Usage

Preparing the input

DGAmore takes as input the result of a DMFT calculation, which currently must be produced with w2dynamics, a continuous-time quantum Monte Carlo solver based on the hybridisation expansion. A w2dynamics run yields both the one-particle quantities and a two-particle output containing the four-point Green’s functions. The two-particle output first has to be converted into the format DGAmore expects. This is done by running the symmetrize_new.py script, which is installed alongside the main entry point; it prompts for the input and output file names and writes only the density and magnetic spin components of the two-particle Green’s function to the output file. The original, unsymmetrised vertex file is not needed afterwards.

In addition to the processed two-particle file, a run requires a configuration file, which describes all run-time parameters, as well as a Hamiltonian. The Hamiltonian can be supplied either as a real-space (Wannier) Hamiltonian or as a momentum-space Hamiltonian generated by wien2wannier, and for interaction types that are not handled directly an additional file specifying the local and non-local interaction can be provided.

Running a calculation

The main entry point of the program is DGAmore.py. Because it is added to the Python environment as a standalone executable, it can be invoked by name without its full path. For single-core execution, which is mainly intended for testing, run:

DGAmore.py

For a parallel run with MPI, use:

mpiexec -np <n_proc> DGAmore.py

Instead of mpiexec you may also use mpirun or, on SLURM-based clusters, srun. The number of processes <n_proc> should be chosen according to the problem size and the available resources; note that increasing the process count adds communication overhead and does not always improve performance.

Two optional command-line arguments are available. The -p argument specifies the directory containing the configuration file, which is useful when several configuration files are kept in different directories; if it is not given, the path defaults to the repository directory. The -c argument specifies the name of the configuration file to load and defaults to dga_config.yaml. As an example, the following command runs the code with eight MPI processes and loads the configuration file my_config.yaml from /configs/:

mpiexec -np 8 DGAmore.py -p /configs/ -c my_config.yaml

On a SLURM-based cluster, a typical job submission script looks as follows:

#!/bin/bash
#SBATCH -N <n nodes>
#SBATCH -J <some job name>
#SBATCH --partition=<some partition>
#SBATCH --qos=<some qos>
#SBATCH --ntasks-per-node=<n proc>
#SBATCH -t <time limit>
#SBATCH -o log.txt
#SBATCH -e log.txt

# Load the necessary modules; here we activate the conda environment that
# has the DGAmore package and its dependencies preinstalled.
module purge
source <path to miniconda>/miniconda3/bin/activate <your conda env>

# Use MPI for parallelisation, so restrict each task to a single thread.
export OMP_NUM_THREADS=1

# Recommended on SLURM-based clusters:
srun DGAmore.py -p "<path to config>" -c "<name of config>.yaml"

# Alternatively, with mpirun or mpiexec:
mpirun -np $SLURM_NTASKS DGAmore.py -p "<path to config>" -c "<name of config>.yaml"

The -o and -e options set the files for the job output and errors; here both are written to the same file, but separate files may be used instead. The results of a completed run are written to a subdirectory of the output folder, whose name encodes run-specific parameters such as the momentum-grid size and the frequency box.

The full set of run-time parameters is described on the Configuration page.