Minimum residual discretization of a semilinear elliptic problem
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.
Ada: We're looking at the paper: "Minimum residual discretization of a semilinear elliptic problem".
Emmy: Here's the gist of the paper. We propose a least-squares penalization as a means to extend the discontinuous Petrov–Galerkin (DPG) method with optimal test functions to a class of semilinear elliptic problems <ref:2603.29863#pg2> The nonlinear contributions are replaced with independent unknowns so that standard DPG techniques apply to the then linear problem with non-trivial kernel <ref:2603.29863#pg15> The nonlinear relations are added as least-squares constraints <ref:2603.29863#pg16> Assuming solvability of the semilinear problem and an Aubin–Nitsche-type approximation property for the primal variable, we prove a Céa estimate for the approximation error in canonical norms Numerical results with uniform and adaptively refined meshes illustrate the performance of the scheme Model Problem Formulation The model problem is given by finding a solution u in H10(omega) to −div(κ∇u + ρ(u)β) + γ(u) = f in omega <ref:2603.29863#pg4> The problem is motivated by the Richards equation <ref:2603.29863#pg5> The right-hand side function f can be zero by a change of the nonlinearity γ For simplicity, we consider a right-hand side function Furthermore, we assume a homogeneous Dirichlet condition for ease of presentation Minimum Residual Formulation The problem is transformed into the minimum residual formulation by representing both nonlinear terms by independent variables, thus giving rise to a linear problem, and add the nonlinear relations as constraints This approach is similar to the approach in
9: for strongly monotone operators, but with least-squares constraints similar to the least-squares DPG-BEM coupling in
30: Globally, it is a finite-element minimum residual method that minimizes the linear residual in a discretized dual test space and nonlinear residuals in L2 Discretization Scheme To discretize minimization problem (3), we consider finite-dimensional subspaces Vh ⊂ V and Uh ⊂ U The minimum residual discretization of (1) reads uh = (uh,̂σh, qh, rh) = arg min wh∈Uh (8) where this is a combination of a DPG method for the linear part of the problem with finite-element least squares for the nonlinear constraints Error Estimation and Adaptivity The quasi-optimal error estimate is proven under the assumption that uh is close to u in the sense that∥u − uh∥ ≤ c(h)∥∇(u − uh)∥ with 0 < c(h) → 0 as h → 0 The global residual Res (17) consists of local terms assigned to elements, in particular Buh − L 2V∗h = ∑T∈T res 2Vh(T)∗ These indicators Res(T) allow to perform adaptive mesh refinements Numerical Experiments Numerical experiments in two space dimensions are presented, and they confirm the expected convergence order O(N−1/2) of both the global residual Res and the error u−uhU The results for adaptive mesh refinements from Figure 3 indicate re-established optimal convergence orders of O(N−1/2) and u−uh = O(N−1), again confirming validity of the Aubin–Nitsche assumption in Theorem 3 References
1: F.
Ada: Bertrand, D.
Emmy: Boffi, and H.
Ada: Schneider, Discontinuous Petrov-Galerkin approximation of eigenvalue problems, Comput.
Title and authors: Ada: Now let's talk about what this paper actually claims in "Minimum residual discretization of a semilinear elliptic problem". They propose a specific way to handle those tricky nonlinear parts.
Emmy: They are essentially doing two main things: first, they replace the nonlinear terms with independent variables to turn it into a linear problem with some extra stuff, and second, they add the original nonlinear relations back in as least-squares constraints.
Kolya: It seems like this whole setup is inspired by approaches used for strongly monotone operators, but they've added these least-squares constraints which are similar to the DPG-BEM coupling you see in some other work.
Ada: That's right. The goal is to apply standard DPG techniques to that now linear problem, even though it has a non-trivial kernel, and then handle the nonlinearity through those constraints.
Emmy: The core idea is turning the nonlinear relations into constraints rather than trying to solve them directly in one go with the DPG framework.
