Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
com
Mathematics and Computers in Simulation 82 (2012) 1185–1218
Numerical study of the KP equation for nonperiodic waves
ChiuYen Kao
∗
, Yuji Kodama
Department of Mathematics, The Ohio State University, 231 West 18th Ave., Columbus, OH 43210, United States
Received 7 February 2010; received in revised form 25 April 2010; accepted 19 May 2010
Available online 11 June 2010
Abstract
The Kadomtsev–Petviashvili (KP) equation describes weakly dispersive and small amplitude waves propagating in a quasitwo
dimensional situation. Recently a large variety of exact soliton solutions of the KP equation has been found and classiﬁed. Those
soliton solutions are localized along certain lines in a twodimensional plane and decay exponentially everywhere else, and they
are called linesoliton solutions in this paper. The classiﬁcation is based on the farﬁeld patterns of the solutions which consist of
a ﬁnite number of linesolitons. In this paper, we study the initial value problem of the KP equation with V and Xshape initial
waves consisting of two distinct linesolitons by means of the direct numerical simulation. We then show that the solution converges
asymptotically to some of those exact soliton solutions. The convergence is in a locally deﬁned L
2
sense. The initial wave patterns
considered in this paper are related to the rogue waves generated by nonlinear wave interactions in shallow water wave problem.
© 2010 IMACS. Published by Elsevier B.V. All rights reserved.
Kadomtsev–Petviashvili equation; Soliton solutions; Chord diagrams; Pseudospectral method; Window technique
1. Introduction
The KdV equation may be obtained in the leading order approximation of an asymptotic perturbation theory for
onedimensional nonlinear waves under the assumptions of weak nonlinearity (small amplitude) and weak dispersion
(long waves). The initial value problem of the KdV equation has been extensively studied by means of the method of
inverse scattering transform (IST). It is then wellknown that a general initial data decaying rapidly for large spatial
variable evolves into a sum of individual solitons and some weak dispersive wave trains separate away from solitons
(see for examples [1,17,19,24,5]).
In 1970, Kadomtsev and Petviashvili [9] proposed a twodimensional dispersive wave equation to study the stability
of one soliton solution of the KdV equation under the inﬂuence of weak transversal perturbations. This equation is now
referred to as the KP equation, and considered to be a prototype of the integrable nonlinear dispersive wave equations in
two dimensions. The KP equation can be also represented in the Lax form, that is, there exists a pair of linear equations
associated with the eigenvalue problem and the evolution of the eigenfunctions. However, unlike the case of the KdV
equation, the method of IST based on the pair of linear equations does not seem to provide a practical method for the
initial value problem with nonperiodic waves considered in this paper. At the present time, there is no feasible analytic
method to solve the initial value problem of the KP equation with initial waves having linesolitons in the far ﬁeld.
∗
Corresponding author. Tel.: +1 614 2928609.
Email address: kao@math.ohiostate.edu (C.Y. Kao).
03784754/$36.00 © 2010 IMACS. Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.matcom.2010.05.025
1186 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
In this paper, we study this type of the initial value problem of the KP equation by means of the direct numerical
simulation. In particular, we consider the following two cases of the initial waves: In the ﬁrst case, the initial wave
consists of two semiinﬁnite linesolitons forming a Vshape pattern, and in the second case, the initial wave is given
by a linear combination of two inﬁnite linesolitons forming Xshape. Those initial waves have been considered in
the study of the generation of large amplitude waves in shallow water [20,22,13,25]. The main result of this paper
is to show that the solutions of the initial value problem with those initial waves asymptotically converge to some of
the exact soliton solutions found in [2,3]. This implies a separation of the (exact) soliton solution from the dispersive
radiations in the similar manner as in the KdV case.
The paper is organized as follows: in Section 2, we provide a brief summary of the soliton solutions of the KP
equation and the classiﬁcation theorem obtained in [2,3] for those soliton solutions as a background necessary for the
present study. In particular, we introduce the parametrization of each soliton solution with a chord diagram which
represents a derangement of the permutation group, i.e. permutation without ﬁxed point.
In Section 3, we present several exact soliton solutions, and describe some properties of those solutions. Each of
those soliton solutions has N
+
numbers of linesolitons in a far ﬁeld on the twodimensional plane, say in y 0, and
N
−
numbers of linesolitons in the far ﬁeld of the opposite side, i.e. y 0. This type of soliton solution is referred to
as a (N
−
, N
+
)soliton solution. Here we consider those solitons with N
−
+ N
+
≤ 4 and N
−
≤ 3.
In Section 4, we describe the numerical scheme used in this paper, which is based on the pseudospectral method
combined with the window technique [21,23]. The window technique is especially used to compute our nonperiodic
problem which is essentially an inﬁnite domain problem.
In Section 5, we present the numerical results of the initial value problems with V and Xshape initial waves,
and show that the solutions asymptotically converge to some of those exact solutions discussed in Section 3. The
convergence is in the sense of locally deﬁned L
2
norm with the form, f
L
2
(D)
:=
___
D
f (x, y)
2
dx dy
_
1/2
, where
D ⊂ R
2
is a compact set which covers the main structure describing the (resonant) interactions in the solution. We also
propose a method to identify an exact solution for a given initial wave with V or Xshape pattern based on the chord
diagrams introduced in the classiﬁcation theory. Then, in Section 6, we conclude the paper with a brief summary of
our results.
2. Background
Here we give a brief summary of the recent result of the classiﬁcation theorem for soliton solutions of the KP
equation (see [12,2,3] for the details). In particular, each soliton solution is then parametrized by a chord diagram
which represents a unique element of the permutation group. We use this parametrization throughout the paper.
2.1. The KP equation
The KP equation is a twodimensional nonlinear dispersive wave equation given by
∂
x
_
4∂
t
u + 6u∂
x
u + ∂
3
x
u
_
+3∂
2
y
u = 0, (2.1)
where ∂
n
x
u:=∂
n
u/∂x
n
, etc. Let us express the solution in the form,
u(x, y, t) = 2 ∂
2
x
ln τ(x, y, t). (2.2)
where the function τ is called the tau function, which plays a central role in the KP theory. In this paper, we consider
a class of the solutions, where each solution can be expressed by τ(x, y, t) in the Wronskian determinant form τ =
Wr(f
1
, . . . , f
N
), i.e.
τ(x, y, t) =
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
f
1
f
1
· · · f
(N−1)
1
f
2
f
2
· · · f
(N−1)
2
.
.
.
.
.
.
.
.
.
.
.
.
f
N
f
N
· · · f
(N−1)
N
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
, (2.3)
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1187
with f
(j)
n
:=∂
j
x
f
n
for n = 1, . . . , N. Here the functions {f
1
, . . . , f
N
} form a set of linearly independent solutions of the
linear equations,
∂
y
f
n
= ∂
2
x
f
n
, ∂
t
f
n
= −∂
3
x
f
n
.
(The fact that (2.2) with (2.3) gives a solution of the KP equation is wellknown and the proof can be found in several
places, e.g. see [8,3].) The solution of those equations can be expressed in the Fourier transform,
f
n
(x, y, t) =
_
C
e
kx+k
2
y−k
3
t
ρ
n
(k) dk, n = 1, 2, . . . , N, (2.4)
with an appropriate contour C in C and the measure ρ
n
(k) dk. In particular, we consider a ﬁnite dimensional solution
with ρ
n
(k) dk =
M
m=1
a
n,m
δ(k − k
m
) dk with a
n,m
∈ R, i.e.
f
n
(x, y, t) =
M
m=1
a
n,m
E
m
(x, y, t) with E
m
= exp(k
m
x + k
2
m
y − k
3
m
t).
Thus this type of solution is characterized by the parameters {k
1
, k
2
, . . . , k
M
} and the N × M matrix A:=(a
n,m
) of
rank(A) = N, that is, we have
(f
1
, f
2
, . . . , f
N
) = (E
1
, E
2
, . . . , E
M
)A
T
. (2.5)
Note that {E
1
, E
2
, . . . , E
M
} gives a basis of R
M
and {f
1
, f
2
, . . . , f
N
} spans an Ndimensional subspace of R
M
. This
means that the Amatrix can be identiﬁed as a point on the real Grassmann manifold Gr(N, M) (see [12,3]). More
precisely, let M
N×M
(R) be the set of all N × M matrices of rank N. Then Gr(N, M) can be expressed as
Gr(N, M) = GL
N
(R) \ M
N×M
(R).
where GL
N
(R) is the general linear group of rank N. This expression means that other basis (g
1
, . . . , g
N
) =
(f
1
, . . . , f
N
)H for any H ∈ GL
N
(R) spans the same subspace, that is, A → H
T
A GL
N
(R) acting from the left).
Notice here that the freedom in the Amatrix with GL
N
(R) can be ﬁxed by expressing A in the reduced row echelon
form (RREF). We then assume throughout this paper that the Amatrix is in the RREF, and show that the Amatrix
plays a crucial role in our discussion on the asymptotic behavior of the initial value problem.
Now using the Binet–Cauchy Lemma for the determinant, the τfunction of (2.2) can be expressed in the form,
τ =
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
⎛
⎜
⎜
⎜
⎜
⎝
E
1
E
2
· · · · · · E
M
k
1
E
1
k
2
E
2
· · · · · · k
M
E
M
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
k
N−1
1
E
1
k
N−1
2
E
2
· · · · · · k
N−1
M
E
M
⎞
⎟
⎟
⎟
⎟
⎠
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
a
11
a
21
· · · a
N1
a
12
a
22
· · · a
N2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
a
1M
a
2M
· · · a
NM
⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
=
1≤j
1
<j
2
<···<j
N
≤M
ξ(j
1
, j
2
, . . . , j
N
)E(j
1
, j
2
, . . . , E
N
), (2.6)
where ξ(j
1
, . . . , j
N
) is the N × N minor of the Amatrix with N columns marked by (j
1
, . . . , j
N
), and E(j
1
, . . . , j
N
)
is given by
E(j
1
, . . . , j
N
) = Wr(E
j
1
, . . . , E
j
N
) =
l<m
(k
j
m
− k
j
l
)E
j
1
· · ·E
j
N
.
From formula (2.6), one can see that for a given Amatrix,
• if a column of A has only zero elements, then the exponential term E
m
with m being the column index never appear
in the τfunction, and
• if a row of A has only the pivot as nonzero element, then E
n
with n being the row index can be factored out from
the τfunction, i.e. E
n
has no contribution to the solution.
1188 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
For Amatrix with no such cases, we call it irreducible, because the τfunction with reducible matrix can be obtained
by a matrix with smaller size.
We are also interested in nonsingular solutions. Since the solution is given by u = 2∂
2
x
(ln τ), the nonsingular
solutions are obtained by imposing the nonnegativity condition on the minors,
ξ(j
1
, j
2
, . . . , j
N
) ≥ 0, for all 1 ≤ j
1
< j
2
< · · · < j
N
≤ M. (2.7)
This condition is not only sufﬁcient but also necessary for the nonsingularity of the solution for any initial data. We
call a matrix having the condition (2.7) totally nonnegative matrix.
2.2. Linesoliton solution and the notations
Let us here present the simplest solution, called one linesoliton solution, and introduce several notations to describe
the solution. A linesoliton solution is obtained by a τfunction with two exponential terms, i.e. the case N = 1 and
M = 2 in (2.5): With the Amatrix of the form A = (1 a), we have
τ = E
1
+ aE
2
= 2
√
ae
(1/2)(θ
1
+θ
2
)/2
cosh
1
2
(θ
1
− θ
2
− ln a),
The parameter a in the Amatrix must be a ≥ 0 for a nonsingular solution (i.e. totally nonnegative Amatrix), and
it determines the location of the soliton solution. Since a = 0 leads to a trivial solution, we consider only a > 0 (i.e.
irreducible Amatrix). Then the solution u = 2∂
2
x
(ln τ) gives
u =
1
2
(k
1
− k
2
)
2
sech
2
1
2
(θ
1
− θ
2
− ln a).
Thus the peak of the solution is located along the line θ
1
− θ
2
= ln a. We here emphasize that the linesoliton appears
at the boundary of two regions, in each of which either E
1
or E
2
becomes the dominant exponential term, and because
of this we also call this soliton [1, 2]soliton solution (or soliton of [1, 2]type). In general, the linesoliton solution of
[i, j]type with i < j has the following structure (sometimes we consider only locally),
u = A
[i,j]
sech
2
1
2
(K
[i,j]
· x − Ω
[i,j]
t + Θ
[i,j]
) (2.8)
with some constant Θ
[i,j]
. The amplitude A
[i,j]
, the wave vector K
[i,j]
and the frequency Ω
[i,j]
are deﬁned by
A
[i,j]
=
1
2
(k
j
− k
i
)
2
K
[i,j]
=
_
k
j
− k
i
, k
2
j
− k
2
i
_
= (k
j
− k
i
)
_
1, k
i
+ k
j
_
,
Ω
[i,j]
= k
3
j
− k
3
i
= (k
j
− k
i
)(k
2
i
+ k
i
k
j
+ k
2
j
).
The direction of the wavevector K
[i,j]
= (K
x
[i,j]
, K
y
[i,j]
) is measured in the counterclockwise from the xaxis, and it is
given by
K
y
[i,j]
K
x
[i,j]
= tan Ψ
[i,j]
= k
i
+ k
j
.
Notice that Ψ
[i,j]
gives the angle between the line K
[i,j]
· x = const and the yaxis (See Fig. 1). Then a linesoliton
(2.8) can be written in the form with three parameters A
[i,j]
, Ψ
[i,j]
and x
0
[i,j]
,
u = A
[i,j]
sech
2
_
A
[i,j]
2
(x + y tan Ψ
[i,j]
− C
[i,j]
t − x
0
[i,j]
), (2.9)
withC
[i,j]
= k
2
i
+ k
i
k
j
+ k
2
j
= (1/2)A
[i,j]
+ (3/4) tan
2
Ψ
[i,j]
. For the parameter x
0
[i,j]
giving the location of the line
soliton, we also use the notation,
x
0
[i,j]
= −
1
k
j
− k
i
Θ
[i,j]
, (2.10)
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1189
Fig. 1. One linesoliton solution of [i, j]type and the corresponding chord diagram. The left ﬁgure shows the soliton solution, u(x, y, t) with
the amplitude A
[i,j]
and the angle Ψ
[i,j]
which are, respectively, given by A
[i,j]
=
1
2
(k
i
−k
j
)
2
and tan Ψ
[i,j]
= k
i
+k
j
. Throughout this paper, the
horizontal and vertical axes in the ﬁgures of soliton solutions are x and y, respectively, and the solutions are shown in the contour lines with colors
indicating the magnitudes. The right panel shows the corresponding chord diagram. The upper oriented chord represents the part of [i, j]soliton
for y 0 and the lower one for y 0. Here the horizontal line is the coordinate for the kparameters, and in this example, k
i
= −0.5 and k
j
= 1,
which give tan Ψ
[i,j]
= 0.5 and A
[i,j]
= 9/8.
with Θ
[i,j]
in (2.8) which is determined by the Amatrix and the kparameters. For multisoliton solutions, one can
only deﬁne the location of each [i, j]soliton using the asymptotic position in the xyplane either x 0 or x 0
(we mainly consider the cases where the solitons are not parallel to the yaxis), and we use the notation x
+
[i,j]
(or
x
−
[i,j]
) which describes the xintercept of the line determined by the wave crest of [i, j]soliton in the region x 0 (or
x 0) at t = 0. In Fig. 1, we illustrate an example of one linesoliton solution. In the right panel of this ﬁgure, we
show a chord diagram which represents this soliton solution. Here the chord diagram indicates the permutation of the
dominant exponential terms E
i
and E
j
in the τfunction, that is, with the ordering k
i
< k
j
, E
i
dominates in x 0,
while E
j
dominates in x 0. This representation of the linesolitons in terms of the chord diagrams is the key concept
throughout the present paper. (See section 2.3 below for the precise deﬁnition of the chord diagrams.)
For each soliton solution of (2.9), the wave vector K
[i,j]
and the frequency Ω
[i,j]
satisfy the solitondispersion
relation, i.e.,
4Ω
[i,j]
K
x
[i,j]
= (K
x
[i,j]
)
4
+ 3(K
y
[i,j]
)
2
. (2.11)
The soliton velocity V
[i,j]
deﬁned by K
[i,j]
· V
[i,j]
= Ω
[i,j]
is given by
V
[i,j]
=
Ω
[i,j]
K
[i,j]

2
K
[i,j]
=
k
2
i
+ k
i
k
j
+ k
2
j
1 + (k
i
+ k
j
)
2
(1, k
i
+ k
j
).
Note in particular that C
[i,j]
= k
2
i
+ k
i
k
j
+ k
2
j
> 0, and this implies that the xcomponent of the velocity is always
positive, that is, any soliton propagates in the positive xdirection. On the other hand, one should note that any small
perturbation propagates in the negative xdirection, that is, the xcomponent of the group velocity is always negative.
This can be seen from the dispersion relation of the KP equation for a linear wave φ = exp(ik · x − iωt) with the
wavevector k = (k
x
, k
y
) and the frequency ω,
ω = −
1
4
k
3
x
+
3
4
k
2
y
k
x
,
from which the group velocity of the wave is given by
v = ∇ω =
_
∂ω
∂k
x
,
∂ω
∂k
y
_
=
_
−
3
4
_
k
2
x
+
k
2
y
k
2
x
_
,
2
3
k
y
k
x
_
.
This is similar to the case of the KdV equation, and we expect that asymptotically soliton separates from dispersive
radiations. Physically this implies that soliton is a supersonic wave due to its nonlinearity (recall that the velocity of
shallowwater wave is proportional to the square root of the water depth, and the KP equation in the form (2.1) describes
the waves in the moving frame with this velocity in the xdirection).
1190 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
Stability of onesoliton solution was shown in the original paper by Kadomtsev and Petviashvii [9], and this may
be stated as follows: For any > 0 and any r > 0, there exists δ > 0 so that if the initial wave u(x, y, 0):=u
0
(x, y)
satisﬁes
u
0
− u
0
exact
L
2
(D
0
r
)
< δ,
for some exact soliton solution, u
t
exact
:=A
0
sech
2
√
A
0
/2(x + y tan Ψ
0
− C
0
t − x
0
) with appropriate constants A
0
, Ψ
0
and x
0
(recall C
0
= (1/2)A
0
+ (3/4) tan
2
Ψ
0
), then the stability implies that the solution u
t
(x, y):=u(x, y, t) satisﬁes
u
t
− u
t
exact
L
2
(D
t
r
)
< , for t → ∞,
where D
t
r
is a circular disc with radius r moving with the soliton, i.e.
D
t
r
= {(x, y) ∈ R
2
: (x − x
0
(t))
2
+ (y − y
0
(t))
2
≤ r
2
},
with (x
0
(t), y
0
(t)) at any point on the soliton, x
0
(t) + y
0
(t) tan Ψ
0
= C
0
t + x
0
. Here f
L
2
(D)
is the usual L
2
norm of
f(x, y) over a compact domain D ⊂ R
2
, i.e.
f
L
2
(D)
:=
___
D
f (x, y)
2
dx dy
_
1/2
.
This stability implies a separation of the soliton from the dispersive radiations (nonsoliton parts) as in the case of the
KdV soliton. We would like to prove the similar statement for more general initial waves. However there are several
difﬁculties for twodimensional stability problem in general. In this paper, we will give a numerical study for some
special cases where the initial waves consist of two semiinﬁnite linesolitons with V or Xshape.
Finally we remark that a linesoliton having the angle Ψ
0
≈ π/2 has an inﬁnite speed, and of course it is beyond
the assumption of the quasitwodimensionality. We also emphasize that the structure of the solution for y 0 can be
different from that in y 0, that is, the set of asymptotic solitons in y 0 can be different from that in y 0. This
difference is a consequence of the resonant interactions among solitons as we can see throughout the paper.
2.3. Classiﬁcation theorems
Now we present the main theorems obtained in [2,3] for the classiﬁcation of soliton solutions generated by the
τfunctions with irreducible and totally nonnegative Amatrices:
Theorem 2.1. Let{e
1
, . . . , e
N
} be the pivot indices, and let{g
1
, . . . , g
M−N
} be nonpivot indices for an N × M
irreducible and totally nonnegative Amatrix. Then the soliton solution generated by the τfunction with the Amatrix
has the following asymptotic structure:
(a) Fory 0, there are N linesolitons of [e
n
, j
n
]type forn = 1, . . . , N.
(b) Fory 0, there areM − N linesolitons of [i
m
, g
m
]type form = 1, . . . , M − N.
Herej
n
(> e
n
) and i
m
(< g
m
) are determined uniquely from the Amatrix.
The unique index pairings [e
n
, j
n
] and [i
m
, g
m
] in Theorem 2.1 have a combinatorial interpretation. Let us deﬁne
the pairing map π : {1, 2, . . . , M} → {1, 2, . . . , M} such that
_
π(e
n
) = j
n
, n = 1, . . . , N,
π(g
m
) = i
m
, m = 1, . . . , M − N,
(2.12)
where e
n
and g
m
are respectively the pivot and nonpivot indices of the Amatrix. Then we have:
Theorem 2.2. The pairing map π is a bijection. That is, π ∈ S
M
, whereS
M
is the group of permutation for the index
set{1, 2, . . . , M}, i.e.
π =
_
e
1
· · ·e
N
g
1
· · ·g
M−N
j
1
· · ·j
N
i
1
· · ·i
M−N
_
.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1191
Fig. 2. Example of the chord diagram. This shows π = (46523817). Each chord joining ith and jth marked points represents the linesoliton of
[i, j]type. Here the numbers in the ﬁgure indicate the kparameters, i.e. j means k
j
.
Note in particular that the corresponding π is the derangement, i.e. π has no ﬁxed point.
Theorem 2.2 shows that the pairing map π for an (M − N, N)soliton solution has N excedances, i.e. π(i) > i for
i = 1, . . . , N, and the excedance set is the set of pivot indices of the Amatrix. We represent each soliton solution with
the chord diagram deﬁned as follows:
(i) There are M marked points on a line, each of the point corresponds to the kparameter.
(ii) On the upper side of the line, there are N chords (pairings), each of them connects two points on the line representing
π(e
n
) = j
n
for n = 1, . . . , N, i.e. the excedance j
n
> e
n
.
(iii) On the lower side of the line, there are M − N chords representing π(g
m
) = i
m
for m = 1, . . . , M − N, i.e. the
deﬁciency i
m
< g
m
.
In Fig. 2, we illustrate an example of the chord diagram which represents the derangement,
π =
_
1 2 3 4 5 6 7 8
4 6 5 2 3 8 1 7
_
or simply π = (46523817).
The diagram then shows that the set of excedances is {1, 2, 3, 6}, and the corresponding soliton solution consists of
the asymptotic linesolutions of [1, 4], [2, 6], [3, 5] and [6, 8]types in y 0 and of [1, 7], [2, 4], [3, 5] and
[7, 8]types in y 0.
3. Exact solutions
Here we present several exact solutions generated by smaller size N × M matrices with N ≤ 3 and M ≤ 4. Those
solutions give a fundamental structure of general solutions, and we will show that some of those solutions appear
naturally as asymptotic solutions of the KP equation for certain classes of initial waves related to the rogue wave
generation [20,22,23,7]. The detailed discussions and the formulae given in this section can be found in [3].
3.1. Yshape solitons: resonant solutions
We ﬁrst discuss the resonant interaction among linesolitons, which is the most important feature of the KP equation
(see e.g. [14,18,10]). To describe resonant solutions, let us consider the τfunction with M = 3, that is, the τfunction
has three exponential terms {E
1
, E
2
, E
3
}. In terms of the τfunction in the form (2.3), those are given by the cases
(N = 1, M = 3) and (N = 2, M = 3).
Let us ﬁrst study the case with N = 1 and M = 3, where the Amatrix is given by
A = ( 1 a b ).
The parameters a, b in the matrix are positive constants, and the positivity implies the irreducibility and the regularity
of the solution. The τfunction is simply given by
τ = E
1
+ aE
2
+ bE
3
.
