Releases

Download QMCPACK v3.6.0

Notes

This release includes a completely new AFQMC implementation, significant performance improvements for large runs, greater functionality in the structure-of-arrays (SoA) code path, support for larger spline data on multiple GPUs, and support for new machines and compilers. The manual has been improved, bugs have been fixed, and source code cleanup continued.

A C++14 and C99 capable compiler, Boost 1.61.0, and CMake 3.6 or greater are now required.

• Completely updated AFQMC implementation including reduced scaling separable density fitting https://arxiv.org/abs/1810.00284 Documentation and examples will be added in v3.7.0. Contact the developers for use instructions in the interim. #1245

• Implementation of delayed updates for CPU. Substantial speedups for runs with 100s of electrons, with increasing gains at larger electron counts. See manual for details. #1170

• Initial support for nested OpenMP to further reduce time-to-solution for large problems. #1082

• Support for splitting/distributing spline orbital data across multiple GPUs on a single node. #1101

• Cusp correction for all electron calculations is implemented in the SoA version. #1172

• Backflow is implemented in the SoA version. #1225

• K-points with real coefficients are supported in periodic LCAO. #1006

• Initial support for Summit at OLCF. Revisions may be needed in January 2019 as the software stack is updated. This will be addressed in a new version as required.

• Initial support for PGI compiler.

• Build instructions for ARM-based systems. #1148

• Setup scripts are python 2 and 3 compatible. #1261

• QMCPACK and NEXUS can now be installed by "make install" after configuring CMake with CMAKE_PREFIX_PATH. #1020

• Significantly reworked test labeling and categorization system. #1155

• Partial transition to a new MPI wrapper implementation for greater compatibility.

• Utilities have been renamed for clarity and to avoid name collisions with other applications. getSupercell is renamed qmc-get-supercell. extract-eshdf-kvectors is renamed qmc-extract-eshdf-kvectors.

Known bugs

Several potentially significant bugs are outstanding and will be addressed in the next release. See list of open bugs.

• LCAO (Gaussian basis) molecular calculations are incorrect with certain diffusion functions. The reason for this bug is currently unclear. #1145

• On NVIDIA Volta GPUs some runs show inconsistencies with the CPU version. Standard carbon diamond and LiH tests pass with good agreement with the CPU implementation. #1054

• QMCPACK will not build with OpenMPI v4.0.0 due to use of deprecated functions. This will be addressed in the next version as the new MPI wrappers are fully adopted. Older OpenMPI libraries are fully capable.

NEXUS

• Interface to and support for PySCF. #1220

• Interface to and support for Quantum Package (QP). #1093

• Support for excited state calculations. #1200

• qfit is renamed qmc-fit.

• ntest, sim, redo are renamed nxs-test, nxs-sim, nxs-redo.

• Many smaller improvements.

Download QMCPACK v3.5.0

Notes

This release includes support for the latest Quantum Espresso version 6.3, an initial implementation of periodic Gaussian support via PySCF, and a new version of the hybrid or "APW" representation of orbitals. Many minor bugs have been fixed, configuration and documentation improved. It is a recommended update for all users.

Note that the PDF manuals are no longer included with the source. Versions are available online via https://qmcpack.org . The PDFs can be built using manual/build_manual.sh and nexus/documentation/user_guide_source/build_nexus_user_guide.sh

Attention developers: This version contains substantially fewer source lines than previous versions due to clean out of old code and unused execution paths. Refactoring to improve the internal structure of QMCPACK is ongoing. Track the develop branch and follow discussion on GitHub closely to avoid difficult merges.

• Support for Quantum Espresso 6.3 and 6.2.1. Check documentation to ensure compiled with required HDF5 support.
• Support for periodic gaussians and PySCF generated wavefunctions. Initial version is limited to Gamma-point.
• Improved hybrid representation of single particle orbitals (APW-like) for significantly reduced memory usage and possible accuracy increase compared to conventional spline representation. https://arxiv.org/abs/1805.07406
• Norms of orbitals are checked inside QMCPACK to catch conversion errors.
• Added verbosity setting to QMCPACK output.
• CUDA can now be enabled with SoA builds.
• Many improvements to QMCPACK manual, including all new features, CIPSI, 3-body jastrow factor description, spack package, and enabling HTML generation.
• CMake configuration improvements, particularly around MKL handling.
• Extensive cleanup of unused source files and unused code paths removed, reducing the number of source lines by over 30 percent.

Known bugs

• Weight of first block of DMC density is incorrect in CPU code. DMC densities in CUDA GPU code are incorrect for all blocks. #934 and #925
• Runs with only a single electron may crash. #945

NEXUS

• Support for GAMESS HDF5 workflows.
• Nexus accepts command line inputs.
• Nexus testing via ntest executable.
• Added GAMESS-NEXUS examples for RHF, CISD, and CASSCF wavefunction.
• Added support for -nojastrow workflows.
• Added support for Stampede supercomputer.
• Added script to build NEXUS user guide.
• Various bugfixes including to GAMESS input parsing.

Download QMCPACK v3.4.0

Notes

This release includes size-consistent t-moves, and improvements to load balancing and memory usage that will be visible in large runs. Significant revisions have been made to the gaussian wavefunction reader and a PySCF interface is in progress. A bug affecting non-git installs (from release tarballs) is fixed. Feedback is particularly welcome on the new features.

