DeePaTB: A Deep Learning-Powered Semi-Empirical Quantum Mechanical Method

Read full text Summarise with VIRAT
Authors: Jin Xiao, Tong Zhu
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Data-driven methods can reproduce the accuracy of high-level quantum methods with reduced computational costs by learning features from low-level electronic structure calculations. We introduce the Deep Atomic Density-Based Tight-Binding (DeePaTB) framework, a novel machine-learning semiempirical quantum mechanical (MLSQM) framework built on our atomic density-based tight-binding (aTB) method. DeePaTB generates the electronic structure feature “eigenvalue of the local density matrix” computed using Amesp software as inputs to a deep model that predicts energies and related properties. Across diverse chemical systems, DeePaTB attains density functional theory (DFT)-level accuracy while retaining SQM computational effi ciency. This is applicable to closed-shell and open-shell species and systems under external electric fi elds, demonstrating the strong transferability of DeePaTB. Together, these results establish DeePaTB as a general, scalable approach that eff ectively balances accuracy and efficiency for large and complex quantum chemical computations.

Fields of Study & Subjects

Machine Learning in Materials Science Advanced Chemical Physics Studies Computational Drug Discovery Methods

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN
Document type
Dataset