Coupled-cluster molecular properties across the main group that extrapolate beyond training size
Abstract
Coupled-cluster theory defines the accuracy standard for molecular electronic-structure properties but scales too steeply for routine application, whereas density-functional theory is affordable yet systematically biased. We resolve this trade-off with a single equivariant network, HARP (Hamiltonian Read-out for Properties), that predicts an effective one-electron Hamiltonian from one inexpensive B3LYP/def2-SVP calculation and derives a broad suite of properties from it (energy, optical gap, dipole, quadrupole, polarizability, Mulliken atomic charges, and Mayer bond orders) at coupled-cluster accuracy across nine main-group elements, including the under-served phosphorus, sulfur, and chlorine chemistries. The model is trained on a new in-house dataset of multi-property labels computed at the CCSD(T) level for all nine elements. On a held-out test set, it reduces the error of every property by a factor of 3.8 to 270 relative to semi-local, hybrid, and double-hybrid DFT (referenced to composite CCSD(T)/cc-pVTZ), while adding only ~0.1 s wall time per molecule, delivering coupled-cluster-quality predictions at the cost of a single DFT calculation. Critically, deriving every property from a predicted Hamiltonian rather than pooling per-atom features builds the correct size-scaling into the model architecture: on pi-conjugated oligothiophenes it matches finite-field CCSD polarizability to ~1% and the EOM-CCSD optical gap to ~3% at the largest sizes where those references remain affordable (44 and 37 atoms, where a single CCSD field point already costs ~500x the model's entire inference) and extrapolates the corrected trends to 58-atom chains, a regime where pooling-based architectures fail by construction. Accurate extrapolation is therefore set by the model's inductive bias rather than by the training data.
Related
- Accelerated "on-the-fly" coupled-cluster path-integral molecular dynamics: Impact of nuclear quantum effects on an asymmetric proton
- Bringing coupled‑cluster theory to the strong-correlation frontier
- Examining Convergence of Cluster Perturbation Theory and Cluster Coupled Cluster for J1-J2 Heisenberg Spin Lattices
- Coupled Cluster Amplitude Extrapolation in Geometry Optimization
- Coupled cluster theory for multireference and open-shell molecules
BibTeX
@misc{he2026coupledcluster,
title = {{Coupled-cluster molecular properties across the main group that extrapolate beyond training size}},
author = {Wenhao He and Xu Chen and Noah Song and Haowei Xu and Tim S. Hindges and Bohan Li and Zihan Lin and Yu Yao and Avetik R. Harutyunyan and Fang Liu and Yao Wang and Hao Tang and Ju Li},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2608.18346},
eprint = {2608.18346},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.18346},
}
RIS
TY - UNPB
TI - Coupled-cluster molecular properties across the main group that extrapolate beyond training size
AU - Wenhao He
AU - Xu Chen
AU - Noah Song
AU - Haowei Xu
AU - Tim S. Hindges
AU - Bohan Li
AU - Zihan Lin
AU - Yu Yao
AU - Avetik R. Harutyunyan
AU - Fang Liu
AU - Yao Wang
AU - Hao Tang
AU - Ju Li
PY - 2026
DA - 2026/10/01
PB - arXiv
DO - 10.48550/arxiv.2608.18346
UR - https://arxiv.org/abs/2608.18346
AB - Coupled-cluster theory defines the accuracy standard for molecular electronic-structure properties but scales too steeply for routine application, whereas density-functional theory is affordable yet systematically biased. We resolve this trade-off with a single equivariant network, HARP (Hamiltonian Read-out for Properties), that predicts an effective one-electron Hamiltonian from one inexpensive B3LYP/def2-SVP calculation and derives a broad suite of properties from it (energy, optical gap, dipole, quadrupole, polarizability, Mulliken atomic charges, and Mayer bond orders) at coupled-cluster accuracy across nine main-group elements, including the under-served phosphorus, sulfur, and chlorine chemistries. The model is trained on a new in-house dataset of multi-property labels computed at the CCSD(T) level for all nine elements. On a held-out test set, it reduces the error of every property by a factor of 3.8 to 270 relative to semi-local, hybrid, and double-hybrid DFT (referenced to composite CCSD(T)/cc-pVTZ), while adding only ~0.1 s wall time per molecule, delivering coupled-cluster-quality predictions at the cost of a single DFT calculation. Critically, deriving every property from a predicted Hamiltonian rather than pooling per-atom features builds the correct size-scaling into the model architecture: on pi-conjugated oligothiophenes it matches finite-field CCSD polarizability to ~1% and the EOM-CCSD optical gap to ~3% at the largest sizes where those references remain affordable (44 and 37 atoms, where a single CCSD field point already costs ~500x the model's entire inference) and extrapolates the corrected trends to 58-atom chains, a regime where pooling-based architectures fail by construction. Accurate extrapolation is therefore set by the model's inductive bias rather than by the training data.
ER -