This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

KNAW

UTwente

**Lectures on Free Surface Waves
**

Brenny van Groesen, Applied Analysis & Mathematical Physics, University of Twente Lecture Notes for course

LABMATH: Mathematical support for Hydrodynamic Laboratories

August – September 2001

Course at Pusat Matematika, P4M, Institut Teknologi Bandung, Indonesia, in bi-national Scientific Cooperation Indonesia-Netherlands, KNAW - project 95-BTM-29

PREFACE, Acknowledgment

Surface waves are phenomena that are characterised by the dynamic interplay between linear interference and nonlinear interactions between individual waves. For understanding these natural phenomena and for many technological applications, such as in hydrodynamic laboratories, a good understanding and description are important. Mathematical modelling of the penomena is a challenge that was started by scientific giants like Huygens and Newton, and in the 19th century by Scott Russell, Boussinesq, Korteweg & de Vries. In the 20th century, dynamical system theory and improved methods from mathematical physics made it possible to incorporate nonlinear effects to a better extent. Investigations of the mathematical models have profited, and are still profiting, from mathematical-physical methods and, on the other hand, stimulated the development of these methods. This part of the Lecture Notes is written in the spirit of the interplay between

**natural phenomena – physics – – math modelling – analytical methods.
**

In the above mentioned cooperation programme between the applied mathematics institutes at ITB and UT with the hydrodynamic laboratories MARIN (Wageningen) and IHL (Surabaya), various people have been involved: senior staff members:

**Andonowati, Andi Jamalludin, Edy Soewono, Rene Huijsmans
**

PhD/S3-students:

**Jaap Harm Westhuis, Edi Cahyono, Toto Nusantara, Helena Margaretha, Agus Suryanto
**

and many participants of the Research WorkShop in 1997 which was the start of this project. I hope that the enthusiasm with which the problems have been tackled and results have been achieved in the best possible, friendly and stimulating atmosphere, is noticeable in these Lecture Notes.

Specific reference

Many results in Sections 2 and 3 is joint work with Andonowati, Edi Cahyono and Agus Suryanto. Thanks to Frits van Beckum for critically reading the notes and improving various errors.

**‘Lectures on Free Surface Waves’
**

Brenny van Groesen Lecturers & Tutors: Edi Cahyono & Helena Margaretha & Frits van Beckum August 5, 2001

Abstract The lecture notes consist of three sections, following a general description and overview of the aims and achievements in the LABMATH-project. In the ﬁrst section, basic notions of wave propagation are introduced starting at an elementary level: inteference, dispersion, nonlinear eﬀects, and progressing to the description by nonlinear pde’s with special mathematical-physical structure. In section 2 the basic ’laws of nature’ are given that describe the actual motion of surface waves. The most complete description is possible, but too diﬃcult to allow a direct investigation, which motivates the design and study of simpliﬁed models, in particular of KdV or NLS-type. In the ﬁnal section 3 a speciﬁc advanced phenomenon is taken: the nonlinear evolution of bi-chromatic waves, the bench-mark problem in LABMATH.

Contents

1 Basic notions Wave Propagation 1.1 Interference . . . . . . . . . . . . . . . . . . . . 1.2 Translation wave . . . . . . . . . . . . . . . . . 1.3 Linear Dispersive wave model . . . . . . . . . . 1.4 Wave groups: modulation . . . . . . . . . . . . 1.5 Nonlinearity: mode-interaction and -generation 1.6 Wave Equations: pde’s . . . . . . . . . . . . . . 1.7 Distortion of linear wavegroups: linear envelope 1.8 Nonlinear wave equation: ”Breaking” waves . . 1.9 Math-physical structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 5 6 7 8 9 11 12 13 17 17 20 20 20 22

2 Model Equations for free Surface Waves (SW) 2.1 Full surface wave equations . . . . . . . . . . . . . . . . . . . . 2.1.1 Variational structure . . . . . . . . . . . . . . . . . . . . 2.2 Physics of wave evolution: energy, momentum . . . . . . . . . . 2.2.1 Intermezzo: Conservation laws and conserved quantities 2.2.2 Conservation properties for full SW equations . . . . . . 1

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

2.3 2.4 2.5

Linearised SW, dispersion . . . . . . . . . . KdV type of equations . . . . . . . . . . . . Focusing/defocusing NLS-models . . . . . . 2.5.1 Linear dispersive envelope equation . 2.5.2 Nonlinear contributions . . . . . . . 2.5.3 Phase-amplitude equations for NLS

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

23 25 27 28 29 32 34 34 35 37 38 39 39 39 40 45 46 47 49 50 51 51 52 55

3 Advanced Surface Wave Phenomena 3.1 Solitons of KdV . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Analysis of solitary wave proﬁles . . . . . . . . . . . 3.2 Special solutions of NLS . . . . . . . . . . . . . . . . . . . . 3.2.1 Nonlinear harmonic . . . . . . . . . . . . . . . . . . 3.2.2 Nonlinear modulated harmonic . . . . . . . . . . . . 3.2.3 Soliton . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.4 Nonlinear bi-harmonic . . . . . . . . . . . . . . . . . 3.3 Intermezzo: Uniform expansion techniques . . . . . . . . . . 3.3.1 Third-order nonlinearity: Duﬃng’s eqn . . . . . . . 3.3.2 Improved perturbation techniques . . . . . . . . . . 3.3.3 Uniform Approximation Procedure . . . . . . . . . . 3.3.4 Second-order nonlinearity . . . . . . . . . . . . . . . 3.3.5 Resonance in dispersive wave equations . . . . . . . 3.4 Distortion of bi-chromatic waves . . . . . . . . . . . . . . . 3.4.1 Nonlinear harmonics . . . . . . . . . . . . . . . . . . 3.4.2 Uniformly valid approximate bi-harmonic evolution . 3.4.3 Graphical representation . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

2

**Schedule — for presentation/exercises
**

• Overview LABMATH, motivation (sheets)

• Basic notions and phenomena in Wave Propagation (LN section 1) — Interference — Dispersion — Wave groups: modulation (bi-chromatic) — Nonlinearity — Math-physical structure

• Model Equations for free Surface Waves (LN section 2) — Full SW equations — Physics of wave evolution: energy, momentum — Linearised SW, dispersion — KdV model — soliton

• Basic approaches to numerical modelling (lecturer: Frits van Beckum)

• Nonlinear evolution of Bi-chromatic waves (LN section 3, lecturer: Edi Cahyono) — Series expansions — Nonlinear Bi-chromatic wave — Analytic Wave Code

3

λ x → cos(kx). For example: 2 1 -4 -2 0 2 x 4 -1 -2 Harmonic function. = 2π /k wavelength Superposition of 2 harmonics leads to (harmonic) modulation of harmonic carrier: ¯ ). and to enlargement (constructive interference) at other points. k 4 . periodic with period (=wave length) k wavenumber. cos(k1 x) + cos(k2 x) = 2 cos(∆kx) cos(kx | {z } | {z } ¯ = (k1 + k2 )/2 with ∆k = (k2 − k1 )/2. destructive interference) at some points.1 1.1 Basic notions Wave Propagation Interference Superposition of ’wiggling’ functions can lead to annihilation (cancellation.

e. get a harmonic wave: cos(kx − ω t) k wave number. is given by Z 1 F (k) = f (x)e−ikx dx. and more generally Z 1 f (x)g (x)dx = 2π Z ˆ(k)ˆ g (k)dk f Z 1 |f (x)| dx = 2π 2 Z |F (k)|2 dk Exercise 2 Show interference (’MAPLE-plot’) for bi-harmonic cos(14x) + cos(16x). 1.t) = p(x − ct) For harmonic proﬁle. ω frequency ω /k phase velocity 5 . With complex notation for Fourier integral: Provided the ’energy’ is ﬁnite.Intermezzo 1 Fourier theory Any (square integrable) function f can be expressed as a superposition of possibly inﬁnite number of harmonics. 2π Some basic properties: 1. i. f is real. Parceval’s indentity: 3. ⇔ F (k) = F (−k) (overbar is complex conjugate) 2.2 Translation wave Fixed Proﬁle shifted with constant speed: η (x. −∞ f can be written like (all integrals over the wole real axis in the following) Z f (x) = F (k)eikx dk where F . Z ∞ |f (x)|2 dx < ∞. the Fourier-transform of f.

Example 4 1. speed ∆ω/∆k ’group velocity’ Interpretation: superposition of two ‘waves’. leads to ‘beat-pattern’. each one propagating with its own phase speed ω k . Animation Exercise 3 Introduce time dependence and animate (’MAPLE-animate’) the bichromatic evolution. 1.e. cos(k1 x − ω1 t) + cos(k2 x − ω2 t) = 2 cos(∆kx − ∆ωt) cos(kx {z }| {z } | ¯ −ω cos(kx ¯ t) ¯ speed ω ¯ /k lead to a modulation of a carrier wave. both the carrier wave and the modulated wave propagate at their own speed: carrier wave modulation: amplitude cos(∆kx − ∆ω t). Z corresponding evolution : η (x.3 Linear Dispersive wave model Uni-directional Dispersive wave model: for each wavenumber the physical model assigns a corresponding frequency k − > Ω(k ). no dispersion: Ω(k) = ck for some constant c.Superposition of two harmonic waves with diﬀerent phase speed: ¯ −ω ¯ t). verify the propagation speeds of carrier wave and of modulation wave. t) = η ˆ(k)ei[kx−Ω(k)t] dk for initial wave proﬁle Observe/deﬁnition: 6 . such that exp(i [kx − Ω(k )t]) is a physical solution. surface water waves (see next section) in normalised variables: r tanh(k ) SW-dispersion: Ω(k ) = k k and approximated for ’long waves’ (i. t = 0) = η ˆ(k)eikx dk. 2. a so-called harmonic ’mode’. k small) Ω(k ) ≈ k(1 − k2 /6) Then Fourier-theory describes the evolution of an initial wave proﬁle: Z : η (x. the relation being the so-called dispersion relation.

meaning η (x. 7 . dΩ(k) .and groupvelocity: 1. t) = η ˆ(k)e the case of translating wave. 4. Plot of SW-dispersion. The group velocity is deﬁned as 1. Ω(k) = ck for some constant c. 0)2 dx. the phase. t) = ei[k0 x−Ω(k0 )t] When there is dispersion. it is found that Z Z i[kx−Ω(k)t] dk = η ˆ(k)ei[k(x−ct] dk = η (x − ct. provided Ω(k) = −Ω(−k ) R R 2. a special example is the bi-chromatic wave considered above. provided Ω(k) is real.3. the evolution may be much more diﬃcult. This example can be rewritten in complex notation with a spectral function consisting of two delta-functions: η ˆ(k ) = δ (k − k1 ) + δ (k − k2 ). A monochromatic wave (complex notation) is obtained for a spectral function that is a delta-function: η ˆ(k) = δ (k − k0 ) ⇒ ηmonochrom (x. t)2 dx = η(x. dk Vgr (k) = when there is dispersion. η (x.4 Wave groups: modulation When there is no dispersion. evolution is conservative. 0). this group velocity is not constant but depends on the wave number. evolution from real initial data is real .

one then talks about a wave packet. a term appears for which the quantity is squared. say N (η) = η 2 The Fourier transform of a product is not a simple multiplication. the solution can be written as a harmonic wave at the central wavenumber modulated by an amplitude Z η (x. the quadratic and higher order terms in the dispersion will lead to some distortion (broadening) of the amplitude’s proﬁle. say η ˆ(k) = G(k − k0 ) and Taylor expansion of the dispersion relation around κ = k − k0 ≈ 0 : ω (k) = Ω0 + V0 κ + βκ2 + . V0 = dk dk 2 Then. However. the evolution can be seen (as in the case of the bichromatic wave) as a modulation of a wave with the main wavelenght. t) = ei[k1 x−Ω(k1 )t] + ei[k2 x−Ω(k2 )t] In general. The details of the statement follow from looking at a sharply peaked spectral function. ¤ £ A(x.. when the waves have almost the same wavelenght. 8 . Below we will investigate this deviation by deriving the equation for the amplitude. dΩ d2 Ω (k0 ). t) = G(k − k0 )ei[kx−Ω(k)t] dk = A(x. which is the group velocity at the central wavenumber. the amplitude has ﬁxed proﬁle and travels at constant speed V0 . G(κ) = the amplitude is a Gaussian itself: £ ¤ 1 exp −κ2 /(4σ 2 ) σπ In the more general case. For instance. t)ei[k0 x=Ω0 t] . in particular the evolution of surface waves. in case of Quadratic Nonlinearity. For instance. β = (k0 ) Ω0 = ω (k0 ). t) = exp −σ 2 (x − V0 t)2 1..5 Nonlinearity: mode-interaction and -generation Most systems. t) = G(κ)ei[κ(x−V0 t)−βκ t] dk Observe that upon neglecting the quadratic terms in the dispersion relation. the solution in the form of the Fourier-integral expression is diﬃcult to grasp..from which η bichrom (x. but a convolution. are non-linear. Z 2 A(x. for a Gaussian wave packet.

