A convergence rate result for front tracking approximations of conservation laws with discontinuous flux
Shyam Sundar Ghoshal, John D. Towers · 2025
Read A convergence rate result for front tracking approximations of conservation laws with discontinuous flux by Shyam Sundar Ghoshal, John D. Towers (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-2509-22952.txt.
Introduction
Shyam Sundar Ghoshal and John D Towers bound the error of a front tracking method for conservation laws. Their law has a flux that jumps at one point in space, the two-flux case. Front tracking follows each wave front in steps. They prove the error falls with a rate of one half or better.
The fluxes need only be smooth, not convex, monotone, or single-peaked. Only a finite count of flux crossings is assumed. When both fluxes rise or both fall, the rate improves to one. Equal fluxes also give rate one, matching classic results. Their proof uses a new route, not the standard Kuznetsov lemma.
Catalog record
| CREATOR | Shyam Sundar Ghoshal, John D. Towers |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 66 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-22952
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A convergence rate result for front tracking approximations of conservation laws with discontinuous flux Shyam Sundar Ghoshal, John D. Towers Abstract We consider the initial value problem for a scalar conservation law in one space dimension with a single spatial flux discontinuity, the so-called two-flux problem. We prove that a well-known front tracking algorithm has a convergence rate of at least one-half. The fluxes are required to be smooth, but are not required to be convex or concave, monotone, or even unimodal. We require that there are no more than finitely many flux crossings, but we do not require that they satisfy the so-called crossing condition. If both fluxes are strictly increasing or strictly decreasing then our analysis yields a convergence rate of one, in agreement with a recent result. Similarly, if the fluxes are equal, i.e., there is no flux discontinuity, we obtain a convergence rate of one in this case also, in agreement with a classical result. The novelty of this paper is that the class of discontinuous-flux conservation laws for which there is a front tracking error estimate is expanded, and that the method of analysis is new; we do not use the Kuznetsov lemma which is commonly used for this type of analysis. 2010 MSC 35L65, 35R05, 65M12, 65M15 1 Introduction The subject of this paper is approximation via front tracking to solutions of an initial value problem for a scalar conservation law having a spatially discontinuous flux: \textrm{Problem $\displaystyle P$:}\quad\left\{\begin{split}&u_{t}+\mathcal{F}(x,u)_{x}=0\quad\textrm{for $\displaystyle(x,t)\in\Pi_{T}:=\mathbb{R}\times(0,T)$,}\\ &\mathcal{F}(x,u):=H(-x)g(u)+H(x)f(u),\\ &u(x,0)=u_{0}(x)\quad\textrm{for $\displaystyle x\in\mathbb{R}$.}\end{split}\right. (1.1) Here \displaystyle H(x) is the Heaviside function. Thus, the flux \displaystyle\mathcal{F}(x,u) of this conservation law has a spatial dependence that is discontinuous at \displaystyle x=0 if the functions \displaystyle f and \displaystyle g are different. The front tracking algorithm [ 32 , 44 ] computes approximate solutions \displaystyle u^{\delta} of Problem P by producing the exact solution to a simpler problem \displaystyle P^{\delta} where the flux \displaystyle\mathcal{F} is replaced by a piecewise linear approximation \displaystyle\mathcal{F}^{\delta} and the initial data \displaystyle u_{0} is replaced by a piecewise constant approximation \displaystyle u_{0}^{\delta} : \textrm{Problem $\displaystyle P^{\delta}$:}\quad\left\{\begin{split}&u^{\delta}_{t}+\mathcal{F}^{\delta}(x,u^{\delta})_{x}=0\quad\textrm{for $\displaystyle(x,t)\in\Pi_{T}$,}\\ &\mathcal{F}^{\delta}(x,u):=H(-x)g^{\delta}(u)+H(x)f^{\delta}(u),\\ &u^{\delta}(x,0)=u^{\delta}_{0}(x)\quad\textrm{for $\displaystyle x\in\mathbb{R}$.