Dataset accompanying machine learning of electronic structure and atomistic properties from the external potential

Electronic structure data accompanying Ref. 1. These archives provide molecular geometries along with atomic-orbital (AO) matrix representations of the external (nuclear attraction) potential V, Fock matrix H, density matrix P (stored as `dm`), and overlap matrix S, computed using PySCF at the DFT/LDA level unless noted otherwise. Matrix data are stored as hickle files: each file contains a Python list of length N (number of structures), with one torch.Tensor per frame, in the same order as the corresponding `.xyz` file. Atomic structures are provided in as .xyz files readable with ASE. An example snippet to load the data in python is as follows: import hickle from ase.io import read frames = read("water_1000.xyz", ":") fock = hickle.load("def2-tzvp/fock.hickle") # list of N tensors H_i = fock[i] # AO Fock matrix for frames[i] anthracene.zip : anthracene.xyz : 1000 frames of anthracene taken from Ref. 2. Fock, density matrix (dm) and overlap matrices as hickle files computed in the def2-SVP and STO-3G basis, respectively (in the def2svp and sto-3g subfolders) For the matrices in the def2-SVP basis, we also provide the projection on the STO-3G basis along with the projected indices following the procedure described in the Supplementary Info of Ref. 1. These matrices can be found in def2svp/projected_onto_sto-3g/. water.zip : water.xyz : 1000 frames of distorted water monomer taken from Ref. 3. Fock, density matrix (dm), and overlap matrices as hickle files computed in the def2-TZVP and STO-3G basis respectively, along with the properties computed from the Fock matrix (energy, eigenvalues, dipole moments and Löwdin charges) The external potential computed in the def2-SVP, cc-PVTZ, and cc-PVQZ basis sets are also provided. Access them at <basis_name>/vext.hickle. water_dimer.zip : water_dimers.xyz : 1000 frames resulting from the trajectories of 50 random pairs of molecules from the water dataset described above, obtained by systematically increasing the intermolecular separation between the two monomers from 3.5 Angstroms to 12 Angstroms in twenty steps. Reference energy and dipole moments (saved respectively as energy.hickle and dipole.hickle) computed using the def2-TZVP basis (DFT/PBE) are provided in the def2-tzvp subfolder. The external potential is computed in the cc-PVTZ basis and included in the ccpvtz subfolder. qm7.zip : qm7.xyz : 1000 frames containing C, H, N, O atoms, sampled randomly from the QM7b data (Ref. 4) Fock, density matrix (dm) and overlap matrices as hickle files computed in the def2-TZVP and STO-3G basis respectively, along with the properties computed from the Fock matrix (energy, eigenvalues, dipole moments, and Löwdin charges). For the matrices in the def2-TZVP basis, we also provide the projection on the STO-3G basis along with the projected indices following the procedure described in the Supplementary Info of Ref. 1. These matrices can be found in def2-tzvp/projected_onto_sto-3g/. The external (nuclear attraction) potential computed in the cc-PVTZ basis are also provided. Access them at ccpvtz/vext.hickle. The code to generate the electronic matrices and use them for machine learning is provided at Ref. 5. If you are using these data or the accompanying code, please cite this data repository along with Ref. 1. References : J. Nigam, T. Smidt, & G. Dusson, Machine learning electronic structure and atomistic properties from the external potential, arXiv:2602.15345 (2026) Materials Cloud Archive, Electronic excited states from physically-constrained machine learning, E. Cignoni et al., https://doi.org/10.24435/materialscloud:j2-58 (2024) Materials Cloud Archive, Dataset Open Symmetry-Adapted Machine Learning for Tensorial Properties of Atomistic Systems, A. Grisafi et al., https://doi.org/10.24435/materialscloud:2018.0009/v1 (2018) G. Montavon et al., Machine learning of molecular electronic properties in chemical compound space, New J. Phys. 15, 095003 (2013) https://github.com/curiosity54/op2op