Kolya: So if we're thinking about what this means for solving these problems, it's about finding a robust way to use existing linear solvers while still respecting the original nonlinear physics through those constraints.
Ada: Right. And assuming the semilinear problem is solvable and that the primal variable has that Aubin-Nitsche type approximation property, they can prove a Céa estimate for the approximation error in canonical norms.
Emmy: So they aren't just proposing a new discretization; they are providing a theoretical guarantee on how accurate their numerical solution is supposed to be, provided those assumptions hold true.
Kolya: That theoretical guarantee is important because it gives us confidence that as we refine our meshes, we're actually getting closer to the real solution in a predictable way.
The paper's summary: Ada: Moving on to the summary of "Minimum residual discretization of a semilinear elliptic problem", they are looking at a specific model problem.
Emmy: The model problem is finding a solution u in H10(Omega) to the equation where you have negative divergence of kappa times the gradient of u plus rho times u beta, plus gamma times u equals f in Omega.
Kolya: That equation comes from the Richards equation, which describes how liquid flows through porous media in unsaturated soil. It's a nonlinear degenerate parabolic type problem.
Ada: The right-hand side function f can be zero by changing the nonlinearity gamma, so they focus on a right-hand side function for simplicity, and they assume homogeneous Dirichlet conditions to make it easier to present the problem.
Emmy: They also specify that kappa is symmetric and uniformly positive definite, and rho and gamma are Lipschitz continuous functions.
Kolya: So you have this nonlinear equation that models water flow, they simplify it by making the right side zero or just setting up a simple case with a right-hand side function f.
Ada: And importantly, they state that for simplicity of manufacturing model solutions, they consider a right-hand side function. Then they assume homogeneous Dirichlet conditions for ease of presentation.
Emmy: So it sets the stage by defining exactly what kind of elliptic problem they are tackling, focusing on the structure and regularity of those functions involved.
Kolya: And then the next section is where they introduce all the required operators and spaces, which is just setting up all the mathematical machinery you need before you can even start discretizing.
The paper's improvements: Ada: So what are the actual improvements they suggest in this paper? It's about how they handle those nonlinear terms.
Emmy: The main improvement is this minimum residual formulation where they represent both nonlinear terms by independent variables, which leads to that linear problem with a non-trivial kernel we talked about before.
Kolya: And then the second part of the improvement is adding the nonlinear relations as least-squares constraints instead of trying to solve them directly in one step.
Ada: So that's it. The transformation into a linear problem, followed by adding those constraints as least-squares terms, is what they are proposing to make this method work for semilinear problems.
Emmy: This approach allows standard DPG techniques to be used on the then linear part of the problem, which is a big structural advantage because it opens up existing tools.
Kolya: So in plain terms, they're not inventing a whole new nonlinear solver; they are cleverly reformulating the problem so that we can use established linear solvers on a related system and handle the nonlinearity through constraints.
Ada: And assuming solvability of the semilinear problem and that Aubin-Nitsche type approximation property for the primal variable, this gives them their key theoretical tool: proving a Céa estimate for approximation error in canonical norms.
Emmy: This Céa estimate is what connects their discretization to the actual solution accuracy we care about, showing that the numerical solution will converge correctly if those initial assumptions hold.
Conclusion: Ada: So to wrap up on "Minimum residual discretization of a semilinear elliptic problem", they show that this approach works for these types of problems.
Emmy: They confirm this through their numerical experiments, showing performance with both uniform and adaptively refined meshes in two space dimensions.
Kolya: And those numerical results confirm the expected convergence order of O(N to the minus one half) for both the global residual and the error in canonical norms.
Ada: But more importantly, they show that with adaptive mesh refinements, they re-establish optimal convergence orders of O(N to the minus one half) and for u-uh it's O(N to the minus one).
Emmy: So if you only listen to this, what you get is a picture of how this method handles these problems on the physical model they defined, showing that it performs as expected in practice when meshes are refined.