But so far no physics (the ’LAWS OF NATURE’) has been made explicit. which is satisﬁed for any function u = p(ξ ). 9 . the successive evolution (positive time) is uniquely deﬁned. Translation wave equation For any proﬁle function p the translation wave: u(x. and possibly higher order derivatives. introduce new variables ξ = x − ct. For instance. and hence the result. υ = x + ct and observe that the pde becomes in new variables: ∂υ u = 0.) We will now show some examples. possibly achieved by increasing the number of variables). will be described by relations between the elevation. but here we just prepare for the general frame work. products of long waves produce short waves. 1. and conversely We will see the drastic consequence below. η . in contrast to ordinary de’s for functions of only one variable. which is the description by partial diﬀerential equations. and its derivatives with respect to time ∂t η and space ∂x η . (The fact that partial derivatives are involved. g the Fourier transform of the product f g is the convolution of the Fourier-transforms: Z c ˆ ˆ(k − `)ˆ fg (k ) = (f ∗ g ˆ)(k) ≡ f g (`)d` eiκ1 x · eik2 x = ei(k1 +k2 )x which expresses that two Fourier components produce. mosttimes we will emphasize the time evolution. t) = p(x − ct) satisﬁes the simple pde ∂t u = −c∂x u and conversely1 !! Linear Dispersive Wave equation 1 To show the converse.6 Wave Equations: pde’s All the topics above are mathetical methods that are useful to describe phenomena that show characteristic wave properties. modes will be mixed up. We will discuss this in the next section. and some smoothness conditions are satisﬁed. t). This has a fundamental reason: if the equation is of ﬁrst order in time (meaning that the highest order time derivative is 1. if such nonlinearity is present. the IVP (Initial Value Problem) is well posed: for any given initial ’state’ (at initial time). that is express ∂t η in terms of the spatial derivatives. ’generate’.In the simplest case (verify) Intermezzo 5 For functions f. Clearly. a third one. say for the wave elevation η(x. As will be relevant for the wave equations. explains the name partial de. The physical laws.

First observe that exp(i(kx − Ω(k)t)) can be written as a time dependent coeﬃcient multiplying exp(ikx). Ak (0) = η ˆ(k ) upon multiplying by eikx and integrating over k there results Z ∂t η (x. Ω waves.A single harmonic function u = exp(i(kx − ω t)) is a special solution (not the only one) of the pde ω ∂t u = − ∂x u k However.t) = −i Ω(k )Ak (t)eikx dk Intermezzo 6 Pseudo-diﬀerential operators The right hand side is an operator action on the function η (x. this is the general form of a linear dispersive wave equation.t) = −i Ω(k)Ak (t)eikx dk can be written like ˇ η (x. t) = η ˆ(k )e dk = η ˆ(k)e e dk = Ak (t)eikx dk Since ∂t Ak = −iΩ(k)Ak . and the function Ω is called the symbol of the operator Ω ˇ is a simple diﬀerential operator. this formulation should be generalised as follows.t) = −iΩ When Ω(k) is real (for real k ) and odd. 10 . this is for each k a solution of the ode ∂t ak = −iΩ(k)ak . for a dispersive wave model. t). and hence Ω(k) = k n . the operator Ω since ˇ = i∂x . ∂t η (x. Ω ˇ = (i∂x )n Ω(k) = k. the equation ∂t η (x. t) by multiplying each Fourier coeﬃcient of η by Ω(k) and then inverse Fourier transformation. This so-called Fourier-integral operator. it is deﬁned as as Ω Z Z ikx ˆ(k)eikx dk ˆ ˇ for f (x) = f (k)e dk by deﬁnition Ωf (x) = Ω(k )f ˇ. ak (0) = 1 This motivates to ﬁnd the equation for the general solution Z Z Z i[kx−Ω(k)t] −iΩ(k)t i[kx] η (x. for which ω = Ω(k) is prescribed. R Using this notation. Example 7 When Ω is a polynomial. will be denoted symbolically ˇ . writing ak (t) = exp −iΩ(k)t. Ω ˇ is also well-deﬁned when it is not a polynomial as in te case of surface However. or pseudo-diﬀerential operator.

. say η ˆ(k) = G(k − k0 ).e. the distortion of the amplitude of the wavegroup is described by this envelope equation. dΩ d2 Ω (k0 ). Z 2 A(ξ . 1.t)dx = 0 Exercise 10 Verify (prove) these two propositions. t) = G(κ)ei[κ(x−V0 t)−βκ t] dk From Taylor expansion of the dispersion relation around κ = k − k0 ≈ 0 : where qubic and higher order terms in the dispersion relation have been neglected. meaning travelling at constant speed V0 in the laboratory frame of reference. τ ) is the solution of a dispersive wave equation that has dispersion relation κ → −βκ2 . V0 = dk dk 2 we found that the solution can be written as a harmonic wave at the central wavenumber modulated by an amplitude η(x.e. Ω(k ) = −Ω(−k ). β = (k0 ) Ω0 = ω (k0 ). Z 2 A(x. 11 . A satisﬁes 2 ∂τ A = −iβ∂ξ A. i. g Z Z ˇ f (x)dx. Observe that for β = 0. the solution with a sharply peaked spectral function. t)ei[k0 x=Ω0 t] . Hence.7 Distortion of linear wavegroups: linear envelope equation Let’s now investigate once again the linear wave group. t) = A(x.. τ ) = G(κ)ei[κξ−βκ τ ] dk Observe that A(ξ . the amplitude is ﬁxed. i..Proposition 8 When Ω(k ) is real (for real k ) and odd. the corresponding pseudo-diﬀerential operator is skew-symmetric in the sense that for any two real functions f. ˇ f (x)Ωg (x)dx = − g (x)Ω Proposition 9 For a linear dispersive wave equation the square energy is a constant of the motion: Z ∂t η2 (x. Using ξ = x − V0 t and τ = t as the basic variables (which means introducing a frame of reference moving with the group velocity). ω (k) = Ω0 + V0 κ + βκ2 + .

dt Therefore. t). t) + u(x. subtract the undeformed evolution at the group velocity. t) = ∂x u(x(t). Example 12 For a Gaussian function as example of an initial proﬁle. The characteristic through this point is the straight line with tangent f (y ): x = y + f (y) t and on this line the function u has the value f (y ). dt dt and hence u(x. t) x(t) + ∂t u(x(t). First observe that on a curve t → x(t) the total derivative of u is given by d d u(x(t). t) = f (y) is the solution in implicit form. which is called a characteristic curve. t plane.8 Nonlinear wave equation: ”Breaking” waves Consider the following prototype of a nonlinear equation: ∂t u(x. is in this case a straight line with slope u. together with a diﬀerent presentation that is obtained by shifting the proﬁles vertically with increasing time.Exercise 11 Study the ’slow’ deformation of the envelope in a wave packet due to dispersion. t) = 0 on x(t) = u. and so u(y + f (y ) t. t) ∂x u(x. approximate the integral by a Riemann sum.) 1. t) = constant on d x(t) = u. t)) = 0 The solution with initial value f can be written analytically in an implicit way as follows. 12 . the solution can be plotted easily using the parametric representation. In detail: Let y be a point at the x-axis of the x. dt dt From this it follows that d d u(x(t). at this point u has the value f (y). a succession of proﬁles at diﬀerent times is given below. the ‘curve’ with ddt x(t) = u. (Hint: make plots in Maple. From this the construction becomes clear by using the information of the initial value.

A special example of a basic property is conservation of ’energy’. etc. That is.. [[ Hint: Note that µ ¶ ∂u d d u(y + f (y ) t.]] 1. Draw the proﬁles at two successive (positive) times of the solution with initial condition the positive step function: ½ = 0 for x < 0. Do the same for the reversed step function. t∗ ) = ∞.e. t. i.. 4. t) = (x. 3. t) 1 + t f (y) dy ∂x dy and show that from this it follows that on the characteristic through y this is the case for the time (if it exists) such that . complete the argument.. Determine the ﬁrst time t∗ at which “breaking” occurs: ∂x u(x. the initial condition step(−x). but it is worthwhile to try to recognise basic properties of physics from the mathematical structure.plane.9 Math-physical structure ’Laws of Nature’ express basic properties and describe special phenomena. 2. sometimes this is not obvious.Exercise 13 Consider the nonlinear “breaking” equation ∂t u + u∂x u = 0. step(x) = = 1 for x > 0. 1. Mathematical equations that express these LofN will/should contain these properties and phenomena in the description. 13 . Draw in each case characteristic curves in the x. such properties will lead to a special structure of the math equations..

Deﬁnition 15 A dynamical system in Rn (or C n ) with state variable u(t) is called a Hamiltonian (Poisson) system if the system of ode’s has the following form: ∂t u = Γ∂u H (u). (q. E = H (q. Newton’s equation with conservative force from potential energy V = V (x): x ¨+ 2. Level sets of the function E give information about the dynamics (why?).e. In complex notation z ˙ = −i ∂H . and where u → H (u) is a given function. E = H (z ) ∂z 1 2 ∂V = 0. Im z ). E = x ˙ + V (x) ∂x 2 3. (Re z. Γv >. ˙ x). p) ∂t 1 0 p ∂p H Actually. Draw th so-called phase portrait. namely all these are examples of ode’s with a Hamiltonian (Poisson) structure. called the Hamiltonian 14 . ½ q ˙=p p ˙ = −ω2 q E= ¢ 1¡ 2 x ˙ + ω 2 x2 2 1 E = |z |2 2 ¢ 1¡ 2 E= p + ω2 q2 2 The phase plane for these equations is the plane (x. z ˙ + iω z = 0.Exercise 14 Show that for the following ode’s in time the given quantity E is conserved (constant in time for solutions) x ¨ + ω2 x = 0. v >= − < u. These examples are special cases of certain classes of equations. the ‘structure map’ meaning that in a suitable innerproduct < Γu.variables (q. all these classes of equations themselves are just representatives of one larger class of equations. p) respecively. i. various levellines in the phase-plane. Classical Hamilton’s equations in so-called canonical. p): ¶µ ¶ µ ¶ µ ∂q H 0 −1 q = . where Γ is a skew-symmetric operator. each of which has ’energy conservation’: 1.

can be recognised as such. when viscous eﬀects which cause dissipation are neglected. As a simple example. just as in the examples above. v >= ¯ dε ε=0 15 . Remark 18 This is an example of the opening sentences of this subsection: the mathematical structure of Hamilton’s equations implies that there is a constant of the motion (H ). i. sum of kinetic and potential energy. v >= u(x)v (x)dx. t). For instance. Exercise 17 Show that for a Hamiltonian system the Hamiltonian is a constant of the motion itself: ∂t H (u) = 0 for solutions. and hence u(t) now belongs to an inﬁnite dimensional state space — the symplectic structure is now Γ = ∂x and is indeed skew-symmetric for the usual L2 -innerproduct of functions of x that decay at inﬁnity:: Z < u. a Hamiltonian structure contains a conservation property. u(t)(x) = u(x. in most cases the Hamiltonian is (related to) the total physical energy. in this case given by Z 1 2 cu (x)dx.e. it is possible to generalise the notion of Hamiltonian system. called functional. — the Hamiltonian is now a function on the function space. H (u) = 2 it can be shown that the ’derivative’ of such a functional H (u). called the ’variational derivative’ and denoted by δ H (u). In Classical Mechanics Hamilton’s equations are studied in detail. all wave equations. i. the translation equation at constant speed ∂t u = −∂x (cu) is of this form: — the state variable t → u(t) now depends on x.Exercise 16 Show that the given equations above are indeed Hamiltonian systems by specifying the structure map and the Hamiltonian. can be deﬁned just as for functions on Rn with the directional derivative as a generalisation of the gradient: ¯ ¯ d H (u + εv )¯ . For systems described by partial diﬀerential equations. < δ H (u).e.

16 . Exercise 19 Verify by direct calculation that this Hamiltonian is conserved for solutions of the translation equation (restrict to functions that vanish at inﬁnity).for the given Hamiltonian H (u) = Z 1 2 cu (x)dx =⇒ δ H (u) = cu 2 from which the Hamiltonian structure of the translation equation follows. Give also a more direct argument for this result. based on the observation of the change of this functional when the function is translated over an arbitrary distance.

