kim, Andrew T. Baczewski, Todd D. Beaudet, Anouar Benali, M. Chandler Bennett, Mark A. Berrill, Nick S. Blunt, Edgar Josué Landinez Borda, Michele Casula, David M. Ceperley, Simone Chiesa, Bryan K. Clark, Raymond C. Clay III, Kris T. Delaney, Mark Dewing, Kenneth P. Esler, Hongxia Hao, Olle Heinonen, Paul R. C. Kent, Jaron T. Krogel, Ilkka Kylänpää, Ying Wai Li, M. Graham Lopez, Ye Luo, Fionn D. Malone, Richard M. Martin, Amrita Mathuriya, Jeremy McMinis, Cody A. Melton, Lubos Mitas, Miguel A. Morales, Eric Neuscamman, William D. Parker, Sergio D. Pineda Flores, Nichols A. Romero, Brenda M. Rubenstein, Jacqueline A. R. Shea, Hyeondeok Shin, Luke Shulenburger, Andreas F. Tillack, Joshua P. Townsend, Norm M. Tubman, Brett Van Der Goetz, Jordan E. Vincent, D. ChangMo Yang, Yubo Yang, Shuai Zhang, and Luning Zhao. QMCPACK : an open source ab initio quantum Monte Carlo package for the electronic structure of atoms, molecules and solids. Journal of Physics: Condensed Matter, 30(19):195901, 2018. doi:10.1088/1361-648X/aab9c3.
Features of QMCPACK¶
Note that besides direct use, most features are also available via Nexus, an advanced workflow tool to automate all aspects of QMC calculation from initial DFT calculations through to final analysis. Use of Nexus is highly recommended for research calculations due to the greater ease of use and increased reproducibility.
Real-space Monte Carlo¶
The following list contains the main production-level features of QMCPACK for real-space Monte Carlo. If you do not see a specific feature that you are interested in, check the remainder of this manual or ask if that specific feature can be made available.
Variational Monte Carlo (VMC).
Diffusion Monte Carlo (DMC).
Reptation Monte Carlo.
Single and multideterminant Slater Jastrow wavefunctions.
Wavefunction updates using optimized multideterminant algorithm of Clark et al.
Backflow wavefunctions.
One, two, and three-body Jastrow factors.
Excited state calculations via flexible occupancy assignment of Slater determinants.
All electron and nonlocal pseudopotential calculations.
Casula T-moves for variational evaluation of nonlocal pseudopotentials (non-size-consistent and size-consistent variants).
Spin-orbit coupling from relativistic pseudopotentials following the approach of Melton, Bennett, and Mitas.
Support for twist boundary conditions and calculations on metals.
Wavefunction optimization using the “linear method” of Umrigar and coworkers, with an arbitrary mix of variance and energy in the objective function.
Blocked, low memory adaptive shift optimizer of Zhao and Neuscamman.
Gaussian, Slater, plane-wave, and real-space spline basis sets for orbitals.
Interface and conversion utilities for plane-wave wavefunctions from Quantum Espresso (Plane-Wave Self-Consistent Field package [PWSCF]).
Interface and conversion utilities for Gaussian-basis wavefunctions from GAMESS, PySCF, and QP2. Many more are supported via the molden format and molden2qmc.
Easy extension and interfacing to other electronic structure codes via standardized XML and HDF5 inputs.
MPI parallelism, with scaling to millions of cores.
Fully threaded using OpenMP.
Highly efficient vectorized CPU code tailored for modern architectures. [MLC+17]
GPU (NVIDIA CUDA) implementation (limited functionality - see Supported GPU features for real space QMC).
Analysis tools for minimal environments (Perl only) through to Python-based environments with graphs produced via matplotlib (included with Nexus).
Auxiliary-Field Quantum Monte Carlo¶
The orbital-space Auxiliary-Field Quantum Monte Carlo (AFQMC) method is now also available in QMCPACK. The main input data are the matrix elements of the Hamiltonian in a given single particle basis set, which must be produced from mean-field calculations such as Hartree-Fock or density functional theory. A partial list of the current capabilities of the code follows. For a detailed description of the available features, see Auxiliary-Field Quantum Monte Carlo.
Phaseless AFQMC algorithm of Zhang et al. [ZK03].
Very efficient GPU implementation for most features.
“Hybrid” and “local energy” propagation schemes.
Hamiltonian matrix elements from (1) Molpro’s FCIDUMP format (which can be produced by Molpro, PySCF, and VASP) and (2) internal HDF5 format produced by PySCF (see AFQMC section below).
AFQMC calculations with RHF (closed-shell doubly occupied), ROHF (open-shell doubly occupied), and UHF (spin polarized broken symmetry) symmetry.
Single and multideterminant trial wavefunctions. Multideterminant expansions with either orthogonal or nonorthogonal determinants.
Fast update scheme for orthogonal multideterminant expansions.
Distributed propagation algorithms for large systems. Enables calculations where data structures do not fit on a single node.
Complex implementation for PBC calculations with complex integrals.
Sparse representation of large matrices for reduced memory usage.
Mixed and back-propagated estimators.
Specialized implementation for solids with k-point symmetry (e.g. primitive unit cells with k-points).
Supported GPU features for real space QMC¶
The “legacy” GPU implementation for real space QMC uses NVIDIA CUDA and achieves very good speedup on NVIDIA GPUs. However, only a very limited subset of features is available. As detailed in QMCPACK Roadmap, a new full-featured GPU implementation is currently being made that is both high performing and portable to other GPU architectures. The existing implementation supports multiple GPUs per node, with MPI tasks assigned in a round-robin order to the GPUs. Currently, for large runs, 1 MPI task should be used per GPU per node. For smaller calculations, use of multiple MPI tasks per GPU might yield improved performance.
Supported GPU features:
VMC, wavefunction optimization, DMC.
Periodic and open boundary conditions. Mixed boundary conditions are not supported.
Wavefunctions:
Single Slater determinants with 3D B-spline orbitals. Twist-averaged boundary conditions and complex wavefunctions are fully supported. Gaussian type orbitals are not supported.
Hybrid mixed basis representation in which orbitals are represented as 1D splines times spherical harmonics in spherical regions (muffin tins) around atoms and 3D B-splines in the interstitial region.
One-body and two-body Jastrow functions represented as 1D B-splines. Three-body Jastrow functions are not supported.
Semilocal (nonlocal and local) pseudopotentials, Coulomb interaction (electron-electron, electron-ion), and model periodic Coulomb (MPC) interaction.
Sharing of spline data across multiple GPUs¶
Sharing of GPU spline data enables distribution of the data across multiple GPUs on a given computational node. For example, on a two-GPU-per-node system, each GPU would have half of the orbitals. This allows use of larger overall spline tables than would fit in the memory of individual GPUs and potentially up to the total GPU memory on a node. To obtain high performance, large electron counts or a high-performing CPU-GPU interconnect is required.
To use this feature, the following needs to be done:
The CUDA Multi-Process Service (MPS) needs to be used (e.g., on OLCF Summit/SummitDev use “-alloc_flags gpumps” for bsub). If MPI is not detected, sharing will be disabled.
CUDA_VISIBLE_DEVICES needs to be properly set to control each rank’s visible CUDA devices (e.g., on OLCF Summit/SummitDev create a resource set containing all GPUs with the respective number of ranks with “jsrun –task-per-rs Ngpus -g Ngpus”).
In the determinant set definition of the <wavefunction> section, the “gpusharing” parameter needs to be set (i.e., <determinantset gpusharing=“yes”>). See Spline basis sets.
- MLC+17
Amrita Mathuriya, Ye Luo, Raymond C. Clay, III, Anouar Benali, Luke Shulenburger, and Jeongnim Kim. Embracing a new era of highly efficient and productive quantum monte carlo simulations. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, SC ‘17, 38:1–38:12. New York, NY, USA, 2017. ACM. URL: http://doi.acm.org/10.1145/3126908.3126952, doi:10.1145/3126908.3126952.
- ZK03
Shiwei Zhang and Henry Krakauer. Quantum Monte Carlo Method using Phase-Free Random Walks with Slater Determinants. Physical Review Letters, 90(13):136401, April 2003. doi:10.1103/PhysRevLett.90.136401.
Obtaining, installing, and validating QMCPACK¶
This section describes how to obtain, build, and validate QMCPACK. This process is designed to be as simple as possible and should be no harder than building a modern plane-wave density functional theory code such as Quantum ESPRESSO, QBox, or VASP. Parallel builds enable a complete compilation in under 2 minutes on a fast multicore system. If you are unfamiliar with building codes we suggest working with your system administrator to install QMCPACK.
Installation steps¶
To install QMCPACK, follow the steps below. Full details of each step are given in the referenced sections.
Download the source code from Obtaining the latest release version or Obtaining the latest development version.
Verify that you have the required compilers, libraries, and tools installed (Prerequisites).
If you will use Quantum ESPRESSO, download and patch it. The patch adds the pw2qmcpack utility (Installing and patching Quantum ESPRESSO).
Run the cmake configure step and build with make (Building with CMake and Quick build instructions (try first)). Examples for common systems are given in Installation instructions for common workstations and supercomputers.
Run the tests to verify QMCPACK (Testing and validation of QMCPACK).
Hints for high performance are in How to build the fastest executable version of QMCPACK. Troubleshooting suggestions are in Troubleshooting the installation.
Note that there are two different QMCPACK executables that can be produced: the general one, which is the default, and the “complex” version, which supports periodic calculations at arbitrary twist angles and k-points. This second version is enabled via a cmake configuration parameter (see Configuration Options). The general version supports only wavefunctions that can be made real. If you run a calculation that needs the complex version, QMCPACK will stop and inform you.
Obtaining the latest release version¶
Major releases of QMCPACK are distributed from http://www.qmcpack.org. Because these versions undergo the most testing, we encourage using them for all production calculations unless there are specific reasons not to do so.
Releases are usually compressed tar files that indicate the version number, date, and often the source code revision control number corresponding to the release. To obtain the latest release:
Download the latest QMCPACK distribution from http://www.qmcpack.org.
Untar the archive (e.g.,
tar xvf qmcpack_v1.3.tar.gz).
Releases can also be obtained from the ‘master’ branch of the QMCPACK git repository, similar to obtaining the development version (Obtaining the latest development version).
Obtaining the latest development version¶
The most recent development version of QMCPACK can be obtained anonymously via
git clone https://github.com/QMCPACK/qmcpack.git
Once checked out, updates can be made via the standard git pull.
The ‘develop’ branch of the git repository contains the day-to-day development source with the latest updates, bug fixes, etc. This version might be useful for updates to the build system to support new machines, for support of the latest versions of Quantum ESPRESSO, or for updates to the documentation. Note that the development version might not be fully consistent with the online documentation. We attempt to keep the development version fully working. However, please be sure to run tests and compare with previous release versions before using for any serious calculations. We try to keep bugs out, but occasionally they crawl in! Reports of any breakages are appreciated.
Prerequisites¶
The following items are required to build QMCPACK. For workstations, these are available via the standard package manager. On shared supercomputers this software is usually installed by default and is often accessed via a modules environment—check your system documentation.
Use of the latest versions of all compilers and libraries is strongly encouraged but not absolutely essential. Generally, newer versions are faster; see How to build the fastest executable version of QMCPACK for performance suggestions. Versions of compilers over two years old are unsupported and untested by the developers although they may still work.
C/C++ compilers such as GNU, Clang, Intel, and IBM XL. C++ compilers are required to support the C++ 14 standard. Use of recent (“current year version”) compilers is strongly encouraged.
An MPI library such as OpenMPI (http://open-mpi.org) or a vendor-optimized MPI.
BLAS/LAPACK, numerical, and linear algebra libraries. Use platform-optimized libraries where available, such as Intel MKL. ATLAS or other optimized open source libraries can also be used (http://math-atlas.sourceforge.net).
CMake, build utility (http://www.cmake.org).
Libxml2, XML parser (http://xmlsoft.org).
HDF5, portable I/O library (http://www.hdfgroup.org/HDF5/). Good performance at large scale requires parallel version \(>=\) 1.10.
BOOST, peer-reviewed portable C++ source libraries (http://www.boost.org). Minimum version is 1.61.0.
FFTW, FFT library (http://www.fftw.org/).
To build the GPU accelerated version of QMCPACK, an installation of NVIDIA CUDA development tools is required. Ensure that this is compatible with the C and C++ compiler versions you plan to use. Supported versions are included in the NVIDIA release notes.
Many of the utilities provided with QMCPACK require Python (v3). The numpy and matplotlib libraries are required for full functionality.
C++ 14 standard library¶
The C++ standard consists of language features—which are implemented in the compiler—and library features—which are implemented in the standard library. GCC includes its own standard library and headers, but many compilers do not and instead reuse those from an existing GCC install. Depending on setup and installation, some of these compilers might not default to using a GCC with C++ 14 headers (e.g., GCC 4.8 is common as a base system compiler, but its standard library only supports C++ 11).
The symptom of having header files that do not support the C++ 14 standard is usually compile errors involving standard include header files. Look for the GCC library version, which should be present in the path to the include file in the error message, and ensure that it is 5.0 or greater. To avoid these errors occurring at compile time, QMCPACK tests for a C++ 14 standard library during configuration and will halt with an error if one is not found.
At sites that use modules, it is often sufficient to simply load a newer GCC.
Intel compiler¶
The Intel compiler version must be 19 or newer due to use of C++14 and bugs and limitations in earlier versions.
If a newer GCC is needed, the -cxxlib option can be used to point to a different
GCC installation. (Alternately, the -gcc-name or -gxx-name options can be used.) Be sure to
pass this flag to the C compiler in addition to the C++ compiler. This
is necessary because CMake extracts some library paths from the C
compiler, and those paths usually also contain to the C++ library. The
symptom of this problem is C++ 14 standard library functions not found
at link time.
Building with CMake¶
The build system for QMCPACK is based on CMake. It will autoconfigure based on the detected compilers and libraries. The most recent version of CMake has the best detection for the greatest variety of systems. The minimum required version of CMake is 3.6, which is the oldest version to support correct application of C++ 14 flags for the Intel compiler. Most computer installations have a sufficiently recent CMake, though it might not be the default.
If no appropriate version CMake is available, building it from source is
straightforward. Download a version from https://cmake.org/download/ and
unpack the files. Run ./bootstrap from the CMake directory, and then run make when that
finishes. The resulting CMake executable will be in the directory. The
executable can be run directly from that location.
Previously, QMCPACK made extensive use of toolchains, but the build system has since been updated to eliminate the use of toolchain files for most cases. The build system is verified to work with GNU, Intel, and IBM XLC compilers. Specific compile options can be specified either through specific environment or CMake variables. When the libraries are installed in standard locations (e.g., /usr, /usr/local), there is no need to set environment or CMake variables for the packages.
Quick build instructions (try first)¶
If you are feeling lucky and are on a standard UNIX-like system such as a Linux workstation, the following might quickly give a working QMCPACK:
The safest quick build option is to specify the C and C++ compilers through their MPI wrappers. Here we use Intel MPI and Intel compilers. Move to the build directory, run CMake, and make
cd build
cmake -DCMAKE_C_COMPILER=mpiicc -DCMAKE_CXX_COMPILER=mpiicpc ..
make -j 8
You can increase the “8” to the number of cores on your system for faster builds. Substitute mpicc and mpicxx or other wrapped compiler names to suit your system. For example, with OpenMPI use
cd build
cmake -DCMAKE_C_COMPILER=mpicc -DCMAKE_CXX_COMPILER=mpicxx ..
make -j 8
If you are feeling particularly lucky, you can skip the compiler specification:
cd build
cmake ..
make -j 8
The complexities of modern computer hardware and software systems are such that you should check that the autoconfiguration system has made good choices and picked optimized libraries and compiler settings before doing significant production. That is, check the following details. We give examples for a number of common systems in Installation instructions for common workstations and supercomputers.
Environment variables¶
A number of environment variables affect the build. In particular they can control the default paths for libraries, the default compilers, etc. The list of environment variables is given below:
CXX C++ compiler
CC C Compiler
MKL_ROOT Path for MKL
HDF5_ROOT Path for HDF5
BOOST_ROOT Path for Boost
FFTW_HOME Path for FFTW
Configuration Options¶
In addition to reading the environment variables, CMake provides a number of optional variables that can be set to control the build and configure steps. When passed to CMake, these variables will take precedent over the environment and default variables. To set them, add -D FLAG=VALUE to the configure line between the CMake command and the path to the source directory.
Key QMCPACK build options
QMC_CUDA Enable legacy CUDA code path for NVIDIA GPU acceleration (1:yes, 0:no) QMC_COMPLEX Build the complex (general twist/k-point) version (1:yes, 0:no) QMC_MIXED_PRECISION Build the mixed precision (mixing double/float) version (1:yes (QMC_CUDA=1 default), 0:no (QMC_CUDA=0 default)). Mixed precision calculations can be signifiantly faster but should be carefully checked validated against full double precision runs, particularly for large electron counts. ENABLE_CUDA ON/OFF(default). Enable CUDA code path for NVIDIA GPU acceleration. Production quality for AFQMC. Pre-production quality for real-space. Use CUDA_ARCH, default sm_70, to set the actual GPU architecture. ENABLE_OFFLOAD ON/OFF(default). Enable OpenMP target offload for GPU acceleration. ENABLE_TIMERS ON(default)/OFF. Enable fine-grained timers. Timers are on by default but at level coarse to avoid potential slowdown in tiny systems. For systems beyond tiny sizes (100+ electrons) there is no risk.
General build options
CMAKE_BUILD_TYPE A variable which controls the type of build (defaults to Release). Possible values are: None (Do not set debug/optmize flags, use CMAKE_C_FLAGS or CMAKE_CXX_FLAGS) Debug (create a debug build) Release (create a release/optimized build) RelWithDebInfo (create a release/optimized build with debug info) MinSizeRel (create an executable optimized for size) CMAKE_SYSTEM_NAME Set value to CrayLinuxEnvironment when cross-compiling in Cray Programming Environment. CMAKE_C_COMPILER Set the C compiler CMAKE_CXX_COMPILER Set the C++ compiler CMAKE_C_FLAGS Set the C flags. Note: to prevent default debug/release flags from being used, set the CMAKE_BUILD_TYPE=None Also supported: CMAKE_C_FLAGS_DEBUG, CMAKE_C_FLAGS_RELEASE, and CMAKE_C_FLAGS_RELWITHDEBINFO CMAKE_CXX_FLAGS Set the C++ flags. Note: to prevent default debug/release flags from being used, set the CMAKE_BUILD_TYPE=None Also supported: CMAKE_CXX_FLAGS_DEBUG, CMAKE_CXX_FLAGS_RELEASE, and CMAKE_CXX_FLAGS_RELWITHDEBINFO CMAKE_INSTALL_PREFIX Set the install location (if using the optional install step) INSTALL_NEXUS Install Nexus alongside QMCPACK (if using the optional install step)
Additional QMCPACK build options
QE_BIN Location of Quantum Espresso binaries including pw2qmcpack.x QMC_DATA Specify data directory for QMCPACK performance and integration tests QMC_INCLUDE Add extra include paths QMC_EXTRA_LIBS Add extra link libraries QMC_BUILD_STATIC ON/OFF(default). Add -static flags to build QMC_SYMLINK_TEST_FILES Set to zero to require test files to be copied. Avoids space saving default use of symbolic links for test files. Useful if the build is on a separate filesystem from the source, as required on some HPC systems. QMC_VERBOSE_CONFIGURATION Print additional information during cmake configuration including details of which tests are enabled.
BLAS/LAPACK related
BLA_VENDOR If set, checks only the specified vendor, if not set checks all the possibilities. See full list at https://cmake.org/cmake/help/latest/module/FindLAPACK.html MKL_ROOT Path to MKL libraries. Only necessary when auto-detection fails or overriding is desired.
libxml2 related
LIBXML2_INCLUDE_DIR Include directory for libxml2 LIBXML2_LIBRARY Libxml2 library
HDF5 related
HDF5_PREFER_PARALLEL TRUE(default for MPI build)/FALSE, enables/disable parallel HDF5 library searching. ENABLE_PHDF5 ON(default for parallel HDF5 library)/OFF, enables/disable parallel collective I/O.
FFTW related
FFTW_INCLUDE_DIRS Specify include directories for FFTW FFTW_LIBRARY_DIRS Specify library directories for FFTW
CTest related
MPIEXEC_EXECUTABLE Specify the mpi wrapper, e.g. srun, aprun, mpirun, etc. MPIEXEC_NUMPROC_FLAG Specify the number of mpi processes flag, e.g. "-n", "-np", etc. MPIEXEC_PREFLAGS Flags to pass to MPIEXEC_EXECUTABLE directly before the executable to run.
LLVM/Clang Developer Options
LLVM_SANITIZE_ADDRES link with the Clang address sanitizer library LLVM_SANITIZE_MEMORY link with the Clang memory sanitizer library
Clang address sanitizer library
Clang memory sanitizer library
See LLVM Sanitizer Libraries for more information.
Notes for OpenMP target offload to accelerators (experimental)¶
QMCPACK is currently being updated to support OpenMP target offload and obtain performance portability across GPUs from different vendors. This is currently an experimental feature and is not suitable for production. Additional implementation in QMCPACK as well as improvements in open-source and vendor compilers is required for production status to be reached. The following compilers have been verified:
LLVM Clang 11. Support NVIDIA GPUs.
-D ENABLE_OFFLOAD=ON -D USE_OBJECT_TARGET=ON
Clang and its downstream compilers support two extra options
OFFLOAD_TARGET for the offload target. default nvptx64-nvidia-cuda. OFFLOAD_ARCH for the target architecture if not using the compiler default.
IBM XL 16.1. Support NVIDIA GPUs.
-D ENABLE_OFFLOAD=ON
AMD AOMP Clang 11.8. Support AMD GPUs.
-D ENABLE_OFFLOAD=ON -D OFFLOAD_TARGET=amdgcn-amd-amdhsa -D OFFLOAD_ARCH=gfx906
Intel oneAPI beta08. Support Intel GPUs.
-D ENABLE_OFFLOAD=ON -D OFFLOAD_TARGET=spir64
HPE Cray 11. Support NVIDIA and AMD GPUs.
-D ENABLE_OFFLOAD=ON
OpenMP offload features can be used together with vendor specific code paths to maximize QMCPACK performance. Some new CUDA functionality has been implemented to improve efficiency on NVIDIA GPUs in conjunction with the Offload code paths: For example, using Clang 11 on Summit.
-D ENABLE_OFFLOAD=ON -D USE_OBJECT_TARGET=ON -D ENABLE_CUDA=ON -D CUDA_ARCH=sm_70 -D CUDA_HOST_COMPILER=`which gcc`
Installation from CMake¶
Installation is optional. The QMCPACK executable can be run from the bin directory in the build location.
If the install step is desired, run the make install command to install the QMCPACK executable, the converter,
and some additional executables.
Also installed is the qmcpack.settings file that records options used to compile QMCPACK.
Specify the CMAKE_INSTALL_PREFIX CMake variable during configuration to set the install location.
Role of QMC_DATA¶
QMCPACK includes a variety of optional performance and integration tests that use research quality wavefunctions to obtain meaningful performance and to more thoroughly test the code. The necessarily large input files are stored in the location pointed to by QMC_DATA (e.g., scratch or long-lived project space on a supercomputer). These multi-gigabyte files are not included in the source code distribution to minimize size. The tests are activated if CMake detects the files when configured. See tests/performance/NiO/README, tests/solids/NiO_afqmc/README, tests/performance/C-graphite/README, and tests/performance/C-molecule/README for details of the current tests and input files and to download them.
Currently the files must be downloaded via https://anl.box.com/s/yxz1ic4kxtdtgpva5hcmlom9ixfl3v3c.
The layout of current complete set of files is given below. If a file is missing, the appropriate performance test is skipped.
QMC_DATA/C-graphite/lda.pwscf.h5
QMC_DATA/C-molecule/C12-e48-pp.h5
QMC_DATA/C-molecule/C12-e72-ae.h5
QMC_DATA/C-molecule/C18-e108-ae.h5
QMC_DATA/C-molecule/C18-e72-pp.h5
QMC_DATA/C-molecule/C24-e144-ae.h5
QMC_DATA/C-molecule/C24-e96-pp.h5
QMC_DATA/C-molecule/C30-e120-pp.h5
QMC_DATA/C-molecule/C30-e180-ae.h5
QMC_DATA/C-molecule/C60-e240-pp.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S1.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S2.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S4.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S8.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S16.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S32.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S64.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S128.h5
QMC_DATA/NiO/NiO-fcc-supertwist111-supershift000-S256.h5
QMC_DATA/NiO/NiO_afm_fcidump.h5
QMC_DATA/NiO/NiO_afm_wfn.dat
QMC_DATA/NiO/NiO_nm_choldump.h5
Configure and build using CMake and make¶
To configure and build QMCPACK, move to build directory, run CMake, and make
cd build
cmake ..
make -j 8
As you will have gathered, CMake encourages “out of source” builds, where all the files for a specific build configuration reside in their own directory separate from the source files. This allows multiple builds to be created from the same source files, which is very useful when the file system is shared between different systems. You can also build versions with different settings (e.g., QMC_COMPLEX) and different compiler settings. The build directory does not have to be called build—use something descriptive such as build_machinename or build_complex. The “..” in the CMake line refers to the directory containing CMakeLists.txt. Update the “..” for other build directory locations.
Example configure and build¶
Set the environments (the examples below assume bash, Intel compilers, and MKL library)
export CXX=icpc export CC=icc export MKL_ROOT=/usr/local/intel/mkl/10.0.3.020 export HDF5_ROOT=/usr/local export BOOST_ROOT=/usr/local/boost export FFTW_HOME=/usr/local/fftw
Move to build directory, run CMake, and make
cd build cmake -D CMAKE_BUILD_TYPE=Release .. make -j 8
Build scripts¶
We recommended creating a helper script that contains the configure line for CMake. This is particularly useful when avoiding environment variables, packages are installed in custom locations, or the configure line is long or complex. In this case it is also recommended to add “rm -rf CMake*” before the configure line to remove existing CMake configure files to ensure a fresh configure each time the script is called. Deleting all the files in the build directory is also acceptable. If you do so we recommend adding some sanity checks in case the script is run from the wrong directory (e.g., checking for the existence of some QMCPACK files).
Some build script examples for different systems are given in the config directory. For example, on Cray systems these scripts might load the appropriate modules to set the appropriate programming environment, specific library versions, etc.
An example script build.sh is given below. It is much more complex than usually needed for comprehensiveness:
export CXX=mpic++
export CC=mpicc
export ACML_HOME=/opt/acml-5.3.1/gfortran64
export HDF5_ROOT=/opt/hdf5
export BOOST_ROOT=/opt/boost
rm -rf CMake*
cmake \
-D CMAKE_BUILD_TYPE=Debug \
-D LIBXML2_INCLUDE_DIR=/usr/include/libxml2 \
-D LIBXML2_LIBRARY=/usr/lib/x86_64-linux-gnu/libxml2.so \
-D FFTW_INCLUDE_DIRS=/usr/include \
-D FFTW_LIBRARY_DIRS=/usr/lib/x86_64-linux-gnu \
-D QMC_EXTRA_LIBS="-ldl ${ACML_HOME}/lib/libacml.a -lgfortran" \
-D QMC_DATA=/projects/QMCPACK/qmc-data \
..
Using vendor-optimized numerical libraries (e.g., Intel MKL)¶
Although QMC does not make extensive use of linear algebra, use of vendor-optimized libraries is strongly recommended for highest performance. BLAS routines are used in the Slater determinant update, the VMC wavefunction optimizer, and to apply orbital coefficients in local basis calculations. Vectorized math functions are also beneficial (e.g., for the phase factor computation in solid-state calculations). CMake is generally successful in finding these libraries, but specific combinations can require additional hints, as described in the following:
Using Intel MKL with non-Intel compilers¶
To use Intel MKL with, e.g. an MPICH wrapped gcc:
cmake \
-DCMAKE_C_COMPILER=mpicc -DCMAKE_CXX_COMPILER=mpicxx \
-DMKL_ROOT=YOUR_INTEL_MKL_ROOT_DIRECTORY \
..
MKL_ROOT is only necessary when MKL is not auto-detected successfully or a particular MKL installation is desired. YOUR_INTEL_MKL_ROOT_DIRECTORY is the directory containing the MKL bin, examples, and lib directories (etc.) and is often /opt/intel/mkl.
Serial or multithreaded library¶
Vendors might provide both serial and multithreaded versions of their libraries. Using the right version is critical to QMCPACK performance. QMCPACK makes calls from both inside and outside threaded regions. When being called from outside an OpenMP parallel region, the multithreaded version is preferred for the possibility of using all the available cores. When being called from every thread inside an OpenMP parallel region, the serial version is preferred for not oversubscribing the cores. Fortunately, nowadays the multithreaded versions of many vendor libraries (MKL, ESSL) are OpenMP aware. They use only one thread when being called inside an OpenMP parallel region. This behavior meets exactly both QMCPACK needs and thus is preferred. If the multithreaded version does not provide this feature of dynamically adjusting the number of threads, the serial version is preferred. In addition, thread safety is required no matter which version is used.
Cross compiling¶
Cross compiling is often difficult but is required on supercomputers with distinct host and compute processor generations or architectures. QMCPACK tried to do its best with CMake to facilitate cross compiling.
On a machine using a Cray programming environment, we rely on compiler wrappers provided by Cray to correctly set architecture-specific flags. Please also add
-DCMAKE_SYSTEM_NAME=CrayLinuxEnvironmentto cmake. The CMake configure log should indicate that a Cray machine was detected.If not on a Cray machine, by default we assume building for the host architecture (e.g., -xHost is added for the Intel compiler and -march=native is added for GNU/Clang compilers).
If -x/-ax or -march is specified by the user in CMAKE_C_FLAGS and CMAKE_CXX_FLAGS, we respect the user’s intention and do not add any architecture-specific flags.
The general strategy for cross compiling should therefore be to
manually set CMAKE_C_FLAGS and CMAKE_CXX_FLAGS for the target
architecture. Using make VERBOSE=1 is a useful way to check the
final compilation options. If on a Cray machine, selection of the
appropriate programming environment should be sufficient.
Installation instructions for common workstations and supercomputers¶
This section describes how to build QMCPACK on various common systems including multiple Linux distributions, Apple OS X, and various supercomputers. The examples should serve as good starting points for building QMCPACK on similar machines. For example, the software environment on modern Crays is very consistent. Note that updates to operating systems and system software might require small modifications to these recipes. See How to build the fastest executable version of QMCPACK for key points to check to obtain highest performance and Troubleshooting the installation for troubleshooting hints.
Installing on Ubuntu Linux or other apt-get–based distributions¶
The following is designed to obtain a working QMCPACK build on, for example, a student laptop, starting from a basic Linux installation with none of the developer tools installed. Fortunately, all the required packages are available in the default repositories making for a quick installation. Note that for convenience we use a generic BLAS. For production, a platform-optimized BLAS should be used.
apt-get cmake g++ openmpi-bin libopenmpi-dev libboost-dev
apt-get libatlas-base-dev liblapack-dev libhdf5-dev libxml2-dev fftw3-dev
export CXX=mpiCC
cd build
cmake ..
make -j 8
ls -l bin/qmcpack
For qmca and other tools to function, we install some Python libraries:
sudo apt-get install python-numpy python-matplotlib
Installing on CentOS Linux or other yum-based distributions¶
The following is designed to obtain a working QMCPACK build on, for example, a student laptop, starting from a basic Linux installation with none of the developer tools installed. CentOS 7 (Red Hat compatible) is using gcc 4.8.2. The installation is complicated only by the need to install another repository to obtain HDF5 packages that are not available by default. Note that for convenience we use a generic BLAS. For production, a platform-optimized BLAS should be used.
sudo yum install make cmake gcc gcc-c++ openmpi openmpi-devel fftw fftw-devel \
boost boost-devel libxml2 libxml2-devel
sudo yum install blas-devel lapack-devel atlas-devel
module load mpi
To set up repoforge as a source for the HDF5 package, go to http://repoforge.org/use. Install the appropriate up-to-date release package for your operating system. By default, CentOS Firefox will offer to run the installer. The CentOS 6.5 settings were still usable for HDF5 on CentOS 7 in 2016, but use CentOS 7 versions when they become available.
sudo yum install hdf5 hdf5-devel
To build QMCPACK:
module load mpi/openmpi-x86_64
which mpirun
# Sanity check; should print something like /usr/lib64/openmpi/bin/mpirun
export CXX=mpiCC
cd build
cmake ..
make -j 8
ls -l bin/qmcpack
Installing on Mac OS X using Macports¶
These instructions assume a fresh installation of macports and use the gcc 10.2 compiler.
Follow the Macports install instructions at https://www.macports.org/.
Install Xcode and the Xcode Command Line Tools.
Agree to Xcode license in Terminal: sudo xcodebuild -license.
Install MacPorts for your version of OS X.
We recommend to make sure macports is updated:
sudo port -v selfupdate # Required for macports first run, recommended in general
sudo port upgrade outdated # Recommended
Install the required tools. For thoroughness we include the current full set of python dependencies. Some of the tests will be skipped if not all are available.
sudo port install gcc10
sudo port select gcc mp-gcc10
sudo port install openmpi-devel-gcc10
sudo port select --set mpi openmpi-devel-gcc10-fortran
sudo port install fftw-3 +gcc10
sudo port install libxml2
sudo port install cmake
sudo port install boost +gcc10
sudo port install hdf5 +gcc10
sudo port install python38
sudo port select --set python python38
sudo port select --set python3 python38
sudo port install py38-numpy +gcc10
sudo port select --set cython cython38
sudo port install py38-scipy +gcc10
sudo port install py38-h5py +gcc10
sudo port install py38-pandas
sudo port install py38-lxml
sudo port install py38-matplotlib #For graphical plots with qmca
QMCPACK build:
cd build
cmake -DCMAKE_C_COMPILER=mpicc -DCMAKE_CXX_COMPILER=mpiCXX ..
make -j 6 # Adjust for available core count
ls -l bin/qmcpack
Run the deterministic tests:
ctest -R deterministic
This recipe was verified on October 26, 2020, on a Mac running OS X 10.15.7 “Catalina” with macports 2.6.3.
Installing on Mac OS X using Homebrew (brew)¶
Homebrew is a package manager for OS X that provides a convenient route to install all the QMCPACK dependencies. The following recipe will install the latest available versions of each package. This was successfully tested under OS X 10.15.7 “Catalina” on October 26, 2020.
Install Homebrew from http://brew.sh/:
/usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
Install the prerequisites:
brew install gcc # 10.2.0 when tested brew install openmpi brew install cmake brew install fftw brew install boost brew install hdf5 export OMPI_CC=gcc-10 export OMPI_CXX=g++-10
Configure and build QMCPACK:
cmake -DCMAKE_C_COMPILER=/usr/local/bin/mpicc \ -DCMAKE_CXX_COMPILER=/usr/local/bin/mpicxx .. make -j 6 # Adjust for available core count ls -l bin/qmcpack
Run the deterministic tests
ctest -R deterministic
Installing on ALCF Theta, Cray XC40¶
Theta is a 9.65 petaflops system manufactured by Cray with 3,624 compute nodes. Each node features a second-generation Intel Xeon Phi 7230 processor and 192 GB DDR4 RAM.
export CRAYPE_LINK_TYPE=dynamic
module load cmake/3.16.2
module unload cray-libsci
module load cray-hdf5-parallel
module load gcc # Make C++ 14 standard library available to the Intel compiler
export BOOST_ROOT=/soft/libraries/boost/1.64.0/intel
cmake -DCMAKE_SYSTEM_NAME=CrayLinuxEnvironment ..
make -j 24
ls -l bin/qmcpack
Installing on ORNL OLCF Summit¶
Summit is an IBM system at the ORNL OLCF built with IBM Power System AC922 nodes. They have two IBM Power 9 processors and six NVIDIA Volta V100 accelerators.
Building QMCPACK¶
Note that these build instructions are preliminary as the software environment is subject to change. As of December 2018, the IBM XL compiler does not support C++14, so we currently use the gnu compiler.
For ease of reproducibility we provide build scripts for Summit.
cd qmcpack
./config/build_olcf_summit.sh
ls bin
Building Quantum Espresso¶
We provide a build script for the v6.4.1 release of Quantum Espresso (QE). The following can be used to build a CPU version of QE on Summit, placing the script in the external_codes/quantum_espresso directory.
cd external_codes/quantum_espresso
./build_qe_olcf_summit.sh
Note that performance is not yet optimized although vendor libraries are used. Alternatively, the wavefunction files can be generated on another system and the converted HDF5 files copied over.
Installing on NERSC Cori, Haswell Partition, Cray XC40¶
Cori is a Cray XC40 that includes 16-core Intel “Haswell” nodes installed at NERSC. In the following example, the source code is cloned in $HOME/qmc/git_QMCPACK and QMCPACK is built in the scratch space.
mkdir $HOME/qmc
mkdir $HOME/qmc/git_QMCPACK
cd $HOME/qmc_git_QMCPACK
git clone https://github.com/QMCPACK/qmcpack.git
cd qmcpack
git checkout v3.7.0 # Edit for desired version
export CRAYPE_LINK_TYPE=dynamic
module unload cray-libsci
module load boost/1.70.0
module load cray-hdf5-parallel
module load cmake/3.14.4
module load gcc/8.3.0 # Make C++ 14 standard library available to the Intel compiler
cd $SCRATCH
mkdir build_cori_hsw
cd build_cori_hsw
cmake -DQMC_SYMLINK_TEST_FILES=0 -DCMAKE_SYSTEM_NAME=CrayLinuxEnvironment $HOME/qmc/git_QMCPACK/qmcpack/
nice make -j 8
ls -l bin/qmcpack
When the preceding was tested on June 15, 2020, the following module and software versions were present:
build_cori_hsw> module list
Currently Loaded Modulefiles:
1) modules/3.2.11.4 13) xpmem/2.2.20-7.0.1.1_4.8__g0475745.ari
2) nsg/1.2.0 14) job/2.2.4-7.0.1.1_3.34__g36b56f4.ari
3) altd/2.0 15) dvs/2.12_2.2.156-7.0.1.1_8.6__g5aab709e
4) darshan/3.1.7 16) alps/6.6.57-7.0.1.1_5.10__g1b735148.ari
5) intel/19.0.3.199 17) rca/2.2.20-7.0.1.1_4.42__g8e3fb5b.ari
6) craype-network-aries 18) atp/2.1.3
7) craype/2.6.2 19) PrgEnv-intel/6.0.5
8) udreg/2.3.2-7.0.1.1_3.29__g8175d3d.ari 20) craype-haswell
9) ugni/6.0.14.0-7.0.1.1_7.32__ge78e5b0.ari 21) cray-mpich/7.7.10
10) pmi/5.0.14 22) craype-hugepages2M
11) dmapp/7.1.1-7.0.1.1_4.43__g38cf134.ari 23) gcc/8.3.0
12) gni-headers/5.0.12.0-7.0.1.1_6.27__g3b1768f.ari 24) cmake/3.14.4
The following slurm job file can be used to run the tests:
#!/bin/bash
#SBATCH --qos=debug
#SBATCH --time=00:10:00
#SBATCH --nodes=1
#SBATCH --tasks-per-node=32
#SBATCH --constraint=haswell
echo --- Start `date`
echo --- Working directory: `pwd`
ctest -VV -R deterministic
echo --- End `date`
Installing on NERSC Cori, Xeon Phi KNL partition, Cray XC40¶
Cori is a Cray XC40 that includes Intel Xeon Phi Knight’s Landing (KNL) nodes. The following build recipe ensures that the code generation is appropriate for the KNL nodes. The source is assumed to be in $HOME/qmc/git_QMCPACK/qmcpack as per the Haswell example.
export CRAYPE_LINK_TYPE=dynamic
module swap craype-haswell craype-mic-knl # Only difference between Haswell and KNL recipes
module unload cray-libsci
module load boost/1.70.0
module load cray-hdf5-parallel
module load cmake/3.14.4
module load gcc/8.3.0 # Make C++ 14 standard library available to the Intel compiler
cd $SCRATCH
mkdir build_cori_knl
cd build_cori_knl
cmake -DQMC_SYMLINK_TEST_FILES=0 -DCMAKE_SYSTEM_NAME=CrayLinuxEnvironment $HOME/qmc/git_QMCPACK/qmcpack/
nice make -j 8
ls -l bin/qmcpack
When the preceding was tested on June 15, 2020, the following module and software versions were present:
build_cori_knl> module list
Currently Loaded Modulefiles:
1) modules/3.2.11.4 13) xpmem/2.2.20-7.0.1.1_4.8__g0475745.ari
2) nsg/1.2.0 14) job/2.2.4-7.0.1.1_3.34__g36b56f4.ari
3) altd/2.0 15) dvs/2.12_2.2.156-7.0.1.1_8.6__g5aab709e
4) darshan/3.1.7 16) alps/6.6.57-7.0.1.1_5.10__g1b735148.ari
5) intel/19.0.3.199 17) rca/2.2.20-7.0.1.1_4.42__g8e3fb5b.ari
6) craype-network-aries 18) atp/2.1.3
7) craype/2.6.2 19) PrgEnv-intel/6.0.5
8) udreg/2.3.2-7.0.1.1_3.29__g8175d3d.ari 20) craype-mic-knl
9) ugni/6.0.14.0-7.0.1.1_7.32__ge78e5b0.ari 21) cray-mpich/7.7.10
10) pmi/5.0.14 22) craype-hugepages2M
11) dmapp/7.1.1-7.0.1.1_4.43__g38cf134.ari 23) gcc/8.3.0
12) gni-headers/5.0.12.0-7.0.1.1_6.27__g3b1768f.ari 24) cmake/3.14.4
Installing on systems with ARMv8-based processors¶
The following build recipe was verified using the ‘Arm Compiler for HPC’ on the ANL JLSE Comanche system with Cavium ThunderX2 processors on November 6, 2018.
# load armclang compiler
module load Generic-AArch64/RHEL/7/arm-hpc-compiler/18.4
# load Arm performance libraries
module load ThunderX2CN99/RHEL/7/arm-hpc-compiler-18.4/armpl/18.4.0
# define path to pre-installed packages
export HDF5_ROOT=</path/to/hdf5/install/>
export BOOST_ROOT=</path/to/boost/install> # header-only, no need to build
Then using the following command:
mkdir build_armclang
cd build_armclang
cmake -DCMAKE_C_COMPILER=armclang -DCMAKE_CXX_COMPILER=armclang++ -DQMC_MPI=0 \
-DLAPACK_LIBRARIES="-L$ARMPL_DIR/lib -larmpl_mp" \
-DFFTW_INCLUDE_DIR="$ARMPL_DIR/include" \
-DFFTW_LIBRARIES="$ARMPL_DIR/lib/libarmpl_mp.a" \
..
make -j 56
Note that armclang is recognized as an ‘unknown’ compiler by CMake v3.13* and below. In this case, we need to force it as clang to apply necessary flags. To do so, pass the following additionals option to CMake:
-DCMAKE_C_COMPILER_ID=Clang -DCMAKE_CXX_COMPILER_ID=Clang \
-DCMAKE_CXX_COMPILER_VERSION=5.0 -DCMAKE_CXX_STANDARD_COMPUTED_DEFAULT=98 \
Installing on Windows¶
Install the Windows Subsystem for Linux and Bash on Windows. Open a bash shell and follow the install directions for Ubuntu in Installing on Ubuntu Linux or other apt-get–based distributions.
Installing via Spack¶
Spack is a package manager for scientific software. One of the primary goals of Spack is to reduce the barrier for users to install scientific software. Spack is intended to work on everything from laptop computers to high-end supercomputers. More information about Spack can be found at https://spack.readthedocs.io/en/latest. The major advantage of installation with Spack is that all dependencies are automatically built, potentially including all the compilers and libraries, and different versions of QMCPACK can easily coexist with each other. The QMCPACK Spack package also knows how to automatically build and patch QE. In principle, QMCPACK can be installed with a single Spack command.
Known limitations¶
The QMCPACK Spack package inherits the limitations of the underlying
Spack infrastructure and its dependencies. The main limitation is that
installation can fail when building a dependency such as HDF5, MPICH,
etc. For spack install qmcpack to succeed, it is very
important to leverage preinstalled packages on your computer or
supercomputer. The other frequently encountered challenge is that the
compiler configuration is nonintuitive. This is especially the case
with the Intel compiler. If you encounter any difficulties, we
recommend testing the Spack compiler configuration on a simpler
packages, e.g. HDF5.
Here are some additional limitations of the QMCPACK Spack package that will be resolved in future releases:
CUDA support in Spack still has some limitations. It will catch only some compiler-CUDA conflicts.
The Intel compiler must find a recent and compatible GCC compiler in its path or one must be explicity set with the
-gcc-nameand-gxx-nameflags.
Setting up the Spack environment¶
Begin by cloning Spack from GitHub and configuring your shell as described at https://spack.readthedocs.io/en/latest/getting_started.html.
The goal of the next several steps is to set up the Spack environment
for building. First, we highly recommend limiting the number of build jobs to
a reasonable value for your machine. This can be
accomplished by modifying your ~/.spack/config.yaml file as follows:
config:
build_jobs: 16
Make sure any existing compilers are properly detected. For many architectures, compilers are properly detected with no additional effort.
your-laptop> spack compilers
==> Available compilers
-- gcc sierra-x86_64 --------------------------------------------
gcc@7.2.0 gcc@6.4.0 gcc@5.5.0 gcc@4.9.4 gcc@4.8.5 gcc@4.7.4 gcc@4.6.4
However, if your compiler is not automatically detected, it is straightforward to add one:
your-laptop> spack compiler add <path-to-compiler>
The Intel compiler, and other commerical compilers like PGI, typically
require extra environment variables to work properly. If you have an
module environment set-up by your system administrators, it is
recommended that you set the module name in
~/.spack/linux/compilers.yaml. Here is an example for the
Intel compiler:
- compiler:
environment:{}
extra_rpaths: []
flags: {}
modules:
- intel/18.0.3
operating_system: ubuntu14.04
paths:
cc: /soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/bin/intel64/icc
cxx: /soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/bin/intel64/icpc
f77: /soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/bin/intel64/ifort
fc: /soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/bin/intel64/ifort
spec: intel@18.0.3
target: x86_64
If a module is not available, you will have to set-up the environment variables manually:
- compiler:
environment:
set:
INTEL_LICENSE_FILE: server@national-lab.doe.gov
extra_rpaths:
['/soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/compiler/lib/intel64',
'/soft/apps/packages/gcc/gcc-6.2.0/lib64']
flags:
cflags: -gcc-name=/soft/apps/packages/gcc/gcc-6.2.0/bin/gcc
fflags: -gcc-name=/soft/apps/packages/gcc/gcc-6.2.0/bin/gcc
cxxflags: -gxx-name=/soft/apps/packages/gcc/gcc-6.2.0/bin/g++
modules: []
operating_system: ubuntu14.04
paths:
cc: /soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/bin/intel64/icc
cxx: /soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/bin/intel64/icpc
f77: /soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/bin/intel64/ifort
fc: /soft/com/packages/intel/18/u3/compilers_and_libraries_2018.3.222/linux/bin/intel64/ifort
spec: intel@18.0.3
target: x86_64
This last step is the most troublesome. Pre-installed packages are not
automatically detected. If vendor optimized libraries are already
installed, you will need to manually add them to your
~/.spack/packages.yaml. For example, this works on Mac OS X
for the Intel MKL package.
your-laptop> cat \~/.spack/packages.yaml
packages:
intel-mkl:
paths:
intel-mkl@2018.0.128: /opt/intel/compilers_and_libraries_2018.0.104/mac/mkl
buildable: False
Some trial-and-error might be involved to set the directories correctly. If you do not include enough of the tree path, Spack will not be able to register the package in its database. More information about system packages can be found at http://spack.readthedocs.io/en/latest/getting_started.html#system-packages.
Beginning with QMCPACK v3.9.0, Python 3.x is required. However,
installing Python with a compiler besides GCC is tricky. We recommend
leveraging your local Python installation by adding an entry in
~/.spack/packages.yaml:
packages:
python:
modules:
python@3.7.4: anaconda3/2019.10
Or if a module is not available
packages:
python:
paths:
python@3.7.4: /nfs/gce/software/custom/linux-ubuntu18.04-x86_64/anaconda3/2019.10/bin/python
buildable: False
Building QMCPACK¶
The QMCPACK Spack package has a number of variants to support different compile time options and different versions of the application. A full list can be displayed by typing:
your laptop> spack info qmcpack
CMakePackage: qmcpack
Description:
QMCPACK, is a modern high-performance open-source Quantum Monte Carlo
(QMC) simulation code.
Homepage: http://www.qmcpack.org/
Tags:
ecp ecp-apps
Preferred version:
3.9.1 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.9.1
Safe versions:
develop [git] https://github.com/QMCPACK/qmcpack.git
3.9.1 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.9.1
3.9.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.9.0
3.8.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.8.0
3.7.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.7.0
3.6.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.6.0
3.5.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.5.0
3.4.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.4.0
3.3.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.3.0
3.2.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.2.0
3.1.1 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.1.1
3.1.0 [git] https://github.com/QMCPACK/qmcpack.git at tag v3.1.0
Variants:
Name [Default] Allowed values Description
build_type [Release] Debug, Release, The build type to build
RelWithDebInfo
afqmc [off] True, False Install with AFQMC support.
NOTE that if used in
combination with CUDA, only
AFQMC will have CUDA.
complex [off] True, False Build the complex (general
twist/k-point) version
cuda [off] True, False Build with CUDA
cuda_arch [none] none, 53, 20, 62, CUDA architecture
60, 61, 50, 75, 70,
72, 32, 52, 30, 35
da [off] True, False Install with support for basic
data analysis tools
gui [off] True, False Install with Matplotlib (long
installation time)
mixed [off] True, False Build the mixed precision
(mixture of single and double
precision) version for gpu and
cpu
mpi [on] True, False Build with MPI support
phdf5 [on] True, False Build with parallel collective
I/O
ppconvert [off] True, False Install with pseudopotential
converter.
qe [on] True, False Install with patched Quantum
Espresso 6.4.0
soa [on] True, False Build with Structure-of-Array
instead of Array-of-Structure
code. Only for CPU codeand
only in mixed precision
timers [off] True, False Build with support for timers
Installation Phases:
cmake build install
Build Dependencies:
blas boost cmake cuda fftw-api hdf5 lapack libxml2 mpi python
Link Dependencies:
blas boost cuda fftw-api hdf5 lapack libxml2 mpi python
Run Dependencies:
py-matplotlib py-numpy quantum-espresso
Virtual Packages:
None
For example, to install the complex-valued version of QMCPACK in mixed-precision use:
your-laptop> spack install qmcpack+mixed+complex%gcc@7.2.0 ^intel-mkl
where
%gcc@7.2.0
specifies the compiler version to be used and
^intel-mkl
specifies that the Intel MKL should be used as the BLAS and LAPACK
provider. The ^ symbol indicates the the package to the right of the
symbol should be used to fulfill the dependency needed by the installation.
It is also possible to run the QMCPACK regression tests as part of the installation process, for example:
your-laptop> spack install --test=root qmcpack+mixed+complex%gcc@7.2.0 ^intel-mkl
will run the unit and short tests. The current behavior of the QMCPACK Spack package is to complete the install as long as all the unit tests pass. If the short tests fail, a warning is issued at the command prompt.
For CUDA, you will need to specify and extra cuda_arch
parameter otherwise, it will default to cuda_arch=61.
your-laptop> spack install qmcpack+cuda%intel@18.0.3 cuda_arch=61 ^intel-mkl
Due to limitations in the Spack CUDA package, if your compiler and CUDA combination conflict, you will need to set a specific verison of CUDA that is compatible with your compiler on the command line. For example,
your-laptop> spack install qmcpack+cuda%intel@18.0.3 cuda_arch=61 ^cuda@10.0.130 ^intel-mkl
Loading QMCPACK into your environment¶
If you already have modules set-up in your enviroment, the Spack modules will be detected automatically. Otherwise, Spack will not automatically find the additional packages. A few additional steps are needed. Please see the main Spack documentation for additional details: https://spack.readthedocs.io/en/latest/module_file_support.html.
Dependencies that need to be compiled with GCC¶
Failing to compile a QMCPACK dependency is the most common reason that a Spack build fails. We recommend that you compile the following dependencies with GCC:
For MPI, using MPICH as the provider, try:
your-laptop> spack install qmcpack%intel@18.0.3 ^boost%gcc ^pkgconf%gcc ^perl%gcc ^libpciaccess%gcc ^cmake%gcc ^findutils%gcc ^m4%gcc
For serial,
your-laptop> spack install qmcpack~mpi%intel@18.0.3 ^boost%gcc ^pkgconf%gcc ^perl%gcc ^cmake%gcc
Installing QMCPACK with Spack on Linux¶
Spack works robustly on the standard flavors of Linux (Ubuntu, CentOS, Ubuntu, etc.) using GCC, Clang, PGI, and Intel compilers.
Installing QMCPACK with Spack on Mac OS X¶
Spack works on Mac OS X but requires installation of a few packages using Homebrew. You will need to install at minimum the GCC compilers, CMake, and pkg-config. The Intel compiler for Mac on OS X is not well supported by Spack packages and will most likely lead to a compile time failure in one of QMCPACK’s dependencies.
Installing QMCPACK with Spack on Cray Supercomputers¶
Spack now works with the Cray environment. To leverage the installed
Cray environment, both a compilers.yaml and
packages.yaml file should be provided by the supercomputing
facility. Additionally, Spack packages compiled by the facility can be
reused by chaining Spack installations
https://spack.readthedocs.io/en/latest/chain.html.
Instructions for DOE supercomputing facilities that support Spack directly will be forthcoming.
Installing Quantum-Espresso with Spack¶
More information about the QE Spack package can be obtained directly from Spack
spack info quantum-espresso
There are many variants available for QE, most, but not all, are compatible with QMCPACK patch. Here is a minimalistic example of the Spack installation command that needs to be invoked:
your-laptop> spack install quantum-espresso+qmcpack~patch@6.4.1%gcc hdf5=parallel
The ~ decorator means deactivate the patch
variant. This refers not to the QMCPACK patch, but to the upstream
patching that is present for some versions of QE. These upstream QE
patches fix specific critical autoconf/configure fixes. Unfortunately,
some of these QE upstream patches are incompatible with the QMCPACK
patch. Note that the Spack package will prevent you from installing
incompatible variants and will emit an error message explaining the
nature of the incompatibility.
A serial (no MPI) installation is also available, but the Spack installation command is non-intuitive for Spack newcomers:
your-laptop> spack install quantum-espresso+qmcpack~patch~mpi~scalapack@6.4.1%gcc hdf5=serial
QE Spack package is well tested with GCC and Intel compilers, but will not work with the PGI compiler or in a cross-compile environment.
Reporting Bugs¶
Bugs with the QMCPACK Spack package should be filed at the main GitHub Spack repo https://github.com/spack/spack/issues.
In the GitHub issue, include @naromero77 to get the attention
of our developer.
Testing and validation of QMCPACK¶
We strongly encourage running the included tests each time QMCPACK is built. A range of unit and integration tests ensure that the code behaves as expected and that results are consistent with known-good mean-field, quantum chemical, and historical QMC results.
The tests include the following:
Unit tests: to check fundamental behavior. These should always pass.
Stochastic integration tests: to check computed results from the Monte Carlo methods. These might fail statistically, but rarely because of the use of three sigma level statistics. These tests are further split into “short” tests, which have just sufficient length to have valid statistics, and “long” tests, to check behavior to higher statistical accuracy.
Converter tests: to check conversion of trial wavefunctions from codes such as QE and GAMESS to QMCPACK’s formats. These should always pass.
Workflow tests: in the case of QE, we test the entire cycle of DFT calculation, trial wavefunction conversion, and a subsequent VMC run.
Performance: to help performance monitoring. Only the timing of these runs is relevant.
The test types are differentiated by prefixes in their names, for example, short-LiH_dimer_ae_vmc_hf_noj_16-1 indicates a short VMC test
for the LiH dime.
QMCPACK also includes tests for developmental features and features that are unsupported on certain platforms. To indicate these, tests that are unstable are labeled with the CTest label “unstable.” For example, they are unreliable, unsupported, or known to fail from partial implementation or bugs.
When installing QMCPACK you should run at least the unit tests:
ctest -R unit
These tests take only a few seconds to run. All should pass. A failure here could indicate a major problem with the installation.
A wider range of deterministic integration tests are being developed. The goal is to test much more of QMCPACK than the unit tests do and to do so in a manner that is reproducible across platforms. All of these should eventually pass 100% reliably and quickly. At present, some fail on some platforms and for certain build types.
ctest -R deterministic -LE unstable
If time allows, the “short” stochastic tests should also be run. The short tests take a few minutes each on a 16-core machine—about 1 hour total depending on the platform. You can run these tests using the following command in the build directory:
ctest -R short -LE unstable # Run the tests with "short" in their name.
# Exclude any known unstable tests.
The output should be similar to the following:
Test project build_gcc
Start 1: short-LiH_dimer_ae-vmc_hf_noj-16-1
1/44 Test #1: short-LiH_dimer_ae-vmc_hf_noj-16-1 .............. Passed 11.20 sec
Start 2: short-LiH_dimer_ae-vmc_hf_noj-16-1-kinetic
2/44 Test #2: short-LiH_dimer_ae-vmc_hf_noj-16-1-kinetic ...... Passed 0.13 sec
..
42/44 Test #42: short-monoO_1x1x1_pp-vmc_sdj-1-16 ............... Passed 10.02 sec
Start 43: short-monoO_1x1x1_pp-vmc_sdj-1-16-totenergy
43/44 Test #43: short-monoO_1x1x1_pp-vmc_sdj-1-16-totenergy ..... Passed 0.08 sec
Start 44: short-monoO_1x1x1_pp-vmc_sdj-1-16-samples
44/44 Test #44: short-monoO_1x1x1_pp-vmc_sdj-1-16-samples ....... Passed 0.08 sec
100% tests passed, 0 tests failed out of 44
Total Test time (real) = 167.14 sec
Note that the number of tests run varies between the standard, complex, and GPU compilations. These tests should pass with three sigma reliability. That is, they should nearly always pass, and when rerunning a failed test it should usually pass. Overly frequent failures suggest a problem that should be addressed before any scientific production.
The full set of tests consist of significantly longer versions of the short tests, as well as tests of the conversion utilities. The runs require several hours each for improved statistics and a much more stringent test of the code. To run all the tests, simply run CTest in the build directory:
ctest -LE unstable # Run all the stable tests. This will take several hours.
You can also run verbose tests, which direct the QMCPACK output to the standard output:
ctest -V -R short # Verbose short tests
The test system includes specific tests for the complex version of the code.
The input data files for the tests are located in the tests directory.
The system-level test directories are grouped into heg, molecules, and solids, with particular physical systems under each (for example molecules/H4_ae 1 ).
Under each physical system directory there might be tests for multiple QMC methods or parameter variations.
The numerical comparisons and test definitions are in the CMakeLists.txt file in each physical system directory.
If all the QMC tests fail it is likely
that the appropriate mpiexec (or mpirun, aprun, srun, jsrun) is not being
called or found. If the QMC runs appear to work but all the other
tests fail, it is possible that Python is not working on your system. We suggest checking some of the test console output in build/Testing/Temporary/LastTest.log
or the output files under build/tests/.
Note that because most of the tests are very small, consisting of only a few electrons, the performance is not representative of larger calculations. For example, although the calculations might fit in cache, there will be essentially no vectorization because of the small electron counts. These tests should therefore not be used for any benchmarking or performance analysis. Example runs that can be used for testing performance are described in Performance tests.
Deterministic and unit tests¶
QMCPACK has a set of deterministic tests, predominantly unit tests. All of these tests can be run with the following command (in the build directory):
ctest -R deterministic -LE unstable
These tests should always pass. Failure could indicate a major problem with the compiler, compiler settings, or a linked library that would give incorrect results.
The output should look similar to the following:
Test project qmcpack/build
Start 1: unit_test_numerics
1/11 Test #1: unit_test_numerics ............... Passed 0.06 sec
Start 2: unit_test_utilities
2/11 Test #2: unit_test_utilities .............. Passed 0.02 sec
Start 3: unit_test_einspline
...
10/11 Test #10: unit_test_hamiltonian ............ Passed 1.88 sec
Start 11: unit_test_drivers
11/11 Test #11: unit_test_drivers ................ Passed 0.01 sec
100% tests passed, 0 tests failed out of 11
Label Time Summary:
unit = 2.20 sec
Total Test time (real) = 2.31 sec
Individual unit test executables can be found in build/tests/bin.
The source for the unit tests is located in the tests directory under each directory in src (e.g. src/QMCWavefunctions/tests).
See Unit Testing for more details about unit tests.
Integration tests with Quantum Espresso¶
As described in Installing and patching Quantum ESPRESSO, it is possible to test entire workflows of trial wavefunction generation, conversion, and eventual QMC calculation. A patched QE must be installed so that the pw2qmcpack converter is available.
By adding -D QE_BIN=your_QE_binary_path in the CMake command line when building your QMCPACK,
tests named with the “qe-” prefix will be included in the test set of your build.
You can test the whole pw > pw2qmcpack > qmcpack workflow by
ctest -R qe
This provides a very solid test of the entire QMC toolchain for plane wave–generated wavefunctions.
Performance tests¶
Performance tests representative of real research runs are included in the tests/performance directory. They can be used for benchmarking, comparing machine performance, or assessing optimizations. This is in contrast to the majority of the conventional integration tests in which the particle counts are too small to be representative. Care is still needed to remove initialization, I/O, and compute a representative performance measure.
The CTest integration is sufficient to run the benchmarks and measure relative performance from version to version of QMCPACK and to assess proposed code changes. Performance tests are prefixed with “performance.” To obtain the highest performance on a particular platform, you must run the benchmarks in a standalone manner and tune thread counts, placement, walker count (etc.). This is essential to fairly compare different machines. Check with the developers if you are unsure of what is a fair change.
For the largest problem sizes, the initialization of spline orbitals might
take a large portion of overall runtime. When QMCPACK is run at scale,
the initialization is fast because it is fully
parallelized. However, the performance tests usually run on a single node.
Consider running QMCPACK once with save_coefs="yes" XML input tag
added to the line of ‘determinantset’ to save the converted spline
coefficients to the disk and load them for later runs in the same folder.
See Spline basis sets for more information.
The delayed update algorithm in Single determinant wavefunctons significantly changes the performance characteristics of QMCPACK. A parameter scan of the maximal number of delays specific to every architecture and problem size is required to achieve the best performance.
NiO performance tests¶
Follow the instructions in tests/performance/NiO/README to enable and run the NiO tests.
The NiO tests are for bulk supercells of varying size. The QMC runs consist of short blocks of (1) VMC
without drift (2) VMC with drift term included, and (3) DMC with
constant population. The tests use spline wavefunctions that must be
downloaded as described in the README file because of their large size. You
will need to set -DQMC_DATA=YOUR_DATA_FOLDER
when running CMake as
described in the README file.
Two sets of wavefunction are tested: spline orbitals with one- and two-body Jastrow functions and a more complex form with an additional three-body Jastrow function. The Jastrows are the same for each run and are not reoptimized, as might be done for research purposes. Runs in the hundreds of electrons up to low thousands of electrons are representative of research runs performed in 2017. The largest runs target future machines and require very large memory.
Performance test name |
Historical name |
Atoms |
Electrons |
Electrons/spin |
|---|---|---|---|---|
performance-NiO-cpu-a32-e384 |
S8 |
32 |
384 |
192 |
performance-NiO-cpu-a64-e768 |
S16 |
64 |
768 |
384 |
performance-NiO-cpu-a128-e1536 |
S32 |
128 |
1536 |
768 |
performance-NiO-cpu-a256-e3072 |
S64 |
256 |
3072 |
1536 |
performance-NiO-cpu-a512-e6144 |
S128 |
512 |
6144 |
3072 |
performance-NiO-cpu-a1024-e12288 |
S256 |
1024 |
12288 |
6144 |
Troubleshooting tests¶
CTest reports briefly pass or fail of tests in printout and also collects all the standard outputs to help investigating how tests fail.
If the CTest execution is completed, look at Testing/Temporary/LastTest.log.
If you manually stop the testing (ctrl+c), look at Testing/Temporary/LastTest.log.tmp.
You can locate the failing tests by searching for the key word “Fail.”
Slow testing with OpenMPI¶
OpenMPI has a default binding policy that makes all the threads run on a single core during testing when there are two or fewer MPI ranks. This significantly increases testing time. If you are authorized to change the default setting, you can just add “hwloc_base_binding_policy=none” in /etc/openmpi/openmpi-mca-params.conf.
Automated testing of QMCPACK¶
The QMCPACK developers run automatic tests of QMCPACK on several different computer systems, many on a continuous basis. See the reports at https://cdash.qmcpack.org/CDash/index.php?project=QMCPACK. The combinations that are currently tested can be seen on CDash and are also listed in https://github.com/QMCPACK/qmcpack/blob/develop/README.md. They include GCC, Clang, Intel, and PGI compilers in combinations with various library versions and different MPI implementations. NVIDIA GPUs are also tested.
Building ppconvert, a pseudopotential format converter¶
QMCPACK includes a utility—ppconvert—to convert between different pseudopotential formats. Examples include effective core potential formats (in Gaussians), the UPF format used by QE, and the XML format used by QMCPACK itself. The utility also enables the atomic orbitals to be recomputed via a numerical density functional calculation if they need to be reconstructed for use in an electronic structure calculation. Use of ppconvert is an expert feature and discouraged for casual use: a poor choice of orbitals for the creation of projectors in UPF can introduce severe errors and inaccuracies.
Example test results for QMCPACK showing data for a workstation (Intel, GCC, both CPU and GPU builds) and for two ORNL supercomputers. In this example, four errors were found. This dashboard is accessible at https://cdash.qmcpack.org¶
Installing and patching Quantum ESPRESSO¶
For trial wavefunctions obtained in a plane-wave basis, we mainly support QE. Note that ABINIT and QBox were supported historically and could be reactivated.
QE stores wavefunctions in a nonstandard internal “save” format. To convert these to a conventional HDF5 format file we have developed a converter—pw2qmcpack—which is an add-on to the QE distribution.
To simplify the process of patching QE we have developed a script that will automatically download and patch the source code. The patches are specific to each version. For example, to download and patch QE v6.3:
cd external_codes/quantum_espresso
./download_and_patch_qe6.3.sh
After running the patch, you must configure QE with the HDF5 capability enabled in either way:
If your system already has HDF5 installed with Fortran, use the -{}-with-hdf5 configuration option.
cd qe-6.3 ./configure --with-hdf5=/opt/local # Specify HDF5 base directoryCheck the end of the configure output if HDF5 libraries are found properly. If not, either install a complete library or use the other scheme. If using a parallel HDF5 library, be sure to use the same MPI with QE as used to build the parallel HDF5 library.
Currently, HDF5 support in QE itself is preliminary. To enable use of pw2qmcpack but use the old non-HDF5 I/O within QE, replace
-D__HDF5with{-D__HDF5_C}in make.inc.
If your system has HDF5 with C only, manually edit make.inc by adding
-D__HDF5_Cand-DH5_USE_16_APIinDFLAGSand provide include and library path inIFLAGSandHDF5_LIB.
The complete process is described in external_codes/quantum_espresso/README.
The tests involving pw.x and pw2qmcpack.x have been integrated into the test suite of QMCPACK.
By adding -D QE_BIN=your_QE_binary_path in the CMake command line when building your QMCPACK,
tests named with the “qe-” prefix will be included in the test set of your build.
You can test the whole pw > pw2qmcpack > qmcpack workflow by
ctest -R qe
See Integration tests with Quantum Espresso and the testing section for more details.
How to build the fastest executable version of QMCPACK¶
To build the fastest version of QMCPACK we recommend the following:
Use the latest C++ compilers available for your system. Substantial gains have been made optimizing C++ in recent years.
Use a vendor-optimized BLAS library such as Intel MKL and AMD ACML. Although QMC does not make extensive use of linear algebra, it is used in the VMC wavefunction optimizer to apply the orbital coefficients in local basis calculations and in the Slater determinant update.
Use a vector math library such as Intel VML. For periodic calculations, the calculation of the structure factor and Ewald potential benefit from vectorized evaluation of sin and cos. Currently we only autodetect Intel VML, as provided with MKL, but support for MASSV and AMD LibM is included via #defines. See, for example, src/Numerics/e2iphi.h. For large supercells, this optimization can gain 10% in performance.
Note that greater speedups of QMC calculations can usually be obtained by carefully choosing the required statistics for each investigation. That is, do not compute smaller error bars than necessary.
Troubleshooting the installation¶
Some tips to help troubleshoot installations of QMCPACK:
First, build QMCPACK on a workstation you control or on any system with a simple and up-to-date set of development tools. You can compare the results of CMake and QMCPACK on this system with any more difficult systems you encounter.
Use up-to-date development software, particularly a recent CMake.
Verify that the compilers and libraries you expect are being configured. It is common to have multiple versions installed. The configure system will stop at the first version it finds, which might not be the most recent. If this occurs, directly specify the appropriate directories and files (Configuration Options). For example,
cmake -DCMAKE_C_COMPILER=/full/path/to/mpicc -DCMAKE_CXX_COMPILER=/full/path/to/mpicxx ..
To monitor the compiler and linker settings, use a verbose build,
make VERBOSE=1. If an individual source file fails to compile you can experiment by hand using the output of the verbose build to reconstruct the full compilation line.
If you still have problems please post to the QMCPACK Google group with full details, or contact a developer.
Footnotes
- 1
The suffix “ae” is short for “all-electron,” and “pp” is short for “pseudopotential.”
Running QMCPACK¶
QMCPACK requires at least one xml input file, and is invoked via:
qmcpack [command line options] <XML input file(s)>
Command line options¶
QMCPACK offers several command line options that affect how calculations are performed. If the flag is absent, then the corresponding option is disabled:
--dryrunValidate the input file without performing the simulation. This is a good way to ensure that QMCPACK will do what you think it will.--enable-timers=none|coarse|medium|fineControl the timer granularity when the build optionENABLE_TIMERSis enabled.helpPrint version information as well as a list of optional command-line arguments.noprintDo not print extra information on Jastrow or pseudopotential. If this flag is not present, QMCPACK will create several.datfiles that contain information about pseudopotentials (one file per PP) and Jastrow factors (one per Jastrow factor). These file might be useful for visual inspection of the Jastrow, for example.--verbosity=low|high|debugControl the output verbosity. The default low verbosity is concise and, for example, does not include all electron or atomic positions for large systems to reduce output size. Use “high” to see this information and more details of initialization, allocations, QMC method settings, etc.versionPrint version information and optional arguments. Same ashelp.
Input files¶
The input is one or more XML file(s), documented in Input file overview.
Output files¶
QMCPACK generates multiple files documented in Output Overview.
Running in parallel with MPI¶
QMCPACK is fully parallelized with MPI. When performing an ensemble job, all the MPI ranks are first equally divided into groups that perform individual QMC calculations. Within one calculation, all the walkers are fully distributed across all the MPI ranks in the group. Since MPI requires distributed memory, there must be at least one MPI per node. To maximize the efficiency, more facts should be taken into account. When using MPI+threads on compute nodes with more than one NUMA domain (e.g., AMD Interlagos CPU on Titan or a node with multiple CPU sockets), it is recommended to place as many MPI ranks as the number of NUMA domains if the memory is sufficient (e.g., one MPI task per socket). On clusters with more than one GPU per node (NVIDIA Tesla K80), it is necessary to use the same number of MPI ranks as the number of GPUs per node to let each MPI rank take one GPU.
Using OpenMP threads¶
Modern processors integrate multiple identical cores even with hardware threads on a single die to increase the total performance and maintain a reasonable power draw. QMCPACK takes advantage of this compute capability by using threads and the OpenMP programming model as well as threaded linear algebra libraries. By default, QMCPACK is always built with OpenMP enabled. When launching calculations, users should instruct QMCPACK to create the right number of threads per MPI rank by specifying environment variable OMP_NUM_THREADS. Assuming one MPI rank per socket, the number of threads should typically be the number of cores on that socket. Even in the GPU-accelerated version, using threads significantly reduces the time spent on the calculations performed by the CPU.
Nested OpenMP threads¶
Nested threading is an advanced feature requiring experienced users to finely tune runtime parameters to reach the best performance.
For small-to-medium problem sizes, using one thread per walker or for multiple walkers is most efficient. This is the default in QMCPACK and achieves the shortest time to solution.
For large problems of at least 1,000 electrons, use of nested OpenMP threading can be enabled to reduce the time to solution further, although at some loss of efficiency. In this scheme multiple threads are used in the computations of each walker. This capability is implemented for some of the key computational kernels: the 3D spline orbital evaluation, certain portions of the distance tables, and implicitly the BLAS calls in the determinant update. Use of the batched nonlocal pseudopotential evaluation is also recommended.
Nested threading is enabled by setting OMP_NUM_THREADS=AA,BB, OMP_MAX_ACTIVE_LEVELS=2 and OMP_NESTED=TRUE where the additional BB is the number of second-level threads. Choosing the thread affinity is critical to the performance.
QMCPACK provides a tool qmc-check-affinity (source file src/QMCTools/check-affinity.cpp for details), which might help users investigate the affinity. Knowledge of how the operating system logical CPU cores (/prco/cpuinfo) are bound to the hardware is also needed.
For example, on Blue Gene/Q with a Clang compiler, the best way to fully use the 16 cores each with 4 hardware threads is
OMP_NESTED=TRUE
OMP_NUM_THREADS=16,4
MAX_ACTIVE_LEVELS=2
OMP_PLACES=threads
OMP_PROC_BIND=spread,close
On Intel Xeon Phi KNL with an Intel compiler, to use 64 cores without using hardware threads:
OMP_NESTED=TRUE
OMP_WAIT_POLICY=ACTIVE
OMP_NUM_THREADS=16,4
MAX_ACTIVE_LEVELS=2
OMP_PLACES=cores
OMP_PROC_BIND=spread,close
KMP_HOT_TEAMS_MODE=1
KMP_HOT_TEAMS_MAX_LEVEL=2
Most multithreaded BLAS/LAPACK libraries do not spawn threads by default when being called from an OpenMP parallel region. See the explanation in Serial or multithreaded library. This results in the use of only a single thread in each second-level thread team for BLAS/LAPACK operations. Some vendor libraries like MKL support using multiple threads when being called from an OpenMP parallel region. One way to enable this feature is using environment variables to override the default behavior. However, this forces all the calls to the library to use the same number of threads. As a result, small function calls are penalized with heavy overhead and heavy function calls are slow for not being able to use more threads. Instead, QMCPACK uses the library APIs to turn on nested threading only at selected performance critical calls. In the case of using a serial library, QMCPACK implements nested threading to distribute the workload wherever necessary. Users do not need to control the threading behavior of the library.
Performance considerations¶
As walkers are the basic units of workload in QMC algorithms, they are loosely coupled and distributed across all the threads. For this reason, the best strategy to run QMCPACK efficiently is to feed enough walkers to the available threads.
In a VMC calculation, the code automatically raises the actual number of walkers per MPI rank to the number of available threads if the user-specified number of walkers is smaller, see “walkers/mpi=XXX” in the VMC output.
In DMC, for typical small to mid-sized calculations choose the total number of walkers to be a significant multiple of the total number of threads (MPI tasks * threads per task). This will ensure a good load balance. e.g., for a calculation on a few nodes with a total 512 threads, using 5120 walkers may keep the load imbalance around 10%. For the very largest calculations, the target number of walkers should be chosen to be slightly smaller than a multiple of the total number of available threads across all the MPI ranks. This will reduce occurrences worse-case load imbalance e.g. where one thread has two walkers while all the others have one.
To achieve better performance, a mixed-precision version (experimental) has been developed in the CPU code. The mixed-precision CPU code uses a mixed of single precision (SP) and double precision (DP) operations, while the default code use DP exclusively. This mixed precision version is more aggressive than the GPU CUDA version in using single precision (SP) operations. The Current implementation uses SP on most calculations, except for matrix inversions and reductions where double precision is required to retain high accuracy. All the constant spline data in wavefunction, pseudopotentials, and Coulomb potentials are initialized in double precision and later stored in single precision. The mixed-precision code is as accurate as the double-precision code up to a certain system size, and may have double the throughput. Cross checking and verification of accuracy is always required but is particularly important above approximately 1,500 electrons.
Memory considerations¶
When using threads, some memory objects are shared by all the threads. Usually these memory objects are read only when the walkers are evolving, for instance the ionic distance table and wavefunction coefficients. If a wavefunction is represented by B-splines, the whole table is shared by all the threads. It usually takes a large chunk of memory when a large primitive cell was used in the simulation. Its actual size is reported as “MEMORY increase XXX MB BsplineSetReader” in the output file. See details about how to reduce it in Spline basis sets.
The other memory objects that are distinct for each walker during random walks need to be associated with individual walkers and cannot be shared. This part of memory grows linearly as the number of walkers per MPI rank. Those objects include wavefunction values (Slater determinants) at given electronic configurations and electron-related distance tables (electron-electron distance table). Those matrices dominate the \(N^2\) scaling of the memory usage per walker.
Running on GPU machines¶
The GPU version for the NVIDIA CUDA platform is fully incorporated into the main source code. Commonly used functionalities for solid-state and molecular systems using B-spline single-particle orbitals are supported. Use of Gaussian basis sets, three-body Jastrow functions, and many observables are not yet supported. A detailed description of the GPU implementation can be found in [EKCS12].
The current GPU implementation assumes one MPI process per GPU. To use nodes with multiple GPUs, use multiple MPI processes per node. Vectorization is achieved over walkers, that is, all walkers are propagated in parallel. In each GPU kernel, loops over electrons, atomic cores, or orbitals are further vectorized to exploit an additional level of parallelism and to allow coalesced memory access.
Performance considerations¶
To run with high performance on GPUs it is crucial to perform some benchmarking runs: the optimum configuration is system size, walker count, and GPU model dependent. The GPU implementation vectorizes operations over multiple walkers, so generally the more walkers that are placed on a GPU, the higher the performance that will be obtained. Performance also increases with electron count, up until the memory on the GPU is exhausted. A good strategy is to perform a short series of VMC runs with walker count increasing in multiples of two. For systems with 100s of electrons, typically 128–256 walkers per GPU use a sufficient number of GPU threads to operate the GPU efficiently and to hide memory-access latency. For smaller systems, thousands of walkers might be required. For QMC algorithms where the number of walkers is fixed such as VMC, choosing a walker count the is a multiple of the number of streaming multiprocessors can be most efficient. For variable population DMC runs, this exact match is not possible.
To achieve better performance, the current GPU implementation uses single-precision operations for most of the calculations. Double precision is used in matrix inversions and the Coulomb interaction to retain high accuracy. The mixed-precision GPU code is as accurate as the double-precision CPU code up to a certain system size. Cross checking and verification of accuracy are encouraged for systems with more than approximately 1,500 electrons. For typical calculations on smaller electron counts, the statistical error bars are much larger then the error introduced by mixed precision.
Memory considerations¶
In the GPU implementation, each walker has a buffer in the GPU’s global memory to store temporary data associated with the wavefunctions. Therefore, the amount of memory available on a GPU limits the number of walkers and eventually the system size that it can process. Additionally, for calculations using B-splines, this data is stored on the GPU in a shared read-only buffer. Often the size of the B-spline data limits the calculations that can be run on the GPU.
If the GPU memory is exhausted, first try reducing the number of walkers per GPU. Coarsening the grids of the B-splines representation (by decreasing the value of the mesh factor in the input file) can also lower the memory usage, at the expense (risk) of obtaining inaccurate results. Proceed with caution if this option has to be considered. It is also possible to distribute the B-spline coefficients table between the host and GPU memory, see option Spline_Size_Limit_MB in Spline basis sets.
- EKCS12
Kenneth P. Esler, Jeongnim Kim, David M. Ceperley, and Luke Shulenburger. Accelerating quantum monte carlo simulations of real materials on gpu clusters. Computing in Science and Engineering, 14(1):40–51, 2012. doi:http://doi.ieeecomputersociety.org/10.1109/MCSE.2010.122.
Units used in QMCPACK¶
Internally, QMCPACK uses atomic units throughout. Unless stated, all inputs and outputs are also in atomic units. For convenience the analysis tools offer conversions to eV, Ry, Angstrom, Bohr, etc.
Input file overview¶
This chapter introduces XML as it is used in the QMCPACK input file. The focus is on the XML file format itself and the general structure of the input file rather than an exhaustive discussion of all keywords and structure elements.
QMCPACK uses XML to represent structured data in its input file. Instead of text blocks like
begin project
id = vmc
series = 0
end project
begin vmc
move = pbyp
blocks = 200
steps = 10
timestep = 0.4
end vmc
QMCPACK input looks like
<project id="vmc" series="0">
</project>
<qmc method="vmc" move="pbyp">
<parameter name="blocks" > 200 </parameter>
<parameter name="steps" > 10 </parameter>
<parameter name="timestep"> 0.4 </parameter>
</qmc>
XML elements start with <element_name>, end with </element_name>}, and can be nested within each other to denote substructure (the trial wavefunction is composed of a Slater determinant and a Jastrow factor, which are each further composed of \(...\)). id and series are attributes of the <project/> element. XML attributes are generally used to represent simple values, like names, integers, or real values. Similar functionality is also commonly provided by <parameter/> elements like those previously shown.
The overall structure of the input file reflects different aspects of the QMC simulation: the simulation cell, particles, trial wavefunction, Hamiltonian, and QMC run parameters. A condensed version of the actual input file is shown as follows:
<?xml version="1.0"?>
<simulation>
<project id="vmc" series="0">
...
</project>
<qmcsystem>
<simulationcell>
...
</simulationcell>
<particleset name="e">
...
</particleset>
<particleset name="ion0">
...
</particleset>
<wavefunction name="psi0" ... >
...
<determinantset>
<slaterdeterminant>
..
</slaterdeterminant>
</determinantset>
<jastrow type="One-Body" ... >
...
</jastrow>
<jastrow type="Two-Body" ... >
...
</jastrow>
</wavefunction>
<hamiltonian name="h0" ... >
<pairpot type="coulomb" name="ElecElec" ... />
<pairpot type="coulomb" name="IonIon" ... />
<pairpot type="pseudo" name="PseudoPot" ... >
...
</pairpot>
</hamiltonian>
</qmcsystem>
<qmc method="vmc" move="pbyp">
<parameter name="warmupSteps"> 20 </parameter>
<parameter name="blocks" > 200 </parameter>
<parameter name="steps" > 10 </parameter>
<parameter name="timestep" > 0.4 </parameter>
</qmc>
</simulation>
The omitted portions ... are more fine-grained inputs such as the axes of the simulation cell, the number of up and down electrons, positions of atomic species, external orbital files, starting Jastrow parameters, and external pseudopotential files.
Project¶
The <project> tag uses the id and series attributes.
The value of id is the first part of the prefix for output file names.
Output file names also contain the series number, starting at the value given by the
series tag. After every <qmc> section, the series value will increment, giving each section a unique prefix.
For the input file shown previously, the output files will start with vmc.s000, for example, vmc.s000.scalar.dat.
If there were another <qmc> section in the input file, the corresponding output files would use the prefix vmc.s001.
Random number initialization¶
The random number generator state is initialized from the random element using the seed attribute:
<random seed="1000"/>
If the random element is not present, or the seed value is negative, the seed will be generated from the current time.
To initialize the many independent random number generators (one per thread and MPI process), the seed value is used (modulo 1024) as a starting index into a list of prime numbers. Entries in this offset list of prime numbers are then used as the seed for the random generator on each thread and process.
If checkpointing is enabled, the random number state is written to an HDF file at the end of each block (suffix: .random.h5).
This file will be read if the mcwalkerset tag is present to perform a restart.
For more information, see the checkpoint element in the QMC methods Quantum Monte Carlo Methods and Checkpoint and restart files on checkpoint and restart files.
Specifying the system to be simulated¶
Specifying the Simulation Cell¶
The simulationcell block specifies the geometry of the cell, how the boundary
conditions should be handled, and how ewald summation should be broken
up.
simulationcell Element:
Parent elements:
qmcsystemChild elements:
None
Attribute:
parameter name
datatype
values
default
description
lattice9 floats
any float
Must be specified
Specification of lattice vectors.
bcondsstring
“p” or “n”
“n n n “
Boundary conditions for each axis.
vacuumfloat
\(\geq 1.0\)
1.0
Vacuum scale.
LR_dim_cutofffloat
float
15
Ewald breakup distance.
LR_tolfloat
float
3e-4
Tolerance in Ha for Ewald ion-ion energy per atom.
An example of a block is given below:
<simulationcell>
<parameter name="lattice">
3.8 0.0 0.0
0.0 3.8 0.0
0.0 0.0 3.8
</parameter>
<parameter name="bconds">
p p p
</parameter>
<parameter name="LR_dim_cutoff"> 20 </parameter>
</simulationcell>
Here, a cubic cell 3.8 bohr on a side will be used. This simulation will use periodic boundary conditions, and the maximum \(k\) vector will be \(20/r_{wigner-seitz}\) of the cell.
Lattice¶
The cell is specified using 3 lattice vectors.
Boundary conditions¶
QMCPACK offers the capability to use a mixture of open and periodic boundary conditions. The parameter expects a single string of three characters separated by spaces, e.g. “p p p” for purely periodic boundary conditions. These characters control the behavior of the \(x\), \(y\), and \(z\), axes, respectively. Non periodic directions must be placed after the periodic ones. Examples of valid include:
“p p p” Periodic boundary conditions. Corresponds to a 3D crystal.
“p p n” Slab geometry. Corresponds to a 2D crystal.
“p n n” Wire geometry. Corresponds to a 1D crystal.
“n n n” Open boundary conditions. Corresponds to an isolated molecule in a vacuum.
Vacuum¶
The vacuum option allows adding a vacuum region in slab or wire boundary
conditions (bconds= p p n or bconds= p n n, respectively). The main use is to save memory with
spline or plane-wave basis trial wavefunctions, because no basis
functions are required inside the vacuum region. For example, a large
vacuum region can be added above and below a graphene sheet without
having to generate the trial wavefunction in such a large box or to have
as many splines as would otherwise be required. Note that the trial
wavefunction must still be generated in a large enough box to
sufficiently reduce periodic interactions in the underlying electronic
structure calculation.
With the vacuum option, the box used for Ewald summation increases along
the axis labeled by a factor of vacuum. Note that all the particles remain in
the original box without altering their positions. i.e. Bond lengths are
not changed by this option. The default value is 1, no change to the
specified axes.
An example of a simulationcell block using is given below. The size of the box along
the z-axis increases from 12 to 18 by the vacuum scale of 1.5.
<simulationcell>
<parameter name="lattice">
3.8 0.0 0.0
0.0 3.8 0.0
0.0 0.0 12.0
</parameter>
<parameter name="bconds">
p p n
</parameter>
<parameter name="vacuum"> 1.5 </parameter>
<parameter name="LR_dim_cutoff"> 20 </parameter>
</simulationcell>
LR_dim_cutoff¶
When using periodic boundary conditions direct calculation of the Coulomb energy is not well behaved. As a result, QMCPACK uses an optimized Ewald summation technique to compute the Coulomb interaction. [NC95]
In the Ewald summation, the energy is broken into short- and long-ranged terms. The short-ranged term is computed directly in real space, while the long-ranged term is computed in reciprocal space. controls where the short-ranged term ends and the long-ranged term begins. The real-space cutoff, reciprocal-space cutoff, and are related via:
where \(r_{c}\) is the Wigner-Seitz radius, and \(k_{c}\) is the length of the maximum \(k\)-vector used in the long-ranged term. Larger values of increase the accuracy of the evaluation. A value of 15 tends to be conservative.
Specifying the particle set¶
The particleset blocks specify the particles in the QMC simulations: their types,
attributes (mass, charge, valence), and positions.
Input specification¶
particleset element:
Parent elements
simulationChild elements
group,attrib
Attribute:
Name
Datatype
Values
Default
Description
name/idText
Any
e
Name of particle set
size\(^o\)Integer
Any
0
Number of particles in set
random\(^o\)Text
Yes/no
No
Randomize starting positions
randomsrc/randomsrc\(^o\)Text
particleset.nameNone
Particle set to randomize
Detailed attribute description¶
Required particleset attributes¶
name/idUnique name for the particle set. Default is “e” for electrons. “i” or “ion0” is typically used for ions. For special cases where an empty particle set is needed, the special name “empty” can be used to bypass the zero-size error check.
Optional particleset attributes¶
sizeNumber of particles in set.
Group element:
Parent elements
particlesetChild elements
parameter,attribAttribute:
Name
Datatype
Values
Default
Description
nameText
Any
e
Name of particle set
size\(^o\)Integer
Any
0
Number of particles in set
mass\(^o\)Real
Any
1
Mass of particles in set
unit\(^o\)Text
au/amu
au
Units for mass of particles
Parameters:
Name
Datatype
Values
Default
Description
chargeReal
Any
0
Charge of particles in set
valenceReal
Any
0
Valence charge of particles in set
atomicnumberInteger
Any
0
Atomic number of particles in set
attrib element:
Parent elements
particleset,groupAttribute:
Name
Datatype
Values
Default
Description
nameString
Any
None
Name of attrib
datatypeString
IntArray, realArray, posArray, stringArray
None
Type of data in attrib
size\(^o\)String
Any
None
Size of data in attrib
randomRandomize starting positions of particles. Each component of each particle’s position is randomized independently in the range of the simulation cell in that component’s direction.randomsrc/random_sourceSpecify source particle set around which to randomize the initial positions of this particle set.
Required name attributes¶
name/idUnique name for the particle set group. Typically, element symbols are used for ions and “u” or “d” for spin-up and spin-down electron groups, respectively.
Optional group attributes¶
massMass of particles in set.unitUnits for mass of particles in set (au[\(m_e\) = 1] or amu[\(\frac{1}{12}m_{\rm ^{12}C}\) = 1]).
Example use cases¶
Particleset elements for ions and electrons randomizing electron start positions.
<particleset name="i" size="2">
<group name="Li">
<parameter name="charge">3.000000</parameter>
<parameter name="valence">3.000000</parameter>
<parameter name="atomicnumber">3.000000</parameter>
</group>
<group name="H">
<parameter name="charge">1.000000</parameter>
<parameter name="valence">1.000000</parameter>
<parameter name="atomicnumber">1.000000</parameter>
</group>
<attrib name="position" datatype="posArray" condition="1">
0.0 0.0 0.0
0.5 0.5 0.5
</attrib>
<attrib name="ionid" datatype="stringArray">
Li H
</attrib>
</particleset>
<particleset name="e" random="yes" randomsrc="i">
<group name="u" size="2">
<parameter name="charge">-1</parameter>
</group>
<group name="d" size="2">
<parameter name="charge">-1</parameter>
</group>
</particleset>
Particleset elements for ions and electrons specifying electron start positions.
<particleset name="e">
<group name="u" size="4">
<parameter name="charge">-1</parameter>
<attrib name="position" datatype="posArray">
2.9151687332e-01 -6.5123272502e-01 -1.2188463918e-01
5.8423636048e-01 4.2730406357e-01 -4.5964306231e-03
3.5228575807e-01 -3.5027014639e-01 5.2644808295e-01
-5.1686250912e-01 -1.6648002292e+00 6.5837023441e-01
</attrib>
</group>
<group name="d" size="4">
<parameter name="charge">-1</parameter>
<attrib name="position" datatype="posArray">
3.1443445436e-01 6.5068682609e-01 -4.0983449009e-02
-3.8686061749e-01 -9.3744432997e-02 -6.0456005388e-01
2.4978241724e-02 -3.2862514649e-02 -7.2266047173e-01
-4.0352404772e-01 1.1927734805e+00 5.5610824921e-01
</attrib>
</group>
</particleset>
<particleset name="ion0" size="3">
<group name="O">
<parameter name="charge">6</parameter>
<parameter name="valence">4</parameter>
<parameter name="atomicnumber">8</parameter>
</group>
<group name="H">
<parameter name="charge">1</parameter>
<parameter name="valence">1</parameter>
<parameter name="atomicnumber">1</parameter>
</group>
<attrib name="position" datatype="posArray">
0.0000000000e+00 0.0000000000e+00 0.0000000000e+00
0.0000000000e+00 -1.4308249289e+00 1.1078707576e+00
0.0000000000e+00 1.4308249289e+00 1.1078707576e+00
</attrib>
<attrib name="ionid" datatype="stringArray">
O H H
</attrib>
</particleset>
Particleset elements for ions specifying positions by ion type.
