The Fokker--Planck approximation for an elementary linear, two-dimensional kinetic model endowed with a mass-preserving integral collision process is numerically studied, along with its diffusive limit. In order to set up a well-balanced discretization relying on an $S$-matrix, exact steady states of the continuous equation are derived. The ability of the scheme to keep these stationary solutions invariant produces the discretization of the local differential operator which mimics the collision process. The aforementioned scheme can be reformulated as an implicit-explicit one, which is proved to be both well-balanced and asymptotic-preserving in the diffusion limit. Several numerical benchmarks, computed on coarse grids, are displayed so as to illustrate the results.

A (2+2)-dimensional kinetic equation, directly inspired by the run-and-tumble modelingof chemotaxis dynamics is studied so as to derive a both ''2D well-balanced'' and''asymptotic-preserving'' numerical approximation. To this end, exact stationary regimesare expressed by means of Laplace transforms of Fourier-Bessel solutions of associatedelliptic equations. This yields a scattering S-matrix which permits to formulate a timemarchingscheme in the form of a convex combination in kinetic scaling. Then, in thediffusive scaling, an IMEX-type discretization follows, for which the ''2D well-balancedproperty'' still holds, while the consistency with the asymptotic drift-diffusion equation ischecked. Numerical benchmarks, involving ''nonlocal gradients'' (or finite samplingradius), carried out in both scalings, assess theoretical findings. Nonlocal gradients appearto inhibit blowup phenomena.

Two-dimensional dissipative and isotropic kinetic models, like the ones used in neutron transport theory, are considered. Especially, steady-states are expressed for constant opacity and damping, allowing to derive a scattering S-matrix and corresponding "truly 2D well-balanced" numerical schemes. A first scheme is obtained by directly implementing truncated Fourier-Bessel series, whereas another proceeds by applying an exponential modulation to a former, conservative, one. Consistency with the asymptotic damped parabolic approximation is checked for both algorithms. A striking property of some of these schemes is that they can be proved to be both 2D well-balanced and asymptotic-preserving in the parabolic limit, even when setting up IMEX time-integrators: see Corollaries 3.4 and A.1. These findings are further confirmed by means of practical benchmarks carried out on coarse Cartesian computational grids.

For a four-stream approximation of the kinetic model of radiative transfer with isotropic scattering, a numerical scheme endowed with both truly 2D well-balanced and diffusive asymptotic-preserving properties is derived, in the same spirit as what was done in [L. Gosse and G. Toscani, C. R. Math. Acad. Sci. Paris, 334 (2002), pp. 337-342] in the 1D case. Building on former results of Birkhoff and Abu-Shumays [J. Math. Anal. Appl., 28 (1969), pp. 211-221], it is possible to express 2D kinetic steady-states by means of harmonic polynomials, and this allows one to build a scattering S-matrix yielding a time-marching scheme. Such an S-matrix can be decomposed, as in [L. Gosse and N. Vauchelet, Numer. Math., 141 (2019), pp. 627-680], so as to deduce another scheme, well-suited for a diffusive approximation of the kinetic model, for which rigorous convergence can be proved. Challenging benchmarks are also displayed on coarse grids.

Dissipative kinetic models inspired by neutron transport are studied in a (1+1)-dimensional context: first, in the two-stream approximation, then in the general case of continuous velocities. Both are known to relax, in the diffusive scaling, toward a damped heat equation. Accordingly, it is shown that "uniformly accurate" L-splines discretizations of this parabolic asymptotic equation emerge from well-balanced schemes involving scattering S-matrices for the kinetic models. Moreover, well-balanced properties are shown to be preserved when applying IMEX time-integrators in the diffusive scaling. Numerical tests confirm these theoretical findings.

This paper is concerned with diffusive approximations of some numerical schemes for several linear (or weakly nonlinear) kinetic models which are motivated by wide-range applications, including radiative transfer or neutron transport, run-and-tumble models of chemotaxis dynamics, and Vlasov-Fokker-Planck plasma modeling. The well-balanced method applied to such kinetic equations leads to time-marching schemes involving a ``{\it scattering $S$-matrix}'', itself derived from a normal modes decomposition of the stationary solution. One common feature these models share is the type of diffusive approximation: their macroscopic densities solve drift-diffusion systems, for which a distinguished numerical scheme is Il'in/Scharfetter-Gummel's ``{\it exponential fitting}'' discretization. We prove that these well-balanced schemes relax, within a parabolic rescaling, towards such type of discretization by means of an appropriate decomposition of the $S$-matrix, hence are {\it asymptotic preserving

Localization phenomena (sometimes called ``{\it flea on the elephant}'') for the operator $L^\varepsilon=-\varepsilon^2 \Delta u + p(\xx) u$, $p(\xx)$ being an asymmetric double-well potential, are studied both analytically and numerically, mostly in two space dimensions within a perturbative framework. Starting from a classical harmonic potential, the effects of various perturbations are retrieved, especially in the case of two asymmetric potential wells. These findings are illustrated numerically by means of an original algorithm, which relies on a discrete approximation of the Steklov-Poincar\'e operator for $L^\varepsilon$, and for which error estimates are established. Such a two-dimensional discretization produces less mesh-imprinting than more standard finite-differences and captures correctly sharp layers.

Well-balanced schemes, nowadays mostly developed for both hyperbolic and kinetic equations, are extended in order to handle linear parabolic equations, too. By considering the variational solution of the resulting stationary boundary-value problem, a simple criterion of uniqueness is singled out: the C1 regularity at all knots of the computational grid. Being easy to convert into a finite-difference scheme, a well-balanced discretization is deduced by defining the discrete time-derivative as the defect of C1 regularity at each node. This meets with schemes formerly introduced in the literature relying on so-called L-spline interpolation of discrete values. Various monotonicity, consistency and asymptotic-preserving properties are established, especially in the under-resolved vanishing viscosity limit. Practical experiments illustrate the outcome of such numerical methods.

A general methodology, which consists in deriving two-dimensional finite-difference schemes which involve numerical fluxes based on Dirichlet-to-Neumann maps (or Steklov-Poincare operators), is first recalled. Then, it is applied to several types of diffusion equations, some being weakly anisotropic, endowed with an external source. Standard finite-difference discretizations are systematically recovered, showing that in absence of any other mechanism, like e.g. convection and/or damping (which bring Bessel and/or Mathieu functions inside that type of numerical fluxes), these well-known schemes achieve a satisfying multi-dimensional character. (C) 2018 Published by Elsevier Ltd.

A genuinely two-dimensional discretization of general drift-diffusion (including incompressible Navier--Stokes) equations is proposed. Its numerical fluxes are derived by computing the radial derivatives of "bubbles" which are deduced from available discrete data by exploiting the stationary Dirichlet--Green function of the convection-diffusion operator. These fluxes are reminiscent of Scharfetter and Gummel's in the sense that they contain modified Bessel functions which allow one to pass smoothly from diffusive to drift-dominating regimes. For certain flows, monotonicity properties are established in the vanishing viscosity limit (``asymptotic monotony'') along with second-order accuracy when the grid is refined. Practical benchmarks are displayed to assess the feasibility of the scheme, including the "western currents" with a Navier--Stokes--Coriolis model of ocean circulation. Read More: https://epubs.siam.org/doi/10.1137/17M1151353

Classical results from spectral theory of stationary linear kinetic equations are applied to efficiently approximate two physically relevant weakly nonlinear kinetic models: a model of chemotaxis involving a biased velocity-redistribution integral term, and a Vlasov-Fokker-Planck (VFP) system. Both are coupled to an attractive elliptic equation producing corresponding mean-field potentials. Spectral decompositions of stationary kinetic distributions are recalled, based on a variation of Case's elementary solutions (for the first model) and on a Sturm-Liouville eigenvalue problem (for the second one). Well-balanced Godunov schemes with strong stability properties are deduced. Moreover, in the stiff hydrodynamical scaling, an hybridized algorithm is set up, for which asymptotic-preserving properties can be established under mild restrictions on the computational grid. Several numerical validations are displayed, including the consistency of the VFP model with Burgers-Hopf dynamics on the velocity field after blowup of the macroscopic density into Dirac masses. (C) 2016 Elsevier Inc. All rights reserved.

Well-balanced schemes, nowadays well-known for 1D hyperbolic equations with source terms and systems of balance laws, are extended to strictly parabolic equations, first in 1D, then in 2D on Cartesian computational grids. The construction heavily relies on a particular type of piecewise-smooth interpolation of discrete data known as -splines. In 1D, several types of widely-used discretizations are recovered as particular cases, like the El-Mistikawy-Werle scheme or Scharfetter- Gummel's. Moreover, a distinctive feature of our 2D scheme is that dimensional-splitting never occurs within its derivation, so that all the multi-dimensional interactions are kept at the discrete level. This leads to improved accuracy, as illustrated on several types of drift-diffusion equations.

A model consisting of a kinetic equation for \run-and-tumble" biased bacteria motion, coupled with two reaction-diusion equations for chemical signals, is studied. It displays time- asymptotic propagation at constant velocity, i.e., aggregated travelling (exponential) layers. To capture them for various parameters, a well-balanced setup is based on both \Case's elementary solutions" and L-spline reconstruction. Far from the diusive regime, waves travelling at dierent velocities (bistability) are proved to coexist. Numerics suggest that they are locally asymptotically stable, so that the resulting bifurcation diagram shows counterintuitive features.

A notion of "2D well-balanced" for drift-diffusion is proposed. Exactness at steady-state, typical in 1D, is weakened by aliasing errors when deriving "truly 2D" numerical fluxes from local Green's function. A main ingredient for proving that such a property holds is the optimality of the trapezoidal rule for periodic functions. In accordance with practical evidence, a "Bessel scheme" previously introduced in [SIAM J. Numer. Anal. 56 (2018), pp. 2845-2870] is shown to be "2D well-balanced" (along with former algorithms known as "discrete weighted means" or "tailored schemes". Some L-2 stability estimates are established, too.

Numerical resolution of two-stream kinetic models in a strong aggregative setting is considered. To illustrate our approach, we consider a one-dimensional kinetic model for chemotaxis in hydrodynamic scaling and the high field limit of the Vlasov-Poisson-Fokker-Planck system. A difficulty is that, in this aggregative setting, weak solutions of the limiting model blow up in finite time, and therefore the scheme should be able to handle Dirac measures. It is overcome thanks to a careful discretization of the macroscopic velocity resulting of Vol'pert calculus: accordingly, a new well-balanced and asymptotic preserving numerical scheme is provided. Numerical simulations confirm a good behavior of solutions.

Sharp and local L-1 a posteriori error estimates are established for so-called "well-balanced" BV (hence possibly discontinuous) numerical approximations of 2 x 2 space-dependent Jin-Xin relaxation systems under sub-characteristic condition. According to the strength of the relaxation process, one can distinguish between two complementary regimes: 1) a weak relaxation, where local L-1 errors are shown to be of first order in Delta x and uniform in time, 2) a strong relaxation, where numerical solutions are kept close to entropy solutions of the reduced scalar conservation law, and for which Kuznetsov's theory indicates a behavior of the L1 error in t center dot root Delta x. The uniformly first-order accuracy in weak relaxation regime is obtained by carefully studying interaction patterns and building up a seemingly original variant of Bressan Liu Yang's functional, able to handle BV solutions of arbitrary size for these particular inhomogeneous systems. The complementary estimate in strong relaxation regime is proven by means of a suitable extension of methods based on entropy dissipation for space-dependent problems. Preliminary numerical illustrations are provided.

Well-balanced schemes were introduced to numerically enforce consistency with longtime behavior of the underlying continuous PDE. When applied to linear kinetic models, like the Goldstein-Taylor system, this construction generates discretizations which are inconsistent with the hydrodynamic stiff limit (despite it captures diffusive limits quite well). A numerical hybridization, taking advantage of both time-splitting (TS) and well-balanced (WB) approaches is proposed in order to fix this defect: numerical results show that resulting composite schemes improve rendering of macroscopic fluxes while keeping a correct hydrodynamic stiff limit.

MUSCL extensions (Monotone Upstream-centered Schemes for Conservation Laws) of the Godunov numerical scheme for scalar conservation laws are shown to admit a rather simple reformulation when recast in the formalism of the Haar multi-resolution analysis of L²(R). By pursuing this wavelet reformulation, a seemingly new MUSCL-WB scheme is derived for advection-reaction equations which is stable for a Courant number up to 1 (instead of roughly 1/2 ). However these highorder reconstructions aren't likely to improve the handling of delicate nonlinear wave interactions in the involved case of systems of Conservation/Balance laws.

An elementary model of (1 + 1)-dimensional general relativity, known as "R = T " and mainly developed by Mann and coworkers in the early 1990s, is set up in various contexts. Its formulation, mostly in isothermal coordinates, is derived and a relativistic Euler system of selfgravitating gas coupled to a Liouville equation for the metric's conformal factor is deduced. First, external field approximations are carried out: both a Klein-Gordon equation is studied along with its corresponding density, and a Dirac one inside a hydrostatic gravitational field induced by a static, piecewise constant mass repartition. Finally, the coupled Euler-Liouville system is simulated, by means of a locally inertial Godunov scheme: the gravitational collapse of a static random initial distribution of density is displayed. Well-balanced discretizations rely on the treatment of source terms at each interface of the computational grid, hence the metric remains flat in every computational cell.

This monograph presents, in an attractive and self-contained form, techniques based on the L1 stability theory derived at the end of the 1990s by A. Bressan, T.-P. Liu and T. Yang that yield original error estimates for so-called well-balanced numerical schemes solving 1D hyperbolic systems of balance laws. Rigorous error estimates are presented for both scalar balance laws and a position-dependent relaxation system, in inertial approximation. Such estimates shed light on why those algorithms based on source terms handled like "local scatterers" can outperform other, more standard, numerical schemes. Two-dimensional Riemann problems for the linear wave equation are also solved, with discussion of the issues raised relating to the treatment of 2D balance laws. All of the material provided in this book is highly relevant for the understanding of well-balanced schemes and will contribute to future improvements.