For waves with small wave lengths. this is a Hamiltoinian structure as brieﬂy discussed in the previous section. it is important to identify some fundamental properties. though diﬀerent from the well known Benjamin-Feir instability). which leads to a non-local version of the KdV-equation and the focusing NLS-equation. and then keep these properties in simpliﬁed models. Although the full SWE’s are well known and describe the complete physics. for wave packets with narrow banded spectra. while. That is. should be accurate enough to capture the phenomenon of interest. and the interior ﬂuid velocity. incompressible (density normalized to unity). and the ﬂow to be irrotational. and that the ﬂow is irrotational. simpliﬁed models are desired that are amenable for theoretical investigations. Mathematically. Using the available results from experiments and simulations. the dispersion for long waves leads to the classical KdV-equation. the full surface wave equations (full SWE) are well known. The combination of dispersive and non-linear eﬀects present major problems in the numerical simulation as well as in the theoretical analysis of the resulting interesting phenomena. 17 . we will carefully model the dispersive properties. which are physically most relevant. The simpliﬁed models will be checked for such properties. this set of equations is too complicated for a direct investigation. the basic assumptions are that the ﬂuid is incompressible and inviscid (such as water in a good approximation). In the ﬁrst case. we distinguish between long waves (shallow water) and short waves (deep water). A class of simpliﬁed models are KdV type of equations for uni-directional wave propagation. The full set of equations (SWE’s) is given ﬁrst. and no surface tension. to varying degree of accuracy. initially smooth wave packets will show large deformations (caused by side band instabilities. Mass conservation and energy conservation are the most important properties to be retained. In all cases. For designing simpler models. at the same time. In this section we present a uniﬁed view on results from the literature and describe models of KdV. since this is essential for the phenomena to be expected. describe. the full dispersive properties should be dealt with. furthermore we mainly restrict to wave propagation in one horizontal direction. and the derivation is recalled as far as necessary for the appreciation of the two diﬀerent cases (focusing and defocusing). comparable to the depth of the ﬂuid. Finally. and to the defocusing NLS-equation for wave packets.1 Full surface wave equations We consider the motion of a layer of ﬂuid under the following simplifying assumptions: • the ﬂuid is inviscid. NLS-type of equations are described. these will be considered both for shallow and deep water waves. Assuming the ﬂuid to be incompressible and inviscid.and NLS-type of equations and their relation. the surface wave problem.2 Model Equations for free Surface Waves (SW) The evolution of waves on the surface of a layer of ﬂuid (such as water) remains a challenging task. Therefore. except from detailed experiments and numerical simulations. to gain insight in interesting characteristic phenomena. These equations describe in detail the evolution of the surface elevation. The mathematical models to be presented. However. 2. In the latter case. the variational structure present in the equations. we investigate the performance of the simpliﬁed NLS-model. is just as important to be retained.

η (x. t) (1) (2) (3) Φz = 0 at z = −H. t). t) and its x-derivative (which is a velocity type of quantity) u(x. W = Φz . w still has to be expressed as a function of η . t). Remark 20 Alternative descriptions 1. u.• the bottom is ﬂat. t). Then incompressibility implies Ux + Wz = 0. t) = ∂x ϕ = Φx + Φz η x Then ∂t ϕ = Φt + Φz η t . z. 1 1 ∂t u = −∂x [g η + u2 − w2 (1 + η 2 x )] 2 2 where w is the vertical velocity at the free surface: w(x. For various reasons it is convenient to introduce the potential at the free surface ϕ(x. ∂t η = −η x Φx + Φz ∂t Φ + at z = η (x. W ) = ∇Φ. U = Φx . Then the governing equations are ∆Φ ≡ Φxx + Φzz = 0. t) = Φ(x. however. t). u in which. ∆Φ = 0. if the horizontal and vertical velocities are denoted by U = U (x. • the ﬂuid motion is irrotational. • the surface elevation is the graph of a function (no overturning waves) η = η (x. This is a set of equations for η . irrotational motion means Uz − Wx = 0 and allows the introduction of the ﬂuid potential Φ such that (U. resulting from Bernoulli’s equation restricted to the free surface. η (x. t) and W = W (x. and the free surface conditions become ∂t η = −uη x + w(1 + η 2 x ). 1 2 (Φ + Φ2 (4) z ) + g η = 0 at z = η (x. 2 x Equation (3) is a kinematic condition. t). t). z. at depth z = −H . t) respectively. − H < z < η (x. t) = W (x. equation (4) is a dynamic condition. 18 (5) (6) . assumed to be uniform in the (horizontal) y-direction and unbounded in the x-direction.

on a spatially unbounded domain. these are an impermeable bottom condition. Alternatively. This also shows that the boundary conditions are in ’conservation form’: (7) and (6). this is nothing but mass-conservation. 4. lateral boundary conditions. t) ≡ ∂t η + ∂x η dt dt For multi-directional waves. absorption at the down stream end of the domain. For a simple interpretation of the SWE’s. • use this information in the right hand side of the dynamic equations. The algorithm for one timestep can be described as follows: • start with a given surface elevation and given potential at the surface. and calculate the velocity (∇Φ) at the free surface. For the applications in hydrodynamic laboratories. and evolve these for one time-step. an initial surface elevation and horizontal ﬂuid velocity may be prescribed. • then solve the Laplace equation. 3. t). t). a simple extension is as follows.2. 19 . Stated diﬀerently. depending on both horizontal variables. Mass conservation within the ﬂuid domain D can now be expressed as Laplace’s equation for the potential ∆Φ ≡ Φxx + Φyy + Φzz = 0 in D. most important is to observe how the interior Laplace problem is linked to the dynamic equations. Another way to rewrite the kinematic surface condition (the continuity equation) is as a local conservation law: Z η Φx dz (7) ∂t η = −∂x −H which shows that the ﬂux equals the integrated horizontal velocity. Due to the irrotationality of the velocity ﬁeld ~ v a scalar potential function Φ can be deﬁned as ∇Φ = ~ v . and some prescribed condition at the upstream position to deﬁne a suitable signalling problem. (8) Now the two conditions that govern the free surface dynamics are 1 Φt = − |∇Φ|2 − g η on z = η (x. • this then provides updated ﬂuid surface and potential. 2 η t = Φz − η x Φx − η y Φy on z = η (x. (9) (10) Additional initial and boundary conditions are required to make the problem well-posed. Yet another interpretation of the kinematic surface condition is that the vertical ﬂuid velocity should equal the vertical displacement of the ﬂuid surface: Φz = d dx η(x(t).

Summarising. momentum Intermezzo: Conservation laws and conserved quantities We start with some general ideas about conservation in pde’s. Then the full surface wave equations can be described as a Hamiltonian system (see for full details [4]) ∂t η = −δ φ H (φ. thereby having eliminated the problem to calculate the interior ﬂow. η . t). 2. t). t) of the spatial variable x and time t. In fact. η ) = K (φ. Numerically this can be done nowadays quite eﬃciently and accurately. t) for its value at (x.derivatives of u at (x. actually the solution of the Dirichlet to Neumann (DtN) problem : given Φ at the surface. η ) + 2 where the kinetic energy is given forRsolutions of the Dirichlet problem for the Laplace probR R 1 2 dxdz = φ [ ∂ Φ lem in the ﬂuid domain: K (φ. t) is determined by u(x. which will after that be applicable for the wave problems. which would require to solve the DtN problem.e.1 Physics of wave evolution: energy. η ) = | ∇ Φ | n ]z =η dx. Let e = e[u] be some local density.• start the next timestep. The main problem in this algorithm is the calculation of the velocity at the free surface. Since this functional 2 cannot be expressed explicitly in terms of φ. ∂t φ = δ η H (11) where H is the Hamiltonian functional and φ = Φ|z =η is the free surface potential. Let the state of a dynamic system be described by a function u = u(x. We will describe this in the next subsection.1 Variational structure Another way to interpret the dynamical structure is using a variational description. ﬁnd ∇Φ at the surface. an expression that depends on u in such a way that the value of e at (x. η ). t) and a ﬁnite number of x. 2. leads to a model that has retained the basic variational structure. which is not guaranteed in a direct approach. Just as for Hamiltonian systems from Classical Mechanics. simpliﬁed models can be obtained by constructing approximations for this functional. We write e = e[u](x.2 2. 20 . various diﬀerent approximations for the solution of the DtN problem are used to arrive at equations on the free surface. and taking as governing equations the system (11) with the approximated Hamiltonian.1. In doing so. In ﬁnding simpliﬁed models for theoretical investigations. it has been observed (independently) by Zakharov (1968) and Broer (1974) and Miles (1977) that SWE can be described as a Hamiltonian system. the Hamiltonian is the sum of kinetic and potential energy Z 1 2 g η dx H (φ.2. i. this can be described by using as variables the ﬂuid potential at the free surface and the surface elevation.

t) = 0. and Q is called the ﬂux density corresponding to e. t) } dx = 0.e. 2. the reverse is equally true (and simpler): integrating (13) along an interval [a. Rewriting the right hand side Z b ∂x Q[u](x.The integral of e along a given spatial interval [a. When for a given solution.b] a the global law can be rewritten like Z b { ∂t e[u](x. dt [a. As described above. Intuitively clear. dt [a. strictly speaking use Lagrange’s Lemma and assume continuity of the expression in brackets in the integral. t) dx. b] deﬁnes the integrated quantity E[a. a If this holds for any interval of integration [a. the local conservation law is derived from the global conservation law. i. This expression is called the local conservation law for the density e. t) dx E[a. the result will be (12). 21 . t) dx = 0 Rb a a and we have a conserved quantity or constant of the motion: 2 e[u](x. the time derivative may depend only on the value of a (local) ﬂux density Q at the end points of the interval: d E [u](t) = −Q[u](b. the net ﬂux through the boundary vanishes. t) + ∂x Q[u](x.b] that will depend on u and time: Z b e[u](x. t) dx −Q[u](b. it follows2 that ∂t e[u](x. t) then ∂t Z b (13) e[u](x. Remark 21 1. when Q[u](a. b]. t) dx.b] [u](t) ≡ a In speciﬁc cases. depending on the evolution equation and the density. t). t) + ∂x Q[u](x. b] (within some given interval).b] (12) This is then called a global conservation law for the density e. t) = Q[u](b. t) ≡ − a and (assuming diﬀerentiability) the left hand side Z b d E [u](t) ≡ ∂t e[u](x. t) + Q[u](a. t) + Q[u](a.

t) it is custom to talk about an evolution equation in conservation form.2 Conservation properties for full SW equations 1. 22 . write x ∈ Ω ⊂ Rn and let Ω be a bounded or unbounded domain with n the outward pointing normal at the boundary ∂ Ω. the existence of a local/global conservation law is special. the balance law becomes a conservation law. In the special case that the evolution equation for the state variable itself is of the form of a local conservation law ∂t u(x. t) = −∂x Q[u](x. The ﬂux of a (scalar) density e is an n-vector Q and the ﬂux through the boundary is determined by the normal component: Z Q · n dA ﬂux through boundary: ∂Ω With a source density S . For a system in more spatial dimensions.3. If S = 0. and a local density e. All these notions become particularly relevant when for e(u) a physically motivated density is taken. a local balance law is obtained in the form ∂t e + div Q = S. Conservation of mass (ﬂuid): The ’continuity equation’ in the description of continua with a mass density ρ and Eulerian velocity v reads ∂t ρ + div (ρv) = 0. ∂Ω Ω and assuming smoothness. t) dx = − Q · n dA + S (x. 2. the global balance law reads Z Z Z d e(x. 4. For a given evolution equation.2. t) dx dt Ω Ω ∂Ω Using Gauss’ theorem: Z Q · n dA ≡ Z div Q dx. in general the density will satisfy a local balance law of the form: ∂t e[u] + ∂x Φ[u] = S [u] where S is the so-called local source density.

2.n ∂t Ω Ω ∂Ω For water. which leads to Laplace equation with the expression for the Eulerian velocity (from irrotationallity assumption) v = ∇Φ. to be denoted by I (x. the total amount of mass can only change by ﬂux though the boundary as is expressed as follows: Z Z Z ρdx = − div (ρv )dx = − ρv. t) = 2 2 2 2 −H −H 2 We can and will neglect the inessential constant 1 2 ρH in the following. t): Z η I= Φx dz −H 3. This explains the various approximations. dispersion As stated above. In particular. the linearised problem can be solved and leads to the basic (linear) dispersion relation between frequency and wavenumber. for inﬁnitesimally small amplitude waves. given by (using normalised variables) p (14) ω = ±Ω(k).expressing mass conservation: in a ﬁxed domain Ω with boundary ∂ Ω with normal n. with Ω(k ) = k tanh(k )/k. which are based (given a horizontal straight bottom) on specifying simplifying properties of the surface elevation. the problem is on a straight strip and can be solved in closed form by Fourier techniques. 2. Without surface elevation and bottom variations. we assume ρ =constant. the same dispersion relation is found with k = |k|. For the depth integration of mass-conservation the kinematic condition follows as we have seen already: Z η Z η dz = ∂t η = −∂x Φx dz ∂t −H −H The ﬂux density is the horizontal momentum. In more dimensions. any simplifying model is based on an approximation of the DtN problem. 23 . for a plane wave with wavevector k. Total Energy 2 The local kinetic energy density is given by 1 2 ρ|∇Φ| . and therefore the total energy integrated over depth is found to be: ¶ µ ¶ ¶ Z η µ Z η µ 1 1 1 1 2 2 2 2 ρ|∇Φ| + ρz dz = ρ|∇Φ| dz + ρη + ρH E (x. The potential energy of mass ρ at height z equals ρz. corresponding to div (v) = 0.3 Linearised SW.

t) = a(z )ei(k1 x+k2 y−ωt) and inserting in the equation there results: azz − |k|2 a = 0. and linearised free surface conditions: kinematic surface condition ∂t η = Φz and dynamic linearised surface condition ∂t Φ + g η = 0 z = 0. With the boundary condition at bottom (16). periodic in the horizontal directions. y. i. both η and Φ of the form ei(k1 x+k2 y−ωt) . y. z. Hence Φ = A sinh(|k|(z + H ))ei(k1 x+k2 y−ωt) . The general solution for Φ then has the form Φ(x. take the boundary in the Laplace problem to be the undisturbed level z = 0. we look for time harmonic solutions. which is Φz |z =0 = A|k | cosh(|k|H )ei(k1 x+k2 y−ωt) Writing η (x.e. we retain only linear terms in the surface equations. (16) (17) (18) To ﬁnd the dispersion relation. and. Then the full set becomes Laplace problem on strip: ∆Φ ≡ Φxx + +Φyy + Φzz = 0. For the kinematic surface condition (17) we need to calculate Φz at z = 0. t) = 0 at z = −H.We now present the details of the derivation. we suppose that the wave elevation η is small. which requires a(z. moreover. First we have to consider the linearised equations. Then consistently in lowest order. at z = 0. and just as well the ﬂuid velocity ∇Φ. This means. t) = Bei(k1 x+k2 y−ωt) the kinematic boundary condition requires: −iωB = A|k| cosh(|k |H ) (19) 24 . the solution becomes: a(z ) = A sinh(|k |(z + H )) with A arbitrary. with boundary condition at bottom: −H < z < 0 (15) Φz = 0 at z = −H.

