Numerical approximations of one-dimensional linear conservation equations with discontinuous coefficients (Q4955858)

The paper is devoted to rather general numerical approximations of the following linear conservation equation: \[ \partial_t \mu+ \partial_x (a\mu)= 0\text{ for }(t,x) \in ]0,T] \times {\mathbb{R}}, \qquad \mu (0,.) = \mu _0 \in M_{\text{loc}} ({\mathbb{R}}), \tag{1} \] \[ a \in L^\infty (]0,T[ \times {\mathbb{R}}),\quad \partial _x( a) \leq \alpha \text{ in }]0,T[ \times {\mathbb{R}},\quad \alpha \in L^1 (]0,T[), \tag{2} \] which appears also explicitly when linearizing a nonlinear hyperbolic equation \[ \partial _t u + \partial _x f(u) = 0 \tag{3} \] with respect to the initial data or the flux \(f.\) It turns out that, most of the cases, \(\mu\) is a measure in the space variable. Because of the very low regularity imposed on the coefficient \(a,\) one cannot treat a priori this Cauchy problem in the theory of distributions. One way out is to understand the solution of (1) in the duality sense. For this purpose the dual problem is considered: \[ \partial_t p + a\partial_x p = 0 \text{ for }(t,x) \in ]0,T] \times {\mathbb{R}}, \quad p(T,.) = p^T \in \text{Lip}( {\mathbb{R}}),\text{ with compact support.} \tag{4} \] It is known that this backward problem admits a Lipschitz continuous solution under condition (2), and this fact has been used already to obtain uniqueness for (3). The point is that there is no uniqueness for (4), and one of the main results of \textit{F. Bouchut} and \textit{F. James} [C. R. Acad. Sci., Paris, Sér. I 320, No. 9, 1097-1102 (1995; Zbl 0829.35139); One-dimensional transport equations with discontinuous coefficients, Nonlinear Analysis, TMA 32, No. 7, 891-933 (1998)] is to characterize a class of solutions, known as reversible solutions, for which existence and uniqueness hold. The duality solution of (1) is then the unique element of the space \( C([0,T];M_{\text{loc}} ({\mathbb{R}}))\) satisfying for all reversible \(p\)'s \[ \frac{d} {dt}\int_{\mathbb{R}} p(t,x)\mu (t,dx). \tag{5} \] Equipped with this characterization, it is possible to give a precise interpretation of the product \((a\mu)\) in the distributional framework. As the numerical approximation of (1) the conservative algorithms of the type \[ \begin{aligned} \mu _j^{n + 1}&= \mu _j^n - \frac{\Delta t}{\Delta x} (\langle {\mathbf A}_{j + \frac{1}{2}}^n ,\vec \mu _{j + \frac{1}{2}}^n\rangle _{{\mathbb{R}}^{2K}}- \langle {\mathbf A}_{j - \frac{1}{2}}^n, \vec \mu _{j - \frac{1}{2}}^n \rangle _{{\mathbb{R}}^{2K} }),\\ \vec \mu _{j + \frac{1}{2}}^n &= (\mu _{j - K + 1}^n , \ldots ,\mu _{j + K}^n) \in {\mathbb{R}}^{2K}, \tag{6}\\ {\mathbf A}_{j + \frac{1}{2}}^n &= (a_{j + \frac{1}{2}, - K + 1}^n , \ldots ,a_{j + \frac{1}{2},K}^n) \in {\mathbb{R}}^{2K}\end{aligned} \] is considered for \(K \in N,\) where \(\mu _j^n \) and \({\mathbf A}_{j + \frac{1} {2}}^n \) denote some approximations of \(\mu(n\Delta t,j\Delta x)\) and \(a(n\Delta t,j\Delta x)\), respectively. Most of the work in the paper is done for estimating the dual scheme which is obtained by a summation by parts (as for the continuous equations). Because of the smoothness of the reversible solutions of (4), the strong properties such as BV, \(L^\infty \) or Lipschitz-like bounds for these backward approximations are sought. It is proved that, under some CFL-type conditions on the space-time grid, there are compactness results and convergence toward the reversible solution associated to every smooth final data. Moreover, property (5) is automatically enforced by the definition of the dual scheme. Finally, as a consequence of the conservative character of (6), there is an uniform bound on the total mass of the approximate solution of (1). All these arguments together gives the convergence result toward the duality solution of the problem (1). Some numerical computations obtained with three-point schemes taken from both these classes are presented.