chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 1 1 1 2 2 CHAPTER 6 3 3 4 4 5 5 6 6 Existence and Stability of Spikes for the Gierer–Meinhardt System 7 8 7 8 9 9 10 10 11 11 12 12 Juncheng Wei 13 13 Department of Mathematics, The Chinese University of Hong Kong, Shatin, Hong Kong E-mail: wei@math.cuhk.edu.hk 14 14 15 15 16 16 17 Contents 18 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Steady states in shadow system case . . . . . . . . . . . . . . . . . . . . 2.1. Reduction to single equation . . . . . . . . . . . . . . . . . . . . . 2.2. Subcritical case: spikes to (2.4) . . . . . . . . . . . . . . . . . . . 2.3. Localized energy method (LEM) . . . . . . . . . . . . . . . . . . . 2.4. Bubbles to (2.4): the critical case . . . . . . . . . . . . . . . . . . . 2.5. Bubbles to (2.4): slightly supercritical case . . . . . . . . . . . . . 2.6. Concentration on higher-dimensional sets . . . . . . . . . . . . . . 2.7. Robin boundary condition . . . . . . . . . . . . . . . . . . . . . . 3. Stability and instability in the shadow system case . . . . . . . . . . . . 3.1. Small eigenvalues for L . . . . . . . . . . . . . . . . . . . . . . . 3.2. A reduction lemma . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Large eigenvalues: NLEP method . . . . . . . . . . . . . . . . . . 3.4. Uniqueness of Hopf bifurcations . . . . . . . . . . . . . . . . . . . 3.5. Finite case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6. The stability of boundary spikes for the Robin boundary condition 4. Full Gierer–Meinhardt system: One-dimensional case . . . . . . . . . . 4.1. Bound states: the case of Ω = R1 . . . . . . . . . . . . . . . . . . 4.2. The bounded domain case: Existence of symmetric K-spikes . . . 4.3. The bounded domain case: existence of asymmetric K-spikes . . . 4.4. Classification of asymmetric patterns . . . . . . . . . . . . . . . . 4.5. The stability of symmetric and asymmetric K-spikes . . . . . . . 5. The full Gierer–Meinhardt system: Two-dimensional case . . . . . . . 5.1. Bound states: spikes on polygons . . . . . . . . . . . . . . . . . . 5.2. Existence of symmetric K-spots . . . . . . . . . . . . . . . . . . . 5.3. Existence of multiple asymmetric spots . . . . . . . . . . . . . . . 5.4. Stability of symmetric K-spots . . . . . . . . . . . . . . . . . . . . 5.5. Stability of asymmetric K-spots . . . . . . . . . . . . . . . . . . . 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 17 42 43 44 45 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489 491 491 492 495 512 514 514 516 519 519 522 523 535 537 541 542 543 543 544 551 555 557 557 561 563 570 575 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 HANDBOOK OF DIFFERENTIAL EQUATIONS Stationary Partial Differential Equations, volume 5 Edited by M. Chipot © 2008 Elsevier B.V. All rights reserved 43 44 45 487 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 488 1 2 3 4 6. High-dimensional case: N 3 7. Conclusions and remarks . . . Acknowledgments . . . . . . . . References . . . . . . . . . . . . F:chipot506.tex; VTEX/Rita p. 2 J. Wei . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 576 580 581 581 1 2 3 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 41 41 42 42 43 43 44 44 45 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 3 Existence and stability of spikes 1 489 1. Introduction 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 2 It is a common belief that diffusion is a smoothing and trivializing process. Indeed, this is the case for a single diffusion equation. Consider the heat equation ⎧ ⎨ ut = u u(x, 0) = u0 (x) 0 ⎩ ∂u ∂ν = 0 6 (1.1) in Ω × (0, +∞), in Ω, on ∂Ω × (0, +∞), 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 9 11 12 13 15 (1.2) 16 17 18 ⎧ u = D u + f (u, v) t u ⎪ ⎪ ⎨ v = D v + g(u, v) t v u(x, 0) = u (x), ⎪ ⎪ ⎩ ∂u ∂v 0 ∂ν = ∂ν = 0 v(x, 0) = v0 (x) in Ω × (0, +∞), in Ω × (0, +∞), in Ω, on ∂Ω × (0, +∞). 19 20 21 22 23 24 25 26 27 28 (1.3) 29 30 31 32 In 1957, Turing [68] proposed a mathematical model for morphogenesis, which describes the development of complex organisms from a single shell. He speculated that localized peaks in concentration of a chemical substance, known as an inducer or morphogen, could be responsible for a group of cells developing differently from the surrounding cells. He then demonstrated, with linear analysis, how a non-linear reaction diffusion system like (1.3) could possibly generate such isolated peaks. Later in 1972, Gierer and Meinhardt [21] demonstrated the existence of such solution numerically for the following (so-called Gierer–Meinhardt system) 33 34 35 36 37 38 39 40 41 42 43 45 10 14 it is known that when Ω is convex, the only stable solutions are constants [5,46]. Thus there are only trivial patterns (constant solutions) for single reaction–diffusion equations (on convex domains). On the other hand, it is important to be able to use diffusion (and reaction) to model pattern formations in various branches of science (e.g., biology and chemistry). One important question is: can we get non-trivial patterns (stable non-trivial solutions) for systems of reaction–diffusion equations? Let us consider the following system of reaction–diffusion equations: 41 44 7 8 18 19 4 5 in Ω × (0, +∞), in Ω, on ∂Ω × (0, +∞). Assume that u0 (x) iscontinuous. It is known that u(x, t) is smooth for t > 0 (smooth1 ing), and u(x, t) → |Ω| Ω u0 (x) dx as t → +∞ (trivializing). A similar result holds when a source/sink term (or a reaction term) is present. Namely, for the problem ⎧ ⎨ ut = u + f (u) u(x, 0) = u0 (x) 0 ⎩ ∂u ∂ν = 0 3 (GM) ⎧ ∂a ap 2 ⎪ ⎨ ∂t = a − a + hq , r τ ∂h = Dh − h + ahs , ∂t ⎪ ⎩ ∂a ∂h ∂ν = ∂ν = 0, x ∈ Ω, t > 0, 42 43 x ∈ Ω, t > 0, 44 x ∈ ∂Ω. 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 490 1 2 3 4 5 6 F:chipot506.tex; VTEX/Rita p. 4 J. Wei Here, the unknowns a = a(x, t) and h = h(x, t) represent the respective concentrations at point x ∈ Ω ⊂ RN and at time t of the biochemical called an activator and an inhibitor; ∂2 > 0, D > 0, τ > 0 are all positive constants; = N 2 is the Laplace operator in j =1 ∂xj RN ; Ω is a smooth bounded domain in RN ; ν(x) is the unit outer normal at x ∈ ∂Ω. The exponents (p, q, r, s) are assumed to satisfy the condition 7 8 p > 1, q > 0, r > 0, 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 s 0, and γ := at = −a + a p / hq , τ ht = −h + a r / hs . 2 3 4 5 6 7 qr > 1. (p − 1)(s + 1) 8 9 Gierer–Meinhardt system was used in [21] to model head formation in the hydra. Hydra, an animal of a few millimeters in length, is made up of approximately 100,000 cells of about fifteen different types. It consists of a “head” region located at one end along its length. Typical experiments on hydra involve removing part of the “head” region and transplanting it to other parts of the body column. Then, a new “head” will form if and only if the transplanted area is sufficiently far from the (old) head. These observations have led to the assumption of the existence of two chemical substances—a slowly diffusing (i.e., 1) activator a and a fast diffusing (i.e., D ) inhibitor h. To understand the dynamics of (GM), it is helpful to consider first its corresponding “kinetic system” 1 10 11 12 13 14 15 16 17 18 19 20 21 (1.4) qr it This system has a unique constant steady state a ≡ 1, h ≡ 1. For 0 < τ < (p−1)(s+1) is easy to see that the constant solution a ≡ 1, h ≡ 1 is stable as a steady state of (ODE). However, if √ is small, it is not hard to see that the constant steady state a ≡ 1, h ≡ D 1 of (GM) becomes unstable and bifurcation may occur. This phenomenon is generally referred to as Turing’s diffusion-driven instability. (A general criteria for this can be found in Murray’s book [47].) There are many other reaction–diffusion systems which exhibit Turing’s diffusiondriven instability: they include Gray–Scott model from chemical reactor theory, Schnakenberg model, Sel’kov model, Lengyl–Epstein model, Thomas model, Keener–Tyson model, Brusselator, Oregonator, etc. For introduction and discussion on these general Turing models, we refer to the book [47]. A survey of mathematical modeling of biological and chemical phenomena using reaction–diffusion systems is given in [38]. Mathematical modeling of patterns in biological morphogenesis using extensions of GM model are discussed in [36] and [48]. Several common characteristics of Turing type reaction–diffusion systems include: first, they are non-variational, i.e., they do not have Lyapunov or energy functional so standard variational (or energy) method cannot be applied; second, they are non-cooperative, i.e., they do not have Maximum Principles so sub-super-solution method cannot be applied; third, they support finite-amplitude spatial-temporal patterns of remarkable diversity and complexity, such as stable spikes, layers, stripes, spot-splitting, traveling waves, etc. (See [63].) The study of these RD systems not only increases our knowledge on Turing patterns, 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 5 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 491 but also induces new tools and techniques to deal with other problems which may share similar characteristics. The most interesting phenomena associated with (GM) is the existence of stable spikes and stripes. The numerical studies of [21] and more recent those of [31] have revealed that in the limit → 0, the (GM) system seems to have stable stationary solutions with the property that the activator concentration is localized around a finite number of points in Ω̄. Moreover, as → 0, the pattern exhibits a “spike layer phenomenon” by which we mean that the activator concentration is localized in narrower and narrower regions around some points and eventually shrinks to a certain number of points as → 0, whereas the maximum value of the activator concentration diverges to +∞. Such kind of point-condensation phenomena has generated a lot of interests both mathematically and biologically in recent years. The purpose of this paper is to report on the current trend and status of such studies (up to June, 2006). We shall not give most of proofs. For more details, please see the references and therein. In the study of spiky patterns (or concentration phenomena), two fundamental methods emerge. The first one is the so-called “Localized Energy Method”, or LEM in short. LEM is a combination of traditional Lyapunov–Schmidt reduction method with variational techniques. This is a very useful tool to construct solutions with various concentration behavior, such as spikes, layers, or vortices. The second method is the so-called “Nonlocal Eigenvalue Problem Method”, or NLEP in short. This deals with eigenvalue problems which are non-selfadjoint. It plays fundamental role in the study of stability of spike patterns. In this survey, I shall illustrate these two methods in details in the hope that they may find applications in other problems. Throughout this paper, unless otherwise stated, we always assume that 25 26 D is finite, τ 0. (1.5) 35 38 39 40 41 42 43 44 45 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 26 30 2.1. Reduction to single equation 31 32 In general, the full (GM) system is very difficult to study. A very useful idea, which goes back to Keener and Nishiura, is to consider the so-called shadow system. Namely, we let D → +∞ first. Suppose that the quantity −h + a p / hq remains bounded, then we obtain 36 37 5 29 32 34 4 28 2. Steady states in shadow system case 30 33 3 27 28 31 2 25 1, 27 29 1 33 34 35 36 h → 0, ∂h = 0 on ∂Ω. ∂ν (2.1) 38 Thus h(x, t) → ξ(t), a constant. To derive the equation for ξ(t), we integrate both sides of the equation for h over Ω and then we obtain the following so-called shadow system ⎧ 2 p q in Ω, ⎪ ⎨ at = a − a + a /ξ 1 r s τ ξt = −ξ + |Ω| Ω a dx/ξ , ⎪ ⎩ a > 0 in Ω and ∂a ∂ν = 0 on ∂Ω. 37 39 40 41 42 43 (2.2) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 492 1 J. Wei The advantage of shadow system is that by a simple scaling, 2 3 4 q a = ξ p−1 u, ξ= 5 6 7 8 9 10 11 12 ur 2 p−1 (p−1)(s+1)−qr 3 , (2.3) Ω 6 7 8 = 0 in Ω, and ∂u ∂ν = 0 on ∂Ω u > 0 in Ω (2.4) 11 whose energy functional is given by J [u] := Ω 12 2 19 20 21 22 23 24 25 26 27 13 1 1 p+1 dx, |∇u|2 + u2 − u 2 2 p+1 + where u+ = max(u, 0), 14 15 (2.5) 32 33 34 35 36 37 38 39 42 43 44 45 19 20 21 22 23 24 25 26 27 29 2.2. Subcritical case: spikes to (2.4) 30 +2 N +2 Let us assume first that 1 < p < ( N N −2 )+ (= N −2 if N 3; = +∞ when N = 1, 2). In this case, problem (2.4) can be studied by traditional variational methods, for example, Mountain-Pass method, or Nehari’s solution manifold method. For Mountain-Pass method, by taking a function e(x) ≡ k for some constant k in Ω, and choosing k large enough, we have J (e) < 0, for all ∈ (0, 1). Then for each ∈ (0, 1), we can define the so-called mountain-pass value c = inf max J h(t) h∈Γ 0t1 40 41 18 28 29 31 16 17 for u ∈ H 1 (Ω). First we give some definitions on solutions to (2.4). A family of solutions {u } to (2.4) are called concentrated solutions if there exists a subset Γ ⊂ Ω̄ such that u → 0 in 0 (Ω̄\Γ ) and max Cloc x∈Γ u (x) c0 > 0. If Γ consists of only points in Ω̄, these kind solutions are called point condensations. Among point condensations, there are two kinds: spikes and bubbles. Spikes are those concentrated solutions such that maxx∈Ω̄ u C, while bubbles are those with maxx∈Ω̄ u → +∞. If the dimension of Γ is positive, concentrated solutions are also called layers. (Similar definitions can also be given for solutions of the full Gierer–Meinhardt system by considering the activator a only.) In the following, we discuss the existence of all kinds of concentrated solutions to (2.4). 28 30 9 10 17 18 4 5 2 u − u + up 15 16 1 |Ω| 1 the stationary shadow system can be reduced to a single equation 13 14 F:chipot506.tex; VTEX/Rita p. 6 inf 32 33 34 35 36 37 38 (2.6) 39 40 41 where Γ = {h: [0, 1] → H 1 (Ω) | |h(t) is continuous, h(0) = 0, h(1) = e}. It is easy to see that (Lemma 2.1 of [57]), c can be characterized by c = 31 sup J [tu], u≡0, u∈H 1 (Ω) t>0 42 43 (2.7) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 7 Existence and stability of spikes 1 2 3 4 5 493 which can be shown to be the least among all non-zero critical values of J . (This formulation (2.7) is sometimes referred to as the Nehari manifold technique.) Moreover, c is attained by some function u which is then called a least-energy solution. In a series of papers [57] and [58], Ni and Takagi studied the so-called least energy solutions and proved the following theorem 6 7 8 9 17 18 = 0, w(0) = maxy∈RN w(y), w − w + w p 22 w > 0 in w → 0 at ∞. (2.8) c = N 27 30 31 32 33 34 35 36 37 1 I [w] − c1 H (P ) + o() 2 40 41 44 45 15 17 20 21 22 25 (2.9) 26 27 28 where I [w] = RN 1 1 |∇w|2 + w 2 − w p+1 2 p+1 29 30 31 32 2 order is given by is the energy of the ground state. A further expansion of c up to the [90] 2 2 N 1 2 I [w] − c1 H (P ) + c2 H (P ) + c3 R(P ) + o (2.10) c = 2 33 34 35 36 37 38 where c1 , c2 , c3 are generic constants and R(P ) is the scalar curvature at P . In particular c1 , c3 > 0. (When N = 2, a further expansion to the order of 3 is also given in [91].) Some applications of the formula (2.10) are given in [90]. 42 43 14 24 38 39 13 23 R EMARK 2.2.2. The proof of Theorem 2.1 is by expansion of energy: 26 29 9 19 R EMARK 2.2.1. The existence of ground state to (2.8) is well known. The radial symmetry of w follows from the famous Gidas–Ni–Nirenberg theorem [22]. The uniqueness of w is proved in [39]. 25 28 8 18 23 24 7 16 RN , 19 21 5 12 where H (P ) is the mean curvature function for P ∈ ∂Ω, and u (P + y) → w(y) uniformly in Ω,P = {y | P + y ∈ Ω}, where w(y) is the unique solution of the following 16 20 4 11 P ∈∂Ω 12 15 3 10 H (P ) → max H (P ), 11 14 2 6 T HEOREM 2.1. (See [57,58].) For sufficiently small, there exists a mountain-pass solution u which is also least-energy solution such that u has only one local maximum point 2 (Ω̄\{P }). Moreover, as → 0, P ∈ ∂Ω and u → 0 in Cloc 10 13 1 39 40 41 42 Since then there has been a lot of studies on problem (2.4). A general principle is that boundary spike solutions are related to the boundary mean-curvature H (P ), P ∈ ∂Ω, while interior spike solutions are related to the distance function d(P , ∂Ω). Note also that 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 494 1 2 3 4 5 6 7 8 9 10 11 F:chipot506.tex; VTEX/Rita p. 8 J. Wei d for boundary spike the order is usually O() while for interior spikes the order is O(e− ) for some d > 0. Let me mention some results on multiple boundary and interior peaked solutions. For single and multiple boundary spikes, Gui [26] first constructed multiple boundary spike solutions at multiple local maximum points of H (P ), using variational method. Wei [73], Wei and Winter [82,83] (independently by Bates, Dancer and Shi [4]) constructed single and multiple boundary spike solutions at multiple non-degenerate critical points of H (P ), using Lyapunov–Schmidt reduction method. Y.Y. Li [41], del Pino, Felmer and Wei [16] constructed single and multiple boundary spikes in the degeneracy case. Using Localized Energy method (LEM), a clustered solution is also constructed by Gui, Wei and Winter [29] (independently by Dancer and Yan [9]). 12 13 min H (P ) > min H (P ). (2.11) Γ ∂Γ 17 24 J [u ] = N 25 26 K I [w] − c1 2 29 j =1 37 38 45 16 18 19 20 27 . (2.12) 28 29 For single and multiple interior peaked solutions, the situation is quite different, as the errors are exponentially small. Wei [79,74] first constructed single interior peak solution at a strictly local maximum point of d(P , ∂Ω). Gui and Wei [27] proved the following 31 32 33 34 T HEOREM 2.3. (See [27].) For any fixed positive integer k, there exists k such that for < k , problem (2.4) has a solution u with k interior local maximum points Pj, ∈ Ω. Moreover, (P1, , . . . , Pk, ) approaches a limiting sphere-packing position, i.e., ϕk (P1, , . . . , Pk, ) → max (P1 ,...,Pk )∈Ω k ϕk (P1 , . . . , Pk ) (2.13) 41 44 15 30 40 43 11 26 i=j 39 42 10 25 34 36 9 24 30 35 8 22 H (Pj, ) |Pi, − Pj, | γ0 + o(1) w − 28 33 7 23 K 27 32 6 21 The energy expansion for K-boundary spikes is 23 31 5 17 Then for any fixed positive integer k, there exists k such that for < k , problem (2.4) has a solution u with k boundary local maximum points Pj, ∈ Γ . Furthermore, H (Pj, ) → minΓ H (P ). 21 22 4 14 16 20 3 13 T HEOREM 2.2. (See [9,29].) Let Γ be a subset of ∂Ω, where it holds 15 19 2 12 14 18 1 35 36 37 38 39 40 41 42 where ϕk (P1 , . . . , Pk ) = min |Pi − Pj |, 2d(Pl , ∂Ω) . i,j,l,i=j 43 (2.14) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 9 Existence and stability of spikes 1 2 3 4 495 The energy expansion for K-interior spikes is 1 2 K 2d(Pj, ,∂Ω) N e− J [u ] = KI [w] − γ0 3 4 j =1 5 |Pi, − Pj, | . − γ1 w 6 7 5 6 (2.15) i=j 8 9 10 11 12 13 16 17 20 23 26 27 28 29 30 31 1K αN,Ω,f N (| ln |)N where αN,Ω,p is a constant depending on N, Ω and p only, there exists a solution with K interior peaks. (An explicit formula for αN,Ω,p is also given.) As a consequence, we obtain α ] number of solutions. Moreover, that for sufficiently small, there exists at least [ N (|N,Ω,p ln |)N for each β ∈ (0, N ) there exists solution with energy in the order of N −β . 32 33 34 12 13 15 16 17 19 20 21 T HEOREM 2.5. (See [44].) There exists an 0 > 0 such that for 0 < < 0 and for each integer K bounded by 24 25 11 18 Theorems 2.2, 2.3 and 2.4 imply that the number of solutions to (2.4) goes to infinity as → 0. Recently, the following lower bound on number of solutions is obtained: 21 22 10 14 T HEOREM 2.4. (See [28].) For any two fixed positive integers k, l, there exists k,l such that for < k,l , problem (2.4) has a solution u with k interior local maximum points and l boundary maximum points. 18 19 8 9 Grossi, Pistoia and Wei [30] further showed that there is an one-to-one correspondence between the (sub-differential) critical points of ϕk and k-interior peaked solutions. Concerning the existence of mixed-boundary-interior-spikes, the following theorem gives a complete answer. 14 15 7 22 23 24 25 26 27 28 29 30 31 32 Theorems 2.2, 2.3, 2.4 and 2.5 can all be proved by the powerful method—Localized Energy Method—which was first introduced in [27]. We shall discuss it next. 33 34 35 35 36 36 37 2.3. Localized energy method (LEM) 38 39 40 41 42 43 44 45 37 38 We illustrate a general method in finding solutions with concentrating behavior—the socalled Localized Energy Method, or LEM in short. The advantage of such method is that it can be applied to subcritical, critical or supercritical problems, as long as the limiting solution is well analyzed. This method was introduced in Gui and Wei [27] in dealing with spikes. In the following, we show how to prove Theorem 2.5 by LEM. We need to introduce some notation first. 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 496 1 2 3 4 5 6 7 8 9 10 11 12 F:chipot506.tex; VTEX/Rita p. 10 J. Wei Theorem 2.5 actually holds for a slightly more general equation than (2.4), namely, ⎧ 2 ⎪ ⎨ u − u + f (u) = 0 u>0 ⎪ ⎩ ∂u = 0 ∂ν 2 3 in Ω, in Ω, on ∂Ω. (2.16) 14 6 17 18 19 20 21 22 23 24 25 26 w > 0 in RN , w → 0 at ∞, (2.17) 29 30 31 ∂w ∂w Kernel − 1 + f (w) = span . ,..., ∂y1 ∂yN 40 +2 One typical example of f is: f (u) = up − auq , where a 0, 1 < q < p < ( N N −2 )+ . For the uniqueness of w, see [39] and [40]. The proof of non-degeneracy is given in [58]. Without loss of generality, we may assume that 0 ∈ Ω. By the following rescaling: x = z, (2.19) 45 19 21 22 23 24 26 27 equation (2.16) becomes 28 29 u − u + f (u) = 0 in Ω , u > 0 in Ω , and ∂u ∂ν = 0 in ∂Ω . (2.20) 30 31 32 For u ∈ H 2 (Ω ), we put 33 34 S [u] = u − u + f (u). (2.21) 35 36 37 Then (2.20) is equivalent to 38 S [u] = 0, u ∈ H 2 (Ω ), u>0 in Ω , ∂u = 0 on ∂Ω . ∂ν 39 (2.22) 1 J˜ [u] = 2 Ω |∇u|2 + u2 − 42 43 F (u), Ω 40 41 Associated with problem (2.20) is the following energy functional 43 44 14 25 z ∈ Ω := {z | |z ∈ Ω}, 41 42 13 20 38 39 11 18 (2.18) 36 37 10 17 34 35 9 16 has a unique solution w(y) and w is non-degenerate, i.e., 32 33 8 15 27 28 7 12 w − w + f (w) = 0, w(0) = maxy∈RN w(y), 15 16 4 5 We will always assume that f : R → R is of class C 1+σ for some 0 < σ 1 and satisfies the following conditions (f1)–(f2): (f1) f (u) ≡ 0 for u 0, f (0) = f (0) = 0. (f2) The following equation 13 1 u ∈ H 1 (Ω ). (2.23) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 11 Existence and stability of spikes 1 2 3 4 We define two inner products: u, v = uv, for u, v ∈ L2 (Ω ); 7 8 9 12 13 14 15 2 (2.24) (u, v) = 5 (∇u∇v + uv), for u, v ∈ H (Ω ). 