(Note that if one of the parameters is zero, then the τfunction consists only two exponential terms and it gives one
linesoliton solution, i.e. reducible case.) With the ordering k
1
< k
2
< k
3
, the solution u = 2∂
2
x
(ln τ) consists of [1, 3]
1192 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
soliton for y 0 and [1, 2] and [2, 3]solitons for y 0. Taking the balance between two exponential terms in the
τfunction, one can see that those linesolitons of [1, 3] and [2, 3]types are localized along the lines given in (2.9), i.e.
x + y tan Ψ
[i,j]
− C
[i,j]
t = x
0
[i,j]
for [i, j] = [1, 3] and [2, 3] where the locations x
0
[i,j]
are determined by the Amatrix
(see (2.10)),
x
0
[1,3]
= −
1
3, 1
ln b, x
0
[2,3]
= −
1
3, 2
ln
b
a
. (3.1)
where i, j:=k
i
− k
j
. The shape of solution generated by τ = E
1
+ aE
2
+ bE
3
with a = b = 1 (i.e. at t = 0 three
linesolitons meet at the origin) is illustrated via the contour plot in the ﬁrst row of Fig. 3. This solution represents a
resonant solution of three linesolitons, and the resonant condition is given by
K
[1,3]
= K
[1,2]
+ K
[2,3]
, Ω
[1,3]
= Ω
[1,2]
+ Ω
[2,3]
,
which are trivially satisﬁed with K
[i,j]
= (k
j
− k
i
, k
2
j
− k
2
i
) and Ω
[i,j]
= k
3
j
− k
3
i
. The chord diagram corresponding
to this soliton is shown in the right panel of the ﬁrst raw in Fig. 3, and it represents the permutation π = (312).
Let us now consider the case with N = 2 and M = 3: We take the Amatrix in the form,
A =
_
1 0 −b
0 1 a
_
,
where a and b are positive constants. Then the τfunction is given by
τ = E(1, 2) + aE(1, 3) + bE(2, 3),
with E(i, j) = (k
j
− k
i
)E
i
E
j
for i < j. In this case we have [1, 2] and [2, 3]solitons for y 0 and [1, 3]soliton for
y 0, and this solution can be labeled by π = (231). Those linesolitons of [1, 2] and [1, 3]types are localized along
the lines, x + y tan Ψ
[i,j]
− C
[i,j]
t = x
0
[i,j]
with
x
0
[1,2]
= −
1
2, 1
ln
_
3, 2
3, 1
b
a
_
, x
0
[1,3]
= −
1
3, 1
ln
_
3, 2
2, 1
b
_
, (3.2)
where i, j:=k
i
− k
j
. In the lower ﬁgures of Fig. 3, we illustrate the solution in this case. Notice that this ﬁgure can
be obtained from (2, 1)soliton in the upper ﬁgure by changing (x, y) → (−x, −y). Here the parameters a and b in the
Amatrix are chosen, so that x
0
[1,2]
= x
0
[1,3]
= 0, that is, all of those soliton solutions meet at the origin at t = 0.
Fig. 3. Examples of (2, 1) and (1, 2)soliton solutions and the chord diagrams. The kparameters are given by (k
1
, k
2
, k
3
) = (−5/4, −1/4, 3/4).
The right panels are the corresponding chord diagrams. For both cases, the parameters in the Amatrices are chosen so that at t = 0 three linesolitons
meet at the origin as shown in the left panels.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1193
3.2. N = 1, 3 and M = 4 cases
Let us ﬁrst discuss the case with N = 1 and M = 4, that is, the Amatrix is given by
A = (1 a b c),
where a, b and c are positive constants. The τfunction is simply written in the form
τ = E
1
+ aE
2
+ bE
3
+ cE
4
.
In this case, we have (3, 1)soliton solution consisting of one linesoliton of [1, 4]type for y 0 and three linesolitons
of [1, 2], [2, 3] and [3, 4]types for y 0. This solution is labeled by π = (4 1 2 3). The upper ﬁgures in Fig. 4 shows
the time evolution of the solution of this type. We set a = b = c = 1 of the Amatrix, so that all four linesolitons meet
at the origin at t = 0.
For N = 3, any irreducible and totally nonnegative Amatrix has the form,
A =
⎛
⎜
⎝
1 0 0 c
0 1 0 −b
0 0 1 a
⎞
⎟
⎠
,
where a, b and c are positive constants. The τfunction is then given by
τ = E(1, 2, 3) + aE(1, 2, 4) + bE(1, 3, 4) + cE(2, 3, 4),
with E(l, m, n) = (k
n
− k
m
)(k
n
− k
l
)(k
m
− k
l
)E
l
E
m
E
n
. This gives (1, 3)soliton solution which is dual to the case of
N = 1, that is, [1, 4]soliton for y 0 and [1, 2], [2, 3] and [3, 4]solitons for y 0. The corresponding label for
this solution is given by π = (2 3 4 1). The lower ﬁgures in Fig. 4 shows the time evolution of this type. Here we set
a, b, c of the Amatrix as
a =
1, 2, 3
1, 2, 4
, b =
1, 2, 3
1, 3, 4
, c =
1, 2, 3
2, 3, 4
,
where l, m, n:=(k
n
− k
m
)(k
n
− k
l
)(k
m
− k
l
), so that all four solitons meet at the origin at t = 0.
Fig. 4. Examples of (3, 1) and (1, 3)soliton solutions. The solution in the upper ﬁgures are generated by the 1 ×4 Amatrix with (k
1
, k
2
, k
3
, k
4
) =
(−
√
3, −1, 1,
√
3). The symmetry in the kparameters implies that [1, 4] and [2, 3]solitons are parallel to the yaxis. The lower ﬁgures are generated
by the 3 ×4 Amatrix with (k
1
, k
2
, k
3
, k
4
) = (−1, −1/(2
√
2), 1/(2
√
2), 1).
1194 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
Fig. 5. The chord diagrams for seven different types of (2, 2)soliton solutions. Each diagram parametrizes a unique cell in the totally nonnegative
Grassmannian Gr(2, 4).
3.3. N = 2 and M = 4 cases
By a direct construction of the derangements of S
4
with two exedances (i.e. N = 2), one can easily see that there
are seven cases with 2 × 4 irreducible and totally nonnegative Amatrices. Then the classiﬁcation theorems imply
that we have a (2, 2)soliton solution associated to each of those Amatrices. We here discuss all of those (2, 2)soliton
solutions with the same kparameters given by (k
1
, k
2
, k
3
, k
4
) = (−7/4, −1/4, 3/4, 3/2), and show how each Amatrix
chooses a particular set of linesolitons. In Fig. 5, we illustrate the corresponding chord diagrams for all those seven
cases, from which one can ﬁnd the asymptotic linesolitons in each case.
3.3.1. The case π = (3412)
Fromthe chord diagram in Fig. 5, one can see that the asymptotic linesolitons are given by [1, 3] and [2, 4]types
for both y → ±∞. The solution in this case corresponds to the top cell of Gr(2, 4), and the Amatrix is given by
A =
_
1 0 −c −d
0 1 a b
_
,
where a, b, c, d > 0 are free parameters with D:=ad − bc > 0. This is the generic solution on the maximal dimensional
cell, and it is called Ttype (after [12]). The most important feature of this solution is the generation of four intermediate
solitons forming a box at the intersection point. Those intermediate solitons are identiﬁed as [1, 2], [2, 3], [1, 4]
and [3, 4]solitons, and they may be obtained by cutting the chord diagram of (3412)type at the crossing points.
For example, if we cut the chords of [1, 3] and [2, 4] at the crossing point, we obtain either pair of {[1, 2], [3, 4]} or
{[1, 4], [2, 3]}. The ﬁrst pair appears in the lower and upper edges of the box, and the second pair in the right and
left edges. Fig. 6 shows the time evolution of the solution of this type. The parameters a, b, c and d in the Amatrix
determine the locations of solitons, their phase shifts and the onset of the boxshaped interaction pattern: Those are
given by
b =
2, 1
4, 1
se
Θ
+
[1,3]
, c =
2, 1
3, 2
se
Θ
+
[2,4]
,
a
d
=
4, 2
3, 1
r, D =
2, 1
4, 3
s, (3.3)
Fig. 6. Example of (3412)type soliton solution (Ttype). The asymptotic solitons are [1, 3] and [2, 4]types for both y →±∞. The intermediate
solitons forming a boxshape pattern are given by [1, 2],[1, 4], [2, 3] and [3, 4]solitons.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1195
where s is for the phase shift, Θ
+
[i,j]
for the location of [i, j]soliton for x 0 (recall the deﬁnition below (2.10), i.e.
x
+
[i,j]
= −1/(j, i)−Θ
+
[i,j]
/j, i gives the xintercept of the line of the [i, j]soliton for x 0 at t = 0), and r for the on
set of the box. The phase shifts for [1, 3] and [2, 4]solitons are given by x
[i,j]
:=x
−
[i,j]
− x
+
[i,j]
= −1/(j, i)−Θ
T
/j, i
with
Θ
T
= ln
_
1, 42, 3
1, 23, 4
bc
D
_
. (3.4)
In Fig. 6, we have chosen those parameters as s = 1, Θ
+
[1,3]
= Θ
+
[2,4]
= 0 and r = 1, so that at t = 0 the solution
forms an Xshape without phase shifts and opening of a box at the origin. In Fig. 6, the amplitudes are A
[1,3]
= 25/8
and A
[2,4]
= 49/36. Among the soliton solutions for N = 2 and M = 4, this solution is the most complicated and
interesting one. As you can see below, this solution contains all other solutions as some parts of this solution, that is,
as explained above, all six possible solitons (i.e. C
4
2
= 6) appear in this solution.
3.3.2. The case π = (4312)
Fromthe chord diagram in Fig. 5, one can see that the asymptotic linesolitons are given by [1, 4] and [2, 3]solitons
for y 0 and [1, 3] and [2, 4]solitons in y 0. The Amatrix in this case is given by
A =
_
1 0 −b −c
0 1 a 0
_
,
where a, b, c > 0 are free parameters. Fig. 7 illustrates an example of this solution. The parameters in the Amatrix
determine the locations of the linesolitons in the following form,
a =
2, 1
3, 1
e
Θ
−
[2,3]
, b =
2, 1
3, 2
e
Θ
−
[1,3]
and c =
3, 1
4, 3
e
Θ
+
[1,4]
.
The Θ
±
[ij]
represent the locations of the [i, j]soliton for x 0 and x 0, and the three linesolitons determine the
location of the other one. We here take the parameters Θ
−
[1,3]
= Θ
+
[1,4]
= Θ
−
[2,3]
= 0, so that all those four solitons meet
at the origin at t = 0. Notice here that the pattern in the lower part (y < 0) of the solution is the same as that of the
previous one, Ttype. This can be also seen by comparing the chord diagrams of those two cases.
3.3.3. The case π = (3421)
The asymptotic linesolitons in this case are given by [1, 3] and [2, 4]solitons for y 0 and [1, 4] and [2, 3]
solitons in y 0. The Amatrix has the form,
A =
_
1 0 0 −c
0 1 a b
_
,
Fig. 7. Example of (4312)type soliton solution. The asymptotic solitons are [1, 4] and [2, 3]types for y 0 and [1, 3] and [2, 4]types for y 0.
The intermediate solitons are [3, 4]type for t = −6 and [1, 2]type for t = 6.
1196 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
Fig. 8. Example of (3421)type soliton solution. The asymptotic solitons are given by [1, 3]and [2, 4]types for y 0 and [1, 4] and [2, 3]types
for y 0. This solution is dual to the previous one of (4312)type, i.e. the duality is given by (x, y, t) →(−x, −y, −t).
where a, b, c > 0 are free parameters. Fig. 8 illustrates the evolution of the solution of this type. The parameters in the
Amatrix are related to the locations of the linesolitons,
a =
2, 1
3, 1
e
Θ
+
[2,3]
, b =
2, 1
4, 1
e
Θ
−
[2,4]
and c =
2, 1
4, 2
e
Θ
−
[1,4]
.
This solution can be considered as a dual of the previous case (b), that is, two sets of linesolitons for y 0 and y 0
are exchanged. Here we take Θ
±
[i,j]
= 0 for those pairs (i, j) = (1, 4), (2, 3) and (2, 4), hence all those four solitons
meet at the origin at t = 0. Note here that now the upper pattern of the solution is the same as that of the Ttype. So if
one cuts Figs. 7 and 8 along the xaxis and glues the lower half of Fig. 7 together with the upper half of Fig. 8, we can
obtain Fig. 6 (Ttype). This can be observed from the corresponding chord diagrams.
One should also note that this solution is “dual” to the previous one in the sense that the patterns of those solutions
are symmetric with respect to the change (x, y, t) ↔ (−x, −y, −t).
3.3.4. The case π = (2413)
The asymptotic linesolitons are given by [1, 2] and [2, 4]solitons for y 0 and [1, 3] and [3, 4]solitons for
y 0. The Amatrix is given by
A =
_
1 0 −c −d
0 1 a b
_
,
where a, b, c, d > 0 with ad − bc = 0. Those parameters are related to the locations of the linesolitons,
b =
2, 1
4, 1
e
Θ
−
[2,4]
, c =
2, 1
3, 2
e
Θ
−
[1,3]
and
d
b
=
4, 1
4, 2
e
Θ
+
[1,2]
,
with a = bc/d. Fig. 9 illustrates the evolution of the solution of this type. For the case that all the solitons meet at
the origin at t = 0, we set all Θ
[i,j]
to be zero. Notice that the nonlinear interaction generates the intermediate soliton
of [1, 4]type in t < 0. This has the maximum amplitude among the (local) solitons appearing in the solution. It is
interesting to note that the maximum amplitude soliton might be expected from the chord diagram, that is, this soliton
is generated by the resonant interactions of [1, 2] + [2, 4] = [1, 4] for y > 0 and [1, 3] + [3, 4] = [1, 4] for y < 0.
Fig. 9. Example of (2413)type soliton solution. The asymptotic solitons are [1, 2] and [2, 4]types for y 0 and [1, 3] and [3, 4]types for y 0.
The intermediate solitons are [1, 4]type in t = −6 and [2, 3]type in t = 6.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1197
Fig. 10. Example of (3142)type soliton solution. The asymptotic solutions are [1, 3] and [3, 4]types for y 0 and [1, 2] and [2, 4]types
fory 0. Notice the duality of this with (2413)type in the previous example.
One should also note that the left patterns in Fig. 9 including [1, 4]soliton for t < 0 and [2, 3]soliton for t > 0 are the
same as those in Fig. 6 (Ttype).
3.3.5. The case π = (3142)
The asymptotic linesolitons are given by [1, 3] and [3, 4]solitons in y 0 and [1, 2] and [2, 4]solitons in y 0.
The Amatrix has the form,
A =
_
1 a 0 −c
0 0 1 b
_
,
where a, b, c > 0. We represent those parameters in the following form which is particularly useful for our stability
problem discussed in the next section,
a =
3, 1
3, 2
se
Θ
+
[2,4]
, b =
3, 1
4, 1
se
Θ
+
[1,3]
and c =
3, 1
4, 3
s. (3.5)
Here Θ
+
[i,j]
represent the location of the [i, j]soliton in x 0, and s represents the location of the other linesolitons
in x 0, that is the generation of the intermediate linesolitons either [1, 4] or [2, 3]type (c.f. the case (c) in Section
5). This solution is dual to the previous one of π = (2413). In fact, Fig. 10 illustrates the dual behavior of the evolution
in the sense of the reverse of time. Here we take s = 1 and Θ
+
[1,3]
= Θ
+
[2,4]
= 0, so that all solitons meet at the origin
at t = 0. As in the previous case, the large amplitude soliton of [1, 4]type is generated by the resonant interaction. In
particular, this solution can be used to explain the Mach reﬂection observed in shallow water [14,7,3].
3.3.6. The case π = (4321)
The asymptotic linesolitons are [1, 4] and [2, 3]types for both y → ±∞. Since those solitons can be placed both
in nearly parallel to the yaxis, this type of solutions ﬁts better in the physical assumption for the derivation of the KP
equation, i.e. a quasitwodimensionality. This is then referred to as Ptype (after [12]). The situation is similar to the
KdV case, for example, two solitons must have different amplitudes, A
[1,4]
> A
[2,3]
. The Amatrix is given by
A =
_
1 0 0 −b
0 1 a 0
_
.
The parameters are expressed by
a =
4, 2
4, 3
e
Θ
+
[2,3]
and b =
3, 1
4, 3
e
Θ
+
[1,4]
. (3.6)
where Θ
+
[i,j]
represents the location of the [i, j]soliton in x > 0 (the wave front) (notice the phase shifts for the solitons).
Fig. 11 illustrates the evolution of the solution of this type. We here set Θ
+
[i,j]
= 0, and hence the solitons in x > 0 (the
wave front) meet at the origin at t = 0. Note here that there is a phase shift due to the interaction, and the shift is in the
1198 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
Fig. 11. Example of (4321)type soliton solution (Ptype). The asymptotic solitons are [1, 4] and [2, 3]types for both y →±∞. The pattern of
the solution is stationary, and the interaction generates a negative phase shift of the solitons.
negative direction due to the repulsive force in the interaction similar to the case of KdV solitons. The phase shifts are
given by x
[i,j]
= −(1/j, i)−Θ
P
/j, i with
Θ
P
= ln
_
4, 23, 1
2, 14, 3
_
> 0. (3.7)
This negative phase shift is due to the repulsive force in the interaction as in the case of the KdV solitons. One should
also note that the interaction point moves in the negative ydirection, because the larger soliton of [1, 4]type propagates
faster than the other one. Those [1, 4] and [2, 3]solitons only appear as the intermediate solitons in the solution of
Ttype, that is, they form a part of the box in Fig. 6.
3.3.7. The case π = (2143)
The asymptotic linesolitons are [1, 2] and [3, 4]types for both y → ±∞. This solution was used originally to
describe two soliton solution, and it is called Otype (after [12]). The interaction property for solitons with equal
amplitudes has been discussed using this solution. However this solution becomes singular, when those solitons are
almost parallel to each other and close to the yaxis, contrary to the assumption of the quasitwo dimensionality for the
KP equation [14]. The Amatrix is given by
A =
_
1 a 0 0
0 0 1 b
_
.
Notice that this Amatrix is the limit of that for (3142)type with c → 0 and k
2
→ k
3
. We will further discuss this
issue in the next section when we study the initial value problem with Vshape initial wave. The parameters a and b
are expressed by
a =
4, 1
4, 2
e
Θ
+
[1,2]
and b =
3, 2
4, 2
e
Θ
+
[3,4]
. (3.8)
where Θ
+
[i, j] represents the location of the wave front consisting of [1, 2] and [3, 4]solitons. Fig. 12 illustrates the
evolution of this type solution. Here we take all Θ
+
[i,j]
= 0, the solitons in x > 0 (the wave front) meet at the origin at
Fig. 12. Example of (2143)type soliton solution (Otype). The asymptotic solitons are [1, 2] and [3, 4]types for both y →±∞. The pattern of
the solution is stationary, and has a positive phase shift in the solitons.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1199
t = 0. The phase shifts of those solitons are positive due to an attractive force in the interaction, and they are given by
x
[i,j]
= −(1/j, i)−Θ
O
/j, i with
Θ
O
= ln
_
4, 13, 2
3, 14, 2
_
< 0. (3.9)
When k
2
approaches to k
3
, the phase shift gets larger, and the interaction part of the solution is getting to be close to
[1, 4]soliton as a result of the resonant interaction with [1, 2] and [3, 4]solitons at k
2
= k
3
. This is the main argument
of the paper [14] about the discussion on the Mach reﬂection of shallow water waves. Note also that those [1, 2] and
[3, 4]solitons appear as the intermediate solitons forming a box of Ttype as shown in Fig. 6.
4. Numerical simulations
The main purpose of the numerical simulation is to study the interaction properties of linesolitons, and we will
show that the solution of the initial value problem with certain types of initial waves asymptotically approaches to
some of the exact solutions discussed in the previous section. This implies a stability of those exact solutions under
the inﬂuence of certain deformations (notice that the deformations in our initial waves are not so small).
The initial value problem considered here is essentially an inﬁnite energy problem in the sense that each linesoliton
in the initial wave is supported asymptotically in either y 0 or y 0, and the interactions occur only in a ﬁnite domain
in the xyplane. In the numerical scheme, we consider the rectangular domain D = {(x, y) : x ≤ L
x
, y ≤ L
y
}, and
each linesoliton is matched with a KdV soliton at the boundaries y = ±L
y
. In this section, we explain the details of
our numerical scheme and give an error estimate of the scheme.
We also remark that for onedimensional problem (e.g. the KdV equation), the accuracy for numerical scheme is
usually tested by using the integrals such as the total mass and energy. However, in our study, we are dealing with
twodimensional problem with nonvanishing boundary values which provide energy ﬂow into the numerical domain.
So we cannot use the integrals for the accuracy test of our numerical scheme. We then check the accuracy with the
maximumerrors in the Fourier domain and the L
2
estimates in comparison with the exact solutions of the KP equation.
4.1. Numerical scheme
Using the socalled window technique [21,23], we ﬁrst transform our nonperiodic solution u having nonvanishing
boundary values into a function v rapidly decaying near the boundaries y = ±L
y
, so that we consider the function v
to be periodic in y. With the window function W(y), this can be expressed by the decomposition of the solution u, i.e.
u = v + (1 − W)u with v = Wu.
We take the window function W(y) in the form of a superGaussian,
W(y) = exp
_
−a
n
¸
¸
¸
¸
y
L
y
¸
¸
¸
¸
n
_
, (4.1)
where a
n
and n are positive constants (we choose a
n
= (1.111)
n
ln 10 with n = 27 [23]). We assume that near the
boundaries y = ±L
y
, the solution can be described by exact linesolitons (i.e. we take L
y
large enough so that the
interaction will not inﬂuence those parts near boundaries up to certain ﬁnite time). We then consider the transformation
in the form,
u = v + (1 − W)u
0
, (4.2)
where u
0
consists of exact linesolitons satisfying
(1 − W)u
0
= (1 − W)u.
This assumption prevents any disturbance from the boundary, and keeps the stability of the far ﬁelds consisting of
wellseparated linesolitons. In x direction, we choose L
x
large enough so that u(±L
x
, y) ≈ v(±L
x
, y) ≈ 0 initially.
We are interested in studying the asymptotic behavior near the interaction point, and we do not expect to see any
strong effect from the boundary region in a short time (recall that the system has a ﬁnite group velocity for waves
1200 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
with nonzero k
x
component and the assumption of quasitwodimensionality, i.e. k
2
y
≈ k
x
, in the derivation of the KP
equation). With the transformation (4.2), we have the equation for v,
∂
x
_
4∂
t
v + 6v∂
x
v + ∂
3
x
v
_
+3∂
2
y
v = 6(1 − W)∂
x
(Wu
0
∂
x
u
0
− ∂
x
(vu
0
)) +3(2W
∂
y
u
0
+ W
u
0
). (4.3)
Assuming the solution v to be periodic in both x and y with zero boundary condition at y = L
y
, we solve (4.3) by
using fast Fourier transform (FFT).
1
For a convenience, let us rescale the domain D = {(x, y) : x ≤ L
x
, y ≤ L
y
}
with D:={(X, Y) : X ≤ π, Y ≤ π}, i.e.
X =
π
L
x
x, Y =
π
L
y
y.
Then (4.3) becomes
∂
X
_
∂
t
v + Pv∂
X
v + Q∂
3
X
v
_
+R∂
2
Y
v = F, (4.4)
where F is the right hand side of (4.3) with the scalings, and
P =
3π
2L
x
, Q =
π
3
4L
3
x
, and R =
3πL
x
4L
2
y
.
Under the periodic assumption of v and the meanfree condition, we have
v(X, Y, t) =
∞
l=−∞
∞
m=−∞
ˆ v(l, m, t)e
i(lX+mY)
.
Here the Fourier transformation ˆ v:=F(v) is given by
ˆ v(l, m, t) =
1
(2π)
2
_
π
−π
_
π
−π
v(X, Y, t)e
−i(lX+mY)
dX dY,
and the equation becomes
∂
t
ˆ v + i
_
Rm
2
l
− Ql
3
_
ˆ v +il
P
2
N(ˆ v) =
1
il
F(F), for l / = 0, (4.5)
with
N(ˆ v) = F(v
2
).
