Doktorov E.V., Leble S.B. A Dressing Method in Mathematical Physics

Подождите немного. Документ загружается.

2.4 Darboux transformation. Generalized Burgers equations 39

The explicit expression for

L can b e o btained in terms of (2.32) and (2.16)

and has the form

L = a



n=1

′

n−1

+ a

− sa

n−1

). (2.35)

Let us write (2.33) explicitly using (2.16). It is established that for the

intertwining relation (2.34) to be valid, it is necessary and suﬃcient that s be

a solution of the equation

s =



n=0

′

(s)+a

n+1

(s) − sa

(s)]. (2.36)

Remark 2.17. Equation (2.36) is nonlinear but linearizable. This equation (in

a diﬀerent form) was introduced in [388]. The form we suggest here is the most

compact and convenient for further investigations, e.g., in the framework of

the bilineraization technique of Hirota [210].

In the case of scalar functions and L = D

equation (2.36) is known as the

Burgers equation. For this reason and owing to the integrability of (2.36) by

the Cole–Hopf transformation, it is natural to refer to (2.36) as a generalized

Burgers equation.

Proposition 2.18. Suppose an invertible function ϕ is a solution to the linear

diﬀerential equation

ϕ = Lϕ.

Then the function s satisﬁes the generalized Burgers equation (2.36).

The obvious corollary of the intertwining relation (2.34) and Proposition 2.18

is as follows:

Theorem 2.19. Let functions ψ and ϕ be solutions of the equations

ψ = Lψ, D

ϕ = Lϕ (2.37)

for an invertible function ϕ. Then the function

ψ = L

ψ = Dψ − sψ, s =(Dϕ)ϕ

−1

(2.38)

is a solution of the equation

ψ =

ψ. (2.39)

The last statement accomplishes the proof of the Matveev theorem for

diﬀerential polynomials [314] in its non-Ab elian version.

40 2 Factorization and classical Darboux transformations

The equality (2.35) gives a representation of the transformed operator in

terms of the generalized Bell polynomials. The explicit expressions for the

transformed coeﬃcients are

[1] = a

, (2.40)

[1] = a



n=k+1

n,n−k

+(a

′

− sa

n−1,n−1−k

], (2.41)

k =0,...,N − 1.

2.5 Iterations and quasideterminants

via Darboux transformation

Here we would like to revisit the non-Abelian iterated DT formulas following

the ideas of the pioneering paper of Matveev [313], where the basic formulas

were derived. Their Abelian counterpart is demonstrated in [324] and dis-

cussed also in [316, 322]. In fact, this approach goes back to the famous paper

of Crum [94]. We will see, in th e framework of a general non-Abelian DT

theory, that the dressing procedure naturally produces the quasideterminants

(Sect. 1.9). In the paper [191] this procedure is also properly analyzed for the

matrix Schr¨odinger operator.

2.5.1 General s tatements

Let R b e a diﬀerential algebra with a derivation D : R → R and φ ∈ R be

an invertible element. Recall that we denote D(g)=g

′

and D

(g)=g

(k)

.In

particular, D

(0)

(g)=g.

For ψ ∈ R deﬁne D(φ; ψ)=ψ

′

−φ

′

−1

ψ. Following [321], we call D(φ; ψ)

the DT of ψ deﬁned by φ.

Theorem 2.20. Let φ

,...,φ

∈ R. Deﬁne by induction the iterated DT

D(φ

,...φ

; ψ) as follows. For N =1, it coincides with the DT deﬁned above.

Assume N>1. The expression D(φ

,...,φ

; ψ) is deﬁned if D(φ

,...,φ

; ψ)

is deﬁned and invertible and D(φ

; ψ) is deﬁned. In this case,

D(φ

,...φ

; ψ)=D[D(φ

,...φ

; ψ); D(φ

; ψ)].

Theorem 2.21. If all square submatrices of matrix (φ

(j)

), i =1,...,N,

j = k − 1,...,0 are invertible, then the Vandermond supermatrix deﬁnes the

quasideterminant:



D(φ

,...,φ

; ψ)=



(k)

...φ

(k)

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

ψφ

... φ



2.5 Iterations and quasideterminants via Darboux transformation 41

Recall that we use the “hat” symbol to denote the quasideterminant. This

general statement ﬁrst appeared in [313].

