betaFeedback

WorksPhysics and Astronomy

Bi-Hamiltonian structure and conserved quantities of the Yajima-Oikawa-Newell system

Abstract

We study the Hamiltonian and bi-Hamiltonian structure of the Yajima-Oikawa-Newell (YON) system, an integrable model of long wave-short wave interaction depending on two real parameters αα and ββ, which reduces to the Yajima-Oikawa system for α=0α=0 and to the Newell system for β=0β=0. For α≠0α\neq0, a constant shift of the long wave maps the YON system into the Newell system, for which two nonlocal Hamiltonian operators are known. We show that, for α≠0α\neq0, the system admits no local Hamiltonian operator homogeneous in its natural grading and polynomial in ββ. A nonlocal gauge transformation built on the potential of the long wave makes the first of the known operators local for a single value of its parameter: the transformed system is Hamiltonian, for every α≠0α\neq0 and every ββ, with respect to a local, non-degenerate Poisson operator, which is the Lie-Poisson structure of the sl(2)\mathfrak{sl}(2) current algebra. From a recursion operator advancing the hierarchy by two steps, given in explicit weakly nonlocal form, we construct a second, weakly nonlocal Hamiltonian structure, whose nonlocal tail is generated by the three autonomous Lie point symmetries of the system, and prove that both structures are compatible. Since the recursion operator admits no square root and the first structure has no Casimirs, the Lenard-Magri chain splits into two families, seeded by the two simplest conserved quantities. Every flow of both families is local, and all the corresponding Hamiltonians are in involution. As by-products, we relate the Lie point symmetries of the system to its conserved quantities via Noether's theorem, and obtain the conserved densities in closed form by asymptotic diagonalisation of the Lax pair.

Introduction

The resonant interaction between a long wave and a short wave, which takes place whenever the group velocity of the short wave matches the phase velocity of the long wave [1], is one of the classical mechanisms of nonlinear wave theory, and two of its best-known integrable models are the Yajima–Oikawa (YO) system [2, 3, 4] and the Newell system [1, 5]. In [6], both models were unified and generalised into the Yajima–Oikawa–Newell (YON) system

iSt+Sxx+(iαLx+α2L2−βL−2α∣S∣2)S=0,(1)\mathrm{i}S_t+S_{xx}+(\mathrm{i}\alpha L_x+\alpha^2L^2-\beta L-2\alpha|S|^2)S=0, \tag*{(1)}
Lt=2(∣S∣2)x,(2)L_t=2(|S|^2)_x, \tag*{(2)}

where SS is a complex variable representing the short-wave envelope, LL is the real amplitude of the long wave, and α\alpha and β\beta are two real parameters.

The YON system is integrable, in the sense of possessing a Lax pair, for every choice of α\alpha and β\beta [6], and it contains the two classical models as particular cases. Namely, taking α=0\alpha= 0 and β=1\beta= 1 in Eqs. (1) we recover the YO system

iSt+Sxx−LS=0,(3)\mathrm{i}S_t+S_{xx}-LS=0, \tag*{(3)}
Lt=2(∣S∣2)x,(4)L_t=2(|S|^2)_x, \tag*{(4)}

whereas taking α=σ\alpha= \sigma, β=0\beta= 0 and L↦σLL \mapsto\sigma L, with σ=±1\sigma= \pm1, we recover the Newell system

iSt+Sxx+(iLx+L2−2σ∣S∣2)S=0(5)\mathrm{i}S_t + S_{xx} + \left(\mathrm{i}L_x + L^2 - 2\sigma|S|^2\right)S = 0 \tag*{(5)}
Lt=2σ(∣S∣2)x.(6)L_t = 2\sigma\left(|S|^2\right)_x. \tag*{(6)}

Let us remark that the choices β=1\beta= 1 and α=σ\alpha= \sigma entail no loss of generality: simple scalings of the fields and of the independent variables allow one to normalise any nonzero value of α\alpha to ±1\pm1 and any nonzero value of β\beta to 11 (see § 2). In fact, for α≠0\alpha\ne0 the parameter β\beta can be removed altogether: a constant shift of LL, together with a time-dependent phase of SS, maps the YON system into the same system with β=0\beta= 0, that is, into the Newell system, at the price of replacing the vanishing boundary condition of LL by a constant background. We will nevertheless keep both parameters arbitrary throughout, as we consider fields that vanish at infinity, and as this makes both the grading of the system and its reductions to YO and Newell more transparent.

Since its introduction, the YON system and its solutions have attracted steady interest. The linear stability of its plane wave solutions was analysed in [6], and its solitons, breathers and rogue waves were obtained by an ansatz approach in [7] and within the Hirota bilinear framework in [8, 9], and further studied in [10, 11]. In [12], the YON system is derived from physical principles, by applying the multiscale method to a modified KdV equation for shallow water waves, and in [13] it appears in connection with the Tolkynay equations.

Some conservation laws of Eqs. (1) were obtained in [7] by means of the multiplier method of Anco and Bluman [14, 15], as no variational formulation of the YON system was known. For its two reductions, on the other hand, a Hamiltonian theory is available. The YO system (2) is well known to be bi-Hamiltonian, being a reduction of the constrained KP hierarchy [16, 17], and Liu [18] showed that the Newell system (3) is a modification of it, the two systems being related by a Miura-type map, and gave two Hamiltonian operators for the Newell hierarchy. Both of them are nonlocal in the variables (S,S∗,L)(S, S^*, L): the first one is weakly nonlocal, with a tail built on the phase symmetry, whereas the second one involves inverses of differential operators. To the best of our knowledge, these results have not been related to the YON system, and it is not known whether the nonlocality of these operators can be avoided. The purpose of this paper is to address both questions.

As we will see, the nonlocality is unavoidable in the original variables, but it is not intrinsic. For α≠0\alpha\ne0, the shift of LL mentioned above reduces the YON system to the Newell system, so that the operators of [18] apply to it. We first show that the natural local candidates, namely operators that are homogeneous in the grading of the system and compatible with its U(1)U(1) invariance, all fail to be Poisson as soon as α≠0\alpha\ne0. A phase gauge on the potential of the long wave, very much in the spirit of the gauge transformations relating the members of the derivative NLS family [19, 20, 21, 22, 23], then turns the first operator of [18] into a local one, and it does so for a single value of the gauge parameter. In that gauge the operator is linear in the fields, and it is in fact the Lie–Poisson structure of the sl(2)\mathfrak{sl}(2) current algebra, which makes its Jacobi identity manifest. Taking this local, non-degenerate operator as the first structure, we construct a recursion operator for the hierarchy in explicit, weakly nonlocal form and, from it, a second Hamiltonian structure, also weakly nonlocal, which lies one step above the pair of [18] and whose nonlocal tail is generated by the three autonomous Lie point symmetries of the system. The resulting picture has some features that are worth pointing out from the start. The recursion operator advances the hierarchy by two steps rather than one, as in the YO case, and admits no square root, so that the hierarchy splits into two families; and, since the first structure has no Casimirs, the Lenard–Magri chain cannot be started, as usual, from a Casimir, but has to be seeded by the two simplest conserved quantities.

The paper is organised as follows. In § 2, we recall the Lax pair of the YON system, introduce its grading and U(1)U(1) charges, and use them to prove that, for α≠0\alpha\ne0, the system admits no local Hamiltonian operator homogeneous in that grading and polynomial in β\beta. In § 3, we show that a nonlocal gauge transformation on the potential of the long wave makes the system Hamiltonian with respect to a local, non-degenerate operator, which we identify both with the first operator of [18] and with the Lie–Poisson structure of the sl(2)\mathfrak{sl}(2) current algebra, construct the conserved densities in closed form by asymptotic diagonalisation of the Lax pair, and take advantage of the Hamiltonian formulation to relate the Lie point symmetries of the system obtained in [24] to its conserved quantities via Noether’s theorem. In Section 4, we construct a recursion operator acting on the symmetries of the system and, from it, a second Hamiltonian structure, which is weakly nonlocal in the sense of Ferapontov, Maltsev and Novikov [25, 26, 27], and prove that both structures are compatible. In fact, we show that the nonlocal tail is generated not just by symmetries of the system, but by its three autonomous Lie point symmetries. In Section 5, we study the Lenard–Magri chain of the pair, show that every flow of both families is local and that the corresponding Hamiltonians are in involution, and discuss the reductions β=0\beta= 0 and α=0\alpha= 0. Finally, Section 6 collects our conclusions.

No-go theorem for local Hamiltonian structures

Lax pair, symmetries and grading

Before looking for a Hamiltonian structure, let us collect the properties of the YON system that will constrain the search, namely its Lax pair, its scaling symmetries and the grading they induce, and its U(1)U(1) invariance. The YON system (1) arises as the compatibility condition Xt−Tx+[X,T]=0X_t-T_x+[X,T]=0 of the Lax pair [6]

X(λ)=iλΣ+Q,T(λ)=(iλ)2T2+(iλ)T1+T0,Σ=diag⁡(1,0,−1),(7)X(\lambda)=\mathrm{i}\lambda\Sigma+Q,\qquad T(\lambda)=(\mathrm{i}\lambda)^2T_2+(\mathrm{i}\lambda)T_1+T_0,\qquad\Sigma=\operatorname{diag}(1,0,-1), \tag*{(7)}

with

Q=(0SiLαS∗0S∗iα2L−iβαS0),(8)Q=\begin{pmatrix} 0 & S & \mathrm{i}L\\ \alpha S^{*} & 0 & S^{*}\\ \mathrm{i}\alpha^{2}L-\mathrm{i}\beta& \alpha S & 0 \end{pmatrix}, \tag*{(8)}
T2=i3(1000−20001),T1=(0iS0iαS∗0−iS∗0−iαS0),(9)T_2=\frac{\mathrm{i}}{3}\begin{pmatrix} 1 & 0 & 0\\ 0 & -2 & 0\\ 0 & 0 & 1 \end{pmatrix},\qquad T_1=\begin{pmatrix} 0 & \mathrm{i}S & 0\\ \mathrm{i}\alpha S^{*} & 0 & -\mathrm{i}S^{*}\\ 0 & -\mathrm{i}\alpha S & 0 \end{pmatrix}, \tag*{(9)}
T0=(−iα∣S∣2−αLS+iSxi∣S∣2−α2LS∗+βS∗−iαSx∗2iα∣S∣2−αLS∗−iSx∗iα2∣S∣2−α2LS+βS+iαSx−iα∣S∣2).(10)T_0=\begin{pmatrix} -\mathrm{i}\alpha\lvert S\rvert^2 & -\alpha LS+\mathrm{i}S_x & \mathrm{i}\lvert S\rvert^2\\ -\alpha^2LS^{*}+\beta S^{*}-\mathrm{i}\alpha S_x^{*} & 2\mathrm{i}\alpha\lvert S\rvert^2 & -\alpha LS^{*}-\mathrm{i}S_x^{*}\\ \mathrm{i}\alpha^2\lvert S\rvert^2 & -\alpha^2LS+\beta S+\mathrm{i}\alpha S_x & -\mathrm{i}\alpha\lvert S\rvert^2 \end{pmatrix}. \tag*{(10)}

eq:1 are invariant under the scaling

(S,L,x,t,α,β)↦(εS,εL,ε−1x,ε−2t,α,εβ),(11)(S,L,x,t,\alpha,\beta)\mapsto(\varepsilon S,\varepsilon L,\varepsilon^{-1}x,\varepsilon^{-2}t,\alpha,\varepsilon\beta), \tag*{(11)}

which induces the grading

[S]=[S∗]=[L]=[∂x]=[β]=1,[α]=0,[∂t]=2,(12)[S]=[S^{*}]=[L]=[\partial_x]=[\beta]=1,\qquad[\alpha]=0,\qquad[\partial_t]=2, \tag*{(12)}

in which the flow (1) is homogeneous of weight 3. Let us note in passing that eq:1 are also invariant under the transformation

(S,L,α,β)↦(aS,a2L,a−2α,a−2β),a∈R∖{0},(13)(S,L,\alpha,\beta)\mapsto(aS,a^2L,a^{-2}\alpha,a^{-2}\beta),\qquad a\in\mathbb{R}\setminus\{0\}, \tag*{(13)}

which leaves xx and tt unchanged. Together, the two scalings allow one to normalise any nonzero value of α\alpha to ±1\pm1 and any nonzero value of β\beta to 1, as anticipated in the Introduction. Moreover, for α≠0\alpha\ne0 the parameter β\beta can be removed altogether: if (S,L)(S,L) solves eq:1, then

S~=exp⁡(iβ2t4α2)S,L~=L−β2α2(14)\widetilde{S}=\exp\left(\frac{\mathrm{i}\beta^2t}{4\alpha^2}\right)S,\qquad\widetilde{L}=L-\frac{\beta}{2\alpha^2} \tag*{(14)}

solve the same equations with β=0\beta=0, that is, the Newell system. In other words, for α≠0\alpha\ne0 the parameter β\beta plays the role of a constant background of the long wave in the Newell system. We will not make use of any of these freedoms: the transformation (8) does not preserve the vanishing of LL at infinity, and the grading (6) relies precisely on β\beta being a parameter of weight one.

Let us now see how the grading constrains a Hamiltonian formulation. We will restrict ourselves to operators and densities that are homogeneous in the grading (6) and polynomial in β\beta, in the fields and in their xx-derivatives, with coefficients that may depend on α\alpha. If P\mathcal{P} is such a Hamiltonian operator and H=∫h dxH=\int h\,dx is the corresponding Hamiltonian, the relation ut=P δHu_t=\mathcal{P}\,\delta H, with u=(S,S∗,L)u=(S,S^*,L), implies that

3=weight⁡(Pab)+[h]−1(15)3=\operatorname{weight}(\mathcal{P}_{ab})+[h]-1 \tag*{(15)}

for each entry Pab\mathcal{P}_{ab} of the operator, since all three fields have weight one and the variational derivative lowers the weight by one. Hence, P\mathcal{P} must be homogeneous of weight p=4−[h]p=4-[h]. Since the entries of P\mathcal{P} have non-negative weight, p≥0p\geq0, while p≥4p\geq4 would require [h]≤0[h]\leq0, that is, a constant density, for which δH=0\delta H=0. Only the weights p=0,1,2,3p=0,1,2,3 remain, and for each of them the space of candidates is finite-dimensional.

We can further restrict the class of candidates by means of the U(1)U(1) symmetry S↦eiϕSS\mapsto e^{i\phi}S, ϕ∈R\phi\in\mathbb{R}. Assigning charges +1+1, −1-1 and 00 to SS, S∗S^* and LL, respectively, the entry Pab\mathcal{P}_{ab} must carry charge qa+qbq_a+q_b, since P\mathcal{P} maps covectors, whose components δH/δub\delta H/\delta u_b carry charge −qb-q_b, to vectors, whose components carry charge qaq_a. Combining this condition with the skew-adjointness of the operator and with the reality condition (that complex conjugation exchanges the roles of SS and S∗S^* and fixes LL), the problem at each admissible weight reduces to a linear problem for finitely many coefficients.

