DGDFT: A massively parallel method for large scale density functional theory calculations.

Area of Science:

• Computational Physics
• Materials Science
• Quantum Chemistry

Background:

• Density Functional Theory (DFT) is crucial for electronic structure calculations.
• Kohn-Sham DFT methods face scalability challenges for large systems.
• Developing efficient and accurate computational methods is essential.

Purpose of the Study:

• To introduce a massively parallel implementation of the Discontinuous Galerkin DFT (DGDFT) method.
• To enable efficient large-scale electronic structure calculations.
• To assess the accuracy and scalability of DGDFT for 2D materials.

Main Methods:

• Utilized adaptive local basis (ALB) functions generated on-the-fly.
• Employed pole expansion and selected inversion techniques.
• Implemented a two-level parallelization scheme for high-performance computing.

Main Results:

• DGDFT exhibits at most quadratic scaling with the number of electrons.
• Achieved high accuracy for 2D phosphorene: 1.3 × 10⁻⁴ Hartree/atom (energy) and 6.2 × 10⁻⁴ Hartree/bohr (force).
• Demonstrated 80% parallel efficiency on 128,000 cores for large phosphorene systems (3500-14,000 atoms).

Conclusions:

• The parallel DGDFT method offers a significant advancement for large-scale electronic structure computations.
• The ALB approach provides systematic accuracy improvement.
• The method is highly scalable and efficient for studying complex 2D materials.