SciLM-ai/MaterialsSaddles

MaterialsSaddles是一个用于固态和表面化学的高通量收敛过渡态库。该数据集包含34,135,597个完全收敛的过渡态，这些过渡态是通过在公共材料和催化数据集上进行大规模并行鞍点搜索计算得到的，使用了`tsearch`包和Meta的`uma-s-1p2`机器学习原子间势能。每个文件中的条目是一个单一结构，三个连续的条目形成一个过渡态事件：反应物最小值、过渡态（一阶鞍点）和产物最小值。端点收敛到0.02 eV/Å（最大|F|），鞍点收敛到0.05 eV/Å。每个鞍点行还存储了其虚频振动模式——一个`(N, 3)`的每原子位移场，给出了鞍点不稳定的方向。该数据集旨在用于训练生成模型以预测过渡态、生成DFT标签以对抗MLIP势垒软化以及热启动DFT鞍点搜索。

MaterialsSaddles is a high-throughput library of converged transition states for solid-state and surface chemistry. The dataset contains 34,135,597 fully converged transition states computed by massively-parallel saddle searches on top of public materials and catalysis datasets, using the `tsearch` package and Metas `uma-s-1p2` machine-learning interatomic potential. Each entry in a file is a single structure, and three consecutive entries form one transition-state event: reactant minimum, transition state (first-order saddle), product minimum. Endpoints are converged to 0.02 eV/Å (max|F|), saddles to 0.05 eV/Å. Each saddle row also stores its imaginary-frequency vibrational mode — an `(N, 3)` per-atom displacement field giving the direction along which the saddle is unstable. The dataset is intended for training generative models for transition-state prediction, generating DFT labels to fight MLIP barrier softening, and warm-starting DFT saddle searches.