Accurate and Fast Geometry Optimization with Time Estimation and Method Switching

TL;DR

This paper introduces a scheme to estimate execution times for quantum chemistry geometry optimizations at various accuracy levels and proposes a method switching technique that dynamically reduces total computation time while maintaining accuracy.

Contribution

It presents a novel time estimation scheme for different accuracy levels and a gradient-based method switching technique to optimize geometry calculations efficiently.

Findings

Estimated optimization times with 29.5% mean error.

GMS reduces execution time by up to 42.7%.

Maintains accuracy despite method switching.

Abstract

Geometry optimization is an important task in quantum chemical calculations to analyze the characteristics of molecules. A top concern on it is a long execution time because time-consuming energy and gradient calculations are repeated across several to tens of steps. In this work, we present a scheme to estimate the execution times of geometry optimization of a target molecule at different accuracy levels (i.e., the combinations of ab initio methods and basis sets). It enables to identify the accuracy levels where geometry optimization will finish in an acceptable time. In addition, we propose a gradient-based method switching (GMS) technique that reduces the execution time by dynamically switching multiple methods during geometry optimization. Our evaluation using 46 molecules in total shows that the geometry optimization times at 20 accuracy levels are estimated with a mean error of 29.5%, and GMS reduces the execution time by up to 42.7% without affecting the accuracy of geometry optimization.