Interpolating moving least-squares methods for fitting potential energy surfaces: a strategy for efficient automatic...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Molecular Modeling

Background:

• Accurate potential energy surfaces (PES) are crucial for understanding molecular behavior.
• Traditional PES construction methods can be computationally expensive and time-consuming.
• Automated and efficient methods are needed for complex molecular systems.

Purpose of the Study:

• To demonstrate an accurate and efficient automated method for molecular global potential energy surface (PES) construction and fitting.
• To develop a flexible interpolating moving least-squares (IMLS) method adaptable to various levels of ab initio data.
• To optimize PES generation for applications such as molecular dynamics and spectroscopy.

Main Methods:

• Developed an interpolating moving least-squares (IMLS) method capable of fitting energies, gradients, and Hessian data.
• Implemented a local IMLS approach storing fitting coefficients at expansion points for computational efficiency.
• Utilized an automatic point selection scheme based on successive IMLS fits and conjugate gradient minimizations for optimal data placement.

Main Results:

• Demonstrated the accuracy and efficiency of the automated IMLS method for PES construction.
• Showcased the method's scalability across one-dimensional (Morse) to nine-dimensional (CH4) molecular systems.
• Validated the efficacy of conjugate gradient minimizations for high-dimensional data point selection.

Conclusions:

• The developed automated IMLS method provides an accurate, efficient, and scalable approach for global PES construction.
• The flexibility in fitting different levels of ab initio data makes it suitable for various computational chemistry applications.
• This method significantly advances the automated generation of high-quality potential energy surfaces for molecular simulations.