Reduced basis solvers for unfitted methods on parameterized domains
Nicholas Mueller †, Santiago Badia †, Yiran Zhao ¶ · 2025
Read Reduced basis solvers for unfitted methods on parameterized domains by Nicholas Mueller †, Santiago Badia †, Yiran Zhao ¶ (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-15320.txt.
Introduction
Nicholas Mueller, Santiago Badia, and Yiran Zhao build faster solvers for physics problems posed on shapes that change with design parameters. Their 2025 paper combines unfitted finite element methods, which let the grid ignore exact boundaries, with reduced basis methods that compress many past solutions into a small set. The challenge is fitting changing shapes into fixed-size solution snapshots.
A deformation step maps each new shape back to a reference shape, solving that mismatch. Local dictionaries of reduced spaces keep the approximation sharp, and a matrix sampling trick cuts evaluation cost. The framework extends to pressure-flow problems without losing stability. Tests on heat, elasticity, and flow equations showed speed with accuracy.
Catalog record
| CREATOR | Nicholas Mueller †, Santiago Badia †, Yiran Zhao ¶ |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 93 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-15320
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Reduced basis solvers for unfitted methods on parameterized domains Nicholas Mueller †, Santiago Badia †, Yiran Zhao ¶ Abstract . In this paper, we present a unified framework for reduced basis approximations of parametrized partial differential equations defined on parameter-dependent domains. Our approach combines unfitted finite element methods with both classical and tensor-based reduced basis techniques – particularly the tensor-train reduced basis method – to enable efficient and accurate model reduction on general geometries. To address the challenge of reconciling geometric variability with fixed-dimensional snapshot representations, we adopt a deformation-based strategy that maps a reference configuration to each parameterized domain. Furthermore, we introduce a localization procedure to construct dictionaries of reduced subspaces and hyper-reduction approximations, which are obtained via matrix discrete empirical interpolation in our work. We extend the proposed framework to saddle-point problems by adapting the supremizer enrichment strategy to unfitted methods and deformed configurations, demonstrating that the supremizer operator can be defined on the reference configuration without loss of stability. Numerical experiments on two- and three-dimensional problems – including Poisson, linear elasticity, incompressible Stokes and Navier-Stokes equations – demonstrate the flexibility, accuracy and efficiency of the proposed methodology. 1. Introduction Projection-based rom are advanced numerical techniques designed to approximate parametric hf ( hf ) models, which typically involve finely resolved spatio-temporal discretizations of pde . These methods aim to capture the hf parameter-to-solution manifold within a carefully chosen vector subspace. The process generally consists of a computationally intensive offline phase, during which the subspace is constructed and a (Petrov-)Galerkin projection of the hf equations onto this subspace is performed. This stage often includes the hyper-reduction of nonaffinely parameterized hf quantities, such as residuals and Jacobians. Once the reduced model is assembled, a subsequent online phase enables the rapid evaluation of accurate solutions for new parameter values, at a cost largely independent of the full-order problem size. The rb ( rb ) method is one of the most prominent projection-based rom , with a broad range of applications, including unsteady [ 57 , 39 ] , nonlinear [ 53 , 18 ] , and saddle-point problems [ 7 , 42 ] . It constructs the projection subspace using a dictionary of high-fidelity solutions – commonly referred to as snapshots. For reducible problems [ 53 ] , i.e., those whose parameter-to-solution maps exhibit low Kolmogorov N -width [ 59 ] , the reduced space can be of much smaller dimension than the hf space, while retaining high accuracy. Subspace extraction techniques include direct compression methods [ 32 , 33 ] and greedy algorithms [ 52 , 51 ] . Direct approaches benefit from a priori estimates, such as those derived from the Schmidt-Eckart-Young Theorem [ 21 , 37 ] , which underpins the widely used tpod ( tpod ) method [ 60 ] . Greedy procedures, in contrast, iteratively build the reduced basis using a