Since our solutions do not satisfy the meanfree condition, i.e.
_
L
x
−L
x
u dx = 0, our scheme based on the FFT generates
some error (the zero mode cannot satisfy (4.5) for those solutions discussed in this paper). However, within our
computational times, we conﬁrmed that the accuracy of our scheme is not contaminated by the error introduced from
the zero modes (see below). We also checked the error due to the violation of the periodicity due to the meanfree
condition even in the case where the zero modes of u is forced to be zero. The algorithm seems to be quite stable within
our computational times. The zero mode effect was also discussed numerically in [11].
2
Let c = Rm
2
/l − Ql
3
and
ˆ
V = e
ict
ˆ v. Then we have
∂
t
ˆ
V + il
P
2
e
ict
N(
ˆ
V e
−ict
) =
1
il
e
ict
F(F), for l / = 0. (4.6)
We write this equation in the form of time evolution,
∂
t
ˆ
V = G(t,
ˆ
V ).
1
The periodic assumption implies that one needs to have the meanfree function in the xdirection, i.e.
_
L
x
−L
x
udx = 0. Although our solutions
do not satisfy this condition, we conﬁrmed that the overall error including the violation of the condition remained small within our accuracy (see
below).
2
We would like to thank a referee for pointing out the issue of the zero mode for the numerical integration of the KP equation based on FFT
method.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1201
Fig. 13. Numerical accuracy test on Ttype solution shown in Fig. 6. The graph shows the estimate of maxF(v
t
) −F(v
t
exact
) as a function of the
number of Fourier modes N ×N/2 for t = 0, 3 and 6. The error is decreasing exponentially with respect to N until it saturates at round of error
≈ 10
−16
for N ≈ 1280, 1472 and 1536 for t = 0, 3 and 6, respectively. The numerical domain is D = [−64, 64] ×[−32, 32].
and use the 4th order RungeKutta (RK4) method to solve the equation. We use a pseudospectral method to evaluate
nonlinear term N(ˆ v) in (4.6). For the function F on the right hand side of (4.3), one needs to evaluate the derivatives
of the exact solution u
0
. However, these derivatives have rather complex formulae, especially those of Ttype solution,
we estimate the F in the 2nd order difference. Thus the numerical solutions are expected to be of 2nd order accuracy
when there are enough Fourier modes in space and the solutions are vanished in large x.
In order to verify the accuracy of the method, we provide two tests. The ﬁrst test intends to answer the ques
tion on how many Fourier modes are needed to ensure machine accuracy in space for the solution v of (4.3).
The second test shows the overall algorithm is 2nd order accurate. In Fig. 13, we show the maximum error in the
Fourier modes of the Ttype solution in the sense that we estimate maxF(v
t
) − F(v
t
exact
) with respect to the num
ber of Fourier modes N in both x and ydirections for the case shown in Fig. 6. On the computational domain
D = [−64, 64] × [−32, 32], machine accuracy ≈ 10
−16
is reached for N ≈ 1280, 1472 and 1536 for t = 0, 3 and
6, respectively. Since the Ttype solution develops a box at later time, it requires more Fourier modes to well repre
sent the solution. From this test with N = 1536, we ﬁnd that we need roughly 1536/(2 × 64) = 12 Fourier modes
per 1 unit in x to reach 10
−16
accuracy. Notice that it might take longer time to run such a ﬁne mesh for a very
large computational domain. We choose to use 8 Fourier modes per unit length in both x and ydirections for all
the simulations discussed in this paper. The spacial error is within 10
−10
and it can also speed up the computa
tion. In Table 1, we summarize the error of the solution given by u
t
− u
t
exact
L
2
(D)
/u
t
exact
L
2
(D)
for two different
time steps t = 0.01 and t = 5 × 10
−3
on D = [−128, 128] × [−16, 16] with 2048 × 256 Fourier modes. It can
be clearly seen that the order of convergence is 2 because the error decreases by a factor around 4 when the time
step is halved. If we take t = 5 × 10
−3
and keep running the simulation up to t = 12, the magnitude of the
error is always within 2.5 × 10
−2
. In the following simulations, the errors of the solutions are at this magnitude or
smaller.
Table 1
The convergence test of our numerical scheme on the domain D = [−128, 128] ×[−16, 16] with 2048 ×256 Fourier modes. The table shows the
error u
t
−u
t
exact
L
2
(D)
/u
t
exact
L
2
(D)
with Ttype exact solution for the time steps t = 10
−2
and 5 ×10
−3
. This indicates that our scheme is of
2nd order accuracy in time.
t = 10
−2
t = 5 ×10
−3
Order
t = 1 1.046 ×10
−2
2.645 ×10
−3
1.98
t = 2 1.893 ×10
−2
4.353 ×10
−3
2.12
t = 3 2.507 ×10
−2
5.986 ×10
−3
2.07
t = 4 2.992 ×10
−2
7.732 ×10
−3
1.95
t = 5 3.523 ×10
−2
9.675 ×10
−3
1.86
t = 6 4.211 ×10
−2
1.194 ×10
−2
1.82
1202 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
5. Numerical results
Now we present the results of the direct numerical simulation of the KP equation with the following two types of
initial waves consisting of two linesolitons. In particular, we ﬁx the amplitude of one of the linesolitons, and place
these linesolitons in the symmetric shape with respect to the xaxis. This then implies that each proﬁle of the initial
wave is determined by two parameters A
0
(the amplitude of other soliton) and Ψ
0
> 0 (the angle of the wavevector
of the soliton in y < 0):
(i) Vshape wave consisting of two semiinﬁnite linesolitons, i.e. u(x, y, 0) = u
0
1
(x, y) + u
0
2
(x, y) with
⎧
⎪
⎨
⎪
⎩
u
0
1
(x, y) = A
0
sech
2
_
A
0
2
(x − y tan Ψ
0
) × H(y),
u
0
2
(x, y) = 2 sech
2
(x + y tan Ψ
0
) × H(−y),
(5.1)
where H(z) is the unit step function,
H(z) =
_
1 if z ≥ 0,
0 if z < 0.
Here one should note that the initial locations x
+
[i,j]
are zero for both linesolitons, that is, they cross the origin at
t = 0. The shifts of x
+
[i,j]
observed in the simulations provide the information of the interaction property.
(ii) Xshape wave of the linear combination of two linesolitons, i.e. u(x, y, 0) = u
0
1
(x, y) + u
0
2
(x, y) with
⎧
⎪
⎨
⎪
⎩
u
0
1
(x, y) = A
0
sech
2
_
A
0
2
(x − y tan Ψ
0
),
u
0
2
(x, y) = 2 sech
2
(x + y tan Ψ
0
).
(5.2)
Again note that the initial locations are x
+
[i,j]
= 0.
These initial waves have been considered in certain physical models of interacting waves [14,7,20,22,3,13,25]. In
particular, in [20,22,13,25], the Vshape initial waves were considered to study the generation of large amplitude waves
in shallow water.
The main result of this section is to show numerically that those initial waves converge asymptotically to some of
the exact solutions presented in Section 3. Here what we mean by the convergence is as follows: We ﬁrst show that
the solution u
t
(x, y) at the intersection point of V or Xshape generates a pattern due to the resonant interaction, and
we then identify the pattern near the resonant point with that given by some of the exact solutions u
t
exact
(x, y). The
convergence is only in a local sense, and we use the following (relative) error estimate,
E(t):=u
t
− u
t
exact
2
L
2
(D
t
r
)
/ u
t
exact
2
L
2
(D
t
r
)
, (5.3)
where D
t
r
is the circular disc deﬁned by
D
t
r
:=
_
(x, y) ∈ R
2
: (x − x
0
(t))
2
+ (y − y
0
(t))
2
≤ r
2
_
.
Here the center (x
0
(t), y
0
(t)) is chosen as the intersection point of two lines determined from the corresponding exact
solution identiﬁed by the following steps:
1. For a given initial wave, we ﬁrst identify the type of exact solution (i.e. the chord diagram) from the formulae,
A
[i,j]
=
1
2
(k
i
− k
j
)
2
and tan Ψ
[i,j]
= k
i
+ k
j
. This then gives the exact solution with the corresponding Amatrix.
Note here that the entries of the Amatrix are not determined yet.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1203
2. We then minimize the error function E(t) of (5.3) at certain large time t = T
0
by changing the entries in the Amatrix,
that is, we consider
Minimize
A−matrix
E(T
0
). (5.4)
The minimization is achieved by adjusting the solution pattern with the exact one, that is, the pattern determines
the Amatrix whose entries give the locations of all linesolitons including the intermediate solitons in the solution.
With this Amatrix, the center (x
0
(t), y
0
(t)) of the circular domain D
t
r
is given by the intersection point of the lines,
x + y tan Ψ
[i
l
,j
l
]
− C
[i
l
,j
l
]
t = x
+
[i
l
,j
l
]
for l = 1, 2, (5.5)
where the locations x
+
[i
l
,j
l
]
are expressed in terms of the Amatrix (see Section 3). Here we have A
[i
1
,j
1
]
=
A
0
, A
[i
2
,j
2
]
= 2 and Ψ
[i
2
,j
2
]
= −Ψ
[i
1
,j
1
]
= Ψ
0
. The time T
0
may be taken as T
0
= 0, but for most of the cases,
we need T
0
to be sufﬁciently large. This implies that the solutions have nontrivial location shifts, i.e. x
+
[i,j]
/ = 0 in
the solution.
3. We then conﬁrm that E(t) further decreases for a larger time t > T
0
upto a time t = T
1
> T
0
, just before the effects
of the boundary enter the disc D
t
r
(those effects include the periodic condition in x and a mismatch on the boundary
patching).
We take the radius r in D
t
r
large enough so that the main interaction area is covered for all t < T
0
, but D
t
r
should be
kept away from the boundary to avoid any inﬂuence coming from the boundaries. The time T
1
> T
0
gives an enough
time to develop a pattern close to the corresponding exact solution, but it is also limited to avoid any disturbance
from the boundaries for t < T
1
. In this paper, we give T
0
and T
1
based on the observation of the numerical results.
Unfortunately, we do not have a systematic way to estimate those values, T
0
, T
1
and the radius r. We are currently
working on this problem, and hope will be able to report the complete analysis in a future communication.
5.1. Vshape initial waves
The initial waveform is illustrated in the left panel of Fig. 14. The right ﬁgure shows the (incomplete) chord diagrams
corresponding to the initial Vshape waves; that is, the upper chord represents the (semiinﬁnite) linesoliton in y > 0
and the lower one represents the (semiinﬁnite) linesoliton in y < 0. We have done the numerical simulations with
six different cases which are marked in Fig. 14 as (a) through (f). The main result in this section is to show that each
incomplete chord diagram representing the initial wave gets a completion with minimum additional chords which
represents the asymptotic pattern of the solution. Here the minimal completion implies that for a given incomplete
chord diagram, one adds extra chords with minimum total length so that the complete chord diagram represents a
derangement (see also [13]). For our examples of the cases (a) through (f), we have the following completions:
Fig. 14. Vshape initial waves. Each line of the Vshape is a semiinﬁnite linesoliton. We set those linesolitons to meet at the origin, and ﬁx
the amplitude of the soliton in y < 0 to be 2. A
0
is the amplitude of the soliton in y > 0. The right panel shows the incomplete chord diagrams
corresponding to the choice of the values A
0
and Ψ
0
. There is no corresponding exact soliton solution on the dotted line of k
2
= k
3
.
1204 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
(a) With k
1
= k
2
, the completion is π =
_
1 3 4
3 4 1
_
, i.e. add [3, 4]chord in y > 0. This implies that [3, 4]soliton
appears in the asymptotic solution in y > 0, and the corresponding exact solution is a (1, 2)soliton (cf. Fig. 3).
(b) With k
3
= k
4
, the completion is π =
_
1 2 3
3 1 2
_
, i.e. add [1, 2]chord in y < 0 which appears in the asymptotic
solution, and the corresponding solution is a (2, 1)soliton (cf. Fig. 3).
(c) The completion is π = (3142), i.e. add [1, 2]chord in y < 0 and [3, 4]chord in y > 0. These corresponding
solitons appear behind those initial solitons (cf. Fig. 10).
(d) The completion is π = (2143), i.e. add [1, 2]chord in y < 0 and [3, 4]chord in y > 0. The soliton solution is of
Otype, and the semiinﬁnite initial solitons extend to generate other half with the phase shift (cf. Fig. 12).
(e) The minimal completion is π = (2341), i.e. add [1, 2] and [3, 4]chords in y > 0. This is a (1, 3)soliton. (Note
that the initial wave is a front half of Ptype soliton solution, but the minimal completion is not of Ptype.) Since
[1, 2]soliton in y > 0 appears ahead of the wavefront consisting of [2, 3]soliton in y > 0 and [1, 4]soliton in
y < 0, the generation of [1, 2]soliton is impossible. The [1, 2]soliton is a virtual one, but it gives an inﬂuence on
bending the [1, 4]soliton in y > 0 (see below for the details).
(f) The minimal completion is π = (4123), i.e. add [1, 2] and [3, 4]chords in y < 0. The is a (3, 1)type, but again
we note that [3, 4]soliton in y < 0 is a virtual one (i.e. it locates ahead of the wavefront similar to the previous
case (e)). We also observe the bending of [1, 4]soliton in y > 0 under the inﬂuence of this virtual soliton (see
again below for the details).
5.1.1. The case (a)
This is a critical case withk
1
= k
2
. The linesolitons of Vshape initial wave are [1, 3]soliton in y > 0 and [1, 4]
soliton in y < 0. We take the following values for the amplitudes and angles of the those solitons,
⎧
⎪
⎨
⎪
⎩
A
[1,3]
= A
0
=
1
2
, tan Ψ
[1,3]
= −
1
2
= − tan Ψ
0
,
A
[1,4]
= 2, tan Ψ
[1,4]
=
1
2
= tan Ψ
0
.
With the formulas for the amplitude A
[i,j]
=
1
2
(k
i
− k
j
)
2
and the angle tan Ψ
[i,k]
= k
i
+ k
j
, we have the kparameters
(k
1
= k
2
, k
3
, k
4
) = (−3/4, 1/4, 5/4). The corresponding exact solution is of (1, 2)type.
The ﬁgures in the ﬁrst row of Fig. 15 illustrate the result of the direct numerical simulation. The domain of
computation is [−192, 192] × [−48, 48] and there are 8 Fourier modes in every unit length. The simulation shows a
generation of third wave at the point of the intersection of those initial solitons. This generation is due to the resonant
interaction, and this may be identiﬁed as [3, 4]soliton of the exact solution, that is, the resonance condition gives
[1, 3] + [3, 4] = [1, 4] and completes the incomplete chord diagram. The generation of [3, 4]soliton may be also
understood by decomposing the exact solution into two parts, say u
exact
= u + v, where u is the solution of the KP
equation with our initial wave of [1, 3] and [1, 4]solitons. Then v satisﬁes,
∂
x
(4∂
t
v + 6v∂
x
v + ∂
3
x
v) + 3∂
2
y
v = −6∂
2
x
(uv),
which is the KP equation with one extra term in the right hand side. If we assume that the initial solitons remain as
the part of the exact solution u
exact
, the product uv in the extra term has nonzero value only near the origin at t = 0
(this assumption is not totally correct, but the shifts of the initial solitons are very small for most of the cases). We then
expect v to be close to the solution of the KP equation with the initial wave of [3, 4]soliton only for y > 0 and x < 0.
Since this [3, 4]soliton is missing in the front half (x > 0), the edge of the wave gets a strong dispersive effect. Then
the edge of the solution v disperses away with a bowshape wake propagating toward the negative xaxis as can be seen
in the simulation (cf. Fig. 15). Recall here that the dispersive waves (i.e. nonsoliton parts) propagate in the negative
xdirection (see Section 3). Then the decay of the v solution at the front edge implies the appearance of the [3, 4]soliton
in the exact solution as one can see from u = u
exact
− v. This argument may be also applied to other cases as well.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1205
Fig. 15. Numerical simulation of the case (a) in Fig. 14: The initial wave consists of [1, 3]soliton in y > 0 and [1, 4]soliton in y < 0. The ﬁgures
in the ﬁrst row show the result of the direct simulation, the ﬁgures in the second row show the corresponding exact solution of (1, 2)type, and the
bottomone represents the error function E(t) of (5.3) which is minimized at t = 10. The domain of the simulation is [−192, 192] ×[−48, 48]. The
circle in the exact solution shows the domain D
t
r
with r = 10 for the error function E(t). A dispersive radiation appearing behind the interaction
region may be considered as a reaction of the generation of the third soliton of [3, 4]type as a result of resonant interaction (see the text for the
details).
Minimizing the error function E(t) at t = T
0
= 10 with (5.4), we obtain the Amatrix of (1, 2)type,
A =
_
1 0 −1.27
0 1 0.63
_
,
The minimization is achieved by adjusting the locations of the [1, 3] and [1, 4]solitons of the wavefront, and it gives
the shifts
x
+
[1,3]
= −0.0079, x
+
[1,4]
= −0.120.
(The Amatrix is then obtained using (3.2)). Thus the new linesoliton of [3, 4]type has a relatively large negative
shift. This may be explained as follows: The [3, 4]soliton is generated as a part of the tail of [1, 4]soliton, and in the
beginning stage of the generation, the [3, 4]soliton has a small amplitude and propagates with slow speed. This causes
a negative shift of this soliton. Also the negative shift in the [1, 4]soliton is due to the generation of the [3, 4]soliton,
i.e. losing its momentum. The ﬁgures in the second row of Fig. 15 show the exact solutions generated by the τfunction
with those parameters (k
1
= k
2
, k
3
, k
4
) = (−3/4, 1/4, 5/4)) and the Amatrix given above. The circles in the ﬁgures
showD
t
r
where we take r = 12 and the center is given by (x
0
(t) = (13/16)t − 0.0637, y
0
(t) = (3/4)t − 0.112). In the
bottom ﬁgures of Fig. 15, we plot the error function E(t) of (5.3). Fig. 15 indicates that the asymptotic solution seems
to converge to the exact solution of (1, 2)type with the above Amatrix and the same kparameters.
1206 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
5.1.2. The case (b)
This is also a critical case with k
3
= k
4
. The linesolitons of Vshape initial wave are [1, 3]soliton in y > 0 and
[2, 3]soliton in y < 0. We take the amplitudes and angles of those solitons to be
⎧
⎪
⎪
⎨
⎪
⎪
⎩
A
[1,3]
= A
0
=
49
8
, tan Ψ
[1,3]
= −
3
4
= − tan Ψ
0
,
A
[2,3]
= 2, tan Ψ
[2,3]
=
3
4
= tan Ψ
0
.
This then gives the kparameters (k
1
, k
2
, k
3
= k
4
) = (−17/8, −5/8, 11/8). Fig. 16 illustrates the result of the numerical
simulation. The ﬁgures in the ﬁrst row show the direct simulation of the KP equation. Notice again that a bowshape
wake behind the interaction point corresponds to the v solution in the decomposition, u = u
exact
− v, and the decay of v
implies the appearance (or generation) of [1, 2]soliton. Notice that v gives the [1, 2]soliton part in the exact solution.
The ﬁgures in the second row show the corresponding exact solution of (2, 1)type, whose Amatrix is obtained by
minimizing the error function E(t) at t = 6,
A =
_
1 1.15 1.21
_
.
The shifts of locations x
+
[i,j]
obtained in the minimization are given by
x
+
[1,3]
= −0.0545, x
+
[2,3]
= −0.0254,
Fig. 16. Numerical simulation of the case (b) in Fig. 14: The initial wave consists of [1, 4] and [2, 4]solitons in y > 0 and y < 0 respectively. The
ﬁgures in the ﬁrst rowshowthe result of the direct simulation, and the ﬁgures in the second rowshowthe corresponding exact solution of (2, 1)type.
The computational domain is [−192, 192] ×[−48, 48], and the circle in the ﬁgures of the exact solution shows D
t
r
with r = 12. The bottom graph
shows the error function E(t) of (5.3) which is minimized at t = 6.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1207
using (3.1). Those negative shifts may be explained in the similar way as in the previous case. In those ﬁgures, the
domain D
t
r
has r = 12 and its center (x
0
(t) = (157/64)t − 0.0399, y
0
(t) = −(11/8)t + 0.0194). The bottom graph in
Fig. 16 shows that the solution converges asymptotically to the exact solution of (2, 1)type with the above Amatrix
and the same kparameters.
5.1.3. The case (c)
The linesolitons of the initial wave are [1, 3]soliton in y > 0 and [2, 4]soliton in y < 0. We consider the case with
the larger soliton in y > 0, and take the amplitudes and angles of those solitons as
_
A
[1,3]
= A
0
= 2, tan Ψ
[1,3]
= −1 = − tan Ψ
0
,
A
[2,4]
= 2, tan Ψ
[2,4]
= 1 = tan Ψ
0
.
The kparameters are then given by (k
1
, k
2
, k
3
, k
4
) = (−3/2, −1/2, 1/2, 3/2). The exact solution obtained by the
completion of the chord diagram is given by (3142)type soliton solution. Fig. 17 illustrates the result of the numerical
simulation. The ﬁgures in the ﬁrst row show the direct simulation of the KP equation. We observe a bowshape wake
behind the interaction point. The wake expands and decays, and then we see the appearance of new solitons which form
resonant interactions with the initial solitons. One should note that the solution generates a large amplitude intermediate
soliton at the interaction point, and this soliton is identiﬁed as [1, 4]soliton with the amplitude A
[1,4]
= 4.5. This
solution has been considered to describe the Mach reﬂection of shallow water waves, and the [1, 4]soliton describes
the wave called the Mach stem [3,13].
Fig. 17. Numerical simulation of the case (c) in Fig. 14: The initial wave consists of [1, 3]soliton in y > 0 and [2, 4]soliton in y < 0. The ﬁgures
in the ﬁrst row show the result of the direct simulation, and the ﬁgures in the second row show the corresponding exact solution of (3142)type.
The computational domain is [−192, 192] ×[−48, 48], and the circle in the middle ﬁgures shows D
t
r
with r = 12. The bottom graph of the error
function E(t) which is minimized at t = 10. A large amplitude intermediate soliton is generated at the intersection point, and it corresponds to the
[1, 4]soliton with the amplitude A
[1,4]
= 4.5.
1208 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
The ﬁgures in the second row in Fig. 17 show the corresponding exact solution of (3142)type whose Amatrix is
found by minimizing E(t) at t = 10,
A =
_
1 1.92 0 −1.96
0 0 1 0.64
_
In the minimization to ﬁnd the Amatrix, we adjust the locations of solitons including the intermediate solitons, and
using (3.5), we obtain
x
+
[1,3]
= x
+
[2,4]
= −0.01, s = 0.980.
Those values indicate that the solution is very close to the exact solution for all the time. The negative value of the shifts
x
+
[i,j]
is due to the generation of a large amplitude soliton [1, 4]type (i.e. the initial solitons loose their momentum), and
s < 1 implies that the [1, 4]soliton is generated after t = 0. Also note that the [1, 4]soliton now resonantly interact
with [1, 3] and [2, 4]solitons to create new solitons [1, 2] and [3, 4]solitons (called the reﬂected waves as considered
in the Mach reﬂection problem [14,3,13]). This process then seems to compensate the shifts of incident waves, even
though we observe a large wake behind the interaction point. Notice that the wake disperses, and the exact solution
starts to appear as the wake decays (recall again the decomposition u
exact
= u + v (where v is the wake and u is the
solution).
The bottom graph in Fig. 17 shows a rapid convergence of the initial wave to (3142)type soliton solution with those
parameters of the Amatrix and k values given above, and for the error function, we use D
t
r
with r = 12.
5.1.4. The case (d)
The linesolitons of the initial wave are [1, 2]soliton in y > 0 and [3, 4]soliton in y < 0. This case corresponds
to the solitons with a large angle interaction. For simplicity, we consider a symmetric initial data with the initial wave
whose amplitudes and angles are given by
⎧
⎪
⎨
⎪
⎩
A
[1,2]
= A
0
= 2, tan Ψ
[1,2]
= −
12
5
= − tan Ψ
0
,
A
[3,4]
= 2, tan Ψ
[3,4]
=
12
5
= tan Ψ
0
.
