Learning neural network potentials from experimental data via Differentiable Trajectory Reweighting

TL;DR

The paper introduces DiffTRe, a novel method for training neural network potentials directly from experimental data by efficiently computing gradients without backpropagating through MD simulations, enabling better integration of experimental info.

Contribution

DiffTRe provides a scalable, stable approach for top-down learning of neural network potentials from experimental data, extending capabilities beyond bottom-up methods.

Findings

Achieves 100x faster gradient computation for top-down learning.

Successfully learns NN potentials for diamond and water models from experimental data.

Generalizes bottom-up structural coarse-graining methods to arbitrary potentials.

Abstract

In molecular dynamics (MD), neural network (NN) potentials trained bottom-up on quantum mechanical data have seen tremendous success recently. Top-down approaches that learn NN potentials directly from experimental data have received less attention, typically facing numerical and computational challenges when backpropagating through MD simulations. We present the Differentiable Trajectory Reweighting (DiffTRe) method, which bypasses differentiation through the MD simulation for time-independent observables. Leveraging thermodynamic perturbation theory, we avoid exploding gradients and achieve around 2 orders of magnitude speed-up in gradient computation for top-down learning. We show effectiveness of DiffTRe in learning NN potentials for an atomistic model of diamond and a coarse-grained model of water based on diverse experimental observables including thermodynamic, structural and mechanical properties. Importantly, DiffTRe also generalizes bottom-up structural coarse-graining methods such as iterative Boltzmann inversion to arbitrary potentials. The presented method constitutes an important milestone towards enriching NN potentials with experimental data, particularly when accurate bottom-up data is unavailable.