Kolya: And from a probabilistic viewpoint, those convergence orders suggest that the tail behavior of the error is governed by something specific related to that Aubin-Nitsche assumption.
Ada: So for someone who only listens to this, this paper shows a practical way to take complex nonlinear elliptic problems and apply existing DPG methods by reformulating them into a linear problem with constraints.
Emmy: And the "Minimum residual discretization of a semilinear elliptic problem" is the title, and it sets up a new way to analyze approximation errors for these systems.
Carlos García Vera, Norbert Heuer, §Dirk Praetorius
Departamento de Ciencias Exactas, Universidad de Los Lagos · Facultad de Matemáticas, Pontificia Universidad Catílica de Chile · Institut für Analysis und Scientific Computing, Technische UniversiƟe Wien
math.NA, cs.NA
Submitted: 2026-03-31
Updated: 2026-10-29
Comments: Updated and extended numerical results, addition of a proof of the Aubin-Nitsche assumption
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 79/100
The gist: We propose a least-squares penalization as a means to extend the discontinuous Petrov–Galerkin (DPG) method with optimal test functions to a class of semilinear elliptic problems…
Key concepts
- Least-Squares Penalization
- This technique handles nonlinear parts of the equation by treating them as independent variables and adding constraints that minimize the squared difference between the actual nonlinear relations and their approximation. This converts a difficult nonlinear problem into a more manageable linear problem suitable for standard numerical methods.
- Minimum Residual Formulation
- The problem is reformulated so that it seeks to minimize the residual of the governing equations in both a discretized dual test space and the $L^2$ space for nonlinear residuals. This approach is similar to methods used for strongly monotone operators and provides a robust framework for discretization.
- Discontinuous Petrov–Galerkin (DPG)
- DPG is a numerical method that uses discontinuous test functions to approximate solutions in Sobolev spaces. It is particularly useful for problems with discontinuities or complex geometries, allowing the use of non-conforming elements and providing stability in certain elliptic problems.
- Aubin–Nitsche Approximation Property
- This assumption relates the error of a solution to the error of its gradient. Assuming this property allows the authors to prove quasi-optimal convergence rates for their numerical scheme, ensuring that as the mesh is refined, the solution accuracy improves predictably.
Terminology
Summary
We propose a least-squares penalization as a means to extend the discontinuous Petrov–Galerkin (DPG) method with optimal test functions to a class of semilinear elliptic problems <ref:2603. p. 029863#pg2> The nonlinear contributions are replaced with independent unknowns so that standard DPG techniques apply to the then linear problem with non-trivial kernel <ref:2603. p. 029863#pg15> The nonlinear relations are added as least-squares constraints <ref:2603. p. 029863#pg16> Assuming solvability of the semilinear problem and an Aubin–Nitsche-type approximation property for the primal variable, we prove a Céa estimate for the approximation error in canonical norms Numerical results with uniform and adaptively refined meshes illustrate the performance of the scheme
Model Problem Formulation
The model problem is given by finding a solution u in H10(omega) to −div(κ∇u + ρ(u)β) + γ(u) = f in omega <ref:2603.29863#pg4> The problem is motivated by the Richards equation <ref:2603. p. 129863#pg5> The right-hand side function f can be zero by a change of the nonlinearity γ For simplicity, we consider a right-hand side function Furthermore, we assume a homogeneous Dirichlet condition for ease of presentation