<particleset name="ion0">
<group name="O" size="1">
<parameter name="charge">6</parameter>
<parameter name="valence">4</parameter>
<parameter name="atomicnumber">8</parameter>
<attrib name="position" datatype="posArray">
0.0000000000e+00 0.0000000000e+00 0.0000000000e+00
</attrib>
</group>
<group name="H" size="2">
<parameter name="charge">1</parameter>
<parameter name="valence">1</parameter>
<parameter name="atomicnumber">1</parameter>
<attrib name="position" datatype="posArray">
0.0000000000e+00 -1.4308249289e+00 1.1078707576e+00
0.0000000000e+00 1.4308249289e+00 1.1078707576e+00
</attrib>
</group>
</particleset>
- NC95
Vincent Natoli and David M. Ceperley. An optimized method for treating long-range potentials. Journal of Computational Physics, 117(1):171 – 178, 1995. URL: http://www.sciencedirect.com/science/article/pii/S0021999185710546, doi:http://dx.doi.org/10.1006/jcph.1995.1054.
Trial wavefunction specification¶
Introduction¶
This section describes the input blocks associated with the specification of the trial wavefunction in a QMCPACK calculation. These sections are contained within the <wavefunction> \(...\) </wavefunction> xml blocks. Users are expected to rely on converters to generate the input blocks described in this section. The converters and the workflows are designed such that input blocks require minimum modifications from users. Unless the workflow requires modification of wavefunction blocks (e.g., setting the cutoff in a multideterminant calculation), only expert users should directly alter them.
The trial wavefunction in QMCPACK has a general product form:
where each \(\Theta_k(\vec{r})\) is a function of the electron coordinates (and possibly ionic coordinates and variational parameters). For problems involving electrons, the overall trial wavefunction must be antisymmetric with respect to electron exchange, so at least one of the functions in the product must be antisymmetric. Notice that, although QMCPACK allows for the construction of arbitrary trial wavefunctions based on the functions implemented in the code (e.g., slater determinants, jastrow functions), the user must make sure that a correct wavefunction is used for the problem at hand. From here on, we assume a standard trial wavefunction for an electronic structure problem
where \(\textit{A}(\vec{r})\)
is one of the antisymmetric functions: (1) slater determinant, (2) multislater determinant, or (3) pfaffian and \(\textit{J}_k\)
is any of the Jastrow functions (described in Jastrow Factors). The antisymmetric functions are built from a set of single particle orbitals (sposet). QMCPACK implements four different types of sposet, described in the following section. Each sposet is designed for a different type of calculation, so their definition and generation varies accordingly.
Single determinant wavefunctons¶
Placing a single determinant for each spin is the most used ansatz for the antisymmetric part of a trial wavefunction.
The input xml block for slaterdeterminant is given in Listing 1. A list of options is given in
Table2.
slaterdeterminant element:
Parent elements |
|
Child elements |
|
Attribute:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
Integer |
>=0 |
1 |
Number of delayed updates. |
|
Text |
yes/no |
yes |
Enable orbital optimization. |
Table 2 Options for the slaterdeterminant xml-block.
<slaterdeterminant delay_rank="32">
<determinant id="updet" size="208">
<occupation mode="ground" spindataset="0">
</occupation>
</determinant>
<determinant id="downdet" size="208">
<occupation mode="ground" spindataset="0">
</occupation>
</determinant>
</slaterdeterminant>
Additional information:
delay_rankThis option enables delayed updates of the Slater matrix inverse when particle-by-particle move is used. By default or ifdelay_rank=0given in the input file, QMCPACK sets 1 for Slater matrices with a leading dimension \(<192\) and 32 otherwise.delay_rank=1uses the Fahy’s variant [FWL90] of the Sherman-Morrison rank-1 update, which is mostly using memory bandwidth-bound BLAS-2 calls. Withdelay_rank>1, the delayed update algorithm [LK18][MDAzevedoL+17] turns most of the computation to compute bound BLAS-3 calls. Tuning this parameter is highly recommended to gain the best performance on medium-to-large problem sizes (\(>200\) electrons). We have seen up to an order of magnitude speedup on large problem sizes. When studying the performance of QMCPACK, a scan of this parameter is required and we recommend starting from 32. The bestdelay_rankgiving the maximal speedup depends on the problem size. Usually the largerdelay_rankcorresponds to a larger problem size. On CPUs,delay_rankmust be chosen as a multiple of SIMD vector length for good performance of BLAS libraries. The bestdelay_rankdepends on the processor microarchitecture. GPU support is under development.
Single-particle orbitals¶
Spline basis sets¶
In this section we describe the use of spline basis sets to expand the sposet.
Spline basis sets are designed to work seamlessly with plane wave DFT code (e.g.,Quantum ESPRESSO as a trial wavefunction generator).
In QMC algorithms, all the SPOs \(\{\phi(\vec{r})\}\) need to be updated every time a single electron moves. Evaluating SPOs takes a very large portion of computation time. In principle, PW basis set can be used to express SPOs directly in QMC, as in DFT. But it introduces an unfavorable scaling because the basis set size increases linearly as the system size. For this reason, it is efficient to use a localized basis with compact support and a good transferability from the plane wave basis.
In particular, 3D tricubic B-splines provide a basis in which only 64 elements are nonzero at any given point in [AlfeG04]. The 1D cubic B-spline is given by
where \(b^{i}(x)\) is the piecewise cubic polynomial basis functions and \(i = \text{floor}(\Delta^{-1} x)\) is the index of the first grid point \(\le x\). Constructing a tensor product in each Cartesian direction, we can represent a 3D orbital as
This allows the rapid evaluation of each orbital in constant time. Furthermore, this basis is systematically improvable with a single spacing parameter so that accuracy is not compromised compared with the plane wave basis.
The use of 3D tricubic B-splines greatly improves computational efficiency. The gain in computation time from a plane wave basis set to an equivalent B-spline basis set becomes increasingly large as the system size grows. On the downside, this computational efficiency comes at the expense of increased memory use, which is easily overcome, however, by the large aggregate memory available per node through OpenMP/MPI hybrid QMC.
The input xml block for the spline SPOs is given in Listing 2. A list of options is given in
table3.
<determinantset type="bspline" source="i" href="pwscf.h5"
tilematrix="1 1 3 1 2 -1 -2 1 0" twistnum="-1" gpu="yes" meshfactor="0.8"
twist="0 0 0" precision="double">
<slaterdeterminant>
<determinant id="updet" size="208">
<occupation mode="ground" spindataset="0">
</occupation>
</determinant>
<determinant id="downdet" size="208">
<occupation mode="ground" spindataset="0">
</occupation>
</determinant>
</slaterdeterminant>
</determinantset>
determinantset element:
Parent elements |
|
Child elements |
|
attribute:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
Text |
Bspline |
Type of |
|
|
Text |
Path to hdf5 file from pw2qmcpack.x. |
||
|
9 integers |
Tiling matrix used to expand supercell. |
||
|
Integer |
Index of the super twist. |
||
|
3 floats |
Super twist. |
||
|
Float |
\(\le 1.0\) |
Grid spacing ratio. |
|
|
Text |
Single/double |
Precision of spline coefficients |
|
|
Text |
Yes/no |
GPU switch. |
|
|
Text |
Yes/no |
No |
Share B-spline table across GPUs. |
|
Integer |
Limit B-spline table size on GPU. |
||
|
Text |
Yes/no |
Yes |
Check norms of orbitals from h5 file. |
|
Text |
Yes/no |
No |
Save the spline coefficients to h5 file. |
|
Text |
Any |
Ion0 |
Particle set with atomic positions. |
Table 3 Options for the determinantset xml-block associated with B-spline single particle orbital sets.
Additional information:
precision. Only effective on CPU versions without mixed precision, “single” is always imposed with mixed precision. Using single precision not only saves memory use but also speeds up the B-spline evaluation. We recommend using single precision since we saw little chance of really compromising the accuracy of calculation.meshfactor. The ratio of actual grid spacing of B-splines used in QMC calculation with respect to the original one calculated from h5. A smaller meshfactor saves memory use but reduces accuracy. The effects are similar to reducing plane wave cutoff in DFT calculations. Use with caution!twistnum. If positive, it is the index. We recommend not taking this way since the indexing might show some uncertainty. If negative, the super twist is referred bytwist.save_coefs. If yes, dump the real-space B-spline coefficient table into an h5 file on the disk. When the orbital transformation from k space to B-spline requires more than the available amount of scratch memory on the compute nodes, users can perform this step on fat nodes and transfer back the h5 file for QMC calculations.gpusharing. If enabled, spline data is shared across multiple GPUs on a given computational node. For example, on a two-GPU-per-node system, each GPU would have half of the orbitals. This enables larger overall spline tables than would normally fit in the memory of individual GPUs to be used, potentially up to the total GPU memory on a node. To obtain high performance, large electron counts or a high-performing CPU-GPU interconnect is required. To use this feature, the following needs to be done:The CUDA Multi-Process Service (MPS) needs to be used (e.g., on Summit/SummitDev use “-alloc_flags gpumps” for bsub). If MPS is not detected, sharing will be disabled.
CUDA_VISIBLE_DEVICES needs to be properly set to control each rank’s visible CUDA devices (e.g., on OLCF Summit/SummitDev one needs to create a resource set containing all GPUs with the respective number of ranks with “jsrun –task-per-rs Ngpus -g Ngpus”).
Spline_Size_Limit_MB. Allows distribution of the B-spline coefficient table between the host and GPU memory. The compute kernels access host memory via zero-copy. Although the performance penalty introduced by it is significant, it allows large calculations to go through.
Gaussian basis tests¶
In this section we describe the use of localized basis sets to expand the sposet. The general form of a single particle orbital in this case is given by:
where \(\{\eta_k(\vec{r})\}\) is a set of M atom-centered basis functions and \(C_{i,k}\) is a coefficient matrix. This should be used in calculations of finite systems employing an atom-centered basis set and is typically generated by the convert4qmc converter. Examples include calculations of molecules using Gaussian basis sets or Slater-type basis functions. Initial support for periodic systems is described in Periodic LCAO for Solids. Even though this section is called “Gaussian basis sets” (by far the most common atom-centered basis set), QMCPACK works with any atom-centered basis set based on either spherical harmonic angular functions or Cartesian angular expansions. The radial functions in the basis set can be expanded in either Gaussian functions, Slater-type functions, or numerical radial functions.
In this section we describe the input sections for the atom-centered basis set and the sposet for a single Slater determinant trial wavefunction. The input sections for multideterminant trial wavefunctions are described in Multideterminant wavefunctions. The basic structure for the input block of a single Slater determinant is given in Listing 3.
A list of options for determinantset associated with this sposet is given in table4.
sposet expanded on an atom-centered basis set.¶ <wavefunction id="psi0" target="e">
<determinantset>
<basisset>
...
</basisset>
<slaterdeterminant>
...
</slaterdeterminant>
</determinantset>
</wavefunction>
The definition of the set of atom-centered basis functions is given by the basisset block, and the sposet is defined within slaterdeterminant. The basisset input block is composed from a collection of atomicBasisSet input blocks, one for each atomic species in the simulation where basis functions are centered. The general structure for basisset and atomicBasisSet are given in Listing 4, and the corresponding lists of options are given in
table5 and table6.
determinantset element:
Parent elements |
|
Child elements |
|
Attribute:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
Text |
Any |
‘’ ‘’ |
Name of determinant set |
|
Text |
See below |
‘’ ‘’ |
Type of |
|
Text |
NMO, GTO, STO |
NMO |
Type of orbital set generated |
|
Text |
Yes/no |
Yes |
Transform to numerical radial functions? |
|
Text |
Any |
Ion0 |
Particle set with the position of atom centers |
|
Text |
Yes/no |
No |
Apply cusp correction scheme to |
Table 4 Options for the determinantset xml-block associated with atom-centered single particle orbital sets.
basisset.¶<basisset name="LCAOBSet">
<atomicBasisSet name="Gaussian-G2" angular="cartesian" elementType="C" normalized="no">
<grid type="log" ri="1.e-6" rf="1.e2" npts="1001"/>
<basisGroup rid="C00" n="0" l="0" type="Gaussian">
<radfunc exponent="5.134400000000e-02" contraction="1.399098787100e-02"/>
...
</basisGroup>
...
</atomicBasisSet>
<atomicBasisSet name="Gaussian-G2" angular="cartesian" type="Gaussian" elementType="C" normalized="no">
...
</atomicBasisSet>
...
</basisset>
basisset element:
Parent elements |
|
Child elements |
|
Attribute:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
Text |
Any |
” “ |
Name of atom-centered basis set |
Table 5 Options for the basisset xml-block associated with atom-centered single particle orbital sets.
AtomicBasisSet element:
Parent elements |
|
Child elements |
|
Attribute:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
Text |
Any |
” “ |
Name of atomic basis set |
|
Text |
See below |
Default |
Type of angular functions |
|
Text |
See below |
Yes |
Expand Ylm shells? |
|
Text |
See below |
Yes |
Add sign for \((-1)^{m}\)? |
|
Text |
Any |
e |
Atomic species where functions are centered |
|
Text |
Yes/no |
Yes |
Are single particle functions normalized? |
Table 6 Options for the atomicBasisSet xml-block.
basicGroup element:
Parent elements |
|
Child elements |
|
Attribute:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
Text |
Any |
‘’ ‘’ |
Name of the basisGroup |
|
Text |
Any |
‘’ ‘’ |
Type of basisGroup |
|
Integer |
Any |
0 |
Quantum numbers of basisGroup |
table7 Options for the basisGroup xml-block.
slaterdeterminant with an atom-centered sposet.¶ <slaterdeterminant>
</slaterdeterminant>
element:
Parent elements: |
|
Child elements: |
Attribute:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
Text |
Any |
‘’ ‘’ |
Name of determinant set |
Text |
Any |
‘’ ‘’ |
Detailed description of attributes:¶
In the following, we give a more detailed description of all the options presented in the various xml-blocks described in this section. Only nontrivial attributes are described. Those with simple yes/no options and whose previous description is enough to explain the intended behavior are not included.
determinantset attributes:
typeType of
sposet. For atom-centered basedsposets, use type=”MolecularOrbital” or type=“MO.” Other options described elsewhere in this manual are “spline,” “composite,” “pw,” “heg,” “linearopt,” etc.
keyword/keyType of basis set generated, which does not necessarily match the type of basis set on the input block. The three possible options are: NMO (numerical molecular orbitals), GTO (Gaussian-type orbitals), and STO (Slater-type orbitals). The default option is NMO. By default, QMCPACK will generate numerical orbitals from both GTO and STO types and use cubic or quintic spline interpolation to evaluate the radial functions. This is typically more efficient than evaluating the radial functions in the native basis (Gaussians or exponents) and allows for arbitrarily large contractions without any additional cost. To force use of the native expansion (not recommended), use GTO or STO for each type of input basis set.
transformRequest (or avoid) a transformation of the radial functions to NMO type. The default and recommended behavior is to transform to numerical radial functions. If
transformis set to yes, the optionkeywordis ignored.
cuspCorrectionEnable (disable) use of the cusp correction algorithm (CASINO REFERENCE) for a
basissetbuilt with GTO functions. The algorithm is implemented as described in (CASINO REFERENCE) and works only with transform=”yes” and an input GTO basis set. No further input is needed.
atomicBasisSet attributes:
name/idName of the basis set. Names should be unique.
angularType of angular functions used in the expansion. In general, two angular basis functions are allowed: “spherical” (for spherical Ylm functions) and “Cartesian” (for functions of the type \(x^{n}y^{m}z^{l}\)).
expandYlmDetermines whether each basis group is expanded across the corresponding shell of m values (for spherical type) or consistent powers (for Cartesian functions). Options:
“No”: Do not expand angular functions across corresponding angular shell.
“Gaussian”: Expand according to Gaussian03 format. This function is compatible only with angular=”spherical.” For a given input (l,m), the resulting order of the angular functions becomes (1,-1,0) for l=1 and (0,1,-1,2,-2,…,l,-l) for general l.
“Natural”: Expand angular functions according to (-l,-l+1,…,l-1,l).
“Gamess”: Expand according to Gamess’ format for Cartesian functions. Notice that this option is compatible only with angular=”Cartesian.” If angular=”Cartesian” is used, this option is not necessary.
expMDetermines whether the sign of the spherical Ylm function associated with m (\(-1^{m}\)) is included in the coefficient matrix or not.
elementType/speciesName of the species where basis functions are centered. Only one
atomicBasisSetblock is allowed per species. Additional blocks are ignored. The corresponding species must exist in theparticlesetgiven as thesourceoption todeterminantset. Basis functions for all the atoms of the corresponding species are included in the basis set, based on the order of atoms in theparticleset.
basisGroup attributes:
typeType of input basis radial function. Note that this refers to the type of radial function in the input xml-block, which might not match the radial function generated internally and used in the calculation (if
transformis set to “yes”). Also note that differentbasisGroupblocks within a givenatomicBasisSetcan have differenttypes.
n/l/m/sQuantum numbers of the basis function. Note that if
expandYlmis set to “yes” inatomicBasisSet, a full shell of basis functions with the appropriate values of “m” will be defined for the corresponding value of “l.” Otherwise a single basis function will be given for the specific combination of “(l,m).”
radfunc attributes for type = “Gaussian”:
TBDoc
slaterdeterminant attributes:
TBDoc
Hybrid orbital representation¶
The hybrid representation of the single particle orbitals combines a localized atomic basis set around atomic cores and B-splines in the interstitial regions to reduce memory use while retaining high evaluation speed and either retaining or increasing overall accuracy. Full details are provided in [LEKS18], and users of this feature are kindly requested to cite this paper. In practice, we have seen that using a meshfactor=0.5 is often possible and achieves huge memory savings. Fig. 3 illustrates how the regions are assigned.
Regular and hybrid orbital representation. Regular B-spline representation (left panel) contains only one region and a sufficiently fine mesh to resolve orbitals near the nucleus. The hybrid orbital representation (right panel) contains near nucleus regions (A) where spherical harmonics and radial functions are used, buffers or interpolation regions (B), and an interstitial region (C) where a coarse B-spline mesh is used.¶
Orbitals within region A are computed as
Orbitals in region C are computed as the regular B-spline basis described in Spline basis sets above. The region B interpolates between A and C as
To enable hybrid orbital representation, the input XML needs to see the tag hybridrep="yes" shown in Listing 6.
<determinantset type="bspline" source="i" href="pwscf.h5"
tilematrix="1 1 3 1 2 -1 -2 1 0" twistnum="-1" gpu="yes" meshfactor="0.8"
twist="0 0 0" precision="single" hybridrep="yes">
...
</determinantset>
Second, the information describing the atomic regions is required in the particle set, shown in Listing 7.
<group name="Ni">
<parameter name="charge"> 18 </parameter>
<parameter name="valence"> 18 </parameter>
<parameter name="atomicnumber" > 28 </parameter>
<parameter name="cutoff_radius" > 1.6 </parameter>
<parameter name="inner_cutoff" > 1.3 </parameter>
<parameter name="lmax" > 5 </parameter>
<parameter name="spline_radius" > 1.8 </parameter>
<parameter name="spline_npoints"> 91 </parameter>
</group>
The parameters specific to hybrid representation are listed as
attrib element
Attribute:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
Real |
>=0.0 |
None |
Cutoff radius for B/C boundary |
|
Integer |
>=0 |
None |
Largest angular channel |
|
Real |
>=0.0 |
Dep. |
Cutoff radius for A/B boundary |
|
Real |
>0.0 |
Dep. |
Radial function radius used in spine |
|
Integer |
>0 |
Dep. |
Number of spline knots |
cutoff_radiusis required for every species. If a species is intended to not be covered by atomic regions, setting the value 0.0 will put default values for all the reset parameters. A good value is usually a bit larger than the core radius listed in the pseudopotential file. After a parametric scan, pick the one from the flat energy region with the smallest variance.lmaxis required ifcutoff_radius\(>\) 0.0. This value usually needs to be at least the highest angular momentum plus 2.inner_cutoffis optional and set ascutoff_radius\(-0.3\) by default, which is fine in most cases.spline_radiusandspline_npointsare optional. By default, they are calculated based oncutoff_radiusand a grid displacement 0.02 bohr. If users prefer inputing them, it is required thatcutoff_radius<=spline_radius\(-\) 2 \(\times\)spline_radius/(spline_npoints\(-\) 1).
In addition, the hybrid orbital representation allows extra optimization to speed up the nonlocal pseudopotential evaluation using the batched algorithm listed in Pseudopotentials.
Plane-wave basis sets¶
Homogeneous electron gas¶
The interacting Fermi liquid has its own special determinantset for filling up a
Fermi surface. The shell number can be specified separately for both spin-up and spin-down.
This determines how many electrons to include of each time; only closed shells are currently
implemented. The shells are filled according to the rules of a square box; if other lattice
vectors are used, the electrons might not fill up a complete shell.
This following example can also be used for Helium simulations by specifying the proper pair interaction in the Hamiltonian section.
<qmcsystem>
<simulationcell name="global">
<parameter name="rs" pol="0" condition="74">6.5</parameter>
<parameter name="bconds">p p p</parameter>
<parameter name="LR_dim_cutoff">15</parameter>
</simulationcell>
<particleset name="e" random="yes">
<group name="u" size="37">
<parameter name="charge">-1</parameter>
<parameter name="mass">1</parameter>
</group>
<group name="d" size="37">
<parameter name="charge">-1</parameter>
<parameter name="mass">1</parameter>
</group>
</particleset>
</qmcsystem>
<qmcsystem>
<wavefunction name="psi0" target="e">
<determinantset type="electron-gas" shell="7" shell2="7" randomize="true">
</determinantset>
<jastrow name="J2" type="Two-Body" function="Bspline" print="no">
<correlation speciesA="u" speciesB="u" size="8" cusp="0">
<coefficients id="uu" type="Array" optimize="yes">
</correlation>
<correlation speciesA="u" speciesB="d" size="8" cusp="0">
<coefficients id="ud" type="Array" optimize="yes">
</correlation>
</jastrow>
Jastrow Factors¶
Jastrow factors are among the simplest and most effective ways of including dynamical correlation in the trial many body wavefunction. The resulting many body wavefunction is expressed as the product of an antisymmetric (in the case of Fermions) or symmetric (for Bosons) part and a correlating Jastrow factor like so:
In this section we will detail the types and forms of Jastrow factor used
in QMCPACK. Note that each type of Jastrow factor needs to be specified using
its own individual jastrow XML element. For this reason, we have repeated the
specification of the jastrow tag in each section, with specialization for the
options available for that given type of Jastrow.
One-body Jastrow functions¶
The one-body Jastrow factor is a form that allows for the direct inclusion of correlations between particles that are included in the wavefunction with particles that are not explicitly part of it. The most common example of this are correlations between electrons and ions.
The Jastrow function is specified within a wavefunction element
and must contain one or more correlation elements specifying
additional parameters as well as the actual coefficients.
Example use cases gives examples of the typical nesting of
jastrow, correlation, and coefficient elements.
Input Specification¶
Jastrow element:
name
datatype
values
defaults
description
name
text
(required)
Unique name for this Jastrow function
type
text
One-body
(required)
Define a one-body function
function
text
Bspline
(required)
BSpline Jastrow
text
pade2
Pade form
text
…
…
source
text
name
(required)
Name of attribute of classical particle set
text
yes / no
yes
Jastrow factor printed in external file?
elements
Correlation
Contents
(None)
To be more concrete, the one-body Jastrow factors used to describe correlations between electrons and ions take the form below:
where I runs over all of the ions in the calculation, i runs over the electrons and \(u_{ab}\) describes the functional form of the correlation between them. Many different forms of \(u_{ab}\) are implemented in QMCPACK. We will detail two of the most common ones below.
Spline form¶
The one-body spline Jastrow function is the most commonly used one-body Jastrow for solids. This form was first described and used in [EKCS12]. Here \(u_{ab}\) is an interpolating 1D B-spline (tricublc spline on a linear grid) between zero distance and \(r_{cut}\). In 3D periodic systems the default cutoff distance is the Wigner Seitz cell radius. For other periodicities, including isolated molecules, the \(r_{cut}\) must be specified. The cusp can be set. \(r_i\) and \(R_I\) are most commonly the electron and ion positions, but any particlesets that can provide the needed centers can be used.
Correlation element:
Name
Datatype
Values
Defaults
Description
ElementType
Text
Name
See below
Classical particle target
SpeciesA
Text
Name
See below
Classical particle target
SpeciesB
Text
Name
See below
Quantum species target
Size
Integer
\(> 0\)
(Required)
Number of coefficients
Rcut
Real
\(> 0\)
See below
Distance at which the correlation goes to 0
Cusp
Real
\(\ge 0\)
0
Value for use in Kato cusp condition
Spin
Text
Yes or no
No
Spin dependent Jastrow factor
Elements
Coefficients
Contents
(None)
Additional information:
elementType, speciesA, speciesB, spinFor a spin-independent Jastrow factor (spin = “no”), elementType should be the name of the group of ions in the classical particleset to which the quantum particles should be correlated. For a spin-dependent Jastrow factor (spin = “yes”), set speciesA to the group name in the classical particleset and speciesB to the group name in the quantum particleset.
rcutThe cutoff distance for the function in atomic units (bohr). For 3D fully periodic systems, this parameter is optional, and a default of the Wigner Seitz cell radius is used. Otherwise this parameter is required.
cuspThe one-body Jastrow factor can be used to make the wavefunction satisfy the electron-ion cusp condition :cite:
kato. In this case, the derivative of the Jastrow factor as the electron approaches the nucleus will be given by
Note that if the antisymmetric part of the wavefunction satisfies the electron-ion cusp condition (for instance by using single-particle orbitals that respect the cusp condition) or if a nondivergent pseudopotential is used, the Jastrow should be cuspless at the nucleus and this value should be kept at its default of 0.
Coefficients element:
Name
Datatype
Values
Defaults
Description
Id
Text
(Required)
Unique identifier
Type
Text
Array
(Required)
Optimize
Text
Yes or no
Yes
if no, values are fixed in optimizations
Elements
(None)
Contents
(No name)
Real array
Zeros
Jastrow coefficients
Example use cases¶
Specify a spin-independent function with four parameters. Because rcut is not specified, the default cutoff of the Wigner Seitz cell radius is used; this Jastrow must be used with a 3D periodic system such as a bulk solid. The name of the particleset holding the ionic positions is “i.”
<jastrow name="J1" type="One-Body" function="Bspline" print="yes" source="i">
<correlation elementType="C" cusp="0.0" size="4">
<coefficients id="C" type="Array"> 0 0 0 0 </coefficients>
</correlation>
</jastrow>
Specify a spin-dependent function with seven up-spin and seven down-spin parameters. The cutoff distance is set to 6 atomic units. Note here that the particleset holding the ions is labeled as ion0 rather than “i,” as in the other example. Also in this case, the ion is lithium with a coulomb potential, so the cusp condition is satisfied by setting cusp=”d.”
<jastrow name="J1" type="One-Body" function="Bspline" source="ion0" spin="yes">
<correlation speciesA="Li" speciesB="u" size="7" rcut="6">
<coefficients id="eLiu" cusp="3.0" type="Array">
0.0 0.0 0.0 0.0 0.0 0.0 0.0
</coefficients>
</correlation>
<correlation speciesA="C" speciesB="d" size="7" rcut="6">
<coefficients id="eLid" cusp="3.0" type="Array">
0.0 0.0 0.0 0.0 0.0 0.0 0.0
</coefficients>
</correlation>
</jastrow>
Pade form¶
Although the spline Jastrow factor is the most flexible and most commonly used form implemented in QMCPACK, there are times where its flexibility can make it difficult to optimize. As an example, a spline Jastrow with a very large cutoff can be difficult to optimize for isolated systems such as molecules because of the small number of samples present in the tail of the function. In such cases, a simpler functional form might be advantageous. The second-order Pade Jastrow factor, given in (11), is a good choice in such cases.
Unlike the spline Jastrow factor, which includes a cutoff, this form has an infinite range and will be applied to every particle pair (subject to the minimum image convention). It also is a cuspless Jastrow factor, so it should be used either in combination with a single particle basis set that contains the proper cusp or with a smooth pseudopotential.
Correlation element:
Name
Datatype
Values
Defaults
Description
ElementType
Text
Name
See below
Classical particle target
Elements
Coefficients
Contents
(None)
Parameter element:
Name
Datatype
Values
Defaults
Description
Id
String
Name
(Required)
Name for variable
Name
String
A or B or C
(Required)
See (11)
Optimize
Text
Yes or no
Yes
If no, values are fixed in optimizations
Elements
(None)
Contents
(No name)
Real
Parameter value
(Required)
Jastrow coefficients
Example use case¶
Specify a spin-independent function with independent Jastrow factors for two different species (Li and H). The name of the particleset holding the ionic positions is “i.”
<jastrow name="J1" function="pade2" type="One-Body" print="yes" source="i">
<correlation elementType="Li">
<var id="LiA" name="A"> 0.34 </var>
<var id="LiB" name="B"> 12.78 </var>
<var id="LiC" name="C"> 1.62 </var>
</correlation>
<correlation elementType="H"">
<var id="HA" name="A"> 0.14 </var>
<var id="HB" name="B"> 6.88 </var>
<var id="HC" name="C"> 0.237 </var>
</correlation>
</jastrow>
Short Range Cusp Form¶
The idea behind this functor is to encode nuclear cusps and other details at very short range around a nucleus in the region that the Gaussian orbitals of quantum chemistry are not capable of describing correctly. The functor is kept short ranged, because outside this small region, quantum chemistry orbital expansions are already capable of taking on the correct shapes. Unlike a pre-computed cusp correction, this optimizable functor can respond to changes in the wave function during VMC optimization. The functor’s form is
in which \(R_0\) acts as a soft cutoff radius (\(u(r)\) decays to zero quickly beyond roughly this distance) and \(A\) determines the cusp condition.
The simple exponential decay is modified by the \(N\) coefficients \(B_k\) that define an expansion in sigmoidal functions, thus adding detailed structure in a short-ranged region around a nucleus while maintaining the correct cusp condition at the nucleus. Note that sigmoidal functions are used instead of, say, a bare polynomial expansion, as they trend to unity past the soft cutoff radius and so interfere less with the exponential decay that keeps the functor short ranged. Although \(A\), \(R_0\), and the \(B_k\) coefficients can all be optimized as variational parameters, \(A\) will typically be fixed as the desired cusp condition is known.
To specify this one-body Jastrow factor, use an input section like the following.
<jastrow name="J1Cusps" type="One-Body" function="shortrangecusp" source="ion0" print="yes">
<correlation rcut="6" cusp="3" elementType="Li">
<var id="LiCuspR0" name="R0" optimize="yes"> 0.06 </var>
<coefficients id="LiCuspB" type="Array" optimize="yes">
0 0 0 0 0 0 0 0 0 0
</coefficients>
</correlation>
<correlation rcut="6" cusp="1" elementType="H">
<var id="HCuspR0" name="R0" optimize="yes"> 0.2 </var>
<coefficients id="HCuspB" type="Array" optimize="yes">
0 0 0 0 0 0 0 0 0 0
</coefficients>
</correlation>
</jastrow>
Here “rcut” is specified as the range beyond which the functor is assumed to be zero. The value of \(A\) can either be specified via the “cusp” option as shown above, in which case its optimization is disabled, or through its own “var” line as for \(R_0\), in which case it can be specified as either optimizable (“yes”) or not (“no”). The coefficients \(B_k\) are specified via the “coefficients” section, with the length \(N\) of the expansion determined automatically based on the length of the array.
Note that this one-body Jastrow form can (and probably should) be used in conjunction with a longer ranged one-body Jastrow, such as a spline form. Be sure to set the longer-ranged Jastrow to be cusp-free!
Two-body Jastrow functions¶
The two-body Jastrow factor is a form that allows for the explicit inclusion of dynamic correlation between two particles included in the wavefunction. It is almost always given in a spin dependent form so as to satisfy the Kato cusp condition between electrons of different spins [Kat51].
The two body Jastrow function is specified within a wavefunction element
and must contain one or more correlation elements specifying additional parameters
as well as the actual coefficients. Example use cases gives
examples of the typical nesting of jastrow, correlation and
coefficient elements.
Input Specification¶
Jastrow element:
name
datatype
values
defaults
description
name
text
(required)
Unique name for this Jastrow function
type
text
Two-body
(required)
Define a one-body function
function
text
Bspline
(required)
BSpline Jastrow
text
yes / no
yes
Jastrow factor printed in external file?
elements
Correlation
Contents
(None)
The two-body Jastrow factors used to describe correlations between electrons take the form
The most commonly used form of two body Jastrow factor supported by the code is a splined Jastrow factor, with many similarities to the one body spline Jastrow.
Spline form¶
The two-body spline Jastrow function is the most commonly used two-body Jastrow for solids. This form was first described and used in [EKCS12]. Here \(u_{ab}\) is an interpolating 1D B-spline (tricublc spline on a linear grid) between zero distance and \(r_{cut}\). In 3D periodic systems, the default cutoff distance is the Wigner Seitz cell radius. For other periodicities, including isolated molecules, the \(r_{cut}\) must be specified. \(r_i\) and \(r_j\) are typically electron positions. The cusp condition as \(r_i\) approaches \(r_j\) is set by the relative spin of the electrons.
Correlation element:
Name
Datatype
Values
Defaults
Description
SpeciesA
Text
U or d
(Required)
Quantum species target
SpeciesB
Text
U or d
(Required)
Quantum species target
Size
Integer
\(> 0\)
(Required)
Number of coefficients
Rcut
Real
\(> 0\)
See below
Distance at which the correlation goes to 0
Spin
Text
Yes or no
No
Spin-dependent Jastrow factor
Elements
Coefficients
Contents
(None)
Additional information:
speciesA, speciesBThe scale function u(r) is defined for species pairs uu and ud. There is no need to define ud or dd since uu=dd and ud=du. The cusp condition is computed internally based on the charge of the quantum particles.
Coefficients element:
Name
Datatype
Values
Defaults
Description
Id
Text
(Required)
Unique identifier
Type
Text
Array
(Required)
Optimize
Text
Yes or no
Yes
If no, values are fixed in optimizations
Elements
(None)
Contents
(No name)
Real array
Zeros
Jastrow coefficients
Example use cases¶
Specify a spin-dependent function with four parameters for each channel. In this case, the cusp is set at a radius of 4.0 bohr (rather than to the default of the Wigner Seitz cell radius). Also, in this example, the coefficients are set to not be optimized during an optimization step.
<jastrow name="J2" type="Two-Body" function="Bspline" print="yes">
<correlation speciesA="u" speciesB="u" size="8" rcut="4.0">
<coefficients id="uu" type="Array" optimize="no"> 0.2309049836 0.1312646071 0.05464141356 0.01306231516</coefficients>
</correlation>
<correlation speciesA="u" speciesB="d" size="8" rcut="4.0">
<coefficients id="ud" type="Array" optimize="no"> 0.4351561096 0.2377951747 0.1129144262 0.0356789236</coefficients>
</correlation>
</jastrow>
User defined functional form¶
To aid in implementing different forms for \(u_{ab}(r)\), there is a
script that uses a symbolic expression to generate the appropriate code
(with spatial and parameter derivatives). The script is located in
src/QMCWaveFunctions/Jastrow/codegen/user_jastrow.py. The script
requires Sympy (www.sympy.org) for symbolic mathematics and code
generation.
To use the script, modify it to specify the functional form and a list of variational parameters. Optionally, there may be fixed parameters - ones that are specified in the input file, but are not part of the variational optimization. Also one symbol may be specified that accepts a cusp value in order to satisfy the cusp condition. There are several example forms in the script. The default form is the simple Padé.
Once the functional form and parameters are specified in the script, run
the script from the codegen directory and recompile QMCPACK. The
main output of the script is the file
src/QMCWaveFunctions/Jastrow/UserFunctor.h. The script also prints
information to the screen, and one section is a sample XML input block
containing all the parameters.
There is a unit test in
src/QMCWaveFunctions/test/test_user_jastrow.cpp to perform some
minimal testing of the Jastrow factor. The unit test will need updating
to properly test new functional forms. Most of the changes relate to the
number and name of variational parameters.
Jastrow element:
name
datatype
values
defaults
description
name
text
(required)
Unique name for this Jastrow function
type
text
One-body
(required)
Define a one-body function
Two-body
(required)
Define a two-body function
function
text
user
(required)
User-defined functor
See other parameters as appropriate for one or two-body functions
elements
Correlation
Contents
(None)
Long-ranged Jastrow factors¶
While short-ranged Jastrow factors capture the majority of the benefit for minimizing the total energy and the energy variance, long-ranged Jastrow factors are important to accurately reproduce the short-ranged (long wavelength) behavior of quantities such as the static structure factor, and are therefore essential for modern accurate finite size corrections in periodic systems.
Below two types of long-ranged Jastrow factors are described. The first (the k-space Jastrow) is simply an expansion of the one and/or two body correlation functions in plane waves, with the coefficients comprising the optimizable parameters. The second type have few variational parameters and use the optimized breakup method of Natoli and Ceperley [NC95] (the Yukawa and Gaskell RPA Jastrows).
Long-ranged Jastrow: k-space Jastrow¶
The k-space Jastrow introduces explicit long-ranged dependence commensurate with the periodic supercell. This Jastrow is to be used in periodic boundary conditions only.
The input for the k-space Jastrow fuses both one and two-body forms into a single element and so they are discussed together here. The one- and two-body terms in the k-Space Jastrow have the form:
Here \(\rho_G\) is the Fourier transform of the instantaneous electron density:
and \(\rho_G^I\) has the same form, but for the fixed ions. In both cases the coefficients are restricted to be real, though in general the coefficients for the one-body term need not be. See Feature: Reciprocal-space Jastrow factors for more detail.
Input for the k-space Jastrow follows the familar nesting of jastrow-correlation-coefficients elements, with attributes unique to the k-space Jastrow at the correlation input level.
jastrow type=kSpace element:
parent elements:
wavefunctionchild elements:
correlation
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
kSpace
must be kSpace
name\(^r\)text
anything
0
Unique name for Jastrow
source\(^r\)text
particleset.nameIon particleset name
correlation element:
parent elements:
jastrow type=kSpacechild elements:
coefficients
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
One-body, Two-Body
Must be One-body/Two-body
kc\(^r\)real
kc \(\ge\) 0
0.0
k-space cutoff in a.u.
symmetry\(^o\)text
crystal,isotropic,none
crystal
symmetry of coefficients
spinDependent\(^o\)boolean
yes,no
no
No current function
coefficients element:
parent elements:
correlationchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
id\(^r\)text
anything
cG1/cG2
Label for coeffs
type\(^r\)text
Array0
Must be Array
body text: The body text is a list of real values for the parameters.
Additional information:
It is normal to provide no coefficients as an initial guess. The number of coefficients will be automatically calculated according to the k-space cutoff + symmetry and set to zero.
Providing an incorrect number of parameters also results in all parameters being set to zero.
There is currently no way to turn optimization on/off for the k-space Jastrow. The coefficients are always optimized.
Spin dependence is currently not implemented for this Jastrow.
kc: Parameters with G vectors magnitudes less thankcare included in the Jastrow. Ifkcis zero, it is the same as excluding the k-space term.symmetry=crystal: Impose crystal symmetry on coefficients according to the structure factor.symmetry=isotropic: Impose spherical symmetry on coefficients according to G-vector magnitude.symmetry=none: Impose no symmetry on the coefficients.
<jastrow type="kSpace" name="Jk" source="ion0">
<correlation kc="4.0" type="One-Body" symmetry="cystal">
<coefficients id="cG1" type="Array">
</coefficients>
</correlation>
<correlation kc="4.0" type="Two-Body" symmetry="crystal">
<coefficients id="cG2" type="Array">
</coefficients>
</correlation>
</jastrow>
<jastrow type="kSpace" name="Jk" source="ion0">
<correlation kc="4.0" type="One-Body" symmetry="crystal">
<coefficients id="cG1" type="Array">
</coefficients>
</correlation>
</jastrow>
<jastrow type="kSpace" name="Jk" source="ion0">
<correlation kc="4.0" type="Two-Body" symmetry="crystal">
<coefficients id="cG2" type="Array">
</coefficients>
</correlation>
</jastrow>
Long-ranged Jastrows: Gaskell RPA and Yukawa forms¶
NOTE: The Yukawa and RPA Jastrows do not work at present and are currently being revived. Please contact the developers if you are interested in using them.
The exact Jastrow correlation functions contain terms which have a form similar to the Coulomb pair potential. In periodic systems the Coulomb potential is replaced by an Ewald summation of the bare potential over all periodic image cells. This sum is often handled by the optimized breakup method [NC95] and this same approach is applied to the long-ranged Jastrow factors in QMCPACK.
There are two main long-ranged Jastrow factors of this type implemented in QMCPACK: the Gaskell RPA [Gas61][Gas62] form and the [Cep78] form. Both of these forms were used by Ceperley in early studies of the electron gas [Cep78], but they are also appropriate starting points for general solids.
The Yukawa form is defined in real space. It’s long-range form is formally defined as
with \(u_Y(r)\) given by
In QMCPACK a slightly more restricted form is used:
here “\(r_s\)” is understood to be a variational parameter.
The Gaskell RPA form—which contains correct short/long range limits and minimizes the total energy of the electron gas within the RPA—is defined directly in k-space:
where $v_k$ is the Fourier transform of the Coulomb potential and \(S_0(k)\) is the static structure factor of the non-interacting electron gas:
When written in atomic units, RPA Jastrow implemented in QMCPACK has the form
Here “\(r_s\)” is again a variational parameter and \(k_F\equiv(\tfrac{9\pi}{4r_s^3})^{1/3}\).
For both the Yukawa and Gaskell RPA Jastrows, the default value for \(r_s\) is \(r_s=(\tfrac{3\Omega}{4\pi N_e})^{1/3}\).
jastrow type=Two-Body function=rpa/yukawa element:
parent elements:
wavefunctionchild elements:
correlationattributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
Two-body
Must be two-body
function\(^r\)text
rpa/yukawa
Must be rpa or yukawa
name\(^r\)text
anything
RPA_Jee
Unique name for Jastrow
longrange\(^o\)boolean
yes/no
yes
Use long-range part
shortrange\(^o\)boolean
yes/no
yes
Use short-range part
parameters:
Name
Datatype
Values
Default
Description
rs\(^o\)rs
\(r_s>0\)
\(\tfrac{3\Omega}{4\pi N_e}\)
Avg. elec-elec distance
kc\(^o\)kc
\(k_c>0\)
\(2\left(\tfrac{9\pi}{4}\right)^{1/3}\tfrac{4\pi N_e}{3\Omega}\)
k-space cutoff
<jastrow name=''Jee'' type=''Two-Body'' function=''rpa''>
</jastrow>
Three-body Jastrow functions¶
Explicit three-body correlations can be included in the wavefunction via the three-body Jastrow factor. The three-body electron-electron-ion correlation function (\(u_{\sigma\sigma'I}\)) currently used in is identical to the one proposed in [DTN04]:
Here \(M_{eI}\) and \(M_{ee}\) are the maximum polynomial orders of the electron-ion and electron-electron distances, respectively, \(\{\gamma_{\ell mn}\}\) are the optimizable parameters (modulo constraints), \(r_c\) is a cutoff radius, and \(r_{ab}\) are the distances between electrons or ions \(a\) and \(b\). i.e. The correlation function is only a function of the interparticle distances and not a more complex function of the particle positions, \(\mathbf{r}\). As indicated by the \(\Theta\) functions, correlations are set to zero beyond a distance of \(r_c/2\) in either of the electron-ion distances and the largest meaningful electron-electron distance is \(r_c\). This is the highest-order Jastrow correlation function currently implemented.
Today, solid state applications of QMCPACK usually utilize one and two-body B-spline Jastrow functions, with calculations on heavier elements often also using the three-body term described above.
Example use case¶
Here is an example of H2O molecule. After optimizing one and two body Jastrow factors, add the following block in the wavefunction. The coefficients will be filled zero automatically if not given.
<jastrow name="J3" type="eeI" function="polynomial" source="ion0" print="yes">
<correlation ispecies="O" especies="u" isize="3" esize="3" rcut="10">
<coefficients id="uuO" type="Array" optimize="yes"> </coefficients>
</correlation>
<correlation ispecies="O" especies1="u" especies2="d" isize="3" esize="3" rcut="10">
<coefficients id="udO" type="Array" optimize="yes"> </coefficients>
</correlation>
<correlation ispecies="H" especies="u" isize="3" esize="3" rcut="10">
<coefficients id="uuH" type="Array" optimize="yes"> </coefficients>
</correlation>
<correlation ispecies="H" especies1="u" especies2="d" isize="3" esize="3" rcut="10">
<coefficients id="udH" type="Array" optimize="yes"> </coefficients>
</correlation>
</jastrow>
Multideterminant wavefunctions¶
Multiple schemes to generate a multideterminant wavefunction are possible, from CASSF to full CI or selected CI. The QMCPACK converter can convert MCSCF multideterminant wavefunctions from GAMESS [SBB+93] and CIPSI [EG13] wavefunctions from Quantum Package [Sce17] (QP). Full details of how to run a CIPSI calculation and convert the wavefunction for QMCPACK are given in CIPSI wavefunction interface.
The script utils/determinants_tools.py can be used to generate
useful information about the multideterminant wavefunction. This script takes, as a required argument, the path of an h5 file corresponding to the wavefunction. Used without optional arguments, it prints the number of determinants, the number of CSFs, and a histogram of the excitation degree.
> determinants_tools.py ./tests/molecules/C2_pp/C2.h5
Summary:
excitation degree 0 count: 1
excitation degree 1 count: 6
excitation degree 2 count: 148
excitation degree 3 count: 27
excitation degree 4 count: 20
n_det 202
n_csf 104
If the --verbose argument is used, the script will print each determinant,
the associated CSF, and the excitation degree relative to the first determinant.
> determinants_tools.py -v ./tests/molecules/C2_pp/C2.h5 | head
1
alpha 1111000000000000000000000000000000000000000000000000000000
beta 1111000000000000000000000000000000000000000000000000000000
scf 2222000000000000000000000000000000000000000000000000000000
excitation degree 0
2
alpha 1011100000000000000000000000000000000000000000000000000000
beta 1011100000000000000000000000000000000000000000000000000000
scf 2022200000000000000000000000000000000000000000000000000000
excitation degree 2
Backflow Wavefunctions¶
One can perturb the nodal surface of a single-Slater/multi-Slater wavefunction through use of a backflow transformation. Specifically, if we have an antisymmetric function \(D(\mathbf{x}_{0\uparrow},\cdots,\mathbf{x}_{N\uparrow}, \mathbf{x}_{0\downarrow},\cdots,\mathbf{x}_{N\downarrow})\), and if \(i_\alpha\) is the \(i\)-th particle of species type \(\alpha\), then the backflow transformation works by making the coordinate transformation \(\mathbf{x}_{i_\alpha} \to \mathbf{x}'_{i_\alpha}\) and evaluating \(D\) at these new “quasiparticle” coordinates. QMCPACK currently supports quasiparticle transformations given by
Here, \(\eta^{\alpha\beta}(|\mathbf{x}_{i_\alpha}-\mathbf{x}_{j_\beta}|)\) is a radially symmetric backflow transformation between species \(\alpha\) and \(\beta\). In QMCPACK, particle \(i_\alpha\) is known as the “target” particle and \(j_\beta\) is known as the “source.” The main types of transformations are so-called one-body terms, which are between an electron and an ion \(\eta^{eI}(|\mathbf{x}_{i_e}-\mathbf{x}_{j_I}|)\) and two-body terms. Two-body terms are distinguished as those between like and opposite spin electrons: \(\eta^{e(\uparrow)e(\uparrow)}(|\mathbf{x}_{i_e(\uparrow)}-\mathbf{x}_{j_e(\uparrow)}|)\) and \(\eta^{e(\uparrow)e(\downarrow)}(|\mathbf{x}_{i_e(\uparrow)}-\mathbf{x}_{j_e(\downarrow)}|)\). Henceforth, we will assume that \(\eta^{e(\uparrow)e(\uparrow)}=\eta^{e(\downarrow)e(\downarrow)}\).
In the following, we explain how to describe general terms such as (24) in a QMCPACK XML file. For specificity, we will consider a particle set consisting of H and He (in that order). This ordering will be important when we build the XML file, so you can find this out either through your specific declaration of <particleset>, by looking at the hdf5 file in the case of plane waves, or by looking at the QMCPACK output file in the section labeled “Summary of QMC systems.”
Input specifications¶
All backflow declarations occur within a single <backflow> ... </backflow> block. Backflow transformations occur in <transformation> blocks and have the following input parameters:
Transformation element:
Name
Datatype
Values
Default
Description
name
Text
(Required)
Unique name for this Jastrow function.
type
Text
“e-I”
(Required)
Define a one-body backflow transformation.
Text
“e-e”
Define a two-body backflow transformation.
function
Text
B-spline
(Required)
B-spline type transformation (no other types supported).
source
Text
“e” if two body, ion particle set if one body.
Just like one- and two-body jastrows, parameterization of the backflow transformations are specified within the <transformation> blocks by <correlation> blocks. Please refer to Spline form for more information.
Example Use Case¶
Having specified the general form, we present a general example of one-body and two-body backflow transformations in a hydrogen-helium mixture. The hydrogen and helium ions have independent backflow transformations, as do the like and unlike-spin two-body terms. One caveat is in order: ionic backflow transformations must be listed in the order they appear in the particle set. If in our example, helium is listed first and hydrogen is listed second, the following example would be correct. However, switching backflow declaration to hydrogen first then helium, will result in an error. Outside of this, declaration of one-body blocks and two-body blocks are not sensitive to ordering.
<backflow>
<!--The One-Body term with independent e-He and e-H terms. IN THAT ORDER -->
<transformation name="eIonB" type="e-I" function="Bspline" source="ion0">
<correlation cusp="0.0" size="8" type="shortrange" init="no" elementType="He" rcut="3.0">
<coefficients id="eHeC" type="Array" optimize="yes">
0 0 0 0 0 0 0 0
</coefficients>
</correlation>
<correlation cusp="0.0" size="8" type="shortrange" init="no" elementType="H" rcut="3.0">
<coefficients id="eHC" type="Array" optimize="yes">
0 0 0 0 0 0 0 0
</coefficients>
</correlation>
</transformation>
<!--The Two-Body Term with Like and Unlike Spins -->
<transformation name="eeB" type="e-e" function="Bspline" >
<correlation cusp="0.0" size="7" type="shortrange" init="no" speciesA="u" speciesB="u" rcut="1.2">
<coefficients id="uuB1" type="Array" optimize="yes">
0 0 0 0 0 0 0
</coefficients>
</correlation>
<correlation cusp="0.0" size="7" type="shortrange" init="no" speciesA="d" speciesB="u" rcut="1.2">
<coefficients id="udB1" type="Array" optimize="yes">
0 0 0 0 0 0 0
</coefficients>
</correlation>
</transformation>
</backflow>
Currently, backflow works only with single-Slater determinant wavefunctions. When a backflow transformation has been declared, it should be placed within the <determinantset> block, but outside of the <slaterdeterminant> blocks, like so:
<determinantset ... >
<!--basis set declarations go here, if there are any -->
<backflow>
<transformation ...>
<!--Here is where one and two-body terms are defined -->
</transformation>
</backflow>
<slaterdeterminant>
<!--Usual determinant definitions -->
</slaterdeterminant>
</determinantset>
Optimization Tips¶
Backflow is notoriously difficult to optimize—it is extremely nonlinear in the variational parameters and moves the nodal surface around. As such, it is likely that a full Jastrow+Backflow optimization with all parameters initialized to zero might not converge in a reasonable time. If you are experiencing this problem, the following pointers are suggested (in no particular order).
Get a good starting guess for \(\Psi_T\):¶
Try optimizing the Jastrow first without backflow.
Freeze the Jastrow parameters, introduce only the e-e terms in the backflow transformation, and optimize these parameters.
Freeze the e-e backflow parameters, and then optimize the e-I terms.
If difficulty is encountered here, try optimizing each species independently.
Unfreeze all Jastrow, e-e backflow, and e-I backflow parameters, and reoptimize.
Optimizing Backflow Terms¶
It is possible that the previous prescription might grind to a halt in steps 2 or 3 with the inability to optimize the e-e or e-I backflow transformation independently, especially if it is initialized to zero. One way to get around this is to build a good starting guess for the e-e or e-I backflow terms iteratively as follows:
Start off with a small number of knots initialized to zero. Set \(r_{cut}\) to be small (much smaller than an interatomic distance).
Optimize the backflow function.
If this works, slowly increase \(r_{cut}\) and/or the number of knots.
Repeat steps 2 and 3 until there is no noticeable change in energy or variance of \(\Psi_T\).
Tweaking the Optimization Run¶
The following modifications are worth a try in the optimization block:
Try setting “useDrift” to “no.” This eliminates the use of wavefunction gradients and force biasing in the VMC algorithm. This could be an issue for poorly optimized wavefunctions with pathological gradients.
Try increasing “exp0” in the optimization block. Larger values of exp0 cause the search directions to more closely follow those predicted by steepest-descent than those by the linear method.
Note that the new adaptive shift optimizer has not yet been tried with backflow wavefunctions. It should perform better than the older optimizers, but a considered optimization process is still recommended.
Gaussian Product Wavefunction¶
The Gaussian Product wavefunction implements (25)
where \(\vec{R}_i\) is the position of the \(i^{\text{th}}\) quantum particle and \(\vec{R}_i^o\) is its center. \(\sigma_i\) is the width of the Gaussian orbital around center \(i\).
This variational wavefunction enhances single-particle density at chosen spatial locations with adjustable strengths. It is useful whenever such localization is physically relevant yet not captured by other parts of the trial wavefunction. For example, in an electron-ion simulation of a solid, the ions are localized around their crystal lattice sites. This single-particle localization is not captured by the ion-ion Jastrow. Therefore, the addition of this localization term will improve the wavefunction. The simplest use case of this wavefunction is perhaps the quantum harmonic oscillator (please see the “tests/models/sho” folder for examples).
Input specification
Gaussian Product Wavefunction (ionwf):
Name
Datatype
Values
Default
Description
Name
Text
ionwf
(Required)
Unique name for this wavefunction
Width
Floats
1.0 -1
(Required)
Widths of Gaussian orbitals
Source
Text
ion0
(Required)
Name of classical particle set
Additional information:
widthThere must be one width provided for each quantum particle. If a negative width is given, then its corresponding Gaussian orbital is removed. Negative width is useful if one wants to use Gaussian wavefunction for a subset of the quantum particles.sourceThe Gaussian centers must be specified in the form of a classical particle set. This classical particle set is likely the ion positions “ion0,” hence the name “ionwf.” However, arbitrary centers can be defined using a different particle set. Please refer to the examples in “tests/models/sho.”
Example Use Case¶
<qmcsystem>
<simulationcell>
<parameter name="bconds">
n n n
</parameter>
</simulationcell>
<particleset name="e">
<group name="u" size="1">
<parameter name="mass">5.0</parameter>
<attrib name="position" datatype="posArray" condition="0">
0.0001 -0.0001 0.0002
</attrib>
</group>
</particleset>
<particleset name="ion0" size="1">
<group name="H">
<attrib name="position" datatype="posArray" condition="0">
0 0 0
</attrib>
</group>
</particleset>
<wavefunction target="e" id="psi0">
<ionwf name="iwf" source="ion0" width="0.8165"/>
</wavefunction>
<hamiltonian name="h0" type="generic" target="e">
<extpot type="HarmonicExt" mass="5.0" energy="0.3"/>
<estimator type="latticedeviation" name="latdev"
target="e" tgroup="u"
source="ion0" sgroup="H"/>
</hamiltonian>
</qmcsystem>
- AlfeG04
D. Alfè and M. J. Gillan. An efficient localized basis set for quantum Monte Carlo calculations on condensed matter. Physical Review B, 70(16):161101, 2004.
- Cep78(1,2)
D. Ceperley. Ground state of the fermion one-component plasma: a monte carlo study in two and three dimensions. Phys. Rev. B, 18:3126–3138, October 1978. URL: https://link.aps.org/doi/10.1103/PhysRevB.18.3126, doi:10.1103/PhysRevB.18.3126.
- DTN04
N. D. Drummond, M. D. Towler, and R. J. Needs. Jastrow correlation factor for atoms, molecules, and solids. Physical Review B - Condensed Matter and Materials Physics, 70(23):1–11, 2004. doi:10.1103/PhysRevB.70.235119.
- EG13
M. Caffarel E. Giner, A. Scemama. Using perturbatively selected configuration interaction in quantum monte carlo calculations. Canadian Journal of Chemistry, 91:9, 2013.
- EKCS12(1,2)
Kenneth P. Esler, Jeongnim Kim, David M. Ceperley, and Luke Shulenburger. Accelerating quantum monte carlo simulations of real materials on gpu clusters. Computing in Science and Engineering, 14(1):40–51, 2012. doi:http://doi.ieeecomputersociety.org/10.1109/MCSE.2010.122.
- FWL90
S. Fahy, X. W. Wang, and Steven G. Louie. Variational quantum Monte Carlo nonlocal pseudopotential approach to solids: Formulation and application to diamond, graphite, and silicon. Physical Review B, 42(6):3503–3522, 1990. doi:10.1103/PhysRevB.42.3503.
- Gas61
T Gaskell. The collective treatment of a fermi gas: ii. Proceedings of the Physical Society, 77(6):1182, 1961. URL: http://stacks.iop.org/0370-1328/77/i=6/a=312.
- Gas62
T Gaskell. The collective treatment of many-body systems: iii. Proceedings of the Physical Society, 80(5):1091, 1962. URL: http://stacks.iop.org/0370-1328/80/i=5/a=307.
- Kat51
T Kato. Fundamental properties of hamiltonian operators of the schrodinger type. Transactions of the American Mathematical Society, 70:195–211, 1951.
- LEKS18
Ye Luo, Kenneth P. Esler, Paul R. C. Kent, and Luke Shulenburger. An efficient hybrid orbital representation for quantum monte carlo calculations. The Journal of Chemical Physics, 149(8):084107, 2018.
- LK18
Ye Luo and Jeongnim Kim. An highly efficient delayed update algorithm for evaluating slater determinants in quantum monte carlo. in preparation, ():, 2018.
- MDAzevedoL+17
T. McDaniel, E. F. D’Azevedo, Y. W. Li, K. Wong, and P. R. C. Kent. Delayed slater determinant update algorithms for high efficiency quantum monte carlo. The Journal of Chemical Physics, 147(17):174107, November 2017. doi:10.1063/1.4998616.
- NC95(1,2)
Vincent Natoli and David M. Ceperley. An optimized method for treating long-range potentials. Journal of Computational Physics, 117(1):171 – 178, 1995. URL: http://www.sciencedirect.com/science/article/pii/S0021999185710546, doi:http://dx.doi.org/10.1006/jcph.1995.1054.
- Sce17
A. Scemamma. Quantum package. https://github.com/LCPQ/quantum_package, 2013–2017.
- SBB+93
Michael W. Schmidt, Kim K. Baldridge, Jerry A. Boatz, Steven T. Elbert, Mark S. Gordon, Jan H. Jensen, Shiro Koseki, Nikita Matsunaga, Kiet A. Nguyen, Shujun Su, Theresa L. Windus, Michel Dupuis, and John A. Montgomery. General atomic and molecular electronic structure system. Journal of Computational Chemistry, 14(11):1347–1363, 1993. URL: http://dx.doi.org/10.1002/jcc.540141112, doi:10.1002/jcc.540141112.
Hamiltonian and Observables¶
QMCPACK is capable of the simultaneous measurement of the Hamiltonian and many other quantum operators. The Hamiltonian attains a special status among the available operators (also referred to as observables) because it ultimately generates all available information regarding the quantum system. This is evident from an algorithmic standpoint as well since the Hamiltonian (embodied in the projector) generates the imaginary time dynamics of the walkers in DMC and reptation Monte Carlo (RMC).
This section covers how the Hamiltonian can be specified, component by component, by the user in the XML format native to qmcpack. It also covers the input structure of statistical estimators corresponding to quantum observables such as the density, static structure factor, and forces.
The Hamiltonian¶
The many-body Hamiltonian in Hartree units is given by
Here, the sums indexed by \(i/j\) are over quantum particles, while \(\ell/m\) are reserved for classical particles. Often the quantum particles are electrons, and the classical particles are ions, though is not limited in this way. The mass of each quantum particle is denoted \(m_i\), \(v^{qq}/v^{qc}/v^{cc}\) are pair potentials between quantum-quantum/quantum-classical/classical-classical particles, and \(v^{ext}\) denotes a purely external potential.
QMCPACK is designed modularly so that any potential can be supported with minimal additions to the code base. Potentials currently supported include Coulomb interactions in open and periodic boundary conditions, the MPC potential, nonlocal pseudopotentials, helium pair potentials, and various model potentials such as hard sphere, Gaussian, and modified Poschl-Teller.
Reference information and examples for the <hamiltonian/> XML
element are provided subsequently. Detailed descriptions of the input
for individual potentials is given in the sections that follow.
hamiltonian element:
parent elements:
simulation, qmcsystemchild elements:
pairpot extpot estimator constant(deprecated)
attributes:
Name
Datatype
Values
Default
Description
name/id\(^o\)text
anything
h0
Unique id for this Hamiltonian instance
type\(^o\)text
generic
No current function
role\(^o\)text
primary/extra
extra
Designate as Hamiltonian or not
source\(^o\)text
particleset.namei
Identify classical
particleset
target\(^o\)text
particleset.namee
Identify quantum
particleset
default\(^o\)boolean
yes/no
yes
Include kinetic energy term implicitly
Additional information:
target: Must be set to the name of the quantum
particleset. The default value is typically sufficient. In normal usage, no other attributes are provided.
<hamiltonian target="e">
<pairpot name="ElecElec" type="coulomb" source="e" target="e"/>
<pairpot name="ElecIon" type="coulomb" source="i" target="e"/>
<pairpot name="IonIon" type="coulomb" source="i" target="i"/>
</hamiltonian>
<hamiltonian target="e">
<pairpot name="ElecElec" type="coulomb" source="e" target="e"/>
<pairpot name="PseudoPot" type="pseudo" source="i" wavefunction="psi0" format="xml">
<pseudo elementType="Li" href="Li.xml"/>
<pseudo elementType="H" href="H.xml"/>
</pairpot>
<pairpot name="IonIon" type="coulomb" source="i" target="i"/>
</hamiltonian>
Pair potentials¶
Many pair potentials are supported. Though only the most commonly used pair potentials are covered in detail in this section, all currently available potentials are listed subsequently. If a potential you desire is not listed, or is not present at all, feel free to contact the developers.
pairpot factory element:
parent elements:
hamiltonianchild elements:
typeattribute
type options
coulomb
Coulomb/Ewald potential
pseudo
Semilocal pseudopotential
mpc
Model periodic Coulomb interaction/correction
cpp
Core polarization potential
skpot
Unknown
shared attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
See above
0
Select pairpot type
name\(^r\)text
Anything
any
Unique name for this pairpot
source\(^r\)text
particleset.name
hamiltonian.targetIdentify interacting particles
target\(^r\)text
particleset.name
hamiltonian.targetIdentify interacting particles
units\(^o\)text
hartree
No current function
Additional information:
type: Used to select the desired pair potential. Must be selected from the list of type options.
name: A unique name used to identify this pair potential. Block averaged output data will appear under this name in
scalar.datand/orstat.h5files.source/target: These specify the particles involved in a pair interaction. If an interaction is between classical (e.g., ions) and quantum (e.g., electrons),
source/targetshould be the name of the classical/quantumparticleset.Only
Coulomb, pseudo, andmpcare described in detail in the following subsections. The older or less-used types (cpp, skpot) are not covered.Available only if
QMC_CUDAis not defined:skpot.Available only if
OHMMS_DIM==3:mpc, vhxc, pseudo.Available only if
OHMMS_DIM==3andQMC_CUDAis not defined:cpp.
Coulomb potentials¶
The bare Coulomb potential is used in open boundary conditions:
When periodic boundary conditions are selected, Ewald summation is used automatically:
The sum indexed by \(L\) is over all nonzero simulation cell lattice vectors. In practice, the Ewald sum is broken into short- and long-range parts in a manner optimized for efficiency (see [NC95]) for details.
For information on how to set the boundary conditions, consult Specifying the system to be simulated.
pairpot type=coulomb element:
parent elements:
hamiltonianchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
coulomb
Must be coulomb
name/id\(^r\)text
anything
ElecElec
Unique name for interaction
source\(^r\)text
particleset.name
hamiltonian.targetIdentify interacting particles
target\(^r\)text
particleset.name
hamiltonian.targetIdentify interacting particles
pbc\(^o\)boolean
yes/no
yes
Use Ewald summation
physical\(^o\)boolean
yes/no
yes
Hamiltonian(yes)/Observable(no)
forcesboolean
yes/no
no
Deprecated
Additional information:
type/source/target: See description for the previous generic
pairpotfactory element.name: Traditional user-specified names for electron-electron, electron-ion, and ion-ion terms are
ElecElec,ElecIon, andIonIon, respectively. Although any choice can be used, the data analysis tools expect to find columns in*.scalar.datwith these names.pbc: Ewald summation will not be performed if
simulationcell.bconds== n n n, regardless of the value ofpbc. Similarly, thepbcattribute can only be used to turn off Ewald summation ifsimulationcell.bconds!= n n n. The default value is recommended.physical: If
physical==yes, this pair potential is included in the Hamiltonian and will factor into theLocalEnergyreported by QMCPACK and also in the DMC branching weight. Ifphysical==no, then the pair potential is treated as a passive observable but not as part of the Hamiltonian itself. As such it does not contribute to the outputtedLocalEnergy. Regardless of the value ofphysicaloutput data will appear inscalar.datin a column headed byname.
<pairpot name="ElecElec" type="coulomb" source="e" target="e"/>
<pairpot name="ElecIon" type="coulomb" source="i" target="e"/>
<pairpot name="IonIon" type="coulomb" source="i" target="i"/>
Pseudopotentials¶
QMCPACK supports pseudopotentials in semilocal form, which is local in the radial coordinate and nonlocal in angular coordinates. When all angular momentum channels above a certain threshold (\(\ell_{max}\)) are well approximated by the same potential (\(V_{\bar{\ell}}\equiv V_{loc}\)), the pseudopotential separates into a fully local channel and an angularly nonlocal component:
Here the electron/ion index is \(i/j\), and only one type of ion is shown for simplicity.
Evaluation of the localized pseudopotential energy \(\Psi_T^{-1}V^{PP}\Psi_T\) requires additional angular integrals. These integrals are evaluated on a randomly shifted angular grid. The size of this grid is determined by \(\ell_{max}\). See [MSC91] for further detail.
uses the FSAtom pseudopotential file format associated with the “Free
Software Project for Atomic-scale Simulations” initiated in 2002. See
http://www.tddft.org/fsatom/manifest.php for more information. The
FSAtom format uses XML for structured data. Files in this format do not
use a specific identifying file extension; instead they are simply
suffixed with “.xml.” The tabular data format of CASINO is also
supported.
In addition to the semilocal pseudopotential above, spin-orbit interactions can also be included through the use of spin-orbit pseudopotentials. The spin-orbit contribution can be written as
Here, \(\vec{s}\) is the spin operator. For each atom with a spin-orbit contribution,
the radial functions \(V_{\ell}^{\rm SO}\) can be included in the pseudopotential
“.xml” file.
pairpot type=pseudo element:
parent elements:
hamiltonianchild elements:
pseudo
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
pseudo
Must be pseudo
name/id\(^r\)text
anything
PseudoPot
No current function
source\(^r\)text
particleset.namei
Ion
particlesetname
target\(^r\)text
particleset.name
hamiltonian.targetElectron
particlesetname
pbc\(^o\)boolean
yes/no
yes*
Use Ewald summation
forcesboolean
yes/no
no
Deprecated
wavefunction\(^r\)text
wavefunction.nameinvalid
Identify wavefunction
format\(^r\)text
xml/table
table
Select file format
algorithm\(^o\)text
batched/default
default
Choose NLPP algorithm
DLA\(^o\)text
yes/no
no
Use determinant localization approximation
physicalSO\(^o\)boolean
yes/no
no
Include the SO contribution in the local energy
Additional information:
type/source/target See description for the generic
pairpotfactory element.name: Ignored. Instead, default names will be present in
*scalar.datoutput files when pseudopotentials are used. The fieldLocalECPrefers to the local part of the pseudopotential. If nonlocal channels are present, aNonLocalECPfield will be added that contains the nonlocal energy summed over all angular momentum channels.pbc: Ewald summation will not be performed if
simulationcell.bconds== n n n, regardless of the value ofpbc. Similarly, thepbcattribute can only be used to turn off Ewald summation ifsimulationcell.bconds!= n n n.format: If
format==table, QMCPACK looks for*.psffiles containing pseudopotential data in a tabular format. The files must be named after the ionic species provided inparticleset(e.g.,Li.psfandH.psf). Ifformat==xml, additionalpseudochild XML elements must be provided (see the following). These elements specify individual file names and formats (both the FSAtom XML and CASINO tabular data formats are supported).algorithm The default algorithm evaluates the ratios of wavefunction components together for each quadrature point and then one point after another. The batched algorithm evaluates the ratios of quadrature points together for each wavefunction component and then one component after another. Internally, it uses
VirtualParticleSetfor quadrature points. Hybrid orbital representation has an extra optimization enabled when using the batched algorithm.DLA Determinant localization approximation (DLA) [ZBMAlfe19] uses only the fermionic part of the wavefunction when calculating NLPP.
physicalSO If the spin-orbit components are included in the
.xmlfile, this flag allows control over whether the SO contribution is included in the local energy.
<pairpot name="PseudoPot" type="pseudo" source="i" wavefunction="psi0" format="psf"/>
<pairpot name="PseudoPot" type="pseudo" source="i" wavefunction="psi0" format="xml">
<pseudo elementType="Li" href="Li.xml"/>
<pseudo elementType="H" href="H.xml"/>
</pairpot>
<pairpot name="PseudoPot" type="pseudo" source="i" wavefunction="psi0" format="xml" physicalSO="yes">
<pseudo elementType="Pb" href="Pb.xml"/>
</pairpot>
Details of <pseudo/> input elements are shown in the following. It
is possible to include (or construct) a full pseudopotential directly in
the input file without providing an external file via href. The full
XML format for pseudopotentials is not yet covered.
pseudo element:
parent elements:
pairpot type=pseudochild elements:
header local grid
attributes:
Name
Datatype
Values
Default
Description
elementType/symbol\(^r\)text
groupe.namenone
Identify ionic species
href\(^r\)text
filepath
none
Pseudopotential file path
format\(^r\)text
xml/casino
xml
Specify file format
cutoff\(^o\)real
Nonlocal cutoff radius
lmax\(^o\)integer
Largest angular momentum
nrule\(^o\)integer
Integration grid order
<pseudo elementType="Li" href="Li.xml"/>
MPC Interaction/correction¶
The MPC interaction is an alternative to direct Ewald summation. The MPC
corrects the exchange correlation hole to more closely match its
thermodynamic limit. Because of this, the MPC exhibits smaller
finite-size errors than the bare Ewald interaction, though a few
alternative and competitive finite-size correction schemes now exist.
The MPC is itself often used just as a finite-size correction in
post-processing (set physical=false in the input).
pairpot type=mpc element:
parent elements:
hamiltonianchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
mpc
Must be MPC
name/id\(^r\)text
anything
MPC
Unique name for interaction
source\(^r\)text
particleset.name
hamiltonian.targetIdentify interacting particles
target\(^r\)text
particleset.name
hamiltonian.targetIdentify interacting particles
physical\(^o\)boolean
yes/no
no
Hamiltonian(yes)/observable(no)
cutoffreal
\(>0\)
30.0
Kinetic energy cutoff
Remarks:
physical: Typically set tono, meaning the standard Ewald interaction will be used during sampling and MPC will be measured as an observable for finite-size post-correction. Ifphysicalisyes, the MPC interaction will be used during sampling. In this case an electron-electron Coulombpairpotelement should not be supplied.Developer note: Currently the
nameattribute for the MPC interaction is ignored. The name is always reset toMPC.
<pairpot type="MPC" name="MPC" source="e" target="e" ecut="60.0" physical="no"/>
General estimators¶
A broad range of estimators for physical observables are available in QMCPACK.
The following sections contain input details for the total number
density (density), number density resolved by particle spin
(spindensity), spherically averaged pair correlation function
(gofr), static structure factor (sk), static structure factor
(skall), energy density (energydensity), one body reduced
density matrix (dm1b), \(S(k)\) based kinetic energy correction
(chiesa), forward walking (ForwardWalking), and force
(Force) estimators. Other estimators are not yet covered.
When an <estimator/> element appears in <hamiltonian/>, it is
evaluated for all applicable chained QMC runs (e.g.,
VMC\(\rightarrow\)DMC\(\rightarrow\)DMC). Estimators are
generally not accumulated during wavefunction optimization sections. If
an <estimator/> element is instead provided in a particular
<qmc/> element, that estimator is only evaluated for that specific
section (e.g., during VMC only).
estimator factory element:
parent elements:
hamiltonian, qmctype selector:
typeattribute
type options
density
Density on a grid
spindensity
Spin density on a grid
gofr
Pair correlation function (quantum species)
sk
Static structure factor
SkAll
Static structure factor needed for finite size correction
structurefactor
Species resolved structure factor
species kinetic
Species resolved kinetic energy
latticedeviation
Spatial deviation between two particlesets
momentum
Momentum distribution
energydensity
Energy density on uniform or Voronoi grid
dm1b
One body density matrix in arbitrary basis
chiesa
Chiesa-Ceperley-Martin-Holzmann kinetic energy correction
Force
Family of “force” estimators (see “Force” estimators)
ForwardWalking
Forward walking values for existing estimators
orbitalimages
Create image files for orbitals, then exit
flux
Checks sampling of kinetic energy
localmoment
Atomic spin polarization within cutoff radius
Pressure
No current function
shared attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
See above
0
Select estimator type
name\(^r\)text
anything
any
Unique name for this estimator
Chiesa-Ceperley-Martin-Holzmann kinetic energy correction¶
This estimator calculates a finite-size correction to the kinetic energy following the formalism laid out in [CCMH06]. The total energy can be corrected for finite-size effects by using this estimator in conjunction with the MPC correction.
estimator type=chiesa element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
chiesa
Must be chiesa
name\(^o\)text
anything
KEcorr
Always reset to KEcorr
source\(^o\)text
particleset.namee
Identify quantum particles
psi\(^o\)text
wavefunction.namepsi0
Identify wavefunction
<estimator name="KEcorr" type="chiesa" source="e" psi="psi0"/>
Density estimator¶
The particle number density operator is given by
The density estimator accumulates the number density on a uniform
histogram grid over the simulation cell. The value obtained for a grid
cell \(c\) with volume \(\Omega_c\) is then the average number
of particles in that cell:
estimator type=density element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
density
Must be density
name\(^r\)text
anything
any
Unique name for estimator
delta\(^o\)real array(3)
\(0\le v_i \le 1\)
0.1 0.1 0.1
Grid cell spacing, unit coords
x_min\(^o\)real
\(>0\)
0
Grid starting point in x (Bohr)
x_max\(^o\)real
\(>0\)
\(|\)
lattice[0]\(|\)Grid ending point in x (Bohr)
y_min\(^o\)real
\(>0\)
0
Grid starting point in y (Bohr)
y_max\(^o\)real
\(>0\)
\(|\)
lattice[1]\(|\)Grid ending point in y (Bohr)
z_min\(^o\)real
\(>0\)
0
Grid starting point in z (Bohr)
z_max\(^o\)real
\(>0\)
\(|\)
lattice[2]\(|\)Grid ending point in z (Bohr)
potential\(^o\)boolean
yes/no
no
Accumulate local potential, deprecated
debug\(^o\)boolean
yes/no
no
No current function
Additional information:
name: The name provided will be used as a label in thestat.h5file for the blocked output data. Postprocessing tools expectname="Density."delta: This sets the histogram grid size used to accumulate the density:delta="0.1 0.1 0.05"\(\rightarrow 10\times 10\times 20\) grid,delta="0.01 0.01 0.01"\(\rightarrow 100\times 100\times 100\) grid. The density grid is written to astat.h5file at the end of each MC block. If you request many \(blocks\) in a<qmc/>element, or select a large grid, the resultingstat.h5file could be many gigabytes in size.*_min/*_max: Can be used to select a subset of the simulation cell for the density histogram grid. For example if a (cubic) simulation cell is 20 Bohr on a side, setting*_min=5.0and*_max=15.0will result in a density histogram grid spanning a \(10\times 10\times 10\) Bohr cube about the center of the box. Use ofx_min, x_max, y_min, y_max, z_min, z_maxis only appropriate for orthorhombic simulation cells with open boundary conditions.When open boundary conditions are used, a
<simulationcell/>element must be explicitly provided as the first subelement of<qmcsystem/>for the density estimator to work. In this case the molecule should be centered around the middle of the simulation cell (\(L/2\)) and not the origin (\(0\) since the space within the cell, and hence the density grid, is defined from \(0\) to \(L\)).
<estimator name="Density" type="density" delta="0.05 0.05 0.05"/>
Spin density estimator¶
The spin density is similar to the total density described previously. In this case, the sum over particles is performed independently for each spin component.
estimator type=spindensity element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
spindensity
Must be spindensity
name\(^r\)text
anything
any
Unique name for estimator
report\(^o\)boolean
yes/no
no
Write setup details to stdout
parameters:
Name
Datatype
Values
Default
Description
grid\(^o\)integer array(3)
\(v_i>\)
Grid cell count
dr\(^o\)real array(3)
\(v_i>\)
Grid cell spacing (Bohr)
cell\(^o\)real array(3,3)
anything
Volume grid exists in
corner\(^o\)real array(3)
anything
Volume corner location
center\(^o\)real array (3)
anything
Volume center/origin location
voronoi\(^o\)text
particleset.nameUnder development
test_moves\(^o\)integer
\(>=0\)
0
Test estimator with random moves
Additional information:
name: The name provided will be used as a label in thestat.h5file for the blocked output data. Postprocessing tools expectname="SpinDensity."grid: The grid sets the dimension of the histogram grid. Input like<parameter name="grid"> 40 40 40 </parameter>requests a \(40 \times 40\times 40\) grid. The shape of individual grid cells is commensurate with the supercell shape.dr: Thedrsets the real-space dimensions of grid cell edges (Bohr units). Input like<parameter name="dr"> 0.5 0.5 0.5 </parameter>in a supercell with axes of length 10 Bohr each (but of arbitrary shape) will produce a \(20\times 20\times 20\) grid. The inputteddrvalues are rounded to produce an integer number of grid cells along each supercell axis. Eithergridordrmust be provided, but not both.cell: Whencellis provided, a user-defined grid volume is used instead of the global supercell. This must be provided if open boundary conditions are used. Additionally, ifcellis provided, the user must specify where the volume is located in space in addition to its size/shape (cell) using either thecornerorcenterparameters.corner: The grid volume is defined as \(corner+\sum_{d=1}^3u_dcell_d\) with \(0<u_d<1\) (“cell” refers to either the supercell or user-provided cell).center: The grid volume is defined as \(center+\sum_{d=1}^3u_dcell_d\) with \(-1/2<u_d<1/2\) (“cell” refers to either the supercell or user-provided cell).corner/centercan be used to shift the grid even ifcellis not specified. Simultaneous use ofcornerandcenterwill cause QMCPACK to abort.
<estimator type="spindensity" name="SpinDensity" report="yes">
<parameter name="grid"> 40 40 40 </parameter>
</estimator>
<estimator type="spindensity" name="SpinDensity" report="yes">
<parameter name="grid">
20 20 20
</parameter>
<parameter name="center">
0.0 0.0 0.0
</parameter>
<parameter name="cell">
10.0 0.0 0.0
0.0 10.0 0.0
0.0 0.0 10.0
</parameter>
</estimator>
Pair correlation function, \(g(r)\)¶
The functional form of the species-resolved radial pair correlation function operator is
where \(N_s\) is the number of particles of species \(s\) and \(V\) is the supercell volume. If \(s=s'\), then the sum is restricted so that \(i_s\ne j_s\).
In QMCPACK, an estimate of \(g_{ss'}(r)\) is obtained as a radial histogram with a set of \(N_b\) uniform bins of width \(\delta r\). This can be expressed analytically as
where the radial coordinate \(r\) is restricted to reside at the bin centers, \(\delta r/2, 3 \delta r/2, 5 \delta r/2, \ldots\).
estimator type=gofr element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
gofr
Must be gofr
name\(^o\)text
anything
any
No current function
num_bin\(^r\)integer
\(>1\)
20
# of histogram bins
rmax\(^o\)real
\(>0\)
10
Histogram extent (Bohr)
dr\(^o\)real
\(0\)
0.5
No current function
debug\(^o\)boolean
yes/no
no
No current function
target\(^o\)text
particleset.name
hamiltonian.targetQuantum particles
source/sources\(^o\)text array
particleset.name
hamiltonian.targetClassical particles
Additional information:
num_bin:This is the number of bins in each species pair radial histogram.rmax:This is the maximum pair distance included in the histogram. The uniform bin width is \(\delta r=\texttt{rmax/num\_bin}\). If periodic boundary conditions are used for any dimension of the simulation cell, then the default value ofrmaxis the simulation cell radius instead of 10 Bohr. For open boundary conditions, the volume (\(V\)) used is 1.0 Bohr\(^3\).source/sources:If unspecified, only pair correlations between each species of quantum particle will be measured. For each classical particleset specified bysource/sources, additional pair correlations between each quantum and classical species will be measured. Typically there is only one classical particleset (e.g.,source="ion0"), but there can be several in principle (e.g.,sources="ion0 ion1 ion2").target:The default value is the preferred usage (i.e.,targetdoes not need to be provided).Data is output to the
stat.h5for each QMC subrun. Individual histograms are named according to the quantum particleset and index of the pair. For example, if the quantum particleset is named “e” and there are two species (up and down electrons, say), then there will be three sets of histogram data in eachstat.h5file namedgofr_e_0_0,gofr_e_0_1, andgofr_e_1_1for up-up, up-down, and down-down correlations, respectively.
<estimator type="gofr" name="gofr" num_bin="200" rmax="3.0" />
<estimator type="gofr" name="gofr" num_bin="200" rmax="3.0" source="ion0" />
Static structure factor, \(S(k)\)¶
Let
\(\rho^e_{\mathbf{k}}=\sum_j e^{i \mathbf{k}\cdot\mathbf{r}_j^e}\)
be the Fourier space electron density, with \(\mathbf{r}^e_j\) being
the coordinate of the j-th electron. \(\mathbf{k}\) is a wavevector
commensurate with the simulation cell. QMCPACK allows the user to
accumulate the static electron structure factor \(S(\mathbf{k})\) at
all commensurate \(\mathbf{k}\) such that
\(|\mathbf{k}| \leq (LR\_DIM\_CUTOFF) r_c\). \(N^e\) is the
number of electrons, LR_DIM_CUTOFF is the optimized breakup
parameter, and \(r_c\) is the Wigner-Seitz radius. It is defined as
follows:
estimator type=sk element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
sk
Must sk
name\(^r\)text
anything
any
Unique name for estimator
hdf5\(^o\)boolean
yes/no
no
Output to
stat.h5(yes) orscalar.dat(no)
Additional information:
name:This is the unique name for estimator instance. A data structure of the same name will appear instat.h5output files.hdf5:Ifhdf5==yes, output data for \(S(k)\) is directed to thestat.h5file (recommended usage). Ifhdf5==no, the data is instead routed to thescalar.datfile, resulting in many columns of data with headings prefixed bynameand postfixed by the k-point index (e.g.,sk_0 sk_1 …sk_1037 …).This estimator only works in periodic boundary conditions. Its presence in the input file is ignored otherwise.
This is not a species-resolved structure factor. Additionally, for \(\mathbf{k}\) vectors commensurate with the unit cell, \(S(\mathbf{k})\) will include contributions from the static electronic density, thus meaning it will not accurately measure the electron-electron density response.
<estimator type="sk" name="sk" hdf5="yes"/>
Static structure factor, SkAll¶
In order to compute the finite size correction to the potential energy,
records of \(\rho(\mathbf{k})\) is required. What sets SkAll
apart from sk is that SkAll records \(\rho(\mathbf{k})\) in
addition to \(s(\mathbf{k})\).
estimator type=SkAll element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
sk
Must be sk
name\(^r\)text
anything
any
Unique name for estimator
source\(^r\)text
Ion ParticleSet name
None
-
target\(^r\)text
Electron ParticleSet name
None
-
hdf5\(^o\)boolean
yes/no
no
Output to
stat.h5(yes) orscalar.dat(no)
writeionion\(^o\)boolean
yes/no
no
Writes file rhok_IonIon.dat containing \(s(\mathbf{k})\) for the ions
Additional information:
name:This is the unique name for estimator instance. A data structure of the same name will appear instat.h5output files.hdf5:Ifhdf5==yes, output data is directed to thestat.h5file (recommended usage). Ifhdf5==no, the data is instead routed to thescalar.datfile, resulting in many columns of data with headings prefixed byrhokand postfixed by the k-point index.This estimator only works in periodic boundary conditions. Its presence in the input file is ignored otherwise.
This is not a species-resolved structure factor. Additionally, for \(\mathbf{k}\) vectors commensurate with the unit cell, \(S(\mathbf{k})\) will include contributions from the static electronic density, thus meaning it wil not accurately measure the electron-electron density response.
<estimator type="skall" name="SkAll" source="ion0" target="e" hdf5="yes"/>
Species kinetic energy¶
Record species-resolved kinetic energy instead of the total kinetic
energy in the Kinetic column of scalar.dat. SpeciesKineticEnergy
is arguably the simplest estimator in QMCPACK. The implementation of
this estimator is detailed in
manual/estimator/estimator_implementation.pdf.
estimator type=specieskinetic element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
specieskinetic
Must be specieskinetic
name\(^r\)text
anything
any
Unique name for estimator
hdf5\(^o\)boolean
yes/no
no
Output to
stat.h5(yes)
<estimator type="specieskinetic" name="skinetic" hdf5="no"/>
Lattice deviation estimator¶
Record deviation of a group of particles in one particle set (target) from a group of particles in another particle set (source).
estimator type=latticedeviation element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
latticedeviation
Must be latticedeviation
name\(^r\)text
anything
any
Unique name for estimator
hdf5\(^o\)boolean
yes/no
no
Output to
stat.h5(yes)
per_xyz\(^o\)boolean
yes/no
no
Directionally resolved (yes)
source\(^r\)text
e/ion0/…
no
source particleset
sgroup\(^r\)text
u/d/…
no
source particle group
target\(^r\)text
e/ion0/…
no
target particleset
tgroup\(^r\)text
u/d/…
no
target particle group
Additional information:
source: The “reference” particleset to measure distances from; actual reference points are determined together withsgroup.sgroup: The “reference” particle group to measure distances from.source: The “target” particleset to measure distances to.sgroup: The “target” particle group to measure distances to. For example, in Listing 32 the distance from the up electron (“u”) to the origin of the coordinate system is recorded.per_xyz: Used to record direction-resolved distance. In Listing 32, the x,y,z coordinates of the up electron will be recorded separately ifper_xyz=yes.hdf5: Used to record particle-resolved distances in the h5 file ifgdf5=yes.
<particleset name="e" random="yes">
<group name="u" size="1" mass="1.0">
<parameter name="charge" > -1 </parameter>
<parameter name="mass" > 1.0 </parameter>
</group>
<group name="d" size="1" mass="1.0">
<parameter name="charge" > -1 </parameter>
<parameter name="mass" > 1.0 </parameter>
</group>
</particleset>
<particleset name="wf_center">
<group name="origin" size="1">
<attrib name="position" datatype="posArray" condition="0">
0.00000000 0.00000000 0.00000000
</attrib>
</group>
</particleset>
<estimator type="latticedeviation" name="latdev" hdf5="yes" per_xyz="yes"
source="wf_center" sgroup="origin" target="e" tgroup="u"/>
Energy density estimator¶
An energy density operator, \(\hat{\mathcal{E}}_r\), satisfies
where the integral is over all space and \(\hat{H}\) is the Hamiltonian. In QMCPACK, the energy density is split into kinetic and potential components
with each component given by
Here, \(r_i\) and \(\tilde{r}_\ell\) represent electron and ion positions, respectively; \(\hat{p}_i\) is a single electron momentum operator; and \(\hat{v}^{ee}(r_i,r_j)\), \(\hat{v}^{eI}(r_i,\tilde{r}_\ell)\), and \(\hat{v}^{II}(\tilde{r}_\ell,\tilde{r}_m)\) are the electron-electron, electron-ion, and ion-ion pair potential operators (including nonlocal pseudopotentials, if present). This form of the energy density is size consistent; that is, the partially integrated energy density operators of well-separated atoms gives the isolated Hamiltonians of the respective atoms. For periodic systems with twist-averaged boundary conditions, the energy density is formally correct only for either a set of supercell k-points that correspond to real-valued wavefunctions or a k-point set that has inversion symmetry around a k-point having a real-valued wavefunction. For more information about the energy density, see [KYKC13].
In QMCPACK, the energy density can be accumulated on piecewise uniform 3D grids in generalized Cartesian, cylindrical, or spherical coordinates. The energy density integrated within Voronoi volumes centered on ion positions is also available. The total particle number density is also accumulated on the same grids by the energy density estimator for convenience so that related quantities, such as the regional energy per particle, can be computed easily.
estimator type=EnergyDensity element:
parent elements:
hamiltonian, qmcchild elements:
reference_points, spacegrid
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
EnergyDensity
Must be EnergyDensity
name\(^r\)text
anything
Unique name for estimator
dynamic\(^r\)text
particleset.nameIdentify electrons
static\(^o\)text
particleset.nameIdentify ions
ion_points\(^o\)text
yes/no
no
Separate ion energy density onto point field
Additional information:
name:Must be unique. A dataset with blocked statistical data for the energy density will appear in thestat.h5files labeled asname.Important: in order for the estimator to work, a traces XML input element (<traces array=”yes” write=”no”/>) must appear following the <qmcsystem/> element and prior to any <qmc/> element.
<estimator type="EnergyDensity" name="EDcell" dynamic="e" static="ion0">
<spacegrid coord="cartesian">
<origin p1="zero"/>
<axis p1="a1" scale=".5" label="x" grid="-1 (.05) 1"/>
<axis p1="a2" scale=".5" label="y" grid="-1 (.1) 1"/>
<axis p1="a3" scale=".5" label="z" grid="-1 (.1) 1"/>
</spacegrid>
</estimator>
<estimator type="EnergyDensity" name="EDatom" dynamic="e" static="ion0">
<reference_points coord="cartesian">
r1 1 0 0
r2 0 1 0
r3 0 0 1
</reference_points>
<spacegrid coord="spherical">
<origin p1="ion01"/>
<axis p1="r1" scale="6.9" label="r" grid="0 1"/>
<axis p1="r2" scale="6.9" label="phi" grid="0 1"/>
<axis p1="r3" scale="6.9" label="theta" grid="0 1"/>
</spacegrid>
<spacegrid coord="spherical">
<origin p1="ion02"/>
<axis p1="r1" scale="6.9" label="r" grid="0 1"/>
<axis p1="r2" scale="6.9" label="phi" grid="0 1"/>
<axis p1="r3" scale="6.9" label="theta" grid="0 1"/>
</spacegrid>
</estimator>
<estimator type="EnergyDensity" name="EDvoronoi" dynamic="e" static="ion0">
<spacegrid coord="voronoi"/>
</estimator>
The <reference_points/> element provides a set of points for later
use in specifying the origin and coordinate axes needed to construct a
spatial histogramming grid. Several reference points on the surface of
the simulation cell (see table8), as well as the
positions of the ions (see the energydensity.static attribute), are
made available by default. The reference points can be used, for
example, to construct a cylindrical grid along a bond with the origin on
the bond center.
reference_pointselement:
parent elements:
estimator type=EnergyDensitychild elements:
None
attributes:
Name
Datatype
Values
Default
Description
coord\(^r\)text
Cartesian/cell
Specify coordinate system
body text: The body text is a line formatted list of points with labels
Additional information:
coord:Ifcoord=cartesian, labeled points are in Cartesian (x,y,z) format in units of Bohr. Ifcoord=cell, then labeled points are in units of the simulation cell axes.body text:The list of points provided in the body text are line formatted, with four entries per line (label coor1 coor2 coor3). A set of points referenced to the simulation cell is available by default (seetable8). Ifenergydensity.staticis provided, the location of each individual ion is also available (e.g., ifenergydensity.static=ion0, then the location of the first atom is available with label ion01, the second with ion02, etc.). All points can be used by label when constructing spatial histogramming grids (see the followingspacegridelement) used to collect energy densities.
|
|
|
|---|---|---|
|
0 0 0 |
Cell center |
|
\(a_1\) |
Cell axis 1 |
|
\(a_2\) |
Cell axis 2 |
|
\(a_3\) |
Cell axis 3 |
|
\(a_1\)/2 |
Cell face 1+ |
|
-\(a_1\)/2 |
Cell face 1- |
|
\(a_2\)/2 |
Cell face 2+ |
|
-\(a_2\)/2 |
Cell face 2- |
|
\(a_3\)/2 |
Cell face 3+ |
|
-\(a_3\)/2 |
Cell face 3- |
|
\((a_1+a_2+a_3)/2\) |
Cell corner +,+,+ |
|
\((a_1+a_2-a_3)/2\) |
Cell corner +,+,- |
|
\((a_1-a_2+a_3)/2\) |
Cell corner +,-,+ |
|
\((-a_1+a_2+a_3)/2\) |
Cell corner -,+,+ |
|
\((a_1-a_2-a_3)/2\) |
Cell corner +,-,- |
|
\((-a_1+a_2-a_3)/2\) |
Cell corner -,+,- |
|
\((-a_1-a_2+a_3)/2\) |
Cell corner -,-,+ |
|
\((-a_1-a_2-a_3)/2\) |
Cell corner -,-,- |
Table 8 Reference points available by default. Vectors \(a_1\), \(a_2\), and \(a_3\) refer to the simulation cell axes. The representation of the cell is centered around zero.
