Skip to main content
Log in

Behavior of periodic solutions of viscous conservation laws under localized and nonlocalized perturbations

  • Published:
Inventiones mathematicae Aims and scope

Abstract

We establish nonlinear stability and asymptotic behavior of traveling periodic waves of viscous conservation laws under localized perturbations or nonlocalized perturbations asymptotic to constant shifts in phase, showing that long-time behavior is governed by an associated second-order formal Whitham modulation system. A key point is to identify the way in which initial perturbations translate to initial data for this formal system, a task accomplished by detailed estimates on the linearized solution operator about the background wave. Notably, our approach gives both a common theoretical treatment and a complete classification in terms of “phase-coupling” or “-decoupling” of general systems of conservation or balance laws, encompassing cases that had previously been studied separately or not at all.

At the same time, our refined description of solutions gives the new result of nonlinear asymptotic stability with respect to localized perturbations in the phase-decoupled case, further distinguishing behavior in the different cases. An interesting technical aspect of our analysis is that for systems of conservation laws the Whitham modulation description is of system rather than scalar form, as a consequence of which renormalization methods such as have been used to treat the reaction-diffusion case in general do not seem to apply.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Similar content being viewed by others

Notes

  1. Verifying formal predictions and rigorous spectral descriptions of [17].

  2. As for example for the Korteweg–de Vries (KdV) equation [26, 29, 62, 68] or Euler–Korteweg system [9].

  3. Concerning the more tractable (since faster-decaying) three and higher dimensional case.

  4. See Remark 1.2 below.

  5. For example, Hamiltonian structure or existence of additional conserved quantities other than \(\bar{q}\) [9, 25, 62].

  6. The needed structural conditions may indeed be deduced from assumptions (H1)–(H3) and (D1)–(D3) below.

  7. These intermediate decay rates require some assumptions about characteristics speeds provided by assumption (H3) below; see Remark 1.21.

  8. Henceforth, in our notation for Lebesgue and Sobolev spaces we will suppress the definition of the range; in particular, we will write \(L^{p}(\mathbb{R})\) for the equivalence class of p-integrable \(\mathbb{R}^{n}\)-valued functions \(L^{p}(\mathbb {R};\mathbb{R}^{n})\).

  9. Mark that the important property we have used through these arguments and obtained from the introduction of Bloch symbols L ξ is compactness, which plays for this periodic setting the role of the finite dimensionality that one obtains with Fourier symbols associated to constant-coefficient operators.

  10. A periodic Evans function computation showing that the (n+1)st derivative of the Evans function at λ=0 is proportional to det α (M,k), hence \(\det\partial_{\alpha}(M,k)_{|(\bar{\alpha},\bar{\beta})}\neq0\) is necessary for (D3) (implicit also in the earlier work [62]).

  11. Full hyperbolicity requiring of course also semisimplicity of a j as eigenvalues of \(\frac{\partial(F-\bar{c}M,-\omega-\bar{c}k)}{\partial(M,k)}|_{(\bar{M} ,\bar{k})}\).

  12. These are not true distances, but rather measures associated with a seminorm.

  13. The proof gives also a \(\delta_{L^{1}\cap H^{K}}-\delta_{H^{K}}\) asymptotic stability.

  14. This issue does not arise in the related analysis [33] of the reaction-diffusion case, as M does not appear.

  15. This follows from the general theory for parabolic systems of conservation laws, see for instance Proposition B.1.

  16. Here we are using the additional fact (not explicitly stated here) that estimate (1.19)(ii) is sharp. On the other hand, we do not expect (1.19)(i) to be sharp (see Remark 2.1).

  17. See Remark 1.2.

  18. See Remark 1.2.

  19. At second order, a linear heat equation κ t = xx , d>0.

  20. Recall that \(\mathcal{M}\) does not appear in the Whitham equation for the reaction-diffusion case.

  21. Up to the contribution of the critical part of the evolution not described by phase modulation but estimated in a similar way.

  22. Here and elsewhere, we denote \(\|g\|_{L^{q}([-\pi,\pi ],L^{p}([0,1]))}:= (\int_{-\pi}^{\pi}\|g(\xi,\cdot)\| _{L^{p}([0,1])}^{q}d\xi )^{1/q}\).

  23. Specifically, [65, Sect. 3] and [66, Sect. 2.2.1] (group invariance and uniqueness), [74, Sect. 3] (translation-invariant center–stable manifold), and [67, Theorem 2.2.0] (Nash–Moser uniqueness theorem).

  24. The first space is the one of U-values, the second one is the one of modulation parameters \(({\mathcal{M}},\kappa)\).

  25. In particular, for the three example systems considered in the introduction, the Whitham system does have large-amplitude solutions for data merely bounded in L 1L 2 since it has an associated convex entropy; see [22].

  26. Compare to the similar but much simpler argument of [32, Proposition 4.1], in the reaction-diffusion case.

  27. Up to the explicit computation of the final convolution kernel, the arguments of the proofs of (3.15)–(3.16) and (3.24)–(3.25) are the ones refined to obtain (4.24).

  28. Compare to the argument of [28, Lemma 4.2], regarding localized perturbations in the decoupled case.

  29. A situation that trivially occurs when some symmetry is present, see Remark 1.27.

  30. Here and elsewhere, we identify as usual bilinear maps with vector-valued matrices, and in particular d 2 g(w ) with Hess(g)(w ). Moreover, for these vector-valued matrices, we use coordinate notations so that for instance \(\varGamma_{*}^{j}\in\mathbb {R}^{n\times n}\) satisfies \(w^{t}\varGamma_{*}^{j}w=(w^{t}\varGamma_{*}w)_{j}=d^{2}g_{j}(w,w)\).

  31. Computing \(\|\theta_{j}(t)\|_{L^{p}(\mathbb{R})}= t^{-1/2}\|\bar{\theta}_{j} (\cdot/\sqrt{t})\|_{L^{p}(\mathbb{R})}\sim t^{-\frac{1}{2}(1-1/p)}\).

  32. Though stated in [33, Lemma 1.2] for scalar equations, the proof applies equally to the system case; see Appendix C.

  33. Linearized analysis suggests that the sharp condition is, rather, some averaged version of this one, which holds trivially by the fact that perfect derivatives have zero mean [4].

  34. It has also been shown numerically that there exist bands of stable periodic Swift–Hohenberg solutions in the parameter space (ω,κ,ε) [7, 44], for |ε| not necessarily small.

References

  1. Balmforth, N.J., Mandre, S.: Dynamics of roll waves. J. Fluid Mech. 514, 1–33 (2004)

    Article  MATH  MathSciNet  Google Scholar 

  2. Barker, B., Johnson, M., Noble, P., Rodrigues, M.: Stability of roll waves of the inclined capillary Saint Venant equations (Work in progress)

  3. Barker, B., Johnson, M., Noble, P., Rodrigues, M., Zumbrun, K.: Nonlinear modulational stability of periodic traveling-wave solutions of the generalized Kuramoto–Sivashinsky equation. Phys. D, Nonlinear Phenom. 258, 11–46 (2013)

    Article  MATH  MathSciNet  Google Scholar 

  4. Barker, B., Johnson, M., Noble, P., Rodrigues, M., Zumbrun, K.: Efficient numerical evaluation of the periodic Evans function of Gardner and spectral stability of periodic viscous roll waves (in preparation)

  5. Barker, B., Johnson, M., Noble, P., Rodrigues, M., Zumbrun, K.: Witham averaged equations and modulational stability of periodic solutions of hyperbolic-parabolic balance laws. In: Journéees Équations aux Dérivées Partielles, June 2010, Port d’Albret, France, pp. 1–24 (2010). Available as http://eudml.org/doc/116384

    Google Scholar 

  6. Barker, B., Lewicka, M., Zumbrun, K.: Existence and stability of viscoelastic shock waves. Arch. Ration. Mech. Anal. 200(2), 491–532 (2011)

    Article  MATH  MathSciNet  Google Scholar 

  7. Barker, B., Johnson, M., Noble, P., Rodrigues, M., Zumbrun, K.: Stability of periodic Kuramoto–Sivashinsky waves. Appl. Math. Lett. 25, 824–829 (2012)

    Article  MATH  MathSciNet  Google Scholar 

  8. Beck, M., Nguyen, T., Sandstede, B., Zumbrun, K.: Toward nonlinear stability of sources via a modified Burgers equation. Physica D 241, 382–392 (2012)

    Article  MATH  Google Scholar 

  9. Benzoni-Gavage, S., Noble, P., Rodrigues, M.: Slow modulations of periodic waves in capillary fluids (submitted). arXiv:1303.6467

  10. Bertozzi, A., Münch, A., Fanton, X., Cazabat, A.M.: Contact line stability and ‘undercompressive shocks’ in driven thin film flow. Phys. Rev. Lett. 8(23), 5169–5172 (1998)

    Article  Google Scholar 

  11. Bertozzi, A., Münch, A., Shearer, M., Zumbrun, K.: Stability of compressive and undercompressive thin film travelling waves. Eur. J. Appl. Math. 12(3), 253–291 (2001)

    Article  MATH  Google Scholar 

  12. Chang, H.C., Demekhin, E.A., Kopelevich, D.I.: Laminarizing effects of dispersion in an active-dissipative nonlinear medium. Physica D 63, 299–320 (1993)

    Article  MATH  Google Scholar 

  13. Collet, P., Eckmann, J.-P.: The stability of modulated fronts. Helv. Phys. Acta 60, 969–991 (1987)

    MATH  MathSciNet  Google Scholar 

  14. Collet, P., Eckmann, J.-P.: Instabilities and Fronts in Extended Systems. Princeton University Press, Princeton (1990)

    MATH  Google Scholar 

  15. Doelman, A., Sandstede, B., Scheel, A., Schneider, G.: The Dynamics of Modulated Wavetrains. Mem. Am. Math. Soc., vol. 199 (2009), no. 934, viii+105 pp. ISBN: 978-0-8218-4293-5

    Google Scholar 

  16. Dressler, R.: Mathematical solution of the problem of roll waves in inclined open channels. Commun. Pure Appl. Math. 2, 149–190 (1949)

    Article  MATH  MathSciNet  Google Scholar 

  17. Eckhaus, W.: Studies in Nonlinear Stability Theory. Springer Tracts in Nat. Phil., vol. 6 (1965)

    Book  Google Scholar 

  18. Frisch, U., She, Z.S., Thual, O.: Viscoelastic behaviour of cellular solutions to the Kuramoto–Sivashinsky model. J. Fluid Mech. 168, 221–240 (1986)

    Article  MATH  MathSciNet  Google Scholar 

  19. Gardner, R.: On the structure of the spectra of periodic traveling waves. J. Math. Pures Appl. 72, 415–439 (1993)

    MATH  MathSciNet  Google Scholar 

  20. Häcker, T., Schneider, G., Zimmermann, D.: Justification of the Ginzburg-Landau approximation in case of marginally stable long waves. J. Nonlinear Sci. 21(1), 93–113 (2011)

    Article  MATH  MathSciNet  Google Scholar 

  21. Henry, D.: Geometric Theory of Semilinear Parabolic Equations. Lecture Notes in Mathematics. Springer, Berlin (1981)

    MATH  Google Scholar 

  22. Hoff, D., Smoller, J.: Global existence for systems of parabolic conservation laws in several space variables. J. Differ. Equ. 68, 210–220 (1987)

    Article  MATH  MathSciNet  Google Scholar 

  23. Hoff, D., Zumbrun, K.: Asymptotic behavior of multidimensional scalar viscous shock fronts. Indiana Univ. Math. J. 49(2), 427–474 (2000)

    Article  MATH  MathSciNet  Google Scholar 

  24. Howard, L.N., Kopell, N.: Slowly varying waves and shock structures in reaction-diffusion equations. Stud. Appl. Math. 56(2), 95–145 (1976/77)

    MathSciNet  Google Scholar 

  25. Johnson, M., Zumbrun, K.: Nonlinear stability of periodic traveling waves of viscous conservation laws in the generic case. J. Differ. Equ. 249(5), 1213–1240 (2010)

    Article  MATH  MathSciNet  Google Scholar 

  26. Johnson, M., Zumbrun, K.: Rigorous justification of the Whitham modulation equations for the generalized Korteweg–de Vries equation. Stud. Appl. Math. 125(1), 69–89 (2010)

    MATH  MathSciNet  Google Scholar 

  27. Johnson, M., Zumbrun, K.: Nonlinear stability of spatially-periodic traveling-wave solutions of systems of reaction diffusion equations. Ann. Inst. Henri Poincaré, Anal. Non Linéaire 28(4), 471–483 (2011)

    Article  MATH  MathSciNet  Google Scholar 

  28. Johnson, M., Zumbrun, K.: Nonlinear stability and asymptotic behavior of periodic traveling waves of multidimensional viscous conservation laws in dimensions one and two. SIAM J. Appl. Dyn. Syst. 10(1), 189–211 (2011)

    Article  MATH  MathSciNet  Google Scholar 

  29. Johnson, M., Zumbrun, K., Bronski, J.: Bloch wave expansion vs. Whitham modulation equations for the generalized Korteweg–de Vries equation. J. Differ. Equ. 249(5), 1213–1240 (2010)

    Article  MATH  Google Scholar 

  30. Johnson, M., Zumbrun, K., Noble, P.: Nonlinear stability of viscous roll waves. SIAM J. Math. Anal. 43(2), 557–611 (2011)

    Article  MathSciNet  Google Scholar 

  31. Johnson, M., Noble, P., Rodrigues, L.M., Zumbrun, K.: Spectral stability of periodic wave trains of the Korteweg–de Vries/Kuramoto–Sivashinsky equation in the Korteweg–de Vries limit. Preprint (2012)

  32. Johnson, M., Noble, P., Rodrigues, L.M., Zumbrun, K.: Nonlocalized modulation of periodic reaction diffusion waves: nonlinear stability. Arch. Ration. Mech. Anal. 207(2), 693–715 (2013)

    Article  MATH  MathSciNet  Google Scholar 

  33. Johnson, M., Noble, P., Rodrigues, L.M., Zumbrun, K.: Nonlocalized modulation of periodic reaction diffusion waves: the Whitham equation. Arch. Ration. Mech. Anal. 207(2), 669–692 (2013)

    Article  MATH  MathSciNet  Google Scholar 

  34. Kato, T.: Perturbation Theory for Linear Operators. Springer, Berlin (1985)

    Google Scholar 

  35. Kawashima, S.: Large-time behaviour of solutions to hyperbolic-parabolic systems of conservation laws and applications. Proc. R. Soc. Edinb., Sect. A 106(1–2), 169–194 (1987)

    Article  MATH  MathSciNet  Google Scholar 

  36. Kotschote, M.: Dynamics of compressible non-isothermal fluids of non-Newtonian Korteweg-type. SIAM J. Math. Anal. 44(1), 74–101 (2012)

    Article  MATH  MathSciNet  Google Scholar 

  37. Kwon, B., Zumbrun, K.: Asymptotic behavior of multidimensional scalar relaxation shocks. J. Hyperbolic Differ. Equ. 6(4), 663–708 (2009)

    Article  MATH  MathSciNet  Google Scholar 

  38. Kuramoto, Y., Tsuzuki, T.: On the formation of dissipative structures in reaction-diffusion systems. Prog. Theor. Phys. 54, 3 (1975)

    Google Scholar 

  39. Lax, P.D.: Hyperbolic Systems of Conservation Laws and the Mathematical Theory of Shock Waves. Conference Board of the Mathematical Sciences Regional Conference Series in Applied Mathematics, vol. 11. Society for Industrial and Applied Mathematics, Philadelphia (1973). v+48 pp.

    Book  MATH  Google Scholar 

  40. Liu, T.-P.: Interaction of nonlinear hyperbolic waves. In: Nonlinear Analysis, Taipei, pp. 171–183 (1989)

    Google Scholar 

  41. Liu, T.-P., Zeng, Y.: Large Time Behavior of Solutions for General Quasilinear Hyperbolic–Parabolic Systems of Conservation Laws. AMS Memoirs, vol. 599 (1997)

    Google Scholar 

  42. Mascia, C., Zumbrun, K.: Stability of large-amplitude viscous shock profiles of hyperbolic–parabolic systems. Arch. Ration. Mech. Anal. 172(1), 93–131 (2004)

    Article  MATH  MathSciNet  Google Scholar 

  43. Matar, O.K., Troian, S.M.: Dynamics and stability of surfactant coated thin spreading films. Mater. Res. Soc. Symp. Proc. 464, 237–242 (1997)

    Article  Google Scholar 

  44. Mielke, A.: A new approach to sideband-instabilities using the principle of reduced instability. In: Nonlinear Dynamics and Pattern Formation in the Natural Environment. Pitman Res. Notes Math. Ser., vol. 335, pp. 206–222 (1995)

    Google Scholar 

  45. Nguyen, T., Zumbrun, K.: Long-time stability of multi-dimensional noncharacteristic viscous boundary layers. Commun. Math. Phys. 299(1), 1–44 (2010)

    Article  MATH  MathSciNet  Google Scholar 

  46. Noble, P.: On the spectral stability of roll waves. Indiana Univ. Math. J. 55, 795–848 (2006)

    Article  MATH  MathSciNet  Google Scholar 

  47. Noble, P.: Linear stability of viscous roll waves. Commun. Partial Differ. Equ. 32(10–12), 1681–1713 (2007)

    Article  MATH  MathSciNet  Google Scholar 

  48. Noble, P., Rodrigues, M.: Whitham’s modulation equations for shallow flows. Unpublished manuscript (2010). arXiv:1011.2296

  49. Noble, P., Rodrigues, M.: Whithams modulation equations and stability of periodic wave solutions of the generalized Kuramoto-Sivashinsky equations. Indiana Univ. Math. J. (to appear)

  50. Oh, M., Zumbrun, K.: Stability of periodic solutions of viscous conservation laws with viscosity—1. Analysis of the Evans function. Arch. Ration. Mech. Anal. 166(2), 99–166 (2003)

    Article  MATH  MathSciNet  Google Scholar 

  51. Oh, M., Zumbrun, K.: Stability of periodic solutions of viscous conservation laws with viscosity—pointwise bounds on the Green function. Arch. Ration. Mech. Anal. 166(2), 167–196 (2003)

    Article  MATH  MathSciNet  Google Scholar 

  52. Oh, M., Zumbrun, K.: Low-frequency stability analysis of periodic traveling-wave solutions of viscous conservation laws in several dimensions. Z. Anal. Anwend. 25, 1–21 (2006)

    Article  MATH  MathSciNet  Google Scholar 

  53. Oh, M., Zumbrun, K.: Stability and asymptotic behavior of traveling-wave solutions of viscous conservation laws in several dimensions. Arch. Ration. Mech. Anal. 196(1), 1–20 (2010). Erratum: Arch. Ration. Mech. Anal. 196(1), 21–23 (2010)

    Article  MATH  MathSciNet  Google Scholar 

  54. Pego, R., Schneider, H., Uecker, H.: Long-time persistence of Korteweg–de Vries solitons as transient dynamics in a model of inclined film flow. Proc. R. Soc. Edinb. A 137, 133–146 (2007)

    Article  MATH  MathSciNet  Google Scholar 

  55. Pogan, A., Scheel, A., Zumbrun, K.: Quasi-gradient systems, modulational dichotomies, and stability of spatially periodic patterns. Preprint (2012)

  56. Prüss, J.: On the spectrum of C 0-semigroups. Trans. Am. Math. Soc. 284(2), 847–857 (1984)

    Article  MATH  Google Scholar 

  57. Sandstede, B., Scheel, A., Schneider, G., Uecker, H.: Diffusive mixing of periodic wave trains in reaction-diffusion systems. J. Differ. Equ. 252(5), 3541–3574 (2012)

    Article  MATH  MathSciNet  Google Scholar 

  58. Schecter, S., Shearer, M.: Transversality for undercompressive shocks in Riemann problems. In: Viscous Profiles and Numerical Methods for Shock Waves, Raleigh, NC, 1990, pp. 142–154. SIAM, Philadelphia (1991)

    Google Scholar 

  59. Schneider, G.: Diffusive stability of spatial periodic solutions of the Swift–Hohenberg equation. Commun. Math. Phys. 178(3), 679–702 (1996) (English summary)

    Article  MATH  Google Scholar 

  60. Schneider, G.: Nonlinear diffusive stability of spatially periodic solutions—abstract theorem and higher space dimensions. In: Proceedings of the International Conference on Asymptotics in Nonlinear Diffusive Systems, Sendai, 1997. Tohoku Math. Publ., vol. 8, pp. 159–167. Tohoku Univ., Sendai (1998)

    Google Scholar 

  61. Schneider, G.: Nonlinear stability of Taylor vortices in infinite cylinders. Arch. Ration. Mech. Anal. 144(2), 121–200 (1998)

    Article  MATH  Google Scholar 

  62. Serre, D.: Spectral stability of periodic solutions of viscous conservation laws: large wavelength analysis. Commun. Partial Differ. Equ. 30(1–3), 259–282 (2005)

    Article  MATH  Google Scholar 

  63. Sivashinsky, G.I.: Nonlinear analysis of hydrodynamic instability in laminar flame. I. Derivation of basic equations. Acta Astron. 4(11–12), 1177–1206 (1977)

    Article  MATH  MathSciNet  Google Scholar 

  64. Smoller, J.: Shock Waves and Reaction–Diffusion Equations, 2nd edn. Grundlehren der Mathematischen Wissenschaften (Fundamental Principles of Mathematical Sciences), vol. 258. Springer, New York (1994). xxiv+632 pp. ISBN: 0-387-94259-9

    Book  MATH  Google Scholar 

  65. Texier, B., Zumbrun, K.: Relative Poincaré–Hopf bifurcation and galloping instability of traveling waves. Methods Appl. Anal. 12(4), 349–380 (2005)

    MATH  MathSciNet  Google Scholar 

  66. Texier, B., Zumbrun, K.: Galloping instability of viscous shock waves. Physica D 237, 1553–1601 (2008)

    Article  MATH  MathSciNet  Google Scholar 

  67. Texier, B., Zumbrun, K.: Nash–Moser iterates and singular perturbations. Ann. Inst. Henri Poincaré, Anal. Non Linéaire 28(4), 499–527 (2011)

    Article  MATH  MathSciNet  Google Scholar 

  68. Whitham, G.B.: Linear and Nonlinear Waves. Pure and Applied Mathematics (New York). Wiley, New York (1999). Reprint of the 1974 original, A Wiley-Interscience Publication

    Book  MATH  Google Scholar 

  69. Yao, J.: Existence and stability of periodic solutions of the equations of viscoelasticity with strain-gradient effects. Preprint (2011)

  70. Zimmermann, D.: PhD thesis, University of Stuttgart (2011)

  71. Zumbrun, K.: Refined wave-tracking and stability of viscous Lax shocks. Methods Appl. Anal. 7, 747–768 (2000)

    MATH  MathSciNet  Google Scholar 

  72. Zumbrun, K.: Stability of large-amplitude shock waves of compressible Navier–Stokes equations. In: Handbook of Mathematical Fluid Dynamics, vol. III, pp. 311–533. North-Holland, Amsterdam (2004). With an appendix by Helge Kristian Jenssen and Gregory Lyng

    Google Scholar 

  73. Zumbrun, K.: Planar stability criteria for viscous shock waves of systems with real viscosity. In: Marcati, P. (ed.) Hyperbolic Systems of Balance Laws. CIME School Lectures Notes. Lecture Note in Mathematics, vol. 1911. Springer, Berlin (2004)

    Google Scholar 

  74. Zumbrun, K.: Conditional stability of unstable viscous shocks. J. Differ. Equ. 247(2), 648–671 (2009)

    Article  MATH  MathSciNet  Google Scholar 

  75. Zumbrun, K.: Stability and dynamics of viscous shock waves. In: Nonlinear Conservation Laws and Applications. IMA Vol. Math. Appl., vol. 153, pp. 123–167. Springer, New York (2011)

    Chapter  Google Scholar 

  76. Zumbrun, K.: Instantaneous shock location and one-dimensional nonlinear stability of viscous shock waves. Q. Appl. Math. 69(1), 177–202 (2011)

    MATH  MathSciNet  Google Scholar 

Download references

Acknowledgements

K.Z. thanks Björn Sandstede for a number of helpful orienting discussions regarding modulation of periodic reaction-diffusion waves, Guido Schneider for bringing to our attention the treatment of Bénard–Marangoni cells in [20, 70], and Denis Serre for his interest in the subject of modulation of periodic solutions and his contributions through [62] and private and public communications. M.J., P.N., and M.R. thank Indiana University, and K.Z. thanks the École Normale Supérieure, Paris, the University of Paris 13, and the Foundation Sciences Mathématiques de Paris for their hospitality during visits in which this work was partially carried out. Special thanks to David Lannes for his careful reading of the manuscript and many helpful suggestions. Finally, thanks to the anonymous referees for suggestions improving the exposition.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to L. Miguel Rodrigues.

Additional information

Research of M.J. was partially supported under NSF grant no. DMS-1211183 and by the University of Kansas General Research Fund allocation 2302278.

Research of P.N. was partially supported by the French ANR Project no. ANR-09-JCJC-0103-01.

Stay of M.R. in Bloomington was supported by French ANR project no. ANR-09-JCJC-0103-01.

Research of K.Z. was partially supported under NSF grant no. DMS-0300487.