It remains, therefore, to analyse the weights 00, 11, 22 and 33. Since the Hamiltonian density, of weight [h]=4−p[h]=4-p, has to be conserved, it is convenient to compute first all the conserved densities of weights one to three, by solving δ(∂tρ)=0\delta(\partial_t\rho)=0 on the graded monomial basis of each weight. Modulo total derivatives, this results in

ρ0=L,ρ1=α2L2−∣S∣2,ρ2=2α2L∣S∣2−β∣S∣2+2αℑ(S∗Sx),(16)\rho_0=L,\qquad\rho_1=\frac{\alpha}{2}L^2-|S|^2,\qquad\rho_2=2\alpha^2L|S|^2-\beta|S|^2+2\alpha\Im(S^*S_x), \tag*{(16)}

which, apart from the products of the lower ones by powers of β\beta, are the only conserved densities of their weight, and which match the ones derived in [7]. Here and in what follows, for a density ρ\rho we write δρ\delta\rho for the variational derivative of ∫ρ dx\int\rho\,dx. A single new density appears at weight four as well, and we will come back to it in Section 3.

Search for local Hamiltonian operators

With the candidates reduced to a finite-dimensional family at each weight, let us now determine which of them, if any, are Poisson operators. To this end, we will check the Jacobi identity via the Schouten criterion [28, 29], that is, a skew-adjoint operator P\mathcal{P} is Hamiltonian if and only if, given arbitrary covectors θ1\theta^1, θ2\theta^2, θ3\theta^3, the Schouten bracket [30, 31, 32]

[P,P]=∑cyc⟨θ1,P′[Pθ2]θ3⟩(17)[\mathcal{P},\mathcal{P}]=\sum_{\mathrm{cyc}}\langle\theta^1,\mathcal{P}'[\mathcal{P}\theta^2]\theta^3\rangle \tag*{(17)}

is a total xx-derivative, where the sum is extended over all cyclic permutations of the indices, ⟨θ,v⟩=∑aθava\langle\theta,v\rangle=\sum_a\theta_av_a is the bilinear covector-vector pairing, with no complex conjugation, SS and S∗S^* being treated as independent fields, and P′[w]\mathcal{P}'[w] is the Fréchet derivative of the coefficients of P\mathcal{P} along the direction of ww. Whether an expression is a total xx-derivative is decided, in turn, by the Euler operator: it is so if and only if its variational derivatives with respect to all the fields and all the components of the covectors vanish identically.

Weight zero is excluded by the U(1)U(1)-charge and skew-adjointness conditions alone. Indeed, the entries are then constants, and the charge rule allows only for P12\mathcal{P}_{12}, P21\mathcal{P}_{21} and P33\mathcal{P}_{33}; but a constant P33\mathcal{P}_{33} is self-adjoint, so skew-adjointness forces it to vanish. Hence, the last row of P\mathcal{P} vanishes identically, and the operator cannot reproduce the equation for LtL_t.

Theorem 2.1 (Weight one). Let P(1)\mathcal{P}^{(1)} be a skew-adjoint operator, homogeneous of weight 11 under the grading (6), satisfying the reality and U(1)U(1)-charge conditions, which generates the flow (1) with a Hamiltonian whose

density is conserved of weight three. Up to multiplication by a constant, P(1)\mathcal{P}^{(1)} is forced to take the form

P(1)=(0−α∂x+iα2L−iβ2−2iαS−α∂x−iα2L+iβ202iαS∗2iαS−2iαS∗4∂x).(18)\mathcal{P}^{(1)} = \begin{pmatrix} 0 & -\alpha\partial_x+\mathrm{i}\alpha^2L-\frac{\mathrm{i}\beta}{2} & -2\mathrm{i}\alpha S\\ -\alpha\partial_x-\mathrm{i}\alpha^2L+\frac{\mathrm{i}\beta}{2} & 0 & 2\mathrm{i}\alpha S^*\\ 2\mathrm{i}\alpha S & -2\mathrm{i}\alpha S^* & 4\partial_x \end{pmatrix}. \tag*{(18)}

The corresponding Hamiltonian density is

h(1)=1αRe(S∗Sx)+L∣S∣2−β2α2∣S∣2+β28α3L,(19)h^{(1)}=\frac{1}{\alpha}\mathop{\mathrm{Re}}(S^*S_x)+L|S|^2-\frac{\beta}{2\alpha^2}|S|^2+\frac{\beta^2}{8\alpha^3}L, \tag*{(19)}

and its Schouten bracket (11) reduces, modulo total derivatives, to a single trivector

[P(1),P(1)]≡−6iα2∑cyc(θ31)x(θ21θ13−θ22θ13),(20)[\mathcal{P}^{(1)},\mathcal{P}^{(1)}]\equiv-6\mathrm{i}\alpha^2\sum_{\mathrm{cyc}}(\theta^1_3)_x(\theta^1_2\theta^3_1-\theta^2_2\theta^3_1), \tag*{(20)}

where the right-hand side is not a total derivative.

Hence, P(1)\mathcal{P}^{(1)} is a Poisson operator if and only if α=0\alpha=0. Note, however, that at α=0\alpha=0 it reduces to the constant-coefficient operator diag⁡{βσ2,4∂x}\operatorname{diag}\{\beta\sigma_2,4\partial_x\}, with σ2\sigma_2 the second Pauli matrix, which, although Poisson, generates the Yajima–Oikawa flow from no Hamiltonian at all, as the density h(1)h^{(1)} is singular there. The Hamiltonian structure of the Yajima–Oikawa system is a different one, which we will recover in § 5.

Theorem 2.2 (Weight two). The family of skew-adjoint operators, homogeneous of weight 2 under the grading (6), satisfying the reality and U(1)U(1)-charge conditions, which generate the flow (1) with Hamiltonian ∫ρ1 dx\int\rho_1\,dx, takes the form, up to multiplication by a constant,

P(2)=(0P12(2)P13(2)−P12(2)†0P13(2)−P13(2)†−P13(2)†0),(21)\mathcal{P}^{(2)}= \begin{pmatrix} 0 & \mathcal{P}^{(2)}_{12} & \mathcal{P}^{(2)}_{13}\\ -\mathcal{P}^{(2)\dagger}_{12} & 0 & \mathcal{P}^{(2)}_{13}\\ -\mathcal{P}^{(2)\dagger}_{13} & -\mathcal{P}^{(2)\dagger}_{13} & 0 \end{pmatrix}, \tag*{(21)}

where

P12(2)=−i∂x2+dαL∂x+i(1+αb)βL+iα(a−α)L2+2iα∣S∣2+α(ic+d2)Lx,\mathcal{P}^{(2)}_{12}=-\mathrm{i}\partial_x^2+d\alpha L\partial_x+\mathrm{i}(1+\alpha b)\beta L+\mathrm{i}\alpha(a-\alpha)L^2+2\mathrm{i}\alpha|S|^2+\alpha\left(\mathrm{i}c+\frac{d}{2}\right)L_x,
P13(2)=(−1+ic+d2)S∂x+dSx+i(aL+bβ)S,(22)\mathcal{P}^{(2)}_{13}=\left(-1+\mathrm{i}c+\frac{d}{2}\right)S\partial_x+dS_x+\mathrm{i}(aL+b\beta)S, \tag*{(22)}

with a,b,c,d∈Ra,b,c,d\in\mathbb{R} arbitrary parameters. The associated Jacobi system, made of 287 distinct polynomial equations in (a,b,c,d)(a,b,c,d), is inconsistent for every value of α\alpha and β\beta; hence, no member of the family is a Poisson operator.

Remark 2.3. Admitting a Hamiltonian density that combines the two weight-two conserved densities ρ1\rho_1 and βρ0\beta\rho_0 extends the class of candidates to a total of five parameters (with 310 distinct equations in the Jacobi system), but it does not change the conclusion of Theorem 2.2.

The last remaining weight is p=3p=3, which implies [h]=1[h]=1 and hence the Hamiltonian H=∫L dxH=\int L\,dx, ρ0\rho_0 being the only weight-one density other than the constant β\beta. For this choice, the deciding test is not the Jacobi identity but a simpler necessary condition. Since a Hamiltonian vector field must preserve its own Poisson structure, any Poisson operator P\mathcal{P} that generates Eqs. (1) must satisfy

LKP=P′[K]−K′P−P(K′)†=0,(23)\mathcal{L}_K\mathcal{P}=\mathcal{P}'[K]-K'\mathcal{P}-\mathcal{P}(K')^\dagger=0, \tag*{(23)}

where L\mathcal{L} denotes the Lie derivative, KK is the right-hand side of Eqs. (1), K′K' is its Fréchet derivative and P′[K]\mathcal{P}'[K] is the derivative of the coefficients of P\mathcal{P} along KK. Relation (16) is a linear identity, allowing us to produce the following result.

Theorem 2.4 (Weight three). The family of skew-adjoint operators, homogeneous of weight 33 under the grading (6), satisfying the reality and U(1)U(1)-charge conditions and the identity PδL dx=K\mathcal{P}\delta_L\,dx=K is a 3131-parameter affine family. Introducing the linear condition (16) on the general 3131-parameter operator results in a linear system of rank 3131 and augmented rank 3232, meaning that the system is incompatible. The incompatibility holds for every value of α\alpha and β\beta. Hence, no member of the family is a Poisson operator.

Since we have obtained a negative answer at every admissible weight, we can condense the results in the following:

Corollary 2.5. For α≠0\alpha\ne0, the YON system (1) admits no local Hamiltonian structure in the variables (S,S∗,L)(S,S^*,L) whose operator and Hamiltonian density are homogeneous in the grading (6) and polynomial in β\beta.

Remark 2.6. The case α=0\alpha=0 is not an exception within this class. The canonical Hamiltonian operator of the YO system, which we will recover in § 5, is homogeneous of weight one, but its Hamiltonian contains the term ∣Sx∣2/β\lvert S_x\rvert^2/\beta, and is therefore not polynomial in β\beta. This reflects the fact that the natural grading of the YO system is a different one, [S]=3/2[S]=3/2, [L]=2[L]=2 and [β]=0[\beta]=0, which for the YON system results from combining the scalings (5) and (7) and assigns weight −1-1 to α\alpha. For α≠0\alpha\ne0, on the other hand, admitting such terms does not change the conclusion at weight one: with densities containing negative powers of β\beta down to β−2\beta^{-2}, the operator of Theorem 2.1 remains the only candidate.

Having exhausted the homogeneous local candidates, we are thus led to look for a Hamiltonian structure beyond the local class in the original variables. This is consistent with the results of [18] for the Newell system, where both Hamiltonian operators are nonlocal in these variables. As we will see in the next section, however, the nonlocality of the first of them can be removed by a change of variables.

Nonlocal Hamiltonian structure

Section 2 leaves us with a negative answer in the original variables. Let us now show that the obstruction found there is tied to the choice of variables, and that it can be removed by a nonlocal change of them. A natural candidate is suggested by the Newell system itself: in [5], Newell showed that a phase gauge whose exponent is the potential of the long wave simplifies system (3) considerably [24]. Let WW satisfy Wx=LW_x=L and Wt=2∣S∣2W_t=2\lvert S\rvert^2, as allowed by Eq. (1b), and consider a gauge transformation of the form S=exp⁡(−iγW)S^S=\exp(-\mathrm{i}\gamma W)\hat{S}, with γ\gamma a real parameter. Introducing the transformation in Eq. (1a) yields

iS^t+S^xx−2iγLS^x+i(α−γ)LxS^+(α2−γ2)L2S^−βLS^+2(γ−α)∣S^∣2S^=0.(24)\mathrm{i}\hat{S}_t+\hat{S}_{xx}-2\mathrm{i}\gamma L\hat{S}_x+\mathrm{i}(\alpha-\gamma)L_x\hat{S}+(\alpha^2-\gamma^2)L^2\hat{S}-\beta L\hat{S}+2(\gamma-\alpha)\lvert\hat{S}\rvert^2\hat{S}=0. \tag*{(24)}

Repeating the procedure of Theorem 2.1 with the flow (17) leads to a Schouten bracket that reduces, modulo total derivatives, to a trivector

[P,P]=−2i(α−γ)(3α+γ)∑cyc(θx1)(θ21θ32−θ22θ31).(25)[\mathcal{P},\mathcal{P}]=-2\mathrm{i}(\alpha-\gamma)(3\alpha+\gamma)\sum_{\mathrm{cyc}}(\theta_x^1)(\theta_2^1\theta_3^2-\theta_2^2\theta_3^1). \tag*{(25)}

The root γ=α\gamma=\alpha here is spurious: there, the whole short-wave part of the operator carries the factor α−γ\alpha-\gamma and vanishes, so that P\mathcal{P} reduces to diag⁡(0,0,4α∂x)\operatorname{diag}(0,0,4\alpha\partial_x) and fails to generate the flow. This is precisely the gauge of Newell, in which (24) becomes linear in S^\hat{S}, namely iS^t+S^xx−2iαLS^x−βLS^=0\mathrm{i}\hat{S}_t+\hat{S}_{xx}-2\mathrm{i}\alpha L\hat{S}_x-\beta L\hat{S}=0; the gauge that linearises the short-wave equation is therefore not the one that makes the system Hamiltonian in local form. This leaves γ=−3α\gamma=-3\alpha as the only genuine root within the weight-one class. Let us remark that the restriction on γ\gamma is a restriction on the gauge transformation and not on the YON system: the transformed system (17) with γ=−3α\gamma=-3\alpha can be endowed with a local Hamiltonian structure, and therefore the YON system (1) admits a nonlocal Hamiltonian structure for every α≠0\alpha\ne0 and β\beta via the gauge transformation S=exp⁡(3iαW)S^S=\exp(3\mathrm{i}\alpha W)\hat{S}.

Explicit Hamiltonian structure and hierarchy

Introducing the choice γ=−3α\gamma=-3\alpha in Eq. (24) leads to the system

iS^t+S^xx+6iαLS^x+4iαLxS^−8α2L2S^−βLS^−8α∣S^∣2S^=0,(26)i\hat{S}_{t}+\hat{S}_{xx}+6i\alpha L\hat{S}_{x}+4i\alpha L_{x}\hat{S}-8\alpha^{2}L^{2}\hat{S}-\beta L\hat{S}-8\alpha|\hat{S}|^{2}\hat{S}=0, \tag*{(26)}
Lt=2(∣S^∣2)x.(27)L_{t}=2\left(|\hat{S}|^{2}\right)_{x}. \tag*{(27)}

We can introduce its Hamiltonian structure through the following result.