often used approximation is the so-called Boussinesq approximation. p gk for k → ∞. Further restricting to waves running mainly in one direction. mostly the restriction is to ‘long’ waves. The general form of these equations is a ﬁrst order in time equation of the form 3 (21) ∂t u = −∂x (Ru + u2 ) 4 3 Korteweg-de Vries equation (1895) Korteweg and de Vries derived in 1895 a model equation for the motion of waves on the surface of a layer of ﬂuid above a ﬂat bottom. the case of ‘rather small. which corresponds to the speciﬁc relation between the wave amplitude ε and wave length λ given by 1/λ2 ∼ ε. B only can give a nontrivial result provided ω 2 = g |k | tanh(|k |H ). Restricting to rather low. This is the basic assumption to arrive at what are called Boussinesq-type of equations. numerical calculations have to be used to ﬁnd the subsequent wave proﬁles. assumptions on the characteristic wave length are commonly made. ﬁrst order in time. then leads to KdV-type of equations3 .4 KdV type of equations One step further than the linear approximation. they derived the equation (21) that now bears their name with RKdV (k). 25 . the odd solution should be taken and there results the dispersion relation r g tanh(kH ) ω=k k Note the limiting behaviour. is to take for small amplitude solutions a ﬁrst order nonlinear eﬀect into account. 2.while the dynamic boundary condition gives −iω A sinh(|k|H ) + gB = 0. which is the desired dispersion relation. It also became clear that many problems in physics and technics are modelled by this equation. ω ∼ (20) In more dimensions. This equation became well known in the sixties since it turned out that from a mathematical point of view it was the ﬁrst partial diﬀerential equation shown to be completely integrable. rather long waves’. allowing the wave number to have diﬀerent signs. the positive value of ω is taken for waves travelling in the wave vector direction. for arbitrary initial proﬁles. the complete integrability makes it possible in principle to write down the time-asymptotic proﬁle. and the negative sign for waves in the opposite direction. At the same time. for small wave numbers (long waves) and large wave numbers (short waves) p ω ∼ k gH for k → 0. A characteristic. Being an evolution equation. given the wave vector. These equations describe both waves running to the right and the left. leading to a huge extension of the theory of nonlinear pde’s. rather long waves. These two conditions for the coeﬃcients A. the initial value problem requires to ﬁnd the evolution of the surface proﬁle from a given initial proﬁle. In one space dimension. This initial value problem for KdV is not easy to solve.

2 the dispersion is correct for long waves. 26 . i. in which a series expansion in the steepness of the wave. In view of the rather good performance of the model (21). given by ∂t u = −∂x δ u H (u) where the symplectic operator is the diﬀerential operator ∂x and the Hamiltonian is now given by ¸ Z · 1 3 1 uRu + u dx H (u) = 2 4 In [4] a derivation of the KdV-equation is described using the variational structure as the guiding principle to obtain consistent simpliﬁed models. as will be shown in the next sections.for the normalised surface elevation u. An improvement of the dispersive properties can be achieved by taking for R the exact dispersion operator for linear waves. The quadratic nonlinearity in (21) is the second order term from the quadratic character of the original problem. the operator is given by 1 1 2 RKdV (k ) = 1 + ∂x ' 1 − k2 6 6 where the symbol ' means the action of the operator in Fourier space. For the classical KdV equation. the so-called Stokes expansion. For modelling waves of shorter wave lengths. as stated above. p R ' tanh(k)/k Then the operator is no longer an ordinary diﬀerential operator and the equation (21) is no longer a standard partial diﬀerential equation.. a model should retain the Hamiltonian character of the full SWE’s. and various improvements have been proposed and studied (for instance Pade-approximations). since Ω(k) = k (1 − 1 6 k + . the KdV-type of equations have a Hamiltonian structure too. Indeed. However. Then the operator is a local operator. Here R is a linear operator that determines the speciﬁc dispersive properties. Finally. Then the equation has also a dispersive operator in the quadratic term.e. With this choice. Exercise 22 The KdV-eqn in normalized form: ∂t u + uxxx + uux = 0. that for RKdV the phase and group velocity become negative for shorter waves. we restrict to the non-dispersive quadratic nonlinearity for simplicity. almost up to the point of breaking waves. as in hydrodynamic laboratories. very diﬀerent from the exact relation for which both tend to zero for shorter waves. the speciﬁc form is just a simpliﬁcation and higher order terms could be added. essentially ka. In the next subsection we will see explicitly the importance of the choice of the dispersion relation. Observe. for instance.) for k → 0. is taken as parameter. In ﬁnite diﬀerence approximations of (21) this non-local behaviour is a serious problem. since the linearised equation has dispersion relation ω = kR(k). in the Boussinesq approximation. and KdV is a partial diﬀerential equation. This is actually done in an alternative model. this improvement turns out to be essential.

In view of the applications of propagation in wave tanks.can be written as a conservation law: 1 ∂t u = −∂x [uxx + u2 ]. (Benjamin. Without the nonlinearity. Determine these conserved quantities. three relevant conserved densities are given below. and with that of the standard-KdVequation. write down the equation for the proﬁle function f . 2. 2 and so u is a conserved density. Then the spatial evolution is expressed in terms of temporal derivatives. This equation can be derived in a certain approximation as follows. 2. 1. What is the relation with the dispersion relation of the full SW eqn’s.5 Focusing/defocusing NLS-models In this subsection we derive the equation for the envelope of a wave group. this is the convenient description for a signalling problem: sending waves from a speciﬁed point (the wave maker) and investigating the down-stream evolution. the global conserved quantities are the integrals over the whole real line (or with the integrals over one period. In fact there are more conserved densities. we will sketch the derivation of the spatial NLSequation. Looking for travelling waves. Bona & Mahony. t) = f (x − V t). Determine the dispersion relation of the linearised equation. mass 2 f (u) = uxx + 1 2u e(u) = u2 momentum 1 3 f (u) = 2uuxx − u2 x + 3u 1 3 1 e(u) = u2 x − u . Exercise 23 Consider the socalled BBM eqn. 1972): 2 )∂t u = −∂x u − u∂x u (1 − ∂x (22) This model equation is a variant of the KdV eqn (normalised variables). in which the evolution parameter is in the down stream direction (instead of in time). do you recognise this (form of the) equation? Find the solution explicitly. the next equations are equivalent ∂t u + iΩ(i∂x )u = 0 ⇔ ∂x u − iK (−i∂t )u = 0 27 . Verify the given corresponding ﬂuxes: e(u) = u. energy 2 6 1 1 2 1 4 f (u) = − u2 + ux uxxx + u2 x u − u uxx − u 2 xx 2 8 For solutions decaying (with all derivatives) at inﬁnity (or for periodic solutions). respectively). u(x.

We will do this in some detail. with iΩ = ∂x R as in (21). 2. assume in lowest order that R ≈ 1. where K is the inverse of Ω : K (ω) = Ω−1 (ω ). leading to ∂x u − iK (−i∂t )u − ∂t (u2 ) = 0. Now. 28 . i. in this approximation we can replace ∂x (u2 ) in (21) by −∂t (u2 ). t) = α(ω )ei[K (ω)x−ωt] dω where α is the spectral function of the ﬁeld at x = 0. ζ = x. To eliminate the ﬁrst order term in the dispersion. Taking an initial spectrum sharply centred at ω ¯ . a frame moving with the group velocity 1/K 0 (¯ ω ) is introduced. Starting point is a given time signal at the initial position. t) = A(x. so that approximately ∂t u + ∂x u = 0 Then. u(x. given by ω) + cc. since the precise relation with the actual wave height u contains some intricate phase eﬀects which are important. τ = t − K 0 (¯ ω )x. (23) where we removed the factor 3 4 for convenience (by rescaling u).5. t) = α(¯ ω + ν )ei[(K (¯ dν 0 and satisﬁes the linear dispersive equation ∂x A − i [K (¯ ω + i∂t ) − K (¯ ω )] A = 0.1 Linear dispersive envelope equation First we deal with dispersive eﬀects upon neglecting the quadratic nonlinearity. The derivation given here follows the description in [6] (apart from some change of notation and corresponding changes in signs). The result will be an NLS-equation with coeﬃcients that are determined by the speciﬁc dispersion relation. centerd at a central frequency ω ¯ . The complex amplitude that describes the modulation is given by Z ∞ ω+ν )−K (¯ ω))x−ν t] A(x. The general solution of the linearised equation can be written down as Z u(x. the wavenumber now taken to depend on the frequency: k = K (ω) ⇔ ω = Ω(k ).e. Then we include eﬀects of the nonlinearity.where the function K is the inverse of Ω. the resulting wave group is a modulation of a harmonic mode. t)eiΘ(¯ where here and in the following we use the notation Θ(ω ) to denote the phase of a mode with frequency ω satisfying the dispersion relation Θ(¯ ω ) = K (¯ ω )x − ω ¯ t.

and the variations in the equilibrium level should satisfy at each position x Z Cdt = 0. ∂ζ A + iβ∂t (24) (25) What is the relevance of the sign of β in this equation? This is a simple dispersive wave equation. the generation of a second order double harmonic bound wave and a variation of the equilibrium level is anticipated: ω) ω) + Be2iΘ(¯ + C + cc. constant of the motion)? 3. in contrast to the so-called diﬀusion equation (when β > 0) for the real function u 2 ∂ζ u = β∂t u R (a parabolic pde) is dissipative: Calculate for instance ∂ζ (∂t u)2 dt and show that it has damping solutions. very diﬀerent from the equation (25) which is not dissipative but dispersive. the linear envelope equation (24) becomes 2 A = 0. Show that K2 is second order in ν : K2 (ν ) = βν 2 + O(ν 3 ). square integrable in time (decaying at inﬁnity) or are periodic with some period T . So. and B and C are of second order in the amplitude. This coeﬃcient is called the group-velocity dispersion coeﬃcient . B and C are allowed to vary slowly in the frame of reference. 2.5.2 Nonlinear contributions To incorporate nonlinearity in the deformation of the envelope amplitude. Taking only the second-order approximation. R In the latter case udt = 0. Show that.Writing ω + ν ) − K (¯ ω) − K 0 (¯ ω )ν K2 (ν ) = K (¯ the equation can then be rewritten like: ∂ζ A − iK2 (i∂τ )A = 0. determine β and show it is positive for surface wave dispersion. can you explain why? 2. Exercise 24 Linear envelope equation 1. It is natural to restrict to solutions u which are. 29 . what is te dispersion relation? Is there a conservation property (local conservation law. either. u = A0 eiΘ(¯ (26) The amplitudes A0 .

(29) The phase shift indicates that the physical solution should actually be seen in a frame of reference moving at a slightly adjusted speed. (28) One can simplify this result by getting rid of the term 4iσ 0 I0 ω ¯ A0 in the equation by introducing an additional phase in A0 : ¯ σ 0 I0 ζ A0 = Ae−4iω leading to the generalised NLS-equation. This deﬁnes the second order coeﬃcients Z £ ¤ 2 2 B = 2σ 2 A0 .Inserting the Ansatz in the equation gives a residue that has to be made as small as possible. This may be less relevant for the envelope equation since then. Then. this eﬀect is part of the nonlinear dispersion relation. for A0 there results the equation ¯ A0 − iK2 (i∂τ )A0 + iγ |A0 |2 A0 = 0 ∂ζ A0 + 4iσ 0 I0 ω with γ = 4¯ ω (σ 0 + σ 2 ) . to avoid resonant terms in third order. we have retained the full dispersive properties of the problem. while (21) is valid for waves with a broad spectrum. the actual physical solution is then up to and including second order given by £ ¤ ω ) − 4¯ ω σ 0 I0 ζ ] + cc (30) u = A exp i [Θ(¯ ω ) − 4¯ ω σ 0 I0 ζ ] + σ 0 |A|2 − I0 + 2σ 2 A2 exp i2 [Θ(¯ This can be rewritten in a more attractive way by writing the complex amplitude A with real amplitude a and phase φ like A = aeiφ . retaining the 30 . However. In the envelope equation above. From a solution of this equation. (31) u = 2a cos(Θ(¯ ω) + φ − 4¯ ωσ 0 I0 x) + 2σ 0 [a2 − I0 ] + 4a2 σ 2 cos(2(Θ(¯ The non-harmonic term with σ 0 modiﬁes locally the equilibrium level of the ﬁeld (keeping the averaged value at zero). and C = σ 0 |A0 | − I0 with I0 = |A0 |2 dτ where σ0 = 1 ω ¯ . then ω ) + φ − 4¯ ω σ 0 I0 x)). the attention is to waves with a sharply peaked spectrum. just as in the KdV-type of equation in the previous subsection. denoted by gNLS: ∂ζ A − iK2 (i∂τ )A + iγ |A|2 A = 0. while the double harmonic term with σ 2 deforms the ﬁrst-order harmonic proﬁle to a cnoidal-type of proﬁle. σ2 = 0 − K (0) 2K (¯ ω ) − K (2¯ ω) (27) K 0 (¯ ω) are (transfer) coeﬃcients from the generation of the second harmonic and non-harmonic term in the solution. from the start on.

