Gradient-based multiconfiguration Shepard interpolation for generating potential energy surfaces for polyatomic...

Area of Science:

• Computational Chemistry
• Theoretical Chemistry
• Chemical Dynamics

Background:

• Constructing accurate potential energy surfaces (PESs) is crucial for chemical dynamics simulations.
• Traditional methods often require computationally expensive electronic structure calculations, including Hessian information.
• Existing methods may be prohibitive for large molecular systems or when analytical Hessians are unavailable.

Purpose of the Study:

• To present a novel gradient-based multiconfiguration Shepard interpolation (MCSI) method for constructing multidimensional PESs.
• To demonstrate the efficacy of MCSI without relying on electronic structure Hessian data.
• To enable accurate and affordable dynamics calculations for complex chemical reactions.

Main Methods:

• Developed a multiconfiguration Shepard interpolation (MCSI) method, also known as multiconfiguration molecular mechanics (MCMM).
• Utilized Shepard interpolation of first-order Taylor series expansions of a diabatic Hamiltonian matrix.
• Focused solely on gradient information, omitting Hessian calculations from electronic structure computations.

Main Results:

• Accurately represented multidimensional PESs for two test reactions: OH+H2 and methyl radical abstraction from alpha-tocopherol.
• Achieved mean unsigned errors within 1 kcal/mol for both reactions, despite using limited gradient data (13 and 11 gradients, respectively).
• Demonstrated the method's applicability to a 108-dimensional PES for a 38-atom system.

Conclusions:

• The gradient-based MCSI (MCMM) method provides an efficient way to represent multidimensional PESs.
• This approach is suitable when analytical Hessians are too costly or unavailable.
• MCSI opens new avenues for employing high-level electronic structure data in dynamics at a reduced computational cost.