Overview

This dataset qm9_op_pdos_fhi-aims.h5 and accompanying Python script op_pdos_analysis.py provide a comprehensive resource for analyzing the electronic structure of organic molecules in the QM9 dataset[L. Ruddigkeit et al. (2012),  R. Ramakrishnan et al. (2014)]. The dataset includes Overlap Population (OP) and Projected Density of States (PDOS) data, calculated using the FHI-aims software package[V. Blum et al. (2009)] with aims_chembond chemical bonding analysis functionality[I. Takahara et al. (2024)]. The Python script, offers functions to parse and visualize this data, facilitating further analysis and research.

Dataset Description

The HDF5 file qm9_op_pdos_fhi-aims.h5 is organized into groups and datasets, each containing specific types of data related to the electronic structure of molecules. Each molecule in the QM9 dataset has a corresponding group in the HDF5 file, identified by its molecule ID. Within each molecule group, the following datasets and subgroups are included:

Top-Level Structure

• Molecule Groups: Each molecule in the QM9 dataset has a corresponding group in the HDF5 file. The name of each group corresponds to the molecule ID in QM9 dataset.

Molecule Group Structure

Within each molecule group, the following datasets and attributes are stored:

• Attributes:

  • n_atoms: The number of atoms in the molecule.
  • n_states: The number of electronic states without spin considered.

• Datasets:

  • z: Atomic numbers of the atoms in the molecule. The length is n_atoms
  • pos: Positions of the atoms in the molecule. Note that the unit is Ångström. The shape is (n_atoms, 3).
  • ks_eigenvalue: Kohn-Sham eigenvalues from the FHI-aims calculations. Note that the unit is Hartree. The energy levels corresponding to the OP and PDOS weights both refer to this value. The energy origin (0) corresponds to the vacuum level. The shape is (k_point, spin_channel, n_states).
  • occ_numbers: Occupation numbers corresponding to the Kohn-Sham eigenvalues. This can be used for determining the HOMO and LUMO. The shape is (k_point, spin_channel, n_states).

• Subgroups:

  • op: Contains datasets related to the overlap population (OP) data. Edge specific.
    • Attributes:
      • bond_index: Index of the bonds considered in the molecule.
    • Datasets:
      • weights: Weights for the OP data. The shape is (bond_index, k_point, spin_channel, eigen_state)
  • pdos: Contains datasets related to the PDOS data. Node specific.
    • Attributes:
      • lmax: List of maximum number of l for each site.
    • Datasets:
      • weights_{i_site}: Weights for the PDOS data for each site. i_site is the index of atom sites starting from 0. The shape is (lmax[i_site], k_point, spin_channel, eigen_state).

Notes:

• length along k_point is 1, only the Γ point since the calcuations are non-periodic.
• length along spin_channel is 2, the up and down channel.
• length along eigen_state is n_states.

Calculation Method and Conditions

The electronic structure calculations were performed using the FHI-aims software package. The Perdew-Burke-Ernzerhof (PBE) generalized gradient approximation (GGA) was used for the exchange-correlation functional. Collinear spin polarization was considered, and no relativistic effects were included. The molecular structures were taken directly from the torch_geometric.datasets.qm9.QM9 object without geometry relaxation.

Python Script

The op_pdos_analysis.py script provides functions to parse and visualize the OP and PDOS data from the HDF5 file. Key features of the script include:

• Gaussian Function: Calculates the Gaussian distribution for data smoothing.
• Smeared Spectrum Function: Computes the smeared spectrum using Gaussian broadening.
• Fetcher Class: Fetches OP and PDOS data for molecules from the HDF5 file.
• Plotter Class: Inherits from the Fetcher class and provides methods to plot and save OP and PDOS data.