Reduced Basis Approach for Convection-Diffusion Equations with Non-Linear Boundary Reaction Conditions
S. Matera 1, C. Merdon 2, D. Runge 2 · 2025
Read Reduced Basis Approach for Convection-Diffusion Equations with Non-Linear Boundary Reaction Conditions by S. Matera 1, C. Merdon 2, D. Runge 2 (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-2510-24632.txt.
Introduction
Sebastian Matera, Christian Merdon, and Daniel Runge seek a fast way to solve drift and spread tasks with hard edge rules. Such tasks show up in work on mixed phase chem acts, where the hard part sits just at the edge. Their plan builds a small set of base maps first.
Those base maps come from the main flow rule and do not depend on the hard edge part. So the full hard task shrinks to a much smaller one at the edge alone. The same base set can then be reused for new edge rules. They show the plan on a test case with a fuel gas change at a surface.
Catalog record
| CREATOR | S. Matera 1, C. Merdon 2, D. Runge 2 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 18 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-24632
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Reduced Basis Approach for Convection-Diffusion Equations with Non-Linear Boundary Reaction Conditions S. Matera 1, C. Merdon 2, D. Runge 2 Abstract This paper aims at an efficient strategy to solve drift-diffusion problems with non-linear boundary conditions as they appear, e.g., in heterogeneous catalysis. Since the non-linearity only involves the degrees of freedom along (a part of) the boundary, a reduced basis ansatz is suggested that computes discrete Green’s-like functions for the present drift-diffusion operator such that the global non-linear problem reduces to a smaller non-linear problem for a boundary method. The computed basis functions are completely independent of the non-linearities. Thus, they can be reused for problems with the same differential operator and geometry. Corresponding scenarios might be inverse problems in heterogeneous catalysis but also modeling the effect of different catalysts in the same reaction chamber. The strategy is explained for a mass-conservative finite volume method and demonstrated on a simple numerical example for catalytic CO oxidation. † † tocauthor: Sebastian Matera, Christian Merdon, Daniel Runge † † institute: Fritz Haber Institute of the Max Planck Society Faradayweg 4-6, 14195 Berlin, Germany and Weierstrass Institute for Applied Analysis and Stochastics Anton-Wilhelm-Amo-Straße 39, 10117 Berlin, Germany This preprint has not undergone peer review or any post-submission improvements or corrections. The Version of Record of this contribution is published in Finite Volumes for Complex Applications X—Volume 1, Elliptic and Parabolic Problems, and is available online at https://doi.org/10.1007/978-3-031-40864-9˙28 . 1 Introduction Reduced basis approaches have gained popularity for the solution of partial differential equations (PDE), especially to address parameter dependencies Quarteroni2015 ; BennerMRA2017 . The idea is to employ only a few simulations with a general purpose discretization and high resolution to determine a (small) set of problem-specific basis functions (offline phase). This then allows for a fast solution of the PDE using the previously computed basis functions (online phase). We present a new reduced basis approach for linear convection-diffusion equations with highly non-linear flux boundary conditions. Such models are often good approximations in problems which involve surface chemistry, e.g., heterogeneous catalysis or electrochemistry MBHZGLR2015 ; fuhrmann_role_2011 . Our approach decomposes the problem into a set of linear problems to obtain the reduced basis and a non-linear problem which only depends on the degrees of freedom on the boundary. Being essentially a discrete representation of a type of Green’s function of the linear operator, the reduced basis is independent of the non-linearity and can be reused for very different scenarios, e.g. inverse problems for parametrizing the non-linear boundary condition, which is a common task in surface chemistry applications. The connection to a type of Green’s functions renders the approach closely related to boundary integral methods like trauble_nonlinear_2007 , but without explicit knowledge of the former. In principle, the problem can be discretized by any method, e.g., finite volumes or finite element methods. Here, we demonstrate the approach with Voronoi finite volume methods, which allow for advantageous structural properties like conservation of mass or the non-negativity of Y FUHRMANN2001201 . An algebraic formulation reveals that the reduced basis solution agrees with the solution of the fully coupled problem discretized with the same underlying method. The rest of this paper is structured as follows. Section 2 introduces the model problem and lays out its finite volume discretization. Section 3 concerns the design and implementation of the set of reduced basis functions. Section 4 demonstrates the approach in one model application. Finally, Section 5 gives an outlook to further aspects and target applications. 