Accelerating Linear-Response Time-Dependent Hybrid Density Functional Theory with Low-Rank Decomposition Techniques...

Area of Science:

• Computational materials science
• Quantum chemistry
• Condensed matter physics

Background:

• Linear-response time-dependent density functional theory (LR-TDDFT) is crucial for studying electronic excitations.
• Hybrid functionals in TDDFT offer improved accuracy but are computationally expensive.
• Efficient methods are needed to study large systems and complex electronic properties.

Purpose of the Study:

• To present an efficient low-rank implementation of LR-TDDFT for hybrid functionals (hybrid-LR-TDDFT).
• To introduce adaptively compressed exchange (ACE) operator and natural transition orbitals (NTOs) for computational acceleration.
• To enable the study of excitonic properties in large material systems.

Main Methods:

• Implementation of a low-rank approximation for the nonlocal exchange operator in hybrid-LR-TDDFT.
• Utilizing the adaptively compressed exchange (ACE) operator.
• Incorporating natural transition orbitals (NTOs) with a cutoff parameter for further acceleration.

Main Results:

• The ACE approximation significantly reduces computational cost without sacrificing accuracy.
• NTO approximation further accelerates calculations, enabling efficient studies.
• Successfully studied excitonic properties of 2D MoS2 (216 atoms) on a single GPU.

Conclusions:

• The developed hybrid-LR-TDDFT method is computationally efficient and accurate.
• This approach facilitates the investigation of excitonic properties in large, complex systems.
• The method opens new possibilities for materials science research using hybrid functionals.