Accelerating Quantum Chemistry Automation: Early Detection and Recovery of SCF Convergence Failure via Time Series Feature Extraction

Read full text Summarise with VIRAT
Authors: Lechen Dong, Fang Liu
Cited by: 0
References: 49

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

Self-consistent field (SCF) convergence failures bottleneck high-throughput quantum chemistry, particularly in open-shell systems. To address this, we introduce an autonomous pipeline treating the electronic structure descriptors from initial spin-unrestricted SCF iterations as a time series. Using statistical descriptors extracted from just the first ten iterations, a lightweight gradient-boosted classifier accurately predicts convergence failures. The model achieves 94.8% accuracy on 55,000 test molecules after training on only 10,000 examples. Coupling this early-warning predictor with a 𝛽-level-shift restart (𝛽-RST) heuristic, derived from Bayesian Optimization of the shift parameters, enables automated recovery. Across six diverse calculation settings, our pipeline successfully rescued 57.5% of inherently difficult-to-converge molecules while maintaining a 94.1% convergence rate for easy cases, resulting in a 460,000-iteration reduction across the 1200 calculations. Ultimately, this framework provides a highly transferable, data-efficient, and autonomous solution for largescale electronic structure calculations.

Fields of Study & Subjects

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

Publication details

Journal / source
ChemRxiv
Publication date
30 March 2026
Language
EN
Document type
Preprint