PyPWDFT: A Lightweight Python Software for Single-Node 10K Atom Plane-Wave Density Functional Theory Calculations.

Area of Science:

• Computational Physics
• Materials Science
• Quantum Chemistry

Background:

• Density Functional Theory (DFT) is a cornerstone of modern computational materials science and quantum chemistry.
• Efficient software is crucial for tackling large-scale DFT calculations, enabling the study of complex systems.
• Existing high-performance DFT codes are often written in Fortran or C++, limiting accessibility for some researchers.

Purpose of the Study:

• To develop a high-performance, user-friendly DFT software package in Python.
• To enable large-scale DFT calculations on single nodes with CPU-GPU heterogeneous computing.
• To demonstrate that a native Python implementation can achieve performance comparable to traditional compiled codes.

Main Methods:

• Development of PyPWDFT using a native Python environment, leveraging NumPy, SciPy, and CuPy.
• Implementation of plane-wave basis sets and density functional approximations (local and hybrid).
• Benchmarking against established Fortran/C++ DFT codes and evaluating performance on consumer and professional-grade GPUs.

Main Results:

• PyPWDFT achieves performance comparable to established Fortran/C++ DFT codes for large systems (up to 10,000 atoms).
• CPU-GPU heterogeneous computing with CuPy provides a two-order-of-magnitude speedup over single-threaded CPU execution.
• Consumer-grade GPUs (e.g., NVIDIA GeForce RTX 4090) offer computational efficiency rivaling professional-grade GPUs (e.g., NVIDIA V100).

Conclusions:

• PyPWDFT demonstrates significant potential as a versatile DFT software package.
• Its efficient performance, scalability, numerical accuracy, and cross-platform compatibility lower barriers to entry for DFT research.
• The software facilitates medium-scale DFT calculations via a GUI on personal computers, broadening access to advanced computational methods.