posteriori error estimators, though often at a higher computational cost. More recently, tensor-based approaches such as the ttrb ( ttrb ) method [ 36 , 35 , 40 ] have been proposed, in which snapshots are treated as multidimensional arrays and compressed along Cartesian axes using ttsvd ( ttsvd ) or ttcross ( ttcross ) [ 46 , 24 , 45 , 10 , 17 , 20 ] . These methods offer cheaper offline costs, but in the existing literature their applicability has been limited to problems posed on tensor-product grids – a restriction that the present work overcomes. A multi-purpose rom framework implementing both classical approaches as well as the more novel ttrb on arbitrarily parameterized domains can be effectively built upon unfitted (or embedded-boundary) fe ( fe ) methods [ 26 , 13 , 22 , 9 ] . These techniques allow for the discretization of pde on complex geometries without requiring body-fitted meshes, which are often challenging to generate. Instead, a background Cartesian mesh is used to define the fe space, while the geometry is specified analytically. Although early unfitted approaches suffered from ill-conditioning due to small cut cells [ 19 ] , more recent developments such as ghost penalty methods [ 13 , 14 ] and the aggregated fe method [ 4 ] have addressed these issues, achieving mesh-independent condition numbers, controlled penalty parameters, and optimal convergence rates in both mesh size and polynomial order. Despite the advantages offered by unfitted fe methods, constructing rom on parameterized geometries remains a challenging task. The first major difficulty lies in reconciling geometric variability with the requirement that all snapshots reside in a fixed-dimensional vector space. Two primary strategies have been developed to address this issue. The first, known as the extension-based approach, involves computing each snapshot on its parameterized domain and subsequently extending it onto the background mesh. This method is widely adopted in embedded-domain frameworks due to its conceptual simplicity and compatibility with structured discretizations [ 44 , 6 , 29 , 62 , 31 , 30 , 28 ] . However, it often requires complex remeshing and may introduce artifacts or degrade accuracy due to the extension process itself [ 28 ] . A more accurate alternative – though geometrically less flexible [ 16 ] – is the deformation-based approach. In this setting, a fixed reference configuration is employed, and each parameterized domain is mapped onto it through a smooth deformation map. The weak form of the pde is then integrated over the reference mesh via pull-back operations. This avoids remeshing and artificial extensions altogether, preserving snapshot fidelity. However, it requires solving additional high-fidelity systems to compute the deformation maps. Common choices include harmonic [ 5 ] , biharmonic [ 27 ] , and elastic deformations [ 56 , 55 , 58 ] , many of which stem from ale ( ale ) formulations in fluid-structure interaction [ 61 , 8 ] . In this work, we adopt the deformation-based approach as our default strategy; for ttrb approximations, we supplement it with an extension-based procedure to ensure that snapshot values are defined at every index of the background Cartesian grid. A second major challenge arises from the limited reducibility of problems posed on parameterized domains. To mitigate this, we employ a localization strategy that enhances the compressibility of the reduced model. The use of local reduced subspaces has recently gained attention as an effective means to achieve meaningful dimensionality reduction [ 1 , 48 , 47 ] . Among the few works incorporating this idea within unfitted discretization frameworks, [ 15 ] constructs local subspaces for both projection and hyper-reduction. Their approach relies on traditional tpod -based bases and local hyper-reduction via rbf ( rbf ) interpolation [ 50 ] . By contrast, we explore both tpod and ttrb subspaces in conjunction with mdeim ( mdeim )-based hyper-reduction. While the latter is more computationally expensive, we find it substantially more accurate than rbf -based methods, making it better suited for the high-fidelity requirements of our applications. Another key distinction from [ 15 ] lies in our reliance on deformation-based rather than extension-based snapshot representations, which, in our experience, results in improved compressibility and overall reducibility of the problem. In summary, the main contributions of the work are the following. • We develop an interface that combines an unfitted fe method with