Datasets, MLIP models, and global-search workflows for medium-sized silicon clusters (Si_n = 16–38)

This dataset contains the structural, energetic, and methodological files associated with the global-minimum study of medium-sized silicon clusters Si, from 16 to 38 atoms. It includes the TPSSh/def2-TZVP and r2SCAN/def2-TZVP reference frames used to train the machine-learned interatomic potentials, the resulting NequIP model for the r2SCAN/def2-TZVP training, and the Python script employed for the genetic-algorithm global search. The repository also provides the xyz files for the r2SCAN reoptimisations of literature reference structures, the structures generated during the searches with the TPSSh- and r2SCAN-trained potentials, and the final PBE0/def2-TZVP optimised candidates, including both validated minima and structures later discarded as non-minima or false positives after higher-level analysis. Together, these files enable reproduction of the MLIP training, structural exploration, and electronic-structure refinement workflow reported in the associated manuscript.