e.full dispersion makes it possible to study the inﬂuence of truncating the dispersive properties. In this case. The speciﬁc value of ωcrit is related to the critical for ω ¯ < ω crit and positive for ω wave number known as the Davey-Stewartson value. the waves constituting a wave group have too little interaction to remain conﬁned in a group. and also for the classical KdV long wave dispersion. it is custom to expand the dispersion operator to second or third order. or better sign(β 2 γ ).g. a well known equation that has been studied extensively (e. [1]). being negative ¯ > ωcrit . determines the character of the NLS equation: • defocusing NLS if sign(β 2 γ ) < 0. γ 0 = V0 γ 31 2 A + iγ 0 |A|2 A = 0 (∂t + V0 ∂x )A + iβ 0 ∂x . the value of γ changes sign. β = 1 K 000 (¯ ω). the quadratic function K2 is even. This has as physical interpretation that when the wave length of the carrier wave is suﬃciently large. the dispersive and the nonlinear eﬀects counterbalance eachother. However. for the surface wave dispersion. This means that essentially diﬀerent behaviour is found for these cases. and then A satisﬁes where β 2 = 1 3 2 K (¯ 6 2 3 ∂ζ A + iβ 2 ∂τ A − β 3 ∂τ A + iγ |A|2 A = 0. In contrast. leading to the defocusing NLS. K2 (ν ) ≈ β 2 ν 2 + β 3 ν 3 00 ω ). (32) Taking β 3 = 0 and performing a simple scaling transforms this equation to the standard form of the NLS-equation 2 ∂ζ A + i∂τ A + isign(β 2 γ )|A|2 A = 0. For the problem under consideration it is important to realise that the sign depends on the value of the central frequency. the sign of β 2 is positive for each ω ¯ . the value of γ is negative for all ω ¯ . in laboratory coordinates the NLS equation reads N LS − spat : (∂x + N LS − temp : 1 2 ∂t )A + iβ∂t A + iγ |A|2 A = 0 V0 In the same way we can derive the temporal NLS: β 0 = V03 β . with the third order dispersive term β 3 included. Summary 25 For the spatial NLS as considered above. The focussing NLS has soliton-type of solutions and more ’conﬁned’ solutions. the equation is known as the Dysthe equation ([3]). In fact. kcrit ≈ 1. as we shall see in the next section. i.363. Indeed. The sign of the coeﬃcients. while in the defocussing NLS the waves will spread. as can be easily veriﬁed. when the long-wave dispersion as in the classical KdV equation is used. for the full dispersion relation. and • focusing NLS if sign(β 2 γ ) > 0. while for suﬃciently short waves the interaction length covers several waves which leads to grouping behaviour. For this standard NLS.

t)) + hot all of the above can also be written in terms of the p • real amplitude a(x. in many cases one uses the ’energy’ E = |A|2 .5. to be determined from the phase. (deﬁning the envelope) . t) = 2a(x.2. in particular. dispersion transfer-coeﬃcient E=a2 V (k ) = Ω0 (k ) . The Variational formulation is found from the Lagrangian functional Z Z 1 4 L= [(ω − Ω(k))a2 + β a2 x − γ a ]dxdt. t) k := ∂x Φ ω := −∂t Φ Notice that here we deﬁne the wavenumber and frequency as local quantites. which has been written above like ω − Ω(k) = γ E + β 2 a ∂x a 32 .vel. 00 β = −1 2 Ω (k ) > 0 γ>0 Remark 26 These equations have a variational structure. t) ∗ cos(Φ(x.3 Writing Phase-amplitude equations for NLS u(x. the phase-amplitude relations: amplitude (energy) phase equation kinematic relation ∂t E + ∂x (V (k )E ) = 0 2 a ω − Ω(k ) = γ E + β ∂x a ∂t k + ∂x ω = 0 where Ω is the standard linear dispersion relation. 2 Observe. Then NLS can be found from the follwoing basic relations. namely the phase local wave number local frequency Φ(x. t) = k0 x − ω 0 t + φ(x. and • quantities that deﬁne the carrier wave. and energy groupvelocity gr. the expression for the Phase equation. t) = |A|2 .

the solutions may be very complicated. with ν = ω − Ω(k).This is to be interpreted as an extension of ‘nonlinear dispersion relation ’ (NDR) which reads for one mode with amplitude q ω − Ω(k) = γ q 2 The present modiﬁcation results from the degeneracy caused by the ‘arbitrary’ choice of 2 a k0 . independent from magnitude but locally deﬁned and dependent on the whole envelope. When ν is constant. Remark 27 The NDR can be written in an alternative form like 2 β∂x a − ν a + γ a3 = 0. more important. This is at the heart of the complicated dynamics of surface waves!! 33 . the solutions are easy to ﬁnd. for which ν = ω − Ω(k) constant. pseudo-periodic and even chaotic behaviour. ω0 . As we shall see later. showing periodic.but. when non-constant. This form is particularly interesting when looking for steady state solutions. NDR now includes an ‘envelope-proﬁle’-dependence β ∂x a from dispersion (β 6= 0). this equation has as simple mechanical analogy the dynamic (now ’time’ is x) evolution of a particle with mass β in a potential force ﬁeld as described by Newton’s equation.

deviations: an (almost periodic) amplitude increase and wave proﬁle distortions that are large on long time or space intervals. This is the basis of the so-called Lindstedt-Poincaré method. but just as well in the frequency. Both for KdV and NLS there are opposing eﬀects: the dispersive eﬀect that tends to inrease the spreading of waves (diﬀerent phase velocities for diﬀerent wave lengths) and the nonlinear eﬀect that tends to steepen the waves. makes it possible that they counterbalance. The total solution is therefore quite complicated. however. the Duﬃng equation. i. which can be written down explicitly for the standard KdV-equation. rather long waves’ in the Boussinesq approximation. Then some explicit solutions will be derived for the NLS equation. but clearly noticeable.1 Solitons of KdV The motivation for Korteweg and de Vries to study the problem of surface waves. only two frequencies are involved initially. with the result that waves (or wave groups) of ’permanent’ form (only for a special form) become possible. Although these examples may easily give the impression that explicit solutions are common. The combined eﬀect.3 Advanced Surface Wave Phenomena In this section we derive increasingly more complex mathematical methods to deal with more and more detailed and complicated wave phenomena. More realistic waves have a continuous spectrum. The nonlinear eﬀects cause slow. Also the method to remedy this artefact of the expansion method is illustrated and motivated: an expansion not only in the amplitude. and leads to uniformly valid (that is. series expansions are a common tool and are based on an expansion in the ’order’ of some quantity that is assumed to be ’small’ (a quantity measuring the amplitude is a standard choice). A naive perturbation expansion will. the equation that describes the evolution of the amplitude of a wave group. the nonlinear analogue of the linear bi-chromatic solution that is the result of linear interference of two mono-chromatic waves.e. or space) approximations. We start with the well known soliton-solutions. Computer-algebraic tools to deal with the Fourier-manipulations can be designed. We show that. 3. That should be kept in mind in the follwoing. and the interaction takes place for each pair of frequencies (an inﬁnite number). i. it is quite diﬃcult to obtain the simplest but one solution. for al time. leading to an Analytical Wave Code (AWC) that can predict correctly the evolution of an irregular wave (with many initial frequencies) over the large distances that are relevant for hydrodynamic laboratories. lead to resonant terms which will ruin the approximation: the deviation from the exact solution increases linearly with time or space.e. The same method will be used to obtain uniformly valid approximations for the nonlinear evolution of bi-chromatic waves. i. this is illustrated for simple ODE’s. when in a delicate balance as described by these equations. The bi-chromatic wave results from interaction of two mono-chromatic waves. was to settle a dispute that continued throughout the nineteenth century about the existence of travelling 34 . even for the KdV-model. the equation that was derived under the assumption of ’rather small. the opposite is true and solutions in closed form are usually impossible to obtain. but can nevertheless be tackled. In a preparatory subsection.e. To study such kind of problems. then this remedy leads to the so-called nonlinear dispersion relation.

but is merely travells at a ﬁxed speed? They showed. suﬃciently fast at inﬁnity (“almost conﬁned”. that the answer is aﬃrmative. 3. since KdV combines nonlinearity (leading to “breaking”phenomenon) and dispersion (“spreading” of initial proﬁle). We shall see that. The remarkable property is that these combined eﬀects make it possible that there exist travelling waves. This is quite unexpected at ﬁrst sight. we substitute this form in the KdV-equation.waves: is it possible that a wave exists that doesn’t change in time. leads to the second order ode for the proﬁle: −V f (ξ ) + 1 2 f (ξ )2 + ∂ξ f (ξ ) = 0 2 35 . the larger the velocity. a solution of the form η (x. 2 Then the pde becomes an ode for the function f in which V enters as a parameter to be determined together with the proﬁle. the higher the amplitude. in fact. for which f is a periodic function of ξ . say f .1 Analysis of solitary wave proﬁles To ﬁnd the solution we have to distinguish two cases: • space-(and time-) periodic solutions. for certain V . That is. We will concentrate on the solitary wave proﬁles. waves with a speciﬁc proﬁle. More so. and • solitary wave solutions : wave proﬁles of a single hump that decay. exponentially small outside a certain interval). noticing that the constant of integration has to vanish as a consequence of the decay at inﬁnity. say V .1. the socalled cnoidal waves (since the proﬁle is expressed with the elliptic cnoidal function). the equation becomes 3 f (ξ ) = 0 −V ∂ξ f (ξ ) + f (ξ ) ∂ξ f (ξ ) + ∂ξ A solution of this equation. produces the wave proﬁle f of the wave that travels undisturbed in shape at speed V . in normalised variables · ¸ 1 2 2 ∂t η + ∂x ∂x η + η = 0. Writing ξ := x − V t. t) = f (x − V t) just as in the translation equation. together with all derivatives. by deriving their (KdV-) equation and analyzing it. there is a whole family of such waves. but now only for speciﬁc proﬁles f and speciﬁc velocity V . that will neither break nor spread (an exact balance between the counteraction of breaking and spreding). and that travel undisturbed in shape at a speciﬁc speed. it is possible to write down the wave shapes and speeds explicitly. Then by integrating the equation above once. To ﬁnd the wave proﬁle f and the velocity V .

at the right for negative values: . 2 Since E should be zero for a solitary wave proﬁle. it is given by f (ξ . 2 2 6 This is a ﬁrst order equation for the proﬁle function and its solution can be given explicitly. Multiplying (33) with ∂ξ f and integrating the equation again. there results: 1 [∂ξ f ]2 + U (f (ξ )) = E. 2 6 The plot of U is qualitatively as shown below. Clearly. with V > 0. this can only be achieved for positive values of V . for the proﬁle equation mechanical-energy conservation holds. at the left for positive values of V . and f as the position. V ) = cosh( 1 2 36 3V √ V ξ )2 . This solution is a solitary wave proﬁle : for each V . we look for the solution that is nontrivial and connects the origin with itself: a homoclinic orbit. the equation describes the motion of a particle of unit mass subject to a potential force with potential energy U according to Newtons law: 2 f (ξ ) + ∂ξ dU =0 df (33) with potential energy 1 1 U (f ) = − V f 2 + f 3 .This equation can be solved in a standard way by observing the mechanical analogue: when ξ is interpreted as the time. Looking for a solitary wave proﬁle f that decays to zero for ξ tending to −∞. In more detail. ∞. . the equation becomes 1 1 1 [∂ξ f (ξ )]2 − V f (ξ )2 + f (ξ )3 = 0.

We restrict the investigation of special solutions to solutions of form A(τ .2 and V = 1. 3. (Note: in laboratory-variables this corresponds to a travelling wave.2 Special solutions of NLS Referring to the previous section. for V = .) 37 . w = ∂ξ f (ξ ) in the following way. The curves of constant energy. w). and the width proportional 1 to √ : the larger the amplitude. and in this phase portrait the solitary wave corresponds to the homoclinic orbit which is the level curve through the origin (for which E = 0). τ which may be time-space variables or conversely: 2 A + iγ |A|2 A = 0. we use the following form of the NLS equation for the amplitude A depending on ζ . γ can be scaled away to arrive at an equation with no parameters. and can therefore be interpreted as ‘standing waves’. are shown below: Observe: with V the velocity. NLS: ∂ζ A + iβ∂τ A simple scaling of ζ . and the larger its speed. the more conﬁned the wave.Two proﬁles. V The solution above can be found in the phase plane z = f. given by 1 2 w + U (z ) = E 2 are sketched in the phase-plane (z. τ shows that the parameters β . Such solutions are a pure harmonic modulation e−iµζ of the ﬁxed proﬁle a(τ ). ζ ) = a(τ )e−iµζ . the amplitude is proportional to V . the signs in this equation only depend on the sign of the product βγ . We will see the relevance in a moment. since NLS is formulated in a moving frame of reference.

