Differentiable Schwarz-Coupled Solvers for Thin Boundary Layer Transport: Inverse Modeling via Agentic AI Optimization
physics.flu-dyn, physics.comp-ph
Submitted: 2026-10-28
Updated: 2026-10-28
License: http://creativecommons.org/licenses/by/4.0/
The gist: High-Péclet-number transport problems give rise to thin boundary layers with steep near-wall gradients whose accurate solution requires high-resolution meshes and could challenge inverse modeling.
Terminology
Abstract
High-Péclet-number transport problems give rise to thin boundary layers with steep near-wall gradients whose accurate solution requires high-resolution meshes and could challenge inverse modeling. To address this challenge, we develop an overlapping Schwarz domain-decomposition framework grounded in perturbation theory, in which an inner equation governs near-wall dynamics and an outer equation governs bulk transport, and both subproblems are discretized using the finite element method and assembled into a fully differentiable multidomain physics solver. The formulation permits coarser mesh resolutions, reducing computational cost relative to a single-domain formulation. Exploiting the differentiability of the coupled solver, we formulate inverse problems in which the governing equations and boundary conditions are imposed as hard constraints, with the Schwarz coupling enforcing interface continuity. To accelerate the resulting adjoint-based optimization, we integrate large language models (LLMs) into the inverse workflow in two distinct configurations: a standalone optimizer that proposes successive parameter updates from the adjoint-computed sensitivities and an agentic artificial intelligence (AI) approach with access to the complete solver workflow, enabling it to perform multistage optimization procedures. The agentic AI configuration is shown to be most advantageous when the spatially varying unknown is represented in a functional basis rather than a high-dimensional nodal field representation. In this case, the AI agent grows the library of candidate basis functions during optimization under a prescribed admission policy, admitting predefined families in stages and initializing their parameters. In short, our proposed framework integrates Schwarz-coupled differentiable physics with agentic AI optimization for the inverse modeling of multiscale transport problems.
Related papers
- Emergent Transfer of a Physics Foundation Model from Simulation to Laboratory Turbulence
- Early onset of secondary shear instability in Kelvin-Helmholtz braids at high Reynolds number
- Rotational covariance, continuous spectrum and non-modal growth of three-dimensional planar weak viscous shock layers
- Deep Learning-based Algebraic Reynolds Stress Closures for RANS Simulations of Turbulent Flows
- Neptuna: A Comprehensive Machine Learning Framework for Benchmarking Complex Multiphase Flows
- Size and shape of condensing steam jets and plumes