Nonlinear wave equations (uncomplete draft) Heinz-Ju¨rgen Schmidt March 12, 2003 Contents 1 Introduction 2 2 Linear equations 4 2.1 Mathematical definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2.2 Physical examples and interpretation . . . . . . . . . . . . . . . . . . . . 5 2.2.1 Harmonical oscillator . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.2.2 Electromagnetic waves in a linear medium . . . . . . . . . . . . . 7 2.2.3 Schro¨dinger equation . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2.4 Linear dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3 Nonlinear equations 7 3.1 Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3.2 Nonlinear approximation to linear equations . . . . . . . . . . . . . . . . 7 3.3 Nonlinear 1-dimensional oscillations . . . . . . . . . . . . . . . . . . . . . 7 3.3.1 Small damping, no driving force . . . . . . . . . . . . . . . . . . . 7 3.3.2 General case: chaotic motion . . . . . . . . . . . . . . . . . . . . . 8 3.3.3 The Duffing oscillator (1918) . . . . . . . . . . . . . . . . . . . . . 8 4 Cubic Schr¨odinger equation (CSE) 10 4.1 Derivation from a Duffing-Lorentz model . . . . . . . . . . . . . . . . . . 10 4.2 General properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 4.3 Conservation laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 4.4 1-soliton solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 4.5 Interaction of 2 solitons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 5 Korteweg-de Vries equation 19 5.1 1-soliton solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 5.2 A physical derivation of the KdV equation for water waves . . . . . . . . 21 1 6 Exact solution methods for non-linear wave equations 29 6.1 An introduction into the inverse scattering transform (IST) . . . . . . . . 30 References 33 7 Appendices 34 7.1 Appendix 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 7.2 Appendix 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 7.3 Appendix 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 7.4 Appendix 4a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 7.5 Appendix 4b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 7.6 Appendix 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 1 Introduction This course addresses itself mainly to the fellows of the graduate college “Nonlinearities of optical materials”. It aims at providing a general physical background rather than dealing with the details of the various projects of this graduate college. It is generally held that most phenomena of nature are, in principle, nonlinear. Lin- earity is considered to occur only under idealized special circumstances. For example, any real oscillation will be nonlinear, and can only be approximated by a linear (or har- monic) oscillation in the case of small amplitudes. In contrast to this general situation, most methods of theoretical physics are confined to the linear case, which is mathemat- ically much easier to handle than the nonlinear case. Yet the abundance of nonlinearity seems somewhat paradoxical in view of the fundamental theories of physics which are, at present, general relativity and quantum field theory, including gauge theories of the fundamental interactions. These are indeed nonlinear theories, but their limiting cases which are believed to describe condensed matter and electromagnetic waves, are linear. Among the limiting cases we have the quantum theory of atoms, molecules and solid states as well as electrodynamics, quantum or classical. What then is the physical origin of nonlinearity? We will frequently come back to this question during this course. The first part of the course will deal with the definition of linearity and nonlinearity of (wave) equations and a couple of physical examples. The definition is not as trivial as it may look at first sight, since a nonlinear transformation can turn a linear equation into a nonlinear one. One wouldn’t like to call the result a nonlinear problem, since it is only a linear equation in some nonlinear disguise. But this leads to very difficult problems. In this part of the course we will consider 3 typical examples of linear equations of motion or wave equations and recall some approaches and methods of solution which are typical for linear equations. Then we will consider 3 nonlinear counter-parts of these equations where the mentioned methods fail. 2 One of the central themes of this course is the question: What are the physical origins of nonlinearity? We will have to give different answers depending on the case under consideration. From the huge class of nonlinear wave equations in physics we will select two important cases: The Korteweg-de Vries (KdV) equation ψ +σψψ +ψ = 0, (1) t x xxx and the cubic Schro¨dinger (CSE) equation iψ +ψ +σ|ψ|2ψ = 0. (2) t xx Here σ denotes a constant and the subscripts denote partial derivatives with respect to x or t. Both equations have numerous physical applications. We will concentrate on two typical applications and present detailed derivations from more fundamental equations in both cases. Thereby we study the physical origins of nonlinearity and, moreover, techniques of approximation. The KdV equation is suited to describe waves in shallow water. Among its solutions are the solitary waves (or “solitons”) first reported by J. Scott Russell in 1843. The nonlinearity of the KdV equation can be traced back to the nonlinearity of the more fundamental Navier-Stokes equations. The CSE appears, for example, in the context of self-focussing of laser beams in nonlinear optical media. It describes the first nonlinear correction to linear polarization and the propagation of certain optical solitons. As a rule, nonlinear equations cannot be solved exactly. Being oneexception to this rule, the exact solutions of some nonlinear wave equations (including the KdV- and the CSE) describing the interaction of N solitons has been one of the high-lights of mathematical physics of the 20th century. Unfortunately, the pertinent techniques (Lax theory, inverse scattering method, Ba¨cklund transforms etc. ) are too complicated to be presented in detail in this course. Rather we will try to give a rough overview about the methods involved and show some results in graphical form. For a selected number of topics we provide numerical and analytical studies in the form of MATHEMATICA notebooks, see the Appendices. Acknowledgement I thank M. Kadiroglu for his aid in preparing this course and F. Homann for providing the material of app. 5 and critical reading of the manuscript. 3 2 Linear equations 2.1 Mathematical definitions The definition of the linearity of any (wave) equation which has to comprise numerous physical cases is necessarily very abstract. We will illustrate this abstract definition by a couple of physical examples. We assume as known the notion of (real or complex) linear space or vector space. Its definition is taught in the introductory courses on linear algebra, see e. g. [1]. We will only consider the complex case, the linear case being analogous. LetV,W beC-linearspacesandD : V −→ W alinearoperator. Thenalinear, homogeneous (wave) equation is of the form Dφ = 0, φ ∈ V. (3) One may think of φ as a field obeying the wave equation. D will also be called the “wave operator”. It follows that the set of solutions of (3), Sol ⊂ V, will be a linear subspace of V. This h means that φ ,φ ∈ Sol and c ,c ∈ C implies 1 2 h 1 2 c φ +c φ ∈ Sol . (4) 1 1 2 2 h This equation expresses some kind of superposition principle for fields satisfying a linear, homogeneous wave equation. The corresponding inhomogeneous linear (wave) equation will be of the form Dφ = ρ, φ ∈ V,ρ ∈ W. (5) Mathematically, the inhomogeneous equation can also be written in the form (3), but its physical interpretation is different: ρ is assumed to be given, and φ is to be determined by ρ, (5) and additional boundary conditions. Typically, ρ is the “source” and φ is the “field” produced by ρ. In general, φ is not uniquely determined by ρ and (5), since Dφ = ρ = Dφ implies D(φ −φ ) = 0, i. e. φ −φ solves the homogeneous equation 1 2 1 2 1 2 (3). Therefore one usually needs additional boundary conditions to single out a unique solution φ. Mathematically, theset ofsolutions of(5)with fixedρ, Sol willbean“affinesubspace” i,ρ of V. If ρ is allowed to vary, we have again a linear subspace of solutions, Sol ⊂ V⊕W, i which means that (φ ,ρ ),(φ ,ρ ) ∈ Sol and c ,c ∈ C implies 1 1 2 2 i 1 2 (c φ +c φ ,c ρ +c ρ ) ∈ Sol , (6) 1 1 2 2 1 1 2 2 i which means that a superposition principle holds for fields and sources. 