1 (2.25) 8 Let σ be the Hölder exponent of f and 6 + 2σ K σ be a fixed positive constant. Now we define a configuration space: Λ := (Q1 , . . . , QK ) ∈ Ω K ϕK (Q1 , . . . , QK ) M| ln | 9 10 (2.26) 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 13 14 (2.27) 43 44 45 15 16 where ϕK is defined at (2.14). Let w be the unique solution of (2.17). By the well-known result of Gidas, Ni and Nirenberg [22], w is radially symmetric: w(y) = w(|y|) and strictly decreasing: w (r) < 0 for r > 0, r = |y|. Moreover, we have the following asymptotic behavior of w: N−1 1 , w(r) = AN r − 2 e−r 1 + O r 1 − N−1 −r 1+O , (2.28) w (r) = −AN r 2 e r 17 for r large, where AN > 0 is a constant. Let K(r) be the fundamental solution of − + 1 centered at 0. Then we have 1 w(r) = A0 + O K(r), r 1 K(r), for r 1, (2.29) w (r) = −A0 + O r 27 K Qj u= +φ w z− 18 19 20 21 22 23 24 25 26 28 29 30 31 32 33 34 35 where A0 is a positive constant. The idea of LEM is to look for solutions of (2.16) of the following type: 36 37 38 (2.30) j =1 41 42 11 12 16 17 6 7 Ω M> 3 4 10 11 1 Ω 5 6 497 39 40 41 where φ is solved first by Lyapunov–Schmidt reduction process, and (Q1 , . . . , QK ) are adjusted so as to achieve a solution. LEM is a method of reducing the infinite-dimensional problem of finding a critical point of J˜ to a finite-dimensional problem of (Q1 , . . . , QK ). In general, it consists of the following five steps: 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 498 1 F:chipot506.tex; VTEX/Rita p. 12 J. Wei S TEP 1. Find out good approximate functions. 1 2 3 4 5 6 7 8 2 This step contains most of the important computations. The idea is to choose good approximate functions such that the error S is small. For Q ∈ Ω, we define w,Q to be the unique solution of ∂v = 0 on ∂Ω . ∂ν (2.31) w,Q = K 12 (2.32) w,Qj . 19 15 We first analyze w,Q . To this end, set 16 x |x − Q| − w,Q . ϕ,Q (x) = w ε 17 18 19 20 21 22 20 We state the following useful lemmas on the properties of ϕ,Q , whose proof can be found in [44]. 23 24 25 26 27 L EMMA 2.6. Assume that have M 2 | ln | |x ϕ,Q = − A0 + o(1) K 28 29 30 31 32 33 34 37 − Q∗ | Ω,j Qj 1 ϕK (Q) , = z − 2 Ω,K+1 = Ω 40 K +O 2M+N +1 45 23 24 27 28 29 30 31 32 33 34 35 j = 1, . . . , K, 36 37 38 Ω,j . (2.34) 39 40 j =1 41 L EMMA 2.7. For z ∈ Ω,j , j = 1, . . . , K, we have 43 44 22 26 (2.33) 41 42 21 25 √ The next lemma analyze w,Q in Ω . To this end, we divide Ω into K + 1-parts: 38 39 d(Q, ∂Ω) δ where δ is sufficiently small. We where K(r) is the (radially symmetric) fundamental solution of − + 1 in RN , Q∗ = Q + 2d(Q, ∂Ω)νQ̄ , νQ̄ denotes the unit outer normal at Q̄ ∈ ∂Ω and Q̄ is the unique point on ∂Ω such that d(Q̄, Q) = d(Q, ∂Ω). 35 36 13 14 15 18 8 10 j =1 14 17 7 11 12 16 5 9 Let Q = (Q1 , . . . , QK ) ∈ Λ. We then define the approximate solution as 11 13 4 6 Q = 0 in Ω , v − v + f w · − 9 10 3 w,Q = w,Qj M M Qj 2 + O K 2 . =w z− + O K 42 43 44 (2.35) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 13 Existence and stability of spikes 1 499 For z ∈ Ω,K+1 , we have 2 1 M w,Q = O K 2 . 3 4 2 (2.36) 4 5 6 7 5 w,Qk 9 10 12 21 22 23 14 w,Qk |Q1 −Q2 | 36 37 38 39 40 43 44 45 19 21 22 23 24 (2.37) 25 26 27 28 γ0 = RN f w(y) e−y1 dy. R EMARK . Note that γ0 > 0. See Lemma 4.7 of [61]. 29 (2.38) 30 31 32 33 P ROOF. By (2.28), we have for |y| |Q1 − Q2 |, N−1 2 Q1 −Q2 Q1 − Q2 = AN + o(1) e−|y+ | w y+ |y + Q1 − Q2 | Q1 −Q2 |Q1 − Q2 | −y, |Q +o(|y|) 1 −Q2 | e =w . 41 42 18 20 where 33 35 large, it holds Q2 Q1 |Q1 − Q2 | w z− = γ0 + o(1) w f w z− RN 31 34 16 29 32 15 17 L EMMA 2.8. For 26 30 12 Next we state a useful lemma about the interactions of two w’s. 25 28 11 M = O K 2 which proves (2.35). The proof of (2.36) is similar. 24 27 10 k=j 17 20 9 13 16 19 8 and so 15 18 7 Qk − ϕ,Qk (z) =w z− Q∗ Qk M k = O e−|z− | + e−|z− | + M+N +1 = O 2 11 14 6 P ROOF. For k = j and z ∈ Ω,j , we have 8 13 3 34 35 36 37 38 39 40 41 Thus by Lebesgue’s Dominated Convergence Theorem Q2 Q1 w z− f w z− RN 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 500 1 2 3 4 5 6 7 F:chipot506.tex; VTEX/Rita p. 14 J. Wei Q1 − Q2 f w(y) w y + RN −y, Q1 −Q2 |Q1 − Q2 | |Q1 −Q2 | = 1 + o(1) w f w(y) e dy RN |Q1 − Q2 | . = γ0 + o(1) w 1 = 2 3 4 5 6 8 9 8 Let us define several quantities for later use: 10 11 12 9 10 B (Qj ) = − Ω f (wj )ϕ,Qj , B (Qi , Qj ) = f (wi )wj . (2.39) Ω 13 14 17 18 19 20 21 22 25 26 27 15 L EMMA 2.9. For Q = (Q1 , . . . , QK ) ∈ Λ, it holds 16 2d(Qj , ∂Ω) + o w M| ln | , B (Qj ) = γ0 + o(1) w |Qi − Qj | + o w M| ln | . B (Qi , Qj ) = γ0 + o(1) w 32 33 34 35 36 37 38 39 40 43 44 45 18 19 20 21 (2.41) 22 24 25 26 27 29 Q∗j − Qj √ + O 2M+N +1 f (wj )w z − Ω |Qj − Q∗j | + o w(M| ln |) + o(1) w 2d(Qj , ∂Ω) + o w M| ln | . + o(1) w B (Qj ) = 1 + o(1) = γ = γ (2.40) follows from Lemma 2.6. To prove (2.41), we note that 41 42 (2.40) 28 and by Lemma 2.6 30 31 17 23 P ROOF. Note that |x − Q∗ | |x − Q∗ | = 1 + o(1) w A0 K 28 29 12 14 23 24 11 13 Then we have 15 16 7 B (Qi , Qj ) = RN − Qi − Qj Qi − Qj f (w)w y − f (w)w y − RN \Ω,Qi 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 15 Existence and stability of spikes 501 d(Qj ,∂Ω) σ d(Qi ,∂Ω) |Qi − Qj | + O e−(1+ 2 ) e− = γ + o(1) w |Qi − Qj | + o w M| ln | . = γ + o(1) w 1 2 3 4 5 1 2 3 6 7 8 11 J˜ 14 K w,Qj i=1 15 11 12 1 B (Qi ) = KI [w] − 2 K 17 18 13 14 i=1 1 − 2 16 15 B (Qi , Qj ) + o w M| ln | , (2.42) i,j =1,...,K, i=j 23 24 20 K w,Qj S j =1 21 CK q+1 q +σ 22 M(1+σ ) 2 (2.43) for any q > 26 N 2. 27 28 28 The proof of Lemma 2.10 is technical and tedious. We refer to [44] for the computations. 30 31 S TEP 2. Obtain a priori estimates for a linear problem. 32 33 34 35 36 33 This is the fundamental step in reducing an infinite-dimensional problem to finitedimensional one. The key result we need here is the non-degeneracy assumption (f2). Fix Q ∈ Λ. We define the following functions 37 40 41 Zi,j ∂wi = ( − 1) χi (z) , ∂zj 2|z − Qi | , where χi (z) = χ (M − 1)| ln | 44 45 35 36 i = 1, . . . , K, j = 1, . . . , N, 38 39 40 (2.44) 42 43 34 37 38 39 29 30 31 32 23 24 Lq (Ω ) 25 26 29 17 19 and 25 27 16 18 19 22 8 10 L EMMA 2.10. For any Q = (Q1 , . . . , QK ) ∈ Λ and sufficiently small we have 13 21 7 9 12 20 5 6 We then have the following which provides the key estimates on the energy expansion and error estimates. 9 10 4 41 42 where χ(t) is a smooth cut-off function such that χ(t) = 1 for |t| < 1 and χ(t) = 0 for 2 |t| > MM2 −1 . Note that the support of Zi,j belongs to B M 2 −1 ( Q i ). 2M | ln | 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 502 1 2 3 4 5 6 7 8 9 F:chipot506.tex; VTEX/Rita p. 16 J. Wei In this step, we consider the following linear problem: Given h ∈ L2 (Ω ), find a function φ satisfying ⎧ ⎪ ⎨ L [φ] := φ − φ + f (w,Q )φ = h + k,l ck,l Zk,l ; φ, Zi,j = 0, i = 1, . . . , K, j = 1, . . . , N, and (2.45) ⎪ ⎩ ∂φ = 0 on ∂Ω , ∂ν 12 13 14 15 16 17 18 19 20 φ∗ = φW 2,q (Ω ) , f ∗∗ = f Lq (Ω ) , (2.46) 27 28 29 30 31 32 33 34 35 36 37 38 39 42 43 44 45 6 7 11 where q > is a fixed number. We have the following result: 13 P ROPOSITION 2.11. Let φ satisfy (2.45). Then for sufficiently small and Q ∈ Λ, we have 16 14 15 φ∗ Ch∗∗ (2.47) 17 18 19 where C is a positive constant independent of , K and Q ∈ Λ. 20 21 22 P ROOF. Arguing by contradiction, assume that 23 φ∗ = 1; h∗∗ = o(1). (2.48) 24 25 We multiply (2.45) by ∂wi ∂zj χi (z) and integrate over Ω to obtain ∂wi ck,l Zk,l , χi (z) ∂zj k,l ∂wi ∂wi = − h, χi (z) + φ − φ + f (w,Q )φ, χi (z) . ∂zj ∂zj 26 27 28 29 30 31 (2.49) Observe that ∂wi ∂zj χi (z) 34 35 36 37 38 39 satisfies 40 ∂wi ∂wi ∂wi χi (z) − χi (z) + f (wi ) χi (z) ∂zj ∂zj ∂zj = 2∇z ∂wi ∂wi ∇z χi + (χi ) . ∂zj ∂zj 32 33 From the exponential decay of w one finds ∂wi χi (z) = o(1). h, ∂zj 40 41 5 12 N 2 25 26 4 10 23 24 3 9 21 22 2 8 for some constants ck,l , k = 1, . . . , K, l = 1, . . . , N . To this purpose, we define two norms 10 11 1 41 42 43 44 (2.50) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 17 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 Integrating by parts and using Lemma 2.7, we deduce 12 13 14 1 2 ∂wi χi (z) φ − φ + f (w,Q )φ, ∂zj M−1 ∂wi = f (w,Q ) − f (wi ) χi (z), φ + O 2 φ∗ ∂zj σ Mσ = O K 2 φ∗ = o φ∗ = o(1) 10 11 503 where we have used the fact that M > 6+2σ σ N f (w,Q ) − f (wi ) ∂wi χi ∂z j ∗∗ 3 4 5 6 7 8 9 10 and that 11 Mσ σ ∂wi C |w,Q − wi | χi K σ 2 . ∂z j ∗ 15 16 17 19 17 (2.51) 20 24 26 27 21 ∂wi Zk,l , χi (z) = 0 ∂zj 22 (2.52) 29 25 26 ∂wi Zi,l , χi (z) = O M . ∂zj 27 (2.53) 33 30 The left hand side of (2.49) becomes ci,j + 31 O M ci,l = o(1) 32 33 l=j 34 34 35 36 37 38 35 and hence ci,j = o(1), 36 37 i = 1, . . . , K, j = 1, . . . , N. (2.54) 39 40 φi = φχi , where χi 43 44 45 38 39 To obtain a contradiction, we define the following cut-off functions: 41 42 28 29 30 32 23 24 and for k = i and l = j , we have 28 31 18 19 On the other hand, for k = i we have 23 25 14 16 ∂wi ∂w 2 Zi,j , χi (z) = − f (w) dy + o(1). ∂zj ∂yj RN 20 22 13 15 It is easy to see that 18 21 12 Note that χi = 1 for z ∈ B M 2 −1 2M 40 2|z − Qi | , i = 1, . . . , K. =χ (M − M −1 )| ln | | ln | 41 (2.55) 42 43 ( Q i ) and the support of φ belongs to B M | ln | ( Q i ). 2 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 504 1 F:chipot506.tex; VTEX/Rita p. 18 J. Wei Then the conditions φ, Zi,j = 0 is equivalent to 1 2 3 2 φi , Zi,j = 0. (2.56) 4 5 4 5 The equation for φi becomes 6 7 8 φi − φi + f (w,Q )φi = 11 12 13 6 ci,j Zi,j + hχi + 2∇φ∇χi + χi φ. (2.57) j 9 10 16 11 f (w,Q )φi = f (wi ) + o M/2−N φi . 12 (2.58) q φi W 2,q (Ω ) q C hχ q i L (Ω ) q + C 2∇φ∇χ + χ φ q i i L (Ω (2.59) 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 20 φ= 22 φi + Φ (2.60) where Φ = φ(1 − 44 45 23 24 i=1 25 K i=1 χi ). Then the equation for Φ becomes 26 27 Φ − Φ + f (w,Q )Φ K K K =h 1− χi − 2 ∇φ∇χi − χi φ. i=1 i=1 28 29 30 (2.61) i=1 33 By Lemma 2.7, f (w,Q )Φ = o(1)Φ. Standard regularity theorem gives K q ΦW 2,q (Ω ) C h 1 − χi i=1 34 q q 35 36 37 L (Ω ) K q K + C 2 ∇φ∇χi + (χi )φ q i=1 31 32 i=1 L Ω 38 39 . (2.62) 42 43 18 21 K 25 26 16 19 Next, we decompose 22 24 15 17 . ) 21 23 13 14 Using (2.56) and (2.58), a contradiction argument similar to that of Proposition 3.2 of [27] gives 19 20 8 10 Lemma 2.7 yields 17 18 7 9 14 15 3 40 41 42 Lp -regularity is independent of < 1. The case of (Observe that the constant C in the Dirichlet boundary condition has been proved in Lemma 6.4 of [61]. The case of Neumann boundary condition can be proved similarly.) 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 19 Existence and stability of spikes 1 Combining (2.60), (2.59) and (2.62), we obtain 2 4 i=1 5 6 C 7 8 K 5 6 q φi W 2,q (Ω ) q + CΦW 2,q (Ω ) i=1 12 7 8 i=1 q q 9 10 11 L (Ω ) 12 K 2∇φ∇χ + (χ )φ q q +C 13 14 i i=1 15 16 17 22 4 ) K K q hχ q h 1 − C + χi i L (Ω ) 11 21 W 2,q (Ω ) i=1 10 20 3 q + CΦW 2,q (Ω 9 19 1 2 K q q φW 2,q (Ω ) C φi 3 18 505 q ChLq (Ω ) i 13 14 L (Ω ) q + O | ln |−1 φW 2,q (Ω 15 16 ) 17 18 since K K q χi + 1 − χi i=1 19 q 2, −1 |∇χ | + |χ | C | ln | . 20 (2.63) i=1 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 23 This gives 24 φW 2,q (Ω ) = o(1). L [φ] = h + (2.65) ci,j Zi,j , i,j φ, Zi,j = 0, i = 1, . . . , K, j = 1, . . . , N, ∂φ = 0 on ∂Ω . ∂ν 28 31 32 33 34 35 36 38 (2.66) 39 40 41 Moreover, we have 42 42 φ∗ Ch∗∗ 44 45 27 37 40 43 26 30 P ROPOSITION 2.12. There exists 0 > 0 such that for any 0 < < 0 the following property holds true. Given h ∈ W 2,q (Ω ), there exist s a unique pair (φ, c) = (φ, {ci,j }i=1,...,K,j =1,...,N ) such that 25 29 From Proposition 2.11, we derive the following existence result: 38 41 (2.64) A contradiction to (2.48). 37 39 21 (2.67) 43 44 for some positive constant C. 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 506 1 2 3 4 5 6 7 J. Wei P ROOF. The bound in (2.67) follows from Proposition 2.11 and (2.54). Let us now prove the existence part. Set 1 where we define the inner product on 12 13 14 15 16 17 (u, v) = 5 6 as 7 8 (∇u∇v + uv). 9 Ω 10 11 Note that, integrating by parts, one has ψ ∈H if and only if 12 ψ, Zi,j = 0, 13 i = 1, . . . , K, j = 1, . . . , N. 14 15 Observe that φ solves (2.65) and (2.66) if and only if φ ∈ H satisfies 18 19 ) Ω 16 17 ! " (∇φ∇ψ + φψ) − f (w,Q )φ, ψ = h, ψ , ∀ψ ∈ H. 18 19 20 21 20 This equation can be rewritten as 21 22 23 22 φ + S(φ) = h̄ in H, (2.68) 24 25 26 27 28 35 36 37 38 39 40 41 42 43 44 45 26 27 28 30 31 φ = A (h). (2.69) 33 34 25 29 In the following, if φ is the unique solution given in Proposition 2.12, we set 31 32 23 24 where h̄ is defined by duality and S : H → H is a linear compact operator. Using Fredholm’s alternative, showing that equation (2.68) has a unique solution for each h̄, is equivalent to showing that the equation has a unique solution for h̄ = 0, which in turn follows from Proposition 2.11 and our proof is complete. 29 30 2 4 10 11 H 1 (Ω 1 3 H = u ∈ H (Ω ) u, ( − 1)−1 Zi,j = 0 8 9 F:chipot506.tex; VTEX/Rita p. 20 32 33 34 Note that (2.67) implies 35 A (h) Ch∗∗ . ∗ (2.70) 37 38 S TEP 3. A non-linear Lyapunov–Schmidt reduction. 39 For small and for Q ∈ Λ, we are going to find a function φ,Q such that for some constants ci,j , j = 1, . . . , N , the following equation holds true (w,Q + φ) − (w,Q + φ) + f (w,Q + φ) = φ, Zi,j = 0, j = 1, . . . , N, 36 ∂φ ∂ν k,l ck,l Zk,l = 0 on ∂Ω . 40 41 42 43 in Ω , (2.71) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 21 Existence and stability of spikes 1 507 The first equation in (2.71) can be written as 2 1 2 φ − φ + f (w,Q )φ = −S [w,Q ] + N [φ] + ci,j Zi,j , 3 4 3 4 i,j 5 6 7 8 5 where 6 N [φ] = − f (w,Q + φ) − f (w,Q ) − f (w,Q )φ . 7 (2.72) 9 10 11 12 13 14 9 L EMMA 2.13. For Q ∈ Λ and sufficiently small, we have for φ∗ + φ1 ∗ + φ2 ∗ 1, N [φ] 17 18 19 20 21 22 23 N [φ1 ] − N [φ2 ] (2.73) C φ1 σ∗ + φ2 σ∗ φ1 − φ2 ∗ . ∗∗ 26 27 28 29 20 21 f (w,Q + φ) − f (w,Q ) − f (w,Q )φ| C φ|1+σ , 22 23 P ROPOSITION 2.14. For Q ∈ Λ and sufficiently small, there exists a unique φ = φ,Q such that (2.71) holds. Moreover, Q → φ,Q is of class C 1 as a map into W 2,q (Ω ) ∩ H, and we have φ,Q ∗ rK q+1 q +σ 34 35 36 37 40 41 (2.75) 44 45 25 26 27 28 29 31 32 for some constant r > 0. 33 34 P ROOF. Let A be as defined in (2.69). Then (2.71) can be written as φ = A −S [w,Q ] + N [φ] . 35 36 (2.76) 37 38 Let r be a positive (large) number, and set 39 q+1 ) +σ M(1+σ 2 . Fr = φ ∈ H ∩ W 2,q (Ω ): φ∗ < rK q 42 43 24 30 M(1+σ ) 2 38 39 17 19 Since f is Hölder continuous with exponent σ , we deduce which implies (2.73). The proof of (2.74) goes along the same way. 16 18 f (w,Q + φ) − f (w,Q ) = f (w,Q + θ φ)φ. 32 33 14 15 30 31 12 13 (2.74) P ROOF. Inequality (2.73) follows from the mean-value theorem. In fact, for all z ∈ Ω there holds 24 25 10 11 Cφ∗1+σ ; ∗∗ 15 16 8 40 41 42 Define now the map G : Fr → H ∩ W 2,q (Ω ) G [φ] = A −S [w,Q ] + N [φ] . as 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 508 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 F:chipot506.tex; VTEX/Rita p. 22 J. Wei Solving (2.71) is equivalent to finding a fixed point for G . By Lemmas 2.10 and 2.13, for sufficiently small and r large we have 4 5 G [φ1 ] − B [φ2 ] C N [φ1 ] − N [φ2 ] < 1 φ1 − φ2 ∗ , ∗ ∗ 2 6 7 which shows that G is a contraction mapping on Fr . Hence there exists a unique φ = φ,Q ∈ Fr such that (2.71) holds. Now we come to the differentiability of φ,Q . Consider the following map H : Λ × H ∩ W 2,q (Ω ) × R N K → H ∩ W 2,q (Ω ) × R N K of class C 1 ⎜ ⎜ ⎜ H (Q, φ, c) = ⎜ ⎜ ⎜ ⎝ ( − 1)−1 (S [w,Q + φ]) − −1 i,j ci,j ( − 1) Zi,j (φ, ( − 1)−1 Z1,1 ) .. . 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 43 44 45 9 10 11 12 14 ⎟ ⎟ ⎟ ⎟. ⎟ ⎟ ⎠ 15 16 17 18 (φ, ( − 1)−1 ZK,N ) 19 (2.77) 20 21 Equation (2.71) is equivalent to H (Q, φ, c) = 0. We know that, given Q ∈ Λ, there is a unique local solution φ,Q , c,Q obtained with the above procedure. We prove that the linear operator ∂H (Q, φ, c) : H ∩ W 2,q (Ω ) × R N K → H ∩ W 2,q (Ω ) × R N K ∂(φ, c) (Q,φ,Q ,c,Q ) 22 is invertible for small. Then the of Q → (φ,Q , c,Q ) follows from the Implicit Function Theorem. Indeed we have ∂H (Q, φ, c) [ψ, d] ∂(φ, c) (Q,φ,Q ,c,Q ) ⎛ ⎞ ( − 1)−1 (S [w,Q + φ,Q ](ψ)) − i,j dij ( − 1)−1 Zi,j ⎜ ⎟ ⎜ ⎟ (ψ, ( − 1)−1 Z1,1 ) ⎜ ⎟ ⎟. =⎜ ⎜ ⎟ .. ⎜ ⎟ . ⎝ ⎠ −1 (ψ, ( − 1) ZK,N ) 29 C 1 -regularity 41 42 8 13 ⎞ 21 22 2 3 q+1 M(1+σ ) G [φ] C S [w,Q ] + C N [φ] < rK q +σ 2 , ∗ ∗∗ ∗∗ ⎛ 1 23 24 25 26 27 28 30 31 32 33 34 35 36 37 38 39 40 41 Since φ,Q ∗ is small, the same proof as in that of Proposition 2.11 shows that ∂H (Q, φ, c) ∂(φ, c) (Q,φ,Q ,c,Q ) 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 23 Existence and stability of spikes 1 2 509 is invertible for small. This concludes the proof of Proposition 2.14. 1 3 4 5 6 3 In some cases (e.g., critical or nearly critical exponent problems), we need to obtain further differentiability of φ,Q (e.g., C 2 in Q). This will be achieved by further reduction. See [13,65] and [66] for such arguments. 7 8 11 14 15 16 17 18 19 20 21 22 Fix Q ∈ Λ. Let φ,Q be the solution given by Proposition 2.14. We define a new functional 27 28 29 30 31 32 M (Q) = J˜ [w,Q + φ,Q ] : Λ → R. (2.78) 35 P ROOF. By Proposition 2.14, there exists 0 such that for 0 < < 0 we have a C 1 map which, to any Q ∈ Λ, associates φ,Q such that S [w,Q + φ,Q ] = (2.79) 20 21 22 26 27 29 30 31 i = 1, . . . , K, j = 1, . . . , N. 32 33 Hence we have Ω 34 ∂(w,Q + φ,Q ) ∇u ∇ ∂Q i,j ∂(w,Q + φ,Q ) − f (u ) ∂Q i,j ∂(w,Q + φ,Q ) +u ∂Q Qi =Qi Qi =Qi i,j 35 36 Qi =Qi 37 38 = 0, 39 40 41 45 19 28 for some constants ckl Let Q ∈ Λ be a critical point of M . Set u = w,Q + φ,Q . Then we have 40 44 18 25 ∈ R KN . 39 43 17 24 φ,Q , Zi,j = 0 38 42 14 23 ckl Zk,l , k=1,...,K; l=1,...,N DQi,j |Qi =Qi M (Q ) = 0, 13 16 L EMMA 2.15. If Q is critical point of M (Q) in Λ, then u = w,Q + φ,Q is a critical point of J˜ [u]. 36 37 11 15 Then we have the following reduction lemma 33 34 10 12 25 26 6 9 23 24 5 8 S TEP 4. A reduction lemma. 12 13 4 7 9 10 2 41 which gives 42 k=1,...,K; l=1,...,N ckl ∂(w,Q + φ,Q ) Zk,l = 0. ∂Qi,j Ω Qi =Q i 43 (2.80) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 510 1 2 3 4 5 F:chipot506.tex; VTEX/Rita p. 24 J. Wei We claim that (2.80) is a diagonally dominant system. In fact, since φ,Q , Zi,j = 0, we have that ∂φ,Q ∂Zk,l Zk,l = − φ,Q = 0 if k = i. ∂Qi,j ∂Qi,j Ω Ω 6 If k = i, we have 8 10 11 ∂φ,Q Zk,l =− ∂Qk,j Ω Ω ∂Zk,l ∂Zk,l φ,Q ∗∗ φ,Q = ∂Qk,j ∂Qk,j ∗∗ M(1+σ ) −( q+1 +σ )N −1 q+1 +σ M(1+σ ) −1 q =O 2 =O K q 2 M =O 2 . 13 14 15 17 18 23 24 25 Zk,l ∂w,Qi ∂Qi,j Ω = Q Ω ∩B M | ln | ( k 2 Zk,l Ω ∂w,Qk Zk,l ) ∂Qk,j 34 35 36 37 38 39 ∂w,Qi ∂Qi,j = O M . 42 = Ω ∩B M | ln | ( 2 = − −1 δlj Qk Zk,l ) 45 13 14 18 19 20 24 ∂Qk,j 25 ∂w 2 f (w) + O(1). ∂yj RN For each (k, l), the off-diagonal term gives M M M O 2 + + O() = O 2 + K M + = o(1) k=i k=i,l=j by our choice of M > 6+2σ σ N. Thus equation (2.80) becomes a system of homogeneous equations for ckl and the matrix of the system is non-singular. So ckl ≡ 0, k = 1, . . . , K, l = 1, . . . , N . Hence u = K i=1 w,Qi + φ,Q1 ,...,QK is a solution of (2.20). 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 S TEP 5. Using variational arguments to find critical points for the finite-dimensional reduced problem. 43 44 12 ∂w,Qk 40 41 11 23 29 33 10 22 28 32 9 21 27 31 8 17 For k = i, we have 26 30 5 16 21 22 4 15 For k = i, we have 19 20 3 7 12 16 2 6 7 9 1 41 42 43 By Lemma 2.15, we just need to find a critical point for the reduced energy functional M (Q). Depending on the asymptotic behavior of the reduced energy functional, 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 25 Existence and stability of spikes 1 2 3 4 5 6 7 one can use either local minimization, or local maximization [29], or saddle point techniques [66]. Here there is no compactness problem since the reduced problem is already finite-dimensional. We first obtain an asymptotic formula for M (Q). In fact for any Q ∈ Λ, we have ˜ M (Q) = J [w,Q ] + (∇w,Q ∇φ,Q + w,Q φ,Q ) 8 − 9 Ω 10 12 16 K 22 23 24 15 16 17 18 19 20 21 22 23 (2.81) 26 27 28 P ROPOSITION 2.16. For small, the following maximization problem 34 35 36 37 38 39 29 max{M (Q): Q ∈ Λ} 31 33 has a solution Q ∈ Λ◦ —the (2.82) 42 31 33 P ROOF. First, we obtain a lower bound for M : Recall that KΩ (r) is the maximum number of non-overlapping balls with equal radius r packed in Ω. Now we choose K such that M + 2N 1 K KΩ | ln | . (2.83) 2 34 35 36 37 38 39 40 = (Q01 , . . . , Q0K ) be the centers of arbitrary balls. Certainly Q0 ∈ Λ. Then we have Q0 Let | ln |) 43 44 45 30 32 interior of Λ. 40 41 24 25 We shall prove 30 32 7 14 28 29 6 13 26 27 5 12 i=j 25 4 11 1 2d(Qi , ∂Ω) γ0 + o(1) w 2 i=1 |Qi − Qj | 1 + o w M| ln | . w − γ0 + o(1) 2 M (Q) = KI [w] − 21 3 10 −S [w,Q ] φ,Q + O φ,Q 2∗ by Lemma 2.10, Proposition 2.14 and the choice of M at (2.26). By Lemma 2.10, we obtain 20 2 9 2+ 2 +2σ M(1+σ ) = J˜ [w,Q ] + O K q = J˜ [w,Q ] + o w M| ln | 15 1 8 = J˜ [w,Q ] + O S [w,Q ]∗∗ φ,Q ∗ + O φ,Q 2∗ 14 19 f (w,Q )φ,Q + O φ,Q 2∗ Ω 13 18 Ω = J˜ [w,Q ] + 11 17 511 w 2d(Q0i , ∂Ω) e 2d(Q0i ,∂Ω) − K balls among those × 41 42 M+2N , KΩ ( M+2N 2 w |Q0i − Q0j | 43 M+2N 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 512 1 F:chipot506.tex; VTEX/Rita p. 26 J. Wei and hence 2 1 K γ0 + o(1) M+2N M Q M Q0 KI [w] − 2 2 K γ0 + o(1) M+2N + o w(M| ln |) − 2 KI [w] − K 2 γ0 + o(1) M+2N + o w M| ln | . 3 4 5 6 7 8 2 3 4 5 6 7 (2.84) 9 10 11 12 9 On the other hand, if Q ∈ ∂Λ, then either there exists (i, j ) such that |Qi − Qj | = M| ln |, or there exists a k such that d(Qk , ∂Ω) = M 2 | ln |. In both cases we have 1 M Q KI [w] − γ0 + o(1) w M| ln | + o w M| ln | . 2 13 14 15 16 (2.86) 33 34 37 38 39 40 Let p = 30 By suitable scaling, (2.4) becomes the following problem 31 32 N+2 N−2 = 0 in Ω, u − μu + u u > 0 in Ω and ∂u ∂ν = 0 on ∂Ω (2.87) 45 33 34 35 where μ = is large. It is well known that the solutions to 36 1 2 37 38 N+2 U + U N−2 = 0 (2.88) 39 40 41 are given by the following 43 44 25 29 N +2 N −2 . 41 42 24 28 35 36 22 27 2.4. Bubbles to (2.4): the critical case 29 32 21 26 27 31 19 23 C OMPLETION OF P ROOF OF T HEOREM 2.5. Theorem 2.5 follows from Proposition 2.16 and the reduction Lemma 2.15. 26 30 18 20 which is impossible. We conclude that Q ∈ Λ. This completes the proof of Proposition 2.16. 23 28 14 17 20 25 12 16 −2N w M| ln | 2K 2 M+2N C M | ln | 19 24 11 15 Combining (2.85) and (2.84), we obtain 18 22 10 13 (2.85) 17 21 8 UΛ,ξ = cN 1 Λ2 + |x − ξ |2 42 43 N−2 2 , where Λ > 0, ξ ∈ RN . 