Theorem 3.1. The skew-adjoint operator

P∗=(0−α∂x−iμ−2iαS^−α∂x+iμ02iαS^∗2iαS^−2iαS^∗∂x),(28)\mathcal{P}_{*}=\begin{pmatrix} 0 & -\alpha\partial_{x}-i\mu& -2i\alpha\hat{S} \\ -\alpha\partial_{x}+i\mu& 0 & 2i\alpha\hat{S}^{*} \\ 2i\alpha\hat{S} & -2i\alpha\hat{S}^{*} & \partial_{x} \end{pmatrix}, \tag*{(28)}

where μ=2α2L+β2\mu=2\alpha^{2}L+\frac{\beta}{2}, is a Hamiltonian operator, and system (19) is its Hamiltonian flow ut=P∗ δH∗u_{t}=\mathcal{P}_{*}\,\delta H_{*}, generated by

H∗=∫[4L∣S^∣2−β2α2∣S^∣2+1αℑ(S^∗S^x)+β2L8α3] dx,(29)H_{*}=\int\left[4L|\hat{S}|^{2}-\frac{\beta}{2\alpha^{2}}|\hat{S}|^{2}+\frac{1}{\alpha}\Im(\hat{S}^{*}\hat{S}_{x})+\frac{\beta^{2}L}{8\alpha^{3}}\right]\,dx, \tag*{(29)}

for every α≠0\alpha\ne0 and β\beta.

The Jacobi identity of P∗\mathcal{P}_{*} can be understood in purely algebraic terms, since the operator is of first order and affine in the fields. Indeed, writing u~=(S^,S^∗,L+β/4α2)\tilde{u}=(\hat{S},\hat{S}^{*},L+\beta/4\alpha^{2}), we have

(P∗)ab=gab ∂x+∑c=13fabcu~c,g=(0−α0−α00001),(30)(\mathcal{P}_{*})_{ab}=g_{ab}\,\partial_{x}+\sum_{c=1}^{3}f_{ab}^{c}\tilde{u}_{c},\qquad g=\begin{pmatrix} 0 & -\alpha& 0 \\ -\alpha& 0 & 0 \\ 0 & 0 & 1 \end{pmatrix}, \tag*{(30)}

where the only nonzero constants fabc=−fbacf_{ab}^{c}=-f_{ba}^{c} are f123=−2iα2f_{12}^{3}=-2i\alpha^{2}, f131=−2iαf_{13}^{1}=-2i\alpha and f232=2iαf_{23}^{2}=2i\alpha. These are the structure constants of the Lie algebra sl(2,C)\mathfrak{sl}(2,\mathbb{C}), and gg is an invariant form on it, namely −1/8α2-1/8\alpha^{2} times its Killing form. Hence, P∗\mathcal{P}_{*} is the Lie–Poisson structure of the centrally extended sl(2)\mathfrak{sl}(2) current algebra [33], and its Jacobi identity reduces to that of the Lie algebra together with the invariance of gg. Three consequences are worth noting. First, the parameter β\beta enters P∗\mathcal{P}_{*} only through the shift of LL in u~\tilde{u}, in agreement with Eq. (14). Second, in the real coordinates (Re⁡S^,Im⁡S^,L)(\operatorname{Re}\hat{S},\operatorname{Im}\hat{S},L) the form gg becomes diag⁡(−α/2,−α/2,1)\operatorname{diag}(-\alpha/2,-\alpha/2,1), so that the real form of the algebra is the compact one, su(2)\mathfrak{su}(2), for α<0\alpha<0, and sl(2,R)\mathfrak{sl}(2,\mathbb{R}) for α>0\alpha>0: the two signs of α\alpha, that is, the two signs σ\sigma of the Newell system (3), correspond to different real forms. Third, the quadratic functional 12∫⟨u~,g−1u~⟩ dx\frac{1}{2}\int\langle\tilde{u},g^{-1}\tilde{u}\rangle\,dx generates the translation, as in the Sugawara construction.

Transforming the local Poisson operator (28) back to the original variables (S,S∗,L)(S,S^{*},L) by means of the gauge transformation, we obtain a nonlocal Poisson structure for the YON system (1), with a weakly nonlocal tail built on the phase symmetry. For β=0\beta=0, this is the first Hamiltonian operator given in [18] for the Newell hierarchy: the modified system of [18, Eq. (2.22)], in the variables (u^,q^,r^)(\hat{u},\hat{q},\hat{r}) and time t2t_{2} used there, is the Newell system under the identification u^=iαL\hat{u}=i\alpha L, q^=aS\hat{q}=aS, r^=2αS∗/a\hat{r}=2\alpha S^{*}/a and t2=2itt_{2}=2it, with aa an arbitrary nonzero constant, and, on covectors that decay at infinity, the operator of [18, Eq. (2.17)] coincides with −1/2α2-1/2\alpha^{2} times the operator P∗\mathcal{P}_{*} carried back to the original variables. The gauge γ=−3α\gamma=-3\alpha thus provides the variables in which that operator becomes local, and Eq. (30) identifies it as a current algebra, whereas in [18] it is reduced to a constant-coefficient operator by two further, non-invertible, Miura-type maps. Together with Eq. (14), this extends the first Hamiltonian structure of the Newell hierarchy to the YON system for every β\beta.

The fact that a Hamiltonian structure can only be achieved after a nonlocal gauge transformation is not exclusive to YON. In fact, the mechanism here is similar to the one at work in the derivative NLS family. The Kaup–Newell, Chen–Lee–Liu and Gerdjikov–Ivanov equations [19, 20, 21] are related by phase gauges whose exponent is the potential of a conserved density [22, 23, 34], in the same way as WW is the potential of the conserved density ρ0=L\rho_0=L in the YON case; there, the gauge parameter changes the Hamiltonian operator and the Hamiltonian together, and it can be chosen according to the member of the family one wants to display. The gauge-equivalent forms of the Newell system, also known as the derivative Yajima–Oikawa system, have themselves been classified into three types in analogy with the derivative NLS family [35]. The key difference in the YON case is that γ\gamma is not a choice: the Jacobi obstruction fixes γ=−3α\gamma=-3\alpha as the only value for which a local Poisson operator exists.

Before studying the hierarchy generated by P∗\mathcal{P}_*, let us fix some notation for its lowest members. We denote by K−1=(iS^,−iS^∗,0)K_{-1}=(i\hat{S},-i\hat{S}^*,0) the phase symmetry, by K0=(S^x,S^x∗,Lx)K_0=(\hat{S}_x,\hat{S}_x^*,L_x) the translation and by K1K_1 the flow (19). In general, P∗\mathcal{P}_* maps a Hamiltonian of weight pp to a flow of weight pp, and the Hamiltonians of these three flows, P∗δHn=Kn\mathcal{P}_*\delta H_n=K_n, are H−1=−∫L/2α dxH_{-1}=-\int L/2\alpha\,\mathrm{d}x, H0=∫(ρ1/α+βρ0/4α2) dxH_0=\int(\rho_1/\alpha+\beta\rho_0/4\alpha^2)\,\mathrm{d}x and H1=H∗H_1=H_*.

Let us now move on to study the degeneracy of the Poisson operator P∗\mathcal{P}_*.

Theorem 3.2 (Non-degeneracy). Provided α≠0\alpha\ne0, the operator P∗\mathcal{P}_* is injective on local covectors, i. e., if the components of gg are differential polynomials in (S^,S^∗,L)(\hat{S},\hat{S}^*,L) and P∗g=0\mathcal{P}_*g=0, then g=0g=0. In particular, P∗\mathcal{P}_* has no nontrivial Casimir at any weight.

Proof. Using Eq. (28), the equation P∗g=0\mathcal{P}_*g=0 reads

αg2,x=−iμg2−2iαS^g3,αg1,x=iμg1+2iαS^∗g3,g3,x=−2iα(S^g1−S^∗g2).(31)\begin{aligned} \alpha g_{2,x} &= -i\mu g_2-2i\alpha\hat{S}g_3, \qquad\alpha g_{1,x}=i\mu g_1+2i\alpha\hat{S}^*g_3, \qquad g_{3,x}=-2i\alpha(\hat{S}g_1-\hat{S}^*g_2). \tag*{(31)} \end{aligned}

Note that each of the three equations features exactly one xx-derivative on the left-hand side and none on the right-hand side.

Let njn_j be the order of gjg_j as a differential polynomial, where j=1,2,3j=1,2,3, and N=max⁡jnjN=\max_j n_j, attained at j=j0j=j_0.

Let us consider the equation corresponding to gj0,xg_{j_0,x}. On the left-hand side, gj0,xg_{j_0,x} is a differential polynomial of order N+1N+1, while the right-hand side only contains the gjg_j’s, of order at most NN. Hence, the coefficients of the terms of order N+1N+1 in gj0,xg_{j_0,x} must vanish, but these are precisely the derivatives of gj0g_{j_0} with respect to the jet variables of order NN, so that gj0g_{j_0} is at most of order N−1N-1, a contradiction. Therefore, no component of gg depends on the fields, that is, gg is constant.

For constant gg, the first two equations in (31) read iμg2+2iαS^g3=0i\mu g_2+2i\alpha\hat{S}g_3=0 and iμg1+2iαS^∗g3=0i\mu g_1+2i\alpha\hat{S}^*g_3=0. Since μ=2α2L+β/2\mu=2\alpha^2L+\beta/2 and α≠0\alpha\ne0, the coefficients of LL give g1=g2=0g_1=g_2=0, and then those of S^\hat{S} give g3=0g_3=0. Hence, g=0g=0.

Since a Casimir is a functional CC satisfying P∗δC=0\mathcal{P}_*\delta C=0, one has δC=0\delta C=0, so CC is the zero functional. □\square

Remark 3.3. The proof divides by α\alpha twice: to solve the first two equations for g1,xg_{1,x} and g2,xg_{2,x}, and through the coefficient 2α22\alpha^2 of LL in μ\mu. For α=0\alpha=0, β≠0\beta\ne0, the condition P∗g=0\mathcal{P}_*g=0 reduces to g1=g2=0g_1=g_2=0, g3,x=0g_{3,x}=0, so that the kernel is spanned by (0,0,1)=δ∫L dx(0,0,1)=\delta\int L\,\mathrm{d}x, and ∫L dx\int L\,\mathrm{d}x becomes a Casimir. This is consistent with P∗δρ0=−2αK−1\mathcal{P}_*\delta\rho_0=-2\alpha K_{-1}, which vanishes at α=0\alpha=0 (see Theorem 3.6). For α=β=0\alpha=\beta=0, when the system reduces to a free Schrödinger equation driving a slaved long wave, the whole short-wave block of P∗\mathcal{P}_* vanishes, and P∗=diag⁡(0,0,∂x)\mathcal{P}_*=\operatorname{diag}(0,0,\partial_x) fails to generate the flow. In fact, no local Hamiltonian operator of weight one or two generates the flow at α=β=0\alpha=\beta=0: the corresponding families have five and six parameters, and their Jacobi conditions are inconsistent.

Theorem 3.2 proves that P∗\mathcal{P}_* is non-degenerate on local covectors. In particular, the Hamiltonian HH solving P∗δH=K\mathcal{P}_*\delta H=K, whenever it exists, is unique for each flow KK.

The space of weight-four symmetries of system (19), obtained by solving the linearised symmetry condition on the graded ansatz, turns out to be four-dimensional. Three of its directions are the lower symmetries multiplied by powers of β\beta, namely β3K−1\beta^3K_{-1}, β2K0\beta^2K_0 and βK1\beta K_1, and the remaining one is the second flow of the hierarchy, K2K_2, whose long-wave component is

K2(L)=−32(βL−α2L2−6α∣S^∣2)Lx+9αL(∣S^∣2)x−3i2(S^xS^∗−S^S^x∗)+14Lxxx.(32)K_2^{(L)}=-\frac{3}{2}\left(\beta L-\alpha^2L^2-6\alpha|\hat{S}|^2\right)L_x+9\alpha L\left(|\hat{S}|^2\right)_x-\frac{3i}{2}\left(\hat{S}_x\hat{S}^*-\hat{S}\hat{S}_x^*\right)+\frac{1}{4}L_{xxx}. \tag*{(32)}

Proposition 3.4. The second flow is Hamiltonian with respect to the same P∗\mathcal{P}_{*}, that is, P∗δH2=K2\mathcal{P}_{*}\delta H_{2}=K_{2}, with

H2=∫[α28L4−β4L3+23α2L2∣S^∣2+12∣S^∣4−18Lx2+1α∣S^x∣2+7Lℑ(S^∗S^x)−βαL∣S^∣2+β24α3∣S^∣2−β2α2ℑ(S^∗S^x)−β316α4L] dx.(33)H_{2}=\int\left[\frac{\alpha^{2}}{8}L^{4}-\frac{\beta}{4}L^{3}+\frac{23\alpha}{2}L^{2}|\hat{S}|^{2}+\frac{1}{2}|\hat{S}|^{4}-\frac{1}{8}L_{x}^{2}+\frac{1}{\alpha}|\hat{S}_{x}|^{2}+7L\Im(\hat{S}^{*}\hat{S}_{x})-\frac{\beta}{\alpha}L|\hat{S}|^{2}+\frac{\beta^{2}}{4\alpha^{3}}|\hat{S}|^{2}-\frac{\beta}{2\alpha^{2}}\Im(\hat{S}^{*}\hat{S}_{x})-\frac{\beta^{3}}{16\alpha^{4}}L\right]\,dx . \tag*{(33)}

whose density is the single new conserved density of weight four.

Conserved densities by asymptotic diagonalisation

We can construct the infinite tower of conserved densities of the system using the Lax pair (4) by the approach of asymptotic diagonalisation. The first step is to construct a recursion of the conserved densities.

Proposition 3.5. Let Γ=ad⁡Σ−1\Gamma=\operatorname{ad}^{-1}_{\Sigma} be the map acting on the space of off-diagonal matrices by (Γ(M))jk=Mjk/(sj−sk)(\Gamma(M))_{jk}=M_{jk}/(s_{j}-s_{k}), with s=(1,0,−1)s=(1,0,-1), as introduced in [6].

We seek a matrix G=I+ΞG=I+\Xi, with Ξ\Xi an off-diagonal matrix of the form Ξ=∑n≥1(iλ)−nΞn\Xi=\sum_{n\geq1}(i\lambda)^{-n}\Xi_{n}, such that G−1(XG−Gx)=DG^{-1}(XG-G_{x})=D is diagonal. By introducing the change of variable ψ=Gφ\psi=G\varphi, that implies that φx=Dφ\varphi_{x}=D\varphi, by direct application of the product rule.

