Molecular relaxation by reverse diffusion with time step prediction
This addresses the problem of efficiently finding equilibrium molecular structures in computational chemistry, offering a novel statistical approach that avoids costly non-equilibrium data, though it is incremental in improving over existing force field methods.
The paper tackles molecular relaxation by proposing MoreRed, a reverse diffusion method that treats non-equilibrium structures as noisy instances of equilibrium states, achieving competitive results with a root-mean-square deviation of 0.15 Å and energy errors within 1 kcal/mol compared to reference structures. It uses a simpler pseudo potential energy surface and requires only unlabeled equilibrium data, reducing dataset size and computational cost.
Molecular relaxation, finding the equilibrium state of a non-equilibrium structure, is an essential component of computational chemistry to understand reactivity. Classical force field (FF) methods often rely on insufficient local energy minimization, while neural network FF models require large labeled datasets encompassing both equilibrium and non-equilibrium structures. As a remedy, we propose MoreRed, molecular relaxation by reverse diffusion, a conceptually novel and purely statistical approach where non-equilibrium structures are treated as noisy instances of their corresponding equilibrium states. To enable the denoising of arbitrarily noisy inputs via a generative diffusion model, we further introduce a novel diffusion time step predictor. Notably, MoreRed learns a simpler pseudo potential energy surface (PES) instead of the complex physical PES. It is trained on a significantly smaller, and thus computationally cheaper, dataset consisting of solely unlabeled equilibrium structures, avoiding the computation of non-equilibrium structures altogether. We compare MoreRed to classical FFs, equivariant neural network FFs trained on a large dataset of equilibrium and non-equilibrium data, as well as a semi-empirical tight-binding model. To assess this quantitatively, we evaluate the root-mean-square deviation between the found equilibrium structures and the reference equilibrium structures as well as their energies.