44 (2.89) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 27 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 513 4 N−2 +2 A notable difference here is that the linearized operator + ( N N −2 )UΛ,ξ has (N + 1)dimensional kernels. Namely, 4 ∂Uλ,ξ ∂UΛ,ξ ∂Uλ,ξ N + 2 N−2 . UΛ,ξ = span , ,..., Kernel + N −2 ∂Λ ∂ξ1 ∂ξN 2 3 4 (2.90) 5 6 Thus when we apply LEM, we need also to take care of the scaling parameters. See [13,43,65,66] and the references therein. Concerning boundary bubbles, the existence of mountain-pass solutions was first proved in Wang [69] and Adimurthi and Mancini [1]. Ni, Takagi and Pan [55] showed the least energy solutions develop a bubble at the maximum point of the mean curvature (thereby establishing results similar to Theorem 2.1). Local mountain-pass solutions concentrating on one or separated boundary points are established in [23]. At non-degenerate critical points of the positive mean curvature, single boundary bubbles exist [2]. Lin, Wang and Wei [43] established results similar to Theorem 2.2 for dimension N 7, at a non-degenerate local minimum point of the mean curvature with positive value: 17 18 1 7 8 9 10 11 12 13 14 15 16 17 18 T HEOREM 2.17. Suppose the following two assumptions hold: 19 19 20 (H1) N 7, 20 21 (H2) Q0 = 0 is a non-degenerate minimum point of H (Q) and H (Q0 ) > 0. 21 22 23 24 25 26 Let K 2 be a fixed integer. Then there exists a μK > 0 such that for μ > μK , problem (2.87) has a non-trivial solution uμ with the following properties (1) 27 28 29 30 31 32 33 34 35 36 37 u(x) = K j =1 U1 μ Λj ,Q0 +μ R[Q̂1 , . . . , Q̂K ] := c1 K j =1 where ϕ(Q) = 45 26 29 ϕ(Q̂j ) + c2 1 i=j |Q̂i − Q̂j |N −2 k,l ∂k ∂l H (Q0 )Qk Ql , c1 31 32 33 34 35 36 37 38 (2.91) 39 40 41 and c2 are two generic constants. 43 44 25 30 40 42 24 28 Q̂j (∗) Find out the optimal configuration (Q̂1 , . . . , Q̂K ) that minimizes the functional R[Q̂1 , . . . , Q̂K ]. Here for Q̂ = (Q̂1 , . . . , Q̂K ) ∈ R (N −1)K , Q̂i = Q̂j , we define 41 23 27 N−4 2 , μ +O μ where Λj → Λ0 := A0 H (Q0 ) > 0, j = 1, . . . , K, and μ μ (2) Q̂μ := (Q̂1 , . . . , Q̂K ) approach an optimal configuration in the following problem: 38 39 3−N N 22 42 43 Theorem 2.17 is proved by LEM. Here the computation is more complicated, since the interaction between bubbles is very involved. 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 514 1 2 3 4 5 F:chipot506.tex; VTEX/Rita p. 28 J. Wei Concerning interior bubbles, under some assumptions, it is proved in [24] and [64] that there are no interior bubble solutions. However interior bubble solutions can be recovered if one add the boundary layers. (The boundary layer solution has been constructed in [50] (see Section 2.6).) The following result establishes the existence of multiple interior bubbles in dimension N = 3, 4, 5. 6 7 8 9 14 15 16 17 20 21 22 23 24 25 26 27 28 29 30 31 32 33 36 37 38 39 40 41 In the slightly supercritical case, we let p = N +2 N −2 u − μu + up = 0 in Ω, u > 0 in Ω and ∂u ∂ν = 0 on ∂Ω. 8 9 14 15 (2.92) 16 17 18 The following result was proved by [66] and [14] through the use of LEM. 19 T HEOREM 2.19. Let N 3. Then δ > 0 sufficiently small, problem (2.92) admits a boundary bubble solution. In fact, in the slightly supercritical case, there is also the phenomena of bubble-towers. A bubble-tower is a sum of bubbles centered at the same point K 20 21 22 23 24 25 26 UΛj ,ξ , j =1 Λj +1 where Λ1 , → +∞, j = 1, . . . , K − 1. Λj (2.93) 27 28 29 This has been discussed in [15] and [25]. It is completely open whether or not point condensation solutions exist for (2.92) when +2 p> N N −2 + δ. In fact, let Ω be the unit ball. Using Pohozaev’s identity, it is not difficult to show that there exists a positive constant c0 , independent of 1, such that 30 31 32 33 34 inf u c0 Ω (2.94) for all radial solution u of (2.4). This marks a basic difference between the behavior of +2 N +2 solutions of these two cases p N N −2 and p > N −2 . It eliminates the possibility of the existence of a radial spiky solution which approaches zero in measure as approaches +2 zero in the supercritical case p > N N −2 . 35 36 37 38 39 40 41 42 2.6. Concentration on higher-dimensional sets 44 45 7 13 + δ where δ > 0. Consider 42 43 5 12 34 35 4 11 2.5. Bubbles to (2.4): slightly supercritical case 18 19 3 10 11 13 2 6 T HEOREM 2.18. (See [71,92].) Let N = 3, 4, 5. For any fixed integer k, then problem (2.87) has a solution (at least along a subsequence k → 0) with k interior bubbles and one boundary layer. 10 12 1 43 44 The following conjecture has been made by Ni [53,54]. 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 29 Existence and stability of spikes 515 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 Fig. 1. Lines intersecting with ∂Ω orthogonally. 12 12 13 14 15 16 13 C ONJECTURE . Given any integer 0 k n − 1, there exists pk ∈ (1, ∞) such that for 1 < p < pk , (2.4) possesses a solution with k-dimensional concentration set, provided that is sufficiently small. 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 15 16 17 Progress in this direction has only been made very recently. In [49] and [50], Malchiodi and Montenegro proved that for N 2, there exists a sequence of numbers εk → 0 such that problem (2.4) has a solution uεk which concentrates at boundary of ∂Ω (or any component of ∂Ω). Such a solution has the following energy bound (2.95) In [48], Malchiodi showed the concentration phenomena for (2.4) along a closed nondegenerate geodesic of ∂Ω in three-dimensional smooth bounded domain Ω. F. Mahmoudi and A. Malchiodi in [51] prove a full general concentration of solutions along k-dimensional (1 k n − 1) non-degenerate minimal sub-manifolds of the boundary for n 3 and 1 < p < n−k+2 n−k−2 . When Ω = B1 (0), there are also multiple (radially symmetric) clustered interfaces near the boundary [52]. For concentrations on lines intersecting with the boundary, Wei and Yang [93] made the first attempt in the two-dimensional case. Let Γ ⊂ Ω ⊂ R2 be a curve satisfying the following assumptions: The curvature of Γ is zero and Γ intersects ∂Ω at exactly two points, saying, γ1 , γ0 and at these points Γ ⊥∂Ω. Let −k1 and k0 are the curvatures of the boundary ∂Ω at the points γ1 and γ0 respectively. A picture of Γ and Ω is as follows: We define a geometric eigenvalue problem −f (θ ) = λf (θ ), 18 19 20 21 22 Jεk [uεk ] ∼ εkN −1 . f (0) + k0 f (0) = 0. 24 25 26 27 28 29 30 31 32 33 34 35 36 38 f (1) + k1 f (1) = 0, 23 37 0 < θ < 1, 39 (2.96) 40 41 We say that Γ is non-degenerate if (2.96) does not have a zero eigenvalue. This is equivalent to the following condition: 44 45 14 42 43 44 k0 − k1 + k0 k1 |Γ | = 0, (2.97) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 516 1 2 3 4 5 F:chipot506.tex; VTEX/Rita p. 30 J. Wei where |Γ | denotes the length of Γ . Moreover, we set up the gap condition that there exists a small constant c > 0 2 2 λ0 − k π ε 2 cε, |Γ |2 1 2 3 ∀k ∈ N. (2.98) 6 7 10 11 12 13 14 15 16 7 8 T HEOREM 2.20. We assume that the line segment Γ satisfies the non-degenerate condition (2.97). Given a small constant c, there exists ε0 such that for all ε < ε0 satisfying the gap condition (2.98), problem (2.4) has a positive solution uε concentrating along a curve Γ near Γ . Moreover, there exists some number c0 such that uε satisfies globally, uε (x) exp −c0 ε −1 dist(x, Γ ) 19 20 21 22 23 24 25 26 27 28 29 30 31 34 10 11 12 13 14 16 17 R EMARK 2.6.1. The geometric eigenvalue problem (2.96) was first introduced by M. Kowalczyk in [37] where he constructed layered solution concentrating on a line for the Allen–Cahn equation. R EMARK 2.6.2. Theorem 2.20 is proved using the infinite-dimensional Lyapunov– Schmidt reduction technique introduced in [18]. R EMARK 2.6.3. One can also constructed multiple clustered line concentrating solutions, using the Toda system. See [94]. This follows from earlier work in [19], where multiple clustered interfaces are constructed at non-minimizing lines for the Allen–Cahn equation. It is quite interesting to see the connection between Toda system qj +e qj −qj +1 −e qj −1 −qj =0 (2.99) 18 19 20 21 22 23 24 25 26 27 28 29 30 31 and clustered interfaces. 32 33 9 15 and the curve Γ will collapse to Γ as ε → 0. 17 18 5 6 In [93], the following result was proved 8 9 4 32 R EMARK 2.6.4. It will be interesting to construct solutions concentrating on surfaces which intersect with ∂Ω orthogonally. 33 34 35 35 36 36 37 37 2.7. Robin boundary condition 38 39 40 41 42 43 44 45 38 Robin boundary conditions are particularly interesting in biological models where they often arise. We refer the reader to [10] for this aspect. In [3], Berestycki and Wei discussed the existence and asymptotic behavior of least energy solution for following singularly perturbed problem with Robin boundary condition: 39 40 41 42 43 2 u − u + up ∂u ∂ν = 0, u > 0 in Ω, + λu = 0 on ∂Ω, (2.100) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 31 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 10 517 where λ > 0. Similar to [57], we can define the following energy functional associated with (2.100): 1 2 1 λ J [u] := |∇u|2 + u2 − F (u) + u2 , 2 Ω 2 Ω 2 Ω ∂Ω u where F (u) = 0 f (s) ds, f (s) = s p , u ∈ H 1 (Ω). Similarly, for ∈ (0, 1), we can define the so-called mountain-pass value c,λ = inf max J h(t) 3 h∈Γ 0t1 (2.101) 13 14 15 16 17 18 19 8 9 (2.102) where Γ = {h: is continuous, h(0) = 0, h(1) = e}. For fixed small, as λ moves from 0 (which is Neumann BC) to +∞ (which is Dirichlet BC), by the results of [57,58] and [61], the asymptotic behavior of u,λ changes dramatically: a boundary spike is displaced to become an interior spike. The question we shall answer is: where is the borderline of λ for spikes to move inwards? Note that when N = 1, by ODE analysis, it is easy to see that the borderline is exactly at λ = 1. In fact, we may assume that Ω = (0, 1), and a s → 0, the least energy solution converges to a homoclinic solution of the following ODE: w(y) → 0 as |y| → +∞. (2.103) 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 13 14 15 16 17 18 19 21 22 Then it follows that 23 (w )2 = w 2 − 26 27 12 20 w − w + w p = 0 in R1 , 24 25 10 11 [0, 1] → H 1 (Ω) | h(t) 22 23 5 7 20 21 4 6 11 12 2 2 w p+1 , p+1 24 |w | < w. (2.104) 26 w (0) − λw(0) = 0. We see As → 0, the limiting boundary condition (2.100) becomes from (2.104) that this is possible if and only if λ < 1. When N = 2, the situation changes dramatically. To understand the location of the spikes at the boundary, an essential role is played by the analogous problem in a half space with Robin boundary condition on the boundary. Thus we first consider u − u + f (u) = 0, u > 0 in RN +, (2.105) ∂u N 1 u ∈ H (R+ ), ∂ν + λu = 0 on ∂RN + N where RN + = {(y , yN ) | yN > 0} and ν is the outer normal on ∂R+ . Let 1 1 2 λ 2 Iλ [u] = F (u) + u2 . |∇u| + u − + + 2 2 2 R+ R ∂R N N N inf v≡0, v∈H 1 (R+ N) sup Iλ [tv]. t>0 27 28 29 30 31 32 33 34 35 36 37 38 39 (2.106) 40 41 42 As before, we define a mountain-pass vale for Iλ : cλ = 25 43 (2.107) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 518 1 F:chipot506.tex; VTEX/Rita p. 32 J. Wei Our first result deals with the half space problem: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 2 T HEOREM 2.21. (1) For λ 1, cλ is achieved by some function wλ , which is a solution of (2.105). (2) For λ large enough, cλ is never achieved. (3) Set λ∗ = inf{λ | cλ is achieved}. 26 27 28 31 32 The proof of Theorem 2.21 is by the method of concentration-compactness, and the method of vanishing viscosity. Now consider the problem in a bounded domain. T HEOREM 2.22. Let λ λ∗ and u,λ be a least energy solution of (2.100). Let x ∈ Ω be a point where u,λ reaches its maximum value. Then after passing to a subsequence, x → x0 ∈ ∂Ω and (1) d(x , ∂Ω)/ → d0 , for some d0 > 0, (2) v,λ (y) = u,λ (x + y) → wλ (y) in C 1 locally, where wλ attains cλ of (2.107) (and thus is a solution of (2.105)), (3) the associated critical value can be estimated as follows: c,λ = N cλ − H̄ (x0 ) + o() (2.109) 37 38 39 40 41 42 43 44 45 8 11 12 13 14 15 16 17 18 19 20 21 22 23 24 where cλ is given by (2.107), and H̄ (x0 ) is given by the following ∂wλ y · ∇ wλ H (x0 ) H̄ (x0 ) = max − ∂yN wλ ∈Sλ R+ N 25 26 (2.110) 27 28 29 where Sλ is the set of all solutions of (2.105) attaining cλ , and y = (y1 , . . . , yN −1 ), ∂ ∇ = ( ∂y∂ 1 , . . . , ∂yN−1 ), (4) H̄ (x0 ) = maxx∈∂Ω H̄ (x). 30 31 32 33 On the other hand, when λ > λ∗ , a different asymptotic behavior appears. 35 36 6 10 33 34 5 9 Then λ∗ > 1 and for λ λ∗ , cλ is achieved, and for λ > λ∗ , cλ is not achieved. 29 30 4 7 (2.108) 24 25 3 34 35 T HEOREM 2.23. Let λ > λ∗ and u,λ be a least energy solution of (2.100). Let x ∈ Ω be a point where u,λ reaches its maximum value. Then after passing a subsequence, we have (1) d(x , ∂Ω) → maxx∈Ω d(x, ∂Ω), (2) v,λ (y) := u,λ (x + y) → w(y) in C 1 locally, where w is the unique solution of (2.8), (3) the associated critical value can be estimated as follows: 2d(x , ∂Ω) 1 + o(1) . (2.111) c,λ = N I [w] + exp − 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 33 Existence and stability of spikes 1 519 3. Stability and instability in the shadow system case 1 2 3 4 5 6 7 2 As we have already seen in Section 2 that there are many single and multiple spike solutions for the shadow system (2.2). The question is: are they all stable with respect to the shadow system (2.2)? Unfortunately, as we will show below, only one of them is stable. Let u be a (boundary or interior) spike solution. Then it is easy to see that (a , ξ ) defined by the following 8 9 a = ξq/(p−1) u , 10 ξ = 1 |Ω| −(p−1)/(qr−(p−1)(s+1)) Ω ur dx 13 14 15 16 17 18 19 20 21 22 23 (3.1) 28 29 34 35 36 37 38 39 40 By using (3.1), it is easy to see that the eigenvalues of problem (3.2) in H 2 (Ω) × L∞ (Ω) are the same as the eigenvalues of the following eigenvalue problem r−1 qr 2 p−1 Ω u φ p φ − φ + pu φ − u = α φ, r s + 1 + τ α Ω u φ ∈ H 2 (Ω). 45 16 17 (3.3) 18 19 20 21 22 23 24 25 26 A simple argument [8] shows that 27 T HEOREM 3.1. Any multiple-spike solution is linearly unstable for the shadow system (2.2). 28 29 30 31 Let 32 L (φ) = 2 φ − φ + pup−1 φ, r−1 qr Ω u φ p u . L (φ) = L (φ) − s + 1 + τ λ Ω ur 33 34 (3.4) Thus we can only concentrate on the study of stability for single-spike solutions. The study of stability and instability of single spike solutions can be divided into two parts: small eigenvalues and large eigenvalues. 35 36 37 38 39 40 41 3.1. Small eigenvalues for L 43 44 10 15 (3.2) 41 42 9 14 32 33 7 13 30 31 6 12 25 27 5 11 is a solution pair of the stationary problem to the shadow system (2.2). In this section, we analyze the following linearized eigenvalue problem ⎧ p−1 p ⎪ ⎨ 2 φ − φ + p a q φ − q aq+1 η = α φ , ∂φ ∂ν = 0 on ∂Ω, ξ ξ ⎪ r ar−1 φ 1+s ⎩ τ |Ω| Ω ξ s dx − τ η = α η. 24 26 4 8 11 12 3 42 43 In [73], it was proved that single boundary spike must concentrate at a critical point of the mean curvature function H (P ). On the other hand, at a non-degenerate critical point of 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 520 1 2 3 4 5 F:chipot506.tex; VTEX/Rita p. 34 J. Wei H (P ), there is also a single boundary spike. Furthermore, in [76], it is proved that the single boundary spike at a non-degenerate critical point of H (P ) is actually non-degenerate. Next we study the eigenvalue estimates associated with the linearized operator at u : p−1 L = 2 − 1 + pu . (Here the domain of L is H 2 (Ω).) We first note the following result. 6 7 φ − φ + pw p−1 φ = μφ in RN , φ ∈ H 1 RN 12 μ2 = · · · = μN +1 = 0, μ1 > 0, μN +2 < 0, . . . . (3.6) P ROOF. This follows from Theorem 2.12 of [42] and Lemma 4.2 of [58]. 15 22 23 24 27 28 20 T HEOREM 3.3. (See [76].) For sufficiently small, the following eigenvalue problem p−1 2 φ − φ + pu φ = τ φ in Ω, (3.7) ∂φ on ∂Ω ∂ν = 0 31 32 33 34 35 36 37 38 39 40 41 42 45 22 23 24 26 27 28 29 j τ 2 30 → η0 λ j , j = 1, . . . , N − 1, (3.8) + 1 (Here w (|z|) denotes the radial derivative of w with respect to |z|.) j Furthermore the eigenfunction corresponding to τ , j = 1, . . . , N − 1, is given by the following: φj = 31 32 where λ1 λ2 · · · λN −1 are the eigenvalues of the matrix Gb (P0 ) := (∂i ∂j H (P0 )), and 2 N − 1 R+N (w (|z|)) zN dz η0 = > 0. (3.9) ∂w 2 N +1 R N ( ∂z ) dz 43 44 21 25 μ1 admits exactly (N −1) eigenvalues τ1 τ2 · · · τN −1 in the interval [ μN+1 2 , 2 ], where μ1 and μN +1 are given by Lemma 3.2. j Moreover, we have the following asymptotic behavior of τ : 29 30 17 19 25 26 16 18 The small eigenvalues for L were characterized completely in [76]. 20 21 13 14 Moreover, the eigenfunction corresponding to μ1 is radial and of constant sign. 18 19 9 11 16 17 5 10 admits the following set of eigenvalues: 14 15 4 8 (3.5) 12 13 3 7 10 11 2 6 L EMMA 3.2. The following eigenvalue problem 8 9 1 N −1 i=1 ∂w,P aij + o(1) ∂τi (P ) 33 34 35 36 37 38 39 40 41 42 43 (3.10) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 35 Existence and stability of spikes 1 2 521 where P is the local maximum point of u , aj = (a1j , . . . , a(N −1)j )T is the eigenvector corresponding to λj , namely 3 j = 1, . . . , N − 1. (3.11) 5 7 8 9 10 12 dμP0 (z) = lim ε→0 13 17 18 19 20 e− ∂Ω 14 16 2|z−P0 | ε e 23 2|z−P | − ε 0 (3.12) . dz ∂Ω 28 29 30 31 32 35 36 37 38 39 40 41 42 ∂Ω ez−P0 ,a (z − P0 )i (z − P0 )j dμP0 (z) := Gi (P0 ) 45 10 12 13 15 16 17 18 19 20 23 is non-singular. (H2) 24 25 Such a vector a is unique. Moreover, Gi (P0 ) is a positive definite matrix. A geometric characterization of a non-degenerate peak point P0 is the following: P0 ∈ interior (convex hull of support dμP0 (z) . For a proof of the above facts, see Theorem 5.1 of [74]. In [75] and [74], the author proved the following theorem. 26 27 28 29 30 31 32 33 T HEOREM 3.4. Suppose that P0 is a non-degenerate peak point. Then for 1, there exists a single interior spike solution u concentrating at P0 . Furthermore, u is locally unique. Namely, if there are two families of single interior spike solutions u,1 and u,2 of (2.4) such that P1 → P0 , P2 → P0 where u,1 P1 = max u (P ), P ∈Ω̄ u,2 P2 = max u,2 (P ), P ∈Ω̄ then P1 = P2 , u,1 = u,2 . Moreover, 43 44 9 22 33 34 8 21 and 25 27 7 14 It is easy to see that the support of dμP0 (z) is contained in B̄d(P0 ,∂Ω) (P0 ) ∩ ∂Ω. A point P0 is called “non-degenerate peak point” if the followings hold: there exists a ∈ RN such that ez−P0 ,a (z − P0 ) dμP0 (z) = 0 (H1) 24 26 6 11 dz 21 22 4 5 For single interior spikes, we obtain similar results. But it becomes more involved since now the error is exponentially small. The existence of interior spike solutions depends highly on the geometry of the domain. In [74] and [75], the author first constructed a single interior spike solution. To state the result, we need to introduce some notations. Let 11 15 2 3 Gb (P0 ) aj = λj aj , 4 6 1 P1 = P2 1 = P0 + d(P0 , ∂Ω)a + o(1) as → 0. 2 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 522 1 2 3 F:chipot506.tex; VTEX/Rita p. 36 J. Wei Let w,P and ϕ,P be defined as in Section 2.3. (It was proved in [75] and [74] that − log[−ϕ,P (P )] → 2d(P , ∂Ω) as → 0.) Similarly, we obtain the following eigenvalue estimates for u 4 5 8 6 φ = τφ 2 φ − φ + pup−1 ∂φ = 0 on ∂Ω ∂ν in Ω, 7 (3.13) 13 14 15 16 19 τj = c0 + o(1) ϕ,P0 (P0 )λj −1 , τl = μl + o(1), l N + 2, 22 23 24 25 26 27 c0 = 2d 30 31 32 33 34 35 12 13 14 15 −2 (P0 , ∂Ω) RN pw p−1 w u∗ (r) ∂w 2 RN ( ∂y1 ) dy 16 17 < 0, (3.14) u − u = 0, u(0) = 1, u = u(r) in R . N N ∂w,P = , aj −1,l + o(1) ∂Pl P =P 22 (3.15) where aj = (aj,1 , . . . , aj,N Gi (P0 ) aj = λj aj , (3.16) is the eigenvector corresponding to λj , namely j = 1, . . . , N. 36 41 44 45 26 27 29 30 31 32 33 34 35 38 39 Let α be an eigenvalue of (3.3). Then the following holds. (The proof of it is routine. See Appendix of [77].) 42 43 25 37 3.2. A reduction lemma 39 40 24 36 37 38 23 28 l=1 )t 19 21 Furthermore, the eigenfunction (suitably normalized) corresponding to τj , j = 2, . . . , N + 1, is given by the following: φj 18 20 where u∗ (r) is the unique radial solution of the following problem 28 29 11 j = 2, . . . , N + 1, where λj , j = 1, . . . , N , are the eigenvalues of Gi (P0 ) and 20 21 10 τ1 = μ1 + o(1), 17 18 8 9 admits the following set of eigenvalues: 11 12 3 5 T HEOREM 3.5. The following eigenvalue problem 9 10 2 4 6 7 1 40 41 42 L EMMA A. (1) α = o(1) if and only if α = (1 + o(1))τj for some j = 2, . . . , N + 1, where τj is given by Theorem 3.3 or Theorem 3.5. 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 37 Existence and stability of spikes 1 523 (2) If α → α0 = 0. Then α0 is an eigenvalue of the following eigenvalue problem 2 qr φ − φ + pw p−1 φ − s + 1 + τ α0 φ ∈ H 2 RN . 3 4 5 6 2 w r−1 φ p w = α0 φ, r RN w 3 RN 4 5 (3.17) 7 8 8 9 C OROLLARY 3.6. For 1, (a , ξ ) is unstable with respect to the shadow system (2.2). 11 12 3.3. Large eigenvalues: NLEP method 13 14 15 16 17 18 19 20 21 22 14 This section is devoted to the study of the non-local eigenvalue problem (3.17). By [77] and [78], if problem (3.17) admits an eigenvalue λ with positive real part, then all single point-condensation solutions are unstable, while if all eigenvalues of problem (3.17) have negative real part, then all single point-condensation solutions are either stable or metastable. (Here we say that a solution is metastable if the eigenvalues of the associated linearized operator either are exponentially small or have strictly negative real parts.) Therefore it is vital to study problem (3.17). We first consider the simple case when τ = 0. Namely, we study the following NLEP: 23 24 φ − φ + pw 25 p−1 φ − γ (p − 1) 26 27 30 31 w r−1 φ p w = λφ, r RN w RN φ ∈ H RN , 2 (3.18) 34 35 36 37 38 39 40 41 γ := 18 19 20 21 22 24 25 28 qr , (s + 1)(p − 1) λ ∈ C, λ = 0, φ(x) = φ |x| . 29 30 (3.19) 31 32 For problem (3.18), it is known that when γ = 0, there exists an eigenvalue λ = μ1 > 0 (Lemma 3.2). An important property of (3.18) is that non-local term can push the eigenvalues of problem (3.18) to become negative so that the point-condensation solutions of the Gierer–Meinhardt system become stable or metastable. A major difficulty in studying problem (3.18) is that the left-hand side operator is not self-adjoint if r = p + 1. (In the classical Gierer–Meinhardt system, r = 2, p = 2.) Therefore it may have complex eigenvalues or Hopf bifurcations. Many traditional techniques do not work here. In [78] and [77], the eigenvalues of problem (3.18) in the following two cases 33 34 35 36 37 38 39 40 41 42 r = 2, or r = p + 1 44 45 17 27 42 43 16 26 32 33 15 23 where 28 29 10 11 12 13 6 7 A direct application of Theorem 3.5 is the following corollary. 9 10 1 43 44 are studied and the following results are proved. 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 524 1 2 J. Wei T HEOREM 3.7. (1) If (p, q, r, s) satisfies 3 4 (A) 5 6 7 1 2 3 qr γ= > 1, (s + 1)(p − 1) 4 5 6 and 8 9 F:chipot506.tex; VTEX/Rita p. 38 (B) r = 2, 1<p1+ 4 N or r = p + 1, 1 < p < 10 11 12 13 14 15 16 8 , +2 N +2 =N N −2 when N 3 and ( N −2 )+ = +∞ when N = 1, 2. Re(λ) < −c1 < 0 for some c1 > 0, where λ = 0 is an eigenvalue of where Then (3.18). (2) If γ < 1, problem (3.18) has an eigenvalue λ1 > 0. (3) If (C) r = 2, 4 p>1+ N 9 + +2 (N N −2 )+ 17 18 N +2 N −2 7 10 11 problem 15 16 17 and 1 < γ < 1 + c0 , 18 19 for some c0 > 0. Then problem (3.18) has an eigenvalue λ1 > 0. 