Proof. Iterations of the DT yield

ψ[N]=ψ

(N)

N−1



m=0

(m)

, (2.42)

which is a replica of an expression in [313, 322, 324] for the Abelian case.

The iterated DT (2.42) as a function of ψ sends to zero any φ

on which the

transformation is constructed:

[N]=φ

(N)

N−1



m=0

(m)

=0,p=1,...,N. (2.43)

The successive excluding of s

as a function of the derivatives ψ

(m)

from the

system (2.43) yields the algorithm that results in the evaluation of s

;the

procedure was p ointed out already in [314], applied in [277] and, as it is

seen from a comparison with the quasideterminant deﬁnition, could deﬁne

the Vandermond quasideterminant.

Let us illustrate the scheme with the case of N = 2. The ﬁrst iteration is

basedonasetofφ

, p =1, 2. The equations for s

, i =0, 1

(2)

+ s

′

=0,φ

(2)

+ s

′

= 0 (2.44)

yield

= −φ

(2)

−1

− s

′

−1

Inserting this into the second relation of (2.44) pro duces the equation for s

(φ

′

− φ

′

−1

)=−φ

(2)

+ φ

(2)

−1

It is solved as

=(−φ

(2)

+ φ

(2)

−1

)(φ

′

− φ

′

−1

)

−1

and

= −φ

(2)

−1

− (−φ

(2)

+ φ

(2)

−1

)(φ

′

− φ

′

−1

)

−1

′

−1

bo th recognized as quasideterminants. The ﬁnal expression for ψ[2] is given

by (2.42).

As mentioned in [322], the comparison of the resulting formula for ψ[N ]

(2.42) and the formula for the DT

D(φ

N−1

,...,φ

; ψ[N −1]) = ψ[N −1]

′

−φ[N −1]

′

φ[N −1]

−1

ψ[N −1], (2.45)

where φ[N − 1] = ψ[N − 1]|

ψ=φ

, yields the non-Abelian Jacobi identity for

quasi-Wronskians “for free.” Recall that Crum [94] used the Jacobi identity

to prove the determinant formulas for the iterated DT for Abelian entries.

42 2 Factorization and classical Darboux transformations

Remark 2.22. The non-Abelian algorithm to exclude s

s hin ts at the deﬁni-

tion of quasideterminants as a function of submatrices a

(compare with the

results in Sect. 1.9.1). Namely, it is enough to change ψ

(m)

→ a

The solution of the system (2.43) with respect to s

may be reinterpreted

as the Vandermond-quasideterminant representation of the iterated DT for

solutions (2.42) (with inserted s

) and, next, linked to the DT for the poten-

tials a

[N].

To check this proposition, let us substitute (2.42) into the evolution equa-

tion (2.37) for ψ[N ] (see Theorem 2.19):

ψ[N]



k=0

a[N]

(N+k)



k=0

a[N]

N−1



m=0

(m)

)

(k)

. (2.46)

On the other hand,

ψ[N]

= ψ

(N)

N−1



m=0

m,t

(m)

+ s

(m)

) (2.47)



k=0

(k)

)

(N)

N−1



m=0

m,t

(m)

N−1



m=0



k=0

(k)

)

(m)

Equating terms with the highest derivative ψ

(N+n)

gives

a[N]

= a

and, subsequently, for ψ

(N+n−1)

produces

a[N]

n−1

= a

n−1

+ Na

′

+ s

N−1

− a

N−1

For ψ

(N+n−2)

we obtain

a[N]

n−2

= a

n−2

+ Na

′

n−1

N(N +1)

′′

− a

N−2

+ ns

′

N−1

) (2.48)

−a[N]

n−1

N−1

+(N − 1)s

N−1

′

+ s

N−1a

n−1

+ s

N−2

preserving the order of diﬀerentiation in (2.47) to keep the non-Abelian char-

acter. Substituting here a[N ]

n−1

yields the explicit form of a[N ]

n−2

a[N]

n−2

= a

n−2

+ Na

′

n−1

N(N +1)

′′

+[s

N−1

n−1

]+[s

N−2

]

−na

′

N−1

− (Na

′

+[s

N−1

])s

N−1

. (2.49)

One could compare the resulting expression (2.49) for commuting entries and

for a

′

= a

′

n−1

=0,

a[N]

n−2

= a

n−2

− na

′

N−1

, (2.50)

with (2.37) [322] when taking into account that s

N−1

= −ln

W (ϕ

,...,ϕ

with ϕ

being solutions of (2.38).

