Abstract. We prove that bounded solutions of the vanishing hyper-viscosity equation, ut + f (u)x + (−1) s ε∂ 2sx u = 0 converge to the entropy solution of the corresponding convex conservation law ut +f (u)x = 0, f > 0. The hyper-viscosity case, s > 1, lacks the monotonicity which underlines the Krushkov BV theory in the viscous case s = 1. Instead we show how to adapt the Tartar-Murat compensated compactness theory together with a weaker entropy dissipation bound to conclude the convergence of the vanishing hyper-viscosity.Key words. Conservation law, hyper-viscosity, entropy dissipation, compensated compactness AMS subject classifications. 35L65, 35B30, 65N35.
Convergence with vanishing hyper-viscosityConsider the convex conservation lawsubject to initial conditions, u(x, 0) = u 0 . We are concerned with the convergence of its hyper-viscosity regularization of order s ≥ 1The viscous case corresponding to s = 1 is well understood within the classical Krushkov theory, which is built on the monotonicity of the associated solution operator, e.g., [Daf00, §VI]. The prototype is Burgers' equation governed by the quadratic flux f (u) = u 2 /2. The hyper-viscosity case for s > 1, however, lacks monotonicity and the Krushkov BV theory seems out of reach. Instead we show how to adapt the Tartar-Murat compensated compactness theory, [Tar75,Mur78] in the present nonmonotone framework. A similar approach originated with [Sch82] for the vanishing diffusion-dispersion problem where the RHS of (1.2) is replaced by εu ε xx + δ ε u ε xxx . In the particular borderline case, δ ε ∼ ε 2 , limit solutions may in fact violate Krushkov entropy condition, [KL02]. Otherwise, entropy solution limits are recovered by compensated arguments as long as diffusion dominates, δ ε ε 2 , [Sch82, KL02]. We should point out that in the present context, hyper-viscosity with s > 1 yields a weaker entropy dissipation bound than in the viscosity dominated case s = 1, consult (1.6) below. We show that this hyper-viscosity entropy dissipation estimate will suffice.To begin with, we rescale the hyper-viscosity amplitude ε ≡ ε N = N −(2s−1) . Denote u N ≡ u εN , then (1.2) reads *