Separating diagonal and off-diagonal part of the relation (XG−Gx)=GD(XG-G_{x})=GD leads to

D=iλΣ+(QΞ)diag,iλ[Σ,Ξ]=Ξx−Q−(QΞ)off+Ξ(QΞ)diag.(34)D=i\lambda\Sigma+(Q\Xi)^{\mathrm{diag}},\qquad i\lambda[\Sigma,\Xi]=\Xi_{x}-Q-(Q\Xi)^{\mathrm{off}}+\Xi(Q\Xi)^{\mathrm{diag}}. \tag*{(34)}

Both equations can be solved order by order by

Ξ1=−Γ(Q),Ξn+1=Γ((Ξn)x−(QΞn)off+∑p+q=nΞp(QΞq)diag).(35)\Xi_{1}=-\Gamma(Q),\qquad\Xi_{n+1}=\Gamma\left((\Xi_{n})_{x}-(Q\Xi_{n})^{\mathrm{off}}+\sum_{p+q=n}\Xi_{p}(Q\Xi_{q})^{\mathrm{diag}}\right). \tag*{(35)}

Since the change of variable GG obtained above also diagonalises the temporal equation, φt=Jφ\varphi_{t}=J\varphi with J=G−1(TG−Gt)J=G^{-1}(TG-G_{t}) diagonal, every diagonal entry of DD is a conserved density of the YON system (1) and its flux is the corresponding entry of JJ.

Note that the recursion here is explicit and can be computed at any given order. In order to pair these densities with the Poisson structure P∗\mathcal{P}_{*}, one needs to carry them into the gauge. Although the gauge map is nonlocal, it acts locally on them: writing S=ES^S=E\hat{S} with Ex=3iαLEE_{x}=3i\alpha LE and ∣E∣=1|E|=1, every xx-derivative of EE is EE times a differential polynomial in LL, and, since the densities are invariant under the phase S↦eiφSS\mapsto e^{i\varphi}S, all the factors of EE cancel out.

Let us also note that the densities obtained by asymptotic diagonalisation coincide with those in Eq. (16), with the caveat that ρ0\rho_{0} and ρ1\rho_{1} do not appear separately: at first order, the diagonal entries of DD are proportional to the combination αρ1−β2ρ0\alpha\rho_{1}-\frac{\beta}{2}\rho_{0}, whereas at second order they are proportional to ρ2\rho_{2} itself.

Noether correspondence

Let us now use P∗\mathcal{P}_{*} to relate the Lie point symmetries of the YON system listed in [24] to its conserved densities. For α≠0\alpha\ne0, these are the translations X1=∂xX_{1}=\partial_{x} and X2=∂tX_{2}=\partial_{t}, the phase rotation X3=−iS∂S+iS∗∂S∗X_{3}=-iS\partial_{S}+iS^{*}\partial_{S^{*}} and the scaling

X4=4t∂t+2x∂x+(−2−iβ2tα2)S∂S+(−2+iβ2tα2)S∗∂S∗−2α2L−βα2∂L,(36)X_{4}=4t\partial_{t}+2x\partial_{x}+\left(-2-\frac{i\beta^{2}t}{\alpha^{2}}\right)S\partial_{S}+\left(-2+\frac{i\beta^{2}t}{\alpha^{2}}\right)S^{*}\partial_{S^{*}}-\frac{2\alpha^{2}L-\beta}{\alpha^{2}}\partial_{L}, \tag*{(36)}

which combines a scaling of xx and tt with a shift of LL by a multiple of β/α2\beta/\alpha^{2} and a time-dependent phase. In fact, X4X_{4} is nothing but the scaling symmetry of the Newell system, 4t∂t+2x∂x−2S∂S−2S∗∂S∗−2L∂L4t\partial_{t}+2x\partial_{x}-2S\partial_{S}-2S^{*}\partial_{S^{*}}-2L\partial_{L}, written in the variables of Eq. (14). For α=0\alpha=0, the symmetry X4X_{4} is replaced by the scaling

X5=4t∂t+2x∂x−3S∂S−3S∗∂S∗−4L∂L.(37)X_{5}=4t\partial_{t}+2x\partial_{x}-3S\partial_{S}-3S^{*}\partial_{S^{*}}-4L\partial_{L}. \tag*{(37)}

and there is an additional generator,

X6=−iβtS∂S+iβtS∗∂S∗+∂L,(38)X_{6}=-i\beta tS\partial_{S}+i\beta tS^{*}\partial_{S^{*}}+\partial_{L}, \tag*{(38)}

which corresponds to a shift of LL combined with a time-dependent phase. However, since the Hamiltonian structure of YO is well known and not the goal of this work, we will focus on the case α≠0\alpha\ne0. For each generator X=τ∂t+ξ∂x+ϕa∂uaX=\tau\partial_{t}+\xi\partial_{x}+\phi_{a}\partial_{u_{a}}, let us denote by KX=−(ϕ−ξux−τut)K_{X}=-(\phi-\xi u_{x}-\tau u_{t}) the flow it generates, so that KX1=K0K_{X_{1}}=K_{0}, KX2=K1K_{X_{2}}=K_{1} and KX3=K−1K_{X_{3}}=K_{-1}. We can gather the Noether correspondence in the following result.

Theorem 3.6. Solving P∗δH=KXP_{*}\delta H=K_{X} on the graded ansatz gives, for the autonomous generators,

X1→H0,X2→H1=H∗,X3→H−1=−∫L2α dx,(39)X_{1}\to H_{0},\qquad X_{2}\to H_{1}=H_{*},\qquad X_{3}\to H_{-1}=-\int\frac{L}{2\alpha}\,dx, \tag*{(39)}

and, in the opposite direction, the three densities in Eq. (16), once carried into the gauge, generate

P∗δρ0=−2αK−1,P∗δρ1=αK0+β2K−1,P∗δρ2=2α2K1+β22K−1,(40)P_{*}\delta\rho_{0}=-2\alpha K_{-1},\qquad P_{*}\delta\rho_{1}=\alpha K_{0}+\frac{\beta}{2}K_{-1},\qquad P_{*}\delta\rho_{2}=2\alpha^{2}K_{1}+\frac{\beta^{2}}{2}K_{-1}, \tag*{(40)}

that is, the three autonomous point symmetries. Additionally, the density

ρ3=tρ2+x(αρ1−β2ρ0)(41)\rho_{3}=t\rho_{2}+x\left(\alpha\rho_{1}-\frac{\beta}{2}\rho_{0}\right) \tag*{(41)}

is conserved, and

P∗δ(2α2∫ρ^3 dx)=KX^4,(42)P_{*}\delta\left(\frac{2}{\alpha^{2}}\int\widehat{\rho}_{3}\,dx\right)=K_{\widehat{X}_{4}}, \tag*{(42)}

where ρ^3\widehat{\rho}_{3} denotes ρ3\rho_{3} carried into the gauge, and

X^4=X4−3βxα(iS^∂S^−iS^∗∂S^∗)(43)\widehat{X}_{4}=X_{4}-\frac{3\beta x}{\alpha}\left(i\widehat{S}\partial_{\widehat{S}}-i\widehat{S}^{*}\partial_{\widehat{S}^{*}}\right) \tag*{(43)}

is the scaling X4X_{4} carried into the gauge, the additional phase arising from the shift of LL through the gauge factor e3iαWe^{3i\alpha W}.

In particular, the conserved quantity ∫ρ0 dx\int\rho_{0}\,dx, with the seemingly trivial density ρ0=L\rho_{0}=L, is actually the Noether charge of the phase symmetry, while ∫ρ1 dx\int\rho_{1}\,dx, with the momentum-like density ρ1\rho_{1}, is, up to a multiple of the phase, the charge of the translation.

A second Hamiltonian structure

Having a non-degenerate first structure at hand, the natural next step is to look for a second one, that is, for a Hamiltonian operator P2P_{2} compatible with P∗P_{*}, [P∗,P2]=0[P_{*},P_{2}]=0, and satisfying a Lenard chain P2δHn=P∗δHn+kP_{2}\delta H_{n}=P_{*}\delta H_{n+k} for some kk. For the Newell hierarchy, a second Hamiltonian operator is given in [18 Eq. (2.18)]. In the chain presented there it generates, from each Hamiltonian, the flow lying two steps below the one generated by the first operator, and its entries involve inverses of differential operators. Here we look instead for the operator lying above P∗P_{*} in the hierarchy, which will turn out to be weakly nonlocal. A first, naïve approach would be to repeat the analysis of § 2 in the gauge variables. This possibility, however, is ruled out: for the weight p=3p=3, which is the one the second structure will turn out to have, the analysis of Theorem 2.4, repeated in the gauge of Theorem 3.1 with the most general target allowed by the grading,

P∗δ∫L dx=c1K1+c0βK0+c−1β2K−1,P_{*}\delta\int L\,dx=c_{1}K_{1}+c_{0}\beta K_{0}+c_{-1}\beta^{2}K_{-1},

leaves only the trivial member β2P∗\beta^{2}P_{*}, so that no local operator can play the role of the second structure. We are thus left with two natural approaches. One can either impose a ‘reasonable’ ansatz for the nonlocality –e.g. a weakly nonlocal tail– and look for a candidate there, which is computationally heavy and may not stumble upon the right operator, or one can try to construct a recursion operator R\mathcal{R} for the symmetries of the system and check whether the composite operator P2=RP∗\mathcal{P}_2=\mathcal{R}\mathcal{P}_* works as a second Hamiltonian structure. Here, we will take the second approach.

As in the YO case, and in agreement with the chain of [18], the YON system has no recursion operator taking each flow to the next one, RKn=Kn+1\mathcal{R}K_n=K_{n+1}; in fact, no such operator of weight one exists, as we will see in § 5. The recursion operator advances the hierarchy by two steps, RKn=Kn+2\mathcal{R}K_n=K_{n+2}, and has the form

R=(∂x2+6iαL∂x−Θ+4iαLx6αS^23iαS^∂x−34(β+24α2L)S^+132iαS^x6αS^∗2∂x2−6iαL∂x−Θ−4iαLx−3iαS^∗∂x−34(β+24α2L)S^∗−132iαS^x∗−32iS^∗∂x+3αLS^∗−12iS^x∗32iS^∂x+3αLS^+12iS^x14∂x2+α2L2+10α∣S^∣2−βL)(44)\mathcal{R}= \begin{pmatrix} \partial_x^2+6\mathrm{i}\alpha L\partial_x-\Theta+4\mathrm{i}\alpha L_x & 6\alpha\hat{S}^{2} & 3\mathrm{i}\alpha\hat{S}\partial_x-\frac{3}{4}(\beta+24\alpha^2L)\hat{S}+\frac{13}{2}\mathrm{i}\alpha\hat{S}_x \\ 6\alpha\hat{S}^{*2} & \partial_x^2-6\mathrm{i}\alpha L\partial_x-\Theta-4\mathrm{i}\alpha L_x & -3\mathrm{i}\alpha\hat{S}^{*}\partial_x-\frac{3}{4}(\beta+24\alpha^2L)\hat{S}^{*}-\frac{13}{2}\mathrm{i}\alpha\hat{S}^{*}_x \\ -\frac{3}{2}\mathrm{i}\hat{S}^{*}\partial_x+3\alpha L\hat{S}^{*}-\frac{1}{2}\mathrm{i}\hat{S}^{*}_x & \frac{3}{2}\mathrm{i}\hat{S}\partial_x+3\alpha L\hat{S}+\frac{1}{2}\mathrm{i}\hat{S}_x & \frac{1}{4}\partial_x^2+\alpha^2L^2+10\alpha|\hat{S}|^2-\beta L \end{pmatrix} \tag*{(44)}

where Θ=βL+8α2L2+8α∣S^∣2\Theta=\beta L+8\alpha^2L^2+8\alpha|\hat{S}|^2, plus a nonlocal tail of nineteen legs of the form vi⊗∂x−1∘wiv_i\otimes\partial_x^{-1}\circ w_i, of weight at most three. Here and in what follows, ∂x−1\partial_x^{-1} denotes the skew-adjoint inverse of ∂x\partial_x,

(∂x−1f)(x)=12(∫−∞xf(y) dy−∫x∞f(y) dy)(45)(\partial_x^{-1}f)(x)=\frac{1}{2}\left(\int_{-\infty}^{x}f(y)\,\mathrm{d}y-\int_{x}^{\infty}f(y)\,\mathrm{d}y\right) \tag*{(45)}

so that ∂x∘∂x−1=1\partial_x\circ\partial_x^{-1}=1, whereas ∂x−1∘∂x=1\partial_x^{-1}\circ\partial_x=1 holds on functions that vanish at infinity, but not on constants. As will also happen for the composite operator below, only three of these legs are independent, and they are generated once again by the point symmetries of the system; the tail is given in closed form in Eq. (54) below.

Invariant form for the nonlocal tail

Composing P2=RP∗\mathcal{P}_2=\mathcal{R}\mathcal{P}_* allows us to write the tail as nineteen legs of the form ∑vi∂x−1wi\sum v_i\partial_x^{-1}w_i. However, this presentation is not invariant, as a decomposition into legs can be performed arbitrarily; only the tensor T=∑ivi⊗wi\mathcal{T}=\sum_i v_i\otimes w_i is. Its antisymmetric part vanishes in an exact monomial decomposition, which ensures the skew-adjointness of the nonlocal part of P2\mathcal{P}_2, since the adjoint of v∂x−1∘wv\partial_x^{-1}\circ w is −w∂x−1∘v-w\partial_x^{-1}\circ v. Moreover, it verifies the following remarkable property.

