Lecture 16: Solitary waves

Lecturer: Roger Grimshaw. Write-up: Daisuke Takagi, Yiping Ma and Andrew Stewart

June 24, 2009

1 Introduction
A solitary wave is a wave which propagates without any temporal evolution in shape or
size when viewed in the reference frame moving with the group velocity of the wave. The
envelope of the wave has one global peak and decays far away from the peak. Solitary waves
arise in many contexts, including the elevation of the surface of water and the intensity of
light in optical fibers. A soliton is a nonlinear solitary wave with the additional property
that the wave retains its permanent structure, even after interacting with another soliton.
For example, two solitons propagating in opposite directions effectively pass through each
other without breaking.
Solitons form a special class of solutions of model equations, including the Korteweg
de-Vries (KdV) and the Nonlinear Schrödinger (NLS) equations. These model equations
are approximations, which hold under a restrictive set of conditions. The soliton solutions
obtained from the model equations provide important insight into the dynamics of solitary
waves. However, they are limited by the conditions under which the model equations hold.
An alternative approach, which deals directly with the exact equations from which the
model equations are derived, provides insight into a larger class of solitary waves than those
obtained from the model equations.
In this lecture, we show that important properties of solitary waves can be determined
directly from the exact equations governing a physical system by an asymptotic perturbation
procedure. Information about the possible existence of certain types of solitary waves is
obtained using a phase-plane formalism, a common technique in dynamical systems. In this
framework, a solitary wave corresponds to a homoclinic orbit in a spatial dynamical system.
Solitary waves exist in a number of different cases to be determined, given that they must
decay far away from their global peak. A close examination of the tail regions, far away
from the global peak, indicates that there are four possible cases of solitary waves. Three of
the four possible cases are presented in detail. In the first case, the steady-state solution of
the KdV equation is obtained. In the second case, a generalized solitary wave is obtained,
which decays to non-zero oscillations of constant amplitude and wavenumber. In the third
case, an envelope solitary wave is obtained, which satisfies the NLS equation.

2 Formulation of solitary waves as a dynamical system

Consider a solitary wave propagating with speed c in the positive direction of the x axis.
The wave amplitude in the moving reference frame is the function A(ξ) where ξ ≡ x − ct.

158
Figure 1: Plot of the dispersion relation for water waves given by (1) with Bond numbers B = 0
(violet), B = 0.2 (red) and B = 0.4 (blue).

Without loss of generality we can select ξ such that the peak of A is at ξ = 0, while A
decays in the tail regions, in the limits as ξ → ±∞. In the tail regions, where a linearized
analysis is assumed to hold, it is fruitful to seek solutions proportional to the real part of
exp(ikξ), where k is a complex-valued wavenumber to be determined. The wavenumber
must not be purely real, given that the wave amplitude must decay in the tail regions. This
leads to the notion that solitary waves with a real phase speed c only exist within a limited
range of wavenumbers in the spectrum, called gaps.
A relationship between the phase speed c and the wavenumber k is obtained by lin-
earizing the governing equations of a physical system of interest, which yields a linearized
dispersion relation. For example, the linearized dispersion relation of surface water waves
is given by
c2 /gh = f (q) = [(1 + Bq 2 )/q] tanh q, (1)
where g is the acceleration due to gravity, h the constant depth of the water when it
is unperturbed and q = kh the dimensionless wavenumber. The dimensionless number
B = σ/ρgh2 is the Bond number, which measures the relative importance of surface tension
and gravity, where σ is the coefficient of surface tension and ρ is the density of water. A
plot of the function f (q) given by equation (1) for three representative values of the Bond
number B is shown in Figure 1.
Solitary waves can exist provided that no real value of k satisfies the dispersion relation,
meaning that k has a non-zero imaginary part. In the case of B = 0 (no surface tension),
solitary waves can exist only when c2 > gh. When effects due to surface tension are

