Mathematics for web-like patterns of solitary waves in shallow water
Yuji Kodama The Ohio State University
Colloquium Lecture National Cheng Kung University Tainan, Taiwan, December 20, 2018
Real Shallow Water Waves
Mexican beach (by Mark Ablowitz)
Real Shallow Water Waves
Mexican beach (by Mark Ablowitz)
Real Shallow Water Waves
Washington State beach (by Bernard Deconick)
Real Shallow Water Waves
Estonian lake (by Ira Didenkulova)
Shallow Water Wave Equation
Three dimensional water wave profile:
Here we have
λ0=typical wavelength∼horizontal scale h0=asymptotic depth∼vertical scale a0 =typical amplitude∼nonlinear scale We consider an irrotational and incompressible fluid:
~
v =∇φ and ∇ ·~v = ∆φ= 0, whereφis the velocity potential.
Shallow Water Wave Equation
Then the 3-dimensional water wave equation is given by the Euler equation (non-dimensional form)
(a) φzz+β∆⊥φ= 0, for 0<z <1 +αη,
(b) φz= 0, at z = 0,
(c)
φt+1
2α|∇⊥φ|2+ 1 2
α
βφ2z+η= 0, ηt+α∇⊥φ· ∇⊥η= 1
βφz,
at z = 1 +αη,
whereη is the surface elevation,∇⊥= (∂x∂ ,∂y∂ ), ∆⊥=∂x2+∂y2. The parametersα,β are defined by
α:= a0 h0
, β :=
h0 λ0
2
Asymptotic perturbation method
We now introduce a small parameter 0<0:
Small amplitude and long waves α:= a0
h0
∼ β:=
h0 λ0
2
=O().
Examples:
(a) Tohoku tsunami in northern Japan on March 11, 2011:
α= 10m 1000m= 1
100 ∼β=
1km
10km 2
.
(b) Rogue wave observed off South African coast on July, 1909:
α= 20m 200m = 1
10 ∼β =
200m
600m 2
.
(c) Rogue wave observed in Lake Superior on Nov. 1975:
α= 11m 160m = 7
100 ∼β =
160m
600m 2
.
Asymptotic perturbation method
With the assumptionα∼β ∼(i.e. small and long wave), we employ theasymptotic perturbation method to derive the approximate equation:
We first note that the velocity potentialφsatisfying (a) and (b) in the water wave equation can be written in the expansion form,
φ(x,y,z,t) = cos(zp
β∆⊥)ψ(x,y,t)
=ψ−βz2
2∆⊥ψ+O(2).
Recall thatz = 0 implies the bottom, and z = 1 is the asymptotic surface. That is, we have
ψ(x,y,t) =φ(x,y,0,t).
Asymptotic perturbation method
Then from (c) of the water wave equation, the equation for the velocity potentialψ (the Bernoulli equation) becomes
ψt+η+α
2 |∇ψ|2−β
2∆ψt =O(2) and the equation for the surface elevationη is
ηt+ ∆ψ−α∇ ·(ψt∇ψ)−β
6∆2ψ=O(2).
Here we simply write∇⊥=∇and ∆⊥= ∆, and note that those equations arerotationally invariant on thexy-plane.
The set of these equations for (ψ, η) is called theBoussinesq-type equation (1872).
The Benny-Luke equation
Eliminatingη from the Boussinesq-type equation, we have the (regularized)Benney-Lukeequation (1964),
1−β
3∆
ψtt −∆ψ+α(∇ψ· ∇ψt+∇ ·(ψt∇ψ)) =O(2).
This can be reduced to theKorteweg-de Vries (KdV) equation (1895) for a far field with a unidirectional approximation: let χ=Xcos Ψ0+Ysin Ψ0 be the coordinate of the propagation direction and usingψχ=η+O(), we have the KdV equation for the surface elevationH=h0η in the physical coordinates,
HT +c0Hχ+3c0 2h0
HHχ+c0h20
6 Hχχχ= 0.
wherec0 =√
gh0 the velocity of the surface wave (g = 9.8m/sec).