This plot gets its meaning from the interpretation of the equation as the phaseequation. since then ’soliton’-type of solutions exist. the parameter µ and the proﬁle a should satisfy the equation β aτ τ − µa + γ a3 = 0. For the waterwave problem. which is a solution if µ > 0 and q= µ γ . potential P (a) = − µa2 + γ a4 ∂a 2 4 2 a=− β∂τ The sign of µ and γ determine the behaviour of the potential.3). where µ deﬁnes the diﬀerence from the linear dispersion plot: µ = ω − Ω(k ) . Having chosen β > 0. which is known as the ’focussing’ NLS. The defocussing case. The study of this equation is. two essentially diﬀerent cases are given by γ > 0 and γ < 0. We will restrict to the case γ > 0.1 Nonlinear harmonic The q simplest solution is a constant amplitude a(τ ) = q . γ < 0 does not have such solutions. approx. 1. the coeﬃcient γ is indeed positive but the value of µ in the NLS-equation depends on whether the local wave number and frequency are above or below the curve of the linear dispersion relation. for suﬃciently large k (larger than the Davey-Stewardson value. 3. In the plot below we sketch the potential proﬁles for the focussing case γ > 0. just as for the investigation of the KdV-soliton.2. as described in section 2. simpliﬁed by looking at its Mechanical analog: the equation is Newtons Law for a mass particle of mass β (> 0) in a ﬁeld with potential P : 1 ∂ 1 P (a).In order to be a solution. Then 38 .

2. the real amplitude crosses the origin. leading to plots for the real part of the complex amplitude as below: 39 .2 Nonlinear modulated harmonic For µ > 0. the NLS-solution is sketched below: 2 3. In a plot of the real part as function of ζ .2.3 Soliton For µ > 0 soliton solution as homoclinic orbit for µ = γ q2 /2: r γ a = qsech(q τ) 2β 3.2. small amplitude periodic motions around minimal potential energy exist: 3.4 Nonlinear bi-harmonic For µ < 0 only periodic solutions exist.A = qe−iγ q ζ . The constant amplitude corresponds to the point of minimal potential.

In most cases this is not possible but here we can exploit the standard analysis for Newtonian systems: x ¨=− ∂V ∂x (35) (34) where V is the potential energy. the solution is simply the motion of the harmonic oscillator: x1 (0) = ² cos t.3 Intermezzo: Uniform expansion techniques In this subsection we consider asymptotic methods for nonlinear equations. all solutions have the same period. phenomena as ‘mode generation’ and ‘resonance’ are essential in the inﬁnite dimensional problems as well. For (34). Note that for the linear oscillator. Neglecting the nonlinearity. the potential V is given by: 1 1 V2 (x) = x2 + x3 2 3 while for (35) the potential is given by 1 1 V3 (x) = x2 + x4 . independent of amplitude. y ˙ = 0. Exercise 29 Before discussing perturbation methods. 2 4 Sketch the graph of these potentials and investigate the phase portraits of solutions. as we shall see in the following subsections. investigate the global properties of solutions by phase plane analysis. Conclude from this which solutions will be bounded for ALL time. x ˙ (0) = 0. In particular. and y ¨ + y + ²2 y3 = 0 This shows more clearly that the contribution of nonlinearity is ’small’ (as long as y remains bounded). both with the initial condition x(0) = ².3. 2π . but now the ’small’ parameter ² appears in the equations: y ¨ + y + ²y2 = 0. 40 . We restrict the analysis to ode’s but all ideas and methods are equally useful for the perturbation analysis of pde’s. As classical examples to which the ideas can be best illustrated we consider the following two nonlinear extensions of the harmonic oscillator (in suitably deﬁned normalized coordinates): the second order nonlinearity x ¨ + x + x2 = 0. Remark 28 Writing y = ²x the initial values are normalised: y(0) = 1. and the third order nonlinearity (Duﬃng’s eqn) x ¨ + x + x3 = 0. For small ² this may well be a good approximation of the nonlinear equations also.

the phenomenon of resonance appears.We conclude that for (34). The solution of this linear eqn can be found explicitly. . t) = ² cos t + ²2 x2 (t) + ²3 x3 (t) + . and observe that (how) the period of the solution depends on the amplitude (i. It can therefore be expected that perturbation-methods will also have to be able to vary the frequency with ². even if the forcing Fk is bounded. Exercise 31 For small-amplitude solutions. the period will be close to that of the linear equation. we try to ﬁnd a more accurate solution by substituting a series expansion for the solution: x(². resp. when the forcing is periodic with the same period as the solution of the homogeneous eqn. This can. will the period be larger or smaller? (Can you write down a formula for the period and analyse this?) Naive perturbation method With ² cos t as a good approximation. however. Substituting this series in the equation leads for each xk . Resonance 41 . all solutions with amplitude ε smaller than a critical value are periodic. ω (x)2 = 1 + x2 for (34) and (35) respectively. however. be expected by observing that both equations can be written like x ¨ + ω (x)2 x = 0 with ω (x)2 = 1 + x. The smooth distortions of the integral curves near ε = 0 compared to the circles of the harmonic oscillator. Exercise 30 Use MAPLE to actually calculate several solutions numerically. while for (35) even all solutions are periodic. in contrast to the linear equation for which every solution has period 2π . gives additional conﬁdence in the regularity of the problems. x ˙ (t = 0) = 0) . One phenomenon that is not directly visible from the phase plane analysis is the fact that the periods of the periodic motion depend on the amplitude (on ²). the solution does not have to be bounded. . j ≤ k − 1 and is therefore known in a successive treatment. take diﬀerent initial conditions (for instance: take x(t = 0) = ε. What can you say about the period for larger amplitude solutions. Here the period depends on the solution and hence is not known in advance. k ≥ 1 to an equation of the form x ¨k + xk = Fk where Fk depends on the functions xj . on ε) of the solution. which is 2π . plot the solution as function of time.e.

The unboundedness is a consequence of the forcing with the same period as the natural frequency ω 0 of the unforced system. this phenomenon is called resonance. f (t) sin ω 0 t dt = 0. Then. and the requirement must be that the coeﬃcients in front of the term with eı ω0 t vanish. the operator is symmetric with respect to the integral inner-product on the set of T0 -periodic functions Z T0 Z T0 v(t) L u(t) dt = L v(t) u(t) dt 0 0 42 . Then the left hand sides vanish. Then the equation can be written like Lu = f Furthermore. Remark 33 Fredholm alternative (analogy with Linear Algebra) A deeper way to interpret the conditions is as follows. Quite diﬀerent is the situation for ω = ω 0 : a t sin ω0 t . Resonance can be useful to obtain solutions of large amplitude (when pushing a swing) but is often undesirable. or. given a force f . and is forced to be periodic with the same period of the driving force. demanding the right hand sides to vanish. where T0 is the fundamental period T0 = 2π / ω 0 . u(t) = 2ω0 then the solution is unbounded. Since sin ω0 t.The explicit solution. 0 0 These conditions on f are known as solvability conditions. This leads to the conditions above as being necessary. of u ¨ + ω2 0 u = a cos ω t is given for ω 6= ω 0 by u(t) = ω2 0 a cos ω t. Write the equation in operator form by introducing the operator L such that L u ≡ u ¨ + ω2 0 u. multiply the equation with sin ω 0 t and cos ω 0 t and integrate over a fundamental period T0 . − ω2 this shows that the solution is bounded. Z T0 Z T0 f (t) cos ω 0 t dt = 0.e. cos ω 0 t are solutions of the homogeneous eqn. stated diﬀerently. Remark 32 A somewhat diﬀerent way to derive these conditions is as follows. how can we investigate if resonance occurs in the equation u ¨ + ω2 0 u = f (t). Now. f can be expanded as a Fourier series. can we give conditions on f that guarantee that the solution is uniformly bounded for all time? The answer will be clear if f is periodic with period T0 . growing linearly in time. i.

F (ω ) = 1 2π Z f (t)eiωt dω The last integral seems to have a singularity at ω = ω0 . but square integrable. Fourier-integral methods can be applied. write ω = ω 0 + σ . dσ = F (ω0 + σ )e dσ = e e dt −iσ we ﬁnd the explicit solution z (t) = e −iω0 t Z t eiω0 s f (s)ds. i. Then observe Z F (ω 0 + σ ) −iσt −iω0 t z (t) = e dσ e −iσ and. This shows that now NO resonance will appear: the solution remains bounded even when t → ∞. Let’s use complex notation for simplicity: z ˙ + iω 0 z = f (t) Writing f (t) = the solution is given by z (t) = Z F (ω) e−iωt dω −iω + iω0 Z F (ω )e−iωt dω .. 43 . z ˙ + iω 0 z = Ae−iω0 t =⇒ z (t) = Ate−iω0 t which also follows from the variation of constant formula: Z t −iω0 t z (t) = e eiω0 s Ae−iω0 s ds = Ate−iω0 t . cos ω0 t } Then the solvability conditions are nothing but the condition that the force f is perpendicular to the kernel of L.e. Compare this with the Fredholm alternative in Linear Algebra: for (symmetric) matrices A the solvability condition for Ax = b is that b ⊥ ker(A∗ ). Of course. Remark 34 In case f is not periodic. with Z Z Z d F (ω 0 + σ ) −iσt −iσ t iω 0 t F (ω )e−iωt dω = eiω0 t f (t). known as the ’variation of constant formula’. and F (ω 0 ) 6= 0. To investigate the behaviour near ω 0 . when the driving is periodic.The kernel of L is given above: ker L = { sin ω 0 t. resonance appears as we have seen above. even when the fundamental frequency ω0 is present in the driving force.

and use Fourier series: Z X π 2π 1 T −i 2 t. Make a T −periodic continuation of the signal. t ∈ [0. provided ˆ0 (n. T ].n dt s(t) = cn e T .n ∆ω n 2. Extend the signal by zero outside the given time-interval. 44 . In a practical situation. t ∈ / [0. T ]. Z Z T 1 1 s0 (t)eiωt dt s0 (t)eiωt dt = 2π 2π 0 Approximating the integrals by Riemann-sums.t. cn = T 0 The relation between the methods will be clear now: the choice ∆ω = agreement. Then. ’given’ only for a ﬁnite time. say s(t). s0 (t) = 0. and therefore make the follwoing remarks.The explanation of this seemingly discrepancy is the diﬀerence in energy-content in the resonant forcing: the energy is ﬁnite for square integrable functions. t ∈ [0.n s(t)ei T t. we are dealing with (results of) signals that were measured. and hence inﬁnitesimally small per unit time. This is also seen from looking at the increase in ’energy’ over a time interval T : Z T Z T 2 2 2 ∂t |z (t)| dt = 2 Re z (t)f (t)dt. while the energy is inﬁnite for the periodic forcing (ﬁnite per unit time).∆ω)e−i∆ω. for which s0 (t) = s ˆ0 (ω ) = Z s ˆ0 (ω)e−iωt dω . We will have to deal with this in the anlysis for the NDR for nonlinear wave equations furtheron. in principle there are two ways to continue the mathematical reasoning: 1. This then leads to an L2 function ½ s(t).∆ω ) · ∆ω cn = s for ∆ω = 2π T 2π T leads to complete which shows the appearance of the length T of the time interval in the relation between the discrete and continuous spectra. T ]. there results X s0 (t) = s ˆ0 (n. |z (T )| − |z (0)| = 0 0 Remark 35 In practical situations the above may lead to some confusion when energy content is not taken into account properly.

since the third order term. However. we would even like to have an approximation that holds for all time. in fact for t in any interval of prescribed length.3. an approximation with this property is called a uniformly valid approximation. there results x ¨1 + x1 = 0. 8 32 cos t indicates Interpretation. 45 . Remark 36 From symmetry of the equation (if x(t) is a solution with x(0) = ². The forcing of the eqn for x3 can then be expanded (after some straightforward calculations) to 3 x3 1 = (cos t) = 3 1 cos 3t + cos t 4 4 3 4 The solution can be written down. t) should be odd in ². . but the appearance of the source term that resonance will appear: 3 1 x3 (t) = − t sin t + cos 3t 8 32 Hence. T ]. . this is a good approximation for ² suﬃciently small. x ˙ (0) = 0 Naive perturbation technique Substituting the series expansion x(². the solution reads 3 1 x(². and a same result holds for the other terms. t) = ² x1 (t) + ²2 x2 (t) + ²3 x3 (t) + . x(0) = ². x ˙ 1 (0) = 0 x ˙ 2 (0) = 0 x ˙ 3 (0) = 0 Clearly the solution for x1 is given by x1 = cos t. Clearly the approximation constructed above does not satisfy this criterion. . x1 (0) = 1. x ¨3 + x3 = −x3 1 . in practice we want to go further and have a solution that is also correct for larger values of time. . For small values of t > 0. up to third order. and hence all terms ²k xk (t) with k even should vanish. x3 (0) = 0. t ∈ [0. in the equation. t) = ² cos t + ²3 [ − t sin t + cos 3t ] + . is bounded by constant ²3 T . . Since the solution is known to be bounded for all time (from phase plane analysis). then y(t) ≡ −x(t) is the solution with y (0) = −²) it follows that the solution x(². The solution for x2 vanishes identically: x2 ≡ 0. x2 (0) = 0. x ¨2 + x2 = 0.1 Third-order nonlinearity: Duﬃng’s eqn We ﬁrst consider the oscillator with third order nonlinearity Eqn ≡ x ¨ + x + x3 = 0.3. since ²3 t will explode for t 7→ ∞ at ﬁxed ². also t 7→ ∞. although increasing linearly in time.