20 21 22 23 24 25 21 We give a complete proof of Theorem 3.7 since this is the key element in all the stability result later on. The proof of Theorem 3.7 is based on the following important inequalities which are new and interesting. 26 27 28 29 30 31 32 33 34 39 40 41 42 43 44 45 23 24 25 27 28 2(p − 1) RN wφ RN w p φ 2 2 p−1 2 |∇φ| + φ − pw φ + 2 RN RN w 2 p+1 RN w −(p − 1) wφ ( RN w 2 )2 RN a1 dL2 2 (RN ) (φ, X1 ), 36 38 22 26 L EMMA 3.8. Let w be the unique solution to (2.8). (1) If 1 < p < 1 + N4 , then there exists a positive constant a1 > 0 such that 35 37 13 14 19 20 12 for all φ ∈ H 1 (RN ), where X1 := span{w, (2) If p = 1 + 4 N, 29 30 31 32 33 34 (3.20) ∂w ∂yj , j = 1, . . . , N }. 35 36 37 38 then there exists a positive constant a2 > 0 such that 39 2(p − 1) RN wφ RN w p φ 2 2 p−1 2 |∇φ| + φ − pw φ + 2 RN RN w 2 p+1 N w − (p − 1) R wφ ( RN w 2 )2 RN 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 39 Existence and stability of spikes 525 a2 dL2 2 (RN ) (φ, X2 ), 1 (3.21) 2 3 4 5 2 1 1 ∂w p−1 w + 2 y∇w(y), ∂yj , for all φ 2 := span{w, (3) There exists a positive constant a3 > 0 such that ∈ H 1 (RN ), where X j = 1, . . . , N}. 5 (p − 1)( RN w p φ)2 |∇φ|2 + φ 2 − pw p−1 φ 2 + p+1 RN RN w a3 dL2 2 (RN ) (φ, X1 ), ∀φ ∈ H 1 RN . 7 8 9 6 7 8 (3.22) 10 12 13 14 15 18 and L0 := −1+pw p−1 . Note that L is not selfadjoint if r = p +1. 21 L0 w = (p − 1)w p , L0 23 1 1 w + x∇w = w p−1 2 (3.23) 28 RN −1 L0 w w = RN w 1 1 w + x∇w = p−1 2 1 N − p−1 4 32 RN −1 p L0 w w = 39 40 41 42 43 44 45 = 36 38 25 27 31 37 24 26 and 30 35 17 21 28 34 15 20 26 33 14 22 25 29 13 19 Then 24 27 12 18 ∂w X0 := kernel(L0 ) = span j = 1, . . . , N . ∂yj 20 23 11 16 qr (p−1)(s+1) where γ = Let 19 22 9 10 P ROOF OF L EMMA 3.8. To this end, we first introduce some notations and make some preparations. Set r−1 φ N w Lφ := L0 φ − γ (p − 1) R w p , φ ∈ H 2 RN r RN w 16 17 3 4 6 11 1 RN RN w p 1 1 w + x∇w p−1 2 −1 L0 w RN 29 w2 , 30 (3.24) 32 33 1 1 L0 w = p−1 p−1 31 34 35 36 2 RN w . Since L is not selfadjoint, we introduce a new operator as follows: p N wφ N w φ p R R L1 φ := L0 φ − (p − 1) w − (p − 1) w 2 2 RN w RN w p+1 RN w RN wφ w. + (p − 1) 2 ( RN w )2 (3.25) 37 38 39 40 41 42 43 (3.26) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 526 1 2 F:chipot506.tex; VTEX/Rita p. 40 J. Wei By (3.26), L1 is selfadjoint. Next we compute the kernel of L1 . It is easy to see that ∂w w, ∂y , j = 1, . . . , N, ∈ kernel(L1 ). On the other hand, if φ ∈ kernel(L1 ), then by (3.23) j 3 5 6 7 9 12 13 14 15 19 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 7 8 9 10 11 12 13 14 15 16 φ − c1 (φ) 18 22 6 Hence 17 21 5 p p+1 N w φ N w R R RN wφ c1 (φ) = (p − 1) − (p − 1) , 2 2 ( R N w )2 RN w N wφ c2 (φ) = (p − 1) R . 2 RN w 11 20 4 where 10 16 2 3 L0 φ = c1 (φ)w + c2 (φ)w p 1 w 1 w + x∇w + c2 (φ)L0 = c1 (φ)L0 p−1 2 p−1 4 8 1 1 1 1 w + x∇w − c2 (φ) w ∈ kernel(L0 ). p−1 2 p−1 17 (3.27) 18 19 20 Note that 21 1 w p ( p−1 w + 12 x∇w) c1 (φ) = (p − 1)c1 (φ) 2 RN w 1 1 p+1 RN w RN w( p−1 w + 2 x∇w) − (p − 1)c1 (φ) ( RN w 2 )2 N RN w p+1 1 − = c1 (φ) − c1 (φ) 2 p−1 4 RN w RN by (3.24) and (3.25). This implies that c1 (φ) = 0. By (3.27) and Lemma 3.2, this shows that the kernel of L1 is exactly X1 . Now we prove (3.20). Suppose (3.20) is not true, then there exists (α, φ) such that (i) α ∂w is real and positive, (ii) φ ⊥ w, φ ⊥ ∂y , j = 1, . . . , N , and (iii) L1 φ = αφ. j We show that this is impossible. From (ii) and (iii), we have p N w φ (L0 − α)φ = (p − 1) R w. (3.28) 2 RN w We first claim that RN w p φ = 0. In fact if RN w p φ = 0, then α > 0 is an eigenvalue of L0 . By Lemma 3.2, α = μ1 and φ has constant sign. This contradicts with the fact that φ ⊥ w. Therefore α = μ1 , 0, and hence L0 − α is invertible in X0⊥ . So (3.28) implies p N w φ (L0 − α)−1 w. φ = (p − 1) R 2 w N R 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 41 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 Thus 1 p N w φ p w φ = (p − 1) R (L0 − α)−1 w w p , 2 RN RN RN w (L0 − α)−1 w w p , w 2 = (p − 1) RN 0= 12 13 14 15 16 w = 2 RN Let h1 (α) = RN 19 3 4 5 6 7 (L0 − α)−1 w (L0 − α)w + αw , (L0 − α)−1 w w. 8 9 10 (3.29) RN = −1 L0 w w = N 1 − p−1 4 13 then RN 1 1 w + x · ∇w w p−1 2 RN 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 15 16 17 w >0 19 20 since 1 < p < 1 + 22 23 14 18 2 20 21 11 12 − α)−1 w)w, 17 18 RN RN ((L0 h1 (0) = 2 RN 10 11 527 h1 (α) = 4 N. Moreover RN −2 L0 − α w w = 21 RN 2 (L0 − α)−1 w > 0. 22 23 24 This implies h1 (α) > 0 for all α ∈ (0, μ1 ). Clearly, also h1 (α) < 0 for α ∈ (μ1 , ∞) (since limα→+∞ h1 (α) = 0). This is a contradiction to (3.29)! This proves the inequality (3.20). The proof of (3.21) is similar. In this case we have −1 1 1 w + x∇w = 0. (3.30) L0 w w = w p−1 2 RN RN 25 Thus the kernel of L1 is X2 . The rest of the proof is exactly the same as before. To prove (3.22), we introduce p N w φ L3 φ := L0 φ − (p − 1) R p+1 w p . (3.31) RN w 33 Similar as before, the kernel of L3 is exactly X1 . Suppose (3.22) is not true, then there exists (α, φ) such that (a) α is real and positive, ∂w (b) φ ⊥ w, φ ⊥ ∂y , j = 1, . . . , N , and (c) L3 φ = αφ. j We show that this is impossible. From (a) and (c), we have (p − 1) RN w p φ p (L0 − α)φ = w . (3.32) p+1 RN w 39 26 27 28 29 30 31 32 34 35 36 37 38 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 528 1 2 3 J. Wei Similar to the proof of (3.20), we have that invertible in X0⊥ . So (3.32) implies 4 φ= 5 6 7 8 F:chipot506.tex; VTEX/Rita p. 42 (p − 1) RN wp φ RN w p+1 w p φ = 0, α = μ1 , 0, and hence L0 − α is RN 2 3 4 (L0 − α)−1 w p . 5 6 7 Thus 8 p N w φ p (L0 − α)−1 w p w p , w φ = (p − 1) R p+1 RN RN RN w (L0 − α)−1 w p w p . w p+1 = (p − 1) 9 10 11 12 RN 13 9 10 11 (3.33) RN 14 15 h3 (α) = (p − 1) RN 18 h3 (0) = (p − 1) 23 28 29 30 17 18 RN p p (L−1 0 w )w − 20 RN 21 w p+1 = 0. 22 23 Moreover 24 25 27 RN 16 w p+1 , 19 22 26 (L0 − α)−1 w p w p − then 21 24 h3 (α) = (p − 1) RN (L0 − α)−2 w p w p = (p − 1) RN 2 (L0 − α)−1 w p > 0. This implies h3 (α) > 0 for all α ∈ (0, μ1 ). Clearly, also h3 (α) < 0 for α ∈ (μ1 , ∞). A contradiction to (3.33)! 31 32 35 36 37 38 39 40 41 42 43 44 45 25 26 27 28 29 30 31 Using Lemma 3.8, we can prove Theorem 3.7(i). 32 33 34 13 15 17 20 12 14 Let 16 19 1 33 P ROOF OF T HEOREM 3.7( I ). We divide the proof into three cases. C ASE 1. r = 2, 1 < p < 1 + 34 35 4 N. 36 Let α0 = αR + iαI and φ = φR + iφI . Since α0 = 0, we can choose φ ⊥ kernel(L0 ). Then we obtain two equations N wφR L0 φR − (p − 1)γ R w p = αR φR − αI φI , 2 RN w N wφI L0 φI − (p − 1)γ R w p = αR φI + αI φR . 2 w N R 37 38 39 40 (3.34) 41 42 43 (3.35) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 43 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 10 11 529 Multiplying (3.34) by φR and (3.35) by φI and adding them together, we obtain 2 φR + φI2 −αR 1 2 3 4 RN = L1 (φR , φR ) + L1 (φI , φI ) p p R + RN wφI RN w φI RN wφR RN w φ + (p − 1)(γ − 2) 2 RN w 2 2 p+1 RN w . + (p − 1) wφR + wφI ( RN w 2 )2 RN RN 5 6 7 8 9 10 11 12 13 14 15 16 17 18 12 Multiplying (3.34) by w and (3.35) by w we obtain N wφR p R (p − 1) w φR − γ (p − 1) w p+1 2 RN RN RN w = αR wφR − αI wφI , RN 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 36 37 38 39 40 41 42 43 44 45 14 15 16 17 (3.36) RN 20 N wφI p R (p − 1) w φI − γ (p − 1) w p+1 2 RN RN RN w = αR wφI + αI wφR . RN 21 22 23 (3.37) RN Multiplying (3.36) by RN wφR and (3.37) by RN wφI and adding them together, we obtain p wφR w φR + (p − 1) wφI w p φI (p − 1) RN RN = αR + γ (p − 1) R N w p+1 RN 18 19 34 35 13 w2 RN RN wφR 2 RN + RN 2 wφI 24 25 26 27 28 29 30 31 32 . Therefore we have 2 −αR φR + φI2 RN = L1 (φR , φR ) + L1 (φI , φI ) p+1 ( 2 2 1 N w N wφR ) + ( RN wφI ) R R + (p − 1)(γ − 2) αR + γ 2 2 p−1 RN w RN w 2 2 p+1 N w . + (p − 1) R wφ + wφ R I ( RN w 2 )2 RN RN 33 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 530 1 F:chipot506.tex; VTEX/Rita p. 44 J. Wei Set 1 2 2 φR = cR w + φR⊥ , φR⊥ ⊥ X1 , 3 φI = cI w + φI⊥ , φI⊥ ⊥ X1 . 3 4 5 4 5 Then 6 7 8 RN wφR = cR RN w2 , RN 2 dL2 2 (RN ) (φR , X1 ) = φR⊥ L2 , 9 10 6 wφI = cI RN 7 w2 , 8 2 dL2 2 (RN ) (φI , X1 ) = φI⊥ L2 . 9 10 11 12 11 By some simple computations we have 12 13 13 L1 (φR , φR ) + L1 (φI , φI ) 2 2 + (γ − 1)αR cR + cI 14 15 16 RN 17 2 w 2 + (p − 1)(γ − 1)2 cR + cI2 2 2 + αR φR⊥ L2 + φI⊥ L2 = 0. 18 14 15 RN w p+1 16 17 18 19 20 19 20 By Lemma 3.8 (1) 21 2 (γ − 1)αR cR + cI2 22 23 24 26 28 31 32 33 34 37 38 39 40 41 42 43 44 45 23 24 RN w p+1 25 26 27 28 29 Since γ > 1, we must have αR < 0, which proves Theorem 3.7 in Case 1. C ASE 2. r = 2, p = 1 + 30 31 4 N. 32 33 Set 34 35 36 22 w2 2 2 + (αR + a1 ) φR⊥ L2 + φI⊥ L2 0. 27 30 RN 2 + (p − 1)(γ − 1)2 cR + cI2 25 29 21 w0 = 35 1 1 w + x∇w. p−1 2 (3.38) 37 We just need to take care of w0 . Suppose that α0 = 0 is an eigenvalue of L. Let α0 = αR + iαI and φ = φR + iφI . Since α0 = 0, we can choose φ ⊥ kernel(L0 ). Then similar to Case 1, we obtain two equations (3.34) and (3.35). We now decompose φR = cR w + bR w0 + φR⊥ , φI = cI w + bI w0 + φI⊥ , 36 φR⊥ ⊥ X1 , φI⊥ ⊥ X1 . 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 45 Existence and stability of spikes 1 531 Similar to Case 1, we obtain 1 2 3 4 5 6 7 8 2 L1 (φR , φR ) + L1 (φI , φI ) 2 + (γ − 1)αR cR + cI2 2 + αR bR RN 9 10 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 RN 2 2 w 2 + (p − 1)(γ − 1)2 cR + cI2 + bI2 w02 RN 2 w02 RN 4 w p+1 5 6 2 2 + φR⊥ L2 + φI⊥ L2 0 7 8 9 10 By Lemma 3.8(2) 11 12 3 2 (γ − 1)αR cR + cI2 2 + αR bR RN RN 2 w 2 + (p − 1)(γ − 1)2 cR + cI2 2 w02 + bI2 RN 2 w02 11 RN 12 w p+1 13 2 2 + (αR + a2 ) φR⊥ L2 + φI⊥ L2 33 34 35 36 37 38 39 40 41 42 43 44 45 15 16 0. 17 If αR 0, then necessarily we have 19 cR = cI = 0, 18 φR⊥ = 0, 20 φI⊥ 21 = 0. 22 23 Hence φR = bR w0 , φI = bI w0 . This implies that bR L0 w0 = (bR − bI )w0 , 24 25 bI L0 w0 = (bR + bI )w0 , 26 27 which is impossible unless bR = bI = 0. A contradiction! 28 29 +2 C ASE 3. r = p + 1, 1 < p < ( N N −2 )+ . 30 31 32 14 31 Let r = p + 1. L becomes 32 p qr N w · R L = L0 − wp . p+1 s + 1 RN w 33 34 35 36 We will follow the proof of Case 1. Let α0 = αR + iαI and φ = φR + iφI . Since α0 = 0, we can choose φ ⊥ kernel(L0 ). Then similarly we obtain two equations p N w φR L0 φR − (p − 1)γ R w p = αR φR − αI φI , p+1 w N R p N w φI L0 φI − (p − 1)γ R w p = αR φI + αI φR . p+1 w N R 37 38 39 40 41 (3.39) 42 43 (3.40) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 532 1 2 3 4 J. Wei Multiplying (3.39) by φR and (3.40) by φI and adding them together, we obtain 2 −αR φR + φI2 = L3 (φR , φR ) + L3 (φI , φI ) RN 6 7 9 10 11 12 13 14 15 16 17 1 2 3 4 ( RN w p φR )2 + ( RN w p φI )2 . + (p − 1)(γ − 1) p+1 RN w 5 8 F:chipot506.tex; VTEX/Rita p. 46 5 6 7 8 By Lemma 3.8(3) 2 αR φR + φI2 + a3 dL2 2 (φ, X1 ) 9 10 11 RN ( N w p φR )2 + ( RN w p φI )2 0 + (p − 1)(γ − 1) R p+1 RN w 12 13 14 15 which implies αR < 0 since γ > 1. Theorem 3.7(i) in Case 3 is thus proved. 16 18 19 20 21 22 18 P ROOF OF T HEOREM 3.7( II ). Assume that γ < 1. To prove Theorem 3.7(ii), we introduce the following function: h4 (λ) := (L0 − λ)−1 w p w r−1 . w r − γ (p − 1) (3.41) RN 23 24 25 26 27 28 29 30 31 32 RN Note that h4 (λ) is well defined in (0, μ1 ), where μ1 is the unique positive eigenvalue of L0 . Let us denote the corresponding eigenfunction by Φ0 . Since μ1 is a principal eigenvalue, we may assume that Φ0 > 0. It is easy to see that to prove Theorem 3.7(ii), it is enough to find a positive zero of h4 (λ). First we have p r−1 h4 (0) = w r − γ (p − 1) L−1 w w = (1 − γ ) w r > 0. (3.42) 0 RN RN RN 39 40 43 44 45 22 24 25 26 27 28 29 30 31 32 34 (3.43) 36 37 Multiplying (3.43) by Φ0 and integrating by parts, we have (μ1 − λ) Φλ Φ0 = Φ0 w p , RN 41 42 21 35 (L0 − λ)Φλ = w p . 37 38 20 33 Set Φλ = (L0 − λ)−1 w p . Then Φλ satisfies 35 36 19 23 33 34 17 which implies that Φλ Φ0 = RN RN 38 39 40 41 42 1 μ1 − λ 43 RN Φ0 w p . 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 47 Existence and stability of spikes 1 Let 1 2 3 4 533 Φλ = 1 (μ1 − λ) RN Φ02 Φ0 w Φ0 + Φλ⊥ , p RN 2 Φλ⊥ ⊥ Φ0 . (3.44) 4 5 6 7 8 9 10 11 5 Then as λ → μ1 , λ < μ1 , we have that Φλ⊥ L2 (RN ) is uniformly bounded and by (3.44) RN 8 Φλ w r−1 → +∞, 9 10 which implies that 11 12 h4 (λ) → −∞ as λ → μ1 , λ < μ1 . (3.45) 14 15 16 19 20 21 15 RN RN 18 19 (3.46) 22 h5 (λ) = (1 − γ ) 27 = (1 − γ ) 28 30 31 23 RN w 2 − γ (p − 1)λ RN 26 29 w − γ (p − 1)λ 2 RN RN w (L0 − λ)−1 (w) 24 25 26 2 wL−1 0 (w) + O λ . 27 28 Since RN wL−1 0 (w) < 0, we see that for 0 < γ − 1 small, there is a small λ0 > 0 such that h5 (λ0 ) > 0. 32 33 36 39 40 41 42 45 31 34 35 36 D(r) := (p − 1) −1 r−1 r−1 w RN L0 w RN ( RN w r )2 where L0 = − 1 + u(|x|)}) and pw p−1 (L−1 0 37 w2 >0 (3.47) 38 39 40 exists in Hr2 (RN ) = {u ∈ H 2 (RN ) | u(x) = 43 44 30 33 T HEOREM 3.9. (1) If 37 38 29 32 For general r, the author in [80] proved the following: 34 35 20 21 We write it as 24 25 16 17 P ROOF OF T HEOREM 3.7( III ). Similarly, we just need to find a zero of 2 w − γ (p − 1) w(L0 − λ)−1 w p . h5 (λ) := 22 23 13 14 By (3.42) and (3.45), there is a λ0 ∈ (0, μ1 ) such that h4 (λ0 ) = 0. This proves (ii) of Theorem 3.7. 17 18 6 7 12 13 3 41 42 43 1 1 1+ √ <γ <1+ √ , 1 + ρ0 1 − ρ0 44 (3.48) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 534 1 2 3 4 5 6 7 8 9 10 11 12 13 F:chipot506.tex; VTEX/Rita p. 48 J. Wei where ρ0 > 0 is given by p+1 N w ρ0 := ) R < 1. 2p 2 RN w RN w 1 2 (3.49) 16 17 Then for any non-zero eigenvalue λ of problem (3.18), we have Re(λ) < −c1 < 0 for some c1 > 0. (2) If (p, q, r, s) satisfies 2r N +2 1+ <p< and γ < 1 + c0 , (3.50) N N −2 + for some c0 > 0. Then problem (3.18) has a real eigenvalue λ1 > 0. 24 25 26 29 30 31 32 33 F (r) = 1 − 36 37 40 41 42 43 12 15 16 17 22 2r N 23 and |γ − 2| * γ −2 F (p + 1) + F (p + 1)(F (p + 1) − F (2)), F (r) γ γ 24 25 (3.51) 26 27 then for any non-zero eigenvalue λ of problem (3.18), we have Re(λ) < −c1 < 0 for some c1 > 0. R EMARK . Condition (3.51) holds if 2 < r < p + 1, F (r) 0 (i.e., 1 < p 1 + 2r N ) and 1 < γ 2. Thus in this case we obtain the stability of the non-zero eigenvalues of (3.18). This is the first explicit result for the case when r ∈ / {2, p + 1}. For γ > 2, we need F (r) * γ − 2 F (p + 1) − F (p + 1)(F (p + 1) − F (2)) . γ Going back to the shadow system case, the following result was proved in [77]. 28 29 30 31 32 33 34 35 36 37 38 T HEOREM 3.11. Assume that 1 and τ is small. If (p, q, r, s) satisfy (A) and (B) in Theorem 3.7, then (1) single boundary spike solution at a non-degenerate local maximum point of mean curvature is stable, and (2) single interior spike solution is metastable. 44 45 11 21 38 39 10 20 p−1 N. 2r Suppose 2 < r < p + 1, 1 < p < 1 + 34 35 9 19 27 28 8 18 T HEOREM 3.10. Let 22 23 7 14 Generally speaking, D(r) is very difficult to compute. A recent result of the author and L. Zhang partially solved this problem and moreover we obtained more general and explicit result. For example the following result are proved [81]. 20 21 6 13 18 19 4 5 14 15 3 39 40 41 42 43 44 Related work can also be found in [59] and [60]. 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 49 Existence and stability of spikes 1 535 3.4. Uniqueness of Hopf bifurcations 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 2 In Section 3.3, we have discussed the NLEP (3.17) when τ = 0. It is easy to see that when τ small, results in Theorem 3.7 still hold. On the other hand, for τ large, it is easy to see that there is an unstable eigenvalue [8] to (3.17). (In fact, as τ → +∞, there is a positive eigenvalue near μ1 > 0.) Therefore, as τ varies from 0 to ∞, Hopf bifurcation may occur. In this section, we show that in some special cases, Hopf bifurcation is actually unique. We consider the following non-local eigenvalue problem (putting r = p = 2, s = 0 in (3.17)) γ N wφ R Lφ := φ − φ + 2wφ − w 2 = λ0 φ, 1 + τ λ0 R 2 w 2 21 22 23 24 25 26 27 28 φ ∈ H 2 RN . 33 34 35 36 37 N wφ w2 = 0 L0 φ − γ R 2 w N R which implies that L0 N wφ φ − γ R w =0 2 RN w and hence by Lemma 3.2 N wφ φ − γ R w ∈ X0 . 2 RN w 38 39 40 41 42 45 7 8 9 12 13 14 15 16 17 18 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 This is impossible since φ is radially symmetric and φ = cw for all c ∈ R. Thus there must be a point √ τ1 at which L has a Hopf bifurcation, i.e., L has a purely imaginary eigenvalue α = −1αI . To prove Theorem 3.12, all we need to show is that τ1 is unique. That is 43 44 6 19 P ROOF OF T HEOREM 3.12. Let γ > 1. As in [8], we may consider radially symmetric functions only. By Theorem 1.4 of [77], for τ = 0 (and by perturbation, for τ small), all eigenvalues lie on the left half plane. By [8], for τ large, there exist unstable eigenvalues. Note that the eigenvalues will not cross through zero: in fact, if λ0 = 0, then we have 30 32 5 11 (3.52) T HEOREM 3.12. Let L be defined by (3.52). Assume that N 3 and γ > 1. Then there exists a unique τ = τ1 > 0 such that for τ < τ1 , (3.52) admits a positive eigenvalue, and for τ > τ1 , all non-zero eigenvalues of problem (3.52) satisfy Re(λ) < 0. At τ = τ1 , L has a Hopf bifurcation. 29 31 4 10 19 20 3 39 40 41 42 43 L EMMA 3.13. Let γ > 1. Then there exists a unique τ1 > 0 such that L has a Hopf bifurcation. 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 536 1 2 3 4 5 6 7 8 Let φ0 = φ0R + √ −1φ0I . Then from (3.53), we obtain the two equations 11 12 13 14 15 (3.55) 28 29 30 31 32 33 40 41 42 45 14 19 21 −1 φ0R = L0 L20 + αI2 w 2 . 22 (3.56) 23 24 Substituting (3.56) into (3.54) and (3.55), we obtain 2 2 −1 2 1 RN [wL0(L0 + αI ) w ] = , 2 γ RN w 2 + α 2 )−1 w 2 ] τ I RN [w(L 0 = . 2 γ w R2 25 26 (3.57) 27 28 29 (3.58) 30 31 32 Let 33 h6 (αI ) = (L20 RN wL0 R2 + αI2 )−1 w 2 . w2 34 35 36 Then integration by parts gives 2 −1 w 2 2 2 N w (L0 + αI ) h6 (αI ) = R . 2 RN w 37 Note that 42 43 44 13 20 36 39 11 17 L0 φ0I = αI φ0R . −1 φ0I = αI L20 + αI2 w 2 , 35 38 7 16 So φ0R = αI−1 L0 φ0I and 34 37 6 15 24 27 5 18 23 26 4 12 L0 φ0R = w 2 − αI φ0I , 22 25 3 10 (3.54) Note that (3.54) is independent of τ . Let us now compute RN wφ0R . Observe that (φ0R , φ0I ) satisfies 19 21 2 8 18 20 1 9 R 1 N wφ0 R = , N γ R2 w I τ αI N wφ0 R . = N γ w 2 R 10 17 J. Wei √ P ROOF. Let λ0 = −1αI be an eigenvalue of L. Without loss of generality, we may √ assume that α > 0. (Note that − −1α is also an eigenvalue of L.) Let φ0 = (L0 − I I √ −1αI )−1 w 2 . Then (3.52) becomes √ 1 + τ −1αI N wφ0 R . (3.53) = 2 γ RN w 9 16 F:chipot506.tex; VTEX/Rita p. 50 h6 (αI ) = −2αI 38 39 40 41 43 2 2 + α 2 )−2 w 2 I RN w (L 0 < 0. 2 w N R 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 51 Existence and stability of spikes 1 So since 1 2 3 h6 (0) = 4 RN 5 6 7 8 537 2 2 w(L−1 0 w ) > 0, 2 RN w 3 4 5 h6 (αI ) → 0 as αI → ∞, and γ > 1, there exists a unique αI > 0 such that (3.57) holds. Substituting this unique αI into (3.58), we obtain a unique τ = τ1 > 0. Lemma 3.13 is thus proved. 9 10 17 18 19 13 14 In all the previous sections, it is always assumed that is small. However, in practical applications, it is vital to know how small should be. The finite case has been completely characterized in one-dimensional case by Wei and Winter [89]. We summarize the results here. Without loss of generality, we may assume that Ω = (0, 1). That is, we consider ⎧ ap 2 ⎪ ⎨ at = axx − a + ξ q , 0 < x < 1, t > 0, 1 τ ξt = −ξ + ξ −s 0 a r dx, ⎪ ⎩ a > 0, ax (0, t) = ax (1, t) = 0. 20 21 22 23 24 25 26 27 29 ξ qr 1+s− p−1 44 45 27 30 2 33 p 34 ux (0) = ux (1) = 0. (3.60) 35 36 Letting 37 38 43 26 29 36 42 25 32 ux (x) < 0, 0 < x < 1, 41 22 31 35 40 19 28 uxx − u + u = 0, 39 18 24 ur (x) dx 34 37 17 23 and 33 16 21 (3.59) 0 30 32 1 = 15 20 The steady-state problem of (3.59) is equivalent to the following problem for the trans− q formed function u given by u (x) = ξ p−1 a(x): 28 31 10 12 3.5. Finite case 14 16 8 11 12 15 7 9 Theorem 3.12 now follows from Lemma 3.13. 11 13 6 38 1 L := (3.61) 40 and rescaling u(x) = wL (y), where y = Lx, we see that wL satisfies the following ODE: p 0 < y < L, 41 42 43 wL − wL + wL = 0, wL (y) < 0, 39 44 wL (0) = wL (L) = 0. (3.62) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 538 1 2 F:chipot506.tex; VTEX/Rita p. 52 J. Wei Since (3.62) is an autonomous ODE, it is easy to see that a non-trivial solution exists if and only if √ p−1 < π 3 4 5 π or L > √ . p−1 8 9 10 (3.63) (3.64) 13 16 17 18 19 q p−1 aL (x) = ξL 21 22 24 25 λ1 > 0, 31 33 34 1 = 0 (3.65) (3.66) 14 15 16 17 18 19 21 23 24 25 27 29 Hence for the map L from to we know that inverse of L. This implies that L−1 wL is well defined. Then we have the following theorem L2 (I ) (3.67) L−1 exists, where L−1 is the 30 31 32 33 34 35 T HEOREM 3.14. Assume that L > r = 2, 39 L √π p−1 36 and either 37 38 wL L −1 wL dy > 0 (3.68) 0 40 39 40 41 or 42 42 r = p + 1. (3.69) 44 45 12 28 j = 2, 3, . . . . λj < 0, H 2 (I ) 38 43 10 26 p−1 + pwL φ. 37 41 9 22 35 36 8 20 wLr (Lx) dx. It is proved [89] that L has the spectrum 30 32 wL (Lx), L[φ] = φ − φ 27 29 qr 1+s− p−1 ξL Before stating our results, we first introduce some notation. Let I = (0, L) and φ ∈ H 2 (I ). We define the following operator: 26 28 7 13 Thus all non-monotone steady-state solutions are linearly unstable. Therefore we focus our attention on monotone solutions. There are two monotone solutions—the monotone increasing one and the monotone decreasing one. Since these two solutions differ by reflection, we consider the monotone decreasing function only. This solution is then called π u and it has the least energy among all positive solutions of (3.60), see [60]. If L √p−1 , then wL = 1. We also denote the corresponding solutions to (3.59) by 20 23 5 11 ax (x, t0 ) = 0 for all (x, t) ∈ (0, 1) × [t0 , +∞). 12 15 4 6 The stability of steady-state solutions to (3.59) has been a subject of study in the last few years. A recent result of [56] (see Theorem 1.1 of [56]) says that a stable solution to (3.59) must be asymptotically monotone. More precisely, if (A(x, t), ξ(t)), t 0 is a solution to (3.59) that is linearly neutrally stable, then there is a t0 > 0 such that 11 14 2 3 6 7 1 43 44 Then (aL , ξL ) (given by (3.65)) is a linearly stable steady state to (3.59) for τ small. 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 53 Existence and stability of spikes 1 2 3 This theorem reduces the issue of stability to the computation of the integral 4 5 6 7 8 539 1 2 L 3 wL L−1 wL dy. 4 0 5 6 This integral is quite difficult to compute for general L. For τ finite, we have the following theorem. 