Accelerating the density-functional tight-binding method using graphical processing units.

Area of Science:

• Computational Chemistry
• Materials Science
• High-Performance Computing

Background:

• The density-functional tight-binding (DFTB) method is crucial for simulating molecular systems.
• Computational bottlenecks in DFTB calculations limit the scale and speed of simulations.
• Accelerating DFTB is essential for advancing materials discovery and chemical process modeling.

Purpose of the Study:

• To accelerate the density-functional tight-binding (DFTB) method on graphical processing units (GPUs).
• To address computational bottlenecks in Hamiltonian matrix diagonalization and density matrix construction within DFTB.
• To evaluate the performance and scalability of the accelerated DFTB code on different GPU architectures.

Main Methods:

• Implementation of DFTB acceleration using the MAGMA linear algebra library on GPUs.
• Benchmarking on high-performance computing systems (SUMMIT supercomputer) and in-house GPU clusters.
• Performance and parallel scalability analysis using 1D, 2D, and 3D molecular models (carbon nanotubes, covalent organic frameworks, water clusters).

Main Results:

• Significant acceleration of DFTB ground-state calculations achieved through GPU implementation.
• Demonstrated performance gains and parallel scalability across various molecular system dimensions.
• Successful identification and mitigation of key computational bottlenecks in DFTB.

Conclusions:

• The GPU-accelerated DFTB method offers a substantial performance improvement for large-scale molecular simulations.
• The MAGMA library effectively facilitates the acceleration of critical DFTB computational steps.
• This advancement enables more efficient and extensive computational studies in chemistry and materials science.