one now chooses A. such that the approximation satisﬁes the equation as good as possible (variation of ‘parameters’). for reasons to become clear soon. the ‘amplitude’ and ‘phase’ are not uniquely deﬁned. since this is caused by the resonance phenomenon in the equation for x3 . A and α have to satisfy A cos(α) = 1. ‘averaging technique’(van der Pol. Let us now investigate the details. All improved perturbation techniques are based on creating more ﬂexibility in the Ansatz for the solution so that this resonance can be prevented. In order to choose the constants such that the initial value is satisﬁed. Observe that in this Ansatz. which will be called the amplitude-phase method. and the phase θ is simply θ = t + α.3. One improved technique is as follows. θ in such a way that Rc and Rs vanish. We will demonstrate one technique. t) cos(θ(². and even ‘multiple time scale’ method (although in this problem no two time scales can be distinguished). ‘renormalization’. or Krylov-Bogoliubov method).2 Improved perturbation techniques The naive method above (series expansion in ²) does not give a possibility to prevent the linear growth in time.3. Substitution in the eqn leads to Eqn = ² Rc cos(θ) + ² Rs sin(θ) − ²3 with ˙ 2 + A + ²2 3 A3 ¨− Aθ Rc = A 4 and ˙ − A¨ ˙θ θ Rs = −2A In order to reduce the residue as much as possible. t)) into the equation. it is often referred to as Lindstedt-Poincaré technique. A straightforward generalization of this form of solution to the nonlinear system would be to let A and θ depend on time (and ²) in a way to be decided. just as above. where α is a phase constant. t) = ² A(². and is closely related to other methods like ‘uniformization’. will lead to equations for A and θ which may then be solved approximately. but not unique. Neglecting the nonlinear term. α = 0 is one solution. the solution is a simple harmonic function: x(t) = ² A cos(θ) where the amplitude A is constant. Substituting the ‘Ansatz’ x(². Of course. A = 1. this leads to the socalled phase-amplitude equations ˙ 2 + A + ²2 3 A3 = 0 ¨ − Aθ A 4 46 1 3 A cos(3 θ) 4 .

]t + ε2 ξ 2 + ε3 ξ 3 + .) Remark 37 Compare the solution obtained above with the solution from series expansion. the approximate solution becomes x(². which can be summarised as follows. (To determine ω 4 one ﬁrst needs to ﬁnd the third order contribution explicitly.3. with ω (²) = 1 + 3 2 ² . ε5 t) which is smaller than O(ε) for t < ε−4 = (ε−2 )2 . a linear dependence on time for θ. The improved approximation has error O(ε3 . In higher order terms. 3... i. hence the solution constructed above reads µµ ¶ ¶ 3 2 4 x(². θ 8 i. other resonances will turn up. t) = ² cos 1 + 8 ² + ω 4 ε + O(²6 ) t .and ˙ − A¨ ˙θ −2A θ = 0. Taking A=1. t) ≈ ² cos(ω(²) t). is of higher order and does not lead to resonance.e. A solution of these eqns can now be found easily: A independent of time and θ such that ˙ 2 = 1 + ²2 3 A2 θ 4 which leads to ˙ = 1 + ²2 3 A2 + O(²4 ). t) = ² cos ( t)+O(²3 .. a considerably longer time interval. Expand both amplitude and argument of an oscillating function (frequency) in a power-series of ε : • take as ’Ansatz’: ¡ ¢ x(ε. i. ’uniformity’ should not be interpreted as ’valid for all time’. 8 3 The term that remains in the residue.3 Uniform Approximation Procedure The above example may have shwn the line of reasoning. strictly speaking.] cos [ω0 + εω 1 + ε2 ω 2 + . tε3 ). although. by carefully choosing ω4 in x(².. 8 This approximate solution is valid on a much larger interval of time then the approximation with the linearly growing resonant term: x(². In a case like this one often says that the improved approximation is uniformly valid.e.. and observe that the series expansion solution is precisely the expansion in ² of the phaseamplitude solution! This explains the nonuniformity of the series expansion solution. These can then be avoided by adjusting once again in the lowest ¢ ¢ ¡¡the frequency 3 2 4 order solution. t) = ² cos 1 + ² + O(² ) t + O(²3 ).e. doing that leads to a solution that is accurate in order ε on a time-interval t < (ε−2 )3 . t << ε−2 . 47 . 1 4 ² cos(3θ ).The last approximation has order term smaller than O(ε) provided tε3 << ε. it is just valid on a much larger time-interval. t) = [εA1 + ε2 A2 + ε3 A3 + .

i. • expand in powers of ε • equate to zero the coeﬃcient for each power of ε • determine successively the constants Ak and ω k such that the functions ξ 2 (ξ 3 .. somewhat deeper. i..e.. t) = [εA1 + . U = a cos(τ ). with a = γ /π : ¸ · Z d 3 1 2 2 4 2 λ = cos(τ ) dτ /π = 1 + γ .. but actually more natural way. to derive the above results. substitute (simplifying somewhat): ¡ ¢ x(ε. 8 48 . to prevent resonance in ξ : Z Z 2 3 −2A1 ω0 ω 2 cos (ω 0 t)dt + A1 cos4 (ω 0 t)dt = 0.] cos [ω0 + εω 1 + ε2 ω 2 + . dγ 2 the correct result is obtained by approximating the minimising value by the value p of the trial function that is the solution of the linear equation.]t + ε3 ξ and ﬁnd O(ε) : O(ε ) : O(ε ) : 3 2 A1 [−ω 0 + 1] = 0 ⇒ ω 0 = 1 2ω 0 ω 1 = 0 ⇒ ω 1 = 0 3 ¨ − 2A1 ω 0 ω 2 cos(ω 0 t) + A3 1 cos (ω 0 t) + ξ + ξ = 0 hence the same result as above: From this last equation follows the condition for boundedness of ξ ..e. γ+ γ dγ 2 4π 3 2 3 =1+ a hence λ ≈ 1 + γ 8π 8 leading to the correct asymptotically valid ﬁrst order solution ¸ ¶ µ· 3 x(t) = a cos 1 + a2 t + O(a3 ). For the Duﬃng equation. R 1 2 cos4 (ω0 t)dt 3 ω 2 = A1 R = A2 2 2 8 1 cos (ω0 t)dt Remark 38 There is another..• substitute in the equation.. For Duﬃng’s equation this starts with the observation that the equation x ¨ + x + x3 = 0 transforms after a time transformation τ = λt ∈ [0.)..) are bounded (from non-resonance conditions for ξ and ξ 2 . 2π ] to λ2 u00 + u + u3 = 0 Interpreting λ2 as the Lagrange multiplier of a constrained variational problem: ¯Z · ½Z ¾¸ d 1 4¯ 2 2 02 ¯ M in λ = u + u ¯ u =γ .

t) = ² x1 (t) + ²2 x2 (t) + ²3 x3 (t) + . . 49 5 6 x ˙ 1 (0) = 0 x ˙ 2 (0) = 0 x ˙ 3 (0) = 0 −x2 1. With x1 = cos t. Continuing with the third order calculation. Substituting the series expansion x(². x2 (0) = 0. in second order the solution is uniformly bounded. x3 (0) = 0.3.4 Second-order nonlinearity Now consider the oscillator with second order nonlinearity ˙ (0) = 0 x ¨ + x + x2 = 0. x(0) = ². but will now appear in the third order term. the equation for x2 becomes 1 x ¨2 + x2 = −(cos t)2 ≡ − (1 + cos 2t) 2 with solution (satisfying the homogeneous initial data) 1 1 1 x2 (t) = − + cos 2t + cos t 2 6 3 So. t) = ² cos t + ²2 [ − + cos 2t + cos t ] 2 6 3 3 5 t sin t + ( periodic terms) ]. . but the appearance of the source term that resonance will appear: 5 t sin t + ( periodic terms) 12 Hence. x1 (0) = 1. cos t indicates . there results x ¨1 + x1 = 0. the forcing of the eqn for x3 can be expanded (after some straightforward calculations) to 1 1 1 5 −2x1 x2 = − + cos t − cos 2t − cos 3t 3 6 3 6 The solution can be written down. series expansion leads to resonance and then to a nonuniformly valid approximation.3. the solution reads x3 (t) = 1 1 1 x(². x ¨2 + x2 = x ¨3 + x3 = −2x1 x2 . x A Naive perturbation technique will show as follows that resonance is absent in second order terms. in the equation. +² [ 12 Just as above for the Duﬃng eqn. up to third order. Exercise 39 Construct a uniformly valid approximation (up to and including third order) by replacing the ’ﬁrst order’ solution ² cos t by ε cos([1 + ε2 ω 2 ]t) for an appropriate value of ω 2 .

ω) = F (k. the same remarks apply as made in the ODE-case. ω )ei(kx−ωt) dkdω F (k. For nonlinear equations (KdV. ω ) ei(kx−ωt) dkdω −i(ω − Ω(k)) Z Z From this formal expression the ’resonance’ is seen to take place at the dispersion curve ω = Ω(k) in the ω . k-plane when the spectrum of the forcing doesn’t vanish there. t) = F (k. t) = = Z Z U (k. ω )ei(kx−ωt) dkdω and for the equation: [−iω + iΩ(k)] U (k. and improved method) will appear in partial diﬀerential equations.3.3. ω ) and hence u(x. As a speciﬁc example. the same is possible for NLS) this will be dealt with in the next subsection. Formally. t. variable) leads to Z Z f (x. or a ’ﬁnite time signal’. For a continuous spectrum. suppose that the forcing is a solution of the homogeneous equation. then the resonant solution is given by ures = Atei(k0 x−Ω0 t) and increases without bound in time. t) = Aei(k0 x−Ω0 t) with Ω0 = Ω(k0 ). 50 . a double Fourier transform (for spatial. say f (x. Consider the forced equation ∂t u + iΩ(−i∂x )u = f (x. and temporal. t).5 Resonance in dispersive wave equations The same phenomena (resonance leading to non-uniform approximations) and methods (direct perturbation method. x. Here we prepare by investigating resonance eﬀects in linear dispersive wave equations.

4 Distortion of bi-chromatic waves This subsection is for the most part contained in a paper with Edi Cahyono and Agus Suryanto [6]. The ﬁrst order solution is the linear harmonic ω) + cc = 2a cos(Θ(¯ ω )) u(1) = aeiΘ(¯ Then for the second order term there results the equation and corresponding solution ³ ´ (∂z − iK )u(2) = ∂t u(1)2 ω) u(2) = −2a2 σ 2 e2iΘ(¯ + cc. (37) or ∂z u + ∂t Ru − u The spatial parameter is denoted by z instead of x. In linear theory the solution is ω) + cc = 2a cos(Θ(¯ ω)) ulin = aeiΘ(¯ where here and in the following we use the notation Θ(ω ) to denote the phase of a mode with frequency ω satisfying the dispersion relation: Θ(¯ ω ) = K (¯ ω)z − ω ¯ t. The correction to a non-linear dispersion relation reads in this case k = K (ω ) + 4¯ ωσ 2 a2 . We use the notation of that paper. The third order equation ³ ´ (∂z − iK )u(3) = 2∂t u(1) u(2) contains a resonant term which can be dealt with by correcting the dispersion relation. and the way how the ’nonlinear dispersion relation’ is introduced to prevent resonance. ω ¯ where σ 2 = 2K (¯ ω)−K (2¯ ω) is the (transfer) coeﬃcient from the generation of the second harmonic term in the solution. of frequency ω ¯ say. but restrict at the end to the dispersion for surface waves. We start with a straightforward series expansion in the amplitude: u = u(1) + u(2) + u(3) where the function u(k) is of the order k in the amplitude. £ ¤ (36) ∂z u − iK (i∂t )u − ∂t u2 = 0 £ ¤ 2 = 0. 3. can already be seen from an investigation of the simplest example.1 Nonlinear harmonics The eﬀect of the nonlinearity. The standard equation to be used is the spatial KdV-equation.4.3. 51 . an initially harmonic mode.

