MACE-mpa-0 relaxation trajectories with ASE optimizers

This dataset contains full geometry-relaxation trajectories generated with the MACE-mpa-0 machine-learning interatomic potential (MLIP) using several ASE optimizers (BFGS, LBFGS, and line-search variants of these, FIRE, and (SciPyFmin) CG).It accompanies the publication Benchmarking Local Geometry Optimization Algorithms for Computational Materials Discovery. For each optimizer, all intermediate relaxation images are stored (not only initial and final structures), together with energies, forces, convergence flags, and timing information. In addition to the main datasets (one HDF5 file per optimizer using ASE defaults), the record includes alpha-test datasets that probe the effect of the initial inverse Hessian scaling for BFGS-type optimizers. The data are provided in HDF5 format with a consistent schema across all files.Detailed documentation and usage examples are included in the accompanying README.md. If you require additional data, such as full atomic coordinates or ASE Atoms objects, please feel free to contact one of the authors.