Mutual information prediction for strongly correlated systems

We have trained a new machine-learning (ML) model which predicts mutual information (MI) for strongly correlated systems. This is a complex quantity, which is much more difficult to predict than one-site entropies, but carries important information about the correlation structure inside electronic s...

Full description

Saved in:
Bibliographic Details
Published in:Chemical physics letters 2023-02, Vol.813 (C), p.140297, Article 140297
Main Authors: Golub, Pavlo, Antalik, Andrej, Beran, Pavel, Brabec, Jiri
Format: Article
Language:English
Subjects:
DMRG
Machine learning
Mutual information
Quantum chemistry
Strong correlation
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We have trained a new machine-learning (ML) model which predicts mutual information (MI) for strongly correlated systems. This is a complex quantity, which is much more difficult to predict than one-site entropies, but carries important information about the correlation structure inside electronic systems. In this work, we replaced the expensive density matrix renormalization group (DMRG) calculations by newly trained ML model for prediction of the mutual information. We show the performance of the model on two important tasks: (a) to determine the correlation structure and (b) to determine ordering of orbitals for accurate DMRG calculations. The results are compared with the MI obtained from accurate DMRG calculations. [Display omitted] •DMRG-based mutual information is predicted by ML model for multireference systems.•Performance is shown on correlation patterns, comparison with exact DMRG results.•ML model is used also for estimation of optimal orbital ordering in active space.
ISSN:0009-2614
1873-4448
DOI:10.1016/j.cplett.2023.140297