Efficient Construction of Excited-State Hessian Matrices with Machine Learning Accelerated Multilayer Energy-Based...

Area of Science:

• Computational chemistry
• Quantum mechanics
• Materials science

Background:

• Describing excited states of large systems is computationally demanding.
• Existing methods struggle with accuracy and efficiency for complex molecules.
• Fragment-based methods offer a promising approach to tackle large systems.

Purpose of the Study:

• To develop an efficient method for constructing excited-state Hessian matrices of large systems.
• To accelerate computational efficiency using machine learning models.
• To explore Hessian-matrix-based excited-state properties of large systems.

Main Methods:

• Developed a multilayer energy-based fragment (MLEBF) method for excited states.
• Derived MLEBF for efficient construction of excited-state Hessian matrices.
• Integrated machine learning (ML) models with MLEBF, creating ML-MLEBF, by replacing the inert region with trained ML models.

Main Results:

• The MLEBF method accurately calculates energies and gradients for large systems.
• The ML-MLEBF method significantly improves computational efficiency for Hessian matrices, especially in large systems.
• Both MLEBF and ML-MLEBF methods are highly parallel and exhibit low-scaling computational costs.

Conclusions:

• The developed ML-MLEBF method offers an efficient and accurate approach for excited-state Hessian calculations in large systems.
• This work demonstrates the potential of combining ML techniques with fragment-based electronic structure methods.
• Future research can explore Hessian-matrix-based excited-state properties of large systems using these advanced computational tools.