7 8 9 10 11 12 13 14 15 16 17 18 19 9 π T HEOREM 3.15. Let (3.68) be true and L > √p−1 . Then there exists a unique τc > 0 such that for τ < τc , (aL , ξL ) is stable and for τ > τc , (aL , ξL ) is unstable. At τ = τc , there exists a unique Hopf bifurcation. Furthermore, the Hopf bifurcation is transversal, namely, we have 22 23 24 25 (3.70) 28 29 30 33 34 35 36 37 38 39 40 41 42 43 44 45 13 15 16 18 where λR is the real part of the eigenvalue. 19 20 Using Weierstrass p(z) functions and Jacobi elliptic integrals, one can show that L −1 0 wL L wL dy > 0 for all L > π in the cases r = 2, p = 2, 3. The original Gierer– Meinhardt system ((p, q, r, s) = (2, 1, 2, 0)) falls into this class. Thus for the shadow system of the original Gierer–Meinhardt system, we have a complete picture of the stability of (aL , ξL ) for any τ > 0 and any L > 0, by the following theorem 21 22 23 24 25 26 π T HEOREM 3.16. Assume that L > √p−1 and r = 2, p = 2 or 3. Then there exists a unique τc > 0 such that for τ < τc , (aL , ξL ) is stable and for τ > τc , (AL , ξL ) is unstable. At τ = τc , there exists a Hopf bifurcation. Furthermore, the Hopf bifurcation is transversal. 31 32 12 17 26 27 11 14 dλR > 0, dτ τ =τc 20 21 10 27 28 29 30 31 Theorem 3.16 gives a complete picture of the stability of non-trivial monotone solutions π we necessarily have wL ≡ 1. Combining this with the in terms of L since for L √p−1 results of [56], we have completely classified stability and instability of all steady-state solutions for all > 0 for the shadow system of the original Gierer–Meinhardt system. We do not know if the Hopf bifurcation in Theorem 3.15 is subcritical or supercritical. This is related to another interesting question: is there time-periodic solution (a(x, t), ξ(x, t)) to (3.59) at the Hopf bifurcation point τ = τc ? If so, is it stable or unstable? We can also extend this idea to general domains in RN , N 2. Namely we consider ⎧ p ⎪ a = a − a + aξ q , x ∈ ΩL , ⎪ ⎨ t τ ξt = −ξ + ξ −s |Ω1L | ΩL a r , ⎪ ⎪ ⎩ a > 0, ∂a ∂ν = 0 on ∂ΩL , 32 33 34 35 36 37 38 39 40 41 42 t > 0, 43 (3.71) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 540 1 2 3 F:chipot506.tex; VTEX/Rita p. 54 J. Wei where we have scaled the into the domain through ΩL = 1 Ω. In this case, let us assume that ΩL ⊂ RN is a smooth and bounded domain, and the exponents (p, q, r, s) satisfy the following condition 4 p > 1, q > 0, r > 0, 6 s 0, qr γ := > 1, (p − 1)(s + 1) 6 7 and p is subcritical: 9 1<p< 10 11 13 14 15 8 9 N +2 N −2 if N 3; 1 < p < +∞ if N = 2. 10 11 The steady state solution of (3.71) is given by q qr 1 1+s− p−1 p−1 a=ξ u, ξ = ur |ΩL | ΩL 12 13 (3.72) 16 17 19 20 u − u + up = 0, ∂u ∂ν = 0 18 u > 0 in ΩL , on ∂ΩL . (3.73) E[wL ] = 24 25 23 inf u∈H 1 (ΩL ), u≡0 E[u] (3.74) 26 27 2 2 Ω (|∇u| + u ) . E[u] = L 2 ( ΩL up+1 ) p+1 29 30 28 29 30 31 35 36 37 38 39 40 31 The corresponding steady-state solution to the shadow system (3.71) is denoted by q qr 1+s− p−1 1 p−1 = wr . (3.75) a L = ξL wL , ξL |ΩL | ΩL L 43 44 45 32 33 34 35 36 Let 37 L[φ] = φ − φ p−1 + pwL φ. 38 39 40 Then we have the following lemma whose proof is similar to Lemma 3.2. 41 42 24 25 28 34 20 22 where 27 33 19 21 We again consider the minimizer solution wL (x) which satisfies (3.73) and 23 32 15 17 21 26 14 16 where u is a solution of the following problem: 18 22 3 5 7 12 2 4 5 8 1 41 L EMMA 3.17. Consider the following eigenvalue problem Lφ = λφ, in ΩL , ∂φ on ∂ΩL . ∂ν = 0 42 43 44 (3.76) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 55 Existence and stability of spikes 1 541 Then λ1 > 0 and λ2 0. 1 2 2 3 3 We now put two important assumptions: We first assume that 4 4 5 5 6 (A1) L 7 8 −1 6 exists. 7 8 Under (A1), we assume that 9 10 10 wL L−1 wL > 0. (A2) 11 9 11 ΩL 12 12 13 13 We can now state the following theorem 14 14 15 16 15 T HEOREM 3.18. Assume that either 16 17 17 r = p + 1, 18 and (A1) holds, 18 19 20 19 or 20 21 22 21 r = 2, 22 and (A1) and (A2) hold. 23 24 25 26 27 28 29 30 31 32 23 Then (aL , ξL ) is linearly stable for τ small. In the case of r = 2, there exists a unique τ = τc such that (aL , ξL ) is stable for τ < τc , unstable for τ > τc , and there is a Hopf bifurcation at τ = τc . Furthermore, the Hopf bifurcation is transversal. The proof of Theorem (3.18) is similar to the one-dimensional case. It remains an interesting and difficult question as to verify (A1) and (A2) analytically. If L is large, the assumption (A1) is verified in [76] and assumption (A2) holds true if 33 34 35 36 37 38 39 1<p<1+ 4 . N This recovers the results of [77]. It is difficult to verify (A1) and (A2) in general domains. One may ask: does (A1) hold true for generic domains? 27 28 29 30 31 32 34 35 36 37 38 39 41 3.6. The stability of boundary spikes for the Robin boundary condition 43 45 26 40 41 44 25 33 (3.77) 40 42 24 42 43 The stability of least energy solution in the Robin boundary condition case is quite complicated. We state the following result which deals with one-dimensional case only: 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 542 1 2 3 4 5 6 7 F:chipot506.tex; VTEX/Rita p. 56 J. Wei T HEOREM 3.19. (See [45].) Consider the following 1 ⎧ p at = 2 axx − a + aξ q , 0 < x < 1, t > 0, ⎪ ⎪ ⎪ 1 ⎨ τ ξt = −ξ + ξ −s 0 a r dx, ⎪ ⎪ ⎪ ⎩ a > 0, ax (0, t) + λa(0, t) = ax (1, t) + λa(1, t) = 0, hx (0, t) = hx (1, t) = 0. 2 3 4 (3.78) 6 7 8 9 10 11 8 Assume that r = 2, 1 < p 3 or r = p + 1, 1 < p < +∞. Then for each λ ∈ (0, 1) the least energy solution is stable for τ < τ1 and unstable for τ > τ1 . At τ1 , there is a Hopf bifurcation. 12 13 14 15 16 17 18 19 22 23 24 The main idea of the proof is similar to that of Theorem 3.14. Here we have to study an NLEP on a half line with Robin boundary condition: ⎧ ∞ ⎨ φ − φ + pw p−1 φ − γ (p − 1) 0 wx0 φ w p = αφ, ∞ 2 x0 x0 0 wx0 ⎩ φ (0) − λφ(0) = 0 27 28 29 30 33 34 35 11 13 14 15 16 0 < y < +∞, 17 (3.79) 18 19 where wx0 = w(y − x0 we need to show that 0 ) with w (−x p−1 0 ) = λw(−x0 ). Let Lx0 (φ) = φ − φ + pwx0 φ. Then 21 22 23 ∞ wx0 L−1 x0 (wx0 ) > 0. 24 (3.80) 25 26 ∞ By some lengthy computations, we can show that the function 0 wx0 [L−1 x0 (wx0 )] is an increasing function in x0 when p < 3, and a constant when p = 3, and an decreasing function when p > 3. 31 32 10 20 25 26 9 12 20 21 5 27 28 29 30 31 R EMARK 3.6.1. An interesting phenomena is the case of 3 < p < 5. In this case, one can show that there exists a a0 ∈ (0, 1) such that the boundary spike is stable when a ∈ (0, a0 ) and unstable when a ∈ (a0 , 1). It is quite interesting to see that the Robin boundary condition can also introduce some instability. 32 33 34 35 36 36 37 37 38 39 38 4. Full Gierer–Meinhardt system: One-dimensional case 40 41 42 43 44 45 39 40 In this section, we study the full Gierer–Meinhardt system in the one-dimensional case. Unlike the shadow system case, where one can reduce the existence of solutions to a variational elliptic problem, there is no variational structure for the full Gierer–Meinhardt system. This is the major problem, which is also the source of all interesting new phenomena. We begin with the steady-state problem in the full space case. 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 57 Existence and stability of spikes 1 543 4.1. Bound states: the case of Ω = R1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 2 Let Ω = R1 . By a change of variables the steady-state problem for (GM) can be conveniently written as follows ⎧ ap ⎪ in R1 , ⎨ a − a + hq = 0, a > 0 r (4.1) h − σ 2 h + ahs = 0, h > 0 in R1 , ⎪ ⎩ a(x), h(x) → 0 as |x| → +∞ 20 21 22 23 24 27 28 29 σ2 = 32 33 34 35 36 37 38 39 D 42 45 8 9 14 The existence of multiple spikes solutions to (4.1) is referred to as “symmetry-breaking” phenomena. This was proved in [12] (by dynamical system techniques) and [7] (by PDE methods). We will sketch the PDE methods in Section 5.1. 15 16 17 18 T HEOREM 4.1. (See [7,12].) For each fixed positive integer k, there exists σk > 0 such that problem (4.1) has a solution (a , h ) with the following properties 19 20 21 k ck w x − ξjσ a (x) ∼ σ 22 (4.2) j =1 23 24 25 where ck > 0 is a generic constant and ξjσ = j − 26 1 1 k+1 log + O log log , 2 σ σ 27 j = 1, . . . , k. (4.3) 28 29 30 4.2. The bounded domain case: Existence of symmetric K-spikes Without loss of generality, we may assume that Ω = (−1, 1). We consider the following elliptic system ⎧ ap 2 ⎪ ⎨ a − a + hq = 0, −1 < x < 1, r (4.4) Dh − h + ahs = 0, −1 < x < 1, ⎪ ⎩ a (±1) = h (±1) = 0. 31 32 33 34 35 36 37 38 39 40 In this case, the existence of multiple-peaked solutions was first obtained by I. Takagi in [67]. 43 44 7 13 40 41 6 12 1. 30 31 5 11 2 25 26 4 10 where 18 19 3 41 42 43 T HEOREM 4.2. (See [67].) Fix any positive integer K. If √ sufficiently small, there exD ists a K-peaked solution (a,K , hK ) to (4.4) such that (a,K , h,K ) has exactly K local 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 544 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 F:chipot506.tex; VTEX/Rita p. 58 J. Wei maximum points −1 < x1 < x2 < · · · < xK < 1 which are equally distributed. In fact, we have 2j − 1 , xj = −1 + K 18 19 20 j = 1, . . . , K. 4 5 Takagi’s proof uses the symmetry of the problems: by reflection, one can reduce the existence of multiple symmetric spikes solutions to studying the existence of one boundary spike solution. Namely, we just need to study the following system ⎧ 2 p 1 a − a + ahq = 0, 0 < x < 2K , ⎪ ⎪ ⎪ r ⎪ ⎨ Dh − h + a s = 0, 0 < x < 1 , h 2K q x ⎪ p−1 ⎪ a(x) ∼ ξ w( ), h(0) = ξ, ⎪ ⎪ ⎩ 1 1 a (0) = a ( 2K ) = h (0) = h ( 2K ) = 0. 27 28 29 30 35 36 37 38 39 40 41 42 43 44 45 13 14 15 17 18 19 20 21 22 4.3. The bounded domain case: existence of asymmetric K-spikes 23 24 In the bounded domain case, as D is getting smaller, more and more new solutions appear. By using the same matched asymptotic analysis in [34], M. Ward and Wei in [70] discovered that for D < DK , where DK is given by (4.67) below, problem (4.4) has asymmetric K-peaked steady-state solutions. Such asymmetric solutions are generated by two types of peaks-called type A and type B, respectively. Type A and type B peaks have different heights. They can be arranged in any given order 25 26 27 28 29 30 31 ABAABBB...ABBBA...B 33 34 9 16 For the one boundary spike solution, one can use the Implicit Function Theorem, since the linearized operator is invertible in the space of functions with Neumann boundary conditions. (The last statement follows from the fact that the kernel of the operator − 1 + pw p−1 consists exactly those of partial derivatives of w. See Lemma 3.2.) 31 32 8 12 (4.5) 24 26 7 11 22 25 6 10 21 23 2 3 16 17 1 32 33 to form an K-peaked solution. The existence of such solutions is surprising. It shows that the solution structure of (4.4) is much more complicated than one would expect. The stability of such asymmetric K-peaked solutions is also studied in [70], through a formal approach. We remark that asymmetric patterns can also be obtained for the Gierer–Meinhardt system on the real line, see [12]. In this and next section, we present a rigorous and unified theoretic foundation for the existence and stability of general K-peaked (symmetric or asymmetric) solutions. In particular, the results of [34] and [70] are rigorously established. Moreover, we show that if the K peaks are separated, then they are generated by peaks of type A and type B, respectively. This implies that there are only two kinds of K-peaked patterns: symmetric K-peaked solutions constructed in [67] and asymmetric K-peaked patterns constructed in [70]. 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 59 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 10 11 12 545 The existence proof is based on Lyapunov–Schmidt reduction. Stability is proved by first separating the problem into the case of large eigenvalues which tend to a non-zero limit and the case of small eigenvalues which tend to zero in the limit → 0. Large eigenvalues are then explored by studying non-local eigenvalue problems using results in Section 3.3 and employing an idea of Dancer [8]. Small eigenvalues are calculated explicitly by an asymptotic analysis with rigorous error estimates. In this section, we present the existence part. Before we state our main results, we introduce some notation. Let GD (x, z) be the Green function of DGD (x, z) − GD (x, z) + δz i(x) = 0 in (−1, 1), (4.6) GD (−1, z) = GD (1, z) = 0. 13 14 16 GD (x, z) = 17 θ sinh(2θ) θ sinh(2θ) cosh[θ (1 + x)] cosh[θ (1 − z)], −1 < x < z, cosh[θ (1 − x)] cosh[θ (1 + z)], z<x<1 (4.7) 27 28 29 30 31 32 33 34 37 38 39 40 41 42 43 44 45 7 8 9 10 11 12 16 19 20 θ = D −1/2 . 21 22 We set 23 24 1 − √1 |x−z| KD |x − z| = √ e D , 2 D (4.8) p−1 (p−1)(s+1)−qr r ξ := w (z) dz . ) ∈ (−1, 1)K , GD (t) = GD (ti , tj ) . 29 30 31 32 34 36 let 37 (4.10) as ∇ti . When i = j , we can define ∇ti G(ti , tj ) in the classical way. When i = j , KD (|ti − tj |) = KD (0) = 28 35 We introduce several matrices for later use: For t = (t1 , . . . , tK ∂ ∂ti 27 33 (4.9) R Let us denote 25 26 to be the singular part of GD (x, z) and by GD = KD − HD we define the regular part HD of GD . Note that HD is C ∞ in both x and z. 0 < 1 be K points in (−1, 1) and w be the unique Let −1 < t10 < · · · < tj0 < · · · < tK solution of (2.8). Put 35 36 6 17 22 26 5 18 21 25 4 15 where 20 24 3 14 18 23 2 13 We can calculate explicitly 15 19 1 1 √ 2 D is a constant and we define ∂ H (x, ti ). ∇ti GD (ti , ti ) := − ∂x x=ti 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 546 1 J. Wei Similarly, we define 1 2 3 ∇ti ∇tj GD (ti , tj ) = 4 5 6 7 10 11 14 N 16 17 if i = j . 3 (4.11) qr −s GD ti0 , tj0 ξ̂j0 p−1 = ξ̂i0 , 6 7 8 (4.12) (4.13) 25 14 15 i = 1, . . . , N. (4.14) 20 B = (bij ). 21 (4.15) 24 25 (4.16) 29 where σ (B) is the set of eigenvalues of B. 30 30 31 31 R EMARK 4.3.1. Since the matrix B is of the form GD D, where GD is symmetric and D is a diagonal matrix, it is easy to see that the eigenvalues of B are real. 34 37 40 qr −s GD (ti , tj )ξˆj p−1 = ξ̂i , 35 36 37 39 i = 1, . . . , K. (4.17) j =1 41 40 41 42 45 33 38 K 39 44 32 34 By the assumption (H2) and the implicit function theorem, for t = (t1 , . . . , tK ) near 0 ), there exists a unique solution ξ̂ (t) = (ξ̂ (t), . . . , ξ̂ (t)) for the following t0 = (t10 , . . . , tK 1 K equation 38 43 27 28 29 36 22 26 p−1 ∈ / σ (B), qr − s(p − 1) 28 35 17 23 27 33 16 19 Our second assumption is the following: (H2) It holds that 26 32 11 13 23 24 10 18 qr −s−1 bij = GD ti0 , tj0 ξ̂j0 p−1 , 22 9 12 Next we introduce the following matrix 21 4 5 j =1 18 20 ∇ti ∇tj GD (ti , tj ) if i = j , We now have our first assumption: (H1) There exists a solution (ξ̂10 , . . . , ξ̂N0 ) of the following equation 15 19 2 ∇GD (t) = ∇ti GD (ti , tj ) , ∇ 2 GD (t) = ∇ti ∇tj GD (ti , tj ) . 9 13 ∂ ∂ ∂x x=ti ∂y y=ti HD (x, y) Now the derivatives of G are defined as follows: 8 12 F:chipot506.tex; VTEX/Rita p. 60 42 Set 43 H(t) = ξ̂i (t)δij . 44 (4.18) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 61 Existence and stability of spikes 1 547 We define the following vector field: 2 1 2 F (t) = F1 (t), . . . , FK (t) , 3 3 4 5 4 where 5 6 6 7 Fi (t) = 8 K ∇ti GD (ti , tl )ξ̂l qr p−1 −s 7 8 l=1 9 9 qr p−1 −s 10 = −∇ti HD (ti , ti )ξ̂i 11 + ∇ti GD (ti , tl )ξ̂l qr p−1 −s 10 , 11 l=i 12 12 i = 1, . . . , K. 13 (4.19) 14 15 16 17 14 Set 15 16 M(t) = ξ̂i−1 ∇tj Fi (t) . (4.20) 18 19 20 21 19 20 21 F (t0 ) = 0, det M(t0 ) = 0. 23 24 (4.21) 30 ∇tj ξ̂i = 33 34 36 38 39 42 ∇tj ξ̂i = 45 32 33 34 35 36 qr −s p−1 N 37 GD (ti , tl )ξ̂l qr p−1 −s−1 qr p−1 −s ∇tj ξ̂l + ∇tj GD (ti , tj )ξ̂j 38 . l=1 ∇tj ξ̂i = 39 40 41 For i = j , we have 43 44 31 For i = j , we have 40 41 30 l=1 37 27 29 K qr −s ∂ GD (ti , tl ) ξ̂lp−1 . + ∂tj 32 26 28 K qr −s−1 qr −s GD (ti , tl )ξ̂lp−1 ∇tj ξ̂l p−1 l=1 31 35 24 25 Let us now calculate M(t0 ): Therefore we first compute the derivatives of ξ̂ . It is easy to see that ξ̂ (t) is C 1 in t. We can calculate: 28 29 22 23 (4.22) 25 27 17 18 Our final assumption concerns the vector field F (t). 0 ): (H3) We assume that at t0 = (t10 , . . . , tK 22 26 13 42 K K qr qr −s ∂ qr p−1 −s−1 −s GD (ti , tl )ξ̂l ∇ti ξ̂l + GD (ti , tl ) ξ̂lp−1 p−1 ∂ti l=1 l=1 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 548 F:chipot506.tex; VTEX/Rita p. 62 J. Wei 1 = 2 K qr qr −s−1 −s qr −s GD (ti , tl )ξ̂lp−1 ∇ti ξ̂l + ∇ti GD (ti , ti )ξ̂ip−1 p−1 1 2 l=1 3 4 + 5 6 K ∇ti GD (ti , tl )ξ̂l qr p−1 −s 3 4 , 5 6 l=1 7 8 9 10 11 7 since ∂t∂ i GD (ti , ti ) = 2∇ti GD (ti , ti ). Note that ∇tj GD (ti , tj ) = (∇GD )T . 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 12 Therefore if we denote the matrix ∇ξ = (∇tj ξ̂i ) (4.23) 25 26 27 28 29 30 31 32 −1 qr qr qr −s−1 −s − s GD H p−1 ∇ξ(t) = I − (∇GD )T H p−1 p−1 K Fj (t) . +O 18 19 20 21 (4.24) j =1 24 We can compute M(t0 ) by using (4.24): qr −s M t0 = H−1 ∇ 2 GD H p−1 qr qr −s−1 − s ∇GD H p−1 + H−1 p−1 −1 qr qr qr −s−1 −s p−1 − s GD H × I− (∇GD )T H p−1 . p−1 37 38 25 26 27 28 29 30 (4.25) 41 42 43 44 45 31 32 33 The existence result is as follows 34 35 T HEOREM 4.3. (See [84].) Assume that assumptions (H1), (H2) and (H3) are satisfied. , Then for 1, problem (4.4) has an K-peaked solution which concentrates at t1 , . . . , tK or more precisely: 39 40 22 23 35 36 15 17 33 34 14 16 then we have 23 24 13 a (x) ∼ K j =1 q p−1 ξ h ti ∼ ξ ξ̂i0 , ti → ti0 , x − tj 0 q , ξ̂j p−1 w 36 37 38 39 (4.26) 40 41 42 i = 1, . . . , K, i = 1, . . . , K. (4.27) 43 44 (4.28) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 63 Existence and stability of spikes 1 2 3 4 549 R EMARK 4.3.2. In the case of symmetric K-peaked solutions, conditions (H2) and (H3) are not needed, as in the construction of solutions one can restrict the function space to the class of symmetric functions (see for example [67]). Note that for small (and not only in the limit → 0) the peaks are placed equidistantly. 5 6 7 10 11 12 13 14 15 16 17 18 19 22 23 24 25 28 31 32 33 36 37 38 39 18 (4.30) 12 13 14 15 16 17 19 22 23 where c(D) is some positive constant depending on D only and the function b(z) is given by 24 25 26 tanhα (z) , b(z) := cosh(z) (p − 1) α := . qr − (s + 1)(p − 1) (4.31) 27 28 29 The idea now is that we fix l and try to find another l¯ = l such that the following holds l b √ D ¯ l =b √ , D 30 31 0 < l < l¯ < 1, 32 (4.32) 33 34 ¯ This shows that if there exists a solution to (4.32), we which will imply that h(l) = h(l). ¯ In other words, we may match up solutions of (4.29) in may match up h(l) and h(l). different intervals. It turns out that for D small, (4.32) is always solvable. Now (4.32) has to be solved along with the following interval constraint: 35 36 37 38 39 40 K1 l + K2 l¯ = 1, K1 + K2 = K. (4.33) 41 42 For a solution l of (4.60) and (4.33) and j = 1, . . . , K we define 43 44 45 11 21 l , h(l) ∼ c(D)b √ D 42 43 10 20 40 41 7 Then the single interior symmetric spike solution is considered which was constructed by I. Takagi [67]. By some simple computations based on (4.6), we have that 34 35 6 9 29 30 4 8 26 27 3 In [70], by using matched asymptotic analysis, Ward and the first author constructed such solutions and studied their stability. We now summarize their main ideas. First (4.4) is solved in a small interval (−l, l): ⎧ p 2 a − a + ahq = 0 in (−l, l), ⎪ ⎪ ⎪ ⎨ ar Dh − h + hs = 0 in (−l, l), (4.29) ⎪ a(x) > 0, h(x) > 0 in (−l, l), ⎪ ⎪ ⎩ a (−l) = a (l) = h (−l) = h (l) = 0. 20 21 2 5 R EMARK 4.3.3. Our results here can be applied to give a rigorous proof for the existence and stability of K-peaked solutions consisting of peaks with different heights. 8 9 1 44 lj = l or lj = l¯ (4.34) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 550 1 2 3 4 0 t − t 0 = lj + lj −1 , j j +1 6 7 9 10 11 12 13 14 15 J. Wei where the number of j ’s such that lj = l is K1 (and consequently the number of j ’s such that lj = l¯ is K2 ). We call the small spike with lj = l type A and the large spike with lj = l¯ type B. Then we choose tj0 such that 5 8 F:chipot506.tex; VTEX/Rita p. 64 24 25 26 27 28 29 30 31 32 33 36 It is difficult to check (H1) directly. Instead we note that (See [34] and [70].) More precisely, we calculate ⎛ ⎞ .. .. . . 0 ⎟ ⎜ γ1 β1 0 ⎜ ⎟ . . ⎜β γ β ⎟ .. .. ⎜ 1 2 2 ⎟ ⎜ ⎟ ⎜ .. .. .. .. .. .. ⎟ √ ⎜ ⎟ . . . . . . −1 ⎟ GD = (aij ) = 2 D ⎜ ⎜. . ⎟ ⎜ .. .. β 0 ⎟ j −1 γj βj ⎜ ⎟ ⎜ ⎟ ⎜ .. .. .. .. .. .. ⎟ . . . .⎟ ⎜ . . ⎝ ⎠ . . 0 . . . . 0 βN −1 γN −1 GD is a tridiagonal matrix. 43 44 45 13 14 15 19 21 23 24 25 26 27 28 29 30 31 32 33 34 35 γj = coth(θj −1 + θj ) + coth(θj + θj +1 ), 42 12 22 where 38 41 11 20 γ1 = coth(θ1 + θ2 ) + tanh(θ1 ), 40 10 18 (4.36) 37 39 9 17 lj θj = √ . D 34 35 8 16 20 23 4 7 where 19 22 3 6 0 where t00 = −1, tK+1 = 1. By using matched asymptotics, we now have K1 type A and K2 type B peaks. This ends the short review of the ideas in [70]. Let us now use Theorem 4.3 to give a rigorous proof of results of [70]. In order to apply Theorem 4.3, we have to check the three assumptions (H1), (H2) and (H3). To this end, let us set √ ξ̂j0 = (2 D) tanh(θj ), j = 1, . . . , K, (4.35) 18 21 2 5 j = 0, . . . , K, 16 17 1 36 37 j = 2, . . . , K − 1, 38 39 γK = coth(θK−1 + θK ) + tanh(θK ), βj = − csch(θj + θj +1 ), 40 j = 1, . . . , N − 1 and θj was defined in (4.36). Note that √ aij = 2 D(βj δi(j −1) + γj δij + βj +1 δi(j +1) ). 41 42 43 44 (4.37) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 65 Existence and stability of spikes 1 551 Verifying (4.14) amounts to checking the following identity 1 2 2 N 3 4 aij ξ̂j0 = ξ̂i0 qr p−1 −s 3 (4.38) , j =1 5 5 6 7 8 9 10 11 12 13 6 which is an easy exercise. It remains to verify (H2) and (H3). To this end, we need to know the eigenvalues of B and M. In the same way as for the matrix GD , one can show that B −1 is a tridiagonal matrix. However, it is almost impossible to obtain an explicit formula for the eigenvalues. Numerical software for solving eigenvalue problems of large matrices is indispensable. Then (H2) has to be checked explicitly. Numerical computations in [70] do suggest that assumption (H3) is always satisfied. 14 15 16 17 18 19 20 21 22 23 24 T HEOREM 4.4. (See [84].) Suppose that for sufficiently small, there are solutions (a , h ) of (4.4) such that a (x) ∼ K q p−1 ξ j =1 (4.39) h ti ∼ ξ ξ̂i , 17 18 19 20 21 22 23 24 26 30 i = 1, . . . , K, (4.40) 31 32 where ξ is given by (4.9), 33 34 34 ξ̂i → ξ̂i0 > 0, 35 ti → ti0 , ti0 = tj0 , i = j, i, j = 1, . . . , K. (4.41) 36 35 36 Then necessarily, we have 38 0 ¯ li := ti0 − ti−1 ∈ {l, l}, 39 37 38 i = 1, . . . , K, (4.42) where = −1, l and l¯ satisfy (4.32) and (4.33) with K1 being the number of i’s for which li = l and K2 being the number of i’s for which li = l¯ (hence K1 + K2 = K). t00 43 45 13 29 32 44 12 28 31 42 11 27 and 30 41 10 25 x − tj q , ξ̂j p−1 w 28 40 9 16 A natural question is the following: Are all K-peaked solution generated by two types of peaks as the solutions which were constructed in [70]? Our next theorem gives an affirmative answer. It completely classifies all K-peaked solutions, provided that the K peaks are separated. 