Appendices

Appendix A: Algebraic relations

We record in this appendix some crucial relations obtained by differentiating the profile equations. In order to differentiate, we here consider variable parameters (M,k) rather than fixed values \((\bar{M},\bar{k})\), imposing dependence implicitly through the profile equations (denoting \(\langle a\rangle:=\int_{0}^{1} a\)):

$$ k^2 U''-k \bigl(f(U)\bigr)'+kc U'=0,\quad\langle U\rangle=M. $$
(A.1)

We expand L ξ =L 0+ikξL (1)+(ikξ)2 L (2) with

$$ \begin{aligned} L_0v&=k^2 v''-k\bigl((d f) (U) v\bigr)'+kc v' ; \\ L^{(1)} v&=2kv'-(d f) (U) v+c v ; \\ L^{(2)} v&=v . \end{aligned} $$
(A.2)

Then, by differentiation of (A.1), we obtain

$$ \begin{aligned} &L_0 U'=0 ,\quad \bigl\langle U'\bigr\rangle=0 ; \\ &L_0\partial_M U+kU'\partial_M c=0 ,\quad \langle\partial_M U\rangle ={\rm Id } ; \\ &L_0\partial_k U+kU'\partial_k c+L^{(1)} U'=0,\quad \langle\partial _k U \rangle=0 . \end{aligned} $$
(A.3)

Accordingly, with ω=−kc, using L (1) U′=kU″, we have

$$\bigl\langle d^2 U\bigr\rangle=0 , $$
$$ L_0 \partial_M^2U+2k (\partial_Mc) ( \partial_MU)'-k \bigl[\bigl(d^2f\bigr) (U) (\partial_MU,\partial _MU)\bigr]'=-U' k \partial_M^2c , $$
(A.4)
$$\begin{aligned} \begin{aligned}[b] &L_0 \partial_{kM}^2U+k (\partial_kc) (\partial_MU)'+k (\partial_Mc) ( \partial_kU)'-k \bigl[\bigl(d^2f\bigr) (U) (\partial_kU,\partial_MU)\bigr]' \\ &\quad +k (\partial_MU)''=-U' k \partial_{kM}^2c, \end{aligned} \end{aligned}$$
(A.5)
$$ L_0 \partial_k^2U+2k (\partial_kc) ( \partial_kU)'-k \bigl[\bigl(d^2f\bigr) (U) (\partial_kU,\partial_kU)\bigr]' +2k ( \partial_kU)''=-U' k \partial_k^2c. $$
(A.6)

Appendix B: The Whitham equations and asymptotic equivalence

In this appendix, we explain how to obtain the needed formal averaged modulation system for comparison to our analytical description of asymptotic behavior. This is performed in three steps.

  1. 1.

    First, we develop a direct WKB-like formal approximation. At this stage we obtain a system that may contain harmless irrelevant terms.

  2. 2.

    Next, we use known results about large-time asymptotic behavior of systems of conservation laws about constant states to get a canonical form for the averaged modulation system.

  3. 3.

    Finally, we adapt the system taking into account the fact that the analysis of the main part of the paper is carried out after an implicit nonlinear change of coordinates.

2.1 B.1 Formal asymptotics

Though the full nonlinear analysis may be carried out without distinction between linearly coupled and linearly uncoupled cases, the formal derivation of averaged equations involves resolutions of systems of the form L 0 g=h and therefore requires knowledge of the kernel of L 0. We are thus compelled to provide two separate derivations.

Besides, there are at least two ways to obtain relevant averaged equations. The first one is to develop a full WKB-type expansion as in [48, 49], extending the procedure in [62] to get higher order equations. This method provides the hyperbolic part of the averaged system in a quick way and a nice form. Its main drawback is that it requires a knowledge of the kernel of L 0 for all waves close to the wave under study essentially reducing the scope of the method to the nondegenerate case or to a fully degenerate case where M c would vanish in a neighborhood of the studied wave.Footnote 29 The second method is designed to study dynamics about a given wave, so that it does not suffer from the same flaws; moreover, it is closer to our nonlinear analysis, and yields a semilinear system.

We derive the system for the generic case with the first method and the one for the linearly uncoupled case with the second one. Note that both methods provide averaged systems with diffusion matrices containing terms that are not relevant for our present analysis.

2.1.1 B.1.1 Generic case

To treat the linearly phase-coupled case, we essentially borrow the derivation of [49] for the Korteweg–de Vries/Kuramoto–Sivashinsky equation, a model for which linear phase-coupling is a consequence of assumptions (H1)–(H2) and (D3). In the present derivation, we assume that all the waves involved in the slow-modulation description satisfy (H1)–(H2) and (D3) and are linearly phase-coupled.

Since in this derivation there is no reference wave, thus no privileged frame, we go back to the original equation

$$ u_t + f(u)_x = u_{xx}. $$
(B.1)

We are looking for a formal expansion of a solution u of Eq. (B.1) according to the two-scale ansatz

$$ u(x,t) = U \biggl(\frac{\varPsi(\varepsilon x,\varepsilon t)}{\varepsilon};\varepsilon x,\varepsilon t \biggr) $$
(B.2)

where

$$ U(y,X,T)=\sum_j \varepsilon^j U_j(y;X,T)\quad\textrm{and}\quad\varPsi (X,T)=\sum_j\varepsilon^j \varPsi_j(X,T) , $$
(B.3)

with the functions U and U j being 1-periodic in the y-variable. We insert the ansatz (B.2), (B.3) into (B.1) and collect terms of the same order in ε.

First this yields, with Ω 0= T ϕ 0 and κ 0= X Ψ 0, \(\varOmega_{0} \partial_{y} U_{0}+\kappa_{0} \partial_{y}(f(U_{0})) =\kappa_{0}^{2} \partial _{y}^{2} U_{0}\), which is solved by

$$ \begin{aligned} \varOmega_0(X,T)&=-k_0(X,T) c\bigl({\mathcal{M}}_0(X,T),\kappa_0(X,T)\bigr), \\ U_0(y;X,T)&= U\bigl(y;{\mathcal{M}}_0(X,T), \kappa_0(X,T)\bigr). \end{aligned} $$
(B.4)

We have disregarded in (B.4) the possibility of a phase shift dependent on (X,T) since this is already encoded by Ψ 1. We will have to rule out similar problems of uniqueness in the following steps. At this stage the compatibility condition T X Ψ 0= X T Ψ 0 already yields the first equation of a Whitham’s modulation system:

$$ \partial_T \kappa_0+ \partial_X \bigl(\kappa_0 c({\mathcal{M}}_0, \kappa_0) \bigr) = 0 . $$
(B.5)

In the rest of the derivation, we will use the notations of Proposition 3.1 and Appendix A, with the convention that operators act in y and are associated to the wave profile \(U( \cdot ;{\mathcal{M}}_{0}(X,T),\kappa_{0}(X,T))\). To fix some of the uniqueness issues of the ansatz, we pick, for any (M,k), u adj(⋅;M,k) a generalized zero eigenfunction of \(L_{0}^{*}\) such that 〈u adj, M U |(M,k)〉=0 and 〈u adj,U′(⋅;M,k)〉=1, set \(u_{0}^{\mathit{adj}}(y;X,T)=u^{\mathit{adj}}( \cdot ;{\mathcal{M}}_{0}(X;T),\kappa _{0}(X,T))\) and add to the ansatz the normalizing condition

$$ \bigl\langle u_0^{\mathit{adj}},U_j( \cdot ;X,T)\bigr\rangle= 0,\quad j\neq0. $$
(B.6)

The next step of the identification process gives, with Ω 1= T Ψ 1 and κ 1= X Ψ 1,

$$ \begin{aligned}[b] &\bigl( \varOmega_1+c({\mathcal{M}}_0,\kappa_0) \kappa_1\bigr)\partial_yU_0- \kappa_1 L^{(1)}\partial_yU_0 - L_0 U_1 - L^{(1)}\partial_X U_0 \\ &\quad - \partial_X\kappa_0 L^{(2)} \partial_yU_0 + \partial_T U_0+c({\mathcal{M}}_0,\kappa _0) \partial_X U_0 = 0, \end{aligned} $$
(B.7)

whose solvability condition reads

$$ \partial_T {\mathcal{M}}_0+ \partial_X\bigl(F({\mathcal{M}}_0,\kappa_0) \bigr) = 0, $$
(B.8)

where F denotes the averaged flux F(M,k)=〈f(U(⋅;M,k))〉. To proceed, for arbitrary (M,k) we introduce g k(⋅;M,k), g M(⋅;M,k) solutions of

$$\begin{aligned} &L_0\bigl(g^{k}( \cdot ;M,k)\bigr) \\ &{}\quad =- L^{(1)}\partial_k U_{|(M,k)}- \partial_k F_{|(M,k)} - L^{(2)}U'_{|(M,k)} \\ &\qquad {}- \partial_k U_{|(M,k)} k \partial_k c_{|(M,k)} - (\partial_M U_{|(M,k)}-{\rm Id }) \partial_k F_{|(M,k)} , \end{aligned}$$
(B.9)
$$\begin{aligned} & L_0\bigl(g^{M}( \cdot ;M,k)\bigr) \\ &{}\quad =-L^{(1)}\partial_M U_{|(M,k)}- \partial_M F_{|(M,k)}+c_{|(M,k)}{\rm Id } \\ &\qquad {}-\partial_k U_{|(M,k)} k \partial_M c_{|(M,k)}-(\partial_M U_{|(M,k)}-{\rm Id })[ \partial_M F_{|(M,k)}-c_{|(M,k)}{\rm Id }] \end{aligned}$$
(B.10)

orthogonal to u adj(⋅;M,k) and set g=(g M g k) and \(g_{0}( \cdot ;X,T)= g( \cdot ;{\mathcal{M}}_{0}(X,T),\kappa_{0}(X,T))\). Then with (B.5)–(B.8) and (A.3) Eq. (B.7) reads

$$\begin{aligned} & L_0 \biggl(U_1 -d U_{|({\mathcal{M}}_0,\kappa_0)} ( \cdot ;\tilde{\mathcal{M}}_1,\kappa_1) -g_0 \begin{pmatrix}\partial_X{\mathcal{M}}_0\\\partial_X\kappa_0 \end{pmatrix} \biggr) \\ &\quad {}=\bigl(\varOmega_1 + \kappa_0d c_{|({\mathcal{M}}_0,\kappa_0)} (\tilde {\mathcal{M}}_1,\kappa_1)+c_{|({\mathcal{M}}_0,\kappa_0)} \kappa_1\bigr) \partial_yU_0 \end{aligned}$$
(B.11)

for any choice of \(\tilde{\mathcal{M}}_{1}\). Let us set \({\mathcal{M}}_{1}=\langle U_{1}\rangle\). Choosing \(\tilde{\mathcal{M}}_{1}\) to get

$$ \varOmega_1+\kappa_0 d c({\mathcal{M}}_0,\kappa_0) [\tilde{\mathcal{M}}_1, \kappa_1]+c({\mathcal{M}}_0,\kappa_0) \kappa_1= 0 $$
(B.12)

and normalizing the parametrization, as in Lemma 4.1, to get, for any (M,k),

$$ \bigl\langle u^{\mathit{adj}}( \cdot ;M,k),\partial_k U_{|(M,k)}\bigr\rangle=0, $$
(B.13)

Eq. (B.11) is reduced to

$$\begin{aligned} U_1 =& d U_{|({\mathcal{M}}_0,\kappa_0)} (\tilde{\mathcal{M}}_1, \kappa_1) + g_0 \begin{pmatrix}\partial_X{\mathcal{M}}_0\\\partial_X\kappa_0 \end{pmatrix}, \\ {\mathcal{M}}_1 =& \tilde{\mathcal{M}}_1 + \langle g_0\rangle \begin{pmatrix}\partial_X{\mathcal{M}}_0\\ \partial_X\kappa_0 \end{pmatrix} . \end{aligned}$$

Then, compatibility condition T κ 1= X Ω 1 yields

$$ \begin{aligned}[b] & \partial_T \kappa_1+ \partial_X\bigl(\kappa_0d c_{|({\mathcal{M}}_0,\kappa_0)} [{\mathcal{M}}_1,\kappa_1]+ c_{|({\mathcal{M}}_0,\kappa_0)} \kappa_1\bigr) \\ &\quad=\partial_X \biggl(\kappa_0\partial_Mc_{|({\mathcal{M}}_0,\kappa _0)} \langle g_0\rangle \begin{pmatrix}\partial_X{\mathcal{M}}_0\\\partial_X\kappa_0 \end{pmatrix} \biggr). \end{aligned} $$
(B.14)

Returning to the identification process, we obtain an equation of the form

$$\partial_T U_1 + \partial_X \bigl(df(U_0) U_1\bigr) - \partial_X^2U_0 - L_0U_2 + \partial_y( \cdots ) = 0, $$

whose solvability condition is

$$ \begin{aligned}[b] &\partial_T{ \mathcal{M}}_1 + \partial_X\bigl(d F_{|({\mathcal{M}}_0,\kappa _0)} [{ \mathcal{M}}_1,\kappa_1]\bigr)\\ &\quad = \partial_X^2{\mathcal{M}}_0 - \partial_X \biggl(\bigl\langle df(U_0) g_0 \bigr\rangle \begin{pmatrix}\partial_X{\mathcal{M}}_0\\\partial_X\kappa_0 \end{pmatrix} \biggr) \\ &\qquad + \partial_X \biggl(\partial_MF_{|({\mathcal{M}}_0,\kappa_0)} \langle g_0\rangle \begin{pmatrix}\partial_X{\mathcal{M}}_0\\\partial_X\kappa_0 \end{pmatrix} \biggr). \end{aligned} $$
(B.15)

To write the second order system in a compact form, let us introduce, for arbitrary (M,k),

$$\begin{aligned} d_{1,1}(M,k)&= {\rm Id }-\bigl\langle df \bigl(U(M,k)\bigr) g^M(M,k)\bigr\rangle+\partial_MF(M,k) \bigl\langle g^M(M,k)\bigr\rangle, \\ d_{1,2}(M,k)&= -\bigl\langle df\bigl(U(M,k)\bigr) g^k(M,k) \bigr\rangle+\partial_MF(M,k) \bigl\langle g^k(M,k)\bigr \rangle, \\ d_{2,1}(M,k)&= k\partial_Mc (M,k) \bigl\langle g^M(M,k)\bigr\rangle, \\ d_{2,2}(M,k)&= k\partial_Mc (M,k) \bigl\langle g^k(M,k)\bigr\rangle. \end{aligned} $$

With these notations, systems (B.5), (B.8), (B.14), (B.15) coincide with the first systems obtained in the formal expansion of a solution \(({\mathcal{M}},\kappa)\) of

$$ \left\{ \begin{aligned} &\partial_t {\mathcal{M}}+ \partial_x\bigl(F({\mathcal{M}},\kappa)\bigr)= \partial_x \bigl(d_{1,1}({\mathcal{M}},\kappa) \partial_x{\mathcal{M}}+d_{1,2}({\mathcal{M}},\kappa) \partial_x\kappa \bigr) \\ &\partial_t \kappa+ \partial_x\bigl(\kappa c({\mathcal{M}},\kappa)\bigr)= \partial_x \bigl(d_{2,1}({\mathcal{M}}, \kappa) \partial_x{\mathcal{M}}+d_{2,2}({\mathcal{M}}, \kappa) \partial_x\kappa \bigr) \end{aligned} \right. $$
(B.16)

according to the slow ansatz

$$ ({\mathcal{M}},\kappa) (x,t) = \sum_j \varepsilon^j ({\mathcal{M}}_j,\kappa_j) ( \varepsilon x,\varepsilon t). $$
(B.17)

We call system (B.16) a (second-order) Whitham’s modulation system.

