Proc. Nati. Acad. Sci. USA Vol. 77, No. 9, pp. 5025-5027, September 1980 Applied Mathematical Sciences Backlund transformations and nonlinear differential difference equations (nonlinear evolution equations/Schrbdinger spectral problem/soliton) D. LEVI* AND R. BENGURIAt Physics Department, The Rockefeller University, New York, New York 10021 Communicated by Kenneth M. Case, June 2,1980 ABSTRACT It is shown that any Bicklund transformation of a nonlinear differential equation integrable by the multichannel Schrbdinger eigenvalue problem can be written in the form V1 = U'V - VU. This allows us to interpret the Bicklund transformation formally as a nonlinear differential difference equation for which we can immediately construct the soliton solutions. Auto-&icklund transfonnations (BT) are an important property of any integrable nonlinear evolution equation and have been derived for a wide class of equations. BT are equations that relate different solutions of the same nonlinear evolution equation; the application of the BT allows one to construct the soliton solutions of the nonlinear evolution equations. In this note we want to show that, for the multichannel (i.e., matrix) Schr6dinger spectral problem (MSP), all the BT defined through the Wronskian technique (1) can be written in a very simple and unique way (see Eq. 12). We consider the MSP only because the proof of our statement becomes simpler and also because MSP includes a wide class of other problems, in particular the Zakharov-Shabat problem, which can be obtained from MSP by means of the reduction techniques.t Eq. 12 seems appropriate for studying the group properties of the BT, and work on this is in progress. On the other hand, it can be formally intepreted as a functional differential equation (DDE) (or elsewhere also called differential difference equations), which is integrable and for which we shall provide the one-soliton solution. Backlund transformations for the MSP We shall summarize some of the results contained in ref. 1. The MSP is defined by the differential equation Vxx = (Q - k2)i, [1] in which Q is an X X q hermitian matrix that depends on x e R and 61 is a vector that depends on x and keC. Both 41 and Q may depend parametrically on some other variable. We assume Q vanishes exponentially as x I co. For Eq. 1, we can define the continuum part of the spectrum by the following asymptotic boundary condition for the matrix solution A, for ke R, built up of X7 linearly independent vector solutions of Eq. 1, e-ikx+ R(k)ekx, 4I(x,k) k) 'I'x [2a] > T(k)e-ikx. x --I - co (-4k2) 3 dxTI'T(x,k)F(x)YI(x,k) = 3' dx I'T(x,k)AF(x)T(x,k), [3] in which he T is the transpose of the matrix erator A is defined by AF(x) = F..(x) - 2[Q'(x)F(x) + F(x)Q(x)] V'. Here the op- +r dx'F(x') [4] with the operator r defined as rF(x) = Q'(x)F(x) + F(x)Qx(x) + 3 cho dx'[Q'(x)Q'(x')F(x') - Q'(x)F(x')Q(x') - [5] Q'(x')F(x')Q(x) + F(x')Q(x')Q(x)]. From the Wronskian identity and the asymptotic behavior of ' and T' given in Eq. 2, we get (see equation 3.2.1 of ref. 1) (2ik)[MR(k) - R'(k)M] = 3 dxI'T(x,k) X [MQ(x) - Q'(x)M]*(x,k) [6a] and (see equation 3.2.3 of ref. 1) (-4k2)[NR(k) + R'(k)N] = f dx IT(x,k)(rN)i&(x,k), [6b] in which M and N are independent of x but otherwise are arbitrary n X n matrices. Let f and g be arbitrary entire scalar functions. § Then from Eqs. 3 and 6 we get (2ik)f(-4k2)[MR(k) - R'(k)M] = 3' dxt'IT(x,k)[f(A){MQ(x) -Q'(x)Mj]4'(x,k) [7a] _co Abbreviations: BT, Bicklund transformation(s); MSP, multichannel Schrodinger spectral problem; DDE, differential difference equation(s). * Permanent