ROM for Viscous, Incompressible Flow in Polygons - exponential n-width bounds and convergence rate

Francesco Romor, Federico Pichi, Giovanni Stabile, Gianluigi Rozza, Christoph SchwabarXiv 2025 · DOI · arXiv

Introduction

Simulating fluid flow via the Navier-Stokes Equations (NSE) is a cornerstone of modern engineering, yet it remains computationally intensive. High-fidelity simulations often require millions of degrees of freedom to resolve complex flow patterns, making real-time control or large-scale optimization nearly impossible. Reduced Order Models (ROMs) offer a solution by projecting these high-dimensional problems onto a significantly smaller linear subspace.

While ROMs have shown remarkable empirical success, a rigorous theoretical understanding of their efficiency in complex geometries has been lacking. Specifically, the presence of sharp corners in polygonal domains introduces mathematical singularities that typically degrade convergence rates. This paper provides a formal theoretical foundation, proving that for stationary, incompressible NSE in 2D polygons, the solution set can be approximated with exponential accuracy relative to the number of basis functions.

The Problem: Kolmogorov n-width and Singularities

The efficiency of a ROM is fundamentally limited by the Kolmogorov n-width, a metric representing the best possible error achievable using an $n$-dimensional linear space. In simple, smooth domains, exponential convergence of the n-width is well-documented. However, real-world engineering geometries are rarely smooth; they involve polygons with corners that create "singularities"—points where the solution's derivatives may become unbounded.

In standard numerical methods, these singularities often force a transition from fast exponential convergence to slow algebraic convergence. The central question addressed by this research is: Can we maintain exponential accuracy in the presence of corner singularities?

Theoretical Framework and Methodology

The authors bridge the gap between functional analysis and practical CFD by combining several advanced mathematical techniques:

1. Corner-Weighted Analytic Regularity

The study leverages the concept of weighted Sobolev spaces. While solutions near corners are technically singular, they belong to specific weighted spaces where they remain "analytic" (expressible as power series). This regularity is the key to unlocking higher-order convergence rates.

2. hp-Finite Element Method (hp-FEM) Theory

The proof utilizes the theoretical existence of $hp$-FEM approximations. $hp$-FEM employs variable mesh sizes and varying polynomial degrees to resolve singularities. By proving that an $hp$-approximation exists that converges exponentially, the authors establish an upper bound on the Kolmogorov $n$-width of the solution manifold.

3. The Small Data Hypothesis

To maintain mathematical tractability and ensure solution uniqueness, the study operates under a "small data" regime. This assumption effectively limits the analysis to stable, laminar flows, avoiding the complexities of bifurcations or fully developed turbulence.

4. POD-Galerkin ROM

For numerical validation, the authors employ the Proper Orthogonal Decomposition (POD). This method extracts optimal basis functions from "snapshots" of high-fidelity simulations. These functions are then used in a Galerkin projection to solve the reduced system, allowing the researchers to compare empirical performance against their theoretical bounds.

Key Insights: The $n^{1/3}$ Convergence Rate

The core theoretical contribution of the paper is the derivation of a specific decay rate: $C \exp(-bn^{1/3})$. This $n^{1/3}$ exponent is a distinct signature of $hp$-approximation in 2D domains with corners.

Crucially, the research demonstrates that corner singularities are not "dealbreakers" for ROM efficiency. If the reduced spaces are chosen correctly—as POD tends to do automatically—the model can recover exponential efficiency even when the underlying geometry is non-smooth. Furthermore, these bounds are shown to be robust across various boundary conditions, including no-slip, slip, and open (outflow) conditions.

Numerical Results and Validation

The theory was tested using two classic computational fluid dynamics benchmarks:

  1. Backward-Facing Step: A channel with a sudden expansion that creates a recirculation zone, testing the model's ability to capture flow separation.
  2. Flow Past a Square: A 2D representation of a cube in a channel, evaluating the resolution of internal boundaries and wake patterns.

In both cases, the POD-based ROMs exhibited the predicted exponential decay. This trend remained consistent regardless of changes in physical parameters (such as viscosity $\nu$) or the resolution of the underlying high-fidelity mesh, confirming that the reduced-order logic is independent of the grid density once the physics are sufficiently resolved.

Limitations and Future Directions

While this work provides a vital theoretical pillar for ROMs, several avenues for future research remain:

  • Reynolds Number Constraints: The "small data" requirement limits the current proof to low-to-moderate Reynolds numbers. Extending this to high-Reynolds-number turbulent flows remains a significant challenge.
  • Dimensionality: The current analysis is strictly 2D. Moving to 3D polyhedra introduces more complex edge and vertex singularities that require further study.
  • Non-Newtonian Fluids: The current theory assumes a linear stress-strain relationship. While the authors anticipate similar patterns in non-Newtonian fluids, formal proofs are yet to be established.
  • Scientific Machine Learning (SciML): The authors suggest that these linear ROM results can serve as a benchmark for non-linear reduction methods, such as autoencoders, which may offer even tighter bounds on solution manifolds.

Conclusion

This paper establishes the first formal proof of exponential Kolmogorov $n$-width bounds for the stationary Navier-Stokes equations in polygonal domains. By proving that the $n^{1/3}$ convergence rate is achievable, the authors provide a rigorous justification for the use of ROMs in complex, real-world geometries, ensuring that the efficiency of these models is backed by solid mathematical theory.

Summary updated

Abstract

We demonstrate exponential convergence of Reduced Order Model (ROM) approximations for mixed boundary value problems of the stationary, incompressible Navier-Stokes equations in plane, polygonal domains $Ω$. Admissible boundary conditions comprise mixed BCs, no-slip, slip and open boundary conditions, subject to corner-weighted analytic boundary data and volume forcing. The small data hypothesis is assumed to ensure existence of a unique weak solution in the sense of Leray-Hopf. Recent results on corner-weighted, analytic regularity of velocity and pressure fields in $Ω$, imply exponential convergence rates of so-called mixed $hp$-Finite Element Methods in $H^1(Ω)^2\times L^2(Ω)$ on sequences of geometric partitions of $Ω$, with corner-refinement. Based on these exponential convergence rate bounds, we infer exponential bounds for the Kolmogorov $n$-widths of solution sets for analytic forcing and boundary data. This implies corresponding exponential convergence rates of POD Galerkin methods that are based on truth solutions which are obtained offline from low-order, divergence stable mixed Finite Element discretizations. Numerical experiments confirm the exponential rates and the theoretical results.
BibTeX
@misc{romor2025rom,
  author = {Francesco Romor and Federico Pichi and Giovanni Stabile and Gianluigi Rozza and Christoph Schwab},
  title = {{ROM for Viscous, Incompressible Flow in Polygons - exponential n-width bounds and convergence rate}},
  year = {2025},
  eprint = {2512.22567},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.22567},
}
Figure