27 37 8 15 4.4. Classification of asymmetric patterns 26 33 7 14 25 29 4 39 40 41 42 43 Theorem 4.4 shows that an K-peaked solution must be generated by exactly two types ¯ This shows that the of peaks—type A with shorter length l and type B with larger length l. 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 552 1 2 3 F:chipot506.tex; VTEX/Rita p. 66 J. Wei solutions constructed in [70] (through a formal approach) exhaust all possible separated K-peaked solutions. In particular, it shows that there are at most 2K K-peaked solutions. If the assumptions (H1)–(H3) are met, then there are exactly 2K K-peaked solutions. 4 5 8 9 10 11 12 13 14 15 16 17 18 19 20 23 24 25 5 a = ξ â , 7 h = ξ ĥ 8 9 where ξ is defined at (4.9). Hence (â , ĥ ) satisfies ⎧ ⎨ 2 â − â + p â q ĥ = 0, 10 11 −1 < x < 1, 12 (4.43) ⎩ Dĥ − ĥ + c âr = 0, −1 < x < 1, s ĥ where c is defined as c = ( R w r )−1 . Now (4.39) and (4.40) imply that 15 16 17 18 K x − tj q , â ∼ ξ̂j p−1 w ĥ tj = ξ̂j . 19 (4.44) j =1 28 tj → tj0 , 22 23 24 j = 1, . . . , K. 25 26 We see that ĥ → h0 (x) where h0 (x) satisfies K 29 Dh0 − h0 + 30 h0 (−1) = h0 (1) = 0. 31 32 27 28 qr 0 p−1 j =1 (ξ̂j ) −s δ(x − tj0 ) = 0, −1 < x < 1, 29 (4.45) 32 In other words, we have 33 34 h0 (x) = 36 34 K 0 qr −s ξ̂j p−1 GD x, tj0 . (4.46) 39 37 Since h0 (tj0 ) = ξ̂j0 , j = 1, . . . , K, we have from (4.46) that (ξ̂10 , . . . , ξ̂K0 ) must satisfy the following identity: 40 41 42 43 38 39 40 K qr −s GD ti0 , tj0 ξ̂j0 p−1 = ξ̂i0 , j =1 44 45 35 36 j =1 37 38 30 31 33 35 20 21 Letting → 0, we assume that ξ̂j → ξ̂j0 , 13 14 26 27 3 6 q p−1 21 22 2 4 P ROOF OF T HEOREM 4.4. First we make the following scaling 6 7 1 41 i = 1, . . . , K. (4.47) 42 43 44 This is the same as (4.14). 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 67 Existence and stability of spikes 1 Define 1 2 3 ã,j = â χ 4 x − tj0 7 8 2 3 4 r̃0 5 6 553 5 where r̃0 is a very small number. Then ã,j is supported in the interval Ij = (−r̃0 + tj , r̃0 + tj ). We may choose r̃0 so small that Ii ∩ Ij = ∅ for i = j . Then 9 â = 11 K 11 12 13 13 Now we multiply the first equation in (4.43) by ã,j and integrate over (−1, 1). We obtain 16 0= 18 19 20 = 21 22 23 = −2 24 p â â p q Ij 25 q(p + 2) = p+1 26 27 ĥ 28 Ij − 18 19 20 21 p+1 ĥ + e.s.t. q+1 ĥ 22 q â 23 24 25 p+1 â ĥ q+1 ĥ + e.s.t. (4.48) ĥ (x) = c 33 30 31 1 −1 GD (x, z) âr 32 ĥs 33 34 34 and thus for x 36 ∈ Ij , 37 38 ĥ (x) = 39 44 45 35 36 K 37 qr −s GD x, tk ξ̂k p−1 + O() 38 39 k=1 40 43 27 29 32 42 26 28 By the equation for ĥ , we have that 31 41 15 17 29 35 14 16 p 1 aˆ p â ã,j q ã,j − q ĥ −1 hˆ p â â + e.s.t. −2 q Ij ĥ 17 30 8 10 ã,j + e.s.t. j =1 12 15 7 9 10 14 6 40 41 and 42 K qr −s ∇tj GD tj , tk ξ̂k p−1 + O(). Ĥ tj = k=1 43 (4.49) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 554 1 F:chipot506.tex; VTEX/Rita p. 68 J. Wei Substituting (4.49) into (4.48) and using (4.44), we obtain the following identity 1 2 3 2 K 4 7 8 9 10 qr −s tj , tk ξ̂k p−1 = o(1) 3 (4.50) k=1 5 6 ∇tj GD 5 6 and hence 7 K qr −s ∇t 0 GD tj0 , tk0 ξ̂k0 p−1 = 0, j 8 j = 1, . . . , K, (4.51) 13 14 15 16 17 18 19 20 21 22 11 which is the same as (4.21). Note that by the expression for h0 in (4.46), (4.51) is equivalent to the following h0 tj0 + + h0 tj0 − = 0, j = 1, . . . , K, (4.52) where h0 (tj0 +) is the right-hand derivative of h0 at tj0 and h0 (tj0 −) is the left-hand derivative of h0 at tj0 . On the other hand, from the equation for h0 , we have that qr −s D h0 tj0 + − h0 tj0 − = − ξ̂j0 p−1 , 25 26 27 28 29 30 31 32 33 34 35 36 37 j = 1, . . . , K. (4.53) 40 41 42 45 14 15 16 17 18 20 21 23 qr 1 0 p−1 −s h0 tj0 + = −h0 tj0 − = − ξ̂j < 0, 2D j = 1, . . . , K. (4.54) Since h0 satisfies Dh0 = h0 > 0 in each interval (tj0−1 , tj0 ), j = 2, . . . , K, we see that there exists a unique point sj −1 ∈ (tj0−1 , tj0 ) such that h0 (sj −1 ) = 0. Since h0 (−1) = 0, by using symmetry, we see that 24 25 26 27 28 29 30 sj −1 + sj = tj0 , 2 j = 1, . . . , K, (4.55) Dh0 − h0 + ξ̂j0 qr p−1 −s δ t − tj0 31 32 33 where we take s0 = −1, sK = 1. Let 2lj = sj − sj −1 , j = 1, . . . , K. Note that on each interval (−lj + tj0 , lj + tj0 ), h0 satisfies 34 35 36 =0 37 38 with Neumann boundary conditions at both ends. Thus from (4.6) it is easy to see that √ 0 qr −s−1 lj ξ̂j p−1 , j = 1, . . . , K, (4.56) = 2 D tanh √ D 43 44 13 22 Solving (4.52) and (4.53), we have that 38 39 12 19 23 24 9 10 k=1 11 12 4 h0 (lj ) = 39 40 41 42 43 ξ̂j0 l cosh( √j ) D . (4.57) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 69 Existence and stability of spikes 1 555 Since h0 is continuous on (−1, 1), we have 1 2 3 2 h0 (l1 ) = h0 (l2 ) = · · · = h0 (lK ). (4.58) 4 5 6 7 8 9 10 11 12 13 4 Using (4.56) and (4.57), we see that (4.58) is equivalent to l1 l2 lK b √ =b √ = ··· = b √ , D D D 5 6 (4.59) 16 17 18 19 20 where the function b was defined in (4.31). Suppose without loss of generality that l1 l2 , ¯ and that lj ∈ {l, l} ¯ for j = 2, . . . , K. then we take l1 = l and (4.59) implies that l2 ∈ {l, l} Thus l must satisfy (4.60) and (4.33). This finishes the proof of Theorem 4.4. 23 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 10 11 12 13 15 4.5. The stability of symmetric and asymmetric K-spikes 16 In this section, we present the stability of the K-peaked solutions constructed in Theorem 4.3. To this end, we need to study the following linearized eigenvalue problem L 24 25 9 14 21 22 7 8 14 15 3 φ ψ 2 φ − φ + p = 1 τ Dψ − ψ + r p−1 a q H ar−1 hs φ − q ψ , φ = λ , ψ ψ p a q+1 h ar hs+1 φ − s (4.60) and furthermore qr − s min σ > 1 p−1 σ ∈σ (B) (4.61) 20 22 23 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 (4.62) 41 42 and σ (M) ⊆ σ Re(σ ) > 0 , 19 24 T HEOREM 4.5. Let (a , h ) be the solutions constructed in Theorem 4.3. Assume that 1. (1) (Stability) If r = p + 1, < p < +∞ 18 21 where (a , h ) is the solution constructed in Theorem 4.3 and λ ∈ C—the set of complex numbers. We say that (a , h ) is linearly stable if the spectrum σ (L ) of L lies in the left half plane {λ ∈ C: Re(λ) < 0}. (a , h ) is called linearly unstable if there exists an eigenvalue λ of L with Re(λ ) > 0. (From now on, we use the notations linearly stable and linearly unstable as defined above.) r = 2, < p < 5 or 17 43 44 (4.63) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 556 1 2 3 4 J. Wei there exists τ0 > 0 such that (a , h ) is linearly stable for 0 τ < τ0 . (2) (Instability) If 5 6 7 8 9 10 11 F:chipot506.tex; VTEX/Rita p. 70 qr −s p−1 1 2 3 4 min σ < 1, σ ∈σ (B) (4.64) 6 7 there exists τ0 > 0 such that (a , h ) is linearly unstable for 0 τ < τ0 . (3) (Instability) If there exists σ ∈ σ (M), 8 9 10 Re(σ ) < 0, (4.65) 12 13 16 13 14 R EMARK 4.5.1. In the original Gierer–Meinhardt model, (p, q, r, s) = (2, 1, 2, 0) or (p, q, r, s) = (2, 4, 2, 0). This means that condition (4.61) is satisfied. 17 18 19 20 21 22 23 24 25 28 29 R EMARK 4.5.2. By Theorems 4.3 and 4.5, the existence and stability of K-peaked solutions are completely determined by the two matrices B and M. They are related to the asymptotic behavior of large eigenvalues which tend to a non-zero limit and small eigenvalues which tend to zero as → 0, respectively. The computations of these two matrices are by no means easy. We refer to [34] and [70] for exact computations and numerics. Combining the results here and the computations in [34], the stability of symmetric K-peaked solutions is completely characterized and the following result is established rigorously. r = p + 1, 1 < p < +∞ (4.66) 36 39 40 D < DK := √ K 2 (log( 1 , √ α + α + 1))2 45 22 23 24 25 27 28 29 31 35 (4.67) 36 37 where α is given by (4.31), then the symmetric K-peaked solution is linearly stable. (b) (Instability) If 38 39 40 41 D > DK , (4.68) 43 44 21 33 41 42 20 34 37 38 19 32 and 34 35 18 30 r = 2, 1 < p < 5 or 32 33 16 26 T HEOREM 4.6. (See [34,84].) Let (a,K , h,K ) be the symmetric K-peaked solutions constructed in [67]. Assume that 1. (a) (Stability) Assume that 0 < τ < τ0 for some τ0 small and that 30 31 15 17 26 27 11 12 then (a , h ) is linearly unstable for all τ > 0. 14 15 5 42 43 where DK is given by (4.67), then the symmetric K-peaked solution is linearly unstable for all τ > 0. 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 71 Existence and stability of spikes 1 2 3 557 The proof of Theorem 4.5 consists of two parts: we have to compute both small and large eigenvalues. For large eigenvalues, we will arrive at the following system of nonlocal eigenvalue problems (NLEPs) 4 5 −1 − qr(I + sB) 7 11 12 13 14 B R w r−1 Φ R w r −1 6 w = λΦ p (4.69) ⎛ φ1 ⎜ φ2 ⎜ Φ = ⎜. ⎝ .. 9 10 ⎞ 11 12 ⎟ K ⎟ ⎟ ∈ H 2 (R) . ⎠ 13 14 φK 15 16 18 19 20 21 22 16 By diagonalization, we may reduce it to K NLEPs of the form (3.17). Using the results of Theorem 3.7, we obtain the stability (or instability) of large eigenvalues. For the study of small eigenvalues, we need to expand the eigenfunction up to the order O( 2 ) term. This computation is quite involved. In the end, the matrix B and M will appear. A similar stability analysis for the Schnakenberg model has been carried out in [35]. 23 28 29 30 37 38 39 40 41 42 43 44 45 21 22 27 28 29 30 31 32 5.1. Bound states: spikes on polygons 33 34 36 20 26 Let us now consider the Gierer–Meinhardt system in a two-dimensional domain. The results are more complicated. To reduce the complexity and grasp the essential difficulties, we assume that (p, q, r, s) = (2, 1, 2, 0) in this section. We start with the bound states. 32 35 19 25 31 33 18 24 5. The full Gierer–Meinhardt system: Two-dimensional case 26 27 17 23 24 25 7 8 where B is given by (4.15) and 15 17 3 5 8 10 2 4 Φ − Φ + pw p−1 Φ 6 9 1 34 We first consider the case when Ω = R2 : ⎧ a2 ⎪ ⎨ a − a + h = 0, a > 0 ⎪ ⎩ h − σ 2 h + a 2 a(x), h(x) → 0 = 0, 35 36 in R2 , R2 , h > 0 in as |x| → +∞. 37 (5.1) As we will see, a notable feature of this ground-state problem in the plane is the presence of solutions with any prescribed number of bumps in the activator as the parameter σ gets smaller. These bumps are separated from each other at a distance O(| log log σ |) and approach a single universal profile given by the unique radial solution of (2.8). The multibump solutions correspond respectively to bumps arranged at the vertices of a k-regular 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 558 1 2 3 4 5 J. Wei polygon and at those of two concentric regular polygons. These arrangements with one extra bump at the origin are also considered. This unveils a new side of the rich and complex structure of the solution set of the Gierer–Meinhardt system in the plane and gives rise to a number of questions. Let us set 6 7 8 τσ = 1 k log 2π σ 9 10 11 12 13 14 15 16 F:chipot506.tex; VTEX/Rita p. 72 R2 w 2 (y) dy 19 20 21 22 25 26 27 (5.2) k lim τσ aσ (x) − w(x − ξi ) = 0, σ →0 32 33 34 35 36 37 38 39 42 43 10 11 12 15 16 17 uniformly in x ∈ R2 . Here the points ξi correspond to the vertices of a regular polygon centered at the origin, with sides of equal length lσ satisfying 1 1 . lσ = log log + O log log log σ σ 18 19 20 21 (5.4) 22 23 Finally, for each 1 j k we have 24 25 lim τσ hσ (ξj + y) − 1 = 0, 26 σ →0 27 28 uniformly on compact sets in y. 29 30 Our second result gives existence of a solution with bumps at vertices of two concentric polygons. T HEOREM 5.2. (See [17].) Let k 1 be a fixed positive integer. There exists σk > 0 such that, for each 0 < σ < σk , problem (5.1) admits a solution (a, h) with the following property: k w(x − ξi ) + w(x − ξi∗ ) = 0, lim τσ aσ (x) − σ →0 31 32 33 34 35 36 37 38 (5.5) i=1 39 40 uniformly in x ∈ R2 . Here the points ξi and ξi∗ are the vertices of two concentric regular polygons. They satisfy 44 45 8 14 (5.3) i=1 40 41 5 13 30 31 4 9 T HEOREM 5.1. (See [17].) Let k 1 be a fixed positive integer. There exists σk > 0 such that, for each 0 < σ < σk , problem (5.1) admits a solution (a, h) with the following property: 28 29 3 7 . 23 24 2 6 −1 17 18 1 41 42 43 44 ξj = ρ σ e 2j π k i , ξj∗ 2πj = ρσ∗ e k i , j = 1, . . . , k, 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 73 Existence and stability of spikes 1 559 where 1 2 2 1 1 ρσ = , log log + O log log log 2πi σ σ |1 − e k | 1 3 4 3 4 5 6 5 and ∗ ρσ = 1 + 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 30 31 8 9 12 11 The method employed in the proof of the above results consists of a Lyapunov–Schmidt type reduction. The basic idea of solving the second equation in (5.1) for h first and then working with a non-local elliptic PDE rather than directly with the system. Let T (a 2 ) be the unique solution of the equation h − σ 2 h + a 2 =0 h(x) → 0 a2 R2 , (5.6) a − a + T (a 2 ) 34 (5.7) 39 40 41 42 43 44 45 17 18 19 20 21 23 25 26 27 29 31 32 2 1 1 log log |ξj − ξi | 2 log log , 3 σ σ 33 for all i = j. 34 35 We look for solutions to (5.7) of the form 37 38 16 30 Fixing m points which satisfy the constraints 35 36 15 28 = 0. 32 33 14 24 for Equation (5.3) can be solved via sub-super-solution method. Solving the second equation for h in (5.1) we get h = T (a 2 ), which leads to the non-local PDE for a a2 13 22 in as |x| → +∞, ∈ L2 (R2 ). 28 29 7 T HEOREM 5.3. (See [17].) Let k 1 be given. Then there exists solutions which are exactly as those in Theorems 5.1 and 5.2 but with an additional bump at the origin. More precisely, with w(x) added to ki=1 w(x − ξi ) in (5.3) and added to ki=1 [w(x − ξi ) + w(x − ξi∗ )] in (5.5). 24 27 1 1 1 log log + O log log log . 2πi σ σ |1 − e k | 10 23 26 6 A similar assertion to (5.4) holds for hσ , around each of the ξi and the ξi∗ ’s. 22 25 1 a(x) = (W + φ), τσ where W = 36 K 37 w(x − ξj ). (5.8) j =1 By using finite-dimensional reduction method, we first solve an auxiliary problem ⎧ +φ)2 ∂W ⎨ (W + φ) − (W + φ) + (W = i,α ciα ∂ξ 1 2 i,α T ( τσ (W +φ) ) (5.9) ⎩ φ ∂W = 0, i = 1, . . . , m, α = 1, 2. R2 ∂ξi,α 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 560 1 2 F:chipot506.tex; VTEX/Rita p. 74 J. Wei Solutions satisfying the required conditions in Theorems 5.1–5.3 will be precisely those satisfying a non-linear system of equations of the form 3 4 5 6 7 8 9 10 13 ciα (ξ1 , ξ2 , . . . , ξm ) = 0, i = 1, . . . , m, α = 1, 2, 4 5 6 where for such a class of points the functions ciα satisfy 7 ∂ ciα (ξ1 , . . . , ξk ) = F |ξj − ξi | + iα , ∂ξiα 8 (5.10) 11 function F : R+ → R is of the form F (r) = 16 c7 log r log 1 σ 12 13 14 + c8 w(r), 15 16 17 18 17 c7 and c8 are universal constants and 19 20 21 iα = O 1 (log σ1 )1+γ 18 19 20 , 21 22 23 24 25 26 27 28 29 30 22 for some γ > 0. Although (5.10) does not have a variationalstructure, solutions of the problem ciα = 0 are close to critical points of the functional i=j F (|ξj − ξi |). In spite of the simple form of this functional, its critical points are highly degenerate because of the invariance under rotations and translations of the problem. Thus, to get solutions using degree theoretical arguments, we need to restrict ourselves to classes of points enjoying symmetry constraints. This is how Theorems 5.1–5.3 are established. On the other hand, we believe strongly that finer analysis may yield existence of more complex patterns, such as honey-comb patterns, or lattice patterns. 31 32 33 34 35 36 37 R EMARK 5.1.1. Similar method can also be used to prove Theorem 4.1. In that case, we have ∂ F1 |ξj − ξi | + O σ 1+γ , ci (ξ1 , . . . , ξk ) = ∂ξi i=j 44 45 25 26 27 28 29 30 32 33 34 36 37 38 function F1 : R+ → R is of the form 39 40 F1 (r) = c9 σ r + c10 w(r), 42 43 24 35 (5.11) 40 41 23 31 38 39 9 10 i=j 14 15 2 3 11 12 1 c9 and c10 are universal constants. It is easy to see that the critical points of ξj |) is non-degenerate (in the class of points with K j =1 ξj = 0). 41 42 i=j F1 (|ξi − 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 75 Existence and stability of spikes 1 561 5.2. Existence of symmetric K-spots 1 2 3 4 2 We look for solutions to the stationary GM on a two-dimensional domain with the following form 5 6 7 a (x) ∼ j =1 8 9 10 11 12 13 14 15 16 17 18 19 20 K x − Pj ξ,j w ξ,i = 1, →0 ξ,j 6 (5.12) 23 24 25 26 9 10 11 12 for i = j, (5.13) 29 30 31 32 33 34 35 36 37 38 39 40 41 42 such solutions are called symmetric K-spots. Otherwise, they are called asymmetric Kspots. In this section, we discuss the existence of symmetric K-spots. It turns out in twodimensional case, we have to discuss two cases: the strong coupling case, D ∼ O(1), and the weak coupling case, D 1. We first have the following existence result in the strong coupling case 45 15 16 17 18 19 20 21 T HEOREM 5.4. (See [86].) Let Ω ⊂ R 2 be a bounded smooth domain and D be a fixed positive constant. Let GD (x, y) be the Green function of −D + 1 in Ω (with Neumann boundary condition). Let HD (x, y) be the regular part of GD (x, y) and set hD (P ) = HD (P , P ). Set 22 23 24 25 26 27 FD (P1 , . . . , PK ) = K i=1 HD (Pi , Pi ) − 28 GD (Pj , Pl ). j =l Assume that (P1 , . . . , PK ) ∈ Ω K is a non-degenerate critical point of FD (P1 , . . . , PK ). Then for sufficiently small, problem (GM) has a steady state solution (a , h ) with the following properties: x−Pj 0 (1) a (x) = ξ ( K j =1 w( ) + o(1)) uniformly for x ∈ Ω̄, Pj → Pj , j = 1, . . . , K, as → 0, and w is the unique solution of the problem (2.8). (2) h (x) = ξ (1 + O( | log1 | )) uniformly for x ∈ Ω̄, where 1 + o(1)) 2 log 1 R2 w 2 . (3) ξ−1 = ( 2π R EMARK 5.2.1. Theorem 5.4 shows that interior peaks solutions are related to the Green function (contrast to shadow system case). Thus in the strong coupling case, the peaks are produced by a different mechanism. It seems that the equation for h controls everything. 43 44 13 14 27 28 7 8 21 22 4 5 where Pj are the locations of the K-spikes and ξ,j is the height of the spike at Pj . If all the heights are asymptotically equal, i.e. lim 3 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 R EMARK 5.2.2. In a general domain, the function FD (P) always has a global maximum point P0 in Ω × · · · × Ω. (A proof of this fact can be found in the Appendix of [86].) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 562 1 2 3 4 5 6 1 Next, we discuss the weak coupling case. We assume that lim→0 D = +∞. We first introduce a Green function G0 which we need to formulate our main results. Let G0 (x, ξ ) be the Green function given by 3 log ⎧ 1 ⎪ 0 (x, ξ ) − |Ω| + δξ (x) = 0 in Ω, ⎨ G Ω G0 (x, ξ ) dx = 0, ⎪ ⎩ ∂G0 (x,ξ ) =0 on ∂Ω ∂ν 8 9 10 11 13 J. Wei The proof of Theorem 5.4 depends on fine estimates in the finite-dimensional reduction: the major problem is to sum up the errors of powers in terms of 1 1 . 7 12 19 11 14 1 1 log − G0 (x, ξ ) 2π |x − ξ | (5.15) 17 18 19 20 21 22 with 23 24 24 Pi = (Pi,1 , Pi,2 ) 25 for i = 1, 2, . . . , K. 25 26 27 26 For P ∈ Ω K we define 27 28 28 29 F0 (P) = 30 31 K H0 (Pk , Pk ) − 29 G0 (Pi , Pj ) (5.16) i,j =1,...,K, i=j k=1 34 35 36 37 38 39 40 32 and 33 34 M0 (P) = ∇P2 F0 (P) . Here M0 (P) is a (2K) × (2K) matrix with components 1, 2 (recall that Pi,j is the j th component of Pi ). Set (5.17) 43 ∂ 2 F (P) ∂Pi,j ∂Pk,l , i, k = 1, . . . , K, j, l = 37 38 39 40 41 1 D = 2, β β 2 |Ω| 1 η := log . 2π 44 45 35 36 41 42 30 31 32 33 15 16 22 23 9 10 P = (P1 , P2 , . . . , PK ) 21 6 8 (5.14) be the regular part of G0 (x, ξ ). Denote P ∈ Ω K , where P is arranged such that 20 5 13 16 18 4 12 H0 (x, ξ ) = 15 2 7 and let 14 17 F:chipot506.tex; VTEX/Rita p. 76 (5.18) 42 43 44 Then D → +∞ is equivalent to β → 0. 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 77 Existence and stability of spikes 1 2 3 4 5 6 563 The stationary system for (GM) is the following system of elliptic equations: ⎧ a2 2 ⎪ ⎨ a − a + h = 0, a > 0 h − β 2 h + β 2 a 2 = 0, h > 0 ⎪ ⎩ ∂a ∂h ∂ν = ∂ν = 0 1 2 3 in Ω, in Ω, on ∂Ω. (5.19) 9 10 11 12 13 14 15 7 The following concerns the existence of symmetric K-peaked solutions in a twodimensional domain which generalizes the one-dimensional result Theorem 4.2. T HEOREM 5.5. (See [87].) Let P0 = (P10 , P20 , . . . , PK0 ) be a non-degenerate critical point of F0 (P) (defined by (5.16)). Moreover, we assume that the following technical condition holds 18 19 20 21 22 (5.20) →0 24 26 ξ = 27 28 29 ⎧1 ⎪ K ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎩ |Ω| 2 R2 w 2 (y) dy 1 |Ω| η 2 2 w 2 (y) dy R 1 |Ω| K+η0 2 2 w 2 (y) dy R 32 33 34 35 36 10 11 12 13 15 if η → ∞, 17 18 19 20 21 22 23 if η → 0, 24 (5.21) if η → η0 , 25 26 27 28 29 and 30 31 9 16 where η is defined by (5.18). Then for sufficiently small and D = β12 sufficiently large, problem (5.19) has a solution (a , h ) with the following properties: x−Pj (1) a (x) = ξ ( K j =1 w( ) + O(k(, β))) uniformly for x ∈ Ω̄. Here w is the unique solution of (2.8) and 23 25 8 14 if K > 1, then lim η = K, 16 17 5 6 7 8 4 30 k(, β) := ξ β . 2 2 (5.22) 31 32 (By (5.21), k(, β) = O(min{ 1 1 , β 2 }).) log Furthermore, Pj → Pj0 as → 0 for j 33 34 = 1, . . . , K. (2) h (x) = ξ (1 + O(k(, β))) uniformly for x ∈ Ω̄. 35 36 37 37 38 38 39 5.3. Existence of multiple asymmetric spots 39 40 41 42 40 Similar to the on dimensional case, there are also multiple asymmetric spots in a twodimensional domain. But the existence of such patterns is only restricted when 43 44 45 41 42 43 lim →0 D log 1 < +∞. (5.23) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 564 1 2 3 4 F:chipot506.tex; VTEX/Rita p. 78 J. Wei We first derive the algebraic equations for the heights (ξ,1 , . . . , ξ,K ). For β > 0 let Gβ (x, ξ ) be the Green function given by 5 6 1 2 3 Gβ − β 2 Gβ ∂Gβ ∂ν = 0 + δξ = 0 in Ω, 4 on ∂Ω. (5.24) 6 7 8 7 Recall that β 2 = and therefore β ∼ 1 D 9 10 11 12 13 14 )1 log 1 . Let G0 (x, ξ ) be the Green function defined β −2 + G0 (x, ξ ) + O β 2 Gβ (x, ξ ) = |Ω| 17 18 19 20 21 10 11 12 (5.25) 24 25 26 27 28 29 x − Pj , χ,Pj (x) = χ δ 32 33 34 35 x ∈ Ω, j = 1, . . . , Km, 38 39 40 41 42 (5.26) where χ is a smooth cut-off function which is equal to 1 in B1 (0) and equal to 0 in B2 (0). Let us assume the following ansatz for a multiple-spike solution (a , h ) of (GM): R2 \ 45 17 18 20 21 23 24 25 26 K x−Pi i=1 ξ,i w( )χ,Pi (x), a ∼ h (Pi ) ∼ ξ,i , 27 (5.27) 28 29 30 where w is the unique solution of (2.8), ξ,i , i = 1, . . . , K, are the heights of the peaks, to be determined later, and P = (P1 , . . . , PK ) are the locations of K peaks. Then we can make the following calculations, which can be made rigorous with error terms of the order O( 1 1 ) in H 2 (Ω). log 31 32 33 34 35 From the equation for h , 36 h − β 2 h + β 2 a2 = 0, h Pi = 37 38 39 we get, using (5.25), 43 44 16 22 36 37 14 19 30 31 13 15 in the operator norm of L2 Ω) → H 2 (Ω). (Note that the embedding of H 2 (Ω) into L∞ (Ω) is compact.) We define cut-off functions as follows: Let P ∈ Ω K . Introduce 22 23 8 9 in (5.14). In Section 2 of [87] a relation between G0 and Gβ is derived as follows 15 16 5 40 Ω Gβ Pi , ξ β 2 a2 (ξ ) dξ = Ω β −2 |Ω| 2 2 + G0 Pi , ξ + O β β 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 79 Existence and stability of spikes 1 × 2 K 2 ξ,j w2 ξ − Pj j =1 3 565 1 χ,Pj (ξ ) dξ 2 3 4 1 2 + β G0 Pi , ξ + O β = Ω |Ω| K ξ − Pj 2 χ,Pj (ξ ) dξ. × ξ,j w2 4 5 6 7 8 4 5 6 7 8 j =1 9 9 10 11 10 Thus 11 12 13 14 ξ,i 15 16 17 2 2 ξ − Pi 2 2 2 2 χ,Pi (ξ ) dξ = ξ,i w (y) dy + ξ,i β G0 Pi , ξ w |Ω| R2 Ω 1 2 2 2 + β G0 Pi , Pj ξ,j w 2 (y) dy + |Ω| R2 j =i 18 + 20 24 25 26 27 28 29 K 2 4 2 O β + O β 42 . ξ,j j =1 21 23 R2 = 2 G0 Pi , Pj 33 + 34 3 2 37 R2 ∂Pj,l = G0 Pi , Pj 36 19 (5.28) 20 R2 24 25 26 27 28 29 31 32 w (y)yl dy + O 4 2 R2 w (y) dy + O 4 . 