2 Model problem and FV discretisation For a given domain \Omega , a (divergence-free) velocity field \bm{v} , a positive diffusion coefficient D\in\mathbb{R} , and a right-hand side f\in L^{2}(\Omega) , the model problem seeks some function Y\in H^{1}(\Omega) such that \displaystyle\mathrm{div}(Y\bm{v}-D\nabla Y)=f. (1) On the boundary \Gamma:=\partial\Omega=\Gamma_{\mathrm{in}}\cup\Gamma_{\mathrm{out}}\cup\Gamma_{\mathrm{nl}}\cup\Gamma_{\mathrm{rest}} , various boundary conditions may apply. Here, we assume some inlet condition \displaystyle Y=Y_{\text{in}}\quad\text{along}\quad\Gamma_{\mathrm{in}}, an outflow condition \displaystyle\bm{n}\cdot(Y\bm{v}-D\nabla Y) \displaystyle=Y\bm{v}\cdot\bm{n}\quad\text{along}\quad\Gamma_{\mathrm{out}}, and a non-linear boundary condition \displaystyle\bm{n}\cdot(Y\bm{v}-D\nabla Y)=R(Y|_{\Gamma_{\mathrm{nl}}})\quad\text{along}\quad\Gamma_{\mathrm{nl}} (2) for some given functional R . On the remaining part of the boundary \Gamma_{\mathrm{rest}} we assume homogeneous Neumann boundary conditions for simplicity. But this approach can easily be extended to linear Robin boundary conditions. Here, we consider a finite volume discretization of the model problem based on some boundary-conforming Delaunay triangulation \mathcal{T} , which computes a piecewise constant approximation Y_{h}\in P_{0}(\mathcal{K}) to Y with respect to the set of open, convex Voronoi cells \mathcal{K} . Each cell K\in\mathcal{K} has some associated collocation point \bm{x}_{K}\in\overline{K} . The subset of cells at the boundary {\Gamma_{\mathrm{nl}}} are denoted by \mathcal{K}_{\mathrm{nl}} and their associated collocation points are located on the boundary \Gamma_{\mathrm{nl}} . Details on the implementation can be found, e.g., in the documentation of the Julia package VoronoiFVM.jl fuhrmann-voronoifvm which was also used as a basis for the implementation of the reduced bases. Note, that we employ an exponential fitting flux discretization that ensures desirable structural properties like non-negativity and a maximum principle for Y under certain conditions, e.g., if R=0 and \bm{v} is divergence-free FUHRMANN2001201 . 3 Reduced basis approach This section describes the main idea to speed up the computation of the solution of the model problem with the help of a reduced basis related to the boundary degrees of freedom of the non-linear boundary \mathcal{K}_{\mathrm{nl}} . For this, observe that the discrete solution Y_{h} can be decomposed into \displaystyle Y_{h}=Y_{0}+Y_{\mathrm{nl}}:=Y_{0}+\sum_{K\in\mathcal{K}_{\mathrm{nl}}}\alpha_{K}Y_{K} where Y_{0} solves the discretized linear sub-problem for R=0 and each Y_{K} solves the discretized fully linear problem \displaystyle\mathrm{div}(Y_{K}\bm{v}-D\nabla Y_{K}) \displaystyle=0 \displaystyle\bm{n}\cdot(Y_{K}\bm{v}-D\nabla Y_{K}) \displaystyle=\chi_{K}\quad\text{along}\quad\Gamma_{\mathrm{nl}} (3) and homogeneous boundary conditions on the remaining boundary \Gamma\setminus\Gamma_{\mathrm{nl}} . Note, that Y_{K} can be interpreted as a type of discrete Green’s function of the drift-diffusion operator for its corressponding part \partial K\subset\Gamma_{\mathrm{nl}} of the boundary. To further investigate this on the algebraic level, let x_{\mathrm{nl}} , x_{0} and x_{K} denote the coefficient vectors of the Y_{\mathrm{nl}} , Y_{0} and Y_{K} parts, respectively. They are given by a solution of the linear systems of equations \displaystyle Ax_{0} \displaystyle=b_{0} (4) \displaystyle Ax_{K} \displaystyle=b_{K}, (5) where A is the finite-volume discretization of the drift-diffusion operator, b_{0} encodes all linear ( Y -independent) boundary and right-hand side data, and b_{K} encodes the boundary condition ( 3 ). Note here, that the matrix A is the same in all computations and it is therefore straightforward to solve for x_{0} and all x_{K} efficiently and in parallel. Also note, that for a constant inlet concentration and f\equiv 0 , it holds Y_{0}\equiv Y_{\text{in}} which allows to avoid the computation of Y_{0} in that case. To determine the coefficients x_{\mathrm{nl}} for Y_{\mathrm{nl}}=\sum_{K\in\mathcal{K}_{\mathrm{nl}}}\alpha_{K}Y_{K} , it is required to solve the non-linear system \displaystyle A(x_{0}+x_{\mathrm{nl}})=b_{0}+b_{\mathrm{nl}}\left(x_{0}+x_{\mathrm{nl}}\right)\quad\Leftrightarrow\quad Ax_{\mathrm{nl}}=b_{\mathrm{nl}}\left(x_{0}+x_{\mathrm{nl}}\right) (6) where b_{\mathrm{nl}}(x) encodes the finite volume discretisation of the (non-linear) catalytic boundary data. Inserting the decomposition into the reduced boundary basis x_{\mathrm{nl}}=\sum_{K\in\mathcal{K}_{\mathrm{nl}}}\alpha_{K}x_{K} , this is equivalent to seeking \alpha_{K} such that \displaystyle\sum_{K\in\mathcal{K}_{\mathrm{nl}}}\alpha_{K}b_{K}=b_{\mathrm{nl}}\left(x_{0}+\sum_{K\in\mathcal{K}_{\mathrm{nl}}}\alpha_{K}x_{K}\right). In a finite volume method that approximates the boundary integrals in the assembly of b_{K} and b_{\mathrm{nl}}(x) by a quadrature rule evaluating in the collocation point, the non-linear system to determine the coefficients \alpha_{K} can be rewritten into \displaystyle\alpha_{L}=R\left(Y_{0}(\bm{x}_{L})+\sum_{K\in\mathcal{K}_{\mathrm{nl}}}\alpha_{K}Y_{K}\left(\bm{x}_{L}\right)\right)\quad\text{for all }L\in\mathcal{K}_{\mathrm{nl}}. (7) Here, \bm{x}_{L}\in\Gamma_{\mathrm{nl}} denotes the collocation point of the cell L\in\mathcal{K}_{\mathrm{nl}} . Note, that this is a much smaller system to solve than the global system ( 6 ). Remark