2.5 Iterations and quasideterminants via Darboux transformation 43

Corollary 2.23. [321] In the commutative case, the iterated Darboux trans-

formation is a ratio of two Wronskians, as follows by direct applic ation of the

Kramer rule to (2.43):

D(φ

,...,φ

; ψ)=

W (φ

,...φ

,ψ)

W (φ

,...,φ

)

. (2.51)

All these results are naturally generalized to the cases when the main prop-

erties of the DTs are valid: the n-iterated transform is a linear function of

ψ and there is an n-dimensional kernel of the transformation operator.

This remark relates ﬁrst of all to the next section (see also [3 21]) and to

the Moutard/Goursat transformations (Chap. 6). The algorithm of the con-

struction given here is easily transferred to the iterated Moutard/Gou rsat

transformation because the “kernel property” (2.43) is also valid.

2.5.2 Positons

An interesting illustration of the application of (2.51) is concerned with posi-

tons. Positons were introduced by Matveev [318, 319] as a class of singular

solutions of the KdV equation,

− 6uu

+ u

xxx

=0, (2.52)

that lead to a trivial scattering matrix for the asso ciated spectral problem

−ψ

+ uψ = λψ. (2.53)

Here we consider this topic in more detail, following [323].

The KdV equation can be written as the compatibility condition [13] of

the linear system of equations comprising the spectral problem (2.53) and the

evolutionary equation

= −4ψ

xxx

+6uψ

+3u

ψ. (2.54)

Note that the spectral problem (2.53) is a representative of the general equa-

tion (2.37).

Let φ(λ) solve the spectral equation (2.53). Diﬀerentiation in λ produces

(in general, linearly independent) solutions φ

[m]

= ∂

φ(λ)/∂λ

of the same

equation. The set of solutions φ

,...,φ

]

,φ

,...,φ

]

,...,φ

]

are integers, generated by the λ-derivatives in the points λ

, λ

,...,λ

yields the iterated DT, which is the Abelian speciﬁcation of the transform

(2.42) and the quasideterminant formula (2.43).

Proposition 2.24. Let u(x, t) be a solution of the KdV equation (2.52). The

Wronskians

= W (φ

,...,φ

]

,φ

,...,φ

]

,...,φ

]

)

44 2 Factorization and classical Darboux transformations

and

= W (φ

,...,φ

]

,φ

,...,φ

]

,...,φ

]

,ψ)

produce the DT

u[N]=u − 2 ∂

ln W

,ψ[N]=W

. (2.55)

The Lax pair equations (2.53) and (2.54) are covariant with respect to the

DT (2.55). In other words, the function ψ[N] satisﬁes (2.53) and (2.54) with

u → u[N] and ψ → ψ[N] and u[N ] is a new solution of the KdV equation.

Now we apply the DT (2.55) to dress the simplest seed solu ti on u =0

with the choice ψ =exp(ikx +4ik

t), k

= λ. As a solution of the spectral

equation with zero potential we choose an oscillating function

φ =sinκ(x + x

(κ)+4κ

t) ≡ sin θ.

Here κ is a real parameter and x

(κ) is an analytic function in the vicinity of

the point κ. Therefore, a new solution to the KdV equation is given by

u[1] = −2∂

ln W (φ, ∂

φ) (2.56)

and is written explicitly as

u =32κ

sin θ − κγ cos θ

(sin 2θ − 2κγ)

sin θ, (2.57)

where

γ = ∂

θ = x + x

+12κ

t, x

= x

+ κ∂

,W(φ, ∂

φ)=sin2θ − 2κγ.

(2.58)

The solution (2.57) is determined by three real parameters x

, x

,andκ

and has a second-order p ole in x. The precise pole position is found by solu-

tion of the nonlinear functional equation W (φ, ∂

φ) = 0. The corresponding

solution of the Lax pair [or of (2.51)] takes the form

ψ(x, k)=

W [φ, ∂

φ, exp(ikx +4ik

t)]

W (φ, ∂

φ)

(2.59)



−k

4ikxsin

sin 2θ − 2κγ

− κ

sin 2θ +2κγ

sin 2θ − 2κγ



ikx+4ik

In the point k = κ, this solution is simpliﬁed:

ψ(k, x)=−4κ

sin θ

sin 2θ − 2κγ

. (2.60)

We see from (2.60) that the function ψ is localized near its pole but is

not square-integrable on the whole x-axis. The point κ

is called the Wigner–

von Neuman resonance [320].

