December 31, 2007

10:11

WSPC - Proceedings Trim Size: 9in x 6in

nontrivial

NONTRIVIAL LARGE-TIME DYNAMICS IN REACTION-DIFFUSION EQUATIONS

M. BAGES , P. MARTINEZ and J.-M. ROQUEJOFFRE Institut de Math ematiques, Universit e Paul Sabatier, F-31062 Toulouse, France E-mail: bages@mip.ups-tlse.fr, E-mail: martinez@mip.ups-tlse.fr E-mail: roque@mip.ups-tlse.fr V. ROUSSIER-MICHON Institut de Math ematiques, INSA de Toulouse, F-31077 Toulouse, France E-mail: roussier@insa-toulouse.fr Solutions of bistable reaction-diusion equations, posed on the whole real line, will converge to travelling waves (Fife, McLeod). When the same equation is posed on the plane, or in the N -dimensional space, suciently wellprepared initial data will still give raise to convergence to 1D travelling waves (Levermore-Xin; Kapitula). We explain here why, when the initial datum is slightly less well-prepared than in the Levermore-Xin setting, the solution will not converge to a single 1D wave; it will instead undergo some diusive behaviour. We also discuss what happens when the level sets of the initial datum are conical-shaped. Finally, we describe what happens for KPP equations: nontrivial behaviour occurs even in one space dimension. Keywords : Reaction-diusion equations; travelling fronts; Pulsating fronts; nontrivial dynamics.

1. Introduction; motivation Consider the reaction-diusion equation ut uxx = f (u) (t > 0, x I R), u(t, ) = 0, u(t, +) = 1
1

(1)

where the function f is of the bistable, unbalanced type: f (0) = f (1) = 0; f (0), f (1) < 0; f < 0 on (0, ), f > 0 on (, 1);
0

f > 0.

It is well-known1 that the solutions of (1) will converge to travelling waves, i.e. solutions of the form 0 (x + c0 t). The prole 0 satises 0 + c0 0 = f (0 ), (x I R); 0 () = 0, 0 (+) = 1 (2)

December 31, 2007

10:11

WSPC - Proceedings Trim Size: 9in x 6in

nontrivial

Suppose now that (1) is an equation in the plane I R2 , i.e. ut u = f (u), t > 0, (x, y ) I R2 (3)

with an initial datum u0 whose level sets are at bounded distance from horizontal straight lines. Under the assumption that := u0 0 H 1 (I R2 ) << 1, we have - see Xin,2 Levermore-Xin,3 Kapitula4 - for some > 0: u(t, x, y ) 0 (y + c0 t) = O(t ), uniformly in (x, y ) I R2 . Notice that the assumption u0 0 H 1 (I R2 ) is somewhat more stringent than simply supposing level sets trapped between straight lines: it says that the level sets of u0 are asymptotic to horizontal straight lines. A natural question is therefore whether something drastically dierent occurs if we relax it. The goal of this paper is to review some recent results obtained by the authors of the present paper. To sum up: unless the initial datum of (3) is really of the Xin, Levermore-Xin or Kapitula form, convergence to a single wave is never likely to occur. The prole of the solution does converge to that of a wave, but with a spatial shift whih does not converge to any value as t +. In Section 2, we review how two-dimensional eects in the bistable case (3) prevent convergence; in Section 3 we describe what happens when f is of the KPP type: here, non-convergence occurs even in the one-dimensional case. 2. Planar and conical waves for bistable equations Let us start with almost planar initial data for (3). Theorem 2.1. (Roquejore, Roussier-Michon 5 ) Given u0 C (I R2 ), assume the existence of two reals y1 y2 such that (x, y ) I R2 : 0 (y + y1 ) u0 (x, y ) 0 (y + y2 ), where 0 (y ) is a solution of (2). Then there is t0 > 0 and a function s(t, x) C 2 ([t0 , +)I R) such that the solution u(t, x, y ) of (3), emanating from u0 , satises, for all (0, 1): sup
tt0 ,(x,y )I R2

|u(t, x, y ) 0 (y + c0 t + s(t, x))| = O(t1 ).