4 2.2 Physical examples and interpretation 2.2.1 Harmonical oscillator We start with a mechanical example which is also used as a simple model for the inter- action between electromagnetic radiation and matter (Lorentz model) to be considered later. 2.2.1.1 The simplest case Let us first consider the equation of motion of a 1- dimensional harmonical oscillator: d2x (t)+ω2x(t) = 0. (7) dt2 0 This is a linear, homogeneous equation in the sense considered in the preceding para- graph, if we choose V = W = the space of smooth functions x : R −→ R. Here and henceforward we will not be too precise about the exact function spaces appropriate for the various (wave) operators. In the case of (7) the (wave) operator is, of course, d2 D = +ω2 . (8) dt2 0 The superposition principle implies in this case that with the special solutions cosω t 0 and sinω t also 0 x(t) = Acosω t+Bsinω t, A,B ∈ C (9) 0 0 will be a solution of (7). Actually, (9) will be the most general solution. Hence the subspace Sol of V is 2-dimensional. h 2.2.1.2 The damped harmonic oscillator with driving force Next we will consider the equation of motion of a damped harmonic oscillator with driving force: d2x dx (t)+2Γ (t)+ω2x(t) = f(t), (10) dt2 dt 0 where Γ > 0 is a constant and f is a general driving force (divided by the mass). This is a linear, inhomogeneous equation with the same V and W as before and d2 d D = +2Γ +ω2 . (11) dt2 dt 0 Again,thesuperpositionprinciple(6)canbeemployedtocomposethegeneralsolutionof (10) from special solutions. Let G(t) denote the special solution of (10) with f(t) = δ(t). Since (10) is invariant under time translations, G(t−s) will also be a solution of (10) with f(t) = δ(t− s) for arbitrary s. Then a solution of (10) with general f(t) can be written as (cid:1) ∞ x(t) = G(t−s)f(s)ds. (12) −∞ 5 G is sometimes called the Greens function of the corresponding (wave) equation. The general solution of (10) is obtained by adding to (12) the general solution of the corre- sponding homogeneous equation. The latter vanishes for t → ∞. It remains to find the Greens function for (10). Since δ(t) = 0 for t > 0 it will be a solution of the homoge- neous equation for t > 0 and chosen as G(t) = 0 for t < 0 (“causality”). The appropriate initial condition is x(0) = 0 and x˙(0) = 1, since the second t-derivative of this kink gives just δ(t). This yields the result (cid:2) (cid:3) √ 1 e−Γtsin ω2 −Γ2t : t > 0 G(t) = ω02−Γ2 0 (13) 0 : t < 0 for the case Γ < ω . Due to the vanishing of G(t−s) for s > t the upper limit of the 0 integral in (12) may be replaced by t. An important case is a harmonic driving force f(t) = Aeiωt, A ∈ C. (14) Due to linearity (and the left hand side of (10) being real) this form can be superposed with f(t) = Ae−iωt to give real sinωt and cosωt expressions for f(t) which have a direct physical interpretation. Instead of solving the integral (12) it is easier in this case to insert the ansatz x(t) = Beiωt, B ∈ C (15) into (10) which gives A B = ≡ Ar(ω)eiϕ(ω), (16) −ω2 +2iΓω +ω2 0 1 r(ω) = (cid:3) : relative amplitude (17) (ω2 −ω2)2 +4Γ2ω2 0 2Γω tanϕ = , ϕ : phase shift. (18) ω2 −ω2 0 √ (cid:3)In the case Γ < ω0/ 2 the relative amplitude has its maximum at the value ωr = ω2 −2Γ2 which is called the “resonance frequency”. Typical graphs for r(ω) and ϕ(ω) 0 are shown in the following figures. 6 r 1 (cid:5)(cid:5)(cid:5)(cid:5)(cid:5)(cid:5)(cid:5)(cid:5)(cid:5) Ω 2 0 (cid:1)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2)(cid:2) Ω Ω (cid:2) Ω 2 (cid:3) 2 (cid:4)2 r 0 Figure 1: Relative amplitude r of forced oscillation as a function of the driving frequency ω. 2.2.2 Electromagnetic waves in a linear medium 2.2.3 Schr¨odinger equation 2.2.4 Linear dispersion 3 Nonlinear equations 3.1 Generalities 3.2 Nonlinear approximation to linear equations 3.3 Nonlinear 1-dimensional oscillations 3.3.1 Small damping, no driving force See app. 1. 