Super-Localized Orthogonal Decomposition Method for Heterogeneous Linear Elasticity

Camilla Belponer, José C. Garay, Peter Munch, Daniel Peterseim · 2025

Read Super-Localized Orthogonal Decomposition Method for Heterogeneous Linear Elasticity by Camilla Belponer, José C. Garay, Peter Munch, Daniel Peterseim (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-2501-05193.txt.

Introduction

Camilla Belponer and co-authors present a faster numerical method for materials whose stiffness varies at fine scales. Their 2025 paper introduces the Super-Localized Orthogonal Decomposition method for linear elasticity problems. The method handles messy microstructures without assuming any repeating pattern or clean scale gap.

Earlier decomposition methods needed generous oversampling around each patch to stay accurate. The new version uses a nonlinear superlocalization trick that makes the system much sparser, so far less oversampling is required. The team extends the approach to vector-valued problems and backs it with analysis. A scalable build inside a standard finite element library shows the method suits high-performance computing.

Catalog record

CREATORCamilla Belponer, José C. Garay, Peter Munch, Daniel Peterseim
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE67 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-05193

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Super-Localized Orthogonal Decomposition Method for Heterogeneous Linear Elasticity Thanks: The research of C. Belponer and D. Peterseim has been funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), grant DFG CA 1159/1-4 and PE 2143/1-6 “Computational multiscale methods for inverse estimation of effective properties of poroelastic tissues”. The work of J.C. Garay is part of a project that has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant agreement No. 865751 – RandomMultiScales). Camilla Belponer, José C. Garay, Peter Munch, Daniel Peterseim Abstract . We present the Super-Localized Orthogonal Decomposition (SLOD) method for the numerical homogenization of linear elasticity problems with multiscale microstructures modeled by a heterogeneous coefficient field without any periodicity or scale separation assumptions. Compared to the established Localized Orthogonal Decomposition (LOD) and its linear localization approach, SLOD achieves significantly improved sparsity properties through a nonlinear superlocalization technique, leading to computationally efficient solutions with significantly less oversampling – without compromising accuracy. We generalize the method to vector-valued problems and provide a supporting numerical analysis. We also present a scalable implementation of SLOD using the deal.II finite element library, demonstrating its feasibility for high-performance simulations. Numerical experiments illustrate the efficiency and accuracy of SLOD in addressing key computational challenges in multiscale elasticity. Key words: superlocalization, numerical homogenization, rough coefficients, multiscale method, linear elasticity AMS subject classifications. 65N12, 65N15, 65N30 1. Introduction The partial differential equations (PDEs) of linear elasticity are fundamental tools in structural analysis, describing how solid objects deform and experience stress under small loads. In this work, we consider solid materials endowed with highly heterogeneous microstructures encoded in spatially varying PDE coefficients. We specifically avoid assumptions of periodicity or scale separation, which can significantly restrict the range of applicability of classical homogenization theories. When the microstructure of the material exhibits multiple characteristic length scales, the finite element discretization of the resulting multiscale PDE often requires extremely fine meshes to capture all relevant features. Such meshes lead to prohibitively large systems of equations — especially in elasticity, where the displacement field is vector-valued and the computational burden is further amplified. Numerical homogenization methods address these challenges by designing generalized finite element spaces whose basis functions capture essential microscopic behavior while operating at a coarse, macroscopic scale. A variety of such methods have been proposed, including the Localized Orthogonal Decomposition (LOD) method [ 32 ] (see also [ 26 , 38 , 23 , 10 ] ), the Multiscale Finite Element Method (MsFEM) [ 24 , 13 ] , the Generalized Multiscale Finite Element Method (GMsFEM) [ 14 ] , the multiscale Generalized Finite Element Method (MS-GFEM) [ 4 , 5 , 28 , 27 ] , Bayesian approaches [ 34 ] , rough polyharmonic splines [ 37 ] , Gamblets [ 35 ] , and multiscale methods inspired by FETI-DP and BDDC frameworks [ 29 , 25 ] . General overviews can be found in [ 36 , 33 , 12 , 6 ] and in the review [ 3 ] . Among these techniques, the LOD approach has proven successful both theoretically and empirically for a range of PDEs beyond the prototypical Poisson problem, including heterogeneous linear elasticity [ 22 ] and more general multiphysics problems such as thermoelasticity [ 31 ] and poroelasticity [ 16 , 2 ] . In LOD, the degrees of freedom are associated with the elements (or other suitable entities such as vertices or edges) of a coarse mesh, chosen independently of the underlying fine-scale heterogeneities. Microscopic details are then incorporated through local patch problems—often called cell problems —in slightly enlarged regions around these coarse elements (oversampling patches). The optimal convergence rates of the error with respect to the coarse mesh size H typically require an oversampling region of diameter \mathcal{O}(H\,\lvert\log H\rvert) . Although the appearance of a logarithmic factor is optimal in existing constructions and common among numerical homogenization methods, the associated moderately increased communication and decreased sparsity still pose a bottleneck in large-scale applications, where the vector-valued nature of the problem further amplifies the computational cost. An improved localization strategy, named Super-Localized Orthogonal Decomposition (SLOD), was introduced in [ 20 ] for a prototype scalar diffusion problem to address precisely this issue. By relaxing the linear structure of the classical LOD construction, SLOD achieves (in practice) superexponential decay of the localization error instead of merely exponential decay. Although superexponential convergence hinges on a conjecture of spectral geometry [ 20 ] , it is provably at least as effective as LOD. More importantly, its sharper localization yields significant computational gains, reducing the diameter of the local patch problems from \mathcal{O}(H\,\lvert\log H\rvert) down to \mathcal{O}(H\sqrt{\lvert\log H\rvert}) . In turn, this improvement enhances the sparsity of the resulting system by a factor of (|\log H|)^{d/2} (where d is the dimension of the spatial domain) and accelerates both the offline assembly and the online solution phases. So far, SLOD has been explored for linear and nonlinear scalar PDEs, such as the nonlinear Schrödinger equation [ 39 ] , parametric problems [ 8 ] , stochastic homogenization [ 21 ] , Helmholtz equations [ 15 ] , convection-dominated diffusion [ 7 ] , and even spatial network models [ 19 ] . In this paper, we extend the superlocalization paradigm to vector-valued PDEs, focusing on heterogeneous linear elasticity. We show that the method retains its superior localization properties in the elasticity context, providing substantial enhancements in practical performance compared to standard LOD [ 22 ] . We also present a basic numerical analysis that demonstrates that SLOD is never worse than LOD in theory, while offering clear advantages in practice. Furthermore, we develop a new implementation of SLOD within the open-source deal.II finite element library. Our experience shows that integrating multiscale methods into modern high-performance finite element codes is feasible and beneficial for addressing the large-scale, heterogeneous elasticity problems commonly encountered as a building block for advanced models in engineering and the sciences. The remainder of this paper is organized as follows. In Section 2 we introduce the linear elasticity equations and discuss their solution challenges in the presence of multiscale microstructures. Section 3 describes the construction of an ideal numerical homogenization framework, serving as a guiding principle for subsequent sections. In Section 4 , we detail the construction of superlocalized functions employed in the definition of the SLOD method for vector-valued problems. Section 5 introduces the practical SLOD method and provides a stability and convergence analysis of it, including rigorous error estimates and condition number bounds. Finally, in Section 6 , we outline the implementation details in deal.II and present numerical experiments that confirm the efficiency and accuracy of SLOD on benchmark problems of heterogeneous linear elasticity. 2. Mathematical model problem In this section, we present the linear elasticity equations and describe the challenges that arise when solving these equations in the presence of multiscale microstructural features. Such features often necessitate exceptionally fine meshes in standard finite element schemes, motivating the development of numerical homogenization methods. An important aspect of these methods is the support locality of the basis functions defining the approximate solution spaces, which helps to capture fine-scale effects at coarse scales without excessive computational overhead. Accordingly, we will later examine rapidly

Read the complete text