School of Civil and Environmental Engineering

Ph.D. Thesis Defense Announcement

HYBRID FUNCTIONAL IN REAL-SPACE DENSITY FUNCTIONAL THEORY

By Xin Jing

Advisor:

Dr. Phanish Suryanarayana (CEE/CSE)

Committee Members:  Dr. Andrew Medford (ChBE)

Dr. Edmond Chow (CSE)

Dr. Victor Fung (CSE)

Dr. John Pask (LLNL)

Date and Time:  November 06, 2024. 4:00pm EST

Location: Ford ES&T L1114, https://gatech.zoom.us/j/94221339623

We present an efficient real-space formalism for hybrid exchange-correlation functionals in generalized Kohn-Sham density functional theory (DFT), leveraging the Kronecker product structure of the finite-difference Laplacian matrix. This approach enables competitive time-to-solution performance compared to the fast Fourier transform (FFT) method, without imposing boundary condition restrictions. Our formalism, implemented for both unscreened and range-separated hybrid functionals, demonstrates up to an order-of-magnitude speedup in time-to-solution compared to established planewave codes, verified for isolated and bulk systems. Furthermore, an accelerated version on NVIDIA GPUs achieves 2-8x speedup on V100 clusters and 10-40x on H100 clusters. We also extend the method to study the structure of liquid water using ab initio molecular dynamics, finding good agreement with literature. Additionally, we introduce O(N) and O(N^2) hybrid functionals within the Spectral Quadrature (SQ) and Discrete Discontinuous Basis Projection (DDBP) frameworks, respectively, significantly reducing computational cost for large bulk system simulations, offering a promising pathway for efficient real-space DFT calculations with hybrid functionals.