Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
TT: Fachverband Tiefe Temperaturen
TT 51: Correlated Electrons: Method Development II
TT 51.5: Talk
Wednesday, March 11, 2026, 16:00–16:15, HSZ/0101
SOLAX: An Open Source Python Package for Neural Network Configuration Interaction — Pavlo Bilous2, Louis Thirion1, •Max Kroesbergen1, Paul Fadler3, and Philipp Hansmann1 — 1Friedrich-Alexander-Universität Erlangen-Nürnberg — 2Max Planck Institute for the Science of Light, Erlangen — 3Universität Bremen
We present a modular Python library, SOLAX [1], designed for configuration interaction (CI) calculations of fermionic quantum many-body systems in high dimensional Hilbert spaces. The provided classes allow convenient encoding of states and operators in second quantization. The JAX-based GPU-accelerated back-end efficiently performs the operations necessary to determine many-body eigenstates in finite-size Hilbert spaces. Along with its core functionalities, SOLAX integrates neural-network (NN) support for the CI calculation of otherwise prohibitively large expansions in Slater determinant basis sets. We show how a NN can be used in CI calculations to identify a priori unknown subsets of the most important Slater determinants and iteratively obtain high-quality approximate many-body eigenstates. Applications involve the paradigmatic Single Anderson Impurity Model in a solid-state physics context [2], as well as computation of molecular ground- [3,4] and excited states [5] in Quantum Chemistry.
[1] L.Thirion, P.Hansmann, P.Bilous, 10.21468/SciPostPhysCodeb.51,
[2] P. Bilous et al., 10.1103/PhysRevB.111.035124
[3] Y.L.A. Schmerwitz et al., 10.1021/acs.jctc.4c01479
[4] L. Thirion et al., arXiv:2510.27665
[5] G. Levi et al., arXiv:2510.26751
Keywords: configuration interaction; machine learning; neural network classifier; Python