(4)

Moreover, for all (0, 1), there is C (u0 ) > 0 such that the function (t, x) := ec0 s(t,x)/2 satises, for t t0 : |t xx | C (u0 ) . (1 + t)22 (5)

December 31, 2007

10:11

WSPC - Proceedings Trim Size: 9in x 6in

nontrivial

Equation (5) indicates that the solution has a local shift that behaves, as t +, as a solution of the heat equation. We know, since ColletEckmann,6 that a lot of solutions of the heat equation do not converge as t +, while all the derivatives go to 0. In our case it is possible5 to construct initial data to (3) that will not converge to a single wave - while the large-time prole is that of the wave. Equation (3) has genuinely nonplanar travelling wave solutions, still propagating downwards; of the form u(t, x, y ) = (x, y + ct), solving + cy = f () in I R2 ; (x, ) = 0, (x, +) = 1 (6)

where the limit is now taken pointwise in x. We restrict ourselves to those waves whose level sets are globally Lipschitz graphs; we then may7,8 classify them all; moreover we have very precise stability properties for them The following properties of these waves are extracted from.79 and9 - see also10 for the existence part: (P1) For a solution of (6) whose level sets are globally Lipschitz graphs we have c c0 - the planar wave speed; moreover 0 < < 1 in I R2 , (P2) (x, y ) = (|x|, y ), |x| 0, y > 0, c0 , and the function satises (P3) There is (0, ] such that c = 2 sin lim sup (1 (x, y )) = 0, lim sup (x, y ) = 0.
A+, y A|x|cot A, y A|x|cot

(P4) the function is decreasing in any unit direction = (x , y ) I R2 such that y < cos , (P5) there is exponential convergence of (x, y ) to the planar fronts 0 (x cos + y sin ) in the directions ( sin , cos ); moreover the slopes of the level lines of converge exponentially, in the same directions, to cot. (P6) Let u0 (x, y ) be a - two-dimensional - Cauchy datum to (3), satisfying |u0 (x, y ) (x, y )| = O(e(|x|+|y|) ), where is some positive number, and (x, y ) a solution of (6) - hence a conical-shaped solution. Then we have, for some > 0 uniformly in (x, y ) I R2 : u(t, x, y ) (x, y + ct) = O(et ). Notice that the assumption on u0 in (P5) is similar to the XinLevermore-Kapitula one, and we wonder what happens if this assumption is slightly relaxed. In what follows, (X , Y ) is the refrence frame obtained by rotating the axes (x, y ) by an angle . Theorem 2.2. Consider a smooth Cauchy datum u0 such that: - there exist a small > 0 and a couple (X1 , X2 ) I R2 I R2 such that ((x, y ) + X1 ) u0 (x, y ) ((x, y ) + X2 ), |X1 X2 | , (7)

December 31, 2007

10:11

WSPC - Proceedings Trim Size: 9in x 6in

nontrivial

- there holds y u0 > 0. Moreover there is > 0, with lim that lim sup X X u0 (X , .)
X + L (I R)

4 .

= 0, such 4

Choose (0, 1), let the set {u0 (x, y ) = } be written as {Y+ = s+ ( 0 X+ )} - resp. {Y = s0 (X )} in the right half-plane {x > 0} - resp. in the left half-plane {x < 0} (the dependence in is deleted for commod ity). Dene the functions 0 (X ) as
0 (X ) =

ec0 s0 (X )/2 if X 1 ec0 s0 (1)/2 if X 1

(8)

Let (t, X ) be the solutions of the advection-diusion equations (t X X c cos X ) = 0,

(0, X ) = 0 (X )

(9)

Let u(t, x, y ) be the solution of (3) emanating from u0 . For a given (0, 1), there exists A > 0 such that the set {u(t, x, y ) = } can be described as of the form {Y+ = + (t, X+ )} in the half-plane {x A} resp. {Y = (t, X )} in the half-plane {x A}. Moreover there is a constant C > 0 - possibly going to + as 0 - and another constant 1 C > 0 independent of , such that there holds, for all (0, ), and 2 uniformly in (t, x, y ) I R+ I R2 : | (t, X ) Log (t, X )| C 1 + e(|x|+|y|) (1 + t)12
2 + C/ .

We believe that the size restrictions on the initial datum can be removed. However, Theorem 2.2 is still sucient to exhibit nontrivial large-time dynamics. 3. Nontrivial dynamics for KPP type equations We consider here the one-dimensional Cauchy problem ut uxx = f (x, u), (t > 0, x I R) u(t, ) = 0, u(t, +) = 1 (10)

where the nonlinear term f is 1-periodic in x, and of KPP (KolmogorovPetrovskii-Piskunov) type: f (x, 0) = f (x, 1) = 0; f (x, u) > 0 if 0 < u < 1, fuu (x, .) < 0 on [0, 1]. Rather than that of travelling waves, the relevant notion is that of pulsating waves, i.e. solutions (t, x) of (10) for which there exists c > 0 such that 1 t (t, x ct) is -periodic. When f does not depend on x, existence c