The <spacegrid/> element is used to specify a spatial histogramming
grid for the energy density. Grids are constructed based on a set of,
potentially nonorthogonal, user-provided coordinate axes. The axes are
based on information available from reference_points. Voronoi grids
are based only on nearest neighbor distances between electrons and ions.
Any number of space grids can be provided to a single energy density
estimator.
spacegrid element:
parent elements:
estimator type=EnergyDensitychild elements:
origin, axis
attributes:
Name
Datatype
Values
Default
Description
coord\(^r\)text
Cartesian
Specify coordinate system
cylindrical
spherical
Voronoi
The <origin/> element gives the location of the origin for a
non-Voronoi grid.
Additional information:
p1/p2/fraction:The location of the origin is set top1+fraction*(p2-p1). If onlyp1is provided, the origin is atp1.
origin element:
parent elements:
spacegridchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
p1\(^r\)text
reference_point.labelSelect end point
p2\(^o\)text
reference_point.labelSelect end point
fraction\(^o\)real
0
Interpolation fraction
The <axis/> element represents a coordinate axis used to construct the, possibly curved, coordinate system for the histogramming grid. Three <axis/> elements must be provided to a non-Voronoi <spacegrid/> element.
axis element:
parent elements:
spacegridchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
label\(^r\)text
See below
Axis/dimension label
grid\(^r\)text
“0 1”
Grid ranges/intervals
p1\(^r\)text
reference_point.labelSelect end point
p2\(^o\)text
reference_point.labelSelect end point
scale\(^o\)real
Interpolation fraction
Additional information:
label:The allowed set of axis labels depends on the coordinate system (i.e.,spacegrid.coord). Labels arex/y/zforcoord=cartesian,r/phi/zforcoord=cylindrical,r/phi/thetaforcoord=spherical.p1/p2/scale:The axis vector is set top1+scale*(p2-p1). If onlyp1is provided, the axis vector isp1.grid:The grid specifies the histogram grid along the direction specified bylabel. The allowed grid points fall in the range [-1,1] forlabel=x/y/zor [0,1] forr/phi/theta. A grid of 10 evenly spaced points between 0 and 1 can be requested equivalently bygrid="0 (0.1) 1"orgrid="0 (10) 1."Piecewise uniform grids covering portions of the range are supported, e.g.,grid="-0.7 (10) 0.0 (20) 0.5."Note that
gridspecifies the histogram grid along the (curved) coordinate given bylabel. The axis specified byp1/p2/scaledoes not correspond one-to-one withlabelunlesslabel=x/y/z, but the full set of axes provided defines the (sheared) space on top of which the curved (e.g., spherical) coordinate system is built.
One body density matrix¶
The N-body density matrix in DMC is \(\hat{\rho}_N=\left|{\Psi_{T}}\rangle{}\langle{\Psi_{FN}}\right|\) (for VMC, substitute \(\Psi_T\) for \(\Psi_{FN}\)). The one body reduced density matrix (1RDM) is obtained by tracing out all particle coordinates but one:
In this formula, the sum is over all electron indices and \(Tr_{R_n}(*)\equiv\int dR_n\langle{R_n}\left|{*}\right|{R_n}\rangle\) with \(R_n=[r_1,...,r_{n-1},r_{n+1},...,r_N]\). When the sum is restricted over spin-up or spin-down electrons, one obtains a density matrix for each spin species. The 1RDM computed by is partitioned in this way.
In real space, the matrix elements of the 1RDM are
A more efficient and compact representation of the 1RDM is obtained by expanding in the SPOs obtained from a Hartree-Fock or DFT calculation, \(\{\phi_i\}\):
The integration over \(r'\) in (41) is inefficient when one is also interested in obtaining matrices involving energetic quantities, such as the energy density matrix of [KKR14] or the related (and more well known) generalized Fock matrix. For this reason, an approximation is introduced as follows:
For VMC, FN-DMC, FP-DMC, and RN-DMC this formula represents an exact sampling of the 1RDM corresponding to \(\hat{\rho}_N^\dagger\) (see appendix A of [KKR14] for more detail).
estimtor type=dm1b element:
parent elements:
hamiltonian, qmcchild elements:
None
attributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
dm1b
Must be dm1b
name\(^r\)text
anything
Unique name for estimator
parameters:
Name
Datatype
Values
Default
Description
basis\(^r\)text array
sposet.name(s)
Orbital basis
integrator\(^o\)text
uniform_grid uniform density
uniform_grid
Integration method
evaluator\(^o\)text
loop/matrix
loop
Evaluation method
scale\(^o\)real
\(0<scale<1\)
1.0
Scale integration cell
center\(^o\)real array(3)
any point
Center of cell
points\(^o\)integer
\(>0\)
10
Grid points in each dim
samples\(^o\)integer
\(>0\)
10
MC samples
warmup\(^o\)integer
\(>0\)
30
MC warmup
timestep\(^o\)real
\(>0\)
0.5
MC time step
use_drift\(^o\)boolean
yes/no
no
Use drift in VMC
check_overlap\(^o\)boolean
yes/no
no
Print overlap matrix
check_derivatives\(^o\)boolean
yes/no
no
Check density derivatives
acceptance_ratio\(^o\)boolean
yes/no
no
Print accept ratio
rstats\(^o\)boolean
yes/no
no
Print spatial stats
normalized\(^o\)boolean
yes/no
yes
basiscomes norm’ed
volume_normed\(^o\)boolean
yes/no
yes
basisnorm is volume
energy_matrix\(^o\)boolean
yes/no
no
Energy density matrix
Additional information:
name:Density matrix results appear instat.h5files labeled according toname.basis:Listsposet.name’s. The total set of orbitals contained in allsposet’s comprises the basis (subspace) onto which the one body density matrix is projected. This set of orbitals generally includes many virtual orbitals that are not occupied in a single reference Slater determinant.integrator:Select the method used to perform the additional single particle integration. Options areuniform_grid(uniform grid of points over the cell),uniform(uniform random sampling over the cell), anddensity(Metropolis sampling of approximate density, \(\sum_{b\in \texttt{basis}}\left|{\phi_b}\right|^2\), is not well tested, please check results carefully!). Depending on the integrator selected, different subsets of the other input parameters are active.evaluator:Select for-loop or matrix multiply implementations. Matrix is preferred for speed. Both implementations should give the same results, but please check as this has not been exhaustively tested.scale:Resize the simulation cell by scale for use as an integration volume (active forintegrator=uniform/uniform_grid).center:Translate the integration volume to center at this point (active forintegrator=uniform/ uniform_grid). Ifcenteris not provided, the scaled simulation cell is used as is.points:Number of grid points in each dimension forintegrator=uniform_grid. For example,points=10results in a uniform \(10 \times 10 \times 10\) grid over the cell.samples:Sets the number of MC samples collected for each step (active forintegrator=uniform/ density).warmup:Number of warmup Metropolis steps at the start of the run before data collection (active forintegrator=density).timestep:Drift-diffusion time step used in Metropolis sampling (active forintegrator=density).use_drift:Enable drift in Metropolis sampling (active forintegrator=density).check_overlap:Print the overlap matrix (computed via simple Riemann sums) to the log, then abort. Note that subsequent analysis based on the 1RDM is simplest if the input orbitals are orthogonal.check_derivatives:Print analytic and numerical derivatives of the approximate (sampled) density for several sample points, then abort.acceptance_ratio:Print the acceptance ratio of the density sampling to the log for each step.rstats:Print statistical information about the spatial motion of the sampled points to the log for each step.normalized:Declare whether the inputted orbitals are normalized or not. Ifnormalized=no, direct Riemann integration over a \(200 \times 200 \times 200\) grid will be used to compute the normalizations before use.volume_normed:Declare whether the inputted orbitals are normalized to the cell volume (default) or not (a norm of 1.0 is assumed in this case). Currently, B-spline orbitals coming from QE and HEG planewave orbitals native to QMCPACK are known to be volume normalized.energy_matrix:Accumulate the one body reduced energy density matrix, and write it tostat.h5. This matrix is not covered in any detail here; the interested reader is referred to [KKR14].
<estimator type="dm1b" name="DensityMatrices">
<parameter name="basis" > spo_u spo_uv </parameter>
<parameter name="evaluator" > matrix </parameter>
<parameter name="integrator" > uniform_grid </parameter>
<parameter name="points" > 4 </parameter>
<parameter name="scale" > 1.0 </parameter>
<parameter name="center" > 0 0 0 </parameter>
</estimator>
<estimator type="dm1b" name="DensityMatrices">
<parameter name="basis" > spo_u spo_uv </parameter>
<parameter name="evaluator" > matrix </parameter>
<parameter name="integrator" > uniform </parameter>
<parameter name="samples" > 64 </parameter>
<parameter name="scale" > 1.0 </parameter>
<parameter name="center" > 0 0 0 </parameter>
</estimator>
<estimator type="dm1b" name="DensityMatrices">
<parameter name="basis" > spo_u spo_uv </parameter>
<parameter name="evaluator" > matrix </parameter>
<parameter name="integrator" > density </parameter>
<parameter name="samples" > 64 </parameter>
<parameter name="timestep" > 0.5 </parameter>
<parameter name="use_drift" > no </parameter>
</estimator>
sposet initialization for density matrix use. Occupied and virtual orbital sets are created separately, then joined (basis="spo_u spo_uv").¶<sposet_builder type="bspline" href="../dft/pwscf_output/pwscf.pwscf.h5" tilematrix="1 0 0 0 1 0 0 0 1" twistnum="0" meshfactor="1.0" gpu="no" precision="single">
<sposet type="bspline" name="spo_u" group="0" size="4"/>
<sposet type="bspline" name="spo_d" group="0" size="2"/>
<sposet type="bspline" name="spo_uv" group="0" index_min="4" index_max="10"/>
</sposet_builder>
sposet initialization for density matrix use. Density matrix orbital basis created separately (basis="dm_basis").¶<sposet_builder type="bspline" href="../dft/pwscf_output/pwscf.pwscf.h5" tilematrix="1 0 0 0 1 0 0 0 1" twistnum="0" meshfactor="1.0" gpu="no" precision="single">
<sposet type="bspline" name="spo_u" group="0" size="4"/>
<sposet type="bspline" name="spo_d" group="0" size="2"/>
<sposet type="bspline" name="dm_basis" size="50" spindataset="0"/>
</sposet_builder>
Forward-Walking Estimators¶
Forward walking is a method for sampling the pure fixed-node distribution \(\langle \Phi_0 | \Phi_0\rangle\). Specifically, one multiplies each walker’s DMC mixed estimate for the observable \(\mathcal{O}\), \(\frac{\mathcal{O}(\mathbf{R})\Psi_T(\mathbf{R})}{\Psi_T(\mathbf{R})}\), by the weighting factor \(\frac{\Phi_0(\mathbf{R})}{\Psi_T(\mathbf{R})}\). As it turns out, this weighting factor for any walker \(\mathbf{R}\) is proportional to the total number of descendants the walker will have after a sufficiently long projection time \(\beta\).
To forward walk on an observable, declare a generic forward-walking
estimator within a <hamiltonian> block, and then specify the
observables to forward walk on and the forward-walking parameters. Here
is a summary.
estimator type=ForwardWalking element:
parent elements:
hamiltonian, qmcchild elements:
Observableattributes:
Name
Datatype
Values
Default
Description
type\(^r\)text
ForwardWalking
Must be “ForwardWalking”
name\(^r\)text
anything
any
Unique name for estimator
Observable element:
parent elements:
estimator, hamiltonian, qmcchild elements:
None
Name
Datatype
Values
Default
Description
name\(^r\)text
anything
any
Registered name of existing estimator on which to forward walk
max\(^r\)integer
\(>0\)
Maximum projection time in steps (
max\(=\beta/\tau\))
frequency\(^r\)text
\(\geq 1\)
Dump data only for every
frequency-th toscalar.datfile
Additional information:
Cost: Because histories of observables up to
maxtime steps have to be stored, the memory cost of storing the nonforward-walked observables variables should be multiplied by \(\texttt{max}\). Although this is not an issue for items such as potential energy, it could be prohibitive for observables such as density, forces, etc.Naming Convention: Forward-walked observables are automatically named
FWE_name_i, whereiis the forward-walked expectation value at time stepi, andnameis whatever name appears in the<Observable>block. This is also how it will appear in thescalar.datfile.
In the following example case, QMCPACK forward walks on the potential energy for 300 time steps and dumps the forward-walked value at every time step.
<estimator name="fw" type="ForwardWalking">
<Observable name="LocalPotential" max="300" frequency="1"/>
<!--- Additional Observable blocks go here -->
</estimator>
“Force” estimators¶
QMCPACK supports force estimation by use of the Chiesa-Ceperly-Zhang (CCZ) estimator. Currently, open and periodic boundary conditions are supported but for all-electron calculations only.
Without loss of generality, the CCZ estimator for the z-component of the force on an ion centered at the origin is given by the following expression:
Z is the ionic charge, \(M\) is the degree of the smoothing polynomial, \(\mathcal{R}\) is a real-space cutoff of the sphere within which the bare-force estimator is smoothed, and \(c_\ell\) are predetermined coefficients. These coefficients are chosen to minimize the weighted mean square error between the bare force estimate and the s-wave filtered estimator. Specifically,
Here, \(m\) is the weighting exponent, \(f_z(r)\) is the unfiltered radial force density for the z force component, and \(\tilde{f}_z(r)\) is the smoothed polynomial function for the same force density. The reader is invited to refer to the original paper for a more thorough explanation of the methodology, but with the notation in hand, QMCPACK takes the following parameters.
estimator type=Force element:
parent elements:
hamiltonian, qmcchild elements:
parameterattributes:
Name
Datatype
Values
Default
Description
mode\(^o\)text
See above
bare
Select estimator type
lrmethod\(^o\)text
ewald or srcoul
ewald
Select long-range potential breakup method
type\(^r\)text
Force
Must be “Force”
name\(^o\)text
Anything
ForceBase
Unique name for this estimator
pbc\(^o\)boolean
yes/no
yes
Using periodic BCs or not
addionion\(^o\)boolean
yes/no
no
Add the ion-ion force contribution to output force estimate
parameters:
Name
Datatype
Values
Default
Description
rcut\(^o\)real
\(>0\)
1.0
Real-space cutoff \(\mathcal{R}\) in bohr
nbasis\(^o\)integer
\(>0\)
2
Degree of smoothing polynomial \(M\)
weightexp\(^o\)integer
\(>0\)
2
\(\chi^2\) weighting exponent :math`m`
Additional information:
Naming Convention: The unique identifier
nameis appended withname_X_Yin thescalar.datfile, whereXis the ion ID number andYis the component ID (an integer with x=0, y=1, z=2). All force components for all ions are computed and dumped to thescalar.datfile.Long-range breakup: With periodic boundary conditions, it is important to converge the lattice sum when calculating Coulomb contribution to the forces. As a quick test, increase the
LR_dim_cutoffparameter until ion-ion forces are converged. The Ewald method converges more slowly than optimized method, but the optimized method can break down in edge cases, eg. too largeLR_dim_cutoff.Miscellaneous: Usually, the default choice of
weightexpis sufficient. Different combinations ofrcutandnbasisshould be tested though to minimize variance and bias. There is, of course, a tradeoff, with largernbasisand smallerrcutleading to smaller biases and larger variances.
The following is an example use case.
<simulationcell>
...
<parameter name="LR_handler"> opt_breakup_original </parameter>
<parameter name="LR_dim_cutoff"> 20 </parameter>
</simulationcell>
<hamiltonian>
<estimator name="F" type="Force" mode="cep" addionion="yes">
<parameter name="rcut">0.1</parameter>
<parameter name="nbasis">4</parameter>
<parameter name="weightexp">2</parameter>
</estimator>
</hamiltonian>
Stress estimators¶
QMCPACK takes the following parameters.
parent elements:
hamiltonianattributes:
Name
Datatype
Values
Default
Description
mode\(^r\)text
stress
bare
Must be “stress”
type\(^r\)text
Force
Must be “Force”
source\(^r\)text
ion0
Name of ion particleset
name\(^o\)text
Anything
ForceBase
Unique name for this estimator
addionion\(^o\)boolean
yes/no
no
Add the ion-ion stress contribution to output
Additional information:
Naming Convention: The unique identifier
nameis appended withname_X_Yin thescalar.datfile, whereXandYare the component IDs (an integer with x=0, y=1, z=2).Long-range breakup: With periodic boundary conditions, it is important to converge the lattice sum when calculating Coulomb contribution to the forces. As a quick test, increase the
LR_dim_cutoffparameter until ion-ion stresses are converged. Check using QE “Ewald contribution”, for example. The stress estimator is implemented only with the Ewald method.
The following is an example use case.
<simulationcell>
...
<parameter name="LR_handler"> ewald </parameter>
<parameter name="LR_dim_cutoff"> 45 </parameter>
</simulationcell>
<hamiltonian>
<estimator name="S" type="Force" mode="stress" source="ion0"/>
</hamiltonian>
- CCMH06
Simone Chiesa, David M. Ceperley, Richard M. Martin, and Markus Holzmann. Finite-size error in many-body simulations with long-range interactions. Phys. Rev. Lett., 97:076404, August 2006. doi:10.1103/PhysRevLett.97.076404.
- KKR14(1,2,3)
Jaron T. Krogel, Jeongnim Kim, and Fernando A. Reboredo. Energy density matrix formalism for interacting quantum systems: quantum monte carlo study. Phys. Rev. B, 90:035125, July 2014. doi:10.1103/PhysRevB.90.035125.
- KYKC13
Jaron T. Krogel, Min Yu, Jeongnim Kim, and David M. Ceperley. Quantum energy density: improved efficiency for quantum monte carlo calculations. Phys. Rev. B, 88:035137, July 2013. doi:10.1103/PhysRevB.88.035137.
- MSC91
Lubos Mitas, Eric L. Shirley, and David M. Ceperley. Nonlocal pseudopotentials and diffusion monte carlo. The Journal of Chemical Physics, 95(5):3467–3475, 1991. doi:10.1063/1.460849.
- NC95
Vincent Natoli and David M. Ceperley. An optimized method for treating long-range potentials. Journal of Computational Physics, 117(1):171–178, 1995. URL: http://www.sciencedirect.com/science/article/pii/S0021999185710546, doi:10.1006/jcph.1995.1054.
- ZBMAlfe19
Andrea Zen, Jan Gerit Brandenburg, Angelos Michaelides, and Dario Alfè. A new scheme for fixed node diffusion quantum monte carlo with pseudopotentials: improving reproducibility and reducing the trial-wave-function bias. The Journal of Chemical Physics, 151(13):134105, October 2019. doi:10.1063/1.5119729.
Quantum Monte Carlo Methods¶
qmc factory element:
Parent elements
simulation, looptype selector
methodattributetype options:
vmc
Variational Monte Carlo
linear
Wavefunction optimization with linear method
dmc
Diffusion Monte Carlo
rmc
Reptation Monte Carlo
shared attributes:
Name
Datatype
Values
Default
Description
methodtext
listed above
invalid
QMC driver
movetext
pbyp, alle
pbyp
Method used to move electrons
gputext
yes/no
dep.
Use the GPU
tracetext
no
???
checkpointinteger
-1, 0, n
-1
Checkpoint frequency
recordinteger
n
0
Save configuration ever n steps
targettext
???
completedtext
???
appendtext
yes/no
no
???
Additional information:
move: There are two ways to move electrons. The more used method is the particle-by-particle move. In this method, only one electron is moved for acceptance or rejection. The other method is the all-electron move; namely, all the electrons are moved once for testing acceptance or rejection.gpu: When the executable is compiled with CUDA, the target computing device can be chosen by this switch. With a regular CPU-only compilation, this option is not effective.checkpoint: This enables and disables checkpointing and specifying the frequency of output. Possible values are:[-1] No checkpoint (default setting).
[0] Dump after the completion of a QMC section.
[n] Dump after every \(n\) blocks. Also dump at the end of the run.
The particle configurations are written to a .config.h5 file.
<qmc method="dmc" move="pbyp" checkpoint="0">
<parameter name="timestep"> 0.004 </parameter>
<parameter name="blocks"> 100 </parameter>
<parameter name="steps"> 400 </parameter>
</qmc>
The checkpoint flag instructs QMCPACK to output walker configurations. This also
works in VMC. This outputs an h5 file with the name projectid.run-number.config.h5.
Check that this file exists before attempting a restart.
To continue a run, specify the mcwalkerset element before your VMC/DMC block:
<mcwalkerset fileroot="BH.s002" version="0 6" collected="yes"/>
<qmc method="dmc" move="pbyp" checkpoint="0">
<parameter name="timestep"> 0.004 </parameter>
<parameter name="blocks"> 100 </parameter>
<parameter name="steps"> 400 </parameter>
</qmc>
BH is the project id, and s002 is the calculation number to read in the walkers from the previous run.
In the project id section, make sure that the series number is different from any existing ones to avoid overwriting them.
Variational Monte Carlo¶
vmc method:
parameters:
Name
Datatype
Values
Default
Description
walkersinteger
\(> 0\)
dep.
Number of walkers per MPI task
blocksinteger
\(\geq 0\)
1
Number of blocks
stepsinteger
\(\geq 0\)
1
Number of steps per block
warmupstepsinteger
\(\geq 0\)
0
Number of steps for warming up
substepsinteger
\(\geq 0\)
1
Number of substeps per step
usedrifttext
yes,no
yes
Use the algorithm with drift
timestepreal
\(> 0\)
0.1
Time step for each electron move
samplesinteger
\(\geq 0\)
0
Number of walker samples for DMC/optimization
stepsbetweensamplesinteger
\(> 0\)
1
Period of sample accumulation
samplesperthreadinteger
\(\geq 0\)
0
Number of samples per thread
storeconfigsinteger
all values
0
Show configurations o
blocks_between_recomputeinteger
\(\geq 0\)
dep.
Wavefunction recompute frequency
spinMovestext
yes,no
no
Whether or not to sample the electron spins
spinMassreal
\(> 0\)
1.0
Effective mass for spin sampling
Additional information:
walkersThe number of walkers per MPI task. The initial default number of ixml{walkers} is one per OpenMP thread or per MPI task if threading is disabled. The number is rounded down to a multiple of the number of threads with a minimum of one per thread to ensure perfect load balancing. One walker per thread is created in the event fewerwalkersthan threads are requested.blocksThis parameter is universal for all the QMC methods. The MC processes are divided into a number ofblocks, each containing a number of steps. At the end of each block, the statistics accumulated in the block are dumped into files, e.g.,scalar.dat. Typically, each block should have a sufficient number of steps that the I/O at the end of each block is negligible compared with the computational cost. Each block should not take so long that monitoring its progress is difficult. There should be a sufficient number ofblocksto perform statistical analysis.warmupsteps-warmupstepsare used only for equilibration. Property measurements are not performed during warm-up steps.steps-stepsare the number of energy and other property measurements to perform per block.substepsFor each substep, an attempt is made to move each of the electrons once only by either particle-by-particle or an all-electron move. Because the local energy is evaluated only at each full step and not each substep,substepsare computationally cheaper and can be used to reduce the correlation between property measurements at a lower cost.usedriftThe VMC is implemented in two algorithms with or without drift. In the no-drift algorithm, the move of each electron is proposed with a Gaussian distribution. The standard deviation is chosen as the time step input. In the drift algorithm, electrons are moved by Langevin dynamics.timestepThe meaning of time step depends on whether or not the drift is used. In general, larger time steps reduce the time correlation but might also reduce the acceptance ratio, reducing overall statistical efficiency. For VMC, typically the acceptance ratio should be close to 50% for an efficient simulation.samplesSeperate from conventional energy and other property measurements, samples refers to storing whole electron configurations in memory (“walker samples”) as would be needed by subsequent wavefunction optimization or DMC steps. A standard VMC run to measure the energy does not need samples to be set.\[\texttt{samples}= \frac{\texttt{blocks}\cdot\texttt{steps}\cdot\texttt{walkers}}{\texttt{stepsbetweensamples}}\cdot\texttt{number of MPI tasks}\]samplesperthreadThis is an alternative way to set the target amount of samples and can be useful when preparing a stored population for a subsequent DMC calculation.\[\texttt{samplesperthread}= \frac{\texttt{blocks}\cdot\texttt{steps}}{\texttt{stepsbetweensamples}}\]stepsbetweensamplesBecause samples generated by consecutive steps are correlated, havingstepsbetweensampleslarger than 1 can be used to reduces that correlation. In practice, using larger substeps is cheaper than usingstepsbetweensamplesto decorrelate samples.storeconfigsIfstoreconfigsis set to a nonzero value, then electron configurations during the VMC run are saved to files.blocks_between_recomputeRecompute the accuracy critical determinant part of the wavefunction from scratch: =1 by default when using mixed precision. =0 (no recompute) by default when not using mixed precision. Recomputing introduces a performance penalty dependent on system size.spinMovesDetermines whether or not the spin variables are sampled following
spinMassIf spin sampling is on usingspinMoves== yes, the spin mass determines the rate
of spin sampling, resulting in an effective spin timestep \(\tau_s = \frac{\tau}{\mu_s}\).
An example VMC section for a simple VMC run:
<qmc method="vmc" move="pbyp">
<estimator name="LocalEnergy" hdf5="no"/>
<parameter name="walkers"> 256 </parameter>
<parameter name="warmupSteps"> 100 </parameter>
<parameter name="substeps"> 5 </parameter>
<parameter name="blocks"> 20 </parameter>
<parameter name="steps"> 100 </parameter>
<parameter name="timestep"> 1.0 </parameter>
<parameter name="usedrift"> yes </parameter>
</qmc>
Here we set 256 walkers per MPI, have a brief initial equilibration of 100 steps, and then have 20 blocks of 100 steps with 5 substeps each.
The following is an example of VMC section storing configurations (walker samples) for optimization.
<qmc method="vmc" move="pbyp" gpu="yes">
<estimator name="LocalEnergy" hdf5="no"/>
<parameter name="walkers"> 256 </parameter>
<parameter name="samples"> 2867200 </parameter>
<parameter name="stepsbetweensamples"> 1 </parameter>
<parameter name="substeps"> 5 </parameter>
<parameter name="warmupSteps"> 5 </parameter>
<parameter name="blocks"> 70 </parameter>
<parameter name="timestep"> 1.0 </parameter>
<parameter name="usedrift"> no </parameter>
</qmc>
Wavefunction optimization¶
Optimizing wavefunction is critical in all kinds of real-space QMC calculations because it significantly improves both the accuracy and efficiency of computation. However, it is very difficult to directly adopt deterministic minimization approaches because of the stochastic nature of evaluating quantities with MC. Thanks to the algorithmic breakthrough during the first decade of this century and the tremendous computer power available, it is now feasible to optimize tens of thousands of parameters in a wavefunction for a solid or molecule. QMCPACK has multiple optimizers implemented based on the state-of-the-art linear method. We are continually improving our optimizers for robustness and friendliness and are trying to provide a single solution. Because of the large variation of wavefunction types carrying distinct characteristics, using several optimizers might be needed in some cases. We strongly suggested reading recommendations from the experts who maintain these optimizers.
A typical optimization block looks like the following. It starts with method=”linear” and contains three blocks of parameters.
<loop max="10">
<qmc method="linear" move="pbyp" gpu="yes">
<!-- Specify the VMC options -->
<parameter name="walkers"> 256 </parameter>
<parameter name="samples"> 2867200 </parameter>
<parameter name="stepsbetweensamples"> 1 </parameter>
<parameter name="substeps"> 5 </parameter>
<parameter name="warmupSteps"> 5 </parameter>
<parameter name="blocks"> 70 </parameter>
<parameter name="timestep"> 1.0 </parameter>
<parameter name="usedrift"> no </parameter>
<estimator name="LocalEnergy" hdf5="no"/>
...
<!-- Specify the correlated sampling options and define the cost function -->
<parameter name="minwalkers"> 0.3 </parameter>
<cost name="energy"> 0.95 </cost>
<cost name="unreweightedvariance"> 0.00 </cost>
<cost name="reweightedvariance"> 0.05 </cost>
...
<!-- Specify the optimizer options -->
<parameter name="MinMethod"> OneShiftOnly </parameter>
...
</qmc>
</loop>
- Loop is helpful to repeatedly execute identical optimization blocks.
- The first part is highly identical to a regular VMC block.
- The second part is to specify the correlated sampling options and
define the cost function.
- The last part is used to specify the options of different optimizers,
which can be very distinct from one to another.
VMC run for the optimization¶
The VMC calculation for the wavefunction optimization has a strict requirement
that samples or samplesperthread must be specified because of the optimizer needs for the stored samples.
The input parameters of this part are identical to the VMC method.
Recommendations:
Run the inclusive VMC calculation correctly and efficiently because this takes a significant amount of time during optimization. For example, make sure the derived
stepsper block is 1 and use largersubstepsto control the correlation betweensamples.A reasonable starting wavefunction is necessary. A lot of optimization fails because of a bad wavefunction starting point. The sign of a bad initial wavefunction includes but is not limited to a very long equilibration time, low acceptance ratio, and huge variance. The first thing to do after a failed optimization is to check the information provided by the VMC calculation via
*.scalar.dat files.
Optimizers¶
QMCPACK implements a number of different optimizers each with different priorities for accuracy, convergence, memory usage, and stability. The optimizers can be switched among “OneShiftOnly” (default), “adaptive,” “descent,” “hybrid,” and “quartic” (old) using the following line in the optimization block:
<parameter name="MinMethod"> THE METHOD YOU LIKE </parameter>
OneShiftOnly Optimizer¶
The OneShiftOnly optimizer targets a fast optimization by moving parameters more aggressively. It works with OpenMP and GPU and can be considered for large systems.
This method relies on the effective weight of correlated sampling rather than the cost function value to justify a new set of parameters.
If the effective weight is larger than minwalkers, the new set is taken whether or not the cost function value decreases.
If a proposed set is rejected, the standard output prints the measured ratio of effective samples to the total number of samples
and adjustment on minwalkers can be made if needed.
linear method:
parameters:
Name
Datatype
Values
Default
Description
shift_ireal
\(> 0\)
0.01
Direct stabilizer added to the Hamiltonian matrix
shift_sreal
\(> 0\)
1.00
Initial stabilizer based on the overlap matrix
Additional information:
shift_iThis is the direct term added to the diagonal of the Hamiltonian matrix. It provides more stable but slower optimization with a large value.shift_sThis is the initial value of the stabilizer based on the overlap matrix added to the Hamiltonian matrix. It provides more stable but slower optimization with a large value. The used value is auto-adjusted by the optimizer.
Recommendations:
Default
shift_i,shift_sshould be fine.For hard cases, increasing
shift_i(by a factor of 5 or 10) can significantly stabilize the optimization by reducing the pace towards the optimal parameter set.If the VMC energy of the last optimization iterations grows significantly, increase
minwalkerscloser to 1 and make the optimization stable.If the first iterations of optimization are rejected on a reasonable initial wavefunction, lower the
minwalkersvalue based on the measured value printed in the standard output to accept the move.
We recommended using this optimizer in two sections with a very small minwalkers in the first and a large value in the second, such as the following.
In the very beginning, parameters are far away from optimal values and large changes are proposed by the optimizer.
Having a small minwalkers makes it much easier to accept these changes.
When the energy gradually converges, we can have a large minwalkers to avoid risky parameter sets.
<loop max="6">
<qmc method="linear" move="pbyp" gpu="yes">
<!-- Specify the VMC options -->
<parameter name="walkers"> 1 </parameter>
<parameter name="samples"> 10000 </parameter>
<parameter name="stepsbetweensamples"> 1 </parameter>
<parameter name="substeps"> 5 </parameter>
<parameter name="warmupSteps"> 5 </parameter>
<parameter name="blocks"> 25 </parameter>
<parameter name="timestep"> 1.0 </parameter>
<parameter name="usedrift"> no </parameter>
<estimator name="LocalEnergy" hdf5="no"/>
<!-- Specify the optimizer options -->
<parameter name="MinMethod"> OneShiftOnly </parameter>
<parameter name="minwalkers"> 1e-4 </parameter>
</qmc>
</loop>
<loop max="12">
<qmc method="linear" move="pbyp" gpu="yes">
<!-- Specify the VMC options -->
<parameter name="walkers"> 1 </parameter>
<parameter name="samples"> 20000 </parameter>
<parameter name="stepsbetweensamples"> 1 </parameter>
<parameter name="substeps"> 5 </parameter>
<parameter name="warmupSteps"> 2 </parameter>
<parameter name="blocks"> 50 </parameter>
<parameter name="timestep"> 1.0 </parameter>
<parameter name="usedrift"> no </parameter>
<estimator name="LocalEnergy" hdf5="no"/>
<!-- Specify the optimizer options -->
<parameter name="MinMethod"> OneShiftOnly </parameter>
<parameter name="minwalkers"> 0.5 </parameter>
</qmc>
</loop>
For each optimization step, you will see
The new set of parameters is valid. Updating the trial wave function!
or
The new set of parameters is not valid. Revert to the old set!
Occasional rejection is fine. Frequent rejection indicates potential
problems, and users should inspect the VMC calculation or change
optimization strategy. To track the progress of optimization, use the
command qmca -q ev *.scalar.dat to look at the VMC energy and
variance for each optimization step.
Adaptive Organizer¶
The default setting of the adaptive optimizer is to construct the linear method Hamiltonian and overlap matrices explicitly and add different shifts to the Hamiltonian matrix as “stabilizers.” The generalized eigenvalue problem is solved for each shift to obtain updates to the wavefunction parameters. Then a correlated sampling is performed for each shift’s updated wavefunction and the initial trial wavefunction using the middle shift’s updated wavefunction as the guiding function. The cost function for these wavefunctions is compared, and the update corresponding to the best cost function is selected. In the next iteration, the median magnitude of the stabilizers is set to the magnitude that generated the best update in the current iteration, thus adapting the magnitude of the stabilizers automatically.
When the trial wavefunction contains more than 10,000 parameters, constructing and storing the linear method matrices could become a memory bottleneck. To avoid explicit construction of these matrices, the adaptive optimizer implements the block linear method (BLM) approach. [ZN17] The BLM tries to find an approximate solution \(\vec{c}_{opt}\) to the standard LM generalized eigenvalue problem by dividing the variable space into a number of blocks and making intelligent estimates for which directions within those blocks will be most important for constructing \(\vec{c}_{opt}\), which is then obtained by solving a smaller, more memory-efficient eigenproblem in the basis of these supposedly important block-wise directions.
linear method:
parameters:
Name
Datatype
Values
Default
Description
max_relative_changereal
\(> 0\)
10.0
Allowed change in cost function
max_param_changereal
\(> 0\)
0.3
Allowed change in wavefunction parameter
shift_ireal
\(> 0\)
0.01
Initial diagonal stabilizer added to the Hamiltonian matrix
shift_sreal
\(> 0\)
1.00
Initial overlap-based stabilizer added to the Hamiltonian matrix
target_shift_ireal
any
-1.0
Diagonal stabilizer value aimed for during adaptive method (disabled if \(\leq 0\))
cost_increase_tolreal
\(\geq 0\)
0.0
Tolerance for cost function increases
chase_lowesttext
yes, no
yes
Chase the lowest eigenvector in iterative solver
chase_closesttext
yes, no
no
Chase the eigenvector closest to initial guess
block_lmtext
yes, no
no
Use BLM
blocksinteger
\(> 0\)
Number of blocks in BLM
noldsinteger
\(> 0\)
Number of old update vectors used in BLM
nkeptinteger
\(> 0\)
Number of eigenvectors to keep per block in BLM
Additional information:
shift_iThis is the initial coefficient used to scale the diagonal stabilizer. More stable but slower optimization is expected with a large value. The adaptive method will automatically adjust this value after each linear method iteration.shift_sThis is the initial coefficient used to scale the overlap-based stabilizer. More stable but slower optimization is expected with a large value. The adaptive method will automatically adjust this value after each linear method iteration.target_shift_iIf set greater than zero, the adaptive method will choose the update whose shift_i value is closest to this target value so long as the associated cost is within cost_increase_tol of the lowest cost. Disable this behavior by setting target_shift_i to a negative number.cost_increase_tolTolerance for cost function increases when selecting the best shift.nblocksThis is the number of blocks used in BLM. The amount of memory required to store LM matrices decreases as the number of blocks increases. But the error introduced by BLM would increase as the number of blocks increases.noldsIn BLM, the interblock correlation is accounted for by including a small number of wavefunction update vectors outside the block. Larger would include more interblock correlation and more accurate results but also higher memory requirements.nkeptThis is the number of update directions retained from each block in the BLM. If all directions are retained in each block, then the BLM becomes equivalent to the standard LM. Retaining five or fewer directions per block is often sufficient.
Recommendations:
Default
shift_i,shift_sshould be fine.When there are fewer than about 5,000 variables being optimized, the traditional LM is preferred because it has a lower overhead than the BLM when the number of variables is small.
Initial experience with the BLM suggests that a few hundred blocks and a handful of and often provide a good balance between memory use and accuracy. In general, using fewer blocks should be more accurate but would require more memory.
<loop max="15">
<qmc method="linear" move="pbyp">
<!-- Specify the VMC options -->
<parameter name="walkers"> 1 </parameter>
<parameter name="samples"> 20000 </parameter>
<parameter name="stepsbetweensamples"> 1 </parameter>
<parameter name="substeps"> 5 </parameter>
<parameter name="warmupSteps"> 5 </parameter>
<parameter name="blocks"> 50 </parameter>
<parameter name="timestep"> 1.0 </parameter>
<parameter name="usedrift"> no </parameter>
<estimator name="LocalEnergy" hdf5="no"/>
<!-- Specify the correlated sampling options and define the cost function -->
<cost name="energy"> 1.00 </cost>
<cost name="unreweightedvariance"> 0.00 </cost>
<cost name="reweightedvariance"> 0.00 </cost>
<!-- Specify the optimizer options -->
<parameter name="MinMethod">adaptive</parameter>
<parameter name="max_relative_cost_change">10.0</parameter>
<parameter name="shift_i"> 1.00 </parameter>
<parameter name="shift_s"> 1.00 </parameter>
<parameter name="max_param_change"> 0.3 </parameter>
<parameter name="chase_lowest"> yes </parameter>
<parameter name="chase_closest"> yes </parameter>
<parameter name="block_lm"> no </parameter>
<!-- Specify the BLM specific options if needed
<parameter name="nblocks"> 100 </parameter>
<parameter name="nolds"> 5 </parameter>
<parameter name="nkept"> 3 </parameter>
-->
</qmc>
</loop>
The adaptive optimizer is also able to optimize individual excited states directly. [ZN16] In this case, it tries to minimize the following function:
The global minimum of this function corresponds to the state whose energy lies immediately above the shift parameter \(\omega\) in the energy spectrum. For example, if \(\omega\) were placed in between the ground state energy and the first excited state energy and the wavefunction ansatz was capable of a good description for the first excited state, then the wavefunction would be optimized for the first excited state. Note that if the ansatz is not capable of a good description of the excited state in question, the optimization could converge to a different state, as is known to occur in some circumstances for traditional ground state optimizations. Note also that the ground state can be targeted by this method by choosing \(\omega\) to be below the ground state energy, although we should stress that this is not the same thing as a traditional ground state optimization and will in general give a slightly different wavefunction. Excited state targeting requires two additional parameters, as shown in the following table.
Excited state targeting:
parameters:
Name
Datatype
Values
Default
Description
targetExcitedtext
yes, no
no
Whether to use the excited state targeting optimization
omegareal
real numbers
none
Energy shift used to target different excited states
Excited state recommendations:
Because of the finite variance in any approximate wavefunction, we recommended setting \(\omega=\omega_0-\sigma\), where \(\omega_0\) is placed just below the energy of the targeted state and \(\sigma^2\) is the energy variance.
To obtain an unbiased excitation energy, the ground state should be optimized with the excited state variational principle as well by setting
omegabelow the ground state energy. Note that using the ground state variational principle for the ground state and the excited state variational principle for the excited state creates a bias in favor of the ground state.
Descent Optimizer¶
Gradient descent algorithms are an alternative set of optimization methods to the OneShiftOnly and adaptive optimizers based on the linear method. These methods use only first derivatives to optimize trial wave functions and convergence can be accelerated by retaining a memory of previous derivative values. Multiple flavors of accelerated descent methods are available. They differ in details such as the schemes for adaptive adjustment of step sizes. [ON19] Descent algorithms avoid the construction of matrices that occurs in the linear method and consequently can be applied to larger sets of optimizable parameters. Parameters for descent are shown in the table below.
descent method:
parameters:
Name
Datatype
Values
Default
Description
flavortext
RMSprop, Random, ADAM, AMSGrad
RMSprop
Particular type of descent method
Ramp_etatext
yes, no
no
Whether to gradually ramp up step sizes
Ramp_numinteger
\(> 0\)
30
Number of steps over which to ramp up step size
TJF_2Body_etareal
\(> 0\)
0.01
Step size for two body Jastrow parameters
TJF_1Body_etareal
\(> 0\)
0.01
Step size for one body Jastrow parameters
F_etareal
\(> 0\)
0.001
Step size for number counting Jastrow F matrix parameters
Gauss_etareal
\(> 0\)
0.001
Step size for number counting Jastrow gaussian basis parameters
CI_etareal
\(> 0\)
0.01
Step size for CI parameters
Orb_etareal
\(> 0\)
0.001
Step size for orbital parameters
collection_stepreal
\(> 0\)
0.01
Step number to start collecting samples for final averages
compute_stepreal
\(> 0\)
0.001
Step number to start computing averaged from stored history
print_derivsreal
yes, no
no
Whether to print parameter derivatives
These descent algortihms have been extended to the optimization of the same excited state functional as the adaptive LM. [LON20] This also allows the hybrid optimizer discussed below to be applied to excited states. The relevant parameters are the same as for targeting excited states with the adaptive optimizer above.
Additional information and recommendations:
It is generally advantageous to set different step sizes for different types of parameters. More nonlinear parameters such as those for number counting Jastrow factors or orbitals typically require smaller steps sizes than those for CI coefficients or traditional Jastrow parameters. There are defaults for several parameter types and a default of .001 has been chosen for all other parameters.
The ability to gradually ramp up step sizes to their input values is useful for avoiding spikes in the average local energy during early iterations of descent optimization. This initial rise in the energy occurs as a memory of past gradients is being built up and it may be possible for the energy to recover without ramping if there are enough iterations in the optimization.
The step sizes chosen can have a substantial influence on the quality of the optimization and the final variational energy achieved. Larger step sizes may be helpful if there is reason to think the descent optimization is not reaching the minimum energy. There are also additional hyperparameters in the descent algorithms with default values. [ON19] They seem to have limited influence on the effectiveness of the optimization compared to step sizes, but users can adjust them within the source code of the descent engine if they wish.
The sampling effort for individual descent steps can be small compared that for linear method iterations as shown in the example input below. Something in the range of 10,000 to 30,000 seems sufficient for molecules with tens of electrons. However, descent optimizations may require anywhere from a few hundred to a few thousand iterations.
For reporting quantities such as a final energy and associated uncertainty, an average over many descent steps can be taken. The parameters for
collection_stepandcompute_stephelp automate this task.After the descent iteration specified by
collection_step, a history of local energy values will be kept for determining a final error and average, which will be computed and given in the output once the iteration specified bycompute_stepis reached. For reasonable results, this procedure should use descent steps near the end of the optimization when the wave function parameters are essentially no longer changing.
In cases where a descent optimization struggles to reach the minimum and a linear method optimization is not possible or unsatisfactory, it may be useful to try the hybrid optimization approach described in the next subsection.
<loop max="2000">
<qmc method="linear" move="pbyp" checkpoint="-1" gpu="no">
<!-- VMC inputs -->
<parameter name="blocks">2000</parameter>
<parameter name="steps">1</parameter>
<parameter name="samples">20000</parameter>
<parameter name="warmupsteps">100</parameter>
<parameter name="timestep">0.05</parameter>
<parameter name="MinMethod">descent</parameter>
<estimator name="LocalEnergy" hdf5="no"/>
<parameter name="usebuffer">yes</parameter>
<estimator name="LocalEnergy" hdf5="no"/>
<!-- Descent Inputs -->
<parameter name="flavor">RMSprop</parameter>
<parameter name="Ramp_eta">no</parameter>
<parameter name="Ramp_num">30</parameter>
<parameter name="TJF_2Body_eta">.02</parameter>
<parameter name="TJF_1Body_eta">.02</parameter>
<parameter name="F_eta">.001</parameter>
<parameter name="Gauss_eta">.001</parameter>
<parameter name="CI_eta">.1</parameter>
<parameter name="Orb_eta">.0001</parameter>
<parameter name="collection_step">500</parameter>
<parameter name="compute_step">998</parameter>
<parameter name="targetExcited"> yes </parameter>
<parameter name="targetExcited"> -11.4 </parameter>
<parameter name="print_derivs">no</parameter>
</qmc>
</loop>
Hybrid Optimizer¶
Another optimization option is to use a hybrid combination of accelerated descent and blocked linear method.
It provides a means to retain the advantages of both individual methods while scaling to large numbers of parameters beyond the traditional 10,000 parameter limit of the linear method. [ON19]
In a hybrid optimization, alternating sections of descent and BLM optimization are used.
Gradient descent is used to identify the previous important directions in parameter space used by the BLM, the number of which is set by the nold input for the BLM.
Over the course of a section of descent, vectors of parameter differences are stored and then passed to the linear method engine after the optimization changes to the BLM.
One motivation for including sections of descent is to counteract noise in linear method updates due to uncertainties in its step direction and allow for a smoother movement to the minimum.
There are two additional parameters used in the hybrid optimization and it requires a slightly different format of input to specify the constituent methods as shown below in the example.
descent method:
parameters:
Name
Datatype
Values
Default
Description
num_updatesinteger
\(> 0\)
Number of steps for a method
Stored_Vectorsinteger
\(> 0\)
5
Number of vectors to transfer to BLM
<loop max="203">
<qmc method="linear" move="pbyp" checkpoint="-1" gpu="no">
<parameter name="Minmethod"> hybrid </parameter>
<optimizer num_updates="100">
<parameter name="blocks">1000</parameter>
<parameter name="steps">1</parameter>
<parameter name="samples">20000</parameter>
<parameter name="warmupsteps">1000</parameter>
<parameter name="timestep">0.05</parameter>
<estimator name="LocalEnergy" hdf5="no"/>
<parameter name="Minmethod"> descent </parameter>
<parameter name="Stored_Vectors">5</parameter>
<parameter name="flavor">RMSprop</parameter>
<parameter name="TJF_2Body_eta">.01</parameter>
<parameter name="TJF_1Body_eta">.01</parameter>
<parameter name="CI_eta">.1</parameter>
<parameter name="Ramp_eta">no</parameter>
<parameter name="Ramp_num">10</parameter>
</optimizer>
<optimizer num_updates="3">
<parameter name="blocks">2000</parameter>
<parameter name="steps">1</parameter>
<parameter name="samples">1000000</parameter>
<parameter name="warmupsteps">1000</parameter>
<parameter name="timestep">0.05</parameter>
<estimator name="LocalEnergy" hdf5="no"/>
<parameter name="Minmethod"> adaptive </parameter>
<parameter name="max_relative_cost_change">10.0</parameter>
<parameter name="max_param_change">3</parameter>
<parameter name="shift_i">0.01</parameter>
<parameter name="shift_s">1.00</parameter>
<parameter name="block_lm">yes</parameter>
<parameter name="nblocks">2</parameter>
<parameter name="nolds">5</parameter>
<parameter name="nkept">5</parameter>
</optimizer>
</qmc>
</loop>
Additional information and recommendations:
In the example above, the input for
loopgives the total number of steps for the full optimization while the inputs fornum_updatesspecify the number of steps in the constituent methods. For this case, the optimization would begin with 100 steps of descent using the parameters in the firstoptimizerblock and then switch to the BLM for 3 steps before switching back to descent for the final 100 iterations of the total of 203.The design of the hybrid method allows for more than two
optimizerblocks to be used and the optimization will cycle through the individual methods. However, the effectiveness of this in terms of the quality of optimization results is unexplored.It can be useful to follow a hybrid optimization with a section of pure descent optimization and take an average energy over the last few hundred iterations as the final variational energy. This approach can achieve a lower statistical uncertainty on the energy for less overall sampling effort compared to what a pure linear method optimization would require. The
collection_stepandcompute_stepparameters discussed earlier for descent are useful for setting up the descent engine to do this averaging on its own.
Quartic Optimizer¶
This is an older optimizer method retained for compatibility. We recommend starting with the newest OneShiftOnly or adaptive optimizers. The quartic optimizer fits a quartic polynomial to 7 values of the cost function obtained using reweighting along the chosen direction and determines the optimal move. This optimizer is very robust but is a bit conservative when accepting new steps, especially when large parameters changes are proposed.
linear method:
parameters:
Name
Datatype
Values
Default
Description
bigchangereal
\(> 0\)
50.0
Largest parameter change allowed
alloweddifferencereal
\(> 0\)
1e-4
Allowed increase in energy
exp0real
any value
-16.0
Initial value for stabilizer
stabilizerscalereal
\(> 0\)
2.0
Increase in value of
exp0between iterations
nstabilizersinteger
\(> 0\)
3
Number of stabilizers to try
max_itsinteger
\(> 0\)
1
Number of inner loops with same samples
Additional information:
exp0This is the initial value for stabilizer (shift to diagonal of H). The actual value of stabilizer is \(10^{\textrm{exp0}}\).
Recommendations:
For hard cases (e.g., simultaneous optimization of long MSD and 3-Body J), set
exp0to 0 and do a single inner iteration (max its=1) per sample of configurations.
<!-- Specify the optimizer options -->
<parameter name="MinMethod">quartic</parameter>
<parameter name="exp0">-6</parameter>
<parameter name="alloweddifference"> 1.0e-4 </parameter>
<parameter name="nstabilizers"> 1 </parameter>
<parameter name="bigchange">15.0</parameter>
General Recommendations¶
All electron wavefunctions are typically more difficult to optimize than pseudopotential wavefunctions because of the importance of the wavefunction near the nucleus.
Two-body Jastrow contributes the largest portion of correlation energy from bare Slater determinants. Consequently, the recommended order for optimizing wavefunction components is two-body, one-body, three-body Jastrow factors and MSD coefficients.
For two-body spline Jastrows, always start from a reasonable one. The lack of physically motivated constraints in the functional form at large distances can cause slow convergence if starting from zero.
One-body spline Jastrow from old calculations can be a good starting point.
Three-body polynomial Jastrow can start from zero. It is beneficial to first optimize one-body and two-body Jastrow factors without adding three-body terms in the calculation and then add the three-body Jastrow and optimize all the three components together.
Optimization of CI coefficients¶
When storing a CI wavefunction in HDF5 format, the CI coefficients and the \(\alpha\) and \(\beta\) components of each CI are not in the XML input file. When optimizing the CI coefficients, they will be stored in HDF5 format. The optimization header block will have to specify that the new CI coefficients will be saved to HDF5 format. If the tag is not added coefficients will not be saved.
<qmc method="linear" move="pbyp" gpu="no" hdf5="yes">
The rest of the optimization block remains the same.
When running the optimization, the new coefficients will be stored in a *.sXXX.opt.h5 file, where XXX coressponds to the series number. The H5 file contains only the optimized coefficients. The corresponding *.sXXX.opt.xml will be updated for each optimization block as follows:
<detlist size="1487" type="DETS" nca="0" ncb="0" nea="2" neb="2" nstates="85" cutoff="1e-2" href="../LiH.orbs.h5" opt_coeffs="LiH.s001.opt.h5"/>
The opt_coeffs tag will then reference where the new CI coefficients are stored.
When restarting the run with the new optimized coeffs, you need to specify the previous hdf5 containing the basis set, orbitals, and MSD, as well as the new optimized coefficients. The code will read the previous data but will rewrite the coefficients that were optimized with the values found in the *.sXXX.opt.h5 file. Be careful to keep the pair of optimized CI coefficients and Jastrow coefficients together to avoid inconsistencies.
Diffusion Monte Carlo¶
Main input parameters are given in table9, additional in table10.
dmc method:
parameters:
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
integer |
\(> 0\) |
dep. |
Overall total number of walkers |
|
integer |
\(\geq 0\) |
1 |
Number of blocks |
|
integer |
\(\geq 0\) |
1 |
Number of steps per block |
|
integer |
\(\geq 0\) |
0 |
Number of steps for warming up |
|
real |
\(> 0\) |
0.1 |
Time step for each electron move |
|
string |
yes, no, v0, v1, v3 |
no |
Run with T-moves |
|
string |
classic/DRV/ZSGMA/YL |
classic |
Branch cutoff scheme |
|
real |
\(\geq 0\) |
3.6e5 |
Maximum allowed walltime in seconds |
|
integer |
\(\geq 0\) |
dep. |
Wavefunction recompute frequency |
|
text |
yes,no |
no |
Whether or not to sample the electron spins |
|
real |
\(> 0\) |
1.0 |
Effective mass for spin sampling |
Table 9 Main DMC input parameters.
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
integer |
\(\geq 0\) |
0 |
Trial energy update interval |
|
real |
all values |
dep. |
Reference energy in atomic units |
|
double |
\(\geq 0\) |
1.0 |
Population feedback on the trial energy |
|
10 |
\(\geq 0\) |
10 |
Parameter to cutoff large weights |
|
string |
yes/other |
no |
Kill or reject walkers that cross nodes |
|
option |
yes,no |
0 |
Warm up with a fixed population |
|
string |
yes/pure/other |
no |
Fixed population technique |
|
integer |
\(\geq 0\) |
1 |
Branching interval |
|
integer |
\(\geq 0\) |
1 |
Branching interval |
|
double |
\(\geq 0\) |
10 |
Kill persistent walkers |
|
double |
\(\geq 0\) |
2 |
Limit population growth |
|
real |
all values |
-1 |
Maximum particle move |
|
string |
yes/other |
yes |
Scale weights (CUDA only) |
|
integer |
\(\geq 0\) |
100 |
Number of steps between walker updates |
|
text |
yes/other |
yes |
Fast gradients |
|
integer |
all values |
0 |
Store configurations |
|
string |
yes/no |
yes |
Using nonblocking send/recv |
|
string |
yes/no |
no |
Disable branching for debugging without correctness guarantee |
Table 10 Additional DMC input parameters.
Additional information:
targetwalkers: A DMC run can be considered a restart run or a new run. A restart run is considered to be any method block beyond the first one, such as when a DMC method block follows a VMC block. Alternatively, a user reading in configurations from disk would also considered a restart run. In the case of a restart run, the DMC driver will use the configurations from the previous run, and this variable will not be used. For a new run, if the number of walkers is less than the number of threads, then the number of walkers will be set equal to the number of threads.blocks: This is the number of blocks run during a DMC method block. A block consists of a number of DMC steps (steps), after which all the statistics accumulated in the block are written to disk.steps: This is the number of DMC steps in a block.warmupsteps: These are the steps at the beginning of a DMC run in which the instantaneous average energy is used to update the trial energy. During regular steps, E\(_{ref}\) is used.timestep: Thetimestepdetermines the accuracy of the imaginary time propagator. Generally, multiple time steps are used to extrapolate to the infinite time step limit. A good range of time steps in which to perform time step extrapolation will typically have a minimum of 99% acceptance probability for each step.checkproperties: When using a particle-by-particle driver, this variable specifies how often to reset all the variables kept in the buffer.maxcpusecs: The default is 100 hours. Once the specified time has elapsed, the program will finalize the simulation even if all blocks are not completed.spinMovesDetermines whether or not the spin variables are sampled following [MZG+16] and [MBM16]. If a relativistic calculation is desired using pseudopotentials, spin variable sampling is required.spinMassIf spin sampling is on usingspinMoves== yes, the spin mass determines the rate of spin sampling, resulting in an effective spin timestep \(\tau_s = \frac{\tau}{\mu_s}\) where \(\tau\) is the normal spatial timestep and \(\mu_s\) is the value of the spin mass.energyUpdateInterval: The default is to update the trial energy at every step. Otherwise the trial energy is updated everyenergyUpdateIntervalstep.
where \(N\) is the current population.
refEnergy: The default reference energy is taken from the VMC run that precedes the DMC run. This value is updated to the current mean whenever branching happens.feedback: This variable is used to determine how strong to react to population fluctuations when doing population control. See the equation in energyUpdateInterval for more details.useBareTau: The same time step is used whether or not a move is rejected. The default is to use an effective time step when a move is rejected.warmupByReconfiguration: Warmup DMC is done with a fixed population.sigmaBound: This determines the branch cutoff to limit wild weights based on the sigma andsigmaBound.killnode: When running fixed-node, if a walker attempts to cross a node, the move will normally be rejected. Ifkillnode= “yes,” then walkers are destroyed when they cross a node.reconfiguration: Ifreconfigurationis “yes,” then run with a fixed walker population using the reconfiguration technique.branchInterval: This is the number of steps between branching. The total number of DMC steps in a block will beBranchInterval*Steps.substeps: This is the same asBranchInterval.nonlocalmoves: Evaluate pseudopotentials using one of the nonlocal move algorithms such as T-moves.no(default): Imposes the locality approximation.
yes/v0: Implements the algorithm in the 2006 Casula paper [Cas06].
v1: Implements the v1 algorithm in the 2010 Casula paper [CMSF10].
v2: Is not implemented and is skipped to avoid any confusion with the v2 algorithm in the 2010 Casula paper [CMSF10].
v3: (Experimental) Implements an algorithm similar to v1 but is much faster. v1 computes the transition probability before each single electron T-move selection because of the acceptance of previous T-moves. v3 mostly reuses the transition probability computed during the evaluation of nonlocal pseudopotentials for the local energy, namely before accepting any T-moves, and only recomputes the transition probability of the electrons within the same pseudopotential region of any electrons touched by T-moves. This is an approximation to v1 and results in a slightly different time step error, but it significantly reduces the computational cost. v1 and v3 agree at zero time step. This faster algorithm is the topic of a paper in preparation.
The v1 and v3 algorithms are size-consistent and are important advances over the previous v0 non-size-consistent algorithm. We highly recommend investigating the importance of size-consistency.
scaleweight: This is the scaling weight per Umrigar/Nightengale. CUDA only.MaxAge: Set the weight of a walker to min(currentweight,0.5) after a walker has not moved forMaxAgesteps. Needed if persistent walkers appear during the course of a run.MaxCopy: When determining the number of copies of a walker to branch, set the number of copies equal to min(Multiplicity,MaxCopy).fastgrad: This calculates gradients with either the fast version or the full-ratio version.maxDisplSq: When running a DMC calculation with particle by particle, this sets the maximum displacement allowed for a single particle move. All distance displacements larger than the max are rejected. If initialized to a negative value, it becomes equal to Lattice(LR/rc).sigmaBound: This determines the branch cutoff to limit wild weights based on the sigma andsigmaBound.storeconfigs: Ifstoreconfigsis set to a nonzero value, then electron configurations during the DMC run will be saved. This option is disabled for the OpenMP version of DMC.blocks_between_recompute: See details in Variational Monte Carlo.branching_cutoff_scheme:Modifies how the branching factor is computed so as to avoid divergences and stability problems near nodal surfaces.classic (default): The implementation found in QMCPACK v3.0.0 and earlier. \(E_{\rm cut}=\mathrm{min}(\mathrm{max}(\sigma^2 \times \mathrm{sigmaBound},\mathrm{maxSigma}),2.5/\tau)\), where \(\sigma^2\) is the variance and \(\mathrm{maxSigma}\) is set to 50 during warmup (equilibration) and 10 thereafter. \(\mathrm{sigmaBound}\) is default to 10.
DRV: Implements the algorithm of DePasquale et al., Eq. 3 in [DRV88] or Eq. 9 of [UNR93]. \(E_{\rm cut}=2.0/\sqrt{\tau}\).
ZSGMA: Implements the “ZSGMA” algorithm of [ZSG+16] with \(\alpha=0.2\). The cutoff energy is modified by a factor including the electron count, \(E_{\rm cut}=\alpha \sqrt{N/\tau}\), which greatly improves size consistency over Eq. 39 of [UNR93]. See Eq. 6 in [ZSG+16] and for an application to molecular crystals [ZBKlimevs+18].
YL: An unpublished algorithm due to Ye Luo. \(E_{\rm cut}=\sigma\times\mathrm{min}(\mathrm{sigmaBound},\sqrt{1/\tau})\). This option takes into account both size consistency and wavefunction quality via the term \(\sigma\). \(\mathrm{sigmaBound}\) is default to 10.
<qmc method="dmc" move="pbyp" target="e">
<parameter name="blocks">100</parameter>
<parameter name="steps">400</parameter>
<parameter name="timestep">0.010</parameter>
<parameter name="warmupsteps">100</parameter>
</qmc>
The time step should be individually adjusted for each problem. Please refer to the theory section on diffusion Monte Carlo.
<qmc method="dmc" move="pbyp" checkpoint="0">
<parameter name="timestep"> 0.004 </parameter>
<parameter name="blocks"> 100 </parameter>
<parameter name="steps"> 400 </parameter>
</qmc>
The checkpoint flag instructs QMCPACK to output walker configurations.
This also works in VMC. This will output an h5 file with the name
projectid.run-number.config.h5. Check that this file exists before
attempting a restart. To read in this file for a continuation run,
specify the following:
<mcwalkerset fileroot="BH.s002" version="0 6" collected="yes"/>
BH is the project id, and s002 is the calculation number to read in the walkers from the previous run.
Combining VMC and DMC in a single run (wavefunction optimization can be combined in this way too) is the standard way in which QMCPACK is typically run. There is no need to run two separate jobs since method sections can be stacked and walkers are transferred between them.
<qmc method="vmc" move="pbyp" target="e">
<parameter name="blocks">100</parameter>
<parameter name="steps">4000</parameter>
<parameter name="warmupsteps">100</parameter>
<parameter name="samples">1920</parameter>
<parameter name="walkers">1</parameter>
<parameter name="timestep">0.5</parameter>
</qmc>
<qmc method="dmc" move="pbyp" target="e">
<parameter name="blocks">100</parameter>
<parameter name="steps">400</parameter>
<parameter name="timestep">0.010</parameter>
<parameter name="warmupsteps">100</parameter>
</qmc>
<qmc method="dmc" move="pbyp" target="e">
<parameter name="warmupsteps">500</parameter>
<parameter name="blocks">50</parameter>
<parameter name="steps">100</parameter>
<parameter name="timestep">0.005</parameter>
</qmc>
Reptation Monte Carlo¶
Like DMC, RMC is a projector-based method that allows sampling of the fixed-node wavefunciton. However, by exploiting the path-integral formulation of Schrödinger’s equation, the RMC algorithm can offer some advantages over traditional DMC, such as sampling both the mixed and pure fixed-node distributions in polynomial time, as well as not having population fluctuations and biases. The current implementation does not work with T-moves.
There are two adjustable parameters that affect the quality of the RMC projection: imaginary projection time \(\beta\) of the sampling path (commonly called a “reptile”) and the Trotter time step \(\tau\). \(\beta\) must be chosen to be large enough such that \(e^{-\beta \hat{H}}|\Psi_T\rangle \approx |\Phi_0\rangle\) for mixed observables, and \(e^{-\frac{\beta}{2} \hat{H}}|\Psi_T\rangle \approx |\Phi_0\rangle\) for pure observables. The reptile is discretized into \(M=\beta/\tau\) beads at the cost of an \(\mathcal{O}(\tau)\) time-step error for observables arising from the Trotter-Suzuki breakup of the short-time propagator.
The following table lists some of the more practical
vmc method:
parameters:
Name
Datatype
Values
Default
Description
betareal
\(> 0\)
dep.
Reptile project time \(\beta\)
timestepreal
\(> 0\)
0.1
Trotter time step \(\tau\) for each electron move
beadsint
\(> 0\)
1
Number of reptile beads \(M=\beta/\tau\)
blocksinteger
\(> 0\)
1
Number of blocks
stepsinteger
\(\geq 0\)
1
Number of steps per block
vmcprestepsinteger
\(\geq 0\)
0
Propagates reptile using VMC for given number of steps
warmupstepsinteger
\(\geq 0\)
0
Number of steps for warming up
maxAgeinteger
\(\geq 0\)
0
Force accept for stuck reptile if age exceeds
maxAge
Additional information:
Because of the sampling differences between DMC ensembles of walkers and
RMC reptiles, the RMC block should contain the following estimator
declaration to ensure correct sampling:
<estimator name="RMC" hdf5="no">.
betaorbeads? One or the other can be specified, and from the Trotter time step, the code will construct an appropriately sized reptile. If both are given,betaoverridesbeads.Mixed vs. pure observables? Configurations sampled by the endpoints of the reptile are distributed according to the mixed distribution \(f(\mathbf{R})=\Psi_T(\mathbf{R})\Phi_0(\mathbf{R})\). Any observable that is computable within DMC and is dumped to the
scalar.datfile will likewise be found in thescalar.datfile generated by RMC, except there will be an appended_mto alert the user that the observable was computed on the mixed distribution. For pure observables, care must be taken in the interpretation. If the observable is diagonal in the position basis (in layman’s terms, if it is entirely computable from a single electron configuration \(\mathbf{R}\), like the potential energy), and if the observable does not have an explicit dependence on the trial wavefunction (e.g., the local energy has an explicit dependence on the trial wavefunction from the kinetic energy term), then pure estimates will be correctly computed. These observables will be found in either thescalar.dat, where they will be appended with a_psuffix, or in thestat.h5file. No mixed estimators will be dumped to the h5 file.Sampling: For pure estimators, the traces of both pure and mixed estimates should be checked. Ergodicity is a known problem in RMC. Because we use the bounce algorithm, it is possible for the reptile to bounce back and forth without changing the electron coordinates of the central beads. This might not easily show up with mixed estimators, since these are accumulated at constantly regrown ends, but pure estimates are accumulated on these central beads and so can exhibit strong autocorrelations in pure estimate traces.
Propagator: Our implementation of RMC uses Moroni’s DMC link action (symmetrized), with Umrigar’s scaled drift near nodes. In this regard, the propagator is identical to the one QMCPACK uses in DMC.
Sampling: We use Ceperley’s bounce algorithm.
MaxAgeis used in case the reptile gets stuck, at which point the code forces move acceptance, stops accumulating statistics, and requilibrates the reptile. Very rarely will this be required. For move proposals, we use particle-by-particle VMC a total of \(N_e\) times to generate a new all-electron configuration, at which point the action is computed and the move is either accepted or rejected.
- Cas06
Michele Casula. Beyond the locality approximation in the standard diffusion Monte Carlo method. Physical Review B - Condensed Matter and Materials Physics, 74:1–4, 2006. doi:10.1103/PhysRevB.74.161102.
- CMSF10(1,2)
Michele Casula, Saverio Moroni, Sandro Sorella, and Claudia Filippi. Size-consistent variational approaches to nonlocal pseudopotentials: Standard and lattice regularized diffusion Monte Carlo methods revisited. Journal of Chemical Physics, 2010. arXiv:1002.0356, doi:10.1063/1.3380831.
- DRV88
Michael F. DePasquale, Stuart M. Rothstein, and Jan Vrbik. Reliable diffusion quantum monte carlo. The Journal of Chemical Physics, 89(6):3629–3637, September 1988. doi:10.1063/1.454883.
- LON20
Leon Otis, Isabel Craig and Eric Neuscamman. A hybrid approach to excited-state-specific variational monte carlo and doubly excited states. arXiv preprint arXiv:2008:03586, 2020. URL: https://arxiv.org/abs/2008.03586.
- MBM16(1,2)
Cody A. Melton, M. Chandler Bennett, and Lubos Mitas. Quantum monte carlo with variable spins. The Journal of Chemical Physics, 144(24):244113, 2016. doi:10.1063/1.4954726.
- MZG+16(1,2)
Cody A. Melton, Minyi Zhu, Shi Guo, Alberto Ambrosetti, Francesco Pederiva, and Lubos Mitas. Spin-orbit interactions in electronic structure quantum monte carlo methods. Phys. Rev. A, 93:042502, Apr 2016. doi:10.1103/PhysRevA.93.042502.
- ON19(1,2,3)
Leon Otis and Eric Neuscamman. Complementary first and second derivative methods for ansatz optimization in variational monte carlo. Phys. Chem. Chem. Phys., 21:14491, 2019. doi:10.1039/C9CP02269D.
- UNR93(1,2)
C J Umrigar, M P Nightingale, and K J Runge. A diffusion Monte Carlo algorithm with very small timestep errors A diffusion Monte Carlo algorithm with very small time-step errors. The Journal of Chemical Physics, 99(4):2865, 1993. doi:10.1063/1.465195.
- ZBKlimevs+18
Andrea Zen, Jan Gerit Brandenburg, Jiř’ı Klimeš, Alexandre Tkatchenko, Dario Alfè, and Angelos Michaelides. Fast and accurate quantum monte carlo for molecular crystals. Proceedings of the National Academy of Sciences, 115(8):1724–1729, February 2018. doi:10.1073/pnas.1715434115.
- ZSG+16(1,2)
Andrea Zen, Sandro Sorella, Michael J. Gillan, Angelos Michaelides, and Dario Alfè. Boosting the accuracy and speed of quantum monte carlo: size consistency and time step. Physical Review B, June 2016. doi:10.1103/physrevb.93.241118.
- ZN16
Luning Zhao and Eric Neuscamman. An efficient variational principle for the direct optimization of excited states. J. Chem. Theory. Comput., 12:3436, 2016. doi:10.1021/acs.jctc.6b00508.
- ZN17
Luning Zhao and Eric Neuscamman. A blocked linear method for optimizing large parameter sets in variational monte carlo. J. Chem. Theory. Comput., 2017. doi:10.1021/acs.jctc.7b00119.
Output Overview¶
QMCPACK writes several output files that report information about the simulation (e.g., the physical properties such as the energy), as well as information about the computational aspects of the simulation, checkpoints, and restarts. The types of output files generated depend on the details of a calculation. The following list is not meant to be exhaustive but rather to highlight some salient features of the more common file types. Further details can be found in the description of the estimator of interest.
The .scalar.dat file¶
The most important output file is the scalar.dat file. This file contains the
output of block-averaged properties of the system such as the local
energy and other estimators. Each line corresponds to an average over
\(N_{walkers}*N_{steps}\) samples. By default, the quantities
reported in the scalar.dat file include the following:
- LocalEnergy
The local energy.
- LocalEnergy_sq
The local energy squared.
- LocalPotential
The local potential energy.
- Kinetic
The kinetic energy.
- ElecElec
The electron-electron potential energy.
- IonIon
The ion-ion potential energy.
- LocalECP
The energy due to the pseudopotential/effective core potential.
- NonLocalECP
The nonlocal energy due to the pseudopotential/effective core potential.
- MPC
The modified periodic Coulomb potential energy.
- BlockWeight
The number of MC samples in the block.
- BlockCPU
The number of seconds to compute the block.
- AcceptRatio
The acceptance ratio.
QMCPACK includes a python utility, qmca, that can be used to process these files. Details and examples are given in Analyzing QMCPACK data.
The .opt.xml file¶
This file is generated after a VMC wavefunction optimization and contains the part of the input file that lists the optimized Jastrow factors. Conveniently, this file is already formatted such that it can easily be incorporated into a DMC input file.
The .qmc.xml file¶
This file contains information about the computational aspects of the simulation, for example, which parts of the code are being executed when. This file is generated only during an ensemble run in which QMCPACK runs multiple input files.
The .dmc.dat file¶
This file contains information similar to the .scalar.dat file but also includes extra information about the details of a DMC calculation, for example, information about the walker population.
- Index
The block number.
- LocalEnergy
The local energy.
- Variance
The variance.
- Weight
The number of samples in the block.
- NumOfWalkers
The number of walkers times the number of steps.
- AvgSentWalkers
The average number of walkers sent. During a DMC simulation, walkers might be created or destroyed. At every step, QMCPACK will do some load balancing to ensure that the walkers are evenly distributed across nodes.
- TrialEnergy
The trial energy. See Diffusion Monte Carlo for an explanation of trial energy.
- DiffEff
The diffusion efficiency.
- LivingFraction
The fraction of the walker population from the previous step that survived to the current step.
The .bandinfo.dat file¶
This file contains information from the trial wavefunction about the band structure of the system, including the available \(k\)-points. This can be helpful in constructing trial wavefunctions.
Checkpoint and restart files¶
The .cont.xml file¶
This file enables continuation of the run. It is mostly a copy of the input XML file with the series number incremented and the mcwalkerset element added to read the walkers from a config file. The .cont.xml file is always created, but other files it depends on are present only if checkpointing is enabled.
The .config.h5 file¶
This file contains stored walker configurations.
The .random.h5 file¶
This file contains the state of the random number generator to allow restarts.
(Older versions used an XML file with a suffix of .random.xml).
Analyzing QMCPACK data¶
Using the qmca tool to obtain total energies and related quantities¶
The qmca tool is the primary means of analyzing scalar-valued data
generated by QMCPACK. Output files that contain scalar-valued data are
*.scalar.dat and *.dmc.dat (see Output Overview for a detailed description of these files).
Quantities that are available for analysis in *.scalar.dat files
include the local energy and its variance, kinetic energy, potential
energy and its components, acceptance ratio, and the average CPU time
spent per block, among others. The *.dmc.dat files provide
information regarding the DMC walker population in addition to the local
energy.
Basic capabilities of qmca include calculating mean values and
associated error bars, processing multiple files at once in batched
fashion, performing twist averaging, plotting mean values by series, and
plotting traces (per block or step) of the underlying data. These
capabilities are explained with accompanying examples in the following
subsections.
To use qmca, installations of Python and NumPy must be present on
the local machine. For graphical plotting, the matplotlib module must
also be available.
An overview of all supported input flags to qmca can be obtained by
typing qmca at the command line with no other inputs (also try
qmca -x for a short list of examples):
>qmca
no files provided, please see help info below
Usage: qmca [options] [file(s)]
Options:
--version show program's version number and exit
-v, --verbose Print detailed information (default=False).
-q QUANTITIES, --quantities=QUANTITIES
Quantity or list of quantities to analyze. See names
and abbreviations below (default=all).
-u UNITS, --units=UNITS
Desired energy units. Can be Ha (Hartree), Ry
(Rydberg), eV (electron volts), kJ_mol (k.
joule/mole), K (Kelvin), J (Joules) (default=Ha).
-e EQUILIBRATION, --equilibration=EQUILIBRATION
Equilibration length in blocks (default=auto).
-a, --average Average over files in each series (default=False).
-w WEIGHTS, --weights=WEIGHTS
List of weights for averaging (default=None).
-b, --reblock (pending) Use reblocking to calculate statistics
(default=False).
-p, --plot Plot quantities vs. series (default=False).
-t, --trace Plot a trace of quantities (default=False).
-h, --histogram (pending) Plot a histogram of quantities
(default=False).
-o, --overlay Overlay plots (default=False).
--legend=LEGEND Placement of legend. None for no legend, outside for
outside legend (default=upper right).
--noautocorr Do not calculate autocorrelation. Warning: error bars
are no longer valid! (default=False).
--noac Alias for --noautocorr (default=False).
--sac Show autocorrelation of sample data (default=False).
--sv Show variance of sample data (default=False).
-i, --image (pending) Save image files (default=False).
-r, --report (pending) Write a report (default=False).
-s, --show_options Print user provided options (default=False).
-x, --examples Print examples and exit (default=False).
--help Print help information and exit (default=False).
-d DESIRED_ERROR, --desired_error=DESIRED_ERROR
Show number of samples needed for desired error bar
(default=none).
-n PARTICLE_NUMBER, --enlarge_system=PARTICLE_NUMBER
Show number of samples needed to maintain error bar on
larger system: desired particle number first, current
particle number second (default=none)
Obtaining a statistically correct mean and error bar¶
A rough guess at the mean and error bar of the local energy can be
obtained in the following way with qmca:
>qmca -q e qmc.s000.scalar.dat
qmc series 0 LocalEnergy = -45.876150 +/- 0.017688
In this case the VMC energy of an 8-atom cell of diamond is estimated to be \(-45.876(2)\) Hartrees (Ha). This rough guess should not be used for production-level or publication-quality estimates.
To obtain production-level results, the underlying data should first be inspected visually to ensure that all data included in the averaging can be attributed to a distribution sharing the same mean. The first steps of essentially any MC calculation (the “equilibration phase”) do not belong to the equilibrium distribution and should be excluded from estimates of the mean and its error bar.
We can plot a data trace (-t) of the local energy in the
following way:
>qmca -t -q e -e 0 qmc.s000.scalar.dat
The -e 0 part indicates that we do not want any data
to be initially excluded from the calculation of averages. The resulting
plot is shown in Fig. 4. The unphysical
equilibration period is visible on the left side of the plot.
Trace of the VMC local energy for an 8-atom cell of diamond generated
with qmca. The x-axis (“samples”) refers to the VMC block index in
this case.¶
Most of the data fluctuates around a well-defined mean (consistent variations around a flat line). This property is important to verify by plotting the trace for each QMC run.
If we exclude none of the equilibration data points, we get an erroneous estimate of \(-45.870(2)\) Ha for the local energy:
>qmca -q e -e 0 qmc.s000.scalar.dat
qmc series 0 LocalEnergy = -45.870071 +/- 0.018072
The equilibration period is typically estimated by eye, though a few conservative values should be checked to ensure that the mean remains unaffected. In this dataset, the equilibration appears to have been reached after 100 or so samples. After excluding the first 100 VMC blocks from the analysis we get
>qmca -q e -e 100 qmc.s000.scalar.dat
qmc series 0 LocalEnergy = -45.877363 +/- 0.017432
This estimate (\(-45.877(2)\) Ha) differs significantly from the
\(-45.870(2)\) Ha figure obtained from the full set of data, but it
agrees with the rough estimate of \(-45.876(2)\) Ha obtained with
the abbreviated command (qmca -q e qmc.s000.scalar.dat). This is
because qmca makes a heuristic guess at the equilibration period and
got it reasonably correct in this case. In many cases, the heuristic
guess fails and should not be relied on for quality results.
We have so far obtained a statistically correct mean. To obtain a statistically correct error bar, it is best to include \(\sim\)100 or more statistically independent samples. An estimate of the number of independent samples can be obtained by considering the autocorrelation time, which is essentially a measure of the number of samples that must be traversed before an uncorrelated/independent sample is reached. We can get an estimate of the autocorrelation time in the following way:
>qmca -q e -e 100 qmc.s000.scalar.dat --sac
qmc series 0 LocalEnergy = -45.877363 +/- 0.017432 4.8
The flag –sac stands for (s)how (a)uto(c)orrelation. In this case,
the autocorrelation estimate is \(4.8\approx 5\) samples. Since the
total run contained 800 samples and we have excluded 100 of them, we can
estimate the number of independent samples as \((800-100)/5=140\).
In this case, the error bar is expected to be estimated reasonably well.
Trace of the local energy during one- and two-body Jastrow optimizations
for an 8-atom cell of diamond generated with qmca. Data for each
optimization cycle (QMCPACK series) is separated by a vertical black
line.¶
Keep in mind that the error bar represents the expected range of the mean with a certainty of only \(\sim 70\%\); i.e., it is a one sigma error bar. The actual mean value will lie outside the range indicated by the error bar in 1 out of every 3 runs, and in a set of 20 runs 1 value can be expected to deviate from its estimate by twice the error bar.
Judging wavefunction optimization¶
Wavefunction optimization is a highly nonlinear and sometimes sensitive process. As such, there is a risk that systematic errors encountered at this stage of the QMC process can be propagated into subsequent (expensive) DMC runs unless they are guarded against with vigilance.
In this section we again consider an 8-atom cell of diamond but now in
the context of Jastrow optimization (one- and two-body terms). In
optimization runs it is often preferable to use a large number of
warmupsteps (\(\sim 100\)) so that equilibration bias does not
propagate into the optimization process. We can check that the added
warm-up has had its intended effect by again checking the local energy
trace:
>qmca -t -q e *scalar*
The resulting plot can be found in Fig. 5. In
this case sufficient warmupsteps were used to exit the equilibration
period before samples were collected and we can proceed without using
the -e option with qmca.
After inspecting the trace, we should inspect the text output from
qmca, now including the total energy and its variance:
>qmca -q ev opt*scalar.dat
LocalEnergy Variance ratio
opt series 0 -44.823616 +/- 0.007430 7.054219 +/- 0.041998 0.1574
opt series 1 -45.877643 +/- 0.003329 1.095362 +/- 0.041154 0.0239
opt series 2 -45.883191 +/- 0.004149 1.077942 +/- 0.021555 0.0235
opt series 3 -45.877524 +/- 0.003094 1.074047 +/- 0.010491 0.0234
opt series 4 -45.886062 +/- 0.003750 1.061707 +/- 0.014459 0.0231
opt series 5 -45.877668 +/- 0.003475 1.091585 +/- 0.021637 0.0238
opt series 6 -45.877109 +/- 0.003586 1.069205 +/- 0.009387 0.0233
opt series 7 -45.882563 +/- 0.004324 1.058771 +/- 0.008651 0.0231
The flags -q ev requested the energy (e) and the variance
(v). For this combination of quantities, a third column (ratio)
is printed containing the ratio of the variance and the absolute value
of the local energy. The variance/energy ratio is an intensive quantity
and is useful to inspect regardless of the system under study.
Successful optimization of molecules and solids of any size generally
result in comparable values for the variance/energy ratio.
The first line of the output (series 0) corresponds to the local
energy and variance of the system without a Jastrow factor (all Jastrow
coefficients were initialized to zero in this case), reflecting the
quality of the orbitals alone. For pseudopotential systems, a
variance/energy ratio \(>0.20\) Ha generally indicates there is a
problem with the input orbitals that needs to be resolved before
performing wavefunction optimization.
The subsequent lines correspond to energies and variances of
intermediate parameterizations of the trial wavefunction during the
optimization process. The output line containing opt series 1, for
example, corresponds to the trial wavefunction parameterized during the
series 0 step (the parameters of this wavefunction would be found in
an output file matching *s000*opt.xml). The first thing to check
about the resulting optimization is again the variance/energy ratio. For
pseudopotential systems, a variance/energy ratio \(<0.03\) Ha is
consistent with a trial wavefunction of production quality, and values
of \(0.01\) Ha are rarely obtainable for standard Slater-Jastrow
wavefunctions. By this metric, all parameterizations obtained for
optimizations performed in series 0-6 are of comparable quality (note
that the quality of the wavefunction obtained during optimization series
7 is effectively unknown).
A good way to further discriminate among the parameterizations is to
plot the energy and variance as a function of series with qmca:
>qmca -p -q ev opt*scalar.dat
The -p option results in plots of means plus error bars
vs. series for all requested quantities.
The resulting plots for the local energy and variance are shown
in Fig. 6. In this case, the resulting energies
and variances are statistically indistinguishable for all optimization
cycles.
A good way to choose the optimal wavefunction for use in DMC is to select the one with the lowest statistically significant energy within the set of optimized wavefunctions with reasonable variance (e.g., among those with a variance/energy ratio \(<0.03\) Ha). For pseudopotential calculations, minimizing according to the total energy is recommended to reduce locality errors in DMC.
Energy and variance vs. optimization series for an 8-atom cell of
diamond as plotted by qmca.¶
Judging diffusion Monte Carlo runs¶
Judging the quality of the DMC projection process requires more care than is needed in VMC. To reduce bias, a small time step is required in the approximate projector but this also leads to slow equilibration and long autocorrelation times. Systematic errors in the projection process can also arise from statistical fluctuations due to pseudopotentials or from trial wavefunctions with larger-than-necessary variance.
To illustrate the problems that can arise with respect to slow equilibration and long autocorrelation times, we consider the 8-atom diamond system with VMC (\(200\) blocks of \(160\) steps) followed by DMC (\(400\) blocks of \(5\) steps) with a small time step (\(0.002\) Ha\(^{-1}\)). A good first step in assessing the quality of any DMC run is to plot the trace of the local energy:
>qmca -t -q e -e 0 *scalar*
Trace of the local energy for VMC followed by DMC with a small time step
(\(0.002\) Ha\(^{-1}\)) for an 8-atom cell of diamond
generated with qmca.¶
The resulting trace plot is shown in Fig. 7. As always, the DMC local energy decreases exponentially away from the VMC value, but in this case it takes a long time to do so. At least half of the DMC run is inefficiently consumed by equilibration. If we are not careful to inspect and remove the transient, the estimated DMC energy will be strongly biased by the transient as shown by the horizontal red line (estimated mean) in the figure. The autocorrelation time is also large (\(\sim 12\) blocks):
>qmca -q e -e 200 --sac *s001.scalar*
qmc series 1 LocalEnergy = -46.045720 +/- 0.004813 11.6
Of the included 200 blocks, fewer than 20 contribute to the estimated error bar, indicating that we cannot trust the reported error bar. This can also be demonstrated directly from the data. If we halve the number of included samples to 100, we expect from Gaussian statistics that the error bar will grow by a factor of \(\sqrt{2}\), but instead we get
>qmca -q e -e 300 *s001.scalar*
qmc series 1 LocalEnergy = -46.048537 +/- 0.009280
which erroneously shows an estimated increase in the error bar by a factor of about 2. Overall, this run is simply too short to gain meaningful information.
Consider the case in which we are interested in the cohesive energy of diamond, and, after having performed a time step study of the cohesive energy, we have found that the energy difference between bulk diamond and atomic carbon converges to our required accuracy with a larger time step of \(0.01\) Ha\(^{-1}\). In a production setting, a small cell could be used to determine the appropriate time step, while a larger cell would subsequently be used to obtain a converged cohesive energy, though for purposes of demonstration we still proceed here with the 8-atom cell. The new time step of \(0.01\) Ha\(^{-1}\) will result in a shorter autocorrelation time than the smaller time step used previously, but we would like to shorten the equilibration time further still. This can be achieved by using a larger time step (say \(0.02\) Ha\(^{-1}\)) in a short intermediate DMC run used to walk down the transient. The rapidly achieved equilibrium with the \(0.02\) Ha\(^{-1}\) time step projector will be much nearer to the \(0.01\) Ha\(^{-1}\) time step we seek than the original VMC equilibrium, so we can expect a shortened secondary equilibration time in the production \(0.01\) Ha\(^{-1}\) time step run. Note that this procedure is fully general, even if having to deal with an even shorter time step (e.g., \(0.002\) Ha\(^{-1}\)) for a particular problem.
We now rerun the previous example but with an intermediate DMC
calculation using \(40\) blocks of \(5\) steps with a time step
of \(0.02\) Ha\(^{-1}\), followed by a production DMC
calculation using \(400\) blocks of \(10\) steps with a time
step of \(0.01\) Ha\(^{-1}\). We again plot the local energy
trace using qmca:
>qmca -t -q e -e 0 *scalar*
with the result shown in Fig. 8. The projection transient has been effectively contained in the short DMC run with a larger time step. As expected, the production run contains only a short equilibration period. Removing the first 20 blocks as a precaution, we obtain an estimate of the total energy in VMC and DMC:
>qmca -q ev -e 20 --sac qmc.*.scalar.dat
LocalEnergy Variance ratio
qmc series 0 -45.881042 +/- 0.001283 1.0 1.076726 +/- 0.007013 1.0 0.0235
qmc series 1 -46.040814 +/- 0.005046 3.9 1.011303 +/- 0.016807 1.1 0.0220
qmc series 2 -46.032960 +/- 0.002077 5.2 1.014940 +/- 0.002547 1.0 0.0220
Trace of the local energy for VMC followed by a short intermediate DMC with a large time step (\(0.02\) Ha\(^{-1}\)) and finally a production DMC run with a time step of \(0.01\) Ha\(^{-1}\). Calculations were performed in an 8-atom cell of diamond.¶
Notice that the variance/energy ratio in DMC (\(0.220\) Ha) is
similar to but slightly smaller than that obtained with VMC
(\(0.235\) Ha). If the DMC variance/energy ratio is ever
significantly larger than with VMC, this is cause to be concerned about
the correctness of the DMC run. Also notice the estimated
autocorrelation time (\(\sim 5\) blocks). This leaves us with an
estimated \(\sim 76\) independent samples, though we should recall
that the autocorrelation time is also a statistical estimate that can be
improved with more data. We can gain a better estimate of the
autocorrelation time by using the *.dmc.dat files, which contain
output data resolved per step rather than per block (there are
\(10\times\) more steps than blocks in this example case):
>qmca -q ev -e 200 --sac qmc.s002.dmc.dat
LocalEnergy Variance ratio
qmc series 2 -46.032909 +/- 0.002068 31.2 1.015781 +/- 0.002536 1.4 0.0221
This results in an estimated autocorrelation time of \(\sim 31\) steps, or \(\sim 3\) blocks, indicating that we actually have \(\sim 122\) independent samples, which should be sufficient to obtain a trustworthy error bar. Our final DMC total energy is estimated to be \(-46.0329(2)\) Ha.
Another simulation property that should be explicitly monitored
is the behavior of the DMC walker population. Data regarding the
walker population is contained in the *.dmc.dat files.
In Fig. 9 we show the trace of the DMC
walker population for the current run:
>qmca -t -q nw *dmc.dat
qmc series 1 NumOfWalkers = 2056.905405 +/- 8.775527
qmc series 2 NumOfWalkers = 2050.164160 +/- 4.954850
Following a DMC run, the walker population should be checked for two qualities: (1) that the population is sufficiently large (a number \(>2,000\) is generally sufficient to reduce population control bias) and (2) that the population fluctuates benignly around its intended target value. In this case the target walker count (provided in the input file) was \(2,048\) and we can confirm from the plot that the population is simply fluctuating around this value. Also, from the text output we have a dynamic population estimate of 2,050(5) walkers. Rapid population reductions or increases—population explosions—are indicative of problems with a run. These issues sometimes result from using a considerably poor wavefunction (see comments regarding variance/energy ratio in the preceding subsections). QMCPACK has internal guards in place that prevent the population from exceeding certain maximum and minimum bounds, so in particularly faulty runs one might see the population “stabilize” to a constant value much larger or smaller than the target. In such cases the cause(s) for the divergent population behavior needs to be investigated and resolved before proceeding further.
Trace of the DMC walker population for an 8-atom cell of diamond
obtained with qmca.¶
Obtaining other quantities¶
A number of other scalar-valued quantities are available with qmca.
To obtain text output for all quantities available, simply exclude the
-q option used in previous examples. The following example shows
output for a DMC calculation of the 8-atom diamond system from the
scalar.dat file:
>qmca -e 20 qmc.s002.scalar.dat
qmc series 2
LocalEnergy = -46.0330 +/- 0.0021
Variance = 1.0149 +/- 0.0025
Kinetic = 33.851 +/- 0.019
LocalPotential = -79.884 +/- 0.020
ElecElec = -11.4483 +/- 0.0083
LocalECP = -22.615 +/- 0.029
NonLocalECP = 5.2815 +/- 0.0079
IonIon = -51.10 +/- 0.00
LocalEnergy_sq = 2120.05 +/- 0.19
BlockWeight = 20514.27 +/- 48.38
BlockCPU = 1.4890 +/- 0.0038
AcceptRatio = 0.9963954 +/- 0.0000055
Efficiency = 71.88 +/- 0.00
TotalTime = 565.80 +/- 0.00
TotalSamples = 7795421 +/- 0
Similarly, for the dmc.dat file we get
>qmca -e 20 qmc.s002.dmc.dat
qmc series 2
LocalEnergy = -46.0329 +/- 0.0020
Variance = 1.0162 +/- 0.0025
TotalSamples = 8201275 +/- 0
TrialEnergy = -46.0343 +/- 0.0023
DiffEff = 0.9939150 +/- 0.0000088
Weight = 2050.23 +/- 4.82
NumOfWalkers = 2050 +/- 5
LivingFraction = 0.996427 +/- 0.000021
AvgSentWalkers = 0.2625 +/- 0.0011
Any subset of desired quantities can be obtained by using the -q
option with either the full names of the quantities just listed
>qmca -q 'LocalEnergy Kinetic LocalPotential' -e 20 qmc.s002.scalar.dat
qmc series 2
LocalEnergy = -46.0330 +/- 0.0021
Kinetic = 33.851 +/- 0.019
LocalPotential = -79.884 +/- 0.020
or with their corresponding abbreviations.
