Discretization and convergence of the ballistic Benamou-Brenier formulation of the porous medium and Burgers’ equations.

Jean-Marie Mirebeau, Erwan Stampfli · 2025

Read Discretization and convergence of the ballistic Benamou-Brenier formulation of the porous medium and Burgers’ equations. by Jean-Marie Mirebeau, Erwan Stampfli (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-2511-02662.txt.

Introduction

Jean-Marie Mirebeau and Erwan Stampfli study how to compute two hard evolution equations by turning them into optimization problems. The porous medium equation governs slow spreading, as in gas through rock, while Burgers' equation models shock-like flow. Their 2025 paper uses a ballistic variant of the Benamou-Brenier formulation, borrowed from optimal transport theory.

The pair discretize the whole space-time problem at once, using harmonic interpolation for the densities, a choice they note is new here. They prove the scheme stays stable for any step sizes and converges at a quadratic rate for the dual solution. A proximal splitting solver with fast Fourier transforms makes it practical. Numerical tests back the theory on both equations, in several dimensions for the porous case.

Catalog record

CREATORJean-Marie Mirebeau, Erwan Stampfli
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE130 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-02662

Browse the full library · About the Tower of Babel

Full text

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

Discretization and convergence of the ballistic Benamou-Brenier formulation of the porous medium and Burgers’ equations. Jean-Marie Mirebeau, Erwan Stampfli Abstract We study the discretization, convergence, and numerical implementation of recent reformulations of the quadratic porous medium equation (multidimensional and anisotropic) and Burgers’ equation (one-dimensional, with optional viscosity), as forward in time variants of the Benamou-Brenier formulation of optimal transport. This approach turns those evolution problems into global optimization problems in time and space, of which we introduce a discretization, one of whose originalities lies in the harmonic interpolation of the densities involved. We prove that the resulting schemes are unconditionally stable w.r.t. the space and time steps, and we establish a quadratic convergence rate for the dual PDE solution, under suitable assumptions. We also show that the schemes can be efficiently solved numerically using a proximal splitting method and a global space-time fast Fourier transform, and we illustrate our results with numerical experiments. 0 0 footnotetext: Mathematics Subject Classification (2020) : 65M06, 65M12, 49M29, 35A15, 35L65 1 Introduction A number of non-linear partial differential equations (PDEs) can be reformulated as global convex optimization problems in time and space [ Bre20 ] , revealing so-called hidden convexity properties. We follow in this paper an approach whose properties are particularly appealing for evolution problems featuring quadratic non-linearities [ Vor22 ] , already investigated numerically for Burgers’ equation [ Bre20 , KA24 ] and for non-linear elasticity [ SGA24 ] . In the continuous setting, its benefits include new notions of weak solutions, and related methods for proving their existence and/or uniqueness. In the discrete setting, it leads to a radically new method for solving evolution problems, of which this paper provides the first numerical convergence analysis, with schemes free of any Courant-Friedrichs-Levy (CFL) condition, second-order accurate in time and space, solved using proximal optimization algorithms and global space-time fast-Fourier transforms. As discussed below, this approach shares similarities with the Benamou-Brenier dynamic formulation of optimal transport [ BB00 ] , whose numerical analysis is studied in e.g. [ Lav19 , NT21 ] or [ LZ24 ] in the matrix-valued setting. Our numerical implementation eventually relies on proximal splitting, inspired by [ PPO14 ] which follows that approach for the dynamic formulation of optimal transport, see [ Lav19 ] for a history and comparison of alternative numerical methods. While [ Vor22 , Bre20 ] mostly focus on systems of PDEs, often related to fluid mechanics, we shall limit in this paper our attention to two scalar PDEs for simplicity, see ( 1 ). Before that, let us further motivate the approach and replace it within the literature. The initial value problem for many non-linear evolution PDEs may admit infinitely many weak solutions for most initial conditions [ DiP79 , LJ09 , LS10 ] . For the Euler equations of incompressible fluids and, also, for the class of hyperbolic systems of conservation laws with a convex entropy, Brenier [ Bre18 ] considered the following problem: given an initial condition u_{0} and a finite time interval [0,T] , find among all weak solutions starting from u_{0} one that minimizes the time integral of the kinetic energy (for Euler) or the entropy (for systems of conservation laws). This problem might have no solution, but it generates a dual variational problem which is automatically convex and usually admits a solution, that we call dual solution . Brenier obtained an explicit formula relating this dual solution to any tentative smooth solution of the initial value problem, at least when T is not too large, see remark 1.5 . It was also pointed out that dual problems of this kind are strongly related to optimal transport problems (in their ‘Benamou-Brenier’ formulation [ BB00 ] ). and can be interpreted as generalized (with matrix-valued densities) Mean Field Games à la Lasry-Lions. The method of [ Bre18 ] was further developped by Vorotnikov [ Vor22 ] for a very large class of quadratic evolution PDEs and a similar duality approach has been used by Acharya and collaborators [ Ach23 , AS24 , ASZ24 ] . Let us finally mention [ BM22 ] where the dual technique is applied to a multi-stream formulation of the Euler-Poisson system and [ Bre20 ] where various parabolic PDEs are considered, including the QPME, for which there is no restriction on the size of T and which is the main subject of the present paper. As announced, this paper is devoted to the discretization and convergence analysis of dual reformulations of two scalar evolution PDEs featuring quadratic non-linearities: the quadratic porous medium equation (QPME, multidimensional and anisotropic), and Burgers’ equation with optional viscosity (one dimensional). \displaystyle\text{QPME : }\partial_{t}u \displaystyle=\tfrac{1}{2}\diver(\mathcal{D}\nabla u^{2}), \displaystyle\text{Burgers' : }\partial_{t}u+\tfrac{1}{2}\partial_{x}u^{2}=\nu\partial_{xx}u. (1) They are posed for simplicity on the periodic unit box \mathbb{T}^{d} , with d=1 for Burgers’ equation, where \mathbb{T}:=\mathbb{R}/\mathbb{Z} is the periodic unit interval. See [ KA24 ] for a closely related approach implementing non-periodic boundary conditions in Burgers’ case. The PDE parameters are a smooth positive definite tensor field \mathcal{D}\in C^{\infty}(\mathbb{T}^{d},\mathcal{S}_{d}^{++}) , following [ Li18 ] , and a non-negative diffusion coefficient \nu\geq 0 respectively. A smooth and positive initial condition u_{0}\in C^{\infty}(\mathbb{T}^{d},]0,\infty[) and a bounded time interval [0,T] are also given. We make these strong assumptions for the sake of the numerical analysis, even though the PDEs of interest admit solutions under weaker assumptions [ Vá07 ] . Our numerical experiments section 4 also show that the proposed schemes are robust and appear to work without these assumptions. The two PDEs ( 1 ) share a common structure, with a quadratic non-linearity: \displaystyle\partial_{t}u+\tfrac{1}{2}L(u^{2})+LAu \displaystyle=0\quad\text{on }[0,T], \displaystyle u(0) \displaystyle=u_{0}, (2) where L=-\diver(\mathcal{D}\nabla\cdot) and A=0 for the QPME, and L=\partial_{x} and A=-\nu\partial_{x} for Burgers’ equation. Following [ Bre20 , Vor22 ] we formally consider the (apparently trivial) problem ( 3 ) of minimizing the L^{2} norm among all weak solutions of ( 2 ); variants featuring a time-dependent weight or a base state can also be considered, see [ Vor25 , ASZ24 ] and remark 1.5 . Formally, again, we exchange in ( 4 ) the \sup and \inf , and perform an integration by parts \displaystyle\inf_{u(0)=u_{0}}\sup_{\phi(T)=0}\int_{[0,T]\times\mathbb{T}^{d}}\tfrac{1}{2}u^{2}+(\partial_{t}u+\tfrac{1}{2}L(u^{2})+LAu)\phi (3) \displaystyle\overset{?}{=}\!\!\sup_{\phi(T)=0}\inf_{u(0)=u_{0}}\int_{[0,T]\times\mathbb{T}^{d}}\tfrac{1}{2}u^{2}(1+L^{*}\phi)-(u-u_{0})\partial_{t}\phi+uA^{*}L^{*}\phi. (4) Here and below, given a bivariate mapping u(t,x) , we denote by u(t):=u(t,\cdot) the univariate function of the space variable alone, where the time argument has been frozen to some given value t . Note that One always has ( 3 ) \geq ( 4 ), and the difference ( 3 ) - ( 4 ) is referred to as the duality gap. We make at this point the critical assumption that 1+L^{*}\phi>0 everywhere on [0,T]\times\mathbb{T}^{d} , which is closely related with the lack of duality gap, see remark 1.5 . Minimizing pointwise ( 4 ) w.r.t. the primal variable u , we obtain the non-trivial dual problem [ Bre20 , Vor22 , KA24 ] : \displaystyle\inf_{\phi(T)=0}\int_{[0,T]\times\mathbb{T}^{d}}\frac{(\partial_{t}\phi-A^{*}L^{*}\phi)^{2}}{2(1+L^{*}\phi)}-u_{0}\partial_{t}\phi. (5) The primal and dual variables u and \phi obey the following optimality relation \displaystyle\partial_{t}\phi-A^{*}L^{*}\phi \displaystyle=(1+L^{*}\phi)u, \displaystyle\phi(T) \displaystyle=0, (6) which can be regarded as a reversed in time linear evolution problem w.r.t. \phi . (For large T , this may contradict our assumption 1+L^{*}\phi>0 , see remark 1.5 .) Defining a pseudo-momentum and pseudo-density as \displaystyle m \displaystyle:=\partial_{t}\phi, \displaystyle\rho \displaystyle:=1+L^{*}\phi, (7) we obtain as announced a

Read the complete text