The Benney-Luke equation
The KdV equation has one-soliton solution in the form, H=a0sech2
s3a0 4h30
χ−c0
1 + a0
2h0
T
. An example of interactions of those solitary waves of the Benney-Luke equation:
Here the left figure is the initial wave given by a linear combination of two KdV solitons propagating different directions. The right figure is the solution after some timet>0.
Numerical simulation of the Benney-Luke equation
A simulation of the ion-acoustic wave (similar to the Benney-Luke) equation by Kako and Yajima (1982):
Numerical simulation of the Benney-Luke equation
Notice the change of interaction, and astem wave starts to develop when the interaction angle takes acriticalvalue.
Waves in a pool
Instantgeneration of those waves with stems in a pool:
Can be alaptop experiment for exotic wave patterns.
Asymptotic perturbation method continued
We make a further assumption:
Quasi-two dimensionality:
ζ :=√
γy, γ =O().
That is, we assume a weak dependence in they-direction. Also note that thisbreaks the rotational symmetry. Then we have
ψt+η+α
2ψ2x+β
2ηxx =O(2) ηt+ψxx−α(ψtψx)x−β
6ψxxxx +γψζζ =O(2).
Asymptotic perturbation method continued
We then consider afar fieldwith the moving coordinate, Unidirectional approximation:
ξ:=x−t, τ :=t.
Then eliminateη and define v :=ψξ, we have 2vτ ξ+ 3α(vvξ)ξ+β
3vξξξξ+γvζζ =O(2) Introduce thenew(scaled) variables (x,y,t,u) defined by
ξ=p
βx, ζ=p
βγy, τ =−3 2p
βt, v = 2 3αu.
The Kadomtsev-Petviashvili (KP) equation (1970) (−4ut+ 6uux+uxxx)x+ 3uyy = 0
The KP equation
We write the solutionu(x,y,t) in the form with theτ-function u(x,y,t) = 2 ∂2
∂x2lnτ(x,y,t),
Then theτ-function satisfies a bilinear equation, called Hirota bilinear form,
−4(τ τxt −τxτt) + (τ τxxxx −4τxτxxx+ 3τxx2 ) + 3(τ τyy−τy2) = 0.
This can be written in the following nice form,
P(Dx,Dy,Dt)τ◦τ := [Dx(−4Dt+Dx3) + 3Dy2]τ◦τ = 0.
whereDzn’s are the Hirota (skew) derivatives defined by Dznf ◦g =
∂
∂z − ∂
∂z0 n
f(z)g(z0) z=z0.
The KP equation
Remark: The KP bilinear operator gives thedispersion relation, P(kx,ky, ω) =−4kxω+kx4+ 3ky3= 0.
This can be parametrized by
kx =κi−κj, ky =κ2i −κ2j, ω =κ3i −κ3j, whereκi’s are arbitrary constants. From this, one can find a solution ofτ in the sum of exponentials,
τ =a1eθ1+a2eθ2+. . .+aMeθM with θi =κix+κ2iy+κ3it,
whereκi’s are arbitrary constants (i.e. P(Dx,Dy,Dt)eθi ◦eθj = 0).
Note here that theτ-function satisfies thelinear equations τy =τxx, τt =τxxx.
The KP equation
In general, we writeτ(x,y,t) in theWronskian form:
τ = Wr(f1,f2, . . . ,fN) =
f1(0) f1(1) · · · f1(N−1) f2(0) f2(1) · · · f2(N−1)
... ... . .. ... fN(0) fN(1) · · · fN(N−1)
,
wherefi’s are arbitrary smooth functions of the variables{x,y,t}, andfj(n)= ∂nfj
∂xn for n= 0,1, . . . ,N−1.
Then we have the following Proposition for the solution of the KP equation:
The KP equation
Proposition
The Wronskian form provides a solution of the KP equation, if those{fi(x,y,t) :i = 1, . . . ,N} satisfy thelinear equations,
∂fi
∂y = ∂2fi
∂x2, ∂fi
∂t = ∂3fi
∂x3.
Then considerfinite dimensional solutions (finite Fourier series):
fi(x,y,t) =
M
X
j=1
aijeθj(x,y,t), i = 1, . . . ,N<M, where θj(x,y,t) :=κjx+κ2jy+κ3jt, j = 1, . . . ,M.
This solutionu(x,y,t) of the KP equation is then completely determined by theκ-parameters and theN×M matrixA= (aij).
The KP equation
Mathematical remark: We have a picture of theGrassmannian Gr(N,M), which is defined as follows:
SpanR{E~j = (1, κj, . . . , κM−1j )Teθj :j = 1, . . . ,M} ∼=RM. SpanR{~fi :i = 1, . . . ,N} forms an N-dimensional subspace in RM,
f~i =
M
X
j=1
ai,jE~j for i = 1, . . . ,N, where the coefficients {ai,j}define the N×M matrix
A=
a11 · · · a1M ... . .. ... ... aN1 · · · aNM
∈MN×M(R), with full rank, i.e. rank(A) =N.
The KP equation
For anyH ∈GLN(R), theτ-function with the new base (~g1, . . . , ~gN) = (~f1, . . . , ~fN)H gives thesame solution, since Wr(g1, . . . ,gN) = det(H)Wr(f1, . . . ,fN). This implies that the τ-function can be identified as a point of the Grassmannian, Gr(N,M), i.e.
Gr(N,M)∼= GLN(R)\MN×M(R), with dim Gr(N,M) =NM−N2=N(M−N).
H ∈GLN(R) gives a row reduction of theA-matrix. For example, a genericA can be written in the reduced row echelon form (RREF),
A=
1 · · · 0 ∗ · · · ∗ ... . .. ... ... ... ...
0 · · · 1 ∗ · · · ∗
Here 1’s are called ”pivot” ones.
One-soliton and Y-soliton
Example 1: N= 1, M = 2. This gives oneline-soliton solution:
WithA= (1a)∈Gr(1,2), the τ-function is τ =eθ1+aeθ2 =eθ1+eθ˜2 =e12(θ1+˜θ2)
e12(θ1−θ˜2)+e−12(θ1−θ˜2)
= 2e12(θ1+˜θ2)cosh1
2(θ1−θ˜2), whereθi’s are given by
θ1=κ1x+κ21y+κ31t, θ˜2 =κ2x+κ22y+κ32t+ lna Then we have
u = 2 ∂2
∂x2lnτ = 1
2(κ1−κ2)2sech21
2(θ1−θ˜2) Denote this solution by[1,2]-soliton. In general, [i,j]-soliton.
One-soliton and Y-soliton
3D figure of the solutionuA, and the contour plot. Numbers(i) represent thedominantexponential term eθj’s withκ1< κ2 .
The line of the wave crest is given byθ1 = ˜θ2 =θ2+ lna, i.e.
x+ (κ1+κ2)y+ (κ21+κ22+κ1κ2)t− 1
κ1−κ2 lna= 0.
One-soliton and Y-soliton
Example 2: N= 1 andM = 3, i.e. A= (1a b)∈Gr(1,3). The τ-function is
τ =eθ1+aeθ2+beθ3, with θi =κix+κ2iy+κ3it.
There are three distinct regions in thexy-plane, and for a fixedt, in each region one of the exponentialeθj dominates. Then the solution has a Y-shape pattern (calledY-type solution):
One-soliton and Y-soliton
As we noted that each line-soliton exists as abalance between two dominant exponentials, and crossing the soliton, the index set of the dominant exponential changes. The following diagram, called chord diagram, represents this change:
One-soliton with {i,j}, i.e. [i,j]-soliton:
ki kj π = i j j i
Y-solitons with {i,j,l}, i.e. [i,j]-, [j,l]- and [i,l]-solitons:
ki kj 0 kl
ki kj 0 kl
These give the permutations π= i j ll i j
andπ= i j lj l i .
Classification Theorem
Now weclassify the soliton solutions generated by the Wronskian form,
τ = Wr(f1,f2, . . . ,fN).
Usingfi =PM
j=1ai,jeθj, this can be expressed as,
τ = Wr(f1,f2, . . . ,fN) =
f1(0) f1(1) · · · f1(N−1) f2(0) f2(1) · · · f2(N−1)
... ... . .. ... fN(0) fN(1) · · · fN(N−1)
=
a1,1 a1,2 · · · a1,M ... ... . .. ... aN,1 aN,2 · · · aN,M
eθ1 κ1eθ1 · · · κN−11 eθ1 eθ2 κ2eθ2 · · · κN−12 eθ2
... . .. ...
eθM κMeθM · · · κN−1M eθM
Classification Theorem
Then using theBinet-Cauchy lemma, we have:
Lemma (the expansion formula of theτ-function) Theτ-function can be expanded as
τ(x,y,t) = X
I={i1<···<iN}
∆I(A)EI(x,y,t),
where∆I(A)is the N ×N minor of the A-matrix with the set of columns I ={i1, . . . ,iN} ⊂ {1,2, . . . ,M}, and
EI = Wr(eθi1. . . ,eθiN) =
Y
1≤j<l≤N
(κij −κil)
eθi1+···+θiN >0, where we have assumed the orderκ1 < κ2<· · ·< κM.
Classification Theorem
Remark: We are interested in the case where the matrixA has
∆I(A)≥0 for all I ={i1<i2 <· · ·<iN}. Such matrix is called totally non-negativematrix. Then the set of all such matrices defines the totally non-negative Grassmannian:
Totally nonnegative Grassmannian
Gr+(N,M):={A∈Gr(N,M) : ∆I(A)≥0,∀I ={i1< . . . <iN}}
Then we have the following theorem:
Theorem (K. -Williams, 2013) The solution u(x,y,t) = 2 ∂2
∂x2lnτ is non-singularif and only if A∈Gr+(N,M).
Classification Theorem
IrreducibleAmatrix
We say that theA-matrix isirreducible, if
in each column, there is at least one nonzero element, in each raw, there is at least one more nonzero element in addition to the pivot.
Example: ForN = 2 andM = 4, there are only two types of irreducibleA-matrices in RREF:
1 0 ∗ ∗ 0 1 ∗ ∗
,
1 ∗ 0 ∗ 0 0 1 ∗
.
Note that other cases can be expressed by a smaller matrix of N0×M0 with either N0<N or M0 <M.
Classification Theorem
Example: ForN = 2 andM = 4, there are totally seven types of theA-matrices in RREF which are bothirreducible andtotally non-negative:
1 0 −c −d
0 1 a b
1 0 −b −c
0 1 a 0
1 0 0 −c
0 1 a b
1 0 0 −b
0 1 a 0
1 a 0 −c
0 0 1 b
1 a 0 0 0 0 1 b
Herea,b,c and d are positive numbers, and for the first one, eitherad −cb>0 or = 0. The total number of nonzero minors
∆i,j(A) is at least four, and the maximal number is six. Note that there aresevenderangements (permutations without fixed points) of the symmetric groupS4 with two excedances.
Classification Theorem
LetAbe an N×M irreducibleand totally non-negativematrix, and let{e1, . . . ,eN} be the pivot indices and {g1, . . . ,gM−N} be thenon-pivotindices. Then we have:
Theorem (Chakravarty-K. 2008)
The soliton solution associated with the A-matrix has (a) for y 0,∃ N solitons of [en,jn]-type for some jn, (b) for y 0,∃ M−N solitons of[im,gm]-type for some im.
Interaction Region (e ,e ,...,e )1 2 N+
[e , j ]N+ N+
x = ∞ x = - ∞
y = ∞
y = - ∞
[e , j ] j j [e , j ] k k
[e , j ]1 1
[ i , g ]N- N-
[ i , g ]1 1
[ i , g ]s s
Classification Theorem
Theorem (Chakravarty-K. 2008)
The set of those solitons[en,jn]and [im,gm]are expressed by a uniquechord diagram which corresponds to aderangementof the symmetric groupSM with N excedances, i.e.
e1 · · · eN g1 · · · gM−N
j1 · · · jN i1 · · · iM−N
∈ SM
Theorem (K.-Williams, 2014)
Conversely, for each chord diagram associated with the derangement, one can reconstruct an A-matrix, and the correspondingτ-function gives the solution of the KP equation having line-solitons expressed by the chord diagram. The entries of the A-matrix give thescattering data, i.e. the locations of those line-solitons and their interaction properties.
Classification Theorem
Example: N= 2,M = 4. We have sevendifferent types of
(2,2)-soliton solution, which are parametrized by the derangements ofS4:
(3412)
(4321)
(2143)
(4312) (3421)
(3142) (2413)
The 4-tuples of the diagrams represent the derangements, π=
1 2 3 4
π(1) π(2) π(3) π(4)
= (π(1), . . . , π(4)).
List of (2, 2)-solitons
The caseπ= (3412): The A-matrix is given by
A=
1 0 −c −d
0 1 a b
The solution contains all possible line-solitons, that is, for any 1≤i <j ≤4, there aresix different solitons of [i,j]-types. In particular, the asymptotic line-solitons are [1,3]- and [2,4]-types.
List of (2, 2)-solitons
The caseπ= (4312): The A-matrix is given by
A=
1 0 −c −d
0 1 a 0
1 2 3 π = (4312)
4
The asymptotic line-solitons are (a) for y 0, [1,4]- and [2,3]-types, (b) for y 0, [1,3]- and [2,4]-types.
List of (2, 2)-solitons
The caseπ= (2413): The A-matrix is given by
A=
1 0 −c −d
0 1 a b
1
2 3
π = (2413) 4
wheread−bc = 0.
The asymptotic line-solitons are (a) for y 0, [1,2]- and [2,4]-types, (b) for y 0, [1,3]- and [3,4]-types.
List of (2, 2)-solitons
The caseπ= (4321): The A-matrix is given by
A=
1 0 0 −d
0 1 a 0
There is no resonant interaction, and this solution is called P-type.
The asymptotic line-solitons are of [1,4]- and [2,4]-types.
List of (2, 2)-solitons
The caseπ= (2143): The A-matrix is given by A=
1 a 0 0 0 0 1 b
There is no resonant interaction, and this solution is called O-type.
The asymptotic line-solitons are of [1,2]- and [3,4]-types.
List of (2, 2)-solitons
Gr(3,6) Example: (9-dimensional solution).
-100 -50 0 50 100
-100 -50 0 50 100
-100 -50 0 50 100
-100 -50 0 50 100
-100 -50 0 50 100
-100 -50 0 50 100
[2,5]
[3,6]
[1,4]
1 0 0 g h k A = 0 1 0 -d -e -f 0 0 1 a b c
1 2 3 4 5 6
t = -10 t = 0 t = 10
π = (456123) = [14][25][36]
List of (2, 2)-solitons
Other example inGr(3,6): (7-dimensional solution).
-100 -50 0 50 100
-100 -50 0 50 100
-100 -50 0 50 100
-100 -50 0 50 100
-100 -50 0 50 100
-100 -50 0 50 100
[5,6]
[2,5]
[1,4]
[3,6]
[2,4]
[1,3]
1 0 -a -b 0 c A = 0 1 d e 0 -f 0 0 0 0 1 g
1 2 3 4
5
6
t = -30 t = 0 t = 30
π = (451263)
Mach reflection phenomena in shallow water
Incident line-soliton propagating into an inclined wall. The right panel is an equivalent system, and illustrates a typical wave pattern after colliding the inclined wall. It turns out that this can be described by one of the exact solutions of the KP equation.
Mach reflection phenomena in shallow water
Harry Yeh’s experiment of the Mach reflection with Ψ0<Ψc:
The wave is similar to a (3142)-type soliton solution. Notice in particular that the stem length is increasing, i.e. the Mach stem.
Mach reflection phenomena in shallow water
Appearance of Mach-stem is an “everyday” occurrence.