• Size consistent t-moves implemented (Casula 2010 algorithm). Enabled via nonlocalmoves parameter, see manual.
• Bugfix: For non-git builds, build process failed on some systems due to git-rev.h handling.
• Optimized load balancing in DMC. Command line option async_swap removed. Parameter use_nonblocking now disables non-blocking MPI load balancing. Non-blocking MPI is now enabled by default.
• Improved memory handling and usage in SoA code, increases performance.
• Improved stability of GPU matrix inversion for large runs.
• Ongoing improvements to output to improve readability.
• Initial interface to PySCF for real space QMC trial wavefunctions.
• Enabled use of HDF5 files for Gaussian based wavefunctions with SoA implementation.
• Added Appendix to manual listing all known QMCPACK publications. This will be updated on an ongoing basis. Please advise of any missing publications.
• Optimized momentum distribution estimator. Supported by SoA and 1,2,3-body Jastrow functions.
• Support for labeled timers in Intel VTune based profiling.

NEXUS

• Minor bugfixes and improvements.

Known limitations

• PySCF interface is preliminary. convert4qmc is updated, but manual entries are not yet provided. This will be improved in later versions. The interface is currently only for isolated molecular systems. A full periodic implementation is in progress.

• Documentation, examples and tutorials are not yet consistent with the updated converter convert4qmc.

Download QMCPACK v3.3.0

Notes

This release includes new methods, converter updates, and many optimizations, feature improvements, and bug fixes. It is a recommended update for all users.

QMCPACK updates

• Support for finite difference linear response (FDLR) method and wavefunctions, developed and contributed by Nick Blunt and Eric Neuscamman, see Journal of Chemical Physics 147, 194101 (2017), https://doi.org/10.1063/1.4998197 and https://arxiv.org/abs/1707.09439 .
• Major update to convert4qmc, conversion from GAMESS and other gaussian basis set codes. HDF5 output is now supported for large wavefunctions with -hdf5 option. Significantly improved example inputs *.qmc.in.xml.
• Gaussian based trial wavefunctions now supported by structure of arrays implementation (ENABLE_SOA=1). A full reimplementation that will also support gaussians in periodic boundary conditions, e.g. from pyscf, is in progress.
• Initialization of multideterminant wavefunctions improved for faster startup and lower memory usage. In practice this significantly raises the usable maximum number of determinants.
• Maximum CPU time setting (maxcpusecs): QMC drivers will not start a new block if there is not enough estimated time remaining to complete the next block and gracefully shut down.
• Homogeneous electron gas wavefunction support and tests.
• New command line verbosity command line flag -verbosity. Output of QMCPACK will be overhauled over the next few releases to support low, high, and debug options, and also to significantly improve readability and utility.
• Bugfix: Umrigar drift diffusion term is now consistent with the Umrigar small time step error algorithm with complex wavefunctions.
• Bugfix: Momentum distribution is now correctly weighted and also correctly signed for twist averaging.
• Renamed performance tests with atom and electron count.
• Removed support for "buffering" of non-local pseudopotential wavefunction components during optimization (useBuffer setting) to reduce memory usage and for simplicity.
• doxygen documentation for developer-level documentation of the code and file structure. Produced via make in qmcpack/doxygen. HTML currently published at http://docs.qmcpack.org/doxygen/doxy/
• Many minor bug fixes and improved tests.

NEXUS

• Improved postprocessing support for Quantum Espresso.
• Various minor bug fixes.

Known issues and limitations

• Documentation, examples and tutorials are not yet consistent with the updated converter convert4qmc.
• Core functionality is largely compatible with ENABLE_SOA but some specialized wavefunctions and observables are not.
• Use of GNU compilers with glibc 2.23 builds will crash due to a bug in libmvec of glibc. The glibc version can be verified by "ldd --version".

Notes

This release provides a significant speed increase for many calculations. A C++11 compiler is now required. It is a recommended update.

QMCPACK updates

• Major speedup for calculations using spline wavefunctions via initial implementation of “Structure of Arrays” data layout and improved algorithms. Enabled via -DENABLE_SOA=1. Benefits all CPU architectures. Many runs are doubled in speed. Not yet available for Gaussian-basis sets or for all observables and QMC methods. See writeup in manual for guidance.
• A compiler supporting C++11 is now required.
• DMC respects MaxCPUSecs parameter and will gracefully shut down and not start a new block if there is not sufficient estimated time to complete it.
• Checkpointing code rewritten for robustness and performance at scale. Parallel as well as serial HDF5 supported and autodetected.
• Improved beta-release of AFQMC code and documentation.
• Backflow documentation and optimization tips added.
• Correlated sampling VMC drivers reactivated.
• Added carbon graphite performance test similar to CORAL benchmark.
• Improvements to CMake and CTest usage.
• Build instructions for NERSC, ALCF, and OLCF machines updated.
• Latest manual PDF now available here

NEXUS

• Significantly improved manual entry for “qmca” analysis tool, the main recommended tool for statistical analysis of QMCPACK data.
• Added time step fitting tool “qfit” for timestep extrapolation. Uses jack-knife statistical technique.
• Improved density file postprocessing.
• Support for Makov-Payne corrections.