}\end{split}\right. (1.2) Here \displaystyle\delta is a parameter that defines the level of discretization. Our main result (Theorem 3.1 ) is that the front tracking algorithm described below produces approximate solutions that converge to the so-called vanishing viscosity solution (Definition 1.3 ) at a rate of at least \displaystyle O(\delta^{1/2}) . If \displaystyle f and \displaystyle g are both strictly monotone, we obtain a convergence rate of \displaystyle O(\delta) (Theorem 4.2 ), in agreement with a recent result of Ruf [ 45 ] . Finally, we also obtain a convergence rate of \displaystyle O(\delta) if \displaystyle f=g (Theorem 4.3 ), in agreement with a classical result due to Lucier [ 39 ] . Our method of analysis is different from the methods found in [ 45 ] and [ 39 ] ; for example we do not invoke Kuznetsov’s lemma [ 32 , Theorem 3.11] . Front tracking is both a theoretical tool and a practical computational algorithm. It is more complex to implement than finite difference / finite volume methods, but it is computationally rapid and devoid of the numerical dissipation produced by finite difference / finite volume methods. Before specifying the data of the problem we will need the following definition: Definition 1.1 (flux crossing) . We call \displaystyle u_{\chi} a flux crossing if \displaystyle f(u_{\chi})=g(u_{\chi}) , and for some \displaystyle\epsilon>0 it holds that \textrm{$\displaystyle\left(g(u_{L})-f(u_{L})\right)\left(g(u_{R})-f(u_{R})\right)<0$ for all $\displaystyle u_{L}\in(u_{\chi}-\epsilon,u_{\chi})$, $\displaystyle u_{R}\in(u_{\chi},u_{\chi}+\epsilon)$}. In less formal terms, a flux crossing \displaystyle u_{\chi} is the location of a zero of the function \displaystyle f(u)-g(u) where the sign of \displaystyle f-g changes. Assumption 1.1 . We make the following assumptions about the data. In what follows \displaystyle\underline{u},\overline{u}\in\mathbb{R} with \displaystyle\underline{u}<\overline{u} . ( A .1) \displaystyle g(\underline{u})=f(\underline{u}) , \displaystyle g(\overline{u})=f(\overline{u}) , ( A .2) \displaystyle g,f\in C^{2}([\underline{u},\overline{u}]) , ( A .3) \displaystyle f and \displaystyle g are not linear on any nondegenerate interval, ( A .4) \displaystyle f(u)\neq g(u) for \displaystyle u\in(\underline{u},\overline{u}) , except for finitely many (possibly zero) flux crossings in the interval \displaystyle(\underline{u},\overline{u}) , ( A .5) \displaystyle u_{0}\in L^{\infty}(\mathbb{R})\cap\textrm{BV}(\mathbb{R}) , and \displaystyle u_{0}(x)\in[\underline{u},\overline{u}] for all \displaystyle x\in\mathbb{R} , ( A .6) for some \displaystyle X>0 , \displaystyle u_{0}(x)=u_{L} for \displaystyle x\leq-X and \displaystyle u_{0}(x)=u_{R} for \displaystyle x\geq X . Remark 1.1 . Assumption (A.4) requires that there be only finitely many flux crossings, but we do not require that any of them satisfy the so-called “crossing condition” [ 34 , 36 ] . Remark 1.2 . Items (A.3) and (A.4) are required so that we can use the results of [ 36 ] concerning the convergence of a certain Godunov scheme (which we use as an analytical tool) to a vanishing viscosity solution. In