>qmca -q ekp -e 20 qmc.s002.scalar.dat
qmc series 2
LocalEnergy = -46.0330 +/- 0.0021
Kinetic = 33.851 +/- 0.019
LocalPotential = -79.884 +/- 0.020
Abbreviations for each quantity can be found by typing qmca at the
command line with no other input. This following is a current list:
Abbreviations and full names for quantities:
ar = AcceptRatio
bc = BlockCPU
bw = BlockWeight
ce = CorrectedEnergy
de = DiffEff
e = LocalEnergy
ee = ElecElec
eff = Efficiency
ii = IonIon
k = Kinetic
kc = KEcorr
l = LocalECP
le2 = LocalEnergy_sq
mpc = MPC
n = NonLocalECP
nw = NumOfWalkers
p = LocalPotential
sw = AvgSentWalkers
te = TrialEnergy
ts = TotalSamples
tt = TotalTime
v = Variance
w = Weight
See the output overview for scalar.dat
(The .scalar.dat file) and dmc.dat
(The .dmc.dat file) for more information about
these quantities. The data analysis aspects for these
quantities are essentially the same as for the local
energy as covered in the preceding subsections.
Quantities that do not belong to an equilibrium distribution
(e.g., BlockCPU) are somewhat different, though they
still exhibit statistical fluctuations.
Processing multiple files¶
Batch file processing is a common use case for qmca. If we consider
an “equation-of-state” calculation involving the 8-atom diamond cell we
have used so far, we might be interested in the total energy for the
various supercell volumes along the trajectory from compression to
expansion. After checking the traces
(qmca -t -q e scale_*/vmc/*scalar*) to settle on a sensible
equilibration cutoff as discussed in the preceding subsections, we can
obtain the total energies all at once:
>qmca -q ev -e 40 scale_*/vmc/*scalar*
LocalEnergy Variance ratio
scale_0.80/vmc/qmc series 0 -44.670984 +/- 0.006051 2.542384 +/- 0.019902 0.0569
scale_0.82/vmc/qmc series 0 -44.982818 +/- 0.005757 2.413011 +/- 0.022626 0.0536
scale_0.84/vmc/qmc series 0 -45.228257 +/- 0.005374 2.258577 +/- 0.019322 0.0499
scale_0.86/vmc/qmc series 0 -45.415842 +/- 0.005532 2.204980 +/- 0.052978 0.0486
scale_0.88/vmc/qmc series 0 -45.570215 +/- 0.004651 2.061374 +/- 0.014359 0.0452
scale_0.90/vmc/qmc series 0 -45.683684 +/- 0.005009 1.988539 +/- 0.018267 0.0435
scale_0.92/vmc/qmc series 0 -45.751359 +/- 0.004928 1.913282 +/- 0.013998 0.0418
scale_0.94/vmc/qmc series 0 -45.791622 +/- 0.005026 1.843704 +/- 0.014460 0.0403
scale_0.96/vmc/qmc series 0 -45.809256 +/- 0.005053 1.829103 +/- 0.014536 0.0399
scale_0.98/vmc/qmc series 0 -45.806235 +/- 0.004963 1.775391 +/- 0.015199 0.0388
scale_1.00/vmc/qmc series 0 -45.783481 +/- 0.005293 1.726869 +/- 0.012001 0.0377
scale_1.02/vmc/qmc series 0 -45.741655 +/- 0.005627 1.681776 +/- 0.011496 0.0368
scale_1.04/vmc/qmc series 0 -45.685101 +/- 0.005353 1.682608 +/- 0.015423 0.0368
scale_1.06/vmc/qmc series 0 -45.615164 +/- 0.005978 1.652155 +/- 0.010945 0.0362
scale_1.08/vmc/qmc series 0 -45.543037 +/- 0.005191 1.646375 +/- 0.013446 0.0361
scale_1.10/vmc/qmc series 0 -45.450976 +/- 0.004794 1.707649 +/- 0.048186 0.0376
scale_1.12/vmc/qmc series 0 -45.371851 +/- 0.005103 1.686997 +/- 0.035920 0.0372
scale_1.14/vmc/qmc series 0 -45.265490 +/- 0.005311 1.631614 +/- 0.012381 0.0360
scale_1.16/vmc/qmc series 0 -45.161961 +/- 0.004868 1.656586 +/- 0.014788 0.0367
scale_1.18/vmc/qmc series 0 -45.062579 +/- 0.005971 1.671998 +/- 0.019942 0.0371
scale_1.20/vmc/qmc series 0 -44.960477 +/- 0.004888 1.651864 +/- 0.009756 0.0367
In this case, we are using a Jastrow factor optimized only at the
equilibrium geometry (scale_1.00) but with radial cutoffs restricted
to the Wigner-Seitz radius of the most compressed supercell
(scale_0.80) to avoid introducing wavefunction cusps at the cell
boundary (had we tried, QMCPACK would have aborted with a warning in
this case). It is clear that this restricted Jastrow factor is not an
optimal choice because it yields variance/energy ratios between
\(0.036\) and \(0.057\) Ha. This issue is largely a result of
our undersized (8-atom) supercell; larger cells should always be used in
real production calculations.
Batch processing is also possible for multiple quantities. If multiple quantities are requested, an additional line is inserted to separate results from different runs:
>qmca -q 'e bc eff' -e 40 scale_*/vmc/*scalar*
scale_0.80/vmc/qmc series 0
LocalEnergy = -44.6710 +/- 0.0061
BlockCPU = 0.02986 +/- 0.00038
Efficiency = 38104.00 +/- 0.00
scale_0.82/vmc/qmc series 0
LocalEnergy = -44.9828 +/- 0.0058
BlockCPU = 0.02826 +/- 0.00013
Efficiency = 44483.91 +/- 0.00
scale_0.84/vmc/qmc series 0
LocalEnergy = -45.2283 +/- 0.0054
BlockCPU = 0.02747 +/- 0.00030
Efficiency = 52525.12 +/- 0.00
scale_0.86/vmc/qmc series 0
LocalEnergy = -45.4158 +/- 0.0055
BlockCPU = 0.02679 +/- 0.00013
Efficiency = 50811.55 +/- 0.00
scale_0.88/vmc/qmc series 0
LocalEnergy = -45.5702 +/- 0.0047
BlockCPU = 0.02598 +/- 0.00015
Efficiency = 74148.79 +/- 0.00
scale_0.90/vmc/qmc series 0
LocalEnergy = -45.6837 +/- 0.0050
BlockCPU = 0.02527 +/- 0.00011
Efficiency = 65714.98 +/- 0.00
...
Twist averaging¶
Twist averaging can be performed straightforwardly for any output
quantity listed in Obtaining other quantities with qmca.
We illustrate these capabilities by repeating the 8-atom diamond DMC
runs performed in Section Judging diffusion Monte Carlo runs at 8 real-valued
supercell twist angles (a \(2\times 2\times 2\) Monkhorst-Pack grid
centered at the \(\Gamma\)-point). Data traces for each twist can be
overlapped on the same plot:
>qmca -to -q e -e '30 20 30' *scalar* --legend outside
The -o option requests the plots to be overlapped; otherwise,
8 separate plots would be generated. The
equilibration input -e '30 20 30' cuts out from
the analyzed data the first 30 blocks for series 0 (VMC),
20 blocks for series 1 (intermediate DMC), and 30 blocks for
series 2 (production DMC). The resulting plot is shown in
Fig. 10.
Overlapped energy traces from VMC to DMC for an 8-supercell diamond
obtained with qmca. Data for each twist appears in a different
color.¶
Twist averaging is performed by providing the -a
option. If provided on its own, uniform weights are applied
to each twist angle. To obtain a trace plot with twist averaging
enforced, use a command similar to the following:
>qmca -a -t -q e -e '30 20 30' *scalar*
The resulting plot is shown in Fig. 11. As
can be seen from the trace plot, the chosen equilibration lengths are
appropriate, and we proceed to obtain the twist-averaged total energy
from the scalar.dat files
>qmca -a -q ev -e 30 --sac *s002.scalar*
LocalEnergy Variance ratio
avg series 2 -45.873369 +/- 0.000753 5.3 1.028751 +/- 0.001056 1.3 0.0224
and also from the dmc.dat files
>qmca -a -q ev -e 300 --sac *s002.dmc*
LocalEnergy Variance ratio
avg series 2 -45.873371 +/- 0.000741 30.5 1.028843 +/- 0.000972 1.6 0.0224
yielding a twist-averaged total energy of \(-45.8733(8)\) Ha.
Twist-averaged energy trace from VMC to DMC for an 8-supercell diamond obtained with qmca.¶
As can be seen from Fig. 10, some of the twist angles are degenerate. This is seen more clearly in the text output
>qmca -q ev -e 30 *s002.scalar*
LocalEnergy Variance ratio
qmc.g000 series 2 -45.264510 +/- 0.001942 1.057065 +/- 0.002318 0.0234
qmc.g001 series 2 -46.035511 +/- 0.001806 1.015992 +/- 0.002836 0.0221
qmc.g002 series 2 -46.035410 +/- 0.001538 1.015039 +/- 0.002661 0.0220
qmc.g003 series 2 -46.047285 +/- 0.001898 1.018219 +/- 0.002588 0.0221
qmc.g004 series 2 -46.034225 +/- 0.002539 1.013420 +/- 0.002835 0.0220
qmc.g005 series 2 -46.046731 +/- 0.002963 1.018337 +/- 0.004109 0.0221
qmc.g006 series 2 -46.047133 +/- 0.001958 1.021483 +/- 0.003082 0.0222
qmc.g007 series 2 -45.476146 +/- 0.002065 1.070456 +/- 0.003133 0.0235
The degenerate twists grouped by set are \(\{0\}\), \(\{1,2,4\}\), \(\{3,5,6\}\), and \(\{7\}\).
Alternatively, the run could have been performed at the four unique (irreducible) twist angles only. We will emulate this situation by analyzing data for twists 0, 1, 3, and 7 only. In a production setting with irreducibly weighted twists, the run would be performed on these twists alone; we reuse the uniform twist data for illustration purposes only.
We can use qmca to perform twist averaging with different
weights applied to each twist:
>qmca -a -w '1 3 3 1' -q ev -e 30 *g000*2*sc* *g001*2*sc* *g003*2*sc* *g007*2*sc*
LocalEnergy Variance ratio
avg series 2 -45.873631 +/- 0.001044 1.028769 +/- 0.001520 0.0224
yielding a total energy value of \(-45.874(1)\) Ha, in agreement with the uniform weighted twist average performed previously.
The decision of whether or not to perform irreducible weighted twist averaging should be made on the basis of efficiency. The relative efficiency of irreducible vs. uniform weighted twist averaging depends on the irreducible weights and the ratio of the lengths of the available sampling and equilibration periods. A formula for the relative efficiency of these two cases is derived and discussed in more detail in Appendix A: Derivation of twist averaging efficiency.
Setting output units¶
Estimates outputted by qmca are in Hartree units by default. The
output units for energetic quantities can be changed by using the -u
option.
Energy in Hartrees:
>qmca -q e -u Ha -e 20 qmc.s002.scalar.dat
qmc series 2 LocalEnergy = -46.032960 +/- 0.002077
Energy in electron volts:
>qmca -q e -u eV -e 20 qmc.s002.scalar.dat
qmc series 2 LocalEnergy = -1252.620565 +/- 0.056521
Energy in Rydbergs:
>qmca -q e -u rydberg -e 20 qmc.s002.scalar.dat
qmc series 2 LocalEnergy = -92.065919 +/- 0.004154
Energy in kilojoules per mole:
>qmca -q e -u kj_mol -e 20 qmc.s002.scalar.dat
qmc series 2 LocalEnergy = -120859.512998 +/- 5.453431
Speeding up trace plotting¶
When working with many files or files with many entries, qmca might
take a long time to produce plots. The time delay is actually due to the
autocorrelation time estimate used to calculate error bars. The
calculation time for the autocorrelation scales as
\(\mathcal{O}(M^2)\), with \(M\) being the number of statistical
samples. If you are interested only in plotting traces and not in the
estimated error bars, the autocorrelation time estimation can be turned
off with the –noac option:
>qmca -t -q e -e 20 --noac qmc.s002.scalar.dat
Note that the resulting error bars printed to the console will be
underestimated and are not meaningful. Do not use –noac in
conjunction with the -p plotting option as these plots are of no use
without meaningful error bars.
Short usage examples¶
Plotting a trace of the local energy:
>qmca -t -q e *scalar*
Applying an equilibration cutoff to VMC data (series 0):
>qmca -q e -e 30 *s000.scalar*
Applying the same equilibration cutoff to VMC and DMC data (series 0, 1, 2):
>qmca -q e -e 20 *scalar*
Applying different equilibration cutoffs to VMC and DMC data (series 0, 1, 2):
>qmca -q e -e '30 20 40' *scalar*
Obtaining the energy, variance, and variance/energy ratio for all series:
>qmca -q ev -e 30 *scalar*
Overlaying plots of mean + error bar for energy and variance for separate two- and three-body Jastrow optimization runs:
>qmca -po -q ev ./optJ2/*scalar* ./optJ3/*scalar*
Obtaining the acceptance ratio:
>qmca -q ar -e 30 *scalar*
Obtaining the average DMC walker population:
>qmca -q nw -e 400 *s002.dmc.dat
Obtaining the MC efficiency:
>qmca -q eff -e 30 *scalar*
Obtaining the total wall clock time per series:
>qmca -q tt -e 0 *scalar*
Obtaining the average wall clock time spent per block:
>qmca -q bc -e 0 *scalar*
Obtaining a subset of desired quantities:
>qmca -q 'e v ar eff' -e 30 *scalar*
Obtaining all available quantities:
>qmca -e 30 *scalar*
Obtaining the twist-averaged total energy with uniform weights:
>qmca -a -q e -e 40 *g*s002.scalar.dat
Obtaining the twist-averaged total energy with specific weights:
>qmca -a -w '1 3 3 1' -q e -e 40 *g*s002.scalar.dat
Obtaining the local, kinetic, and potential energies in eV:
>qmca -q ekp -e 30 -u eV *scalar*
Production quality checklist¶
Inspect the trace plots (
-toption) for any oddities in the data. Typical behavior is a short equilibration period followed by benign fluctuations around a clear mean value. There should not be any large spikes in the data. This applies to all runs (VMC, optimization, DMC, etc.).Remove all equilibration steps (
-eoption) from the data by inspecting the trace plot.Check the quality of the orbitals (standalone Jastrow-less VMC or sometimes the first
scalarfile produced during optimization) by inspecting the variance/energy ratioqmca -q ev *scalar*. For pseudopotential systems without a Jastrow, the variance/energy ratio should not exceed \(0.2\) Ha; otherwise, there is a problem with the orbitals.Check the quality of the optimized Jastrow factor by inspecting the variance/energy ratio. For pseudopotential systems with a Jastrow, the variance/energy ratio should not exceed \(0.04\) Ha for pseudopotential systems. A good Jastrow is indicated by a variance/energy ratio in the range of \(0.01-0.03\) Ha. A value less than \(0.01\) Ha is difficult to achieve.
Confirm that the optimization has converged by plotting the energy and variance vs. optimization series (
qmca -p -q ev *scalar*). Do not assume that optimization has converged in only a few cycles. Use at least 10 cycles with about 100,000 samples unless you already have experience with the system in question.Optimize Jastrow factors according to energy minimization to reduce locality errors arising from the use of nonlocal pseudopotentials in DMC. A good approach is to optimize with a few cycles of variance minimization followed by several cycles of energy minimization.
Occasionally try optimizing with more samples and/or cycles to see if improved results are obtained.
If using a B-spline representation of the orbitals, converge the VMC energy and variance with respect to the mesh size (controlled via meshfactor). This is best done in the presence of any Jastrow factor to reduce noise. Consider using the hybrid LMTO representation of the orbitals as this can reduce both the VMC/DMC variance and the DMC time step error, in addition to saving memory.
Check the variance/energy ratio of all production VMC and DMC calculations. In all cases, the DMC ratio should be slightly less than the VMC ratio and both should abide the preceding guidelines, i.e., the ratio should be less than \(0.04\) Ha for pseudopotential systems. The production ratio should also be consistent with what is observed during wavefunction optimization.
Be aware of population control bias in DMC. Run with a population of \(\sim 2,000\) or greater. Occasionally repeat a run using a larger population to explicitly confirm that population control bias is small.
Check the stability of the DMC walker population by plotting the trace of the population size (
qmca -t -q nw *dmc.dat). Verify that the average walker population is consistent with the requested value provided in the input.In DMC, perform a time step study to obtain either (1) extrapolated results or (2) a time step for future production where an energy difference shows convergence (e.g., a band gap or defect formation energy). For pseudopotential systems, converged time steps for many systems are in the range of \(0.002-0.01\) Ha\(^{-1}\), but the actual converged time step must be explicitly checked.
In periodic systems, converge the total energy with respect to the size of the twist/k-point grid. Results for smaller systems can easily be transferred to larger ones (e.g., a \(2 \times 2 \times 2\) twist grid in a \(2 \times 2 \times 2\) tiled cell is equivalent to a \(1 \times 1 \times 1\) twist grid in a \(4 \times 4 \times 4\) tiled cell).
In periodic systems, perform finite-size extrapolation including two body corrections (needed for cohesive energy/phase stability studies) unless it can be shown that finite-size effects cancel for the energy difference in question (e.g., some defect formation energies).
Using the qmc-fit tool for statistical time step extrapolation and curve fitting¶
The qmc-fit tool is used to provide statistical estimates of
curve-fitting parameters based on QMCPACK data. Although qmc-fit
will eventually support many types of fitted curves (e.g., Morse
potential binding curves and various equation-of-state fitting curves),
it is currently limited to estimating fitting parameters related to time
step extrapolation.
The jackknife statistical technique¶
The qmc-fit tool obtains estimates of fitting parameter means and
associated error bars via the “jack-knife” technique. This technique is
a powerful and general tool to obtain meaningful error bars for any
quantity that is related in a nonlinear fashion to an underlying set of
statistical data. For this reason, we give a brief overview of the
jackknife technique before proceeding with usage instructions for the
qmc-fit tool.
Consider \(N\) statistical variables \(\{x_n\}_{n=1}^N\) that have been outputted by one or more simulation runs. If we have \(M\) samples of each of the \(N\) variables, then the mean values of each these variables can be estimated in the standard way, that is, \(\bar{x}_n\approx \tfrac{1}{M}\sum_{m=1}^Mx_{nm}\).
Suppose we are interested in \(P\) statistical quantities \(\{y_p\}_{p=1}^P\) that are related to the original \(N\) variables by a known multidimensional function \(F\):
The relationship implied by \(F\) is completely general. For example, the \(\{x_n\}\) might be elements of a matrix with \(\{y_p\}\) being the eigenvalues, or \(F\) might be a fitting procedure for \(N\) energies at different time steps with \(P\) fitting parameters. An approximate guess at the mean value of \(\vec{y}\) can be obtained by evaluating \(F\) at the mean value of \(\vec{x}\) (i.e. \(F(\bar{x}_1\ldots\bar{x}_N)\)), but with this approach we have no way to estimate the statistical error bar of any \(\bar{y}_p\).
In the jackknife procedure, the statistical variability intrinsic to the underlying data \(\{x_n\}\) is used to obtain estimates of the mean and error bar of \(\{y_p\}\). We first construct a new set of \(x\) statistical data by taking the average over all samples but one:
The result is a distribution of approximate \(x\) mean values. These are used to construct a distribution of approximate means for \(y\):
Estimates for the mean and error bar of the quantities of interest can finally be obtained using the following formulas:
Performing time step extrapolation¶
In this section, we use a 32-atom supercell of MnO as an example system
for time step extrapolation. Data for this system has been collected in
DMC using the following sequence of time steps:
\(0.04,~0.02,~0.01,~0.005,~0.0025,~0.00125\) Ha\(^{-1}\). For
a typical production pseudopotential study, time steps in the range of
\(0.02-0.002\) Ha\(^{-1}\) are usually sufficient and it is
recommended to increase the number of steps/blocks by a factor of two
when the time step is halved. To perform accurate statistical fitting,
we must first understand the equilibration and autocorrelation
properties of the inputted local energy data. After plotting the local
energy traces (qmca -t -q e -e 0 ./qmc*/*scalar*), it is clear that
an equilibration period of \(30\) blocks is reasonable. Approximate
autocorrelation lengths are also obtained with qmca:
>qmca -e 30 -q e --sac ./qmc*/qmc.g000.s002.scalar.dat
./qmc_tm_0.00125/qmc.g000 series 2 LocalEnergy = -3848.234513 +/- 0.055754 1.7
./qmc_tm_0.00250/qmc.g000 series 2 LocalEnergy = -3848.237614 +/- 0.055432 2.2
./qmc_tm_0.00500/qmc.g000 series 2 LocalEnergy = -3848.349741 +/- 0.069729 2.8
./qmc_tm_0.01000/qmc.g000 series 2 LocalEnergy = -3848.274596 +/- 0.126407 3.9
./qmc_tm_0.02000/qmc.g000 series 2 LocalEnergy = -3848.539017 +/- 0.075740 2.4
./qmc_tm_0.04000/qmc.g000 series 2 LocalEnergy = -3848.976424 +/- 0.075305 1.8
The autocorrelation must be removed from the data before jackknifing, so we will reblock the data by a factor of 4.
The qmc-fit tool can be used in the following way to obtain a linear
time step fit of the data:
>qmc-fit ts -e 30 -b 4 -s 2 -t '0.00125 0.0025 0.005 0.01 0.02 0.04' ./qmc*/*scalar*
fit function : linear
fitted formula: (-3848.193 +/- 0.037) + (-18.95 +/- 1.95)*t
intercept : -3848.193 +/- 0.037 Ha
The input arguments are as follows: ts indicates we are performing a
time step fit, -e 30 is the equilibration period removed from each
set of scalar data, -b 4 indicates the data will be reblocked by a
factor of 4 (e.g., a file containing 400 entries will be block averaged
into a new set of 100 before jackknife fitting), -s 2 indicates that
the time step data begins with series 2 (scalar files matching
*s000* or *s001* are to be excluded), and -t ‘0.00125 0.0025
0.005 0.01 0.02 0.04’ provides a list of time step values corresponding
to the inputted scalar files. The -e and -b options can receive
a list of file-specific values (same format as -t) if desired. As
can be seen from the text output, the parameters for the linear fit are
printed with error bars obtained with jackknife resampling and the zero
time step “intercept” is \(-3848.19(4)\) Ha. In addition to text
output, the previous command will result in a plot of the fit with the
zero time step value shown as a red dot, as shown in the top panel of
Fig. 12.
Different fitting functions are supported via the -f option.
Currently supported options include linear (\(a+bt\)),
quadratic (\(a+bt+ct^2\)), and sqrt
(\(a+b\sqrt{t}+ct\)). Results for a quadratic fit are shown
subsequently and in the bottom panel of Fig. 12.
>qmc-fit ts -f quadratic -e30 -b4 -s2 -t '0.00125 0.0025 0.005 0.01 0.02 0.04' ./qmc*/*scalar*
fit function : quadratic
fitted formula: (-3848.245 +/- 0.047) + (-7.25 +/- 8.33)*t + (-285.00 +/- 202.39)*t^2
intercept : -3848.245 +/- 0.047 Ha
In this case, we find a zero time step estimate of \(-3848.25(5)\) Ha\(^{-1}\). A time step of \(0.04\) Ha\(^{-1}\) might be on the large side to include in time step extrapolation, and it is likely to have an outsize influence in the case of linear extrapolation. Upon excluding this point, linear extrapolation yields a zero timestep value of \(-3848.22(4)\) Ha\(^{-1}\). Note that quadratic extrapolation can result in intrinsically larger uncertainty in the extrapolated value. For example, when the \(0.04\) Ha\(^{-1}\) point is excluded, the uncertainty grows by 50% and we obtain an estimated value of \(-3848.28(7)\) instead.
Linear (top) and quadratic (bottom) time step fits to DMC data for a 32-atom supercell of MnO obtained with qmc-fit. Zero time step estimates are indicated by the red data point on the left side of either panel.¶
Using the qdens tool to obtain electron densities¶
The qdens tool is provided to post-process the heavy density data
produced by QMCPACK and output the mean density (with and without
errorbars) in file formats viewable with, e.g., XCrysDen or VESTA. The
tool currently works only with the SpinDensity estimator in QMCPACK.
Note: this tool is provisional and may be changed or replaced at any
time. The planned successor to this tool (qstat) will expand access
to other observables and will retain at least the non-plotting
capabilities of qdens.
To use qdens, Nexus must be installed along with NumPy and H5Py. A
short list of example use cases are covered in the next section. Current
input flags are:
>qdens
Usage: qdens [options] [file(s)]
Options:
--version show program's version number and exit
-h, --help Print help information and exit (default=False).
-v, --verbose Print detailed information (default=False).
-f FORMATS, --formats=FORMATS
Format or list of formats for density file output.
Options: dat, xsf, chgcar (default=None).
-e EQUILIBRATION, --equilibration=EQUILIBRATION
Equilibration length in blocks (default=0).
-r REBLOCK, --reblock=REBLOCK
Block coarsening factor; use estimated autocorrelation
length (default=None).
-a, --average Average over files in each series (default=False).
-w WEIGHTS, --weights=WEIGHTS
List of weights for averaging (default=None).
-i INPUT, --input=INPUT
QMCPACK input file containing structure and grid
information (default=None).
-s STRUCTURE, --structure=STRUCTURE
File containing atomic structure (default=None).
-g GRID, --grid=GRID Density grid dimensions (default=None).
-c CELL, --cell=CELL Simulation cell axes (default=None).
--lineplot=LINEPLOT Produce a line plot along the selected dimension: 0,
1, or 2 (default=None).
--noplot Do not show plots interactively (default=False).
Usage examples¶
Process a single file, excluding the first 40 blocks, and produce XSF files:
qdens -v -e 40 -f xsf -i qmc.in.xml qmc.s000.stat.h5
Process files for all available series:
qdens -v -e 40 -f xsf -i qmc.in.xml *stat.h5
Combine groups of 10 adjacent statistical blocks together (appropriate if the estimated autocorrelation time is about 10 blocks):
qdens -v -e 40 -r 10 -f xsf -i qmc.in.xml qmc.s000.stat.h5
Apply different equilibration lengths and reblocking factors to each
series (below is appropriate if there are three series, e.g. s000,
s001, and s002):
qdens -v -e '20 20 40' -r '4 4 8' -f xsf -i qmc.in.xml *stat.h5
Produce twist averaged densities (also works with multiple series and reblocking):
qdens -v -a -e 40 -f xsf -i qmc.g000.twistnum_0.in.xml qmc.g*.s000.stat.h5
Twist averaging with arbitrary weights can be performed via the -w
option in a fashion identical to qmca.
Files produced¶
Look for files with names and extensions similar to:
qmc.s000.SpinDensity_u.xsf
qmc.s000.SpinDensity_u-err.xsf
qmc.s000.SpinDensity_u+err.xsf
qmc.s000.SpinDensity_d.xsf
qmc.s000.SpinDensity_d-err.xsf
qmc.s000.SpinDensity_d+err.xsf
qmc.s000.SpinDensity_u+d.xsf
qmc.s000.SpinDensity_u+d-err.xsf
qmc.s000.SpinDensity_u+d+err.xsf
qmc.s000.SpinDensity_u-d.xsf
qmc.s000.SpinDensity_u-d-err.xsf
qmc.s000.SpinDensity_u-d+err.xsf
Files postfixed with u relate to the up electron density, d to
down, u+d to the total charge density, and u-d to the difference
between up and down electron densities.
Files without err in the name contain only the mean, whereas files
with +err/-err in the name contain the mean plus/minus the
estimated error bar. Please use caution in interpreting the error bars
as their accuracy depends crucially on a correct estimation of the
autocorrelation time by the user (see -r option) and having a
sufficient number of blocks remaining following any reblocking.
When twist averaging, the group tag (e.g. g000 or similar) will be
replaced with avg in the names of the outputted files.
Periodic LCAO for Solids¶
Introduction¶
QMCPACK implements the linear combination of atomic orbitals (LCAO) and Gaussian basis sets in periodic boundary conditions. This method uses orders of magnitude less memory than the real-space spline wavefunction. Although the spline scheme enables very fast evaluation of the wavefunction, it might require too much on-node memory for a large complex cell. The periodic Gaussian evaluation provides a fallback that will definitely fit in available memory but at significantly increased computational expense. Well-designed Gaussian basis sets should be used to accurately represent the wavefunction, typically including both diffuse and high angular momentum functions.
The current implementation is not highly optimized for efficiency but can handle real and complex trial wavefunctions generated by PySCF [SBB+18], but other codes such as Crystal can be interfaced on request. Supercell tiling is handled outside QMCPACK through a proper PySCF input generated by Nexus and the Supercell geometry and coefficients of the molecular orbotals are constructed in the converter provided by QMCPACK. This is different from the plane wave/spline route where the tiling is provided in QMCPACK.
LCAO schemes use physical considerations to construct a highly efficient basis set compared with plane waves. Typically only a few tens of basis functions per atom are required compared with thousands of plane waves. Many forms of LCAO schemes exist and are being implemented in QMCPACK. The details of the already-implemented methods are described in the following section.
GTOs: The Gaussian basis functions follow a radial-angular decomposition of
where \(Y_{{lm}}(\theta ,\phi )\) is a spherical harmonic, \(l\) and \(m\) are the angular momentum and its \(z\) component, and \(r, \theta, \phi\) are spherical coordinates. In practice, they are atom centered and the \(l\) expansion typically includes 1–3 additional channels compared with the formally occupied states of the atom (e.g., 4–6 for a nickel atom with occupied \(s\), \(p\), and \(d\) electron shells.
The evaluation of GTOs within PBC differs slightly from evaluating GTOs
in open boundary conditions (OBCs). The orbitals are evaluated at a
distance \(r\) in the primitive cell (similar to OBC), and then the
contributions of the periodic images are added by evaluating the orbital
at a distance \(r+T\), where T is a translation of the cell lattice
vector. This requires loops over the periodic images until the
contributions are orbitals \(\Phi\). In the current implementation,
the number of periodic images is an input parameter named PBCimages,
which takes three integers corresponding to the number of periodic
images along the supercell axes (X, Y and Z axes for a cubic cell). By
default these parameters are set to PBCimages= 8 8 8, but they
require manual convergence checks. Convergence checks can be
performed by checking the total energy convergence with respect to
PBCimages, similar to checks performed for plane wave cutoff energy
and B-spline grids. Use of diffuse Gaussians might require these
parameters to be increased, while sharply localized Gaussians might
permit a decrease. The cost of evaluating the wavefunction increases
sharply as PBCimages is increased. This input parameter will be
replaced by a tolerance factor and numerical screening in the future.
Generating and using periodic Gaussian-type wavefunctions using PySCF¶
Similar to any QMC calculation, using periodic GTOs requires the generation of a periodic trial wavefunction. QMCPACK is currently interfaced to PySCF, which is a multipurpose electronic structure written mainly in Python with key numerical functionality implemented via optimized C and C++ libraries [SBB+18]. Such a wavefunction can be generated according to the following example for a \(2 \times 1 \times 1\) supercell using tiling (kpoints) and a supertwist shifted away from \(\Gamma\), leading to a complex wavefunction.
#! /usr/bin/env python3
import numpy
import h5py
from pyscf.pbc import gto, scf, dft, df
from pyscf.pbc import df
cell = gto.Cell()
cell.a = '''
3.37316115 3.37316115 0.00000000
0.00000000 3.37316115 3.37316115
3.37316115 0.00000000 3.37316115'''
cell.atom = '''
C 0.00000000 0.00000000 0.00000000
C 1.686580575 1.686580575 1.686580575
'''
cell.basis = 'bfd-vdz'
cell.ecp = 'bfd'
cell.unit = 'B'
cell.drop_exponent = 0.1
cell.verbose = 5
cell.charge = 0
cell.spin = 0
cell.build()
sp_twist=[0.07761248, 0.07761248, -0.07761248]
kmesh=[2,1,1]
kpts=[[ 0.07761248, 0.07761248, -0.07761248],[ 0.54328733, 0.54328733, -0.54328733]]
mf = scf.KRHF(cell,kpts)
mf.exxdiv = 'ewald'
mf.max_cycle = 200
e_scf=mf.kernel()
ener = open('e_scf','w')
ener.write('%s\n' % (e_scf))
print('e_scf',e_scf)
ener.close()
title="C_diamond-tiled-cplx"
from PyscfToQmcpack import savetoqmcpack
savetoqmcpack(cell,mf,title=title,kmesh=kmesh,kpts=kpts,sp_twist=sp_twist)
Note that the last three lines of the file
title="C_diamond-tiled-cplx"
from PyscfToQmcpack import savetoqmcpack
savetoqmcpack(cell,mf,title=title,kmesh=kmesh,kpts=kpts,sp_twist=sp_twist)
contain the title (name of the HDF5 to be used in QMCPACK) and the call to the converter. The title variable will be the name of the HDF5 file where all the data needed by QMCPACK will be stored. The function savetoqmcpack will be called at the end of the calculation and will generate the HDF5 similarly to the nonperiodic PySCF calculation in convert4qmc. The function is distributed with QMCPACK and is located in the qmcpack/src/QMCTools directory under the name PyscfToQmcpack.py. Note that you need to specify the supertwist coordinates that was used with the provided kpoints. The supertwist must match the coordinates of the K-points otherwise the phase factor for the atomic orbital will be incorrect and incorrect results will be obtained. (For more details on how to generate tiling with PySCF and Nexus, refer to the Nexus guide or the 2019 QMCPACK Workshop material available on github: https://github.com/QMCPACK/qmcpack_workshop_2019 under qmcpack_workshop_2019/day2_nexus/pyscf/04_pyscf_diamond_hf_qmc/
For the converter in the script to be called properly, you need to specify the path to the file in your PYTHONPATH such as
export PYTHONPATH=QMCPACK_PATH/src/QMCTools:$PYTHONPATH
To generate QMCPACK input files, you will need to run convert4qmc exactly as specified in convert4qmc for both cases:
convert4qmc -pyscf C_diamond-tiled-cplx
This tool can be used with any option described in convert4qmc. Since the HDF5 contains all the information needed, there is no need to specify any other specific tag for periodicity. A supercell at \(\Gamma\)-point or using multiple k-points will work without further modification.
Running convert4qmc will generate 3 input files:
<?xml version="1.0"?>
<qmcsystem>
<simulationcell>
<parameter name="lattice">
6.74632230000000e+00 6.74632230000000e+00 0.00000000000000e+00
0.00000000000000e+00 3.37316115000000e+00 3.37316115000000e+00
3.37316115000000e+00 0.00000000000000e+00 3.37316115000000e+00
</parameter>
<parameter name="bconds">p p p</parameter>
<parameter name="LR_dim_cutoff">15</parameter>
</simulationcell>
<particleset name="ion0" size="4">
<group name="C">
<parameter name="charge">4</parameter>
<parameter name="valence">4</parameter>
<parameter name="atomicnumber">6</parameter>
</group>
<attrib name="position" datatype="posArray">
0.0000000000e+00 0.0000000000e+00 0.0000000000e+00
1.6865805750e+00 1.6865805750e+00 1.6865805750e+00
3.3731611500e+00 3.3731611500e+00 0.0000000000e+00
5.0597417250e+00 5.0597417250e+00 1.6865805750e+00
</attrib>
<attrib name="ionid" datatype="stringArray">
C C C C
</attrib>
</particleset>
<particleset name="e" random="yes" randomsrc="ion0">
<group name="u" size="8">
<parameter name="charge">-1</parameter>
</group>
<group name="d" size="8">
<parameter name="charge">-1</parameter>
</group>
</particleset>
</qmcsystem>
As one can see, for both examples, the two-atom primitive cell has been expanded to contain four atoms in a \(2 \times 1 \times 1\) carbon cell.
<?xml version="1.0"?>
<qmcsystem>
<wavefunction name="psi0" target="e">
<determinantset type="MolecularOrbital" name="LCAOBSet" source="ion0" transform="yes" twist="0.07761248 0.07761248 -0.07761248" href="C_diamond-tiled-cplx.h5" PBCimages="8 8 8">
<slaterdeterminant>
<determinant id="updet" size="8">
<occupation mode="ground"/>
<coefficient size="52" spindataset="0"/>
</determinant>
<determinant id="downdet" size="8">
<occupation mode="ground"/>
<coefficient size="52" spindataset="0"/>
</determinant>
</slaterdeterminant>
</determinantset>
<jastrow name="J2" type="Two-Body" function="Bspline" print="yes">
<correlation size="10" speciesA="u" speciesB="u">
<coefficients id="uu" type="Array"> 0 0 0 0 0 0 0 0 0 0</coefficients>
</correlation>
<correlation size="10" speciesA="u" speciesB="d">
<coefficients id="ud" type="Array"> 0 0 0 0 0 0 0 0 0 0</coefficients>
</correlation>
</jastrow>
<jastrow name="J1" type="One-Body" function="Bspline" source="ion0" print="yes">
<correlation size="10" cusp="0" elementType="C">
<coefficients id="eC" type="Array"> 0 0 0 0 0 0 0 0 0 0</coefficients>
</correlation>
</jastrow>
</wavefunction>
</qmcsystem>
This file contains information related to the trial wavefunction. It is identical to the input file from an OBC calculation to the exception of the following tags:
*.wfj.xml specific tags:
tag
tag type
default
description
twist3 doubles
(0 0 0)
Coordinate of the twist to compute
hrefstring
default
Name of the HDF5 file generated by PySCF and used for convert4qmc
PBCimages3 Integer
8 8 8
Number of periodic images to evaluate the orbitals
Other files containing QMC methods (such as optimization, VMC, and DMC blocks) will be generated and will behave in a similar fashion regardless of the type of SPO in the trial wavefunction.
- SBB+18(1,2)
Qiming Sun, Timothy C. Berkelbach, Nick S. Blunt, George H. Booth, Sheng Guo, Zhendong Li, Junzi Liu, James D. McClain, Elvira R. Sayfutyarova, Sandeep Sharma, Sebastian Wouters, and Garnet Kin-Lic Chan. Pyscf: the python-based simulations of chemistry framework. Wiley Interdisciplinary Reviews: Computational Molecular Science, 8(1):n/a–n/a, 2018. doi:10.1002/wcms.1340.
Selected Configuration Interaction¶
A direct path towards improving the accuracy of a QMC calculation is through a better trial wavefunction. Although using a multireference wavefunction can be straightforward in theory, in actual practice methods such as CASSCF are not always intuitive and often require being an expert in either the method or the code generating the wavefunction. An alternative is to use a selected configuration of interaction method (selected CI) such as CIPSI (configuration interaction using a perturbative selection done iteratively). This provides a direct route to systematically improving the wavefunction.
Theoretical background¶
The principle behind selected CI is rather simple and was first published in 1955 by R. K. Nesbet [Nes55]. The first calculations on atoms were performed by Diner, Malrieu, and Claverie [DMC67] in 1967 and became computationally viable for larger molecules in 2013 by Caffarel et al. [EG13].
As described by Caffarel et al. in [EG13], multideterminantal expansions of the ground-state wavefunction \(\Psi_T\) are written as a linear combination of Slater determinants
where each determinant corresponds to a given occupation by the \(N_{\alpha}\) and \(N_{\beta}\) electrons of \(N=N_{\alpha}+N_{\beta}\) orbitals among a set of M spin-orbitals \(\{\phi_1,.,\phi_M\}\) (restricted case). When no symmetries are considered, the maximum number of such determinants is
a number that grows factorially with M and N. The best representation of the exact wavefunction in the determinantal basis is the full configuration interaction (FCI) wavefunction written as
where \(c_i\) are the ground-state coefficients obtained by diagonalizing the matrix, \(H_{ij}=\langle{D_i}|H|{D_j}\rangle\), within the full orthonormalized set \(\langle{D_i}||{D_j}\rangle=\delta_{ij}\) of determinants \(|{D_i}\rangle\). CIPSI provides a convenient method to build up to this full wavefunction with a single criteria.
A CIPSI wavefunction is built iteratively starting from a reference wavefunction, usually Hartree-Fock or CASSCF, by adding all single and double excitations and then iteratively selecting relevant determinants according to some criteria. Detailed iterative steps can be found in the reference by Caffarel et al. and references within [EG13], [SAGC16], [SGCL0] and [GSLC17] and are summarized as follows:
Step 1: Define a reference wavefunction:
Step 2: Generate external determinants \(|{\alpha}\rangle\): New determinants are added by generating all single and double excitations from determinants \(i \in D\) such as:
Step 3: Evaluate the second-order perturbative contribution to each determinant \(|{\alpha}\rangle\):
Step 4: Select the determinants with the largest contributions and add them to the Hamiltonian.
Step 5: Diagonalize the Hamiltonian within the new added determinants and update the wavefunction and the the value of \(E_{var}\).
Step 6: Iterate until reaching convergence.
Repeating this process leads to a multireference trial wavefunction of high quality that can be used in QMC.
The linear coefficients \(c_k\) are then optimized with the presence of the Jastrow function.
Note the following:
When all determinants \(|{\alpha}\rangle\) are selected, the full configuration interaction result is obtained.
CIPSI can be seen as a deterministic counterpart of FCIQMC.
In practice, any wavefunction method can be made multireference with CIPSI. For instance, a multireference coupled cluster (MRCC) with CIPSI is implemented in QP. [GGMS17]
At any time, with CIPSI selection, \(E_{PT_2}=\sum_\alpha \Delta E_\alpha\) estimates the distance to the FCI solution.
CIPSI wavefunction interface¶
The CIPSI method is implemented in the QP code:cite:QP developed by the Caffarel group. Once the trial wavefunction is generated, QP is able to produce output readable by the QMCPACK converter as described in convert4qmc. QP can be installed with multiple plugins for different levels of theory in quantum chemistry. When installing the “QMC” plugin, QP can save the wavefunction in a format readable by the QMCPACK converter.
In the following we use the \(C_2O_2H_3N\) molecule (Fig. 13) as an example of how to run a multireference calculation with CIPSI as a trial wavefunction for . The choice of this molecule is motivated by its multireference nature. Although the molecule remains small enough for CCSD(T) calculations with aug-cc-pVTZ basis set, the D1 diagnostic shows a very high value for \(C_2O_2H_3N\), suggesting a multireference character. Therefore, an accurate reference for the system is not available, and it becomes difficult to trust the quality of a single-determinant wavefunction even when using the DFT-B3LYP exchange and correlation functional. Therefore, in the following, we show an example of how to systematically improve the nodal surface by increasing the number of determinants in the trial wavefunction.
\(C_2O_2H_3N\) molecule.¶
The following steps show how to run from Hartree-Fock to selected CI using QP2, convert the wavefunction to a QMCPACK trial wavefunction, and analyze the result.
Step 1: Generate the QP input file. QP takes for input an XYZ file containing the geometry of the molecule such as:
8
C2O2H3N
C
1.067070
-0.370798
0.020324
C
-1.115770
-0.239135
0.081860
O
-0.537581
1.047619
-0.091020
N
0.879629
0.882518
0.046830
H
-1.525096
-0.354103
1.092299
H
-1.868807
-0.416543
-0.683862
H
2.035229
-0.841662
0.053363
O
-0.025736
-1.160835
-0.084319
The input file is generated through the following command line:
qp_create_ezfio C2O2H3N.xyz -b cc-pvtz
This means that we will be simulating the molecule in all electrons within the cc-pVTZ basis set. Other options are, of course, possible such as using ECPs, different spin multiplicities, etc. For more details, see the QP tutorial at https://quantumpackage.github.io/qp2/
A directory called
C2O2H3N.ezfiois created and contains all the relevant data to run the SCF Hartree-Fock calculation. Note that because of the large size of molecular orbitals (MOs) (220), it is preferable to run QP in parallel. QP parallelization is based on a master/slave process that allows a master node to manage the work load between multiple MPI processes through the LibZMQ library. In practice, the run is submitted to one master node and is then submitted to as many nodes as necessary to speed up the calculations. If a slave node dies before the end of its task, the master node will resubmit the workload to another available node. If more nodes are added at any time during the simulation, the master node will use them to reduce the time to solution.Step 2: Run Hartree-Fock. To save the integrals on disk and avoid recomputing them later, edit the
ezfiodirectory with the following command:qp_edit C2O2H3N.ezfio
This will generate a temporary file showing all the contents of the simulation and opens an editor to allow modification of their values. Look for
io_ao_one_e_integralsand modify its value fromNonetoWrite.To run a simulation with QP, use the binary texttt{qp_run} with the desired level of theory, in this case Hartree-Fock (scf).
mpirun -np 1 qp_run scf C2O2H3N.ezfio &> C2O2H3N-SCF.out
If run in serial, the evaluation of the integrals and the Hamiltonian diagonalization would take a substantial amount of computer time. We recommend adding a few more slave nodes to help speed up the calculation.
mpirun -np 20 qp_run -s scf C2O2H3N.ezfio &> C2O2H3N-SCF-Slave.out
The total Hartree-Fock energy of the system in cc-pVTZ is :math:`E_{HF}=-283.0992`Ha.
Step 3: Freeze core electrons. To avoid making excitation from the core electrons, freeze the core electrons and do only the excitations from the valence electrons.
qp_set_frozen_core C2O2H3N.ezfio
This will will automatically freeze the orbitals from 1 to 5, leaving the remaining orbitals active.
Step 4: Transform atomic orbitals (AOs) to MOs. This step is the most costly, especially given that its implementation in QP is serial. We recommend completing it in a separate run and on one node.
qp_run four_idx_transform C2O2H3N.ezfio
The MO integrals are now saved on disk, and unless the orbitals are changed, they will not be recomputed.
Step 5: CIPSI At this point the wavefunction is ready for the selected CI. By default, QP has two convergence criteria: the number of determinants (set by default to 1M) or the value of PT2 (set by default to \(1.10^{-4}\)Ha). For this molecule, the total number of determinants in the FCI space is \(2.07e+88\) determinants. Although this number is completely out of range of what is possible to compute, we will set the limit of determinants in QP to 5M determinants and see whether the nodal surface of the wavefunction is converged enough for the DMC. At this point it is important to remember that the main value of CIPSI compared with other selected CI methods, is that the value of PT2 is evaluated directly at each step, giving a good estimate of the error to the FCI energy. This allows us to conclude that when the E+PT2 energy is converged, the nodal surface is also probably converged. Similar to the SCF runs, FCI runs have to be submitted in parallel with a master/slave process:
mpirun -np 1 qp_run fci C2O2H3N.ezfio &> C2O2H3N-FCI.out & sleep 300 mpirun -np 199 qp_run -s fci C2O2H3N.ezfio &> C2O2H3N-FCI-Slave.out wait
Step 6 (optional): Natural orbitals Although this step is optional, it is important to note that using natural orbitals instead of Hartree-Fock orbitals will always improve the quality of the wavefunction and the nodal surface by reducing the number of needed determinants for the same accuracy. When a full convergence to the FCI limit is attainable, this step will not lead to any change in the energy but will only reduce the total number of determinants. However, if a full convergence is not possible, this step can significantly increase the accuracy of the calculation at the same number of determinants.
qp_run save_natorb C2O2H3N.ezfio
At this point, the orbitals are modified, a new AO \(\rightarrow\)MO transformation is required, and steps 3 and 4 need to be run again.
Step 7: Analyze the CIPSI results. Fig. 14 shows the evolution of the variational energy and the energy corrected with PT2 as a function of the number of determinants up to 4M determinants. Although it is clear that the raw variational energy is far from being converged, the Energy + PT2 appears converged around 0.4M determinants.
Evolution of the variational energy and the Energy + PT2 as a function of the number of determinants for the \(C_2O_2H_3N\) molecule.¶
Step 8: Truncate the number of determinants. Although using all the 4M determinants from CIPSI always guarantees that all important determinants are kept in the wavefunction, practically, such a large number of determinants would make any QMC calculation prohibitively expensive because the cost of evaluating a determinant in DMC grows as \(\sqrt[]{N_{det}}\), where \(N_{det}\) is the number of determinants in the trial wavefunction. To truncate the number of determinants, we follow the method described by Scemama et al. [SGCL0] where the wavefunction is truncated by independently removing spin-up and spin-down determinants whose contribution to the norm of the wavefunction is below a user-defined threshold, \(\epsilon\). For this step, we choose to truncate the determinants whose coefficients are below, \(1.10^{-3}\), \(1.10^{-4}\), \(1.10^{-5}\), and \(1.10^{-6}\), translating to 239, 44539, 541380, and 908128 determinants, respectively.
To truncate the determinants in QP, edit the
ezfiofile as follows:qp_edit C2O2H3N.ezfio
Then look for
ci\_thresholdand modify the value according to the desired threshold. Use the following run to truncate the determinants:qp_run truncate_wf_spin C2O2H3N.ezfio
Method |
N_det |
Energy |
|---|---|---|
Hartree-Fock |
1 |
-281.6729 |
Natural orbitals |
1 |
-281.6735 |
E_Variational |
438,753 |
-282.2951 |
E_Variational |
4,068,271 |
-282.4882 |
E+PT2 |
438,753 |
-282.6809 |
E+PT2 |
4,068,271 |
-282.6805 |
table11 Energies of \(C_2O_2H_3N\) using orbitals from
Hartree-Fock, natural orbitals, and 0.4M and 4M determinants
Save the wavefunction for QMCPACK. The wavefunction in QP is now ready to be converted to QMCPACK format. Save the wavefunction into QMCPACK format and then convert the wavefunction using the
convert4qmctool.qp_run save_for_qmcpack C2O2H3N.ezfio convert4qmc -orbitals QP2QMCPACK.h5 -multidets QP2QMCPACK.h5 -addCusp -production
Note that QP2 produces an HDF5 file in the QMCPACK format, named QP2QMCPACK. Such file can be used fir single determinants or multideterminants calculations. Since we are running all-electron calculations, orbitals in QMC need to be corrected for the electron-nuclearcusp condition. This is done by adding the option
-addCusptoconvert4qmc, which adds a tag forcing QMCPACK to run the correction or read them from a file if pre-computed. When running multiple DMC runs with different truncation thresholds, only the number of determinants is varied and the orbitals remain unchanged from one calculation to another and the cusp correction needs to be run only once.Step 10: Run QMCPACK. At this point, running a multideterminant DMC becomes identical to running a regular DMC with QMCPACK; After correcting the orbitals for the cusp, optimize the Jastrow functions and then run the DMC. It is important, however, to note a few items:
QMCPACK allows reoptimization of the coefficients of the determinants during the Jastrow optimization step. Although this has proven to lower the energy significantly when the number of determinants is below 10k, a large number of determinants from CIPSI is often too large to optimize conveniently. Keeping the coefficients of the determinants from CIPSI unoptimized is an alternative strategy.
The large determinant expansion and the Jastrows are both trying to recover the missing correlations from the system. When optimizing the Jastrows, we recommend first optimizing J1 and J2 without the J3, and then with the added J3. Trying to initially optimize J1, J2, and J3 at the same time could lead to numerical instabilities.
The parameters of the Jastrow function will need to be optimized for each truncation scheme and usually cannot be reused efficiently from one truncation scheme to another.
Step 11: Analyze the DMC results from QMCPACK. From
table12, we can see that increasing the number of determinants from 0.5M to almost 1M keeps the energy within error bars and does not improve the quality of the nodal surface. We can conclude that the DMC energy is converged at 0.54M determinants. Note that this number of determinants also corresponds to the convergence of E+PT2 in CIPSI calculations, confirming for this case that the convergence of the nodal surface can follow the convergence of E+PT2 instead of the more difficult variational energy.
N_det |
DMC |
CISPI |
|---|---|---|
1 |
-283.0696 (6) |
-283.0063 |
239 |
-283.0730 (9) |
-282.9063 |
44,539 |
-283.078 (1) |
-282.7339 |
541,380 |
-283.088 (1) |
-282.6772 |
908,128 |
-283.089 (1) |
-282.6775 |
Table 12 DMC Energies and CIPSI(E+PT2) of \(C_2O_2H_3N\) in function of the number of determinants in the trial wavefunction.
As mentioned in previous sections, DMC is variational relative to the exact nodal surface. A nodal surface is “better” if it lowers DMC energy. To assess the quality of the nodal surface from CIPSI, we compare these DMC results to other single-determinant calculations from multiple nodal surfaces and theories. Fig. 15 shows the energy of the \(C_2O_2H_3N\) molecule as a function of different single-determinant trial wavefunctions with an aug-cc-pVTZ basis set, including Hartree-Fock, DFT-PBE, and hybrid functionals B3LYP and PBE0. The last four points in the plot show the systematic improvement of the nodal surface as a function of the number of determinants.
DMC energy of the \(C_2O_2H_3N\) molecule as a function of different single-determinant trial wavefunctions with aug-ccp-VTZ basis set using nodal surfaces from Hartree-Fock, DFT-PBE, and DFT with hybrid functionals PBE0 and P3LYP. As indicated, the CIPSI trial wavefunction contains 239, 44539, 514380, and 908128 determinants (D).¶
When the DMC-CIPSI energy is converged with respect to the number of determinants, its nodal surface is still lower than the best SD-DMC (B3LYP) by 6(1) mHa. When compared with CCSD(T) with the same basis set, \(E_{CCSD(T)}\) is 4 mHa higher than DMC-CIPSI and 2 mHa lower than DMC-B3LYP. Although 6 (1) mHa can seem very small, it is important to remember that CCSD(T) cannot correctly describe multireference systems; therefore, it is impossible to assess the correctness of the single-determinant–DMC result, making CIPSI-DMC calculations an ideal benchmark tool for multireference systems.
- DMC67
S. Diner, J. P. Malrieu, and P. Claverie. The use of perturbation methods for the study of the effects of configuration interaction. Theoretica chimica acta, 8(5):390–403, 1967. doi:10.1007/BF00529454.
- EG13(1,2,3)
M. Caffarel E. Giner, A. Scemama. Using perturbatively selected configuration interaction in quantum monte carlo calculations. Canadian Journal of Chemistry, 91:9, 2013.
- GGMS17
Yann Garniron, Emmanuel Giner, Jean-Paul Malrieu, and Anthony Scemama. Alternative definition of excitation amplitudes in multi-reference state-specific coupled cluster. The Journal of Chemical Physics, 146(15):154107, 2017. arXiv:https://doi.org/10.1063/1.4980034, doi:10.1063/1.4980034.
- GSLC17
Yann Garniron, Anthony Scemama, Pierre-François Loos, and Michel Caffarel. Hybrid stochastic-deterministic calculation of the second-order perturbative contribution of multireference perturbation theory. The Journal of Chemical Physics, 147(3):034101, 2017. arXiv:https://doi.org/10.1063/1.4992127, doi:10.1063/1.4992127.
- Nes55
R. K. Nesbet. Configuration interaction in orbital theories. Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, 230(1182):312–321, 1955. URL: http://rspa.royalsocietypublishing.org/content/230/1182/312, arXiv:http://rspa.royalsocietypublishing.org/content/230/1182/312.full.pdf, doi:10.1098/rspa.1955.0134.
- SAGC16
Anthony Scemama, Thomas Applencourt, Emmanuel Giner, and Michel Caffarel. Quantum monte carlo with very large multideterminant wavefunctions. Journal of Computational Chemistry, 37(20):1866–1875, 2016. doi:10.1002/jcc.24382.
- SGCL0(1,2)
Anthony Scemama, Yann Garniron, Michel Caffarel, and Pierre-Francois Loos. Deterministic construction of nodal surfaces within quantum monte carlo: the case of fes. Journal of Chemical Theory and Computation, 0(ja):null, 0. arXiv:http://dx.doi.org/10.1021/acs.jctc.7b01250, doi:10.1021/acs.jctc.7b01250.
- Sce17
A. Scemamma. Quantum package. https://github.com/LCPQ/quantum_package, 2013–2017.
Spin-Orbit Calculations in QMC¶
Introduction¶
In order to introduce relativistic effects in real materials, in principle the full Dirac equation must be solved where the resulting wave function is a four-component spinor. For the valence electrons that participate in chemistry, the single particle spinors can be well approximated by two-component spinors as two of the components are negligible. Note that this is not true for the deeper core electrons, where all four components contribute. In light of this fact, relativistic pseudopotentials have been developed to remove the core electrons while providing an effective potential for the valence electrons [DC12]. This allows relativistic effects to be studied in QMC methods using two-component spinor wave functions.
In QMCPACK, spin-orbit interactions have been implemented following the methodology described in [MZG+16] and [MBM16]. We briefly describe some of the details below.
Single-Particle Spinors¶
The single particle spinors used in QMCPACK take the form
where \(s\) is the spin variable and using the complex spin representation.
In order to carry out spin-orbit calculations in solids, the single-particle spinors
can be obtained using Quantum Espresso. After carrying out the spin-orbit calculation in QE
(with flags noncolin = .true., lspinorb = .true., and a relativistic .UPF pseudopotential),
the spinors can be obtained by using the converter convertpw4qmc:
convertpw4qmc data-file-schema.xml
where the data-file-schema.xml file is output from your QE calculation.
This will produce an eshdf.h5 file which contains the up and down components of the spinors per k-point.
Trial Wavefunction¶
Using the generated single particle spinors, we build the many-body wavefunction in a similar fashion to the normal non-relativistic calculations, namely
where we now utilize determinants of spinors, as opposed to the usual product of up and down determinants. An example xml input block for the trial wave function is show below:
<?xml version="1.0"?>
<qmcsystem>
<wavefunction name="psi0" target="e">
<sposet_builder name="spo_builder" type="spinorbspline" href="eshdf.h5" tilematrix="100010001" twistnum="0" source="ion0" size="10">
<sposet type="bspline" name="myspo" size="10">
<occupation mode="ground"/>
</sposet>
</sposet_builder>
<determinantset>
<slaterdeterminant>
<determinant id="det" group="u" sposet="myspo" size="10"\>
</slaterdeterminant>
</determinantset>
<jastrow type="One-Body" name="J1" function="bspline" source="ion0" print="yes">
<correlation elementType="O" size="8" cusp="0.0">
<coefficients id="eO" type="Array">
</coefficients>
</correlation>
</jastrow>
<jastrow type="Two-Body" name="J2" function="bspline" print="yes">
<correlation speciesA="u" speciesB="u" size="8">
<coefficients id="uu" type="Array">
</coefficients>
</correlation>
</jastrow>
</wavefunction>
</qmcsystem>
We note that we only specify an “up” determinant, since we no longer need a product of up and down determinants. In the Jastrow specification, we only need to provide the jastrow terms for the same spin as there is no longer a distinction between the up and down spins.
We also make a small modification in the particleset specification:
<particleset name="e" random="yes" randomsrc="ion0">
<group name="u" size="10" mass="1.0">
<parameter name="charge" > -1 </parameter>
<parameter name="mass" > 1.0 </parameter>
</group>
</particleset>
Note that we only provide a single electron group to represent all electrons in the system, as opposed to the usual separation of up and down electrons.
note: In the current implementation, spinor wavefunctions are only supported at the single determinant level. Multideterminant spinor wave functions will be supported in a future release.
QMC Methods¶
In this formalism, the spin degree of freedom becomes a continuous variable similar to the spatial degrees of freedom. In order to sample the spins, we introduce a spin kinetic energy operator
where \(\mu_s\) is a spin mass. This operator vanishes when acting on an arbitrary spinor or anti-symmetric product of spinors due to the offset. This avoids any unphysical contribution to the local energy. However, this does contribute to the Green’s function in DMC,
where \(G(\mathbf{R}'\leftarrow\mathbf{R}; \tau)\) is the usual Green’s function for the spatial evolution and the spin kinetic energy operator introduces a Green’s function for the spin variables. Note that this includes a contribution from the spin drift \(\mathbf{v}_{\mathbf{S}}(\mathbf{S}) = \nabla_{\mathbf{S}} \ln \Psi_T(\mathbf{S})\).
In both the VMC and DMC methods, the spin sampling is controlled by two input
parameters in the xml blocks.
<qmc method="vmc/dmc">
<parameter name="steps" > 50 </parameter>
<parameter name="blocks" > 50 </parameter>
<parameter name="walkers" > 10 </parameter>
<parameter name="timestep" > 0.01 </parameter>
<parameter name="spinMoves"> yes </parameter>
<parameter name="spinMass" > 1.0 </parameter>
</qmc>
The spinMoves flag turns on the spin sampling, which is off by default.
The spinMass flag sets the \(\mu_s\) parameter used in the
particle updates, and effectively controls the rate of sampling for the spin
coordinates relative to the spatial coordinates.
A larger/smaller spin mass corresponds to slower/faster spin sampling relative to the spatial coordinates.
Spin-Orbit Effective Core Potentials¶
The spin-orbit contribution to the Hamiltonian can be introduced through the use of Effective Core Potentials (ECPs). As described in [MBM16], the relativistic (semilocal) ECPs take the general form
where the projectors \(|\ell j m_j\rangle\) are the so-called spin spherical harmonics. An equivalent formulation is to decouple the fully relativistic effective core potential (RECP) into averaged relativistic (ARECP) and spin-orbit (SORECP) contributions:
Note that the \(W^{\rm ARECP}\) takes exactly the same form as
the semilocal pseudopotentials used in standard QMC calculations.
In the pseudopotential .xml file format, the \(W^{\rm ARECP}_\ell(r)\) terms are tabulated as usual.
If spin-orbit terms are included in the .xml file, the file must tabulate the entire radial spin-orbit prefactor \(\frac{2}{2\ell + 1}\Delta W^{\rm SORECP}_\ell(r)\).
We note the following relations between the two representations of the relativistic potentials
The structure of the spin-orbit .xml is
<?xml version="1.0" encoding="UTF-8"?>
<pseudo>
<header ... relativistic="yes" ... />
<grid ... />
<semilocal units="hartree" format="r*V" npots-down="4" npots-up="0" l-local="3" npots="2">
<vps l="s" .../>
<vps l="p" .../>
<vps l="d" .../>
<vps l="f" .../>
<vps_so l="p" .../>
<vps_so l="d" .../>
</semilocal>
</pseudo>
This is included in the Hamiltonian in the same way as the usual pseudopotentials.
If the <vps_so> elements are found, the spin-orbit contributions will be present in the calculation.
By default, the spin-orbit terms will not be included in the local energy, but will be accumulated as an estimator.
In order to include the spin-orbit directly in the local energy (and therefore propogated into the walker weights in DMC for example),
the physicalSO flag should be set to yes in the Hamiltonian input, for example
<hamiltonian name="h0" type="generic" target="e">
<pairpot name="ElecElec" type="coulomb" source="e" target="e" physical="true"/>
<pairpot name="IonIon" type="coulomb" source=ion0" target="ion0" physical="true"/>
<pairpot name="PseudoPot" type="pseudo" source="i" wavefunction="psi0" format="xml" physicalSO="yes">
<pseudo elementType="Pb" href="Pb.xml"/>
</pairpot>
</hamiltonian>
The contribution from the spin-orbit will be printed to the .stat.h5 and .scalar.dat files for post-processing.
An example output is shown below
LocalEnergy = -3.4419 +/- 0.0014
Variance = 0.1132 +/- 0.0013
Kinetic = 1.1252 +/- 0.0027
LocalPotential = -4.5671 +/- 0.0028
ElecElec = 1.6881 +/- 0.0025
LocalECP = -6.5021 +/- 0.0062
NonLocalECP = 0.3286 +/- 0.0025
LocalEnergy_sq = 11.9601 +/- 0.0086
SOECP = -0.08163 +/- 0.0003
The NonLocalECP represents the \(W^{\rm ARECP}\), SOECP represents the \(W^{\rm SORECP}\), and the sum is the full \(W^{\rm RECP}\) contribution.
- DC12
Michael Dolg and Xiaoyan Cao. Relativistic pseudopotentials: their development and scope of applications. Chemical Reviews, 112(1):403–480, 2012. PMID: 21913696. URL: https://doi.org/10.1021/cr2001383, arXiv:https://doi.org/10.1021/cr2001383, doi:10.1021/cr2001383.
- MBM16(1,2)
Cody A. Melton, M. Chandler Bennett, and Lubos Mitas. Quantum monte carlo with variable spins. The Journal of Chemical Physics, 144(24):244113, 2016. URL: https://doi.org/10.1063/1.4954726, arXiv:https://doi.org/10.1063/1.4954726, doi:10.1063/1.4954726.
- MZG+16
Cody A. Melton, Minyi Zhu, Shi Guo, Alberto Ambrosetti, Francesco Pederiva, and Lubos Mitas. Spin-orbit interactions in electronic structure quantum monte carlo methods. Phys. Rev. A, 93:042502, Apr 2016. URL: https://link.aps.org/doi/10.1103/PhysRevA.93.042502, doi:10.1103/PhysRevA.93.042502.
Auxiliary-Field Quantum Monte Carlo¶
The AFQMC method is an orbital-space formulation of the imaginary-time propagation algorithm. We refer the reader to one of the review articles on the method [Zha13][ZK03][PZ04] for a detailed description of the algorithm. It uses the Hubbard-Stratonovich transformation to express the imaginary-time propagator, which is inherently a 2-body operator, as an integral over 1-body propagators, which can be efficiently applied to an arbitrary Slater determinant. This transformation allows us to represent the interacting many-body system as an average over a noninteracting system (e.g., Slater determinants) in a time-dependent fluctuating external field (the Auxiliary fields). The walkers in this case represent nonorthogonal Slater determinants, whose time average represents the desired quantum state. QMCPACK currently implements the phaseless AFQMC algorithm of Zhang and Krakauer [ZK03], where a trial wavefunction is used to project the simulation to the real axis, controlling the fermionic sign problem at the expense of a bias. This approximation is similar in spirit to the fixed-node approximation in real-space DMC but applied in the Hilbert space where the AFQMC random walk occurs.
Input¶
The input for an AFQMC calculation is fundamentally different to the
input for other real-space algorithms in QMCPACK. The main source of
input comes from the Hamiltonian matrix elements in an appropriate
single particle basis. This must be evaluated by an external code and
saved in a format that QMCPACK can read. More details about file formats
follow. The input file has six basic xml-blocks: AFQMCInfo,
Hamiltonian, Wavefunction, WalkerSet, Propagator, and
execute. The first five define input structures required for various
types of calculations. The execute block represents actual
calculations and takes as input the other blocks. Nonexecution blocks
are parsed first, followed by a second pass where execution blocks are
parsed (and executed) in order. Listing 51 shows an example of a
minimal input file for an AFQMC calculation.
table13 shows a brief description of the most
important parameters in the calculation. All xml sections contain a
“name” argument used to identify the resulting object within QMCPACK.
For example, in the example, multiple Hamiltonian objects with different
names can be defined. The one actually used in the calculation is the
one passed to “execute” as ham.
<?xml version="1.0"?>
<simulation method="afqmc">
<project id="Carbon" series="0"/>
<AFQMCInfo name="info0">
<parameter name="NMO">32</parameter>
<parameter name="NAEA">16</parameter>
<parameter name="NAEB">16</parameter>
</AFQMCInfo>
<Hamiltonian name="ham0" info="info0">
<parameter name="filename">fcidump.h5</parameter>
</Hamiltonian>
<Wavefunction name="wfn0" type="MSD" info="info0">
<parameter name="filetype">hdf5</parameter>
<parameter name="filename">wfn.h5</parameter>
</Wavefunction>
<WalkerSet name="wset0">
<parameter name="walker_type">closed</parameter>
</WalkerSet>
<Propagator name="prop0" info="info0">
</Propagator>
<execute wset="wset0" ham="ham0" wfn="wfn0" prop="prop0" info="info0">
<parameter name="timestep">0.005</parameter>
<parameter name="blocks">10000</parameter>
<parameter name="nWalkers">20</parameter>
<Estimator name="back_propagation">
<parameter name="naverages">4</parameter>
<parameter name="nsteps">400</parameter>
<parameter name="path_restoration">true</parameter>
<onerdm/>
<diag2rdm/>
<twordm/>
<ontop2rdm/>
<realspace_correlators/>
<correlators/>
<genfock/>
</Estimator>
</execute>
</simulation>
The following list includes all input sections for AFQMC calculations, along with a detailed explanation of accepted parameters. Since the code is under active development, the list of parameters and their interpretation might change in the future.
AFQMCInfo: Input block that defines basic information about the
calculation. It is passed to all other input blocks to propagate the
basic information: <AFQMCInfo name="info0">
NMO. Number of molecular orbitals, i.e., number of states in the single particle basis.
NAEA. Number of active electrons-alpha, i.e., number of spin-up electrons.
NAEB. Number of active electrons-beta, i.e., number of spin-down electrons.
Hamiltonian: Controls the object that reads, stores, and manages the
hamiltonian.
<Hamiltonian name="ham0" type="SparseGeneral" info="info0">
filename. Name of file with the
Hamiltonian. This is a required parameter.cutoff_1bar. Cutoff applied to integrals during reading. Any term in the Hamiltonian smaller than this value is set to zero. (For filetype=“hdf5”, the cutoff is applied only to the 2-electron integrals). Default: 1e-8
cutoff_decomposition. Cutoff used to stop the iterative cycle in the generation of the Cholesky decomposition of the 2-electron integrals. The generation of Cholesky vectors is stopped when the maximum error in the diagonal reaches this value. In case of an eigenvalue factorization, this becomes the cutoff applied to the eigenvalues. Only eigenvalues above this value are kept. Default: 1e-6
nblocks. This parameter controls the distribution of the 2-electron integrals among processors. In the default behavior (nblocks=1), all nodes contain the entire list of integrals. If nblocks \(>\) 1, the of nodes in the calculation will be split in nblocks groups. Each node in a given group contains the same subset of integrals and subsequently operates on this subset during any further operation that requires the hamiltonian. The maximum number of groups is NMO. Currently only works for filetype=“hdf5” and the file must contain integrals. Not yet implemented for input hamiltonians in the form of Cholesky vectors or for ASCII input. Coming soon! Default: No distribution
printEig. If “yes”, prints additional information during the Cholesky decomposition. Default: no
fix_2eint. If this is set to “yes”, orbital pairs that are found not to be positive definite are ignored in the generation of the Cholesky factorization. This is necessary if the 2-electron integrals are not positive definite because of round-off errors in their generation. Default: no
Wavefunction: controls the object that manages the trial
wavefunctions. This block expects a list of xml-blocks defining actual
trial wavefunctions for various roles.
<Wavefunction name="wfn0" type="MSD/PHMSD" info="info0">
filename. Name of file with wavefunction information.
cutoff. cutoff applied to the terms in the calculation of the local energy. Only terms in the Hamiltonian above this cutoff are included in the evaluation of the energy. Default: 1e-6
nnodes. Defines the parallelization of the local energy evaluation and the distribution of the
Hamiltonianmatrix (not to GPU)nbatch_qr. This turns on(>=1)/off(==0) batched QR calculation. -1 means all the walkers in the batch. Default: 0 (CPU) / -1 (GPU)
WalkerSet: Controls the object that handles the set of walkers.
<WalkerSet name="wset0">
walker_type. Type of walker set: closed or collinear. Default: collinear
pop_control. Population control algorithm. Options: “simple”: Uses a simple branching scheme with a fluctuating population. Walkers with weight above max_weight are split into multiple walkers of weight reset_weight. Walkers with weight below min_weight are killed with probability (weight/min_weight); “pair”: Fixed-population branching algorithm, based on QWalk’s branching algorithm. Pairs of walkers with weight above/below max_weight/min_weight are combined into 2 walkers with weights equal to \((w_1+w_2)/2\). The probability of replicating walker w1 (larger weight) occurs with probability \(w_1/(w_1+w_2)\), otherwise walker w2 (lower weight) is replicated; “comb”: Fixed-population branching algorithm based on the Comb method. Will be available in the next release. Default: “pair”
min_weight. Weight at which walkers are possibly killed (with probability weight/min_weight). Default: 0.05
max_weight. Weight at which walkers are replicated. Default: 4.0
reset_weight. Weight to which replicated walkers are reset to. Default: 1.0
Propagator: Controls the object that manages the propagators.
<Propagator name="prop0" info="info0">
cutoff. Cutoff applied to Cholesky vectors. Elements of the Cholesky vectors below this value are set to zero. Only meaningful with sparse hamiltonians. Default: 1e-6
substractMF. If “yes”, apply mean-field subtraction based on the ImpSamp trial wavefunction. Must set to “no” to turn it off. Default: yes
vbias_bound. Upper bound applied to the vias potential. Components of the vias potential above this value are truncated there. The bound is currently applied to \(\sqrt{\tau} v_{bias}\), so a larger value must be used as either the time step or the fluctuations increase (e.g. from running a larger system or using a poor trial wavefunction). Default: 3.0
apply_constrain. If “yes”, apply the phaseless constrain to the walker propagation. Currently, setting this to “no” produces unknown behavior, since free propagation algorithm has not been tested. Default: yes
hybrid. If “yes”, use hybrid propagation algorithm. This propagation scheme doesn’t use the local energy during propagation, leading to significant speed ups when its evaluation cost is high. The local energy of the ImpSamp trial wavefunction is never evaluated. To obtain energy estimates in this case, you must define an Estimator xml-block with the
Wavefunctionblock. The local energy of this trial wavefunction is evaluated and printed. It is possible to use a previously defined trial wavefunction in the Estimator block, just set its “name” argument to the name of a previously defined wavefunction. In this case, the same object is used for both roles. Default: nonnodes. Controls the parallel propagation algorithm. If nnodes \(>\) 1, the nodes in the simulation are split into groups of nnodes nodes, each group working collectively to propagate their walkers. Default: 1 (Serial algorithm)
nbatch. This turns on(>=1)/off(==0) batched calculation of density matrices and overlaps. -1 means all the walkers in the batch. Default: 0 (CPU) / -1 (GPU)
nbatch_qr. This turns on(>=1)/off(==0) batched QR calculation. -1 means all the walkers in the batch. Default: 0 (CPU) / -1 (GPU)
execute: Defines an execution region.
<execute wset="wset0" ham="ham0" wfn="wfn0" prop="prop0" info="info0">
nWalkers. Initial number of walkers per core group (see ncores). This sets the number of walkers for a given group of “ncores” on a node; the total number of walkers in the simulation depends on the total number of nodes and on the total number of cores on a node in the following way: \(\#_walkers_total = nWalkers * \#_nodes * \#_cores_total / ncores\). Default: 5
timestep. Time step in 1/a.u. Default: 0.01
blocks. Number of blocks. Slow operations occur once per block (e.g., write to file, slow observables, checkpoints), Default: 100
step. Number of steps within a block. Operations that occur at the step level include load balance, orthogonalization, branching, etc. Default: 1
substep. Number of substeps within a step. Only walker propagation occurs in a substep. Default: 1
ortho. Number of steps between orthogonalization. Default: 1
ncores. Number of nodes in a task group. This number defines the number of cores on a node that share the parallel work associated with a distributed task. This number is used in the
WavefunctionandPropagatortask groups. The walker sets are shares by the ncores on a given node in the task group.checkpoint. Number of blocks between checkpoint files are generated. If a value smaller than 1 is given, no file is generated. If hdf_write_file is not set, a default name is used. Default: 0
hdf_write_file. If set (and checkpoint>0), a checkpoint file with this name will be written.
hdf_read_file. If set, the simulation will be restarted from the given file.
Within the Estimators xml block has an argument name: the type
of estimator we want to measure. Currently available estimators include:
“basic”, “energy”, “mixed_one_rdm”, and “back_propagation”.
The basic estimator has the following optional parameters:
timers. print timing information. Default: true
The back_propagation estimator has the following parameters:
ortho. Number of back-propagation steps between orthogonalization. Default: 10
nsteps. Maximum number of back-propagation steps. Default: 10
naverages. Number of back propagation calculations to perform. The number of steps will be chosed equally distributed in the range 0,nsteps. Default: 1
block_size. Number of blocks to use in the internal average of the back propagated estimator. This is used to block data and reduce the size of the output. Default: 1
nskip. Number of blocks to skip at the start of the calculation for equilibration purposes. Default: 0
path_restoration. Use full path restoration. Can result in better back propagated results. Default false.
The following observables can be computed with the back_propagated estimator
onerdm. One-particle reduced density matrix.
twordm. Full Two-particle reduced density matrix.
diag2rdm. Diagonal part of the two-particle reduced density matrix.
ontop2rdm. On top two-particle reduced density matrix.
realspace_correlators. Charge-Charge, and spin-spin correlation functions in real space.
correlators. Charge-Charge, and spin-spin correlation functions in real space centered about atomic sites.
genfock. Generalized Fock matrix.
Real space correlation functions require a real space grid. Details coming soon..
Hamiltonian File formats¶
QMCPACK offers three factorization approaches which are appropriate in different settings. The most generic approach implemented is based on the modified-Cholesky factorization [BL77][KdMerasP03][ADVFerre+09][PKVZ11][PZK13] of the ERI tensor:
where the sum is truncated at \(N_{\mathrm{chol}} = x_c M\), \(x_c\) is typically between \(5\) and \(10\), \(M\) is the number of basis functions and we have assumed that the single-particle orbitals are in general complex. The storage requirement is thus naively \(\mathcal{O}(M^3)\). Note we follow the usual definition of \(v_{pqrs} = \langle pq | rs \rangle = (pr|qs)\). With this form of factorization QMCPACK allows for the integrals to be stored in either dense or sparse format.
The dense case is the simplest and is only implemented for Hamiltonians with real integrals (and basis functions, i.e. not the homegeneous electron gas which has complex orbitals but real integrals). The file format is given as follows:
$ h5dump -n afqmc.h5
HDF5 "afqmc.h5" {
FILE_CONTENTS {
group /
group /Hamiltonian
group /Hamiltonian/DenseFactorized
dataset /Hamiltonian/DenseFactorized/L
dataset /Hamiltonian/dims
dataset /Hamiltonian/hcore
dataset /Hamiltonian/Energies
}
}
where the datasets are given by the following
/Hamiltonian/DenseFactorized/LContains the \([M^2,N_\mathrm{nchol}]\) dimensional matrix representatation of \(L_{pr,n}\)./Hamiltonian/dimsDescriptor array of length 8 containing \([0,0,0,M,N_\alpha,N_\beta,0,N_\mathrm{nchol}]\). Note that \(N_\alpha\) and \(N_\beta\) are somewhat redundant and will be read from the input file and wavefunction. This allows for the Hamiltonian to be used with different (potentially spin polarized) wavefunctions./Hamiltonian/hcoreContains the \([M,M]\) dimensional one-body Hamiltonian matrix elements \(h_{pq}\)./Hamiltonian/EnergiesArray containing \([E_{II}, E_{\mathrm{core}}]\). \(E_{II}\) should contain ion-ion repulsion energy and any additional constant terms which have to be added to the total energy. \(E_{\mathrm{core}}\) is deprecated and not used.
Typically the Cholesky matrix is sparse, particularly if written in the non-orthogonal AO basis (not currently supported in QMCPACK). In this case only a small number of non-zero elements (denoted \(nnz\) below) need to be stored which can reduce the memory overhead considerably. Internally QMCPACK stores this matrix in the CSR format, and the HDF5 file format is reflective of this. For large systems and, more generally when running in parallel, it is convenient to chunk the writing/reading of the Cholesky matrix into blocks of size \([M^2,\frac{N_{\mathrm{chol}}}{N_{\mathrm{blocks}}}]\) (if interpreted as a dense array). This is achieved by writing these blocks to different data sets in the file. For the sparse case the Hamtiltonian file format is given as follows:
$ h5dump -n afqmc.h5
HDF5 "afqmc.h5" {
FILE_CONTENTS {
group /
group /Hamiltonian
group /Hamiltonian/Factorized
dataset /Hamiltonian/Factorized/block_sizes
dataset /Hamiltonian/Factorized/index_0
dataset /Hamiltonian/Factorized/vals_0
dataset /Hamiltonian/ComplexIntegrals
dataset /Hamiltonian/dims
dataset /Hamiltonian/hcore
dataset /Hamiltonian/Energies
}
}
/Hamiltonian/Factorized/block_sizesContains the number of elements in each block of the sparse representation of the Cholesky matrix \(L_{pr,n}\). In this case there is 1 block./Hamiltonian/Factorized/index_0\([2\times nnz]\) dimensional array, containing the indices of the non-zero values of \(L_{ik,n}\). The row indices are stored in the even entries, and the column indices in the odd entries./Hamiltonian/Factorized/vals_0\([nnz]\) length array containing non-zero values of \(L_{pr,n}\) for chunk 0./Hamiltonian/dimsDescriptor array of length 8 containing \([0,nnz,N_{\mathrm{block}},M,N_\alpha,N_\beta,0,N_\mathrm{nchol}]\)./Hamiltonian/ComplexIntegralsLength 1 array that specifies if integrals are complex valued. 1 for complex integrals, 0 for real integrals./Hamiltonian/hcoreContains the \([M,M]\) dimensional one-body Hamiltonian matrix elements \(h_{pq}\). Due to its small size this is written as a dense 2D-array./Hamiltonian/EnergiesArray containing \([E_{II}, E_{\mathrm{core}}]\). \(E_{II}\) should contain ion-ion repulsion energy and any additional constant terms which have to be added to the total energy. \(E_{\mathrm{core}}\) is deprecated and not used.
To reduce the memory overhead of storing the three-index tensor we recently adapted the tensor-hypercontraction [HPMartinez12][PHMartinezS12][HPSMartinez12] (THC) approach for use in AFQMCcite{MaloneISDF2019}. Within the THC approach we can approximate the orbital products entering the ERIs as
where \(\varphi_p(\mathbf{r})\) are the one-electron orbitals and \(\mathbf{r}_\mu\) are a set of specially selected interpolating points, \(\zeta_\mu(\mathbf{r})\) are a set of interpolating vectors and \(N_\mu = x_\mu M\). We can then write the ERI tensor as a product of rank-2 tensors
where
We also require the half-rotated versions of these quantities which live on a different set of \(\tilde{N}_\mu\) interpolating points \(\tilde{\mathbf{r}}_\mu\) (see [MZM19]). The file format for THC factorization is as follows:
$ h5dump -n afqmc.h5
HDF5 "afqmc.h5" {
FILE_CONTENTS {
group /
group /Hamiltonian
group /Hamiltonian/THC
dataset /Hamiltonian/THC/Luv
dataset /Hamiltonian/THC/Orbitals
dataset /Hamiltonian/THC/HalfTransformedMuv
dataset /Hamiltonian/THC/HalfTransformedFullOrbitals
dataset /Hamiltonian/THC/HalfTransformedOccOrbitals
dataset /Hamiltonian/THC/dims
dataset /Hamiltonian/ComplexIntegrals
dataset /Hamiltonian/dims
dataset /Hamiltonian/hcore
dataset /Hamiltonian/Energies
}
}
/Hamiltonian/THC/LuvCholesky factorization of the \(M_{\mu\nu}\) matrix given in (67)./Hamiltonian/THC/Orbitals\([M,N_\mu]\) dimensional array of orbitals evaluated at chosen interpolating points \(\varphi_i(\mathbf{r}_\mu)\)./Hamiltonian/THC/HalfTransformedMuv\([\tilde{N}_\mu,\tilde{N}_\mu]\) dimensional array containing half-transformed \(\tilde{M}_{\mu\nu}\)./Hamiltonian/THC/HalfTransformedFullOrbitals\([M,\tilde{N}_\mu]\) dimensional array containing orbital set computed at half-transformed interpolating points \(\varphi_i(\tilde{\mathbf{r}}_\mu)\)./Hamiltonian/THC/HalfTransformedOccOrbitals\([N_\alpha+N_\beta,\tilde{N}_\mu]\) dimensional array containing half-rotated orbital set computed at half-transformed interpolating points \(\varphi_a(\tilde{\mathbf{r}}_\mu) = \sum_{p} A_{pa}^* \varphi_{p}(\tilde{\mathbf{r}}_\mu)\), where \(\mathbf{A}\) is the Slater-Matrix of the (currently single-determinant) trial wavefunction./Hamiltonian/THC/dimsDescriptor array containing \([M, N_\mu, \tilde{N}_\mu]\)./Hamiltonian/ComplexIntegralsLength 1 array that specifies if integrals are complex valued. 1 for complex integrals, 0 for real integrals./Hamiltonian/dimsDescriptor array of length 8 containing \([0,0,0,M,N_\alpha,N_\beta,0,0]\)./Hamiltonian/hcoreContains the \([M,M]\) dimensional one-body Hamiltonian matrix elements \(h_{ij}\)./Hamiltonian/EnergiesArray containing \([E_{II}, E_{\mathrm{core}}]\). \(E_{II}\) should contain ion-ion repulsion energy and any additional constant terms which have to be added to the total energy (such as the electron-electron interaction Madelung contribution of \(\frac{1}{2} N \xi )\). \(E_{\mathrm{core}}\) is deprecated and not used.
Finally, we have implemented an explicitly \(k\)-point dependent factorization for periodic systems [MZC19][MZM20]
where \(\textbf{k}\), \(\textbf{k}'\) and \(\textbf{Q}\) are vectors in the first Brillouin zone. The one-body Hamiltonian is block diagonal in \(\textbf{k}\) and in (68) we have used momentum conservation \((\textbf{k}_p - \textbf{k}_r + \textbf{k}_q - \textbf{k}_s) = \textbf{G}\) with \(\textbf{G}\) being some vector in the reciprocal lattice of the simulation cell. The convention for the Cholesky matrix \(L^{\textbf{Q},\textbf{k}}_{pr,\gamma}\) is as follows: \(\textbf{k}_r = \textbf{k}_p - \textbf{Q}\), so the vector \(\textbf{k}\) labels the k-point of the first band index, \(\textit{p}\), while the k-point vector of the second band index, \(\textit{r}\), is given by \(\textbf{k} - \textbf{Q}\). Electron repulsion integrals at different \(\textbf{Q}\) vectors are zero by symmetry, resulting in a reduction in the number of required \(\mathbf{Q}\) vectors. For certain \(\textbf{Q}\) vectors that satisfy \(\textbf{Q} \ne -\textbf{Q}\) (this is not satisfied at the origin and at high symmetry points on the edge of the 1BZ), we have \({L^{\textbf{Q},\textbf{k}}_{sq,\gamma}}^{*} = {L^{-\textbf{Q},\textbf{k}-\textbf{Q}}_{qs,\gamma}}\), which requires us to store Cholesky vectors for either one of the \((\textbf{Q},-\textbf{Q})\) pair, but not both.
In what follows let \(m_{\mathbf{k}}\) denote the number of basis functions for basis functions of a given \(k\)-point (these can in principle differ for different \(k\)-points due to linear dependencies), \(n^{\alpha}_{\mathbf{k}}\) the number of \(\alpha\) electrons in a given \(k\)-point and \(n_{\mathrm{chol}}^{\mathbf{Q}_n}\) the number of Cholesky vectors for momentum transfer \(\mathbf{Q}_n\). The file format for this factorization is as follows (for a \(2\times2\times2\) \(k\)-point mesh, for denser meshes generally there will be far fewer symmetry inequivalent momentum transfer vectors than there are \(k\)-points):
$ h5dump -n afqmc.h5
HDF5 "afqmc.h5" {
FILE_CONTENTS {
group /
group /Hamiltonian
group /Hamiltonian/KPFactorized
dataset /Hamiltonian/KPFactorized/L0
dataset /Hamiltonian/KPFactorized/L1
dataset /Hamiltonian/KPFactorized/L2
dataset /Hamiltonian/KPFactorized/L3
dataset /Hamiltonian/KPFactorized/L4
dataset /Hamiltonian/KPFactorized/L5
dataset /Hamiltonian/KPFactorized/L6
dataset /Hamiltonian/KPFactorized/L7
dataset /Hamiltonian/NCholPerKP
dataset /Hamiltonian/MinusK
dataset /Hamiltonian/NMOPerKP
dataset /Hamiltonian/QKTok2
dataset /Hamiltonian/H1_kp0
dataset /Hamiltonian/H1_kp1
dataset /Hamiltonian/H1_kp2
dataset /Hamiltonian/H1_kp3
dataset /Hamiltonian/H1_kp4
dataset /Hamiltonian/H1_kp5
dataset /Hamiltonian/H1_kp6
dataset /Hamiltonian/H1_kp7
dataset /Hamiltonian/ComplexIntegrals
dataset /Hamiltonian/KPoints
dataset /Hamiltonian/dims
dataset /Hamiltonian/Energies
}
}
/Hamiltonian/KPFactorized/L[n]This series of datasets store elements of the Cholesky tensors \(L[\mathbf{Q}_n,\mathbf{k},pr,n]\). Each data set is of dimension \([N_k,m_{\mathbf{k}}\times m_{\mathbf{k}'},n^{\mathbf{Q}_n}_\mathrm{chol}]\), where, again, \(k\) is the \(k\)-point associated with basis function \(p\), the \(k\)-point of basis function \(r\) is defined via the mappingQKtok2./Hamiltonian/NCholPerKP\(N_k\) length array giving number of Cholesky vectors per \(k\)-point./Hamiltonian/MinusK: \(N_k\) length array mapping a \(k\)-point to its inverse: \(\mathbf{k}_i+\)MinusK[i]\(= \mathbf{0} \mod \mathbf{G}\)./Hamiltonian/NMOPerKP: \(N_k\) length array listing number of basis functions per \(k\)-point./Hamiltonian/QKTok2: \([N_k,N_k]\) dimensional array.QKtok2[i,j]yields the \(k\) point index satisfying \(\mathbf{k}=\mathbf{Q}_i-\mathbf{k}_j+\mathbf{G}\)./Hamiltonian/dims: Descriptor array of length 8 containing \([0,0,0,M,N_\alpha,N_\beta,0,0]\)./Hamiltonian/H1_kp[n]Contains the \([m_{\mathbf{k}_n},m_{\mathbf{k}_n}]\) dimensional one-body Hamiltonian matrix elements \(h_{(\mathbf{k}_{n}p)(\mathbf{k}_{n}q)}\)./Hamiltonian/ComplexIntegralsLength 1 array that specifies if integrals are complex valued. 1 for complex integrals, 0 for real integrals./Hamiltonian/KPoints\([N_k,3]\) Dimensional array containing \(k\)-points used to sample Brillouin zone./Hamiltonian/dimsDescriptor array of length 8 containing \([0,0,N_k,M,N_\alpha,N_\beta,0,N_\mathrm{nchol}]\). Note that \(M\) is the total number of basis functions, i.e. \(M=\sum_\mathbf{k} m_\mathbf{k}\), and likewise for the number of electrons./Hamiltonian/EnergiesArray containing \([E_{II}, E_{\mathrm{core}}]\). \(E_{II}\) should contain ion-ion repulsion energy and any additional constant terms which have to be added to the total energy (such as the electron-electron interaction Madelung contribution of \(\frac{1}{2} N \xi )\). \(E_{\mathrm{core}}\) is deprecated and not used.
Complex integrals should be written as an array with an additional dimension, e.g., a 1D array should be written as a 2D array with array_hdf5[:,0]=real(1d_array) and array_hdf5[:,1]=imag(1d_array). The functions afqmctools.utils.misc.from_qmcpack_complex and afqmctools.utils.misc.to_qmcpack_complex can be used to transform qmcpack format to complex valued numpy arrays of the appropriate shape and vice versa.
Finally, if using external tools to generate this file format, we provide a sanity checker script in utils/afqmctools/bin/test_afqmc_input.py which will raise errors if the format does not conform to what is being used internally.
Wavefunction File formats¶
AFQMC allows for two types of multi-determinant trial wavefunctions: non-orthogonal multi Slater determinants (NOMSD) or SHCI/CASSCF style particle-hole multi Slater determinants (PHMSD).
The file formats are described below
NOMSD¶
h5dump -n wfn.h5
HDF5 "wfn.h5" {
FILE_CONTENTS {
group /
group /Wavefunction
group /Wavefunction/NOMSD
dataset /Wavefunction/NOMSD/Psi0_alpha
dataset /Wavefunction/NOMSD/Psi0_beta
group /Wavefunction/NOMSD/PsiT_0
dataset /Wavefunction/NOMSD/PsiT_0/data_
dataset /Wavefunction/NOMSD/PsiT_0/dims
dataset /Wavefunction/NOMSD/PsiT_0/jdata_
dataset /Wavefunction/NOMSD/PsiT_0/pointers_begin_
dataset /Wavefunction/NOMSD/PsiT_0/pointers_end_
group /Wavefunction/NOMSD/PsiT_1
dataset /Wavefunction/NOMSD/PsiT_1/data_
dataset /Wavefunction/NOMSD/PsiT_1/dims
dataset /Wavefunction/NOMSD/PsiT_1/jdata_
dataset /Wavefunction/NOMSD/PsiT_1/pointers_begin_
dataset /Wavefunction/NOMSD/PsiT_1/pointers_end_
dataset /Wavefunction/NOMSD/ci_coeffs
dataset /Wavefunction/NOMSD/dims
}
}
Note that the \(\alpha\) components of the trial wavefunction are stored under
PsiT_{2n} and the \(\beta\) components are stored under PsiT_{2n+1}.