2 38 33 34 35 36 37 38 (Note that we have set y = ξ −Pj and we have used the relation 42 R2 39 40 41 w 2 (y)yl dy = 0 44 45 17 30 w 2 (y) dy K ∂G (P , P ) 0 i j l=1 35 43 16 23 2 ξ − Pj χ,Pj (ξ ) dξ G0 Pi , ξ w Ω G0 Pi , y + Pj w 2 (y) dy + e.s.t. = 2 32 41 15 22 31 40 14 21 Here we have used that for j = i 30 39 13 18 19 22 12 42 43 44 which holds since w is radially symmetric.) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 566 1 J. Wei Using (5.15) in (5.28) gives 2 3 4 F:chipot506.tex; VTEX/Rita p. 80 ξ,i 2 2 = ξ,i |Ω| 5 1 2 R2 3 w 2 (y) dy 4 ξ − Pi 1 1 log − H0 Pi , ξ w 2 2π |P − ξ | Ω i 1 2 2 + + β 2 G0 Pi , Pj ξ,j w 2 (y) dy |Ω| R2 2 2 + ξ,i β 6 7 8 9 5 χ ,Pi (ξ ) dξ 7 8 9 j =i 10 11 + 12 13 K 2 ξ,j j =1 14 15 = 16 2 ξ,i 2 |Ω| 10 11 2 4 O β + O β 42 2 R2 w (y) dy 2 + ξ,i 12 13 β2 2 1 log 2π 14 15 2 R2 w (y) dy 16 2 1 2 β w 2 (y) log w 2 (y) dy + ξ,i 2 dy − 2 H0 Pi , Pi 2π |y| R2 R2 1 2 2 + w 2 (y) dy + β 2 G0 Pi , Pj ξ,j |Ω| R2 17 18 19 20 21 j =i 22 23 + 24 K (5.29) 28 ξ,i = K 2 ξ,j j =1 32 33 34 + 35 K 2 |Ω| 2 ξ,j O R2 2 w 2 (y) dy + ξ,i β2 2π 2 log 1 R2 30 w 2 (y) dy 31 32 (5.30) 45 34 35 36 37 Let us rescale 38 ξ,i = ξ ξ̂,i , where ξ = |Ω| . 2 R2 w 2 ξ,i = 39 (5.31) 40 41 42 Then from (5.30) we get 43 44 24 33 2 2 β . 41 42 22 28 38 40 21 27 j =1 36 39 20 29 30 37 19 26 Recall that H0 ∈ C 2 (Ω̄ × Ω). Considering only the leading terms in (5.29) we get following 29 31 18 25 26 27 17 23 2 4 2 O β + O β 42 . ξ,j j =1 25 6 η 1 + ξ 2 2 |Ω| |Ω| ,i 43 R2 w 2 (y) dy 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 81 Existence and stability of spikes 1 + 2 5 ξ̂,i → ξ̂i , 7 9 2 |Ω| R2 w 2 (y) dy + K 2 ξ,j O β 22 , 1 2 j =1 3 4 where η was introduced in (5.18). Assuming that 6 8 2 ξ,j j =i 3 4 567 η → η0 , 5 (5.32) ξ̂,i = 11 K 9 10 2 2 + ξ̂,i η0 , ξ̂,j i = 1, . . . , K, (5.33) j =1 12 15 13 which can be determined completely. In fact, let 16 14 15 16 ρ(t) = t − η0 t 2 . 17 (5.34) 18 19 21 ρ(ξ̂i ) = 22 K 19 20 ξ̂j2 , i = 1, . . . , K, (5.35) j =1 23 25 ρ(ξ̂i ) = ρ(ξ̂j ) for i = j. (5.36) 27 30 31 32 28 (ξ̂i − ξ̂j ) 1 − η0 (ξ̂i + ξ̂j ) = 0. 37 38 39 40 41 42 43 44 45 29 (5.37) ξ̂i − ξ̂j = 0 or 32 ξ̂i + ξ̂j = 1 . η0 33 (5.38) 34 35 The case of symmetric solutions (ξ̂i = ξ̂1 , i = 2, . . . , N ) has been studied in [86] and [87]. Let us now consider asymmetric solutions, i.e., we assume that there exists an i ∈ {2, . . . , N} such that ξ̂i = ξ̂1 . Without loss of generality, let us assume that 36 37 38 39 40 ξ̂2 = ξ̂1 , 41 42 which implies that ξ̂1 + ξ̂2 = 30 31 Hence for i = j we have 35 36 26 27 That is 33 34 22 24 which implies that 26 29 21 23 25 28 17 18 Then (5.33) is equivalent to 20 24 11 12 13 14 7 8 we obtain the following system of algebraic equations 10 6 43 1 . η0 44 (5.39) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 568 1 2 3 4 5 6 7 F:chipot506.tex; VTEX/Rita p. 82 J. Wei Let us calculate ξ̂j , j = 3, . . . , K. If ξ̂j = ξ̂1 , then by (5.38), ξ̂j + ξ̂1 = 1 η0 , which implies 1 that ξ̂j = ξ̂2 . Thus for j 3, we have either ξ̂j = ξ̂1 or ξ̂j = ξ̂2 . Let k1 be the number of ξ̂1 ’s in {ξ̂1 , . . . , ξ̂K } and k2 be the number of ξ̂2 ’s in {ξ̂1 , . . . , ξ̂K }. Then we have k1 1, k2 1, k1 + k2 = K. This gives 2 8 9 10 13 14 15 16 ξ̂1 − η0 ξ̂12 = K ξ̂j2 = k1 ξ̂12 + k2 ξ̂22 , 19 24 = k1 ξ̂12 + k2 1 − ξ̂1 η0 16 2 17 18 19 20 and therefore 21 22 (k1 + k2 + η0 )ξ̂12 2k2 + η0 k2 − ξ̂1 + 2 = 0. η0 η0 (5.42) 29 30 33 34 35 27 k2 (2k2 + η0 )2 4 2 (k1 + k2 + η0 ). η02 η0 28 (5.43) 38 39 40 41 42 43 44 45 29 30 31 The strict inequality of (5.43) is equivalent to * η0 > 2 k1 k2 . 32 33 (5.44) 36 37 24 26 Equation (5.42) has a solution if and only if 31 32 23 25 27 28 13 15 Substituting (5.41) into (5.40), we obtain ξ̂1 − η0 ξ̂12 10 14 25 26 7 12 (5.41) 22 23 6 11 1 − ξ̂1 . ξ̂2 = η0 20 21 5 9 (5.40) j =1 17 18 4 8 11 12 3 34 35 36 It is easy to see that if (5.44) holds, then there are two different solutions to (5.42) which are given by (ρ± , η± ). Therefore we arrive at the following conclusion. √ L EMMA 5.6. Let η0 2 k1 k2 . Then the solutions of (5.33) are given by (ξ̂1 , . . . , ξ̂N ) ∈ ({ρ± , η± })K√where the number of ρ± ’s is k1 and the number of η± ’s is k2 . If η0 > 2√k1 k2 , there exist two solutions (ρ± , η± ). If η0 = 2√k1 k2 , there exists one solution (ρ± , ρ± ). If η0 < 2 k1 k2 , there are no solutions (ρ± , ρ± ). 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 83 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 10 11 √ Let η0 > 2 k1 k2 where k1 + k2 = K, k1 , k2 1. By Lemma 5.6, there are two solutions to (5.33). In fact, we can solve ) ) 2k2 + η0 + η02 − 4k1 k2 2k2 + η0 − η02 − 4k1 k2 ρ+ = , ρ− = , (5.45) 2η0 (η0 + K) 2η0 (η0 + K) ) ) 2k1 + η0 − η02 − 4k1 k2 2k1 + η0 + η02 − 4k1 k2 , η− = . (5.46) η+ = 2η0 (η0 + K) 2η0 (η0 + K) Note that ρ+ + η+ = 13 14 16 17 21 22 26 For P ∈ ΩK 40 41 42 43 44 45 9 10 13 17 (5.48) 18 19 20 21 22 23 24 for i = 1, 2, . . . , K. 27 28 29 K H0 (Pk , Pk )ξ̂k4 − k=1 31 G0 (Pi , Pj )ξ̂i2 ξ̂j2 (5.49) i,j,=1,...,K, i=j 32 33 34 39 8 16 we define F̂0 (P) = 33 38 7 30 32 37 6 26 31 36 5 15 30 35 4 25 Pi = (Pi,1 , Pi,2 ) 28 3 14 with 27 29 (5.47) P = (P1 , P2 , . . . , PK ) 24 2 12 1 . η0 Then there are k2 η’s in (ξ̂1 , . . . , ξ̂K ). Let P = (P1 , . . . , PK ) ∈ Ω K , where P is arranged such that 23 25 ρ − + η− = ξ̂j ∈ {ρ, η}, and the number of ρ’s in (ξ̂1 , . . . , ξ̂K ) is k1 . 19 20 1 , η0 Let (ρ, η) = (ρ+ , η+ ) or (ρ, η) = (ρ− , η− ). We drop “±” if there is no confusion. K be such that Let (ξ̂1 , . . . , ξ̂K ) ∈ R+ 18 1 11 12 15 569 34 and 35 M̂0 (P) = ∇P2 F̃0 (P). (5.50) Then we have the following theorem, which is on the existence of asymmetric K-peaked solutions. T HEOREM 5.7. (See [88].) Let K 2 be a positive integer. Let k1 , k2 1 be two integers such that k1 + k2 = K. Let √ |Ω| 1 β 2 |Ω| β2 = , η = log , D 2π 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 570 1 F:chipot506.tex; VTEX/Rita p. 84 J. Wei √ where |Ω| denotes the area of Ω, Assume that η0 = lim→0 η > 2 k1 k2 , 1 2 2 η0 = K (T1) 3 3 4 5 4 and that 5 6 7 6 P0 = P10 , P20 , . . . , PK0 is a non-degenerate critical point of F̂0 (P) (T2) 7 8 9 10 11 12 13 8 (defined by (5.49)). Then for sufficiently small the stationary (GM) has a solution (a , h ) with the following properties: x−Pj 1 (1) a (x) = K j =1 ξ,j (w( ) + O( D )) uniformly for x ∈ Ω̄, where w is the unique solution of (2.8) and 14 15 16 19 20 21 ξ,j = ξ ξ̂,j , |Ω| ξ = 2 . R2 w 2 Further, (ξ̂,1 , . . . , ξ̂,K ) → (ξ̂1 , . . . , ξ̂K ) which is given by (5.48). (2) h (Pj ) = ξ,j (1 + D1 ) in H 2 (Ω), j = 1, . . . , K. (3) Pj → Pj0 as → 0 for j = 1, . . . , K. 28 29 30 31 32 33 36 37 38 39 40 41 19 20 21 25 Next we study the stability and instability of the symmetric K-peaked solutions constructed in Theorems 5.4 and 5.5. In the strong coupling case, it turns out all solutions are stable: T HEOREM 5.8. (See [86].) Suppose D = O(1). Let P0 and (a , h ) be defined as in Theorem 5.4. Then for and τ sufficiently small (a , h ) is stable if all eigenvalues of the matrix MD (P0 ) = (∇P20 FD (P0 )) are negative. (a , h ) is unstable if one of the eigenvalues of the matrix MD (P0 ) is positive. 26 27 28 29 30 31 32 33 34 In the weak coupling case, the stability of symmetric K-peaked solutions in a bounded two-dimensional domain can be summarized as follows. T HEOREM 5.9. (See [87].) Let P0 be a non-degenerate critical point of F0 (P) and for sufficiently small and D = β12 sufficiently large let (a , h ) be the K-peaked solutions constructed in Theorem 5.5 whose peaks approach Assume (5.20) holds and further that P0 . 35 36 37 38 39 40 41 42 (∗) 0 P is a non-degenerate local maximum point of F0 (P). 44 45 18 24 42 43 16 23 5.4. Stability of symmetric K-spots 34 35 13 22 25 27 12 17 23 26 11 15 (5.51) 22 24 10 14 17 18 9 43 44 Then we have 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 85 Existence and stability of spikes 1 Table 1 1 2 3 4 5 6 7 8 571 K = 1, τ K = 1, τ K = 1, τ K > 1, τ K > 1, τ K > 1, τ Case 1 Case 2 Case 3 (η0 < K) Case 3 (η0 > K) stable ? unstable unstable unstable unstable stable stable stable stable stable stable stable ? unstable unstable unstable unstable stable ? stable stable ? stable 3 small finite large small finite large 9 14 15 2πD |Ω| log 1 ). 2πD |Ω| log 1 ). 24 25 26 27 30 33 20 C ASE 3. η → η0 ∈ (0, +∞) (i.e., 2πD |Ω| ∼ 1 η0 log 1 ). 36 37 38 39 42 43 44 45 23 24 25 26 27 28 The statement of Theorem 5.9 is rather long. Let us therefore explain the results by the following remarks. 29 30 31 R EMARK 5.4.1. Assuming that condition (∗) holds, then for small the stability behavior of (a , h ) can be summarized in the following table: 32 33 34 R EMARK 5.4.2. The condition (∗) on the locations P0 arises in the study of small (o(1)) eigenvalues. For any bounded smooth domain Ω, the functional F0 (P), defined by (5.16), always admits a global maximum at some P0 ∈ Ω K . The proof of this fact is similar to the Appendix in [87]. We believe that in generic domains, this global maximum point P0 is non-degenerate. 40 41 21 22 If K > 1 and η0 < K, then (a , h ) is linearly unstable for any τ > 0. If η0 > K, then there exist 0 < τ2 τ3 such that (a , h ) is linearly stable for τ < τ2 and τ > τ3 . If K = 1, η0 < 1, then there exist 0 < τ4 τ5 such that (a , h ) is linearly stable for τ < τ4 and linearly unstable for τ > τ5 . 34 35 15 19 31 32 14 18 (a , h ) is linearly stable for any τ > 0. 28 29 13 17 22 23 8 16 C ASE 2. η → +∞ (i.e., 20 21 7 12 If K = 1 then there exists a unique τ1 > 0 such that for τ < τ1 , (a , h ) is linearly stable, while for τ > τ1 , (a , h ) is linearly unstable. If K > 1, (a , h ) is linearly unstable for any τ 0. 18 19 6 11 16 17 5 10 C ASE 1. η → 0 (i.e., 12 13 4 9 10 11 2 35 36 37 38 39 40 It is an interesting open question to numerically compute the critical points of F0 (P) and link them explicitly to the geometry of the domain Ω. We believe that for other types of critical points of F0 (P), such as saddle points, the solution constructed in Theorem 5.5 should be linearly unstable. We are not able to prove this at the moment, since the operator L is not self-adjoint. 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 572 1 2 F:chipot506.tex; VTEX/Rita p. 86 J. Wei R EMARK 5.4.3. Case 1 and Case 3 with η0 < K resemble the shadow system and Case 2 and Case 3 with η0 > K are similar to the strong coupling case. 3 4 5 6 9 10 13 14 15 18 19 22 − 23 2[(1 + η0 (1 + τ λ0 )) 24 wφi + j =i (K + η0 )(1 + τ λ0 ) R2 w 2 R2 R2 wφj ] 2η0 φ − φ + 2wφ − (K + η0 ) R2 w 2 30 31 37 40 (5.52) 43 44 45 15 17 18 19 25 27 R2 28 w(y)φ(y) dy w 2 = λφ, 29 (5.53) 30 31 32 and 2(K + η0 (1 + τ λ0 )) R2 wφ 2 w = λ0 φ, φ − φ + 2wφ − (K + η0 )(1 + τ λ0 ) R2 w 2 φ ∈ H 2 R2 , 33 34 35 36 (5.54) 37 38 where 0 < η0 < +∞ and 0 τ < +∞. Problem (5.53) is the same as (3.7). For problem (5.54), we have the following result 41 42 14 24 38 39 13 23 By diagonalization, we obtain two NELPs: 29 36 12 26 28 35 10 22 w 2 = λ0 φi , 26 34 9 21 i = 1, . . . , K. 25 33 8 20 φi − φi + 2wφi 21 32 6 16 The proof of Theorem 5.9 is again divided by two parts: large eigenvalues and small eigenvalues. For small eigenvalues, we relate them to the functional F (P). For large eigenvalues, we obtain a system of NLEPs: 20 27 5 11 R EMARK 5.4.5. Roughly speaking, assuming that condition (∗) holds and that τ is small, |Ω| log 1 is the critical threshold for the asymptotic behavior then for 1, DK () = 2πK of the diffusion coefficient of the inhibitor which determines the stability of K-peaked solutions. 16 17 4 7 R EMARK 5.4.4. We conjecture that in Case 3, τ2 = τ3 . This will imply that for any τ 0 and η0 > K, multiple spikes are stable, provided condition (∗) is satisfied. (It is possible to obtain explicit values for τ2 and τ3 .) 11 12 2 3 From Case 2 and Case 3 of Theorem 5.9, we see that for multiple spikes (K > 1 ) large τ may increase stability, provided that η0 > K. This is a new phenomenon in R 2 . It is known that in R 1 , large τ implies linear instability for multiple spikes [8,34,59,60]. 7 8 1 39 40 41 T HEOREM 5.10. (1) If η0 < K, then for τ small problem (5.54) is stable while for τ large it is unstable. (2) If η0 > K, then there exists 0 < τ2 τ3 such that problem (5.54) is stable for τ < τ2 or τ > τ3 . 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 87 Existence and stability of spikes 1 2 3 4 573 P ROOF. Let us set 1 2 2(K + η0 (1 + τ λ)) . f (τ λ) = (K + η0 )(1 + τ λ) (5.55) 4 5 6 5 We note that 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 lim f (τ λ) = τ λ→+∞ R2 26 29 30 31 32 33 34 35 36 37 38 8 9 |∇φ|2 + φ 2 − 2wφ 2 + 2 R2 wφ 2 3 2w − R 2 2 wφ 0, ( R2 w ) R2 25 28 7 2η0 =: f∞ . K + η0 If η0 < K, then by Theorem 3.12(2), problem (3.52) with μ = f∞ has a positive eigenvalue α1 . Now by perturbation arguments (similar to those in [8]), for τ large, problem (5.54) has an eigenvalue near α1 > 0. This implies that for τ large, problem (5.54) is unstable. Now we show that problem (5.54) has no non-zero eigenvalues with non-negative real part, provided that either τ is small or η0 > K and τ is large. (It is immediately seen that 0 > 1 as τ λ → +∞ if η0 > K. Then Theorem f (τ λ) → 2 as τ λ → 0 and f (τ λ) → η02η+K 3.12 should apply. The problem is that we do not have control on τ λ. Here we provide a rigorous proof.) We apply the following inequality (Lemma 3.8(1)): for any (real-valued function) φ ∈ Hr2 (R 2 ), we have 24 27 R2 R2 w2 w2 φ = −λ0 40 44 45 12 13 14 15 16 17 18 19 20 22 23 25 26 27 28 29 30 (5.57) 2 R2 wφ |φ| − f (τ λ0 ) w 2 φ̄. 2 R2 R2 R2 w 31 32 φ—and integrating over R 2 , we obtain 33 34 35 36 37 38 (5.58) 39 40 41 43 11 24 Multiplying (5.57) by φ̄—the conjugate function of that |∇φ|2 + |φ|2 − 2w|φ|2 R2 10 21 (5.56) where equality holds if√and only if φ is a multiple of w. √ Now let λ0 = λR + −1λI , φ = φR + −1φI satisfy (5.54). Then we have 2 wφ L0 φ − f (τ λ0 ) R 2 w 2 = λ0 φ. R2 w 39 42 3 41 Multiplying (5.57) by w and integrating over R2 , we obtain that 3 2w 2 w φ = λ0 + f (τ λ0 ) R 2 wφ. R2 R2 R2 w 42 43 (5.59) 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 574 1 2 3 4 5 6 7 8 9 J. Wei Taking the conjugate of (5.59) we have 3 2w 2 R ¯ w φ̄ = λ0 + f (τ λ̄0 ) w φ̄. 2 R2 R2 R2 w 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 1 2 (5.60) R2 = −λ0 3 | 2 2w 2 wφ| 2 R R |φ| − f (τ λ0 ) λ̄0 + f (τ λ̄0 ) . 2 2 R2 R2 w R2 w 7 8 9 10 (5.61) 35 36 37 38 41 42 43 44 45 13 14 15 16 17 18 19 20 21 22 23 24 25 (5.62) 26 27 By the usual Pohozaev’s identity for (2.8) (multiplying (2.8) by y ·∇w(y) and integrating by parts), we obtain that 3 w3 = w2 . (5.63) 2 2 2 R R 28 29 30 31 32 33 Substituting (5.63) and the expression (5.55) for f (τ λ) into (5.62), we have 34 35 2 3 η0 + K + (η0 − K)τ λ + Re (η0 + K)(1 + τ λ̄0 ) (η0 + K)λ̄0 2 + (η0 − K)τ |λ0 |2 0 36 37 38 39 40 11 12 33 34 4 6 We just need to consider the real part of (5.61). Now applying the inequality (5.56) and using (5.60) we arrive at 3 R2 w −λR Re f (τ λ0 ) λ̄0 + f (τ λ̄0 ) 2 R2 w 3 3 2w 2w − 2 Re λ̄0 + f (τ λ̄0 ) R 2 + R 2 , R2 w R2 w √ where we recall λ0 = λR + −1λI with λR , λI ∈ R. Assuming that λR 0, then we have 3 2w f (τ λ0 ) − 12 + Re λ̄0 f (τ λ0 ) − 1 0. R 2 R2 w 3 5 Substituting (5.60) into (5.58), we have that |∇φ|2 + |φ|2 − 2w|φ|2 10 11 F:chipot506.tex; VTEX/Rita p. 88 39 which is equivalent to 3 (1 + μ0 τ λR )2 + λR + μ0 τ + τ + μ0 τ 2 |λ0 |2 λR 2 3 2 2 μ τ + μ0 τ − τ λ2I 0 + 2 0 40 41 42 43 44 (5.64) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 89 Existence and stability of spikes 1 2 575 where we have introduced μ0 := ηη00 −K +K . If η0 > K (i.e., μ0 > 0) and τ is large, then 1 2 3 3 3 2 2 μ τ + μ0 τ − τ 0. 2 0 4 5 (5.65) 6 7 8 9 λI 12 13 8 9 2 wφ 2 2 R |φ| = Im −f (τ λ0 ) w φ̄ . 2 R2 R2 R2 w 10 11 12 13 14 Hence 15 17 7 11 16 15 + 2 |λI | f (τ λ0 ) R R2 18 w4 w2 16 C (5.66) 21 19 where C is independent of λ0 . Substituting (5.66) into (5.64), we see that (5.64) cannot hold for λR 0, if τ is small. 22 27 24 25 Finally we study the stability or instability of the asymmetric K-peaked solutions constructed in Theorem 5.7. 28 29 30 33 34 35 36 37 T HEOREM 5.11. Let (a , h ) be the K-peaked solutions constructed in Theorem 5.7 for sufficiently small, whose peaks are located near P0 . Further assume that (∗) P is a non-degenerate local maximum point of F̂ (P). * 34 35 36 37 38 2 k1 k2 < η0 < K (5.67) 39 40 and 41 42 k1 > k 2 , (ρ, η) = (ρ+ , η+ ). 44 45 30 33 Then we have: (a) (Stability) Assume that 42 43 29 32 0 40 41 27 31 38 39 26 28 31 32 21 23 5.5. Stability of asymmetric K-spots 25 26 20 22 23 24 17 18 19 20 5 6 So (5.64) does not hold for λR 0. To consider the case when τ is small, we have to obtain an upper bound for λI . From (5.58), we have 10 14 4 43 44 Then, for τ small enough, (a , h ) is stable. 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 576 1 2 F:chipot506.tex; VTEX/Rita p. 90 J. Wei (b) (Instability) Assume that either 1 2 3 3 η0 > K 4 4 5 6 5 or 6 7 8 7 τ is large enough. 8 9 10 9 Then (a , h ) is linearly unstable. 10 11 12 13 11 A consequence of Theorem 5.11 is stable asymmetric patterns can exist in a twodimensional domain for a very narrow range of D, namely for 14 15 16 1 log 2πK √ √ D 1 |Ω| |Ω| < < log √ |Ω| 4π k1 k2 19 (5.68) 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 18 19 21 6. High-dimensional case: N 3 22 23 25 16 20 21 24 15 17 and small enough, where k1 and k2 are two integers satisfying k1 + k2 = K, k1 1, k2 1. In most cases, asymmetric patterns are unstable. 20 22 13 14 17 18 12 23 When N 3, there are very few results on the full Gierer–Meinhardt system. The difference between N 3 and N 2 lies on the behavior of the Green function: when N 2, the Green function is locally constant (when N = 2, it is locally ∞). The limiting problem is still a single equation (2.8). But when N 3, the Green function is like |x−y|1 N−2 . The limiting problem when N 3 becomes ⎧ ap ⎪ ⎨ a − a + hq = 0 r h + ahs = 0 ⎪ ⎩ a, h > 0, a, h → 0 25 26 27 28 29 in RN , in RN , as |y| → +∞. 30 (6.1) 31 32 33 Problem (6.1) seems out of reach at this moment. We believe that there should a radially symmetric solution to (6.1) which is also stable. As far as the author knows, the only result in higher-dimensional case is the existence of radially symmetric layer solutions [62]. Let Ω = BR be a ball of radius R in RN . By scaling, we may take D = 1 and obtain formally the following elliptic system ⎧ p 2 a − a + ahq = 0 ⎪ ⎪ ⎪ ⎨ m h − h + ahs = 0 ⎪ vsa > 0, h > 0 ⎪ ⎪ ⎩ ∂a ∂a ∂ν = ∂ν = 0 24 34 35 36 37 38 39 40 41 in BR , in BR in BR , on BR , 42 (6.2) 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 91 Existence and stability of spikes 1 577 where (p, q, m, s) satisfies 1 2 3 4 2 p > 1, q > 0, m > 0, s 0, qm > 1. (p − 1)(s + 1) (6.3) 5 6 7 8 9 10 13 J (0) = 0, J1 (0) = 1, J1 > 0. (6.4) 18 21 22 N −1 + J2 − J2 + δ0 = 0, r J2 > 0, J2 (+∞) = 0, (6.5) 27 28 J1 (r) = c1 r 2−N 2 Iν (r), J2 (r) = c2 r 2−N 2 Kν (r), J1 = 31 32 33 sinh r , r J2 (r) = J2,R (r) = J2 (r) − 36 37 38 39 40 41 N −2 2 44 45 13 17 18 e−r 4πr 20 21 22 24 25 26 27 28 29 (6.7) . 32 33 34 J2 (R) J1 (r) J1 (R) (6.8) 35 36 37 and a Green function GR (r; r ) N −1 GR − GR + δr = 0, GR + r 30 31 38 39 GR (0; r ) = 0, GR (R; r ) = 0. (6.9) 40 41 42 43 12 23 (6.6) Let w(y) be the unique solution for ODE 2.103. Let R > 0 be a fixed constant. We define 34 35 ν= where c1 , c2 are two positive constants and Iν , Kν are modified Bessel functions of order ν. In the case of N = 3, J1 , J2 can be computed explicitly: 29 30 10 19 where δ0 is the Dirac measure at 0. The functions J1 (r) and J2 (r) can be written in terms of modified Bessel’s functions. In fact 25 26 9 16 J2 23 24 8 15 19 20 7 14 The second function, called J2 (r), satisfies 16 17 6 11 N −1 J1 − J1 = 0, J1 + r 14 15 4 5 (The case of the whole RN is also included here, by taking R = +∞.) Note that in (6.2), we have replaced a r by a m since we will use r = |x| to denote the radial variable. We first define two functions, to be used later: let J1 (r) be the radially symmetric solutions of the following problem 11 12 3 42 Note that J2,R (R) = 0, 43 lim J2,R (r) = J2 (r). R→+∞ 44 (6.10) 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 578 1 2 3 4 F:chipot506.tex; VTEX/Rita p. 92 J. Wei For t ∈ (0, R), set 1 2 (N − 1)(p − 1) J1 (t) J2,R (t) MR (t) := + + . qt J1 (t) J2,R (t) 3 (6.11) 5 6 7 5 When R = +∞, J2,+∞ (r) = J2 (r). We denote G+∞ M(t). That is, 8 9 10 G(r; r ) = c0 (r )N −1 (r; r ) as G(r; r ) and M+∞ (t) as 13 14 15 8 J2 (r )J1 (r), J1 (r )J2 (r), for r < r , for r > r , 9 (6.12) 18 12 (6.13) 15 Then we have the following existence result on layered solutions. 16 T HEOREM 6.1. (See [62].) Let N 2. Assume that there exist two radii 0 < r1 < r2 < R such that 23 24 25 26 27 28 29 30 31 32 33 34 35 36 MR (r1 )MR (r2 ) < 0. (6.14) Then for sufficiently small, problem (6.2) has a solution (a,R , h,R ) with the following properties: (1) a,R , h,R are radially symmetric, q p−1 (2) a,R (r) = ξ,R w( r−t )(1 + o(1)), (3) a,R (r) = ξ,R (GR (t ; t ))−1 GR (r; t )(1 + o(1)), where GR (r; t ) satisfies (6.9), ξ,R is defined by the following (1+s)(p−1)−qm qm m ξ,R = w GR (t ; t ) 41 42 43 44 45 20 R 22 23 24 25 26 27 28 29 30 (6.15) 31 32 and t ∈ (r1 , r2 ) satisfies lim→0 MR (t ) = 0. 33 34 It remains to check condition (6.14), which can be verified numerically. Under some conditions on p, q, we can obtain the following corollary. 