SlaterGPU

Library for numerically computing Slater-type orbital integrals. For running on GPU, OpenACC is required. A compatible compiler should be used. Testing has been done with NVHPC SDK version 24.9. A wrapper library for Libcint (https://github.com/sunqm/libcint) is also provided for Gaussian integrals. It should be noted that version 5.3.0 of libcint is required at the moment.

This library uses the resolution of the identity (RI) approximation. Thus, auxiliary and main basis sets must be specified.

For STOs, functions up to 6h (5g for derivatives) are available for the auxiliary basis set. With up to 6h auxiliary functions, it is not recommended to go beyond 4f in the main basis. The library requires user supplied basis sets. See the examples folder for formatting inputs. Atoms up to Zn are currently supported.

Note: This repository is an experimental research project, and functionality related to the prolate spheroidal coordinate system is still in progress and particularly unstable.

Example of getting dependencies on Zimmerman group cluster

Source the script that automatically loads modules and adds dependencies to PATH:

• Note: Paul's nvhpc install seems to have the math libraries in a different location which requires a cmake modification

source env.set.local0

For other users, this bash script may be a helpful example to see how to install / load necessary dependencies.

For compiling run :

cmake -Bbuild
cmake --build build -j 16

The executable will be generated at build/examples/sgpu.exe

Testing:

Go into one of the example directories inside the build directory.

cd build/examples/lih_VK1

Then execute SlaterGPU within this directory to run the test.

../sgpu.exe

After running, if successful the directory should contain the output files "A", "Ciap", "SENT", and "pVp".

Notes:

If running on multiple GPUs, it's advised to have the number of OpenMP threads equal to the number of GPUs. i.e. set the following environment variable

export OMP_NUM_THREADS=<ngpu>

There are example calculations in SlaterGPU/examples/ with integral files denoted generally as INT_ref. It should be noted that normally, the three center coefficients, Ciap, are generated, but were omitted due to Git's file size limit.

Please see LICENSE file for licensing information.

Experimental pixi build:

• Clone this repository
• Install pixi if you don't already have it (curl -fsSL https://pixi.sh/install.sh | sh)
• Ensure NVIDIA HPC SDK is available on the path
  • On athena: module load nvidia-sdk/25.5
  • On perlmutter:

    module use /global/cfs/cdirs/m4957/joshkamm/hpc_sdk/modulefiles
    module unload cudatoolkit
    module load nvhpc/25.5
    

• Install dependencies and build SlaterGPU with pixi -v install (-v to see output from pixi and cmake build)
• If the install has succeeded, activate the pixi environment with pixi shell inside the folder where this repository is cloned
• Run the executable using sgpu.exe in a directory with the appropriate input files
• Experimental global installation to streamline routine usage when not modifying the code and remove the need for pixi shell: pixi -v global install --path . Requires Aug 2025 pixi update: pixi self-update
• Running jobs interactively:
  • pixi -v run srun -p zimA10 -N1 -n2 --gpus=1 --pty /bin/bash on athena will obtain a slurm allocation and return an interactive bash terminal on a compute node inside of the pixi configured environment with sgpu.exe on PATH

Configuration of pixi can be found in the pixi.toml file.