Research map / Architectures and efficiency / RNNs and normalization
Smooth Dynamic Cutoffs for Machine Learning Interatomic Potentials
Kevin Han, Haolin Cong, Bowen Deng, Amir Barati Farimani
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use dynamic cutoff instead of fixed cutoff (mu=40) | MACE, Nequip, Orbv3, TensorNet · MatPES-r2scan | energy and force MAE: MACE force 173 -> 190; Nequip force 167 -> 168 | Mixed |
| use dynamic cutoff instead of fixed cutoff (mu=20) | MACE, Nequip, Orbv3, TensorNet · MD22 | energy and force MAE: DWNT force MACE 35.60 -> 41.22 | Hurt |
| use dynamic cutoff to induce graph sparsity | four MLIPs · MOF and protein benchmarks | memory and inference time: up to 2.26x less memory, 2.04x faster | Improved |
| use dynamic cutoff to induce graph sparsity | MLIP edge count · MOF material system | number of edges: 677000 -> 297000 | Improved |
| use dynamic cutoff to induce graph sparsity | MLIP edge count · protein system | number of edges: 805040 -> 588141 | Improved |
| replace hard max-neighbor limit with smooth dynamic cutoff | TensorNet · NVE MD stability | energy drift / PES smoothness | Improved |
| use dynamic cutoff in 100 ps NVE simulations (DWNT, LiFePO4 at 3000K) | four MLIPs · NVE MD | total energy drift | Improved |
| increase target neighbors mu from 5 to 40 (ablation) | TensorNet 456k · MD22 buckyball catcher and DWNT | force MAE | Improved |
| use dynamic cutoff instead of reduced fixed cutoff with equal average neighbors (ablation) | TensorNet · MatPES-r2scan | energy and force MAE: energy 53 -> 43; force 171 -> 160 | Improved |
| use dynamic cutoff instead of reduced fixed cutoff (ablation) | copper melting 4500K · 864-atom Cu melt | average neighbors per atom | Improved |
| dynamic cutoff on elements with large atomic radii | TensorNet · MatPES per-element force error | relative force MAE increase | Hurt |
Closest papers by meaning
- MLIP Arena: Advancing Fairness and Transparency in Machine Learning Interatomic Potentials via an Open, Accessible Benchmark Platform (NeurIPS 2025 · similarity 0.90)
- Physics-Informed Weakly Supervised Learning For Interatomic Potentials (ICML 2025 · similarity 0.90)
- FlashTP: Fused, Sparsity-Aware Tensor Product for Machine Learning Interatomic Potentials (ICML 2025 · similarity 0.90)
- Stein Kernelized Molecular Dynamics for Active Learning of Interatomic Potentials (NeurIPS 2026 · similarity 0.89)
- Colour me shocked: Exact Molecular Hessians from MLIPs in O(N) time using sparse differentiation! (NeurIPS 2026 · similarity 0.89)
- Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction (ICML 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
