Foundation Neural-Network Quantum States for Molecular Potential Energy Surfaces in Second Quantization

Abstract

Second-quantized neural-network quantum states have achieved accurate molecular energies, but extending them across molecular geometries requires a shared representation of the geometry-dependent wavefunction coefficients. We introduce geometry-conditioned foundation neural-network quantum states for molecular electronic structure in second quantization. A single autoregressive model learns a family of ground states from sparse anchor geometries and provides wavefunctions at untrained geometries without further optimization. Orbital alignment matches orbital identities and transports their phases, establishing an aligned orbital basis across geometries. Frozen energies reach chemical accuracy at every untrained query geometry for N₂, CO, and H₄. On additional molecular paths, the energy-trained wavefunctions yield dipoles, quadrupoles, and natural occupations without property labels. Across three paired N₂ training seeds, orbital alignment lowers the mean absolute energy error over all untrained query geometries from 34-37 mHa to 0.049-0.085 mHa. At approximately 1 mHa mean absolute error, frozen evaluation reduces the per-geometry cost by 986× relative to independent optimization, yielding an estimated 25.8× end-to-end GPU-cost reduction on a 161-point N₂ grid.

From Algebraic Completeness to Orbit Separation: Recursive Separating Invariants for Global Neural Network Potential Energy Surfaces
Yiping Hao, Jing Lou, Bina Fu +1
arXiv15 Oct 2026PreprintCC BY-NC-ND 4.0Full textPhysical chemistryarXiv
Isothermal-Isobaric Molecular Dynamics with a Foundation Interatomic Potential in LAMMPS across NVIDIA and Intel GPUs
Xiaoli Yan, Eliu Huerta
arXiv2 Nov 2026PreprintCC BY 4.0Full textPhysical chemistryarXiv
Protocol for analyzing potential targets of environmental pollutants in human diseases using network toxicology and molecular docking
Yanggang Hong, Jiajun Li, Jie Zhou +4
STAR Protocols14 May 2026Peer-reviewedCC BY5 citedMedicinal chemistry & drug discoveryOpenAlex
Elucidating the MultiPathway Therapeutic Potential of Trichosanthes kirilowii Bioactives in Intestinal Obstruction via Network Pharmacology and Molecular Dynamics
Tianying Wu, Xiaofen Lin, Wencong Lin +2
ChemistryOpen5 Oct 2026Peer-reviewedCC BY 4.0Full textGeneral chemistryCrossref
Network Pharmacology and Molecular Docking-Based Approach Investigating Secondary Metabolites from Aquilaria malaccensis as Potential Hepatitis B Agents
Darmadi Darmadi, Imelda Rey, OK Yulizal +1
Tropical Journal of Natural Product Research1 Oct 2026Peer-reviewedCC BY-NC-ND5 citedOrganic chemistryOpenAlex

Cite

BibTeX RIS

BibTeX

@misc{fu2026foundation,
  title = {{Foundation Neural-Network Quantum States for Molecular Potential Energy Surfaces in Second Quantization}},
  author = {Lizhong Fu and Jianan Wei and Wenguan Wang and Honghui Shang},
  howpublished = {arXiv},
  year = {2026},
  month = sep,
  doi = {10.48550/arxiv.2609.37733},
  eprint = {2609.37733},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.37733},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Foundation Neural-Network Quantum States for Molecular Potential Energy Surfaces in Second Quantization
AU  - Lizhong Fu
AU  - Jianan Wei
AU  - Wenguan Wang
AU  - Honghui Shang
PY  - 2026
DA  - 2026/09/29
PB  - arXiv
DO  - 10.48550/arxiv.2609.37733
UR  - https://arxiv.org/abs/2609.37733
AB  - Second-quantized neural-network quantum states have achieved accurate molecular energies, but extending them across molecular geometries requires a shared representation of the geometry-dependent wavefunction coefficients. We introduce geometry-conditioned foundation neural-network quantum states for molecular electronic structure in second quantization. A single autoregressive model learns a family of ground states from sparse anchor geometries and provides wavefunctions at untrained geometries without further optimization. Orbital alignment matches orbital identities and transports their phases, establishing an aligned orbital basis across geometries. Frozen energies reach chemical accuracy at every untrained query geometry for N_2, CO, and H_4. On additional molecular paths, the energy-trained wavefunctions yield dipoles, quadrupoles, and natural occupations without property labels. Across three paired N_2 training seeds, orbital alignment lowers the mean absolute energy error over all untrained query geometries from 34-37 mHa to 0.049-0.085 mHa. At approximately 1 mHa mean absolute error, frozen evaluation reduces the per-geometry cost by 986× relative to independent optimization, yielding an estimated 25.8× end-to-end GPU-cost reduction on a 161-point N_2 grid.
ER  -

Abstract

Related

Cite