The kparameters are then given by (k
1
, k
2
, k
3
, k
4
) = (−11/5, −1/5, 1/5, 11/5). The completion of the chord diagram
gives Otype soliton solution, i.e. (2143)type. This implies that the solution extends the initial solitons to the negative
xdirection with the same types, i.e. [1, 2] and [3, 4]types in y < 0 and y > 0, respectively. One should note that the
solution also generates the phase shift determined uniquely by the kparameters. Fig. 18 illustrates the result of the
numerical simulation. The ﬁgures in the ﬁrst row show the direct simulation of the KP equation. The wake behind the
interaction point has a large negative amplitude, and this corresponds to the v solution in u
exact
= u + v. The ﬁgures in
the second row show the corresponding Otype exact solution whose Amatrix is determined by minimizing the error
function E(t) at t = 10,
A =
_
1 1.91 0 0
0 0 1 0.17
_
Because of our symmetric initial data, we consider the shifts x
+
[1,2]
= x
+
[3,4]
for the minimization, and using (3.8), we
obtain
x
+
[1,2]
= x
+
[3,4]
= −0.020.
The negative shifts imply the slowdown of the incidence waves due to the generation of the solitons extending the
initial solitons in the negative xdirection. Those new parts of solitons have the locations x
−
[i,j]
. The phase shifts
x
[i,j]
= x
−
[i,j]
− x
+
[i,j]
for the linesolitons are calculated from (3.9), and they are
x
[1,2]
= x
[3,4]
= 0.593.
The positivity of the phase shifts is due to the attractive force between the linesolitons, and this explains the slowdown
of the initial solitons, i.e. the small negative shifts of x
+
[i,j]
. The bottom graph in Fig. 18 shows E(t) which is minimized
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1209
Fig. 18. Numerical simulation of the case (d) in Fig. 14: The initial wave consists of [1, 2]soliton in y > 0 and [3, 4]soliton in y < 0. The upper
ﬁgures show the result of the direct simulation, and the ﬁgures in the second row show the corresponding exact solution of (2143)type (Otype).
The domain of the simulation is [−160, 160] ×[−32, 32] and the circle in the exact solution shows D
t
r
with r = 12. Notice a large wake behind
the interaction point which extends the initial solitons. The bottom graph shows the error function E(t) which is minimized at t = 6.
at t = 6, and we take r = 12 for the domain D
t
r
. One can see a rapid convergence of the solution to the Otype exact
solution with those parameters. One should however remark that for a case near the critical one with k
2
= k
3
, there
exists a large phase shift in the soliton solution, and the convergence is very slow. Note that the case with k
2
= k
3
also corresponds to the boundary of the previous case (c) (See Fig. 14), and the amplitude of the intermediate soliton
generated at the intersection point reaches four times larger than the initial solitons at this limit. This large amplitude
wave generation has been considered as the Mach reﬂection problem of shallow water wave [14,20,3,13,23].
5.1.5. The case (e)
The linesolitons of the initial wave are [2, 3]soliton in y > 0 and [1, 4]soliton in y < 0. We consider the solitons
parallel to the yaxis with the amplitudes,
_
A
[2,3]
= A
0
= 0.25, tan Ψ
[2,3]
= 0 = − tan Ψ
0
,
A
[1,4]
= 2, tan Ψ
[1,4]
= 0 = tan Ψ
0
.
Then the kparameters are given by (k
1
, k
2
, k
3
, k
4
) = (−1, −1/(2
√
2), 1/(2
√
2), 1). We also considered the case with
nonzero angle (i.e. oblique to the yaxis with Ψ
0
> 0), and we expected to see a Ptype soliton as the asymptotic solution.
However, the numerical simulation suggests that we do not see a solution of Ptype, but the asymptotic solution is
somewhat close to the exact solution of (2341)type as obtained by the minimal completion of the chord diagram.
The (2341)type is a (1, 3)type soliton solution with [1, 4]soliton in y < 0 and [1, 2], [2, 3] and [3, 4]solitons in
y > 0. Here one should note that the [1, 2]soliton in y > 0 locates ahead of the wavefront consisting of [1, 4] and
[2, 3]solitons, so that this soliton cannot be generated by the resonant interaction of those [1, 4] and [2, 4]solitons.
1210 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
Fig. 19. Numerical simulation of the case (e) in Fig. 14: The initial wave consists of [2, 3]soliton in y > 0 and [1, 4]soliton in y < 0. The upper
ﬁgures show the result of the direct simulation, and the lower ﬁgures show the corresponding exact solution of (2341)type obtained by the minimal
completion of the chord diagram given by the initial wave. The computational domain is [−192, 192] ×[−32, 32]. Notice that the bending on the
initial [1, 4]soliton appears near the point where the virtual[1, 2]soliton intersects. The dispersive radiation behind the intersection point may be
the part of this [1, 4]soliton.
Fig. 19 illustrates the result of the numerical simulation. The ﬁgures in the ﬁrst row show the direct simulation of
the KP equation. We observe that the [1, 4]soliton in y < 0 propagates faster and its tail extends to y > 0. Also note
that a large dispersive wake appearing in y < 0 has negative amplitude and disperses in the negative xdirection. On
the other hand, the tail part in y > 0 seems to have a resonant interaction with the [2, 3]initial soliton and becomes a
newsoliton. This behavior of the solution may be explained by the chord diagram of (2341)type whose corresponding
exact solution is illustrated in the bottom ﬁgures in Fig. 19: The new soliton generated in y > 0 may be identiﬁed as
[3, 4]soliton. Also we note that the bending of the [1, 4]soliton observed in the simulation seems to appear at the point
where the [1, 2]soliton has a resonant interaction with [1, 4]soliton. This means that the intermediate soliton in the
exact solution is [2, 4]soliton, and this soliton may give a good approximation of the bending part of the [1, 4]soliton.
It is interesting to consider this [1, 2]soliton as a virtual one, which is not visible but has an inﬂuence to the initial
solitons. This can be also understood by the similar argument using the decomposition u
exact
= u + v. In this case, the
initial data for v is given by the [1, 2] and [3, 4]solitons in u
exact
. Then the [3, 4]soliton in u
exact
appears when the
solution v decays in this part. On the other hand, the decay of v in the [1, 2]soliton part seems to occur only in the
region behind the front solitons as seen at t = 5. Nondecaying of the [1, 2]soliton implies that [1, 2]soliton never
appears in the solution u, because of the cancellation of u
exact
and v on this soliton part. We will give more details in a
future communication.
5.1.6. The case (f)
The linesolitons of the initial wave are [1, 4]soliton in y > 0 and [2, 3]soliton in y < 0. Contrary to the previous
case, we have now the larger soliton in y > 0 and the smaller one in y < 0. In a comparison, we also consider the
solitons parallel to the yaxis. We then take the amplitudes and angles of the those solitons as
_
A
[1,4]
= A
0
= 6, tan Ψ
[1,4]
= 0 = − tan Ψ
0
,
A
[2,3]
= 2, tan Ψ
[2,3]
= 0 = tan Ψ
0
.
Thenthe kparameters are given by (k
1
, k
2
, k
3
, k
4
) = (−
√
3, −1, 1,
√
3). Since the upper [1, 4]soliton propagates faster
than the lower one, the lower tail of this soliton appears and then resonates with the lower [2, 3]soliton. Following the
similar arguments as in the previous case, we expect that (3, 1)soliton is the corresponding exact solution which is of
(4123)type as given by the minimal completion of the chord diagram. Fig. 20 illustrates the result of the numerical
simulation. The upper ﬁgures show the direct simulation, and the lower ﬁgures show the exact solution of (4123)type.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1211
Fig. 20. Numerical simulation of the case (f) in Fig. 14: The initial wave consists of [1, 4]soliton in y > 0 and [2, 3]soliton in y < 0. The upper
ﬁgures show the result of the direct simulation, and the lower ﬁgures show the corresponding exact solution of (4123)type obtained by the minimal
completion of the chord diagram. The computational domain is [−192, 192] ×[−32, 32]. Notice that the bending in the [1, 4]soliton appears near
the intersection point with the v irtual [3, 4]soliton, and the dispersive radiation observed in the simulation is a part of this [3, 4]soliton.
As in the previous case (e), this case shows that the [3, 4]soliton part in the exact solution does not appear in the solution
of the simulation, that is, the [3, 4]soliton is a virtual one. This can be also explained with the similar argument as
in the previous case. Notice again that the [1, 4]soliton in y > 0 has a virtual interaction with [3, 4]soliton to get
bending near (x = 18, y = 7) at t = 6.
5.2. Xshape initial waves
The initial waveform is illustrated in the left panel of Fig. 21. The right ﬁgure shows the chord diagrams corresponding
to the initial waves with the parameters (A
0
, tan Ψ
0
). We carried out four different cases marked with (g) through (j)
in Fig. 21. The main result in this section is to show that each simple sum of linesolitons gets asymptotically to the
exact solution corresponding to the chord diagram given by the initial wave. One should however note that those initial
waves are not the exact solutions, and the simulations provide stability analysis for those exact solutions. For example,
Fig. 21. Xshape initial waves. Each line of the Xshape is an inﬁnite linesoliton solution. We set those linesolitons to meet at the origin, and ﬁx
the amplitude of the soliton with positive angle Ψ
0
to be 2 (the corresponding chord is shown in the lighter color in each region). A
0
is the amplitude
of the soliton with negative angle −Ψ
0
. The right panel shows the corresponding chord diagrams with the choice of the values A
0
and Ψ
0
. There is
no exact solution for the degenerate cases marked by the dotted lines.
1212 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
if the exact solution has the phase shift, the asymptotic solution actually converges to this exact solution by generating
the correct phase shifts.
5.2.1. The case (g)
The initial wave consists of [1, 2] and [3, 4]soliton. To emphasize the phase shift appearing in the exact solution
of (2143)type, we consider the initial solitons to be close to the boundary at k
2
= k
3
(but we cannot take the values
so close, since the generation of a large phase shift needs a long time computation). For simplicity, we also take a
symmetric initial wave whose amplitudes and angles of the initial solitons are given by
_
A
[1,2]
= A
0
= 2, tan Ψ
[1,2]
= −2.02 = − tan Ψ
0
A
[3,4]
= 2 tan Ψ
[3,4]
= 2.02 = tan Ψ
0
The kparameters are then given by (k
1
, k
2
, k
3
, k
4
) = (−2.01, −0.01, 0.01, 2.01). The numerical simulation in Fig. 22
demonstrates the generation of the phase shifts after the interaction, that is, the linesolitons in the left side are accelerated
by their interaction. Note here that the interaction generates the dispersive wake behind the interaction point which
propagates in the negative xdirection.
Minimizing the error function E(t) at t = 12, we obtain the Amatrix,
A =
_
1 3.029 0 0
0 0 1 0.015
_
Fig. 22. Numerical simulation of the case (g) in Fig. 21: The initial wave is the sum of [1, 2] and [3, 4]solitons. The ﬁgures in the ﬁrst row show
the numerical simulation, and the ﬁgures in the second row show the corresponding exact solution of (2143)type, i.e. Otype. The domain of the
simulation is [−192, 192] ×[−32, 32]. The bottom graph shows E(t) of (5.3) which is minimized at t = 12, and the circle of the exact solution
shows D
t
r
with r = 12. Notice that a small dispersive radiation appears to generate the correct phase shifts.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1213
In the minimization by adjusting the locations of the solitons, we use x
+
[1,2]
= x
+
[3,4]
(due to the symmetry of the
solution), and then we obtain
x
+
[1,2]
= x
+
[3,4]
= −0.21.
The negative values are due to the generation of the positive phase shifts given by (3.9),
x
[1,2]
= x
[3,4]
= 1.963 > 0.
Namely the accelerations of the back parts of the linesolitons (i.e. the positive phase shifts) imply the deceleration of
the front parts of solitons. The ﬁgures in the second row in Fig. 22 shows the corresponding Otype soliton solution
with this Amatrix. The circle shows the domain D
t
r
with r = 12. Notice the dispersive wake behind the interaction
point in the simulation. The bottom graph in Fig. 22 shows the convergence of the solution to the Otype solution given
by those Amatrix and the same kparameters.
5.2.2. The case (h)
The initial wave is the sum of [1, 3] and [2, 4]soliton. The corresponding chord diagram is of (3412)type (i.e.
Ttype), and we expect the generation of a box at the intersection point of those solitons. For simplicity, we consider a
symmetric initial wave, and take the amplitudes and angles of those solitons to be
_
A
[1,3]
= A
0
= 2, tan Ψ
[1,3]
= −1 = − tan Ψ
0
A
[2,4]
= 2 tan Ψ
[2,4]
= 1 = tan Ψ
0
The kparameters are then given by (k
1
, k
2
, k
3
, k
4
) = (−3/2, −1/2, 1/2, 3/2). Although Ttype soliton appears for
smaller angle Ψ
0
, one should not take so small value. For an example of the symmetric case, if we take Ψ
0
= 0, we
obtain the KdV 2soliton solution with different amplitudes. So for the case with very small angle Ψ
0
, we expect to see
those KdV solitons near the intersection point. However, the solitons expected from the chord diagram have almost
the same amplitude as the incidence solitons for the case with a small angle. The detailed study also shows that near
the intersection point for Ttype solution at the time when all four solitons meet at this point (i.e Xshape), the solution
has a small amplitude due to the repulsive force similar to the KdV solitons (see also [4], where the initial Xshape
wave with small angle generates a large soliton at the intersection point). This then implies that our initial wave given
by the sum of two linesolitons creates a large dispersive perturbation at the intersection point (this can be also seen in
the next cases where we discuss the Ptype solutions).
The ﬁgures in the ﬁrst row in Fig. 23 illustrate the numerical simulation, which clearly shows an opening of a
resonant box as expected by the chord diagram of Ttype. The corresponding exact solution is illustrated in the ﬁgures
in the second row, where the Amatrix of the solution is obtained by minimizing the error function E(t) of (5.3) at
t = 6,
A =
_
1 0 −0.368 −0.330
0 1 1.198 0.123
_
In the minimization, we take x
±
[1,3]
= x
±
[2,4]
due to the symmetric proﬁle of the solution, and adjust the onset of the
box (recall (3.3), and note that the symmetry reduces the number of free parameters to three). We obtain
x
+
[1,3]
= x
+
[2,4]
= 0.025, r = 3.63, s = 0.350.
The positive shifts of those [1, 3] and [2, 4]solitons in the wavefront indicate also the positive shift of the newly
generated soliton of [1, 4]type at the front. This is due to the repulsive force exists in the KdV type interaction
explained above, that is, the interaction part in the initial wave has a larger amplitude than that of the exact solution,
so that this part of the solution moves faster than that in the exact solution. This difference may result as a shift of the
location of the [1, 4]soliton. The phase shifts of the initial waves are obtained from (3.4),
x
[1,3]
= x
[2,4]
= 0.549.
The relatively large value r > 1 indicates that the onset of the box is actually much earlier that t = 0, and s < 1 shows
the positive phase shifts as calculated above.
1214 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
Fig. 23. Numerical simulation of the case (h) in Fig. 21: The initial wave is the sumof [1, 3] and [2, 4]solitons. The ﬁgures in the ﬁrst rowshowthe
numerical simulation, and the ﬁgures in the second rowshowthe corresponding exact solution of (3412)type, i.e. Ttype. The computational domain
is [−192, 192] ×[−48, 48], and the circle in the exact solution shows the domain D
t
r
with r = 22. The bottom graph shows the error function E(t)
of (5.3) which is minimized at t = 6.
The bottom graph in Fig. 23 shows the evolution of the error function E(t) of (5.3) which is minimized at t = 6.
Note here that the circular domain D
t
r
with r = 22 covers well the main feature of the interaction patterns for all the
time computed for t ≤ 7.
5.2.3. The case (i)
The initial wave consists of [1, 4] and [2, 3]soliton. The corresponding diagram is of (4321)type (Ptype), and
the exact solution of this type has negative (repulsive) phase shift. Here we show how the phase shift is generated in
the simulation. We take the amplitudes and angles of those solitons to be
⎧
⎨
⎩
A
[2,3]
= A
0
=
8
25
, tan Ψ
[2,3]
= −0.4 = − tan Ψ
0
A
[1,4]
= 2 tan Ψ
[1,4]
= 0.4 = tan Ψ
0
The kparameters are then given by (k
1
, k
2
, k
3
, k
4
) = (−4/5, −3/5, 1/5, 6/5). In this case, we also note that taking a
very small angle Ψ
0
may give a large perturbation to the exact solution, and one should have a large computation time to
see the convergence. The top ﬁgures in Fig. 24 show the numerical simulation. The simulation indicates the generation
of the phase shift in the lower part of the solution. This may be explained as follows: Since the [1, 4]soliton propagates
faster than the [2, 3]soliton, the intersection point moves in the upper direction, i.e. the positive ydirection. Then the
effect of the nonlinear interaction may be observed in the lower half (i.e. behind the interactions) of the solution, and it
generates the phase shifts. One should also note that the large bending of the [1, 4]soliton in y < 0. This is due to the
interaction with the dispersive wake generated around the origin at t = 0, and notice that this bending part disperses
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1215
Fig. 24. Numerical simulation of the case (i) in Fig. 21: The initial wave is the sum of [1, 4] and [2, 3]solitons. The ﬁgures in the ﬁrst row show
the numerical simulation, and the ﬁgures in the second row show the corresponding exact solution of (4321)type, i.e. Ptype. The computational
domain is [−192, 192] ×[−48, 48], and the circle in the exact solutions shows D
t
r
with r = 10. The bottom graph shows the error function E(t)
of (5.3) which is minimized at t = 12. Note that the bending on the solitons disperses and the interaction pattern converges locally to that of the
corresponding exact solution.
away. The ﬁgures in the second row in Fig. 24 show the corresponding exact solution of Ptype whose Amatrix is
obtained by minimizing the error function E(t) of (5.3) at t = 12,
A =
_
1 0 0 −0.038
0 1 1.629 0
_
In the minimization, we adjust the location of the linesolitons at the wavefront, and using (3.6), we obtain
x
+
[2,3]
= −0.125, x
+
[1,4]
= −0.54.
Notice that the larger shifts in the [1, 4]soliton, which can be also observed in the simulation. This shift is due to the
repulsive force appearing at the intersection point (recall that the exact solution has a smaller amplitude at this point).
The phase shifts of the Ptype solution is given by (3.7),
x
[1,4]
= −1.10, x
[2,3]
= −2.75.
Those large phase shifts are also observed in the simulation, and the bending regions are getting separated from the
intersection point with the correct phase shifts. The bottom graph in Fig. 24 shows the decay of the error function E(t)
with D
t
r
for r = 10. We also note the saturation E(t) after t = 11, which may be due to the large deformation of the
solution, and we may need to perform a larger domain computation.
1216 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
5.2.4. The case (j)
This case is similar to the previous one with [1, 4] and [2, 3]soliton in the initial wave. However in this case, the
larger soliton of [1, 4]type has a negative angle (propagating in the negative ydirection), and we expect that the phase
shift now appear in the upper half of the solution. We take the amplitudes and angles of those solitons to be
⎧
⎪
⎪
⎨
⎪
⎪
⎩
A
[1,4]
= A
0
=
169
32
, tan Ψ
[1,4]
= −
1
4
= − tan Ψ
0
A
[2,3]
= 2 tan Ψ
[2,3]
=
1
4
= tan Ψ
0
The kparameters are then given by (k
1
, k
2
, k
3
, k
4
) = (−7/4, −7/8, 9/8, 3/2). The ﬁgures in the ﬁrst row in Fig. 25
show the numerical simulation. One can observe a dispersive wake is generated by a large perturbation in the initial
wave near the origin, and then the wake pushes the [1, 4]soliton forward. Also the large bending of the [2, 3]soliton
near the origin is due to the interaction with the wake. Since the [1, 4]soliton propagates much faster than the other one,
the intersection point moves in the negative ydirection. We can see a clear separation between the dispersive waves
causing the deformation of solitons and the intersection points of Ptype. The ﬁgures in the second row in Fig. 25
illustrate the corresponding Ptype solution whose Amatrix is obtained by minimizing the error function E(t) at t = 8.
Fig. 25. Numerical simulation of the case (j) in Fig. 21: The initial wave is the sum of [1, 4] and [2, 3]solitons. The ﬁgures in the ﬁrst row
show the numerical simulation, and the ﬁgures in the second row show the corresponding exact solution of (4321)type, i.e. Ptype. The domain
of the simulation is [−192, 192] ×[−64, 64], and the circle in the middle ﬁgures shows D
t
r
with r = 10. The bottom graph illustrates the error
function E(t) of (5.3) which is minimized at t = 8. The graph shows a further reduction of the error, that is, the convergence of the solution to the
corresponding exact solution even with a drastic change of the error at the initial stage.
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1217
We take the radius r = 10 for a good separation of the dispersive perturbation from the steady Ptype soliton part. With
this minimization, we get
A =
_
1 0 0 −0.299
0 1 8.549 0
_
In the minimization, we adjust the locations of the linesolitons at the wavefront, and using (3.6), we obtain
x
+
[1,4]
= 0.627, x
+
[2,3]
= 0.923.
The positive shifts are due to the repulsive interaction at the intersection point as observed in the simulation. The phase
shifts of those incidence solitons are given by (3.7),
x
[1,4]
= −0.934, x
[2,3]
= −1.518.
Notice that the [1, 4]soliton behind the [2, 3]soliton shifts x
−
[1,4]
= x
+
[1,4]
+ x
[1,4]
= −0.307, and it gets a larger shift
in the upper part. The large bending in the simulation is generated by a strong perturbation generated at the intersection
point at t = 0. However, the bending part disperses and the intersection point gets away from this part.
The bottom graph in Fig. 25 shows the decay of the error. We emphasize that the error function gets a rapid change
in the early times. This implies that a drastic change in the proﬁle of the solution due to a large perturbation to the
exact solution. However after t = 3 the solution has a monotone convergence to the exact solution, implying that the
dispersive waves are getting away from the domain D
t
r
as we expected.
6. Conclusion
In this paper, we studied numerically the interaction feature of the KP equation for the solutions consisting of small
number of linesolitons. We started to provide a brief review of the classiﬁcation theorem for soliton solutions of the
KP equation. In particular, we remarked that each soliton solution can be expressed by a unique chord diagram which
contains the information of the asymptotic linesolitons of the solution in y ± ∞. We then performed the numerical
study of the initial value problems using a pseudospectral method with window technique. We considered a special
class of the initial waves with V and Xshape which consist of two distinct linesolitons. This class of initial data was
considered to study some physical problem related to the generation of large amplitude waves, e.g. the Mach reﬂection
phenomena in shallow water. We observed the separation of the exact soliton solution from dispersive radiations,
similar to the solitons in the KdV equation where the initial wave evolves into a sum of solitons and dispersive wave
trains propagating opposite direction to the solitons. This leads to a convergence of the initial wave to certain exact
solution. Here we proposed a notion of the local convergence in the sense that the L
2
error is measured in a compact
domain which covers the main part of the interaction of two soliton solutions forming V or Xshape in the solution.
Then the separation implies that dispersive radiations eventually escape from the compact domain which is moving
along the exact soliton solution. We also conﬁrmed that the exact solution in the convergence is predicted by the
minimal completion of the (partial) chord diagram describing the initial Vshape wave. Here the minimal completion
means that the resulting chord diagram should give the unique derangement (permutation without ﬁxed points) and
the corresponding solution has the smallest total length of the chords.
Our present study indicates the stability of some of the exact solutions, and we expect that our study can provide
useful information even in the case of the solutions including a large number of linesolitons.
Acknowledgements
We would like to thank Professor Thiab Taha for giving us the opportunity to present our result at the Sixth
IMACS International Conference on March 23–26, 2009. We would also like to thank Professors Masayuki Oikawa,
Hidekazu Tsuji and Sarvarish Chakravarty for various useful discussions related to the present work. C.Y.K. is partially
supported by NSF grant DMS0811003 and the Alfred P. Sloan fellowship. Y.K. is partially supported by NSF grant
DMS0806219.
1218 C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
References
[1] M.J. Ablowitz, H. Segur, Solitons and the inverse scattering transform, in: SIAMStudies in Applied Mathematics, SIAM, Philadelphia, 1981.
[2] S. Chakravarty, Y. Kodama, Classiﬁcation of the linesolitons of KPII, J. Phys. A: Math. Theor. 41 (2008) 275209 (33pp).
[3] S. Chakravarty, Y. Kodama, Soliton solutions of the KP equation and application to shallow water waves, Stud. Appl. Math. (2009) 83–151.
[4] M. Chen, Oblique interaction of solitary waves, preprint, 2009.
[5] W. Craig, P. Guyenne, J. Hammack, D. Henderson, C. Sulem, Solitary water wave interactions, Phys. Fluids 18 (2006) 057106.
[7] M. Funakoshi, Reﬂection of obliquely incident solitary waves, J. Phys. Soc. Jpn. 49 (1980) 2371–2379.
[8] R. Hirota, The Direct Method in Soliton Theory, Cambridge University Press, Cambridge, 2004.
[9] B.B. Kadomtsev, V.I. Petviashvili, On the stability of solitary waves in weakly dispersive media, Sov. Phys.Dokl. 15 (1970) 539–541.
[10] F. Kako, N. Yajima, Interaction of ionacoustic solitons in twodimensional space, J. Phys. Soc. Jpn. 49 (1980) 2063–2071.
[11] C. Klein, C. Sparber, P. Markowich, Numerical study of oscillatory regimes in the kadomtsev–petviashvili equation, J. Nonlin. Sci. 17 (2007)
429–470.
[12] Y. Kodama, Young diagrams and Nsoliton solutions of the KP equation, J. Phys. A: Math. Gen. 37 (2004) 11169–11190.
[13] Y. Kodama, M. Oikawa, H. Tsuji, Soliton solutions of the KP equation with Vshape initial waves, J. Phys. A: Math. Theor. 42 (2009) 312001
(9pp).
[14] J.W. Miles, Resonantly interacting solitary waves, J. Fluid Mech. 79 (1977) 171–179.
[17] A.C. Newell, Solitons in Mathematics and Physics, CBMSNSF, 48, SIAM, Philadelphia, 1985.
[18] A.C. Newell, L.G. Redekopp, Breakdown of Zakharov–Shabat theory and soliton creation, Phys. Rev. Lett. 38 (1977) 377–380.
[19] S. Novikov, S.V. Manakov, L.P. Pitaevskii, V.E. Zakharov, Theory of Solitons: The Inverse Scattering Method, Contemporary Soviet Mathe
matics, Consultants Bureau, New York and London, 1984.
[20] A.V. Porubov, H. Tsuji, I.L. Lavrenov, M. Oikawa, Formation of the rogue wave due to nonlinear twodimensional waves interaction, Wave
Motion 42 (2005) 202–210.
[21] P. Schlatter, N.A. Adams, L. Kleiser, A windowing method for periodic inﬂow/outﬂow boundary treatment of nonperiodic ﬂows, J. Comput.
Phys. 206 (2005) 505–535.
[22] H. Tsuji, M. Oikawa, Oblique interaction of solitons in an extended Kadomtsev–Petviashvili equation, J. Phys. Soc. Jpn. 76 (2007) 84401–84408.
[23] H. Tsuji, M. Oikawa, K. Maruno, Twodimensional interaction of long solitary waves and web solution, in: Presented at the Annual Meeting
of The Japan Society of Fluid Mechanics, Kobe, Japan, September 4–7, 2008.
[24] G.B. Whitham, Linear and Nonlinear Waves, A WileyInterscience Publication, John Wiley & Sons, New York, 1974.
[25] H. Yeh, W. Li, Y. Kodama, Mach reﬂection and KP solitons in shallow water (arXiev:1004.0370), EPJ ST, (2010) in press.
1186
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
In this paper, we study this type of the initial value problem of the KP equation by means of the direct numerical simulation. In particular, we consider the following two cases of the initial waves: In the ﬁrst case, the initial wave consists of two semiinﬁnite linesolitons forming a Vshape pattern, and in the second case, the initial wave is given by a linear combination of two inﬁnite linesolitons forming Xshape. Those initial waves have been considered in the study of the generation of large amplitude waves in shallow water [20,22,13,25]. The main result of this paper is to show that the solutions of the initial value problem with those initial waves asymptotically converge to some of the exact soliton solutions found in [2,3]. This implies a separation of the (exact) soliton solution from the dispersive radiations in the similar manner as in the KdV case. The paper is organized as follows: in Section 2, we provide a brief summary of the soliton solutions of the KP equation and the classiﬁcation theorem obtained in [2,3] for those soliton solutions as a background necessary for the present study. In particular, we introduce the parametrization of each soliton solution with a chord diagram which represents a derangement of the permutation group, i.e. permutation without ﬁxed point. In Section 3, we present several exact soliton solutions, and describe some properties of those solutions. Each of those soliton solutions has N+ numbers of linesolitons in a far ﬁeld on the twodimensional plane, say in y 0, and N− numbers of linesolitons in the far ﬁeld of the opposite side, i.e. y 0. This type of soliton solution is referred to as a (N− , N+ )soliton solution. Here we consider those solitons with N− + N+ ≤ 4 and N− ≤ 3. In Section 4, we describe the numerical scheme used in this paper, which is based on the pseudospectral method combined with the window technique [21,23]. The window technique is especially used to compute our nonperiodic problem which is essentially an inﬁnite domain problem. In Section 5, we present the numerical results of the initial value problems with V and Xshape initial waves, and show that the solutions asymptotically converge to some of those exact solutions discussed in Section 3. The 1/2 2 , where convergence is in the sense of locally deﬁned L2 norm with the form, f L2 (D) := D f (x, y) dx dy D ⊂ R2 is a compact set which covers the main structure describing the (resonant) interactions in the solution. We also propose a method to identify an exact solution for a given initial wave with V or Xshape pattern based on the chord diagrams introduced in the classiﬁcation theory. Then, in Section 6, we conclude the paper with a brief summary of our results. 2. Background Here we give a brief summary of the recent result of the classiﬁcation theorem for soliton solutions of the KP equation (see [12,2,3] for the details). In particular, each soliton solution is then parametrized by a chord diagram which represents a unique element of the permutation group. We use this parametrization throughout the paper. 2.1. The KP equation The KP equation is a twodimensional nonlinear dispersive wave equation given by
3 2 ∂x 4∂t u + 6u∂x u + ∂x u + 3∂y u = 0, n where ∂x u:=∂n u/∂xn , etc. Let us express the solution in the form, 2 u(x, y, t) = 2 ∂x ln τ(x, y, t).
(2.1)
(2.2)
where the function τ is called the tau function, which plays a central role in the KP theory. In this paper, we consider a class of the solutions, where each solution can be expressed by τ(x, y, t) in the Wronskian determinant form τ = Wr(f1 , . . . , fN ), i.e. f1 f2 τ(x, y, t) = . . . fN f1 f2 . . . fN · · · f1
(N−1)
· · · f2 . .. . . . · · · fN
(N−1)
,
(2.3)
(N−1)
C.Y. Kao, Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218
(j) j
1187
with fn :=∂x fn for n = 1, . . . , N. Here the functions {f1 , . . . , fN } form a set of linearly independent solutions of the linear equations,
2 ∂y fn = ∂x fn , 3 ∂t fn = −∂x fn .
(The fact that (2.2) with (2.3) gives a solution of the KP equation is wellknown and the proof can be found in several places, e.g. see [8,3].) The solution of those equations can be expressed in the Fourier transform, fn (x, y, t) =
C
ekx+k
2 y−k 3 t
ρn (k) dk,
n = 1, 2, . . . , N,
(2.4)
with an appropriate contour C in C and the measure ρn (k) dk. In particular, we consider a ﬁnite dimensional solution with ρn (k) dk = M an,m δ(k − km ) dk with an,m ∈ R, i.e. m=1
M
fn (x, y, t) =
m=1
an,m Em (x, y, t)
with
2 3 Em = exp(km x + km y − km t).
Thus this type of solution is characterized by the parameters {k1 , k2 , . . . , kM } and the N × M matrix A:=(an,m ) of rank(A) = N, that is, we have (f1 , f2 , . . . , fN ) = (E1 , E2 , . . . , EM )AT . RM RM . (2.5) and {f1 , f2 , . . . , fN } spans an Ndimensional subspace of This Note that {E1 , E2 , . . . , EM } gives a basis of means that the Amatrix can be identiﬁed as a point on the real Grassmann manifold Gr(N, M) (see [12,3]). More precisely, let MN×M (R) be the set of all N × M matrices of rank N. Then Gr(N, M) can be expressed as Gr(N, M) = GLN (R) \ MN×M (R). where GLN (R) is the general linear group of rank N. This expression means that other basis (g1 , . . . , gN ) = (f1 , . . . , fN )H for any H ∈ GLN (R) spans the same subspace, that is, A → H T A GLN (R) acting from the left). Notice here that the freedom in the Amatrix with GLN (R) can be ﬁxed by expressing A in the reduced row echelon form (RREF). We then assume throughout this paper that the Amatrix is in the RREF, and show that the Amatrix plays a crucial role in our discussion on the asymptotic behavior of the initial value problem. Now using the Binet–Cauchy Lemma for the determinant, the τfunction of (2.2) can be expressed in the form, ⎛ ⎞ ⎛ ⎞ a11 a21 · · · aN1 ⎜a ⎟ E1 E2 ··· ··· EM ⎜ 12 a22 · · · aN2 ⎟ ⎜ k1 E 1 ⎟ ⎟⎜ k2 E2 · · · · · · kM E M ⎟ ⎜ . ⎜ ⎟ . . .. . . ⎜ ⎟ ⎟⎜. . . . τ= ⎜ . . . .. .. ⎟ ⎟⎜. . . . . . ⎟ ⎝ ⎠⎜. . . . . . .. ⎜. ⎟ . . . . . ⎝. ⎠ N−1 N−1 N−1 k1 E1 k 2 E2 · · · · · · k M EM a1M a2M · · · aNM =
1≤j1 <j2 <···<jN ≤M
ξ(j1 , j2 , . . . , jN )E(j1 , j2 , . . . , EN ),
(2.6)
where ξ(j1 , . . . , jN ) is the N × N minor of the Amatrix with N columns marked by (j1 , . . . , jN ), and E(j1 , . . . , jN ) is given by E(j1 , . . . , jN ) = Wr(Ej1 , . . . , EjN ) =
l<m
(kjm − kjl )Ej1 · · ·EjN .
From formula (2.6), one can see that for a given Amatrix, • if a column of A has only zero elements, then the exponential term Em with m being the column index never appear in the τfunction, and • if a row of A has only the pivot as nonzero element, then En with n being the row index can be factored out from the τfunction, i.e. En has no contribution to the solution.
j]type with i < j has the following structure (sometimes we consider only locally).j] giving the location of the linesoliton. 3 3 2 2 Ω[i. We here emphasize that the linesoliton appears at the boundary of two regions.j] ). and it determines the location of the soliton solution.j] ) 2 with some constant Θ[i.j] = ki + ki kj + kj = (1/2)A[i. 2 We are also interested in nonsingular solutions. .j] . We call a matrix having the condition (2.j] y = tan Ψ[i. ki + kj .j] .2.5): With the Amatrix of the form A = (1 a). the linesoliton solution of [i.j] sech2 (K[i. Y. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 For Amatrix with no such cases.j] and x[i.j] sech2 A[i. .1188 C. in each of which either E1 or E2 becomes the dominant exponential term. the wave vector K[i. and because of this we also call this soliton [1.j] ) is measured in the counterclockwise from the xaxis.Y. called one linesoliton solution. For the parameter x[i.j] .e. .j] = 1 (kj − ki )2 2 (2.j] = ki + kj . Linesoliton solution and the notations Let us here present the simplest solution.j] t + Θ[i. for all 1 ≤ j1 < j2 < · · · < jN ≤ M. j2 . 2. Then a linesoliton 0 (2. kj − ki (2. kj − ki = (kj − ki ) 1. u= 1 u = A[i. Notice that Ψ[i.10) .j] = kj − ki . we also use the notation. 2 2 Thus the peak of the solution is located along the line θ1 − θ2 = ln a. 2 The parameter a in the Amatrix must be a ≥ 0 for a nonsingular solution (i. 2]soliton solution (or soliton of [1. Then the solution u = 2∂x (ln τ) gives 1 1 (k1 − k2 )2 sech2 (θ1 − θ2 − ln a).j] 0 (x + y tan Ψ[i. .j] .7) This condition is not only sufﬁcient but also necessary for the nonsingularity of the solution for any initial data.8) can be written in the form with three parameters A[i. 1).j] .j] t − x[i. The amplitude A[i.j] gives the angle between the line K[i.j] . Since the solution is given by u = 2∂x (ln τ). because the τfunction with reducible matrix can be obtained by a matrix with smaller size.j] . totally nonnegative Amatrix). 2 (2.j] = − 1 Θ[i. x The direction of the wavevector K[i.j] + (3/4) tan2 Ψ[i. In general.j] − C[i.e.j] are deﬁned by A[i. A linesoliton solution is obtained by a τfunction with two exponential terms. the case N = 1 and M = 2 in (2.7) totally nonnegative matrix. Ψ[i. Since a = 0 leads to a trivial solution. the nonsingular solutions are obtained by imposing the nonnegativity condition on the minors. u = A[i. we have √ 1 τ = E1 + aE2 = 2 ae(1/2)(θ1 +θ2 )/2 cosh (θ1 − θ2 − ln a). ξ(j1 .e.j] = kj − ki = (kj − ki )(ki + ki kj + kj ).j] = (K[i. and introduce several notations to describe the solution. jN ) ≥ 0.j] · x − Ω[i. Kao. 2 irreducible Amatrix). we call it irreducible.j] and the frequency Ω[i. i. we consider only a > 0 (i. 0 x[i. 2]type).j] x K[i. K[i.9) 0 2 2 withC[i.8) 2 2 K[i.j] · x = const and the yaxis (See Fig. (2. and it is given by y K[i.
y.5 and A[i. ki = −0.j] )4 + 3(K[i. K[i.j] K[i.9). given by A[i. and the KP equation in the form (2. In Fig. Here the chord diagram indicates the permutation of the dominant exponential terms Ei and Ej in the τfunction.j] ) . Physically this implies that soliton is a supersonic wave due to its nonlinearity (recall that the velocity of shallow water wave is proportional to the square root of the water depth. the xcomponent of the group velocity is always negative. This is similar to the case of the KdV equation. On the other hand. and we expect that asymptotically soliton separates from dispersive radiations. .j] in (2.j] = 9/8.j] = 2 (ki − kj )2 and tan Ψ[i. and we use the notation x[i. u(x. Throughout this paper.. any soliton propagates in the positive xdirection.j] = (K[i.1) describes the waves in the moving frame with this velocity in the xdirection). one should note that any small perturbation propagates in the negative xdirection. the wave vector K[i. that is. 2 ky 3 kx .5 and kj = 1.j] which are.3 below for the precise deﬁnition of the chord diagrams. ki + kj ).j] and the angle Ψ[i. we illustrate an example of one linesoliton solution.) For each soliton solution of (2.j] = Ω[i. ∂kx ∂ky = − 3 4 2 kx + 2 ky 2 kx . Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1189 Fig. j]soliton 0 and the lower one for y 0. that is. In the right panel of this ﬁgure.C. that is. with Θ[i. j]soliton in the region x 0) at t = 0. This representation of the linesolitons in terms of the chord diagrams is the key concept while Ej dominates in x throughout the present paper. we x show a chord diagram which represents this soliton solution.j] 2 1 + (ki + kj )2 2 2 Note in particular that C[i.8) which is determined by the Amatrix and the kparameters.j] = 2 2 k i + ki kj + k j Ω[i. for y which give tan Ψ[i.j] is given by V[i. 1.e.j] = ki + ki kj + kj > 0.j] = ki + kj . t) with 1 the amplitude A[i. One linesoliton solution of [i. and this implies that the xcomponent of the velocity is always positive.j] and the frequency Ω[i. j]type and the corresponding chord diagram. The upper oriented chord represents the part of [i.j] K[i. The right panel shows the corresponding chord diagram. and the solutions are shown in the contour lines with colors indicating the magnitudes. the horizontal and vertical axes in the ﬁgures of soliton solutions are x and y. i. one can only deﬁne the location of each [i.11) The soliton velocity V[i. Y. The left ﬁgure shows the soliton solution. For multisoliton solutions. This can be seen from the dispersion relation of the KP equation for a linear wave φ = exp(ik · x − iωt) with the wavevector k = (kx . Here the horizontal line is the coordinate for the kparameters. 2 1 3 3 ky .j] deﬁned by K[i. and in this example.Y. ω = − kx + 4 4 kx from which the group velocity of the wave is given by v = ∇ω = ∂ω ∂ω . 1. 0. respectively. (See section 2.j] · V[i. Ei dominates in x 0. y 2 (2.j] ) which describes the xintercept of the line determined by the wave crest of [i. ky ) and the frequency ω. respectively.j] = 0. Kao.j] (or − 0 (or x[i. j]soliton using the asymptotic position in the xyplane either x 0 or x 0 + (we mainly consider the cases where the solitons are not parallel to the yaxis).j] = (1. with the ordering ki < kj .j] satisfy the solitondispersion relation. x x 4Ω[i.
.Y. Let{e1 .1190 C. 1/2 L2 (D) is the usual L2 norm of f L2 (D) := D f (x. . Classiﬁcation theorems Now we present the main theorems obtained in [2. M} → {1. 2. Finally we remark that a linesoliton having the angle Ψ0 ≈ π/2 has an inﬁnite speed. jn ]type forn = 1. . We also emphasize that the structure of the solution for y 0 can be 0. y. for t → ∞. However there are several difﬁculties for twodimensional stability problem in general. that is. .e. i. gm ] in Theorem 2. . y) satisﬁes √ for some exact soliton solution. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 Stability of onesoliton solution was shown in the original paper by Kadomtsev and Petviashvii [9]. . . The pairing map π is a bijection. In this paper. y)2 dx dy . and this may be stated as follows: For any > 0 and any r > 0. y. N. and let{g1 . . . M} such that π(en ) = jn . M}. . jn ] and [im . . whereSM is the group of permutation for the index set{1. < . 2.1 have a combinatorial interpretation. . M − N. gM−N } be nonpivot indices for an N × M irreducible and totally nonnegative Amatrix. Then the soliton solution generated by the τfunction with the Amatrix has the following asymptotic structure: (a) Fory (b) Fory 0. π(gm ) = im . . t) satisﬁes ut − ut exact where t Dr t L2 (Dr ) u0 − u0 exact 0 L2 (Dr ) < δ. . . Herejn (> en ) and im (< gm ) are determined uniquely from the Amatrix. y):=u(x. t Dr = {(x. . That is. The unique index pairings [en .3] for the classiﬁcation of soliton solutions generated by the τfunctions with irreducible and totally nonnegative Amatrices: Theorem 2. 0. n = 1. we will give a numerical study for some special cases where the initial waves consist of two semiinﬁnite linesolitons with V. there are N linesolitons of [en . This stability implies a separation of the soliton from the dispersive radiations (nonsoliton parts) as in the case of the KdV soliton. 2.or Xshape. x0 (t) + y0 (t) tan Ψ0 = C0 t + x0 . is a circular disc with radius r moving with the soliton.e. We would like to prove the similar statement for more general initial waves. M − N. N.1. y) ∈ R2 : (x − x0 (t))2 + (y − y0 (t))2 ≤ r 2 }. ut :=A0 sech2 A0 /2(x + y tan Ψ0 − C0 t − x0 ) with appropriate constants A0 . . i. . y) over a compact domain D ⊂ R2 . there exists δ > 0 so that if the initial wave u(x. 0):=u0 (x. where en and gm are respectively the pivot and nonpivot indices of the Amatrix. . . π ∈ SM . with (x0 (t). 2. . . Kao. Then we have: Theorem 2. Ψ0 exact and x0 (recall C0 = (1/2)A0 + (3/4) tan2 Ψ0 ). there areM − N linesolitons of [im . gm ]type form = 1.3. Let us deﬁne the pairing map π : {1. m = 1. . . y0 (t)) at any point on the soliton. the set of asymptotic solitons in y 0 can be different from that in y 0. . . . This different from that in y difference is a consequence of the resonant interactions among solitons as we can see throughout the paper. and of course it is beyond the assumption of the quasitwodimensionality.e. . . eN } be the pivot indices.12) . . Y. π= e1 · · ·eN j1 · · ·jN g1 · · ·gM−N i1 · · ·iM−N . . . Here f f (x.2. (2. then the stability implies that the solution ut (x. . i. . . .
2. and the corresponding soliton solution consists of the asymptotic linesolutions of [1. .C. . . i. . E3 }. [14. then the τfunction consists only two exponential terms and it gives one 2 linesoliton solution.Y. the excedance jn > en . .e. Let us ﬁrst study the case with N = 1 and M = 3. (iii) On the lower side of the line. The diagram then shows that the set of excedances is {1. 3] . each of the point corresponds to the kparameter.g. [2. 4]. [3.e. there are N chords (pairings). We represent each soliton solution with the chord diagram deﬁned as follows: (i) There are M marked points on a line. Each chord joining ith and jth marked points represents the linesoliton of [i. 8]types in y 3. The parameters a. i. π= 1 4 2 6 3 5 4 2 5 3 6 8 7 8 1 7 or simply π = (46523817). . Those solutions give a fundamental structure of general solutions. b in the matrix are positive constants.e.10]). 2. j]type.22. and we will show that some of those solutions appear naturally as asymptotic solutions of the KP equation for certain classes of initial waves related to the rogue wave generation [20. N. Y. 5]. π has no ﬁxed point. 3. M = 3) and (N = 2.and 0. the solution u = 2∂x (ln τ) consists of [1. j means kj .7]. and the excedance set is the set of pivot indices of the Amatrix. M − N. 2. i. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1191 Fig. . Note in particular that the corresponding π is the derangement. π(i) > i for i = 1. where the Amatrix is given by A = (1 a b ). Exact solutions Here we present several exact solutions generated by smaller size N × M matrices with N ≤ 3 and M ≤ 4. each of them connects two points on the line representing π(en ) = jn for n = 1. [7. . 5]. the τfunction has three exponential terms {E1 .and [6. [3.3). reducible case. Yshape solitons: resonant solutions We ﬁrst discuss the resonant interaction among linesolitons. N)soliton solution has N excedances. Example of the chord diagram. To describe resonant solutions. 4].18. Theorem 2. i. M = 3). This shows π = (46523817).) With the ordering k1 < k2 < k3 . the deﬁciency im < gm . Kao. . there are M − N chords representing π(gm ) = im for m = 1.2 shows that the pairing map π for an (M − N. [2. 3. (ii) On the upper side of the line. 6}. . The τfunction is simply given by τ = E1 + aE2 + bE3 .23. which is the most important feature of the KP equation (see e.e. 6]. . Here the numbers in the ﬁgure indicate the kparameters. i. and the positivity implies the irreducibility and the regularity of the solution. those are given by the cases (N = 1.1. E2 . we illustrate an example of the chord diagram which represents the derangement. let us consider the τfunction with M = 3. 8]types in y 0 and of [1.e. . N. 7]. that is. The detailed discussions and the formulae given in this section can be found in [3]. i. In Fig.e. (Note that if one of the parameters is zero. In terms of the τfunction in the form (2.