159
significant, such that B > 1/3, q is an increasing function of c with q ≥ 0 for c2 /gh ≥ 1. An
example with B = 0.4 is shown in figure 1 (blue curve). Thus, solitary waves with B > 1/3
can only exist when c2 /gh < 1. Finally, when 0 < B < 1/3, a real-valued q satisfies (1)
for any speed |c| > cm for some positive cm . Figure 1 shows a plot of (1) with B = 0.2,
which attains a local minimum c2m /gh at some non-zero wavenumber km . This indicates
that solitary waves with 0 < B < 1/3 can exist only when c2 < c2m .
Solitary waves are found by reformulating the problem in the framework of a dynamical
system. Let’s assume the system is conservative, which is the common and traditional
scenario for solitary waves. The underlying physical system is Hamiltonian and reversible,
meaning that the system conserves energy and is symmetric under the transformation ξ →
−ξ. This implies that if k is a solution then −k is also a solution due to reversibility. If
k has an imaginary part (as is required for a solitary wave), then k∗ is a solution because
the dispersion relation for real-valued phase speed c has real coefficients. The solutions
generically form a quartet (k, k∗ , −k, −k∗ ) with an associated four-dimensional subspace
for the corresponding wave mode. In the general case when the imaginary part of k is non-
zero, there are typically two roots in the limit as ξ → ∞ and two other roots as ξ → −∞.
In the limit as ξ → ∞, the imaginary part of k must be positive provided that the solitary
wave decays in the tail regions. Likewise, the imaginary part of k must be negative as
ξ → −∞.
Let us now study the amplitude A(ξ) in the framework of a dynamical system. Con-
sider the phase plane where the ξ-derivative of the wave amplitude is plotted against the
amplitude. Trajectories in this phase planes are the curves (A(ξ), Aξ (ξ)). This can either
represent the whole system (for a 2D problem) or a projection of the whole system (for a
higher-dimensional problem). Given the form of A at infinity discussed above, we deduce
that the trajectory is a homoclinic orbit from the origin. Indeed, the origin corresponds to
the tail regions as ξ → ±∞, where both A and Aξ tend to 0. Meanwhile, the peak of the
wave (ξ = 0 point) lies somewhere on the Aξ = 0 axis. If k is pure imaginary then the
trajectories satisfy Aξ = −|k|A near the origin, and approach it directly. If k has a real
part, then the trajectories spiral into the origin instead.
Consider now how the trajectory in the phase plane may change as parameters governing
the original system are varied. Changes in the trajectory far from the origin of the phase
plane mean that the overall shape of the wave A(ξ) is qualitatively changed, but that its
asymptotic behavior at |ξ| → ∞ remains the same. However, much more dramatic changes
in A(ξ) occur if the behavior near the phase plane origin changes (for example when k goes
from being real to complex to pure imaginary, or vice-versa).
This leads us to consider how the quartet structure (k, k∗ , −k, −k∗ ) evolves as some
global system parameter is varied. Clearly bifurcations arise when two solutions for k
coalesce, for which the necessary condition is that ∂c/∂k = 0. This condition is equivalent
to the condition that the phase speed c is equal to the group velocity cg (i.e. the weak
dispersion limit), when the bifurcation occurs at a real value of k. The equivalence follows
immediately from the equation cg = c + k∂c/∂k. For example, from Figure 1 we see that
bifurcations arise in water waves when (k, c2 ) takes the values (0, gh) or (km , c2m ).
Generically, there are two possible quartet structures at each of the bifurcation points:
• at (k = 0, c2 = gh): we either have the quartet (0, 0, iγ, −iγ) or (0, 0, β, −β), (two
roots have coalesced at the origin, the other two are either real or pure imaginary)

160
 • at (k = km , c2 = c2m ): we either have (β, β, −β, −β) (cc pairs coalesce on the real
axis) or (iγ, iγ, −iγ, −iγ) (pairs coalesce on the imaginary axis), where β and γ are
real-valued.
The first three of the four possible cases will be examined in detail later. The fourth case,
for the quartet (iγ, iγ, −iγ, −iγ), has only rarely been studied and will not be considered
here.
Consider a projection of the full system onto the appropriate four-dimensional subspace.
The resulting bifurcation is analyzed within the framework of this subspace. A 4-vector
W(ξ), representing the structure of the subspace, satisfies a first-order differential equation
of the form
dW/dξ = L(W; ǫ) + N (W), (2)
where L(W; ǫ) is a linear operator and N (W) contains all nonlinear terms. The parameter
ǫ represents the distance from the bifurcation point. Near the bifurcation (which takes
place at ǫ = 0) the linear operator L can be written as L(W; ǫ) = L0 (W) + ǫL1 (W) + ...,
where the eigenvalues λ = ik of the operator L0 reproduce one of the four possible quartet
structures described above. In each of the first three cases, the structural form of the small-
amplitude solutions W is examined to identify and describe the nature of corresponding
solitary waves.

3 The three cases

The three types of bifurcations may now be studied with the standard techniques of center
manifold reduction and normal form analysis in bifurcation theory. For more details about
these techniques, we refer the reader to the review article by Crawford [3].

