Physics-Guided Dimension Reduction for Simulation-Free Operator Learning of Stiff Differential--Algebraic Systems

2026-04-21

AI Summary

This work addresses the challenge of high errors in neural surrogate models for stiff differential-algebraic equations (DAEs), which arise due to algebraic residuals being amplified by stiffness or reliance on costly numerical integration. The authors propose an extended Newton implicit layer that jointly enforces algebraic consistency and quasi-steady-state dimensionality reduction within a single differentiable solve. By predicting only the slow-varying states, the method accurately recovers both fast dynamics and algebraic variables while reducing output dimensionality to the slow subspace. This is the first approach to integrate physics-guided dimensionality reduction with implicit differential operator learning without simulation-based training. Leveraging the implicit function theorem, it derives gradients accounting for stiff coupling, supports compositional modeling of multi-component cascaded systems, and offers theoretical convergence guarantees. Evaluated on a 21-state power inverter DAE, it achieves a mere 1.42% error—significantly outperforming penalty methods (39.3%)—and composes two models into a 44-state system without retraining, yielding 0.72–1.16% error, zero algebraic residual, and 90% in-distribution coverage via conformal prediction.

Technology Category

• Intelligent Robots: State Estimation
• Reasoning under Uncertainty: Stochastic Optimization
• Machine Learning: Deep Neural Architectures and Foundation Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Neural surrogates for stiff differential-algebraic equations (DAEs) face two key challenges: soft-constraint methods leave algebraic residuals that stiffness amplifies into large errors, while hard-constraint methods require trajectory data from computationally expensive stiff integrators. We introduce an extended Newton implicit layer that enforces algebraic consistency and quasi-steady-state reduction within a single differentiable solve. Given slow-state predictions from a physics-informed DeepONet, the proposed layer recovers fast and algebraic states, eliminates the stiffness-amplification pathway within each time window, and reduces the output dimension to the slow states alone. Gradients derived via the implicit function theorem capture a stiffness-scaled coupling term that is absent in penalty-based approaches. Cascaded implicit layers further extend the framework to multi-component systems with provable convergence. On a grid-forming inverter DAE (21 states), the proposed method (7 outputs, 1.42 percent error) significantly outperforms penalty methods (39.3 percent), standard Newton approaches (57.0 percent), and augmented Lagrangian or feedback linearization baselines, which fail to converge. Two independently trained models compose into a 44-state system without retraining, achieving 0.72 to 1.16 percent error with zero algebraic residual. Conformal prediction further provides 90 percent coverage in-distribution and enables automatic out-of-distribution detection.

Problem

• stiff differential-algebraic equations
• neural surrogates
• algebraic residuals
• operator learning
• simulation-free learning

Innovation

• physics-guided dimension reduction
• implicit neural layers
• stiff differential-algebraic equations
• operator learning
• quasi-steady-state reduction