Proposition 4.1. The tail tensor of P2=RP∗\mathcal{P}_2=\mathcal{R}\mathcal{P}_*, T=∑ivi⊗wi\mathcal{T}=\sum_i v_i\otimes w_i, has rank 3, so that the nineteen legs of the composition collapse into three. A basis of its column space is

u1=(S^,−S^∗,0),u2=i8αK1,u3=(S^x,S^x∗,Lx)(46)u_1=(\hat{S},-\hat{S}^{*},0),\qquad u_2=\frac{\mathrm{i}}{8\alpha}K_1,\qquad u_3=(\hat{S}_x,\hat{S}^{*}_x,L_x) \tag*{(46)}

that is, it is generated precisely by the phase symmetry, the flow itself and the translation symmetry. The tail is then exactly

∑ivi∂x−1wi=∑k,l=13ckluk∂x−1ul(47)\sum_i v_i\partial_x^{-1}w_i=\sum_{k,l=1}^{3}c_{kl}u_k\partial_x^{-1}u_l \tag*{(47)}

with cc the symmetric matrix

c=(−3β2448α3−3iαβ248α300−3iαβ20−α2).(48)c= \begin{pmatrix} -\frac{3\beta^2}{4} & 48\alpha^3 & -\frac{3\mathrm{i}\alpha\beta}{2} \\ 48\alpha^3 & 0 & 0 \\ -\frac{3\mathrm{i}\alpha\beta}{2} & 0 & -\alpha^2 \end{pmatrix}. \tag*{(48)}

Together with the local part,

(P2)11=−12α2S^S^x−12α2S^2∂x,(49)(\mathcal{P}_2)_{11}=-12\alpha^2\hat{S}\hat{S}_x-12\alpha^2\hat{S}^2\partial_x, \tag*{(49)}
(P2)12=−α∂x3−i(16α2L+β)2∂x2+α(20α2L2+14α∣S^∣2+4βL−8iαLx)∂x+16iα4L3+6iα2βL2+L(20α3Lx+76iα3∣S^∣2+iβ22)+2α(7αS^∗S^x+βLx+2iβ∣S^∣2−iαLxx),(50)\begin{aligned} (\mathcal{P}_2)_{12}={}&-\alpha\partial_x^3-\frac{\mathrm{i}(16\alpha^2L+\beta)}{2}\partial_x^2+\alpha(20\alpha^2L^2+14\alpha|\hat{S}|^2+4\beta L-8\mathrm{i}\alpha L_x)\partial_x\\ &+16\mathrm{i}\alpha^4L^3+6\mathrm{i}\alpha^2\beta L^2+L\left(20\alpha^3L_x+76\mathrm{i}\alpha^3|\hat{S}|^2+\frac{\mathrm{i}\beta^2}{2}\right)\\ &+2\alpha(7\alpha\hat{S}^*\hat{S}_x+\beta L_x+2\mathrm{i}\beta|\hat{S}|^2-\mathrm{i}\alpha L_{xx}), \tag*{(50)} \end{aligned}
(P2)13=iαS^∂x2+[52iαS^x−34(8α2L+β)S^]∂x+iαS^xx−8iα3L2S^−20iα2S^2S^∗−5α2LS^x+12iαβLS^−β2S^x−4α2S^Lx,(51)\begin{aligned} (\mathcal{P}_2)_{13}={}&\mathrm{i}\alpha\hat{S}\partial_x^2+\left[\frac{5}{2}\mathrm{i}\alpha\hat{S}_x-\frac{3}{4}(8\alpha^2L+\beta)\hat{S}\right]\partial_x+\mathrm{i}\alpha\hat{S}_{xx}-8\mathrm{i}\alpha^3L^2\hat{S}\\ &-20\mathrm{i}\alpha^2\hat{S}^2\hat{S}^*-5\alpha^2L\hat{S}_x+\frac{1}{2}\mathrm{i}\alpha\beta L\hat{S}-\frac{\beta}{2}\hat{S}_x-4\alpha^2\hat{S}L_x, \tag*{(51)} \end{aligned}
(P2)33=14∂x3+(α2L2−βL+4α∣S^∣2)∂x+(α2L−β2)Lx+2α(∣S^∣2)x,(52)(\mathcal{P}_2)_{33}=\frac{1}{4}\partial_x^3+(\alpha^2L^2-\beta L+4\alpha|\hat{S}|^2)\partial_x+\left(\alpha^2L-\frac{\beta}{2}\right)L_x+2\alpha(|\hat{S}|^2)_x, \tag*{(52)}

whose remaining entries are fixed by skew-adjointness and the reality condition, the tail fully characterises the operator P2\mathcal{P}_2.

The three legs found in Proposition 4.1 make P2\mathcal{P}_2 a weakly nonlocal structure in the sense of Ferapontov, Maltsev and Novikov [25, 26, 27]. That is not by chance: as conjectured by Maltsev and Novikov and later proved by Sergyeyev [36], the higher Hamiltonian, symplectic and recursion operators of a (1+1)(1+1)-dimensional integrable system are weakly nonlocal. In the present case, the three legs moreover commute pairwise and are Hamiltonian with respect to P∗\mathcal{P}_*, namely u1=−iK−1=P∗δ(−iH−1)u_1=-\mathrm{i}K_{-1}=\mathcal{P}_*\delta(-\mathrm{i}H_{-1}), u2=i8αK1=P∗δ(i8αH1)u_2=\frac{\mathrm{i}}{8\alpha}K_1=\mathcal{P}_*\delta(\frac{\mathrm{i}}{8\alpha}H_1) and u3=K0=P∗δH0u_3=K_0=\mathcal{P}_*\delta H_0, so that P2\mathcal{P}_2 has the canonical form local+∑k,lcklKk∂x−1Kl\mathrm{local}+\sum_{k,l}c_{kl}K_k\partial_x^{-1}K_l, with commuting Hamiltonian symmetries as legs. This is the same shape as the second Hamiltonian structure of the NLS equation, whose tail is (q,−r)∂x−1(q,−r)(q,-r)\partial_x^{-1}(q,-r).

The tail of the recursion operator takes the same form. Denoting by ηl\eta_l the gradients of the Hamiltonians of the legs, P∗ηl=ul\mathcal{P}_*\eta_l=u_l, that is,

η1=δ(−iH−1)=(0,0,i2α),η2=i8αδH1,η3=δH0,(53)\eta_1=\delta(-\mathrm{i}H_{-1})=\left(0,0,\frac{\mathrm{i}}{2\alpha}\right),\qquad\eta_2=\frac{\mathrm{i}}{8\alpha}\delta H_1,\qquad\eta_3=\delta H_0, \tag*{(53)}

the recursion operator reads

R=Rloc−∑k,l=13ckluk⊗∂x−1∘ηl,(54)\mathcal{R}=\mathcal{R}_{\mathrm{loc}}-\sum_{k,l=1}^{3}c_{kl}u_k\otimes\partial_x^{-1}\circ\eta_l, \tag*{(54)}

with Rloc\mathcal{R}_{\mathrm{loc}} the local part displayed above and cc the matrix of Proposition 4.1. Note that the pairing between legs and gradients is not diagonal: for instance, the leg of the flow, u2u_2, is paired only with η1\eta_1, the gradient of ∫L dx\int L\,\mathrm{d}x.

In the YON case, however, the nonlocal tail is not generated by arbitrary commuting symmetries, but by the three autonomous Lie point symmetries of the system: the phase, the translation and the flow itself. Though remarkable, this property is not exclusive of this system: tails built from point symmetries also occur for certain Boussinesq-type systems [37], for the NLS equation written in real components, where the leg is the phase rotation [38], and for the group-invariant NLS systems on Hermitian symmetric spaces, where the tail is built from the action of the isotropy group [39].

Compatibility of the Hamiltonian structures

Let us now move on to check the compatibility of the structures P∗\mathcal{P}_* and P2\mathcal{P}_2, that is, that P∗+λP2\mathcal{P}_*+\lambda\mathcal{P}_2 is a Hamiltonian operator for every λ\lambda. Since the Schouten bracket is quadratic in the operator, and P∗\mathcal{P}_* and P2\mathcal{P}_2 are Hamiltonian separately (Theorem 3.1 and Theorem 4.3 below), what is left is the term linear in λ\lambda, the mixed bracket

[P∗,P2]=∑cyc(⟨θ1,P∗′[P2θ2]θ3⟩+⟨θ1,P2′[P∗θ2]θ3⟩),(55)[\mathcal{P}_*,\mathcal{P}_2]=\sum_{\mathrm{cyc}}\left(\left\langle\theta^1,\mathcal{P}'_*\left[\mathcal{P}_2\theta^2\right]\theta^3\right\rangle+\left\langle\theta^1,\mathcal{P}'_2\left[\mathcal{P}_*\theta^2\right]\theta^3\right\rangle\right), \tag*{(55)}

which has to be a total derivative. Now, with P2\mathcal{P}_2 being nonlocal, the mixed bracket, which we denote by Ψ\Psi, contains potentials ∂x−1(…)\partial_x^{-1}(\ldots). Since the Euler operator does not act on such expressions, the condition Ψ≡0\Psi\equiv0 has to be read in integrated form, ∫Ψ dx=0\int\Psi\,dx=0 for all covectors.

In the first term of Eq. (55), we have that P∗\mathcal{P}_* is local, and the Fréchet derivative P∗′\mathcal{P}_*' is acting on the nonlocal vector P2θ2\mathcal{P}_2\theta^2, which, due to the properties of P2\mathcal{P}_2 in the previous subsection, has a tail ∑k,lckluk∂x−1⟨ul,θ2⟩\sum_{k,l}c_{kl}u_k\partial_x^{-1}\langle u_l,\theta^2\rangle. Since the coefficients of P∗\mathcal{P}_* depend on S^\hat{S}, S^∗\hat{S}^* and LL but not on their derivatives, the potentials enter bare.

In the second term, the direction P∗θ2\mathcal{P}_*\theta^2 is local, and differentiating the tail of P2\mathcal{P}_2 gives ckl(uk′[P∗θ2]∂x−1⟨ul,θ3⟩+uk∂x−1⟨ul′[P∗θ2],θ3⟩)c_{kl}\left(u'_k[\mathcal{P}_*\theta^2]\partial_x^{-1}\langle u_l,\theta^3\rangle+u_k\partial_x^{-1}\langle u'_l[\mathcal{P}_*\theta^2],\theta^3\rangle\right), whose arguments are again local. That means that two different kinds of potential occur:

Nlm=∂x−1⟨ul,θm⟩,Nlbc=∂x−1⟨ul′[P∗θb],θc⟩,(56)N_{lm}=\partial_x^{-1}\langle u_l,\theta^m\rangle,\qquad N_{lbc}=\partial_x^{-1}\langle u'_l[\mathcal{P}_*\theta^b],\theta^c\rangle, \tag*{(56)}

where l=1,2,3l=1,2,3 labels the legs, m=1,2,3m=1,2,3 the covectors and (b,c)(b,c) runs over the cyclic pairs (2,3)(2,3), (3,1)(3,1) and (1,2)(1,2), for a total of nine potentials of each kind. Then, the bracket can be written as Ψ=Ψ0+∑νΦνNν\Psi=\Psi_0+\sum_\nu\Phi_\nu N_\nu, with Ψ0\Psi_0 and Φν\Phi_\nu all local. Note that no product of two potentials occurs in any of the terms.

It would be too strong to demand each individual term Φν\Phi_\nu to be a total derivative—here, that fails for fifteen of the eighteen terms—because distinct nonlocal terms cancel each other. On local functions, the form

ω(f,g)=∫f∂x−1g dx(57)\omega(f,g)=\int f\partial_x^{-1}g\,dx \tag*{(57)}

is antisymmetric, ω(f,g)=−ω(g,f)\omega(f,g)=-\omega(g,f), and it is local as soon as either argument is a total derivative, ω(∂xF,g)=−∫Fg dx\omega(\partial_xF,g)=-\int Fg\,dx and ω(f,∂xG)=∫fG dx\omega(f,\partial_xG)=\int fG\,dx. Denoting bν=∂xNνb_\nu=\partial_xN_\nu, the Schouten relation can be written as

∫Ψ dx=∫Ψ0 dx+Ω,Ω=∑νω(Φν,bν),(58)\int\Psi\,dx=\int\Psi_0\,dx+\Omega,\qquad\Omega=\sum_\nu\omega(\Phi_\nu,b_\nu), \tag*{(58)}

and the nonlocal content of Ω\Omega can be written as

∑νq(Φν)∧q(bν)∈Λ2(W/(W∩Im⁡∂x)),W=span⁡{Φν,bν},(59)\sum_\nu q(\Phi_\nu)\wedge q(b_\nu)\in\Lambda^2\left(W/(W\cap\operatorname{Im}\partial_x)\right),\qquad W=\operatorname{span}\{\Phi_\nu,b_\nu\}, \tag*{(59)}

where qq is the projection to the quotient and where being in Im⁡∂x\operatorname{Im}\partial_x (that is, being a total derivative when considering a local expression) is decided by the Euler operator. Checking symbolically over a monomial basis of 405 elements, of which 93 independent combinations are total derivatives, it is found that

∑νq(Φν)∧q(bν)=0(60)\sum_\nu q(\Phi_\nu)\wedge q(b_\nu)=0 \tag*{(60)}

for every α\alpha and β\beta. That means that the nonlocal part cancels exactly. Additionally, by direct application of the Euler operator the remaining local part is checked to be a total derivative. Therefore, we have the following:

Theorem 4.2. [P∗,P2]=0[\mathcal{P}_*,\mathcal{P}_2]=0 modulo total derivatives for every choice of α\alpha and β\beta.

Jacobi identity

The only step left is to check that P2\mathcal{P}_2 is indeed a Poisson operator by verifying it satisfies the Schouten criterion of [P2,P2][\mathcal{P}_2,\mathcal{P}_2] being a total derivative. As opposed to the computation for compatibility, here the nonlocal tail enters the computation twice: both in the direction P2θ2\mathcal{P}_2\theta^2 and in the Fréchet derivative P2′\mathcal{P}_2', so nested potentials ∂x−1[…∂x−1(…)]\partial_x^{-1}[\ldots\partial_x^{-1}(\ldots)] appear, making it necessary to use a different machinery.

Writing P2=P2loc+P2tail\mathcal{P}_2=\mathcal{P}_2^{\mathrm{loc}}+\mathcal{P}_2^{\mathrm{tail}}, with P2loc\mathcal{P}_2^{\mathrm{loc}} the local part and P2tail\mathcal{P}_2^{\mathrm{tail}} the nonlocal tail, we have

[P2,P2]=[P2loc,P2loc]+[P2loc,P2tail]+[P2tail,P2tail],(61)[\mathcal{P}_2,\mathcal{P}_2]=[\mathcal{P}_2^{\mathrm{loc}},\mathcal{P}_2^{\mathrm{loc}}]+[\mathcal{P}_2^{\mathrm{loc}},\mathcal{P}_2^{\mathrm{tail}}]+[\mathcal{P}_2^{\mathrm{tail}},\mathcal{P}_2^{\mathrm{tail}}], \tag*{(61)}

where [P2loc,P2tail][P_2^{\mathrm{loc}},P_2^{\mathrm{tail}}] is the mixed bracket (44) with (P∗,P2)(P_*,P_2) replaced by (P2loc,P2tail)(P_2^{\mathrm{loc}},P_2^{\mathrm{tail}}). The first term is purely local, and the second one is free of nested potentials because P2locP_2^{\mathrm{loc}} is local, exactly the situation of Section 4.2. Neither of them vanishes on its own—in particular, the local part P2locP_2^{\mathrm{loc}} alone is not a Poisson operator—and all the difficulty is confined to [P2tail,P2tail][P_2^{\mathrm{tail}},P_2^{\mathrm{tail}}], which we treat together with them.

