Reproduce “Distilling Large-Scale Machine Learning Interatomic Potentials to Predict Long-Timescale Electrolyte Transport Properties” 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
Ishan Amin, Yue Jian, Nitesh Kumar, Christine Zhang, Jianwei Lai, Casey S. Mezerkor, Aaron M. Melemed, Jonathan R. Thurston and 5 more
The authors train a compact PaiNN student on energy, force, and stress labels from the UMA-OMol teacher for electrolyte simulations. The student reaches 9.5 ns/day versus 0.23 for the teacher, and its conductivity median percentage error is 36% across nine systems at concentrations of 0.5 M or higher, versus 67% for OPLS-AA. It inherits teacher biases, so this does not establish unbiased predictions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.