2.1.2 B.1.2 Phase-decoupled case

For the phase-decoupled case, we propose an alternative derivation that would also work for the uncoupled case. We pick a wave of parameters \((\bar{M},\bar{k})\) and assume that it satisfies (H1)–(H2) and (D3) and is linearly phase-decoupled.

We again insert the ansatz (B.2), (B.3) into (B.1) and collect terms of the same order in ε but this time we specialize to \(({\mathcal{M}}_{0},\kappa_{0})=(\bar {M},\bar{k} )\). We keep (B.6) as ansatz normalization and (B.13) as parametrization normalization. The first nontrivial equation is with (Ω 1,κ 1)=( T Ψ 1, X Ψ 1)

$$ \bigl(\varOmega_1+c(\bar{M},\bar{k}) \kappa_1\bigr)\bar{U}'- \kappa_1 L^{(1)}\bar{U}' - L_0 U_1 = 0 $$
(B.18)

which may also be written as

$$L_0 \bigl(U_1 -d U_{|(\bar{M},\bar{k})} ( \cdot ;{\mathcal{M}}_1,\kappa_1) \bigr) =\bigl(\varOmega_1 + \bar{k}\partial_k c(\bar{M},\bar{k})\kappa _1+c(\bar{M}, \bar{k})\kappa _1\bigr)\bar{U}' $$

for any \({\mathcal{M}}_{1}\). Solvability yields

$$\varOmega_1 + \bar{k}\partial_k c(\bar{M},\bar{k}) \kappa _1+c(\bar{M},\bar{k}) \kappa _1 = 0 $$

and with our normalization choices (B.18) reduces to

$$U_1 = d U_{|(\bar{M},\bar{k})} ({\mathcal{M}}_1, \kappa_1),\quad {\mathcal{M}}_1 = \langle U_1 \rangle. $$

Compatibility condition T κ 1= X Ω 1 already gives

$$ \partial_T\kappa_1+ \partial_X\bigl(\bar{k}\partial_k c_{|(\bar {M},\bar{k})} \kappa_1+c(\bar{M},\bar{k} ) \kappa_1\bigr) = 0. $$
(B.19)

At the next step of the identification, we get with (Ω 2,κ 2)=( T Ψ 2, X Ψ 2)

$$\begin{aligned} &\bigl(\varOmega_2+c(\bar{M},\bar{k}) \kappa_2\bigr) \bar{U}'- \kappa_2 L^{(1)}\bar{U}' - L_0 U_2 - L^{(1)}\partial_X U_1 \\ &\quad - (\kappa_1)^2 L^{(2)}\bar{U}'' - \kappa_1 L^{(1)}U_1' - \partial _X\kappa_1 L^{(2)}\bar{U}' \\ &\quad + \bigl(\varOmega_1+c(\bar{M},\bar{k})\kappa_1\bigr) U_1' + \partial_T U_1+c( \bar{M},\bar{k})\partial_X U_1 \\ &\quad + \bar{k}\partial_y \biggl(\frac{1}{2}d^2f(\bar{U}) (U_1,U_1) \biggr) + \kappa_1 \partial_y \bigl(df(\bar{U}) \bigr) U_1 = 0, \end{aligned} $$

which may also be written

$$\begin{aligned} &L_0 \biggl(U_2 - d U_{|(\bar{M},\bar{k})} ( \cdot ;{\mathcal{M}}_2,\kappa_2) -\frac{1}{2}d^2 U_{|(\bar{M},\bar{k})} \bigl( \cdot ;({\mathcal{M}}_1,\kappa _1),({\mathcal{M}}_1,\kappa_1)\bigr) \biggr) \\ &\quad{} = \partial_T U_1+c(\bar{M},\bar{k}) \partial_X U_1- \partial_X\kappa _1 L^{(2)}\bar{U}'- L^{(1)} \partial_X U_1 \\ &\qquad {}+ \biggl(\varOmega_2- \partial_k \omega( \bar{M},\bar{k})\kappa_2 -\frac{1}{2}d^2\omega(\bar{M}, \bar{k}) \bigl(({\mathcal{M}}_1,\kappa_1),({\mathcal{M}}_1,\kappa_1)\bigr) \biggr)\bar{U}' \end{aligned}$$
(B.20)

for any \({\mathcal{M}}_{1}\). Solvability then reads

$$\begin{aligned} &\partial_T{\mathcal{M}}_1+ \partial_X\bigl(d F_{|(\bar{M},\bar {k})} ({\mathcal{M}}_1, \kappa_1)\bigr) = 0, \end{aligned}$$
(B.21)
$$\begin{aligned} &\varOmega_2-\partial_k \omega(\bar{M}, \bar{k}) \kappa_2 - \frac{1}{2}d^2\omega (\bar{M},\bar{k}) \bigl(({\mathcal{M}}_1,\kappa_1),({\mathcal{M}}_1,\kappa _1)\bigr) \\ &{}\quad =\partial_X\kappa_1+\bigl\langle u^{\mathit{adj}}(\bar{M},\bar{k}), L^{(1)}\partial_kU_{|(\bar{M},\bar{k} )} \bigr\rangle\partial_X\kappa_1 \\ &{}\qquad + \bigl\langle u^{\mathit{adj}}(\bar{M},\bar{k}), L^{(1)}\partial_MU_{|(\bar {M},\bar{k})} \bigr\rangle\partial _X{\mathcal{M}}_1. \end{aligned}$$
(B.22)

Note that the latter equation yields

$$\begin{aligned} &\partial_T\kappa_2- \partial_X \biggl(\partial_k \omega(\bar {M},\bar{k}) \kappa_2 + \frac{1}{2}d^2\omega(\bar{M},\bar{k}) \bigl(({ \mathcal{M}}_1,\kappa_1),({\mathcal{M}}_1, \kappa_1)\bigr) \biggr) \\ &{}\quad =\partial^2_X\kappa_1+ \partial_X \bigl(\bigl\langle u^{\mathit{adj}}(\bar {M},\bar{k}), L^{(1)}\partial _kU_{|(\bar{M},\bar{k})}\bigr\rangle \partial_X\kappa_1 \\ &{}\qquad + \bigl\langle u^{\mathit{adj}}( \bar{M},\bar{k}), L^{(1)}\partial_MU_{|(\bar {M},\bar{k})}\bigr \rangle\partial _X{\mathcal{M}}_1 \bigr). \end{aligned}$$
(B.23)

To proceed, we introduce \(\tilde{g}^{k}\), \(\tilde{g}^{M}\), the solutions of

$$\begin{aligned} L_0 \tilde{g}^{k} =&- L^{(1)} \partial_k U_{|(\bar{M},\bar {k})}-\partial_M U_{|(\bar{M} ,\bar{k})}\partial_k F_{|(\bar{M},\bar{k})} \\ &{}+ \bar{U}'\bigl\langle u^{\mathit{adj}}(\bar{M},\bar{k}), L^{(1)}\partial _kU_{|(\bar{M},\bar{k} )}\bigr\rangle - \partial_k U_{|(\bar{M},\bar{k})} \bar{k} \partial_k c_{|(\bar {M},\bar{k})}, \\ L_0 \tilde{g}^{M} =&-L^{(1)} \partial_M U_{|(\bar{M},\bar {k})}-\partial_M U_{|(\bar{M} ,\bar{k})}(\partial_M F_{|(\bar{M},\bar{k})}-c_{|(\bar{M},\bar {k})}{\rm Id }) \\ &{}+ \bar{U}'\bigl\langle u^{\mathit{adj}}(\bar{M},\bar{k}), L^{(1)}\partial _MU_{|(\bar{M},\bar{k} )}\bigr\rangle, \end{aligned}$$

mean free and orthogonal to \(u^{\mathit{adj}}( \cdot ;\bar{M},\bar{k})\) and set . With (B.21) and (B.22), setting \({\mathcal{M}}_{2}=\langle U_{2}\rangle\), Eq. (B.20) becomes

$$U_2 = d U_{|(\bar{M},\bar{k})} ({\mathcal{M}}_2, \kappa_2) + \frac{1}{2}d^2 U_{|(\bar{M} ,\bar{k})} \bigl( \cdot ;({ \mathcal{M}}_1,\kappa_1),({\mathcal{M}}_1, \kappa_1)\bigr) + \tilde{g} \begin{pmatrix}\partial_X{\mathcal{M}}_1\\\partial_X\kappa_1 \end{pmatrix} . $$

Finally, substituting (B.2), (B.3) into (B.1), and comparing terms of order ε 3, we obtain an equation of the form

$$\partial_T U_2 + \partial_X \biggl(df( \bar{U}) U_2+\frac{1}{2}d^2f(\bar{U}) (U_1,U_1) \biggr) - \partial_X^2U_1 - L_0U_3 + \partial_y( \cdots ) = 0, $$

whose solvability implies

$$\begin{aligned} &\partial_T{\mathcal{M}}_2+ \partial_X \biggl(d F_{|(\bar{M},\bar {k})} ({\mathcal{M}}_2, \kappa_2)+\frac{1}{2}d^2 F_{|(\bar{M},\bar{k})} \bigl[({\mathcal{M}}_1,\kappa _1),({\mathcal{M}}_1, \kappa_1)\bigr] \biggr) \\ &{}\quad =\partial_X^2{\mathcal{M}}_1 - \partial_X \biggl(\bigl\langle df(\bar{U}) \tilde{g}\bigr\rangle \begin{pmatrix}\partial_X{\mathcal{M}}_1\\\partial_X\kappa_1 \end{pmatrix} \biggr). \end{aligned}$$
(B.24)

To write the second order system in a compact form, let us introduce, for arbitrary (M,k),

$$\begin{aligned} \tilde{d}_{1,1}&= {\rm Id }-\bigl\langle df\bigl(U(M,k)\bigr) \tilde{g}^M\bigr\rangle,\qquad \tilde{d}_{1,2}= -\bigl\langle df\bigl(U(M,k)\bigr) g^k\bigr \rangle, \\ \tilde{d}_{2,1}&= \bigl\langle u^{\mathit{adj}}(\bar{M},\bar{k}), L^{(1)}\partial_MU_{|(\bar {M},\bar{k})}\bigr\rangle,\\ \tilde{d}_{2,2}&= 1+\bigl\langle u^{\mathit{adj}}(\bar{M},\bar{k}), L^{(1)}\partial_kU_{|(\bar {M},\bar{k})}\bigr\rangle. \end{aligned} $$

With these notations, systems (B.19), (B.21), (B.23), (B.24) coincide with the first nontrivial systems obtained in the formal expansion of a solution \(({\mathcal{M}},\kappa)\) of

$$ \left\{ \begin{aligned} & \partial_t {\mathcal{M}}+ \partial_x\bigl(F({\mathcal{M}}, \kappa)\bigr)= \partial_x (\tilde{d}_{1,1} \partial_x{\mathcal{M}}+\tilde{d}_{1,2} \partial _x\kappa ) \\ &\partial_t \kappa+ \partial_x\bigl(\kappa c({\mathcal{M}},\kappa)\bigr)= \partial_x (\tilde{d}_{2,1} \partial_x{\mathcal{M}}+\tilde{d}_{2,2} \partial _x\kappa ) \end{aligned} \right. $$
(B.25)

according to the slow ansatz

$$({\mathcal{M}},\kappa) (x,t) = \sum_j \varepsilon^j ({\mathcal{M}}_j,\kappa_j) ( \varepsilon x,\varepsilon t),\qquad({\mathcal{M}}_0, \kappa_0) = (\bar{M},\bar{k}). $$

We call system (B.25), likewise, a (second-order) Whitham’s modulation system.

As should be clear from the formal derivations, there is some freedom in the choice of the diffusion matrices. This reflects the fact that many systems of conservation laws share the same asymptotic behavior about constant states. We recall next how to classify these systems according to their asymptotic behavior; this will provide a canonical modulation system for our nonlinear analysis.