7 (cid:6) Ω Ω 0 Π (cid:3) (cid:5)(cid:5)(cid:5)(cid:5) 2 (cid:3)Π Figure 2: Phase shift ϕ of forced oscillation as a function of the driving frequency ω.. 3.3.2 General case: chaotic motion See app. 2. 3.3.3 The Duffing oscillator (1918) See [5] 3.3.3.1 Bistable response See app. 3. 3.3.3.2 Perturbation series solution We consider a Duffing oscillator with small cubic nonlinearity and a harmonic driving force. Its equation of motion reads x¨+2γx˙ +αx+λx3 = F exp(i(ωt+φ))+CC, (19) where, as usual, CC denotes the complex conjugate of the preceding term. It is impor- tant to use a real driving force, since the use of complex forces and taking the real part of the resulting solution presupposes linearity. We assume that λx3 is small compared with the other terms of (19) and look for solutions which are close to the corresponding 8 solutions of the linear equation with λ = 0. Hence it appears natural to expand the solution into a perturbation series, i. e. a power series with respect to λ: (cid:4)∞ i x(t) = x (t)λ . (20) i i=0 Inserting this series into the equation of motion (19) and equating terms with the same powersofλyieldsahierarchyoflinear,inhomogeneousequationswhichcanbeexplicitely solved. It can be shown by induction that all functions of t which are involved in this hierarchy of equations are periodic functions with frequency ω. Hence they can be expanded into Fourier series, especially (cid:4) imωt x (t) = x e . (21) n nm m∈Z Since x(t) is a real function we can write x = x and need only consider non- n,−m nm negative n,m. Using computer algebra software it is possible to calculate a certain number of terms of the perturbation series, see app. 4a and 4b. Of course they will become more and more complex. Here we will only explicitely calculate the first few terms in order to see the underlying principle. λ0 terms It suffices to reproduce the result (16) of the linear problem, which reads, using the slightly different notation of the present section, iωt x (t) = x e +CC, (22) 0 01 iφ x = h(ω)Fe , (23) 01 1 h(ω) ≡ . (24) α−ω2+2iγω λ1 terms The cubic term gives λ(x +x λ+...)3 = λx3 +O(λ2) = λ(x3 e3iωt +3x2 x eiωt +CC)+O(λ2). (25) 0 1 0 01 01 01 9 Since the driving force is O(λ0), the O(λ1) terms of (19) are (cid:5) (cid:6) x¨ +2γx˙ +αx = −x3 = − x3 e3iωt +3x2 x eiωt +CC , (26) 1 1 1 0 01 01 01 1 x = −3F3eiφh2(ω)h(ω), (27) 11 h(ω) x = −3F3eiφh3(ω)h(ω), (28) 11 1 x = −(Feiφh(ω))3, (29) 13 h(3ω) x = −F3e3iφh3(ω)h(3ω). (30) 13 Some remarks are in order. We see in (26) that the cube of the solution in O(λ0) ap- proximation acts as a driving force for the next order approximation. Since x3 contains 0 the frequencies ω and 3ω, also x (t) will contain terms with these frequencies. Both 1 terms are proportional to F3, F being essentially the amplitude of the driving force. Hence in the lowest order the nonlinear term λx3 produces a correction to the linear response of the oscillator with the same frequency as well as a correction with triple frequency (third harmonic generation). Both terms could be split into a relative am- plitude and a phase shift part, similar as but more complicated than in the linear theory. λ2 terms Similar as above we conclude x¨ +2γx˙ +αx = −3x2x (31) 2 2 2 (cid:5)0 1 (cid:6) (cid:5) (cid:6) = −3 x eiωt +CC 2 x eiωt +x e3iωt +CC (32) 01 11 13 1 x = −3x2 x (33) h(5ω) 25 01 13 x = 3F5e5iφh5(ω)h(3ω)h(5ω) (34) 25 ... (35) It will not be necessary to give more terms. We can already see by the given example that the O(λ2) terms will contain the frequencies ω,3ω,5ω and that the corrections will be of order F5. We stress this finding in order not to generate the wrong impression that a cubic nonlinearity would at most produce 3ω terms proportional to F3. At the contrary, it seems fairlyclear that thehigher order corrections conclude allodd multiples of ω and odd powers of F. 4 Cubic Schr¨odinger equation (CSE) 4.1 Derivation from a Duffing-Lorentz model The well-known Lorentz model is a simple mechanical model to describe the linear re- sponse of matter to electromagnetic waves. The electrons in a piece of matter (insulator) 10