Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Outline:
1. Kortewegde Vries shallow water wave equation (1895).
Solitons waves interacting like particles (1965). Solvable by Inverse Scattering Transform (inverse spectral problem for Schrdinger equation).
2. CamassaHolm shallow water wave equation (1993).
Peakons peak-shaped solitons. Discrete string inverse spectral problem (orthogonal polynomials, Stieltjes continued fractions).
3. DegasperisProcesi equation (1998).
The evil twin of the CamassaHolm equation! Peakons. Discrete cubic string. Shockpeakons discontinuous solitons.
2
cx ct
Zabusky and Kruskal discovered numerically that several solitary waves can interact and yet maintain their identity solitons (1965). Later: Inverse Scattering Transform, explicit formula for the n-soliton solution, innitely many conservation laws, bi-Hamiltonian formulation, Lax pair, etc.
4
2
t
1
t 0
The individual solitons are blurred during the interaction; its hard to tell exactly where they are. If the solitons are nearly equal in size, the two local maxima will not merge into one. Several decompositions and interpretations have been proposed. Does the faster soliton overtake the slower one? Or does it slow down and stay behind? A nice review paper by Benes, Kasman, and Young in J. Nonlinear Sci. 2006 suggests an exchange soliton that transfers energy from the faster soliton to the slower.
MathSciNet says: Among top 10 cited papers 2005 & 2006. In a shallow water wave approximation to higher order than KdV, they obtained this equation:
ut
utxx 3uux
or equivalently mt mx u 2mux
uxx
Remarkable new feature: peakons. C & H derived ODEs which describe the n-peakon solution, 1 and n 2. They also found a and solved the cases n Lax pair for the CH equation.
7
They searched for integrable equations similar to the Camassa Holm equation. In the family
ut
2 uxxt uxxx c0 ux
c1 u c2 ux c3 uuxx x
only KdV, CH, and one new equation satisfy the necessary condition of asymptotic integrability to third order.
utxx 4uux
or equivalently mt mx u 3mux
8
uxx
Antonio Degasperis, Darryl Holm, and Andrew Hone A new integrable equation with peakon solutions Theoretical and Mathematical Physics (2002)
They proved that the DP equation is indeed integrable, by nding a Lax pair and conservation laws. Moreover, the showed that this equation also admits peakon solutions, and solved the ODEs for n 1 and n 2.
uxx ,
3 (DP case).
It admits a particular class of solutions called peakons. A single peakon is a travelling wave of the following shape: u x, t c e xct x This corresponds to m x, t For c 2c xct (Dirac delta).
u x, t
i 1
mi t e x x i t
m x, t
2 mi t xxi t
i 1
u x, t x
m2 t m1 t
1 2
m x, t
x1 t x2 t
m3 t x3 t
11
The n-peakon superposition u mi e xxi is a solution of the b-equation iff the positions xk t and momenta mk t satisfy the following system of ODEs: xk mk
mi e x x
k
i 1
1 mk mi sgnxk xi e x x
k
i 1
b 1 m k u x x k
1 ux xk ux xk : 2
(Note that the speed xk of the kth peakon equals the height of the wave at that point.)
12
n n
1:
x1 m1
m1 0
Travelling wave x1 t
ct, m1 t
c.
x2 and m1 m2.
The integrable cases b 2 (CH peakons) and b 3 (DP peakons) have been solved for arbitrary n using inverse spectral methods. Typical two-peakon interaction (from the CH n formulas that we will see very soon): 2 solution
x Asymptotically (as t ) the peakons separate and behave like free particles (travelling waves).
13
14
CamassaHolm peakons
The solution for n 2 is x1 t x2 t m1 t m2 t log
1
2 b1 2 b2 2 1 logb1 b2 2 b1 2 b2 2 1 1 2 1 b1 2 b2 b1 b2 1 b1 2 b2
2 2 b1 b2
where bk t bk 0et/k . The constants 1 , 2 , b1 0, b2 0 are uniquely determined by initial conditions. (This is the form that the solution takes when one uses inverse spectral methods. Camassa & Holm wrote it a little differently.)
15
The eigenvalues k are real, simple, nonzero. The number of positive eigenvalues equals the number of positive mk s. The quantities bk (residues of the Weyl function) are always positive.
The terminology comes from the inverse spectral solution method, and will be explained a little later. Reference: Richard Beals, David Sattinger, and Jacek Szmigielski Multipeakons and the classical moment problem Advances in Mathematics (2000)
16
j 2 b1 2 b2 2 b3 2 3 1 logb1 b2 b3 j k 2 2 j k 2 b j b k j k
2 2 1 3 2 2 3 2 b1 b2 b3 j k 2 2 j k 2 b j b k j k 2 k j k b j bk k 2 bj bk
3 is
1 2 3 j
j k j
2 b1 2 b2 2 b3 j k j k 2 b j b k 2 3 1 k 2 b j b k 1 b1 2 b2 3 b3 j k j k j b1 b2 b3 1 b1 2 b2 3 b3
The solution for general n looks similar, but to write it down one needs a bit of notation for symmetric functions.