the special case where both \displaystyle f and \displaystyle g are strictly increasing or strictly decreasing, the vanishing viscosity theory simplifies greatly, and those assumptions are not needed. Similarly, if \displaystyle f=g we do not need (A.3) or (A.4) . The front tracking algorithm. To describe the front track algorithm we must specify the construction of \displaystyle f^{\delta} , \displaystyle g^{\delta} , and \displaystyle u_{0}^{\delta} for fixed \displaystyle\delta>0 . To specify \displaystyle f^{\delta} and \displaystyle g^{\delta} , choose a positive integer \displaystyle K , and breakpoints \displaystyle\mathsf{u}_{0},\mathsf{u}_{1},\ldots,\mathsf{u}_{K} such that \underline{u}=\mathsf{u}_{0}<\mathsf{u}_{1}<\cdots<\mathsf{u}_{K}=\overline{u},\quad\textrm{$\displaystyle\mathsf{u}_{k+1}-\mathsf{u}_{k}\leq\delta$ for $\displaystyle k=0,\ldots,K-1$.} (1.3) Then \displaystyle f^{\delta} and \displaystyle g^{\delta} are defined to be the piecewise linear interpolants to \displaystyle f and \displaystyle g such that \displaystyle f^{\delta}(\mathsf{u}_{k})=f(\mathsf{u}_{k}) and \displaystyle g^{\delta}(\mathsf{u}_{k})=g(\mathsf{u}_{k}) for \displaystyle k=0,\ldots,K . The front tracking initial data \displaystyle u_{0}^{\delta} is constructed satisfying the following conditions: \begin{split}&\textrm{1. $\displaystyle u_{0}^{\delta}$ is piecewise constant with finitely many jumps, and such that}\\ &u_{0}^{\delta}(x)\in[\underline{u},\overline{u}]\,\,\forall x\in\mathbb{R},\quad\TV(u_{0}^{\delta})\leq\TV(u_{0}),\\ &\textrm{2. condition~\ref{init_data_large_x} is satisfied with the same $\displaystyle X$ as for $\displaystyle u_{0}$, and}\\ &\textrm{$\displaystyle\left\|u_{0}^{\delta}-u_{0}\right\|_{L^{1}(\mathbb{R})}=O(\delta)$ as $\displaystyle\delta\rightarrow 0$.}\end{split} (1.4) Remark 1.3 . \displaystyle f^{\delta} and \displaystyle g^{\delta} clearly satisfy Assumption (A.1) , and for \displaystyle\delta sufficiently small, Assumption (A.4) also holds for \displaystyle f^{\delta} , \displaystyle g^{\delta} . It follows from Assumption (A.2) that \displaystyle f,g\in\textrm{Lip}([\underline{u},\overline{u}]) . Let \displaystyle L_{f} and \displaystyle L_{g} denote Lipschitz constants for \displaystyle f and \displaystyle g . It is clear that also \displaystyle f^{\delta},g^{\delta}\in\textrm{Lip}([\underline{u},\overline{u}]) , and L_{f^{\delta}}\leq L_{f},\quad L_{g^{\delta}}\leq L_{g}. (1.5) Moreover, it is clear from ( 1.4 ) that \displaystyle u_{0}^{\delta} satisfies Assumptions (A.5) and (A.6) . Thus all parts of Assumption 1.1 are satisfied for the data associated with the front tracking algorithm, with exception of Assumption (A.3) . This assumption is present to guarantee the existence of one sided traces along the interface at \displaystyle x=0 . With the addition of Assumption 1.2 below the required traces of \displaystyle u^{\delta} exist without Assumption (A.3) . In view of Remark 1.3 , with \displaystyle\delta>0 fixed and \displaystyle f^{\delta} , \displaystyle g^{\delta} , and \displaystyle u_{0}^{\delta} in hand, the front tracking algorithm then computes the exact vanishing viscosity solution, defined below (Definition 1.3 ), of problem \displaystyle P^{\delta} . The solution of problem \displaystyle P^{\delta} is piecewise constant, with discontinuities (fronts), some of which are moving. The algorithm keeps track of the location of each front as time advances. Each time two fronts collide, the piecewise constant solution is updated by solving a Riemann problem. Each such Riemann problem