KP solitons in Mach reflection
We perform anumerical simulation with V-shape initial wave profile: First recall that one-line soliton of [i,j]-type,
u(x,y,t) =A[i,j]sech2 pA[i,j]
2 (x+ytan Ψ[i,j]+C[i,j]t), where the amplitudeA[i,j] and the slope tan Ψ[i,j] are given by
A[i,j]= 1
2(κi −κj)2, tan Ψ[i,j]=κi+κj. We consider asymmetric V-shape initial data as shown in
0 0
−Ψ0
Ψ0
Solitons fory → ±∞have A0 = 12(ki±−kj±)2 tan Ψ0 =∓(ki±+kj±)
KP solitons in Mach reflection
Numerical result for the case (a), Ψ0 >Ψc (regular reflection):
0 2 4 6
0 0.1 0.2 0.3 0.4
t
E(t)
KP solitons in Mach reflection
3-D figure of the (2143)- (i.e. O-)type completion: The solution consists of [1,2]-soliton for y >0, and [3,4]-soliton for y <0.
Notice that the maximum amplitude is achieved rather quickly, and the interaction point becomes steady to give aconstant phase shift. Also the bow-shape wakes start to leave the main part of the interaction point, and this can be considered to be aseparation of thesoliton from the radiations .
KP solitons in Mach reflection
Numerical result for the case (b), 0<Ψ0<Ψc:
[1,3]
[2,4]
[3,4]
[1,2]
[1,4]
0 5 10
0 0.02 0.04 0.06
t
E(t)
KP solitons in Mach reflection
3-D figure of the (3142)-type completion.
Notice that the amplitude is slowly growing to the maximum, and the interaction part generates a stem wave,Mach stem. Again the bow-shape wakes start to leave the main part of the interaction point, and this can be considered to be a separation of thesoliton and theradiations .
Other examples
Example 1: Waves in an Estonian beach (the photo (video) courtesy by Ira Didenkulova):
Here we estimate the angles,
Ψ1 = Ψ[3,4] = 37◦, Ψ2= Ψ[2,4]= 14◦, Ψ3 = Ψ[2,3]= 0◦, Ψ4 = Ψ[1,3] =−20◦, Ψ5 = Ψ[1,2]=−42◦.
which gives
(κ1, κ2, κ3, κ4) = (−0.632,−0.268,0.268,0.517)
Other examples
Then we get the KP solution which approximates the video:
The video on the third line is another seen of the same kind.
Other examples
Example 2: The KP exact solution is of π= (2,4,1,5,3).
The photo is taken by Ablowitz at an Mexican beach.
References (mathematical physics and engineering)
1. Y. K., Young diagrams and N-soliton solutions of the KP equation, J. Phys. A: Math. Gen. 37(2004) 11169-11190.
2. S. Chakravarty and Y. K., Classification of the line-solitons of KPII, J. Phys. A: Math. Theor. 41 (2008) 275209 (33pp).
3. S. Chakravarty and Y. K., Soliton solutions of the KP equation and application to shallow water waves, Stud. in Appl. Math. 123 (2009) 83-151.
4. Y. K, KP solitons in shallow water, J. Phys. A: Math. Theor.
43 (2010) 434004 (54pp).
5. W. Li, H. Yeh and Y. K., On the Mach reflection of a solitary wave: revisited, J. Fluid Mech. 672(2011) 326-357.
6. H. Yeh, W. Li and Y.K.,Mach reflection and KP solitons in shallow water, Eur. Phys. J. Special Topics 185 (2010) 97-111.
References (mathematics)
7. Y. K. and L. Williams,KP solitons, total positivity, and cluster algebras, PNAS,108 (2011) 8984-8989 (arXiv:1105.4170).
8. Y. K. and L. Williams, KP solitons and total positivity for the Grassmannian, Invent. Math. 195(2014) (arXiv:1106.0023).
9. Y. K. and L. Williams, The Deodhar decomposition of the Grassmannian and the regularity of KP solitons, Adv. Math.
244 (2013) 979-1032 (arXiv:1204.6446).
10. Y. K. KP solitons and the Grassmannians: Springer Brief of Mathematical Physics vol.22, Springer (2017).
11. Y. K. Solitons in Two-Dimensional Shallow Water, SIAM CBMS-NSF Series 92(2018).