address: Instituto di Fisica dell'Universita di Roma, 00185 Roma, Italy. t On leave from the Department of Physics, Universidad de Chile, Santiago, Chile. *Calogero, F. & Degasperis, A. (1979) Reduction Technique for Matrix Nonlinear Evolution Equations Solvable by the Spectral Transform, preprint, Istituto di Fisica, Universita di Rotna. § One can also considerf and g as being matrix-valued entire functions. For simplicity we consider the case when they are scalar func- -D + 4f(x,k) (equation 3.1.3 in ref. 1) one obtains, for an arbitrary matrix F(x), which vanishes asymptotically together with all its derivatives, the following relationship, 0 Let A' and T be two matrix solutions of Eq. 1 (with potential Q' and Q, respectively, and reflection coefficients R' and R, respectively). Then through the generalized Wronskian identity The publication costs of this article were defrayed in part by page charge payment. This article must therefore be hereby marked "advertisement" in accordance with 18 U. S. C. §1734 solely to indicate this fact. tions. 5025 5026 Applied Mathematical Sciences: Levi and Benguria and (-4k2)g(-4k2)[NR(k) + R'(k)N] Proc. Natl. Acad. Sci. USA 77 (1980) Eq. 17b is identically satisfied when we substitute the expressions for A and B given by Eqs. 16. In fact, we get clo dx" *T(x,k)[g(A)rN]T(x,k). [7b] = 5 By using Eqs. 7, Calogero and Degasperis (1) have shown that the BT for the MSP can be written as [8a] f(A)[MQ(x) - Q'(x)M] + g(A)rN = 0, and the corresponding reflection coefficients R and R' are related by R'(k) = [f(-4k2)M - 2ik g(-4k2)N]R(k) X [f(-4k2)M + 2ik g(-4k2)N]-l. [8b] One can write the MSP given by Eq. 1 in the form ox(xk)=U(Qk)=(xk), in which U is a 2,q X 2n matrix defined by U(Qk) = (-kI Q) -2D(x) - 2Q'(x) dyD(y) dyD(y)Q(x) - 2g(-4k2)Q'(x)N + 2 + 2g(-4k2)NQ(x) -2g(-4k2)[NQ(x) -Q'(x)N] + -2 3 2D,(x) dy[D(y)Q(x) - Q'(x)D(y)] 0. As for Eq. 17a, when we substitute the expression for A and B given by Eqs. 16, it becomes Dx(x) - 2[D(x)Q(x) + Q'(x)D(x)] [9] + 3 + 3J dy[D(y)Qx(x) + Q'(x)D(y)] dy dz[D(z)Q(y)Q(x) - Q'(y)D(z)Q(x) [10] Q'(x)D(z)Q(y) + Q'(x)Q'(y)D(z)] 4k2D(x) - g(-4k2)[NQx(x) - Q'(x)N] - in which I is the identity i X t7 matrix and k is the 27 vector constructed from i in the following way = (1 - ik¢,) g, can be written as [12] VX = U(Q',k)V - VU(Q,k). Proof: Let us write V in terms of two q X i matrices A and B,Ax + ikBx + BQ [13] V (A + Bx-ikB B ~~A + ikB in which A and B are given by 00 00c dy 3 dz $D(z)Q(y) - Q'(y)D(z)l f(-4k2)M + g(-4k2) 3 dy [NQ(y) - Q'(y)N] [14a] and B = 23 dy D(y) + 2g(-4k2)N, [14b] with + j(x)(rN), [15] = f(A) - f(-4k2) [16a] D(x) = Q(x)[MQ(x) - Q'(x)M] in which A + 4k2 + [11] THEOREM. Let U(Q,k) [and U(Q',k)] be the 2q X 2i1 matrix defined by Eq. 10, in which Q (and Q', respectively) is the matrix potential and k the eigenvalue for the MSP given by Eq. 1; then we can explicitly construct a 2q X 2q matrix V(Q,Q',f,gk) such that the BT (Eq. 8a), parametrizd by f and A -D(x) + J + and - - f(-4k2)[Q'(x)M - MQ(x)] g(-4k2) 3 dy[Q'(x)NQ(y) - Q'(x)Q'(y)N NQ(y)Q(x) + Q'(y)NQ(x)] + 2g(-4k2)NQx(x) = 0. Recalling Eqs. 4 and 5, we can write the previous expression as [A + 4k2]D(x) + f(-4k2)[Q'(x)M - MQ(x)] + g(-4k2)rN = 0. [18] Recalling Eqs. 16, Eq. 18 finally reads f(A)[MQ(x) - Q'(x)M] + g(A)rN = 0, which is exactly the BT given by Eq. 8. Differential difference equations It is known (2) that one can apply the generalized inverse method of Zakharov and Shabat to solve some DDE. In order to do that, one writes the DDE as the compatibility condition for a linear problem [19a] 4I',(n,x) = U(n,x)%t(n,x) [19b] 