Research map / Probabilistic and scientific modeling / Dynamical systems
Learning Hamiltonian Flow Maps: Mean Flow Consistency for Large-Timestep Molecular Dynamics
Winfried Ripken, Michael Plainer, Gregor Lied, Thorben Frank, Oliver T. Unke, Stefan Chmiela, Frank Noe, Klaus Robert Muller
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 |
|---|---|---|---|
| increase HFM integration timestep from 0.5 fs to 9 fs (ablation) | HFM · Aspirin NVT | h(r) MAE: 0.026 -> 0.046 | Hurt |
| increase HFM integration timestep from 0.5 fs to 9 fs (ablation) | HFM · Ethanol NVT | h(r) MAE: 0.075 -> 0.115 | Hurt |
| replace MLFF+Velocity Verlet with HFM at 0.5 fs | MLFF · Aspirin NVT | h(r) MAE: 0.030 -> 0.026 | Improved |
| use HFM instead of Velocity Verlet for N-body rollout at coarse discretization | Gravitational N-body (100 bodies) · N-body t=3 to t*=4 | MSE: 4x fewer steps than VV for same accuracy | Improved |
| use HFM with dt=9 fs instead of MLFF+VV at 0.5 fs for fixed step budget | Paracetamol NVE · Paracetamol NVE exploration | state space exploration | Improved |
| distill CG ScoreMD into HFM with dt=45 fs | Chignolin CG · Chignolin NVT | speedup vs reference force field: exceeding 20x | Improved |
| use Nose-Hoover thermostat instead of Langevin | HFM · NVT simulations | simulation artifacts | Hurt |
| train on sparse 256 samples instead of 85k samples | HFM · Paracetamol NVT | free energy surface and spectrum | No clear effect |
| raise alanine dipeptide dt to about 10 fs | HFM · Alanine dipeptide NVT | rollout stability | Hurt |
| sample dt=0 in 75% of training cases | HFM training · MLFF datasets | training stability | Improved |
| use stop-gradient on consistency target | HFM training · training cost | training overhead: approximately 30% overhead vs MLFF | Improved |
| train HFM on biased data with a missing conformational mode | HFM · MD22 Ac-Ala3-NHMe NVT | free energy landscape | Improved |
Closest papers by meaning
- Hamiltonian Score Matching and Generative Flows (NeurIPS 2024 · similarity 0.88)
- FlashMD: long-stride, universal prediction of molecular dynamics (NeurIPS 2025 · similarity 0.87)
- Conditional Normalizing Flows for Active Learning of Coarse-Grained Molecular Representations (ICML 2024 · similarity 0.87)
- Learning Macroscopic Dynamics from Partial Microscopic Observations (NeurIPS 2024 · similarity 0.86)
- Generative Modeling of Molecular Dynamics Trajectories (NeurIPS 2024 · similarity 0.86)
- How to build a consistency model: Learning flow maps via self-distillation (NeurIPS 2025 · similarity 0.85)
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.
