Searcharxiv⌕ Search

arXiv subjects

Kenneth Hvistendahl Karlsen

Publications and source records attributed to Kenneth Hvistendahl Karlsen.

5 recordsLinked to original sources

Upwind filtering of scalar conservation laws

We study a class of multi-dimensional non-local conservation laws of the form $\partial_t u = \operatorname{div}^Φ \mathbf{F}(u)$, where the standard local divergence $\operatorname{div}$ of the flux vector $\mathbf{F}(u)$ is replaced by an average upwind divergence operator $\operatorname{div}^Φ$ acting on the flux along a continuum of directions given by a reference measure and a filter $Φ$. The non-local operator $\operatorname{div}^Φ$ applies to a general non-monotone flux $\mathbf{F}$, and is constructed by decomposing the flux into monotone components according to wave speeds determined by $\mathbf{F}'$. Each monotone component is then consistently subjected to a non-local derivative operator that utilizes an anisotropic kernel supported on the "correct" half of the real axis. We establish well-posedness, derive a priori and entropy estimates, and provide an explicit continuous dependence result on the kernel. This stability result is robust with respect to the "size" of the kernel, allowing us to specify $Φ$ as a Dirac delta $δ_0$ to recover entropy solutions of the local conservation law $\partial_t u = \operatorname{div} \mathbf{F}(u)$ (with an error estimate). Other choices of $Φ$ (and the reference measure) recover known numerical methods for (local) conservation laws. This work distinguishes itself from many others in the field by developing a consistent non-local approach capable of handling non-monotone fluxes.

math.AP↗

Numerical methods for conservation laws with rough flux

Finite volume methods are proposed for computing approximate pathwise entropy/kinetic solutions to conservation laws with a rough path dependent flux function. For a convex flux, it is demonstrated that rough path oscillations may lead to "cancellations" in the solution. Making use of this property, we show that for $α$-H{ö}lder continuous rough paths the convergence rate of the numerical methods can improve from $\mathcal{O}(\text{COST}^{-γ})$, for some $γ\in \left[α/(12-8α), α/(10-6α)\right]$, with $α\in (0, 1)$, to $\mathcal{O}(\text{COST}^{-\min(1/4,α/2)})$. Numerical examples support the theoretical results.

math.NA↗

Path-dependent convex conservation laws

For scalar conservation laws driven by a rough path $z(t)$, in the sense of Lions, Perthame and Souganidis in arXiv:1309.1931, we show that it is possible to replace $z(t)$ by a piecewise linear path, and still obtain the same solution at a given time, under the assumption of a convex flux function in one spatial dimension. This result is connected to the spatial regularity of solutions. We show that solutions are spatially Lipschitz continuous for a given set of times, depending on the path and the initial data. Fine properties of the map $z \mapsto u(τ)$, for a fixed time $τ$, are studied. We provide a detailed description of the properties of the rough path $z(t)$ that influences the solution. This description is extracted by a "factorization" of the solution operator (at time $τ$). In a companion paper, we make use of the observations herein to construct computationally efficient numerical methods.

math.AP↗

On stochastic conservation laws and Malliavin calculus

For stochastic conservation laws driven by a semilinear noise term, we propose a generalization of the Kružkov entropy condition by allowing the Kružkov constants to be Malliavin differentiable random variables. Existence and uniqueness results are provided. Our approach sheds some new light on the stochastic entropy conditions put forth by Feng and Nualart [J. Funct. Anal., 2008] and Bauzet, Vallet, and Wittbold [J. Hyperbolic Differ. Equ., 2012].

math.AP↗

On the convergence rate of finite difference methods for degenerate convection-diffusion equations in several space dimensions

We analyze upwind difference methods for strongly degenerate convection-diffusion equations in several spatial dimensions. We prove that the local $L^1$-error between the exact and numerical solutions is $\mathcal{O}(Δx^{2/(19+d)})$, where $d$ is the spatial dimension and $Δx$ is the grid size. The error estimate is robust with respect to vanishing diffusion effects. The proof makes effective use of specific kinetic formulations of the difference method and the convection-diffusion equation.

math.AP↗