遇见数据集

RGD1-TSopt-GFN2: a transition-state optimization benchmark on the GFN2-xTB surface

收藏
Zenodo2026-06-01 更新2026-06-05 收录
官方服务:

资源简介:

RGD1-TSopt-GFN2 is a benchmark for evaluating transition-state (TS) geometry optimizers on the GFN2-xTB potential energy surface. It contains 1500 distinct transition states sampled from the RGD1 reaction database, each provided with three distorted starting geometries of increasing difficulty (easy, med, hard), for a total of 4500 optimizer inputs. Reference transition states. Each reference is a GFN2-xTB first-order saddle point with exactly one imaginary vibrational mode, computed with the tblite implementation of GFN2-xTB. The geometries are tightly optimized: the maximum Cartesian gradient component is below 1×10⁻⁶ Hartree/Bohr at tight SCF convergence (tblite accuracy 0.01). The GFN2-xTB energy and the Cartesian Hessian are provided at the same setting. (At the default tblite/xtb accuracy the residual gradient is on the order of 1×10⁻⁵ Hartree/Bohr of SCF noise.) All systems are neutral, closed-shell molecules of C, H, N, and O (4–24 atoms). The 1500 reactions are distinct: each has a unique pair of atom-mapped reactant/product SMILES, and no two reference structures coincide. Distorted starting geometries. For each reference TS, three starts are produced by displacing the saddle along normal modes of the mass-weighted GFN2-xTB Hessian: a component (alpha) along the imaginary (reaction) mode and a component (beta) along a random combination of the remaining (non-reactive) modes. The contributions are orthogonal, so the total RMSD is √(alpha²+beta²); each displacement is deterministic. Tiers (RMSD components, Å): easy 0.06/0.12, med 0.11/0.12, hard 0.15/0.12. Difficulty is set by the reaction-mode component alpha. Contents. One directory per case (cases/<id>/): reference TS (coord, xyz, energy, Hessian), the three distorted starts (coord, xyz), the reaction endpoints (reactant.xyz, product.xyz), and a provenance file with the RGD1 id and atom-mapped SMILES. A manifest.tsv lists distortion parameters per structure, and README.md documents all files and columns. Geometries are in Å (.xyz) and atomic units (.coord, TURBOMOLE format). Intended use. For each distorted start, run a TS optimization on the GFN2-xTB surface and compare the result to the reference (Cartesian RMSD or energy), optionally recording optimizer cycles/gradient calls. The graded tiers let success rate and cost be reported as a function of the initial displacement along the reaction coordinate. Source and references. Transition states sampled from RGD1 (Zhao, Q.; Savoie, B. M. Sci. Data 2023, 10, 145). GFN2-xTB method (Bannwarth, C.; Ehlert, S.; Grimme, S. J. Chem. Theory Comput. 2019, 15, 1652).

提供机构:
Zenodo
创建时间:
2026-06-01
二维码
社区交流群
二维码
科研交流群
商业服务