Learning the self-energy through equivariant neural networks

Abstract

The electronic self-energy determines the single-particle Green's function and, through it, quasiparticle energies and spectral functions, but computing it from first principles is expensive and it is a difficult machine-learning target: complex-valued, matrix-valued in a local-orbital basis, and dependent on a continuous frequency. We predict the dynamical correlation part of the imaginary-frequency G_0W_0 self-energy matrix of molecules from atomic species and coordinates with an E(3)-equivariant message-passing network, the rotational covariance of the orbital-space matrix being imposed by the architecture itself and the static exchange part taken from the density-functional reference. Frequency enters either as an explicit network input or through frequency-specific output heads, and the predicted matrices are continued to the real axis in a representation in which causality and a non-negative spectral density hold by construction. The frequency-conditioned model reproduces matrix elements to below 1 meV on the imaginary axis for the systems studied here, and that accuracy survives the continuation: for water clusters, including a tetramer absent from the training set, and for chemically diverse molecules from QM9, the reconstructed spectral functions follow the G_0W_0 reference peak by peak and the frontier quasiparticle levels are recovered to a few tens of meV. Learning the self-energy on the imaginary axis is therefore a practical route to real-frequency spectra: once the model is trained, and given the density-functional reference it is built on, a new geometry costs no explicit G_0W_0 calculation.

Emergent Loop Geometry in Gauge-Equivariant Graph Neural Networks
Ali Rayat, Gia-Wei Chern
arXiv13 Oct 2026PreprintarXiv licenceFull textCondensed matter High-energy & particle physicsarXiv
Accelerating dynamic polarizability calculations of organic molecules using equivariant graph neural networks
Houssam Metni, Maria Kraus, Marie Louise Schubert +3
arXiv7 Oct 2026PreprintCC BY 4.0Full textPhysical chemistry Condensed matterarXiv
From Vacuum Geometry to Solution-Phase Electronic Properties and Thermodynamics via Equivariant Neural Networks
Mathias Hilfiker, Lorenzo D’Amore, Dries Van Rompaey +4
ChemRxiv11 Oct 2026PreprintCC BY-NC-NDFull textAtomic, molecular & optical physicsChemRxiv
Learning Low Modes of the Wilson Dirac Operator with Gauge-Equivariant Networks
Minjae Choi, Hiroshi Ohno
arXiv22 Oct 2026PreprintCC BY-NC-ND 4.0Full textHigh-energy & particle physicsarXiv
NEMORA: Neural Equivariant Multipole Operators for Long-Range Atomistic Learning
Jay L. Kaplan, Samuel Varner, Rebecca Willett +1
arXiv7 Oct 2026PreprintCC BY 4.0Full textPhysical chemistry Computational physicsarXiv

Cite

BibTeX RIS

BibTeX

@misc{shen2026learning,
  title = {{Learning the self-energy through equivariant neural networks}},
  author = {Wudong Shen and Yang Zhong and Min-Ye Zhang and Xinyang Dong and Hongjun Xiang and Xinguo Ren},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.23717},
  eprint = {2610.23717},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.23717},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Learning the self-energy through equivariant neural networks
AU  - Wudong Shen
AU  - Yang Zhong
AU  - Min-Ye Zhang
AU  - Xinyang Dong
AU  - Hongjun Xiang
AU  - Xinguo Ren
PY  - 2026
DA  - 2026/10/16
PB  - arXiv
DO  - 10.48550/arxiv.2610.23717
UR  - https://arxiv.org/abs/2610.23717
AB  - The electronic self-energy determines the single-particle Green's function and, through it, quasiparticle energies and spectral functions, but computing it from first principles is expensive and it is a difficult machine-learning target: complex-valued, matrix-valued in a local-orbital basis, and dependent on a continuous frequency. We predict the dynamical correlation part of the imaginary-frequency G_0W_0 self-energy matrix of molecules from atomic species and coordinates with an E(3)-equivariant message-passing network, the rotational covariance of the orbital-space matrix being imposed by the architecture itself and the static exchange part taken from the density-functional reference. Frequency enters either as an explicit network input or through frequency-specific output heads, and the predicted matrices are continued to the real axis in a representation in which causality and a non-negative spectral density hold by construction. The frequency-conditioned model reproduces matrix elements to below 1 meV on the imaginary axis for the systems studied here, and that accuracy survives the continuation: for water clusters, including a tetramer absent from the training set, and for chemically diverse molecules from QM9, the reconstructed spectral functions follow the G_0W_0 reference peak by peak and the frontier quasiparticle levels are recovered to a few tens of meV. Learning the self-energy on the imaginary axis is therefore a practical route to real-frequency spectra: once the model is trained, and given the density-functional reference it is built on, a new geometry costs no explicit G_0W_0 calculation.
ER  -

Abstract

Related

Cite