Reproduce “Reweighting free energy profiles between universal machine learning interatomic potentials for fast consensus building” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sauradeep Majumdar, Miguel Steiner, Johannes C. B. Dietschreit, Swagata Roy, Daniel Willimetz, Lukaš Grajciar, Rafael Gómez-Bombarelli
A reweighting framework from researchers at MIT and the University of Vienna allows efficient transfer of free energy profiles between universal machine learning interatomic potentials (MLIPs), with an energy-only approximation robustly recovering thermodynamic properties for ion transport, even when phase-space overlap is low.
Key insights
- The 'energy-only' reweighting approximation effectively overcomes the statistical collapse of direct exponential reweighting, providing robust PMF estimates even when phase-space overlap between MLIPs is critically low.
- MLIPs cluster into distinct groups based on their underlying DFT exchange-correlation functionals and training datasets, dictating the topology of their predicted free energy landscapes.
- The reweighting entropy score effectively quantifies phase-space overlap, serving as a diagnostic tool for predicting the robustness of reweighting estimates for different MLIP pairs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.