NeuroSQP: Learning-accelerated sequential quadratic programming for large-scale equality-constrained nonlinear optimization

doi:10.1016/j.neucom.2026.134568

NeuroSQP: Learning-accelerated sequential quadratic programming for large-scale equality-constrained nonlinear optimization

2026-07-22

0

PRE

AI

Mohammad Reza Eslahchi

M

Masoud Hajarian *

DOI：10.1016/j.neucom.2026.134568

Original

Original request for help

Save

Abstract

En 中文

• Introduces NeuroSQP, an SQP framework enhanced with graph neural networks. • Accelerates KKT system solves using neural preconditioning and Krylov methods. • Employs a hybrid Hessian strategy combining exact second-order and BFGS updates. • Demonstrates strong performance on CUTEst with improved scalability. • Successfully solves large-scale equality-constrained problems where others fail.

NeuroSQP: Learning-accelerated sequential quadratic programming for large-scale equality-constrained nonlinear optimization

Abstract

Journal

Organization