colabfit/Massive_Atomic_Diversity_MAD-1.5_PBE

MAD-1.5 PBE数据集是MAD-1.5（大规模原子多样性版本1.5）的一个子集，使用PBE GGA泛函重新计算。它覆盖了MAD-1的子集（包括MC3D、MC3D-rattled、MC3D-random、MC3D-surface、MC3D-cluster、MC2D、SHIFTML-molcrys、SHIFTML-molfrags）以及新MAD-1.5子集中的单体和MC3D-random-extended。所有DFT设置与r2SCAN计算保持一致：使用FHI-aims（版本250806）全电子代码和紧致NAO基组（物种默认值2020），周期性系统的k点密度为8 Angstrom^-1，高斯展宽为0.05 eV，SCF收敛阈值为1e-6 eV（能量）、1e-4 eV/Angstrom（力）和1e-5 e*a0^-3（电子密度）。交叉验证分割与r2SCAN的训练/验证/测试分割一致；此文件包含所有三个分割的组合。PBE目标在PET-MAD-1.5模型训练中与r2SCAN目标一起使用，通过单独的预测头，相对于仅使用r2SCAN训练，力精度提高了约25%。作为较低理论级别的数据集，其整理不如主要的r2SCAN数据集仔细；PBE头在最终发布的模型中被丢弃。A subset of the MAD-1.5 (Massive Atomic Diversity version 1.5) structures recomputed with the PBE GGA functional, covering the MAD-1 subsets (MC3D, MC3D-rattled, MC3D-random, MC3D-surface, MC3D-cluster, MC2D, SHIFTML-molcrys, SHIFTML-molfrags) plus monomers and MC3D-random-extended from the new MAD-1.5 subsets. All DFT settings are consistent with the r2SCAN calculations: FHI-aims (version 250806) all-electron code with tight NAO basis sets (species defaults 2020), 8 Angstrom^-1 k-point density for periodic systems, Gaussian smearing of 0.05 eV, and SCF convergence thresholds of 1e-6 eV (energy), 1e-4 eV/Angstrom (forces), and 1e-5 e*a0^-3 (electron density). Cross-validation splits are consistent with the r2SCAN train/val/test splits; this file contains all three splits combined. PBE targets were used in PET-MAD-1.5 model training with separate prediction heads alongside r2SCAN targets, improving force accuracy by approximately 25% relative to r2SCAN-only training. As a lower level of theory, this dataset is less carefully curated than the primary r2SCAN dataset; PBE heads are discarded from the final released models.