The term that needs a new argument is then [P2tail,P2tail][P_2^{\mathrm{tail}},P_2^{\mathrm{tail}}]. However, the nesting of potentials here can be removed by using the properties of the form ω\omega introduced before. Given a local Φ\Phi and any argument XX, we have that

∫Φ(∂x−1X) dx=−∫(∂x−1Φ)X dx,(62)\int\Phi(\partial_x^{-1}X)\,dx=-\int(\partial_x^{-1}\Phi)X\,dx, \tag*{(62)}

and, in particular, if XX is a nested potential, this transforms it into a computation with two first-level potentials. Applied to the nonlocal tails in [P2tail,P2tail][P_2^{\mathrm{tail}},P_2^{\mathrm{tail}}], this property removes all nine nested potentials, leaving a polynomial of degree two in first-level potentials. Rewriting the 27 resulting potentials over a basis of their arguments—legitimate since ∂x−1\partial_x^{-1} is linear—leaves 18 independent ones, and all 171 coefficients of degree two vanish identically. Note that this is not a feature of the method employed but a remarkable property of the operator itself: at degree two there is no antisymmetry to exploit, since NjNkN_jN_k is symmetric, so nothing imposes that these coefficients must vanish.

What remains is linear in the potentials, Ψ0+∑jΦjNj\Psi_0+\sum_j\Phi_jN_j with ∂xNj=bj\partial_xN_j=b_j local, and the argument of Eq. (60) applies verbatim, with W=span⁡{Φj,bj}W=\operatorname{span}\{\Phi_j,b_j\} of dimension 27, of which 9 independent combinations are total derivatives. Checking symbolically again, it is found that

∑jq(Φj)∧q(bj)=0(63)\sum_j q(\Phi_j)\wedge q(b_j)=0 \tag*{(63)}

for every α\alpha and β\beta, so ∫Ψ dx\int\Psi\,dx reduces to the integral of an explicit local object and the Euler operator annihilates it. With that, we have the following:

Theorem 4.3. [P2,P2]=0[P_2,P_2]=0 modulo total derivatives for every choice of α\alpha and β\beta.

Together with Theorems 3.1 and 4.2, this proves the following.

Corollary 4.4. (P∗,P2)(P_*,P_2) is a Hamiltonian pair for every α≠0\alpha\ne0 and every β\beta.

Lenard–Magri chain

Now that we have found a bi-Hamiltonian structure for the system, we can use it to study the hierarchy it generates by using the theory of Lenard–Magri chains [40, 41, 42]. Since P2=RP∗P_2=RP_*, one would expect the recursion operator to build the whole hierarchy through the chain

P2δHn=RP∗δHn=RKn=Kn+2=P∗δHn+2,(64)P_2\delta H_n=RP_*\delta H_n=RK_n=K_{n+2}=P_*\delta H_{n+2}, \tag*{(64)}

so that each Kn+2K_{n+2} would be Hamiltonian with respect to both P∗P_* and P2P_2. Upon deeper inspection, however, it turns out that relation (64) is a simplification of the behaviour of the hierarchy and, strictly speaking, not its actual behaviour. On covectors whose components decay at infinity, the composition does give P2=RP∗P_2=RP_*, but the composition that produced P2P_2 used that ∂x−1∘∂x=1\partial_x^{-1}\circ\partial_x=1, which is not true on constants, and δ∫L dx=(0,0,1)\delta\int L\,dx=(0,0,1) is constant. There,

P2δ∫L dx=αK1−β2K0,RP∗δ∫L dx=−2αK1.(65)P_2\delta\int L\,dx=\alpha K_1-\frac{\beta}{2}K_0,\qquad RP_*\delta\int L\,dx=-2\alpha K_1. \tag*{(65)}

The first one is the genuine Hamiltonian vector field of ∫L dx\int L\,dx with respect to P2P_2, and the bracket {F,∫L dx}P2\{F,\int L\,dx\}_{P_2} is antisymmetric. The same caveat applies to the first structure when written in nonlocal form in the original variables: on δ∫L dx\delta\int L\,dx, the operator of [18 Eq. (2.17)] yields −1/2-1/2 times the vector field obtained from the local operator P∗P_*, in the normalisation in which the two agree on every covector that decays at infinity. In the gauge variables, P∗P_* being local, no such ambiguity arises. Now, each HnH_n of the indexing of Section 3 is a part with decaying gradient plus κn∫L dx\kappa_n \int L\,dx, with κ−1=−1/2α\kappa_{-1}=-1/2\alpha, κ0=β/4α2\kappa_0=\beta/4\alpha^2 and κ1=β2/8α3\kappa_1=\beta^2/8\alpha^3. Since P∗δ∫L dx=−2αK−1P_*\delta\int L\,dx=-2\alpha K_{-1} and RK−1=K1R K_{-1}=K_1, the pair actually yields, for every nn,

P2δHn=Kn+2+κn(3αK1−β2K0),Kn+2=P∗δHn+2,(66)P_2\delta H_n=K_{n+2}+\kappa_n\left(3\alpha K_1-\frac{\beta}{2}K_0\right),\qquad K_{n+2}=P_*\delta H_{n+2}, \tag*{(66)}

so the correction lies in the span of flows already present. That means that the Lenard chains of the explicit pair are triangular recombinations of the HnH_n. Starting from the two simplest conserved quantities, ∫ρ0 dx\int\rho_0\,dx and ∫ρ1 dx\int\rho_1\,dx (both of them invariant under the gauge, ρ0\rho_0 trivially and ρ1\rho_1 because it carries no derivatives), we get

A1=∫ρ0 dx,Ak=αH2k−3−β2H2k−4 (k≥2),Bk=αH2k−2+β2H2k−3 (k≥1),(67)A_1=\int\rho_0\,dx,\qquad A_k=\alpha H_{2k-3}-\frac{\beta}{2}H_{2k-4}\ (k\ge2),\qquad B_k=\alpha H_{2k-2}+\frac{\beta}{2}H_{2k-3}\ (k\ge1), \tag*{(67)}

with P∗δAk+1=P2δAkP_*\delta A_{k+1}=P_2\delta A_k, P∗δBk+1=P2δBkP_*\delta B_{k+1}=P_2\delta B_k and B1=∫ρ1 dxB_1=\int\rho_1\,dx. Note that every AkA_k with k≥2k\ge2 and every BkB_k has a decaying gradient, so beyond the first step of the first chain only the regime P2=RP∗P_2=RP_* is involved. The content of the chains is therefore that each RKnRK_n is again Hamiltonian with respect to P∗P_*, and later on we will show that it has a local Hamiltonian. Also, since the tail arguments ⟨ul,δF⟩\langle u_l,\delta F\rangle are total derivatives at every step, P2δFP_2\delta F is local despite the operator being nonlocal.

Let us remark that Eq. (66) also determines which flows are Hamiltonian with respect to both structures. Since ∫ρ0 dx\int\rho_0\,dx is the only conserved quantity of weight one, Eq. (65) implies that, for β≠0\beta\ne0, no local functional HH satisfies P2δH=K1P_2\delta H=K_1: what the pair generates at weight three is the combination P∗δA2=P2δA1=αK1−β2K0P_*\delta A_2=P_2\delta A_1=\alpha K_1-\frac{\beta}{2}K_0. As the translation K0K_0 commutes with K1K_1 and neither structure depends explicitly on xx or tt, this combination is the YON system written in a moving frame: if uu solves Eqs. (19), then v(x,t)=u(x−βt/2α,t)v(x,t)=u(x-\beta t/2\alpha,t) solves

vt=K1−β2αK0=P∗δ(A2α)=P2δ(A1α),(68)v_t=K_1-\frac{\beta}{2\alpha}K_0=P_*\delta\left(\frac{A_2}{\alpha}\right)=P_2\delta\left(\frac{A_1}{\alpha}\right), \tag*{(68)}

so that the YON system is bi-Hamiltonian up to a Galilean change of frame. For β=0\beta=0, that is, for the Newell system, the frame is at rest, and K1=P2δ(∫L dx/α)K_1=P_2\delta(\int L\,dx/\alpha) directly.

Note also that the step in the recursion is double: it advances the hierarchy by two, splitting it into two interleaved families, the orbits of K−1K_{-1} and of K0K_0 under RR, with Hamiltonians {H−1,H1,H3,…}\{H_{-1},H_1,H_3,\ldots\} and {H0,H2,H4,…}\{H_0,H_2,H_4,\ldots\}. The chains (56) recombine them with β\beta-multiples of lower members, but each of them still contributes one new conserved quantity every other weight, so two seeds are needed either way. This splitting cannot be undone by taking the square root of RR, as no weakly nonlocal operator of weight one maps the flows to symmetries while advancing the hierarchy by a single step. The reason for this is apparent: weight and charge would force such an operator to be of first order with vanishing (1,2)(1,2) and (2,1)(2,1) entries, since no monomial of weight one carries charge ±2\pm2. Applying it to the phase symmetry K−1K_{-1}, which has no long-wave component, it can only produce a multiple of ∣S^∣2|\hat S|^2 in the long-wave slot, never the needed LxL_x. A graded search over all admissible weakly nonlocal operators confirms that an operator achieving the step K0↦K1K_0\mapsto K_1 is attainable, but it cannot send K−1K_{-1} and K1K_1 to symmetries.

Furthermore, for α≠0\alpha\ne0 the operator P∗P_* has no Casimirs (Theorem 3.2). Hence, the hierarchy needs pre-existing conserved quantities to use as seeds, as opposed to the usual way of generating the chain using Casimirs as a seed [43]. Even for α=0\alpha=0, only one Casimir of Q0Q_0 is available, ∫L dx\int L\,dx, so one of the two families has to be generated using a conserved quantity anyway.

None of these properties is unique to YON on its own. In the original formulations of the chain [40, 44, 45], one of the structures is taken to be symplectic and the seed is supplied from an external source [43]. A well-known example of this is the NLS equation [40]: its canonical structure is constant and invertible, so it has no Casimirs, and its chain starts from the mass ∫∣q∣2 dx\int|q|^2\,dx, the Hamiltonian of the phase rotation, the role played here by H−1H_{-1}. In cases where no recursion operator is available, a master symmetry plays the role of generating mechanism, although it also needs a starting flow [46]. Having a recursion that advances more than one step at a time is not exclusive to YON either. Without gaps in the hierarchy, it happens for the Yajima–Oikawa system itself, read as the 2-constrained KP hierarchy, where the odd and even flows come with separate seeds [16, 47, 17]. With gaps, it happens whenever the step is the order of a Lax operator [48], and it is the case for the Sawada–Kotera and Kaup–Kupershmidt equations, bi-Hamiltonian in the generalised sense of [49], whose hierarchies contain only the flows of orders 1, 5, 7, 11, 13, . . ., so that the chains 1 → 7 → 13 and 5 → 11 → 17 run in parallel [49, 50]. Several seeds, with commutativity across their orbits enforced by conditions on the seeds rather than by Casimirs, are treated in [51] for the two-dimensional periodic Volterra chain, and an even more extreme case is a Burgers-type system with infinitely many local higher symmetries but no recursion operator nor master symmetry [52], where the symmetries are produced by a nonlocal two-term recursion relation. Against this background, the double step of the YON hierarchy is inherited from its Yajima–Oikawa limit, and it is already present in the chain given in [18] for the Newell hierarchy; what the deformation by α\alpha changes is the anchor: at α=0\alpha= 0 the first structure has the Casimir ∫L dx\int L\,dx (Remark 3.3), which is lost for α≠0\alpha\ne0 while the double step survives. What we add here is that, for α≠0\alpha\ne0, P∗\mathcal{P}_* has no Casimirs and R\mathcal{R} admits no weakly nonlocal square root, so that, the hierarchy having no gaps, the two seeds are necessary and are not a byproduct of the construction.

Let us now look at the Lenard chain used as a generator, and check whether it returns the tower obtained via asymptotic diagonalisation before. Indeed, it does return the same conserved quantities and slightly more. One step of the process consists of constructing K′=P2δHK' = \mathcal{P}_2\delta H and then solving P∗δH′=K′\mathcal{P}_*\delta H' = K', the second part having at most one solution up to a constant, since P∗\mathcal{P}_* has no Casimirs. Starting from the two seeds, this returns the two interleaved families and, with them, the conserved densities obtained via asymptotic diagonalisation, plus a single new one: ρ0=L\rho_0 = L itself, which the Lax recursion does not return on its own but only inside the combination αρ1−β2ρ0\alpha\rho_1 - \frac{\beta}{2}\rho_0.

Let us now recall the closed form (54) of the recursion operator, which we write as

R=Rloc+∑b=13Gb⊗∂x−1∘γb,Gb=ub,γb=−∑l=13cblηl.(69)\mathcal{R} = \mathcal{R}_{\mathrm{loc}} + \sum_{b=1}^{3} G_b \otimes\partial_x^{-1} \circ\gamma_b,\qquad G_b = u_b,\qquad\gamma_b = -\sum_{l=1}^{3} c_{bl}\eta_l. \tag*{(69)}

Its three legs are thus generated by the commuting point symmetries,

G1=(S^,−S^∗,0),G2=i8αK1,G3=(S^x,S^x∗,Lx),(70)G_1 = (\hat{S},-\hat{S}^{*},0),\qquad G_2 = \frac{\mathrm{i}}{8\alpha}K_1,\qquad G_3 = (\hat{S}_x,\hat{S}^{*}_x,L_x), \tag*{(70)}

while the γb\gamma_b are gradients of conserved quantities, namely linear combinations of δH−1\delta H_{-1}, δH0\delta H_0 and δH1\delta H_1. The operator carries no explicit xx, and each of its pieces –the local entries, the legs GbG_b and the γb\gamma_b– is U(1)U(1)-covariant. Hence, LK0R=LK−1R=0\mathcal{L}_{K_0}\mathcal{R} = \mathcal{L}_{K_{-1}}\mathcal{R} = 0.