Kao. 1 0 x[2. The shape of solution generated by τ = E1 + aE2 + bE3 with a = b = 1 (i. so that x[1. one can see that those linesolitons of [1.10)). Ω[1.3] . Then the τfunction is given by τ = E(1. In this case we have [1. 3) + bE(2. 3]types are localized along the lines given in (2. kj − ki ) and Ω[i. i. Examples of (2.3] = K[1. 2) + aE(1.2] = − 1 ln 2.3] . 2 a (3. A= 1 0 0 1 −b a . This solution represents a resonant solution of three linesolitons.j] t = x[i. and it represents the permutation π = (312).and [2. that is. 3. where a and b are positive constants. Here the parameters a and b in the 0 0 Amatrix are chosen. 0 x[1. and the resonant condition is given by K[1.j] with 0 x[1. with E(i. . For both cases. 3]solitons for y 0.Y.j] for [i. y) → (−x. −1/4.3] = − 1 ln 3.2] = x[1.3] = − 1 b ln . j:=ki − kj . 2 b 3.and (1. j] = [1. all of those soliton solutions meet at the origin at t = 0. j) = (kj − ki )Ei Ej for i < j. 3. Y.2) where i. 2]. 3.e. In the lower ﬁgures of Fig.j] t = x[i.j] − C[i.j] − C[i. 1)soliton in the upper ﬁgure by changing (x. The chord diagram corresponding to this soliton is shown in the right panel of the ﬁrst raw in Fig. 2 b . 2. 1 3. Let us now consider the case with N = 2 and M = 3: We take the Amatrix in the form.3] = Ω[1. 0 x[1.3] = − 1 ln b.and [2. k3 ) = (−5/4. we illustrate the solution in this case. 2]. 3. 1 3.1192 C.2] + Ω[2. Taking the balance between two exponential terms in the τfunction.1) where i. 3]. Fig.3] = 0. 3 3 2 2 which are trivially satisﬁed with K[i.and [2.e. 3] and [2.2] + K[2. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 soliton for y 0 and [1. 3.j] = (kj − ki . 3/4).j] are determined by the Amatrix (see (2. 3. 3]types are localized along y 0 the lines. 3]solitons for y 0 and [1. x + y tan Ψ[i. 1 (3.and [1. 0 0 x + y tan Ψ[i. at t = 0 three linesolitons meet at the origin) is illustrated via the contour plot in the ﬁrst row of Fig. 1 a . The kparameters are given by (k1 . and this solution can be labeled by π = (231). −y). 3] where the locations x[i. 2]. 3]soliton for 0. j:=ki − kj . 1). Notice that this ﬁgure can be obtained from (2.j] = kj − ki . 2)soliton solutions and the chord diagrams.9). The right panels are the corresponding chord diagrams. k2 . Those linesolitons of [1. the parameters in the Amatrices are chosen so that at t = 0 three linesolitons meet at the origin as shown in the left panels. 3).
C. 2. −1. 3)soliton solutions.2. 1). 4]. . The corresponding label for this solution is given by π = (2 3 4 1). This solution is labeled by π = (4 1 2 3). 3)soliton solution which is dual to the case of N = 1.Y. The lower ﬁgures in Fig. m. 4 c= 1. 3). so that all four solitons meet at the origin at t = 0. 2. that is. We set a = b = c = 1 of the Amatrix. 2. k2 .and [2. b and c are positive constants. For N = 3. 4]soliton for y 0 and [1. the Amatrix is given by A = (1 a b c).and [3. k3 . c of the Amatrix as a= 1. 3. [1. −1/(2 2). 3. 3 . 4 shows the time evolution of the solution of this type. where a. 3 and M = 4 cases Let us ﬁrst discuss the case with N = 1 and M = 4. [2. 1/(2 2). Y. k2 . 3].and (1. [2. The τfunction is then given by τ = E(1. 2. that is. 3]. any irreducible and totally nonnegative Amatrix has the form. This gives (1. 1. Here we set a. m. In this case. 1)soliton solution consisting of one linesoliton of [1. Kao. The upper ﬁgures in Fig. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1193 3. 1. 0 0 1 a where a. b. 4]type for y 0 and three linesolitons of [1. 1). 4) + bE(1. 3 . The lower ﬁgures are generated √ √ by the 3 × 4 Amatrix with (k1 . 3. k4 ) = (−1. 2. 4. k3 . 4 where l. n:=(kn − km )(kn − kl )(km − kl ). ⎛ ⎞ 1 0 0 c ⎜ ⎟ A = ⎝ 0 1 0 −b ⎠ . N = 1. Fig. 4 b= 1. n) = (kn − km )(kn − kl )(km − kl )El Em En .and [3. 2]. we have (3. 4). 1. 4]solitons for y 0. 2]. k4 ) = √ √ (− 3. 4 shows the time evolution of this type. 2. b and c are positive constants. with E(l. 4) + cE(2. 3]solitons are parallel to the yaxis. 2. 3. Examples of (3. The solution in the upper ﬁgures are generated by the 1 × 4 Amatrix with (k1 . so that all four linesolitons meet at the origin at t = 0. 4]types for y 0. The τfunction is simply written in the form τ = E1 + aE2 + bE3 + cE4 . 3 . The symmetry in the kparameters implies that [1. 3) + aE(1.
−1/4. 4]. Kao.Y. 4]solitons. 3/4. b. 1 Θ[1.3. 3]}.and [3. 2)soliton solutions. [2. their phase shifts and the onset of the boxshaped interaction pattern: Those are given by b= + 2. The ﬁrst pair appears in the lower and upper edges of the box. Then the classiﬁcation theorems imply that we have a (2.4] se . 5. 1 Θ[2.3] se . 3] and [2. 2 4. 3]. c. The parameters a. 3. 4). and the second pair in the right and left edges. The most important feature of this solution is the generation of four intermediate solitons forming a box at the intersection point. This is the generic solution on the maximal dimensional cell. we obtain either pair of {[1. 2]. 2)soliton solution associated to each of those Amatrices. [2.1194 C. if we cut the chords of [1. b. N = 2).and [2.1. . Example of (3412)type soliton solution (Ttype). The chord diagrams for seven different types of (2. 1 D= 2. 3]. 2]. For example. 3 (3. c and d in the Amatrix determine the locations of solitons. 4]types for both y → ±∞. Y. The intermediate solitons forming a boxshape pattern are given by [1. 4]solitons. 5. and it is called Ttype (after [12]). [1. Fig. 2 a = r. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 Fig. k4 ) = (−7/4. 4]and [3.and [2. and the Amatrix is given by A= 1 0 0 1 −c a −d b . We here discuss all of those (2. from which one can ﬁnd the asymptotic linesolitons in each case. 4. d 3. 6 shows the time evolution of the solution of this type. and they may be obtained by cutting the chord diagram of (3412)type at the crossing points. we illustrate the corresponding chord diagrams for all those seven cases. 4]} or {[1. 3/2). 1 c= + 2. 2]. The solution in this case corresponds to the top cell of Gr(2. one can easily see that there are seven cases with 2 × 4 irreducible and totally nonnegative Amatrices. 1 s. 2)soliton solutions with the same kparameters given by (k1 . 6.3) Fig. one can see that the asymptotic linesolitons are given by [1.e.[1. Each diagram parametrizes a unique cell in the totally nonnegative Grassmannian Gr(2. and show how each Amatrix chooses a particular set of linesolitons. [2. 4]types for both y → ±∞. Those intermediate solitons are identiﬁed as [1.3. The asymptotic solitons are [1. The case π = (3412) From the chord diagram in Fig. 5. 4] at the crossing point. 3]. 3. 3. 4]. where a. 3]. 4. 4). In Fig. [3. d > 0 are free parameters with D:=ad − bc > 0. k3 . N = 2 and M = 4 cases By a direct construction of the derangements of S4 with two exedances (i. k2 .
The phase shifts for [1.j] /j. The Amatrix has the form.4] = 0 and r = 1. 6. one can see that the asymptotic linesolitons are given by [1.e.and [2.and [2.3] = Θ[2.and [2. 1 Θ[2. + + 0 at t = 0). 0 and [1. 3. 3 bc 1. Kao. 1 Θ[1. Fig. j]soliton for x 0 and x 0. e 4.3. j]soliton for x 0 (recall the deﬁnition below (2. As you can see below.2. A= 1 0 0 1 0 a −c b . 5. 7 illustrates an example of this solution. (3. Y. this solution is the most complicated and interesting one. so that at t = 0 the solution forms an Xshape without phase shifts and opening of a box at the origin. i)−Θ[i. 4]. i. 4 D . i)−ΘT /j.j] for the location of [i. 4 as explained above. Notice here that the pattern in the lower part (y < 0) of the solution is the same as that of the previous one. 3]. 3]types for y The intermediate solitons are [3. 3.Y. 7. Example of (4312)type soliton solution. 1 Θ[1. 3]solitons 0 and [1. The parameters in the Amatrix determine the locations of the linesolitons in the following form. C2 = 6) appear in this solution. 3 ± The Θ[ij] represent the locations of the [i. c > 0 are free parameters. Θ[1.3.10).3. . 0 and [1. Among the soliton solutions for N = 2 and M = 4. The asymptotic solitons are [1. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1195 + where s is for the phase shift.e.j] − x[i. we have chosen those parameters as s = 1.and [2.j] = −1/(j. The case π = (4312) From the chord diagram in Fig. 1 b= − 2. 4]solitons are given by x[i.and [2. that is. 23. a= − 2. 6. 4]types for y 0. so that all those four solitons meet at the origin at t = 0.j] :=x[i. and the three linesolitons determine the − + − location of the other one. all six possible solitons (i. 4]type for t = −6 and [1. 3] Fig. 3]. 4]solitons for y solitons in y 0.4] . and r for the onx[i.3] = 0.3] = 25/8 and A[2. i gives the xintercept of the line of the [i. In Fig.j] = −1/(j. We here take the parameters Θ[1. this solution contains all other solutions as some parts of this solution.4) + + In Fig. i with ΘT = ln 1.4] = 49/36. 4]solitons in y 0. 4]. The Amatrix in this case is given by for y A= 1 0 0 1 −b a −c 0 .and [2.3] .3] = Θ[1. This can be also seen by comparing the chord diagrams of those two cases. the amplitudes are A[1. 2 and c= + 3.3] e 3. 42. where a. b. 3]. Ttype. Θ[i. j]soliton for x − + set of the box. 4].C. 2]type for t = 6.and [2. 3]. The case π = (3421) The asymptotic linesolitons in this case are given by [1.4] = Θ[2. e 3.
4] e 4. The asymptotic solitons are given by [1. y. One should also note that this solution is “dual” to the previous one in the sense that the patterns of those solutions are symmetric with respect to the change (x. −y.e. This solution is dual to the previous one of (4312)type. b. It is interesting to note that the maximum amplitude soliton might be expected from the chord diagram. 2]. 2 with a = bc/d. a= + 2.4] . Fig. c. 2 and + d 4. 4]types for y 0. 7 and 8 along the xaxis and glues the lower half of Fig. 3] + [3. 8 illustrates the evolution of the solution of this type.and [2. This can be observed from the corresponding chord diagrams. 1 Θ[2. 1 Θ[1. d > 0 with ad − bc = 0. This has the maximum amplitude among the (local) solitons appearing in the solution. 1 and c= − 2. −t). Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 Fig. Notice that the nonlinear interaction generates the intermediate soliton of [1.4. For the case that all the solitons meet at the origin at t = 0. The Amatrix is given by A= 1 0 0 1 −c a −d b . t) → (−x. 2] + [2. 3]and [2.and [3. Y. 1 b= − 2. 4). Kao. 4] = [1.and [3. 3) and (2. 1 Θ[2. 8. 4] = [1.and [2. 6 (Ttype). 7 together with the upper half of Fig. t) ↔ (−x. The case π = (2413) The asymptotic linesolitons are given by [1. 2 0 and y 0 This solution can be considered as a dual of the previous case (b). 8.and [2.Y. 2]. e 4. 9. The asymptotic solitons are [1. c > 0 are free parameters. 4]. = e b 4. 4]types for y The intermediate solitons are [1. Example of (3421)type soliton solution. 9 illustrates the evolution of the solution of this type. Note here that now the upper pattern of the solution is the same as that of the Ttype. Those parameters are related to the locations of the linesolitons. hence all those four solitons meet at the origin at t = 0. where a. b= − 2. j) = (1. 1 c= − 2. 0 and [1. 3]types for y 0. two sets of linesolitons for y ± are exchanged. b. 1 Θ[2. 3]. y.j] to be zero. 4]solitons for y y 0.3] . 4). Here we take Θ[i.4] . The parameters in the Amatrix are related to the locations of the linesolitons. 3]type in t = 6.j] = 0 for those pairs (i. 4]type in t < 0. Fig.3. Fig. e 4. −t). 4] for y < 0. we set all Θ[i.3] e 3. 4]type in t = −6 and [2. 1 Θ[1. that is. Example of (2413)type soliton solution. 1 Θ[1.2] . So if one cuts Figs. 4] for y > 0 and [1. . we can obtain Fig. 3. 3]. that is. 0 and [1. e 3. 4]types for y 0 and [1. −y. this soliton is generated by the resonant interactions of [1.1196 C. the duality is given by (x. i. 4]solitons for where a. (2.
3] = Θ[2. where a. 2 b= + 3.e. 4]type is generated by the resonant interaction. 2 Θ[2. fory 0 and [1. Kao. 3]. The parameters are expressed by a= + 4. In fact.7. + Fig. 3]soliton for t > 0 are the same as those in Fig.3]. a= + 3.4] e . two solitons must have different amplitudes. Since those solitons can be placed both in nearly parallel to the yaxis. b. 1 Θ[2. and s represents the location of the other linesolitons Here Θ[i.5) + 0. The Amatrix is given by A= 1 0 0 1 0 a −b 0 . This solution is dual to the previous one of π = (2413). so that all solitons meet at the origin at t = 0. Example of (3142)type soliton solution. The asymptotic solutions are [1. and hence the solitons in x > 0 (the wave front) meet at the origin at t = 0.6) + where Θ[i.3] e 4. 4]. We here set Θ[i. 10. j]soliton in x > 0 (the wave front) (notice the phase shifts for the solitons).3. 11 illustrates the evolution of the solution of this type. this solution can be used to explain the Mach reﬂection observed in shallow water [14. This is then referred to as Ptype (after [12]). Here we take s = 1 and Θ[1. 6 (Ttype). 3]type (c.4] > A[2. 1 and c= 3. 4]soliton for t < 0 and [2. 4]solitons in y 0. 1 Θ[1.4] . se 3. 1 Θ[1. Notice the duality of this with (2413)type in the previous example. 3 (3. 4]solitons in y The Amatrix has the form. and the shift is in the . 3. The case π = (3142) The asymptotic linesolitons are given by [1. 2].3.and [2.3] . the large amplitude soliton of [1.Y. The case π = (4321) The asymptotic linesolitons are [1. c > 0. As in the previous case. the case (c) in Section 5). Y.and [3. 9 including [1.and [2. Fig. 3]types for both y → ±∞. for example. We represent those parameters in the following form which is particularly useful for our stability problem discussed in the next section. The situation is similar to the KdV case.5. A[1.4] = 0. 10 illustrates the dual behavior of the evolution + + in the sense of the reverse of time.6. A= 1 0 a 0 0 1 −c b . 4.C. In particular.f. 2]. that is the generation of the intermediate linesolitons either [1. 4]types for y 0.j] represent the location of the [i.or [2. i. 3.and [3. j]soliton in x in x 0.j] represents the location of the [i.and [2. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1197 Fig. a quasitwodimensionality. 4]types One should also note that the left patterns in Fig. Note here that there is a phase shift due to the interaction. 3]. 3 (3. this type of solutions ﬁts better in the physical assumption for the derivation of the KP equation.3] se 4. 3 and b= + 3.j] = 0. 1 s. 4]. 4. 0 and [1.
Kao. 2]. 4. when those solitons are almost parallel to each other and close to the yaxis. i with ΘP = ln 4. 2]. The interaction property for solitons with equal amplitudes has been discussed using this solution. 2 and b= + 3. contrary to the assumption of the quasitwo dimensionality for the KP equation [14]. and the interaction generates a negative phase shift of the solitons. 4]types for both y → ±∞. 6.7. The phase shifts are given by x[i.1198 C. and it is called Otype (after [12]). i)−ΘP /j. they form a part of the box in Fig. 2]. j] represents the location of the wave front consisting of [1. Example of (2143)type soliton solution (Otype). 3. 1 2. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 Fig. Example of (4321)type soliton solution (Ptype). Fig. Y. The pattern of the solution is stationary. 14. The pattern of the solution is stationary.8) where Θ+ [i. 3]solitons only appear as the intermediate solitons in the solution of Ttype. The asymptotic solitons are [1. 3 > 0. because the larger soliton of [1. However this solution becomes singular. The asymptotic solitons are [1.j] = −(1/j. The case π = (2143) The asymptotic linesolitons are [1. 1 Θ[1.Y.and [3. Here we take all Θ[i. that is. 23.and [2. 4]solitons. We will further discuss this issue in the next section when we study the initial value problem with Vshape initial wave. 2 Θ[3.j] = 0. 3]types for both y → ±∞. Notice that this Amatrix is the limit of that for (3142)type with c → 0 and k2 → k3 . 11. The parameters a and b are expressed by a= + 4. . (3.2] e 4. and has a positive phase shift in the solitons.and [3. the solitons in x > 0 (the wave front) meet at the origin at Fig. 4]type propagates faster than the other one. negative direction due to the repulsive force in the interaction similar to the case of KdV solitons. 4].and [3. 4]types for both y → ±∞.and [2. 2 (3. 4].7) This negative phase shift is due to the repulsive force in the interaction as in the case of the KdV solitons. 12. This solution was used originally to describe two soliton solution. One should also note that the interaction point moves in the negative ydirection. Those [1. 12 illustrates the + evolution of this type solution.3. The Amatrix is given by A= 1 0 a 0 0 1 0 b .4] e .
we are dealing with twodimensional problem with nonvanishing boundary values which provide energy ﬂow into the numerical domain. 13. y) ≈ v(±Lx .e.23]. We assume that near the boundaries y = ±Ly . y) ≈ 0 initially. This assumption prevents any disturbance from the boundary.j] = −(1/j.Y. W(y) = exp −an y Ly . i. and the interactions occur only in a ﬁnite domain in the xyplane. The phase shifts of those solitons are positive due to an attractive force in the interaction. and they are given by x[i. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1199 t = 0. u = v + (1 − W)u with n v = Wu.g. the accuracy for numerical scheme is usually tested by using the integrals such as the total mass and energy.and [3. 2]. where u0 consists of exact linesolitons satisfying (1 − W)u0 = (1 − W)u. 14. so that we consider the function v to be periodic in y. in our study. (3.1) where an and n are positive constants (we choose an = (1. the KdV equation). the solution can be described by exact linesolitons (i. and keeps the stability of the far ﬁelds consisting of wellseparated linesolitons. 4]solitons at k2 = k3 . y ≤ Ly }.and [3.C. This is the main argument of the paper [14] about the discussion on the Mach reﬂection of shallow water waves. we explain the details of our numerical scheme and give an error estimate of the scheme. However. we take Ly large enough so that the interaction will not inﬂuence those parts near boundaries up to certain ﬁnite time). In the numerical scheme. and we do not expect to see any strong effect from the boundary region in a short time (recall that the system has a ﬁnite group velocity for waves (4. 2].e.111)n ln 10 with n = 27 [23]). Note also that those [1. 4. Y. We are interested in studying the asymptotic behavior near the interaction point. The initial value problem considered here is essentially an inﬁnite energy problem in the sense that each linesoliton in the initial wave is supported asymptotically in either y 0 or y 0. the phase shift gets larger. u = v + (1 − W)u0 . 2 3. 4. With the window function W(y). this can be expressed by the decomposition of the solution u. 2 < 0. We take the window function W(y) in the form of a superGaussian. we choose Lx large enough so that u(±Lx . This implies a stability of those exact solutions under the inﬂuence of certain deformations (notice that the deformations in our initial waves are not so small). y) : x ≤ Lx . and we will show that the solution of the initial value problem with certain types of initial waves asymptotically approaches to some of the exact solutions discussed in the previous section. 6. We then consider the transformation in the form. Numerical simulations The main purpose of the numerical simulation is to study the interaction properties of linesolitons. 4]soliton as a result of the resonant interaction with [1. i)−ΘO /j. (4. In this section. and each linesoliton is matched with a KdV soliton at the boundaries y = ±Ly . We then check the accuracy with the maximum errors in the Fourier domain and the L2 estimates in comparison with the exact solutions of the KP equation. In x direction. we consider the rectangular domain D = {(x. Kao.2) . i with ΘO = ln 4.9) When k2 approaches to k3 . 4]solitons appear as the intermediate solitons forming a box of Ttype as shown in Fig. we ﬁrst transform our nonperiodic solution u having nonvanishing boundary values into a function v rapidly decaying near the boundaries y = ±Ly . and the interaction part of the solution is getting to be close to [1. We also remark that for onedimensional problem (e. Numerical scheme Using the socalled window technique [21.1. So we cannot use the integrals for the accuracy test of our numerical scheme.
m.3) becomes 3 2 ∂X ∂t v + Pv∂X v + Q∂X v + R∂Y v = F.3) by using fast Fourier transform (FFT). Y. m. t) = ˆ v(l. (4. Y. V ). v l 2 il for l = 0.3) with the scalings. 4L2 y Under the periodic assumption of v and the meanfree condition. our scheme based on the FFT generates some error (the zero mode cannot satisfy (4. in the derivation of the KP equation). i. we have ∞ v(X. With the transformation (4. Y ) : X ≤ π. 2 We would like to thank a referee for pointing out the issue of the zero mode for the numerical integration of the KP equation based on FFT method. i. π π X= x.e. we conﬁrmed that the accuracy of our scheme is not contaminated by the error introduced from the zero modes (see below).e.3) Assuming the solution v to be periodic in both x and y with zero boundary condition at y = Ly . Although our solutions x do not satisfy this condition.e.6) The periodic assumption implies that one needs to have the meanfree function in the xdirection. −L udx = 0. (4. we solve (4. t)ei(lX+mY ) .Y. Lx Ly Then (4. t)e−i(lX+mY ) dX dY. The algorithm seems to be quite stable within our computational times. y) : x ≤ Lx . and the equation becomes ˆ ∂t v + i with N(ˆ ) = F(v2 ).5) L P 1 ˆ ˆ ∂t V + il eict N(V e−ict ) = eict F(F ). / (4. Y. for 2 il We write this equation in the form of time evolution. However. 4L3 x ∞ and R= 3πLx . l=−∞m=−∞ ˆ Here the Fourier transformation v:=F(v) is given by ˆ v(l. We also checked the error due to the violation of the periodicity due to the meanfree condition even in the case where the zero modes of u is forced to be zero.2). ky ≈ kx . v x Since our solutions do not satisfy the meanfree condition. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 2 with nonzero kx component and the assumption of quasitwodimensionality. Then we have Rm2 P 1 ˆ − Ql3 v + il N(ˆ ) = F(F ). 1 Lx . The zero mode effect was also discussed numerically in [11]. we have the equation for v. y ≤ Ly } with D:={(X. i. 3 2 ∂x 4∂t v + 6v∂x v + ∂x v + 3∂y v = 6(1 − W)∂x (Wu0 ∂x u0 − ∂x (vu0 )) + 3(2W ∂y u0 + W u0 ). ˆ ˆ ∂t V = G(t. / (4.e. Y  ≤ π}. i. −Lx u dx = 0. and P= 3π . l = 0.2 ˆ ˆ Let c = Rm2 / l − Ql3 and V = eict v.1200 C.4) where F is the right hand side of (4.5) for those solutions discussed in this paper). we conﬁrmed that the overall error including the violation of the condition remained small within our accuracy (see below).1 For a convenience. let us rescale the domain D = {(x. within our computational times. t) = 1 (2π)2 π −π π −π v(X. 2Lx Q= π3 . Kao. Y = y.
01 and t = 5 × 10−3 on D = [−128. especially those of Ttype solution. these derivatives have rather complex formulae. 32].194 × 10−2 Order 1.07 1. We use a pseudospectral method to evaluate nonlinear term N(ˆ ) in (4. It can be clearly seen that the order of convergence is 2 because the error decreases by a factor around 4 when the time step is halved.98 2.82 . In Fig. we summarize the error of the solution given by ut − ut exact L2 (D) / uexact L2 (D) for two different time steps t = 0. The error is decreasing exponentially with respect to N until it saturates at round of error ≈ 10−16 for N ≈ 1280. For the function F on the right hand side of (4. Since the Ttype solution develops a box at later time. 6. machine accuracy ≈ 10−16 is reached for N ≈ 1280. From this test with N = 1536.3).Y.95 1.992 × 10−2 3.6). 13. In Table 1. Thus the numerical solutions are expected to be of 2nd order accuracy when there are enough Fourier modes in space and the solutions are vanished in large x. If we take t = 5 × 10−3 and keep running the simulation up to t = 12. Kao. 16] with 2048 × 256 Fourier modes.353 × 10−3 5.046 × 10−2 1. 6. The ﬁrst test intends to answer the question on how many Fourier modes are needed to ensure machine accuracy in space for the solution v of (4. 16] with 2048 × 256 Fourier modes.and ydirections for all the simulations discussed in this paper.3). However. 128] × [−16. we show the maximum error in the Fourier modes of the Ttype solution in the sense that we estimate maxF(vt ) − F(vt ) with respect to the numexact ber of Fourier modes N in both x.12 2. 64] × [−32. In order to verify the accuracy of the method.507 × 10−2 2. The table shows the t −2 and 5 × 10−3 . 128] × [−16. 1472 and 1536 for t = 0.675 × 10−3 1. 13.C.732 × 10−3 9.893 × 10−2 2. it requires more Fourier modes to well represent the solution.523 × 10−2 4. On the computational domain D = [−64.86 1. one needs to evaluate the derivatives v of the exact solution u0 . The numerical domain is D = [−64. and use the 4th order RungeKutta (RK4) method to solve the equation. respectively. We choose to use 8 Fourier modes per unit length in both x. Numerical accuracy test on Ttype solution shown in Fig. The graph shows the estimate of maxF(vt ) − F(vt ) as a function of the exact number of Fourier modes N × N/2 for t = 0. 1472 and 1536 for t = 0.211 × 10−2 t = 5 × 10−3 2. In the following simulations. 64] × [−32. the errors of the solutions are at this magnitude or smaller. The spacial error is within 10−10 and it can also speed up the computat tion. the magnitude of the error is always within 2. 3 and 6. 3 and 6. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1201 Fig.5 × 10−2 . Notice that it might take longer time to run such a ﬁne mesh for a very large computational domain. t = 10−2 t t t t t t =1 =2 =3 =4 =5 =6 1. The second test shows the overall algorithm is 2nd order accurate. Table 1 The convergence test of our numerical scheme on the domain D = [−128. we estimate the F in the 2nd order difference. we provide two tests. respectively. 32]. This indicates that our scheme is of error ut − ut exact L2 (D) / uexact L2 (D) with Ttype exact solution for the time steps t = 10 2nd order accuracy in time.645 × 10−3 4.986 × 10−3 7. we ﬁnd that we need roughly 1536/(2 × 64) = 12 Fourier modes per 1 unit in x to reach 10−16 accuracy. 3 and 6.and ydirections for the case shown in Fig. Y.
z < 0. in [20. Y. Numerical results Now we present the results of the direct numerical simulation of the KP equation with the following two types of initial waves consisting of two linesolitons. 0) = u0 (x. u(x.25].20. u(x. y) = 2 sech2 (x + y tan Ψ ) × H(−y).j] observed in the simulations provide the information of the interaction property.j] = ki + kj . y.13. Here what we mean by the convergence is as follows: We ﬁrst show that the solution ut (x. y) = 2 sech2 (x + y tan Ψ ). In particular. (5. and place these linesolitons in the symmetric shape with respect to the xaxis. Kao.j] = 2 (ki − kj )2 and tan Ψ[i. y) = A0 sech2 (x − y tan Ψ0 ) × H(y).13. . (5.e. 0 2 + Again note that the initial locations are x[i. we ﬁx the amplitude of one of the linesolitons. i. The exact convergence is only in a local sense. The shifts of x[i. 1 0 if if z ≥ 0. y) with 1 2 ⎧ ⎪ 0 A0 ⎨ u1 (x. For a given initial wave. the Vshape initial waves were considered to study the generation of large amplitude waves in shallow water. y) with 1 2 ⎧ ⎪ 0 A0 ⎨ u1 (x.3. and we use the following (relative) error estimate. y0 (t)) is chosen as the intersection point of two lines determined from the corresponding exact solution identiﬁed by the following steps: 1. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 5.e.1202 C.25]. 2 ⎪ 0 ⎩ u (x.7. This then gives the exact solution with the corresponding Amatrix. 0 2 where H(z) is the unit step function. y) + u0 (x.1) H(z) = + Here one should note that the initial locations x[i. y. y). y) ∈ R2 : (x − x0 (t))2 + (y − y0 (t))2 ≤ r 2 . the chord diagram) from the formulae. In particular. that is. 1 A[i. This then implies that each proﬁle of the initial wave is determined by two parameters A0 (the amplitude of other soliton) and Ψ0 > 0 (the angle of the wavevector of the soliton in y < 0): (i) Vshape wave consisting of two semiinﬁnite linesolitons.or Xshape generates a pattern due to the resonant interaction. they cross the origin at + t = 0. The main result of this section is to show numerically that those initial waves converge asymptotically to some of the exact solutions presented in Section 3. E(t):= ut − ut exact 2 t L2 (Dr ) / ut exact 2 t L2 (Dr ) .j] = 0. Note here that the entries of the Amatrix are not determined yet. Here the center (x0 (t). These initial waves have been considered in certain physical models of interacting waves [14. i. y) = A0 sech2 (x − y tan Ψ0 ).e. 0) = u0 (x.j] are zero for both linesolitons. y) + u0 (x.3) t where Dr is the circular disc deﬁned by t Dr := (x.2) 2 ⎪ 0 ⎩ u (x. (ii) Xshape wave of the linear combination of two linesolitons. (5.Y. we ﬁrst identify the type of exact solution (i.22. and we then identify the pattern near the resonant point with that given by some of the exact solutions ut (x.22. y) at the intersection point of V.
We have done the numerical simulations with six different cases which are marked in Fig.jl ] are expressed in terms of the Amatrix (see Section 3). Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1203 2. we have the following completions: Fig. + x + y tan Ψ[il . 3. we give T0 and T1 based on the observation of the numerical results. Vshape initial waves The initial waveform is illustrated in the left panel of Fig.C. 14. but for most of the cases. T1 and the radius r.j2 ] = 2 and Ψ[i2 .jl ] t = x[il . We then conﬁrm that E(t) further decreases for a larger time t > T0 upto a time t = T1 > T0 . that is. . one adds extra chords with minimum total length so that the complete chord diagram represents a derangement (see also [13]). This implies that the solutions have nontrivial location shifts. We are currently working on this problem. x[i. (5. 2. the center (x0 (t). The right panel shows the incomplete chord diagrams corresponding to the choice of the values A0 and Ψ0 . y0 (t)) of the circular domain Dr is given by the intersection point of the lines. i.5) + where the locations x[il .Y. 5.j1 ] = Ψ0 . The time T0 may be taken as T0 = 0. but Dr should be kept away from the boundary to avoid any inﬂuence coming from the boundaries.jl ] for l = 1. that is. A[i2 . 14. Kao. but it is also limited to avoid any disturbance from the boundaries for t < T1 . and hope will be able to report the complete analysis in a future communication. We set those linesolitons to meet at the origin.j] = 0 in / the solution. Each line of the Vshape is a semiinﬁnite linesoliton. just before the effects t of the boundary enter the disc Dr (those effects include the periodic condition in x and a mismatch on the boundary patching). The time T1 > T0 gives an enough time to develop a pattern close to the corresponding exact solution. A−matrix (5. the upper chord represents the (semiinﬁnite) linesoliton in y > 0 and the lower one represents the (semiinﬁnite) linesoliton in y < 0. we do not have a systematic way to estimate those values. that is. For our examples of the cases (a) through (f).e.4) The minimization is achieved by adjusting the solution pattern with the exact one. We then minimize the error function E(t) of (5. There is no corresponding exact soliton solution on the dotted line of k2 = k3 . The main result in this section is to show that each incomplete chord diagram representing the initial wave gets a completion with minimum additional chords which represents the asymptotic pattern of the solution. t t We take the radius r in Dr large enough so that the main interaction area is covered for all t < T0 . and ﬁx the amplitude of the soliton in y < 0 to be 2. Here we have A[i1 . the pattern determines the Amatrix whose entries give the locations of all linesolitons including the intermediate solitons in the solution. Unfortunately.jl ] − C[il . In this paper. T0 .3) at certain large time t = T0 by changing the entries in the Amatrix. Y. + we need T0 to be sufﬁciently large. The right ﬁgure shows the (incomplete) chord diagrams corresponding to the initial Vshape waves.j1 ] = A0 .1.j2 ] = −Ψ[i1 . we consider MinimizeE(T0 ). t With this Amatrix. Here the minimal completion implies that for a given incomplete chord diagram. Vshape initial waves. A0 is the amplitude of the soliton in y > 0. 14 as (a) through (f).
and this may be identiﬁed as [3. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 (a) With k1 = k2 . 2]. 5/4). 2]soliton in y > 0 appears ahead of the wavefront consisting of [2. 2]chord in y < 0 and [3. it locates ahead of the wavefront similar to the previous case (e)). 2]chord in y < 0 which appears in the asymptotic 3 1 2 solution. k4 ) = (−3/4. 4] and completes the incomplete chord diagram. and the semiinﬁnite initial solitons extend to generate other half with the phase shift (cf. i. 4]soliton in y < 0. i. 4]soliton 3 4 1 appears in the asymptotic solution in y > 0. 1)type. 4]chord in y > 0. the resonance condition gives [1. 4]chords in y > 0. i. add [1. 3). the product uv in the extra term has nonzero value only near the origin at t = 0 (this assumption is not totally correct. that is.4] = = tan Ψ0 . The linesolitons of Vshape initial wave are [1. We then expect v to be close to the solution of the KP equation with the initial wave of [3.and [3. tan Ψ[1. Since this [3. 2]. 4]soliton in y > 0 under the inﬂuence of this virtual soliton (see again below for the details). tan Ψ[1. 4]soliton in y < 0 is a virtual one (i. Kao. 3 2 2 ∂x (4∂t v + 6v∂x v + ∂x v) + 3∂y v = −6∂x (uv). 1 2 3 With k3 = k4 . 2]chord in y < 0 and [3. ⎧ ⎪ A[1. The simulation shows a generation of third wave at the point of the intersection of those initial solitons. i. add [1. the completion is π = (b) (c) (d) (e) (f) .) Since [1.e. The case (a) This is a critical case withk1 = k2 . Y. nonsoliton parts) propagate in the negative xdirection (see Section 3). The minimal completion is π = (2341).j] = 2 (ki − kj )2 and the angle tan Ψ[i. (Note that the initial wave is a front half of Ptype soliton solution. The soliton solution is of Otype.3] = − 1 = − tan Ψ0 .1. 2)soliton (cf. These corresponding solitons appear behind those initial solitons (cf. 4]soliton in y > 0 (see below for the details).Y. 3)soliton.e.e. add [1. The ﬁgures in the ﬁrst row of Fig. 15).e.k] = ki + kj . This implies that [3. The completion is π = (2143). but again we note that [3. 2)type. but the shifts of the initial solitons are very small for most of the cases). and the corresponding solution is a (2. but it gives an inﬂuence on bending the [1. Fig. The [1. Fig. Then v satisﬁes. Fig. 1 3 4 5. We also observe the bending of [1.e. k3 . 3]. 4]soliton in y < 0. 4]chord in y > 0. 4]soliton of the exact solution. add [1. 15 illustrate the result of the direct numerical simulation. Recall here that the dispersive waves (i. add [1. the generation of [1. 12). The domain of computation is [−192. . This is a (1. 1)soliton (cf.e. 3]soliton in y > 0 and [1.and [1. we have the kparameters (k1 = k2 . 192] × [−48. This argument may be also applied to other cases as well. 2 1 With the formulas for the amplitude A[i. 10). The minimal completion is π = (4123). 3). and the corresponding exact solution is a (1. The is a (3. i.and [3.e. add [3. The generation of [3.1204 C. 4]chord in y > 0. 3] + [3. 1/4.3] = A0 = 1 . 4]soliton in the exact solution as one can see from u = uexact − v. 4]soliton is missing in the front half (x > 0).e. Then the edge of the solution v disperses away with a bowshape wake propagating toward the negative xaxis as can be seen in the simulation (cf. The corresponding exact solution is of (1. 4]solitons. Fig. 3]soliton in y > 0 and [1. where u is the solution of the KP equation with our initial wave of [1. 4]soliton may be also understood by decomposing the exact solution into two parts. Then the decay of the v solution at the front edge implies the appearance of the [3. This generation is due to the resonant interaction. the completion is π = . If we assume that the initial solitons remain as the part of the exact solution uexact . The completion is π = (3142).1. say uexact = u + v. i. the edge of the wave gets a strong dispersive effect. but the minimal completion is not of Ptype. ⎨ 2 2 1 ⎪ ⎩ A[1.4] = 2. 2]soliton is impossible. which is the KP equation with one extra term in the right hand side. 2]soliton is a virtual one. 48] and there are 8 Fourier modes in every unit length. 4] = [1. 4]chords in y < 0. 4]soliton only for y > 0 and x < 0. We take the following values for the amplitudes and angles of the those solitons. Fig.
3]soliton in y > 0 and [1.C. Y.3). we obtain the Amatrix of (1. we plot the error function E(t) of (5. Also the negative shift in the [1.4). 4]soliton is due to the generation of the [3. The minimization is achieved by adjusting the locations of the [1. This causes a negative shift of this soliton.3) which is minimized at t = 10.112).2)). The circles in the ﬁgures t show Dr where we take r = 12 and the center is given by (x0 (t) = (13/16)t − 0. The ﬁgures in the second row of Fig. 1/4. i. 2)type. 14: The initial wave consists of [1. and it gives the shifts + x[1. 15. Kao.and [1. A dispersive radiation appearing behind the interaction region may be considered as a reaction of the generation of the third soliton of [3. 4]soliton. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1205 Fig. 15 show the exact solutions generated by the τfunction with those parameters (k1 = k2 . 5/4)) and the Amatrix given above. 4]soliton in y < 0. 15. 48]. 4]soliton is generated as a part of the tail of [1.0079. the ﬁgures in the second row show the corresponding exact solution of (1. and in the beginning stage of the generation.Y. A= 1 0 0 1 −1. (The Amatrix is then obtained using (3. + x[1. Minimizing the error function E(t) at t = T0 = 10 with (5. The t circle in the exact solution shows the domain Dr with r = 10 for the error function E(t). 4]type as a result of resonant interaction (see the text for the details). 2)type. 2)type with the above Amatrix and the same kparameters. The ﬁgures in the ﬁrst row show the result of the direct simulation. .e. 4]soliton has a small amplitude and propagates with slow speed. 15 indicates that the asymptotic solution seems to converge to the exact solution of (1. 4]type has a relatively large negative shift. k4 ) = (−3/4.4] = −0. Fig.63 . 3]. Numerical simulation of the case (a) in Fig. losing its momentum. Thus the new linesoliton of [3. the [3. 192] × [−48. k3 . 4]solitons of the wavefront. This may be explained as follows: The [3. and the bottom one represents the error function E(t) of (5. In the bottom ﬁgures of Fig. The domain of the simulation is [−192. 4]soliton.120. y0 (t) = (3/4)t − 0.3] = −0.0637.27 0.
14: The initial wave consists of [1. t The computational domain is [−192. The bottom graph shows the error function E(t) of (5. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 5. + The shifts of locations x[i.3] = 2.3] = −0. 4]solitons in y > 0 and y < 0 respectively.2. + x[2. 16 illustrates the result of the numerical simulation.3] = − 3 = − tan Ψ0 . u = uexact − v.3] = A0 = 49 . The case (b) This is also a critical case with k3 = k4 . The ﬁgures in the second row show the corresponding exact solution of (2. The ﬁgures in the ﬁrst row show the direct simulation of the KP equation.Y.0545.1. whose Amatrix is obtained by minimizing the error function E(t) at t = 6.3] = 3 = tan Ψ0 . tan Ψ[2. Kao. 4 This then gives the kparameters (k1 .1206 C. 2]soliton.3] = −0. 1)type. Fig. 3]soliton in y < 0. 48]. 11/8). −5/8. 2]soliton part in the exact solution. Notice again that a bowshape wake behind the interaction point corresponds to the v solution in the decomposition. 1)type. We take the amplitudes and angles of those solitons to be ⎧ ⎪ A[1.21 . and the decay of v implies the appearance (or generation) of [1. A= 1 1. 16. The linesolitons of Vshape initial wave are [1. Notice that v gives the [1.3) which is minimized at t = 6. Y.15 1. tan Ψ[1. The ﬁgures in the ﬁrst row show the result of the direct simulation. Fig.0254. and the circle in the ﬁgures of the exact solution shows Dr with r = 12. 3]soliton in y > 0 and [2.and [2. k2 . ⎪ ⎨ 8 4 ⎪ ⎪ ⎩A [2. 192] × [−48. 4]. and the ﬁgures in the second row show the corresponding exact solution of (2. k3 = k4 ) = (−17/8. .j] obtained in the minimization are given by + x[1. Numerical simulation of the case (b) in Fig.
4]soliton in y < 0.4] = 4. We observe a bowshape wake behind the interaction point. 192] × [−48. Fig. Kao. 4]soliton with the amplitude A[1.C.4] = 4. 3]soliton in y > 0 and [2.3.1). 3]soliton in y > 0 and [2.3] = −1 = − tan Ψ0 . The wake expands and decays. 48]. y0 (t) = −(11/8)t + 0. The bottom graph in Fig. 3/2). 4]soliton describes the wave called the Mach stem [3. A large amplitude intermediate soliton is generated at the intersection point. tan Ψ[1. and then we see the appearance of new solitons which form resonant interactions with the initial solitons. Numerical simulation of the case (c) in Fig.4] = 2. The exact solution obtained by the completion of the chord diagram is given by (3142)type soliton solution.Y. 4]soliton with the amplitude A[1. tan Ψ[2. 5. and this soliton is identiﬁed as [1. In those ﬁgures. the t domain Dr has r = 12 and its center (x0 (t) = (157/64)t − 0. Y.5.1. We consider the case with the larger soliton in y > 0.13]. k2 . 17. The case (c) The linesolitons of the initial wave are [1. k4 ) = (−3/2. A[2. 4]soliton in y < 0. The kparameters are then given by (k1 . and it corresponds to the [1. and the ﬁgures in the second row show the corresponding exact solution of (3142)type.0399. t The computational domain is [−192. . and the circle in the middle ﬁgures shows Dr with r = 12.0194).3] = A0 = 2.5. −1/2. 14: The initial wave consists of [1.4] = 1 = tan Ψ0 . k3 . Fig. 16 shows that the solution converges asymptotically to the exact solution of (2. The ﬁgures in the ﬁrst row show the result of the direct simulation. The bottom graph of the error function E(t) which is minimized at t = 10. 1/2. and take the amplitudes and angles of those solitons as A[1. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1207 using (3. The ﬁgures in the ﬁrst row show the direct simulation of the KP equation. 17 illustrates the result of the numerical simulation. 1)type with the above Amatrix and the same kparameters. and the [1. Those negative shifts may be explained in the similar way as in the previous case. One should note that the solution generates a large amplitude intermediate soliton at the interaction point. This solution has been considered to describe the Mach reﬂection of shallow water waves.
4]type (i. k2 . ⎨ 5 12 ⎪ ⎩ A[3. 3]. A= 1 0 1. The negative shifts imply the slowdown of the incidence waves due to the generation of the solitons extending the − initial solitons in the negative xdirection.2] = x[3. The ﬁgures in the second row show the corresponding Otype exact solution whose Amatrix is determined by minimizing the error function E(t) at t = 10.17 + + Because of our symmetric initial data.Y. and using (3.j] .e. Fig.4] = −0. The case (d) The linesolitons of the initial wave are [1.j] − x[i.and [2. we adjust the locations of solitons including the intermediate solitons. Notice that the wake disperses. 4]soliton now resonantly interact with [1.3] = x[2. This implies that the solution extends the initial solitons to the negative xdirection with the same types.4] = −0. −1/5. The negative value of the shifts + x[i. and this corresponds to the v solution in uexact = u + v.4] = 2. i.980.3.and [3.4] for the minimization. 18 illustrates the result of the numerical simulation. The ﬁgures in the ﬁrst row show the direct simulation of the KP equation. 2]. 1/5. The bottom graph in Fig.e.9).4] = = tan Ψ0 . This case corresponds to the solitons with a large angle interaction.2] = x[3. tan Ψ[1.5).j] = x[i. and they are x[1.2] = x[3.j] . the small negative shifts of x[i.020. and the exact solution starts to appear as the wake decays (recall again the decomposition uexact = u + v (where v is the wake and u is the solution). 2]. The bottom graph in Fig.64 In the minimization to ﬁnd the Amatrix. Y. and using (3. we obtain + + x[1. 4]types in y < 0 and y > 0. tan Ψ[3. The wake behind the interaction point has a large negative amplitude.2] = A0 = 2. i. we use Dr with r = 12. even though we observe a large wake behind the interaction point. and for the error function.2] = − 12 = − tan Ψ0 . 5 The kparameters are then given by (k1 . we consider a symmetric initial data with the initial wave whose amplitudes and angles are given by ⎧ ⎪ A[1. we consider the shifts x[1. The phase shifts − + x[i. 11/5).1208 C. (2143)type. 17 shows a rapid convergence of the initial wave to (3142)type soliton solution with those t parameters of the Amatrix and k values given above.01.j] is due to the generation of a large amplitude soliton [1.e. The completion of the chord diagram gives Otype soliton solution. we obtain + + x[1. i.91 0 0 1 0 0.4. Those values indicate that the solution is very close to the exact solution for all the time. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 The ﬁgures in the second row in Fig.92 0 0 1 −1. 2]soliton in y > 0 and [3. [1. 4]solitons to create new solitons [1.1. respectively.96 0. 5. Those new parts of solitons have the locations x[i. The positivity of the phase shifts is due to the attractive force between the linesolitons.e.8).593. k4 ) = (−11/5. and this explains the slowdown + of the initial solitons. One should note that the solution also generates the phase shift determined uniquely by the kparameters. 18 shows E(t) which is minimized . 4]solitons (called the reﬂected waves as considered in the Mach reﬂection problem [14. A= 1 0 1.4] = 0. the initial solitons loose their momentum).and [3.j] for the linesolitons are calculated from (3. k3 . 4]soliton in y < 0. 4]soliton is generated after t = 0. and s < 1 implies that the [1. For simplicity. Also note that the [1. 17 show the corresponding exact solution of (3142)type whose Amatrix is found by minimizing E(t) at t = 10. Kao. This process then seems to compensate the shifts of incident waves. s = 0.13]).
4] = 2.23]. [2. and the ﬁgures in the second row show the corresponding exact solution of (2143)type (Otype). The bottom graph shows the error function E(t) which is minimized at t = 6.3] = 0 = − tan Ψ0 . oblique to the yaxis with Ψ0 > 0). Note that the case with k2 = k3 also corresponds to the boundary of the previous case (c) (See Fig. The upper ﬁgures show the result of the direct simulation. 3]. 3)type soliton solution with [1. This large amplitude wave generation has been considered as the Mach reﬂection problem of shallow water wave [14. A[2.and [3. there exists a large phase shift in the soliton solution. and we expected to see a Ptype soliton as the asymptotic solution. One can see a rapid convergence of the solution to the Otype exact solution with those parameters.13.25.and [2. . 3]soliton in y > 0 and [1. and the amplitude of the intermediate soliton generated at the intersection point reaches four times larger than the initial solitons at this limit. However. 4]. 2]soliton in y > 0 and [3.3.20. 14). Numerical simulation of the case (d) in Fig. 4]solitons in y > 0. √ √ Then the kparameters are given by (k1 . k2 .e. 1/(2 2). 18. The case (e) The linesolitons of the initial wave are [2. −1/(2 2). Y. 4]soliton in y < 0 and [1. 4]solitons. Here one should note that the [1. 2].4] = 0 = tan Ψ0 . and we take r = 12 for the domain Dr . 4]soliton in y < 0. k4 ) = (−1. k3 .5. 2]soliton in y > 0 locates ahead of the wavefront consisting of [1.C. 4]soliton in y < 0. tan Ψ[1. Notice a large wake behind the interaction point which extends the initial solitons.3] = A0 = 0. 160] × [−32. but the asymptotic solution is somewhat close to the exact solution of (2341)type as obtained by the minimal completion of the chord diagram. One should however remark that for a case near the critical one with k2 = k3 . the numerical simulation suggests that we do not see a solution of Ptype. 5. so that this soliton cannot be generated by the resonant interaction of those [1. 3]solitons. Kao.and [2. tan Ψ[2. A[1. We consider the solitons parallel to the yaxis with the amplitudes. 1). 4]. 32] and the circle in the exact solution shows Dr with r = 12. and the convergence is very slow.1.Y. t The domain of the simulation is [−160. The (2341)type is a (1. We also considered the case with nonzero angle (i. t at t = 6. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1209 Fig. 14: The initial wave consists of [1.