We emphasize however that the classification below has been so far as we know verified only for symmetrizable hyperbolic-parabolic systems satisfying a Kawashima structural condition [35]. It is unclear to us whether our formally obtained Whitham systems satisfy such conditions. Yet, we know that, by applying to them, on formal grounds, the asymptotic-equivalence reduction, we do obtain a system satisfying such conditions and providing the correct asymptotic behavior. Hence we may safely ignore these technical details.

2.2 B.2 Asymptotic equivalence of systems of conservation laws

2.2.1 B.2.1 General theory

We now recall the notion of asymptotic equivalence and behavior of solutions of systems of conservation laws near a constant state, useful in our context since, being able to prove modulational behavior, we reduce the dynamics about a periodic wave to motion of parameters near a constant state. Given a general system of conservation laws

$$ w_t +\bigl(g(w)\bigr)_x= \bigl( B(w)w_x\bigr)_x $$
(B.26)

and a reference state w at which dg(w ) has distinct eigenvalues, so that L dg(w )R is diagonal for some

$$L_*= \begin{pmatrix}l_1^*\\ \vdots\\ l_n^* \end{pmatrix} ,\ R_*= \begin{pmatrix}r_1^* & \dots& r_n^* \end{pmatrix} ,\quad L_*R_*={\rm Id }, $$

define the quadratic approximant

$$ y_t +A_* y_x +\frac{1}{2} \bigl(y^t\varGamma_* y\bigr)_x = \tilde{B}_* y_{xx}, $$
(B.27)

and the decoupled quadratic approximant

$$ z_t +A_* z_x +\frac{1}{2} \bigl(z^t\tilde{\varGamma}_* z\bigr)_x = \tilde{B}_* z_{xx}, $$
(B.28)

about w , whereFootnote 30

$$ A_*=dg(w_*),\qquad \varGamma_*:= d^2g(w_*), \quad\textrm{and} \quad B_*:=B(w_*), $$
(B.29)
$$ \tilde{\varGamma}_*:= L_*^t{\rm diag }\bigl\{ R_*^t \varGamma_*R_*\bigr\}L_*, \quad\textrm{and}\quad \tilde{B}_*:= R_*{\rm diag }\{ L_* B_*R_*\}L_*. $$
(B.30)

Assume the parabolicity condition, \({\rm diag }\{ L_{*} B(w_{*})R_{*}\}\) is positive, and define the self-similar nonlinear (resp. linear if γ j =0) diffusion waves \(\theta_{j}(x,t)=t^{-1/2}\bar{\theta}_{j}(x/\sqrt{t})\) to be the solutions of the Burgers equations (resp. heat equations if γ j =0)

$$ \theta_t+ \frac{1}{2}\bigl( \gamma_j^* \theta^2\bigr)_x= \theta_{xx}, \quad \gamma_j^*:=\bigl[l_j^* \bigl(r_j^*\bigr)^t\varGamma_* r_j^*\bigr]/ \bigl[l_j^*B_*r_j^*\bigr], $$
(B.31)

with delta-function initial data \(l_{j}^{*} m_{0} \delta(\cdot)\), where m 0:=∫z 0(x)dx. Then, we have the following fundamental result describing behavior of (B.26)–(B.28) with respect to localized initial perturbations.

Proposition B.1

([35, 41])

Let η>0. Let w and z be solutions of (B.26) and (B.28) with initial data w 0 and z 0=w 0w such that \(E_{1}:=\|z_{0}\|_{L^{1}(\mathbb{R})\cap H^{4}(\mathbb{R})}+\||\cdot|z_{0}\| _{L^{1}(\mathbb{R})}\) is sufficiently small. Then, for 1≤p≤∞, \(m_{0}:=\int_{\mathbb{R}}z_{0}\), and θ j as in (B.31),

$$ \bigl\|w(t)-w_*-z(t)\bigr\|_{L^p(\mathbb{R})}\lesssim E_1(1+t)^{-\frac {1}{2}(1-1/p)-\frac{1}{4}+\eta} ; $$
(B.32)

and

$$ \biggl\|z(t)-\sum_j \theta_j \bigl(\cdot-a_j^*(1+t),b_j^* (1+t)\bigr) r_j^*\biggr\| _{L^p(\mathbb{R})}\lesssim E_1(1+t)^{-\frac{1}{2}(1-1/p)-\frac {1}{4}+\eta} , $$
(B.33)

with \(a_{j}^{*}:=l_{j}^{*}A_{*}r_{j}^{*}\), \(b_{j}^{*}:=l_{j}^{*}B_{*}r_{j}^{*}\), whence,Footnote 31 if η<1/4,

$$\bigl\|w(t)-w_*\bigr\|_{L^p(\mathbb{R})}, \bigl\|z(t)\bigr\|_{L^p(\mathbb{R})} \gtrsim|m_0| (1+t)^{-\frac{1}{2}(1-1/p)} . $$

Proposition B.1 asserts that (B.26) and (B.28) (hence also (B.27)) are asymptotically equivalent with respect to small localized initial data \(w_{0}-w_{*}=z_{0}\in L^{1}(\mathbb{R},(1+|x|)dx)\cap H^{3}(\mathbb{R})\), in the sense that the difference between solutions z(t) and w(t)−w decays at rate \((1+t)^{-\frac{1}{2}(1-1/p)-\frac{1}{4}+\eta}\) approximately \((1+t)^{-\frac{1}{4}}\) faster than the (Gaussian) rate \(|m_{0}| (1+t)^{-\frac{1}{2}(1-1/p)}\) at which either one typically (i.e., for data with small L 1 first moment) decays. Moreover, through (B.33), it gives a simple description of asymptotic behavior as the linear superposition of scalar diffusion waves θ j moving with characteristic speeds (eigenvalues \(a_{j}^{*}\)) in the characteristic modes (eigendirections \(r_{j}^{*}\)) of dg(w ), satisfying Burgers equations (B.31).

We have also the following more elementary result comparing to the full quadratic approximant.

Proposition B.2

([33]Footnote 32)

Let η>0. Let w and y be solutions of (B.26) and (B.27) with initial data w 0 and y 0=w 0w such that \(E_{0}:=\|y_{0}\|_{L^{1}(\mathbb{R})\cap H^{4}(\mathbb{R})}\) is sufficiently small. Then, for 1≤p≤∞,

$$\bigl\|(w-w_*-y) (t)\bigr\|_{L^p(\mathbb{R})}\lesssim E_0 (1+t)^{-\frac{1}{2}(1-1/p)-\frac{1}{2}+\eta}. $$

An important consequence of Proposition B.2 is that only the quadratic order quantities appearing in (B.27) need be taken into account in the study of asymptotic behavior of (B.26) to the order of approximation considered in Theorem 1.12. Finally, we note the following result following from a proof similar to but much simpler than the one for Proposition B.2 given in [33, Appendix A].

Lemma B.3

Let k satisfy k(0)=0 and

$$ k_t + ak_x + \bigl(\gamma k^2\bigr)_x- dk_{xx}= (Fk)_x, $$
(B.34)

where a,γ,d are constant, d>0 and F is a given function such that \(\|F(t)\|_{L^{2}(\mathbb{R})}\le E_{0}(1+t)^{-\frac{1}{4}}\). Then, for any η>0, provided E 0 is small enough, for 1≤p≤∞,

$$\bigl\|k(t)\bigr\|_{L^p(\mathbb{R})}\lesssim E_0(1+t)^{-\frac {1}{2}(1-1/p)-\frac {1}{2}+\eta}. $$

For the sake of completeness, we recall the proof of the previous Proposition in Appendix C.

2.2.2 B.2.2 A first application

As an immediate application, we may now establish the improved decay bounds (1.22)–(1.21) of Corollary 1.18. We will use these tools again in establishing (1.18).

Proof of Corollary 1.18

Bound (1.21) follows from the assumption \(\bar{k}\partial _{x}h_{0}=0\). For, a solution (M W ,k W ), with an initial data (∗,0), of the decoupled approximating equations (B.28) to (1.7) satisfies k W (t)≡0, since the k equation decouples in (B.28) for the linearly phase-decoupled case. Comparing to the actual solution of (1.7) using (B.32), we obtain the result. Bound (1.22) goes similarly, observing that in the quadratically decoupled case, the k equation in the full quadratic approximating system (B.27) to (1.7), though it does not completely decouple, is of the form (B.34) with \(F=\mathcal{O}(M)\). □

Remark B.4

Analogous to (B.34) in the quadratically decoupled case, the rate-determining bound in the linearly decoupled case of Proposition B.1 is the key estimate

$$\bigl\|k(t)\bigr\|_{L^p}\lesssim E_0(1+t)^{-\frac{1}{2}(1-1/p)-\frac{1}{4}} $$

established by Liu [40] for quadratic coupling terms involving different modes, thus obeying k(0)=0, \(k_{t} + ak_{x} + (\gamma k^{2})_{x}- dk_{xx}= (\tilde{\theta}^{2})_{x}\), where \(\tilde{\theta}(x,t)=\theta(x-\tilde{a}t,\tilde{b} t)\) with \(\tilde{a}\ne a\) and θ a self similar solution of a Burgers equation (B.31). The anomalous rate \((1+t)^{\frac{1}{4}}\) is different from the powers of \((1+t)^{\frac{1}{2}}\) arising in scalar convection–diffusion processes, reflecting the additional complications present in the system case.

2.2.3 B.2.3 Quadratic approximants of modulation systems

For later reference, let us write, in the original frame (and not the co-moving one), as

$$\begin{aligned} & \partial_t \begin{pmatrix}M\\ k \end{pmatrix} + \partial_x \begin{pmatrix}dF_{|(\bar{M},\bar{k})}(M,k)\\d\omega_{|(\bar{M} ,\bar{k})}(M,k) \end{pmatrix} + \frac{1}{2}\partial_x \begin{pmatrix}d^2F_{|(\bar{M},\bar{k})}(M,k)\\d^2\omega_{|(\bar {M},\bar{k} )}((M,k),(M,k)) \end{pmatrix} \\ &\quad = \tilde{B}_*\partial_x^2 \begin{pmatrix}M\\k \end{pmatrix} \end{aligned}$$
(B.35)

the quadratic approximant of (1.7) (obtained as (B.16) and (B.25) above). As pointed out in Remark 4.11, it follows from Lemma 4.10 that this system is independent of the choices made in the course of the formal derivation.

From the general theory, we know that instead of comparing \(({\mathcal{M}},\kappa)\) in Theorem 1.12 to a solution \(({\mathcal{M}}_{W},\kappa_{W})\) of (1.7), we only need to compare it with \((\bar{M},\bar{k})+(M_{W},k_{W})\) with (M W ,k W ) a solution of (B.35) expressed in the co-moving frame.

2.3 B.3 Implicit change of variables

Our nonlinear analysis begins with an implicit nonlinear change of variable (2.5). We explain now how the modulation system is affected by this change of variables. We could have first performed this implicit change of variables then carried out the formal modulation process, but we find more enlightening to change the system a posteriori.

Since our diffeomorphism is close to identity, only nonlinear terms should be changed, and from the asymptotic equivalence theory we know that nonlinear terms are relevant only in the hyperbolic part. Therefore it is enough to investigate how (B.5), (B.8) is altered. Let us introduce Φ 0 such that Φ 0(Ψ 0(X,T),T)=X. Recall that \(\partial_{T}\varPsi_{0}=\omega({\mathcal{M}}_{0},\partial _{X}\varPsi_{0})\). Therefore if A,B are such that T A+ X B=0 then \((\tilde{A},\tilde{B})(X,T)=(A,B)(\varPhi_{0}(X,T),T)\) implies

$$\partial_T\tilde{A} - \frac{\partial_T\varPhi_0}{\partial_X\varPhi_0} \partial_X \tilde{A} + \frac{1}{\partial_X\varPhi_0} \partial_X\tilde{B} = 0 $$

also written \(\partial_{T} (\partial_{X}\varPhi_{0}\tilde{A} )+\partial_{X} (\tilde{B}-\partial_{T}\varPhi_{0}\tilde{A} )=0\) or

$$\partial_T (\partial_X\varPhi_0\tilde{A} ) + \partial_X \biggl(\tilde{B}-c \biggl(\tilde{\mathcal{M}}_0,\frac{1}{\partial_X\varPhi_0} \biggr)\tilde{A} \biggr) = 0 $$

with \(\tilde{\mathcal{M}}_{0}(X,T)={\mathcal{M}}_{0}(\varPhi_{0}(X,T),T)\). Note that this kind of manipulation is completely similar to the ones needed to perform usual Lagrangian change of coordinates and of course closely related to the computations involved in the proof of Lemma 2.3. As expected, applying this to (B.5) leads to a trivial equation while an application on the trivial equation T (1)+ x (0)=0 gives

$$\partial_T (\partial_X\varPhi_0 ) - \partial_X \biggl(c \biggl(\tilde {\mathcal{M}}_0, \frac{1}{\partial_X\varPhi_0} \biggr) \biggr) = 0. $$

Equation (B.5) is changed into

$$\partial_T (\partial_X\varPhi_0\tilde{ \mathcal{M}}_0 ) + \partial_X \biggl(F \biggl(\tilde{ \mathcal{M}}_0,\frac{1}{\partial_X\varPhi_0} \biggr)-c \biggl(\tilde{\mathcal{M}}_0,\frac{1}{\partial_X\varPhi_0} \biggr)\tilde {\mathcal{M}}_0 \biggr) = 0. $$