/Wavefunction/NOMSD/Psi0_alpha\([M,N_\alpha]\) dimensional array \(\alpha\) component of initial walker wavefunction./Wavefunction/NOMSD/Psi0_beta\([M,N_\beta]\) dimensional array for \(\beta\) initial walker wavefunction./Wavefunction/NOMSD/PsiT_{2n}/data_Array of length \(nnz\) containing non-zero elements of \(n\)-th \(\alpha\) component of trial wavefunction walker wavefunction. Note the conjugate transpose of the Slater matrix is stored./Wavefunction/NOMSD/PsiT_{2n}/dimsArray of length 3 containing \([M,N_{\alpha},nnz]\) where \(nnz\) is the number of non-zero elements of this Slater matrix/Wavefunction/NOMSD/PsiT_{2n}/jdata_CSR indices array./Wavefunction/NOMSD/PsiT_{2n}/pointers_begin_CSR format begin index pointer array./Wavefunction/NOMSD/PsiT_{2n}/pointers_end_CSR format end index pointer array./Wavefunction/NOMSD/ci_coeffs\(N_D\) length array of ci coefficients. Stored as complex numbers./Wavefunction/NOMSD/dimsInteger array of length 5 containing \([M,N_\alpha,N_\beta,\) walker_type \(,N_D]\)
PHMSD¶
h5dump -n wfn.h5
HDF5 "wfn.h5" {
FILE_CONTENTS {
group /
group /Wavefunction
group /Wavefunction/PHMSD
dataset /Wavefunction/PHMSD/Psi0_alpha
dataset /Wavefunction/PHMSD/Psi0_beta
dataset /Wavefunction/PHMSD/ci_coeffs
dataset /Wavefunction/PHMSD/dims
dataset /Wavefunction/PHMSD/occs
dataset /Wavefunction/PHMSD/type
}
}
/Wavefunction/NOMSD/Psi0_alpha\([M,N_\alpha]\) dimensional array \(\alpha\) component of initial walker wavefunction./Wavefunction/NOMSD/Psi0_beta\([M,N_\beta]\) dimensional array for \(\beta\) initial walker wavefunction./Wavefunction/PHMSD/ci_coeffs\(N_D\) length array of ci coefficients. Stored as complex numbers./Wavefunction/PHMSD/dimsInteger array of length 5 containing \([M,N_\alpha,N_\beta,\) walker_type \(,N_D]\)/Wavefunction/PHMSD/occsInteger array of length \((N_\alpha+N_\beta)*N_D\) describing the determinant occupancies. For example if \((N_\alpha=N_\beta=2)\) and \(N_D=2\), \(M=4\), and if \(|\Psi_T\rangle = |0,1\rangle|0,1\rangle + |0,1\rangle|0,2\rangle>\) then occs = \([0, 1, 4, 5, 0, 1, 4, 6]\). Note that \(\beta\) occupancies are displacd by \(M\)./Wavefunction/PHMSD/typeinteger 0/1. 1 implies trial wavefunction is written in different basis than the underlying basis used for the integrals. If so a matrix of orbital coefficients is required to be written in the NOMSD format. If 0 then assume wavefunction is in same basis as integrals.
Current Feature Implementation Status¶
The current status of features available in QMCPACK is as follows:
Hamiltonian |
SD |
NOMSD |
PHMSD |
Real Build |
Complex Build |
|---|---|---|---|---|---|
Sparse |
Yes |
Yes |
Yes |
Yes |
Yes |
Dense |
Yes |
Yes |
No |
Yes |
No |
k-point |
Yes |
No |
No |
No |
Yes |
THC |
Yes |
No |
No |
Yes |
Yes |
Hamiltonian |
SD |
NOMSD |
PHMSD |
Real Build |
Complex Build |
|---|---|---|---|---|---|
Sparse |
No |
No |
No |
No |
No |
Dense |
Yes |
No |
No |
Yes |
No |
k-point |
Yes |
No |
No |
No |
Yes |
THC |
Yes |
No |
No |
Yes |
Yes |
Advice/Useful Information¶
AFQMC calculations are computationally expensive and require some care to obtain reasonable performance. The following is a growing list of useful advice for new users, followed by a sample input for a large calculation.
Generate Cholesky-decomposed integrals with external codes instead of the 2-electron integrals directly. The generation of the Cholesky factorization is faster and consumes less memory.
Use the hybrid algorithm for walker propagation. Set steps/substeps to adequate values to reduce the number of energy evaluations. This is essential when using large multideterminant expansions.
Adjust cutoffs in the wavefunction and propagator bloxks until desired accuracy is reached. The cost of the calculation will depend on these cutoffs.
Adjust ncores/nWalkers to obtain better efficiency. Larger nWalkers will lead to more efficient linear algebra operations but will increase the time per step. Larger ncores will reduce the time per step but will reduce efficiency because of inefficiencies in the parallel implementation. For large calculations, values between 6–12 for both quantities should be reasonable, depending on architecture.
...
<Hamiltonian name="ham0" type="SparseGeneral" info="info0">
<parameter name="filename">fcidump.h5</parameter>
<parameter name="cutoff_1bar">1e-6</parameter>
<parameter name="cutoff_decomposition">1e-5</parameter>
</Hamiltonian>
<Wavefunction name="wfn0" type="MSD" info="info0">
<parameter name="filetype">ascii</parameter>
<parameter name="filename">wfn.dat</parameter>
</Wavefunction>
<WalkerSet name="wset0">
<parameter name="walker_type">closed</parameter>
</WalkerSet>
<Propagator name="prop0" info="info0">
<parameter name="hybrid">yes</parameter>
</Propagator>
<execute wset="wset0" ham="ham0" wfn="wfn0" prop="prop0" info="info0">
<parameter name="ncores">8</parameter>
<parameter name="timestep">0.01</parameter>
<parameter name="blocks">10000</parameter>
<parameter name="steps">10</parameter>
<parameter name="substeps">5</parameter>
<parameter name="nWalkers">8</parameter>
<parameter name="ortho">5</parameter>
</execute>
afqmc method
parameters in AFQMCInfo
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
integer |
\(\geq 0\) |
no |
Number of molecular orbitals |
|
integer |
\(\geq 0\) |
no |
Number of active electrons of spin-up |
|
integer |
\(\geq 0\) |
no |
Number of active electrons of spin-down |
parameters in Hamiltonian
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
argument |
Name of |
||
|
string |
no |
Name of file with the hamiltonian |
|
|
string |
hdf5 |
yes |
Native HDF5-based format of QMCPACK |
parameters in Wavefunction
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
argument |
name of |
||
|
argument |
MSD, PHMSD |
no |
Linear combination of (assumed non-orthogonal) Slater determinants |
|
string |
ascii, hdf5 |
no |
CI-type multi-determinant wave function |
parameters in WalkerSet
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
string |
collinear |
yes |
Request a collinear walker set. |
closed |
no |
Request a closed shell (doubly-occupied) walker set. |
parameters in Propagator
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
argument |
afqmc |
afqmc |
Type of propagator |
|
argument |
Name of |
||
|
string |
yes |
Use hybrid propagation algorithm. |
|
no |
Use local energy based propagation algorithm. |
parameters in execute
Name |
Datatype |
Values |
Default |
Description |
|---|---|---|---|---|
|
argument |
|||
|
argument |
|||
|
argument |
|||
|
argument |
|||
|
argument |
Name of |
||
|
integer |
\(\geq 0\) |
5 |
Initial number of walkers per task group |
|
real |
\(> 0\) |
0.01 |
Time step in 1/a.u. |
|
integer |
\(\geq 0\) |
100 |
Number of blocks |
|
integer |
\(> 0\) |
1 |
Number of steps within a block |
|
integer |
\(> 0\) |
1 |
Number of substeps within a step |
|
integer |
\(> 0\) |
1 |
Number of steps between walker orthogonalization. |
AFQMCTOOLS¶
The afqmctools library found in qmcpack/utils/afqmctools provides a number of
tools to interface electronic structure codes with AFQMC in QMCPACK. Currently PYSCF is
the best supported package and is capable of generating both molecular and solid state
input for AFQMC.
In what follows we will document the most useful routines from a user’s perspective.
afqmctools has to be in your PYTHONPATH.
pyscf_to_afqmc.py¶
This is the main script to convert PYSCF output into QMCPACK input. The command line options are as follows:
> pyscf_to_afqmc.py -h
usage: pyscf_to_afqmc.py [-h] [-i CHK_FILE] [-o HAMIL_FILE] [-w WFN_FILE]
[-q QMC_INPUT] [-t THRESH] [-k] [--density-fit] [-a]
[-c CAS] [-d] [-n NDET_MAX] [-r] [-p]
[--low LOW_THRESH] [--high HIGH_THRESH] [--dense]
[-v]
optional arguments:
-h, --help show this help message and exit
-i CHK_FILE, --input CHK_FILE
Input pyscf .chk file.
-o HAMIL_FILE, --output HAMIL_FILE
Output file name for QMCPACK hamiltonian.
-w WFN_FILE, --wavefunction WFN_FILE
Output file name for QMCPACK wavefunction. By default
will write to hamil_file.
-q QMC_INPUT, --qmcpack-input QMC_INPUT
Generate skeleton QMCPACK input xml file.
-t THRESH, --cholesky-threshold THRESH
Cholesky convergence threshold.
-k, --kpoint Generate explicit kpoint dependent integrals.
--density-fit Use density fitting integrals stored in input pyscf
chkpoint file.
-a, --ao, --ortho-ao Transform to ortho AO basis. Default assumes we work
in MO basis
-c CAS, --cas CAS Specify a CAS in the form of N,M.
-d, --disable-ham Disable hamiltonian generation.
-n NDET_MAX, --num-dets NDET_MAX
Set upper limit on number of determinants to generate.
-r, --real-ham Write integrals as real numbers.
-p, --phdf Use parallel hdf5.
--low LOW_THRESH Lower threshold for non-integer occupanciesto include
in multi-determinant exansion.
--high HIGH_THRESH Upper threshold for non-integer occupanciesto include
in multi-determinant exansion.
--dense Write dense Hamiltonian.
-v, --verbose Verbose output.
examples on how to generate AFQMC input from PYSCF simulations are available in AFQMC Tutorials
afqmc_to_fcidump.py¶
This script is useful for converting AFQMC hamiltonians to the FCIDUMP format.
> afqmc_to_fcidump.py
usage: afqmc_to_fcidump.py [-h] [-i INPUT_FILE] [-o OUTPUT_FILE] [-s SYMM]
[-t TOL] [-c] [--complex-paren] [-v]
optional arguments:
-h, --help show this help message and exit
-i INPUT_FILE, --input INPUT_FILE
Input AFQMC hamiltonian file.
-o OUTPUT_FILE, --output OUTPUT_FILE
Output file for FCIDUMP.
-s SYMM, --symmetry SYMM
Symmetry of integral file (1,4,8).
-t TOL, --tol TOL Cutoff for integrals.
-c, --complex Whether to write integrals as complex numbers.
--complex-paren Whether to write FORTRAN format complex numbers.
-v, --verbose Verbose output.
fcidump_to_afqmc.py¶
This script is useful for converting Hamiltonians in the FCIDUMP format to the AFQMC file format.
> fcidump_to_afqmc.py -h
usage: fcidump_to_afqmc.py [-h] [-i INPUT_FILE] [-o OUTPUT_FILE]
[--write-complex] [-t THRESH] [-s SYMM] [-v]
optional arguments:
-h, --help show this help message and exit
-i INPUT_FILE, --input INPUT_FILE
Input FCIDUMP file.
-o OUTPUT_FILE, --output OUTPUT_FILE
Output file name for PAUXY data.
--write-complex Output integrals in complex format.
-t THRESH, --cholesky-threshold THRESH
Cholesky convergence threshold.
-s SYMM, --symmetry SYMM
Symmetry of integral file (1,4,8).
-v, --verbose Verbose output.
Writing a Hamiltonian¶
write_qmcpack_sparse and write_qmcpack_dense can be used to write either sparse or
dense qmcpack Hamiltonians.
import numpy
from afqmctools.hamiltonian.io import write_qmcpack_sparse, write_qmcpack_dense
nmo = 50
nchol = 37
nelec = (3,3)
enuc = -108.3
# hcore and eri should obey the proper symmetry in real applications
# h_ij
hcore = numpy.random.random((nmo,nmo))
# L_{(ik),n}
chol = numpy.random.random((nmo*nmo, nchol))
write_qmcpack_dense(hcore, chol, nelec, nmo, enuc,
real_chol=True,
filename='hamil_dense.h5')
write_qmcpack_sparse(hcore, chol, nelec, nmo, enuc,
real_chol=True,
filename='hamil_sparse.h5')
Note the real_chol parameter controls whether the integrals are written as real or
complex numbers. Complex numbers should be used if -DENABLE_QMC_COMPLEX=1, while the
dense Hamiltonian is only available for real builds.
Writing a wavefunction¶
write_qmcpack_wfn can be used to write either NOMSD or PHMSD wavefunctions:
import numpy
from afqmctools.wavefunction.mol import write_qmcpack_wfn
# NOMSD
ndet = 100
nmo = 50
nelec = (3, 7)
wfn = numpy.array(numpy.random.random((ndet, nmo, sum(nelec))), dtype=numpy.complex128)
coeffs = numpy.array(numpy.random.random((ndet)), dtype=numpy.complex128)
uhf = True
write_qmcpack_wfn('wfn.h5', (coeffs, wfn), uhf, nelec, nmo)
By default the first term in the expansion will be used as the initial walker
wavefunction. To use another wavefunction we can pass a value to the init parameter:
init = numpy.array(numpy.random.random((nmo,sum(nelec)), dtype=numpy.complex128)
write_qmcpack_wfn('wfn.h5', (coeffs, wfn), uhf, nelec, nmo, init=[init,init])
Particle-hole wavefunction (PHMSD) from SHCI or CASSCF calculations are also written using the same function:
import numpy
from afqmctools.wavefunction.mol import write_qmcpack_wfn
# PHMSD
ndet = 2
nmo = 4
nelec = (2,2)
uhf = True
# |psi_T> = 1/sqrt(2)(|0,1>|0,1> + |0,1>|0,2>)
coeffs = numpy.array([0.707,0.707], dtype=numpy.complex128)
occa = numpy.array([(0,1), (0,1)])
occb = numpy.array([(0,1), (0,2)])
write_qmcpack_wfn('wfn.h5', (coeffs, occa, occb), uhf, nelec, nmo)
Analyzing Estimators¶
The afqmctools.analysis.average module can be used to perform simple error analysis
for estimators computed with AFQMC.
Warning
Autocorrelation is not accounted for. Use with caution.
- average_one_rdm
Returns P[s,i,j] = \(\langle c_{is}^{\dagger} c_{js}\rangle\) as a (nspin, M, M) dimensional array.
- average_two_rdm
Gamma[s1s2,i,k,j,l] = \(\langle c_{i}^{\dagger} c_{j}^{\dagger} c_{l} c_{k} \rangle\). For closed shell systems, returns [(a,a,a,a),(a,a,b,b)]. For collinear systems, returns [(a,a,a,a),(a,a,b,b),(b,b,b,b)].
- average_diag_two_rdm
Returns \(\langle c_{is}^+ c_{jt}^+ c_{jt} c_{is}\rangle\) as a (2M,2M) dimensional array.
- average_on_top_pdm
Returns \(n_2(\mathbf{r},\mathbf{r})\) for a given real space grid.
- average_realspace_correlations
Returns \(\langle C(\mathbf{r}_1)C(\mathbf{r}_2) \rangle\) and \(\langle S(\mathbf{r}_1)S(\mathbf{r}_2) \rangle\) for a given set of points in real space. \(\hat{C} = (\hat{n}_\uparrow+ \hat{n}_\downarrow)\), \(\hat{S}=(\hat{n}_\uparrow-\hat{n}_\downarrow)\)
- average_atom_correlations
Returns \(\langle C(I) \rangle\), \(\langle S(I) \rangle\), \(\langle C(I) C(J) \rangle\), \(\langle S(I) S(J) \rangle\) for a given set of atomic sites \(I,J\). \(\hat{C} = (\hat{n}_\uparrow+ \hat{n}_\downarrow)\), \(\hat{S}=(\hat{n}_\uparrow-\hat{n}_\downarrow)\)
- average_gen_fock
Returns generalized Fock matrix \(F_{\pm}\). The parameter
fock_typeis used to specify \(F_{+}\) (fock_type='plus') or \(F_{-}\) (fock_type='minus')- get_noons
Get natural orbital occupation numbers from one-rdm.
As an example the following will extract the back propagated one rdm for the maximum propagation time, and skip 10 blocks as the equilibration phase.
from afqmctools.analysis.average import average_one_rdm
P, Perr = average_one_rdm('qmc.s000.stat.h5', estimator='back_propagated', eqlb=10)
- ADVFerre+09
Francesco Aquilante, Luca De Vico, Nicolas Ferré, Giovanni Ghigo, Per-åke Malmqvist, Pavel Neogrády, Thomas Bondo Pedersen, Michal Pitoňák, Markus Reiher, Björn O. Roos, Luis Serrano-Andrés, Miroslav Urban, Valera Veryazov, and Roland Lindh. MOLCAS 7: The Next Generation. J. Comput. Chem., 31(1):224, 2009. URL: https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.21318.
- BL77
Nelson H. F. Beebe and Jan Linderberg. Simplifications in the generation and transformation of two-electron integrals in molecular calculations. Int. J. Quantum Chem., 12(4):683, 1977. URL: https://onlinelibrary.wiley.com/doi/abs/10.1002/qua.560120408.
- HPMartinez12
Edward G. Hohenstein, Robert M. Parrish, and Todd J. Martínez. Tensor hypercontraction density fitting. I. Quartic scaling second- and third-order Møller-Plesset perturbation theory. The Journal of Chemical Physics, 137(4):044103, 2012. URL: https://doi.org/10.1063/1.4732310, doi:10.1063/1.4732310.
- HPSMartinez12
Edward G. Hohenstein, Robert M. Parrish, C. David Sherrill, and Todd J. Martínez. Communication: Tensor hypercontraction. III. Least-squares tensor hypercontraction for the determination of correlated wavefunctions. The Journal of Chemical Physics, 137(22):221101, 2012. URL: https://doi.org/10.1063/1.4768241, doi:10.1063/1.4768241.
- KdMerasP03
Henrik Koch, Alfredo Sánchez de Merás, and Thomas Bondo Pedersen. Reduced scaling in electronic structure calculations using Cholesky decompositions. The Journal of Chemical Physics, 118(21):9481, 2003. URL: https://doi.org/10.1063/1.1578621.
- MZM20
Fionn D Malone, Shuai Zhang, and Miguel A Morales. Accelerating auxiliary-field quantum monte carlo simulations of solids with graphical processing unit. arXiv preprint arXiv:2003.09468, 2020.
- MZM19
Fionn D. Malone, Shuai Zhang, and Miguel A. Morales. Overcoming the Memory Bottleneck in Auxiliary Field Quantum Monte Carlo Simulations with Interpolative Separable Density Fitting. J. Chem. Theory. Comput., 15(1):256, 2019. URL: https://doi.org/10.1021/acs.jctc.8b00944, doi:10.1021/acs.jctc.8b00944.
- MZC19
Mario Motta, Shiwei Zhang, and Garnet Kin-Lic Chan. Hamiltonian symmetries in auxiliary-field quantum Monte Carlo calculations for electronic structure. Physical Review B, 100:045127, July 2019. URL: https://link.aps.org/doi/10.1103/PhysRevB.100.045127, doi:10.1103/PhysRevB.100.045127.
- PHMartinezS12
Robert M. Parrish, Edward G. Hohenstein, Todd J. Martínez, and C. David Sherrill. Tensor hypercontraction. II. Least-squares renormalization. The Journal of Chemical Physics, 137(22):224106, 2012. URL: https://doi.org/10.1063/1.4768233, doi:10.1063/1.4768233.
- PKVZ11
Wirawan Purwanto, Henry Krakauer, Yudistira Virgus, and Shiwei Zhang. Assessing weak hydrogen binding on Ca+ centers: An accurate many-body study with large basis sets. The Journal of Chemical Physics, 135(16):164105, 2011. URL: https://doi.org/10.1063/1.3654002, doi:10.1063/1.3654002.
- PZ04
Wirawan Purwanto and Shiwei Zhang. Quantum monte carlo method for the ground state of many-boson systems. Phys. Rev. E, 70:056702, November 2004. doi:10.1103/PhysRevE.70.056702.
- PZK13
Wirawan Purwanto, Shiwei Zhang, and Henry Krakauer. Frozen-Orbital and Downfolding Calculations with Auxiliary-Field Quantum Monte Carlo. Journal of Chemical Theory and Computation, 9(11):4825–4833, 2013. URL: https://doi.org/10.1021/ct4006486.
- Zha13
Shiwei Zhang. Auxiliary-field quantum monte carlo for correlated electron systems. Modeling and Simulation, 3:, 2013. URL: http://hdl.handle.net/2128/5389, doi:.
- ZK03(1,2)
Shiwei Zhang and Henry Krakauer. Quantum monte carlo method using phase-free random walks with slater determinants. Phys. Rev. Lett., 90:136401, April 2003. doi:10.1103/PhysRevLett.90.136401.
Examples¶
WARNING: THESE EXAMPLES ARE NOT CONVERGED! YOU MUST CONVERGE PARAMETERS (SIMULATION CELL SIZE, JASTROW PARAMETER NUMBER/CUTOFF, TWIST NUMBER, DMC TIME STEP, DFT PLANE WAVE CUTOFF, DFT K-POINT MESH, ETC.) FOR REAL CALCUATIONS!
The following examples should run in serial on a modern workstation in a few hours.
Using QMCPACK directly¶
In examples/molecules are the following examples.
Each directory also contains a README file with more details.
Directory |
Description |
|
H2O molecule from GAMESS orbitals |
|
Helium atom with simple wavefunctions |
Using Nexus¶
For more information about Nexus, see the User Guide in nexus/documentation.
For Python to find the Nexus library, the PYTHONPATH environment variable should be set to <QMCPACK source>/nexus/library.
For these examples to work properly, the executables for QE and QMCPACK either
need to be on the path, or the paths in the script should be adjusted.
These examples can be found under the nexus/examples/qmcpack directory.
Directory |
Description |
|
Bulk diamond with VMC |
|
Graphene sheet with DMC |
|
C20 cage molecule |
|
Binding curve for O\(_2\) molecule |
|
H\(_2\)O molecule with QE orbitals |
|
LiH crystal with QE orbitals |
Lab 1: MC Statistical Analysis¶
Topics covered in this lab¶
This lab focuses on the basics of analyzing data from MC calculations. In this lab, participants will use data from VMC calculations of a simple 1-electron system with an analytically soluble system (the ground state of the hydrogen atom) to understand how to interpret an MC situation. Most of these analyses will also carry over to DMC simulations. Topics covered include:
Averaging MC variables
The statisical error bar of mean values
The effects of autocorrelation and variance on the error bar
The relationship between MC time step and autocorrelation
The use of blocking to reduce autocorrelation
The significance of the acceptance ratio
The significance of the sample size
How to determine whether an MC run was successful
The relationship between wavefunction quality and variance
Gauging the efficiency of MC runs
The cost of scaling up to larger system sizes
Lab directories and files¶
labs/lab1_qmc_statistics/
│
├── atom - H atom VMC calculation
│ ├── H.s000.scalar.dat - H atom VMC data
│ └── H.xml - H atom VMC input file
│
├── autocorrelation - varying autocorrelation
│ ├── H.dat - data for gnuplot
│ ├── H.plt - gnuplot for time step vs. E_L, tau_c
│ ├── H.s000.scalar.dat - H atom VMC data: time step = 10
│ ├── H.s001.scalar.dat - H atom VMC data: time step = 5
│ ├── H.s002.scalar.dat - H atom VMC data: time step = 2
│ ├── H.s003.scalar.dat - H atom VMC data: time step = 1
│ ├── H.s004.scalar.dat - H atom VMC data: time step = 0.5
│ ├── H.s005.scalar.dat - H atom VMC data: time step = 0.2
│ ├── H.s006.scalar.dat - H atom VMC data: time step = 0.1
│ ├── H.s007.scalar.dat - H atom VMC data: time step = 0.05
│ ├── H.s008.scalar.dat - H atom VMC data: time step = 0.02
│ ├── H.s009.scalar.dat - H atom VMC data: time step = 0.01
│ ├── H.s010.scalar.dat - H atom VMC data: time step = 0.005
│ ├── H.s011.scalar.dat - H atom VMC data: time step = 0.002
│ ├── H.s012.scalar.dat - H atom VMC data: time step = 0.001
│ ├── H.s013.scalar.dat - H atom VMC data: time step = 0.0005
│ ├── H.s014.scalar.dat - H atom VMC data: time step = 0.0002
│ ├── H.s015.scalar.dat - H atom VMC data: time step = 0.0001
│ └── H.xml - H atom VMC input file
│
├── average - Python scripts for average/std. dev.
│ ├── average.py - average five E_L from H atom VMC
│ ├── stddev2.py - standard deviation using (E_L)^2
│ └── stddev.py - standard deviation around the mean
│
├── basis - varying basis set for orbitals
│ ├── H__exact.s000.scalar.dat - H atom VMC data using STO basis
│ ├── H_STO-2G.s000.scalar.dat - H atom VMC data using STO-2G basis
│ ├── H_STO-3G.s000.scalar.dat - H atom VMC data using STO-3G basis
│ └── H_STO-6G.s000.scalar.dat - H atom VMC data using STO-6G basis
│
├── blocking - varying block/step ratio
│ ├── H.dat - data for gnuplot
│ ├── H.plt - gnuplot for N_block vs. E, tau_c
│ ├── H.s000.scalar.dat - H atom VMC data 50000:1 blocks:steps
│ ├── H.s001.scalar.dat - " " " " 25000:2 blocks:steps
│ ├── H.s002.scalar.dat - " " " " 12500:4 blocks:steps
│ ├── H.s003.scalar.dat - " " " " 6250: 8 blocks:steps
│ ├── H.s004.scalar.dat - " " " " 3125:16 blocks:steps
│ ├── H.s005.scalar.dat - " " " " 2500:20 blocks:steps
│ ├── H.s006.scalar.dat - " " " " 1250:40 blocks:steps
│ ├── H.s007.scalar.dat - " " " " 1000:50 blocks:steps
│ ├── H.s008.scalar.dat - " " " " 500:100 blocks:steps
│ ├── H.s009.scalar.dat - " " " " 250:200 blocks:steps
│ ├── H.s010.scalar.dat - " " " " 125:400 blocks:steps
│ ├── H.s011.scalar.dat - " " " " 100:500 blocks:steps
│ ├── H.s012.scalar.dat - " " " " 50:1000 blocks:steps
│ ├── H.s013.scalar.dat - " " " " 40:1250 blocks:steps
│ ├── H.s014.scalar.dat - " " " " 20:2500 blocks:steps
│ ├── H.s015.scalar.dat - " " " " 10:5000 blocks:steps
│ └── H.xml - H atom VMC input file
│
├── blocks - varying total number of blocks
│ ├── H.dat - data for gnuplot
│ ├── H.plt - gnuplot for N_block vs. E
│ ├── H.s000.scalar.dat - H atom VMC data 500 blocks
│ ├── H.s001.scalar.dat - " " " " 2000 blocks
│ ├── H.s002.scalar.dat - " " " " 8000 blocks
│ ├── H.s003.scalar.dat - " " " " 32000 blocks
│ ├── H.s004.scalar.dat - " " " " 128000 blocks
│ └── H.xml - H atom VMC input file
│
├── dimer - comparing no and simple Jastrow factor
│ ├── H2_STO___no_jastrow.s000.scalar.dat - H dimer VMC data without Jastrow
│ └── H2_STO_with_jastrow.s000.scalar.dat - H dimer VMC data with Jastrow
│
├── docs - documentation
│ ├── Lab_1_MC_Analysis.pdf - this document
│ └── Lab_1_Slides.pdf - slides presented in the lab
│
├── nodes - varying number of computing nodes
│ ├── H.dat - data for gnuplot
│ ├── H.plt - gnuplot for N_node vs. E
│ ├── H.s000.scalar.dat - H atom VMC data with 32 nodes
│ ├── H.s001.scalar.dat - H atom VMC data with 128 nodes
│ └── H.s002.scalar.dat - H atom VMC data with 512 nodes
│
├── problematic - problematic VMC run
│ └── H.s000.scalar.dat - H atom VMC data with a problem
│
└── size - scaling with number of particles
├── 01________H.s000.scalar.dat - H atom VMC data
├── 02_______H2.s000.scalar.dat - H dimer " "
├── 06________C.s000.scalar.dat - C atom " "
├── 10______CH4.s000.scalar.dat - methane " "
├── 12_______C2.s000.scalar.dat - C dimer " "
├── 16_____C2H4.s000.scalar.dat - ethene "
├── 18___CH4CH4.s000.scalar.dat - methane dimer VMC data
├── 32_C2H4C2H4.s000.scalar.dat - ethene dimer " "
├── nelectron_tcpu.dat - data for gnuplot
└── Nelectron_tCPU.plt - gnuplot for N_elec vs. t_CPU
Atomic units¶
QMCPACK operates in Ha atomic units to reduce the number of factors in the Schrödinger equation. Thus, the unit of length is the bohr (5.291772 \(\times 10^{-11}\) m = 0.529177 Å); the unit of energy is the Ha (4.359744 \(\times 10^{-18}\) J = 27.211385 eV). The energy of the ground state of the hydrogen atom in these units is -0.5 Ha.
Reviewing statistics¶
We will practice taking the average (mean) and standard deviation of some MC data by hand to review the basic definitions.
Enter Python’s command line by typing python [Enter]. You will see a
prompt “>>>.”
The mean of a dataset is given by:
To calculate the average of five local energies from an MC calculation
of the ground state of an electron in the hydrogen atom, input (truncate
at the thousandths place if you cannot copy and paste; script versions
are also available in the average directory):
(
(-0.45298911858) +
(-0.45481953564) +
(-0.48066105923) +
(-0.47316713469) +
(-0.46204733302)
)/5.
Then, press [Enter] to get:
>>> ((-0.45298911858) + (-0.45481953564) + (-0.48066105923) +
(-0.47316713469) + (-0.4620473302))/5.
-0.46473683566800006
To understand the significance of the mean, we also need the standard deviation around the mean of the data (also called the error bar), given by:
To calculate the standard deviation around the mean (-0.464736835668) of these five data points, put in:
( (1./(5.*(5.-1.))) * (
(-0.45298911858-(-0.464736835668))**2 + \\
(-0.45481953564-(-0.464736835668))**2 +
(-0.48066105923-(-0.464736835668))**2 +
(-0.47316713469-(-0.464736835668))**2 +
(-0.46204733302-(-0.464736835668))**2 )
)**0.5
Then, press [Enter] to get:
>>> ( (1./(5.*(5.-1.))) * ( (-0.45298911858-(-0.464736835668))**2 +
(-0.45481953564-(-0.464736835668))**2 + (-0.48066105923-(-0.464736835668))**2 +
(-0.47316713469-(-0.464736835668))**2 + (-0.46204733302-(-0.464736835668))**2
) )**0.5
0.0053303187464332066
Thus, we might report this data as having a value -0.465 +/- 0.005 Ha. This calculation of the standard deviation assumes that the average for this data is fixed, but we can continually add MC samples to the data, so it is better to use an estimate of the error bar that does not rely on the overall average. Such an estimate is given by:
To calculate the standard deviation with this formula, input the following, which includes the square of the local energy calculated with each corresponding local energy:
( (1./(5.-1.)) * (
(0.60984565298-(-0.45298911858)**2) + \\
(0.61641291630-(-0.45481953564)**2) +
(1.35860151160-(-0.48066105923)**2) + \\
(0.78720769003-(-0.47316713469)**2) +
(0.56393677687-(-0.46204733302)**2) )
)**0.5
and press [Enter] to get:
>>> ((1./(5.-1.))*((0.60984565298-(-0.45298911858)**2)+
(0.61641291630-(-0.45481953564)**2)+(1.35860151160-(-0.48066105923)**2)+
(0.78720769003-(-0.47316713469)**2)+(0.56393677687-(-0.46204733302)**2))
)**0.5
0.84491636672906634
This much larger standard deviation, acknowledging that the mean of this small data set is not the average in the limit of infinite sampling, more accurately reports the value of the local energy as -0.5 +/- 0.8 Ha.
Type quit() and press [Enter] to exit the Python command line.
Inspecting MC Data¶
QMCPACK outputs data from MC calculations into files ending in
scalar.dat. Several quantities are calculated and written for each
block of MC steps in successive columns to the right of the step index.
Change directories to atom, and open the file ending in
scalar.dat with a text editor (e.g., vi *.scalar.dat or emacs
*.scalar.dat. If possible, adjust the terminal so that lines do not
wrap. The data will begin as follows (broken into three groups to fit on
this page):
# index LocalEnergy LocalEnergy_sq LocalPotential ...
0 -4.5298911858e-01 6.0984565298e-01 -1.1708693521e+00
1 -4.5481953564e-01 6.1641291630e-01 -1.1863425644e+00
2 -4.8066105923e-01 1.3586015116e+00 -1.1766446209e+00
3 -4.7316713469e-01 7.8720769003e-01 -1.1799481122e+00
4 -4.6204733302e-01 5.6393677687e-01 -1.1619244081e+00
5 -4.4313854290e-01 6.0831516179e-01 -1.2064503041e+00
6 -4.5064926960e-01 5.9891422196e-01 -1.1521370176e+00
7 -4.5687452611e-01 5.8139614676e-01 -1.1423627617e+00
8 -4.5018503739e-01 8.4147849706e-01 -1.1842075439e+00
9 -4.3862013841e-01 5.5477715836e-01 -1.2080979177e+00
The first line begins with a #, indicating that this line does not contain MC data but rather the labels of the columns. After a blank line, the remaining lines consist of the MC data. The first column, labeled index, is an integer indicating which block of MC data is on that line. The second column contains the quantity usually of greatest interest from the simulation: the local energy. Since this simulation did not use the exact ground state wavefunction, it does not produce -0.5 Ha as the local energy although the value lies within about 10%. The value of the local energy fluctuates from block to block, and the closer the trial wavefunction is to the ground state the smaller these fluctuations will be. The next column contains an important ingredient in estimating the error in the MC average—the square of the local energy—found by evaluating the square of the Hamiltonian.
... Kinetic Coulomb BlockWeight ...
7.1788023352e-01 -1.1708693521e+00 1.2800000000e+04
7.3152302871e-01 -1.1863425644e+00 1.2800000000e+04
6.9598356165e-01 -1.1766446209e+00 1.2800000000e+04
7.0678097751e-01 -1.1799481122e+00 1.2800000000e+04
6.9987707508e-01 -1.1619244081e+00 1.2800000000e+04
7.6331176120e-01 -1.2064503041e+00 1.2800000000e+04
7.0148774798e-01 -1.1521370176e+00 1.2800000000e+04
6.8548823555e-01 -1.1423627617e+00 1.2800000000e+04
7.3402250655e-01 -1.1842075439e+00 1.2800000000e+04
7.6947777925e-01 -1.2080979177e+00 1.2800000000e+04
The fourth column from the left consists of the values of the local potential energy. In this simulation, it is identical to the Coulomb potential (contained in the sixth column) because the one electron in the simulation has only the potential energy coming from its interaction with the nucleus. In many-electron simulations, the local potential energy contains contributions from the electron-electron Coulomb interactions and the nuclear potential or pseudopotential. The fifth column contains the local kinetic energy value for each MC block, obtained from the Laplacian of the wavefunction. The sixth column shows the local Coulomb interaction energy. The seventh column displays the weight each line of data has in the average (the weights are identical in this simulation).
... BlockCPU AcceptRatio
6.0178991748e-03 9.8515625000e-01
5.8323097461e-03 9.8562500000e-01
5.8213412744e-03 9.8531250000e-01
5.8330412549e-03 9.8828125000e-01
5.8108362256e-03 9.8625000000e-01
5.8254170264e-03 9.8625000000e-01
5.8314813086e-03 9.8679687500e-01
5.8258469971e-03 9.8726562500e-01
5.8158433545e-03 9.8468750000e-01
5.7959401123e-03 9.8539062500e-01
The eighth column shows the CPU time (in seconds) to calculate the data in that line. The ninth column from the left contains the acceptance ratio (1 being full acceptance) for MC steps in that line’s data. Other than the block weight, all quantities vary from line to line.
Exit the text editor ([Esc] :q! [Enter] in vi, [Ctrl]-x [Ctrl]-c in emacs).
Averaging quantities in the MC data¶
QMCPACK includes the qmca Python tool to average quantities in the
scalar.dat file (and also the dmc.dat file of DMC simulations).
Without any flags, qmca will output the average of each column with a
quantity in the scalar.dat file as follows.
Execute qmca by qmca *.scalar.dat, which for this data outputs:
H series 0
LocalEnergy = -0.45446 +/- 0.00057
Variance = 0.529 +/- 0.018
Kinetic = 0.7366 +/- 0.0020
LocalPotential = -1.1910 +/- 0.0016
Coulomb = -1.1910 +/- 0.0016
LocalEnergy_sq = 0.736 +/- 0.018
BlockWeight = 12800.00000000 +/- 0.00000000
BlockCPU = 0.00582002 +/- 0.00000067
AcceptRatio = 0.985508 +/- 0.000048
Efficiency = 0.00000000 +/- 0.00000000
After one blank, qmca prints the title of the subsequent data, gleaned
from the data file name. In this case, H.s000.scalar.dat became “H
series 0.” Everything before the first “.s” will be interpreted as
the title, and the number between “.s” and the next “.” will be
interpreted as the series number.
The first column under the title is the name of each quantity qmca averaged. The column to the right of the equal signs contains the average for the quantity of that line, and the column to the right of the plus-slash-minus is the statistical error bar on the quantity. All quantities calculated from MC simulations have and must be reported with a statistical error bar!
Two new quantities not present in the scalar.dat file are computed
by qmca from the data—variance and efficiency. We will look at these
later in this lab.
To view only one value, qmca takes the -q (quantity) flag. For
example, the output of qmca -q LocalEnergy *.scalar.dat in this
directory produces a single line of output:
H series 0 LocalEnergy = -0.454460 +/- 0.000568
Type qmca –help to see the list of all quantities and their
abbreviations.
Evaluating MC simulation quality¶
There are several aspects of a MC simulation to consider in deciding how well it went. Besides the deviation of the average from an expected value (if there is one), the stability of the simulation in its sampling, the autocorrelation between MC steps, the value of the acceptance ratio (accepted steps over total proposed steps), and the variance in the local energy all indicate the quality of an MC simulation. We will look at these one by one.
Tracing MC quantities¶
Visualizing the evolution of MC quantities over the course of the simulation by a trace offers a quick picture of whether the random walk had the expected behavior. qmca plots traces with the -t flag.
Type qmca -q e -t H.s000.scalar.dat, which produces a graph of the
trace of the local energy:
The solid black line connects the values of the local energy at each MC block (labeled “samples”). The average value is marked with a horizontal, solid red line. One standard deviation above and below the average are marked with horizontal, dashed red lines.
The solid black line connects the values of the local energy at each MC block (labeled “samples”). The average value is marked with a horizontal, solid red line. One standard deviation above and below the average are marked with horizontal, dashed red lines.
The trace of this run is largely centered on the average with no large-scale oscillations or major shifts, indicating a good-quality MC run.
Try tracing the kinetic and potential energies, seeing that their behavior is comparable with the total local energy.
Change to directory problematic and type
qmca -q e -t H.s000.scalar.dat to produce this graph:
Here, the local energy samples cluster around the expected -0.5 Ha for the first 150 samples or so and then begin to oscillate more wildly and increase erratically toward 0, indicating a poor-quality MC run.
Again, trace the kinetic and potential energies in this run and see how their behavior compares with the total local energy.
Blocking away autocorrelation¶
Autocorrelation occurs when a given MC step biases subsequent MC steps, leading to samples that are not statistically independent. We must take this autocorrelation into account to obtain accurate statistics. qmca outputs autocorrelation when given the –sac flag.
Change to directory autocorrelation and type
qmca -q e --sac H.s000.scalar.dat.
H series 0 LocalEnergy = -0.454982 +/- 0.000430 1.0
The value after the error bar on the quantity is the autocorrelation (1.0 in this case).
Proposing too small a step in configuration space, the MC time step,
can lead to autocorrelation since the new samples will be in the
neighborhood of previous samples. Type grep timestep H.xml to see
the varying time step values in this QMCPACK input file (H.xml):
<parameter name="timestep">10</parameter>
<parameter name="timestep">5</parameter>
<parameter name="timestep">2</parameter>
<parameter name="timestep">1</parameter>
<parameter name="timestep">0.5</parameter>
<parameter name="timestep">0.2</parameter>
<parameter name="timestep">0.1</parameter>
<parameter name="timestep">0.05</parameter>
<parameter name="timestep">0.02</parameter>
<parameter name="timestep">0.01</parameter>
<parameter name="timestep">0.005</parameter>
<parameter name="timestep">0.002</parameter>
<parameter name="timestep">0.001</parameter>
<parameter name="timestep">0.0005</parameter>
<parameter name="timestep">0.0002</parameter>
<parameter name="timestep">0.0001</parameter>
Generally, as the time step decreases, the autocorrelation will increase
(caveat: very large time steps will also have increasing
autocorrelation). To see this, type qmca -q e --sac *.scalar.dat to
see the energies and autocorrelation times, then plot with gnuplot by
inputting gnuplot H.plt:
The error bar also increases with the autocorrelation.
Press q [Enter] to quit gnuplot.
To get around the bias of autocorrelation, we group the MC steps into
blocks, take the average of the data in the steps of each block, and
then finally average the averages in all the blocks. QMCPACK outputs the
block averages as each line in the scalar.dat file. (For DMC
simulations, in addition to the scalar.dat, QMCPACK outputs the
quantities at each step to the dmc.dat file, which permits
reblocking the data differently from the specification in the input
file.)
Change directories to blocking. Here we look at the time step of the
last dataset in the autocorrelation directory. Verify this by typing
grep timestep H.xml to see that all values are set to 0.001. Now to
see how we will vary the blocking, type grep -A1 blocks H.xml. The
parameter “steps” indicates the number of steps per block, and the
parameter “blocks” gives the number of blocks. For this comparison, the
total number of MC steps (equal to the product of “steps” and “blocks”)
is fixed at 50,000. Now check the effect of blocking on
autocorrelation—type qmca -q e --sac *scalar.dat to see the data and
gnuplot H.plt to visualize the data:
The greatest number of steps per block produces the smallest autocorrelation time. The larger number of blocks over which to average at small step-per-block number masks the corresponding increase in error bar with increasing autocorrelation.
Press q [Enter] to quit gnuplot.
Balancing autocorrelation and acceptance ratio¶
Adjusting the time step value also affects the ratio of accepted steps to
proposed steps. Stepping nearby in configuration space implies that the
probability distribution is similar and thus more likely to result in an
accepted move. Keeping the acceptance ratio high means the algorithm is
efficiently exploring configuration space and not sticking at particular
configurations. Return to the autocorrelation directory. Refresh your
memory on the time steps in this set of simulations by grep timestep
H.xml. Then, type qmca -q ar *scalar.dat to see the acceptance ratio
as it varies with decreasing time step:
H series 0 AcceptRatio = 0.047646 +/- 0.000206
H series 1 AcceptRatio = 0.125361 +/- 0.000308
H series 2 AcceptRatio = 0.328590 +/- 0.000340
H series 3 AcceptRatio = 0.535708 +/- 0.000313
H series 4 AcceptRatio = 0.732537 +/- 0.000234
H series 5 AcceptRatio = 0.903498 +/- 0.000156
H series 6 AcceptRatio = 0.961506 +/- 0.000083
H series 7 AcceptRatio = 0.985499 +/- 0.000051
H series 8 AcceptRatio = 0.996251 +/- 0.000025
H series 9 AcceptRatio = 0.998638 +/- 0.000014
H series 10 AcceptRatio = 0.999515 +/- 0.000009
H series 11 AcceptRatio = 0.999884 +/- 0.000004
H series 12 AcceptRatio = 0.999958 +/- 0.000003
H series 13 AcceptRatio = 0.999986 +/- 0.000002
H series 14 AcceptRatio = 0.999995 +/- 0.000001
H series 15 AcceptRatio = 0.999999 +/- 0.000000
By series 8 (time step = 0.02), the acceptance ratio is in excess of 99%.
Considering the increase in autocorrelation and subsequent increase in error bar as time step decreases, it is important to choose a time step that trades off appropriately between acceptance ratio and autocorrelation. In this example, a time step of 0.02 occupies a spot where the acceptance ratio is high (99.6%) and autocorrelation is not appreciably larger than the minimum value (1.4 vs. 1.0).
Considering variance¶
Besides autocorrelation, the dominant contributor to the error bar is
the variance in the local energy. The variance measures the
fluctuations around the average local energy, and, as the fluctuations
go to zero, the wavefunction reaches an exact eigenstate of the
Hamiltonian. qmca calculates this from the local energy and local energy
squared columns of the scalar.dat.
Type qmca -q v H.s009.scalar.dat to calculate the variance on the
run with time step balancing autocorrelation and acceptance ratio:
H series 9 Variance = 0.513570 +/- 0.010589
Just as the total energy does not tell us much by itself, neither does
the variance. However, comparing the ratio of the variance with the
energy indicates how the magnitude of the fluctuations compares with the
energy itself. Type qmca -q ev H.s009.scalar.dat to calculate the
energy and variance on the run side by side with the ratio:
LocalEnergy Variance ratio
H series 0 -0.454460 +/- 0.000568 0.529496 +/- 0.018445 1.1651
The very high ration of 1.1651 indicates the square of the fluctuations is on average larger than the value itself. In the next section, we will approach ways to improve the variance that subsequent labs will build on.
Reducing statistical error bars¶
Increasing MC sampling¶
Increasing the number of MC samples in a dataset reduces the error bar as the inverse of the square root of the number of samples. There are two ways to increase the number of MC samples in a simulation: (1) running more samples in parallel and (2) increasing the number of blocks (with fixed number of steps per block, this increases the total number of MC steps).
To see the effect of running more samples in parallel, change to the
directory nodes. The series here increases the number of nodes by
factors of four from 32 to 128 to 512. Type qmca -q ev *scalar.dat
and note the change in the error bar on the local energy as the number
of nodes. Visualize this with gnuplot H.plt:
Increasing the number of blocks, unlike running in parallel, increases the total CPU time of the simulation.
Press q [Enter] to quit gnuplot.
To see the effect of increasing the block number, change to the
directory blocks. To see how we will vary the number of blocks, type
grep -A1 blocks H.xml. The number of steps remains fixed, thus
increasing the total number of samples. Visualize the tradeoff by
inputting gnuplot H.plt:
Press q [Enter] to quit gnuplot.
Improving the basis sets¶
In all of the previous examples, we are using the sum of two Gaussian functions (STO-2G) to approximate what should be a simple decaying exponential for the wavefunction of the ground state of the hydrogen atom. The sum of multiple copies of a function varying each copy’s width and amplitude with coefficients is called a basis set. As we add Gaussians to the basis set, the approximation improves, the variance goes toward zero, and the energy goes to -0.5 Ha. In nearly every other case, the exact function is unknown, and we add basis functions until the total energy does not change within some threshold.
Change to the directory basis and look at the total energy and
variance as we change the wavefunction by typing qmca -q ev H_:
LocalEnergy Variance ratio
H_STO-2G series 0 -0.454460 +/- 0.000568 0.529496 +/- 0.018445 1.1651
H_STO-3G series 0 -0.465386 +/- 0.000502 0.410491 +/- 0.010051 0.8820
H_STO-6G series 0 -0.471332 +/- 0.000491 0.213919 +/- 0.012954 0.4539
H__exact series 0 -0.500000 +/- 0.000000 0.000000 +/- 0.000000 -0.0000
qmca also puts out the ratio of the variance to the local energy in a column to the right of the variance error bar. A typical high-quality value for this ratio is lower than 0.1 or so—none of these few-Gaussian wavefunctions satisfy that rule of thumb.
Use qmca to plot the trace of the local energy, kinetic energy, and potential energy of H__exact. The total energy is constantly -0.5 Ha even though the kinetic and potential energies fluctuate from configuration to configuration.
Adding a Jastrow factor¶
Another route to reducing the variance is the introduction of a Jastrow factor to account for electron-electron correlation (not the statistical autocorrelation of MC steps but the physical avoidance that electrons have of one another). To do this, we will switch to the hydrogen dimer with the exact ground state wavefunction of the atom (STO basis)—this will not be exact for the dimer. The ground state energy of the hydrogen dimer is -1.174 Ha.
Change directories to dimer and put in qmca -q ev *scalar.dat to
see the result of adding a simple, one-parameter Jastrow to the STO
basis for the hydrogen dimer at experimental bond length:
LocalEnergy Variance
H2_STO___no_jastrow series 0 -0.876548 +/- 0.005313 0.473526 +/- 0.014910
H2_STO_with_jastrow series 0 -0.912763 +/- 0.004470 0.279651 +/- 0.016405
The energy reduces by 0.044 +/- 0.006 HA and the variance by 0.19 +/- 0.02. This is still 20% above the ground state energy, and subsequent labs will cover how to improve on this with improved forms of the wavefunction that capture more of the physics.
Scaling to larger numbers of electrons¶
Calculating the efficiency¶
The inverse of the product of CPU time and the variance measures the
efficiency of an MC calculation. Use qmca to calculate efficiency by
typing qmca -q eff *scalar.dat to see the efficiency of these two
H \(_2\) calculations:
H2_STO___no_jastrow series 0 Efficiency = 16698.725453 +/- 0.000000
H2_STO_with_jastrow series 0 Efficiency = 52912.365609 +/- 0.000000
The Jastrow factor increased the efficiency in these calculations by a factor of three, largely through the reduction in variance (check the average block CPU time to verify this claim).
Scaling up¶
To see how MC scales with increasing particle number, change directories
to size. Here are the data from runs of increasing numbers of
electrons for H, H\(_2\), C, CH\(_4\), C\(_2\),
C\(_2\)H\(_4\), (CH\(_4\))\(_2\), and
(C\(_2\)H\(_4\))\(_2\) using the STO-6G basis set
for the orbitals of the Slater determinant. The file names begin with
the number of electrons simulated for those data.
Use qmca -q bc *scalar.dat to see that the CPU time per block
increases with the number of electrons in the simulation; then plot the
total CPU time of the simulation by gnuplot Nelectron_tCPU.plt:
The green pluses represent the CPU time per block at each electron number. The red line is a quadratic fit to those data. For a fixed basis set size, we expect the time to scale quadratically up to 1,000s of electrons, at which point a cubic scaling term may become dominant. Knowing the scaling allows you to roughly project the calculation time for a larger number of electrons.
Press q [Enter] to quit gnuplot.
This is not the whole story, however. The variance of the energy also
increases with a fixed basis set as the number of particles increases at
a faster rate than the energy decreases. To see this, type
qmca -q ev *scalar.dat:
LocalEnergy Variance
01________H series 0 -0.471352 +/- 0.000493 0.213020 +/- 0.012950
02_______H2 series 0 -0.898875 +/- 0.000998 0.545717 +/- 0.009980
06________C series 0 -37.608586 +/- 0.020453 184.322000 +/- 45.481193
10______CH4 series 0 -38.821513 +/- 0.022740 169.797871 +/- 24.765674
12_______C2 series 0 -72.302390 +/- 0.037691 491.416711 +/- 106.090103
16_____C2H4 series 0 -75.488701 +/- 0.042919 404.218115 +/- 60.196642
18___CH4CH4 series 0 -58.459857 +/- 0.039309 498.579645 +/- 92.480126
32_C2H4C2H4 series 0 -91.567283 +/- 0.048392 632.114026 +/- 69.637760
The increase in variance is not uniform, but the general trend is upward with a fixed wavefunction form and basis set. Subsequent labs will address how to improve the wavefunction to keep the variance manageable.
Lab 2: QMC Basics¶
Topics covered in this lab¶
This lab focuses on the basics of performing quality QMC calculations. As an example, participants test an oxygen pseudopotential within DMC by calculating atomic and dimer properties, a common step prior to production runs. Topics covered include:
Converting pseudopotentials into QMCPACK’s FSATOM format
Generating orbitals with QE
Converting orbitals into QMCPACK’s ESHDF format with pw2qmcpack
Optimizing Jastrow factors with QMCPACK
Removing DMC time step errors via extrapolation
Automating QMC workflows with Nexus
Testing pseudopotentials for accuracy
Lab outline¶
Download and conversion of oxygen atom pseudopotential
DMC time step study of the neutral oxygen atom
DFT orbital generation with QE
Orbital conversion with
Optimization of Jastrow correlation factor with QMCPACK
DMC run with multiple time steps
DMC time step study of the first ionization potential of oxygen
Repetition of a-d above for ionized oxygen atom
Automated DMC calculations of the oxygen dimer binding curve
Lab directories and files¶
%
labs/lab2_qmc_basics/
│
├── oxygen_atom - oxygen atom calculations
│ ├── O.q0.dft.in - Quantum ESPRESSO input for DFT run
│ ├── O.q0.p2q.in - pw2qmcpack.x input for orbital conversion run
│ ├── O.q0.opt.in.xml - QMCPACK input for Jastrow optimization run
│ ├── O.q0.dmc.in.xml - QMCPACK input file for neutral O DMC
│ ├── ip_conv.py - tool to fit oxygen IP vs timestep
│ └── reference - directory w/ completed runs
│
├── oxygen_dimer - oxygen dimer calculations
│ ├── dimer_fit.py - tool to fit dimer binding curve
│ ├── O_dimer.py - automation script for dimer calculations
│ ├── pseudopotentials - directory for pseudopotentials
│ └── reference - directory w/ completed runs
│
└── your_system - performing calculations for an arbitrary system (yours)
├── example.py - example nexus file for periodic diamond
├── pseudopotentials - directory containing C pseudopotentials
└── reference - directory w/ completed runs
Obtaining and converting a pseudopotential for oxygen¶
First enter the oxygen_atom directory:
cd labs/lab2_qmc_basics/oxygen_atom/
Throughout the rest of the lab, locations are specified with respect to labs/lab2_qmc_basics (e.g., oxygen_atom).
We use a potential from the Burkatzki-Filippi-Dolg pseudopotential database.
Although the full database is available in QMCPACK distribution (trunk/pseudopotentials/BFD/),
we use a BFD pseudopotential to illustrate the process of converting and testing an
external potential for use with QMCPACK. To obtain the pseudopotential, go to
http://www.burkatzki.com/pseudos/index.2.html
and click on the “Select Pseudopotential” button. Next click on oxygen in the
periodic table. Click on the empty circle next to “V5Z” (a large Gaussian
basis set) and click on “Next.” Select the Gamess format and click on
“Retrive Potential.” Helpful information about the pseudopotential will be
displayed. The desired portion is at the bottom (the last 7 lines). Copy
this text into the editor of your choice (e.g., emacs or vi)
and save it as O.BFD.gamess
(be sure to include a new line at the end of the file). To transform the
pseudopotential into the FSATOM XML format used by QMCPACK, use the ppconvert
tool:
ppconvert --gamess_pot O.BFD.gamess --s_ref "1s(2)2p(4)" \
--p_ref "1s(2)2p(4)" --d_ref "1s(2)2p(4)" --xml O.BFD.xml
Observe the notation used to describe the reference valence configuration for this helium-core PP: 1s(2)2p(4). The ppconvert tool uses the following convention for the valence states: the first $s$ state is labeled 1s (1s, 2s, 3s, \(\ldots\)), the first \(p\) state is labeled 2p (2p, 3p, \(\ldots\)), and the first \(d\) state is labeled 3d (3d, 4d, \(\ldots\)). Copy the resulting xml file into the oxygen_atom directory.
Note: The command to convert the PP into QE’s UPF format is similar (both formats are required):
ppconvert --gamess_pot O.BFD.gamess --s_ref "1s(2)2p(4)" \
--p_ref "1s(2)2p(4)" --d_ref "1s(2)2p(4)" --log_grid --upf O.BFD.upf
For reference, the text of O.BFD.gamess should be:
O-QMC GEN 2 1
3
6.00000000 1 9.29793903
55.78763416 3 8.86492204
-38.81978498 2 8.62925665
1
38.41914135 2 8.71924452
The full QMCPACK pseudopotential is also included in oxygen_atom/reference/O.BFD.*.
DFT with QE to obtain the orbital part of the wavefunction¶
With the pseudopotential in hand, the next step toward a QMC calculation is to obtain the
Fermionic part of the wavefunction, in this case a single Slater determinant constructed
from DFT-LDA orbitals for a neutral oxygen atom. If you had trouble with the pseudopotential conversion
step, preconverted pseudopotential files are located in the oxygen_atom/reference directory.
QE input for the DFT-LDA ground state of the neutral oxygen atom can be found in O.q0.dft.in
and also in Listing 58. Setting wf_collect=.true. instructs QE to write the
orbitals to disk at the end of the run. Option wf_collect=.true. could be a potential problem
in large simulations; therefore, we recommend avoiding it and using the converter pw2qmcpack in parallel
(see details in pw2qmcpack.x).
Note that the plane-wave energy cutoff has been set to a reasonable value of 300 Ry here (ecutwfc=300).
This value depends on the pseudopotentials used, and, in general,
should be selected by running DFT \(\rightarrow\) (orbital conversion) \(\rightarrow\) VMC with
increasing energy cutoffs until the lowest VMC total energy and variance is reached.
O.q0.dft.in)¶&CONTROL
calculation = 'scf'
restart_mode = 'from_scratch'
prefix = 'O.q0'
outdir = './'
pseudo_dir = './'
disk_io = 'low'
wf_collect = .true.
/
&SYSTEM
celldm(1) = 1.0
ibrav = 0
nat = 1
ntyp = 1
nspin = 2
tot_charge = 0
tot_magnetization = 2
input_dft = 'lda'
ecutwfc = 300
ecutrho = 1200
nosym = .true.
occupations = 'smearing'
smearing = 'fermi-dirac'
degauss = 0.0001
/
&ELECTRONS
diagonalization = 'david'
mixing_mode = 'plain'
mixing_beta = 0.7
conv_thr = 1e-08
electron_maxstep = 1000
/
ATOMIC_SPECIES
O 15.999 O.BFD.upf
ATOMIC_POSITIONS alat
O 9.44863067 9.44863161 9.44863255
K_POINTS automatic
1 1 1 0 0 0
CELL_PARAMETERS cubic
18.89726133 0.00000000 0.00000000
0.00000000 18.89726133 0.00000000
0.00000000 0.00000000 18.89726133
Run QE by typing
mpirun -np 4 pw.x -input O.q0.dft.in >&O.q0.dft.out&
The DFT run should take a few minutes to complete. If desired, you can track the progress of the DFT run by typing “tail -f O.q0.dft.out.” Once finished, you should check the LDA total energy in O.q0.dft.out by typing “grep '! ' O.q0.dft.out.” The result should be close to
! total energy = -31.57553905 Ry
The orbitals have been written in a format native to QE in the O.q0.save directory. We will convert them into the ESHDF format expected by QMCPACK by using the pw2qmcpack.x tool. The input for pw2qmcpack.x can be found in the file O.q0.p2q.in and also in Listing 59.
pw2qmcpack.x input file for orbital conversion (O.q0.p2q.in)¶&inputpp
prefix = 'O.q0'
outdir = './'
write_psir = .false.
/
Perform the orbital conversion now by typing the following:
mpirun -np 1 pw2qmcpack.x<O.q0.p2q.in>&O.q0.p2q.out&
Upon completion of the run, a new file should be present containing the orbitals for QMCPACK: O.q0.pwscf.h5. Template XML files for particle (O.q0.ptcl.xml) and wavefunction (O.q0.wfs.xml) inputs to QMCPACK should also be present.
DMC timestep extrapolation I: neutral oxygen atom¶
The DMC algorithm contains two biases in addition to the fixed node and pseudopotential approximations that are important to control: time step and population control bias. In this section we focus on estimating and removing time step bias from DMC calculations. The essential fact to remember is that the bias vanishes as the time step goes to zero, while the needed computer time increases inversely with the time step.
In the same directory you used to perform wavefunction optimization (oxygen_atom) you will find a sample DMC input file for the neutral oxygen atom named O.q0.dmc.in.xml. Open this file in a text editor and note the differences from the optimization case. Wavefunction information is no longer included from pw2qmcpack but instead should come from the optimization run:
<!-- OPT_XML is from optimization, e.g. O.q0.opt.s008.opt.xml -->
<include href="OPT_XML"/>
Replace “OPT_XML” with the opt.xml file corresponding to the best Jastrow parameters you found in the last section (this is a file name similar to O.q0.opt.s008.opt.xml).
The QMC calculation section at the bottom is also different. The linear optimization blocks have been replaced with XML describing a VMC run followed by DMC. Descriptions of the input keywords follow.
- timestep
Time step of the VMC/DMC random walk. In VMC choose a time step corresponding to an acceptance ratio of about 50%. In DMC the acceptance ratio is often above 99%.
- warmupSteps
Number of MC steps discarded as a warmup or equilibration period of the random walk.
- steps
Number of MC steps per block. Physical quantities, such as the total energy, are averaged over walkers and steps.
- blocks
Number of blocks. This is also the number of average energy values written to output files. The number should be greater than 200 for meaningful statistical analysis of output data (e.g., via
qmca). The total number of MC steps each walker takes isblocks\(\times\)steps.- samples
VMC only. This is the number of walkers used in subsequent DMC runs. Each DMC walker is initialized with electron positions sampled from the VMC random walk.
- nonlocalmoves
DMC only. If yes/no, use the locality approximation/T-moves for nonlocal pseudopotentials. T-moves generally improve the stability of the algorithm and restore the variational principle for small systems (T-moves version 1).
The purpose of the VMC run is to provide initial electron positions for each DMC walker. Setting \(\texttt{walkers}=1\) in the VMC block ensures there will be only one VMC walker per execution thread. There will be a total of 4 VMC walkers in this case (see O.q0.dmc.qsub.in). We want the electron positions used to initialize the DMC walkers to be decorrelated from one another. A VMC walker will often decorrelate from its current position after propagating for a few Ha \(^{-1}\) in imaginary time (in general, this is system dependent). This leads to a rough rule of thumb for choosing blocks and steps for the VMC run (\(\texttt{vwalkers}=4\) here):
Fill in the VMC XML block with appropriate values for these parameters. There should be more than one DMC walker per thread and enough walkers in total to avoid population control bias. The general rule of thumb is to have more than \(\sim 2,000\) walkers, although the dependence of the total energy on population size should be explicitly checked from time to time.
To study time step bias, we will perform a sequence of DMC runs over a range of time steps (\(0.1\) Ha\(^{-1}\) is too large, and time steps below \(0.002\) Ha\(^{-1}\) are probably too small). A common approach is to select a fairly large time step to begin with and then decrease the time step by a factor of two in each subsequent DMC run. The total amount of imaginary time the walker population propagates should be the same for each run. A simple way to accomplish this is to choose input parameters in the following way
Each DMC run will require about twice as much computer time as the one preceding it. Note that the number of blocks is kept fixed for uniform statistical analysis. \(\texttt{blocks}\times\texttt{steps}\times\texttt{timestep}\sim 60~\mathrm{Ha}^{-1}\) is sufficient for this system.
Choose an initial DMC time step and create a sequence of \(N\) time steps according to (75). Make \(N\) copies of the DMC XML block in the input file.
<qmc method="dmc" move="pbyp">
<parameter name="warmupSteps" > DWARMUP </parameter>
<parameter name="blocks" > DBLOCKS </parameter>
<parameter name="steps" > DSTEPS </parameter>
<parameter name="timestep" > DTIMESTEP </parameter>
<parameter name="nonlocalmoves" > yes </parameter>
</qmc>
Fill in DWARMUP, DBLOCKS, DSTEPS, and DTIMESTEP for each DMC run according to (75). Start the DMC time step extrapolation run by typing:
mpirun -np 4 qmcpack O.q0.dmc.in.xml >&O.q0.dmc.out&
The run should take only a few minutes to complete.
QMCPACK will create files prefixed with O.q0.dmc. The log file is O.q0.dmc.out. As before, block-averaged data is written to scalar.dat files. In addition, DMC runs produce dmc.dat files, which contain energy data averaged only over the walker population (one line per DMC step). The dmc.dat files also provide a record of the walker population at each step.
Linear fit to DMC timestep data from PlotTstepConv.pl.¶
Use the PlotTstepConv.pl to obtain a linear fit to the time step data (type “PlotTstepConv.pl O.q0.dmc.in.xml 40”). You should see a plot similar to Fig. 16. The tail end of the text output displays the parameters for the linear fit. The “a” parameter is the total energy extrapolated to zero time step in Hartree units.
...
Final set of parameters Asymptotic Standard Error
======================= ==========================
a = -15.8925 +/- 0.0007442 (0.004683%)
b = -0.0457479 +/- 0.0422 (92.24%)
...
Questions and Exercises
What is the \(\tau\rightarrow 0\) extrapolated value for the total energy?
What is the maximum time step you should use if you want to calculate the total energy to an accuracy of \(0.05\) eV? For convenience, \(1~\textrm{Ha}=27.2113846~\textrm{eV}\).
What is the acceptance ratio for this (bias \(<0.05\) eV) run? Does it follow the rule of thumb for sensible DMC (acceptance ratio \(>99\)%) ?
Check the fluctuations in the walker population (
qmca -t -q nw O.q0.dmc*dmc.dat –noac). Does the population seem to be stable?(Optional) Study population control bias for the oxygen atom. Select a few population sizes. Copy
O.q0.dmc.in.xmlto a new file and remove all but one DMC run (select a single time step). Make one copy of the new file for each population, set “textttsamples,” and choose a uniqueidin<project/>. Useqmcato study the dependence of the DMC total energy on the walker population. How large is the bias compared with time step error? What bias is incurred by following the “rule of thumb” of a couple thousand walkers? Will population control bias generally be an issue for production runs on modern parallel machines?
DMC time step extrapolation II: oxygen atom ionization potential¶
In this section, we will repeat the calculations of the previous two sections (optimization, time step extrapolation) for the \(+1\) charge state of the oxygen atom. Comparing the resulting first ionization potential (IP) with experimental data will complete our first test of the BFD oxygen pseudopotential. In actual practice, higher IPs could also be tested before performing production runs.
Obtaining the time step extrapolated DMC total energy for ionized oxygen should take much less (human) time than for the neutral case. For convenience, the necessary steps are summarized as follows.
Obtain DFT orbitals with QE.
Copy the DFT input (
O.q0.dft.in) toO.q1.dft.inEdit
O.q1.dft.into match the +1 charge state of the oxygen atom.... prefix = 'O.q1' ... tot_charge = 1 tot_magnetization = 3 ...
Perform the DFT run:
mpirun -np 4 pw.x -input O.q1.dft.in >&O.q1.dft.out&
Convert the orbitals to ESHDF format.
Copy the pw2qmcpack input (
O.q0.p2q.in) toO.q1.p2q.inEdit
O.q1.p2q.into match the file prefix used in DFT.... prefix = 'O.q1' ...
Perform the orbital conversion run:
mpirun -np 1 pw2qmcpack.x<O.q1.p2q.in>&O.q1.p2q.out&
Optimize the Jastrow factor with QMCPACK.
Copy the optimization input (
O.q0.opt.in.xml) toO.q1.opt.in.xmlEdit
O.q1.opt.in.xmlto match the file prefix used in DFT.... <project id="O.q1.opt" series="0"> ... <include href="O.q1.ptcl.xml"/> ... <include href="O.q1.wfs.xml"/> ...
Edit the particle XML file (
O.q1.ptcl.xml) to have open boundary conditions.<parameter name="bconds"> n n n </parameter>
Add cutoffs to the Jastrow factors in the wavefunction XML file (
O.q1.wfs.xml)... <correlation speciesA="u" speciesB="u" size="8" rcut="10.0"> ... <correlation speciesA="u" speciesB="d" size="8" rcut="10.0"> ... <correlation elementType="O" size="8" rcut="5.0"> ...
Perform the Jastrow optimization run:
mpirun -np 4 qmcpack O.q1.opt.in.xml >&O.q1.opt.out&Identify the optimal set of parameters with
qmca([your opt.xml]).
DMC time step study with QMCPACK
Copy the DMC input (
O.q0.dmc.in.xml) toO.q1.dmc.in.xmlEdit
O.q1.dmc.in.xmlto use the DFT prefix and the optimal Jastrow.... <project id="O.q1.dmc" series="0"> ... <include href="O.q1.ptcl.xml"/> ... <include href="[your opt.xml]"/> ...
Perform the DMC run:
mpirun -np 4 qmcpack O.q1.dmc.in.xml >&O.q1.dmc.out&Obtain the DMC total energy extrapolated to zero time step with
PlotTstepConv.pl.
The aforementioned process, which excludes additional steps for orbital generation and conversion, can become tedious to perform by hand in production settings where many calculations are often required. For this reason, automation tools are introduced for calculations involving the oxygen dimer in Automated binding curve of the oxygen dimer of the lab.
Questions and Exercises
What is the \(\tau\rightarrow 0\) extrapolated DMC value for the first ionization potential of oxygen?
How does the extrapolated value compare with the experimental IP? Go to http://physics.nist.gov/PhysRefData/ASD/ionEnergy.html and enter “
O I” in the box labeled “Spectra” and click on the “Retrieve Data” button.What can we conclude about the accuracy of the pseudopotential? What factors complicate this assessment?
Explore the sensitivity of the IP to the choice of time step. Type
./ip_conv.pyto view three time step extrapolation plots: two for the \(q=0,\) one for total energies, and one for the IP. Is the IP more, less, or similarly sensitive to time step than the total energy?What is the maximum time step you should use if you want to calculate the ionization potential to an accuracy of \(0.05\) eV? What factor of CPU time is saved by assessing time step convergence on the IP (a total energy difference) vs. a single total energy?
Are the acceptance ratio and population fluctuations reasonable for the \(q=1\) calculations?
DMC workflow automation with Nexus¶
Production QMC projects are often composed of many similar workflows. The simplest of these is a single DMC calculation involving four different compute jobs:
Orbital generation via QE or GAMESS.
Conversion of orbital data via
pw2qmcpack.xorconvert4qmc.Optimization of Jastrow factors via QMCPACK.
DMC calculation via QMCPACK.
Simulation workflows quickly become more complex with increasing costs in terms of human time for the researcher. Automation tools can decrease both human time and error if used well.
The set of automation tools we will be using is known as Nexus [Kro16], which is distributed with QMCPACK. Nexus is capable of generating input files, submitting and monitoring compute jobs, passing data between simulations (relaxed structures, orbital files, optimized Jastrow parameters, etc.), and data analysis. The user interface to Nexus is through a set of functions defined in the Python programming language. User scripts that execute simple workflows resemble input files and do not require programming experience. More complex workflows require only basic programming constructs (e.g. for loops and if statements). Nexus input files/scripts should be easier to navigate than QMCPACK input files and more efficient than submitting all the jobs by hand.
Nexus is driven by simple user-defined scripts that resemble keyword-driven input files. An example Nexus input file that performs a single VMC calculation (with pregenerated orbitals) follows. Take a moment to read it over and especially note the comments (prefixed with “\#”) explaining most of the contents. If the input syntax is unclear you may want to consult portions of Appendix A: Basic Python constructs, which gives a condensed summary of Python constructs. An additional example and details about the inner workings of Nexus can be found in the reference publication [Kro16].
#! /usr/bin/env python3
# import Nexus functions
from nexus import settings,job,get_machine,run_project
from nexus import generate_physical_system
from nexus import generate_qmcpack,vmc
settings( # Nexus settings
pseudo_dir = './pseudopotentials', # location of PP files
runs = '', # root directory for simulations
results = '', # root directory for simulation results
status_only = 0, # show simulation status, then exit
generate_only = 0, # generate input files, then exit
sleep = 3, # seconds between checks on sim. progress
machine = 'ws4', # workstation with 4 cores
)
qmcjob = job( # specify job parameters
cores = 4, # use 4 MPI tasks
threads = 1, # 1 OpenMP thread per node
app = 'qmcpack' # use QMCPACK executable (assumed in PATH)
)
qmc_calcs = [ # list QMC calculation methods
vmc( # VMC
walkers = 1, # 1 walker
warmupsteps = 50, # 50 MC steps for warmup
blocks = 200, # 200 blocks
steps = 10, # 10 steps per block
timestep = .4 # 0.4 1/Ha timestep
)]
dimer = generate_physical_system( # make a dimer system
type = 'dimer', # system type is dimer
dimer = ('O','O'), # dimer is two oxygen atoms
separation = 1.2074, # separated by 1.2074 Angstrom
Lbox = 15.0, # simulation box is 15 Angstrom
units = 'A', # Angstrom is dist. unit
net_spin = 2, # nup-ndown is 2
O = 6 # pseudo-oxygen has 6 valence el.
)
qmc = generate_qmcpack( # make a qmcpack simulation
identifier = 'example', # prefix files with 'example'
path = 'scale_1.0', # run in ./scale_1.0 directory
system = dimer, # run the dimer system
job = qmcjob, # set job parameters
input_type = 'basic', # basic qmcpack inputs given below
pseudos = ['O.BFD.xml'], # list of PP's to use
orbitals_h5 = 'O2.pwscf.h5', # file with orbitals from DFT
bconds = 'nnn', # open boundary conditions
jastrows = [], # no jastrow factors
calculations = qmc_calcs # QMC calculations to perform
)
run_project(qmc) # write input file and submit job
Automated binding curve of the oxygen dimer¶
In this section we will use Nexus to calculate the DMC total energy of the oxygen dimer over a series of bond lengths. The equilibrium bond length and binding energy of the dimer will be determined by performing a polynomial fit to the data (Morse potential fits should be preferred in production tests). Comparing these values with corresponding experimental data provides a second test of the BFD pseudopotential for oxygen.
Enter the oxygen_dimer directory. Copy your BFD pseudopotential from the atom runs into oxygen_dimer/pseudopotentials (be sure to move both files: .upf and .xml). Open O_dimer.py with a text editor. The overall format is similar to the example file shown in the last section. The main difference is that a full workflow of runs (DFT orbital generation, orbital conversion, optimization and DMC) are being performed rather than a single VMC run.
As in the example in the last section, the oxygen dimer is generated with the generate_physical_
system function:
dimer = generate_physical_system(
type = 'dimer',
dimer = ('O','O'),
separation = 1.2074*scale,
Lbox = 10.0,
units = 'A',
net_spin = 2,
O = 6
)
Similar syntax can be used to generate crystal structures or to specify systems with arbitrary atomic configurations and simulation cells. Notice that a “scale” variable has been introduced to stretch or compress the dimer.
Next, objects representing a QE (PWSCF) run and subsequent orbital conversion step are constructed with respective generate_* functions:
dft = generate_pwscf(
identifier = 'dft',
...
input_dft = 'lda',
...
)
sims.append(dft)
# describe orbital conversion run
p2q = generate_pw2qmcpack(
identifier = 'p2q',
...
dependencies = (dft,'orbitals'),
)
sims.append(p2q)
Note the dependencies keyword. This keyword is used to construct workflows out of otherwise separate runs. In this case, the dependency indicates that the orbital conversion run must wait for the DFT to finish before starting.
Objects representing QMCPACK simulations are then constructed with the generate_qmcpack function:
opt = generate_qmcpack(
identifier = 'opt',
...
jastrows = [('J1','bspline',8,5.0),
('J2','bspline',8,10.0)],
calculations = [
loop(max=12,
qmc=linear(
energy = 0.0,
unreweightedvariance = 1.0,
reweightedvariance = 0.0,
timestep = 0.3,
samples = 61440,
warmupsteps = 50,
blocks = 200,
substeps = 1,
nonlocalpp = True,
usebuffer = True,
walkers = 1,
minwalkers = 0.5,
maxweight = 1e9,
usedrift = False,
minmethod = 'quartic',
beta = 0.025,
exp0 = -16,
bigchange = 15.0,
alloweddifference = 1e-4,
stepsize = 0.2,
stabilizerscale = 1.0,
nstabilizers = 3,
)
)
],
dependencies = (p2q,'orbitals'),
)
sims.append(opt)
qmc = generate_qmcpack(
identifier = 'qmc',
...
jastrows = [],
calculations = [
vmc(
walkers = 1,
warmupsteps = 30,
blocks = 20,
steps = 10,
substeps = 2,
timestep = .4,
samples = 2048
),
dmc(
warmupsteps = 100,
blocks = 400,
steps = 32,
timestep = 0.01,
nonlocalmoves = True,
)
],
dependencies = [(p2q,'orbitals'),(opt,'jastrow')],
)
sims.append(qmc)
Shared details such as the run directory, job, pseudopotentials, and orbital file have been omitted (...). The “opt” run will optimize a 1-body B-spline Jastrow with 8 knots having a cutoff of 5.0 Bohr and a B-spline Jastrow (for up-up and up-down correlations) with 8 knots and cutoffs of 10.0 Bohr. The Jastrow list for the DMC run is empty, and the previous use of dependencies indicates that the DMC run depends on the optimization run for the Jastrow factor. Nexus will submit the “opt” run first, and upon completion it will scan the output, select the optimal set of parameters, pass the Jastrow information to the “qmc” run, and then submit the DMC job. Independent job workflows are submitted in parallel when permitted. No input files are written or job submissions made until the “run_project” function is reached:
run_project(sims)
All of the simulation objects have been collected into a list (sims) for submission.
As written, O_dimer.py will perform calculations only at the equilibrium separation distance of 1.2074 {AA} since the list of scaling factors (representing stretching or compressing the dimer) contains only one value (scales = [1.00]). Modify the file now to perform DMC calculations across a range of separation distances with each DMC run using the Jastrow factor optimized at the equilibrium separation distance. Specifically, you will want to change the list of scaling factors to include both compression (scale<1.0) and stretch (scale>1.0):
scales = [1.00,0.90,0.95,1.05,1.10]
Note that “1.00” is left in front because we are going to optimize the Jastrow factor first at the equilibrium separation and reuse this Jastrow factor for all other separation distances. This procedure is used because it can reduce variations in localization errors (due to pseudopotentials in DMC) along the binding curve.
Change the status_only parameter in the “settings” function to 1 and type “./O_dimer.py” at the command line. This will print the status of all simulations:
Project starting
checking for file collisions
loading cascade images
cascade 0 checking in
cascade 10 checking in
cascade 4 checking in
cascade 13 checking in
cascade 7 checking in
checking cascade dependencies
all simulation dependencies satisfied
cascade status
setup, sent_files, submitted, finished, got_output, analyzed
000000 dft ./scale_1.0
000000 p2q ./scale_1.0
000000 opt ./scale_1.0
000000 qmc ./scale_1.0
000000 dft ./scale_0.9
000000 p2q ./scale_0.9
000000 qmc ./scale_0.9
000000 dft ./scale_0.95
000000 p2q ./scale_0.95
000000 qmc ./scale_0.95
000000 dft ./scale_1.05
000000 p2q ./scale_1.05
000000 qmc ./scale_1.05
000000 dft ./scale_1.1
000000 p2q ./scale_1.1
000000 qmc ./scale_1.1
setup, sent_files, submitted, finished, got_output, analyzed
In this case, five simulation “cascades” (workflows) have been identified, each one starting and ending with “dft” and “qmc” runs, respectively. The six status flags setup, sent_files, submitted, finished, got_output, analyzed) each shows 0, indicating that no work has been done yet.
Now change “status_only” back to 0, set “generate_only” to 1, and run O_dimer.py again. This will perform a dry run of all simulations. The dry run should finish in about 20 seconds:
Project starting
checking for file collisions
loading cascade images
cascade 0 checking in
cascade 10 checking in
cascade 4 checking in
cascade 13 checking in
cascade 7 checking in
checking cascade dependencies
all simulation dependencies satisfied
starting runs:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
poll 0 memory 91.03 MB
Entering ./scale_1.0 0
writing input files 0 dft
Entering ./scale_1.0 0
sending required files 0 dft
submitting job 0 dft
...
poll 1 memory 91.10 MB
...
Entering ./scale_1.0 0
Would have executed:
export OMP_NUM_THREADS=1
mpirun -np 4 pw.x -input dft.in
poll 2 memory 91.10 MB
Entering ./scale_1.0 0
copying results 0 dft
Entering ./scale_1.0 0
analyzing 0 dft
...
poll 3 memory 91.10 MB
Entering ./scale_1.0 1
writing input files 1 p2q
Entering ./scale_1.0 1
sending required files 1 p2q
submitting job 1 p2q
...
Entering ./scale_1.0 1
Would have executed:
export OMP_NUM_THREADS=1
mpirun -np 1 pw2qmcpack.x<p2q.in
poll 4 memory 91.10 MB
Entering ./scale_1.0 1
copying results 1 p2q
Entering ./scale_1.0 1
analyzing 1 p2q
...
poll 5 memory 91.10 MB
Entering ./scale_1.0 2
writing input files 2 opt
Entering ./scale_1.0 2
sending required files 2 opt
submitting job 2 opt
...
Entering ./scale_1.0 2
Would have executed:
export OMP_NUM_THREADS=1
mpirun -np 4 qmcpack opt.in.xml
poll 6 memory 91.16 MB
Entering ./scale_1.0 2
copying results 2 opt
Entering ./scale_1.0 2
analyzing 2 opt
...
poll 7 memory 93.00 MB
Entering ./scale_1.0 3
writing input files 3 qmc
Entering ./scale_1.0 3
sending required files 3 qmc
submitting job 3 qmc
...
Entering ./scale_1.0 3
Would have executed:
export OMP_NUM_THREADS=1
mpirun -np 4 qmcpack qmc.in.xml
...
poll 17 memory 93.06 MB
Project finished
Nexus polls the simulation status every 3 seconds and sleeps in between.
The “scale_” directories should now contain several files:
scale_1.0
dft.in
O.BFD.upf
O.BFD.xml
opt.in.xml
p2q.in
pwscf_output
qmc.in.xml
sim_dft/
analyzer.p
input.p
sim.p
sim_opt/
analyzer.p
input.p
sim.p
sim_p2q/
analyzer.p
input.p
sim.p
sim_qmc/
analyzer.p
input.p
sim.p
Take a minute to inspect the generated input (dft.in, p2q.in, opt.in.xml, qmc.in.xml). The pseudopotential files (O.BFD.upf and O.BFD.xml) have been copied into each local directory. Four additional directories have been created: sim_dft, sim_p2q, sim_opt and sim_qmc. The sim.p files in each directory contain the current status of each simulation. If you run O_dimer.py again, it should not attempt to rerun any of the simulations:
Project starting
checking for file collisions
loading cascade images
cascade 0 checking in
cascade 10 checking in
cascade 4 checking in
cascade 13 checking in
cascade 7 checking in
checking cascade dependencies
all simulation dependencies satisfied
starting runs:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
poll 0 memory 64.25 MB
Project finished
This way you can continue to add to the O_dimer.py file (e.g., adding more separation distances) without worrying about duplicate job submissions.
Now submit the jobs in the dimer workflow. Reset the state of the simulations by removing the sim.p files (“rm ./scale*/sim*/sim.p”), set “generate_only” to 0, and rerun O_dimer.py. It should take about 20 minutes for all the jobs to complete. You may wish to open another terminal to monitor the progress of the individual jobs while the current terminal runs O_dimer.py in the foreground. You can begin the following first exercise once the optimization job completes.
Questions and Exercises
Evaluate the quality of the optimization at
scale=1.0using theqmcatool. Did the optimization succeed? How does the variance compare with the neutral oxygen atom? Is the wavefunction of similar quality to the atomic case?Evaluate the traces of the local energy and the DMC walker population for each separation distance with the
qmcatool. Are there any anomalies in the runs? Is the acceptance ratio reasonable? Is the wavefunction of similar quality across all separation distances?Use the
dimer_fit.pytool located inoxygen_dimerto fit the oxygen dimer binding curve. To get the binding energy of the dimer, we will need the DMC energy of the atom. Before performing the fit, answer: What DMC time step should be used for the oxygen atom results? The tool accepts three arguments (./dimer_fit.py P N E Eerr),Pis the prefix of the DMC input files (should be “qmc” at this point),Nis the order of the fit (use 2 to start),``E`` andEerrare your DMC total energy and error bar, respectively, for the oxygen atom (in electron volts). A plot of the dimer data will be displayed, and text output will show the DMC equilibrium bond length and binding energy as well as experimental values. How accurately does your fit to the DMC data reproduce the experimental values? What factors affect the accuracy of your results?Refit your data with a fourth-order polynomial. How do your predictions change with a fourth-order fit? Is a fourth-order fit appropriate for the available data?
Add new “
scale” values to the list inO_dimer.pythat interpolate between the original set (e.g., expand to ). Perform the DMC calculations and redo the fits. How accurately does your fit to the DMC data reproduce the experimental values? Should this pseudopotential be used in production calculations?(Optional) Perform optimization runs at the extreme separation distances corresponding to
scale=[0.90,1.10]. Are the individually optimized wavefunctions of significantly better quality than the one imported fromscale=1.00? Why? What form of Jastrow factor might give an even better improvement?
(Optional) Running your system with QMCPACK¶
This section covers a fairly simple route to get started on QMC calculations of an arbitrary system of interest using the Nexus workflow management system to set up input files and optionally perform the runs. The example provided in this section uses QE (PWSCF) to generate the orbitals forming the Slater determinant part of the trial wavefunction. PWSCF is a natural choice for solid-state systems, and it can be used for surface/slab and molecular systems as well, albeit at the price of describing additional vacuum space with plane waves.
To start out, you will need PPs for each element in your system in both
the UPF (PWSCF) and FSATOM/XML (QMCPACK) formats. A good place to start
is the BFD pseudopotential database
(http://www.burkatzki.com/pseudos/index.2.html), which we have already
used in our study of the oxygen atom. The database does not contain PPs
for the fourth and fifth row transition metals or any of the lanthanides
or actinides. If you need a PP that is not in the BFD database, you may
need to generate and test one manually (e.g., with OPIUM,
http://opium.sourceforge.net/). Otherwise, use ppconvert as outlined in
Obtaining and converting a pseudopotential for oxygen to obtain PPs in the formats used by PWSCF
and QMCPACK. Enter the your_system lab directory and place the converted PPs in your_system/pseudopotentials.
Before performing production calculations (more than just the initial setup in this section), be sure to converge the plane-wave energy cutoff in PWSCF as these PPs can be rather hard, sometimes requiring cutoffs in excess of 300 Ry. Depending on the system under study, the amount of memory required to represent the orbitals (QMCPACK uses 3D B-splines) can become prohibitive, forcing you to search for softer PPs.
Beyond PPs, all that is required to get started are the atomic positions and the dimensions/shape of the simulation cell. The Nexus file example.py illustrates how to set up PWSCF and QMCPACK input files by providing minimal information regarding the physical system (an 8-atom cubic cell of diamond in the example). Most of the contents should be familiar from your experience with the automated calculations of the oxygen dimer binding curve in Automated binding curve of the oxygen dimer (if you have skipped ahead you may want to skim that section for relevant information). The most important change is the expanded description of the physical system:
# details of your physical system (diamond conventional cell below)
my_project_name = 'diamond_vmc' # directory to perform runs
my_dft_pps = ['C.BFD.upf'] # pwscf pseudopotentials
my_qmc_pps = ['C.BFD.xml'] # qmcpack pseudopotentials
# generate your system
# units : 'A'/'B' for Angstrom/Bohr
# axes : simulation cell axes in cartesian coordinates (a1,a2,a3)
# elem : list of atoms in the system
# pos : corresponding atomic positions in cartesian coordinates
# kgrid : Monkhorst-Pack grid
# kshift : Monkhorst-Pack shift (between 0 and 0.5)
# net_charge : system charge in units of e
# net_spin : # of up spins - # of down spins
# C = 4 : (pseudo) carbon has 4 valence electrons
my_system = generate_physical_system(
units = 'A',
axes = [[ 3.57000000e+00, 0.00000000e+00, 0.00000000e+00],
[ 0.00000000e+00, 3.57000000e+00, 0.00000000e+00],
[ 0.00000000e+00, 0.00000000e+00, 3.57000000e+00]],
elem = ['C','C','C','C','C','C','C','C'],
pos = [[ 0.00000000e+00, 0.00000000e+00, 0.00000000e+00],
[ 8.92500000e-01, 8.92500000e-01, 8.92500000e-01],
[ 0.00000000e+00, 1.78500000e+00, 1.78500000e+00],
[ 8.92500000e-01, 2.67750000e+00, 2.67750000e+00],
[ 1.78500000e+00, 0.00000000e+00, 1.78500000e+00],
[ 2.67750000e+00, 8.92500000e-01, 2.67750000e+00],
[ 1.78500000e+00, 1.78500000e+00, 0.00000000e+00],
[ 2.67750000e+00, 2.67750000e+00, 8.92500000e-01]],
kgrid = (1,1,1),
kshift = (0,0,0),
net_charge = 0,
net_spin = 0,
C = 4 # one line like this for each atomic species
)
my_bconds = 'ppp' # ppp/nnn for periodic/open BC's in QMC
# if nnn, center atoms about (a1+a2+a3)/2
If you have a system you would like to try with QMC, make a copy of example.py and fill in the relevant information about the PPs, simulation cell axes, and atomic species/positions. Otherwise, you can proceed with example.py as it is.
Set “generate_only” to 1 and type “./example.py” or similar to generate the input files. All files will be written to “./diamond_vmc” (“./[my_project_name]” if you have changed “my_project_name” in the file). The input files for PWSCF, pw2qmcpack, and QMCPACK are scf.in, pw2qmcpack.in, and vmc.in.xml, respectively. Take some time to inspect the generated input files. If you have questions about the file contents, or run into issues with the generation process, feel free to consult with a lab instructor.
If desired, you can submit the runs directly with example.py. To do this, first reset the Nexus simulation record by typing “rm ./diamond_vmc/sim*/sim.p” or similar and set “generate_only” back to 0. Next rerun example.py (you may want to redirect the text output).
Alternatively the runs can be submitted by hand:
mpirun -np 4 pw.x<scf.in>&scf.out&
(wait until JOB DONE appears in scf.out)
mpirun -np 1 pw2qmcpack.x<p2q.in>&p2q.out&
Once the conversion process has finished, the orbitals should be located in the file diamond_vmc/pwscf_output/pwscf.pwscf.h5. Open diamond_vmc/vmc.in.xml and replace “MISSING.h5” with “./pwscf_output/pwscf.pwscf.h5”. Next submit the VMC run:
mpirun -np 4 qmcpack vmc.in.xml>&vmc.out&
Note: If your system is large, the preceding process may not complete within the time frame of this lab. Working with a stripped down (but relevant) example is a good idea for exploratory runs.
Once the runs have finished, you may want to begin exploring Jastrow optimization and DMC for your system. Example calculations are provided at the end of example.py in the commented out text.
Appendix A: Basic Python constructs¶
Basic Python data types (int, float, str, tuple, list, array, dict, obj) and programming constructs (if statements, for loops, functions w/ keyword arguments) are briefly overviewed in the following. All examples can be executed interactively in Python. To do this, type “python” at the command line and paste any of the shaded text below at the >>> prompt. For more information about effective use of Python, consult the detailed online documentation: https://docs.python.org/2/.
Intrinsic types: int, float, str¶
#this is a comment
i=5 # integer
f=3.6 # float
s='quantum/monte/carlo' # string
n=None # represents "nothing"
f+=1.4 # add-assign (-,*,/ also): 5.0
2**3 # raise to a power: 8
str(i) # int to string: '5'
s+'/simulations' # joining strings: 'quantum/monte/carlo/simulations'
'i={0}'.format(i) # format string: 'i=5'
Container types: tuple, list, array, dict, obj¶
from numpy import array # get array from numpy module
from generic import obj # get obj from Nexus' generic module
t=('A',42,56,123.0) # tuple
l=['B',3.14,196] # list
a=array([1,2,3]) # array
d={'a':5,'b':6} # dict
o=obj(a=5,b=6) # obj
# printing
print(t) # ('A', 42, 56, 123.0)
print(l) # ['B', 3.1400000000000001, 196]
print(a) # [1 2 3]
print(d) # {'a': 5, 'b': 6}
print(o) # a = 5
# b = 6
len(t),len(l),len(a),len(d),len(o) #number of elements: (4, 3, 3, 2, 2)
t[0],l[0],a[0],d['a'],o.a #element access: ('A', 'B', 1, 5, 5)
s = array([0,1,2,3,4]) # slices: works for tuple, list, array
s[:] # array([0, 1, 2, 3, 4])
s[2:] # array([2, 3, 4])
s[:2] # array([0, 1])
s[1:4] # array([1, 2, 3])
s[0:5:2] # array([0, 2, 4])
# list operations
l2 = list(l) # make independent copy
l.append(4) # add new element: ['B', 3.14, 196, 4]
l+[5,6,7] # addition: ['B', 3.14, 196, 4, 5, 6, 7]
3*[0,1] # multiplication: [0, 1, 0, 1, 0, 1]
b=array([5,6,7]) # array operations
a2 = a.copy() # make independent copy
a+b # addition: array([ 6, 8, 10])
a+3 # addition: array([ 4, 5, 6])
a*b # multiplication: array([ 5, 12, 21])
3*a # multiplication: array([3, 6, 9])
# dict/obj operations
d2 = d.copy() # make independent copy
d['c'] = 7 # add/assign element
d.keys() # get element names: ['a', 'c', 'b']
d.values() # get element values: [5, 7, 6]
# obj-specific operations
o.c = 7 # add/assign element
o.set(c=7,d=8) # add/assign multiple elements
An important feature of Python to be aware of is that assignment is most often by reference, that is, new values are not always created. This point is illustrated with an obj instance in the following example, but it also holds for list, array, dict, and others.