This leads to a uniformly valid solution given by u = 2a cos[KNDR z − ω ¯ t] + hot. In fact. implicitly.2 Uniformly valid approximate bi-harmonic evolution We now consider the evolution of an initially bi-chromatic wave group. The primary motivation is to study properties of the distortions during the evolution. Θ± = Θ(ω ± ν ). with ν small. ω ) + 4¯ ω σ 2 a2 . Linear dispersive theory would lead to the solution ¡ ¢ ¯ −ω u(z.4. and therfore the correct position of the wave (in particular relevant when the solution is determined over longer distances). The signal at z = 0 is the interference of two mono-chromatic waves of the same amplitude ¯ ± ν . the frequency diﬀerence can be arbitrary. for the standard NLS-equation it was shown in [10] that the distortion of the bi-chromatic wave is determined q : the larger this quotient the larger the by the characteristic parameter which is the quotient ν 52 . slightly diﬀerent phase speed) that depends on the amplitude. t) = qe−iω1 t + qe−iω−1 t + cc = 4q cos(ν t) cos (¯ ω t) . q and slightly diﬀerent frequencies ω ±1 = ω u(0. 3. The eﬀect of the phase correction is that this nonlinear harmonic has slightly diﬀerent wave number (and as a consequence. κ = (K (¯ ¯ k = (K (¯ ω + ν ) + K (¯ ω − ν ))/2 (38) where ¡ ¢ ¯ −ω This solution describes the characteristic modulation of the carrier wave cos kz ¯ t by the envelope 4q cos(κz − ν t) which travels with the group velocity ν /κ. shows that special care is required in ﬁnding accurate analytical approximations. with KNDR = K (¯ where hot denote quadratic and higher order terms. just as well by the frequency diﬀerence. t) = qeiΘ+ + qeiΘ− + cc = 4q cos(κz − ν t) cos kz ¯t ω + ν ) − K (¯ ω − ν ))/2. which show several peculiarities and. This correction for the dispersion relation to a nonlinear dispersion relation (NDR) is important to obtain the correct speed. for this speciﬁc example. It can be observed that the modulation in the moving frame of reference is given by 4q cos(κz − ν t) = 4q cos(δζ − ντ ). but most interesting will be the case of small frequency diﬀerence. but. 1 ω )ν ≈ K 000 (¯ ω )ν 3 where δ = κ − K 0 (¯ 6 The nonlinear eﬀects will be determined in ﬁrst instance by the value of the amplitude q .

Each of these terms is multiplied by a speciﬁc coeﬃcient. Kcorr ∼ 4q 2 ω 4 (40) In [5] it was shown by using exact formula for the NLS-bi-soliton solution that the opposite is the case for the interaction of solitons. The coeﬃcient in front of these terms are inversely frequencies. which are ω ±3 = ω 2 proportional to ν . 53 . s(±) → σ 2 and s0 → σ 0 where σ0 = 1 ω ¯ . the same second order contribution is found. but now there is a nonvanishing long wave component 2σ 0 ei(Θ+ −Θ− ) which has to be taken into account in calculating the third order. In a direct series expansion method. with a = 2q . the ﬁrst order solution is the linear solution ( 38). We will now derive the approximate solution in detail. σ2 = K 0 (¯ ω) − K 0 (0) 2K (¯ ω ) − K (2¯ ω) (39) and the second order solution becomes h i u(2) ≈ −q 2 8σ 2 e2iΘ + 2σ 0 ei(Θ+ −Θ− ) + cc. clearly this third order term can dominate the second order term when q/ν the form q ( ν is not small. K (ω +1 ) + K (ω −1 ) − K (2¯ ω) 2K (ω (±1) ) − K (2ω (±1) ) K (ω1 ) − K (ω −1 ) − K (2ν ) Remark 40 Observe that in the limit ν → 0 these (transfer) coeﬃcients become s. this will lead eﬀectively to terms of order q. s(±) = . Compared to the result for a nonlinear harmonic. 1. leading to the two-parameter expansion described above. s0 = . the resonant term in third order is taken care of by modifying the dispersion relation with suitable nonlinear terms. The second order solution is found to be h i u(2) = −q 2 2s+ e2iΘ+ + 2s− e2iΘ− + 4sei(Θ+ +Θ− ) + 2s0 ei(Θ+ −Θ− ) + cc where s= ω (±1) 2ν ω ¯ . the non-periodic variant of the bi-chromatic interference. but the value of these coeﬃcients diﬀers substantially depending on ν . The equation in third order contains resonant terms which leads to deﬁnition of the nonlinear dispersion relation: replace K (ω) by K KdV K KdV = K (ω ) + Kcorr (ω) ω ± ν ) = 4q 2 (¯ ω ± ν )(s0 + s± + 2s) where Kcorr (¯ Note that in the limit for ν → 0 this becomes ¯ (σ 0 + 3σ 2 ).distortion4 . The other terms in third order that are most dominant are the phases corresponding to the nearest side band ¯ ± 3ν . The quadratic nonlinearity generates second order terms with amplitudes q 2 and phases ± [Θ+ ± Θ− ] and third order terms with amplitudes q 3 and phases ± [[Θ+ ± Θ− ] ± [Θ+ ± Θ− ]]. As before. Diﬀerent from a direct series expansion in powers of q. see Fig. q 2 (the ‘second order’ contribution) and the ‘third order’ contribution of q 2 ) .

As stated above. The third order contribution to the solution contains terms from side band frequencies ω ± 3ν and low and high frequency components. that in the limit ν → 0 a±3 ∼ ω ± 3ν 2K (ω ± ν ) − K (ω ∓ ν ) − K (ω ± 3ν ) −4¯ ω (σ 0 + σ 2 ) . see Figure 1. although being small of order q 3 . contains some surprise. the bounded part of the third order solution. β −3 ∼ 2 K 002 ν . as we show now. We will restrict to the most important terms which are the side bands. and therefore easily neglected in most cases.ω 0 ω 0 ω1 ω2 First order Second order Third order ω Figure 1: Schematic diagram of mode generation from nonlinearity for the mono-chromatic (above) and the bi-chromatic wave (below). 00 4 [K ν 2 ± K 000 ν 3 ] γ 1 γK which is why we introduce α±3 = ν 2 a±3 and β ±3 = α3 ± α−3 . The corrected ’ﬁrst’ order linear solution then becomes u(1) = qeiΘ+ eiKcorr (ω+ν )z + qeiΘ− eiKcorr (ω−ν )z + cc which in a good approximation reads ¯ −ω u(1) = 4q cos(kz ¯ t + Kcorr z ) cos(κz − ν t). and get ³ q ´2 (3) usb = 2q [α3 cos(2Θ+ − Θ− ) + α−3 cos(2Θ− − Θ+ )] ν ³ q ´2 £ ¤ ¯ −ω ¯ −ω ¯ t) cos(3(δζ − ντ )) + β −3 sin((kz ¯ t) sin(3(δζ − ντ )) β 3 cos(kz =q ν 000 54 . These are given by h i (3) usb = q 3 a3 ei(2Θ+ −Θ− ) + a−3 ei(2Θ− −Θ+ ) + cc where a±3 = 4(s0 + s± ) Observe. hence β 3 ∼ 2− K 00 .

e.) In the ﬁgure 2 we show the evolution of the bi-chromatic as given by the analytical approximation. the amplitude of this last contribution may become substantial ν for small ν . The interference of the bound and the free third order carrier waves determines the spatial recurrence length of the total solution. The large deformation of the envelope during the propagation can be clearly seen : the initially sinusoidal proﬁle is 55 . Remark 41 Actually.3 Graphical representation π .05. with ω ¯ = 2 and q = 0. (Actually. Observe the z -dependent shift of the carrier wave that is where ∆3 = K2 (3ν )− 6 diﬀerent from that of the nonlinear dispersion relation.The eﬀect of the side band contributions is clearly seen: Summarising so far. ν = 0. when K 000 (¯ ω ) 6= 0 this term will cause skew-symmetric distortions of the envelope.4. This case is taken since it shows the deformation and break of symmetry more profound than in realsitic cases. i. Zrecurr = 3. to satisfy the correct initial signal.06. the ﬁrst and third order side band contributions together can be written (if we introduce in third order the higher order correction from the nonlinear dispersion relation in the carrier wave) like µ ¶ ³ q ´2 (3) (1) ¯ ¯ t + Kcorr z ) 4 cos(δζ − ντ ) + β 3 cos(3(δζ − ντ )) u + usb = q cos(kz − ω ν ³ q ´2 ¯ −ω +q β −3 sin(kz ¯ t) sin(3(δζ − ντ )) ν ¯ −ω • the carrier wave of the linear solution with phase correction cos(kz ¯ t + Kcorr z ) is modulated by the linear modulation 4 cos(δζ − ντ ) and a triple periodic contribution ¡ q ¢2 β 3 cos(3(δζ − ντ )). this case is rather extreme. 2K 00 ν 2 We will now show a plot of a characteristic situation in surface waves. given by 4K 00 ν 2 Zrecurr = 2π . plotted as time signal at various positions. the contribution of the third order term at z = 0 has to be compensated by harmonic modes (‘free waves’) uf ree = −2q ³ q ´2 ν [α3 cos Θ(¯ ω + 3ν ) + α−3 cos Θ(¯ ω − 3ν )] which can be re-written as a modiﬁcation of the linear solution as ¸ ³ q ´2 · β cos(kz ¯ −ω ¯ t + 4K 00 ν 2 z ) cos 3(∆3 ζ − ντ ) 3 uf ree = −q ¯ −ω +β −3 sin(kz ¯ t + 4K 00 ν 2 z ) sin 3(∆3 ζ − ντ ) ν K2 (−3ν ) ≈ 9δ . since q/ν = 5/6 rather large. This last eﬀect is proportional to q 2 K 000 (¯ ω)/ν . • in addition. the signal contains a contribution from the out-of phase carrier wave that is modulated by a same triple periodic function.

The results are correct for ﬁxed ν and small q . 56 .3 -0 .6 0 .0 .6 820 0 . the asymptotic approximations show (starting at third order) a series expansion in the two parameters.6 0 . Summary 42 From a theoretical point of view.3 z = 100 0 .3 0 .3 -0 .2 0 .6 -0 .3 .6 z = 400 0 . (Comparison with numerical calculations show good agreement.0 . q and q/ν .6 z = 700 0 .6 -3 0 -1 0 10 30 50 70 .0 .0 .3 0 .6 1242 1262 1282 1302 1322 1342 -0 .3 0 0 .0 . Observe the ‘breathing’ of the packets with periodically large intervals of small amplitude waves between large amplitude packets.3 . and the asymmetrization results from the third order term in the dispersion.6 840 860 880 900 920 0 400 420 440 460 480 500 z = 600 0 . including the correctness of the recurrence length theoretically predicted to be Zrecurr = 824). perturbed with a third harmonic of considerable amplitude.6 1460 1480 1500 1520 1540 1560 Figure 2: Analytic approximation for q/ν ≈ 5/6 (which is rather large).3 .4 z = 200 0 . when q/ν is of the order unity.0 .3 0 0 -0 . the series expansion will not converge and a more detailed two-parameter analysis is required.0 .z=0 0 .4 .2 -0 .6 0 .6 182 202 222 242 262 282 0 .

397-415. J. Soc. A369(1979)105-114. Diﬀerence Scheme’s for Solving the Generalized Nonlinear Schrodinger Equation. [5] E. Elsevier North-Holland. Comput. London. Uni-directional models for narrow and broad pulse propagation in second order nonlinear media.V. Chang. E. Hasegawa and Y. E.M. [4] E.C. in press 2001 [7] A. Nusantara & E. [6] E. 1995. Comput. [8] A. 1999. ISBN 0-444-82151-1. Phys. Mathematical structures in continuous dynamical systems. Deformation of modulated wave groups in third-order nonlinear media.B. Solitons in optical communications. Note on a modiﬁcation to the Nonlinear Schrodinger equation for application to deep water waves. Springer Verlag. Moloney. J. van Groesen. de Jager. accepted 2000.1992. van Groesen & E. Ablowitz. Oxford. 55(1984)203. Soewono. Large variations in NLS bi-soliton wave groups. Sun. [10] A. E. Amsterdam (1994) 631 pages. [2] Q. [11] T. Nonlinear pulses and beams. R. Optical and Quantum Electronics. Canada. van Groesen. Optical and Quantum Electronics. Proc. Ankiewicz. Chapman & Hall. accepted 2000. Akhmediev & A. Sulem & P-L Sulem. J. Newell and J. R. van Groesen & H. [3] K. Addison-Wesley. Solitons. T. Nonlinear optics. Taha & M.References [1] N. Cahyono & A.Suryanto. Clarendon Press. 148(1999). Hoekstra. Jia & W. 1997. Phys. 57 . Suryanto. [9] C. The Nonlinear Schrödinger Equation. Dysthe. Analytical and Numerical Aspects of Certain Nonlinear Evolution Equations II. Kodama. Optical and Quantum Electronics.

hamiltonian method for water wave simulations

hamiltonian method for water wave simulations

- Lecture 1
- CTI_Ch06.pdf
- Propagation of Vectorial Laser Beams
- Vls Curve Diffraction Grating
- bose einstein liquid helium.pdf
- Lab 4
- Navier-Stokes Equations.pdf
- A Note on Group Velocity
- Lorentz Attractor
- H.K. Moffatt- Fluid Mechanics, Topology, Cusp Singularities, and Related Matters
- Noncommutativity in non-uniform magnetic field and Quantum Mechanics
- Berry-phase Treatment of the Homogeneous Electric Field
- Signals and System 6.003Fall 2003
- Kausel Forced Circular
- airfoil
- 03507_01
- 1301.1588v18
- Classical Mechanics ( Physics)
- Qchem2 l(Ecturenotes)
- tmp4D6E.tmp
- Physics Mechanics 110A notes UCSD
- signal_and_system_5th.pdf
- Next
- FUNDAMENTALS OF GENERALIZED RIGIDITY MATRICES
- Trig No Me Try 1
- 110A Formulas Solutions
- String theory
- Lecture Notes
- Marsden J.E. Lectures on Mechanics (2ed., Web Draft, 1997)(233s)
- Classical Mechanics

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd