Path-integral Monte Carlo estimator for the dipole polarizability of quantum plasma: data repository

Juha Tiihonen, David Trejo-Garcia, Tapio T. Rantala, and Marco Ornigotti
Faculty of Engineering and Natural Sciences, Tampere University, Tampere, Finland

For support and guidance beyond this overview, contact: [e-mail]

## Original work

The data repository contains numerical data, source code and modest instructions
for reproducing the original work, titled "Path-integral Monte Carlo estimator
for the dipole polarizability of quantum plasma" by the same authors.

## Software

### PIMC3

The raw data is based on path-integral Monte Carlo (PIMC) simulation as
implemented in the PIMC3 software suite found in 'pimc3/' directory. The 'pimc3'
program is written in Fortran and can be built and operated according to
instructions in 'pimc3/README.md'. The documentation may be partially outdated
and incomplete.

The PIMC3 suite also has extensive data management features written in Python,
found in 'pimc3/pyfiles/', which are used to preprocess the raw simulation data
into convenient arrays.

### Additional Python libraries and scripts

Additional data processing features particular for this project are written in
Python and found in 'lib/'. This includes data processing, analyses, reference
models. Furthermore, Python scripts for producing figures and tables are found
in 'analyze/'. The files are laid out to be called from the repository root,
'/', e.g.

> python3 analyze/plot_Giw_vs_theta.py

## Data repository

The data archive contains four types of data
1. Raw data
1. Preprocessed data
1. Postprocessed data
1. Reference data

The raw data is produced by the 'pimc3' software and stored in the Hierarchical
Data Format (HDF5). The raw PIMC observables feature a binning sequence, i.e., a
series of averages over a selected number of measurements. Note: the original
submission scripts, inputs and outputs are preserved as is but only one
trajectory file (for MPI core 0) is archived.

The preprocessed data is produced by the 'p3load' of the PIMC3 suite, which
calculates the statistical averages and uncertainties of the binned raw data and
stores it to disk in ASCII.

The postprocessed data is the preprocessed data after selected manipulations,
such as rescaling, to conform with the physical analysis. The postprocessing
only applies to the correlation functions and their Matsubara series.

The reference data is based on a reference model (here: analytically continued
Drude model), evaluated using Python tools and stored in ASCII for convenience.

## Usage

This overview is an idealized summary of necessary steps, excluding details of
installation, configuration, troubleshooting and methodological practices. Since
all data are already in place, none of the steps are strictly necessary unless
more data points or other adaptations should be needed.

### PIMC simulations

To re-compute or adapt the PIMC simulations:
1. Build and install 'pimc3' on a supercomputer of choice.
1. Launch selected simulation jobs
11. Manually: edit CONF/PARTICLES in a selected subdirectory and submit 'pimc3'
11. Nexus: edit 'simulations/run_*.py' and execute it using Python
1. Once complete, move on or repeat if needed.

### Data preprocessing

The raw data is preprocessed using 'pimc3/pytools/p3load'. For instance, the
following commands yield data similar to that in 'data/1_preproc/'

> p3load simulations/rs_*/theta_*/N*/tau_*/lr_*.0/obs/???/dynpol.h5 -o mu2_11,mu2pp_11 -w data/1_preproc/dynpol -a time

> p3load simulations/rs_*/theta_*/N*/tau_*/lr_*.0/obs/???/dynpol.h5 -o mu2_11,mu2pp_11 -F -w data/1_preproc/dynpol_fourier -a time -I 16

> p3load simulations/rs_*/theta_*/N*/tau_*/lr_*.0/obs/???/energy.h5 -w data/1_preproc/energy

Note 1: The Matsubara data of the autocorrelation function is loaded independently
to measure its apparent statistical uncertainty, which is hard to produce from
the uncertainties of the autocorrelation function.

Note 2: It is possible to do the preprocessing on a supercomputer and then pull
the (lightweight) preprocessing archive to local machine for postprocessing.

### Data postprocessing

Finally, the postprocessed data and the reference data are dynamically generated
as requested by postprocessing scripts. That is, for example running

> python3 analyze/plot_Giw_vs_theta.py

queries the preprocessing archive for selected data (based on the numerical
parameters), and if found, migrates or creates them to the postprocessing
archive with proper manipulations.

Where applicable, the provided scripts accept '-s' switch to 'save the figure'
to disk and '-ow' to 'overwrite' the postprocessing cache. They can be softly
edited for different parameters; run any script with '-h' for details.