3.1 Case (1)

Let us first consider case (1). At the bifurcation point (ǫ = 0) the linearized system
(2) has eigenvalues (0, 0, ∓γ). Since the λ = 0 eigenvalue is degenerate, there will be a
corresponding single eigenvector V0 , and a single generalized eigenvector V1 such that
L0 V1 = λV1 + V0 = V0 . Small-amplitude solutions are then sought in the form
W = A(ξ)V0 + B(ξ)V1 + W(2) . (3)
Here A, B are real variables of O(α), α ≪ 1, where α measures the wave amplitude. The
leading terms form a two-dimensional subspace (A, B), while W(2) is a small error term
of O(α2 , αǫ), where ǫ, α ≪ 1 are both small parameters. Note that the two remaining
eigenvalues ∓γ play no role at the leading order here, since they correspond to strong
exponential decay at infinity, and their effects are included in the small error term W(2) .
Substituting (3) into (2) yields
d  
Aξ V0 + Bξ V1 + W(2) = L0 A(ξ)V0 + B(ξ)V1 + W(2)
dξ
 
+ǫL1 A(ξ)V0 + B(ξ)V1 + W(2)
 
+N A(ξ)V0 + B(ξ)V1 + W(2) (4)

161
Projection onto the two-dimensional subspace (V0 , V1 ) yields, to lowest order in ǫ and α,

Aξ = B + O(ǫ, α2 , ...)
h i
Bξ = V1 · ǫAL1 (V0 ) + ǫBL1 (V1 ) + L0 (W(2) ) + N (AV0 + BV1 ) + O(αǫ2 , ǫα2 , ǫ2 , α3 , ..)

The first two terms in the B equation are linear terms of O(ǫα), and their projection onto
V1 yields ǫ(c1 A + c2 B), where c1 and c2 are the projection coefficients. Similarly, the linear
and nonlinear terms of O(α2 ) yield contributions of the kind A2 , AB or B 2 , so that

Aξ = B
Bξ = ǫ(c1 A + c2 B) + (d1 A2 + d2 AB + d3 B 2 ) + ...

where the omitted terms are O(αǫ2 , α2 ǫ, α3 ). Finally, a normal form analysis reveals the
normal form of the system near the bifurcation :

Aξ = B ,
Bξ = ǫA + µA2 + ... (5)

where µ is a real-valued coefficient, specific to the system being considered.

The eigenvalues λ = ik of this system are ±ǫ1/2 if ǫ > 0, and ±i|ǫ|1/2 if ǫ < 0. As
discussed earlier, only the former case, ǫ > 0, yields the solitary wave solution, with k =
±iǫ1/2 . Finally, comparison with the dispersion relation expressed near k = 0

k2 d2 c 

c(k) = c(0) + + ... (6)
2 dk2 k=0

leads to the identification of ǫ as

2(c − c(0))
ǫ=− . (7)
ckk (0)

Hence, for solitary waves to exist (ǫ > 0), we either need c > c(0) if ckk (0) < 0, or c < c(0)
if ckk (0) > 0. These inequalities recover the graphical argument discussed in the previous
section.
When the error terms in (5) are omitted, we can eliminate B, and the resulting ODE for
A can be recognized as the steady-state KdV equation, which has the well-known “sech2 ”
solution. It is then a delicate and intricate task to establish that this solitary wave solution
persists when the error terms are restored.
This dynamical systems approach to the problem has therefore established the steady-
state KdV equation as the normal form for weakly nonlinear, weakly dispersive solitary
waves (see also Lecture 6) whenever the dispersion relation satisfies dc/dk → 0 when k → 0,
and has provided mathematical conditions on their propagation speed. The KdV is thus
seen to be strictly correct for any solitary wave of the kind described above, far from the
wave peak (i.e. in the low-amplitude tails). Whether this statement still holds nearer the
peak ξ = 0 depends on the peak amplitude A(0).

