Quantum Chemistry · foundry-ml
foundry_qmc_ml_v1-1
Quantum Monte Carlo and electronic structure benchmark dataset for 1,175 small molecules.
What foundry_qmc_ml_v1-1 is
foundry_qmc_ml_v1-1 provides high-accuracy electronic structure reference calculations for machine learning benchmarks across chemical space. Created by Bing Huang, O. Anatole von Lilienfeld, Jaron T. Krogel, and Anouar Benali, the dataset accompanies research on scalable quantum machine learning designed to reach Quantum Monte Carlo (QMC) accuracy levels.
The dataset contains molecular structures alongside electronic energy calculations for 1,175 small molecules. These benchmarks combine baseline quantum chemical methods with diffusion Monte Carlo (DMC) simulations, supporting comparative analysis and surrogate model training in quantum chemistry and materials science.
How it was made
Molecular geometries are provided both as standard XYZ strings and as pymatgen Molecule JSON strings, paired with a fragment identifier. Electronic ground-state energies were computed using all-electron calculations with a cc-pVTZ basis set across multiple levels of theory, including Hartree-Fock (HF) and density functional theory approximations (PBE, PBE0, and B3LYP).
High-accuracy reference energies were obtained via diffusion Monte Carlo (DMC) based on trial wavefunctions derived from HF, PBE, PBE0, and B3LYP methods. For every DMC energy, the dataset records the corresponding statistical error estimate in hartrees (Ha).
Key facts
- Dataset size
- 1,175 molecules
- Data format
- Parquet, tabular
- Coordinate formats
- XYZ string, pymatgen Molecule JSON
- Basis set
- cc-pVTZ (all-electron)
- Energy units
- Hartree (Ha)
- Reference methods
- HF, PBE, PBE0, B3LYP, DMC(HF), DMC(PBE), DMC(PBE0), DMC(B3LYP)
- Authors
- Bing Huang, O. Anatole von Lilienfeld, Jaron T. Krogel, Anouar Benali
- Release year
- 2022
What you can use it for
- Machine learning potential benchmarking
- Train and evaluate quantum machine learning models to reproduce high-level diffusion Monte Carlo molecular energies directly from 3D coordinates.
- Delta-learning and multi-fidelity modeling
- Learn energy corrections between lower-level methods (HF, PBE, PBE0, B3LYP) and high-fidelity DMC reference targets across small molecules.
- Trial wavefunction sensitivity analysis
- Compare how different initial determinants (HF, PBE, PBE0, and B3LYP) influence the final energy and associated uncertainty in DMC simulations.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from foundry-ml/foundry_qmc_ml_v1-1 at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 1,175 rows of default/train
B3LYP
float- min
- -437.50
- median
- -244.57
- mean
- -238.98
- max
- -40.50
DMC(B3LYP)
float- min
- -437.46
- median
- -244.57
- mean
- -238.98
- max
- -40.51
DMC(B3LYP)_err
float- min
- 2.7e-4
- median
- 7.1e-4
- mean
- 7.2e-4
- max
- 2.0e-3
DMC(HF)
float- min
- -437.46
- median
- -244.56
- mean
- -238.97
- max
- -40.51
DMC(HF)_err
float- min
- 2.5e-4
- median
- 7.3e-4
- mean
- 7.5e-4
- max
- 9.5e-3
DMC(PBE)
float- min
- -437.47
- median
- -244.57
- mean
- -238.98
- max
- -40.51
DMC(PBE)_err
float- min
- 3.1e-4
- median
- 7.1e-4
- mean
- 8.9e-4
- max
- 0.19
DMC(PBE0)
float- min
- -437.46
- median
- -244.57
- mean
- -238.98
- max
- -40.51
DMC(PBE0)_err
float- min
- 2.8e-4
- median
- 7.0e-4
- mean
- 7.2e-4
- max
- 3.3e-3
HF
float- min
- -435.81
- median
- -243.18
- mean
- -237.68
- max
- -40.21
PBE
float- min
- -437.25
- median
- -244.41
- mean
- -238.82
- max
- -40.46
PBE0
float- min
- -437.25
- median
- -244.42
- mean
- -238.84
- max
- -40.47
fragment
int- min
- 1
- median
- 588
- mean
- 588
- max
- 1,175
pymatgen
string_text- min
- 542
- median
- 1,655
- mean
- 1593.59
- max
- 2,497
xyz
string_text- min
- 240
- median
- 782
- mean
- 751.34
- max
- 1,189
Strengths and limits
Good at
- Supplies benchmark-grade diffusion Monte Carlo electronic energies alongside statistical error estimates for all QMC calculations.
- Provides multi-method coverage on identical geometries, encompassing both standard mean-field/DFT methods and DMC.
- Includes geometries in both standard XYZ format and pymatgen-compatible JSON strings for streamlined integration with Python tools.
Watch out for
- The dataset is limited in size to 1,175 small molecule entries.
- Only a single predefined train split is provided within the record.
- Property outputs are confined to electronic ground-state energies and errors, without forces or other electronic observables.
Questions
Can I look inside foundry_qmc_ml_v1-1 without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is foundry_qmc_ml_v1-1 under?
Its authors declare other. Read the terms on the original page before reusing the data.
What units are used for the energies and error calculations?
All reported energies (HF, PBE, PBE0, B3LYP, and their corresponding DMC calculations) and DMC statistical errors are provided in hartrees (Ha).
What input representations of the molecules are included?
Each row contains a fragment ID, a standard XYZ-formatted coordinate string, and a pymatgen Molecule representation serialized as a JSON string.
What QMC variants are included in the dataset?
The dataset provides diffusion Monte Carlo results evaluated with trial wavefunctions from Hartree-Fock and three density functionals: DMC(HF), DMC(PBE), DMC(PBE0), and DMC(B3LYP).
How to cite it
Huang, Bing and von Lilienfeld, O. and Krogel, Jaron T and Benali, Anouar (2022). Foundry - Approaching QMC quality energetics throughout chemical space using scalable quantum machine learning. Materials Data Facility. https://doi.org/10.18126/wg30-95z0
Written from foundry_qmc_ml_v1-1's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.