The involution of the conserved quantities of the chains does not have to be checked order by order: it follows from the pair. Let FiF_i and F~j\widetilde{F}_j be two members of the chains. The skew-adjointness and the chain itself give

{Fi,F~j}P∗=∫⟨δFi,P∗δF~j⟩ dx=∫⟨δFi,P2δF~j−1⟩ dx=−∫⟨δF~j−1,P2δFi⟩ dx={Fi+1,F~j−1}P∗,(71)\begin{aligned} \{F_i,\widetilde{F}_j\}_{\mathcal{P}_*} &= \int\langle\delta F_i,\mathcal{P}_*\delta\widetilde{F}_j\rangle\,dx = \int\langle\delta F_i,\mathcal{P}_2\delta\widetilde{F}_{j-1}\rangle\,dx \\ &= -\int\langle\delta\widetilde{F}_{j-1},\mathcal{P}_2\delta F_i\rangle\,dx = \{F_{i+1},\widetilde{F}_{j-1}\}_{\mathcal{P}_*}, \tag*{(71)} \end{aligned}

so the bracket depends only on i+ji+j and is invariant under (i,j)↦(i+1,j−1)(i,j) \mapsto(i+1,j-1). Within one chain, the telescope terminates either at {Fk,Fk}=0\{F_k,F_k\}=0 or at {Fk,Fk+1}=∫⟨δFk,P2δFk⟩ dx=0\{F_k,F_{k+1}\} = \int\langle\delta F_k,\mathcal{P}_2\delta F_k\rangle\,dx = 0. Any bracket including A1=∫ρ0 dxA_1 = \int\rho_0\,dx, whose gradient is constant, is never moved but evaluated directly, as below. For brackets including members of both chains, it always terminates at one of the seeds, either by having A1A_1 directly or by lowering the index of BkB_k to B1=∫ρ1 dxB_1 = \int\rho_1\,dx, leaving

{A1,Bm}P∗=2α∫⟨δBm,K−1⟩ dx,{Am,B1}P∗=∫⟨δAm,αK0+β2K−1⟩ dx,(72)\{A_1,B_m\}_{\mathcal{P}_*} = 2\alpha\int\langle\delta B_m,K_{-1}\rangle\,dx,\qquad\{A_m,B_1\}_{\mathcal{P}_*} = \int\left\langle\delta A_m,\alpha K_0+\frac{\beta}{2}K_{-1}\right\rangle\,dx, \tag*{(72)}

that is, derivatives along the phase rotation and the translation. These vanish by Noether rather than by a Casimir, since the flows of the two seeds are symmetries of every member of the hierarchy. Indeed, LK−1R=LK0R=0\mathcal{L}_{K_{-1}}R=\mathcal{L}_{K_0}R=0, and the Lie derivative has the property that, whenever LYR=0\mathcal{L}_Y R=0, then [Y,RZ]=R[Y,Z][Y,RZ]=R[Y,Z]. With that, we have that [K−1,RjK]=Rj[K−1,K]=0[K_{-1},R^jK]=R^j[K_{-1},K]=0, where KK can be K−1K_{-1} or K0K_0 depending on the family, and likewise for K0K_0, since the phase and the translation commute. Hence, K−1K_{-1} and K0K_0 commute with every KnK_n, and, since P∗δAm\mathcal{P}_*\delta A_m and P∗δBm\mathcal{P}_*\delta B_m lie in the span of the KnK_n, we have P∗δ{F,Hb}P∗=[P∗δF,Kb]=0\mathcal{P}_*\delta\{F,H_b\}_{\mathcal{P}_*}=[\mathcal{P}_*\delta F,K_b]=0 for b=−1,0b=-1,0 and every member FF of the chains, so that Theorem 3.2 leaves only the zero functional. With that, we conclude that the two families are in involution.

The existence of the chains at every order, which is assumed in the computation above, follows from the same symmetries.

Proposition 5.1. Every member of the Lenard chains (56) exists and is a local functional. Hence, every flow KnK_n is local and Hamiltonian with respect to P∗\mathcal{P}_* with a local Hamiltonian HnH_n.

Proof. Let us consider RR in its invariant form (58), whose left legs are, up to constant factors, K−1K_{-1}, K1K_1 and K0K_0. On covectors with a decaying gradient, P2=RP∗\mathcal{P}_2=R\mathcal{P}_*, so a chain step is δFk+1=R†δFk\delta F_{k+1}=R^\dagger\delta F_k, where R†R^\dagger is the formal adjoint of RR, acting on covectors as R†w=Rloc†w−∑bγb∂x−1⟨Gb,w⟩R^\dagger w=R_{\mathrm{loc}}^\dagger w-\sum_b\gamma_b\partial_x^{-1}\langle G_b,w\rangle.

The three arguments ⟨Gb,δFk⟩\langle G_b,\delta F_k\rangle are, after integration, the derivatives of FkF_k along the phase, the flow and the translation, all of which vanish because P∗δFk\mathcal{P}_*\delta F_k lies in the span of the KnK_n, and the three symmetries K−1K_{-1}, K0K_0 and K1K_1 commute with every KnK_n. The commutation with K−1K_{-1} and K0K_0 was already shown above, while for K1K_1 heredity, the classical consequence of compatibility with P∗\mathcal{P}_* non-degenerate [45], gives LK1R=LRK−1R=RLK−1R=0\mathcal{L}_{K_1}R=\mathcal{L}_{RK_{-1}}R=R\mathcal{L}_{K_{-1}}R=0, and the flow commutes with the phase and the translation. Hence, P∗δ{Fk,Hb}P∗=0\mathcal{P}_*\delta\{F_k,H_b\}_{\mathcal{P}_*}=0, Theorem 3.2 gives {Fk,Hb}P∗=0\{F_k,H_b\}_{\mathcal{P}_*}=0 for b=−1,0,1b=-1,0,1, and the integrands, vanishing on integration for all fields, are total derivatives. Therefore, R†δFkR^\dagger\delta F_k is local.

It is closed by the closure lemma of the Lenard–Magri scheme: for a compatible pair with non-degenerate first structure, if a 1-form ω\omega and its image R†ωR^\dagger\omega are closed, then so is (R†)2ω(R^\dagger)^2\omega [40, 45], [28 Theorem 7.27], and a closed local covector, that is, one with a self-adjoint Fréchet derivative, is always the variational derivative of a local functional [28 Theorem 5.92]. Note that the closure lemma by itself does not guarantee that the functionals obtained are local, which is why the locality of R†δFkR^\dagger\delta F_k has been established separately above. That means that, if a covector δFk−1\delta F_{k-1} and its image δFk=R†δFk−1\delta F_k=R^\dagger\delta F_{k-1} are closed, then (R†)2δFk−1(R^\dagger)^2\delta F_{k-1} is also closed, and it is the variational derivative of a local functional, Fk+1F_{k+1}. We can start the induction from A1A_1, A2A_2 (given by Eq. (54)), B1B_1 and B2=αH2+β2H1B_2=\alpha H_2+\frac{\beta}{2}H_1 (Proposition 3.4), all of which are explicit local functionals, so every member of the chains exists and is local.

Finally, since P∗δAk\mathcal{P}_*\delta A_k and P∗δBk\mathcal{P}_*\delta B_k are triangular in the KnK_n with diagonal coefficient α\alpha, every KnK_n is P∗\mathcal{P}_*-Hamiltonian with a local Hamiltonian. □

We can put all of the above together in the following result.

Theorem 5.2. Every flow KnK_n of the hierarchy is local and Hamiltonian with respect to P∗\mathcal{P}_*, with a local Hamiltonian HnH_n. The Hamiltonians are in involution, {Hm,Hn}P∗=0\{H_m,H_n\}_{\mathcal{P}_*}=0, and the flows commute, [Km,Kn]=0[K_m,K_n]=0, for all n,m≥−1n,m\geq-1.

Proof. Proposition 5.1 gives existence and locality. As shown before, all the members of the chains (56) are in involution, and they are triangular in the HnH_n with diagonal coefficient α≠0\alpha\ne0, so the HnH_n are in involution too. Finally, Kj=P∗δHjK_j=\mathcal{P}_*\delta H_j and P∗\mathcal{P}_* is Hamiltonian, so [Km,Kn]=P∗δ{Hm,Hn}P∗=0[K_m,K_n]=\mathcal{P}_*\delta\{H_m,H_n\}_{\mathcal{P}_*}=0. □

Remark 5.3. The locality and commutativity of all flows also follow, without the closure lemma, from Theorem 1 of [36], which applies to hereditary recursion operators of the shape (58), with linearly independent GbG_b, which are normal, meaning that γb′=γb′†\gamma_b'=\gamma_b'{}^\dagger, ζb=ζb†\zeta_b=\zeta_b^\dagger with ζb=R†(γb)\zeta_b=R^\dagger(\gamma_b), and LGb(γb′)=0\mathcal{L}_{G_b}(\gamma_{b'})=0 for all b,b′b,b'. Then, for any seed QQ with LQR=0\mathcal{L}_Q R=0 and RjQR^jQ local, it concludes that the RjQR^jQ are all local and pairwise commute. We have verified the normality conditions for RR: the three γb\gamma_b have a self-adjoint Fréchet derivative, each ζb\zeta_b is local and self-adjoint, and all nine LGb(γb′)\mathcal{L}_{G_b}(\gamma_{b'}) vanish. Applied with the seeds K−1K_{-1} and K0K_0, the theorem makes each family local and commutative, while across the families heredity gives LKnR=0\mathcal{L}_{K_n}R=0 for all $n, and then [RjK−1,RkK0]=Rj+k[K−1,K0]=0[\mathcal{R}^{j}K_{-1},\mathcal{R}^{k}K_{0}]=\mathcal{R}^{j+k}[K_{-1},K_{0}]=0. This gives an alternative path for the proof. Note that Theorem 1 of [36] concerns a single seed, and commutativity across the orbits of several seeds, enforced by conditions on the seeds, is treated in [51].

The structure in the (α,β)(\alpha,\beta) plane

The roles of α\alpha and β\beta in the bi-Hamiltonian structure are very different. At β=0\beta=0, corresponding to the Newell system, nothing degenerates, as expected from Eq. (14): P∗\mathcal{P}_{*} stays regular, with μ=2α2L\mu=2\alpha^{2}L and Theorem 3.2 intact, and P2\mathcal{P}_{2} is regular and nonzero, its tail matrix tending to

c→β→0(048α3048α30000−α2),(73)c \xrightarrow{\beta\to0} \begin{pmatrix} 0 & 48\alpha^{3} & 0 \\ 48\alpha^{3} & 0 & 0 \\ 0 & 0 & -\alpha^{2} \end{pmatrix}, \tag*{(73)}

still of rank three. Hence, the Newell system is bi-Hamiltonian with the restriction of the same pair.

On the other hand, the reduction α=0\alpha=0 yielding Yajima–Oikawa shows a genuine degeneration affecting both structures, as follows. Note that the shift (8) is itself singular as α→0\alpha\to0: the YO system is not a member of the family of systems equivalent to the Newell one, but a limit of it.

At α=0\alpha=0, the gauge factor e3iα∂x−1Le^{3i\alpha\partial_x^{-1}L} is trivial, so S^=S\hat{S}=S and the two structures can be compared to the known bi-Hamiltonian structure of YO directly. In the normalisation chosen for P∗\mathcal{P}_{*} and P2\mathcal{P}_{2}, every one of their entries is polynomial in α\alpha, so the limit can be taken termwise with no rescaling needed.

For the first structure,

P∗→α→0Q0=(0iβ20iβ20000∂x),(74)\mathcal{P}_{*} \xrightarrow{\alpha\to0} Q_{0}=\begin{pmatrix} 0 & \frac{i\beta}{2} & 0 \\ \frac{i\beta}{2} & 0 & 0 \\ 0 & 0 & \partial_x \end{pmatrix}, \tag*{(74)}

which is the canonical YO structure, in which the short-wave block becomes ultralocal and the coupling to the long wave disappears entirely. What is lost in this limit is precisely what made the YON case hard, that is, the derivative term α∂x\alpha\partial_x in the (1, 2) entry and the off-diagonal terms ∓2iαS^\mp2i\alpha\hat{S}. Solving Q0δH=(flow)Q_{0}\delta H=(\mathrm{flow}) on the graded ansatz returns

HYO=∫[2L∣S∣2+2β∣Sx∣2] dx.(75)H_{\mathrm{YO}}=\int\left[2L|S|^{2}+\frac{2}{\beta}|S_{x}|^{2}\right]\,dx. \tag*{(75)}

As apparent from this, the reduction does not work directly for the Hamiltonian. The Hamiltonian H∗H_{*} of Eq. (29) features α\alpha in the denominators, up to α−3\alpha^{-3} through the term β2L/8α3\beta^{2}L/8\alpha^{3}, so its YO reduction diverges. Therefore, the pair (P∗,H∗)(\mathcal{P}_{*},H_{*}) defining the Hamiltonian structure has no finite limit for YO, even though P∗\mathcal{P}_{*} and the flow it generates do. The reason is clear from the form of Q0Q_{0}: since its long-wave corner is ∂x\partial_x and its long-short coupling is gone, δ∫L dx=(0,0,1)\delta\int L\,dx=(0,0,1) lies in its kernel. As α\alpha approaches 0, the Hamiltonian runs off in the direction of the Casimir.

The point symmetries change at the same place: X4X_{4} is lost, and the scaling X5X_{5} of the YO grading and the shift of LL with a time-dependent phase, X6X_{6}, appear instead, the latter being Hamiltonian with respect to Q0Q_{0}: Q0δ∫(xL+2t∣S∣2) dx=(−iβtS,iβtS∗,1)Q_{0}\delta\int(xL+2t|S|^{2})\,dx=(-i\beta tS,i\beta tS^{*},1).

For the second structure, the limit is again finite and nonzero, with 14∂x3\frac{1}{4}\partial_{x}^{3} in the long-wave corner and −iβ2∂x2-\frac{i\beta}{2}\partial_{x}^{2} in the short-wave block. Most interestingly, for α=0\alpha=0 the nonlocal tail of P2\mathcal{P}_{2} collapses to rank one: out of the six nonzero entries of cc only the (1,1) entry survives the limit, leaving a tail of the form −3β24u1∂x−1u1-\frac{3\beta^{2}}{4}u_{1}\partial_{x}^{-1}u_{1}, with u1=(S,−S∗,0)u_{1}=(S,-S^{*},0), a single leg on the phase symmetry. This is exactly the shape of the tail of the second Hamiltonian structure of the NLS equation. The rank-three tail of Section 4.1 is therefore a genuinely two-parameter phenomenon, the legs of the flow and the translation being switched on by α\alpha.

Finally, the conserved tower does not degenerate: the fifteen nontrivial densities produced by the recursion (27) through order six span a space of gradients of rank six both for generic α\alpha and at α=0\alpha=0, so that the six independent ones remain independent in the limit. This is worth stating because a natural-looking basis does collapse: the low-weight densities obey ρ1→−∣S∣2\rho_{1}\to-|S|^{2} and ρ2→−β∣S∣2\rho_{2}\to-\beta|S|^{2} at α=0\alpha=0 and become proportional there. That is a defect of the weight-by-weight basis for the densities and not of the tower itself; the Lax recursion produces a basis that does survive the limit.

Conclusions

To close out the paper, let us give a brief tally of the results presented and an outlook of the questions they leave open.

In this work we have studied the Hamiltonian and bi-Hamiltonian structure of the Yajima–Oikawa–Newell system, which is integrable for every choice of its two parameters. In Section 2, we noted that, for α≠0\alpha\ne0, a constant shift of the long wave reduces the YON system to the Newell system, for which two Hamiltonian operators, both nonlocal, were given in [18], and we used the grading induced by the scaling symmetry and the U(1)U(1) invariance of the system to reduce the search for local Hamiltonian operators to finitely many candidates at each weight, none of which is a Poisson operator for α≠0\alpha\ne0. In Section 3, we showed that the nonlocality of the first of those operators is tied to the choice of variables: a nonlocal gauge transformation on the potential of the long wave turns it into a local, non-degenerate operator P∗\mathcal{P}_* for every α≠0\alpha\ne0 and β\beta, which is the Lie–Poisson structure of the sl(2)\mathfrak{sl}(2) current algebra, the two signs of α\alpha corresponding to its two real forms. The gauge parameter is not a choice, but is fixed to γ=−3α\gamma=-3\alpha by the Jacobi identity, and it differs from the gauge of Newell, γ=α\gamma=\alpha, which linearises the short-wave equation. The Hamiltonian formulation allowed us to make explicit the Noether correspondence between the Lie point symmetries of the system and its conserved quantities, the scaling symmetry included, and we also obtained the conserved densities of the system in closed form by asymptotic diagonalisation of the Lax pair. In Section 4, we constructed, in explicit weakly nonlocal form, a recursion operator advancing the hierarchy by two steps and, from it, a second Hamiltonian structure P2=RP∗\mathcal{P}_2=R\mathcal{P}_*, lying above the pair of [18] in the hierarchy, whose nonlocal tail is generated precisely by the three autonomous Lie point symmetries of the system, and we proved that P∗\mathcal{P}_* and P2\mathcal{P}_2 form a Hamiltonian pair for every α≠0\alpha\ne0 and β\beta. Finally, in Section 5, we studied the Lenard–Magri chain of the pair. Since the recursion operator admits no square root and P∗\mathcal{P}_* has no Casimirs, the chain splits into two interleaved families, which have to be seeded by the two simplest conserved quantities, ∫L dx\int L\,dx and ∫ρ1 dx\int\rho_1\,dx. The flows of these seeds, the phase rotation and the translation, are symmetries of every member of the hierarchy, and that is enough to make every flow local and every Hamiltonian local and in involution, to all orders. For β=0\beta=0, that is, for the Newell system, the flow is bi-Hamiltonian with respect to this pair, while for β≠0\beta\ne0 the YON system itself is bi-Hamiltonian up to a Galilean change of frame.

Some of these features are worth highlighting beyond the YON system. The double step of the recursion is inherited from the Yajima–Oikawa limit, but the deformation by α\alpha removes the Casimir that anchors the chain there: at α=0\alpha=0, ∫L dx\int L\,dx is a Casimir of the first structure, whereas for α≠0\alpha\ne0 it is not. The combination of a structure without Casimirs, a hierarchy without gaps and a recursion operator without square root makes the Lenard–Magri scheme a propagation mechanism rather than a generating one, the role of the Casimir being played by the symmetries generated by the seeds. Let us also note that the composition P2=RP∗\mathcal{P}_2=R\mathcal{P}_* fails on the constant gradient δ∫L dx\delta\int L\,dx, a point that is easy to overlook and that affects the Lenard relations of every Hamiltonian with a component along ∫L dx\int L\,dx, which flows are Hamiltonian with respect to both structures, and the nonlocal form of the first structure in the original variables.

Several questions remain open. Whether the two families exhaust all the conserved densities of the system remains open. The reduction α=β=0\alpha=\beta=0, a free Schrödinger equation driving a slaved long wave, admits no local Hamiltonian operator of weight one or two, and whether it admits a Hamiltonian structure at all is not known. It would also be natural to relate the pair to the Lax-operator brackets of the constrained KP and modified KP hierarchies [16], and in particular to determine the image of P2\mathcal{P}_2 under the Miura map of [18], which takes P∗\mathcal{P}_* to the second Hamiltonian structure of the YO system; to understand the sl(2)\mathfrak{sl}(2) current algebra of the first structure in terms of the 3×33\times3 Lax pair (4); and to extend the construction to multicomponent versions of the YON system. Finally, the bi-Hamiltonian structure may prove useful in the analysis of the solutions of the system, in particular in connection with its inverse scattering problem, to which the standard techniques do not apply directly, the matrix Σ\Sigma in the Lax pair (4) being singular [24].

Acknowledgments

The author is grateful to Matteo Sommacal and Sara Lombardo for insightful discussions and for their help in proofreading the manuscript. This work has been carried out under the auspices of the Italian GNFM (Gruppo Nazionale Fisica Matematica), INdAM (Istituto Nazionale di Alta Matematica), which is gratefully acknowledged.

References

  1. [1]Benney, D. J. A general theory for interactions between short and long waves. Stud. Appl. Math. 56 (1977), 81–94.DOI
  2. [2]Yajima, N., and Oikawa, M. Formation and interaction of sonic–Langmuir solitons: Inverse scattering method. Prog. Theor. Phys. 56, 6 (1976), 1719–1739.DOI
  3. [3]Djordjevic, V. D., and Redekopp, L. G. On two-dimensional packets of capillary-gravity waves. J. Fluid Mech. 79 (1977), 703–714.
  4. [4]Ma, Y.-C. The complete solution of the long wave–short wave resonance equations. Stud. Appl. Math. 59, 3 (1978), 201–221.DOI
  5. [5]Newell, A. C. Long waves–short waves: A solvable model. SIAM J. Appl. Math. 35 (1978), 650–664.DOI
  6. [6]Caso-Huerta, M., Degasperis, A., Lombardo, S., and Sommacal, M. A new integrable model of long wave–short wave interaction and linear stability spectra. Proc. R. Soc. A 477 (2021), 20210408.
  7. [7]Caso-Huerta, M., Degasperis, A., Leal da Silva, P., Lombardo, S., and Sommacal, M. Periodic and solitary wave solutions of the long wave–short wave Yajima–Oikawa–Newell model. Fluids 7, 7 (2022), 227.
  8. [8]Kirane, M., Stalin, S., and Lakshmanan, M. Bright, dark and breather soliton solutions of the generalized long-wave short-wave resonance interaction system. Nonlinear Dyn. 110, 1 (2022), 771–790.
  9. [9]Caso-Huerta, M., Feng, B.-F., Lombardo, S., Maruno, K., and Sommacal, M. Solitons, breathers and rogue waves of the Yajima-Oikawa-Newell long wave-short wave system. Wave Motion 134 (2025), 103511.
  10. [10]Bai, S., and Gegen, H. Rogue wave patterns and hybrid solutions of rogue wave and breather for the Yajima-Oikawa-Newell system. Phys. Scr. 100, 7 (2025), 075239.DOI
  11. [11]Ankiewicz, A., Kanna, T., and Rao, J. Volumes for 3-component rogue waves of long wave–short wave resonant interaction system of Newell form. Chaos Solitons Fractals 208, 2 (2026), 118192.DOI
  12. [12]Caso-Huerta, M., Sommacal, M., Lombardo, S., Horikis, T. P., and Franzeskakis, D. J. Long-short-wave interactions in shallow water: Physical derivation and integrable reductions. Submitted for publication, 2026.
  13. [13]Myrzakulova, Z., Nugmanova, G., Serikbayev, N., Yesmakhanova, K., and Myrzakulov, R. Integrable Tolkynay equations and related Yajima-Oikawa type equations. Ufa Math. J. 15, 1 (2023), 122–134.
  14. [14]Anco, S. C., and Bluman, G. W. Direct construction of conservation laws from field equations. Phys. Rev. Lett. 78 (1997), 2869–2873.DOI
  15. [15]Anco, S. C., and Bluman, G. W. Direct construction method for conservation laws of partial differential equations. Part II: General treatment. Eur. J. Appl. Math. 13 (2002), 567–585.
  16. [16]Oevel, W., and Strampp, W. Constrained KP hierarchy and bi-Hamiltonian structures. Commun. Math. Phys. 157 (1993), 51–81.DOI
  17. [17]Li, H., Li, Y., and Chen, Y. Bi-Hamiltonian structure of multi-component Yajima–Oikawa hierarchy. Z. Naturforsch. A 70, 11 (2015), 929–934.DOI
  18. [18]Liu, Q. P. Modifications of k-constrained KP hierarchy. Phys. Lett. A 187, 5–6 (1994), 373–381.DOI
  19. [19]Kaup, D. J., and Newell, A. C. An exact solution for a derivative nonlinear Schrödinger equation. J. Math. Phys. 19 (1978), 798–801.DOI
  20. [20]Chen, H. H., Lee, Y. C., and Liu, C. S. Integrability of nonlinear Hamiltonian systems by inverse scattering method. Phys. Scr. 20, 3–4 (1979), 490–492.DOI
  21. [21]Gerdjikov, V. S., and Ivanov, M. I. A quadratic pencil of general type and nonlinear evolution equations. II. Hierarchies of Hamiltonian structures. Bulgar. J. Phys. 10 (1983), 130–143.
  22. [22]Wadati, M., and Sogo, K. Gauge transformations in soliton theory. J. Phys. Soc. Jpn. 52 (1983), 394–398.DOI
  23. [23]Kundu, A. Landau–Lifshitz and higher-order nonlinear systems gauge generated from nonlinear Schrödinger-type equations. J. Math. Phys. 25 (1984), 3433–3438.DOI
  24. [24]Caso-Huerta, M. A New Model of Long Wave-Short Wave Interaction Generalising the Yajima–Oikawa and Newell Systems: Integrability and Linear Stability Spectra. PhD thesis, Northumbria University, Newcastle upon Tyne, UK, October 2022.
  25. [25]Ferapontov, E. V. Differential geometry of nonlocal Hamiltonian operators of hydrodynamic type. Funct. Anal. Appl. 25 (1991), 195–204.DOI
  26. [26]Maltsev, A. Y., and Novikov, S. P. On the local systems Hamiltonian in the weakly non-local Poisson brackets. Physica D 156 (2001), 53–80.DOI
  27. [27]Casati, M., Lorenzoni, P., and Vitolo, R. Three computational approaches to weakly nonlocal Poisson brackets. Stud. Appl. Math. 144 (2020), 412–448.DOI
  28. [28]Olver, P. J. Applications of Lie Groups to Differential Equations. No. 107 in Graduate Texts in Mathematics. Springer, New York, NY, US, 1993.
  29. [29]Dorfman, I. Dirac Structures and Integrability of Nonlinear Evolution Equations. Wiley, Chichester, UK, 1993.
  30. [30]Schouten, J. A. Über Differentialkomitanten zweier kontravarianter Grössen. Indag. Math. 2 (1940), 449–452.
  31. [31]Schouten, J. A. On the differential operators of first order in tensor calculus. In Convegno Internazionale di Geometria Differenziale, Italia, 20–26 settembre 1953 (Rome, Italy, 1954), Edizioni Cremonese, pp. 1–7.
  32. [32]Nijenhuis, A. Jacobi-type identities for bilinear differential concomitants of certain tensor fields. I. Indag. Math. 17 (1955), 390–397.DOI
  33. [33]Drinfel’d, V. G., and Sokolov, V. V. Lie algebras and equations of Korteweg–de Vries type. J. Soviet Math. 30, 2 (1985), 1975–2036.
  34. [34]Hisakado, M., and Wadati, M. Gauge transformations among generalised nonlinear Schrödinger equations. J. Phys. Soc. Jpn. 63 (1994), 3962–3966.DOI
  35. [35]Chen, J., Feng, B.-F., Maruno, K.-I., and Ohta, Y. The derivative Yajima–Oikawa system: Bright, dark soliton and breather solutions. Stud. Appl. Math. 141, 2 (2018), 145–185.DOI
  36. [36]Sergyeyev, A. Why nonlocal recursion operators produce local symmetries: New results and applications. J. Phys. A 38 (2005), 3397–3407.
  37. [37]Hernández Heredero, R., and Sokolov, V. Integrability conditions for Boussinesq type systems. Partial Differ. Equ. Appl. Math. 12 (2024), 100959.
  38. [38]Sergyeyev, A. Weakly nonlocal Hamiltonian structures: Lie derivative and compatibility. Symmetry Integr. Geom.: Methods Appl. 3 (2007), 062.
  39. [39]Anco, S. C., and Asadi, E. Hasimoto variables, generalized vortex filament equations, Heisenberg models and Schrödinger maps arising from group-invariant NLS systems. J. Geom. Phys. 144 (2019), 324–357.DOI
  40. [40]Magri, F. A simple model of the integrable Hamiltonian equation. J. Math. Phys. 19 (1978), 1156–1162.DOI
  41. [41]Magri, F. Lenard chains for classical integrable systems. Theor. Math. Phys. 137, 3 (2003), 1716–1722.DOI
  42. [42]Praught, J., and Smirnov, R. G. Andrew Lenard: A mystery unraveled. Symmetry Integr. Geom.: Methods Appl. 1 (2005), 005.DOI
  43. [43]Gelfand, I. M., and Zakharevich, I. Webs, Lenard schemes, and the local geometry of bi-Hamiltonian Toda and Lax structures. Selecta Math. 6 (2000), 131–183.DOI
  44. [44]Gelfand, I. M., and Dorfman, I. Y. Hamiltonian operators and algebraic structures related to them. Funct. Anal. Appl. 13, 4 (1979), 248–262.
  45. [45]Fuchssteiner, B., and Fokas, A. S. Symplectic structures, their Bäcklund transformations and hereditary symmetries. Phys. D 4, 1 (1981), 47–66.DOI
  46. [46]Fuchssteiner, B. Mastersymmetries, higher order time-dependent symmetries and conserved densities of nonlinear evolution equations. Prog. Theor. Phys. 70, 6 (1983), 1508–1522.DOI
  47. [47]Liu, Q. P. Bi-Hamiltonian structures of the coupled AKNS hierarchy and the coupled Yajima–Oikawa hierarchy. J. Math. Phys. 37, 5 (1996), 2307–2314.DOI
  48. [48]Gürses, M., Karasu, A., and Sokolov, V. V. On construction of recursion operators from Lax representation. J. Math. Phys. 40, 12 (1999), 6473–6490.DOI
  49. [49]Fuchssteiner, B., and Oevel, W. The bi-Hamiltonian structure of some nonlinear fifth- and seventh-order differential equations and recursion formulas for their symmetries and conserved covariants. J. Math. Phys. 23, 3 (1982), 358–363.DOI
  50. [50]Wang, J. P. A list of 1 + 1 dimensional integrable equations and their properties. J. Nonlinear Math. Phys. 9, Suppl. 1 (2002), 213–233.DOI
  51. [51]Wang, J. P. Lenard scheme for two-dimensional periodic Volterra chain. J. Math. Phys. 50, 2 (2009), 023506.
  52. [52]Sergyeyev, A. Infinitely many local higher symmetries without recursion operator or master symmetry: Integrability of the Foursov–Burgers system revisited. Acta Appl. Math. 109, 1 (2010), 273–281.

Paper details

vlnc:W_8xs8sqfs
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents