Stabilizing Reduced Order Models for Fluid Dynamics

44 concepts22 papers1 proposal

When simulating complex fluid flows governed by the Navier-Stokes or advection-diffusion equations, Full-Order Models (FOMs) like the Finite Element Method (FEM) often become computationally prohibitive. Reduced-Order Models (ROMs) solve this by projecting the high-dimensional equations onto a low-dimensional subspace constructed from a few representative "snapshots" of the system.

However, this projection introduces a fundamental mathematical hazard. In incompressible flows, velocity and pressure are coupled via a saddle-point problem that requires the approximation spaces to satisfy the inf-sup (or Ladyzhenskaya-Babuška-Brezzi) condition. Even if the underlying FOM is perfectly stable, a naive Galerkin projection onto a reduced basis can break this delicate balance, resulting in spurious pressure oscillations and numerical instabilities. Similarly, in advection-dominated regimes, the lack of proper upwinding in the reduced space triggers non-physical wiggles. To build robust ROMs, we must explicitly restore this stability.

Restoring Balance via Supremizer Enrichment

How do we guarantee that a reduced velocity space is rich enough to balance any pressure field? The most direct solution is supremizer enrichment.

To understand why this works, consider the inf-sup condition as a game: for every possible pressure field, there must be a corresponding velocity field that prevents the system from degenerating. When we compress our velocity space down to a few dominant Proper Orthogonal Decomposition (POD) modes, we inadvertently discard the specific velocity components that counteracted the pressure gradients.

As analyzed by Michael Schneier in Error Analysis of Supremizer Pressure Recovery for POD based Reduced-Order Models of the Time-Dependent Navier-Stokes Equations, supremizer stabilization fixes this by explicitly solving an auxiliary problem during the offline phase. For each pressure mode, we compute a "supremizer" function—the velocity field that maximizes the inf-sup stability condition against that pressure. By appending these supremizer functions directly into the reduced velocity basis, we mathematically force the reduced space to satisfy the discrete inf-sup condition.

This enrichment strategy is highly versatile. For instance, Gianluigi Rozza demonstrates its efficacy across diverse fluid problems, including geometric variations via unfitted meshes in A Reduced Order Cut Finite Element method for geometrically parametrized steady and unsteady Navier-Stokes problems and hierarchical model reduction in Hierarchical Model Reduction Techniques for Flow Modeling in a Parametrized Setting. By expanding the velocity space, supremizer enrichment ensures robust and accurate pressure recovery without altering the underlying structure of the fluid equations.

The Necessity of Consistent Offline-Online Stabilization

An alternative to expanding the approximation space is modifying the governing equations themselves using residual-based stabilization techniques, such as the Streamline Upwind Petrov-Galerkin (SUPG) or Pressure-Stabilized Petrov-Galerkin (PSPG) methods. In the full-order finite element community, these techniques are widely used to handle high Reynolds or Péclet numbers and enable equal-order interpolations (e.g., using identical linear elements for both velocity and pressure), as shown by Thomas M. Smith et al. and John N. Shadid et al..

However, when porting these stabilized methods to reduced-order modeling, a critical trap emerges: offline-only stabilization is insufficient.

In an offline-only approach, one generates the reduced basis snapshots using a stabilized FOM, but then projects the equations onto the reduced space using a standard, unstabilized Galerkin operator during the online stage. Gianluigi Rozza demonstrates in Reduced Basis Approximation of Parametrized Advection-Diffusion PDEs with High Péclet Number why this fails. Because the online stage employs an unstabilized operator, the underlying numerical instabilities are re-introduced; the reduced basis itself cannot carry the stabilization properties if the projection mechanism discards them.

To resolve this, stabilization must be applied consistently across both the offline and online stages. In A Streamline Upwind Petrov-Galerkin Reduced Order Method for Advection-Dominated Partial Differential Equations Under Optimal Control and Stabilized weighted reduced order methods for parametrized advection-dominated optimal control problems governed by partial differential equations with random inputs, Gianluigi Rozza shows that evaluating the stabilization terms online suppresses numerical oscillations and brings the reduced-order error close to the theoretical minimum. Furthermore, as detailed in Stabilized reduced basis methods for parametrized steady Stokes and Navier-Stokes equations, this dual offline-online stabilization can act as a direct alternative to supremizer enrichment, enforcing the inf-sup condition for equal-order spaces without the computational overhead of expanding the reduced basis size.

Alternative Paths: Variational Multiscale and Least-Squares Projections

Beyond explicit space enrichment and residual-based terms, researchers have explored structural changes to the projection framework that inherently preserve stability.

One such paradigm is the Variational Multiscale (VMS) framework. VMS splits the solution into resolvable large scales and unresolvable small scales, modeling the effect of the small scales as an intrinsic stabilization term. In A reduced order variational multiscale approach for turbulent flows, Gianluigi Rozza shows that if the VMS stabilization terms from an inf-sup stable FOM are consistently projected into the ROM, the stability properties are naturally inherited. This consistent projection can completely obviate the need for explicit pressure supremizers.

Another elegant approach shifts away from Galerkin projection entirely. Standard Galerkin projection forces the residual to be orthogonal to the reduced space, which can be poorly conditioned for stiff or highly convective systems. In Model reduction for multi-scale transport problems using model-form preserving least-squares projections with variable transformation, Cheng Huang et al. demonstrate that minimizing the residual of the governing equations in a least-squares sense inherently stabilizes the reduced-order model. By transforming the problem into an optimization that minimizes the global residual norm, the resulting system avoids the stability failures of standard projections, offering a robust mathematical foundation for highly non-linear, multi-scale transport phenomena.

Key Foundational Work

Reduced Basis Approximation of Parametrized Advection-Diffusion PDEs with High Péclet Number (2013) — contributed 6 concept elements to this direction.
Reactor Core Sub-Assembly Simulations Using a Stabilized Finite Element Method. (2011) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_supremizer_enrichment_2b0f7717,
  title = {supremizer enrichment},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/supremizer-enrichment/rdr_2b0f771776a84a5a91c32cfc1a8ec751},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_supremizer_enrichment_2b0f7717_v1,
  title = {supremizer enrichment},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/supremizer-enrichment/rdr_2b0f771776a84a5a91c32cfc1a8ec751/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake