GPU Accelerated Minimal Auxiliary Basis Approach TDDFT for Large Organic Molecules.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Accurate excited-state calculations are crucial for understanding molecular properties and reactions.
• Traditional methods struggle with large systems due to high computational cost.
• Developing efficient algorithms for large-scale quantum chemistry is an ongoing challenge.

Purpose of the Study:

• To introduce a GPU-accelerated implementation of time-dependent density functional theory with the minimal auxiliary basis approach (TDDFT-risp).
• To demonstrate the capability of TDDFT-risp for large molecular systems using the Tamm-Dancoff approximation (TDA-risp).
• To assess the accuracy and efficiency of the new method compared to existing approaches.

Main Methods:

• GPU-accelerated evaluation of three-center integrals and tensor contractions.
• Implementation of exchange-space truncation and omission of hydrogen atoms from the auxiliary basis.
• Utilizing a host memory-assisted Davidson solver for large eigenvalue problems.
• Employing the Tamm-Dancoff approximation (TDA-risp) for excited-state calculations.

Main Results:

• TDDFT-risp achieves excitation-energy errors of ~0.03-0.05 eV for low-lying states with a 40 eV exchange cutoff.
• Calculations for systems of 300-3000 atoms on a single A100 GPU range from minutes to hours.
• GPU TDA-risp shows 140-340× speedups over conventional methods for medium-sized systems.

Conclusions:

• GPU-TDDFT-risp offers a practical and efficient approach for excited-state calculations in large organic and biomolecular systems.
• The method significantly reduces computational time, enabling studies of previously intractable systems.
• This advancement opens new possibilities for computational chemistry research on complex molecular architectures.