A neural network LDA functional overfit to water data achieves 1 kcal/mol errors on ionization and atomization energies and matches PBE/B3LYP on WATER27 binding energies after transfer learning from one datum.
Zuo, et al., Performance and Cost Assessment of Machine Learning Interatomic Potentials
8 Pith papers cite this work, alongside 927 external citations. Polarity classification is still indexing.
8
Pith papers citing it
927
external citations · Crossref
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
A new atomistic simulation framework uses GPP-relaxed variances and on-the-fly NEB barriers to evolve concentrations over diffusion time scales of seconds to years, validated against Cu self-diffusion and Al segregation benchmarks.
EquiformerV2 universal machine learning potentials predict energies and forces of metal and alloy defects with errors below 5 meV/atom and 100 meV/A on most benchmark datasets, approaching DFT accuracy.
Materials phenomena such as fatigue crack growth are framed as conditional probability landscapes over competing unit mechanisms, to be inferred from multiscale simulation and multimodal data and then optimized toward desired emergent outcomes.
ML model using ideal entropy plus simulation features (energy above hull, heat capacity change, icosahedral fraction) predicts metallic glass critical cooling rates with R²=0.78 in leave-one-chemical-system-out cross-validation on 34 alloys.
Higher model accuracy improves uncertainty-error correlation and novelty detection in MLIP UQ, and clustering-enhanced local D-optimality better detects novel environments on heterogeneous datasets.
Three NMC811-derived cathode materials proposed by an expert-guided LLM pipeline were synthesized and showed 160 to 174 mAh/g reversible capacity versus 135 mAh/g for NMC811.
MLACS is a production Python package that iteratively trains linear MLIP surrogates with active learning and MBAR reweighting to sample the DFT canonical ensemble at 50 to 100 times lower DFT cost.
citing papers explorer
-
Overfitting by design: neural network density functionals for water
A neural network LDA functional overfit to water data achieves 1 kcal/mol errors on ionization and atomization energies and matches PBE/B3LYP on WATER27 binding energies after transfer learning from one datum.