Minimum Residual Formulation
The problem is transformed into the minimum residual formulation by representing both nonlinear terms by independent variables, thus giving rise to a linear problem, and add the nonlinear relations as constraints This approach is similar to the approach in [9] for strongly monotone operators, but with least-squares constraints similar to the least-squares DPG-BEM coupling in [30] Globally, it is a finite-element minimum residual method that minimizes the linear residual in a discretized dual test space and nonlinear residuals in L2
Discretization Scheme
To discretize minimization problem (3), we consider finite-dimensional subspaces Vh ⊂ V and Uh ⊂ U The minimum residual discretization of (1) reads uh = (uh,̂σh, qh, rh) = arg min wh∈Uh (8) where this is a combination of a DPG method for the linear part of the problem with finite-element least squares for the nonlinear constraints
Error Estimation and Adaptivity
The quasi-optimal error estimate is proven under the assumption that uh is close to u in the sense that∥u − uh∥ ≤ c(h)∥∇(u − uh)∥ with 0 < c(h) → 0 as h → 0 The global residual Res (17) consists of local terms assigned to elements, in particular Buh − L 2V∗h = ∑T∈T res 2Vh(T)∗ These indicators Res(T) allow to perform adaptive mesh refinements
Numerical Experiments
Numerical experiments in two space dimensions are presented, and they confirm the expected convergence order O(N−1/2) of both the global residual Res and the error u−uhU The results for adaptive mesh refinements from Figure 3 indicate re-established optimal convergence orders of O(N−1/2) and u−uh = O(N−1), again confirming validity of the Aubin–Nitsche assumption in Theorem 3
References
[1] F. Bertrand, D. p. 9 Boffi, and H. p. 9 Schneider, Discontinuous Petrov-Galerkin approximation of eigenvalue problems, Comput. p. 9 Methods Appl. p. 9 p. 9 p. 9 Math., 23 (2023), pp. p. 9 1–17 <ref:2603. p. 029863#pg2>
[4] J. Bramwell, L. p. 9 F. Demkowicz, J. p. 1 Gopalakrishnan, and W. p. 9 Qiu, A locking-free hp DPG method for linear elasticity with symmetric stresses, Numer. p. 9 Math., 122 (2012), pp. p. 9 671–707 <ref:2603. p. 029863#pg5>
[5] P. Bringmann, M. p. 9 Brunner, and D. p. 9 Praetorius, Newton’s method in adaptive iteratively linearized FEM, arXiv: 2512. p. 919357, 2025
[6] P. Bringmann and D. p. 9 Praetorius, Global convergence of adaptive least-squares finite element methods for nonlinear PDEs, arXiv: 2509. p. 901531, 2025 Accepted for publication in SIAM J. p. 9 Numer. Anal., 2603.29864
[7] T. Bui-Thanh and O. p. 9 Ghattas, A PDE-constrained optimization approach to the discontinuous Petrov–Galerkin method with a trust region inexact Newton-CG solver, Comput. p. 9 Methods Appl. Mech. Engrg., 278 (2014), pp. p. 9 20–40
[14] L. Demkowicz and J. p. 1 Gopalakrishnan, The discontinuous Petrov-Galerkin method, Acta Numer. p. 0, 34 (2025), pp. p. 9 293–384
[17] A primal DPG method without a first-order reformulation, Comput. Math. Appl., 66 (2013), pp. 1058–1064
[29] T. Führer and N. p. 1 Heuer, Robust DPG test spaces and Fortin operators — the H1 and H(div) cases, SIAM J. p. 1 Numer. Anal., 62 (2024), pp. 718–748
[30] T. Führer, N. p. 1 Heuer, and M. p. 0 Karkulik, On the coupling of DPG and BEM, Math. p. 1 Comp., 86 (2017), pp. 2261–2284
[31] Analysis of backward Euler primal DPG methods, Comput. Methods Appl. Math., 21 (2021), pp. 811–826
[47] I. Muga and K. G. van der Zee, Discretization of linear problems in Banach spaces: residual minimization, nonlinear Petrov-Galerkin, and monotone mixed methods, SIAM J. p. 1 Numer. Anal., 58 (2020), pp. 3406–3426
[48] A. H. Niemi, Approximation properties of primal discontinuous Petrov-Galerkin method on general quadrilateral meshes, Comput. p. 9 Math. Appl., 151 (2023), pp. p. 9 300–303