>>> o = obj(a=5,b=6)
>>>
>>> p=o
>>>
>>> p.a=7
>>>
>>> print(o)
a = 7
b = 6
>>> q=o.copy()
>>>
>>> q.a=9
>>>
>>> print(o)
a = 7
b = 6
Here p is just another name for o, while q is a fully independent copy of it.
Conditional Statements: if/elif/else¶
a = 5
if a is None:
print('a is None')
elif a==4:
print('a is 4')
elif a<=6 and a>2:
print('a is in the range (2,6]')
elif a<-1 or a>26:
print('a is not in the range [-1,26]')
elif a!=10:
print('a is not 10')
else:
print('a is 10')
#end if
The “\#end if” is not part of Python syntax, but you will see text like this throughout Nexus for clear encapsulation.
Iteration: for¶
from generic import obj
l = [1,2,3]
m = [4,5,6]
s = 0
for i in range(len(l)): # loop over list indices
s += l[i] + m[i]
#end for
print(s) # s is 21
s = 0
for v in l: # loop over list elements
s += v
#end for
print(s) # s is 6
o = obj(a=5,b=6)
s = 0
for v in o: # loop over obj elements
s += v
#end for
print(s) # s is 11
d = {'a':5,'b':4}
for n,v in o.items():# loop over name/value pairs in obj
d[n] += v
#end for
print(d) # d is {'a': 10, 'b': 10}
Functions: def, argument syntax¶
def f(a,b,c=5): # basic function, c has a default value
print(a,b,c)
#end def f
f(1,b=2) # prints: 1 2 5
def f(*args,**kwargs): # general function, returns nothing
print(args) # args: tuple of positional arguments
print(kwargs) # kwargs: dict of keyword arguments
#end def f
f('s',(1,2),a=3,b='t') # 2 pos., 2 kw. args, prints:
# ('s', (1, 2))
# {'a': 3, 'b': 't'}
l = [0,1,2]
f(*l,a=6) # pos. args from list, 1 kw. arg, prints:
# (0, 1, 2)
# {'a': 6}
o = obj(a=5,b=6)
f(*l,**o) # pos./kw. args from list/obj, prints:
# (0, 1, 2)
# {'a': 5, 'b': 6}
f( # indented kw. args, prints
blocks = 200, # ()
steps = 10, # {'steps': 10, 'blocks': 200, 'timestep': 0.01}
timestep = 0.01
)
o = obj( # obj w/ indented kw. args
blocks = 100,
steps = 5,
timestep = 0.02
)
f(**o) # kw. args from obj, prints:
# ()
# {'timestep': 0.02, 'blocks': 100, 'steps': 5}
- Kro16(1,2)
Jaron T. Krogel. Nexus: a modular workflow management system for quantum simulation codes. Computer Physics Communications, 198:154–168, 2016. URL: http://www.sciencedirect.com/science/article/pii/S0010465515002982, doi:10.1016/j.cpc.2015.08.012.
Lab 3: Advanced molecular calculations¶
Topics covered in this lab¶
This lab covers molecular QMC calculations with wavefunctions of increasing sophistication. All of the trial wavefunctions are initially generated with the GAMESS code. Topics covered include:
Generating single-determinant trial wavefunctions with GAMESS (HF and DFT)
Generating multideterminant trial wavefunctions with GAMESS (CISD, CASCI, and SOCI)
Optimizing wavefunctions (Jastrow factors and CSF coefficients) with QMC
DMC time step and walker population convergence studies
Systematic progressions of Jastrow factors in VMC
Systematic convergence of DMC energies with multideterminant wavefunctions
Influence of orbitals basis choice on DMC energy
Lab directories and files¶
abs/lab3_advanced_molecules/exercises
│
├── ex1_first-run-hartree-fock - basic work flow from Hatree-Fock to DMC
│ ├── gms - Hatree-Fock calculation using GAMESS
│ │ ├── h2o.hf.inp - GAMESS input
│ │ ├── h2o.hf.dat - GAMESS punch file containing orbitals
│ │ └── h2o.hf.out - GAMESS output with orbitals and other info
│ ├── convert - Convert GAMESS wavefunction to QMCPACK format
│ │ ├── h2o.hf.out - GAMESS output
│ │ ├── h2o.ptcl.xml - converted particle positions
│ │ └── h2o.wfs.xml - converted wave function
│ ├── opt - VMC optimization
│ │ └── optm.xml - QMCPACK VMC optimization input
│ ├── dmc_timestep - Check DMC timestep bias
│ │ └── dmc_ts.xml - QMCPACK DMC input
│ └── dmc_walkers - Check DMC population control bias
│ └── dmc_wk.xml - QMCPACK DMC input template
│
├── ex2_slater-jastrow-wf-options - explore jastrow and orbital options
│ ├── jastrow - Jastrow options
│ │ ├── 12j - no 3-body Jastrow
│ │ ├── 1j - only 1-body Jastrow
│ │ └── 2j - only 2-body Jastrow
│ └── orbitals - Orbital options
│ ├── pbe - PBE orbitals
│ │ └── gms - DFT calculation using GAMESS
│ │ └── h2o.pbe.inp - GAMESS DFT input
│ ├── pbe0 - PBE0 orbitals
│ ├── blyp - BLYP orbitals
│ └── b3lyp - B3LYP orbitals
│
├── ex3_multi-slater-jastrow
│ ├── cisd - CISD wave function
│ │ ├── gms - CISD calculation using GAMESS
│ │ │ ├── h2o.cisd.inp - GAMESS input
│ │ │ ├── h2o.cisd.dat - GAMESS punch file containing orbitals
│ │ │ └── h2o.cisd.out - GAMESS output with orbitals and other info
│ │ └── convert - Convert GAMESS wavefunction to QMCPACK format
│ │ └── h2o.hf.out - GAMESS output
│ ├── casci - CASCI wave function
│ │ └── gms - CASCI calculation using GAMESS
│ └── soci - SOCI wave function
│ ├── gms - SOCI calculation using GAMESS
│ ├── thres0.01 - VMC optimization with few determinants
│ └── thres0.0075 - VMC optimization with more determinants
│
└── pseudo
├── H.BFD.gamess - BFD pseudopotential for H in GAMESS format
├── O.BFD.CCT.gamess - BFD pseudopotential for O in GAMESS format
├── H.xml - BFD pseudopotential for H in QMCPACK format
└── O.xml - BFD pseudopotential for H in QMCPACK format
Exercise #1: Basics¶
The purpose of this exercise is to show how to generate wavefunctions for QMCPACK using GAMESS and to optimize the resulting wavefunctions using VMC. This will be followed by a study of the time step and walker population dependence of DMC energies. The exercise will be performed on a water molecule at the equilibrium geometry.
Generation of a Hartree-Fock wavefunction with GAMESS¶
From the top directory, go to “ex1_first-run-hartree-fock/gms.” This
directory contains an input file for a HF calculation of a water
molecule using BFD ECPs and the corresponding cc-pVTZ basis set. The
input file should be named: “h2o.hf.inp.” Study the input file. See
Section Appendix A: GAMESS input for a
more detailed description of the GAMESS input syntax. However, there
will be a better time to do this soon, so we recommend continuing with
the exercise at this point. After you are done, execute GAMESS with this
input and store the standard output in a file named “h2o.hf.output.”
Finally, in the “convert” folder, use convert4qmc to generate the
QMCPACK particleset and wavefunction files. It is always useful
to rename the files generated by convert4qmc to something meaningful
since by default they are called sample.Gaussian-G2.xml and
sample.Gaussian-G2.ptcl.xml. In a standard computer (without
cross-compilation), these tasks can be accomplished by the following
commands.
cd ${TRAINING TOP}/ex1_first-run-hartree-fock/gms
jobrun_vesta rungms h2o.hf
cd ../convert
cp ../gms/h2o.hf.output
jobrun_vesta convert4qmc -gamess h2o.hf.output -add3BodyJ
mv sample.Gaussian-G2.xml h2o.wfs.xml
mv sample.Gaussian-G2.ptcl.xml h2o.ptcl.xml
The HF energy of the system is -16.9600590022 Ha. To search for the energy in the output file quickly, you can use
grep "TOTAL ENERGY =" h2o.hf.output
As the job runs on VESTA, it is a good time to review Section :ref`lab-adv-mol-convert4qmc`, “Appendix B: convert4qmc,” which contains a description on the use of the converter.
Optimize the wavefunction¶
When execution of the previous steps is completed, there should be two
new files called h2o.wfs.xml and h2o.ptcl.xml. Now we will use
VMC to optimize the Jastrow parameters in the wavefunction. From the top
directory, go to “ex1_first-run-hartree-fock/opt.” Copy the xml
files generated in the previous step to the current directory. This
directory should already contain a basic QMCPACK input file for an
optimization calculation (optm.xml) Open optm.xml with your
favorite text editor and modify the name of the files that contain the
wavefunction and particleset XML blocks. These files are
included with the commands:
<include href=ptcl.xml/>
<include href=wfs.xml/>
(the particle set must be defined before the wavefunction). The name of
the particle set and wavefunction files should now be h2o.ptcl.xml
and h2o.wfs.xml, respectively. Study both files and submit when you
are ready. Notice that the location of the ECPs has been set for you; in
your own calculations you have to make sure you obtain the ECPs from the
appropriate libraries and convert them to QMCPACK format using
ppconvert. While these calculations finish is a good time to study
Appendix C: Wavefunction optimization XML block, which contains a review of the main parameters
in the optimization XML block. The previous steps can be accomplished by
the following commands:
cd ${TRAINING TOP}/ex1_first-run-hartree-fock/opt
cp ../convert/h2o.wfs.xml ./
cp ../convert/h2o.ptcl.xml ./
# edit optm.xml to include the correct ptcl.xml and wfs.xml
jobrun_vesta qmcpack optm.xml
Use the analysis tool qmca to analyze the results of the
calculation. Obtain the VMC energy and variance for each step in the
optimization and plot it using your favorite program. Remember that
qmca has built-in functions to plot the analyzed data.
qmca -q e *scalar.dat -p
The resulting energy as a function of the optimization step should look
qualitatively similar to Fig. 17. The energy should
decrease quickly as a function of the number of optimization steps.
After 6–8 steps, the energy should be converged to \(\sim\)2–3
mHa. To improve convergence, we would need to increase the number of
samples used during optimization (You can check this for yourself
later.). With optimized wavefunctions, we are in a position to perform
VMC and DMC calculations. The modified wavefunction files after each
step are written in a file named ID.sNNN.opt.xml, where ID is the
identifier of the calculation defined in the input file (this is defined
in the project XML block with parameter “id”) and NNN is a series number
that increases with every executable xml block in the input file.
VMC energy as a function of optimization step.¶
Time-step study¶
Now we will study the dependence of the DMC energy with time step. From
the top directory, go to “ex1_first-run-hartree-fock/dmc_timestep.”
This folder contains a basic XML input file (dmc_ts.xml) that
performs a short VMC calculation and three DMC calculations with varying
time steps (0.1, 0.05, 0.01). Link the particleset and the last
optimization file from the previous folder (the file called
jopt-h2o.sNNN.opt.xml with the largest value of NNN). Rename the
optimized wavefunction file to any suitable name if you wish (for
example, h2o.opt.xml) and change the name of the particleset and
wavefunction files in the input file. An optimized wavefunction can
be found in the reference files (same location) in case it is needed.
The main steps needed to perform this exercise are:
cd \$\{TRAINING TOP\}/ex1_first-run-hartree-fock/dmc_timestep
cp ../opt/h2o.ptcl.xml ./
cp ../opt/jopt-h2o.s007.opt.xml h2o.opt.wfs.xml
# edit dmc_ts.xml to include the correct ptcl.xml and wfs.xml
jobrun_vesta qmcpack dmc_ts.xml
While these runs complete, go to Appendix D: VMC and DMC XML block and review the basic VMC and DMC input blocks. Notice that in the current DMC blocks the time step is decreased as the number of blocks is increased. Why is this?
When the simulations are finished, use qmca to analyze the output
files and plot the DMC energy as a function of time step. Results should
be qualitatively similar to those presented in
Fig. 18; in this case we present more time
steps with well converged results to better illustrate the time step
dependence. In realistic calculations, the time step must be chosen
small enough so that the resulting error is below the desired accuracy.
Alternatively, various calculations can be performed and the results
extrapolated to the zero time-step limit.
DMC energy as a function of time step.¶
Walker population study¶
Now we will study the dependence of the DMC energy with the number of walkers in the simulation. Remember that, in principle, the DMC distribution is reached in the limit of an infinite number of walkers. In practice, the energy and most properties converge to high accuracy with \(\sim\)100–1,000 walkers. The actual number of walkers needed in a calculation will depend on the accuracy of the VMC wavefunction and on the complexity and size of the system. Also notice that using too many walkers is not a problem; at worse it will be inefficient since it will cost more computer time than necessary. In fact, this is the strategy used when running QMC calculations on large parallel computers since we can reduce the statistical error bars efficiently by running with large walker populations distributed across all processors.
From the top directory, go to
“ex1_first-run-hartree-fock/dmc_walkers.” Copy the optimized
wavefunction and particleset files used in the previous
calculations to the current folder; these are the files generated during
step 2 of this exercise. An optimized wavefunction file can be found
in the reference files (same location) in case it is needed. The
directory contains a sample DMC input file and submission script. Create
three directories named NWx, with x values of 120,240,480, and copy the
input file to each one. Go to “NW120,” and, in the input file, change
the name of the wavefunction and particleset files (in this case
they will be located one directory above, so use
“../dmc_timestep/h2.opt.xml,” for example); change the PP directory
so that it points to one directory above; change “targetWalkers” to 120;
and change the number of steps to 100, the time step to 0.01, and the
number of blocks to 400. Notice that “targetWalkers” is one way to set
the desired (average) number of walkers in a DMC calculation. One can
alternatively set “samples” in the <qmc method="vmc" block to carry over de-correlated VMC
configurations as DMC walkers. For your own simulations, we generally
recommend setting \(\sim\)2*(#threads) walkers per node (slightly
smaller than this value).
The main steps needed to perform this exercise are
cd ${TRAINING TOP}/ex1_first-run-hartree-fock/dmc_walkers
cp ../opt/h2o.ptcl.xml ./
cp ../opt/jopt-h2o.s007.opt.xml h2o.opt.wfs.xml
# edit dmc_wk.xml to include the correct ptcl.xml and wfs.xml and
# use the correct pseudopotential directory
mkdir NW120
cp dmc_wk.xml NW120
# edit dmc_wk.xml to use the desired number of walkers,
# and collect the desired amount of statistics
jobrun_vesta qmcpack dmc_wk.xml
# repeat for NW240, NW480
Repeat the same procedure in the other folders by setting
(targetWalkers=240, steps=100, timestep=0.01, blocks=200) in NW240 and
(targetWalkers=480, steps=100, timestep=0.01, blocks=100) in NW480. When
the simulations complete, use qmca to analyze and plot the energy as
a function of the number of walkers in the calculation. As always,
Fig. 19 shows representative results of the
energy dependence on the number of walkers for a single water molecule.
As shown, less than 240 walkers are needed to obtain an accuracy of 0.1
mHa.
DMC energy as a function of the average number of walkers.¶
Exercise #2: Slater-Jastrow wavefunction options¶
From this point on in the tutorial we assume familiarity with the basic
parameters in the optimization, VMC, and DMC XML input blocks of
QMCPACK. In addition, we assume familiarity with the submission system.
As a result, the folder structure will not contain any prepared input or
submission files, so you will need to generate them using input files
from exercise 1. In the case of QMCPACK sample files, you will find
optm.xml, vmc dmc.xml, and submit.csh files. Some of the
options in these files can be left unaltered, but many of them will need
to be tailored to the particular calculation.
In this exercise we will study the dependence of the DMC energy on the choices made in the wavefunction ansatz. In particular, we will study the influence/dependence of the VMC energy with the various terms in the Jastrow. We will also study the influence of the VMC and DMC energies on the SPOs used to form the Slater determinant in single-determinant wavefunctions. For this we will use wavefunctions generated with various exchange-correlation functionals in DFT. Finally, we will optimize a simple multideterminant wavefunction and study the dependence of the energy on the number of configurations used in the expansion. All of these exercises will be performed on the water molecule at equilibrium.
Influence of Jastrow on VMC energy with HF wavefunction¶
In this section we will study the dependence of the VMC energy on the
various Jastrow terms (e.g., 1-body, 2-body and 3-body. From the top
directory, go to “ex2_slater-jastrow-wf-options/jastrow.” We will
compare the single-determinant VMC energy using a 2-body Jastrow term,
both 1- and 2-body terms, and finally 1-, 2- and 3-body terms. Since we
are interested in the influence of the Jastrow, we will use the HF
orbitals calculated in exercise #1. Make three folders named 2j, 12j,
and 123j. For both 2j and 12j, copy the input file optm.xml from
“ex1_first-run-hartree-fock/opt.” This input file performs both
wavefunction optimization and a VMC calculation. Remember to correct
relative paths to the PP directory. Copy the un-optimized HF
wavefunction and particleset files from
“ex1_first-run-hartree-fock/convert”; if you followed the
instructions in exercise #1 these should be named h2o.wfs.xml and
h2o.ptcl.xml. Otherwise, you can obtained them from the REFERENCE
files. Modify the h2o.wfs.xml file to remove the appropriate Jastrow
blocks. For example, for a 2-body Jastrow (only), you need to eliminate
the Jastrow blocks named <jastrow name="J1" and <jastrow name="J3." In
the case of 12j, remove only <jastrow name="J3." Recommended settings
for the optimization run are nodes=32, threads=16, blocks=250,
samples=128000, time-step=0.5, 8 optimization loops. Recommended
settings in the VMC section are walkers=16, blocks=1000, steps=1,
substeps=100. Notice that samples should always be set to blocks*threads
per node*nodes = 32*16*250=128000. Repeat the process in both 2j and 12j
cases. For the 123j case, the wavefunction has already been optimized in
the previous exercise. Copy the optimized HF wavefunction and the
particleset from “ex1_first-run-hartree-fock/opt.” Copy the input
file from any of the previous runs and remove the optimization block
from the input, just leave the VMC step. In all three cases, modify the
submission script and submit the run.
Because these simulations will take several minutes to complete, this is
an excellent opportunity to go to
Appendix E: Wavefunction XML block and review the wavefunction XML block used by QMCPACK. When the
simulations are completed, use qmca to analyze the output files.
Using your favorite plotting program (e.g., gnu plot), plot the energy
and variance as a function of the Jastrow form.
Fig. 20 shows a typical result for this
calculation. As can be seen, the VMC energy and variance depends
strongly on the form of the Jastrow. Since the DMC error bar is directly
related to the variance of the VMC energy, improving the Jastrow will
always lead to a reduction in the DMC effort. In addition, systematic
approximations (time step, number of walkers, etc.) are also reduced
with improved wavefunctions.
VMC energy as a function of Jastrow type.¶
Generation of wavefunctions from DFT using GAMESS¶
In this section we will use GAMESS to generate wavefunctions for QMCPACK
from DFT calculations. From the top folder, go to
“ex2_slater-jastrow-wf-options/orbitals.” To demonstrate the
variation in DMC energies with the choice of DFT orbitals, we will
choose the following set of exchange-correlation functionals (PBE, PBE0,
BLYP, B3LYP). For each functional, make a directory using your preferred
naming convention (e.g., the name of the functional). Go into each
folder and copy a GAMESS input file from
“ex1_first-run-hartree-fock/gms.” Rename the file with your
preferred naming convention; we suggest using h2o.[dft].inp, where
[dft] is the name of the functional used in the calculation. At this
point, this input file should be identical to the one used to generate
the HF wavefunction in exercise #1. To perform a DFT calculation we only
need to add “DFTTYP” to the $CONTRL ... $END section and set it to the desired functional
type, for example, “DFTTYP=PBE” for a PBE functional. This variable must
be set to (PBE, PBE0, BLYP, B3LYP) to obtain the appropriate functional
in GAMESS. For a complete list of implemented functionals, see the
GAMESS input manual.
Optimization and DMC calculations with DFT wavefunctions¶
In this section we will optimize the wavefunction generated in the
previous step and perform DMC calculations. From the top directory, go
to “ex2_slater-jastrow-wf-options/orbitals.” The steps required to
achieve this are identical to those used to optimize the wavefunction
with HF orbitals. Make individual folders for each calculation and
obtain the necessary files to perform optimization, for example, VMC and
DMC calculations from “for ex1_first-run-hartree-fock/opt” and
“ex1_first-run-hartree-fock/dmc_ts.” For each functional, make the
appropriate modifications to the input files and copy the
particleset and wavefunction files from the appropriate
directory in “ex2_slater-jastrow-wf-options/orbitals/[dft].” We
recommend the following settings: nodes=32, threads=16, (in
optimization) blocks=250, samples=128000, timestep=0.5, 8 optimization
loops, (in VMC) walkers=16, blocks=100, steps=1, substeps=100, (in DMC)
blocks 400, targetWalkers=960, and timestep=0.01. Submit the runs and
analyze the results using qmca.
How do the energies compare against each other? How do they compare against DMC energies with HF orbitals?
Exercise #3: Multideterminant wavefunctions¶
In this exercise we will study the dependence of the DMC energy on the set of orbitals and the type of configurations included in a multideterminant wavefunction.
Generation of a CISD wavefunctions using GAMESS¶
In this section we will use GAMESS to generate a multideterminant wavefunction with configuration interaction with single and double excitations (CISD). In CISD, the Schrodinger equation is solved exactly on a basis of determinants including the HF determinant and all its single and double excitations.
Go to “ex3_multi-slater-jastrow/cisd/gms” and you will see input and
output files named h2o.cisd.inp and h2o.cisd.out. Because of
technical problems with GAMESS in the BGQ architecture of VESTA, we are
unable to use CISD properly in GAMESS. Consequently, the output of the
calculation is already provided in the directory.
There will be time in the next step to study the GAMESS input files and the description in Appendix A: GAMESS input. Since the output is already provided, the only action needed is to use the converter to generate the appropriate QMCPACK files.
jobrun_vesta convert4qmc h2o.cisd.out -ci h2o.cisd.out \
-readInitialGuess 57 -threshold 0.0075
We used the PRTMO=.T. flag in the GUESS section to include orbitals in
the output file. You should read these orbitals from the output
(-readInitialGuess 40). The highest occupied orbital in any determinant
should be 34, so reading 40 orbitals is a safe choice. In this case, it
is important to rename the XML files with meaningful names, for example,
h2o.cisd.wfs.xml. A threshold of 0.0075 is sufficient for the
calculations in the training.
Optimization of a multideterminant wavefunction¶
In this section we will optimize the wavefunction generated in the
previous step. There is no difference in the optimization steps if a
single determinant and a multideterminant wavefunction. QMCPACK will
recognize the presence of a multideterminant wavefunction and will
automatically optimize the linear coefficients by default. Go to
“ex3_multi-slater-jastrow/cisd” and make a folder called
thres0.01. Copy the particleset and wavefunction files
created in the previous step to the current directory. With your
favorite text editor, open the wavefunction file h2o.wfs.xml.
Look for the multideterminant XML block and change the “cutoff”
parameter in detlist to 0.01. Then follow the same steps used in Section
9.4.3, “Optimization and DMC calculations with DFT wavefunctions” to
optimize the wavefunction. Similar to this case, design a QMCPACK input
file that performs wavefunction optimization followed by VMC and DMC
calculations. Submit the calculation.
This is a good time to review the GAMESS input file description in
Appendix A: GAMESS input, go to the previous directory and make a new folder
named thres0.0075. Repeat the previous steps to optimize the
wavefunction with a cutoff of 0.01, but use a cutoff of 0.0075 this
time. This will increase the number of determinants used in the
calculation. Notice that the “cutoff” parameter in the XML should be
less than the “-threshold 0.0075” flag passed to the converted, which is
further bounded by the PRTTOL flag in the GAMESS input.
After the wavefunction is generated, we are ready to optimize. Instead of starting from an un-optimized wavefunction, we can start from the optimized wavefunction from thres0.01 to speed up convergence. You will need to modify the file and change the cutoff in detlist to 0.0075 with a text editor. Repeat the optimization steps and submit the calculation.
When you are done, use qmca to analyze the results. Compare the energies at these two
coefficient cutoffs with the energies obtained with DFT orbitals. Because of the time limitations of this tutorial, it is not practical to optimize the wavefunctions with a smaller cutoff since this would require more samples and longer runs due to the larger number of optimizable parameters. Fig. 21 shows the results of such exercise: the DMC energy as a function of the cutoff in the wavefunction. As can be seen, a large improvement in the energy is obtained as the number of configurations is increased.
DMC energy as a function of the sum of the square of CI coefficients from CISD.¶
CISD, CASCI, and SOCI¶
Go to “ex3_multi-slater-jastrow” and inspect the folders for the
remaining wavefunction types: CASCI and SOCI. Follow the steps in the
previous exercise and obtain the optimized wavefunctions for these
determinant choices. Notice that the SOCI GAMESS output is not included
because it is large. Already converted XML inputs can be found in
“ex3_multi-slater-jastrow/soci/thres*.”
A CASCI wavefunction is produced from a CI calculation that includes all the determinants in a complete active space (CAS) calculation, in this case using the orbitals from a previous CASSCF calculation. In this case we used a CAS(8,8) active space that includes all determinants generated by distributing 8 electrons in the lowest 8 orbitals. A SOCI calculation is similar to the CAS-CI calculation, but in addition to the determinants in the CAS it also includes all single and double excitations from all of them, leading to a much larger determinant set. Since you now have considerable experience optimizing wavefunctions and calculating DMC energies, we will leave it to you to complete the remaining tasks on your own. If you need help, refer to previous exercises in the tutorial. Perform optimizations for both wavefunctions using cutoffs in the CI expansion of 0.01 an 0.0075. If you have time, try to optimize the wavefunctions with a cutoff of 0.005. Analyze the results and plot the energy as a function of cutoff for all three cases: CISD, CAS-CI, and SOCI.
Fig. 21 shows the result of similar calculations using more samples and smaller cutoffs. The results should be similar to those produced in the tutorial. For reference, the exact energy of the water molecule with ECPs is approximately -17.276 Ha. From the results of the tutorial, how does the selection of determinants relate to the expected DMC energy? What about the choice in the set of orbitals?
Appendix A: GAMESS input¶
In this section we provide a brief description of the GAMESS input needed to produce
trial wavefunction for QMC calculations with QMCPACK. We assume basic familiarity
with GAMESS input structure, particularly regarding the input of atomic coordinates and
the definition of Gaussian basis sets. This section focuses on generation of the output
files needed by the converter tool, convert4qmc. For a description of the converter, see Appendix B: convert4qmc.
Only a subset of the methods available in GAMESS can be used to generate wavefunctions for QMCPACK, and we restrict our description to these. For a complete description of all the options and methods available in GAMESS, please refer to the official documentation at “http://www.msg.ameslab.gov/gamess/documentation.html.”
Currently, convert4qmc can process output for the following methods in GAMESS (in
SCFTYP): RHF, ROHF, and MCSCF. Both HF and DFT calculations (any DFT
type) can be used in combination with RHF and ROHF calculations. For MCSCF and CI
calculations, ALDET, ORMAS, and GUGA drivers can be used (details follow).
HF input¶
The following input will perform a restricted HF calculation on a closed-shell singlet
(multiplicity=1). This will generate RHF orbitals for any molecular system defined in
$DATA ... $END.
$CONTRL SCFTYP=RHF RUNTYP=ENERGY MULT=1
ISPHER=1 EXETYP=RUN COORD=UNIQUE MAXIT=200 $END
$SYSTEM MEMORY=150000000 $END
$GUESS GUESS=HUCKEL $END
$SCF DIRSCF=.TRUE. $END
$DATA
...
Atomic Coordinates and basis set
...
$END
Main options:
SCFTYP: Type of SCF method, options: RHF, ROHF, MCSCF, UHF and NONE.
RUNTYP: Type of run. For QMCPACK wavefunction generation this should always be ENERGY.
MULT: Multiplicity of the molecule.
ISPHER: Use spherical harmonics (1) or Cartesian basis functions (-1).
COORD: Input structure for the atomic coordinates in $DATA.
DFT calculations¶
The main difference between the input for a RHF/ROHF calculation and a DFT calculation is the definition of the DFTTYP parameter. If this is set in the $CONTROL section, a DFT calculation will be performed with the appropriate functional. Notice that although the default values are usually adequate, DFT calculations have many options involving the integration grids and accuracy settings. Make sure you study the input manual to be aware of these. Refer to the input manual for a list of the implemented exchange-correlation functionals.
MCSCF¶
MCSCF calculations are performed by setting SCFTYP=MCSCF in the CONTROL section. If this option is set, an MCSCF section must be added to the input file with the options for the calculation. An example section for the water molecule used in the tutorial follows.
$MCSCF CISTEP=GUGA MAXIT=1000 FULLNR=.TRUE. ACURCY=1.0D-5 $END
The most important parameter is CISTEP, which defines the CI package used. The only options compatible with QMCPACK are: ALDET, GUGA, and ORMAS. Depending on the package used, additional input sections are needed.
CI¶
Configuration interaction (full CI, truncated CI, CAS-CI, etc) calculations are performed
by setting SCFTYP=NONE and CITYP=GUGA,ALDET,ORMAS. Each one of these packages
requires further input sections, which are typically slightly different from the input sections
needed for MCSCF runs.
GUGA: Unitary group CI package¶
The GUGA package is the only alternative if one wants CSFs with GAMESS. We subsequently provide a very brief description of the input sections needed to perform MCSCF, CASCI, truncated CI, and SOCI with this package. For a complete description of these methods and all the options available, please refer to the GAMESS input manual.
GUGA-MCSCF¶
The following input section performs a CASCI calculation with a CAS that includes 8 electrons in 8 orbitals (4 DOC and 4 VAL), for example, CAS(8,8). NMCC is the number of frozen orbitals (doubly occupied orbitals in all determinants), NDOC is the number of double occupied orbitals in the reference determinant, NVAL is the number of singly occupied orbitals in the reference (for spin polarized cases), and NVAL is the number of orbitals in the active space. Since FORS is set to .TRUE., all configurations in the active space will be included. ISTSYM defines the symmetry of the desired state.
$MCSCF CISTEP=GUGA MAXIT=1000 FULLNR=.TRUE. ACURCY=1.0D-5 $END
$DRT GROUP=C2v NMCC=0 NDOC=4 NALP=0 NVAL=4 ISTSYM=1 MXNINT= 500000 FORS=.TRUE. $END
GUGA-CASCI¶
The following input section performs a CASCI calculation with a CAS that includes 8 electrons in 8 orbitals (4 DOC and 4 VAL), for example, CAS(8,8). NFZC is the number of frozen orbitals (doubly occupied orbitals in all determinants). All other parameters are identical to those in the MCSCF input section.
$CIDRT GROUP=C2v NFZC=0 NDOC=4 NALP=0 NVAL=4 NPRT=2 ISTSYM=1 FORS=.TRUE. MXNINT= 500000 $END
$GUGDIA PRTTOL=0.001 CVGTOL=1.0E-5 ITERMX=1000 $END
GUGA-truncated CI¶
The following input sections will lead to a truncated CI calculation. In this particular case it will perform a CISD calculation since IEXCIT is set to 2. Other values in IEXCIT will lead to different CI truncations; for example, IEXCIT=4 will lead to CISDTQ. Notice that only the lowest 30 orbitals will be included in the generation of the excited determinants in this case. For a full CISD calculation, NVAL should be set to the total number of virtual orbitals.
$CIDRT GROUP=C2v NFZC=0 NDOC=4 NALP=0 NVAL=30 NPRT=2 ISTSYM=1 IEXCIT=2 MXNINT= 500000 $END
$GUGDIA PRTTOL=0.001 CVGTOL=1.0E-5 ITERMX=1000 $END
GUGA-SOCI¶
The following input section performs a SOCI calculation with a CAS that includes 8 electrons in 8 orbitals (4 DOC and 4 VAL), for example, CAS(8,8). Since SOCI is set to .TRUE., all single and double determinants from all determinants in the CAS(8,8) will be included.
$CIDRT GROUP=C2v NFZC=0 NDOC=4 NALP=0 NVAL=4 NPRT=2 ISTSYM=1 SOCI=.TRUE. NEXT=30 MXNINT= 500000 $END
$GUGDIA PRTTOL=0.001 CVGTOL=1.0E-5 ITERMX=1000 $END
ECP¶
To use ECPs in GAMESS, you must define a {$ECP ... $END}
block. There must be a definition of a potential for every atom in the system, including
symmetry equivalent ones. In addition, they must appear in the particular order expected
by GAMESS. The following example shows an ECP input block for a single water molecule using
BFD ECPs. To turn on the use of ECPs, the option “ECP=READ” must be added to the
CONTROL input block.
$ECP
O-QMC GEN 2 1
3
6.00000000 1 9.29793903
55.78763416 3 8.86492204
-38.81978498 2 8.62925665
1
38.41914135 2 8.71924452
H-QMC GEN 0 0
3
1.000000000000 1 25.000000000000
25.000000000000 3 10.821821902641
-8.228005709676 2 9.368618758833
H-QMC
$END
Appendix B: convert4qmc¶
To generate the particleset and wavefunction XML blocks required by QMCPACK in
calculations with molecular systems, the converter convert4qmc must be used. The converter
will read the standard output from the appropriate quantum chemistry calculation and will
generate all the necessary input for QMCPACK. In the following, we describe the main options of the
converter for GAMESS output. In general, there are three ways to use the converter depending
on the type of calculation performed. The minimum syntax for each option is shown subsequently.
For a description of the XML files produced by the converter, see Appendix E: Wavefunction XML block.
For all single-determinant calculations (HF and DFT with any DFTTYP):
convert4qmc -gamess single det.out
single det.out is the standard output generated by GAMESS.
(This option is not recommended. Use the following option to avoid mistakes.) For multideterminant calculations where the orbitals and configurations are read from different files (e.g., when using orbitals from a MCSCF run and configurations from a subsequent CI run):
convert4qmc -gamess orbitals multidet.out -ci cicoeff multidet.out
orbitals_multidet.out is the standard output from the calculation that generates the orbitals. cicoeff multidet.out is the standard output from the calculation that calculates the CI expansion.
For multideterminant calculations where the orbitals and configurations are read from the same file, using PRTMO=.T. in the GUESS input block:
convert4qmc -gamess multi det.out -ci multi det.out -readInitialGuess Norb
multi_det.out is the standard output from the calculation that calculates the CI expansion.
Options:
-gamess file.out: Standard output of GAMESS calculation. With the exception of determinant configurations and coefficients in multideterminant calculations, everything else is read from this file including atom coordinates, basis sets, SPOs, ECPs, number of electrons, multiplicity, etc.
-ci file.out: In multideterminant calculations, determinant configurations and coefficients are read from this file. Notice that SPOs are NOT read from this file. Recognized CI packages are ALDET, GUGA, and ORMAS. Output produced with the GUGA package MUST have the option “NPRT=2” in the CIDRT or DRT input blocks.
-threshold cutoff: Cutoff in multideterminant expansion. Only configurations with coefficients above this value are printed.
-zeroCI: Sets to zero the CI coefficients of all determinants, with the exception of the first one.
-readInitialGuess Norb: Reads Norb initial orbitals (“INITIAL GUESS ORBITALS”) from GAMESS output. These are orbitals generated by the GUESS input block and printed with the option “PRTMO=.T.”. Notice that this is useful only in combination with the option “GUESS=MOREAD” and in cases where the orbitals are not modified in the GAMESS calculation, e.g. CI runs. This is the recommended option in all CI calculations.
-NaturalOrbitals Norb: Read Norb “NATURAL ORBITALS” from GAMESS output. The natural orbitals must exists in the output, otherwise the code aborts.
-add3BodyJ: Adds 3-body Jastrow terms (e-e-I) between electron pairs (both same spin and opposite spin terms) and all ion species in the system. The radial function is initialized to zero, and the default cutoff is 10.0 bohr. The converter will add a 1- and 2-body Jastrow to the wavefunction block by default.
Useful notes¶
The type of SPOs read by the converter depends on the type of calculation and on the options used. By default, when neither -readInitialGuess nor -NaturalOrbitals are used, the following orbitals are read in each case (notice that -readInitialGuess or -NaturalOrbitals are mutually exclusive):
RHF and ROHF: “EIGENVECTORS”
MCSCF: “MCSCF OPTIMIZED ORBITALS”
GUGA, ALDET, ORMAS: Cannot read orbitals without -readInitialGuess or -NaturalOrbitals options.
The SPOs and printed CI coefficients in MCSCF calculations are not consistent in GAMESS. The printed CI coefficients correspond to the next-to-last iteration; they are not recalculated with the final orbitals. So to get appropriate CI coefficients from MCSCF calculations, a subsequent CI (no SCF) calculation is needed to produce consistent orbitals. In principle, it is possible to read the orbitals from the MCSCF output and the CI coefficients and configurations from the output of the following CI calculations. This could lead to problems in principle since GAMESS will rotate initial orbitals by default to obtain an initial guess consistent with the symmetry of the molecule. This last step is done by default and can change the orbitals reported in the MCSCF calculation before the CI is performed. To avoid this problem, we highly recommend using the preceding option #3 to read all the information from the output of the CI calculation; this requires the use of “PRTMO=.T.” in the GUESS input block. Since the orbitals are printed after any symmetry rotation, the resulting output will always be consistent.
Appendix C: Wavefunction optimization XML block¶
<loop max="10">
<qmc method="linear" move="pbyp" checkpoint="-1" gpu="no">
<parameter name="blocks"> 10 </parameter>
<parameter name="warmupsteps"> 25 </parameter>
<parameter name="steps"> 1 </parameter>
<parameter name="substeps"> 20 </parameter>
<parameter name="timestep"> 0.5 </parameter>
<parameter name="samples"> 10240 </parameter>
<cost name="energy"> 0.95 </cost>
<cost name="unreweightedvariance"> 0.0 </cost>
<cost name="reweightedvariance"> 0.05 </cost>
<parameter name="useDrift"> yes </parameter>
<parameter name="bigchange">10.0</parameter>
<estimator name="LocalEnergy" hdf5="no"/>
<parameter name="usebuffer"> yes </parameter>
<parameter name="nonlocalpp"> yes </parameter>
<parameter name="MinMethod">quartic</parameter>
<parameter name="exp0">-6</parameter>
<parameter name="alloweddifference"> 1.0e-5 </parameter>
<parameter name="stepsize"> 0.15 </parameter>
<parameter name="nstabilizers"> 1 </parameter>
</qmc>
</loop>
Options:
bigchange: (default 50.0) Largest parameter change allowed
usebuffer: (default no) Save useful information during VMC
nonlocalpp: (default no) Include nonlocal energy on 1-D min
MinMethod: (default quartic) Method to calculate magnitude of parameter change quartic: fit quartic polynomial to four values of the cost function obtained using reweighting along chosen direction linemin: direct line minimization using reweighting rescale: no 1-D minimization. Uses Umrigars suggestions.
stepsize: (default 0.25) Step size in either quartic or linemin methods.
alloweddifference: (default 1e-4) Allowed increase in energy
exp0: (default -16.0) Initial value for stabilizer (shift to diagonal of H). Actual value of stabilizer is 10 exp0
nstabilizers: (default 3) Number of stabilizers to try
stabilizaterScale: (default 2.0) Increase in value of exp0 between iterations.
max its: (default 1) Number of inner loops with same sample
minwalkers: (default 0.3) Minimum value allowed for the ratio of effective samples to actual number of walkers in a reweighting step. The optimization will stop if the effective number of walkers in any reweighting calculation drops below this value. Last set of acceptable parameters are kept.
maxWeight: (defaul 1e6) Maximum weight allowed in reweighting. Any weight above this value will be reset to this value.
Recommendations:
Set samples to equal to (#threads)*blocks.
Set steps to 1. Use substeps to control correlation between samples.
For cases where equilibration is slow, increase both substeps and warmupsteps.
For hard cases (e.g., simultaneous optimization of long MSD and 3-Body J), set exp0 to 0 and do a single inner iteration (max its=1) per sample of configurations.
Appendix D: VMC and DMC XML block¶
<qmc method="vmc" move="pbyp" checkpoint="-1">
<parameter name="useDrift">yes</parameter>
<parameter name="warmupsteps">100</parameter>
<parameter name="blocks">100</parameter>
<parameter name="steps">1</parameter>
<parameter name="substeps">20</parameter>
<parameter name="walkers">30</parameter>
<parameter name="timestep">0.3</parameter>
<estimator name="LocalEnergy" hdf5="no"/>
</qmc>
<qmc method="dmc" move="pbyp" checkpoint="-1">
<parameter name="nonlocalmoves">yes</parameter>
<parameter name="targetWalkers">1920</parameter>
<parameter name="blocks">100</parameter>
<parameter name="steps">100</parameter>
<parameter name="timestep">0.1</parameter>
<estimator name="LocalEnergy" hdf5="no"/>
</qmc>
General Options:
move: (default “walker”) Type of electron move. Options: “pbyp” and “walker.”
checkpoint: (default “-1”) (If > 0) Generate checkpoint files with given frequency. The calculations can be restarted/continued with the produced checkpoint files.
useDrift: (default “yes”) Defines the sampling mode. useDrift = “yes” will use Langevin acceleration to sample the VMC and DMC distributions, while useDrift=“no” will use random displacements in a box.
warmupSteps: (default 0) Number of steps warmup steps at the beginning of the calculation. No output is produced for these steps.
blocks: (default 1) Number of blocks (outer loop).
steps: (default 1) Number of steps per blocks (middle loop).
sub steps: (default 1) Number of substeps per step (inner loop). During substeps, the local energy is not evaluated in VMC calculations, which leads to faster execution. In VMC calculations, set substeps to the average autocorrelation time of the desired quantity.
time step: (default 0.1) Electronic time step in bohr.
samples: (default 0) Number of walker configurations saved during the current calculation.
walkers: (default #threads) In VMC, sets the number of walkers per node. The total number of walkers in the calculation will be equal to walkers*(# nodes).
Options unique to DMC:
targetWalkers: (default #walkers from previous calculation, e.g., VMC). Sets the target number of walkers. The actual population of walkers will fluctuate around this value. The walkers will be distributed across all the nodes in the calculation. On a given node, the walkers are split across all the threads in the system.
nonlocalmoves: (default “no”) Set to “yes” to turns on the use of Casula’s T-moves.
Appendix E: Wavefunction XML block¶
<wavefunction name="psi0" target="e">
<determinantset type="MolecularOrbital" name="LCAOBSet"
source="ion0" transform="yes">
<basisset name="LCAOBSet">
<atomicBasisSet name="Gaussian-G2" angular="cartesian" type="Gaussian" elementType="O" normalized="no">
...
</atomicBasisSet>
</basisset>
<slaterdeterminant>
<determinant id="updet" size="4">
<occupation mode="ground"/>
<coefficient size="57" id="updetC">
...
</coefficient>
</determinant>
<determinant id="downdet" size="4">
<occupation mode="ground"/>
<coefficient size="57" id="downdetC">
...
</coefficient>
</determinant>
</slaterdeterminant>
</determinantset>
<jastrow name="J2" type="Two-Body" function="Bspline" print="yes">
...
</jastrow>
</wavefunction>
In this section we describe the basic format of a QMCPACK wavefunction XML block. Everything listed in this section is generated by the appropriate converter tools. Little to no modification is needed when performing standard QMC calculations. As a result, this section is meant mainly for illustration purposes. Only experts should attempt to modify these files (with very few exceptions like the cutoff of CI coefficients and the cutoff in Jastrow functions) since changes can lead to unexpected results.
A QMCPACK wavefunction XML block is a combination of a determinantset, which contains the antisymmetric part of the wavefunction and one or more Jastrow blocks. The syntax of the antisymmetric block depends on whether the wavefunction is a single determinant or a multideterminant expansion. Listing 62 shows the general structure of the single-determinant case. The determinantset block is composed of a basisset block, which defines the atomic orbital basis set, and a slaterdeterminant block, which defines the SPOs and occupation numbers of the Slater determinant. Listing 63 shows a (piece of a) sample of a slaterdeterminant block. The slaterdeterminant block consists of two determinant blocks, one for each electron spin. The parameter “size” in the determinant block refers to the number of SPOs present while the “size” parameter in the coefficient block refers to the number of atomic basis functions per SPO.
<slaterdeterminant>
<determinant id="updet" size="5">
<occupation mode="ground"/>
<coefficient size="134" id="updetC">
9.55471000000000e-01 -3.87000000000000e-04 6.51140000000000e-02 2.17700000000000e-03
1.43900000000000e-03 4.00000000000000e-06 -4.58000000000000e-04 -5.20000000000000e-05
-2.40000000000000e-05 6.00000000000000e-06 -0.00000000000000e+00 -0.00000000000000e+00
-0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00
-0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00
-0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00
-0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00
-0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00
-0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00
-0.00000000000000e+00 -5.26000000000000e-04 2.63000000000000e-04 2.63000000000000e-04
-0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00 -1.27000000000000e-04
6.30000000000000e-05 6.30000000000000e-05 -0.00000000000000e+00 -0.00000000000000e+00
-0.00000000000000e+00 -3.20000000000000e-05 1.60000000000000e-05 1.60000000000000e-05
-0.00000000000000e+00 -0.00000000000000e+00 -0.00000000000000e+00 7.00000000000000e-06
Listing 64 shows the general structure of the multideterminant case. Similar to the single-determinant case, the determinantset must contain a basisset block. This definition is identical to the one described previously. In this case, the definition of the SPOs must be done independently from the definition of the determinant configurations; the latter is done in the sposet block, while the former is done on the multideterminant block. Notice that two sposet sets must be defined, one for each electron spin. The name of each sposet set is required in the definition of the multideterminant block. The determinants are defined in terms of occupation numbers based on these orbitals.
<wavefunction id="psi0" target="e">
<determinantset name="LCAOBSet" type="MolecularOrbital" transform="yes" source="ion0">
<basisset name="LCAOBSet">
<atomicBasisSet name="Gaussian-G2" angular="cartesian" type="Gaussian" elementType="O" normalized="no">
...
</atomicBasisSet>
...
</basisset>
<sposet basisset="LCAOBSet" name="spo-up" size="8">
<occupation mode="ground"/>
<coefficient size="40" id="updetC">
...
</coefficient>
</sposet>
<sposet basisset="LCAOBSet" name="spo-dn" size="8">
<occupation mode="ground"/>
<coefficient size="40" id="downdetC">
...
</coefficient>
</sposet>
<multideterminant optimize="yes" spo_up="spo-up" spo_dn="spo-dn">
<detlist size="97" type="CSF" nca="0" ncb="0" nea="4" neb="4" nstates="8" cutoff="0.001">
<csf id="CSFcoeff_0" exctLvl="0" coeff="0.984378" qchem_coeff="0.984378" occ="22220000">
<det id="csf_0-0" coeff="1" alpha="11110000" beta="11110000"/>
</csf>
...
</detlist>
</multideterminant>
</determinantset>
<jastrow name="J2" type="Two-Body" function="Bspline" print="yes">
...
</jastrow>
</wavefunction>
There are various options in the multideterminant block that users should be aware of.
cutoff: (IMPORTANT! ) Only configurations with (absolute value) “qchem coeff” larger than this value will be read by QMCPACK.
optimize: Turn on/off the optimization of linear CI coefficients.
coeff: (in csf ) Current coefficient of given configuration. Gets updated during wavefunction optimization.
qchem coeff: (in csf ) Original coefficient of given configuration from GAMESS calculation. This is used when applying a cutoff to the configurations read from the file. The cutoff is applied on this parameter and not on the optimized coefficient.
nca and nab: Number of core orbitals for up/down electrons. A core orbital is an orbital that is doubly occupied in all determinant configurations, not to be confused with core electrons. These are not explicitly listed on the definition of configurations.
nea and neb: Number of up/down active electrons (those being explicitly correlated).
nstates: Number of correlated orbitals.
size (in detlist ): Contains the number of configurations in the list.
The remaining part of the determinantset block is the definition of Jastrow factor. Any
number of these can be defined. Listing 65 shows a sample Jastrow
block including 1-, 2- and 3-body terms. This is the standard block produced by
convert4qmc with the option -add3BodyJ (this particular example is for a water molecule).
Optimization of individual radial functions can be turned on/off using the “optimize”
parameter. It can be added to any coefficients block, even though it is currently not
present in the J1 and J2 blocks.
<jastrow name="J2" type="Two-Body" function="Bspline" print="yes">
<correlation rcut="10" size="10" speciesA="u" speciesB="u">
<coefficients id="uu" type="Array">0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0</coefficients>
</correlation>
<correlation rcut="10" size="10" speciesA="u" speciesB="d">
<coefficients id="ud" type="Array">0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0</coefficients>
</correlation>
</jastrow>
<jastrow name="J1" type="One-Body" function="Bspline" source="ion0" print="yes">
<correlation rcut="10" size="10" cusp="0" elementType="O">
<coefficients id="eO" type="Array">0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0</coefficients>
</correlation>
<correlation rcut="10" size="10" cusp="0" elementType="H">
<coefficients id="eH" type="Array">0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0</coefficients>
</correlation>
</jastrow>
<jastrow name="J3" type="eeI" function="polynomial" source="ion0" print="yes">
<correlation ispecies="O" especies="u" isize="3" esize="3" rcut="10">
<coefficients id="uuO" type="Array" optimize="yes">
</coefficients>
</correlation>
<correlation ispecies="O" especies1="u" especies2="d" isize="3" esize="3" rcut="10">
<coefficients id="udO" type="Array" optimize="yes">
</coefficients>
</correlation>
<correlation ispecies="H" especies="u" isize="3" esize="3" rcut="10">
<coefficients id="uuH" type="Array" optimize="yes">
</coefficients>
</correlation>
<correlation ispecies="H" especies1="u" especies2="d" isize="3" esize="3" rcut="10">
<coefficients id="udH" type="Array" optimize="yes">
</coefficients>
</correlation>
</jastrow>
This training assumes basic familiarity with the UNIX operating system. In particular, we use simple scripts written in “csh.” In addition, we assume you have obtained all the necessary files and executables and that the training files are located at ${TRAINING TOP}.
The goal of this training is not only to familiarize you with the execution and options in QMCPACK but also to introduce you to important concepts in QMC calculations and many-body electronic structure calculations.
Lab 4: Condensed Matter Calculations¶
Topics covered in this lab¶
Tiling DFT primitive cells into QMC supercells
Reducing finite-size errors via extrapolation
Reducing finite-size errors via averaging over twisted boundary conditions
Using the B-spline mesh factor to reduce memory requirements
Using a coarsely resolved vacuum buffer region to reduce memory requirements
Calculating the DMC total energies of representative 2D and 3D extended systems
Lab directories and files¶
labs/lab4_condensed_matter/
├── Be-2at-setup.py - DFT only for prim to conv cell
├── Be-2at-qmc.py - QMC only for prim to conv cell
├── Be-16at-qmc.py - DFT and QMC for prim to 16 atom cell
├── graphene-setup.py - DFT and OPT for graphene
├── graphene-loop-mesh.py - VMC scan over orbital bspline mesh factors
├── graphene-final.py - DMC for final meshfactor
└── pseudopotentials - pseudopotential directory
├── Be.ncpp - Be PP for Quantum ESPRESSO
├── Be.xml - Be PP for QMCPACK
├── C.BFD.upf - C PP for Quantum ESPRESSO
└── C.BFD.xml - C PP for QMCPACK
The goal of this lab is to introduce you to the somewhat specialized problems involved in performing DMC calculations on condensed matter as opposed to the atoms and molecules that were the focus of the preceding labs. Calculations will be performed on two different systems. Firstly, we will perform a series of calculations on BCC beryllium, focusing on the necessary methodology to limit finite-size effects. Secondly, we will perform calculations on graphene as an example of a system where QMCPACK’s capability to handle cases with mixed periodic and open boundary conditions is useful. This example will also focus on strategies to limit memory usage for such systems. All of the calculations performed in this lab will use the Nexus workflow management system, which vastly simplifies the process by automating the steps of generating trial wavefunctions and performing DMC calculations.
Preliminaries¶
For any DMC calculation, we must start with a trial wavefunction. As is typical for our calculations of condensed matter, we will produce this wavefunction using DFT. Specifically, we will use QE to generate a Slater determinant of SPOs. This is done as a three-step process. First, we calculate the converged charge density by performing a DFT calculation with a fine grid of k-points to fully sample the Brillouin zone. Next, a non-self- consistent calculation is performed at the specific k-points needed for the supercell and twists needed in the DMC calculation (more on this later). Finally, a wavefunction is converted from the binary representation used by QE to the portable hdf5 representation used by QMCPACK.
The choice of k-points necessary to generate the wavefunctions depends on both the supercell chosen for the DMC calculation and by the supercell twist vectors needed. Recall that the wavefunction in a plane-wave DFT calculation is written using Bloch’s theorem as:
where \(\vec{k}\) is confined to the first Brillouin zone of the cell chosen and \(u(\vec{r})\) is periodic in this simulation cell. A plane-wave DFT calculation stores the periodic part of the wavefunction as a linear combination of plane waves for each SPO at all k-points selected. The symmetry of the system allows us to generate an arbitrary supercell of the primitive cell as follows: Consider the set of primitive lattice vectors, \(\{ \mathbf{a}^p_1, \mathbf{a}^p_2, \mathbf{a}^p_3\}\). We may write these vectors in a matrix, \(\mathbf{L}_p\), the rows of which are the primitive lattice vectors. Consider a nonsingular matrix of integers, \(\mathbf{S}\). A corresponding set of supercell lattice vectors, \(\{\mathbf{a}^s_1, \mathbf{a}^s_2, \mathbf{a}^s_3\}\), can be constructed by the matrix product
If the primitive cell contains \(N_p\) atoms, the supercell will then contain \(N_s = |\det(\mathbf{S})| N_p\) atoms.
Now, the wavefunciton at any point in this new supercell can be related to the wavefunction in the primitive cell by finding the linear combination of primitive lattice vectors that maps this point back to the primitive cell:
where \(x, y, z\) are integers. Now the wavefunction in the supercell at point \(\vec{r}'\) can be written in terms of the wavefunction in the primitive cell at \(\vec{r}'\) as:
where \(\vec{k}\) is confined to the first Brillouin zone of the primitive cell. We have also chosen the supercell twist vector, which places a constraint on the form of the wavefunction in the supercell. The combination of these two constraints allows us to identify family of N k-points in the primitive cell that satisfy the constraints. Thus, for a given supercell tiling matrix and twist angle, we can write the wavefunction everywhere in the supercell by knowing the wavefunction a N k-points in the primitive cell. This means that the memory necessary to store the wavefunction in a supercell is only linear in the size of the supercell rather than the quadratic cost if symmetry were neglected.
Total energy of BCC beryllium¶
When performing calculations of periodic solids with QMC, it is essential to work with a reasonable size supercell rather than the primitive cells that are common in mean field calculations. Specifically, all of the finite-size correction schemes discussed in the morning require that the exchange-correlation hole be considerably smaller than the periodic simulation cell. Additionally, finite-size effects are lessened as the distance between the electrons in the cell and their periodic images increases, so it is advantageous to generate supercells that are as spherical as possible to maximize this distance. However, a competing consideration is that when calculating total energies we often want to extrapolate the energy per particle to the thermodynamic limit by means of the following formula in three dimensions:
This formula derived assuming the shape of the supercells is consistent (more specifically that the periodic distances scale uniformly with system size), meaning we will need to do a uniform tiling, that is, \(2\times2\times2\), \(3\times3\times3\), etc. As a \(3\times3\times3\) tiling is 27 times larger than the supercell and the practical limit of DMC is on the order of 200 atoms (depending on Z), sometimes it is advantageous to choose a less spherical supercell with fewer atoms rather than a more spherical one that is too expensive to tile.
In the case of a BCC crystal, it is possible to tile the one atom primitive cell to a cubic supercell only by doubling the number of electrons. This is the best possible combination of a small number of atoms that can be tiled and a regular box that maximizes the distance between periodic images. We will need to determine the tiling matrix S that generates this cubic supercell by solving the following equation for the coefficients of the S matrix:
We will now use Nexus to generate the trial wavefunction for this BCC beryllium.
Fortunately, the Nexus will handle determination of the proper k-vectors given the tiling matrix. All that is needed is to place the tiling matrix in the Be-2at-setup.py file. Now the definition of the physical system is
bcc_Be = generate_physical_system(
lattice = 'cubic',
cell = 'primitive',
centering = 'I',
atoms = 'Be',
constants = 3.490,
units = 'A',
net_charge = 0,
net_spin = 0,
Be = 2,
tiling = [[a,b,c],[d,e,f],[g,h,i]],
kgrid = kgrid,
kshift = (.5,.5,.5)
)
where the tiling line should be replaced with the preceding row major
tiling matrix. This script file will now perform a converged DFT
calculation to generate the charge density in a directory called
bcc-beryllium/scf and perform a non-self-consistend DFT calculation
to generate SPOs in the directory bcc-beryllium/nscf. Fortunately,
Nexus will calculate the required k-points needed to tile the
wavefunction to the supercell, so all that is necessary is the
granularity of the supercell twists and whether this grid is shifted
from the origin. Once this is finished, it performs the conversion from
pwscf’s binary format to the hdf5 format used by QMCPACK. Finally, it
will optimize the coefficients of 1-body and 2-body Jastrow factors in
the supercell defined by the tiling matrix.
Run these calculations by executing the script Be-2at-setup.py. You
will notice the small calculations required to generate the wavefunction
of beryllium in a one-atom cell are rather inefficient to run on a
high-performance computer such as vesta in terms of the time spent doing
calculations versus time waiting on the scheduler and booting compute
nodes. One of the benefits of the portable HDF format that is used by
QMCPACK is that you can generate data like wavefunctions on a local
workstation or other convenient resource and use high-performance
clusters for the more expensive QMC calculations.
In this case, the wavefunction is generated in the directory
bcc-beryllium/nscf-2at_222/pwscf_ output in a file called
pwscf.pwscf.h5. For debugging purposes, it can be useful to verify
that the contents of this file are what you expect. For instance, you
can use the tool h5ls to check the geometry of the cell where the
DFT calculations were performed or the number of k-points or electrons
in the calculation. This is done with the command h5ls -d
pwscf.pwscf.h5/supercell or h5ls -d pwscf.pwscf.h5/electrons.
In the course of running Be-2at-setup.py, you will get an error when
attempting to perform the VMC and wavefunction optimization
calculations. This is because the wavefunction has generated supercell
twists of the form (+/- 1/4, +/- 1/4, +/- 1/4). In the case that the
supercell twist contains only 0 or 1/2, it is possible to operate
entirely with real arithmetic. The executable that has been indicated in
Be-2at-setup.py was compiled for this case. Note that where
possible, the memory use is a factor of two less than the general case
and the calculations are somewhat faster. However, it is often necessary
to perform calculations away from these special twist angles to reduce
finite-size effects. To fix this, delete the directory
bcc-beryllium/opt-2at, change the line near the top of
Be-2at-setup.py from
qmcpack = '/soft/applications/qmcpack/Binaries/qmcpack'
to
qmcpack = '/soft/applications/qmcpack/Binaries/qmcpack_comp'
and rerun the script.
When the optimization calculation has finished, check that everything has proceeded correctly by looking at the output in the opt-2at directory. Firstly, you can grep the output file for Delta to see if the cost function has indeed been decreasing during the optimization. You should find something like this:
OldCost: 4.8789147e-02 NewCost: 4.0695360e-02 Delta Cost:-8.0937871e-03
OldCost: 3.8507795e-02 NewCost: 3.8338486e-02 Delta Cost:-1.6930674e-04
OldCost: 4.1079105e-02 NewCost: 4.0898345e-02 Delta Cost:-1.8076319e-04
OldCost: 4.2681333e-02 NewCost: 4.2356598e-02 Delta Cost:-3.2473514e-04
OldCost: 3.9168577e-02 NewCost: 3.8552883e-02 Delta Cost:-6.1569350e-04
OldCost: 4.2176276e-02 NewCost: 4.2083371e-02 Delta Cost:-9.2903058e-05
OldCost: 4.3977361e-02 NewCost: 4.2865751e-02 Delta Cost:-1.11161830-03
OldCost: 4.1420944e-02 NewCost: 4.0779569e-02 Delta Cost:-6.4137501e-04
which shows that the starting wavefunction was fairly good and that most
of the optimization occurred in the first step. Confirm this by using
qmca to look at how the energy and variance changed over the course
of the calculation with the command: qmca -q ev -e 10 *.scalar.dat
executed in the opt-2at directory. You should get output like the
following:
LocalEnergy Variance ratio
opt series 0 -2.159139 +/- 0.001897 0.047343 +/- 0.000758 0.0219
opt series 1 -2.163752 +/- 0.001305 0.039389 +/- 0.000666 0.0182
opt series 2 -2.160913 +/- 0.001347 0.040879 +/- 0.000682 0.0189
opt series 3 -2.162043 +/- 0.001223 0.041183 +/- 0.001250 0.0190
opt series 4 -2.162441 +/- 0.000865 0.039597 +/- 0.000342 0.0183
opt series 5 -2.161287 +/- 0.000732 0.039954 +/- 0.000498 0.0185
opt series 6 -2.163458 +/- 0.000973 0.044431 +/- 0.003583 0.0205
opt series 7 -2.163495 +/- 0.001027 0.040783 +/- 0.000413 0.0189
Now that the optimization has completed successfully, we can perform DMC
calculations. The first goal of the calculations will be to try to
eliminate the 1-body finite-size effects by twist averaging. The script
Be-2at-qmc.py has the necessary input. Note that on line 42 two
twist grids are specified, (2,2,2) and (3,3,3). Change the tiling matrix
in this input file as in Be-2at-qmc.py and start the calculations.
Note that this workflow takes advantage of QMCPACK’s capability to group
jobs. If you look in the directory dmc-2at_222 at the job submission
script (dmc.qsub.in), you will note that rather than operating on an
XML input file, qmcapp is targeting a text file called dmc.in.
This file is a simple text file that contains the names of the eight XML
input files needed for this job, one for each twist. When operated in
this mode, QMCPACK will use MPI groups to run multiple copies of itself
within the same MPI context. This is often useful both in terms of
organizing calculations and for taking advantage of the large job sizes
that computer centers often encourage.
The DMC calculations in this case are designed to complete in a few
minutes. When they have finished running, first look at the
scalar.dat files corresponding to the DMC calculations at the
various twists in dmc-2at_222. Using a command such as
qmca -q ev -e 32 *.s001.scalar.dat (with a suitably chosen number of
blocks for the equilibration), you will see that the DMC energy in each
calculation is nearly identical within the statistical uncertainty of
the calculations. In the case of a large supercell, this is often
indicative of a situation where the Brillouin zone is so small that the
1-body finite-size effects are nearly converged without any twist
averaging. In this case, however, this is because of the symmetry of the
system. For this cubic supercell, all of the twist angles chosen in this
shifted \(2\times2\times2\) grid are equivalent by symmetry. In the
case where substantial resources are required to equilibrate the DMC
calculations, it can be beneficial to avoid repeating such twists and
instead simply weight them properly. In this case, however, where the
equilibration is inexpensive, there is no benefit to adding such
complexity as the calculations can simply be averaged together and the
result is equivalent to performing a single longer calculation.
Using the command qmc -a -q ev -e 16 *.s001.scalar.dat, average the
DMC energies in dmc-2at_222 and dmc-2at_333 to see whether the
1-body finite-size effects are converged with a \(3\times3\times3\)
grid of twists. When using beryllium as a metal, the convergence is
quite poor (0.025 Ha/Be or 0.7 eV/Be). If this were a production
calculation it would be necessary to perform calculations on much larger
grids of supercell twists to eliminate the 1-body finite-size effects.
In this case there are several other calculations that would warrant a
high priority. Script Be-16at-qmc.py has been provided in which you
can input the appropriate tiling matrix for a 16-atom cell and perform
calculations to estimate the 2-body finite-size effects, which will also
be quite large in the 2-atom calculations. This script will take
approximately 30 minutes to run to completion, so depending on your
interest, you can either run it or work to modify the scripts to address
the other technical issues that would be necessary for a production
calculation such as calculating the population bias or the time step
error in the DMC calculations.
Another useful exercise would be to attempt to validate this PP by calculating the ionization potential and electron affinity of the isolated atom and compare it with the experimental values: IP = 9.3227 eV , EA = 2.4 eV.
Handling a 2D system: graphene¶
In this section we examine a calculation of an isolated sheet of
graphene. Because graphene is a 2D system, we will take advantage of
QMCPACK’s capability to mix periodic and open boundary conditions to
eliminate and spurious interaction of the sheet with its images in the z
direction. Run the script graphene-setup.py, which will generate the
wavefunction and optimize one and two body jastrow factors. In the
script; notice line 160: bconds = ’ppn’ in the generate_qmcpack
function, which specifies this mix of open and periodic boundary
conditions. Consequently, the atoms will need to be kept away from this
open boundary in the z direction as the electronic wavefunction will not
be defined outside of the simulation box in this direction. For this
reason, all of the atom positions at the beginning of the file have z
coordinates 7.5. At this point, run the script graphene-setup.py.
Aside from the change in boundary conditions, the main thing that
distinguishes this kind of calculation from the previous beryllium
example is the large amount of vacuum in the cell. Although this is a
very small calculation designed to run quickly in the tutorial, in
general a more converged calculation would quickly become memory limited
on an architecture like BG/Q. When the initial wavefunction optimization
has completed to your satisfaction, run the script
graphene-loop-mesh.py. This examines within VMC an approach to
reducing the memory required to store the wavefunction. In
graphene-loop-mesh.py, the spacing between the B-spline points is
varied uniformly. The mesh spacing is a prefactor to the linear spacing
between the spline points, so the memory use goes as the cube of the
meshfactor. When you run the calculations, examine the
.s000.scalar.dat files with qmca to determine the lowest
possible mesh spacing that preserves both the VMC energy and the
variance.
Finally, edit the file graphene-final.py, which will perform two DMC
calculations. In the first, (qmc1) replace the following lines:
meshfactor = xxx,
precision = '---',
with the values you have determined will perform the calculation with as small as possible wavefunction. Note that we can also use single precision arithmetic to store the wavefunction by specifying precision=`single.’ When you run the script, compare the output of the two DMC calculations in terms of energy and variance. Also, see if you can calculate the fraction of memory that you were able to save by using a meshfactor other than 1 and single precision arithmetic.
Conclusion¶
Upon completion of this lab, you should be able to use Nexus to perform DMC calculations on periodic solids when provided with a PP. You should also be able to reduce the size of the wavefunction in a solid-state calculation in cases where memory is a limiting factor.
Lab 5: Excited state calculations¶
Topics covered in this lab¶
Tiling DFT primitive cells into optimal QMC supercells
Fundamentals of between neutral and charged calculations
Calculating quasiparticle excitation energies of condensed matter systems
Calculating optical excitation energies of condensed matter systems
Lab directories and files¶
labs/lab5_excited_properties/
├── band.py - Band structure calculation for Carbon Diamond
├── optical.py - VMC optical gap calculation using the tiling matrix from band.py
├── quasiparticle.py - VMC quasiparticle gap calculation using the tiling matrix from band.py
└── pseudopotentials - pseudopotential directory
├── C.BFD.upf - C PP for Quantum ESPRESSO
└── C.BFD.xml - C PP for QMCPACK
The goal of this lab is to perform neutral and charged excitation calculations in condensed matter systems using QMCPACK. Throughout this lab, a working knowledge of Lab4 Condensed Matter Calculations is assumed. First, we will introduce the concepts of neutral and charged excitations. We will briefly discuss these in relation to the specific experimental studies that must be used to benchmark DMC results. Secondly, we will perform charged (quasiparticle) and neutral (optical) excitations calculations on C-diamond.
Basics and excited state experiments¶
Although VMC and DMC methods are better suited for studying ground state properties of materials, they can still provide useful information regarding the excited states. Unlike the applications of band structure theory such as DFT and GW, it is more challenging to obtain the complete excitation spectra using DMC. However, it is relatively straightforward to calculate the band gap minimum of a condensed matter system using DMC.
We will briefly discuss the two main ways of obtaining the band gap minimum through experiments: photoemission and absorption studies. The energy required to remove an electron from a neutral system is called the IP (ionization potential), which is available from direct photoemission experiments. In contrast, the emission energy of a negatively charged system (or the energy required to convert a negatively charged system to a neutral system), known as electron affinity (EA), is available from inverse photoemission experiments. Outlines of these experiments are shown in Fig. 22.
Direct and inverse photoemission experiments involve charged excitations, whereas optical absorption experiments involve excitations that are just enough to be excited to the conduction band. From [ORR02]¶
Following the explanation in the previous paragraph and Fig. 22, the quasiparticle band gap of a material can be defined as:
where \(N\) is the number of electrons in the neutral system and \(E_{N}\) is the ground state energy of the neutral system. CBM and VBM stand for the conduction band minimum and valence band maximum, respectively. K’ can formally be arbitrary at the infinite limit. However, in practical calculations, a supertwist that accommodates both CBM and VBM can be more efficient in terms of computational time and systematic finite-size error cancellation. In the literature, the quasiparticle gap is also called the electronic gap. The term electronic comes from the fact that in both photoemission experiments, it is assumed that the perturbed electron is not interacting with the sample.
Additionally, absorption experiments can be performed in which electrons are perturbed at relatively lower energies, just enough to be excited into the conduction band. In absorption experiments, electrons are perturbed at lower energies. Therefore, they are not completely free and the system is still considered neutral. Since a quasihole and quasielectron are formed simultaneously, a bound state is created, unlike the free electron in the quasiparticle gap as described previously. This process is also known as optical excitation, which is schematically shown in Fig. 22, under “Absorption.” The optical gap can be formulated as follows:
where \(E^{K_1 {\rightarrow} K_2}\) is the energy of the system when a valence electron at wavevector \(K_1\) is promoted to the conduction band at wavevector \(K_2\). Therefore, the \(E_g^{K_1 {\rightarrow} K_2}\) is called the optical gap for promoting an electron at \(K_1\) to \(K_2\). If both CBM and VBM are on the same k-vector then the material is called direct band gap since it can directly emit photons without any external perturbation (phonons). However, if CBM and VBM share different k-vectors, then the photon-emitting electron has to transfer some of its momenta to the crystal lattice and then decay to the ground state. As this process involves an intermediate step, this property is called the indirect band gap. The difference between the optical and electronic band gaps is called the exciton binding energy. Exciton binding energy is very important for optoelectronic applications such as lasers. Since the recombination usually occurs between free holes and free electrons, a bound electron and hole state means that the spectrum of emission energies will be narrower. In the examples that follow, we will investigate the optical excitations of C-diamond.
Preparation for the excited state calculations¶
In this section, we will study the preparation steps to perform excited state calculations with QMC. Here, the most basic steps are listed in the implementation order:
Identify the high-symmetry k-points of the standardized primitive cell.
Perform DFT band structure calculation along high-symmetry paths.
Find a supertwist that includes all the k-points of interest.
Identify the indexing of k-points in the supertwist to be used in QMCPACK.
Identifying high-symmetry k-points¶
Primitive cell is the most basic, nonunique repeat unit of a crystal in real space. However, the translations of the repeat unit, the Bravais lattice is unique for each crystal and can be represented using discrete translation operations, \(R_n\):
\(a_n\) are the real-space lattice vectors in three dimensions. Thanks to the periodicity of the Bravais lattice, a crystal can also be represented using periodic functions in the reciprocal space:
where \(G_m\) are called as the reciprocal lattice vectors. (85) also satisfies the equality \(G_m\cdot{R_n}=2{\pi}N\). High-symmetry structures can be represented using a subspace of the BZ, which is called as the irreducible Brillouin Zone (iBZ). If we choose a series of paths of high-symmetry k-points that encapsulates the iBZ, we can determine the band gap and electronic structure of the material. For more discussion, please refer to any solid-state physics textbook.
There are multiple practical ways to find the high-symmetry k-point path. For example, pymatgen, [ORJ+13] XCRYSDEN [Kok99] or SeeK-path [HPK+17] can be used.
Fig. 23 shows the procedure for visualizing the Brillouin Zone using XCRYSDEN after the structure file is loaded. However, the primitive cell is not unique, and the actual shape of the BZ can depend on the structure used. In our example, we use the Python libraries of SeeK-path, using a wrapper written in Nexus.
Visualizing the Brillouin Zone using XCRYSDEN.¶
SeeK-path includes routines to standardize primitive cells, which will be useful for our work.
SeeK-path can be installed easily using pip:
>pip install --user seekpath
In the band.py script, identification of high-symmetry k-points and band structure calculations are done within the workflow.
In the script, where the dia PhysicalSystem object is used as the input structure, dia2_structure is the standardized primitive cell and dia2_kpath is the respective k-path around the iBZ.
dia2_kpath has a dictionary of the k-path in various coordinate systems; please make sure you are using the right one.
from structure import get_primitive_cell, get_kpath
dia2_structure = get_primitive_cell(structure=dia.structure)['structure']
dia2_kpath = get_kpath(structure=dia2_structure)
DFT band structure calculation along high-symmetry paths¶
After the high-symmetry k-points are identified, band structure calculations can be performed in DFT.
For an insulating structure, DFT can provide VBM and CBM wavevectors, which would be of interest to the DMC calculations.
However, if available, CBM and VBM from DFT would need to be compared with the experiments.
Basically, band.py will do the following:
Perform an SCF calculation in QE using a high-density reciprocal grid.
Identify the high-symmetry k-points on the iBZ and provide a k-path.
Perform a “band” calculation in QE, explicitly writing all the k-points on the path. (Make sure to add extra unoccupied bands.)
Plot the band structure curves and location of VBM/CBM if available.
In Fig. 24, C-diamond is shown to have an indirect
band gap between the red and green dots (CBM and VBM, respectively). VBM
is located at \(\Gamma\). CBM is not located on a high-symmetry
k-point in this case. Therefore, we can use the symbol \(\Delta\) to
denote the CBM wavevector in the rest of this document. In band.py
script, once the band structure calculation is finished, you can use the
following lines to get the exact location of VBM and CBM using
p = band.load_analyzer_image()
print "VBM:\n{0}".format(p.bands.vbm)
print "CBM:\n{0}".format(p.bands.cbm)
Output must be the following:
VBM:
band_number = 3
energy = 13.2874
index = 0
kpoint_2pi_alat = [0. 0. 0.]
kpoint_rel = [0. 0. 0.]
pol = up
CBM:
band_number = 4
energy = 17.1545
index = 51
kpoint_2pi_alat = [0. 0.1095605 0. ]
kpoint_rel = [0.3695652 0. 0.3695652]
pol = up
Band structure calculation of C-diamond performed at the DFT-LDA level. CBMs are shown with red points, and the VBMs are shown with the green points, both at \(\Gamma\). DFT-LDA calculations suggest that the material has an indirect band gap from \(\Gamma\rightarrow{\Delta}\). However, \(\Gamma\rightarrow{\Gamma}\) transition can also be investigated for more complete check.¶
DFT band structure calculation along high-symmetry paths¶
After the high-symmetry k-points are identified, band structure calculations can be performed in DFT.
For an insulating structure, DFT can provide VBM and CBM wavevectors, which would be of interest to the DMC calculations.
However, if available, CBM and VBM from DFT would need to be compared with the experiments.
Basically, band.py will do the following:
Perform an SCF calculation in QE using a high-density reciprocal grid.
Identify the high-symmetry k-points on the iBZ and provide a k-path.
Perform a “band” calculation in QE, explicitly writing all the k-points on the path. (Make sure to add extra unoccupied bands.)
Plot the band structure curves and location of VBM/CBM if available.
In Fig. 24, C-diamond is shown to have an indirect band gap between the red and green dots (CBM and VBM, respectively). VBM is located at \(\Gamma\). CBM is not located on a high-symmetry k-point in this case. Therefore, we can use the symbol \(\Delta\) to denote the CBM wavevector in the rest of this document. In script, once the band structure calculation is finished, you can use the following lines to get the exact location of VBM and CBM using
p = band.load_analyzer_image()
print "VBM:\n{0}".format(p.bands.vbm)
print "CBM:\n{0}".format(p.bands.cbm)
Output must be the following:
VBM:
band_number = 3
energy = 13.2874
index = 0
kpoint_2pi_alat = [0. 0. 0.]
kpoint_rel = [0. 0. 0.]
pol = up
CBM:
band_number = 4
energy = 17.1545
index = 51
kpoint_2pi_alat = [0. 0.1095605 0. ]
kpoint_rel = [0.3695652 0. 0.3695652]
pol = up
Finding a supertwist that includes all the k-points of interest¶
Using the VBM and CBM wavevectors defined in the previous section, we now construct the supertwist, which will hopefully contain both VBM and CBM. In Fig. 25, we provide a simple example using 2D rectangular lattice. Let us assume that we are interested in the indirect transition, \(\Gamma \rightarrow X_1\). In Fig. 25 a, the first BZ of the primitive cell is shown as the square centered on \(\Gamma\), which is drawn using dashed lines. Because of the periodicity of the lattice, this primitive cell BZ repeats itself with spacings equal to the reciprocal lattice vectors: (2\(\pi\)/a, 0) and (0, 2\(\pi\)/a) or (1,0) and (0,1) in crystal coordinates. We are interested in the first BZ, where \(X_1\) is at (0,0.5). In Fig. 25 b, the first BZ of the \(2\times2\) supercell is the smaller square, drawn using solid lines. In Fig. 25 c, the BZ of the \(2\times2\) supercell also repeats in the space, similar to Fig. 25 a. Therefore, in the \(2\times2\) supercell, \(X_1\), \(X_2\), and \(R\) are only the periodic images of \(\Gamma\). The \(2\times2\) supercell calculation can be performed in reciprocal space using a [2,2] tiling matrix. Therefore, individual k-points (twists) of the primitive cell are combined in the supercell calculation, which are then called as supertwists. In more complex primitive cells (hence BZ), more general criteria would be constructing a set of supercell reciprocal lattice vectors that contain the \(\Gamma \rightarrow X_1\) (e.g., \(G_1\) in Fig. 25) vector within their convex hull. Under this constraint, the Wigner-Seitz radius of the simulation cell can be maximized in an effort to reduce finite-size errors.
a) First BZ of the primitive cell centered on \(\Gamma\). Dashed lines indicate zone boundaries. b) First BZ of the \(2\times2\) supercell inside the first BZ of the primitive cell. First BZ boundaries of the supercell are shown using solid lines. c) Periodic translations of the first BZ of the supercell showing that \(\Gamma\) and \(X_1\) are periodic images of each other given the supercell BZ.¶
For the case of the indirect band gap in Diamond, several approximations might be needed to generate a supertwist that corresponds to a reasonable simulation cell. In the Diamond band gap, \(\Delta\) is at . In your calculations, the \(\Delta\) wavevector and the eigenvalues you find can be slightly different in value. The closest simple fraction to this number with the smallest denominator is 1/3. If we use \(\Delta'=[1/3, 0., 1/3]\), we could use a \(3\times1\times3\) supercell as the simple choice and include both \(\Delta'\) and \(\Gamma\) in the same supertwist exactly. Near \(\Delta\), the LDA band curvature is very low and using \(\Delta'\) can be a good approximation. We can compare the eigenvalues using their index numbers:
>>> print p.bands.up[51] ## CBM, $\Delta$ ##
eigs = [-3.2076 4.9221 7.5433 7.5433 17.1545 19.7598 28.3242 28.3242]
index = 51
kpoint_2pi_alat = [0. 0.1095605 0. ]
kpoint_rel = [0.3695652 0. 0.3695652]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
>>> print p.bands.up[46] ## $\Delta'$ ##
eigs = [-4.0953 6.1376 7.9247 7.9247 17.1972 20.6393 27.3653 27.3653]
index = 46
kpoint_2pi_alat = [0. 0.0988193 0. ]
kpoint_rel = [0.3333333 0. 0.3333333]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
This shows that the eigenvalues of the first unoccupied bands in \(\Delta\) and \(\Delta'\) are 17.1545 and 17.1972 eV, respectively, meaning that according to LDA, a correction of nearly -40 meV is obtained. After electronic transitions between \(\Gamma\) and \(\Delta'\) are studied using DMC, the LDA correction can be applied to extrapolate the results to \(\Gamma\) and \(\Delta\) transitions.
Identifying the indexing of k-points of interest in the supertwist¶
At this stage, we must have performed an scf calculation using a
converged k-point grid and then an nscf calculation using the
supertwist k-points given previously. We will be using the orbitals from
neutral DFT calculations; therefore, we need to explicitly define the
band and twist indexes of the excitations in QMCPACK (e.g., to define
electron promotion). In C-diamond, we can give an example by finding the
band and twist indexes of \(\Gamma\) and \(\Delta'\). For this
end, a mock VMC calculation can be run and the einspline.tile_300010003 .spin_0.tw_0.g0.bandinfo.dat file read. The Einspline
file prints out the eigenstates information from DFT calculations.
Therefore, we can obtain the band and the state index from this file,
which can later be used to define the electron promotion. You can see in
the following an explanation of how the band and twist indexes are
defined using a portion of the einspline.tile_300010003.spin_0.tw_0.g0.bandinfo.dat file. Spin_0 in the file name suggests
that we are reading the spin-up eigenstates. Band, state, twistindex,
and bandindex numbers all start from zero. We know we have 72 electrons
in the simulation cell, with 36 of them spin-up polarized. Since the
state number starts from zero, state number 35 must be occupied while
state 36 should be unoccupied. States 35 and 36 have the same reciprocal
crystal coordinates (K1,K2,K3) as \(\Gamma\) and \(\Delta'\),
respectively. Therefore, an electron should be promoted from state
number 35 to 36 to study the indirect band gap here.
# Band State TwistIndex BandIndex Energy Kx Ky Kz K1 K2 K3 KmK
33 33 0 1 0.488302 0.0000 0.0000 0.0000 -0.0000 -0.0000 -0.0000 1
34 34 0 2 0.488302 0.0000 0.0000 0.0000 -0.0000 -0.0000 -0.0000 1
35 35 0 3 0.488302 0.0000 0.0000 0.0000 -0.0000 -0.0000 -0.0000 1
36 36 4 4 0.631985 0.0000 -0.6209 0.0000 -0.3333 -0.0000 -0.3333 1
37 37 8 4 0.631985 0.0000 -1.2418 0.0000 -0.6667 -0.0000 -0.6667 1
38 38 0 4 0.691907 0.0000 0.0000 0.0000 -0.0000 -0.0000 -0.0000 1
However, we should always check whether this is really what we want. It can be seen that bands 33, 34, and 35 are degenerate (energy eigenvalues are listed in the 5th column), but they also have the same reciprocal coordinates in (K1,K2,K3). This is actually expected as one can see from Fig. 24, in the band diagram the band structure is threefold degenerate at \(\Gamma\). Here, we can choose the state with the largest band index: (0,3). Following the (twistindex, bandindex) notation, we can say that \(\Gamma\) to \(\Delta'\) transition can be defined as from (0,3) to (4,4).
Alternatively, we can also read the band and twist indexes using PwscfAnalyzer and determine the band/twist indexes on the go:
p = nscf.load_analyzer_image()
print 'band information'
print p.bands.up
print 'twist 0 k-point:',p.bands.up[0].kpoint_rel
print 'twist 4 k-point:',p.bands.up[4].kpoint_rel
print 'twist 0 band 3 eigenvalue:',p.bands.up[0].eigs[3]
print 'twist 4 band 4 eigenvalue:',p.bands.up[4].eigs[4]
Giving output:
0
eigs = [-8.0883 13.2874 13.2874 13.2874 18.8277 18.8277 18.8277 25.9151]
index = 0
kpoint_2pi_alat = [0. 0. 0.]
kpoint_rel = [0. 0. 0.]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
1
eigs = [-5.0893 3.8761 10.9518 10.9518 21.5031 21.5031 21.5361 28.2574]
index = 1
kpoint_2pi_alat = [-0.0494096 0.0494096 0.0494096]
kpoint_rel = [0.3333333 0. 0. ]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
2
eigs = [-5.0893 3.8761 10.9518 10.9518 21.5031 21.5031 21.5361 28.2574]
index = 2
kpoint_2pi_alat = [-0.0988193 0.0988193 0.0988193]
kpoint_rel = [0.6666667 0. 0. ]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
3
eigs = [-5.0893 3.8761 10.9518 10.9518 21.5031 21.5031 21.5361 28.2574]
index = 3
kpoint_2pi_alat = [ 0.0494096 0.0494096 -0.0494096]
kpoint_rel = [0. 0. 0.3333333]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
4
eigs = [-4.0954 6.1375 7.9247 7.9247 17.1972 20.6393 27.3652 27.3652]
index = 4
kpoint_2pi_alat = [0. 0.0988193 0. ]
kpoint_rel = [0.3333333 0. 0.3333333]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
5
eigs = [-0.6681 2.3791 3.7836 8.5596 19.3423 26.2181 26.6666 28.0506]
index = 5
kpoint_2pi_alat = [-0.0494096 0.1482289 0.0494096]
kpoint_rel = [0.6666667 0. 0.3333333]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
6
eigs = [-5.0893 3.8761 10.9518 10.9518 21.5031 21.5031 21.5361 28.2574]
index = 6
kpoint_2pi_alat = [ 0.0988193 0.0988193 -0.0988193]
kpoint_rel = [0. 0. 0.6666667]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
7
eigs = [-0.6681 2.3791 3.7836 8.5596 19.3423 26.2181 26.6666 28.0506]
index = 7
kpoint_2pi_alat = [ 0.0494096 0.1482289 -0.0494096]
kpoint_rel = [0.3333333 0. 0.6666667]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
8
eigs = [-4.0954 6.1375 7.9247 7.9247 17.1972 20.6393 27.3652 27.3652]
index = 8
kpoint_2pi_alat = [0. 0.1976385 0. ]
kpoint_rel = [0.6666667 0. 0.6666667]
occs = [1. 1. 1. 1. 0. 0. 0. 0.]
pol = up
twist 0 k-point: [0. 0. 0.]
twist 4 k-point: [0.3333333 0. 0.3333333]
twist 0 band 3 eigenvalue: 13.2874
twist 4 band 4 eigenvalue: 17.1972
Quasiparticle (electronic) gap calculations¶
In quasiparticle calculations, it is essential to work with reasonably large sized supercells to avoid spurious “1/N effects.” Since quasiparticle calculations involve charged cells, large simulation cells ensure that the extra charge is diluted over the simulation cell. Coulombic interactions are conditionally convergent for neutral periodic systems, but they are divergent for the charged systems. A typical workflow for a quasiparticle calculation includes the following:
Run an SCF calculation in a neutral charged cell with QE using a high-density reciprocal grid.
Choose a tiling matrix that will at least approximately include VBM and CBM k-points.
Run ‘nscf’/‘p2q’ calculations using the tiling matrix.
Run VMC/DMC calculations for the neutral and positively and negatively charged cells in QMCPACK.
Check the convergence of the quasiparticle gap with respect to the simulation cell size.
<particleset name="e" random="yes">
<group name="u" size="36" mass="1.0"> ##Change size to 35
<parameter name="charge" > -1 </parameter>
<parameter name="mass" > 1.0 </parameter>
</group>
...
...
<determinantset>
<slaterdeterminant>
<determinant id="updet" group="u" sposet="spo_u" size="36"> ##Change size to 35
<occupation mode="ground" spindataset="0"/>
</determinant>
<determinant id="downdet" group="d" sposet="spo_d" size="36">
<occupation mode="ground" spindataset="1"/>
</determinant>
</slaterdeterminant>
</determinantset>
Going back to (83), we can see that it is essential to include VBM and CBM wavevectors in the same twist for quasiparticle calculations as well. Therefore, the added electron will sit at CBM while the subtracted electron will be removed from VBM. However, for the charged cell calculations, we may need to make changes in the input files for the fourth step. Alternatively, in the quasiparticle.py file, the changes in the QMC input are shown for a negatively charged system:
qmc.input.simulation.qmcsystem.particlesets.e.groups.u.size +=1
(qmc.input.simulation.qmcsystem.wavefunction.determinantset
.slaterdeterminant.determinants.updet.size += 1)
Here, the number of up electrons are increased by one (negatively charged system), and QMCPACK is instructed to read more one orbital in the up channel from the .h5 file.
QE uses symmetry to reduce the number of k-points required for the calculation.
Therefore, all symmetry tags in QE (nosym, noinv, and nosym_evc) must be set to false.
An easy way to check whether this is the case is to see that all KmK values einspline files are equal to 1.
Previously, the input for the neutral cell is given, while the changes are denoted as comments for the positively charged cell.
Note that we have used det_format = "old" in the vmc_+/-e.py files.
Optical gap calculations¶
Routines for the optical gap calculations are very similar to the quasiparticle gap calculations. The first three items in the quasiparticle band gap calculations can be reused for the optical gap calculations. However, at the VMC/DMC level, the electronic transitions performed should be explicitly stated. Therefore, compared with the quasiparticle calculations, only item number 4 is different for optical gap calculations. Here, the modified input file is given for the \(\Gamma\rightarrow\Delta'\) transition, which can be compared with the ground state input file in the previous section.
<determinantset>
<slaterdeterminant>
<determinant id="updet" group="u" sposet="spo_u" size="36">
<occupation mode="excited" spindataset="0" format="band" pairs="1" >
0 3 4 4
</occupation>
</determinant>
<determinant id="downdet" group="d" sposet="spo_d" size="36">
<occupation mode="ground" spindataset="1"/>
</determinant>
</slaterdeterminant>
</determinantset>
We have used the (twistindex, bandindex) notation in the annihilaion/creation order for the up-spin electrons.
After resubmitting the batch job, in the output, you should be able to see the following lines in the vmc.out file:
Sorting the bands now:
Occupying bands based on (ti,bi) data.
removing orbital 35
adding orbital 36
We will read 36 distinct orbitals.
There are 0 core states and 36 valence states.
And the einspline.tile_300010003.spin_0.tw_0.g0.bandinfo.dat file must be changed in the following way:
# Band State TwistIndex BandIndex Energy Kx Ky Kz K1 K2 K3 KmK
33 33 0 1 0.499956 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1
34 34 0 2 0.500126 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1
35 35 4 4 0.637231 0.0000 -0.6209 0.0000 -0.3333 0.0000 -0.3333 1
36 36 0 3 0.502916 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1
37 37 8 4 0.637231 0.0000 -1.2418 0.0000 -0.6667 0.0000 -0.6667 1
38 38 0 4 0.699993 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1
Alternatively, the excitations within Nexus can be defined as shown in the optical.py file:
qmc = generate_qmcpack(
...
excitation = ['up', '0 3 4 4'], # (ti, bi) notation
#excitation = ['up', '-35 + 36'], # Orbital (state) index notation
...
)
- HPK+17
Yoyo Hinuma, Giovanni Pizzi, Yu Kumagai, Fumiyasu Oba, and Isao Tanaka. Band structure diagram paths based on crystallography. Computational Materials Science, 128:140–184, 2017. URL: http://www.sciencedirect.com/science/article/pii/S0927025616305110, doi:10.1016/j.commatsci.2016.10.015.
- Kok99
Anton Kokalj. XCrySDen—a new program for displaying crystalline structures and electron densities. Journal of Molecular Graphics and Modelling, 17(3):176–179, 1999. URL: http://www.sciencedirect.com/science/article/pii/S1093326399000285, doi:10.1016/S1093-3263(99)00028-5.
- ORJ+13
Shyue Ping Ong, William Davidson Richards, Anubhav Jain, Geoffroy Hautier, Michael Kocher, Shreyas Cholia, Dan Gunter, Vincent L. Chevrier, Kristin A. Persson, and Gerbrand Ceder. Python Materials Genomics (pymatgen): A robust, open-source python library for materials analysis. Computational Materials Science, 68:314–319, 2013. URL: http://www.sciencedirect.com/science/article/pii/S0927025612006295, doi:10.1016/j.commatsci.2012.10.028.
- ORR02
Giovanni Onida, Lucia Reining, and Angel Rubio. Electronic excitations: density-functional versus many-body Green’s-function approaches. Reviews of Modern Physics, 74(2):601–659, 2002. doi:10.1103/RevModPhys.74.601.
