Introduction
 342 Downloads
Abstract
In this chapter we introduce the topic of the book, namely the class of partial differential equations on which it is focused and an Outline of the presented material.
Keywords
Hyperbolic system of balance laws Accuracy of schemes for balance laws1.1 Some General Perspective
Even if postWWII supersonic aircrafts development considerably stimulated both the understanding and practical approximation schemes for inviscid fluid mechanics equations, multidimensional systems of conservation laws [9], the decision to massively invest into the computational treatment of Euler equations appears to be a bit more recent, roughly speaking the 80s, when the shortcomings of popular widelyused models based on potential flow hypotheses were revealed during the socalled “Stockholm Olympics”, see [29]. A previous success for (hypersonic) computational fluid dynamics was the correct design of the thermal shield on the Apollo spatial vehicles for the atmosphere reentry phase, based on a multiD extension of the Godunov scheme produced in 1959 for inviscid Euler equations, [14, 35]. Oppositely, during the 80s, the Airbus 320 transonic airfoils were still designed by means of a potential flow approximation, i.e. a scalar equation for which general conservation of mass, momentum and energy is not completely ensured.
1.1.1 Inviscid Systems of Conservation Laws
Obviously, an heavy price inherent to switching from the computational simulation of an elementary, possibly fitted, model to the handling of a considerably more complex system of equations is the design of reliable, realizable and efficient numerical algorithms. Presently, multidimensional inviscid Euler equations appear to be replete of difficulty of all kinds, like the mathematical ones in order to give some rigorous sense to weak discontinuous solutions containing shock waves, [5, 26], but also some unexpected ones, like several nonuniqueness and illposedness phenomena recently established for 2D inviscid systems containing vorticity , [8, 12]. It came with no surprise that finescale structures concomitant with vorticity developing in the solution may result into a destabilizing factor in numerical computations; less expected was the fact that it overcomes the stabilizing effect of entropy dissipation, even at the mathematical level of continuous equations. Paradoxically, some authors begin to take a step back and advocate some potential flow models: see [11].
Besides, left apart illposedness pathologies, numerical issues may pop up immediately when one plans to compute multidimensional flows endowed with discontinuous shock waves by means of finitedifferences on a fixed grid: P.L. Roe calls the example of a static shear flow not aligned with the grid (i.e. an oblique contact discontinuity) the “case against upwinding”: see [19, p. 345] and Fig. 6.7. Examples of this kind sparkled many developments and algorithmic advances, some of which were consigned in the book [20], see also [36]. Even if many pathologies still remain in existing algorithms, some signs suggest that algorithmic innovation slowed down so much that it practically came to a stop, [31], the main driving force behind recent computational progress being mostly an increase in CPU power.
1.1.2 OneDimensional Systems
Enormous difficulties coming from the multidimensional characters of compressible flows [13] can be somewhat circumvented by following Godunov’s ideas, namely narrowing the scope only to onedimensional models within a dimensionalsplitting perspective (see e.g. [17, 24]). Necessity knows no law. A salient feature in onedimensional inviscid models is the availability of an efficient tool for the resolution of discontinuities, the Riemann problem and its “simple waves” endowed with neat directions of propagation; such a buildingblock gave rises to many timemarching approximation procedures, which all rely on resolving discontinuities (possibly induced through wave interactions ) by correctly distinguishing the nature of left and rightgoing disturbances. A piecewiseconstant approximation of initial data furnishes a convenient input allowing to set up many Riemannbased algorithms, leading to significant theoretical stability results, too [5]. Of course, one should always keep in mind that, apart from specific problems, onedimensional models are a drastic simplification which cannot be thought as truly reflecting the complexity of certain multidimensional inviscid flows [10, 16, 27, 30, 40].
Even in such a simplified context, both theoretical and computational issues remain: left aside the scalar conservation law for which a satisfying theory was proposed by Kružkov [23] and the socalled Temple class systems [6], general 1D strictly hyperbolic systems of conservation laws are usually subject to restrictive smallness assumptions on their (initial, boundary) data [39]. In the realm of such a \(L^1\)stability theory of (small) \({\textit{BV}}\) entropy solutions, one still faces several important obstructions when considering numerical approximation processes, except for a few notable cases, like Godunov’s scheme for Temple class systems [7] (in contrast with [2, 3]), or the wavefront tracking algorithm [28], a convenient convergence theory of widelyused schemes is still not available. Worse, obstacles appear to be not just technical, as some quite practical examples were produced, displaying for instance wavecurve deformations as a result of numerical viscosity [1, 22, 37].
1.1.3 Source Terms Inclusion

Dissipative mechanisms, like relaxation appearing in discrete models of more complex kinetic equations, which effect is to push the system onto an equilibrium manifold where its dynamics are described by a reduced number of variables.

Possibly accretive terms of bounded extent thus naturally positiondependent, yielding a scattering mechanism as they are somewhat negligible at infinity, but create strong interactions with convective waves in the vicinity of the origin.
Numerical techniques for both these categories are intrinsically different: uniform in space, dissipative terms are well handled by operatorsplitting in time algorithms, consisting essentially in alternating the treatment of the homogeneous equations and the ordinary differential system resulting of the presence of the source (which acts more like a sink ) [17]. Oppositely, positiondependent, boundedextent source terms lead, in large times, to a scattering (i.e. noninteracting) state where homogeneous hyperbolic waves propagate far away, but a steadystate wave, resulting from a delicate balancing of convection forces and sources, stands close to the origin. Such a balance is usually not correctly reproduced by a timesplitting strategy, as was already well understood in the 80s [18, 34]. A scattering mechanism is present in homogeneous systems of conservation laws: by strict hyperbolicity, the solution decays toward a set of noninteracting waves. This set of waves identifies with the Riemann problem at infinity, obtained by a “zoomout” rescaling. Yet, in presence of a bounded extent source, such a rescaling wipes off everything except for a Dirac measure in zero inside which all the effects are concentrated. A Smatrix relates incoming and outgoing hyperbolic waves, locally scattered by a “Dirac source” [13].
1.2 This Book in a Nutshell
1.2.1 Outline of the Contents
This book is mostly concerned with quantifying the accuracy of specific numerical approximations of 1D systems of balance laws endowed with positiondependent, possibly accretive source terms, hence decaying in large time onto a scattering state involving a stationary wave. These approximations follow Glimm’s canvas, namely reducing a positiondependent source term to a countable collection of “local scattering centers” by means of Dirac measures (for instance, located at the edges of computational cells in a Cartesian mesh), and proceeding by iteratively solving all the resulting Riemann problems, but this time, including the “standing wave ” (in the terminology of [21]) which results of the “Dirac source scattering ”. Within a Godunov scheme, this procedure is mostly referred to as wellbalanced scheme [15] whereas in the context of homogenization of periodically forced scalar laws, one speaks more willingly about a generalized Glimm scheme [38]. During the last decade, such a numerical strategy was very successful on practical computations for shallow water equations endowed with steep topography source terms (see [4] and ( 2.11); hence the natural question “is there a rigorous mathematical explanation for this class of schemes outperforming the standard ones ?” to which the forthcoming chapters aim at (partly) answer in a unified, selfcontained manner.

Chapter 2 deals with several forms of error estimates for 1D systems of balance laws. Especially, the classical socalled Local Truncation Error (L.T.E.) is introduced, following an analysis of K. Morton. Most “highorder schemes” for conservation laws rely on this (formal) notion of local error for their construction. Since it usually employs Taylor expansions of the solution, its relevance in the context of weak, possibly discontinuous, solutions isn’t obvious: two simple examples (the deformation of shock curves by numerical viscosity and a nonlinear shockrarefaction interaction ) reveal some of its shortcomings. Moreover, such a local error bound should be integrated in time, with the help of convenient uniform bounds, in order to produce a global error estimate which quantifies the actual error separating the approximate solution from its exact counterpart in some norm (usually \(L^1\)): by assuming infinite smoothness for solutions of a linearized shallow water system, it is shown that handling the topography term a(x) as a collection of “local scattering centers” furnishes a much more robust approximation. This elementary calculation is checked on a practical example.

Chapter 3 focuses onto the 1D, possibly accretive , positiondependent scalar balance law ( 3.1). For simplicity, only one numerical algorithm is considered, the wavefronttracking (WFT) [5, 10], for which a general \(L^1\) stability theory was established by means of a specific Lyapunov functional . Such a functional keeps track of the timeevolution of the distance between two approximate WFT solutions emerging from different \({\textit{BV}}\) initial data. It’s easy to see that under some mild hypotheses (one of these being the nonresonance assumption [21, 25]), this functional being uniformly equivalent to the \(L^1\) norm furnishes a reliable error estimate allowing to bypass the Gronwall lemma, hence free from any exponential dependence in time! Another feature of such an analysis is revealed when considering the particular case of a periodically forced scalar law ( 3.42), for which an original variant of the stability functional allows to improve significantly previous estimates. Practical tests are displayed for both cases, too.

Chapter 4 is probably the most ambitious; it explores the possibility to extend the good error estimates of the former one, from the scalar equation toward a semilinear, positiondependent system endowed with a relaxationtype source term. Systems of this type can be met in many areas of application, left aside wellknown relaxation approximations to scalar conservation laws. Being semilinear, it allows for the derivation of an original, timedecaying, Lyapunov functional which uniformly controls the global \(L^1\) error of the numerical process. Such a functional is absolutely specific to the “local scattering centers” procedure and it’s rather easy to produce examples for which one sees without doubt that it outperforms the Kuznetsovtype estimates which govern the error of more standard discretizations. This fact is checked numerically, too, in Chap. 5.

Last but not least, Chap. 6 aims at giving hints about more complex models for which our analysis may be extended: such systems may include weakly nonlinear models involving a coupling with a selfconsistent Poisson equation (rendering electrostatic interactions, biological confinement or gravity forces) or socalled Temple class systems met for instance in the context of traffic flow modeling. A widely open question remains 2D applications: as Chap. 4 was largely concerned with a semilinear problems being lowerorder perturbations of the 1D wave equation, the Riemann problem for the 2D linear wave equation in variables p, u, v is studied: an analytical expression of its exact solution is given and displayed on Fig. 6.5. Very complex wave interactions appear inside the Kirchhoff disc as soon as initial data are endowed with some vorticity . This is the basic building block in order to construct a genuinely 2D Godunov scheme.
1.2.2 The Ariadne’s Thread
 1.
Glimm’s method of concentrating bounded extent, positiondependent source terms as “local scattering centers” arranged on a discrete lattice consists essentially in treating them by means of supplementary wave interactions .
 2.
If the source term reads k(x)g(u), it is convenient to rewrite it \(g(u)\partial _x a\), being a(x) an antiderivative of k satisfying the trivial equation \(\partial _t a=0\). One immediately deduces that u, a solve an homogeneous system, with a supplementary (immobile) characteristic family, the “standing wave ” in [21] (see also [25, 33]).
 3.
Under the nonresonance assumption, such an augmented system is strictly hyperbolic hence admits a Lyapunov functional decaying along convenient pair of \({\textit{BV}}\) solutions, say u, a and v, b. This decay is proved without invoking the Gronwall lemma, so no timeexponentials are involved. The source term’s effects are generally handled by means of an interaction potential, a classical object in Glimm’s theory which may impose smallness restrictions, though.
 4.
Being that Lyapunov functional uniformly equivalent with the \(L^1\) norm, a global error estimate is easily deduced. However, since it depends only on the \({\textit{BV}}\) norms of the data, mechanically it perceives only the \(L^1\) norm of the positiondependent coefficient. For instance, the presence of k(x)g(u) produces an error depending only of \(\Vert k\Vert _{L^1}\), but not on any of its derivatives. This is most probably the origin of the accuracy on shallow water models endowed with a steeply varying topography term: the global error are insensitive to its oscillations (such a remark is relevant in homogenization of oscillating balance laws [38], too).
 5.
The nonresonance assumption is not to be taken lightly: a formal analysis of constants showing up in the error estimate for balance laws reveals a term like \(g(u)/f'(u)\), being \(f'(u)\) the velocity field. Such a quantity blows up if the augmented system ceases to be strictly hyperbolic [21], so one must expect the accuracy to decrease in the vicinity of sonic or stagnation points. See for instance Fig. 3.3, where an “error spike” is located at a point where \(f'(u)\simeq 0\).
References
 1.M. Arora, P.L. Roe, On postshock oscillations due to capturing schemes in unsteady flows. J. Comput. Phys. 130, 25–40 (1997)Google Scholar
 2.P. Baiti, A. Bressan, H.K. Jenssen, An instability of the Godunov scheme. Commun. Pure Appl. Math. 59, 1604–1638 (2006)MathSciNetCrossRefzbMATHGoogle Scholar
 3.S. Bianchini, \(BV\) solutions of the semidiscrete upwind scheme. Arch. Rational Mech. Anal. 167, 1–81 (2003)MathSciNetCrossRefzbMATHGoogle Scholar
 4.F. Bouchut, Nonlinear Stability of Finite Volume Methods for Hyperbolic Conservation Laws, and Wellbalanced Schemes for Sources. Frontiers in Mathematics Series (Birkhäuser, Basel, 2004). ISBN 3764366656Google Scholar
 5.A. Bressan, Hyperbolic Systems of Conservation Laws—The OneDimensional Cauchy Problem. Oxford Lecture Series in Mathematics and its Applications, vol. 20 (Oxford University Press, Oxford, 2000)Google Scholar
 6.A. Bressan, P. Goatin, Stability of \(L^\infty \) solutions of Temple class systems. Differ. Integral Equ. 13, 1503–1528 (2000)MathSciNetzbMATHGoogle Scholar
 7.A. Bressan, H.K. Jenssen, On the convergence of Godunov scheme for straight line nonlinear hyperbolic systems. Chin. Ann. Math. (CAM) 21, 269–284 (2000)CrossRefzbMATHGoogle Scholar
 8.E. Chiodaroli, A counterexample to wellposedness of entropy solutions to the compressible Euler system. J. Hyperbolic Differ. Equ. 11(2014), 493–519 (2014)MathSciNetCrossRefzbMATHGoogle Scholar
 9.R. Courant, K.O. Friedrichs, Supersonic Flow and Shock Waves (Springer, New York, 1976)CrossRefzbMATHGoogle Scholar
 10.C.M. Dafermos, Hyperbolic Conservation Laws in Continuum Physics, 3rd edn. (Springer, Heidelberg, 2010)CrossRefzbMATHGoogle Scholar
 11.V. Elling, Relative entropy and compressible potential flow. Acta Math. Sci. Ser. B Engl. Ed. 35(4), 763–776 (2015)Google Scholar
 12.V. Elling, The carbuncle phenomenon is incurable. Acta Math. Scientia 29B, 1647–1656 (2009)MathSciNetCrossRefzbMATHGoogle Scholar
 13.J. Glimm, D.H. Sharp, An \(S\)matrix theory for classical nonlinear physics. Found. Phys. 16, 125–141 (1986)MathSciNetCrossRefGoogle Scholar
 14.S.K. Godunov, , V.S. Ryabenkii, Difference Schemes; An Introduction to the Underlying Theory, Studies in Mathematics and Its Applications 19, North Holland (1987)Google Scholar
 15.J.M. Greenberg, A.Y. LeRoux, A wellbalanced scheme for the numerical processing of source terms in hyperbolic equations. SIAM J. Numer. Anal. 33, 1–16 (1996)MathSciNetCrossRefGoogle Scholar
 16.J. Guckenheimer, Shocks and rarefactions in two space dimensions. Arch. Rational Mech. Anal. 59, 281–291 (1975)MathSciNetCrossRefzbMATHGoogle Scholar
 17.H. Holden, K.H. Karlsen, K.A. Lie, N.H. Risebro, Splitting Methods for Partial Differential Equations with Rough Solutions. Analysis and MATLAB Programs. Series of Lectures in Mathematics (European Mathematical Society (EMS), Zürich, 2010)Google Scholar
 18.L. Huang, T.P. Liu, A conservative, piecewisesteady difference scheme for transonic nozzle flow. Comput. Math. Appl. 12A, 377–388 (1986)MathSciNetCrossRefzbMATHGoogle Scholar
 19.M.Y. Hussaini, A. Kumar, M.D. Salas (eds.), Algorithmic Trends in Computational Fluid Dynamics (English) (Springer, Berlin, 1993)Google Scholar
 20.M.Y. Hussaini, B. van Leer, J. Van Rosendale (eds.), Upwind and HighResolution Schemes (English) (Springer, Berlin, 1997)Google Scholar
 21.E. Isaacson, B. Temple, Convergence of the \(2 \times 2\) Godunov method for a general resonant nonlinear balance law. SIAM J. Appl. Math. 55, 625–640 (1995)MathSciNetCrossRefzbMATHGoogle Scholar
 22.S. Jin, J.G. Liu, The effects of numerical viscosities. I. Slowly moving shocks. J. Comput. Phys. 126, 373–389 (1996)MathSciNetCrossRefzbMATHGoogle Scholar
 23.S.N. Kružkov, First order quasilinear equations with several independent variables. Math. Sbornik (N.S.) 81(123), 228255 (1970)MathSciNetGoogle Scholar
 24.R.J. LeVeque, Numerical Methods for Conservation Laws. ETH Zurich (Birkhauser, Basel, 1992)Google Scholar
 25.C. Li, T.P. Liu, Asymptotic states for hyperbolic conservation laws with a moving source. Adv. Appl. Math. 4, 353–379 (1983)MathSciNetCrossRefzbMATHGoogle Scholar
 26.A. Majda, Compressible Flow in Several Space Variables (Springer, New York, 1984)CrossRefzbMATHGoogle Scholar
 27.J. Rauch, BV estimates fail for most quasilinear hyperbolic systems in dimensions greater than one. Commun. Math. Phys. 106, 481–484 (1986)MathSciNetCrossRefzbMATHGoogle Scholar
 28.N.H. Risebro, A front tracking alternative to the random choice method. Proc. Am. Math. Soc. 117, 1125–1139 (1993)MathSciNetCrossRefzbMATHGoogle Scholar
 29.A. Rizzi, H. Viviand (eds.), Numerical Methods for the Computation of Inviscid Transonic Flows with Shock Waves. Notes on Numerical Fluid Mechanics Series, vol. 3 (Vieweg Verlag, Braunschweig, 1981)Google Scholar
 30.P.L. Roe, Discrete models for the numerical analysis of timedependent multidimensional gas dynamics. J. Comput. Phys. 63, 458 (1986)MathSciNetCrossRefzbMATHGoogle Scholar
 31.P.L. Roe, Computational fluid dynamics—retrospective and prospective. Int. J. Comput. Fluid Dyn. 19(8), 581–594 (2005)MathSciNetCrossRefzbMATHGoogle Scholar
 32.
 33.G.A. Sod, A random choice method with application to reactiondiffusion systems in combustion. Comput. Math. Appl. 11, 129–144 (1985)MathSciNetCrossRefGoogle Scholar
 34.G.A. Sod, A numerical study of oxygen diffusion in a spherical cell with the MichaelisMenten oxygen uptake kinetics. J. Math. Biol. 24, 279–289 (1986)CrossRefzbMATHGoogle Scholar
 35.T.D. Taylor, B.S. Masson, Application of the unsteady numerical method of Godunov to computation of supersonic flows past bell shaped bodies. J. Comput. Phys. 5, 443–454 (1970)CrossRefzbMATHGoogle Scholar
 36.E.F. Toro, Riemann Solvers and Numerical Methods for Fluid Dynamics: A Practical Introduction, 3rd edn. (Springer, Berlin, 2009)CrossRefzbMATHGoogle Scholar
 37.J. Von Neumann, R.D. Richtmyer, A method for the numerical calculation of hydrodynamic shocks. J. Appl. Phys. 21, 232–237 (1950)MathSciNetCrossRefzbMATHGoogle Scholar
 38.E. Weinan, Homogenization of scalar conservation laws with oscillatory forcing terms. SIAM J. Appl. Math. 52, 959–972 (1992)MathSciNetCrossRefzbMATHGoogle Scholar
 39.R. Young, On elementary interactions for hyperbolic conservation laws. Unpublished (1993). http://www.math.umass.edu/~young/Research/misc/elem.pdf
 40.Y. Zheng, Systems of Conservation Laws: TwoDimensional Riemann Problems. Progress in Nonlinear Differential Equations and Their Applications, vol. 38 (Birkhauser Verlag, Boston, 2001)Google Scholar