162
3.2 Case (2)
Next consider case (2), where the linearized system (2) at the bifurcation point (ǫ = 0)
has eigenvalues (0, 0, ±iβ). Again, the degeneracy of λ = 0 implies that there is a single
eigenvector V0 , and a single generalized eigenvector V1 . However, account must now be
taken of the other two eigenvalues ±iβ, with their associated eigenvectors V2 , V2∗ , since they
do not now lead to decaying solutions at infinity. Small-amplitude solutions are sought in
the form
W = A(ξ)V0 + B(ξ)V1 + C(ξ)V2 + C ∗ (ξ)V2∗ + W(2) . (8)
Here C is a complex-valued variable, and hence the leading terms form a four-dimensional
subspace (A, B, C), while W(2) is a small error term. Projection onto this four-dimensional
subspace, followed by normal form analysis now reveals that (A, B, C) satisfy the system

Aξ = B ,
Bξ = ǫA + µA2 + ν|C|2 + · · · ,
Cξ = iγ(1 + δA)C + · · · . (9)

Here µ, ν, δ are real-valued coefficients specific to the system being considered, and the
omitted terms are small error terms as above.
When the error terms are omitted the system is integrable. In that limit, it is easy
to verify that |C|2ξ = 0 by constructing the quantity C ∗ Cξ + Cξ∗ C. Hence, |C| = C0 is a
constant. By a change of origin from A → A + A0 with ǫA0 + µA20 + νC02 = 0, the system
reduces to the same form as (5) in case (1). Thus, for the case ǫ > 0 (when case (1) is a
KdV-type solitary wave), the solution is a one-parameter family of homoclinic-to-periodic
solutions, with |C| = C0 constant and (A, B) → (A0 , 0) as ξ → ±∞. The solution is a
generalized solitary wave which typically has a “sech2 ” core, and decays at infinity to non-
zero oscillations of constant amplitude C0 and wavenumber γ (Figure 2). A delicate analysis
of the full system (2) with the small error terms shows that at least two of these solutions
persist; the minimal amplitude C0 being exponentially small, that is O(exp (−K/|ǫ|1/2 ))
where K is a positive real constant. Although such waves are permissible as solutions of
the steady-state equations, they have infinite energy and their associated group velocity
is inevitably inward at one end and outward at the other end. Hence, they cannot be
realized in a physical system from any localized initial condition because energy is finite.
Instead localized initial conditions will typically generate a one-sided generalized solitary
wave, whose central core is accompanied by small-amplitude outgoing waves on one side
only. Such waves cannot be steady, and instead will slowly decay with time as energy is
radiated away by outgoing waves from the core.

3.3 Case (3)

Finally we consider case (3), when there is a double eigenvalue λ = iβ with generically
a corresponding single eigenvector V0 , and a single generalized eigenvector V1 , while the
complex conjugate double eigenvalue λ = −iβ has corresponding complex conjugate eigen-
vectors. Small-amplitude solutions are now sought in the form

W = A(ξ)V0 + B(ξ)V1 + A∗ (ξ)V0∗ + B ∗ (ξ)V1∗ + W(2) . (10)

163
 Figure 2: Generalized solitary wave.

Here A and B are complex-valued variables, forming a four-dimensional subspace while

W(2) is again a small error term. Projection onto this subspace and a normal form analysis
reveal that

Aξ = iβA + B + iAP (ǫ, |A|2 , K) + · · · ,

Bξ = iβB + iBP (ǫ, |A|2 , K) + AQ(ǫ, |A|2 , K) + · · · , (11)

where K = i(AB ∗ − A∗ B) (note that K is real). Here P and Q are real-valued polynomials
of degree 1, which take the form

P (ǫ, |A|2 , K) = ǫ + ν1 |A|2 + ν2 K ,

Q(ǫ, |A|2 , K) = 2ǫβ + µ1 |A|2 + µ2 K , (12)

where all coefficients are real-valued. When the error terms in (11) are omitted, the resulting
system is integrable. There are two constants of motion. The first one is K, which is verified
by a direct construction of Kξ . The second one is H, where
µ1 4
H = |B|2 − (2ǫβ|A|2 + |A| + µ2 K|A|2 ) . (13)
2
To prove that H is constant, note that (|A|2 )ξ = A∗ B + B ∗ A and that (|B|2 )ξ = Q(|A|2 )ξ .
Hence, (|B|2 )ξ − Q(|A|2 )ξ = 0, so that the integral based on this quantity is constant. This
integral is H.

