DMQMC-GPR

Code for applying Gaussian process regression to DMQMC data from HANDE-QMC.

Python packages

The initial implementation was developed with the following packages and corresponding versions. This is not intended to be comprehensive but instead is focused on those required by GPy as indicated by pip. Additionally, packages related to running the example and test scripts.

0. Python 3.9.12
1. GPy 1.10.0
2. paramz 0.9.5
3. numpy 1.21.5
4. scipy 1.7.3
5. six 1.16.0
6. Cython 0.29.28
7. matplotlib 3.5.1
8. pandas 1.4.1

Installation and running

For installation it is recommended to use anaconda or miniforge. The required libraries corresponding to the original implementation are listed within the conda_env.yml file. To make an install using anaconda we can run the following commands:

1. conda env create -f conda_env.yml
2. conda activate dmqmcgpr
3. export PYTHONPATH=$(pwd)/qmcgpr/:$PYTHONPATH

Note the final step is only required to import the library without appending the path (see qmcgpr/examples for more information).

After following the first two steps above using anaconda 24.1.2, we are able to run the simple test suite from the qmcgpr/tests folder using the command:

1. export OMP_NUM_THREADS=1
2. bash ./runtests.sh *simple.py

Note that a more extensive test suite can be run with the following command:

1. bash ./runtests.sh

However our testing has shown that the results for the tests (and the GPR code used within) are hardware, library, and thread count dependent. Additionally, there are likely several factors which are yet to be discovered that can impact the results.

As such, the simple tests are meant to serve as a way to check that the code is functioning as a whole. As well as demonstrate how one can take steps to ensure that the relevant data are reproduced at a late date. That is to say it is important to store the original data set used to apply the code, as well as the output for the code in the form of the parameter files etc.

For the full test suite, these were run and fully passed on an intel based 2016 MacBook Pro from which the data were originally produced.

The simple tests are able to reproduce the data relevant to figures within the manuscript (see link below) on a MacBook Pro M3 using a single OMP thread.

Citation

For referencing the original work this repository is based on use:

@article{vanbenschoten_electronic_2023,
    author = {Van Benschoten, William Z. and Weiler, Laura and Smith, Gabriel J. and Man, Songhang and DeMello, Taylor and Shepherd, James J.},
    title = {Electronic specific heat capacities and entropies from density matrix quantum Monte Carlo using Gaussian process regression to find gradients of noisy data},
    journal = {The Journal of Chemical Physics},
    volume = {158},
    number = {21},
    pages = {214115},
    year = {2023},
    month = {06},
    issn = {0021-9606},
    doi = {10.1063/5.0150702},
    url = {https://doi.org/10.1063/5.0150702},
}

Useful links

• The GPy github page:
  • https://github.com/SheffieldML/GPy
• The HANDE-QMC github page:
  • https://github.com/hande-qmc/hande
• The original manuscript this repository was generated from:
  • https://doi.org/10.1063/5.0150702

Funding

• Research was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences Early Career Research Program (ECRP) under Award Number DE-SC0021317.