Deep Learning of ab initio Hessians for Transition State Optimization

TL;DR

This paper introduces a machine learning approach using an equivariant neural network to efficiently predict Hessians for transition state optimization, significantly reducing computational costs and improving robustness over traditional DFT methods.

Contribution

The authors develop and implement a fully differentiable ML-based Hessian prediction method for transition state searches, outperforming traditional quasi-Newton methods in accuracy and efficiency.

Findings

Successfully optimized transition states for 240 unseen reactions

Reduced optimization steps by 2-3 times compared to DFT-based methods

Demonstrated robustness even with poor initial guesses

Abstract

Identifying transition states -- saddle points on the potential energy surface connecting reactant and product minima -- is central to predicting kinetic barriers and understanding chemical reaction mechanisms. In this work, we train an equivariant neural network potential, NewtonNet, on an ab initio dataset of thousands of organic reactions from which we derive the analytical Hessians from the fully differentiable machine learning (ML) model. By reducing the computational cost by several orders of magnitude relative to the Density Functional Theory (DFT) ab initio source, we can afford to use the learned Hessians at every step for the saddle point optimizations. We have implemented our ML Hessian algorithm in Sella, an open source software package designed to optimize atomic systems to find saddle point structures, in order to compare transition state optimization against quasi-Newton Hessian updates using DFT or the ML model. We show that the full ML Hessian robustly finds the transition states of 240 unseen organic reactions, even when the quality of the initial guess structures are degraded, while reducing the number of optimization steps to convergence by 2--3$\times$ compared to the quasi-Newton DFT and ML methods. All data generation, NewtonNet model, and ML transition state finding methods are available in an automated workflow.