Because κ

> 0, the solution (2.57) is called

positon, as distinct from the soliton solution fo r which κ

< 0 (Sect. 8.7).

Generic asp ects of the scattering theory of the potentials leading to the Wigner–

von Neumann resonances are discussed in [312, 311].

2.6 Darboux transformations at associative ring with automorphism 45

Asymptotic behavior of the function ψ is given by

ψ → (−k

+ κ

ikx+4ik

[1 + o(1)],x→±∞. (2.61)

Let us compare (2.61) with the standard Jost solution J(x, k, t) asymptotic

of the linear Schr¨odinger equation with a decreasing potential:

J(x, k, t) → e

ikx

(1 + o(1)),x→ +∞,

J(x, k, t) → a(k, t)e

ikx

+ b(k, t)e

−ikx

,x→−∞,

where a(k, t)andb(k, t) are the transmission and reﬂection coeﬃcients. For

the positon p otential we obtain

a(k, t)=1,b(k, t)=0.

Potentials for which b(k, t) = 0 are called reﬂectionless. The well-known exam-

ple of the reﬂectionless potentials is provided by solitons. However, for solitons

we have a(k,t) = 1. Hence, positons give a unique example of supertranspar-

ent (or superreﬂectionless) lo n g-range potentials.

A two-positon solution is generated by the evident extension of (2.56),

u = −2∂

ln W (φ

,∂

,φ

,∂

with

=sinκ

(x + x

+4κ

t),φ

=sin(x + x

+4κ

t),

and is determined by six real parameters. For x →∞the two-positon solution

is decomp osed into a sum of two free positons. It should be stressed that the

positon scattering is not accompanied with a phase shift typical for the soliton

scattering.

Interesting suggestions concerni n g physical applications of positons can by

found in the paper by Matveev [323].

2.6 Darboux transformations at associative

ring with automorphism

In this section we reformulate and analyze the results from the paper of

Matveev [321] for further use in the derivation of chain equations and joint

covariance of operator pairs [265, 267, 271]. We begin with general notations.

Let R be an associative ring with an auto morphism, implying that there exists

For singular potentials the scattering data are not uniquely deﬁned. Diﬀerent

self-adjoint extensions of the same diﬀerential operators might lead to diﬀerent

scattering operators. The deﬁnition of the scattering coeﬃcients given above is in

agreement with the nonlinear picture of interaction between positons and solitons,

although the latter can be analyzed independently of this deﬁnition.

46 2 Factorization and classical Darboux transformations

a linear invertible map T, R→R such that for any ψ (x, t)andϕ (x, t) ∈ R,

x ∈ R

, t ∈ R we have

T (ψϕ)=T (ψ)T (ϕ),T(1) = 1. (2.62)

The automorphism with the deﬁning property (2.62) allows us to write down

a wide class of functional-diﬀerential-diﬀerence and diﬀerence-diﬀerence equa-

tions starting from

(x, t)=



m=−M

ψ, (2.63)

where M and N are integers. For example, the op erator T canbechosenas

Tψ(x, t)=ψ(qx + δ, t),

where q ∈ GL(n, C), δ ∈ R

. Another choice gives

Tψ(x)=Wψ(x)W

−1

,W∈ GL(n, C).

We will save the notations and conditions of the paper [321] discussing other

potentials until the end of Sect. 4.9.

Let us consider two DTs for solutions of (2.63),

f = f − σ

±1

f, σ

= ϕ



±1



−1

, (2.64)

where ϕ is a particular solution of the same equation (2.63). For the case of

a diﬀerential ring and for Tf(x, t)=f (x + δ, t), x, δ ∈ R the limit ∂f =

lim

δ→0

(T − 1)f (x, t) gives the link to the classical DT.

To derive the DT of potentials U

, it is necessary to evaluate the deriva-

tive of the elements σ

with respect to the variable t (say, time). We shall do

it by introducing the special functions (analog of the diﬀerential Bell polyno-

mials), similar to [467]. Let us start from the ﬁrst version of the DT deﬁnition

, expressing Tϕ from (2.64) ; hence, Tϕ =(σ

)

−1

ϕ. Acting on this re-

lation by T and taking into account (2.62) yields T

ϕ = T



(σ

)

−1

Tϕ =



(σ

)

−1

(σ

)

−1

ϕ . Repeating the action, we arrive at

ϕ =

m−1



k=0





]

−1

ϕ = B





ϕ. (2.65)

Here and below the product is ordered by the index k running from right to

left.

