Dataset for Unified Deep Learning Framework for Many-Body Quantum Chemistry via Green's Functions

Quantum chemistry data for QM7/QM9 molecules, silicon nanoclusters, azobenzene derivatives, and ethanol/methanol/ethane bond stretching.

The underlying DFT/Hartree-Fock matrices were calculated with PySCF (https://github.com/pyscf) and any available ab initio MBGF results (quasiparticle energies, electron density, self-energy matrices) were calculated with fcdmft (https://github.com/ZhuGroup-Yale/fcdmft). The h5-formatted data in each .chk file can be loaded with pyscf.lib or with the code associated with this publication (https://zenodo.org/records/15178073 or https://github.com/ZhuGroup-Yale/mlgf), which offers additional data processing functionality. Examples of data loading can be found in the included tests.py.

All the provided data have DFT (or Hartree-Fock) matrices - the detailed breakdown of available MBGF is given below:

QM9 training data (2000 molecules + 660 AIMD conformers) have all GW self-energy tensors

QM9 training data (13,165 molecules) have GW density matrix, GW quasiparticle energies, and GW photoemission spectra

Silicon nanoclusters training data (159 clusters) and test cases (14 clusters) have GW quasiparticle energies, and GW photoemission spectra

Azobenzene derivatives training data (100 conformers each of 6 derivatives) and test cases (13 each of 2 derivatives) have GW quasiparticle energies, and GW photoemission spectra

Ethane (50 conformers), methanol (117 conformers), and ethanol (13 conformers) have all coupled cluster self-energy tensors