December 31, 2007

10:11

WSPC - Proceedings Trim Size: 9in x 6in

nontrivial

and stability of travelling waves is an old problem dating back to;11 see for instance.12 In the x-dependent case, existence theory qualitative theory are very well developped.13,14 Essentially: there is c > 0 such that (11) has pulsating wave solutions c of velocity c if and only if c c . Moreover, t c > 0. The minimal speed c is characterised as follows: set (x) := fu (x, 0) > 0; let k () be the least eigenvalue of the operator L := dxx 2dx (2 + (x)), acting on the space of 1-periodic functions. Then c = k () min>0 . Moreover each c behaves at like ec (x+ct) (x), where c the smallest (positive) root of k () = c, and (L c) = 0. To study the large-time behaviour of (10), we need a slightly more precise information. Proposition 3.1. 15,16 For c > c , (10) has a unique (pulsating) wave solution c (t, x) such that, as x + ct : c (t, x) = ec (x+ct) (x) + O(e(c +)(x+ct) ). Consider the case c > c . An easy follow-up of the argument of the preceding proposition proves the following: consider an initial datum u0 (x) for (10), such that there is a pulsating wave c and > 0 satisfying u0 (x) = c (0, x)(1 + O(ex )) as x . (11)

Then, as t + we have, uniformly in x: u(t, x) = c (t, x)(1 + O(et ) for some > 0. We wish to understand what happens when the - quite stringent - assumption (11) is slightly relaxed. A natural way to loosen it is to assume that u0 is trapped between two translates - in time - of a pulsating wave; the beaviour of u(t, x) is given by the following theorem - whose proof is much more dicult than that of Proposition 3.1: Theorem 3.1. Assume there is M > 0 such that c (M, x) u0 (x) c (M, x). Let m(t, x) solve mt mxx 2(c + c )mx cc m2 x = 0; c m(0, x) = c (., x)1 (u0 (x))

Then, as t +: [i] we have mt (t, .) = O(t1 ), mx (t, .) = O(t1/2 ); [ii] we have, uniformly in x I R: |u(t, x) c (t + m(t, x), x)| = O(t1/2 ). The following remarks are in order. First, we point out that Theorem 3.1 is new, even in the x-independent case. Then, let us mention that nontrivial

December 31, 2007

10:11

WSPC - Proceedings Trim Size: 9in x 6in

nontrivial

dynamics does occur: once again in the x-independent case, the ColletEckman example can be adapted. Part [i] of this theorem is the most involved: it relies on very precise asymptotics of a heat kernel that the current literature17 cannot treat. Let us nally point out a consequence of this part: although we cannot treat the case c = c in the same generality, we have Theorem 3.2. 15,16 Assume that, as x : u0 (x) = c (0, x)(1 + O(ex )) for some > 0. Then, as t +: u(t, x) c (t, x) = O(t1/2 ). The stability of the wave with speed c is, as opposed to the case c > c , a delicate matter. In the x-independent case, the optimal result is due to Gallay.18 When the initial datum does not decay exactly as the wave, investigation of the dynamics is currently going on. Acknowledgment. The third author is grateful to Prof. E. Yanagida for inviting him in the minisymposium Qualitative properies of parabolic equations. References
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. Fife, P.C. and McLeod, J.B., Arch. Rat. Mech. Anal. 65, 335 (2003). Xin, J.X., C.P.D.E. 17 1889 (1992). Levermore, C. D. and Xin, J.X., C.P.D.E. 17 1901 (1992). Kapitula, T., Trans. Amer. Math. Soc. 349 257 (1997) Roquejore, J.-M. and Roussier-Michon, V., to appear in Ann. Mat. Pura Appl. Collet, P. and Eckmann, J.-P. Nonlinearity 5 1265 (1992). Hamel, F; Monneau, R. and Roquejore, J.-M. Ann. Sci. Ec. Norm. Sup. 37 469 (2004). Hamel, F; Monneau, R. and Roquejore, J.-M. D.C.D.S. A, 13 1096 (2005). Hamel, F; Monneau, R. and Roquejore, J.-M. D.C.D.S. A, 14 75 (2006). Ninomiya, H. and Taniguchi, M. J. Di. Eq., 213 204 (2005). Kolmogorov, A.; Petrovskii, I. and N. Piskunov, Bjul. Mosk. Gos. Univ., 17 1 (1937). Uchiyama, K. J. Math. Kyoto Univ., 18 453 (1978). Berestycki, H. and Hamel, F. C.P.A.M., 55 949 (2002). Hamel, F. To appear in J. Math Pures Appl. Bages, M. PhD Thesis (2007). Bages, M.; Martinez, P. and Roquejore, J.-M., in preparation. Norris, J.R., Arch. Rat. Mech. Anal., 140 161 (1997). Gallay, T. Nonlinearity 7 741 (1994).