We consider entropy solutions to the eikonal equation $|\nabla u|=1$ in two-space dimensions. These solutions are motivated by a class of variational problems and fail in general to have bounded variation. Nevertheless, they share several of their fine properties with BV functions: we show in particular that the set of non-Lebesgue points has at least one co-dimension.

In this work we deal with the numerical solution of aHamilton-Jacobi-Bellman (HJB) equation with infinitely manysolutions. To compute the maximal solution – the optimalcost of the original optimal control problem – we present acomplete discrete method based on the use of some finite elementsand penalization techniques.

We study the Dirichlet boundary value problem for eikonal type equations of ray light propagation in an inhomogeneous medium with discontinuous refraction index. We prove a comparison principle that allows us to obtain existence and uniqueness of a continuous viscosity solution when the Lie algebra generated by the coefficients satisfies a Hörmander type condition. We require the refraction index to be piecewise continuous across Lipschitz hypersurfaces. The results characterize the value function of the generalized minimum time problem with discontinuous running cost.

In order to accommodate solutions with multiple phases, corresponding to crossing rays, weformulate geometrical optics for the scalar wave equation as a kinetic transport equation set in phase space. If the maximum number of phases is finite and known a priori we can recover the exact multiphase solution from anassociated system of moment equations, closed by an assumption on the form of the density function in the kinetic equation. We consider two different closure assumptions based ondelta and Heaviside functions and analyze the resultingequations. They form systems of nonlinear conservation laws with source terms. In contrast to the classicaleikonal equation, theseequations will incorporate a "finite" superposition principle in the sense that while the maximum number of phasesis not exceeded a sum of solutions is also a solution. We present numerical results for a varietyof homogeneous and inhomogeneous problems.

In the shape from shading problem of computer vision oneattempts to recover the three-dimensional shape of an object orlandscape from the shading on a single image. Under theassumptions that the surface is dusty, distant, and illuminatedonly from above, the problem reduces to that of solving theeikonal equation |Du|=f on a domain in $\mathbb{R}^2$ . Despitevarious existence and uniqueness theorems for smooth solutions,we show that this problem is unstable, which is catastrophic forgeneral numerical algorithms.

We consider the evolution of a set $\Lambda\subset \mathbb R^2$ according to theHuygens principle: i.e. the domain at time t>0, Λt, is the set ofthe points whose distance from Λ is lower thant. We give some general results for this evolution,with particular care given to the behavior of the perimeter of theevoluted set as a function of time. We define a class of sets (non-trapping sets) for which the perimeter is a continuous function of t, andwe give an algorithm to approximate the evolution.Finally we restrict our attention to the class of sets for which the turning angle of the boundary is greater than -π (see [2]).For this class of sets we prove that the perimeter is a Lipschitz-continuous function of t. This evolution problem is relevant for the applicationsbecause it is used as a model for solid fuel combustion.