eDMFT

If you want your jobs to run faster in the queue

#SBATCH -A mendoza_q

#SBATCH --exclude=amr-163,amr-178,amr-179

this should submit to our queue, exclude your buyin nodes and flip the jobs to run in the general partition, avoiding incurring cpu.

Installing and running eDMFT

This will download a Singularity Image to your current directory for EDMFT

singularity pull docker://haulek/edmft:amd64

This will load the image to the current session

singularity shell edmft_amd64.sif

Tutorial for MnO

http://hauleweb.rutgers.edu/tutorials/Tutorial1a.html

Tutorial for FeSe (superonductor)

http://hauleweb.rutgers.edu/tutorials/Tutorial2a.html

Inside the shell, you can run eDMFT commands like you had a module/environment loaded. You can also run eDMFT via MPI from inside the container, but you need to set certain options, which you can do with:

      echo "mpirun --mca btl '^openib' -n [number_of_processes]" > mpi_prefix.dat 

#!/bin/bash --login

#SBATCH --job-name=eDMFT_MnO       # Name of your job

#SBATCH --nodes=1                  # Number of nodes

#SBATCH --ntasks-per-node=1        # Number of tasks (usually 1 for single-processor runs)

#SBATCH --cpus-per-task=1          # Number of CPU cores (adjust as needed)

#SBATCH --mem-per-cpu=4G           # Memory per node (adjust as needed)

#SBATCH --time=02:00:00            # Wall clock time limit

#SBATCH --partition=your_partition # Specify the correct partition for your cluster

# --- Environment Setup ---

echo "Setting up the environment..."

# 1. Start with a clean slate. This is crucial to avoid conflicts.

# 2. Load the eDMFT module. This sets up paths for the compiled Fortran/C++ parts.

module purge

module load eDMFT/Oct-2024-intel-2023a

module load WIEN2k/23.2-intel-2023a

# 3. Manually FIX the incorrect WIEN_DMFT_ROOT set by the module.

#    This is the temporary workaround until the HPC team fixes the module file.

export WIEN_DMFT_ROOT=/opt/software-current/2023.06/x86_64/generic/software/eDMFT/Oct-2024-intel-2023a

#export EBROOTEDMFT="/opt/software-current/2023.06/x86_64/generic/software/eDMFT/Oct-2024-intel-2023a/bin"

# 4. Activate YOUR personal Python virtual environment.

#    This is the key step that provides the correct Python and packages like numba.

#echo "Activating Python virtual environment..."

#source ~/my-material-science-env/bin/activate

source ~/py36-edmft-env/bin/activate

# 5. (Optional but Recommended) Verify the environment is correct.

#    This is very useful for debugging. Check the output file.

echo "--- Environment Verification ---"

echo "WIEN_DMFT_ROOT is: $WIEN_DMFT_ROOT"

echo "Which python: $(which python)"

echo "Which run_dmft.py: $(which run_dmft.py)"

echo "Python Version: $(python --version)"

echo "NumPy Version: $(python -c 'import numpy; print(numpy.__version__)')"

echo "----------------------------"

# --- Run the Calculation ---

echo "Starting eDMFT calculation..."

# Now, run your script. It will use the Python from your virtual environment.

run_dmft.py

echo "Calculation finished."

#!/bin/bash

set -e

# One OpenMP thread per MPI rank is correct for the CTQMC solver.

export OMP_NUM_THREADS=1

# Scratch for large Wien2k vector files. "." keeps them in the run directory,

# which is bound into the container and persists. For heavy I/O you may point

# this at the node-local disk instead (confirm the local path with ICER).

export SCRATCH="."

# Single-node MPI launch command consumed internally by run_dmft.py.

# --mca btl '^openib' disables the deprecated InfiniBand-verbs transport, which

# is unavailable/erroring inside the container; on one node OpenMPI uses shared

# memory. -n 192 fills every core of an amd24 node.

echo "mpirun --mca btl '^openib' -n 60" > mpi_prefix.dat

# Run the full DFT+embedded-DMFT charge-self-consistent loop.

run_dmft.py >& nohup.dat

Make sure that the number of cpus in your mpi command

echo "mpirun --mca btl '^openib' -n 60" > mpi_prefix.dat

matches what you request in slurm

#SBATCH --ntasks=60

https://docs.icer.msu.edu/Cluster_Resources/