4I'(n + 1,x) = V(nx)q(nx), in which n is an integer, x is a real variable, and 4V, U, and V are matrix-valued functions depending on a spectral parameter k. The compatibility condition VX(n,x) = U(n + 1,x)V(n,x) - V(n,x)U(n,x) [20] is equivalent to the DDE. We note that Eq. 20 has the same structure as Eq. 12. In particular, if we formally substitute Q = qn [21] in Eqs. 8 (i.e., in the BT), we get a DDE that can be solved by the inverse method; the U and V corresponding to this DDE are exactly the ones defined by Eqs. 1 and 13, respectively. The solutions that one obtains for the MSP through the inverse spectral technique are also solutions to the DDE when the evolution in time is determined by the matrix V or, that is, the Q= qn+I (ajx) = g(A) - g(-4k2) [16b] A + 4k2[1b By Eqs. 10 and 13, Eq. 12 becomes AXX-Q'A + AQ + 2Bx(Q-k2)+ BQx = 0 [17a] BXX-Q'B + BQ + 2AX =0. [17b] Proc. Natl. Acad. Sci. USA 77 (1980) Applied Mathematical Sciences: Levi and Benguria reflection coefficient evolves according to the following equation: R(n + 1) = [f(-4k2)M - 2ikg(-4k2)N]R(n) X [f(-4k2)M + 2ikg(-4k2)N]-l. [22] Then, following ref. 1, the one-soliton solution to the DDE corresponding to a purely imaginary eigenvalue k = ip of the MSP, given by Eq. 1, reads q(n,x) = -2 p2 Icosh[pfx - t(n)1]-2 P(n), [23] in which P(n) is an idempotent matrix given by (CoIE(n- n)ICo) fM + 2pgN]n-nop(no) X JM - 2pgN]-(n-no), [23a] with (Col =-ICo) P~o(ColICo) P(no) and [23b] E(n) = LfM + 2pgN]-[fM - 2pgN]n, [23c] 4=o+-1Ilog{(coIEln no)J. [23d] - The vector ICo) is given by the asymptotic behavior of i1, as x - + co, corresponding to the initial problem. It is important to notice that the value k = ip is a pole of the reflection coefficient of the BT of the MSP but it is not a pole of Eq. 22, which gives us the evolution of the DDE. An example To conclude this note, we exhibit the matrix V corresponding to the simplest BT contained in Eq. 8; then we write down the corresponding DDE together with its one-soliton solution. The simplest BT is obtained by setting f = constant = p, g = constant = 1/2, and M = N = l: W'2(x) + Wx(x) + 2pfW'(x) - W(x)j + 1/2[W'(x) - W(x)]2 = 0, [24] in which we have introduced W defined by W(x) dx'Q(x'). From Eq. 14 we derive A(x) = -pI + 1/2fW(x) - W'(x)j [25] [26a] 5027 and B(x) = I, since D = 0. Therefore, V reads [26b] [27] -1/2(Wx + W') (-p + ik)I + 1/2(W - W')J One can easily check at this point that V. = U'V - VU, with -(P + ik)I + 1/2(W - + W9) I U given by Eq. 10, is equivalent to Eq. 24. The DDE that one obtains from Eq. 24 is [w(n + 1,x) + w(n,x)]x + 2h1w(n + 1,x) - w(n,x)} + 1/2fw(n + 1,x) - w(n,x)12 = 0. [28] As we have said in the general framework, the DDE given by Eq. 28 can be put into the Zakharov-Shabat form, with U and V being - kx(nx) U(n,x) = (-ikI [29] ikW V(nx) = -(h + ik)I + 1/2{w(n,x) - w(n + 1,x)l -1/2fw.(n,x) + w.(n + 1,x) -(+ h - ik)I + 1/2tw(n,x) - w(n + 1,x)j< [30] The one-soliton solution to Eq. 28 is given by w(n,x) = -2 1i - tghfp(x -(n))JjP(no), in which P(no) is given by Eq. 23b and t(n) = to + 2p (n -nO) log { pJ This work was supported in part by the National Science Foundation under Grant MCS 78-20455 (to R.B.) and in part by a North Atlantic Treaty Organization Fellowship (to D.L.). 1. Calogero, F. & Degasperis, A. (1977) II Nuovo Cimento, 39B, 1-54. 2. Levi, D., Ragnisco, 0. & Bruschi, M. (1980) "Extension of the Zakharov-Shabat generalized inverse method to solve differential difference and difference-difference equations," presented at the Jadwisin Workshop on Solitons (Warsaw, 1979), Il Nuovo Cimento, in press.