[51] I. S. Pop, F. Radu, and P. Knabner, Mixed finite elements for the Richards’ equation: linearization procedure, J. p. 1 Comput Appl. p. 9 Math., 168 (2004), pp. 365–373
[52] F. Radu, I. S. Pop, and P. Knabner, Order of convergence estimates for an Euler implicit, mixed finite element discretization of Richards’ equation, SIAM J. Numer. Anal., 42 (2004), pp. 1452–1478
[53] L. A. Richards, Capillary conduction of liquids through porous mediums, Physics, 1 (1931), pp. p. 1 318–333
[54] N. V. Roberts, L. F. Demkowicz, and R. p. 1 Moser, A discontinuous Petrov-Galerkin methodology for adaptive solutions to the incompressible Navier-Stokes equations, J. p. 0 Comput. Phys., 301 (2015), pp. 456–483
[55] E. Schneid, P. Knabner, and F. Radu, A priori error estimates for a mixed finite element discretization of the Richards’ equation, Numer. p. 1 Math., 98 (2004), pp. 353–370
[11] C. Carstensen, L. p. 1 F. Demkowicz, and J. p. 1 Gopalakrishnan, A posteriori error control for DPG methods, SIAM J. p. 8 Numer. Anal., 52 (2014), pp. p. 9 p. 9 1335–1353
[46] W. F. Mitchell, 30 years of newest vertex bisection, J. p. 8 Numer. Anal Ind Appl Math. p. 0, 11 (2017), pp. 11–22
[49] A. H. Niemi, J. A. Bramwell, and L. p. 9 F. Demkowicz, Discontinuous Petrov-Galerkin method with optimal test functions for thin-body problems in solid mechanics, Comput. p. 0 Methods Appl Mech Eng. p. 9, 200 (2011), pp. 1291–1300
[57] T. Führer and F. p. 1 Fuica, A DPG method for Signorini problems, IMA J. p. 0 Numer. Anal., 38 (2018), pp. 1893–1926
[44] N. Heuer and F. p. 0 Pinochet, Ultra-weak formulation of a hypersingular integral equation on polygons and DPG method with optimal test functions, SIAM J. p. 0 Numer. Anal., 52 (2014), pp.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Adaptive Mesh Refinement for Semilinear Problems: The improved system can perform adaptive mesh refinement by using local terms assigned to elements, which allows for computable error indicators such as Res(T) based on local residuals. This enables marking the smallest set of elements T̃ where sum over T∈T̃ Res(T)2 is greater than θRes2, facilitating re-established optimal convergence orders.
-
Solving Semilinear Equations via Least-Squares Penalization: The system can solve semilinear elliptic problems by replacing nonlinear contributions with independent unknowns and adding the nonlinear relations as least-squares constraints. This transforms the problem into a linear problem with a non-trivial kernel, allowing standard DPG techniques to apply efficiently while maintaining the structure of minimum residual discretization.
-
Robust Error Estimation for Nonlinear Solutions: The system can provide quasi-optimal error estimates in canonical norms by assuming an Aubin–Nitsche type error estimate for the primal variable, which is crucial when problems are motivated by equations like the Richards equation. This ensures that solutions on refined meshes exhibit convergence orders such as u - uh = O(N-1) and grad(u - uh) = O(N-1/2).
Sources
- Least-Squares Finite Element Methods for nonlinear problems: A unified framework
- Newton's method in adaptive iteratively linearized FEM
- Global convergence of adaptive least-squares finite element methods for nonlinear PDEs
Related papers
- Do physics-informed neural networks (PINNs) need to be deep? Shallow PINNs using the Levenberg-Marquardt algorithm
- A Neural-preconditioned Poisson Solver for Mixed Dirichlet and Neumann Boundary Conditions
- Advanced Linear Algebra with Applications - Part I (Numerical linear algebra for PDEs, machine learning, and data assimilation)
- Operator learning meets inverse problems: A probabilistic perspective
- Second-order consistency for learning chaotic dynamics via randomized Jacobian matching
- Windowed thinning and query complexity for the bouncy particle and Zigzag samplers