Deﬁnition 2.25. Equation (2.65) deﬁnes the function

(σ)=

m−1



k=0



(σ)



−1

2.6 Darboux transformations at associative ring with automorphism 47

It is convenient to wr it e down the t-deriva tive of σ by means of the functions

(σ

) that are connected with the generalized Bell polynomials [467, 271]:



m=−M







− σ

T (U

) B

m+1







. (2.66)

The resulting equation (2.66) is a nonlinear equation associated with (2.63)

that is reduced to a generalized Miura tran sformation in the stationary case

(Sect. 2.3).

The Matveev theorem for polynomials of the automorphism T provides

far-reaching generalizations of the conventional Darboux theorem proved orig-

inally for the second-order diﬀerential equation (for generalizations see [324]

as well) and can be formulated by means of the introduced entries in the

following way:

Theorem 2.26. L et the functions ϕ ∈R and ψ ∈R satisfy (2.63). Then the

function ψ

= D

ψ satisﬁes the equation

(x, t)=



m=−M

where the coeﬃcients are evaluated from the recurrence relations

−M

= U

−M

, (2.67)

− U

= U

− σ

, (2.68)

− U

m−1

= U

− σ

m−1

, (2.69)

= σ

(TU

)





−1

. (2.70)

Equations (2.67)–(2.70) deﬁne recursively the DTs of the coeﬃcients (po-

tentials) of the diﬀerential equation (2.63). Solving the re currenc e (2.69) by

means of (2.65) yields

m+M



l=0

−M+l

− σ

(TU

−M+l−1

) B

−M+l









−1

(2.71)

= σ

(TU

)





−1

. (2.72)

Proof. For the proof it is necessary to check the additional equality that ap-

pears from the term T

ψ with essential use of the expression for σ

from

(2.66).

This theorem establishes the covariance (form invariance) of (2.63) with

respect to the DT (2.64).

48 2 Factorization and classical Darboux transformations

The formalism for the second DT from (2.64) may be similarly constructed

on the ground of the identity

ϕ =



k=0



−



−1

ϕ = B

−



−



−1

ϕ. (2.73)

The deﬁnition o f the lattice Bell polynomials of the second type B

−

(σ

−

)can

be extracted from (2.73). The evolution equation for σ

−

is similar to (2.66):

−



m=−M



−



−



− σ

−

−1

) B

−

m−1



−



It may be considered as a further generalization of the Burgers equation (2.36)

and gives the second generalized Miura map for stationary solutions of (2.63).

Explicit formulas for U

−

are similar to (2.68)–(2.72).

2.7 Joint covariance of equations and nonlinear

problems. Necessity conditions of covariance

If a pair o f linear problems is simultaneously covariant with respect to a

Darboux transformation, it generates B¨acklund transformations of the corre-

sponding compatibility condition, or a nonlinear integrable equation. In the

context of such an integrability, th e joint covariance principle, used to con-

struct solutions of nonlinear problems from the very beginning [313], can be

considered as the origin of a classiﬁcation scheme [265, 267]. In this book,

we examine realizations of this scheme and seek the covariant form of equa-

tions and an appropriate basis with the simplest transformation properties.

Note that a proof of the covariance theorems for the linear operators incor-

porates the generalized Burgers equations that i n stationary versions reduce

to the generalized Miura transformation. We give and examine the explicit

form of the Miura equality in both the general and the stationary cases (see

also [270]). This equality gives an additional nonlinear equation that is auto-

matically solved by the Cole–Hopf substitution and is used to generate dress-

ing t-chain equations [79]. We show how the form of the covariant operator

can be found by comparing some kind of Frech´et derivatives of the operator

coeﬃcients and the transforms.

2.7.1 Towards the classiﬁcation scheme: joint

covariance of one-ﬁeld Lax pairs

The basis of the formalism introduced here has been elab orated in [265, 267]

and the compact formulas with the generalized Bell po lynomials are given in

Sect. 2.2. The formalism is valid for non-Abelian coeﬃcients a

as well, and