This dissertation is dedicated to the development and analysis of finite volume numericals chemes for convection-diffusion equations, which notably occur in models arising from physics and biology. We are more particularly interested in preserving asymptotic behavior at the discrete level. This dissertation is composed of three parts, each one including two chapters. In the first part, we consider the discretization of the linear drift-diffusion system for semiconductors with the implicit Scharfetter-Gummel scheme. We focus on preserving two kinds of asymptotics with this scheme : the long-time asymptotic and the quasineutral limit. We show discrete energy–energy dissipation estimates which constitute the main point to prove first the large time convergence of the approximate solution to an approximation of the thermal equilibrium, and then the stability at the quasineutral limit. In the second part, we are interested in designing finite volume schemes which preserve the long time behavior in a more general framework. More precisely, we consider nonlinear convection-diffusion equations arising in various physical models : porous media equation, drift-diffusion system for semiconductors... We propose two spatial discretizations which preserve the long time behavior of the approximate solutions. We first generalize the Scharfetter-Gummel flux for a nonlinear diffusion. This scheme provides satisfying numerical results if the diffusion term does not degenerate. Then we propose a discretization which takes into account together the convection and diffusion terms by rewriting the flux as an advective flux. The numerical flux is then defined in such a way that equilibrium states are preserved, and we use a slope limiters method so as to obtain second order space accuracy, even in the degenerate case. Finally, the third part is devoted to the study of a numerical scheme for a chemotaxis model with cross diffusion, for which the solutions do not blow up in finite time, even for large initial data. The proof of convergence is based on a discrete entropy estimate which requires the use of discrete functional inequalities such as Poincaré-Sobolev and Gagliardo-Nirenberg-Sobolev inequalities. The demonstration of these inequalities is the subject of an independent chapter in which we propose a study in quite a general framework, including mixed boundary conditions and generalization to DDFV schemes.