AFQMC Tutorials¶
Below we will run through some full AFQMC workflow examples. The necessary scripts can be
found in the qmcpack/examples/afqmc directory.
Example 1: Neon atom¶
In this example we will go through the basic steps necessary to generate AFQMC input from a pyscf scf calculation on a simple closed shell molecule (neon/aug-cc-pvdz).
The pyscf scf script is given below (scf.py in the current directory):
from pyscf import gto, scf, cc
from pyscf.cc import ccsd_t
import h5py
mol = gto.Mole()
mol.basis = 'aug-cc-pvdz'
mol.atom = (('Ne', 0,0,0),)
mol.verbose = 4
mol.build()
mf = scf.RHF(mol)
mf.chkfile = 'scf.chk'
ehf = mf.kernel()
ccsd = cc.CCSD(mf)
ecorr_ccsd = ccsd.kernel()[0]
ecorr_ccsdt = ccsd_t.kernel(ccsd, ccsd.ao2mo())
print("E(CCSD(T)) = {}".format(ehf+ecorr_ccsd+ecorr_ccsdt))
The most important point above is that we create a scf checkpoint file by specifying the mf.chkfile mol member variable. Note we will also compute the CCSD and CCSD(T) energies for comparison puposes since this system is trivially small.
We next run the pyscf calculation using
python scf.py > scf.out
which will yield a converged restricted Hartree–Fock total energy of -128.496349730541 Ha, a CCSD value of -128.7084878405062 Ha, and a CCSD(T) value of -128.711294157 Ha.
The next step is to generate the necessary qmcpack input from this scf calculation. To this we do (assuming afqmctools is in your PYTHONPATH):
/path/to/qmcpack/utils/afqmctools/bin/pyscf_to_afqmc.py -i scf.chk -o afqmc.h5 -t 1e-5 -v
which will peform the necessary AO to MO transformation of the one and two electron integrals and perform a modified cholesky transormation of the two electron integrals. A full explanation of the various options available for pyscf_to_afqmc.py you can do
pyscf_to_afqmc.py -h
In the above example, -i designates the input pyscf checkpoint file, -o speficies the output filename to write the qmcpack hamiltonian/wavefunction to, -t specifies the convergence threshold for the Cholesky decomposition, -v increases verbosity. You can optionally pass the -q/–qmcpack-input to generate a qmcpack input file which is based on the hamiltonian and wavefunction generated. Greater control over input file generation can be achieved using the write_xml_input function provided with afqmctools. Run gen_input.py after the integrals/wavefunction have been generated to generate the input file afqmc.xml.
Running the above will generate one file: afqmc.h5. The plain text wavefunction files are deprecated and will be removed in later releases. The qmcpack input file afqmc.xml is a skeleton input file, meaning that it’s created from the information in hamil.h5 and is meant as a convenience, not as a guarantee that the convergeable parameters (timestep, walker number, bias bound etc. are converged or appropriate).
We will next run through the relevant sections of the input file afqmc.xml below:
<project id="qmc" series="0"/>
<random seed="7"/>
<AFQMCInfo name="info0">
<parameter name="NMO">23</parameter>
<parameter name="NAEA">5</parameter>
<parameter name="NAEB">5</parameter>
</AFQMCInfo>
We first specify how to name the output file. We also have fixed the random number seed so that the results of this tutorial can be reproducible (if run on the same number of cores).
Next comes the system description, which is mostly a sanity check, as these parameters will be read from the hamiltonian file. They specify the number of single-particle orbitals in the basis set (NMO) and the number of alpha (NAEA) and beta (NAEB) electrons respectively.
Next we specify the Hamiltonian and wavefunction to use:
<Hamiltonian name="ham0" info="info0">
<parameter name="filetype">hdf5</parameter>
<parameter name="filename">afqmc.h5</parameter>
</Hamiltonian>
<Wavefunction name="wfn0" type="NOMSD" info="info0">
<parameter name="filetype">hdf5</parameter>
<parameter name="filename">afqmc.h5</parameter>
</Wavefunction>
The above should be enough for most calculations. A NOMSD (non-orthogonal multi-Slater determinant) wavefunction allows for a generalised wavefunction input in the form of a single (or multiple) matrix (matrices) of molecular orbital coefficients for the RHF calculation we perform here.
We next set the walker options:
<WalkerSet name="wset0" type="shared">
<parameter name="walker_type">CLOSED</parameter>
</WalkerSet>
The important point here is that as we are using a RHF trial wavefunction we must specify that the walker_type is CLOSED. For a UHF trial wavefunction one would set this to COLLINEAR.
And now the propagator options:
<Propagator name="prop0" info="info0">
<parameter name="hybrid">yes</parameter>
</Propagator>
In the above we specify that we will be using the hybrid approach for updating the walker weights. If you wish to use the local energy approximation you should set this flag to false.
Finally comes the execute block which controls how the simulation is run:
<execute wset="wset0" ham="ham0" wfn="wfn0" prop="prop0" info="info0">
<parameter name="ncores">1</parameter>
<parameter name="timestep">0.01</parameter>
<parameter name="nWalkers">10</parameter>
<parameter name="blocks">100</parameter>
<parameter name="steps">10</parameter>
</execute>
The time step (timestep), number of Monte Carlo samples (blocks`*`steps), and number of walkers (nWalkers) should be adjusted as appropriate. Note that nWalkers sets the number of walkers per ncores. For example, if we wanted to use 100 walkers we could run the above input file on 10 cores. If the problem size is very large we may want distribute the workload over more cores per walker, say 10. In this case we would require 100 cores to maintain the same number of walkers. Typically in this case you want to specify fewer walkers per core anyway.
We can now run the qmcpack simulation:
qmcpack afqmc.xml > qmcpack.out
Assuming the calculation finishes successfully, the very first thing you should do is check the information in qmcpack.out to see confirm no warnings were raised. The second thing you should check is that the energy of the starting determinant matches the Hartree–Fock energy you computed earlier from pyscf to within roughly the error threshold you specified when generating the Cholesky decomposition. This check is not very meaningful if using, say, DFT orbitals. However if this energy is crazy it’s a good sign something went wrong with either the wavefunction or integral generation. Next you should inspect the qmc.scalar.s000.dat file which contains the mixed estimates for various quantities. This can be plotted using gnuplot. EnergyEstim__nume_real contains the block averaged values for the local energy, which should be the 7th column.
Assuming everything worked correctly we need to analyse the afqmc output using:
/path/to/qmcpack/nexus/bin/qmca -e num_skip -q el qmc.s000.scalar.dat
where num_skip is the number of blocks to skip for the equilibration stage. For a practical calculation you may want to use more walkers and run for longer to get meaningful statistics.
See the options for qmca for further information. Essentially we discarded the first 100 blocks as equilibaration and only computed the mixed estimate for the local energy internally called EnergyEstim__nume_real, which can be specified with -q el. We see that the ph-AFQMC energy agrees well with the CCSD(T) value. However, we probably did not run the simulation for long enough to really trust the error bars.
Example 2: Frozen Core¶
In this example we show how to perform a frozen core calculation, which only affects the integral generation step. We will use the the previous Neon example and freeze 2 core electrons. The following only currently works for RHF/ROHF trial wavefunctions.
mpirun -n 1 /path/to/qmcpack/utils/afqmctools/bin/pyscf_to_afqmc.py -i scf.chk -o afqmc.h5 -t 1e-5 -v -c 8,22
Again, run gen_input.py to generate the input file afqmc.xml.
Comparing the above to the previous example we see that we have added the -c or –cas option followed by a comma separated list of the form N,M defining a (N,M) CAS space containing 8 electrons in 22 spatial orbitals (freezing the lowest MO).
The rest of the qmcpack process follows as before.
Example 3: UHF Trial¶
In this example we show how to use a unrestricted Hartree–Fock (UHF) style wavefunction to find the ph-AFQMC (triplet) ground state energy of the carbon atom (cc-pvtz). Again we first run the scf (scf.py) calculation followed by the integral generation script:
mpirun -n 1 /path/to/qmcpack/utils/afqmctools/bin/pyscf_to_afqmc.py -i scf.chk -o afqmc.h5 -t 1e-5 -v -a
Note the new flag -a/–ao which tells the script to transform the integrals to an orthogonalised atomic orbital basis, rather that the more typical MO basis. This is necessary as qmcpack does not support spin dependent two electron integrals.
Running qmcpack as before should yield a mixed estimate for the energy of roughly: -37.78471 +/- 0.00014.
Example 4: NOMSD Trial¶
In this example we will show how to format trial different wavefunctions in such a way that qmcpack can read them.
Rather than use the pyscf_to_afqmc.py, script we will break up the process to allow for more flexibility and show what is going on under the hood.
The qmcpack input can be generated with the scf.py script. See the comments in scf.py for a breakdown of the steps involved.
Currently QMCPACK can deal with trial wavefunctions in two forms: Non-orthogonal multi slater determinant trial wavefunctions (NOMSD) and particle-hole style trial wavefunctions (PHMSD). The NOMSD trial wavefunctions are the most general form and expect Slater determinants in the form of M X N matrices of molecular orbital coefficients, where N is the number of electrons and M is the number of orbitals, along with a list of ci coefficients. Importantly the Slater determinants must be non-orthogonal.
Example 5: CASSCF Trial¶
In this example we will show how to format a casscf trial wavefunction.
Rather than use the pyscf_to_afqmc.py, script we will break up the process to allow for more flexibility and show what is going on under the hood.
The qmcpack input can be generated with the scf.py script followed by gen_input.py.
See the relevant code below for a breakdown of the steps involved.
The first step is to run a CASSCF calculation. Here we’ll consider N:sub:2. This replicates the calculations from Al-Saidi et al J. Chem. Phys. 127, 144101 (2007). They find a CASSCF energy of -108.916484 Ha, and a ph-AFQMC energy of -109.1975(6) Ha with a 97 determinant CASSCF trial.
mol = gto.M(atom=[['N', (0,0,0)], ['N', (0,0,3.0)]],
basis='cc-pvdz',
unit='Bohr')
nalpha, nbeta = mol.nelec
rhf = scf.RHF(mol)
rhf.chkfile = 'scf.chk'
rhf.kernel()
M = 12
N = 6
nmo = rhf.mo_coeff.shape[-1]
mc = mcscf.CASSCF(rhf, M, N)
mc.chkfile = 'scf.chk'
mc.kernel()
Next we unpack the wavefunction
nalpha = 3
nbeta = 3
ci, occa, occb = zip(*fci.addons.large_ci(mc.ci, M, (nalpha,nbeta),
tol=tol, return_strs=False))
and sort the determinants by the magnitude of their weight:
ixs = numpy.argsort(numpy.abs(coeff))[::-1]
coeff = coeff[ixs]
occa = numpy.array(occa)[ixs]
occb = numpy.array(occb)[ixs]
Next we reinsert the frozen core as the AFQMC simulation is not run using an active space:
core = [i for i in range(mc.ncore)]
occa = [numpy.array(core + [o + mc.ncore for o in oa]) for oa in occa]
occb = [numpy.array(core + [o + mc.ncore for o in ob]) for ob in occb]
Next we need to generate the one- and two-electron integrals. Note that we need to use the CASSCF MO coefficients to rotate the integrals.
scf_data = load_from_pyscf_chk_mol('scf.chk', 'mcscf')
write_hamil_mol(scf_data, 'afqmc.h5', 1e-5, verbose=True)
Finally we can write the wavefunction to the QMCPACK format:
ci = numpy.array(ci, dtype=numpy.complex128)
uhf = True # UHF always true for CI expansions.
write_qmcpack_wfn('afqmc.h5', (ci, occa, occb), uhf, mol.nelec, nmo)
To generate the input file we again run gen_input.py. Note the rediag option which
is necessary if the CI code used uses a different convention for ordering creation and
annihilation operations when defining determinant strings.
Example 6: Back Propagation¶
Note
matplotlib is required to generate the figure in this example.
The basic estimators printed out in the qmcpack *.scalar.dat files are mixed estimates. Unless the operator for which the mixed estimate is computed commutes with the Hamiltonian this result will generally be biased. To obtain pure estimates we can use back propagation as outlined in: Motta & Zhang, JCTC 13, 5367 (2017). For this example we will look at computing the one-body energy of a methane molecule (see Fig. 2 of M&Z).
As before run scf.py and generate the integrals using pyscf_to_afmqc.py:
mpirun -n 1 /path/to/qmcpack/utils/afqmctools/bin/pyscf_to_afqmc.py -i scf.chk -o afqmc.h5 -t 1e-5 -v
Note we are working in the MO basis. The input file is generated using gen_input.py and comparing to the previous examples we can now see the estimator block:
<Estimator name="back_propagation">
<parameter name="naverages">4</parameter>
<parameter name="block_size">2</parameter>
<parameter name="ortho">1</parameter>
<OneRDM />
<parameter name="nsteps">200</parameter>
</Estimator>
Which will tell QMCPACK to compute the back propagated one-rdm. In the above we set block_size to be 2 meaning that we average the back propagated estimates into bins of length 2 in this case. This helps reduce the size of the hdf5 files. We also specify the option nsteps: We see that it is set to 200, meaning that we will back propagated the bra wavefunction in the estimator by 200*.01 = 2 a.u., where the timestep has been set to 0.01 a.u. Finally naverages allows us to split the full path into naverages chunks, so we will have averaged data at \(\tau_{BP}=[0.5, 1.0, 1.5, 2.0]\) au. This allows us to monitor the convergence of the estimator with back propagation time.
Running QMCPACK as before we will notice that in addition to the qmc.s000.scalar.dat file we have generated a new file qmc.s000.scalar.h5. This file will contain the back propagated estimates, which, for the time being, means the back propagated one-particle reduced density matrix (1RDM), given as
Before we analyse the output we should question why we chose a back propagation time of 2 au. The back propagation time represents yet another parameter which must be carefully converged.
In this example we will show how this is done. In this directory you will find a script check_h1e_conv.py which shows how to use various helper scripts provided in afqmctools/analysis/average.py. The most of important of which are:
from afqmctools.analysis.extraction import get_metadata
metadata = get_metadata(filename)
which returns a dict containing the RDM metadata,
from afqmctools.analysis.average import average_one_rdm
rdm_av, rdm_errs = average_one_rdm(f, name='back_propagated', eqlb=3, ix=2)
which computes the average of the 1RDM, where ‘i’ specifies the index for the length of back propagation time desired (e.g. \(i=2 \rightarrow \tau_{BP} = 1.5\) au). eqlb is the equilibration time, and here we skip 10 blocks of length 2 au.
from afqmctools.analysis.extraction import extract_observable
dm = extract_observable(filename,
estimator='back_propagated'
name='one_rdm',
ix=2)
which extracts the 1RDM for all blocks and finally,
from afqmctools.analysis.extraction import extract_observable
dm, weights = extract_observable(filename,
estimator='back_propagated'
name='one_rdm',
ix=2,
sample=index)
which extracts a single density matrix for block index.
Have a look through check_h1e_conv.py and run it. A plot should be produced which shows the back propagated AFQMC one-body energy as a function of back propagation time, which converges to a value of roughly -78.888(1). This system is sufficiently small to perform FCI on. How does ph-AFQMC compare? Why are the error bars getting bigger with back propagation time?
Finally, we should mention that the path restoration algorithm introduced in M&Z is also implemented and can be turned on using the path_restoration parameter in the Estimator block.
In QMCPACK path restoration restores both the cosine projection and phase along the back propagation path. In general it was found in M&Z that path restoration always produced better results than using the standard back propagation algorithm, and it is recommended that it is always used. Does path restoration affect the results for methane?
Example 7: 2x2x2 Diamond supercell¶
In this example we will show how to generate the AFQMC input from a pbc pyscf calculation for a 2x2x2 supercell of diamond using a RHF trial wavefunction.
Again the first step is to run a pyscf calculation using the scf.py script in this directory.
The first part of the pyscf calculation is straightforward. See pyscf/examples/pbc for more examples on how to set up Hartree–Fock and DFT simulations.
import h5py
import numpy
import sys
from pyscf.pbc import scf, dft, gto
cell = gto.Cell()
cell.verbose = 5
alat0 = 3.6
cell.a = (numpy.ones((3,3))-numpy.eye(3))*alat0 / 2.0
cell.atom = (('C',0,0,0), ('C',numpy.array([0.25,0.25,0.25])*alat0))
cell.basis = 'gth-szv'
cell.pseudo = 'gth-pade'
cell.mesh = [28,28,28]
cell.build()
nk = [2,2,2]
kpts = cell.make_kpts(nk)
mf = scf.KRHF(cell, kpts=kpts)
mf.chkfile = 'scf.chk'
mf.kernel()
In addition to a standard pyscf calculation, we add the following lines:
from afqmctools.utils.linalg import get_ortho_ao
hcore = mf.get_hcore()
fock = (hcore + mf.get_veff())
X, nmo_per_kpt = get_ortho_ao(cell,kpts)
with h5py.File(mf.chkfile) as fh5:
fh5['scf/hcore'] = hcore
fh5['scf/fock'] = fock
fh5['scf/orthoAORot'] = X
fh5['scf/nmo_per_kpt'] = nmo_per_kpt
essentially, storing the fock matrix, core Hamiltonian and transformation matrix to the orthogonalised AO basis. This is currently required for running PBC AFQMC calculations.
Once the above (scf.py) script is run we will again use the pyscf_to_afqmc.py script to generate the necessary AFQMC input file.
mpirun -n 8 /path/to/qmcpack/utils/afqmctools/bin/pyscf_to_afqmc.py -i scf.chk -o afqmc.h5 -t 1e-5 -v -a
Note that the commands necessary to generate the integrals are identical to those for the molecular calculations, except now we accelerate their calculation using MPI. Note that if the number of tasks > number of kpoints then the number of MPI tasks must be divisible by the number of kpoints.
Once this is done we will again find a Hamiltonian file and afqmc input xml file. Inspecting these you will notice that their structure is identical to the molecular calculations seen previously. This is because we have not exploited k-point symmetry and are writing the integrals in a supercell basis. In the next example we will show how exploiting k-point symmetry can be done explicitly, which leads to a faster and lower memory algorithm for AFQMC.
Example 8: 2x2x2 Diamond k-point symmetry¶
In this example we will show how to run an AFQMC simulation that exploits k-point symmetry which is much more efficient that running in the supercell way discussed in the previous example. We will again look at the same 2x2x2 cell of diamond. We assume you have run the scf calculation in the previous example.
Essentially all that changes in the integral generation step is that we pass the -k/–kpoint flag to pyscf_to_afqmc.py.
mpirun -n 8 /path/to/qmcpack/utils/afqmctools/bin/pyscf_to_afqmc.py -i ../07-diamond_2x2x2_supercell/scf.chk -o afqmc.h5 -t 1e-5 -v -a -k
You will notice that now the Cholesky decomposition is done for each momentum transfer independently and the the form of the hamiltonian file has changed to be k-point dependent.
Apart from these changes, running the AFQMC simulation proceeds as before, however you should see a significant performance boost relative to the supercell simulations, particularly on GPU machines.
Additional Tools¶
QMCPACK provides a set of lightweight executables that address certain
common problems in QMC workflow and analysis. These range from conversion utilities between
different file formats and QMCPACK (e.g., ppconvert and convert4qmc),
(qmc-extract-eshdf-kvectors) to postprocessing utilities (trace-density and qmcfinitesize) to many others. In this section, we cover the use cases, syntax, and features of all additional tools provided with QMCPACK.
Postprocessing¶
qmca¶
qmca is a versatile tool to analyze and plot the raw data from QMCPACK *.scalar.dat files.
It is a Python executable and part of the Nexus suite of tools. It can be found in
qmcpack/nexus/executables. For details, see Using the qmca tool to obtain total energies and related quantities.
qmc-fit¶
qmc-fit is a curve fitting tool used to obtain statistical error bars on fitted parameters.
It is useful for DMC time step extrapolation. For details, see Using the qmc-fit tool for statistical time step extrapolation and curve fitting.
qdens¶
qdens is a command line tool to produce density files from QMCPACK’s stat.h5 output files. For details, see Using the qdens tool to obtain electron densities.
qmcfinitesize¶
qmcfinitesize is a utility to compute many-body finite-size corrections to the energy. It
is a C++ executable that is built alongside the QMCPACK executable. It can be found in
build/bin.
Converters¶
convert4qmc¶
Convert4qmc allows conversion of orbitals and wavefunctions from
quantum chemistry output files to QMCPACK XML and HDF5 input files.
It is a small C++ executable that is built alongside the QMCPACK
executable and can be found in build/bin.
To date, convert4qmc supports the following codes:
GAMESS [SBB+93], PySCF [SBB+18] and QP2 [GAG+19] natively, and NWCHEM [ApraBdJ+20], TURBOMOLE [FAH+14], PSI4 [TSP+12], CFOUR 2.0beta [MCH+20], ORCA 3.X - 4.X [Nee18], DALTON2016 [AAB+14], MOLPRO [WKK+12] and QCHEM 4.X [SGE+15] through the molden2qmc converter (see molden2qmc).
General use¶
General use of convert4qmc can be prompted by running with no options:
>convert4qmc
Defaults : -gridtype log -first 1e-6 -last 100 -size 1001 -ci required -threshold 0.01 -TargetState 0 -prefix sample
convert [-gaussian|-casino|-gamess|-orbitals]
filename
[-nojastrow -hdf5 -prefix title -addCusp -production -NbImages NimageX NimageY NimageZ]
[-psi_tag psi0 -ion_tag ion0 -gridtype log|log0|linear -first ri -last rf]
[-size npts -ci file.out -threshold cimin -TargetState state_number
-NaturalOrbitals NumToRead -optDetCoeffs]
Defaults : -gridtype log -first 1e-6 -last 100 -size 1001 -ci required
-threshold 0.01 -TargetState 0 -prefix sample
When the input format is missing, the extension of filename is used to determine
the format
*.Fchk -> gaussian; *.out -> gamess; *.data -> casino; *.h5 -> hdf5 format
As an example, to convert a GAMESS calculation using a single determinant, the following use is sufficient:
convert4qmc -gamess MyGamessOutput.out
By default, the converter will generate multiple files:
convert4qmcoutput:
output
file type
default
description
*.qmc.in-wfs.xmlXML
default
Main input file for QMCPACK
*.qmc.in-wfnoj.xmlXML
default
Main input file for QMCPACK
*.structure.xmlXML
default
File containing the structure of the system
*.wfj.xmlXML
default
Wavefunction file with 1-, 2-, and 3-body Jastrows
*.wfnoj.xmlXML
default
Wavefunction file with no Jastrows
*.orbs.h5HDF5
with -hdf5
HDF5 file containing all wavefunction data
If no -prefix option is specified, the prefix is taken from
the input file name. For instance, if the GAMESS output file is
Mysim.out, the files generated by convert4qmc will use the
prefix Mysim and output files will be
Mysim.qmc.in-wfs.xml, Mysim.structure.xml, and so on.
Files
.in-wfs.xmland.in-wfnoj.xmlThese are the input files for QMCPACK. The geometry and the wavefunction are stored in external files
*.structure.xmland*.wfj.xml(referenced from*.in-wfs.xml) or*.qmc.wfnoj.xml(referenced from*.qmc.in-wfnoj.xml). The Hamiltonian section is included, and the presence or lack of presence of an ECP is detected during the conversion. If use of an ECP is detected, a default ECP name is added (e.g.,H.qmcpp.xml), and it is the responsibility of the user to modify the ECP name to match the one used to generate the wavefunction.<?xml version="1.0"?> <simulation> <!-- Example QMCPACK input file produced by convert4qmc It is recommend to start with only the initial VMC block and adjust parameters based on the measured energies, variance, and statistics. --> <!--Name and Series number of the project.--> <project id="gms" series="0"/> <!--Link to the location of the Atomic Coordinates and the location of the Wavefunction.--> <include href="gms.structure.xml"/> <include href="gms.wfnoj.xml"/> <!--Hamiltonian of the system. Default ECP filenames are assumed.--> <hamiltonian name="h0" type="generic" target="e"> <pairpot name="ElecElec" type="coulomb" source="e" target="e" physical="true"/> <pairpot name="IonIon" type="coulomb" source="ion0" target="ion0"/> <pairpot name="PseudoPot" type="pseudo" source="ion0" wavefunction="psi0" format="xml"> <pseudo elementType="H" href="H.qmcpp.xml"/> <pseudo elementType="Li" href="Li.qmcpp.xml"/> </pairpot> </hamiltonian> The ``qmc.in-wfnoj.xml`` file will have one VMC block with a minimum number of blocks to reproduce the HF/DFT energy used to generate the trial wavefunction. :: <qmc method="vmc" move="pbyp" checkpoint="-1"> <estimator name="LocalEnergy" hdf5="no"/> <parameter name="warmupSteps">100</parameter> <parameter name="blocks">20</parameter> <parameter name="steps">50</parameter> <parameter name="substeps">8</parameter> <parameter name="timestep">0.5</parameter> <parameter name="usedrift">no</parameter> </qmc> </simulation>If the
qmc.in-wfj.xmlfile is used, Jastrow optimization blocks followed by a VMC and DMC block are included. These blocks contain default values to allow the user to test the accuracy of a system; however, they need to be updated and optimized for each system. The initial values might only be suitable for a small molecule.<loop max="4"> <qmc method="linear" move="pbyp" checkpoint="-1"> <estimator name="LocalEnergy" hdf5="no"/> <parameter name="warmupSteps">100</parameter> <parameter name="blocks">20</parameter> <parameter name="timestep">0.5</parameter> <parameter name="walkers">1</parameter> <parameter name="samples">16000</parameter> <parameter name="substeps">4</parameter> <parameter name="usedrift">no</parameter> <parameter name="MinMethod">OneShiftOnly</parameter> <parameter name="minwalkers">0.0001</parameter> </qmc> </loop> <!-- Example follow-up VMC optimization using more samples for greater accuracy: --> <loop max="10"> <qmc method="linear" move="pbyp" checkpoint="-1"> <estimator name="LocalEnergy" hdf5="no"/> <parameter name="warmupSteps">100</parameter> <parameter name="blocks">20</parameter> <parameter name="timestep">0.5</parameter> <parameter name="walkers">1</parameter> <parameter name="samples">64000</parameter> <parameter name="substeps">4</parameter> <parameter name="usedrift">no</parameter> <parameter name="MinMethod">OneShiftOnly</parameter> <parameter name="minwalkers">0.3</parameter> </qmc> </loop> <!-- Production VMC and DMC: Examine the results of the optimization before running these blocks. For example, choose the best optimized jastrow from all obtained, put in the wavefunction file, and do not reoptimize. --> <qmc method="vmc" move="pbyp" checkpoint="-1"> <estimator name="LocalEnergy" hdf5="no"/> <parameter name="warmupSteps">100</parameter> <parameter name="blocks">200</parameter> <parameter name="steps">50</parameter> <parameter name="substeps">8</parameter> <parameter name="timestep">0.5</parameter> <parameter name="usedrift">no</parameter> <!--Sample count should match targetwalker count for DMC. Will be obtained from all nodes.--> <parameter name="samples">16000</parameter> </qmc> <qmc method="dmc" move="pbyp" checkpoint="20"> <estimator name="LocalEnergy" hdf5="no"/> <parameter name="targetwalkers">16000</parameter> <parameter name="reconfiguration">no</parameter> <parameter name="warmupSteps">100</parameter> <parameter name="timestep">0.005</parameter> <parameter name="steps">100</parameter> <parameter name="blocks">100</parameter> <parameter name="nonlocalmoves">yes</parameter> </qmc> </simulation>File
.structure.xmlThis file will be referenced from the main QMCPACK input. It contains the geometry of the system, position of the atoms, number of atoms, atomic types and charges, and number of electrons.
Files
.wfj.xmland.wfnoj.xmlThese files contain the basis set detail, orbital coefficients, and the 1-, 2-, and 3-body Jastrow (in the case of
.wfj.xml). If the wavefunction is multideterminant, the expansion will be at the end of the file. We recommend using the option-hdf5when large molecules are studied to store the data more compactly in an HDF5 file.File
.orbs.h5This file is generated only if the option-hdf5is added as follows:convert4qmc -gamess MyGamessOutput.out -hdf5
In this case, the
.wfj.xmlor.wfnoj.xmlfiles will point to this HDF file. Information about the basis set, orbital coefficients, and the multideterminant expansion is put in this file and removed from the wavefunction files, making them smaller.
convert4qmc input type:
option name
description
-orbitalsGeneric HDF5 input file. Mainly automatically generated from QP2, Pyscf and all codes in molden2qmc
Actively maintained
-gamessGamess code
Maintained
-gaussianGaussian code
Obsolete/untested
-casinoCasino code
Obsolete/untested
Command line options¶
convert4qmccommand line options:
Option Name
Value
default
description
-nojastrow
Force no Jastrow.
qmc.in.wfjwill not be generated
-hdf5
Force the wf to be in HDF5 format
-prefixstring
All created files will have the name of the string
-multidetstring
HDF5 file containing a multideterminant expansion
-addCusp
Force to add orbital cusp correction (ONLY for all-electron)
-production
Generates specific blocks in the input
-psi_tagstring
psi0
Name of the electrons particles inside QMCPACK
-ion_tagstring
ion0
Name of the ion particles inside QMCPACK
-multidetThis option is to be used when a multideterminant expansion (mainly a CI expansion) is present in an HDF5 file. The trial wavefunction file will not display the full list of multideterminants and will add a path to the HDF5 file as follows (full example for the C2 molecule in qmcpack/tests/molecules/C2_pp).
<?xml version="1.0"?> <qmcsystem> <wavefunction name="psi0" target="e"> <determinantset type="MolecularOrbital" name="LCAOBSet" source="ion0" transform="yes" href="C2.h5"> <sposet basisset="LCAOBSet" name="spo-up" size="58"> <occupation mode="ground"/> <coefficient size="58" spindataset="0"/> </sposet> <sposet basisset="LCAOBSet" name="spo-dn" size="58"> <occupation mode="ground"/> <coefficient size="58" spindataset="0"/> </sposet> <multideterminant optimize="no" spo_up="spo-up" spo_dn="spo-dn"> <detlist size="202" type="DETS" nca="0" ncb="0" nea="4" neb="4" nstates="58" cutoff="1e-20" href="C2.h5"/> </multideterminant> </determinantset> </wavefunction> </qmcsystem>To generate such trial wavefunction, the converter has to be invoked as follows:
> convert4qmc -orbitals C2.h5 -multidet C2.h5
-nojastrowThis option generates only an input file,
*.qmc.in.wfnoj.xml, containing no Jastrow optimization blocks and references a wavefunction file,*.wfnoj.xml, containing no Jastrow section.-hdf5This option generates the
*.orbs.h5HDF5 file containing the basis set and the orbital coefficients. If the wavefunction contains a multideterminant expansion from QP2, it will also be stored in this file. This option minimizes the size of the*.wfj.xmlfile, which points to the HDF file, as in the following example:<?xml version="1.0"?> <qmcsystem> <wavefunction name="psi0" target="e"> <determinantset type="MolecularOrbital" name="LCAOBSet" source="ion0" transform="yes" href="test.orbs.h5"> <slaterdeterminant> <determinant id="updet" size="39"> <occupation mode="ground"/> <coefficient size="411" spindataset="0"/> </determinant> <determinant id="downdet" size="35"> <occupation mode="ground"/> <coefficient size="411" spindataset="0"/> </determinant> </slaterdeterminant> </determinantset> </wavefunction> </qmcsystem>Jastrow functions will be included if the option “-nojastrow” was not specified. Note that when initially optimization a wavefunction, we recommend temporarily removing/disabling the 3-body Jastrow.
-prefix
Sets the prefix for all output generated by
convert4qmc. If not specified,convert4qmcwill use the defaults for the following:Gamess If the Gamess output file is named “Name.out” or “Name.output,” all files generated by
convert4qmcwill carry Name as a prefix (i.e., Name.qmc.in.xml).Generic HDF5 input If a generic HDF5 file is named “Name.H5,” all files generated by
convert4qmcwill carry Name as a prefix (i.e., Name.qmc.in.xml).
-addCusp
This option is very important for all-electron (AE) calculations. In this case, orbitals have to be corrected for the electron-nuclear cusp. The cusp correction scheme follows the algorithm described by Ma et al. [MTDN05] When this option is present, the wavefunction file has a new set of tags:
qmcsystem> <wavefunction name="psi0" target="e"> <determinantset type="MolecularOrbital" name="LCAOBSet" source="ion0" transform="yes" cuspCorrection="yes"> <basisset name="LCAOBSet">
The tag “cuspCorrection” in the
wfj.xml(orwfnoj.xml) wavefunction file will force correction of the orbitals at the beginning of the run. In the “orbitals“ section of the wavefunction file, a new tag “cuspInfo” will be added for orbitals spin-up and orbitals spin-down:<slaterdeterminant> <determinant id="updet" size="2" cuspInfo="../updet.cuspInfo.xml"> <occupation mode="ground"/> <coefficient size="135" id="updetC"> <determinant id="downdet" size="2" cuspInfo="../downdet.cuspInfo.xml"> <occupation mode="ground"/> <coefficient size="135" id="downdetC">
These tags will point to the files
updet.cuspInfo.xmlanddowndet.cuspInfo.xml. By default, the converter assumes that the files are located in the relative path../. If the files are not present in the parent directory, QMCPACK will run the cusp correction algorithm to generate both files in the current run directory (not in../). If the files exist, then QMCPACK will apply the corrections to the orbitals.Important notes:
The cusp correction implementations has been parallelized and performance improved. However, since the correction needs to be applied for every ion and then for every orbital on that ion, this operation can be costly (slow) for large systems. We recommend saving and reusing the computed cusp correction files
updet.cuspInfo.xmlanddowndet.cuspInfo.xml, and transferring them between computer systems where relevant.-psi_tag
QMCPACK builds the wavefunction as a named object. In the vast majority of cases, one wavefunction is simulated at a time, but there may be situations where we want to distinguish different parts of a wavefunction, or even use multiple wavefunctions. This option can change the name for these cases.
<wavefunction name="psi0" target="e">
-ion_tag
Although similar to -psi_tag, this is used for the type of ions.
<particleset name="ion0" size="2">
-production
Without this option, input files with standard optimization, VMC, and DMC blocks are generated. When the “-production” option is specified, an input file containing complex options that may be more suitable for large runs at HPC centers is generated. This option is for users who are already familiar with QMC and QMCPACK. We encourage feedback on the standard and production sample inputs.
The following options are specific to using MCSCF multideterminants from Gamess.
convert4qmcMCSCF arguments:
Option Name
Value
default
description
-ciString
none
Name of the file containing the CI expansion
-thresholddouble
1e-20
Cutoff of the weight of the determinants
-TargetStateint
none
?
-NaturalOrbitalsint
none
?
-optDetCoeffs
no
Enables the optimization of CI coefficients
keyword -ci Path/name of the file containing the CI expansion in a Gamess Format.
keyword -threshold The CI expansion contains coefficients (weights) for each determinant. This option sets the maximum coefficient to include in the QMC run. By default it is set to 1e-20 (meaning all determinants in an expansion are taken into account). At the same time, if the threshold is set to a different value, for example \(1e-5\), any determinant with a weight \(|weight| < 1e-5\) will be discarded and the determinant will not be considered.
keyword -TargetState ?
keyword -NaturalOrbitals ?
keyword -optDetCoeffs This flag enables optimization of the CI expansion coefficients. By default, optimization of the coefficients is disabled during wavefunction optimization runs.
Examples and more thorough descriptions of these options can be found in the lab section of this manual: Lab 3: Advanced molecular calculations.
Grid options¶
These parameters control how the basis set is projected on a grid. The default parameters are chosen to be very efficient. Unless you have a very good reason, we do not recommend modifying them.
Tags |
|||
keyword |
Value |
default |
description |
|
log|log0|linear |
log |
Grid type |
|
double |
1e-6 |
First point of the grid |
|
double |
100 |
Last point of the grid |
|
int |
1001 |
Number of point in the grid |
-gridtype Grid type can be logarithmic, logarithmic base 10, or linear
-first First value of the grid
-last Last value of the grid
-size Number of points in the grid between “first” and “last.”
Supported codes¶
PySCF
PySCF [SBB+18] is an all-purpose quantum chemistry code that can run calculations from simple Hartree-Fock to DFT, MCSCF, and CCSD, and for both isolated systems and periodic boundary conditions. PySCF can be downloaded from https://github.com/sunqm/pyscf. Many examples and tutorials can be found on the PySCF website, and all types of single determinants calculations are compatible with , thanks to active support from the authors of PySCF. A few additional steps are necessary to generate an output readable by
convert4qmc.This example shows how to run a Hartree-Fock calculation for the \(LiH\) dimer molecule from PySCF and convert the wavefunction for QMCPACK.
Python path
PySCF is a Python-based code. A Python module named PyscfToQmcpack containing the function savetoqmcpack is provided by and is located at
qmcpack/src/QMCTools/PyscfToQmcpack.py. To be accessible to the PySCF script, this path must be added to the PYTHONPATH environment variable. For the bash shell, this can be done as follows:export PYTHONPATH=/PATH_TO_QMCPACK/qmcpack/src/QMCTools:\$PYTHONPATH
PySCF Input File
Copy and paste the following code in a file named LiH.py.
#! /usr/bin/env python3 from pyscf import gto, scf, df import numpy cell = gto.M( atom =''' Li 0.0 0.0 0.0 H 0.0 0.0 3.0139239778''', basis ='cc-pv5z', unit="bohr", spin=0, verbose = 5, cart=False, ) mf = scf.ROHF(cell) mf.kernel() ###SPECIFIC TO QMCPACK### title='LiH' from PyscfToQmcpack import savetoqmcpack savetoqmcpack(cell,mf,title)
The arguments to the function savetoqmcpack are:
cell This is the object returned from gto.M, containing the type of atoms, geometry, basisset, spin, etc.
mf This is an object representing the PySCF level of theory, in this example, ROHF. This object contains the orbital coefficients of the calculations.
title The name of the output file generated by PySCF. By default, the name of the generated file will be “default” if nothing is specified.
By adding the three lines below the “SPECIFIC TO QMCPACK” comment in the input file, the script will dump all the necessary data for QMCPACK into an HDF5 file using the value of “title” as an output name. PySCF is run as follows:
>python LiH.py
The generated HDF5 can be read by
convert4qmcto generate the appropriate QMCPACK input files.Generating input files
As described in the previous section, generating input files for PySCF is as follows:
> convert4qmc -pyscf LiH.h5
The HDF5 file produced by “savetoqmcpack” contains the wavefunction in a form directly readable by QMCPACK. The wavefunction files from
convert4qmcreference this HDF file as if the “-hdf5” option were specified (converting from PySCF implies the “-hdf5” option is always present).
Periodic boundary conditions with Gaussian orbitals from PySCF is fully supported for Gamma point and kpoints.
Quantum Package
QP2 [GAG+19] is a quantum chemistry code developed by the LCPQ laboratory in Toulouse, France, and Argonne National Laboratory for the PBC version. It can be downloaded from https://github.com/QuantumPackage/qp2, and the tutorial within is quite extensive. The tutorial section of QP2 can guide you on how to install and run the code.
After a QP2 calculation, the data needed for
convert4qmccan be generated throughqp_run save_for_qmcpack Myrun.ezfio
This command will generate an HDF5 file in the QMCPACK format named
QP2QMCPACK.h5convert4qmccan read this file and generate the*.structure.xml,*.wfj.xmland other files needed to run QMCPACK. . For example:convert4qmc -orbitals QP2QMCPACK.h5 -multidet QP2QMCPACK.h5 -prefix MySystem
The main reason to use QP2 is to access the CIPSI algorithm to generate a multideterminant wavefunction. CIPSI is the preferred choice for generating a selected CI trial wavefunction for QMCPACK. An example on how to use QP2 for Hartree-Fock and selected CI can be found in CIPSI wavefunction interface of this manual. The converter code is actively maintained and codeveloped by both QMCPACK and QP2 developers.
Using -hdf5 tag
convert4qmc -gamess Myrun.out -hdf5
This option is only used/usefull with the gamess code as it is the onlycode not providing an HDF5 output The result will create QMCPACK input files but will also store all key data in the HDF5 format.
Mixing orbitals and multideterminants
Note that the
QP2QMCPACK.h5combined with the tags-orbitalsand-multidetallows the user to choose orbitals from a different code such as PYSCF and the multideterminant section from QP2. These two codes are fully compatible, and this route is also the only possible route for multideterminants for solids.convert4qmc -orbitals MyPyscfrun.h5 -multidet QP2QMCPACK.h5
GAMESS
QMCPACK can use the output of GAMESS [SBB+93] for any type of single determinant calculation (HF or DFT) or multideterminant (MCSCF) calculation. A description with an example can be found in the Advanced Molecular Calculations Lab (Lab 3: Advanced molecular calculations).
pw2qmcpack.x¶
pw2qmcpack.x is an executable that converts PWSCF wavefunctions to QMCPACK readable
HDF5 format. This utility is built alongside the QE postprocessing utilities.
This utility is written in Fortran90 and is distributed as a patch of the QE
source code. The patch, as well as automated QE download and patch scripts, can be found in
qmcpack/external_codes/quantum_espresso.
pw2qmcpack can be used in serial in small systems and should be used in parallel with large systems for best performance. The K_POINT gamma optimization is not supported.
pw2qmcpack.x input file p2q.in¶&inputpp
prefix = 'bulk_silicon'
outdir = './'
write_psir = .false.
/
This example will cause pw2qmcpack.x to convert wavefunctions saved from
PWSCF with the prefix “bulk_silicon.” Perform the conversion via, for
example:
mpirun -np 1 pw2qmcpack.x < p2q.in>& p2q.out
Because of the large plane-wave energy cutoffs in the pw.x calculation required by accurate PPs and the large system sizes of interest, one limitation of QE can be easily reached:
that wf_collect=.true. results in problems of writing and loading correct plane-wave coefficients on disks by pw.x because of the 32 bit integer limits. Thus, pw2qmcpack.x fails to convert the orbitals for QMCPACK. Since the release of QE v5.3.0, the converter has been fully parallelized to overcome this limitation completely.
By setting wf_collect=.false. (by default .false. in v6.1 and before and .true. since v6.2), pw.x does not collect the whole wavefunction into individual files for each k-point but instead writes one smaller file for each processor.
By running pw2qmcpack.x in the same parallel setup (MPI tasks and k-pools) as the last scf/nscf calculation with pw.x,
the orbitals distributed among processors will first be aggregated by the converter into individual temporal HDF5 files for each k-pool and then merged into the final file.
In large calculations, users should benefit from a significant reduction of time in writing the wavefunction by pw.x thanks to avoiding the wavefunction collection.
pw2qmcpack has been included in the test suite of QMCPACK (see instructions about how to activate the tests in Installing and patching Quantum ESPRESSO).
There are tests labeled “no-collect” running the pw.x with the setting wf_collect=.false.
The input files are stored at examples/solids/dft-inputs-polarized-no-collect.
The scf, nscf, and pw2qmcpack runs are performed on 16, 12, and 12 MPI tasks with 16, 2, and 2 k-pools respectively.
convertpw4qmc¶
Convertpw4qmc is an executable that reads xml from a plane wave based DFT code and produces a QMCPACK readable HDF5 format wavefunction. For the moment, this supports both QBox and Quantum Epresso
In order to save the wavefunction from QBox so that convertpw4qmc can work on it, one needs to add a line to the QBox input like
save -text -serial basename.sample
after the end of a converged dft calculation. This will write an ascii wavefunction file and will avoid QBox’s optimized parallel IO (which is not currently supported).
After the wavefunction file is written (basename.sample in this case) one can use convertpw4qmc as follows:
convertpw4qmc basename.sample -o qmcpackWavefunction.h5
This reads the Qbox wavefunction and performs the Fourier transform before saving to a QMCPACK eshdf format wavefunction. Currently multiple k-points are supported, but due to difficulties with the qbox wavefunction file format, the single particle orbitals do not have their proper energies associated with them. This means that when tiling from a primitive cell to a supercell, the lowest n single particle orbitals from all necessary k-points will be used. This can be problematic in the case of a metal and this feature should be used with EXTREME caution.
In the case of quantum espresso, QE must be compiled with HDF support. If this is the case, then an eshdf file can be generated by targeting the data-file-schema.xml file generated in the output of quantum espresso. For example, if one is running a calculation with outdir = ‘out’ and prefix=’Pt’ then the converter can be invoked as:
convertpw4qmc out/Pt.save/data-file-schema.xml -o qmcpackWavefunction.h5
Note that this method is insensitive to parallelization options given to quantum espresso. Additionally, it supports noncollinear magnetism and can be used to generate wavefunctions suitable for qmcpack calculations with spin-orbit coupling.
ppconvert¶
ppconvert is a utility to convert PPs between different commonly used formats.
It is a stand-alone C++ executable that is not built by default but that is accessible via adding
-DBUILD_PPCONVERT=1 to CMake and then typing make ppconvert.
Currently it converts CASINO, FHI, UPF (generated by OPIUM), BFD, and GAMESS formats to several other formats
including XML (QMCPACK) and UPF (QE). See all the formats via ppconvert -h.
For output formats requiring Kleinman-Bylander projectors, the atom will be solved with DFT
if the projectors are not provided in the input formats.
This requires providing reference states and sometimes needs extra tuning for heavy elements.
To avoid ghost states, the local channel can be changed via the --local_channel option. Ghost state considerations are similar to those of DFT calculations but could be worse if ghost states were not considered during the original PP construction.
To make the self-consistent calculation converge, the density mixing parameter may need to be reduced
via the --density_mix option.
Note that the reference state should include only the valence electrons.
One reference state should be included for each channel in the PP.
For example, for a sodium atom with a neon core, the reference state would be “1s(1).”
--s_ref needs to include a 1s state, --p_ref needs to include a 2p state,
--d_ref needs to include a 3d state, etc. If not specified, a corresponding state with zero occupation is added.
If the reference state is chosen as the neon core, setting empty reference states “” is technically correct.
In practice, reasonable reference states should be picked with care.
For PP with semi-core electrons in the valence, the reference state can be long.
For example, Ti PP has 12 valence electrons. When using the neutral atom state,
--s_ref, --p_ref, and --d_ref are all set as “1s(2)2p(6)2s(2)3d(2).”
When using an ionized state, the three reference states are all set as “1s(2)2p(6)2s(2)” or “1s(2)2p(6)2s(2)3d(0).”
Unfortunately, if the generated UPF file is used in QE, the calculation may be incorrect because of the presence of “ghost” states. Potentially these can be removed by adjusting the local channel (e.g., by setting --local_channel 1, which chooses the p channel as the local channel instead of d.
For this reason, validation of UPF PPs is always required from the third row and is strongly encouraged in general. For example, check that the expected ionization potential and electron affinities are obtained for the atom and that dimer properties are consistent with those obtained by a quantum chemistry code or a plane-wave code that does not use the Kleinman-Bylander projectors.
molden2qmc¶
molden2qmc is a tool used to convert molden files into an HDF5 file with the QMCPACK format.
Molden2qmc is a single program that can use multiple different quantum chemistry codes.
It is python code developed by Vladimir Konjkov originally for the CASINO code but then extended to QMCPACK.
This tool can be found at https://github.com/gjohnson3/molden2qmc.git.
Using molden2qmc¶
General use of molden2qmc can be prompted by running molden2qmc.py and entering the corresponding quantum chemistry code number and the molden file name:
number corresponding to the quantum chemistry code used to produce this MOLDEN file:
0 -- TURBOMOLE
1 -- PSI4
2 -- CFOUR 2.0beta
3 -- ORCA 3.X - 4.X
4 -- DALTON2016
5 -- MOLPRO
6 -- NWCHEM
7 -- QCHEM 4.X
Use the --qmcpack flag to create the file as an hdf5 file, suitable for QMCPACK.
Without the --qmcpack flag, the file will become a gwfn file for CASINO.
Example: molden2qmc.py 5 n4.molden --qmcpack.
Obtaining pseudopotentials¶
Pseudopotentiallibrary.org¶
An open website collecting community developed and tested pseudopotentials for QMC and other many-body calculations is being developed at https://pseudopotentiallibrary.org. This site includes potentials in QMCPACK format and an increasing range of electronic structure and quantum chemistry codes. We recommend using potentials from this site if available and suitable for your science application.
Opium¶
Opium is a pseudopotential generation code available from the website http://opium.sourceforge.net/. Opium can generate pseudopotentials with either Hartree-Fock or DFT methods. Once you have a useable pseudopotential param file (for example, Li.param), generate pseudopotentials for use in Quantum ESPRESSO with the upf format as follows:
This generates a UPF-formatted pseudopotential (Li.upf, in this case) for use in Quantum ESPRESSO. The pseudopotential conversion tool ppconvert can then convert UPF to FSAtom xml format for use in QMCPACK:
ppconvert --upf_pot Li.upf --xml Li.xml
Burkatzki-Filippi-Dolg¶
Burkatzki et al. developed a set of energy-consistent pseudopotenitals for use in QMC [BFD07][BFD08], available at http://www.burkatzki.com/pseudos/index.2.html. To convert for use in QMCPACK, select a pseudopotential (choice of basis set is irrelevant to conversion) in GAMESS format and copy the ending (pseudopotential) lines beginning with(element symbol)-QMC GEN:
Li-QMC GEN 2 1
3
1.00000000 1 5.41040609
5.41040609 3 2.70520138
-4.60151975 2 2.07005488
1
7.09172172 2 1.34319829
Save these lines to a file (here, named Li.BFD.gamess; the exact name may be anything as long as it is passed to ppconvert after –gamess_pot). Then, convert using ppconvert with the following:
ppconvert --gamess_pot Li.BFD.gamess --s_ref "2s(1)" --p_ref "2p(0)" --xml Li.BFD.xml
ppconvert --gamess_pot Li.BFD.gamess --s_ref "2s(1)" --p_ref "2p(0)" --log_grid --upf Li.BFD.upf
CASINO¶
The QMC code CASINO also makes available its pseudopotentials available at the website https://vallico.net/casinoqmc/pplib/. To use one in QMCPACK, select a pseudopotential and download its summary file (summary.txt), its tabulated form (pp.data), and (for ppconvert to construct the projectors to convert to Quantum ESPRESSO’s UPF format) a CASINO atomic wavefunction for each angular momentum channel (awfn.data_*). Then, to convert using ppconvert, issue the following command:
ppconvert --casino_pot pp.data --casino_us awfn.data_s1_2S --casino_up awfn.data_p1_2P --casino_ud awfn.data_d1_2D --upf Li.TN-DF.upf
QMCPACK can directly read in the CASINO-formated pseudopotential (pp.data), but four parameters found in the pseudopotential summary file must be specified in the pseudo element (l-local, lmax, nrule, cutoff)[see Pseudopotentials for details]:
<pairpot type="pseudo" name="PseudoPot" source="ion0" wavefunction="psi0" format="xml">
<pseudo elementType="Li" href="Li.pp.data" format="casino" l-local="s" lmax="2" nrule="2" cutoff="2.19"/>
<pseudo elementType="H" href="H.pp.data" format="casino" l-local="s" lmax="2" nrule="2" cutoff="0.5"/>
</pairpot>
wftester¶
While not really a stand-alone application, wftester (short for “Wave Function Tester”) is a helpful tool for testing pre-existing and experimental estimators and observables. It provides the user with derived quantities from the Hamiltonian and wave function, but evaluated at a small set of configurations.
The wftester is implemented as a QMCDriver, so one invokes QMCPACK in the normal manner with a correct input XML, the difference being the addition of an additional qmc input block. This is the main advantage of this tool–it allows testing of realistic systems and realistic combinations of observables. It can also be invoked before launching into optimization, VMC, or DMC runs, as it is a valid <qmc> block.
As an example, the following code generates a random walker configuration and compares the trial wave function ratio computed in two different ways:
<qmc method="wftester">
<parameter name="ratio"> yes </parameter>
</qmc>
Here’s a summary of some of the tests provided:
Ratio Test. Invoked with
<parameter name="ratio">yes</parameter>
This computes the implemented wave function ratio associated with a single-particle move using two different methods.
Clone Test. Invoked with
<parameter name="clone">yes</parameter>
This checks the cloning of TrialWaveFunction, ParticleSet, Hamiltonian, and Walkers.
Elocal Test. Invoked with
<parameter name="printEloc">yes</parameter>
For an input electron configuration (can be random), print the value of TrialWaveFunction, LocalEnergy, and all local observables for this configuration.
Derivative Test. Invoked with
<parameter name="ratio">deriv</parameter>}
Computes electron gradients, laplacians, and wave function parameter derivatives using implemented calls and compares them to finite-difference results.
Ion Gradient Test. Invoked with
<parameter name="source">ion0</parameter>
Calls the implemented evaluateGradSource functions and compares them against finite-difference results.
“Basic Test”. Invoked with
<parameter name="basic">yes</parameter>
Performs ratio, gradient, and laplacian tests against finite-difference and direct computation of wave function values.
The output of the various tests will be to standard out or “wftest.000” after successful execution of qmcpack.
- AAB+14
Kęstutis Aidas, Celestino Angeli, Keld L. Bak, Vebjørn Bakken, Radovan Bast, Linus Boman, Ove Christiansen, Renzo Cimiraglia, Sonia Coriani, Pål Dahle, Erik K. Dalskov, Ulf Ekström, Thomas Enevoldsen, Janus J. Eriksen, Patrick Ettenhuber, Berta Fernández, Lara Ferrighi, Heike Fliegl, Luca Frediani, Kasper Hald, Asger Halkier, Christof Hättig, Hanne Heiberg, Trygve Helgaker, Alf Christian Hennum, Hinne Hettema, Eirik Hjertenæs, Stinne Høst, Ida-Marie Høyvik, Maria Francesca Iozzi, Branislav Jansík, Hans Jørgen \relax Aa. Jensen, Dan Jonsson, Poul Jørgensen, Joanna Kauczor, Sheela Kirpekar, Thomas Kjærgaard, Wim Klopper, Stefan Knecht, Rika Kobayashi, Henrik Koch, Jacob Kongsted, Andreas Krapp, Kasper Kristensen, Andrea Ligabue, Ola B Lutnæs, Juan I. Melo, Kurt V. Mikkelsen, Rolf H. Myhre, Christian Neiss, Christian B. Nielsen, Patrick Norman, Jeppe Olsen, Jógvan Magnus H. Olsen, Anders Osted, Martin J. Packer, Filip Pawlowski, Thomas B. Pedersen, Patricio F. Provasi, Simen Reine, Zilvinas Rinkevicius, Torgeir A. Ruden, Kenneth Ruud, Vladimir V. Rybkin, Pawel Sałek, Claire C. M. Samson, Alfredo Sánchez de Merás, Trond Saue, Stephan P. A. Sauer, Bernd Schimmelpfennig, Kristian Sneskov, Arnfinn H. Steindal, Kristian O. Sylvester-Hvid, Peter R. Taylor, Andrew M. Teale, Erik I. Tellgren, David P. Tew, Andreas J. Thorvaldsen, Lea Thøgersen, Olav Vahtras, Mark A. Watson, David J. D. Wilson, Marcin Ziolkowski, and Hans Ågren. The Dalton quantum chemistry program system. WIREs Comput. Mol. Sci., 4(3):269–284, 2014. doi:10.1002/wcms.1172.
- ApraBdJ+20
E. Aprá, E. J. Bylaska, W. A. de Jong, N. Govind, K. Kowalski, T. P. Straatsma, M. Valiev, H. J. J. van Dam, Y. Alexeev, J. Anchell, V. Anisimov, F. W. Aquino, R. Atta-Fynn, J. Autschbach, N. P. Bauman, J. C. Becca, D. E. Bernholdt, K. Bhaskaran-Nair, S. Bogatko, P. Borowski, J. Boschen, J. Brabec, A. Bruner, E. Cauët, Y. Chen, G. N. Chuev, C. J. Cramer, J. Daily, M. J. O. Deegan, T. H. Dunning, M. Dupuis, K. G. Dyall, G. I. Fann, S. A. Fischer, A. Fonari, H. Früchtl, L. Gagliardi, J. Garza, N. Gawande, S. Ghosh, K. Glaesemann, A. W. Götz, J. Hammond, V. Helms, E. D. Hermes, K. Hirao, S. Hirata, M. Jacquelin, L. Jensen, B. G. Johnson, H. Jónsson, R. A. Kendall, M. Klemm, R. Kobayashi, V. Konkov, S. Krishnamoorthy, M. Krishnan, Z. Lin, R. D. Lins, R. J. Littlefield, A. J. Logsdail, K. Lopata, W. Ma, A. V. Marenich, J. Martin del Campo, D. Mejia-Rodriguez, J. E. Moore, J. M. Mullin, T. Nakajima, D. R. Nascimento, J. A. Nichols, P. J. Nichols, J. Nieplocha, A. Otero-de-la-Roza, B. Palmer, A. Panyala, T. Pirojsirikul, B. Peng, R. Peverati, J. Pittner, L. Pollack, R. M. Richard, P. Sadayappan, G. C. Schatz, W. A. Shelton, D. W. Silverstein, D. M. A. Smith, T. A. Soares, D. Song, M. Swart, H. L. Taylor, G. S. Thomas, V. Tipparaju, D. G. Truhlar, K. Tsemekhman, T. Van Voorhis, Á. Vázquez-Mayagoitia, P. Verma, O. Villa, A. Vishnu, K. D. Vogiatzis, D. Wang, J. H. Weare, M. J. Williamson, T. L. Windus, K. Woliński, A. T. Wong, Q. Wu, C. Yang, Q. Yu, M. Zacharias, Z. Zhang, Y. Zhao, and R. J. Harrison. Nwchem: past, present, and future. The Journal of Chemical Physics, 152(18):184102, 2020. doi:10.1063/5.0004997.
- BFD07
M. Burkatzki, C. Filippi, and M. Dolg. Energy-consistent pseudopotentials for quantum monte carlo calculations. The Journal of Chemical Physics, 126(23):–, 2007. doi:10.1063/1.2741534.
- BFD08
M. Burkatzki, Claudia Filippi, and M. Dolg. Energy-consistent small-core pseudopotentials for 3d-transition metals adapted to quantum monte carlo calculations. The Journal of Chemical Physics, 129(16):–, 2008. doi:10.1063/1.2987872.
- FAH+14
Filipp Furche, Reinhart Ahlrichs, Christof Hättig, Wim Klopper, Marek Sierka, and Florian Weigend. Turbomole. WIREs Computational Molecular Science, 4(2):91–100, 2014. doi:10.1002/wcms.1162.
- GAG+19(1,2)
Yann Garniron, Thomas Applencourt, Kevin Gasperich, Anouar Benali, Anthony Ferté, Julien Paquier, Barthélémy Pradines, Roland Assaraf, Peter Reinhardt, Julien Toulouse, Pierrette Barbaresco, Nicolas Renon, Grégoire David, Jean-Paul Malrieu, Mickaël Véril, Michel Caffarel, Pierre-François Loos, Emmanuel Giner, and Anthony Scemama. Quantum package 2.0: an open-source determinant-driven suite of programs. Journal of Chemical Theory and Computation, 15(6):3591–3609, 2019. doi:10.1021/acs.jctc.9b00176.
- MTDN05
A. Ma, M. D. Towler, N. D. Drummond, and R. J. Needs. Scheme for adding electron–nucleus cusps to gaussian orbitals. The Journal of Chemical Physics, 122(22):224322, 2005. doi:10.1063/1.1940588.
- MCH+20
Devin A. Matthews, Lan Cheng, Michael E. Harding, Filippo Lipparini, Stella Stopkowicz, Thomas-C. Jagau, Péter G. Szalay, Jürgen Gauss, and John F. Stanton. Coupled-cluster techniques for computational chemistry: the cfour program package. The Journal of Chemical Physics, 152(21):214108, 2020. doi:10.1063/5.0004837.
- Nee18
Frank Neese. Software update: the orca program system, version 4.0. WIREs Computational Molecular Science, 8(1):e1327, 2018. doi:10.1002/wcms.1327.
- SBB+93(1,2)
Michael W. Schmidt, Kim K. Baldridge, Jerry A. Boatz, Steven T. Elbert, Mark S. Gordon, Jan H. Jensen, Shiro Koseki, Nikita Matsunaga, Kiet A. Nguyen, Shujun Su, Theresa L. Windus, Michel Dupuis, and John A. Montgomery. General atomic and molecular electronic structure system. Journal of Computational Chemistry, 14(11):1347–1363, 1993. doi:10.1002/jcc.540141112.
- SGE+15
Yihan Shao, Zhengting Gan, Evgeny Epifanovsky, Andrew T.B. Gilbert, Michael Wormit, Joerg Kussmann, Adrian W. Lange, Andrew Behn, Jia Deng, Xintian Feng, Debashree Ghosh, Matthew Goldey, Paul R. Horn, Leif D. Jacobson, Ilya Kaliman, Rustam Z. Khaliullin, Tomasz Kuś, Arie Landau, Jie Liu, Emil I. Proynov, Young Min Rhee, Ryan M. Richard, Mary A. Rohrdanz, Ryan P. Steele, Eric J. Sundstrom, H. Lee Woodcock III, Paul M. Zimmerman, Dmitry Zuev, Ben Albrecht, Ethan Alguire, Brian Austin, Gregory J. O. Beran, Yves A. Bernard, Eric Berquist, Kai Brandhorst, Ksenia B. Bravaya, Shawn T. Brown, David Casanova, Chun-Min Chang, Yunqing Chen, Siu Hung Chien, Kristina D. Closser, Deborah L. Crittenden, Michael Diedenhofen, Robert A. DiStasio Jr., Hainam Do, Anthony D. Dutoi, Richard G. Edgar, Shervin Fatehi, Laszlo Fusti-Molnar, An Ghysels, Anna Golubeva-Zadorozhnaya, Joseph Gomes, Magnus W.D. Hanson-Heine, Philipp H.P. Harbach, Andreas W. Hauser, Edward G. Hohenstein, Zachary C. Holden, Thomas-C. Jagau, Hyunjun Ji, Benjamin Kaduk, Kirill Khistyaev, Jaehoon Kim, Jihan Kim, Rollin A. King, Phil Klunzinger, Dmytro Kosenkov, Tim Kowalczyk, Caroline M. Krauter, Ka Un Lao, Adèle D. Laurent, Keith V. Lawler, Sergey V. Levchenko, Ching Yeh Lin, Fenglai Liu, Ester Livshits, Rohini C. Lochan, Arne Luenser, Prashant Manohar, Samuel F. Manzer, Shan-Ping Mao, Narbe Mardirossian, Aleksandr V. Marenich, Simon A. Maurer, Nicholas J. Mayhall, Eric Neuscamman, C. Melania Oana, Roberto Olivares-Amaya, Darragh P. O’Neill, John A. Parkhill, Trilisa M. Perrine, Roberto Peverati, Alexander Prociuk, Dirk R. Rehn, Edina Rosta, Nicholas J. Russ, Shaama M. Sharada, Sandeep Sharma, David W. Small, Alexander Sodt, Tamar Stein, David Stück, Yu-Chuan Su, Alex J.W. Thom, Takashi Tsuchimochi, Vitalii Vanovschi, Leslie Vogt, Oleg Vydrov, Tao Wang, Mark A. Watson, Jan Wenzel, Alec White, Christopher F. Williams, Jun Yang, Sina Yeganeh, Shane R. Yost, Zhi-Qiang You, Igor Ying Zhang, Xing Zhang, Yan Zhao, Bernard R. Brooks, Garnet K.L. Chan, Daniel M. Chipman, Christopher J. Cramer, William A. Goddard III, Mark S. Gordon, Warren J. Hehre, Andreas Klamt, Henry F. Schaefer III, Michael W. Schmidt, C. David Sherrill, Donald G. Truhlar, Arieh Warshel, Xin Xu, Alán Aspuru-Guzik, Roi Baer, Alexis T. Bell, Nicholas A. Besley, Jeng-Da Chai, Andreas Dreuw, Barry D. Dunietz, Thomas R. Furlani, Steven R. Gwaltney, Chao-Ping Hsu, Yousung Jung, Jing Kong, Daniel S. Lambrecht, WanZhen Liang, Christian Ochsenfeld, Vitaly A. Rassolov, Lyudmila V. Slipchenko, Joseph E. Subotnik, Troy Van Voorhis, John M. Herbert, Anna I. Krylov, Peter M.W. Gill, and Martin Head-Gordon. Advances in molecular quantum chemistry contained in the q-chem 4 program package. Molecular Physics, 113(2):184–215, 2015. doi:10.1080/00268976.2014.952696.
- SBB+18(1,2)
Qiming Sun, Timothy C. Berkelbach, Nick S. Blunt, George H. Booth, Sheng Guo, Zhendong Li, Junzi Liu, James D. McClain, Elvira R. Sayfutyarova, Sandeep Sharma, Sebastian Wouters, and Garnet Kin-Lic Chan. Pyscf: the python-based simulations of chemistry framework. Wiley Interdisciplinary Reviews: Computational Molecular Science, 8(1):n/a–n/a, 2018. doi:10.1002/wcms.1340.
- TSP+12
Justin M. Turney, Andrew C. Simmonett, Robert M. Parrish, Edward G. Hohenstein, Francesco A. Evangelista, Justin T. Fermann, Benjamin J. Mintz, Lori A. Burns, Jeremiah J. Wilke, Micah L. Abrams, Nicholas J. Russ, Matthew L. Leininger, Curtis L. Janssen, Edward T. Seidl, Wesley D. Allen, Henry F. Schaefer, Rollin A. King, Edward F. Valeev, C. David Sherrill, and T. Daniel Crawford. Psi4: an open-source ab initio electronic structure program. WIREs Computational Molecular Science, 2(4):556–565, 2012. doi:10.1002/wcms.93.
- WKK+12
Hans-Joachim Werner, Peter J. Knowles, Gerald Knizia, Frederick R. Manby, and Martin Schütz. Molpro: a general-purpose quantum chemistry program package. WIREs Computational Molecular Science, 2(2):242–253, 2012. doi:10.1002/wcms.82.
External Tools¶
This chapter provides some information on using QMCPACK with external tools.
LLVM Sanitizer Libraries¶
Using CMake, set one of these flags for using the clang sanitizer libraries with or without lldb.
-DLLVM_SANITIZE_ADDRESS link with the Clang address sanitizer library
-DLLVM_SANITIZE_MEMORY link with the Clang memory sanitizer library
These set the basic flags required to build with either of these sanitizer libraries. They require a build of clang with dynamic libraries somehow visible (i.e., through LD_FLAGS=-L/your/path/to/llvm/lib). You must link through clang, which is generally the default when building with it. Depending on your system and linker, this may be incompatible with the “Release” build, so set -DCMAKE_BUILD_TYPE=Debug. They have been tested with the default spack install of llvm 7.0.0 and been manually built with llvm 7.0.1. See the following links for additional information on use, run time, and build options of the sanitizers: https://clang.llvm.org/docs/AddressSanitizer.html & https://clang.llvm.org/docs/MemorySanitizer.html.
In general, the address sanitizer libraries will catch most pointer-based errors. ASAN can also catch memory links but requires that additional options be set. MSAN will catch more subtle memory management errors but is difficult to use without a full set of MSAN-instrumented libraries.
Intel VTune¶
Intel’s VTune profiler has an API that allows program control over collection (pause/resume) and can add information to the profile data (e.g., delineating tasks).
VTune API¶
If the variable USE_VTUNE_API is set, QMCPACK will check that the
include file (ittnotify.h) and the library (libittnotify.a) can
be found.
To provide CMake with the VTune paths, add the include path to CMAKE_CXX_FLAGS and the library path to CMAKE_LIBRARY_PATH.
An example of options to be passed to CMake:
-DCMAKE_CXX_FLAGS=-I/opt/intel/vtune_amplifier_xe/include \
-DCMAKE_LIBRARY_PATH=/opt/intel/vtune_amplifier_xe/lib64
NVIDIA Tools Extensions¶
NVIDIA’s Tools Extensions (NVTX) API enables programmers to annotate their source code when used with the NVIDIA profilers.
NVTX API¶
If the variable USE_NVTX_API is set, QMCPACK will add the library (libnvToolsExt.so) to the QMCPACK target. To add NVTX annotations
to a function, it is necessary to include the nvToolsExt.h header file and then make the appropriate calls into the NVTX API. For more information
about the NVTX API, see https://docs.nvidia.com/cuda/profiler-users-guide/index.html#nvtx. Any additional calls to the NVTX API should be guarded by
the USE_NVTX_API compiler define.
Timers as Tasks¶
To aid in connecting the timers in the code to the profile data, the start/stop of
timers will be recorded as a task if USE_VTUNE_TASKS is set.
In addition to compiling with USE_VTUNE_TASKS, an option needs to be set at run time to collect the task API data.
In the graphical user interface (GUI), select the checkbox labeled “Analyze user tasks” when setting up the analysis type.
For the command line, set the enable-user-tasks knob to true. For example,
amplxe-cl -collect hotspots -knob enable-user-tasks=true ...
Collection with the timers set at “fine” can generate too much task data in the profile. Collection with the timers at “medium” collects a more reasonable amount of task data.
Scitools Understand¶
Scitools Understand (https://scitools.com/) is a tool for static code analysis. The easiest configuration route is to use the JSON output from CMake, which the Understand project importer can read directly:
Configure QMCPACK by running CMake with
CMAKE_EXPORT_COMPILE_COMMANDS=ON, for example:cmake -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DQMC_MPI=0 -DCMAKE_EXPORT_COMPILE_COMMANDS=ON ../qmcpack/
Run Understand and create a new C++ project. At the import files and settings dialog, import the
compile_commands.jsoncreated by CMake in the build directory.
Contributing to the Manual¶
This section briefly describes how to contribute to the manual and is primarily “by developers, for developers.” This section should iterate until a consistent view on style/contents is reached.
Desirable:
Use the following table templates when describing XML input.
Instead of
\textttor\verbuse\ishellfor shell text\ixmlfor xml text\icodefor C++ text
Except within tabularx or math environments
Instead of
\begin{verbatim}environments, use the appropriate\begin{lstlisting}[style=<see qmcpack_listings.sty>].\begin{shade}can be used in place of\begin{lstlisting}[style=SHELL].Unicode rules
Do not use characters for which well-established latex idiom exists, especially dashes, quotes, and apostrophes.
Use math mode markup instead of unicode characters for equations.
Be cautious of WYSIWYG word processors; cutting and pasting can pickup characters promoted to unicode by the program.
Take a look at your text multibyte expanded; that is open it in (emacs and ‘esc-x toggle-enable-multibyte-characters’)—see any unicode you did not intend?
Place unformatted text targeted at developers working on the LaTeX in comments. Include generously.
Encapsulate formatted text aimed at developers (like this entire chapter), in
\dev{}. Text encapsulated in this way will be removed from the user version of the manual by editing the definition of\dev{}inqmcpack_manual.tex. Existing but deprecated or partially functioning features fall in this category.Newly added entries to a Bib file should be as complete as possible. Use a tool such as JabRef or Zotero that can automate creation of these entries from just a DOI.
Forbidden:
Including images instead of using lstlisting sections for text.
Using packages the LaTeX community considers deprecated.
Using packages, features, or fonts not included in texlive 2017 unless you ensure they degrade reasonably for 2017.
Adding packages unless they bring great value and are supported by tex4ht (unless you are willing to add the support).
Saving Tex files and Bib files in encodings other than UTF8. Some may report being ASCII encoded since they contain no unicode characters.
Missing sections (these are opinions, not decided priorities):
Description of XML input in general. Discuss XML format, use of attributes and
<parameter/>s in general, case sensitivity (input is generally case sensitive), and behavior of when unrecognized XML elements are encountered (they are generally ignored without notification).Overview of the input file in general, broad structure, and at least one full example that works in isolation.
Information currently missing for a complete reference specification:
Noting how many instances of each child element are allowed. Examples:
simulation–1 only,method–1 or more,jastrow–0 or more.
Table templates follow for describing XML elements in reference fashion.
A number of examples can be found in, for example,
Hamiltonian and Observables. Preliminary style is (please weigh in with
opinions): typewriter text (\texttt\{}) for XML elements, attributes, and
parameter names; normal text for literal information in the datatype,
values, and default columns; bold (\textbf{}) text if an attribute or parameter
must take on a particular value (values column); italics (\textit{}) for
descriptive (nonliteral) information in the values column (e.g.,
anything, non-zero); and required/optional attributes or parameters
noted by some_attr \(^r\)/some_attr \(^r\) superscripts. Valid datatypes
are text, integer, real, Boolean, and arrays of each. Fixed length
arrays can be noted, for example, by “real array(3).”
Template for a generic XML element:
generic element:
parent elements: |
|
child elements: |
|
attributes:
Name
Datatype
Values
Default
Description
attr1\(^r\)text
attr2\(^r\)integer
attr3\(^r\)real
attr4\(^r\)boolean
attr5\(^r\)text array
attr6\(^r\)integer array
attr7\(^r\)real array
attr8\(^r\)boolean array
parameters:
Name
Datatype
Values
Default
Description
param1\(^r\)text
param2\(^r\)integer
param3\(^r\)real
param4\(^r\)boolean
param5\(^r\)text array
param6\(^r\)integer array
param7\(^r\)real array
param8\(^r\)boolean array
body text: Long form description of body text format
“Factory” elements are XML elements that share a tag but whose contents
change based on the value an attribute, or sometimes multiple
attributes, take. The attribute(s) that determines the allowed content
is subsequently referred to as the “type selector” (e.g., for
<estimator/> elements, the type selector is usually the type
attribute). These types of elements are frequently encountered as they
correspond (sometimes loosely, sometimes literally) to polymorphic
classes in QMCPACK that are built in “factories.” This name is true to
the underlying code but may be obscure to the general user (is there a
better name to retain the general meaning?).
The following template should be provided each time a new “factory” type
is encountered (such as <estimator/>). The table lists all types of
possible elements (see “type options” in the template) and any
attributes that are common to all possible related elements. Specific
“derived” elements are then described one at a time with the previous
template, noting the type selector in addition to the XML tag (e.g.,
“estimator type=density element”).
Template for shared information about “factory” elements.
generic factory element:
parent elements: |
|
child elements: |
|
type selector |
|
type options |
Selection 1 |
Selection 2 |
|
Selection 3 |
|
… |
shared attributes:
Name
Datatype
Values
Default
Description
attr1\(^r\)text
attr2\(^r\)integer
…
Unit Testing¶
Unit testing is a standard software engineering practice to aid in ensuring a quality product. A good suite of unit tests provides confidence in refactoring and changing code, furnishes some documentation on how classes and functions are used, and can drive a more decoupled design.
If unit tests do not already exist for a section of code, you are encouraged to add them when modifying that section of code. New code additions should also include unit tests. When possible, fixes for specific bugs should also include a unit test that would have caught the bug.
Unit testing framework¶
The Catch framework is used for unit testing. See the project site for a tutorial and documentation: https://github.com/philsquared/Catch.
Catch consists solely of header files. It is distributed as a single include file about 400 KB in size. In QMCPACK, it is stored in external_codes/catch.
Unit test organization¶
The source for the unit tests is located in the tests directory under each directory in src (e.g., src/QMCWavefunctions/tests).
All of the tests in each tests directory get compiled into an executable.
After building the project, the individual unit test executables can be found in build/tests/bin.
For example, the tests in src/QMCWavefunctions/tests are compiled into build/tests/bin/test_wavefunction.
All the unit test executables are collected under ctest with the unit label.
When checking the whole code, it is useful to run through CMake (cmake -L unit).
When working on an individual directory, it is useful to run the individual executable.
Some of the tests reference input files. The unit test CMake setup places those input files in particular locations under the tests directory (e.g., tests/xml_test). The individual test needs to be run from that directory to find the expected input files.
Command line options are available on the unit test executables. Some of the more useful ones are
List command line options.
List all the tests in the executable.
A test name can be given on the command line to execute just that test. This is useful when iterating on a particular test or when running in the debugger. Test names often contain spaces, so most command line environments require enclosing the test name in single or double quotes.
Example¶
The first example is one test from src/Numerics/tests/test_grid_functor.cpp.
TEST_CASE("double_1d_grid_functor", "[numerics]")
{
LinearGrid<double> grid;
OneDimGridFunctor<double> f(&grid);
grid.set(0.0, 1.0, 3);
REQUIRE(grid.size() == 3);
REQUIRE(grid.rmin() == 0.0);
REQUIRE(grid.rmax() == 1.0);
REQUIRE(grid.dh() == Approx(0.5));
REQUIRE(grid.dr(1) == Approx(0.5));
}
The test function declaration is
TEST_CASE("double_1d_grid_functor","[numerics]").
The first argument is the test name, and it must be unique in the test suite.
The second argument is an optional list of tags. Each tag is a name surrounded by brackets ("[tag1][tag2]"). It can also be the empty string.
The REQUIRE macro accepts expressions with C++ comparison operators and records an error if the value of the expression is false.
Floating point numbers may have small differences due to roundoff, etc. The Approx class adds some tolerance to the comparison. Place it on either side of the comparison (e.g., Approx(a) == 0.3 or a = Approx(0.3)). To adjust the tolerance, use the epsilon and scale methods to Approx (REQUIRE(Approx(a).epsilon(0.001) = 0.3);.
Expected output¶
When running the test executables individually, the output of a run with no failures should look like
===============================================================================
All tests passed (26 assertions in 4 test cases)
A test with failures will look like
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
test_numerics is a Catch v1.4.0 host application.
Run with -? for options
-------------------------------------------------------------------------------
double_1d_grid_functor
-------------------------------------------------------------------------------
/home/user/qmcpack/src/Numerics/tests/test_grid_functor.cpp:29
...............................................................................
/home/user/qmcpack/src/Numerics/tests/test_grid_functor.cpp:39: FAILED:
REQUIRE( grid.dh() == Approx(0.6) )
with expansion:
0.5 == Approx( 0.6 )
===============================================================================
test cases: 4 | 3 passed | 1 failed
assertions: 25 | 24 passed | 1 failed
Adding tests¶
Three scenarios are covered here: adding a new test in an existing file, adding a new test file, and adding a new test directory.
Adding a test to existing file¶
Copy an existing test or from the example shown here. Be sure to change the test name.
Adding a test file¶
When adding a new test file,
create a file in the test directory, or copy from an existing file. Add the file name to the ADD_EXECUTABLE in the CMakeLists.txt file in that directory.
One (and only one) file must define the main function for the test executable by defining CATCH_CONFIG_MAIN before including the Catch header. If more than one file defines this value, there will be linking errors about multiply defined values.
Some of the tests need to shut down MPI properly to avoid extraneous error messages. Those tests include Message/catch_mpi_main.hpp instead of defining CATCH_CONFIG_MAIN.
Adding a test directory¶
Copy the CMakeLists.txt file from an existing tests directory.
Change the SRC_DIR name and the files in the ADD_EXECUTABLES line. The libraries to link in TARGET_LINK_LIBRARIES may need to be updated.
Add the new test directory to src/CMakeLists.txt in the BUILD_UNIT_TESTS section near the end.
Testing with random numbers¶
Many algorithms and parts of the code depend on random numbers, which makes validating the results difficult. One solution is to verify that certain properties hold for any random number. This approach is valuable at some levels of testing, but is unsatisfying at the unit test level.
The Utilities directory contains a “fake” random number generator that can be used for deterministic tests of these parts of the code.
Currently it outputs a single, fixed value every time it is called, but it could be expanded to produce more varied, but still deterministic, sequences.
See src/QMCDrivers/test_vmc.cpp for an example of using the fake random number generator.
QMCPACK Design and Feature Documentation¶
This section contains information on the overall design of QMCPACK. Also included are detailed explanations/derivations of major features and algorithms present in the code.
QMCPACK design¶
TBD.
Feature: Optimized long-range breakup (Ewald)¶
Consider a group of particles interacting with long-range central potentials, \(v^{\alpha \beta}(|r^{\alpha}_i - r^{\beta}_j|)\), where the Greek superscripts represent the particle species (e.g., \(\alpha=\text{electron}\), \(\beta=\text{proton}\)), and Roman subscripts refer to particle number within a species. We can then write the total interaction energy for the system as
The long-range problem¶
Consider such a system in periodic boundary conditions in a cell defined by primitive lattice vectors \(\mathbf{a}_1\), \(\mathbf{a}_2\), and \(\mathbf{a}_3\). Let \(\mathbf{L}\equiv n_1 \mathbf{a}_1 + n_2 \mathbf{a}_2 + n_3\mathbf{a}_3\) be a direct lattice vector. Then the interaction energy per cell for the periodic system is given by
where \(N^\alpha\) is the number particles of species \(\alpha\). If the potentials \(v^{\alpha\beta}(r)\) are indeed long-range, the summation over direct lattice vectors will not converge in this naive form. A solution to the problem was posited by Ewald. We break the central potentials into two pieces—a short-range and a long-range part defined by
We will perform the summation over images for the short-range part in real space, while performing the sum for the long-range part in reciprocal space. For simplicity, we choose \(v^{\alpha \beta}_s(r)\) so that it is identically zero at the half-the-box length. This eliminates the need to sum over images in real space.
Reciprocal-space sums¶
Heterologous terms¶
We begin with (87), starting with the heterologous terms (i.e., the terms involving particles of different species). The short-range terms are trivial, so we neglect them here.
We insert the resolution of unity in real space twice:
Here, the \(\mathbf{k}\) summations are over reciprocal lattice vectors given by \(\mathbf{k}= m_1 \mathbf{b}_1 + m_2\mathbf{b}_2 + m_3\mathbf{b}_3\), where
We note that \(\mathbf{k}\cdot \mathbf{L}= 2\pi(n_1 m_1 + n_2 m_2 + n_3 m_3)\).
where \(\Omega\) is the volume of the cell. Here we have used the fact that summing over all cells of the integral over the cell is equivalent to integrating over all space.
We have
Then, performing the integrations we have
We now separate the summations, yielding
Summing over \(\mathbf{k}\) and \(\mathbf{k}'\), we have
We can simplify the calculation a bit further by rearranging the sums over species:
Homologous terms¶
We now consider the terms involving particles of the same species interacting with each other. The algebra is very similar to the preceding, with the slight difficulty of avoiding the self-interaction term.
Madelung terms¶
Let us now consider the Madelung term for a single particle of species \(\alpha\). This term corresponds to the interaction of a particle with all of its periodic images.
\(\mathbf{k}=\mathbf{0}\) terms¶
Thus far, we have neglected what happens at the special point \(\mathbf{k}= \mathbf{0}\). For many long-range potentials, such as the Coulomb potential, \(v_k^{\alpha \alpha}\) diverges for \(k=0\). However, we recognize that for a charge-neutral system, the divergent part of the terms cancel each other. If all the potential in the system were precisely Coulomb, the \(\mathbf{k}=\mathbf{0}\) terms would cancel precisely, yielding zero. For systems involving PPs, however, it may be that the resulting term is finite, but nonzero. Consider the terms from \(\mathbf{k}=\mathbf{0}\):
Next, we must compute \(v^{\alpha \beta}_{k=0}\).
We recognize that this integral will not converge because of the large-\(r\) behavior. However, we recognize that when we do the sum in (102), the large-\(r\) parts of the integrals will cancel precisely. Therefore, we define
where \(r_{\text{end}}\) is some cutoff value after which the potential tails precisely cancel.
Neutralizing background terms¶
For systems with a net charge, such as the one-component plasma (jellium), we add a uniform background charge, which makes the system neutral. When we do this, we must add a term that comes from the interaction of the particle with the neutral background. It is a constant term, independent of the particle positions. In general, we have a compensating background for each species, which largely cancels out for neutral systems.
where \(v^{\alpha \beta}_{s\mathbf{0}}\) is given by
Combining terms¶
Here, we sum all of the terms we computed in the previous sections:
Computing the reciprocal potential¶
Now we return to (92). Without loss of generality, we define for convenience \(\mathbf{k}= k\hat{\mathbf{z}}\).
We do the angular integral first. By inversion symmetry, the imaginary part of the integral vanishes, yielding
The Coulomb potential¶
For the case of the Coulomb potential, the preceding integral is not formally convergent if we do the integral naively. We may remedy the situation by including a convergence factor, \(e^{-k_0 r}\). For a potential of the form \(v^{\text{coul}}(r) = q_1 q_2/r\), this yields
Allowing the convergence factor to tend to zero, we have
For more generalized potentials with a Coulomb tail, we cannot evaluate (109) numerically but must handle the coulomb part analytically. In this case, we have
Efficient calculation methods¶
Fast computation of \(\rho_\mathbf{k}\)¶
We wish to quickly calculate the quantity
First, we write
Now, we note that
This allows us to recursively build up an array of the \(C^{i\alpha}\)s and then compute \(\rho_\mathbf{k}\) for all \(\mathbf{k}\)-vectors by looping over all k-vectors, requiring only two complex multiplies per particle per \(\mathbf{k}\).
Algorithm to quickly calculate \(\rho_\mathbf{k}^\alpha\).
Gaussian charge screening breakup¶
This original approach to the short- and long-range breakup adds an opposite screening charge of Gaussian shape around each point charge. It then removes the charge in the long-range part of the potential. In this potential,
where \(\alpha\) is an adjustable parameter used to control how short ranged the potential should be. If the box size is \(L\), a typical value for \(\alpha\) might be \(7/(Lq_1 q_2)\). We should note that this form for the long-range potential should also work for any general potential with a Coulomb tail (e.g., pseudo-Hamiltonian potentials. For this form of the long-range potential, we have in \(k\)-space
Optimized breakup method¶
In this section, we undertake the task of choosing a long-range/short-range partitioning of the potential, which is optimal in that it minimizes the error for given real and \(k\)-space cutoffs \(r_c\) and \(k_c\). Here, we slightly modify the method introduced by Natoli and Ceperley [NC95]. We choose \(r_c = \frac{1}{2}\min\{L_i\}\) so that we require the nearest image in real-space summation. \(k_c\) is then chosen to satisfy our accuracy requirements.
Here we modify our notation slightly to accommodate details not previously required. We restrict our discussion to the interaction of two particle species (which may be the same), and drop our species indices. Thus, we are looking for short- and long-range potentials defined by
Define \(v^s_k\) and \(v^\ell_k\) to be the respective Fourier transforms of the previous equation. The goal is to choose \(v_s(r)\) such that its value and first two derivatives vanish at \(r_c\), while making \(v^\ell(r)\) as smooth as possible so that \(k\)-space components, \(v^\ell_k\), are very small for \(k>k_c\). Here, we describe how to do this in an optimal way.
Define the periodic potential, \(V_p\), as
where \(\mathbf{r}\) is the displacement between the two particles and \(\mathbf{l}\) is a lattice vector. Let us then define our approximation to this potential, \(V_a\), as
Now, we seek to minimize the RMS error over the cell,
We may write
where
We now need a basis in which to represent the broken-up potential. We may choose to represent either \(v^s(r)\) or \(v^\ell(r)\) in a real-space basis. Natoli and Ceperley chose the former in their paper. We choose the latter for a number of reasons. First, singular potentials are difficult to represent in a linear basis unless the singularity is explicitly included. This requires a separate basis for each type of singularity. The short-range potential may have an arbitrary number of features for \(r<r_c\) and still be a valid potential. By construction, however, we desire that \(v^\ell(r)\) be smooth in real-space so that its Fourier transform falls off quickly with increasing \(k\). We therefore expect that, in general, \(v^\ell(r)\) should be well represented by fewer basis functions than \(v^s(r)\). Therefore, we define
where the \(h_n(r)\) are a set of \(J\) basis functions. We require that the two cases agree on the value and first two derivatives at \(r_c\). We may then define
Similarly, we define
Therefore,
Because \(v^s(r)\) goes identically to zero at the box edge, inside the cell we may write
We then write
We see that if we define
Then
which then cancels out all terms for \(|\mathbf{k}| < k_c\). Then we have
We expand the summation,
We take the derivative w.r.t. \(t_{m}\):
We integrate w.r.t. \(\mathbf{r}\), yielding a Kronecker \(\delta\).
Summing over \(\mathbf{k}'\) and equating the derivative to zero, we find the minimum of our error function is given by
which is equivalent in form to Equation 19 in [NC95], where we have \(x_k\) instead of \(V_k\). Thus, we see that we can optimize the short- or long-range potential simply by choosing to use \(V_k\) or \(x_k\) in the preceding equation. We now define
Thus, it becomes clear that our minimization equations can be cast in the canonical linear form
Solution by SVD¶
In practice, we note that the matrix \(\mathbf{A}\) frequently becomes singular in practice. For this reason, we use the singular value decomposition to solve for \(t_n\). This factorization decomposes \(A\) as
where \(\mathbf{U}^T\mathbf{U}= \mathbf{V}^T\mathbf{V}= 1\) and \(\mathbf{S}\) is diagonal. In this form, we have
where the parenthesized subscripts refer to columns. The advantage of this form is that if \(\mathbf{S}_{ii}\) is zero or very near zero, the contribution of the \(i^{\text{th}}\) of \(\mathbf{V}\) may be neglected since it represents a numerical instability and has little physical meaning. It represents the fact that the system cannot distinguish between two linear combinations of the basis functions. Using the SVD in this manner is guaranteed to be stable. This decomposition is available in LAPACK in the DGESVD subroutine.
Constraining Values¶
Often, we wish to constrain the value of \(t_n\) to have a fixed value to enforce a boundary condition, for example. To do this, we define
We then define \(\mathbf{A}^*\) as \(\mathbf{A}\) with the \(n^{\text{th}}\) row and column removed and \(\mathbf{b}^*\) as \(\mathbf{b}'\) with the \(n^{\text{th}}\) element removed. Then we solve the reduced equation \(\mathbf{A}^* \mathbf{t}^* = \mathbf{b}^*\) and finally insert \(t_n\) back into the appropriate place in \(\mathbf{t}^*\) to recover the complete, constrained vector \(\mathbf{t}\). This may be trivially generalized to an arbitrary number of constraints.
The LPQHI basis¶
The preceding discussion is general and independent of the basis used to represent \(v^\ell(r)\). In this section, we introduce a convenient basis of localized interpolant functions, similar to those used for splines, which have a number of properties that are convenient for our purposes.
First, we divide the region from 0 to \(r_c\) into \(M-1\) subregions, bounded above and below by points we term knots, defined by \(r_j \equiv j\Delta\), where \(\Delta \equiv r_c/(M-1)\). We then define compact basis elements, \(h_{j\alpha}\), which span the region \([r_{j-1},r_{j+1}]\), except for \(j=0\) and \(j=M\). For \(j=0\), only the region \([r_0,r_1]\), while for \(j=M\), only \([r_{M-1}, r_M]\). Thus, the index \(j\) identifies the knot the element is centered on, while \(\alpha\) is an integer from 0 to 2 indicating one of three function shapes. The dual index can be mapped to the preceding single index by the relation \(n = 3j + \alpha\). The basis functions are then defined as
where the matrix \(S_{\alpha n}\) is given by
Fig. 26 shows plots of these function shapes.
Basis functions \(h_{j0}\), \(h_{j1}\), and \(h_{j2}\) are shown. We note that at the left and right extremes, the values and first two derivatives of the functions are zero; while at the center, \(h_{j0}\) has a value of 1, \(h_{j1}\) has a first derivative of 1, and \(h_{j2}\) has a second derivative of 1.¶
The basis functions have the property that at the left and right extremes (i.e., \(r_{j-1}\) and \(r_{j+1}\)) their values and first two derivatives are zero. At the center, \(r_j\), we have the properties
These properties allow the control of the value and first two derivatives of the represented function at any knot value simply by setting the coefficients of the basis functions centered around that knot. Used in combination with the method described in Constraining Values, boundary conditions can easily be enforced. In our case, we wish require that
This ensures that \(v^s\) and its first two derivatives vanish at \(r_c\).
Fourier coefficients¶
We wish now to calculate the Fourier transforms of the basis functions, defined as
We may then write,
where
We then further make the definition that
It can then be shown that
Note that these equations correct typographical errors present in [NC95].
Enumerating \(k\)-points¶
We note that the summations over \(k\), which are ubiquitous in this paper, require enumeration of the \(k\)-vectors. In particular, we should sum over all \(|\mathbf{k}| > k_c\). In practice, we must limit our summation to some finite cutoff value \(k_c < |\mathbf{k}| < k_{\text{max}}\), where \(k_{\text{max}}\) should be on the order of \(3,000/L\), where \(L\) is the minimum box dimension. Enumerating these vectors in a naive fashion even for this finite cutoff would prove quite prohibitive, as it would require \(\sim10^9\) vectors.
Our first optimization comes in realizing that all quantities in this calculation require only \(|\mathbf{k}|\) and not \(\mathbf{k}\) itself. Thus, we may take advantage of the great degeneracy of \(|\mathbf{k}|\). We create a list of \((k,N)\) pairs, where \(N\) is the number of vectors with magnitude \(k\). We make nested loops over \(n_1\), \(n_2\), and \(n_3\), yielding \(\mathbf{k}= n_1 \mathbf{b}_1 + n_2 \mathbf{b}_2 + n_3 \mathbf{b}_3\). If \(|\mathbf{k}|\) is in the required range, we check to see whether there is already an entry with that magnitude on our list and increment the corresponding \(N\) if there is, or create a new entry if not. Doing so typically saves a factor of \(\sim200\) in storage and computation.