17
(CamassaHolm n
2 continued)
10
x2 t for all t.
CH peakon-antipeakon collision:
1 ( 1
1.5,
1 2
0.86)
m2 )
m2 )
The individual peakon amplitudes m1 t and m2 t both blow up at the instant of collision, one to and the other to , but in such a way that the innities cancel and u mi e x x i remains continuous. (However, ux blows up.)
19
yt yields the strings Separation of variables U y, t vibrational modes via the spectral problem
y 1
z g y y 0 1
for 0
The usual case that we teach our students is when the density gy is constant; then there is an innite sequence of eigenvalues, and the eigenfunctions are sinusoidal. (Think of the harmonics of a guitar string.) Here well consider the opposite extreme: isolated point masses.
20
yi
tanh xi /2
gi
2mi /1
y2 i
(Point masses gi at positions yi connected by massless string.) 1 , . . . , n Such a discrete string has exactly n eigenvalues z and the corresponding eigenfunctions k y are piecewise linear. The quantities bk in the peakon solution formulas are the residues of the (modied) Weyl function: W z z 1; z z 1; z
n 1 bk 2z k 1 z k
In other words, bk is the coupling coefcient k 1/k 1 of the kth eigenfunction, divided by the factor 2 j k 1 k / j . So the bk s encode some information about the shape of the eigenfunctions.
21
Example:
m1 2
m2
m3 x1 y1 y2 y3
2
g1 g2 g3 9.52 6 11.07
x2
x3
3 g3
0.762
0 0.905
g1 y1 y2
g2 y3
1, 1
0.279
0.673
1.08
1 y
2 y
3 y
22
Crucial fact (thanks to the Lax pair associated with the CH eqn): The CH peakons move in such a way that the spectral data of the corresponding discrete string satisfy k 0 bk bk /k
So we determine the spectral data of the string corresponding to the initial peakon conguration (at time t 0), and let it evolve in time as above. Now if we can recover the string data from 0, then we also obtain the the spectral data at a later time t peakon conguration at that time, which is what we seek. This inverse problem of determining the mass distribution of a discrete string given the spectral data was solved long ago (analytic continued fractions T. Stieltjes 1895, string interpretation M. Krein 1951).
23
Here is how Stieltjes continued fractions enter: Let lk yk1 yk . Propagate y; z from the left endpoint 1 using y z gy y. Then is continuous y and piecewise linear, with jumps in the slope where the point masses are. Keeping track of and , one nds at the right endpoint y 1 that
W z z
1; z z 1; z
1 zln 1
gn
zln1 . ..
1 1 1
g2
zl1
1 1 1 g1 zl
0
24
Stieltjes gave formulas for the coefcients in such a continued fraction expansion of a meromorphic function f z, in terms of the coefcients in its expansion f z j 0 1 j A j z j 1 around z . Since we know k and bk in
W z z
n 1 bk 2z k 1 z k
we can expand each term in a geometric series to get the expansion around z . Then Stieltjes formulas give us the coefcients lk and gk in the continued fraction for W z/z. Using this, one obtains explicit formulas for the general peakon solution xk t, mk t for any n.
J. Moser (1975) showed (in the case n 3) how Stieltjes results give the solution of the n-particle nonperiodic Toda lattice. The Toda lattice and the CH peakons are special cases of a more general construction due to Beals SattingerSzmigielski (2001). 25
DegasperisProcesi peakons
The solution for n x1 t x2 t m1 t m2 t 2 is
1
log
2 2 b b 1 2 1 2
1 b1 2 b2 logb1 b2 1 b1 2 b2 2
2 2 1 2 1 b1 2 b2 41 2 bb 1 2 1 2
b1 b2 2
2 2 1 b1 2 b2 41 2 bb 1 2 1 2
with bk t bk 0et/k as before, but with the spectral data now coming from a discrete cubic string instead of an ordinary string.
26
References: Hans Lundmark and Jacek Szmigielski Multi-peakon solutions of the DegasperisProcesi equation Inverse Problems (2003) Hans Lundmark and Jacek Szmigielski DegasperisProcesi peakons and the discrete cubic string International Mathematics Research Papers (2005) Jennifer Kohlenberg, Hans Lundmark, and Jacek Szmigielski The inverse spectral problem for the discrete cubic string Inverse Problems (2007)
27
3 is log U2 V1 x3 t m3 t log U1
U1 2
log
U3 x2 t V2 U3 V2 2 m2 t V3 W2
V1
1
U2 2 V1 2
W2W1
W1
1 b1 2 b2 3 b3
1 2 1 3 2 3 2 2 2 3 3 2 1 1 2 1 2 b1 b2 1 3 b1 b3 2 3 b2 b3 1 2 1 3 2 3 1 2 2 1 3 2 2 3 2 b1 b2 b3 V3 1 2 3 U3 1 2 1 3 2 3
b1 b2
3 2 b b
1 3
3 2 b b
2 3
U2 U3V1
2 2 2 1 b1 2 b2 3 b3
41 2 41 3 42 3 b1 b2 b1 b3 b2 b3 1 2 1 3 2 3 2 41 2 3 1 2 2 1 3 2 b1 b2 b3 1 2 4 ... 1 2 b1 b2 2 1 2 2 1 2 1 3 2 3
28
y 1
z gy y 0 1
for 0
The discrete cubic string associated to a DP peakon conguration n xk , mk has gy 1 gi yi with yi tanh xi 2 gi 8mi 1 y2 2 i
The eigenfunctions are now piecewise quadratic polynomials in y, 0 away from the supsince port of g.
1 2 2 y 1
y y
29
Example:
m1 2
m2
m3 x1
2
g1 g2 g3 90.7 24 245
x2
x3
3 g3
y1 y2 y3
0.762
0 0.905
g1 g2 y1 0.807 y2 y3 3
(different) 2
1, 1
1.20
1 y
2 y
3 y
30
The DP peakons move such that the spectral data of the corresponding discrete cubic string satisfy k 0 bk bk /k
2 j
k 1
k /j .)
The solution formulas for xk t and mk t hence follow from the solution of the inverse problem for the discrete cubic string, which is much more involved than for the ordinary string. Even the forward spectral problem is more complicated, since it is not selfadjoint. (The GantmacherKrein theory of oscillatory kernels shows that the spectrum is positive and simple, at least for positive mass distributions.)
31
(DegasperisProcesi n
t
x2 t for all t.
x t
The solution formulas are only valid up to the time of collision tc since they were derived under the assumption that x1 x2 can be replaced by x2 in the ODEs. Can the solution be continued past the collision? 32
x1
DP peakon-antipeakon collision:
tc .
Why is the DP case different from the CH case? How does the solution continue?
33
References: Giuseppe Coclite and Kenneth Hvistendahl Karlsen On the well-posedness of the DegasperisProcesi equation Journal of Functional Analysis (2006) Hans Lundmark Formation and dynamics of shock waves in the DegasperisProcesi equation Journal of Nonlinear Science (2007)
34
1 Inverting m u uxx as u e x , one 2 G m where G x can formally rewrite the b-equation as a conservation law:
ut x
b u2 1 G 2 u2 3 b u2 x 2 2 2
After multiplying by a test function and integrating by parts, one obtains a rigorous denition of what weak solutions (including peakons) really mean for this family of equations.
1 2 1 u t x 2 u 2 G 3 u2 2
u2 1 G u2 1 u2 2 2 2 x
0 0
(CH, b (DP, b
2) 3)
Since DP does not involve ux explicitly it is reasonable that it also admits solutions where u (and not just ux ) has jumps.
Coclite and Karlsen: For initial data u0 L1 R BV R there is a unique u L R ; L2 R which satises DP (in the above weak sense) together
DP shockpeakons
Here is the unique entropy solution with the shock formed at the DP peakon-antipeakon collision as initial data:
x s sgn x e
ms m m s
m m
s e x
0 0 0
x
36
Natural idea: try superposition! Superposition (solid curve) of two shockpeakons (dashed curves) 3 1 1 with x1 1, s1 1, m2 1 looks 2 , m1 4 and x2 2 , s2 like this:
-3
-2
-1
Plug a shockpeakon superposition Ansatz into the DP eqn and compute, and you will get. . .
37
i 1
mk t Gx xk t sk t G x xk t
i 1
sk
mk
ux xk
i 1
m i G xk xi si G xk xi
i 1 n i 1
u x xk
i 1
m i G xk xi si G xk xi
38
For n
1 we get x1 m1 m1 0 s1
s2 1
which is a shock wave with constant speed (equal to the average height m1 at the jump; cf. RankineHugoniot condition). The jump is u
2s1 where
s1 t
s1 t0 1 t t0 s1 t0
x
39
The totally symmetric DP peakon-antipeakon collision results in a stationary shockpeakon (zero momentum):
x1 Before collision
x2 ,
m1
m2
x
x1 After collision
0,
m1
0,
s1
1/t
40
The only obvious constant of motion is M mk so we are still quite far from nding an explicit solution of the shock-peakon ODEs, even in the case n 2: x1 x2 m1 m2 s1 s2 m1 m2 s 2 R m2 m1 s 1 R 2m1 s1 m2 s2 R 2 m1 s1 m2 s2 R (Assume x1 x2 and set R ex1 x2 )
s 2 s 1 m2 s 2 R 1 s 2 s 2 m1 s 1 R 2
Is this system even integrable? u (The DP Lax pair involves m make sense in this weak setting.)
41
Numerical experiments show: small shocks business as usual, new phenomena appear. A little bit more can be large shocks said in particular cases:
Test case used in CocliteKarlsenRisebro: Numerical schemes for computing discontinuous solutions of the DegasperisProcesi equation (preprint 2006).
42
THE END