35 36 37 38 C OROLLARY 6.2. Assume that the following condition holds: 39 40 18 21 37 38 17 19 21 22 13 14 19 20 10 11 (N − 1)(p − 1) J1 (t) J2 (t) M(t) := + + . qt J1 (t) J2 (t) 16 17 6 7 11 12 4 39 (N − 2)q + 1 < p < q + 1. N −1 40 (6.16) Then there exists an R0 > 0 such that for R > R0 and sufficiently small, problem (6.2) i , hi ) concentrating on sphere {r = t } with has two radially symmetric solutions (a,R i ,R MR (ti ) = 0, i = 1, 2, and 0 < t1 < t2 < R, i = 1, 2. 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 93 Existence and stability of spikes 1 2 3 4 5 6 7 8 579 We remark that Corollary 6.2 is the first rigorous result on the existence to (6.2) of positive solutions in dimension N 3. Next we consider the existence of bound states. That is, we consider the following elliptic system in RN : ⎧ ap 2 N ⎪ ⎨ a − a + hq = 0 in R , m in RN , h − h + ahs = 0 ⎪ ⎩ a, h > 0, a, h → 0 as |x| → +∞. 13 14 6 (6.17) 9 10 We have the following result. 11 T HEOREM 6.3. (See [62].) Let N 2. Assume that there exist two radii 0 < r1 < r2 < +∞ such that 19 20 21 22 23 24 25 26 27 28 29 (6.18) Then for sufficiently small, problem (6.17) has a solution (a , h ) with the following properties: (1) a , h are radially symmetric, q p−1 (2) a (r) = ξ w( r−r )(1 + o(1)), (3) h (r) = ξ (G(r ; r ))−1 G(r; r )(1 + o(1)), where ξ is defined at the following (1+s)(p−1)−qm qm m ξ = w G(r ; r ) 34 35 36 39 40 41 42 43 44 45 16 R 18 19 20 21 22 23 25 (6.19) and r ∈ (r1 , r2 ) satisfying lim→0 M(r ) = 0. 26 27 28 29 30 Similarly we have the following corollary. 31 32 C OROLLARY 6.4. Assume that N 2 and that the condition (6.16) holds. Then for sufficiently small, problem (6.2) has a radially symmetric bound state solution (a , h ) which concentrates on a sphere {r = t0 } where M(t0 ) = 0. 37 38 14 24 32 33 13 17 30 31 12 15 M(r1 )M(r2 ) < 0. 17 18 7 8 15 16 3 5 11 12 2 4 9 10 1 33 34 35 36 37 By using the same method, it is not difficult to generalize the results of Theorem 6.1 to other symmetric domains, such as annulus or the exterior of a ball. We omit the details. Several interesting questions are left open. First, can multiple layered solutions to (6.2) exist? Second, it would be an interesting question to study the stability of these “ring-like” solutions. Numerical computations in two dimension indicate that the “ring-like” solutions constructed in Theorem 6.1 are unstable and will break into several spots due to angular fluctuations. Third, if we vary R from 0 to +∞, what is the relation between the layered solution constructed in [52] for the single equation (2.4) and the solutions in Theorem 6.1? 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 580 1 F:chipot506.tex; VTEX/Rita p. 94 J. Wei 7. Conclusions and remarks 1 2 3 4 5 6 7 8 9 10 2 In this chapter, I have surveyed the most recent results on the study of Gierer–Meinhardt system. First, we consider the case D = +∞. In this case, the state-state problem becomes a singularly perturbed elliptic Neumann problem (2.4). Using the LEM, we established various existence results on concentrating solutions. In particular, Theorem 2.5 gives a lower bound on the number of solutions to (2.4). Several interesting questions are associated with (2.4). First, is there a lower bound on the number of boundary spikes? What is the optimal bound on the number of solutions to (2.4)? The followings are just some related conjectures 11 12 13 14 15 16 17 20 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 6 7 8 9 10 12 13 15 l(N −1) 16 17 number of boundary spikes to (2.4). 18 C ONJECTURE 2. Suppose the distance function d(P , ∂Ω) has l local maximum points. Then there is at least 19 20 21 C N l 22 23 24 number of interior spikes to (2.4). 25 C ONJECTURE 3. Suppose we have the energy bound J [u ] for some m N . Assume that the concentration set Γ = {u > 12 } is connected. Then the limiting set Γ = lim→0 Γ has Hausdorff dimension N − m. 26 Second, we consider the stability of spike solutions to the shadow system (2.2). By studying both small and large eigenvalues, we have completely characterized the stability (or instability) in the case of r = 2, 1 < p < 1 + N4 or r = p + 1. The study of the NLEP (3.52) is not complete yet. Many interesting questions are still open: the case of general r, the case of large τ , the uniqueness of Hopf bifurcation, etc. The non-linear metastability of interior spike solutions is studied in [6]. The stability of boundary spikes is studied in [32], through a formal approach. It can be proved that when D > D0 () 1, the full Gierer– Meinhardt system converges to the shadow system [59,60,77,78]. However, the critical threshold D0 () seems unknown. Third, we consider the one- and two-dimensional Gierer–Meinhardt systems. For steady states, we established the existence of symmetric and asymmetric K-peaked spikes. In 1D, the bifurcation of asymmetric K-spikes occur when D < DK . In 2D, the bifurcation of asymmetric K-spikes occur when D ∼ log 1 . We also obtain critical thresholds for the stability of K-peaked solutions: If 1 there are stability thresholds 30 C m 44 45 5 14 C 21 22 4 11 C ONJECTURE 1. Suppose the mean curvature function H (P ) has l local minimum points. Then there is at least 18 19 3 27 28 29 31 32 33 34 35 36 37 38 39 40 41 42 43 44 D1 () > D2 () > D3 () > · · · > DK () > · · · 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 95 Existence and stability of spikes 1 581 such that if 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 2 DK () lim >1 →0 D 3 4 5 then the K-peaked solution is stable, and if 6 7 DK () <1 →0 D lim 8 9 In 2D, 10 the critical threshold is 2πK . In 1D, the small eigenvalues determine the critical thresholds, while in 2D, the large eigenvalues give the critical thresholds. An interesting question is to obtain the next order term in the critical threshold for 2D (which should be O(1) and location-dependent). The dynamics of multiple spikes in 1D and 2D is completely open. In 1D, the dynamical equation for the positions of the spikes is a system of algebraic-differential-equations (ADE). A matched asymptotic analysis is given in [33]. In 2D, the dynamics of two well-separated spots is studied in [20] and it is shown that the two spots will repel each other, provided that the initial distance between the two spots is large enough. In a general two-dimensional domain, the dynamics of multiple spots should be governed by ∇FD (P) or ∇F0 (P). Finally, it is almost completely open as regards to three-dimensional Gierer–Meinhardt system. The main difficulty is the study of the coupled system (6.1) which requires some new insights. A layered bound state is constructed, but most likely it is unstable. An interesting question is to generalize Theorem 6.1 to general domains. Although the analysis in this paper was carried out for the Gierer–Meinhardt system, the results can certainly be generalized to a much wide class of non-local reaction diffusion systems that have localized spike solutions. 12 then the K-peaked solution is unstable. In 1D, the critical threshold is DK ∼ √ log |Ω| 1 . K2 29 34 35 36 37 38 45 17 18 19 20 21 22 23 24 25 26 27 28 31 33 34 35 36 37 38 40 References 42 44 16 39 40 43 15 32 I would like to thank Professor W.-M. Ni for introducing me to this subject and for his hearty encouragements and many valuable discussions. I also like to thank many of my collaborators for their patience and sharing with me their wonderful ideas: Professors H. Berestycki, S.-I. Ei, EN Dancer, M. del Pino, F.H. Lin, C. Gui, M. Kowalczyk, T. Kolokolnikov, D. Iron, W.-M. Ni, M. Ward, M. Winter, S. Yan and many others. The research is supported by an Earmarked Grant from RGC of HK. 39 41 14 30 Acknowledgments 32 33 13 29 30 31 11 41 42 [1] Adimurthi and G. Mancini, Geometry and topology of the boundary in the critical Neumann problem, J. Reine Angew. Math. 456 (1994), 1–18. [2] Adimurthi, F. Pacella and S.L. Yadava, Interaction between the geometry of the boundary and positive solutions of a semilinear Neumann problem with critical nonlinearity, J. Funct. Anal. 113 (1993), 318–350. 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 582 1 2 3 4 5 6 7 8 9 10 11 12 13 <uncited> 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 F:chipot506.tex; VTEX/Rita p. 96 J. Wei [3] H. Berestycki and J. Wei, On singular perturbation problems with Robin boundary condition, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (5) II (2003), 199–230. [4] P. Bates, E.N. Dancer and J. Shi, Multi-spike stationary solutions of the Cahn–Hilliard equation in higherdimension and instability, Adv. Diff. Eqns. 4 (1999), 1–69. [5] R.G. Casten and C.J. Holland, Instability results for a reaction diffusion equation with Neumann boundary conditions, J. Diff. Eqns. 27 (1978), 266–273. [6] X. Chen and M. Kowalczyk, Slow dynamics of interior spikes in the shadow Gierer–Meinhardt system, SIAM J. Math. Anal. 33 (1) (2001), 172–193. [7] X. Chen, M. del Pino and M. Kowalczyk, The Gierer–Meinhardt system: the breaking of homoclinics and multi-bump ground states, Comm. Contemp. Math. 3 (3) (2001), 419–439. [8] E.N. Dancer, On stability and Hopf bifurcations for chemotaxis systems, Methods Appl. Anal. 8 (2) (2001), 245–256. [9] E.N. Dancer and S. Yan, Multipeak solutions for a singular perturbed Neumann problem, Pacific J. Math. 189 (1999), 241–262. [10] A. Dillon, P.K. Maini and H.G. Othmer, Pattern formation in generalized Turing systems, I. Steady-state patterns in systems with mixed boundary conditions, J. Math. Biol. 32 (1994), 345–393. [11] A. Doelman, R.A. Gardner and T.J. Kaper, Large stable pulse solutions in reaction–diffusion equations, Indiana Univ. Math. J. 49 (4) (2000). [12] A. Doelman, T.J. Kaper and H. van der Ploeg, Spatially periodic and aperiodic multi-pulse patterns in the one-dimensional Gierer–Meinhardt equation, Methods Appl. Anal. 8 (3) (2001), 387–414. [13] M. Del Pino, P. Felmer and M. Musso, Two-bubble solutions in the super-critical Bahri–Coron’s problem, Calc. Var. Part. Diff. Equ. 16 (2003) 113–145. [14] M. del Pino, M. Musso and A. Pistoia, Super-critical boundary bubbling in a semilinear Neumann problem, Ann. Inst. H. Poincaré Anal. Non Linéaire 22 (1) (2005), 45–82. [15] M. del Pino, J. Dolbeault and M. Musso, The Brezis–Nirenberg problem near criticality in dimension 3, J. Math. Pures Appl. (9) 83 (12) (2004), 1405–1456. [16] M. del Pino, P. Felmer and J. Wei, On the role of mean curvature in some singularly perturbed Neumann problems, SIAM J. Math. Anal. 31 (1999), 63–79. [17] M. del Pino, M. Kowalczyk and J. Wei, Multi-bump ground states of the Gierer–Meinhardt system in R 2 , Ann. Non Linéaire Ann. Inst. H. Poincaré 20 (2003), 53–85. [18] M. del Pino, M. Kowalczyk and J. Wei, Concentration on curve for nonlinear Schrödinger equation, Comm. Pure Appl. Math. (2006), in press. [19] M. del, Pino, M. Kowalczyk and J. Wei, The Toda system and clustering interfaces in the Allen–Cahn equation, preprint, 2006. [20] S.-I. Ei and J. Wei, Dynamics of interior spike solutions, Japan J. Industr. Appl. Math. 19 (2002), 181–226. [21] A. Gierer and H. Meinhardt, A theory of biological pattern formation, Kybernetik (Berlin) 12 (1972), 30–39. [22] B. Gidas, W.M. Ni and L. Nirenberg, Symmetry of positive solutions of nonlinear elliptic equations in R N , Mathematical Analysis and Applications, Part A, L. Nachbin, ed., Adv. Math. Suppl. Stud., vol. 7, Academic Press, New York (1981), 369–402. [23] C. Gui and N. Ghoussoub, Multi-peak solutions for a semilinear Neumann problem involving the critical Sobolev exponent, Math. Z. 229 (1998), 443–474. [24] N. Ghoussoub, C. Gui and M. Zhu, On a singularly perturbed Neumann problem with the critical exponent, Comm. Part. Diff. Equ. 26 (2001), 1929–1946. [25] Y. Ge, R. Jing and F. Pacard, Bubble towers for supercritical semilinear elliptic equations, J. Funct. Anal. 221 (2) (2005), 251–302. [26] C. Gui, Multi-peak solutions for a semilinear Neumann problem, Duke Math. J. 84 (1996), 739–769. [27] C. Gui and J. Wei, Multiple interior peak solutions for some singular perturbation problems, J. Diff. Eqns. 158 (1999), 1–27. [28] C. Gui and J. Wei, On multiple mixed interior and boundary peak solutions for some singularly perturbed Neumann problems, Can. J. Math. 52 (2000), 522–538. [29] C. Gui, J. Wei and M. Winter, Multiple boundary peak solutions for some singularly perturbed Neumann problems, Ann. Inst. H. Poincaré Anal. Non Linéaire 17 (2000), 47–82. [30] P. Grossi and J. Wei, Existence of multiple-peaked solutions in a semilinear elliptic Neumann problem via nonsmooth critical point theory, Cal. Var. PDE 11 (2000), 143–175. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 97 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 583 [31] D.M. Holloway, Reaction–diffusion theory of localized structures with application to vertebrate organogenesis, PhD thesis, University of British Columbia (1995). [32] D. Iron and M.J. Ward, A metastable spike solution for a non-local reaction–diffusion model, SIAM J. Appl. Math. 60 (3) (2000), 778–802. [33] D. Iron and M. Ward, The dynamics of multispike solutions to the one-dimensional Gierer–Meinhardt model, SIAM J. Appl. Math. 62 (6) (2002), 1924–1951. [34] D. Iron, M. Ward and J. Wei, The stability of spike solutions to the one-dimensional Gierer–Meinhardt model, Physica D: Nonlinear Phenomena 150 (1–2) (2001), 25–62. [35] D. Iron, J. Wei and M. Winter, Stability analysis of Turing patterns generated by the Schnakenberg model, J. Math. Biology 49 (4) (2004), 358–390. [36] A.J. Koch and H. Meinhardt, Biological pattern formation from basic mechanisms to complex structures, Rev. Mod. Phys. 66 (4) (1994), 1481–1507. [37] M. Kowalcayk, On the existence and Morse index of solutions to the Allen–Cahn equation in two dimensions, Ann. Mat. Pura Appl. (4) 184 (1) (2005), 17–52. [38] P.K. Maini, K.J. Painter and H. Chau, Spatial pattern formation in biological and chemical systems, J. Chem. Soc. Faraday Transactions 93 (20) (1997), 3601–3610. [39] M.K. Kwong, Uniqueness of positive solutions of u − u + up = 0 in R N , Arch. Ration. Mech. Anal. 105 (1991), 243–266. [40] M.K. Kwong and L. Zhang, Uniqueness of the positive solution of u + f (u) = 0 in an annulus, Diff. Int. Eqns. 4 (3) (1991), 583–599. [41] Y.-Y. Li, On a singularly perturbed equation with Neumann boundary condition, Comm. PDE 23 (1998), 487–545. [42] C.-S. Lin and W.-M. Ni, On the diffusion coefficient of a semilinear Neumann problem, Calculus of Variations and Partial Differential Equations, Trento, 1986, Lecture Notes in Math. 1340, Springer, Berlin (1988), 160–174. [43] C.S. Lin, L. Wang and J. Wei, Bubble accumulations in an elliptic Neumann problem with critical Sobolev exponent, preprint, 2005. [44] F.H. Lin, W.M. Ni and J.C. Wei, On the number of solutions for some singularly perturbed Neumann problems, Comm. Pure Appl. Math. (2006), in press. [45] P. Maini, J. Wei and M. Winter, On the Gierer–Meinhardt system with Robin boundary conditions, preprint (2006). [46] H. Matano, Asymptotic behavior and stability of solutions of semilinear diffusion equations, Pub. Res. Inst. Math. Sci. 15 (1979), 224–243. [47] J. Murray, Mathematical Biology, I: An Introduction, II: Spatial Models and Biomedical Applications, third ed., Springer-Verlag, New York (2001). [48] A. Malchiodi, Concentration at curves for a singularly perturbed Neumann problem in three-dimensional domains, preprint (2004). [49] A. Malchiodi and M. Montenegro, Boundary concentration phenomena for a singularly perturbed elliptic problem, Comm. Pure Appl. Math. 55 (2002), 1507–1508. [50] A. Malchiodi and M. Montenegro, Multidimensional boundary layers for a singularly perturbed Neumann problem, Duke Math. J. 124 (1) (2004), 105-143. [51] F. Mahmoudi and A. Malchiodi, Concentration on minimal submanifolds for a singularly perturbed Neumann problem, preprint, 2006. [52] A. Malchiodi, W.M. Ni and J. Wei, Multiple clustered layer solutions for semilinear Neumann problems on a ball, Ann. Inst. H. Poincaré Anal. Non Linéaire 22 (2005), 143–163. [53] W.-M. Ni, Diffusion, cross-diffusion, and their spike-layer steady states, Notices Amer. Math. Soc. 45 (1998), 9–18. [54] W.-M. Ni, Qualitative properties of solutions to elliptic problems, Stationary Partial Differential Equations, vol. I, Handb. Differ. Equ., North-Holland, Amsterdam (2004), 157–233. [55] W.N. Ni, X.B. Pan and I. Takagi, Singular behavior of least-energy solutions of a semi-linear Neumann problem involving critical Sobolev exponents, Duke Math. J. 67 (1992), 1–20. [56] W.-M. Ni, P. Polacik and E. Yanagida, Monotonicity of stable solutions in shadow systems, Trans. Amer. Math. Soc. 353 (12) (2001), 5057–5069. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 584 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 <uncited> 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 <uncited> 44 45 F:chipot506.tex; VTEX/Rita p. 98 J. Wei [57] W.-M. Ni and I. Takagi, On the shape of least energy solution to a semilinear Neumann problem, Comm. Pure Appl. Math. 41 (1991), 819–851. [58] W.-M. Ni and I. Takagi, Locating the peaks of least energy solutions to a semilinear Neumann problem, Duke Math. J. 70 (1993), 247–281. [59] W.-M. Ni, I. Takagi and E. Yanagida, Stability analysis of point-condensation solutions to a reaction– diffusion system proposed by Gierer and Meinhardt, Tohoku Math. J., in press. [60] W.-M. Ni, I. Takagi and E. Yanagida, Stability of least energy patterns of the shadow system for an activatorinhibitor mo del, Japan J. Industr. Appl. Math., in press. [61] W.-M. Ni and J. Wei, On the location and profile of spike-layer solutions to singularly perturbed semilinear Dirichlet problems, Comm. Pure Appl. Math. 48 (1995), 731–768. [62] W.-M. Ni and J. Wei, On positive solutions concentrating on spheres for the Gierer–Meinhardt system, J. Diff. Eqns. 221 (2006), 158–189. [63] J.E. Pearson, Complex patterns in a simple system, Science 261 (1993), 189–192. [64] O. Rey, The question of interior blow-up points for an elliptic Neumann problem: the critical case, J. Math. Pures Appl. 81 (2002) 655–696. [65] O. Rey and J. Wei, Blow-up solutions for an elliptic Neumann problem with sub-or-supcritical nonlinearity, II: N 4, Ann. Inst. H. Poincaré Anal. Non Linéaire 22 (4) (2005), 459–484. [66] O. Rey and J. Wei, Arbitrary number of positive solutions for an elliptic problem with critical nonlinearity, J. Eur. Math. Soc. 7 (4) (2005) 449–476. [67] I. Takagi, Point-condensation for a reaction–diffusion system, J. Diff. Eqns. 61 (1986), 208–249. [68] A. M. Turing, The chemical basis of morphogenesis, Phil. Trans. Roy. Soc. Lond. B 237 (1952), 37–72. [69] X.J. Wang, Neumann problem of semilinear elliptic equations involving critical Sobolev exponents, J. Diff. Equ. 93 (1991) 283–310. [70] M.J. Ward and J. Wei, Asymmetric spike patterns for the one-dimensional Gierer–Meinhardt model: equilibria and stability, Europ. J. Appl. Math. 13 (2002), 283–320. [71] L. Wang and J. Wei, Interior bubbles for a singularly perturbed problem in lower dimensions, preprint. [72] L. Wang and J. Wei, Arbitrary many bubbles for Lin–Ni’s problem in dimension four and six, preprint. [73] J. Wei, On the boundary spike layer solutions of singularly perturbed semilinear Neumann problem, J. Diff. Eqns. 134 (1997), 104–133. [74] J. Wei, On the interior spike layer solutions of singularly perturbed semilinear Neumann problem, Tohoku Math. J. 50 (2) (1998), 159–178. [75] J. Wei, On the interior spike layer solutions for some singular perturbation problems, Proc. Royal Soc. Edinburgh Sect. A (Mathematics) 128 (1998), 849–874. [76] J. Wei, Uniqueness and eigenvalue estimates of boundary spike solutions, Proc. Royal Soc. Edin. A 131 (2001), 1457–1480. [77] J. Wei, On single interior spike solutions of Gierer–Meinhardt system: uniqueness and spectrum estimates, Eur. J. Appl. Math. 10 (1999), 353–378. [78] J. Wei, Existence, stability and metastability of point condensation patterns generated by Gray–Scott system, Nonlinearity 12 (1999), 593–616. [79] J. Wei, On the construction of interior spike layer solutions to a singularly perturbed semilinear Neumann problem, Partial Differential Equations: Theory and Numerical Solution, CRC Press LLC (1998), 336–349. [80] J. Wei, On a nonlocal eigenvalue problem and its applications to point-condensations in reaction–diffusion systems, Int. J. Bifur. Chaos 10 (6) (2000), 1485–1496. [81] J. Wei and L. Zhang, On a nonlocal eigenvalue problem, Ann. Sc. Norm. Sup. Pisa Cl. Sci. XXX (2001), 41–62. [82] J. Wei and M. Winter, Stationary solutions for the Cahn–Hilliard equation, Ann. Inst. H. Poincaré Anal. Non Linéaire 15 (1998), 459–492. [83] J. Wei and M. Winter, Multiple boundary spike solutions for a wide class of singular perturbation problems, J. London Math. Soc. 59 (1999), 585–606. [84] J. Wei and M. Winter, Existence, classification and stability analysis of multiple-peaked solutions for the Gierer–Meinhardt system in R 1 , preprint (2001). [85] J. Wei and M. Winter, On the two-dimensional Gierer–Meinhardt system with strong coupling, SIAM J. Math. Anal. 30 (1999), 1241–1263. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 chipot5 v.2007/07/10 Prn:17/08/2007; 15:55 F:chipot506.tex; VTEX/Rita p. 99 Existence and stability of spikes 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 585 [86] J. Wei and M. Winter, On multiple spike solutions for the two-dimensional Gierer–Meinhardt system; the strong coupling case, J. Diff. Eqns. 178 (2002), 478–518. [87] J. Wei and M. Winter, On multiple spike solutions for the two-dimensional Gierer–Meinhardt system; the weak coupling case, J. Nonlinear Sci. 6 (2001), 415–458. [88] J. Wei and M. Winter, Asymmetric Patterns for the Gierer–Meinhardt system, J. Math. Pures Appl. (9) 83 (4) (2004), 433–476. [89] J. Wei and M. Winter, Stability of monotone solutions for the shadow Gierer–Meinhardt system with finite diffusivity, Diff. Int. Eqns. 16 (2003), 1153–1180. [90] J. Wei and M. Winter, Higher-order energy expansions and spike locations, Cal. Var. PDE 20 (2004), 403– 430. [91] J. Wei, M. Winter and W. Yeung, A higher-order energy expansion to two-dimensional singularly perturbed Neumann problems, Asymptotic Analysis 43 (1–2) (2005), 75–110. [92] J. Wei and S. Yan, Solutions with interior bubble and boundary layer for an elliptic Neumann problem with critical nonlinearity, submitted for publication. [93] J. Wei and J. Yang, Concentration on curves for a singularly perturbed Neumann problem in twodimensional domain, preprint (2006). [94] J. Wei and J. Yang, Clustered line condensations for a singularly perturbed Neumann problem in twodimensional domain, preprint (2006). 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 37 38 38 39 39 40 40 41 41 42 42 43 43 44 44 45 45