At the hyperbolic level, we are thus lead to the system

$$\begin{gathered} \partial_T p-\partial_X \biggl(c \biggl(\frac{{\mathcal{M}}}{p},\frac{1}{p} \biggr) \biggr) = 0, \\ \partial_T {\mathcal{M}}+\partial_X \biggl(F \biggl( \frac{{\mathcal{M}}}{p},\frac{1}{p} \biggr)-c \biggl(\frac{{\mathcal{M}}}{p},\frac{1}{p} \biggr) \frac {{\mathcal{M}}}{p} \biggr) = 0 \end{gathered} $$

whose quadratic expansion in

$$(p,{\mathcal{M}}) = \biggl(\frac{1}{\bar{k}},\frac{\bar{M}}{\bar {k}} \biggr)+ \biggl(\frac{-k}{\bar{k}}\frac{1}{\bar{k}},\frac{-k}{\bar {k}} \frac{\bar{M} }{\bar{k}}+\frac{M}{\bar{k}} \biggr) $$

gives

$$\begin{aligned} &\partial_T k-\bar{k} c(\bar{M}, \bar{k})\partial_Xk-\bar {k}\partial_X \biggl(d \omega_{|(\bar{M},\bar{k} )}(M,k)+\frac{1}{2}d^2\omega_{|(\bar{M},\bar{k})} \bigl((M,k),(M,k)\bigr) \biggr)\\ &\quad = 0, \\ &\partial_T M-\bar{k} c(\bar{M},\bar{k})\partial_XM \\ &\quad +\partial _X \biggl(dF_{|(\bar{M},\bar{k} )}(M,k)+\frac{1}{2}d^2F_{|(\bar{M},\bar{k})} \bigl((M,k),(M,k)\bigr) \biggr) \\ &\quad +\bar{k}\partial_X \biggl(\frac{k}{\bar{k}} \bigl(dF_{|(\bar {M},\bar{k})}(M,k)-c(\bar{M} ,\bar{k})M \bigr)-dc_{|(\bar{M},\bar{k})}(M,k)M \biggr) = 0. \end{aligned} $$

Two main comments are in order: 1. We end up naturally with equations expressed in a co-moving frame thus no further change is needed. 2. The wavenumber equation remains unaltered at this level of description. This explains why the fact that the implicit change of variables could change the modulation equations was not revealed by previous studies [33, 57] focusing on situations where no other wave parameter is involved,

Remark B.5

Though we do not need it for the present semilinear analysis, let us describe for the sake of generality what would happen for a full quasilinear parabolic system. For

$$\partial_t \begin{pmatrix}{\mathcal{M}}\\\varPsi_x \end{pmatrix} + \partial_x A({\mathcal{M}},\varPsi_x) = \partial_x \biggl(D({\mathcal{M}},\varPsi_x) \partial_x \begin{pmatrix}{\mathcal{M}}\\\varPsi_x \end{pmatrix} \biggr) $$

with

$$\partial_t\varPsi+ A_{n+1}({\mathcal{M}},\varPsi_x) = D_{n+1}({\mathcal{M}},\varPsi_x)\partial_x \begin{pmatrix}{\mathcal{M}}\\\varPsi_x \end{pmatrix}, $$

where A n+1=e n+1A, D n+1=e n+1D, the transformation Φ=Ψ −1, \(\tilde{\mathcal{M}}={\mathcal{M}}\circ\varPhi\) leads to

$$\partial_t\varPhi - \varPhi_x A_{n+1}(\tilde{ \mathcal{M}},1/\varPhi_x) = -D_{n+1}(\tilde{\mathcal{M}},1/ \varPhi_x)\partial_x \begin{pmatrix}\tilde{\mathcal{M}}\\ 1/\varPhi_x \end{pmatrix} $$

with

$$\begin{aligned} &\partial_t\tilde{\mathcal{M}}+ \partial_x \bigl(A_\perp(\tilde {\mathcal{M}},1/ \varPhi_x)-\partial_x\varPhi A_{n+1}(\tilde{ \mathcal{M}},1/\varPhi_x) \tilde{\mathcal{M}} \bigr) \\ &\quad =\partial_x \biggl(\frac{1}{\partial_x\varPhi} D_\perp( \tilde {\mathcal{M}},1/\varPhi_x)\partial_x \begin{pmatrix}\tilde{\mathcal{M}}\\1/\varPhi_x \end{pmatrix}\\ &\qquad -D_{n+1}(\tilde {\mathcal{M}},1/ \varPhi_x)\partial_x \begin{pmatrix}\tilde{\mathcal{M}}\\1/\varPhi_x \end{pmatrix} \tilde{\mathcal{M}} \biggr), \\ &\partial_t(\partial_x\varPhi)-\partial_x \bigl(\varPhi_x A_{n+1}(\tilde{\mathcal{M}},1/ \varPhi_x) \bigr) = -\partial_x \biggl(D_{n+1}( \tilde{\mathcal{M}},1/\varPhi_x)\partial_x \begin{pmatrix}\tilde{\mathcal{M}}\\1/\varPhi_x \end{pmatrix} \biggr), \end{aligned} $$

where

$$A_\perp= \begin{pmatrix}{\rm Id }_{d\times d}& \begin{array}{c}0\\ \vdots\\ 0 \end{array} \end{pmatrix} A,\qquad D_\perp= \begin{pmatrix}{\rm Id }_{d\times d}& \begin{array}{c}0\\ \vdots\\ 0 \end{array} \end{pmatrix} D. $$

Collecting the results of this appendix, we find that to validate the formal Whitham modulation approximation, we only need to compare the couple \((M,\bar{k}\psi_{x})\) of Theorem 1.12 to a solution (M W ,k W ) of

$$\begin{aligned} &\partial_t \begin{pmatrix}M\\ k \end{pmatrix} + \bar{k}A_*\partial_x \begin{pmatrix}M\\k \end{pmatrix} \\ &\quad+ \frac{1}{2}\bar{k}\partial_x \begin{pmatrix}d^2F_{|(\bar{M},\bar{k})}((M,k),(M,k))\\d^2\omega _{|(\bar{M},\bar{k})}((M,k),(M,k)) \end{pmatrix} - \bar{k}^2\tilde{B}_*\partial_x^2 \begin{pmatrix}M\\k \end{pmatrix} \\ &\quad + \bar{k}\partial_x \begin{pmatrix}\frac{k}{\bar{k}} (dF_{|(\bar{M},\bar {k})}(M,k)-c(\bar{M} ,\bar{k})M )-dc_{|(\bar{M},\bar{k})}(M,k)M\\0 \end{pmatrix} = 0, \end{aligned} $$

where \(A_{*}=\partial_{(M,k)} (F-\bar{c}M,-\omega-\bar{c}k)|_{(\bar {M},\bar{k})}\). For writing convenience, we denote this system by

$$ \partial_t \begin{pmatrix}M\\ k \end{pmatrix} + \bar{k}A_*\partial_x \begin{pmatrix}M\\k \end{pmatrix} - \partial_x \biggl(\frac{1}{2} \begin{pmatrix}M\\ k \end{pmatrix} ^T \varGamma_* \begin{pmatrix}M\\ k \end{pmatrix} \biggr) = \bar{k}^2\tilde{B}_*\partial_x^2 \begin{pmatrix}M\\k \end{pmatrix} . $$
(B.36)

Likewise, ψ in Theorem 1.12 needs then to be compared with ψ W a solution of

$$ \begin{aligned}[b] &\partial_t\psi+ e_{n+1} \cdot A_* \begin{pmatrix}M_W\\k_W \end{pmatrix} - \frac{1}{\bar{k} }e_{n+1} \cdot \biggl(\frac{1}{2} \begin{pmatrix}M_W\\ k_W \end{pmatrix} ^T \varGamma_* \begin{pmatrix}M_W\\ k_W \end{pmatrix} \biggr)\\ &\quad = \bar{k}e_{n+1}\cdot\tilde{B}_*\partial_x \begin{pmatrix}M_W\\k_W \end{pmatrix} .\end{aligned} $$
(B.37)

Appendix C: Asymptotic equivalence of quadratic approximants

For completeness, we include here a proof of Proposition B.2 including the treatment of off-diagonal diffusion terms not arising in the scalar case considered in [32].

Proof of Proposition B.2

(Case \(B_{*}=\tilde{B}_{*}\).) We first review the case \(B_{*}=\tilde{B}_{*}\) treated in [32]. By the general results of [35], provided \(E_{0}:=\|y_{0}\|_{L^{1}\cap H^{3}(\mathbb{R})}\) is sufficiently small, we have for 1≤p≤∞

$$\begin{aligned} \bigl\|w(t)-w_*\bigr\|_{L^p(\mathbb{R})}, \bigl\|y(t) \bigr\|_{L^p(\mathbb{R})} &\lesssim E_0 (1+t)^{-\frac{1}{2}(1-\frac {1}{p})}, \\ \bigl\|w_x(t)\bigr\|_{H^1(\mathbb{R})}, \bigl\|y_x(t) \bigr\|_{H^1(\mathbb{R})} &\lesssim E_0 (1+t)^{-\frac{3}{4}}. \end{aligned} $$

Setting δ:=w +yw, we have, subtracting and rearranging,

$$\begin{aligned} &\delta_t +A_* \delta_x- B_*\delta_{xx}= \partial_x \mathcal {F},\\ &\mathcal{F}=\mathcal{O} \bigl(\bigl(|w-w_*|+|y|\bigr)\delta\bigr) +\mathcal{O}\bigl(|w-w_*|^3\bigr)+ \mathcal{O}\bigl(|w-w_*| |w_x|\bigr), \end{aligned} $$

with δ| t=0=0 and A and B as in (B.29)–(B.30). By Duhamel’s formula,

$$\delta(t)= \int_0^t \sigma(t-s) \partial_x\mathcal{F}(s) ds, $$

where σ is the solution operator of the parabolic system of conservation laws u t +A u x B u xx =0. Applying the standard bounds [41] \(\|\sigma(t)\partial_{x}^{r} h\|_{L^{p}(\mathbb{R})}\lesssim t^{-\frac{1}{2}(\frac{1}{q}-\frac{1}{p})-\frac{r}{2}}\|h\|_{L^{q}(\mathbb{R})}\), 1≤qp≤∞, together with

$$\begin{aligned} \bigl\|\mathcal{F}(t)\bigr\|_{L^q(\mathbb{R})} &\lesssim E_0 (1+t)^{-\frac{1}{2}(1-1/q)-\frac{1}{4}} \bigl(\bigl\|\delta(t)\bigr\|_{L^2(\mathbb{R})}\\ &\quad +\bigl\|w_x(t)\bigr\|_{L^2(\mathbb{R})} \bigr)+E_0^2(1+t)^{-\frac{1}{2}(1-1/q)-1},\end{aligned} $$

1≤q≤2, we find, defining \(\nu(t){:=} \sup_{0\le s\le t}\|\delta(s)\|_{L^{2}(\mathbb{R})}(1+s)^{\frac{1}{2}(1-1/p)+\frac{1}{2} -\eta}\), that, for all 1≤p≤∞,

$$\begin{aligned} \bigl\|\delta(t)\bigr\|_{L^p(\mathbb{R})}&\lesssim \int _0^{t/2} (t-s)^{-\frac{1}{2}(1-1/p)-\frac{1}{2}}\bigl\| \mathcal{F}(s)\bigr\| _{L^1(\mathbb{R} )} ds \\ &\quad+ \int_{t/2}^t (t-s)^{-\frac{1}{2}(1/(\min(2,p))-1/p)-\frac{1}{2}} \bigl\| \mathcal{F}(s)\bigr\|_{L^{\min(2,p)}(\mathbb{R})} ds \\ &\lesssim \int_0^{t/2} (t-s)^{-\frac{1}{2}(1-1/p)-\frac{1}{2}} \bigl(\nu (t)E_0+E_0^2\bigr) (1+s)^{-1+\eta}ds \\ &\quad +\int_{t/2}^t (t-s)^{-\frac{1}{2}(1/(\min(2,p))-1/p)-\frac{1}{2}}\bigl(\nu (t)E_0+E_0^2\bigr)\\ &\quad \times (1+s)^{-1+\eta-\frac{1}{2}(1-1/(\min(2,p)))}ds \\ &\lesssim E_0\bigl(E_0 +\nu(t)\bigr) (1+t)^{-\frac{1}{2}(1-1/p)-\frac{1}{2} +\eta}, \end{aligned} $$

whence ν(t)≤C η E 0(E 0+ν(t)). This implies that \(\nu(t)\le2C_{\eta}E_{0}^{2}\) for E 0<1/(2C η ), giving

$$\bigl\|\delta(t)\bigr\|_{L^p(\mathbb{R})}\leq2C_\eta E_0^2 (1+t)^{-\frac{1}{2}(1-1/p)-\frac{1}{2} +\eta} , \quad1\le p\le\infty. $$

(General case.) We treat now the general case that \(B(w)=B_{*}+\mathcal{O}(w-w_{*})\) with B constant but not equal to \(\tilde{B}_{*}\). Defining again δ:=w +yw, and denoting by \(\tilde{\sigma}(t)\) the solution operator of linear system \(u_{t} + A_{*} u_{x}-\tilde{B}_{*} u_{xx}=0\), we have by Duhamel’s principle

$$\begin{aligned} \delta(t)&=(\tilde{\sigma}-\sigma) (t) w_0 +\int _0^t (\tilde{\sigma}-\sigma) (t-s) \partial_x \mathcal{O} \bigl(|w-w_*|^2\bigr) (s)ds \\ &\quad +\int_0^t \tilde{\sigma}(t-s) \partial_x \mathcal{O}\bigl(|\delta |\bigl(|w-w_*|+|y|\bigr)\bigr) (s)ds \\ &\quad +\int_0^t \sigma(t-s) \partial_x \mathcal{O}\bigl(|w-w_*|^3+|w-w_*||w_x| \bigr) (s)ds. \end{aligned}$$

From [35], provided \(E_{0}:=\|y_{0}\|_{L^{1}\cap H^{4}(\mathbb{R})}\) is sufficiently small, we have for 1≤p≤∞

$$\begin{aligned} \bigl\|w(t)-w_*\bigr\|_{L^p(\mathbb{R})}, \bigl\|y(t) \bigr\|_{L^p(\mathbb{R})} &\lesssim E_0 (1+t)^{-\frac{1}{2}(1-\frac {1}{p})}, \\ \bigl\|w_x(t)\bigr\|_{H^2(\mathbb{R})}, \bigl\|y_x(t) \bigr\|_{H^2(\mathbb{R})} &\lesssim E_0 (1+t)^{-\frac{3}{4}}. \end{aligned} $$

Applying the bounds [35, 41] \(\|(\tilde{\sigma}-\sigma)(t)\partial_{x}^{r} h\|_{L^{p}(\mathbb {R})}\lesssim t^{-\frac{1}{2}(\frac{1}{q}-\frac{1}{p})-\frac{r}{2}}(1+t)^{-\frac{1}{2}}\| h\|_{L^{q}(\mathbb{R})}+e^{-\theta t}\|\partial_{x}^{r}h\|_{L^{p}(\mathbb{R})}\), 1≤qp≤∞ for r=0,1 (and some θ>0), estimating

$$\begin{aligned} &\biggl\|\int_0^t ( \tilde{\sigma}-\sigma) (t-s) \partial_x \mathcal{O} \bigl(|w-w_*|^2\bigr) (s)ds\biggr\|_{L^p(\mathbb{R})} \\ &\quad \lesssim \int_0^{t/2} (t-s)^{-\frac{1}{2}(1-1/p)-1}\bigl\| |w-w_*|^2 (s)\bigr\|_{L^1(\mathbb{R})} ds\\ &\qquad +\int_0^t e^{-\theta (t-s)}\bigl\| |w-w_*|^2(s)\bigr\|_{L^p(\mathbb{R})} ds \\ &\qquad+\int_{t/2}^t (t-s)^{-\frac{1}{2}}(1+t-s)^{-\frac{1}{2}} \bigl\| |w-w_*|^2(s)\bigr\|_{L^p(\mathbb{R})} ds \\ &\quad \lesssim E_0\int_0^{t/2} (t-s)^{-\frac{1}{2}(1-1/p)-1}(1+s)^{-\frac{1}{2}} ds\\ &\qquad +E_0\int _0^t e^{-\theta (t-s)}(1+s)^{-\frac{1}{2}(1-1/p)-\frac{1}{2}} ds \\ &\qquad+ E_0\int_{t/2}^t (t-s)^{-\frac{1}{2}}(1+t-s)^{-\frac{1}{2}} (1+s)^{-\frac{1}{2}(1-1/p)-\frac{1}{2}} ds \\ &\quad \lesssim E_0(1+t)^{-\frac{1}{2}(1-1/p)-\frac{1}{2}}\log(2+t), \end{aligned} $$

and other terms either similarly or similarly as in the previous case, we obtain the result. □

Appendix D: Generalizations

We conclude in this appendix by describing briefly extensions to more general types of equations arising in applications, and the modifications in our arguments that are needed to accomplish this, discussing also, when possible, the verification of (H1)–(H3) and (D1)–(D3) in specific cases.

4.1 D.1 Extensions in type: quasilinear and partially parabolic systems

Our analysis carries over in straightforward fashion to divergence-form systems of general quasilinear 2r-parabolic type. For example, the spectral preparation results of Lemma 1.5, Proposition 1.7, and Proposition 3.1 all go through essentially as written, depending on no special structure other than divergence form. From these low-frequency/Bloch number descriptions, we obtain the same linear bounds on the critical modes \(s^{\rm p}\) as described here in the 2-parabolic semilinear case. The high-frequency and or high Bloch number analysis also go through unchanged, the former depending again only on the spectral preparation results and the latter depending only (through Prüss’ Theorem) on high-frequency resolvent bounds following from (but not requiring) sectoriality of the linearized operator L about the wave. This completes the linear analysis.

Likewise, by Remark 2.4, we obtain the useful representation (2.6) of the nonlinear perturbation equations stated in Lemma 2.3, with sources \(\mathcal{Q}\), \(\mathcal{R}\), \(\mathcal{S}\) of quadratic order in v, ψ x , ψ t , and a finite number of their derivatives, which was all that was needed for our nonlinear arguments. To obtain the nonlinear damping estimate of Proposition 2.5, we note that (2.14) becomes

$$(1-\psi_x) v_t +(-1)^r \bar{k}^{2r} \partial_x\bigl( B\bigl(\tilde{U},\dots, \partial _x^{2r-2}\tilde{U}\bigr)\partial_x^{2r-1} v\bigr)= \mbox{lower order terms}, $$

\(\tilde{U}=\bar{U}+v\). Thus, taking the \(L^{2}(\mathbb{R})\) inner product against \(\sum_{j=0}^{K} \frac{(-1)^{j}\partial_{x}^{2j}v}{1-\psi_{x}}\), integrating by parts, and rearranging, we obtain \(\frac{d}{dt} \|v\|_{H^{K}(\mathbb{R})}^{2}(t) \leq-\tilde{\theta} \|\partial_{x}^{K+r} \cdot v(t)\|_{L^{2}(\mathbb{R})}^{2} + \hbox{lower order terms} \), similarly as in the second-order semilinear case, leading thereby to

$$\frac{d}{dt}\|v\|_{H^K(\mathbb{R})}^2(t) \leq-\theta\bigl\|v(t)\bigr\| _{H^K(\mathbb{R})}^2 + C \bigl( \bigl\|v(t)\bigr\|_{L^2(\mathbb{R})}^2+ \bigl\|(\psi_t, \psi_x) (t)\bigr\| _{H^K(\mathbb{R})}^2 \bigr) $$

and (by Gronwall’s inequality) the result. See the proof of [3, Proposition 3.4], for full details in the fourth-order semilinear case.

Combining these ingredients, we obtain, modulo an appropriate increase in the integer K encoding regularity requirements, stability, as stated in Theorem 1.10, and refined stability, as stated in Proposition 4.5, yielding the first part (1.16) of description of asymptotic behavior in Theorem 1.12. By Remark 4.15, we get also a partial version of the second part (1.17)–(1.19) of Theorem 1.12, but describing comparisons not to the Whitham system, but only to a second-order hyperbolic-parabolic system agreeing with the Whitham system in its linearization about the constant state \((\bar{M},\bar{k})\). This in turn yields the conclusions of (1.21), Corollary 1.18, regarding decay with respect to localized perturbations for linearly phase-decoupled systems.

Finally, to recover the full result (1.17)–(1.19) of Theorem 1.12, comparing to the exact Whitham system, and thus the sharpened decay rate (1.22) for localized data in the quadratically decoupled case, we have only to observe that performing the same computations as in Appendix A (differentiating the traveling-wave ODE), and in the proof of Lemma 4.9 (pulling out quadratic order parts of nonlinear term \(\mathcal{N}\)) while carrying along the additional higher-order terms arising in the general case, we obtain a higher-order analog of Lemma 4.14, expressing the resulting quadratic coupling constants (means) in terms of derivatives of first-order terms arising in the Whitham system, after which computations go as before to yield the result; see the proof of Theorem 1.12, Sect. 4.6.

This completes the treatment of the quasilinear 2r-parabolic case. Reviewing the above discussion, but omitting algebraic considerations on which we focus in the next section, we find that the two ingredients needed to treat more general divergence-form systems are the nonlinear damping estimate used to control higher-derivative by lower-derivative norms, and the high-frequency linearized resolvent bounds used to apply Prüss’ Theorem. For, these were the only two places where we used the parabolic form of the equations; the rest of the argument was completely general, Moreover, the second, linearized, estimate can typically be obtained by a linearized version of the same energy estimate that is used to obtain the first, damping-type estimate. This allows us, in particular, to treat (partially parabolic) symmetric hyperbolic–parabolic equations such as arise in continuum mechanics, using “Kawashima-type” energy estimates as described in [35], and variants thereof. See, for example, Proposition 4.4 (proved in Appendix A) and Lemma B.1 in [30].

Remark D.6

The strategy of using a common energy estimate to get, simultaneously, damping high-frequency resolvent, and high-frequency decay estimates, with derivative gains in the first compensating for derivative losses in the third, originates in the study of viscous shock stability; see [73, Sect. 4.2.1]. For simpler, and somewhat sharpened, versions in this context, see [37, 45].

4.2 D.2 Extensions in form: an abstract continuum of models

Still more generally, we may treat the full class of systems

$$ u_t+f(u)_x=g(u)+\bigl(B^1(u)u_x \bigr)_x +\bigl(B^2(u,u_x)u_{xx} \bigr)_x + \cdots , $$
(D.1)

\(u,f, g\in\mathbb{R}^{n}\), \(B_{j}\in\mathbb{R}^{n\times n}\), with , , , , \(u_{2}\in\mathbb{R}^{r}\), including both divergence- and nondivergence-type equations. Note that this includes both reaction diffusion and conservation law cases as limits f≡0 and g≡0, but also many cases in between: for example, the viscous relaxation case n=2, r=1 occurring for the Saint-Venant equations (1.28), or the case n=3, r=1 occurring for the Bénard–Marangoni model (D.4) below.

For such models, integrating the conservative u 1 equation in the traveling-wave ODE, and writing as an N×N first-order system, we obtain from the requirement of periodicity N constraints, while we have N+nr+2 degrees of freedom consisting of the initial condition u(0), the wave number \(k\in\mathbb{R}\), the speed \(c\in \mathbb{R}\), and the constant of integration \(q_{1}\in\mathbb{R}^{n-r}\) arising from integration of the u 1 equation; thus, we expect generically a manifold of periodic solutions of dimension nr+2. In the reaction-diffusion case r=n, this returns the familiar value 2, or, up to translation, a one-dimensional family (generically) indexed by wave number k. In the conservation law case r=0, it returns the value n+2, leading, up to translation, to an (n+1)-dimensional family as in hypothesis (H2) of the introduction.

Substituting this value nr+1 in hypotheses (H2) and (D3), therefore, we readily obtain by the same derivation as for (1.7) a modified Whitham system consisting of the (nr+1)×(nr+1) system of viscous conservation laws

$$ \begin{aligned} {\mathcal{M}}_t+ \bar{k}(F-\bar{c}{\mathcal{M}})_x &=\bar{k}^2(d_{11} {\mathcal{M}}_x + d_{12}\kappa_x)_x, \\ \kappa_t+\bar{k}(-\omega-\bar{c}\kappa)_x&=\bar {k}^2(d_{21}{\mathcal{M}}_x + d_{22}\kappa_x)_x, \end{aligned} $$
(D.2)

where \(\omega({\mathcal{M}},\kappa)=-\kappa c({\mathcal{M}},\kappa)\) denotes time frequency, \({\mathcal{M}}:= \int_{0}^{1} U_{1}^{{\mathcal{M}},\kappa}(x) dx\) and

$$\begin{aligned} &F({\mathcal{M}},\kappa)\\ &\quad :=\int_0^1 \biggl( f_1\bigl(U^{{\mathcal{M}},\kappa}(x)\bigr) -\sum _j \bigl(B^j_{11},B^j_{12} \bigr) \bigl( U^{{\mathcal{M}},\kappa, \dots}(x)\bigr) \bigl(U^{{\mathcal{M}},\kappa} \bigr)'(x)\biggr) dx\end{aligned} $$

denote mean and mean “total flux” in the u 1 coordinate, and \(d_{ij}({\mathcal{M}},\kappa)\) are determined by higher-order corrections. (Note that, for B j≡constant, the terms involving B j are perfect derivatives, so disappear; this explains the fact that they were not present in the discussion of the second-order semilinear case.)

Likewise, we obtain in straightforward fashion analogs of the spectral preparation results of Lemma 1.5, Proposition 1.7, and Proposition 3.1, thus yielding corresponding linear bounds on critical modes \(s^{\rm p}\). Note that the slow decay rates that may arise at the linear level from a possible Jordan block will still be compensated by the special structure of the nonlinear terms, coming now in the form

$$\mathcal{N}= \partial_t\mathcal{N}_0+ \partial_x\mathcal{N}_1 + \begin{pmatrix}0_{(n-r)\times (n-r)}&0_{(n-r)\times r}\\0_{r\times(n-r)}&{\rm Id }_{r\times r}\\ \end{pmatrix} \mathcal{N}_2. $$

See for example [48] for a careful derivation of the second-order derivative Whitham system up to linear and quadratic order in first-order derivative terms, and [30] for a proof of the needed spectral preparation results in the Saint-Venant case (1.28). Indeed, so long as the nonlinear structure of the equations permits a nonlinear damping estimate as in Proposition 2.5, and high-frequency linearized resolvent estimates as needed to apply Prüss’ Theorem in estimating high-frequency linearized behavior as in the proof of (3.18) and (3.19) above, we obtain again (modulo increase in the exponent of regularity K) the stability results of Theorem 1.10 and Proposition 4.5, and a partial version of Theorem 1.12 describing comparisons to a second-order hyperbolic-parabolic system agreeing with (D.2) in its linearization about the constant state \((\bar{M},\bar{k})\), yielding again the result (1.21), of Corollary 1.18 asserting decay with respect to localized perturbations for linearly phase-decoupled systems.