1. 2]soliton implies that [1. 19. and this soliton may give a good approximation of the bending part of the [1.3] = 2. 2]soliton as a virtual one. 4]soliton observed in the simulation seems to appear at the point where the [1. 4]soliton in uexact appears when the solution v decays in this part. and the lower ﬁgures show the corresponding exact solution of (2341)type obtained by the minimal completion of the chord diagram given by the initial wave. 4]soliton. the tail part in y > 0 seems to have a resonant interaction with the [2. The dispersive radiation behind the intersection point may be the part of this [1. This behavior of the solution may be explained by the chord diagram of (2341)type whose corresponding exact solution is illustrated in the bottom ﬁgures in Fig. 2]soliton never appears in the solution u. −1.6.4] = 0 = − tan Ψ0 . 4]soliton in y > 0 and [2. 3]soliton in y < 0. 2]. because of the cancellation of uexact and v on this soliton part. 4]soliton. The upper ﬁgures show the result of the direct simulation. 192] × [−32. Y. Fig. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 Fig. 32]. 2]soliton part seems to occur only in the region behind the front solitons as seen at t = 5. 19 illustrates the result of the numerical simulation. 4]soliton in y < 0. which is not visible but has an inﬂuence to the initial solitons. A[2. Then the [3. The upper ﬁgures show the direct simulation. 20 illustrates the result of the numerical simulation. The computational domain is [−192. On the other hand. Notice that the bending on the initial [1. 2]soliton has a resonant interaction with [1. 3]initial soliton and becomes a new soliton. Also we note that the bending of the [1.4] = A0 = 6. This can be also understood by the similar argument using the decomposition uexact = u + v. On the other hand. 4]soliton. We will give more details in a future communication. 4]soliton propagates faster than the lower one. .1210 C. 1)soliton is the corresponding exact solution which is of (4123)type as given by the minimal completion of the chord diagram. Numerical simulation of the case (e) in Fig. 19: The new soliton generated in y > 0 may be identiﬁed as [3. In this case. 3]soliton. We then take the amplitudes and angles of the those solitons as A[1. we expect that (3. k2 . Nondecaying of the [1. Since the upper [1. 2]soliton intersects. tan Ψ[1. It is interesting to consider this [1. k4 ) = (− 3. we have now the larger soliton in y > 0 and the smaller one in y < 0. 3]soliton in y > 0 and [1. Following the similar arguments as in the previous case. 4]soliton. 4]soliton. Contrary to the previous case. This means that the intermediate soliton in the exact solution is [2. the decay of v in the [1. we also consider the solitons parallel to the yaxis. 5. 4]solitons in uexact . and the lower ﬁgures show the exact solution of (4123)type. the initial data for v is given by the [1. The case (f) The linesolitons of the initial wave are [1.and [3. 4]soliton in y < 0 propagates faster and its tail extends to y > 0. Kao. Fig.Y. In a comparison. √ √ Then the kparameters are given by (k1 .1. We observe that the [1. tan Ψ[2. the lower tail of this soliton appears and then resonates with the lower [2. 3).3] = 0 = tan Ψ0 . The ﬁgures in the ﬁrst row show the direct simulation of the KP equation. k3 . Also note that a large dispersive wake appearing in y < 0 has negative amplitude and disperses in the negative xdirection. 14: The initial wave consists of [2. 4]soliton appears near the point where the virtual[1.
4]soliton part in the exact solution does not appear in the solution of the simulation. 5. and the lower ﬁgures show the corresponding exact solution of (4123)type obtained by the minimal completion of the chord diagram. Xshape initial waves The initial waveform is illustrated in the left panel of Fig. Kao. 4]soliton. Numerical simulation of the case (f) in Fig. For example.C. that is. 4]soliton appears near the intersection point with the v irtual [3. We set those linesolitons to meet at the origin. and the simulations provide stability analysis for those exact solutions. As in the previous case (e). 4]soliton in y > 0 and [2. The computational domain is [−192. A0 is the amplitude of the soliton with negative angle −Ψ0 . 4]soliton is a virtual one. 21. 21.Y. Fig. This can be also explained with the similar argument as in the previous case. 3]soliton in y < 0. y = 7) at t = 6. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1211 Fig. . 4]soliton. 20. and the dispersive radiation observed in the simulation is a part of this [3. 32]. The main result in this section is to show that each simple sum of linesolitons gets asymptotically to the exact solution corresponding to the chord diagram given by the initial wave. We carried out four different cases marked with (g) through (j) in Fig. 4]soliton to get bending near (x = 18. Y. 14: The initial wave consists of [1. the [3. tan Ψ0 ). Notice that the bending in the [1. Notice again that the [1. 21. 192] × [−32. One should however note that those initial waves are not the exact solutions. and ﬁx the amplitude of the soliton with positive angle Ψ0 to be 2 (the corresponding chord is shown in the lighter color in each region). There is no exact solution for the degenerate cases marked by the dotted lines.2. Xshape initial waves. this case shows that the [3. Each line of the Xshape is an inﬁnite linesoliton solution. The right ﬁgure shows the chord diagrams corresponding to the initial waves with the parameters (A0 . The right panel shows the corresponding chord diagrams with the choice of the values A0 and Ψ0 . 4]soliton in y > 0 has a virtual interaction with [3. The upper ﬁgures show the result of the direct simulation.
21: The initial wave is the sum of [1. and the circle of the exact solution t shows Dr with r = 12. Y. Minimizing the error function E(t) at t = 12. we obtain the Amatrix. 4]solitons. tan Ψ[1. we also take a symmetric initial wave whose amplitudes and angles of the initial solitons are given by A[1. 4]soliton.and [3. The ﬁgures in the ﬁrst row show the numerical simulation.015 Fig. since the generation of a large phase shift needs a long time computation).01. Kao. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 if the exact solution has the phase shift.4] = 2 tan Ψ[3. The numerical simulation in Fig. The domain of the simulation is [−192. and the ﬁgures in the second row show the corresponding exact solution of (2143)type.e. 22. To emphasize the phase shift appearing in the exact solution of (2143)type.029 0 0 1 0 0.01. 2]. the linesolitons in the left side are accelerated by their interaction. 5. 32].01). Notice that a small dispersive radiation appears to generate the correct phase shifts. 192] × [−32.1. 22 demonstrates the generation of the phase shifts after the interaction. The bottom graph shows E(t) of (5.4] = 2. that is. 2. i. k2 . k3 .Y. we consider the initial solitons to be close to the boundary at k2 = k3 (but we cannot take the values so close.01.2.and [3.2] = A0 = 2. For simplicity. Otype. 2]. Note here that the interaction generates the dispersive wake behind the interaction point which propagates in the negative xdirection. A= 1 0 3.02 = − tan Ψ0 A[3.1212 C.3) which is minimized at t = 12. the asymptotic solution actually converges to this exact solution by generating the correct phase shifts.2] = −2. . −0. The case (g) The initial wave consists of [1. k4 ) = (−2. 0. Numerical simulation of the case (g) in Fig.02 = tan Ψ0 The kparameters are then given by (k1 .
r = 3. tan Ψ[1. k4 ) = (−3/2. the interaction part in the initial wave has a larger amplitude than that of the exact solution.4). the positive phase shifts) imply the deceleration of the front parts of solitons.3). The positive shifts of those [1. and s < 1 shows the positive phase shifts as calculated above.and [2. The bottom graph in Fig. if we take Ψ0 = 0. x[1. 4]type at the front. 4]soliton.4] = 0. k3 . we take x[1.198 −0. The detailed study also shows that near the intersection point for Ttype solution at the time when all four solitons meet at this point (i. 22 shows the corresponding Otype soliton solution t with this Amatrix. The ﬁgures in the second row in Fig. the solution has a small amplitude due to the repulsive force similar to the KdV solitons (see also [4]. −1/2.350.4] = 1.e Xshape).368 1. Ttype). we obtain the KdV 2soliton solution with different amplitudes. so that this part of the solution moves faster than that in the exact solution.963 > 0. This is due to the repulsive force exists in the KdV type interaction explained above.and [2.2.3) at t = 6. The circle shows the domain Dr with r = 12. which clearly shows an opening of a resonant box as expected by the chord diagram of Ttype.C. 4]soliton.4] = 2 The kparameters are then given by (k1 . s = 0. This difference may result as a shift of the location of the [1. We obtain + + x[1. So for the case with very small angle Ψ0 . Notice the dispersive wake behind the interaction point in the simulation. 5.2. This then implies that our initial wave given by the sum of two linesolitons creates a large dispersive perturbation at the intersection point (this can be also seen in the next cases where we discuss the Ptype solutions).e.4] = −0.Y.3] = −1 = − tan Ψ0 tan Ψ[2. k2 .2] = x[3.4] = 1 = tan Ψ0 A[2. we consider a symmetric initial wave. the solitons expected from the chord diagram have almost the same amplitude as the incidence solitons for the case with a small angle. For simplicity. Namely the accelerations of the back parts of the linesolitons (i. However. 23 illustrate the numerical simulation.4] = 0. and adjust the onset of the box (recall (3.3] = x[2. and note that the symmetry reduces the number of free parameters to three). 3/2). and then we obtain + + x[1. Although Ttype soliton appears for smaller angle Ψ0 . A= 1 0 0 1 −0. 3]. 1/2.21. The phase shifts of the initial waves are obtained from (3. x[1. .4] (due to the symmetry of the solution). Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1213 + + In the minimization by adjusting the locations of the solitons. The ﬁgures in the ﬁrst row in Fig.2] = x[3. we expect to see those KdV solitons near the intersection point.4] due to the symmetric proﬁle of the solution. For an example of the symmetric case. one should not take so small value. 4]solitons in the wavefront indicate also the positive shift of the newly generated soliton of [1. 3]. where the initial Xshape wave with small angle generates a large soliton at the intersection point).025.3] = x[2. that is. The case (h) The initial wave is the sum of [1. where the Amatrix of the solution is obtained by minimizing the error function E(t) of (5.3] = x[2.330 0. Kao. The relatively large value r > 1 indicates that the onset of the box is actually much earlier that t = 0.549. 22 shows the convergence of the solution to the Otype solution given by those Amatrix and the same kparameters.2] = x[3. The corresponding chord diagram is of (3412)type (i. and take the amplitudes and angles of those solitons to be A[1. and we expect the generation of a box at the intersection point of those solitons. Y.3] = A0 = 2.e. The negative values are due to the generation of the positive phase shifts given by (3.63. The corresponding exact solution is illustrated in the ﬁgures in the second row.123 ± ± In the minimization.9). we use x[1.
Y. 4].3] = −0. 21: The initial wave is the sum of [1. 24 show the numerical simulation. Then the effect of the nonlinear interaction may be observed in the lower half (i. We take the amplitudes and angles of those solitons to be ⎧ 8 ⎨A .3] = A0 = 25 ⎩ A[1. 192] × [−48. Kao. 5. and the circle in the exact solution shows the domain Dr with r = 22.2.1214 C. The case (i) The initial wave consists of [1.e. and one should have a large computation time to see the convergence. The computational domain t is [−192. behind the interactions) of the solution. k2 . t Note here that the circular domain Dr with r = 22 covers well the main feature of the interaction patterns for all the time computed for t ≤ 7.4] = 0.and [2. One should also note that the large bending of the [1.e. tan Ψ[2. This is due to the interaction with the dispersive wake generated around the origin at t = 0. 4]soliton propagates faster than the [2. i. and notice that this bending part disperses . the positive ydirection. and the ﬁgures in the second row show the corresponding exact solution of (3412)type. The ﬁgures in the ﬁrst row show the numerical simulation. Here we show how the phase shift is generated in the simulation. −3/5. 23.e.3) which is minimized at t = 6. 6/5).3. 3]soliton. 4]soliton in y < 0. k4 ) = (−4/5. The corresponding diagram is of (4321)type (Ptype).4 = − tan Ψ0 [2. k3 . This may be explained as follows: Since the [1. 23 shows the evolution of the error function E(t) of (5. 1/5. and it generates the phase shifts. Numerical simulation of the case (h) in Fig. In this case. the intersection point moves in the upper direction.4] = 2 tan Ψ[1. i. 4]solitons. The bottom graph in Fig. 3]soliton. Ttype. 48]. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 Fig. and the exact solution of this type has negative (repulsive) phase shift.4 = tan Ψ0 The kparameters are then given by (k1 . we also note that taking a very small angle Ψ0 may give a large perturbation to the exact solution. The top ﬁgures in Fig.and [2.3) which is minimized at t = 6. The bottom graph shows the error function E(t) of (5. The simulation indicates the generation of the phase shift in the lower part of the solution. 3].Y.
we obtain + x[2.3) at t = 12. Note that the bending on the solitons disperses and the interaction pattern converges locally to that of the corresponding exact solution.3] = −2.54. Y. 24 show the corresponding exact solution of Ptype whose Amatrix is obtained by minimizing the error function E(t) of (5. 3]solitons. and the circle in the exact solutions shows Dr with r = 10.10. and using (3.629 −0. x[2. The bottom graph in Fig. which can be also observed in the simulation.Y. The bottom graph shows the error function E(t) of (5. Notice that the larger shifts in the [1.3] = −0. A= 1 0 0 1 0 1. and the bending regions are getting separated from the intersection point with the correct phase shifts. This shift is due to the repulsive force appearing at the intersection point (recall that the exact solution has a smaller amplitude at this point). 24 shows the decay of the error function E(t) t with Dr for r = 10.and [2. The ﬁgures in the second row in Fig. and the ﬁgures in the second row show the corresponding exact solution of (4321)type. Kao. + x[1. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1215 Fig. and we may need to perform a larger domain computation.4] = −1.4] = −0. which may be due to the large deformation of the solution.75. 24. Those large phase shifts are also observed in the simulation. The phase shifts of the Ptype solution is given by (3. The ﬁgures in the ﬁrst row show the numerical simulation. we adjust the location of the linesolitons at the wavefront. away. 48].3) which is minimized at t = 12.038 0 In the minimization. The computational t domain is [−192. 192] × [−48. 4]soliton.e.7). i. Ptype. . x[1.125. 21: The initial wave is the sum of [1.6). 4]. Numerical simulation of the case (i) in Fig. We also note the saturation E(t) after t = 11.C.
The bottom graph illustrates the error function E(t) of (5.and [2. 4]. . Kao. We can see a clear separation between the dispersive waves causing the deformation of solitons and the intersection points of Ptype. and the ﬁgures in the second row show the corresponding exact solution of (4321)type. The ﬁgures in the second row in Fig. 3/2). and we expect that the phase shift now appear in the upper half of the solution. and then the wake pushes the [1. and the circle in the middle ﬁgures shows Dr with r = 10. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 5.e. We take the amplitudes and angles of those solitons to be ⎧ ⎪ A[1. 3]solitons. the larger soliton of [1. 4]soliton forward. The graph shows a further reduction of the error. Since the [1. tan Ψ[1.3] = = tan Ψ0 [2. i. The domain t of the simulation is [−192. that is. the intersection point moves in the negative ydirection. 64]. Numerical simulation of the case (j) in Fig. 3]soliton near the origin is due to the interaction with the wake. 3]soliton in the initial wave. the convergence of the solution to the corresponding exact solution even with a drastic change of the error at the initial stage. 25. k4 ) = (−7/4.3] = 2 4 The kparameters are then given by (k1 .2. Y. The case (j) This case is similar to the previous one with [1.3) which is minimized at t = 8.and [2. Also the large bending of the [2. The ﬁgures in the ﬁrst row show the numerical simulation. 25 show the numerical simulation.4] = A0 = 169 .Y. k3 .4. 192] × [−64. −7/8. 4]type has a negative angle (propagating in the negative ydirection). The ﬁgures in the ﬁrst row in Fig. k2 . 4]soliton propagates much faster than the other one. 4].1216 C. However in this case. 9/8. 21: The initial wave is the sum of [1.4] = − 1 = − tan Ψ0 ⎪ ⎨ 32 4 ⎪ 1 ⎪ ⎩A tan Ψ[2. One can observe a dispersive wake is generated by a large perturbation in the initial wave near the origin. 25 illustrate the corresponding Ptype solution whose Amatrix is obtained by minimizing the error function E(t) at t = 8. Ptype. Fig.
We started to provide a brief review of the classiﬁcation theorem for soliton solutions of the KP equation. We considered a special class of the initial waves with V.627.3] = −1. We then performed the numerical study of the initial value problems using a pseudospectral method with window technique.Y. With this minimization. Our present study indicates the stability of some of the exact solutions. we get A= 1 0 0 1 0 8. is partially supported by NSF grant DMS0811003 and the Alfred P. Sloan fellowship. Here the minimal completion means that the resulting chord diagram should give the unique derangement (permutation without ﬁxed points) and the corresponding solution has the smallest total length of the chords. we adjust the locations of the linesolitons at the wavefront.4] = 0.C. and we expect that our study can provide useful information even in the case of the solutions including a large number of linesolitons.923. we studied numerically the interaction feature of the KP equation for the solutions consisting of small number of linesolitons. the Mach reﬂection phenomena in shallow water. 25 shows the decay of the error. 3]soliton shifts x[1.4] = −0.or Xshape in the solution. + x[2.7). We also conﬁrmed that the exact solution in the convergence is predicted by the minimal completion of the (partial) chord diagram describing the initial Vshape wave. 4]soliton behind the [2.4] + x[1. Acknowledgements We would like to thank Professor Thiab Taha for giving us the opportunity to present our result at the Sixth IMACS International Conference on March 23–26. The positive shifts are due to the repulsive interaction at the intersection point as observed in the simulation.3] = 0. Then the separation implies that dispersive radiations eventually escape from the compact domain which is moving along the exact soliton solution. Y. The bottom graph in Fig. and using (3. . However after t = 3 the solution has a monotone convergence to the exact solution. x[1. 6. x[2.g.518. similar to the solitons in the KdV equation where the initial wave evolves into a sum of solitons and dispersive wave trains propagating opposite direction to the solitons. This leads to a convergence of the initial wave to certain exact solution.4] = −0.4] = x[1.299 0 In the minimization. This class of initial data was considered to study some physical problem related to the generation of large amplitude waves. is partially supported by NSF grant DMS0806219. the bending part disperses and the intersection point gets away from this part. we remarked that each soliton solution can be expressed by a unique chord diagram which contains the information of the asymptotic linesolitons of the solution in y ± ∞. The large bending in the simulation is generated by a strong perturbation generated at the intersection point at t = 0. Y. Here we proposed a notion of the local convergence in the sense that the L2 error is measured in a compact domain which covers the main part of the interaction of two soliton solutions forming V. We emphasize that the error function gets a rapid change in the early times. We would also like to thank Professors Masayuki Oikawa. Kao.934.549 −0.K. In particular. and it gets a larger shift in the upper part. C.K. We observed the separation of the exact soliton solution from dispersive radiations. 2009. implying that the t dispersive waves are getting away from the domain Dr as we expected. Conclusion In this paper. we obtain + x[1. e. − + Notice that the [1. However.6).Y.and Xshape which consist of two distinct linesolitons. The phase shifts of those incidence solitons are given by (3. Hidekazu Tsuji and Sarvarish Chakravarty for various useful discussions related to the present work. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 1217 We take the radius r = 10 for a good separation of the dispersive perturbation from the steady Ptype soliton part.307. This implies that a drastic change in the proﬁle of the solution due to a large perturbation to the exact solution.
Markowich.Dokl. Phys. 15 (1970) 539–541. Chen. Sci. Newell. Kobe. 1984.J. 2008. A: Math. 2009. A. in: SIAM Studies in Applied Mathematics. Chakravarty. Y. in: Presented at the Annual Meeting of The Japan Society of Fluid Mechanics. Henderson. A: Math. Kadomtsev. Whitham. N. J. Segur. Kodama / Mathematics and Computers in Simulation 82 (2012) 1185–1218 References [1] [2] [3] [4] [5] [7] [8] [9] [10] [11] [12] [13] [14] [17] [18] [19] [20] [21] [22] [23] [24] [25] M. Phys. Tsuji. Math. P. J. 76 (2007) 84401–84408. Tsuji. Novikov. Sov. Kodama. Phys. Kako. Guyenne. Kao. Oikawa.V. I. Funakoshi. J. Cambridge University Press. 17 (2007) 429–470.Y. Kleiser.G. H. New York. Gen. (2010) in press. R. V.I. John Wiley & Sons. Hirota. preprint. 49 (1980) 2371–2379. Rev. H. Interaction of ionacoustic solitons in twodimensional space. 42 (2009) 312001 (9pp). Theory of Solitons: The Inverse Scattering Method. A: Math. On the stability of solitary waves in weakly dispersive media. L. 1981. Y. Oikawa. Phys. P. September 4–7. J. Sulem. Soliton solutions of the KP equation and application to shallow water waves. EPJ ST. Contemporary Soviet Mathematics. Jpn. S. Soliton solutions of the KP equation with Vshape initial waves. Japan.A. Y. Klein. 206 (2005) 505–535. Craig. Breakdown of Zakharov–Shabat theory and soliton creation.C. Phys. Maruno. Phys. H. Newell. Tsuji. Petviashvili. Phys. 1985. M.C. S. Jpn. Adams. Consultants Bureau. Classiﬁcation of the linesolitons of KPII. J. Stud. D. H. Phys. Yeh. Theor. Wave Motion 42 (2005) 202–210. Young diagrams and Nsoliton solutions of the KP equation. Twodimensional interaction of long solitary waves and web solution. Zakharov. L. 38 (1977) 377–380. Oblique interaction of solitons in an extended Kadomtsev–Petviashvili equation. Reﬂection of obliquely incident solitary waves. Manakov. M. Oikawa. Kodama.L. Jpn. A windowing method for periodic inﬂow/outﬂow boundary treatment of nonperiodic ﬂows. Soc. Pitaevskii. 41 (2008) 275209 (33pp). Schlatter. Cambridge. 49 (1980) 2063–2071. Chakravarty.P. Solitons and the inverse scattering transform. Phys. Appl. H. 37 (2004) 11169–11190. 48. B. G. Y. Hammack.W. Porubov. H. L. Y. P.B. Sparber. C. Linear and Nonlinear Waves. Solitons in Mathematics and Physics. 1974. M. Lavrenov. Resonantly interacting solitary waves. Kodama. M. S. Fluids 18 (2006) 057106. Soc. V.0370). A. M.1218 C. 2004. J. J. Ablowitz. Soc.V. S. Philadelphia. Fluid Mech. SIAM. Formation of the rogue wave due to nonlinear twodimensional waves interaction. Kodama. CBMSNSF. N. 79 (1977) 171–179. Y. New York and London. Mach reﬂection and KP solitons in shallow water (arXiev:1004. Lett. Kodama. Solitary water wave interactions. Phys. A.E. Comput. J. Oikawa. SIAM. W. Tsuji. Numerical study of oscillatory regimes in the kadomtsev–petviashvili equation. Li. Yajima. Redekopp. J. Philadelphia.B. Theor. C. J. K. (2009) 83–151. A WileyInterscience Publication. Oblique interaction of solitary waves. W. Nonlin. J. The Direct Method in Soliton Theory. . C. Miles. M. F.
This action might not be possible to undo. Are you sure you want to continue?
Use one of your book credits to continue reading from where you left off, or restart the preview.