This reduction is not sufficient for large \(k_max\) since it requires that we still look over \(10^9\) entries. To further reduce costs, we may pick an intermediate cutoff, \(k_{\text{cont}}\), above which we will approximate the degeneracy assuming a continuum of \(k\)-points. We stop our exact enumeration at \(k_{\text{cont}}\) and then add \(\sim1,000\) points, \(k_i\), uniformly spaced between \(k_{\text{cont}}\) and \(k_{\text{max}}\). We then approximate the degeneracy by
where \(k_b = (k_i + k_{i+1})/2\) and \(k_a = (k_i + k_{i-1})\). In doing so, we typically reduce our total number of k-points to sum more than \(\sim2,500\) from the \(10^9\) we had to start.
Calculating \(x_k\)’s¶
The Coulomb potential¶
For \(v(r) = \frac{1}{r}\), \(x_k\) is given by
The \(1/r^2\) potential¶
For \(v(r) = \frac{1}{r^2}\), \(x_k\) is given by
where the sin integral, \(\text{Si}(z)\), is given by
Feature: Optimized long-range breakup (Ewald) 2¶
Given a lattice of vectors \(\mathbf{L}\), its associated reciprocal lattice of vectors \(\mathbf{k}\) and a function \(\psi(\mathbf{r})\) periodic on the lattice we define its Fourier transform \(\widetilde{\psi}(\mathbf{k})\) as
where we indicated both the cell domain and the cell volume by \(\Omega\). \(\psi(\mathbf{r})\) can then be expressed as
The potential generated by charges sitting on the lattice positions at a particular point \(\mathbf{r}\) inside the cell is given by
and its Fourier transform can be explicitly written as a function of \(V\) or \(v\)
where \(\mathbb{R}^3\) denotes the whole 3D space. We now want to find the best (“best” to be defined later) approximate potential of the form
where \(W(r)\) has been chosen to go smoothly to \(0\) when \(r=r_c\), being \(r_c\) lower or equal to the Wigner-Seitz radius of the cell. Note also the cutoff \(k_c\) on the momentum summation.
The best form of \(\widetilde{Y}(k)\) and \(W(r)\) is given by minimizing
or the reciprocal space equivalent
(164) follows from (163) and the unitarity (norm conservation) of the Fourier transform.
This last condition is minimized by
We now use a set of basis function \(c_i(r)\) vanishing smoothly at \(r_c\) to expand \(W(r)\); that is,
Inserting the reciprocal space expansion of \(\widetilde{W}\) in the second condition of (165) and minimizing with respect to \(t_i\) leads immediately to the linear system \(\mathbf{A}\mathbf{t}=\mathbf{b}\) where
Basis functions¶
The basis functions are splines. We define a uniform grid with \(N_{\text{knot}}\) uniformly spaced knots at position \(r_i=i\frac{r_c}{N_{\text{knot}}}\), where \(i\in[0,N_{\text{knot}}-1]\). On each knot we center \(m+1\) piecewise polynomials \(c_{i\alpha}(r)\) with \(\alpha\in[0,m]\), defined as
These functions and their derivatives are, by construction, continuous and odd (even) (with respect to \(r-r_i\rightarrow r_i-r\)) when \(\alpha\) is odd (even). We further ask them to satisfy
(The parity of the functions guarantees that the second constraint is satisfied at \(r_{i-1}\) as well). These constraints have a simple interpretation: the basis functions and their first \(m\) derivatives are \(0\) on the boundary of the subinterval where they are defined; the only function to have a nonzero \(\beta\)-th derivative in \(r_i\) is \(c_{i\beta}\). These \(2(m+1)\) constraints therefore impose \(\mathcal{N}=2m+1\). Inserting the definitions of (168) in the constraints of (169) leads to the set of \(2(m+1)\) linear equation that fixes the value of \(S_{\alpha n}\):
We can further simplify inserting the first of these equations into the second and write the linear system as
Fourier components of the basis functions in 3D¶
\(k\ne 0\), non-Coulomb case¶
We now need to evaluate the Fourier transform \(\widetilde{c}_{i\alpha}(k)\). Let us start by writing the definition
Because \(c_{i\alpha}\) is different from zero only inside the spherical crown defined by \(r_{i-1}<r<r_i\), we can conveniently compute the integral in spherical coordinates as
where we used the definition \(w_{\text{knot}}=1-\delta_{i0}\) and
obtained by integrating the angular part of the Fourier transform. Using the identity
and the definition
we rewrite Equation (174) as
Finally, using integration by part, we can define \(E^\pm_{in}\) recursively as
Starting from the \(n=0\) term,
\(k\ne 0\), Coulomb case¶
To efficiently treat the Coulomb divergence at the origin, it is convenient to use a basis set \(c_{i\alpha}^{\text{coul}}\) of the form
An equation identical to (174) holds but with the modified definition
which can be simply expressed using \(E^\pm_{in}(k)\) as
\(k=0\) Coulomb and non-Coulomb case¶
The definitions of \(D_{in}(k)\) given so far are clearly incompatible with the choice \(k=0\) (they involve division by \(k\)). For the non-Coulomb case, the starting definition is
Using the definition \(I_n^\pm=(\pm)^{n+1}\Delta/(n+1)\), we can express this as
For the Coulomb case, we get
Fourier components of the basis functions in 2D¶
(173) still holds provided we define
where \(C=1(=0)\) for the Coulomb(non-Coulomb) case. (185) is obtained using the integral definition of the zero order Bessel function of the first kind:
and the binomial expansion for \((r-r_i)^n\). The integrals can be computed recursively using the following identities:
The bottom equation of (187) is obtained using the second equation in the same set, integration by part, and the identity \(\int J_1(z) dz =-J_0(z)\). In the top equation, \(H_0\) and \(H_1\) are Struve functions.
Construction of the matrix elements¶
Using the previous equations, we can construct the matrix elements in (167) and proceed solving for \(t_i\). It is sometimes desirable to put some constraints on the value of \(t_i\). For example, when the Coulomb potential is concerned, we might want to set \(t_{0}=1\). If the first \(g\) variable is constrained by \(t_{m}=\gamma_m\) with \(m=[1,g]\), we can simply redefine (167) as
Feature: Cubic spline interpolation¶
We present the basic equations and algorithms necessary to construct and evaluate cubic interpolating splines in one, two, and three dimensions. Equations are provided for both natural and periodic boundary conditions.
One dimension¶
Let us consider the problem in which we have a function \(y(x)\) specified at a discrete set of points \(x_i\), such that \(y(x_i) = y_i\). We wish to construct a piecewise cubic polynomial interpolating function, \(f(x)\), which satisfies the following conditions:
\(f(x_i) = y_i\).
\(f'(x_i^-) = f'(x_i^+)\).
\(f''(x_i^-) = f''(x_i+)\).
Hermite interpolants¶
In our piecewise representation, we wish to store only the values \(y_i\) and first derivatives, \(y'_i\), of our function at each point \(x_i\), which we call knots. Given this data, we wish to construct the piecewise cubic function to use between \(x_i\) and \(x_{i+1}\), which satisfies the preceding conditions. In particular, we wish to find the unique cubic polynomial, \(P(x)\), satisfying
We then define the basis functions,
On the interval, \((x_i, x_{i+1}]\), we define the interpolating function
It can be easily verified that \(P(x)\) satisfies conditions of equations 1 through 3 of (189). It is now left to determine the proper values for the \(y'_i\,\)s such that the continuity conditions given previously are satisfied.
By construction, the value of the function and derivative will match at the knots; that is,
Then we must now enforce only the second derivative continuity:
Let us define
Then we may rearrange
This equation holds for all \(0<i<(N-1)\), so we have a tridiagonal set of equations. The equations for \(i=0\) and \(i=N-1\) depend on the boundary conditions we are using.
Periodic boundary conditions¶
For periodic boundary conditions, we have
Or, in matrix form, we have
The system is tridiagonal except for the two elements in the upper right and lower left corners. These terms complicate the solution a bit, although it can still be done in \(\mathcal{O}(N)\) time. We first proceed down the rows, eliminating the the first non-zero term in each row by subtracting the appropriate multiple of the previous row. At the same time, we eliminate the first element in the last row, shifting the position of the first non-zero element to the right with each iteration. When we get to the final row, we will have the value for \(y'_{N-1}\). We can then proceed back upward, backsubstituting values from the rows below to calculate all the derivatives.
Complete boundary conditions¶
If we specify the first derivatives of our function at the end points, we have what is known as complete boundary conditions. The equations in that case are trivial to solve:
This system is completely tridiagonal, and we may solve trivially by performing row eliminations downward, then proceeding upward as before.
Natural boundary conditions¶
If we do not have information about the derivatives at the boundary conditions, we may construct a natural spline, which assumes the second derivatives are zero at the end points of our spline. In this case our system of equations is the following:
with
Bicubic splines¶
It is possible to extend the cubic spline interpolation method to functions of two variables, that is, \(F(x,y)\). In this case, we have a rectangular mesh of points given by \(F_{ij} \equiv F(x_i,y_j)\). In the case of 1D splines, we needed to store the value of the first derivative of the function at each point, in addition to the value. In the case of bicubic splines, we need to store four quantities for each mesh point:
Consider the point \((x,y)\) at which we wish to interpolate \(F\). We locate the rectangle that contains this point, such that \(x_i <= x < x_{i+1}\) and \(y_i <= x < y_{i+1}\). Let
Then, we calculate the interpolated value as
Construction bicubic splines¶
We now address the issue of how to compute the derivatives that are needed for the interpolation. The algorithm is quite simple. For every \(x_i\), we perform the tridiagonal solution as we did in the 1D splines to compute \(F^y_{ij}\). Similarly, we perform a tridiagonal solve for every value of \(F^x_{ij}\). Finally, to compute the cross-derivative we may either to the tridiagonal solve in the \(y\) direction of \(F^x_{ij}\), or solve in the \(x\) direction for \(F^y_{ij}\) to obtain the cross-derivatives \(F^{xy}_{ij}\). Hence, only minor modifications to the \(1D\) interpolations are necessary.
Tricubic splines¶
Bicubic interpolation required two 4-component vectors and a \(4 \times 4\) matrix. By extension, tricubic interpolation requires three 4-component vectors and a \(4 \times 4 \times 4\) tensor. We summarize the forms of these vectors in the following:
Now, we can write
The appropriate derivatives of \(F\) may be computed by a generalization of the previous method used for bicubic splines.
Feature: B-spline orbital tiling (band unfolding)¶
In continuum QMC simulations, it is necessary to evaluate the electronic orbitals of a system at real-space positions hundreds of millions of times. It has been found that if these orbitals are represented in a localized, B-spline basis, each evaluation takes a small, constant time that is independent of system size.
Unfortunately, the memory required for storing the B-spline grows with the second power of the system size. If we are studying perfect crystals, however, this can be reduced to linear scaling if we tile the primitive cell. In this approach, a supercell is constructed by tiling the primitive cell \(N_1 \times N_2 \times N_3\) in the three lattice directions. The orbitals are then represented in real space only in the primitive cell and an \(N_1 \times N_2 \times N_3\) k-point mesh. To evaluate an orbital at any point in the supercell, it is only necessary to wrap that point back into the primitive cell, evaluate the spline, and then multiply the phase factor, \(e^{-i\mathbf{k}\cdot\mathbf{r}}\).
Here, we show that this approach can be generalized to a tiling constructed with a \(3\times 3\) nonsingular matrix of integers, of which the preceding approach is a special case. This generalization brings with it a number of advantages. The primary reason for performing supercell calculations in QMC is to reduce finite-size errors. These errors result from three sources: (1) the quantization of the crystal momentum, (2) the unphysical periodicity of the exchange-correlation (XC) hole of the electron, and (3) the kinetic-energy contribution from the periodicity of the long-range Jastrow correlation functions. The first source of error can be largely eliminated by twist averaging. If the simulation cell is large enough that XC hole does not “leak” out of the simulation cell, the second source can be eliminated either through use of the MPC interaction or the a postiori correction of Chiesa et al.
The satisfaction of the leakage requirement is controlled by whether the minimum distance, \(L_{\text{min}}\), from one supercell image to the next is greater than the width of the XC hole. Therefore, given a choice, it is best to use a cell that is as nearly cubic as possible since this choice maximizes \(L_{\text{min}}\) for a given number of atoms. Most often, however, the primitive cell is not cubic. In these cases, if we wish to choose the optimal supercell to reduce finite-size effects, we cannot use the simple primitive tiling scheme. In the generalized scheme we present, it is possible to choose far better supercells (from the standpoint of finite-size errors), while retaining the storage efficiency of the original tiling scheme.
The mathematics¶
Consider the set of primitive lattice vectors, \(\{\mathbf{a}^{\text{p}}_1, \mathbf{a}^{\text{p}}_2, \mathbf{a}^{\text{p}}_3\}\). We may write these vectors in a matrix, \(\mathbf{L}_p\), whose rows are the primitive lattice vectors. Consider a nonsingular matrix of integers, \(\mathbf{S}\). A corresponding set of supercell lattice vectors, \(\{\mathbf{a}^{\text{s}}_1, \mathbf{a}^{\text{s}}_2, \mathbf{a}^{\text{s}}_3\}\), can be constructed by the matrix product
If the primitive cell contains \(N_p\) atoms, the supercell will then contain \(N_s = |\det(\mathbf{S})| N_p\) atoms.
Example: FeO¶
As an example, consider the primitive cell for antiferromagnetic FeO (wustite) in the rocksalt structure. The primitive vectors, given in units of the lattice constant, are given by
This primitive cell contains two iron atoms and two oxygen atoms. It is a very elongated cell with acute angles and, thus, has a short minimum distance between adjacent images.
The smallest cubic cell consistent with the AFM ordering can be constructed with the matrix
This cell has \(2|\det(\mathbf{S})| = 32\) iron atoms and 32 oxygen atoms. In this example, we may perform the simulation in the 32-iron supercell, while storing the orbitals only in the 2-iron primitive cell, for a savings of a factor of 16.
The k-point mesh¶
To be able to use the generalized tiling scheme, we need to have the appropriate number of bands to occupy in the supercell. This may be achieved by appropriately choosing the k-point mesh. In this section, we explain how these points are chosen.
For simplicity, let us assume that the supercell calculation will be performed at the \(\Gamma\)-point. We can easily lift this restriction later. The fact that supercell calculation is performed at \(\Gamma\) implies that the k-points used in the primitive-cell calculation must be \(\mathbf{G}\)-vectors of the superlattice. This still leaves us with an infinite set of vectors. We may reduce this set to a finite number by considering that the orbitals must form a linearly independent set. Orbitals with k-vectors \(\mathbf{k}^p_1\) and \(\mathbf{k}^p_2\) will differ by at most a constant factor of \(\mathbf{k}^p_1 - \mathbf{k}^p_2 = \mathbf{G}^p\), where \(\mathbf{G}^p\) is a reciprocal lattice vector of the primitive cell.
Combining these two considerations gives us a prescription for generating our k-point mesh. The mesh may be taken to be the set of k-point which are G-vectors of the superlattice, reside within the first Brillouin zone (FBZ) of the primitive lattice, whose members do not differ a G-vector of the primitive lattice. Upon constructing such a set, we find that the number of included k-points is equal to \(|\det(\mathbf{S})|\), precisely the number we need. This can by considering the fact that the supercell has a volume \(|\det(\mathbf{S})|\) times that of the primitive cell. This implies that the volume of the supercell’s FBZ is \(|\det(\mathbf{S})|^{-1}\) times that of the primitive cell. Hence, \(|\det(\mathbf{S})|\) G-vectors of the supercell will fit in the FBZ of the primitive cell. Removing duplicate k-vectors, which differ from another by a reciprocal lattice vector, avoids double-counting vectors that lie on zone faces.
Formulae¶
Let \(\mathbf{A}\) be the matrix whose rows are the direct lattice vectors, \(\{\mathbf{a}_i\}\). Then, let the matrix \(\mathbf{B}\) be defined as \(2\pi(\mathbf{A}^{-1})^\dagger\). Its rows are the primitive reciprocal lattice vectors. Let \(\mathbf{A}_p\) and \(\mathbf{A}_s\) represent the primitive and superlattice matrices, respectively, and similarly for their reciprocals. Then we have
Consider a k-vector, \(\mathbf{k}\). It may alternatively be written in basis of reciprocal lattice vectors as \(\mathbf{t}\).
We may then express a twist vector of the primitive lattice, \(\mathbf{t}_p\), in terms of the superlattice.
This gives the simple result that twist vectors transform in precisely the same way as direct lattice vectors.
Feature: Hybrid orbital representation¶
where \(u_{lm}(r)\) are complex radial functions represented in some radial basis (e.g., splines).
Real spherical harmonics¶
If \(\phi(\mathbf{r})\) can be written as purely real, we can change the representation so that
where \(\bar{Y}_\ell^m\) are the real spherical harmonics defined by
We need then to relate \(\bar{u}_{\ell m}\) to \(u_{\ell m}\). We wish to express
in terms of \(\bar{u}_{\ell m}(r)\) and \(Y_{\ell m}\).
For \(m>0\),
For \(m<0\),
Then for \(m > 0\),
Projecting to atomic orbitals¶
Inside a muffin tin, orbitals are represented as products of spherical harmonics and 1D radial functions, primarily represented by splines. For a muffin tin centered at \(\mathbf{I}\),
Let use consider the case that our original representation for \(\phi(\mathbf{r})\) is of the form
Recall that
Conjugating,
Setting \(\mathbf{k}\rightarrow -k\),
Then,
Then
Comparing with (222),
If we had adopted the opposite sign convention for Fourier transforms (as is unfortunately the case in wfconvert), we would have
Feature: Electron-electron-ion Jastrow factor¶
The general form of the 3-body Jastrow we describe here depends on the three interparticle distances, \((r_{ij}, r_{iI}, r_{jI})\).
Note that we constrain the form of \(U\) such that \(U(r_{ij}, r_{iI},r_{jI}) = U(r_{ij}, r_{jI},r_{iI})\) to preserve the particle symmetry of the wavefunction. We then compute the gradient as
To compute the Laplacian, we take
We now wish to compute the gradient of these terms w.r.t. the ion position, \(I\).
For the gradient w.r.t. \(i\) of the gradient w.r.t. \(I\), the result is a tensor:
For the Laplacian,
Feature: Reciprocal-space Jastrow factors¶
Two-body Jastrow¶
This may be rewritten as
The \(-1\) is just a constant term and may be subsumed into the \(a_\mathbf{G}\) coefficient by a simple redefinition. This leaves a simple, but general, form:
We may now further constrain this on physical grounds. First, we recognize that \(J_2\) should be real. Since \(\rho_{-\mathbf{G}} = \rho_\mathbf{G}^*\), it follows that \(\rho_{\mathbf{G}}\rho_{-\mathbf{G}} = |\rho_\mathbf{G}|^2\) is real, so that \(a_\mathbf{G}\) must be real. Furthermore, we group the \(\mathbf{G}\)’s into \((+\mathbf{G}, -\mathbf{G})\) pairs and sum over only the positive vectors to save time.
One-body Jastrow¶
The 1-body Jastrow has a similar form but depends on the displacement from the electrons to the ions in the system.
where \(\alpha\) denotes the different ionic species. We may rewrite this in terms of \(\rho^{\alpha}_\mathbf{G}\):
where
We note that in the preceding equation, for a single configuration of the ions, the sum in brackets can be rewritten as a single constant. This implies that the per-species 1-body coefficients, \(b^\alpha_\mathbf{G}\), are underdetermined for single configuration of the ions. In general, if we have \(N\) species, we need \(N\) linearly independent ion configurations to uniquely determine \(b^{\alpha}_\mathbf{G}\). For this reason, we will drop the \(\alpha\) superscript of \(b_\mathbf{G}\) for now.
If we do desire to find a reciprocal space 1-body Jastrow that is transferable to systems with different ion positions and \(N\) ionic species, we must perform compute \(b_\mathbf{G}\) for \(N\) different ion configurations. We may then construct \(N\) equations at each value of \(\mathbf{G}\) to solve for the \(N\) unknown values, \(b^\alpha_\mathbf{G}\).
In the 2-body case, \(a_\mathbf{G}\) was constrained to be real by the fact that \(\rho_\mathbf{G}\rho_{-\mathbf{G}}\) was real. However, in the 1-body case, there is no such guarantee about \(\rho^\alpha_\mathbf{G}\rho_\mathbf{G}\). Therefore, in general, \(b_\mathbf{G}\) may be complex.
Symmetry considerations¶
For a crystal, many of the \(\mathbf{G}\)-vectors will be equivalent by symmetry. It is useful then to divide the \(\mathbf{G}\)-vectors into symmetry-related groups and then require that they share a common coefficient. Two vectors, \(\mathbf{G}\) and \(\mathbf{G}'\), may be considered to be symmetry related if, for all \(\alpha\) and \(\beta\)
For the 1-body term, we may also omit from our list of \(\mathbf{G}\)-vectors those for which all species structure factors are zero. This is equivalent to saying that if we are tiling a primitive cell we should include only the \(\mathbf{G}\)-vectors of the primitive cell and not the supercell. Note that this is not the case for the 2-body term since the XC hole should not have the periodicity of the primitive cell.
Gradients and Laplacians¶
The Laplacian is then given by
- NC95(1,2,3)
Vincent Natoli and David M. Ceperley. An optimized method for treating long-range potentials. Journal of Computational Physics, 117(1):171–178, 1995. URL: http://www.sciencedirect.com/science/article/pii/S0021999185710546, doi:10.1006/jcph.1995.1054.
Development Guide¶
The section gives guidance on how to extend the functionality of QMCPACK. Future examples will likely include topics such as the addition of a Jastrow function or a new QMC method.
QMCPACK coding standards¶
This chapter presents what we collectively have agreed are best practices for the code. This includes formatting style, naming conventions, documentation conventions, and certain prescriptions for C++ language use. At the moment only the formatting can be enforced in an objective fashion.
New development should follow these guidelines, and contributors are expected to adhere to them as they represent an integral part of our effort to continue QMCPACK as a world-class, sustainable QMC code. Although some of the source code has a ways to go to live up to these ideas, new code, even in old files, should follow the new conventions not the local conventions of the file whenever possible. Work on the code with continuous improvement in mind rather than a commitment to stasis.
The current workflow conventions for the project are described in the wiki on the GitHub repository. It will save you and all the maintainers considerable time if you read these and ask questions up front.
A PR should follow these standards before inclusion in the mainline. You can be sure of properly following the formatting conventions if you use clang-format. The mechanics of clang-format setup and use can be found at https://github.com/QMCPACK/qmcpack/wiki/Source-formatting.
The clang-format file found at qmcpack/src/.clang-format should be run over all code touched in a PR before a pull request is prepared. We also encourage developers to run clang-tidy with the qmcpack/src/.clang-tidy configuration over all new code.
As much as possible, try to break up refactoring, reformatting, feature, and bugs into separate, small PRs. Aim for something that would take a reviewer no more than an hour. In this way we can maintain a good collective development velocity.
Files¶
Each file should start with the header.
//////////////////////////////////////////////////////////////////////////////////////
// This file is distributed under the University of Illinois/NCSA Open Source License.
// See LICENSE file in top directory for details.
//
// Copyright (c) 2020 QMCPACK developers
//
// File developed by: Name, email, affiliation
//
// File created by: Name, email, affiliation
//////////////////////////////////////////////////////////////////////////////////////
If you make significant changes to an existing file, add yourself to the list of “developed by” authors.
File organization¶
Header files should be placed in the same directory as their implementations.
Unit tests should be written for all new functionality. These tests should be placed in a tests subdirectory below the implementations.
File names¶
Each class should be defined in a separate file with the same name as the class name. Use separate .cpp implementation files whenever possible to aid in incremental compilation.
The filenames of tests are composed by the filename of the object tested and the prefix test_.
The filenames of fake and mock objects used in tests are composed by the prefixes fake_ and mock_, respectively, and the filename of the object that is imitated.
Header files¶
All header files should be self-contained (i.e., not dependent on following any other header when it is included). Nor should they include files that are not necessary for their use (i.e., headers needed only by the implementation). Implementation files should not include files only for the benefit of files they include.
There are many header files that currently violate this.
Each header must use #define guards to prevent multiple inclusion.
The symbol name of the #define guards should be NAMESPACE(s)_CLASSNAME_H.
Includes¶
Header files should be included with the full path based on the src directory.
For example, the file qmcpack/src/QMCWaveFunctions/SPOSet.h should be included as
#include "QMCWaveFunctions/SPOSet.h"
Even if the included file is located in the same directory as the including file, this rule should be obeyed. Header files from external projects and standard libraries should be includes using the <iostream> convention, while headers that are part of the QMCPACK project should be included using the "our_header.h" convention.
For readability, we suggest using the following standard order of includes:
related header
std C library headers
std C++ library headers
Other libraries’ headers
QMCPACK headers
In each section the included files should be sorted in alphabetical order.
Naming¶
The balance between description and ease of implementation should be balanced such that the code remains self-documenting within a single terminal window. If an extremely short variable name is used, its scope must be shorter than \(\sim 40\) lines. An exception is made for template parameters, which must be in all CAPS.
Namespace names¶
Namespace names should be one word, lowercase.
Type and class names¶
Type and class names should start with a capital letter and have a capital letter for each new word.
Underscores (_) are not allowed.
Variable names¶
Variable names should not begin with a capital letter, which is reserved for type and class names. Underscores (_) should be used to separate words.
Class data members¶
Class private/protected data members names should follow the convention of variable names with a trailing underscore (_).
(Member) function names¶
Function names should start with a lowercase character and have a capital letter for each new word.
Lambda expressions¶
Named lambda expressions follow the naming convention for functions:
auto myWhatever = [](int i) { return i + 4; };
Macro names¶
Macro names should be all uppercase and can include underscores (_).
The underscore is not allowed as first or last character.
Test case and test names¶
Test code files should be named as follows:
class DiracMatrix;
//leads to
test_dirac_matrix.cpp
//which contains test cases named
TEST_CASE("DiracMatrix_update_row","[wavefunction][fermion]")
where the test case covers the updateRow and [wavefunction][fermion] indicates the test belongs to the fermion wavefunction functionality.
Comments¶
Comment style¶
Use the // Comment syntax for actual comments.
Use
/** base class for Single-particle orbital sets
*
* SPOSet stands for S(ingle)P(article)O(rbital)Set which contains
* a number of single-particle orbitals with capabilities of
* evaluating \f$ \psi_j({\bf r}_i)\f$
*/
or
///index in the builder list of sposets
int builder_index;
Documentation¶
Doxygen will be used for source documentation. Doxygen commands should be used when appropriate guidance on this has been decided.
File docs¶
Do not put the file name after the \file Doxygen command. Doxygen will fill it in for the file the tag appears in.
/** \file
* File level documentation
*/
Class docs¶
Every class should have a short description (in the header of the file) of what it is and what is does. Comments for public class member functions follow the same rules as general function comments. Comments for private members are allowed but are not mandatory.
Function docs¶
For function parameters whose type is non-const reference or pointer to non-const memory, it should be specified if they are input (In:), output (Out:) or input-output parameters (InOut:).
Example:
/** Updates foo and computes bar using in_1 .. in_5.
* \param[in] in_3
* \param[in] in_5
* \param[in,out] foo
* \param[out] bar
*/
//This is probably not what our clang-format would do
void computeFooBar(Type in_1, const Type& in_2, Type& in_3,
const Type* in_4, Type* in_5, Type& foo,
Type& bar);
Variable documentation¶
Name should be self-descriptive. If you need documentation consider renaming first.
Golden rule of comments¶
If you modify a piece of code, also adapt the comments that belong to it if necessary.
Formatting and “style”¶
Use the provided clang-format style in src/.clang-format to format .h, .hpp, .cu, and .cpp files. Many of the following rules will be applied to the code by clang-format, which should allow you to ignore most of them if you always run it on your modified code.
You should use clang-format support and the .clangformat file with your editor, use a Git precommit hook to run clang-format or run clang-format manually on every file you modify. However, if you see numerous formatting updates outside of the code you have modified, first commit the formatting changes in a separate PR.
Indentation¶
Indentation consists of two spaces. Do not use tabs in the code.
Line length¶
The length of each line of your code should be at most 120 characters.
Horizontal spacing¶
No trailing white spaces should be added to any line.
Use no space before a comma (,) and a semicolon (;), and add a space after them if they are not at the end of a line.
Preprocessor directives¶
The preprocessor directives are not indented. The hash is the first character of the line.
Binary operators¶
The assignment operators should always have spaces around them.
Unary operators¶
Do not put any space between an unary operator and its argument.
Types¶
The using syntax is preferred to typedef for type aliases.
If the actual type is not excessively long or complex, simply use it; renaming simple types makes code less understandable.
Pointers and references¶
Pointer or reference operators should go with the type. But understand the compiler reads them from right to left.
Type* var;
Type& var;
//Understand this is incompatible with multiple declarations
Type* var1, var2; // var1 is a pointer to Type but var2 is a Type.
Templates¶
The angle brackets of templates should not have any external or internal padding.
template<class C>
class Class1;
Class1<Class2<type1>> object;
Vertical spacing¶
Use empty lines when it helps to improve the readability of the code, but do not use too many. Do not use empty lines after a brace that opens a scope or before a brace that closes a scope. Each file should contain an empty line at the end of the file. Some editors add an empty line automatically, some do not.
Variable declarations and definitions¶
Avoid declaring multiple variables in the same declaration, especially if they are not fundamental types:
int x, y; // Not recommended Matrix a("my-matrix"), b(size); // Not allowed // Preferred int x; int y; Matrix a("my-matrix"); Matrix b(10);
Use the following order for keywords and modifiers in variable declarations:
// General type [static] [const/constexpr] Type variable_name; // Pointer [static] [const] Type* [const] variable_name; // Integer // the int is not optional not all platforms support long, etc. [static] [const/constexpr] [signedness] [size] int variable_name; // Examples: static const Matrix a(10); const double* const d(3.14); constexpr unsigned long l(42);
Function declarations and definitions¶
The return type should be on the same line as the function name. Parameters should also be on the same line unless they do not fit on it, in which case one parameter per line aligned with the first parameter should be used.
Also include the parameter names in the declaration of a function, that is,
// calculates a*b+c
double function(double a, double b, double c);
// avoid
double function(double, double, double);
//dont do this
double function(BigTemplatedSomething<double> a, BigTemplatedSomething<double> b,
BigTemplatedSomething<double> c);
//do this
double function(BigTemplatedSomething<double> a,
BigTemplatedSomething<double> b,
BigTemplatedSomething<double> c);
Conditionals¶
Examples:
if (condition)
statement;
else
statement;
if (condition)
{
statement;
}
else if (condition2)
{
statement;
}
else
{
statement;
}
Switch statement¶
Switch statements should always have a default case.
Example:
switch (var)
{
case 0:
statement1;
statement2;
break;
case 1:
statement1;
statement2;
break;
default:
statement1;
statement2;
}
Loops¶
Examples:
for (statement; condition; statement)
statement;
for (statement; condition; statement)
{
statement1;
statement2;
}
while (condition)
statement;
while (condition)
{
statement1;
statement2;
}
do
{
statement;
}
while (condition);
Class format¶
public, protected, and private keywords are not indented.
Example:
class Foo : public Bar
{
public:
Foo();
explicit Foo(int var);
void function();
void emptyFunction() {}
void setVar(const int var)
{
var_ = var;
}
int getVar() const
{
return var_;
}
private:
bool privateFunction();
int var_;
int var2_;
};
Constructor initializer lists¶
Examples:
// When everything fits on one line:
Foo::Foo(int var) : var_(var)
{
statement;
}
// If the signature and the initializer list do not
// fit on one line, the colon is indented by 4 spaces:
Foo::Foo(int var)
: var_(var), var2_(var + 1)
{
statement;
}
// If the initializer list occupies more lines,
// they are aligned in the following way:
Foo::Foo(int var)
: some_var_(var),
some_other_var_(var + 1)
{
statement;
}
// No statements:
Foo::Foo(int var)
: some_var_(var) {}
Namespace formatting¶
The content of namespaces is not indented. A comment should indicate when a namespace is closed. (clang-format will add these if absent). If nested namespaces are used, a comment with the full namespace is required after opening a set of namespaces or an inner namespace.
Examples:
namespace ns
{
void foo();
} // ns
namespace ns1
{
namespace ns2
{
// ns1::ns2::
void foo();
namespace ns3
{
// ns1::ns2::ns3::
void bar();
} // ns3
} // ns2
namespace ns4
{
namespace ns5
{
// ns1::ns4::ns5::
void foo();
} // ns5
} // ns4
} // ns1
QMCPACK C++ guidance¶
The guidance here, like any advice on how to program, should not be treated as a set of rules but rather the hard-won wisdom of many hours of suffering development. In the past, many rules were ignored, and the absolute worst results of that will affect whatever code you need to work with. Your PR should go much smoother if you do not ignore them.
Encapsulation¶
A class is not just a naming scheme for a set of variables and functions. It should provide a logical set of methods, could contain the state of a logical object, and might allow access to object data through a well-defined interface related variables, while preserving maximally ability to change internal implementation of the class.
Do not use struct as a way to avoid controlling access to the class. Only in rare cases where a class is a fully public data structure struct is this appropriate. Ignore (or fix one) the many examples of this in QMCPACK.
Do not use inheritance primarily as a means to break encapsulation. If your class could aggregate or compose another class, do that, and access it solely through its public interface. This will reduce dependencies.
Casting¶
In C++ source, avoid C style casts; they are difficult to search for and imprecise in function. An exception is made for controlling implicit conversion of simple numerical types.
Explicit C++ style casts make it clear what the safety of the cast is and what sort of conversion is expected to be possible.
int c = 2;
int d = 3;
double a;
a = (double)c / d; // Ok
const class1 c1;
class2* c2;
c2 = (class2*)&c1; // NO
SPOSetAdvanced* spo_advanced = new SPOSetAdvanced();
SPOSet* spo = (SPOSet*)spo_advanced; // NO
SPOSet* spo = static_cast<SPOSet*>(spo_advanced); // OK if upcast, dangerous if downcast
Pre-increment and pre-decrement¶
Use the pre-increment (pre-decrement) operator when a variable is incremented (decremented) and the value of the expression is not used. In particular, use the pre-increment (pre-decrement) operator for loop counters where i is not used:
for (int i = 0; i < N; ++i)
{
doSomething();
}
for (int i = 0; i < N; i++)
{
doSomething(i);
}
The post-increment and post-decrement operators create an unnecessary copy that the compiler cannot optimize away in the case of iterators or other classes with overloaded increment and decrement operators.
Alternative operator representations¶
Alternative representations of operators and other tokens such as and, or, and not instead of &&, ||, and ! are not allowed.
For the reason of consistency, the far more common primary tokens should always be used.
Use of const¶
Add the
constqualifier to all function parameters that are not modified in the function body.For parameters passed by value, add only the keyword in the function definition.
Member functions should be specified const whenever possible.
// Declaration int computeFoo(int bar, const Matrix& m) // Definition int computeFoo(const int bar, const Matrix& m) { int foo = 42; // Compute foo without changing bar or m. // ... return foo; } class MyClass { int count_ ... int getCount() const { return count_;} }
Scalar estimator implementation¶
Introduction: Life of a specialized OperatorBase¶
Almost all observables in QMCPACK are implemented as specialized derived classes of the OperatorBase base class. Each observable is instantiated in HamiltonianFactory and added to QMCHamiltonian for tracking. QMCHamiltonian tracks two types of observables: main and auxiliary. Main observables contribute to the local energy. These observables are elements of the simulated Hamiltonian such as kinetic or potential energy. Auxiliary observables are expectation values of matrix elements that do not contribute to the local energy. These Hamiltonians do not affect the dynamics of the simulation. In the code, the main observables are labeled by “physical” flag; the auxiliary observables have “physical” set to false.
Initialization¶
When an <estimator type="est_type" name="est_name" other_stuff="value"/> tag is present in the <hamiltonian/> section, it is first read by HamiltonianFactory. In general, the type of the estimator will determine which specialization of OperatorBase should be instantiated, and a derived class with myName="est_name" will be constructed. Then, the put() method of this specific class will be called to read any other parameters in the <estimator/> XML node. Sometimes these parameters will instead be read by HamiltonianFactory because it can access more objects than OperatorBase.
Cloning¶
When OpenMP threads are spawned, the estimator will be cloned by the CloneManager, which is a parent class of many QMC drivers.
// In CloneManager.cpp
#pragma omp parallel for shared(w,psi,ham)
for(int ip=1; ip<NumThreads; ++ip)
{
wClones[ip]=new MCWalkerConfiguration(w);
psiClones[ip]=psi.makeClone(*wClones[ip]);
hClones[ip]=ham.makeClone(*wClones[ip],*psiClones[ip]);
}
In the preceding snippet, ham is the reference to the estimator on the master thread. If the implemented estimator does not allocate memory for any array, then the default constructor should suffice for the makeClone method.
// In SpeciesKineticEnergy.cpp
OperatorBase* SpeciesKineticEnergy::makeClone(ParticleSet& qp, TrialWaveFunction& psi)
{
return new SpeciesKineticEnergy(*this);
}
If memory is allocated during estimator construction (usually when parsing the XML node in the put method), then the makeClone method should perform the same initialization on each thread.
OperatorBase* LatticeDeviationEstimator::makeClone(ParticleSet& qp, TrialWaveFunction& psi)
{
LatticeDeviationEstimator* myclone = new LatticeDeviationEstimator(qp,spset,tgroup,sgroup);
myclone->put(input_xml);
return myclone;
}
Evaluate¶
After the observable class (derived class of OperatorBase) is constructed and prepared (by the put() method), it is ready to be used in a QMCDriver. A QMCDriver will call H.auxHevaluate(W,thisWalker) after every accepted move, where H is the QMCHamiltonian that holds all main and auxiliary Hamiltonian elements, W is a MCWalkerConfiguration, and thisWalker is a pointer to the current walker being worked on. As shown in the following, this function goes through each auxiliary Hamiltonian element and evaluates it using the current walker configuration. Under the hood, observables are calculated and dumped to the main particle set’s property list for later collection.
// In QMCHamiltonian.cpp
// This is more efficient.
// Only calculate auxH elements if moves are accepted.
void QMCHamiltonian::auxHevaluate(ParticleSet& P, Walker_t& ThisWalker)
{
#if !defined(REMOVE_TRACEMANAGER)
collect_walker_traces(ThisWalker,P.current_step);
#endif
for(int i=0; i<auxH.size(); ++i)
{
auxH[i]->setHistories(ThisWalker);
RealType sink = auxH[i]->evaluate(P);
auxH[i]->setObservables(Observables);
#if !defined(REMOVE_TRACEMANAGER)
auxH[i]->collect_scalar_traces();
#endif
auxH[i]->setParticlePropertyList(P.PropertyList,myIndex);
}
}
For estimators that contribute to the local energy (main observables), the return value of evaluate() is used in accumulating the local energy. For auxiliary estimators, the return value is not used (sink local variable above); only the value of Value is recorded property lists by the setObservables() method as shown in the preceding code snippet. By default, the setObservables() method will transfer auxH[i]->Value to P.PropertyList[auxH[i]->myIndex]. The same property list is also kept by the particle set being moved by QMCDriver. This list is updated by auxH[i]->setParticlePropertyList(P.PropertyList,myIndex), where myIndex is the starting index of space allocated to this specific auxiliary Hamiltonian in the property list kept by the target particle set P.
Collection¶
The actual statistics are collected within the QMCDriver, which owns an EstimatorManager object. This object (or a clone in the case of multithreading) will be registered with each mover it owns. For each mover (such as VMCUpdatePbyP derived from QMCUpdateBase), an accumulate() call is made, which by default, makes an accumulate(walkerset) call to the EstimatorManager it owns. Since each walker has a property set, EstimatorManager uses that local copy to calculate statistics. The EstimatorManager performs block averaging and file I/O.
Single scalar estimator implementation guide¶
Almost all of the defaults can be used for a single scalar observable. With any luck, only the put() and evaluate() methods need to be implemented. As an example, this section presents a step-by-step guide for implementing a verb|SpeciesKineticEnergy| estimator that calculates the kinetic energy of a specific species instead of the entire particle set. For example, a possible input to this estimator can be:
<estimator type="specieskinetic" name="ukinetic" group="u"/>
<estimator type="specieskinetic" name="dkinetic" group="d"/>.
This should create two extra columns in the scalar.dat file that contains the kinetic energy of the up and down electrons in two separate columns. If the estimator is properly implemented, then the sum of these two columns should be equal to the default Kinetic column.
Barebone¶
The first step is to create a barebone class structure for this simple scalar estimator. The goal is to be able to instantiate this scalar estimator with an XML node and have it print out a column of zeros in the scalar.dat file.
To achieve this, first create a header file “SpeciesKineticEnergy.h” in the QMCHamiltonians folder, with only the required functions declared as follows:
// In SpeciesKineticEnergy.h
#ifndef QMCPLUSPLUS_SPECIESKINETICENERGY_H
#define QMCPLUSPLUS_SPECIESKINETICENERGY_H
#include <Particle/WalkerSetRef.h>
#include <QMCHamiltonians/OperatorBase.h>
namespace qmcplusplus
{
class SpeciesKineticEnergy: public OperatorBase
{
public:
SpeciesKineticEnergy(ParticleSet& P):tpset(P){ };
bool put(xmlNodePtr cur); // read input xml node, required
bool get(std::ostream& os) const; // class description, required
Return_t evaluate(ParticleSet& P);
inline Return_t evaluate(ParticleSet& P, std::vector<NonLocalData>& Txy)
{ // delegate responsity inline for speed
return evaluate(P);
}
// pure virtual functions require overrider
void resetTargetParticleSet(ParticleSet& P) { } // required
OperatorBase* makeClone(ParticleSet& qp, TrialWaveFunction& psi); // required
private:
ParticleSet& tpset;
}; // SpeciesKineticEnergy
} // namespace qmcplusplus
#endif
Notice that a local reference tpset to the target particle set P is saved in the constructor. The target particle set carries much information useful for calculating observables. Next, make “SpeciesKineticEnergy.cpp,” and make vacuous definitions.
// In SpeciesKineticEnergy.cpp
#include <QMCHamiltonians/SpeciesKineticEnergy.h>
namespace qmcplusplus
{
bool SpeciesKineticEnergy::put(xmlNodePtr cur)
{
return true;
}
bool SpeciesKineticEnergy::get(std::ostream& os) const
{
return true;
}
SpeciesKineticEnergy::Return_t SpeciesKineticEnergy::evaluate(ParticleSet& P)
{
Value = 0.0;
return Value;
}
OperatorBase* SpeciesKineticEnergy::makeClone(ParticleSet& qp, TrialWaveFunction& psi)
{
// no local array allocated, default constructor should be enough
return new SpeciesKineticEnergy(*this);
}
} // namespace qmcplusplus
Now, head over to HamiltonianFactory and instantiate this observable if an XML node is found requesting it. Look for “gofr” in HamiltonianFactory.cpp, for example, and follow the if block.
// In HamiltonianFactory.cpp
#include <QMCHamiltonians/SpeciesKineticEnergy.h>
else if(potType =="specieskinetic")
{
SpeciesKineticEnergy* apot = new SpeciesKineticEnergy(*target_particle_set);
apot->put(cur);
targetH->addOperator(apot,potName,false);
}
The last argument of addOperator() (i.e., the false flag) is crucial. This tells QMCPACK that the observable we implemented is not a physical Hamiltonian; thus, it will not contribute to the local energy. Changes to the local energy will alter the dynamics of the simulation. Finally, add “SpeciesKineticEnergy.cpp” to HAMSRCS in “CMakeLists.txt” located in the QMCHamiltonians folder. Now, recompile QMCPACK and run it on an input that requests <estimator type="specieskinetic" name="ukinetic"/> in the hamiltonian block. A column of zeros should appear in the scalar.dat file under the name “ukinetic.”
Evaluate¶
The evaluate() method is where we perform the calculation of the desired observable. In a first iteration, we will simply hard-code the name and mass of the particles.
// In SpeciesKineticEnergy.cpp
#include <QMCHamiltonians/BareKineticEnergy.h> // laplaician() defined here
SpeciesKineticEnergy::Return_t SpeciesKineticEnergy::evaluate(ParticleSet& P)
{
std::string group="u";
RealType minus_over_2m = -0.5;
SpeciesSet& tspecies(P.getSpeciesSet());
Value = 0.0;
for (int iat=0; iat<P.getTotalNum(); iat++)
{
if (tspecies.speciesName[ P.GroupID(iat) ] == group)
{
Value += minus_over_2m*laplacian(P.G[iat],P.L[iat]);
}
}
return Value;
// Kinetic column has:
// Value = -0.5*( Dot(P.G,P.G) + Sum(P.L) );
}
Voila—you should now be able to compile QMCPACK, rerun, and see that the values in the “ukinetic” column are no longer zero. Now, the only task left to make this basic observable complete is to read in the extra parameters instead of hard-coding them.
Parse extra input¶
The preferred method to parse extra input parameters in the XML node is to implement the put() function of our specific observable. Suppose we wish to read in a single string that tells us whether to record the kinetic energy of the up electron (group=“u”) or the down electron (group=“d”). This is easily achievable using the OhmmsAttributeSet class,
// In SpeciesKineticEnergy.cpp
#include <OhmmsData/AttributeSet.h>
bool SpeciesKineticEnergy::put(xmlNodePtr cur)
{
// read in extra parameter "group"
OhmmsAttributeSet attrib;
attrib.add(group,"group");
attrib.put(cur);
// save mass of specified group of particles
SpeciesSet& tspecies(tpset.getSpeciesSet());
int group_id = tspecies.findSpecies(group);
int massind = tspecies.getAttribute("mass");
minus_over_2m = -1./(2.*tspecies(massind,group_id));
return true;
}
where we may keep “group” and “minus_over_2m” as local variables to our specific class.
// In SpeciesKineticEnergy.h
private:
ParticleSet& tpset;
std::string group;
RealType minus_over_2m;
Notice that the previous operations are made possible by the saved reference to the target particle set. Last but not least, compile and run a full example (i.e., a short DMC calculation) with the following XML nodes in your input:
<estimator type="specieskinetic" name="ukinetic" group="u"/>
<estimator type="specieskinetic" name="dkinetic" group="d"/>
Make sure the sum of the “ukinetic” and “dkinetic” columns is exactly the same as the Kinetic columns at every block.
For easy reference, a summary of the complete list of changes follows:
// In HamiltonianFactory.cpp
#include "QMCHamiltonians/SpeciesKineticEnergy.h"
else if(potType =="specieskinetic")
{
SpeciesKineticEnergy* apot = new SpeciesKineticEnergy(*targetPtcl);
apot->put(cur);
targetH->addOperator(apot,potName,false);
}
// In SpeciesKineticEnergy.h
#include <Particle/WalkerSetRef.h>
#include <QMCHamiltonians/OperatorBase.h>
namespace qmcplusplus
{
class SpeciesKineticEnergy: public OperatorBase
{
public:
SpeciesKineticEnergy(ParticleSet& P):tpset(P){ };
// xml node is read by HamiltonianFactory, eg. the sum of following should be equivalent to Kinetic
// <estimator name="ukinetic" type="specieskinetic" target="e" group="u"/>
// <estimator name="dkinetic" type="specieskinetic" target="e" group="d"/>
bool put(xmlNodePtr cur); // read input xml node, required
bool get(std::ostream& os) const; // class description, required
Return_t evaluate(ParticleSet& P);
inline Return_t evaluate(ParticleSet& P, std::vector<NonLocalData>& Txy)
{ // delegate responsity inline for speed
return evaluate(P);
}
// pure virtual functions require overrider
void resetTargetParticleSet(ParticleSet& P) { } // required
OperatorBase* makeClone(ParticleSet& qp, TrialWaveFunction& psi); // required
private:
ParticleSet& tpset; // reference to target particle set
std::string group; // name of species to track
RealType minus_over_2m; // mass of the species !! assume same mass
// for multiple species, simply initialize multiple estimators
}; // SpeciesKineticEnergy
} // namespace qmcplusplus
#endif
// In SpeciesKineticEnergy.cpp
#include <QMCHamiltonians/SpeciesKineticEnergy.h>
#include <QMCHamiltonians/BareKineticEnergy.h> // laplaician() defined here
#include <OhmmsData/AttributeSet.h>
namespace qmcplusplus
{
bool SpeciesKineticEnergy::put(xmlNodePtr cur)
{
// read in extra parameter "group"
OhmmsAttributeSet attrib;
attrib.add(group,"group");
attrib.put(cur);
// save mass of specified group of particles
int group_id = tspecies.findSpecies(group);
int massind = tspecies.getAttribute("mass");
minus_over_2m = -1./(2.*tspecies(massind,group_id));
return true;
}
bool SpeciesKineticEnergy::get(std::ostream& os) const
{ // class description
os << "SpeciesKineticEnergy: " << myName << " for species " << group;
return true;
}
SpeciesKineticEnergy::Return_t SpeciesKineticEnergy::evaluate(ParticleSet& P)
{
Value = 0.0;
for (int iat=0; iat<P.getTotalNum(); iat++)
{
if (tspecies.speciesName[ P.GroupID(iat) ] == group)
{
Value += minus_over_2m*laplacian(P.G[iat],P.L[iat]);
}
}
return Value;
}
OperatorBase* SpeciesKineticEnergy::makeClone(ParticleSet& qp, TrialWaveFunction& psi)
{ //default constructor
return new SpeciesKineticEnergy(*this);
}
} // namespace qmcplusplus
Multiple scalars¶
It is fairly straightforward to create more than one column in the scalar.dat file with a single observable class. For example, if we want a single SpeciesKineticEnergy estimator to simultaneously record the kinetic energies of all species in the target particle set, we only have to write two new methods: addObservables() and setObservables(), then tweak the behavior of evaluate(). First, we will have to override the default behavior of addObservables() to make room for more than one column in the scalar.dat file as follows,
// In SpeciesKineticEnergy.cpp
void SpeciesKineticEnergy::addObservables(PropertySetType& plist, BufferType& collectables)
{
myIndex = plist.size();
for (int ispec=0; ispec<num_species; ispec++)
{ // make columns named "$myName_u", "$myName_d" etc.
plist.add(myName + "_" + species_names[ispec]);
}
}
where “num_species” and “species_name” can be local variables initialized in the constructor. We should also initialize some local arrays to hold temporary data.
// In SpeciesKineticEnergy.h
private:
int num_species;
std::vector<std::string> species_names;
std::vector<RealType> species_kinetic,vec_minus_over_2m;
// In SpeciesKineticEnergy.cpp
SpeciesKineticEnergy::SpeciesKineticEnergy(ParticleSet& P):tpset(P)
{
SpeciesSet& tspecies(P.getSpeciesSet());
int massind = tspecies.getAttribute("mass");
num_species = tspecies.size();
species_kinetic.resize(num_species);
vec_minus_over_2m.resize(num_species);
species_names.resize(num_species);
for (int ispec=0; ispec<num_species; ispec++)
{
species_names[ispec] = tspecies.speciesName[ispec];
vec_minus_over_2m[ispec] = -1./(2.*tspecies(massind,ispec));
}
}
Next, we need to override the default behavior of setObservables() to transfer multiple values to the property list kept by the main particle set, which eventually goes into the scalar.dat file.
// In SpeciesKineticEnergy.cpp
void SpeciesKineticEnergy::setObservables(PropertySetType& plist)
{ // slots in plist must be allocated by addObservables() first
copy(species_kinetic.begin(),species_kinetic.end(),plist.begin()+myIndex);
}
Finally, we need to change the behavior of evaluate() to fill the local vector “species_kinetic” with appropriate observable values.
SpeciesKineticEnergy::Return_t SpeciesKineticEnergy::evaluate(ParticleSet& P)
{
std::fill(species_kinetic.begin(),species_kinetic.end(),0.0);
for (int iat=0; iat<P.getTotalNum(); iat++)
{
int ispec = P.GroupID(iat);
species_kinetic[ispec] += vec_minus_over_2m[ispec]*laplacian(P.G[iat],P.L[iat]);
}
Value = 0.0; // Value is no longer used
return Value;
}
That’s it! The SpeciesKineticEnergy estimator no longer needs the “group” input and will automatically output the kinetic energy of every species in the target particle set in multiple columns. You should now be able to run with
<estimator type="specieskinetic" name="skinetic"/> and check that the sum of all columns that start with “skinetic” is equal to the default “Kinetic” column.
HDF5 output¶
If we desire an observable that will output hundreds of scalars per simulation step (e.g., SkEstimator), then it is preferred to output to the stat.h5 file instead of the scalar.dat file for better organization. A large chunk of data to be registered in the stat.h5 file is called a “Collectable” in QMCPACK. In particular, if a OperatorBase object is initialized with UpdateMode.set(COLLECTABLE,1), then the “Collectables” object carried by the main particle set will be processed and written to the stat.h5 file, where “UpdateMode” is a bit set (i.e., a collection of flags) with the following enumeration:
// In OperatorBase.h
///enum for UpdateMode
enum {PRIMARY=0,
OPTIMIZABLE=1,
RATIOUPDATE=2,
PHYSICAL=3,
COLLECTABLE=4,
NONLOCAL=5,
VIRTUALMOVES=6
};
As a simple example, to put the two columns we produced in the previous section into the stat.h5 file, we will first need to declare that our observable uses “Collectables.”
// In constructor add:
hdf5_out = true;
UpdateMode.set(COLLECTABLE,1);
Then make some room in the “Collectables” object carried by the target particle set.
// In addObservables(PropertySetType& plist, BufferType& collectables) add:
if (hdf5_out)
{
h5_index = collectables.size();
std::vector<RealType> tmp(num_species);
collectables.add(tmp.begin(),tmp.end());
}
Next, make some room in the stat.h5 file by overriding the registerCollectables() method.
// In SpeciesKineticEnergy.cpp
void SpeciesKineticEnergy::registerCollectables(std::vector<observable_helper*>& h5desc, hid_t gid) const
{
if (hdf5_out)
{
std::vector<int> ndim(1,num_species);
observable_helper* h5o=new observable_helper(myName);
h5o->set_dimensions(ndim,h5_index);
h5o->open(gid);
h5desc.push_back(h5o);
}
}
Finally, edit evaluate() to use the space in the “Collectables” object.
// In SpeciesKineticEnergy.cpp
SpeciesKineticEnergy::Return_t SpeciesKineticEnergy::evaluate(ParticleSet& P)
{
RealType wgt = tWalker->Weight; // MUST explicitly use DMC weights in Collectables!
std::fill(species_kinetic.begin(),species_kinetic.end(),0.0);
for (int iat=0; iat<P.getTotalNum(); iat++)
{
int ispec = P.GroupID(iat);
species_kinetic[ispec] += vec_minus_over_2m[ispec]*laplacian(P.G[iat],P.L[iat]);
P.Collectables[h5_index + ispec] += vec_minus_over_2m[ispec]*laplacian(P.G[iat],P.L[iat])*wgt;
}
Value = 0.0; // Value is no longer used
return Value;
}
There should now be a new entry in the stat.h5 file containing the same columns of data as the stat.h5 file. After this check, we should clean up the code by
making “hdf5_out” and input flag by editing the put() method and
disabling output to
scalar.datwhen the “hdf5_out” flag is on.
Estimator output¶
Estimator definition¶
For simplicity, consider a local property \(O(\bf{R})\), where \(\bf{R}\) is the collection of all particle coordinates. An estimator for \(O(\bf{R})\) is a weighted average over walkers:
\(N^{tot}_{walker}\) is the total number of walkers collected in the entire simulation. Notice that \(N^{tot}_{walker}\) is typically far larger than the number of walkers held in memory at any given simulation step. \(w_i\) is the weight of walker \(i\).
In a VMC simulation, the weight of every walker is 1.0. Further, the number of walkers is constant at each step. Therefore, (248) simplifies to
Each walker \(\bf{R}_{s,e}\) is labeled by step index s and ensemble index e.
In a DMC simulation, the weight of each walker is different and may change from step to step. Further, the ensemble size varies from step to step. Therefore, (248) simplifies to
We will refer to the average in the \(\{\}\) as ensemble average and to the remaining averages as block average. The process of calculating \(O(\mathbf{R})\) is evaluate.
Class relations¶
A large number of classes are involved in the estimator collection process. They often have misleading class or method names. Check out the document gotchas in the following list:
EstimatorManageris an unused copy ofEstimatorManagerBase.EstimatorManagerBaseis the class used in the QMC drivers. (PR #371 explains this.)EstimatorManagerBase::Estimatorsis completely different fromQMCDriver::Estimators, which is subtly different fromOperatorBase::Estimators. The first is a list of pointers toScalarEstimatorBase. The second is the master estimator (one per MPI group). The third is the slave estimator that exists one per OpenMP thread.QMCHamiltonianis NOT a parent class ofOperatorBase. Instead,QMCHamiltonianowns two lists ofOperatorBasenamedHandauxH.QMCDriver::His NOT the same asQMCHamiltonian::H. The first is a pointer to aQMCHamiltonian.QMCHamiltonian::His a list.EstimatorManager::stopBlock(std::vector)is completely different fromEstimatorManager::stopBlock(RealType), which is the same asstopBlock(RealType, true)but that is subtly different fromstopBlock(RealType, false). The first three methods are intended to be called by the master estimator, which exists one per MPI group. The last method is intended to be called by the slave estimator, which exists one per OpenMP thread.
Estimator output stages¶
Estimators take four conceptual stages to propagate to the output files: evaluate, load ensemble, unload ensemble, and collect. They are easier to understand in reverse order.
Collect stage¶
File output is performed by the master EstimatorManager owned by QMCDriver. The first 8+ entries in EstimatorManagerBase::AverageCache will be written to scalar.dat. The remaining entries in AverageCache will be written to stat.h5. File writing is triggered by EstimatorManagerBase ::collectBlockAverages inside EstimatorManagerBase::stopBlock.
// In EstimatorManagerBase.cpp::collectBlockAverages
if(Archive)
{
*Archive << std::setw(10) << RecordCount;
int maxobjs=std::min(BlockAverages.size(),max4ascii);
for(int j=0; j<maxobjs; j++)
*Archive << std::setw(FieldWidth) << AverageCache[j];
for(int j=0; j<PropertyCache.size(); j++)
*Archive << std::setw(FieldWidth) << PropertyCache[j];
*Archive << std::endl;
for(int o=0; o<h5desc.size(); ++o)
h5desc[o]->write(AverageCache.data(),SquaredAverageCache.data());
H5Fflush(h_file,H5F_SCOPE_LOCAL);
}
EstimatorManagerBase::collectBlockAverages is triggered from the master-thread estimator via either stopBlock(std::vector) or stopBlock(RealType, true). Notice that file writing is NOT triggered by the slave-thread estimator method stopBlock(RealType, false).
// In EstimatorManagerBase.cpp
void EstimatorManagerBase::stopBlock(RealType accept, bool collectall)
{
//take block averages and update properties per block
PropertyCache[weightInd]=BlockWeight;
PropertyCache[cpuInd] = MyTimer.elapsed();
PropertyCache[acceptInd] = accept;
for(int i=0; i<Estimators.size(); i++)
Estimators[i]->takeBlockAverage(AverageCache.begin(),SquaredAverageCache.begin());
if(Collectables)
{
Collectables->takeBlockAverage(AverageCache.begin(),SquaredAverageCache.begin());
}
if(collectall)
collectBlockAverages(1);
}
// In ScalarEstimatorBase.h
template<typename IT>
inline void takeBlockAverage(IT first, IT first_sq)
{
first += FirstIndex;
first_sq += FirstIndex;
for(int i=0; i<scalars.size(); i++)
{
*first++ = scalars[i].mean();
*first_sq++ = scalars[i].mean2();
scalars_saved[i]=scalars[i]; //save current block
scalars[i].clear();
}
}
At the collect stage, calarEstimatorBase::scalars must be populated with ensemble-averaged data. Two derived classes of ScalarEstimatorBase are crucial: LocalEnergyEstimator will carry Properties, where as CollectablesEstimator will carry Collectables.
Unload ensemble stage¶
LocalEnergyEstimator::scalars are populated by
ScalarEstimatorBase::accumulate, whereas
CollectablesEstimator::scalars are populated by
CollectablesEstimator:: accumulate_all. Both accumulate methods
are triggered by EstimatorManagerBase::accumulate. One confusing
aspect about the unload stage is that
EstimatorManagerBase::accumulate has a master and a slave call
signature. A slave estimator such as QMCUpdateBase::Estimators
should unload a subset of walkers. Thus, the slave estimator should call
accumulate(W,it,it_end). However, the master estimator, such as
SimpleFixedNodeBranch::myEstimator, should unload data from the
entire walker ensemble. This is achieved by calling accumulate(W).
void EstimatorManagerBase::accumulate(MCWalkerConfiguration& W)
{ // intended to be called by master estimator only
BlockWeight += W.getActiveWalkers();
RealType norm=1.0/W.getGlobalNumWalkers();
for(int i=0; i< Estimators.size(); i++)
Estimators[i]->accumulate(W,W.begin(),W.end(),norm);
if(Collectables)//collectables are normalized by QMC drivers
Collectables->accumulate_all(W.Collectables,1.0);
}
void EstimatorManagerBase::accumulate(MCWalkerConfiguration& W
, MCWalkerConfiguration::iterator it
, MCWalkerConfiguration::iterator it_end)
{ // intended to be called slaveEstimator only
BlockWeight += it_end-it;
RealType norm=1.0/W.getGlobalNumWalkers();
for(int i=0; i< Estimators.size(); i++)
Estimators[i]->accumulate(W,it,it_end,norm);
if(Collectables)
Collectables->accumulate_all(W.Collectables,1.0);
}
// In LocalEnergyEstimator.h
inline void accumulate(const Walker_t& awalker, RealType wgt)
{ // ensemble average W.Properties
// expect ePtr to be W.Properties; expect wgt = 1/GlobalNumberOfWalkers
const RealType* restrict ePtr = awalker.getPropertyBase();
RealType wwght= wgt* awalker.Weight;
scalars[0](ePtr[WP::LOCALENERGY],wwght);
scalars[1](ePtr[WP::LOCALENERGY]*ePtr[WP::LOCALENERGY],wwght);
scalars[2](ePtr[LOCALPOTENTIAL],wwght);
for(int target=3, source=FirstHamiltonian; target<scalars.size(); ++target, ++source)
scalars[target](ePtr[source],wwght);
}
// In CollectablesEstimator.h
inline void accumulate_all(const MCWalkerConfiguration::Buffer_t& data, RealType wgt)
{ // ensemble average W.Collectables
// expect data to be W.Collectables; expect wgt = 1.0
for(int i=0; i<data.size(); ++i)
scalars[i](data[i], wgt);
}
At the unload ensemble stage, the data structures Properties and Collectables must be populated by appropriately normalized values so that the ensemble average can be correctly taken. QMCDriver is responsible for the correct loading of data onto the walker ensemble.
Load ensemble stage¶
Properties in the MC ensemble of walkers QMCDriver::W is
populated by QMCHamiltonian::saveProperties. The master QMCHamiltonian::LocalEnergy,
::KineticEnergy, and ::Observables must be properly populated
at the end of the evaluate stage.// In QMCHamiltonian.h
template<class IT>
inline
void saveProperty(IT first)
{ // expect first to be W.Properties
first[LOCALPOTENTIAL]= LocalEnergy-KineticEnergy;
copy(Observables.begin(),Observables.end(),first+myIndex);
}
Collectables’s load stage is combined with its evaluate stage.
Evaluate stage¶
QMCHamiltonian::Observables is populated by slave
OperatorBase ::setObservables. However, the call signature
must be OperatorBase::setObservables (QMCHamiltonian::Observables). This call signature is enforced by
QMCHamiltonian::evaluate and QMCHamiltonian::auxHevaluate.// In QMCHamiltonian.cpp
QMCHamiltonian::Return_t
QMCHamiltonian::evaluate(ParticleSet& P)
{
LocalEnergy = 0.0;
for(int i=0; i<H.size(); ++i)
{
myTimers[i]->start();
LocalEnergy += H[i]->evaluate(P);
H[i]->setObservables(Observables);
#if !defined(REMOVE_TRACEMANAGER)
H[i]->collect_scalar_traces();
#endif
myTimers[i]->stop();
H[i]->setParticlePropertyList(P.PropertyList,myIndex);
}
KineticEnergy=H[0]->Value;
P.PropertyList[WP::LOCALENERGY]=LocalEnergy;
P.PropertyList[LOCALPOTENTIAL]=LocalEnergy-KineticEnergy;
// auxHevaluate(P);
return LocalEnergy;
}
// In QMCHamiltonian.cpp
void QMCHamiltonian::auxHevaluate(ParticleSet& P, Walker_t& ThisWalker)
{
#if !defined(REMOVE_TRACEMANAGER)
collect_walker_traces(ThisWalker,P.current_step);
#endif
for(int i=0; i<auxH.size(); ++i)
{
auxH[i]->setHistories(ThisWalker);
RealType sink = auxH[i]->evaluate(P);
auxH[i]->setObservables(Observables);
#if !defined(REMOVE_TRACEMANAGER)
auxH[i]->collect_scalar_traces();
#endif
auxH[i]->setParticlePropertyList(P.PropertyList,myIndex);
}
}
Estimator use cases¶
VMCSingleOMP pseudo code¶
bool VMCSingleOMP::run()
{
masterEstimator->start(nBlocks);
for (int ip=0; ip<NumThreads; ++ip)
Movers[ip]->startRun(nBlocks,false); // slaveEstimator->start(blocks, record)
do // block
{
#pragma omp parallel
{
Movers[ip]->startBlock(nSteps); // slaveEstimator->startBlock(steps)
RealType cnorm = 1.0/static_cast<RealType>(wPerNode[ip+1]-wPerNode[ip]);
do // step
{
wClones[ip]->resetCollectables();
Movers[ip]->advanceWalkers(wit, wit_end, recompute);
wClones[ip]->Collectables *= cnorm;
Movers[ip]->accumulate(wit, wit_end);
} // end step
Movers[ip]->stopBlock(false); // slaveEstimator->stopBlock(acc, false)
} // end omp
masterEstimator->stopBlock(estimatorClones); // write files
} // end block
masterEstimator->stop(estimatorClones);
}
DMCOMP pseudo code¶
bool DMCOMP::run()
{
masterEstimator->setCollectionMode(true);
masterEstimator->start(nBlocks);
for(int ip=0; ip<NumThreads; ip++)
Movers[ip]->startRun(nBlocks,false); // slaveEstimator->start(blocks, record)
do // block
{
masterEstimator->startBlock(nSteps);
for(int ip=0; ip<NumThreads; ip++)
Movers[ip]->startBlock(nSteps); // slaveEstimator->startBlock(steps)
do // step
{
#pragma omp parallel
{
wClones[ip]->resetCollectables();
// advanceWalkers
} // end omp
//branchEngine->branch
{ // In WalkerControlMPI.cpp::branch
wgt_inv=WalkerController->NumContexts/WalkerController->EnsembleProperty.Weight;
walkers.Collectables *= wgt_inv;
slaveEstimator->accumulate(walkers);
}
masterEstimator->stopBlock(acc) // write files
} // end for step
} // end for block
masterEstimator->stop();
}
Summary¶
Two ensemble-level data structures, ParticleSet::Properties and
::Collectables, serve as intermediaries between evaluate classes and
output classes to scalar.dat and stat.h5. Properties appears
in both scalar.dat and stat.h5, whereas Collectables appears
only in stat.h5. Properties is overwritten by
QMCHamiltonian::Observables at the end of each step.
QMCHamiltonian::Observables is filled upon call to
QMCHamiltonian::evaluate and ::auxHevaluate. Collectables is
zeroed at the beginning of each step and accumulated upon call to
::auxHevaluate.
scalar.dat in four stages: evaluate, load,
unload, and collect. In the evaluate stage,
QMCHamiltonian::Observables is populated by a list of
OperatorBase. In the load stage, QMCHamiltonian::Observables
is transfered to Properties by QMCDriver. In the unload stage,
Properties is copied to LocalEnergyEstimator::scalars. In the
collect stage, LocalEnergyEstimator::scalars is block-averaged to
EstimatorManagerBase::AverageCache and dumped to file. For Collectables, the
evaluate and load stages are combined in a call to
QMCHamiltonian::auxHevaluate. In the unload stage,
Collectables is copied to CollectablesEstimator::scalars. In
the collect stage, CollectablesEstimator::scalars is block-averaged to
EstimatorManagerBase::AverageCache and dumped to file.Appendix: dmc.dat¶
ParticleSet::EnsembleProperty, that is managed by
WalkerControlBase::EnsembleProperty and directly dumped to
dmc.dat via its own averaging procedure. dmc.dat is written by
WalkerControlBase::measureProperties, which is called by
WalkerControlBase::branch, which is called by
SimpleFixedNodeBranch::branch, for example.Slater-backflow wavefunction implementation details¶
For simplicity, consider \(N\) identical fermions of the same spin (e.g., up electrons) at spatial locations \(\{\mathbf{r}_1,\mathbf{r}_2,\dots,\mathbf{r}_{N}\}\). Then the Slater determinant can be written as
where each entry in the determinant is an SPO evaluated at a particle position
When backflow transformation is applied to the determinant, the particle coordinates \(\mathbf{r}_i\) that go into the SPOs are replaced by quasi-particle coordinates \(\mathbf{x}_i\):
where
\(r_{ij}=\vert\mathbf{r}_i-\mathbf{r}_j\vert\). The integers i,j label the particle/quasi-particle. There is a one-to-one correspondence between the particles and the quasi-particles, which is simplest when \(\eta=0\).
Value¶
The evaluation of the Slater-backflow wavefunction is almost identical
to that of a Slater wavefunction. The only difference is that the
quasi-particle coordinates are used to evaluate the SPOs. The actual
value of the determinant is stored during the inversion of the matrix
\(M\) (cgetrf\(\rightarrow\)cgetri). Suppose
\(M=LU\), then \(S=\prod\limits_{i=1}^N L_{ii} U_{ii}\).
// In DiracDeterminantWithBackflow::evaluateLog(P,G,L)
Phi->evaluate(BFTrans->QP, FirstIndex, LastIndex, psiM,dpsiM,grad_grad_psiM);
psiMinv = psiM;
LogValue=InvertWithLog(psiMinv.data(),NumPtcls,NumOrbitals
,WorkSpace.data(),Pivot.data(),PhaseValue);
QMCPACK represents the complex value of the wavefunction in polar
coordinates \(S=e^Ue^{i\theta}\). Specifically, LogValue
\(U\) and PhaseValue \(\theta\) are handled separately. In
the following, we will consider derivatives of the log value only.
Gradient¶
To evaluate particle gradient of the log value of the Slater-backflow wavefunction, we can use the \(\log\det\) identity in (255). This identity maps the derivative of \(\log\det M\) with respect to a real variable \(p\) to a trace over \(M^{-1}dM\):
Following Kwon, Ceperley, and Martin [KCM93], the particle gradient
where the quasi-particle gradient matrix
and the intermediate matrix
with the SPO derivatives (w.r. to quasi-particle coordinates)
Notice that we have made the name change of \(\phi\rightarrow M\)
from the notations of ref. [KCM93]. This
name change is intended to help the reader associate M with the QMCPACK
variable psiM.
// In DiracDeterminantWithBackflow::evaluateLog(P,G,L)
for(int i=0; i<num; i++) // k in above formula
{
for(int j=0; j<NumPtcls; j++)
{
for(int k=0; k<OHMMS_DIM; k++) // alpha in above formula
{
myG(i) += dot(BFTrans->Amat(i,FirstIndex+j),Fmat(j,j));
}
}
}
(256) is still relatively simple to understand. The \(A\) matrix maps changes in particle coordinates \(d\mathbf{r}\) to changes in quasi-particle coordinates \(d\mathbf{x}\). Dotting A into F propagates \(d\mathbf{x}\) to \(dM\). Thus \(F\cdot A\) is the term inside the trace operator of (255). Finally, performing the trace completes the evaluation of the derivative.
Laplacian¶
The particle Laplacian is given in [KCM93] as
where the quasi-particle Laplacian matrix
with the second derivatives of the single-particles orbitals being
Schematically, \(L_i\) has contributions from three terms of the form \(BF, AAFF, and tr(AA,Md2M)\), respectively. \(A, B, M ,d2M,\) and \(F\) can be calculated and stored before the calculations of \(L_i\). The first \(BF\) term can be directly calculated in a loop over quasi-particle coordinates \(j\alpha\).
// In DiracDeterminantWithBackflow::evaluateLog(P,G,L)
for(int j=0; j<NumPtcls; j++)
for(int a=0; a<OHMMS_DIM; k++)
myL(i) += BFTrans->Bmat_full(i,FirstIndex+j)[a]*Fmat(j,j)[a];
Notice that \(B_{ij}^\alpha\) is stored in Bmat_full, NOT
Bmat.
The remaining two terms both involve \(AA\). Thus, it is best to define a temporary tensor \(AA\):
which we will overwrite for each particle \(i\). Similarly, define \(FF\):
which is simply the outer product of \(F\otimes F\). Then the \(AAFF\) term can be calculated by fully contracting \(AA\) with \(FF\).
// In DiracDeterminantWithBackflow::evaluateLog(P,G,L)
for(int j=0; j<NumPtcls; j++)
for(int k=0; k<NumPtcls; k++)
for(int i=0; i<num; i++)
{
Tensor<RealType,OHMMS_DIM> AA = dot(transpose(BFTrans->Amat(i,FirstIndex+j)),BFTrans->Amat(i,FirstIndex+k));
HessType FF = outerProduct(Fmat(k,j),Fmat(j,k));
myL(i) -= traceAtB(AA,FF);
}
Finally, define the SPO derivative term:
then the last term is given by the contraction of \(Md2M\) (q_j)
with the diagonal of \(AA\).
for(int j=0; j<NumPtcls; j++)
{
HessType q_j;
q_j=0.0;
for(int k=0; k<NumPtcls; k++)
q_j += psiMinv(j,k)*grad_grad_psiM(j,k);
for(int i=0; i<num; i++)
{
Tensor<RealType,OHMMS_DIM> AA = dot(
transpose(BFTrans->Amat(i,FirstIndex+j)),
BFTrans->Amat(i,FirstIndex+j)
);
myL(i) += traceAtB(AA,q_j);
}
}
Wavefunction parameter derivative¶
To use the robust linear optimization method of [TU07], the trial wavefunction needs to know its contributions to the overlap and hamiltonian matrices. In particular, we need derivatives of these matrices with respect to wavefunction parameters. As a consequence, the wavefunction \(\psi\) needs to be able to evaluate \(\frac{\partial}{\partial p} \ln \psi\) and \(\frac{\partial}{\partial p} \frac{\mathcal{H}\psi}{\psi}\), where \(p\) is a parameter.
When 2-body backflow is considered, a wavefunction parameter \(p\) enters the \(\eta\) function only (equation (254)). \(\mathbf{r}\), \(\phi\), and \(M\) do not explicitly dependent on \(p\). Derivative of the log value is almost identical to particle gradient. Namely, (256) applies upon the substitution \(r_i^\alpha\rightarrow p\).
where the quasi-particle derivatives are stored in Cmat
The change in local kinetic energy is a lot more difficult to calculate
where \(L_i\) is the particle Laplacian defined in (260) To evaluate (268), we need to calculate parameter derivatives of all three terms defined in the Laplacian evaluation. Namely \((B)(F)\), \((AA)(FF)\), and \(\text{tr}(AA,Md2M)\), where we have put parentheses around previously identified data structures. After \(\frac{\partial}{\partial p}\) hits, each of the three terms will split into two terms by the product rule. Each smaller term will contain a contraction of two data structures. Therefore, we will need to calculate the parameter derivatives of each data structure defined in the Laplacian evaluation:
X and Y are stored as Xmat and Ymat_full (NOT Ymat) in the
code. dF is dFa. \(AA'\) is not fully stored; intermediate
values are stored in Aij_sum and a_j_sum. \(FF'\) is
calculated on the fly as \(dF\otimes F+F\otimes dF\). \(Md2M'\)
is not stored; intermediate values are stored in q_j_prime.
Particles and distance tables¶
ParticleSets¶
The ParticleSet class stores particle positions and attributes
(charge, mass, etc).
The R member stores positions. For calculations, the R variable
needs to be transferred to the structure-of-arrays (SoA) storage in
RSoA. This is done by the update method. In the future the
interface may change to use functions to set and retrieve positions so
the SoA transformation of the particle data can happen automatically.
A particular distance table is retrieved with getDistTable. Use
addTable to add a ParticleSet and return the index of the
distance table. If the table already exists the index of the existing
table will be returned.
The mass and charge of each particle is stored in Mass and Z.
The flag, SameMass, indicates if all the particles have the same
mass (true for electrons).
Groups¶
Particles can belong to different groups. For electrons, the groups are
up and down spins. For ions, the groups are the atomic element. The
group type for each particle can be accessed through the GroupID
member. The number of groups is returned from groups(). The total
number particles is accessed with getTotalNum(). The number of
particles in a group is groupsize(int igroup).
The particle indices for each group are found with first(int igroup)
and last(int igroup). These functions only work correctly if the
particles are packed according to group. The flag, IsGrouped,
indicates if the particles are grouped or not. The particles will not be
grouped if the elements are not grouped together in the input file. This
ordering is usually the responsibility of the converters.
Code can be written to only handle the grouped case, but put an assert or failure check if the particles are not grouped. Otherwise the code will give wrong answers and it can be time-consuming to debug.
Distance tables¶
Distance tables store distances between particles. There are symmetric (AA) tables for distance between like particles (electron-electron or ion-ion) and asymmetric (BA) tables for distance between unlike particles (electron-ion)
The Distances and Displacements members contain the data. The
indexing order is target index first, then source. For electron-ion
tables, the sources are the ions and the targets are the electrons.
Looping over particles¶
Some sample code on how to loop over all the particles in an electron-ion distance table:
// d_table is an electron-ion distance table
for (int jat = 0; j < d_table.targets(); jat++) { // Loop over electrons
for (int iat = 0; i < d_table.sources(); iat++) { // Loop over ions
d_table.Distances[jat][iat];
}
}
Interactions sometimes depend on the type of group of the particles. The
code can loop over all particles and use GroupID[idx] to choose the
interaction. Alternately, the code can loop over the number of groups
and then loop from the first to last index for those groups. This method
can attain higher performance by effectively hoisting tests for group ID
out of the loop.
An example of the first approach is
// P is a ParticleSet
for (int iat = 0; iat < P.getTotalNum(); iat++) {
int group_idx = P.GroupID[iat];
// Code that depends on the group index
}
An example of the second approach is
// P is a ParticleSet
assert(P.IsGrouped == true); // ensure particles are grouped
for (int ig = 0; ig < P.groups(); ig++) { // loop over groups
for (int iat = P.first(ig); iat < P.last(ig); iat++) { // loop over elements in each group
// Code that depends on group
}
}
Adding a wavefunction¶
The total wavefunction is stored in TrialWaveFunction as a product
of all the components. Each component derives from
WaveFunctionComponent. The code contains an example of a
wavefunction component for a Helium atom using a simple form and is
described in Helium Wavefunction Example
Mathematical preliminaries¶
The wavefunction evaluation functions compute the log of the wavefunction, the gradient and the Laplacian of the log of the wavefunction. Expanded, the gradient and Laplacian are
However, the local energy formula needs \(\frac{{\nabla ^2} \psi}{\psi}\).
The conversion from the Laplacian of the log of the wavefunction to the
local energy value is performed in
QMCHamiltonians/BareKineticEnergy.h (i.e. \(L + G \cdot G\).)
Wavefunction evaluation¶
The process for creating a new wavefunction component class is to derive from WaveFunctionComponent and implement a number pure virtual functions. To start most of them can be empty.
The following four functions evaluate the wavefunction values and spatial derivatives:
evaluateLog Computes the log of the wavefunction and the gradient
and Laplacian (of the log of the wavefunction) for all particles. The
input is theParticleSet(P) (of the electrons). The log of
the wavefunction should be stored in the LogValue member variable,
and used as the return value from the function. The gradient is stored
in G and the Laplacian in L.
ratio Computes the wavefunction ratio (not the log) for a single
particle move (\(\psi_{new}/\psi_{old}\)). The inputs are the
ParticleSet(P) and the particle index (iat).
evalGrad Computes the gradient for a given particle. The inputs are
the ParticleSet(P) and the particle index (iat).
ratioGrad Computes the wavefunction ratio and the gradient at the
new position for a single particle move. The inputs are the
ParticleSet(P) and the particle index (iat). The output
gradient is in grad_iat;
The updateBuffer function needs to be implemented, but to start it
can simply call evaluateLog.
The put function should be implemented to read parameter specifics
from the input XML file.
Function use¶
For debugging it can be helpful to know the under what conditions the various routines are called.
The VMC and DMC loops initialize the walkers by calling evaluateLog.
For all-electron moves, each timestep advance calls evaluateLog. If
the use_drift parameter is no, then only the wavefunction value is
used for sampling. The gradient and Laplacian are used for computing the
local energy.
For particle-by-particle moves, each timestep advance
calls
evalGradcomputes a trial move
calls
ratioGradfor the wavefunction ratio and the gradient at the trial position. (If theuse_driftparameter is no, theratiofunction is called instead.)
The following example shows part of an input block for VMC with all-electron moves and drift.
<qmc method="vmc" target="e" move="alle">
<parameter name="use_drift">yes</parameter>
</qmc>
Particle distances¶
The ParticleSet parameter in these functions refers to the
electrons. The distance tables that store the inter-particle distances
are stored as an array.
To get the electron-ion distances, add the ion ParticleSet using
addTable and save the returned index. Use that index to get the
ion-electron distance table.
const int ei_id = elecs.addTable(ions); // in the constructor only
const auto& ei_table = elecs.getDistTable(ei_id); // when consuming a distance table
Getting the electron-electron distances is very similar, just add the
electron ParticleSet using addTable.
Only the lower triangle for the electron-electron table should be used. It is the only part of the distance table valid throughout the run. During particle-by-particle move, there are extra restrictions. When a move of electron iel is proposed, only the lower triangle parts [0,iel)[0,iel) [iel, Nelec)[iel, Nelec) and the row [iel][0:Nelec) are valid. In fact, the current implementation of distance based two and three body Jastrow factors in QMCPACK only needs the row [iel][0:Nelec).
In ratioGrad, the new distances are stored in the Temp_r and
Temp_dr members of the distance tables.
Setup¶
A builder processes XML input, creates the wavefunction, and adds it to
targetPsi. Builders derive from WaveFunctionComponentBuilder.
The new builder hooks into the XML processing in
WaveFunctionFactory.cpp in the build function.
Caching values¶
The acceptMove and restore methods are called on accepted and
rejected moves for the component to update cached values.
Threading¶
The makeClone function needs to be implemented to work correctly
with OpenMP threading. There will be one copy of the component created
for each thread. If there is no extra storage, calling the copy
constructor will be sufficient. If there are cached values, the clone
call may need to create space.
Parameter optimization¶
The checkInVariables, checkOutVariables, and resetParameters
functions manage the variational parameters. Optimizable variables also
need to be registered when the XML is processed.
Variational parameter derivatives are computed in the
evaluateDerivatives function. The first output value is an array
with parameter derivatives of log of the wavefunction. The second output
values is an array with parameter derivatives of the Laplacian divided
by the wavefunction (and not the Laplacian of the log of the
wavefunction) The kinetic energy term contains a \(-1/2m\) factor.
The \(1/m\) factor is applied in TrialWaveFunction.cpp, but the
\(-1/2\) is not and must be accounted for in this function.
Helium Wavefunction Example¶
The code contains an example of a wavefunction component for a Helium atom using STO orbitals and a Pade Jastrow.
to The wavefunction is
where \(Z = 2\) is the nuclear charge, \(A=1/2\) is the
electron-electron cusp, and \(B\) is a variational parameter. The
electron-ion distances are \(r_1\) and \(r_2\), and
\(r_{12}\) is the electron-electron distance. The wavefunction is
the same as the one expressed with built-in components in
examples/molecules/He/he_simple_opt.xml.
The code is in src/QMCWaveFunctions/ExampleHeComponent.cpp. The
builder is in src/QMCWaveFunctions/ExampleHeBuilder.cpp. The input
file is in examples/molecules/He/he_example_wf.xml. A unit test
compares results from the wavefunction evaluation functions for
consistency in src/QMCWaveFunctions/tests/test_example_he.cpp.
The recommended approach for creating a new wavefunction component is to copy the example and the unit test. Implement the evaluation functions and ensure the unit test passes.
Linear Algebra¶
Like in many methods which solve the Schrödinger equation, linear algebra plays a critical role in QMC algorithms and thus is crucial to the performance of QMCPACK. There are a few components in QMCPACK use BLAS/LAPACK with their own characteristics.
Real space QMC¶
Single particle orbitals¶
Spline evaluation as commonly used in solid-state simulations does not use any dense linear algebra library calls. LCAO evaluation as commonly used in molecular calculations relies on BLAS2 GEMV to compute SPOs from a basis set.
Slater determinants¶
Slater determinants are calculated on \(N \times N\) Slater matrices. \(N\) is the number of electrons for a given spin. In the actually implementation, operations on the inverse matrix of Slater matrix for each walker dominate the computation. To initialize it, DGETRF and DGETRI from LAPACK are called. The inverse matrix can be stored out of place. During random walking, inverse matrices are updated by either Sherman-Morrison rank-1 update or delayed update. Update algorithms heavily relies on BLAS. All the BLAS operations require S,C,D,Z cases.
Sherman-Morrison rank-1 update uses BLAS2 GEMV and GER on \(N \times N\) matrices.
Delayed rank-K update uses
BLAS1 SCOPY on \(N\) array.
BLAS2 GEMV, GER on \(k \times N\) and \(k \times k\) matrices. \(k\) ranges from 1 to \(K\) when updates are delayed and accumulated.
BLAS3 GEMM at the final update.
’T’, ’N’, K, N, N
’N’, ’N’, N, K, K
’N’, ’N’, N, N, K
The optimal K depends on the hardware but it usually ranges from 32 to 256.
QMCPACK solves systems with a few to thousands of electrons. To make all the BLAS/LAPACK operation efficient on accelerators. Batching is needed and optimized for \(N < 2000\). Non-batched functions needs to be optimized for \(N > 500\). Note: 2000 and 500 are only rough estimates.
Wavefunction optimizer¶
to be added.
Auxiliary field QMC¶
The AFQMC implementation in QMCPACK relies heavily on linear algebra operations from BLAS/LAPACK. The performance of the code is netirely dependent on the performance of these libraries. See below for a detailed list of the main routines used from BLAS/LAPACK. Since the AFQMC code can work with both single and double precision builds, all 4 versions of these routines (S,C,D,Z) are generally needed, for this reason we omit the data type label.
BLAS1: SCAL, COPY, DOT, AXPY
BLAS2: GEMV, GER
BLAS3: GEMM
LAPACK: GETRF, GETRI, GELQF, UNGLQ, ORGLQ, GESVD, HEEVR, HEGVX
While the dimensions of the matrix operations will depend entirely on the details of the calculation, typical matrix dimensions range from the 100s, for small system sizes, to over 20000 for the largest calculations attempted so far. For builds with GPU accelerators, we make use of batched and strided implementations of these routines. Batched implementations of GEMM, GETRF, GETRI, GELQF and UNGLQ are particularly important for the performance of the GPU build on small to medium size problems. Batched implementations of DOT, AXPY and GEMV would also be quite useful, but they are not yet generally available. On GPU builds, the code uses batched implementations of these routines when available by default.
- KCM93(1,2,3)
Yongkyung Kwon, D. M. Ceperley, and Richard M. Martin. Effects of three-body and backflow correlations in the two-dimensional electron gas. Phys. Rev. B, 48:12037–12046, October 1993. doi:10.1103/PhysRevB.48.12037.
- TU07
Julien Toulouse and C. J. Umrigar. Optimization of quantum monte carlo wave functions by energy minimization. The Journal of Chemical Physics, 126(8):084102, 2007. arXiv:http://dx.doi.org/10.1063/1.2437215, doi:10.1063/1.2437215.
Appendices¶
Appendix A: Derivation of twist averaging efficiency¶
In this appendix we derive the relative statistical efficiency of twist averaging with an irreducible (weighted) set of k-points versus using uniform weights over an unreduced set of k-points (e.g., a full Monkhorst-Pack mesh).
Consider the weighted average of a set of statistical variables \(\{x_m\}\) with weights \(\{w_m\}\):
If produced by a finite QMC run at a set of twist angles/k-points \(\{k_m\}\), each variable mean \(\langle{x_m}\rangle\) has a statistical error bar \(\sigma_m\), and we can also obtain the statistical error bar of the mean of the twist-averaged quantity \(\langle{x_{TA}\rangle}\):
The error bar of each individual twist \(\sigma_m\) is related to the autocorrelation time \(\kappa_m\), intrinsic variance \(v_m\), and the number of postequilibration MC steps \(N_{step}\) in the following way:
In the setting of twist averaging, the autocorrelation time and variance for different twist angles are often very similar across twists, and we have
If we define the total weight as \(W\), that is, \(W\equiv\sum_{m=1}^Mw_m\), for the weighted case with \(M\) irreducible twists, the error bar is
For uniform weighting with \(w_m=1\), the number of twists is \(W\) and we have
We are interested in comparing the efficiency of choosing weights uniformly or based on the irreducible multiplicity of each twist angle for a given target error bar \(\sigma_{target}\). The number of MC steps required to reach this target for uniform weighting is
while for nonuniform weighting we have
The MC efficiency is defined as
where \(\sigma\) is the error bar and \(t\) is the total CPU time required for the MC run.
The main advantage made possible by irreducible twist weighting is to reduce the equilibration time overhead by having fewer twists and, hence, fewer MC runs to equilibrate. In the context of twist averaging, the total CPU time for a run can be considered to be
where \(N_{twist}\) is the number of twists, \(N_{eq}\) is the number of MC steps required to reach equilibrium, \(N_{step}\) is the number of MC steps included in the statistical averaging as before, and \(t_{step}\) is the wall clock time required to complete a single MC step. For uniform weighting \(N_{twist}=W\); while for irreducible weighting \(N_{twist}=M\).
We can now calculate the relative efficiency (\(\eta\)) of irreducible vs. uniform twist weighting with the aim of obtaining a target error bar \(\sigma_{target}\):
In this last expression, \(f\) is the ratio of the number of usable MC steps to the number that must be discarded during equilibration (\(f=N_{step}^{uniform}/N_{eq}\)); and as before, \(W=\sum_mw_m\), which is the number of twist angles in the uniform weighting case. It is important to recall that \(N_{step}^{uniform}\) in \(f\) is defined relative to uniform weighting and is the number of MC steps required to reach a target accuracy in the case of uniform twist weights.
The formula for \(\eta\) in the preceding can be easily changed with the help of (279) to reflect the number of MC steps obtained in an irreducibly weighted run instead. A good exercise is to consider runs that have already completed with either uniform or irreducible weighting and calculate the expected efficiency change had the opposite type of weighting been used.
The break even point \((\eta=1)\) can be found at a usable step fraction of
The relative efficiency \((\eta)\) is useful to consider in view of certain scenarios. An important case is where the number of required sampling steps is no larger than the number of equilibration steps (i.e., \(f\approx 1\)). For a very simple case with eight uniform twists with irreducible multiplicities of \(w_m\in\{1,3,3,1\}\) (\(W=8\), \(M=4\)), the relative efficiency of irreducible vs. uniform weighting is \(\eta=\frac{8}{4}\frac{2}{1+20/8}\approx 1.14\). In this case, irreducible weighting is about \(14\)% more efficient than uniform weighting.
Another interesting case is one in which the number of sampling steps you can reach with uniform twists before wall clock time runs out is small relative to the number of equilibration steps (\(f\rightarrow 0\)). In this limit, \(\eta\approx W/M\). For our eight-uniform-twist example, this would result in a relative efficiency of \(\eta=8/4=2\), making irreducible weighting twice as efficient.
A final case of interest is one in which the equilibration time is short relative to the available sampling time \((f\rightarrow\infty)\), giving \(\eta\approx W^2/(M\sum_{m=1}^Mw_m^2)\). Again, for our simple example we find \(\eta=8^2/(4\times 20)\approx 0.8\), with uniform weighting being \(25\)% more efficient than irreducible weighting. For this example, the crossover point for irreducible weighting being more efficient than uniform weighting is \(f<2\), that is, when the available sampling period is less than twice the length of the equilibration period. The expected efficiency ratio and crossover point should be checked for the particular case under consideration to inform the choice between twist averaging methods.
QMCPACK website: http://www.qmcpack.org
Releases & source code: https://github.com/QMCPACK
Google Group: https://groups.google.com/forum/#!forum/qmcpack