That is, we obtain in this case exactly the conclusions cited in the examples of the introduction, obtained by examination of the linearization of the first-order part of the Whitham equations.

To recover the full result (1.17)–(1.19) of Theorem 1.12 showing convergence to the exact Whitham system, one also needs an analog of Lemma 4.14. But, the only difference between the (formal) computations of the derivation in Sect. B.1.2 and the ones of Lemma 4.14 is that the former are carried out before the implicit change of variables, while the latter are carried out after. Thus, analogs of Lemma 4.14 essentially follow by commutation of an implicit change of variables and expansions to a desired order.

4.3 D.3 Verification of (H1)–(H3), (D1)–(D3)

Regarding verification of our stability hypotheses, we recall that, assuming the trivial regularity hypothesis (H1), hypothesis (H2) is implied by (D1)–(D3), by Lemma 1.6, while (H3) by Proposition 1.7 can generally be verified by the same spectral expansion process needed to verify (D2). Meanwhile, (D1)–(D3) can be verified numerically by Galerkin approximation or numerical Evans function computation (see, e.g., [3, 4, 18]), or, in some cases, analytically, using bifurcation theory (see, e.g., [59]) or singular perturbations (see, e.g., [31]). For general discussion, see [4, 5, 30].

4.4 D.4 Applications revisited

We now discuss previous examples and some new ones in a bit more depth.

4.4.1 D.4.1 The Korteweg–de Vries/Kuramoto–Sivashinsky equation

A more canonical form of (1.26) is

$$u_t+\gamma\partial_x^4u+\varepsilon \partial_x^3 u + \delta\partial_x^2u+ \partial_x f(u) =0, \quad \gamma,\delta>0, $$

modeling phenomena from plasma and flame-front instabilities to inclined thin-film flow [38, 48, 54, 63]. As a fourth-order parabolic equation, this fits the framework of Sect. D.1, so that all of the results of this paper apply. Spectral stability has been studied in detail in [3, 12], indicating the existence of both spectrally stable and unstable waves; in particular, (H1)–(H3) and (D1)–(D3) have been shown in [3] to hold for a wide variety of waves. We note that stability under these hypotheses has been proven for localized perturbations in [3]; the new observations here are asymptotic behavior, and decay for nonlocalized perturbations.

4.4.2 D.4.2 The Saint-Venant equations

Recall, in Lagrangian coordinates, the Saint-Venant equations

$$ \begin{aligned} \tau_t - u_x&= 0, \\ u_t+ \bigl(\bigl(2F^2\bigr)^{-1}\tau^{-2} \bigr)_x&= 1- \tau u^2 +\nu\bigl(\tau^{-2}u_x \bigr)_x , \end{aligned} $$
(D.3)

where τ:=h −1, h is fluid height, u is fluid velocity, and x is a Lagrangian marker. These are not parabolic, yet nonlinear damping and high-frequency resolvent estimates can still be carried out, yielding by the discussion of Sect. D.2 all of the results of this paper. Specifically, nonlinear damping is established in [30, Proposition 4.4] (proved in Appendix A of the reference), under the “slope condition” \(\nu\bar{u}_{x}<F^{-1}\), where \(\bar{U}=(\bar{\tau}, \bar{u})\), a technical condition that appears to hold in most cases of interest, but which we expect can be dropped.Footnote 33 Again, the new observation here is asymptotic behavior, and also stability under nonlocalized perturbations, stability under localized data having been established in [30]. As noted earlier, (D.3) is a balance law rather than a conservation law, with nonconservative source term g.

4.4.3 D.4.3 The capillary Saint-Venant equations

With capillary pressure effects, (D.3) becomes

$$ \begin{aligned} \tau_t - u_x&= 0, \\ u_t+ \bigl(\bigl(2F^2\bigr)^{-1}\tau^{-2} \bigr)_x&= 1- \tau u^2 +\nu\bigl(\tau^{-2}u_x \bigr)_x -\sigma\biggl(\tau^{-5}\tau_{xx}- \frac{5}{2} \tau^{-6}(\tau_x)^2 \biggr)_x, \end{aligned} $$

where σ>0 is the coefficient of capillarity. These equations can be reduced by Kotschote’s [36] method of auxiliary variables (introducing z:=τ x ) to a 3×3 second-order quasilinear parabolic system

$$ \begin{aligned} \tau_t - u_x+z_x&= \tau_{xx}, \\ z_t &= u_{xx}, \\ u_t+ \bigl((2F)^{-1}\tau^{-2} \bigr)_x&= 1- \tau u^2 +\nu\bigl(\tau^{-2}u_x \bigr)_x -\sigma\biggl(\tau^{-5}z_x- \frac{5}{2} \tau^{-6}z^2\biggr)_x \end{aligned} $$

to which standard techniques can be applied [69, 75]. This fits the framework of Sect. D.1, yielding all of the results of this paper, the only change being in the regularity assumptions on data, which must be incremented by one to accommodate the new variable z=τ x . Existence and spectral stability or instability of these waves is a topic of ongoing investigation [2].

4.4.4 D.4.4 Bénard–Marangoni flow

A qualitative model introduced in [20] for Bénard–Marangoni flow, or flow driven by temperature-induced surface tension variation, is

$$ \begin{aligned} u_t &= -(1+u_{xx})_{xx} + \varepsilon ^2 u + f(u,v_x, w_x), \\ v_t&= v_{xx}+v_x + g_1(u,v,w)_x, \\ w_t&= w_{xx} -w_x + g_2(u,v,w)_x, \end{aligned} $$
(D.4)

with f(u,v x ,w x )=−u 3+γ(uv x +uw x ), g 1(u,v,w)=−uv, g 2(u,v,w)=−uw. Though of mixed fourth-order parabolic/second-order parabolic form, it is readily seen that these equations are both sectorial and admit a nonlinear damping estimate; moreover, they are of the mixed conservative/nonconservative form (D.1). Thus, by the discussion of Sects. D.1 and D.2, the main results of this paper apply, giving stability and behavior in terms of a 3×3 hyperbolic–parabolic system agreeing with the Whitham system (D.2).

Let us now discuss existence, the form of the Whitham equations, and validation of (D1)–(D3). Setting vw≡0, we find that the equations reduce to the Swift–Hohenberg equation (1.23) for u, with bifurcation parameter r=ε 2 restricted to the positive side of the bifurcation point r=0 at which periodic solutions appear. Thus, we inherit from the Swift–Hohenberg equations a special class of periodic solutions with (v,w) vanishing. Up to translation, such solutions are given by the 2-parameter family of zero-speed \(\frac{ 2\pi}{1+\varepsilon \omega}\)-periodic Swift–Hohenberg solutions

$$ \begin{aligned} &\bar{U}^{\omega,\varepsilon }(x)= \frac{2 \varepsilon (\sqrt{1-4\omega^2}\,)}{\sqrt{3}}\cos \bigl((1+ \varepsilon \omega)x\bigr) +\mathcal{O}\bigl(\varepsilon ^2\bigr),\\ & \bigl(\bar{V}^{\omega,\varepsilon },\bar{W}^{\omega,\varepsilon }\bigr) (x)\equiv(0,0),\end{aligned} $$
(D.5)

where ε, recall, is the bifurcation parameter, a fixed constant in (D.4). However, there are many other solutions for which \((\bar{V},\bar{W}) \not\equiv(0,0)\), yielding an additional two parameters in the description of nearby periodic traveling waves. Moreover, though the Swift–Hohenberg solutions are zero speed as a result of reflection symmetry (see Remark 1.27), reflection symmetry of (D.4) is broken as soon as \((v,w)\not\equiv(0,0)\), and so in general these waves may have arbitrary speed. It is our expectation, therefore, that the Whitham system is not phase-decoupled even about such special waves.

Numerical experiment by Galerkin approximation in [70] indicate that solutions (D.5) satisfy stability conditions (D1)–(D3) for ω=0 and ε>0 in a moderate range. Here, we demonstrate the same conclusion for \(|\omega|< 1/2\sqrt{3}\) and ε≪1, using decoupling of the equations and known analytical results for the Swift–Hohenberg equation, at the same time obtaining the limiting ε→0 coefficients of the linearized Whitham system about \((\bar{M},\bar{k})=(0,\bar{k})\). It would be interesting to carry out a systematic numerical stability investigation as in [3, 7] on the entire parameter range, and in particular to determine phase-coupling or decoupling of the associated Whitham system.

Proof of (D1)–(D3)

About the special solutions (D.5), the linearized eigenvalue equations are

$$ \begin{aligned} \lambda u&= L^0u + Mv + Nw, \\ \lambda v&= L^+ v, \\ \lambda w&= L^-w, \end{aligned} $$
(D.6)

where L 0 is the linearized operator of the Swift–Hohenberg equation about \(\bar{U}\) and

$$L^\pm:= \partial_x^2 \pm \partial_x -\partial_x \bar{U}^{\omega ,\varepsilon }. $$

By upper triangular form of (D.6), the eigenvalues of L ξ , counted by algebraic multiplicity, consist of the union of the eigenvalues of \(L^{0}_{\xi}\) and \(L^{\pm}_{\xi}\). Let us first consider the eigenvalues of the Swift–Hohenberg operator \(L^{0}_{\xi}\). In [13, 14, 17] (see also [44, 59]) it was analytically verifiedFootnote 34 that for ε≪1, solutions \(\bar{U}^{\omega,\varepsilon }\) in (D.5) are spectrally stable for

$$\bigl \vert 4\omega^2\bigr \vert <\frac{1}{3}+\mathcal{O}( \varepsilon ) $$

(in particular, for ω=0). From the fact that the waves are of speed c≡0, we find that the characteristic speed of the associated scalar Whitham equation is a 0≡0, and the associated critical mode has expansion λ 0(ξ)=−d 0 ξ 2. Turning to the operators L ±, and noting that \(\bar{U}^{\omega,\varepsilon }\to0\) uniformly in all derivatives as ε→0, we find that as ε→0 their eigenvalues approach uniformly the eigenvalues of the limiting constant-coefficient operators

$$\bar{L}^\pm_\xi:= (\partial_x+i \xi)^2 \pm(\partial_x+i\xi), $$

which, by direct (discrete Fourier transform) computation, are

$$\bar{\lambda}^\pm(\xi)= -(j+\xi)^2 \pm i(j+\xi), $$

where the Fourier frequency j runs through the integers. By continuity, these are therefore spectrally stable for |ε|≪1, with approximate critical mode expansions (obtained at j=0) of ±ξ 2. Combining these facts, we find that the limiting linearized Whitham system has characteristic speeds a j =0,±1, with corresponding (diagonal) viscosity coefficients d, 1, 1. This verifies (D1)–(D3) and (by distinctness of a j ) (H3) for |ε| sufficiently small, yielding spectral stability by the discussion of Sect. D.3. □

4.4.5 D.4.5 Inclined Marangoni flow

The related inclined thin-film equation

$$ H_t + \bigl(H^2-H^3 \bigr)_x=-\bigl(H^3H_{xxx} \bigr)_x $$
(D.7)

models Marangoni flow driven by a thermal gradient up an inclined silicon wafer, where H denotes fluid height [10, 11]. As a cousin of the Kuramoto–Sivashinsky equation, it would be interesting to investigate whether this model too supports stable periodic traveling-waves solutions.

4.4.6 D.4.6 Surfactant-driven Marangoni flow

Finally, we mention the surfactant model [43]

$$ \begin{aligned} H_t + \frac{1}{2} \bigl(H^2 \sigma'(\varGamma) \varGamma_x \bigr)_x&=0, \\ \partial_t \varGamma+ \partial_x\bigl(\varGamma H \sigma'(\varGamma) \partial_x \varGamma\bigr)&= {\rm Pe}_{\rm s}^{-1} \partial_x^4 \varGamma, \end{aligned} $$
(D.8)

modeling flow in a thin horizontal film driven by surfactant induced gradients in surface tension, where H is fluid height and Γ surface surfactant concentration, and \({\rm Pe}_{\rm s}\) is the modified Peclet number, a dimensionless constant, and σ(Γ)=1−Γ is an equation of state encoding the dependence of surface tension on surfactant density. Like (D.7), this appears to be an interesting example for study by the methods developed here and in [3, 25, 28, 30]. Note, as the second equation is conservative, that the associated Whitham approximation is indeed of system form.

Rights and permissions

Reprints and permissions

About this article

Cite this article

Johnson, M.A., Noble, P., Rodrigues, L.M. et al. Behavior of periodic solutions of viscous conservation laws under localized and nonlocalized perturbations. Invent. math. 197, 115–213 (2014). https://doi.org/10.1007/s00222-013-0481-0

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00222-013-0481-0

Mathematics Subject Classification (2010)

Navigation