HIP neural potential predicts Hessians directly, cutting error ~60% vs differentiation
bravo_abad · x · 2026-10-19
Burger and colleagues introduce the Hessian Interatomic Potential (HIP), a faster approach to AI molecular calculations:
- Problem: computing vibrational modes or reaction transition states requires knowing how moving one atom changes forces on others (the Hessian), which usually still means expensive differentiation even when energies/forces are predicted quickly by AI.
- Method: HIP adds a prediction head to a graph neural network trained on reference Hessians, baking in the matrix's symmetry and rotational transformation rules — no differentiation of energy/force outputs needed.
- Results: versus the corresponding EquiformerV2 baseline using differentiation, HIP reduces Hessian error by 60% while being faster.
Figures compare DFT, HIP and autodiff baselines on potential energy surfaces and Hessian eigenvalues along reaction coordinates, showing HIP closely matching DFT references.
More from Research
- Probe repo: GPT-4, with pre-GRPO cutoff, predicts DeepSeekMath's GRPO from anonymized text — CatAstro_Piyush · 2026-10-20
- Mistral launches Timestral time-series foundation models, zero-shot SOTA on GIFT-Eval — victormustar · 2026-10-20
- MD simulations reveal striking hydration dynamics in daraxonrasib tri-complex pockets — CatAstro_Piyush · 2026-10-20
- Marburg's Sebastian Wild releases full Advanced Algorithms course videos and materials — caglar_ee · 2026-10-20
- ICLR paper's Future Work section just points to a NEXT_VERSION.md in the supplement — CSProfKGD · 2026-10-20
- Sabi open-sources 27B BCI model with 15% hallucination rate, far below GPT-6 — lukas_m_ziegler · 2026-10-20