mingxue2026/adc2_qm8

QM8 ADC(2) dataset

This repository contains ADC(2) excited-state calculations for QM8-like molecular geometries. Each molecule is stored as a compressed NumPy .npz file with ground-state quantities, transition densities, transition dipoles, oscillator strengths, and convergence diagnostics.

Repository layout

.
├── README.md
├── scripts/
│   ├── run_adcc_mol.py
│   └── run_adcc_mol_usage.md
├── adc2_qm_out/
│   ├── 00000_04999/
│   │   ├── mol_00000.npz
│   │   └── ...
│   ├── 05000_09999/
│   └── ...
└── geometries_xyz/
    ├── 00000_04999/
    ├── 05000_09999/
    ├── 10000_14999/
    ├── 15000_19999/
    └── 20000_21527/

Both NPZ files and XYZ geometry files are split into chunk folders to keep each folder small enough for convenient browsing and upload.

NPZ file schema

Each mol_*.npz file follows the same key order:

[
    "symbols",
    "xyz",
    "exc_energies_eV",
    "mu_trans_au",
    "osc_strengths",
    "rdm1_ground",
    "tdm_ao",
    "X_ia",
    "mo_coeff",
    "S_ao",
    "basis",
    "nocc",
    "nvirt",
    "nAO",
    "dav_niter_total",
    "scf_energy_eV",
    "mo_energy",
    "mo_occ",
    "r_ao",
    "tdm_mo",
    "adc_max_residual",
    "adc_converged",
]

Quick inspection

To inspect one molecule:

import numpy as np

path = "adc2_qm_out/00000_04999/mol_00000.npz"
with np.load(path, allow_pickle=False) as data:
    print(data.files)
    print("symbols:", data["symbols"])
    print("xyz shape:", data["xyz"].shape)
    print("excitation energies / eV:", data["exc_energies_eV"])
    print("oscillator strengths:", data["osc_strengths"])
    print("ADC converged:", data["adc_converged"])
    print("ADC max residual:", data["adc_max_residual"])

Reproducing calculations

Scripts for reproducing individual ADC(2) calculations are in scripts/.

• scripts/run_adcc_mol.py: reproduce one molecule from an XYZ file or geometry index.
• scripts/run_adcc_mol_usage.md: detailed usage guide plus helper scripts for printing NPZ contents and eta-threshold diagnostics.

Example:

python scripts/run_adcc_mol.py \
  --xyz geometries_xyz/00000_04999/10001.xyz \
  --output mol_00003.npz \
  --basis def2-svp \
  --nroots 3 \
  --adc-conv-tol 1e-6 \
  --adc-max-iter 100

The reproduction script writes the same NPZ schema as the dataset files.

Data chunks

The NPZ and XYZ collections are split into folders of at most 5,000 files. The folder names indicate the sorted index range contained in each chunk.

NPZ chunks:

00000_04999
05000_09999
10000_14999
15000_19999
20000_21527

XYZ chunks:

00000_04999
05000_09999
10000_14999
15000_19999
20000_21527

Convergence diagnostics

The main ADC convergence fields are:

• adc_converged: boolean convergence flag.
• adc_max_residual: maximum final ADC residual in Hartree.
• dav_niter_total: total Davidson iterations reported by ADCC.

For residual/gap diagnostics used in the analysis notebook:

gap12_eV = E2 - E1
gap23_eV = E3 - E2
eta12 = 2 * residual_eV / abs(gap12_eV)
eta23 = 2 * residual_eV / abs(gap23_eV)
residual_eV = adc_max_residual * 27.211386245988

See scripts/run_adcc_mol_usage.md for a copy-pasteable eta-threshold script.

Notes

• Energies in exc_energies_eV and scf_energy_eV are in electronvolts.
• Coordinates in xyz are in Angstrom.
• The basis set used for the provided calculations is def2-svp.
• Transition dipoles mu_trans_au are stored in atomic units.