materialyze/matpes

MatPES是一个用于训练基础势能（FPs）的数据集，覆盖周期表中几乎所有元素的机器学习原子间势能（MLIPs）。该数据集由Materialyze实验室和Materials Project合作开发，旨在解决材料势能表面数据集中的关键缺陷。数据集的特点包括高精度（使用静态DFT计算）、全面性（通过两阶段DIRECT采样方法从大量MD结构中采样）和高质量（包含PBE和r2SCAN功能的数据）。初始版本v2025.1包含约400,000个结构，来自300K MD模拟，并在训练FPs中表现出优异的性能和可靠性。MatPES是MatML生态系统的一部分，还包括MatGL、maml、MatPES和MatCalc等组件。

MatPES is a dataset used to train foundational potentials (FPs), i.e., machine learning interatomic potentials (MLIPs) with near-complete coverage of the periodic table. The dataset is an initiative by the Materialyze Lab and the Materials Project to address critical deficiencies in such PES datasets for materials. Key features of the dataset include accuracy (computed using static DFT calculations), comprehensiveness (sampled using a 2-stage DIRECT method from a large configuration of MD structures), and quality (includes data from both PBE and r2SCAN functionals). The initial v2025.1 release comprises ~400,000 structures from 300K MD simulations and demonstrates comparable or improved performance and reliability in training FPs. MatPES is part of the MatML ecosystem, which also includes MatGL, maml, MatPES, and MatCalc.