Parallel Implementation of Large-Scale Linear Scaling Density Functional Theory Calculations With Numerical Atomic...

Area of Science:

• Computational Chemistry
• Materials Science
• Condensed Matter Physics

Background:

• Conventional density functional theory (DFT) calculations face high computational costs (cubic-scaling).
• Linear-scaling DFT methods offer efficiency by exploiting the nearsightedness principle.
• Efficient algorithms are crucial for studying large-scale material properties.

Purpose of the Study:

• To present a parallel implementation of a linear-scaling density matrix purification algorithm.
• To address the computational bottleneck of sparse matrix multiplication in linear-scaling DFT.
• To enable accurate and efficient electronic structure calculations for large systems.

Main Methods:

• Implementation of a linear-scaling density matrix trace correcting (TC) purification algorithm.
• Utilizing numerical atomic orbitals within the HONPAS package.
• Employing MPI_Allgather for parallel sparse matrix multiplication in CSR format.
• Leveraging Kohn's nearsightedness principle for sparse Hamiltonian matrices.

Main Results:

• Demonstrated computational accuracy and efficiency of the parallel algorithm.
• Successfully performed large-scale DFT calculations on boron nitrogen nanotubes.
• The method scales effectively to hundreds of processing cores on heterogeneous supercomputers.

Conclusions:

• The parallel linear-scaling DFT algorithm significantly enhances computational efficiency.
• This approach is suitable for large-scale electronic structure calculations of complex materials.
• The implementation provides a scalable solution for modern supercomputing environments.