164
 For a solitary wave solution, conditions at infinity require K = H = 0. It then follows
that
µ1 4
|B|2 = 2ǫβ|A|2 +
|A|
2
2
(|A|2 )ξ A∗ B 2
   
µ1
(|A|ξ )2 = = = |B|2 = 2ǫβ|A|2 + |A|4 (14)
2|A| |A| 2

using the fact that (|A|2 )ξ = A∗ B + B ∗ A and that A∗ B = B ∗ A is real. Thus solitary
wave solutions exist provided that ǫ > 0, and that the nonlinear coefficient µ1 < 01 . The
condition ǫ > 0 implies that the perturbed eigenvalues, λ ≈ iβ ± (2ǫβ)1/2 have split off
the imaginary axis, and so provide the conditions needed for exponential decay at infinity.
The normal form coefficient µ1 must be computed from the physical parameters through a
reduction procedure [3], so the condition µ1 < 0 is problem-specific.
The solution of the truncated system governed by (14) is

A = a exp (i[β + ǫ]ξ)sech(γξ) , (15)

where γ = (2ǫβ)1/2 , |a|2 = −4ǫβ/µ1 . This solution describes an envelope solitary wave,
with a carrier wavenumber β + ǫ and an envelope described by the “sech”-function. As we
saw in Lecture 14, these solitary waves can also be obtained from the soliton solutions of
the NLS equation, in the special case when the phase velocity equals the group velocity,
c = cg , or more precisely when c + Ω/K = cg + V , where V is the soliton speed, Ω the
frequency and K the wavenumber correction. Note that the solution (15) contains an
arbitrary phase in the complex amplitude a, meaning that the location of the crests of the
carrier wave vis-a-vis the maximum of the envelope (here located at ξ = 0) is arbitrary.
However, restoration of the error terms in (11) leads to the result that only two of these
solutions persist, namely, those for which a carrier wave crest or trough is placed exactly
at ξ = 0, so that the resulting solitary wave is either one of elevation or depression. This
result requires very delicate analysis, but could be anticipated by noting that these are the
only two solutions which persist under the symmetry transformation ξ → −ξ.

4 Applications to water waves

The linearized dispersion relation holds the key to finding solitary waves. For water waves,
for which the dispersion relation is (1), these two cases (1) and (2) imply that pure solitary
waves of elevation exist for B = 0, and of depression for B > 1/3, while generalized solitary
waves arise whenever 0 < B < 1/3. For the case of generalized solitary waves, there is
always the possibility that the amplitude of the oscillations is zero, and the solution then
reduces to a pure solitary wave, called an “embedded” solitary wave. There are now many
examples of such embedded solitary waves arising in various physical systems, notably for
internal waves. This “dynamical-systems” approach to finding solitary waves has also been
applied to interfacial waves, where again the linear dispersion relation holds the key to
1
This may be understood by regarding |A| as space, ξ as time, and |A|ξ as velocity in (14). Then (14)
describes a velocity field on the positive real line. In order for an orbit starting from and returning to the
origin as ξ → ∓∞ to exist, the velocity field must have a stagnation point, which occurs only when µ1 < 0.

165
determining where solitary waves can be found. However, various numerical and analytical
studies suggest that embedded solitary waves do not arise in the context of water waves.
Instead, case (3) implies that envelope solitary waves arise for capillary-gravity waves with
0 < B < 1/3, where it can be shown that the coefficient µ1 in (12) is negative as required.
Finally, we remark that the method of treating solitary waves as homoclinic orbits in
a spatial dynamical system has many applications (cf. [2]) in general evolution equations,
most of which are not integrable. This point of view, coupled with numerical continuation
techniques, turns out to be fruitful in studying spatially localized states in both one (cf. [1])
and two (cf. [7]) dimensional pattern forming systems.

References
[1] Burke, J. and Knobloch, E., Localized states in the generalized Swift-Hohenberg
equation, Physical Review E, 73 (2006), 056211.

[2] Champneys, A. R., Homoclinic orbits in reversible systems and their applications in
mechanics, fluids and optics, Physica D, 112, (1998) pp. 158-186.

[3] Crawford, J. D., Introduction to bifurcation theory, Review of Modern Physics, 63,
(1991) pp. 991-1037.

[4] Dias, F. and Iooss, G. , Water waves as a spatial dynamical system, Handbook of
Mathematical Fluid Dynamics, ed. S. Friedlander and D. Serre, Elsevier, North Holland,
Chapter 10, (2003), pp. 443-499.

[5] Grimshaw, R. and Iooss, G, Solitary waves of a coupled Korteweg-de Vries system,
Mathematics and Computers in Simulation, 62, (2003) pp. 31-40.

[6] Solitary Waves in Fluids, ed. Grimshaw R. 2007, Advances in Fluid Mechanics, Vol
47, WIT press, UK, Chapter 1, 1-17 and Chapter 7, pp. 159-179.

[7] ,Lloyd, D. J. B., Sandstede B., Avitabile D. and Champneys A. R., Localized
hexagon patterns of the planar Swift-Hohenberg equation, Siam J. Applied Dynamical
Systems, 7, (2008), pp. 1049-1100.

166