Multidimensional nonhomogeneous quasi-linear systems and their Hamiltonian structure

Abstract

In this paper we investigate multidimensional first-order quasi-linear systems and find necessary conditions for them to admit Hamiltonian formulation. The insufficiency of the conditions is related to the Poisson cohomology of the admissible Hamiltonian operators. We present in detail the examples of two-dimensional, two-components systems of hydrodynamic type and of a real reduction of the 3-waves system.

1 Introduction

Hamiltonian systems are mathematical models used to describe the dynamical behavior of physical systems. They are based on Hamiltonian principle and Hamiltonian equations, which utilize generalized coordinates and generalized momenta to characterize the state of the system. Hamiltonian systems find wide-ranging applications in classical mechanics, quantum mechanics, and other fields. An autonomous system of evolutionary PDEs

$$u^{i}_{t}=F^{i}(u^{j},u^{j}_{x},\dots,u^{j}_{kx}),\qquad i,j=1,2,\dots,n$$ (1)

in two dependent variables $(x,t)$ is referred to as Hamiltonian, with a Hamiltonian structure $P$ and a Hamiltonian functional $H$, if it can be expressed as

$$u^{i}_{t}=P^{ij}\frac{\delta H}{\delta u^{j}},$$ (2)

where $P$ is a (pseudo) differential operator and $H=\int h(u^{j},u^{j}_{x},\dots)\,dx$ is a local functional.

To date, many researchers have applied advanced mathematical methods and techniques, such as Lie algebra [21], Poisson geometry [26], symplectic geometry [18], etc., to investigate the Hamiltonian structures of nonlinear evolutionary equations, including the KdV equation, the KN equation [18], the Sine-Gordon equation, the KP equation [20, 7], and so on.

In the paper by Vergallo [24], an investigation has been conducted on a class of evolutionary systems known as first-order quasi-linear systems (it was firstly investigated in [23])

$$u^{i}_{t}=V^{i}_{l}(u)u^{l}_{x}+W^{i}(u),\qquad i=1,2,\dots,n.$$ (3)

Through the application of cotangent covering theory [14, 25], the author conducted a study on the compatibility conditions for 1+0 order nonhomogeneous quasi-linear systems to be Hamiltonian systems under the nonhomogeneous hydrodynamic operator.

A multidimensional Hamiltonian system is an evolutionary system

$$u^{i}_{t}=F^{i}(u^{j},u^{j}_{\alpha},u^{j}_{\alpha\beta},\dots),\qquad i,j=1,2,\dots,n$$ (4)

which can be written as $u^{i}_{t}=P^{ij}\frac{\delta H}{\delta u^{j}}$, where $P^{ij}$ is a Hamiltonian operator and $H=\int h(u^{j},u^{j}_{\alpha},u^{j}_{\alpha\beta},\dots)\,dx^{D}$ and $\alpha,\beta,\dots$ is a component of variable $x$ in $D$-dimensional space. $u^{j}_{\alpha}$ is a shorthand notation for $\frac{\partial u^{j}}{\partial x^{\alpha}}$, $u^{j}_{\alpha\beta}$ for $\frac{\partial^{2}u^{j}}{\partial x^{\alpha}\partial x^{\beta}}$, etc.

Therefore, the idea of this paper is to extend the study of quasi-linear systems done by Vergallo to include the $D$-dimensional case ($D>1$), namely

$$u^{i}_{t}=V^{i\alpha}_{l}(u)u^{l}_{\alpha}+W^{i}(u),\qquad i=1,2,\dots,n.$$ (5)

The question is: which quasi-linear systems admit Hamiltonian formulations in the multidimensional case? The main approach involves the application of PVA theory to compute the compatibility conditions of $D$-dimensional nonhomogeneous Hamiltonian operators, along with the application of cotangent covering theory to find compatible quasi-linear systems under such operators. Finally, the compatibility conditions for quasi-linear systems to be Hamiltonian are computed in the $D=N=2$ case, and we relate the compatibility conditions with the first Poisson cohomology group of the Hamiltonian operator.

2 Nonhomogeneous multidimensional Hamiltonian structures

2.1 PVAs and Hamiltonian structures

Let us consider the space of maps $\mathcal{M}$ from a $D$ dimensional manifold $X$ to a $N$ dimensional target manifold $U$. Chosen the coordinate systems ${u^{i}}$ and ${x^{\alpha}}$ , the quotient space of local functionals is defined as $\mathcal{F}:=\frac{\mathcal{A}}{\partial_{1}\mathcal{A}+\partial_{2}\mathcal{A}+\dots+\partial_{D}\mathcal{A}}$ [21], where $\mathcal{A}=\mathcal{A}(U)$ is the space of differential polynomials. Now, we endow the space $\mathcal{F}$ with a Lie bracket $\{\ ,\ \}$, also known as the local Poisson bracket. A Lie bracket on $F$ can equivalently defined in terms of a local Poisson bivector, $P\in\Lambda^{2}$, by

$$P(\delta F,\delta G)=\{F,G\}=\int\frac{\delta F}{\delta u^{i}}P^{ij}\frac{\delta G}{\delta u^{j}},$$ (6)

where $P^{ij}=P^{ij}_{S}\partial^{S}$ ($P^{ij}_{S}\in\mathcal{A}$) is a skewadjoint differential operator such that $[P,P]=0$, referred to as a Hamiltonian structure, here we denote $S=(s_{1},s_{2},\dots,s_{D})$, $\partial^{S}=\left(\frac{d}{dx^{1}}\right)^{s_{1}}\left(\frac{d}{dx^{2}}\right)^{s_{2}}\dots\left(\frac{d}{dx^{D}}\right)^{s_{D}}$, $\frac{\delta F}{\delta u^{i}}=\sum\limits_{S}(-1)^{s_{1}+s_{2}+\dots+s_{D}}\partial^{S}\left(\frac{\partial F}{\partial(\partial^{S}u^{i})}\right)$. The skewadjointness of $P$ is equivalent to the skewsymmetry of the Poisson bracket and the vanishing of the Schouten torsion $[P,P]$ [21] corresponds to the Jacobi identity.

Next, we will briefly demonstrate the application of Poisson Vertex Algebra (PVA) theory to compute the conditions for an operator to be Hamiltonian. Detailed information about PVA can be found in [2]. A multidimensional Poisson Vertex Algebra (mPVA) is a differential algebra $(\mathcal{V},\partial)$ endowed with a bilinear operation $\mathcal{V}\times\mathcal{V}\rightarrow\mathbb{R}[\lambda_{1},\dots,\lambda_{D}]\otimes\mathcal{V}$ called the $\lambda$-bracket and denoted $\{f_{\lambda}g\}=C_{s_{1},s_{2},\dots,s_{D}}(f,g)\lambda^{s_{1}}_{1}\lambda^{s_{2}}_{2}\dots\lambda^{s_{D}}_{D}=C_{S}(f,g)\lambda^{S}$. We take $\mathcal{V}$ to be the space $\mathcal{A}$ of differential polynomials and $\partial_{\alpha}=u^{i}_{S+\xi_{\alpha}}\frac{\partial}{\partial u^{i}_{S}}$, where $u^{i}_{S}$ is a shorthand notation for $\partial^{S}u^{i}$ and $\xi_{\alpha}=(0,\ldots,0,\underbrace{1}_{\alpha},0,\ldots)$ is the canonical basic vectors in $\mathbb{Z}^{D}$.

We can explicitly compute a $\lambda$-bracket between to differential polynomials using the so-called master formula for a multidimensional Poisson Vertex Algebra [5] as it follows

$$\{f_{\lambda}g\}=\sum_{i,j=1,\dots,N}\sum_{S,M\in\mathbb{Z}^{D}_{\geq 0}}\frac{\partial g}{\partial u^{j}_{S}}(\lambda+\partial)^{S}\{u^{i}_{\lambda+\partial}u^{j}\}(-\lambda-\partial)^{M}\frac{\partial f}{\partial u^{i}_{M}}.$$ (7)

The following two theorems [5] are presented to illustrate the relationship between Poisson Vertex Algebras and local Poisson brackets.

Theorem 1.

Let us define a bilinear bracket on $\mathcal{F}$

$$\{F,G\}:=\int{\{f_{\lambda}g\}}\mid_{\lambda=0},$$

where $f,g\in\mathcal{A}$ are the densities of $F,G$. If the $\lambda$-bracket satisfies the axioms of a PVA, then the bracket we defined is a local Poisson bracket.

Theorem 2.

Given a Hamiltonian structure $P=P^{ij}_{S}\partial^{S}$ on $\mathcal{A}$, the $\lambda$-bracket defined on generators as

$$\{u^{i}_{\lambda}u^{j}\}:=\sum P^{ji}_{S}\lambda^{S}$$

and extended to the full algebra using the master formula satisfies the axioms of the mPVA.

These two theorems demonstrate the equivalence between mPVAs and multidimensional Hamiltonian structures. In particular, the Jacobi identity is equivalent to the so-called PVA-Jacobi identity. Thus, we can obtain a Hamiltonian structure by computing the Jacobi identity of the $D$-dimensional $\lambda$-bracket, among the generators (it is equivalent to the Jacobi identity of the generic densities), namely

$$\{u^{i}_{\lambda}\{u^{j}_{\mu}u^{k}\}\}-\{u^{j}_{\mu}\{u^{i}_{\lambda}u^{k}\}\}=\{\{u^{i}_{\lambda}u^{j}\}_{\mu+\lambda}u^{k}\}.$$ (8)

2.2 The homogeneous case

Let us consider homogeneous differential operators of order $m$ with the following form

$$P^{ij}=g^{ij}_{S_{1}}(u)\partial^{S_{1}}+b^{ij\alpha}_{kS_{2}}(u)u^{k}_{\alpha}\partial^{S_{2}}+c^{ij\alpha\beta}_{kS_{3}}(u)u^{k}_{\alpha\beta}\partial^{S_{3}}+c^{ij\alpha\beta}_{klS_{3}}(u)u^{k}_{\alpha}u^{l}_{\beta}\partial^{S_{3}}+\dots,$$ (9)

where

$$S_{n}=(s_{1},s_{2},\dots,s_{D}),\qquad n=1,2,\dots$$

$$\partial^{S_{n}}=\left(\frac{d}{dx^{1}}\right)^{s_{1}}\left(\frac{d}{dx^{2}}\right)^{s_{2}}\dots\left(\frac{d}{dx^{D}}\right)^{s_{D}},\qquad s_{1}+s_{2}+\dots+s_{D}=m-n+1.$$

One important class of Poisson brackets is the Poisson bracket for which a first-order homogeneous differential operator

$$P^{ij}=g^{ij\alpha}(u)\partial_{\alpha}+b^{ij\alpha}_{k}(u)u^{k}_{\alpha}$$ (10)

satisfies $P^{*}=-P$ and $[P,P]=0$. Such brackets, defined by Dubrovin and Novikov [8], are called Poisson brackets of hydrodynamic type or D-N brackets.

We recall the conditions satisfied by the coefficients $(g^{ij\alpha},b^{ij\alpha}_{k})$ in the multidimensional D-N brackets, provided by Mokhov [17].

Theorem 3.

The operator (10) is Hamiltonian if and only if

$$g^{ij\alpha}=g^{ji\alpha}$$ (11a)

$$\frac{\partial g^{ij\alpha}}{\partial u^{k}}=b^{ij\alpha}_{k}+b^{ji\alpha}_{k}$$ (11b)

$$\sum_{(\alpha,\beta)}\left(g^{li\alpha}b^{jk\beta}_{l}-g^{lj\beta}b^{ik\alpha}_{l}\right)=0$$ (11c)

$$\sum_{(i,j,k)}\left(g^{li\alpha}b^{jk\beta}_{l}-g^{lj\beta}b^{ik\alpha}_{l}\right)=0$$ (11d)

$$\sum_{(\alpha,\beta)}\left[g^{li\alpha}\left(\frac{\partial b^{jk\beta}_{l}}{\partial u^{r}}-\frac{\partial b^{jk\beta}_{r}}{\partial u^{l}}\right)+b^{ij\alpha}_{l}b^{lk\beta}_{r}-b^{ik\alpha}_{l}b^{lj\beta}_{r}\right]=0$$ (11e)

$$g^{li\beta}\frac{\partial b^{jk\alpha}_{r}}{\partial u^{l}}-b^{ij\beta}_{l}b^{lk\alpha}_{r}-b^{ik\beta}_{l}b^{jl\alpha}_{r}=g^{lj\alpha}\frac{\partial b^{ik\beta}_{r}}{\partial u^{l}}-b^{ji\alpha}_{l}b^{lk\beta}_{r}-b^{jk\alpha}_{l}b^{il\beta}_{r}$$ (11f)

$$\frac{\partial}{\partial u^{s}}\left[g^{li\alpha}\left(\frac{\partial b^{jk\beta}_{l}}{\partial u^{r}}-\frac{\partial b^{jk\beta}_{r}}{\partial u^{l}}\right)+b^{ij\alpha}_{l}b^{lk\beta}_{r}-b^{ik\alpha}_{l}b^{lj\beta}_{r}\right]$$

$$+\frac{\partial}{\partial u^{r}}\left[g^{li\beta}\left(\frac{\partial b^{jk\alpha}_{l}}{\partial u^{s}}-\frac{\partial b^{jk\alpha}_{s}}{\partial u^{l}}\right)+b^{ij\beta}_{l}b^{lk\alpha}_{s}-b^{ik\beta}_{l}b^{lj\alpha}_{s}\right]$$

$$+\sum_{(i,j,k)}\left[b^{li\beta}_{r}\left(\frac{\partial b^{jk\alpha}_{s}}{\partial u^{l}}-\frac{\partial b^{jk\alpha}_{l}}{\partial u^{s}}\right)\right]+\sum_{(i,j,k)}\left[b^{li\alpha}_{s}\left(\frac{\partial b^{jk\beta}_{r}}{\partial u^{l}}-\frac{\partial b^{jk\beta}_{l}}{\partial u^{r}}\right)\right]=0,$$ (11g)

where $\sum$ represents the cyclic summation over indices.

In analogy to the $D=1$ case, the coefficients $g^{ij\alpha}$ are components of a symmetric (2,0)-tensor. Assuming that $g^{\alpha}$ is non-degenerate and defining $b^{ij\alpha}_{k}=-g^{il\alpha}\Gamma^{j\alpha}_{lk}$, we observe that $\Gamma^{\alpha}$ are Christoffels symbols of the Levi-Civita connection of $g^{\alpha}$.

2.3 The nonhomogeneous case

Following the content of the previous section, we extend the first-order homogeneous Hamiltonian operator (10) to the nonhomogeneous operator of type 1+0

$$P^{ij}=g^{ij\alpha}(u)\partial_{\alpha}+b^{ij\alpha}_{k}(u)u^{k}_{\alpha}+\omega^{ij}(u).$$ (12)

We need to find the conditions for which the operator (12) is Hamiltonian, namely the bracket $\{F,G\}=\int\frac{\delta F}{\delta u^{i}}P^{ij}\frac{\delta G}{\delta u^{j}}$ satisfies the property of skewsymmetry and Jacobi identity.

Firstly, let us recall the condition for the 0-order ultralocal operator $\omega^{ij}(u)$ to be Hamiltonian.

Theorem 4.

The operator $\omega^{ij}(u)$ is Hamiltonian if and only if

$$\omega^{ij}=-\omega^{ji}$$ (13a)

$$\omega^{il}\omega^{jk}_{,l}+\omega^{jl}\omega^{ki}_{,l}+\omega^{kl}\omega^{ij}_{,l}=0,$$ (13b)

where $\omega^{ij}_{,l}$ is a shorthand notation for $\frac{\partial\omega^{ij}}{\partial u^{l}}$.

Remark.

The condition is the same as for a Poisson bivector on a finite dimensional manifold.

In the previous section, we have already stated the Hamiltonian condition for a first-order homogeneous operator. For the nonhomogeneous operator (12), the Hamiltonian condition is the following:

Theorem 5.

The operator (12) is Hamiltonian if and only if the operator $g^{ij\alpha}\partial_{\alpha}+b^{ij\alpha}_{k}u^{k}_{\alpha}$ satisfies the conditions (11), $\omega^{ij}$ satisfies the conditions (13), and the following additional conditions hold true.

$$T^{ikj\alpha}=T^{kji\alpha}$$ (14a)

$$T_{,s}^{kji\alpha}=\sum_{(i,i,k)}b_{s}^{li\alpha}\omega_{,l}^{kj}+\left(b_{s,l}^{ik\alpha}-b_{l,s}^{ik\alpha}\right)\omega^{lj},$$ (14b)

where $T^{ikj\alpha}=g^{il\alpha}\omega^{kj}_{,l}-b^{ik\alpha}_{l}\omega^{lj}-b^{ij\alpha}_{l}\omega^{kl}$.

Proof.

The $\lambda$-bracket corresponding to $P$ is $\{u^{i}_{\lambda}u^{j}\}=P^{ji}(\lambda)=g^{ji\alpha}\lambda_{\alpha}+b^{ji\alpha}_{k}u^{k}_{\alpha}+\omega^{ji}$ by applying the theory of mPVA. We compute the PVA-Jacobi identity of the generators.

$$\{u^{i}_{\lambda}\{u^{j}_{\mu}u^{k}\}\}=\sum_{\begin{subarray}{c}h,s,l=1,\dots,N\\ \alpha,\beta=1,\dots,D\\ M\in\mathbb{Z}^{D}_{\geq 0}\end{subarray}}\frac{\partial\left(g^{kj\alpha}\mu_{\alpha}+b^{kj\alpha}_{h}u^{h}_{\alpha}+\omega^{kj}\right)}{\partial u^{l}_{(M)}}(\lambda+\partial)^{M}\left(g^{li\beta}\lambda_{\beta}+b^{li\beta}_{s}u^{s}_{\beta}+\omega^{li}\right)$$ (15a)

$$\{u^{j}_{\mu}\{u^{i}_{\lambda}u^{k}\}\}=\sum_{\begin{subarray}{c}h,s,l=1,\dots,N\\ \alpha,\beta=1,\dots,D\\ M\in\mathbb{Z}^{D}_{\geq 0}\end{subarray}}\frac{\partial\left(g^{ki\alpha}\lambda_{\alpha}+b^{ki\alpha}_{h}u^{h}_{\alpha}+\omega^{ki}\right)}{\partial u^{l}_{(M)}}(\mu+\partial)^{M}\left(g^{lj\beta}\mu_{\beta}+b^{lj\beta}_{s}u^{s}_{\beta}+\omega^{lj}\right)$$ (15b)

$$\{\{u^{i}_{\lambda}u^{j}\}_{\mu+\lambda}u^{k}\}=\sum_{\begin{subarray}{c}h,s,l=1,\dots,N\\ \alpha,\beta=1,\dots,D\\ M\in\mathbb{Z}^{D}_{\geq 0}\end{subarray}}\left[g^{kl\beta}\left(\lambda_{\beta}+\mu_{\beta}+\partial_{\beta}\right)+b^{kl\beta}_{h}u^{h}_{\beta}+\omega^{kl}\right](-\lambda-\mu-\partial)^{M}\frac{\partial\left(g^{ji\alpha}\lambda_{\alpha}+b^{ji\alpha}_{s}u^{s}_{\alpha}+\omega^{ji}\right)}{\partial u^{l}_{(M)}}.$$ (15c)

The equation (15a)-(15b)=(15c) is PVA-Jacobi (8), extracting the term $\lambda_{\alpha}$ leads to (14a), and extracting the term $u^{s}_{\alpha}$ leads to (14b). The remaining terms satisfy the Jacobi identity for the first-order homogeneous operator and the 0-order operator. ∎

3 Quasi-linear systems and compatibility in the multidimensional case

3.1 Quasi-linear systems

In this section, we focus on $(D+1)$-dimensional first-order quasilinear systems $(D>1)$

$$u^{i}_{t}=V^{i\alpha}_{l}(u)u^{l}_{\alpha}+W^{i}(u),\qquad i=1,2,\dots,n.$$ (16)

The systems (16) (also referred to as the type 1+0) are the sum of 1-order homogeneous systems and 0-order systems. As an example, consider the $N$-waves system [1] in two dimensions:

$$u_{ij,t}=\alpha_{ij}u_{ij,x}+\beta_{ij}u_{ij,y}+\sum_{k\neq i,j}(\alpha_{ik}-\alpha_{kj})u_{ik}u_{kj}$$ (17)

$$\text{for $i\neq j$ and $u_{ii}=0$, \qquad$i,j=1,\dots,N$},$$

where $\alpha_{ij}=\alpha_{ji}\in\mathbb{R},\beta_{ij}=\beta_{ji}\in\mathbb{R}$.

More in general, nonhomogeneous quasilinear systems can arise from scalar equations of arbitrary order. For example, consider the (2+1) dimensional heat equation $u_{t}=u_{xx}+u_{yy}$. By performing coordinate transformations $u^{1}=u,u^{2}=u^{1}_{x},u^{3}=u^{1}_{y}$ and interchanging variables x with t (or y with t) [22], we obtain the following system.

$$\left\{\begin{aligned} &u^{1}_{t}=u^{2}\\ &u^{2}_{t}=-u^{1}_{x}+u^{3}_{y}\\ &u^{3}_{t}=u^{2}_{y}\end{aligned}\right.$$ (18)

3.2 The cotangent covering

The core content of this paper is to investigate the necessary conditions for the multidimensional quasilinear system (16) to be the Hamiltonian system. Similarly to [24], we define these conditions as compatibility conditions and apply the theory of differential coverings [14] for computation. Since in [25] the author has already presented the procedure, we will provide a brief overview of the theoretical framework in the multidimensional case.

Let us consider a multidimensional first-order quasilinear system

$$F^{i}=u^{i}_{t}-V^{i\alpha}_{l}u^{l}_{\alpha}-W^{i}=0.$$ (19)

The vector function $\varphi=\varphi^{i}$ is said to be a symmetry of the system (16) if it satisfies the condition $l_{F}(\varphi)=0$, where $l_{F}$ is the Frechet derivative, expressed as

$$(l_{F})^{i}_{j}=\delta^{i}_{j}D_{t}-(V^{i\alpha}_{l,j}u^{l}_{\alpha}+W^{i}_{,j})-V^{i\alpha}_{j}D_{\alpha}.$$ (20)

Moreover, the adjoint of $l_{F}$ denotes :

$$(l^{*}_{F})^{i}_{j}=((l_{F})^{j}_{i})^{*}=-\delta^{j}_{i}D_{t}+(V^{j\alpha}_{i,l}u^{l}_{\alpha}-V^{j\alpha}_{l,i}u^{l}_{\alpha}-W^{j}_{,i})+V^{j\alpha}_{i}D_{\alpha}.$$ (21)

The corresponding vector function $\psi=\psi_{i}=\frac{\delta b}{\delta u^{i}}$ such that $l^{*}_{F}(\psi)=0$ is called a cosymmetry of the system (16).

We then introduce the so-called cotangent covering $T^{*}$ for the system (19), which is the system $F=0$ and $l^{*}_{F}(p)=0$ where $p$ is an auxiliary odd dependent variable corresponding to a cosymmetry. Explicitly, this is:

$$\left\{\begin{aligned} &u^{i}_{t}=V^{i\alpha}_{l}u^{l}_{\alpha}+W^{i}\\ &p_{i,t}=(V^{j\alpha}_{i,l}u^{l}_{\alpha}-V^{j\alpha}_{l,i}u^{l}_{\alpha}-W^{j}_{,i})p_{j}+V^{j\alpha}_{i}p_{j,\alpha}\end{aligned}\right.$$ (22)

where $p_{i,\alpha}=D_{\alpha}\psi_{i}$ and so on.

In [13], the following theorem is presented.

Theorem 6.

If the system (16) admits the Hamiltonian formulation, then

$$l_{F}\circ P=P^{*}\circ l_{F}^{*},$$ (23)

where the operator $P$ is a Hamiltonian structure for $F=0$.

The proof in [13] is independent from the dimension of $X$, namely on the number of spatial variables, so it is not only true for one-dimensional systems but also in our case. Therefore, the following result immediately follows.

Proposition 7.

If (16) is a Hamiltonian system with Hamiltonian structure $P^{ij}=P^{ij}_{S}\partial^{S}$, then

$$(l_{F})^{i}_{j}P^{jl}(p_{l})=0$$ (24)

for $(u,p)$ the cotangent covering of $F$.

Proof.

Apply the identity (23) to $p$, the RHS vanishes and the LHS is (24). ∎

We call (24) the compatibility condition. It serves as a necessary condition (but not sufficient) for the multidimensional quasilinear system (16) to be Hamiltonian. The condition (24) cannot guarantee that the operator $P$ is Hamiltonian nor it does ensure that, even in the case when $P$ is a Hamiltonian operator, the system $F$ is Hamiltonian.

However, by computing the condition (24), we can exclude certain operators. In other words, for a given quasilinear system, when searching for possible Hamiltonian structures, all the operators (Hamiltonian or not) that do not satisfy (24) should be discharged. Moreover, for a given operator satisfying the Hamiltonian conditions, it is also possible to search for compatible systems by computing condition (24).

3.3 Compatibility in the multidimensional case

The next step is to compute the compatibility conditions. We assume that the operator $P$ is Hamiltonian and consider the Hamiltonian formulation of the first-order quasilinear system

$$u^{i}_{t}=V^{i\alpha}_{l}u^{l}_{\alpha}+W^{i}=P^{ij}\frac{\delta H}{\delta u^{j}}.$$ (25)

Since we consider quasi-linear systems, we assume that $P$ is at most of differential order 1, and consequently assume that $H$ depends only on the variables $u$ (and not on jet variables). For a multidimensional nonhomogeneous quasilinear system of type 1+0, the corresponding Hamiltonian structure must be the form of $P=A+\omega$, where deg($A$)=1 and deg($\omega$)=0. Similarly, the 1-order homogeneous quasilinear system corresponds to the homogeneous operator $P$ such that deg($P$)= 1, and the 0-order quasilinear system corresponds to the 0-degree operator.

We directly consider the compatibility conditions of 1+0 order nonhomogeneous systems

$$u^{i}_{t}=V^{i\alpha}_{l}u^{l}_{\alpha}+W^{i},\qquad i=1,2,\dots n$$ (26)

which admit nonhomogeneous operators (satisfying the conditions (11), (13) and (14))

$$P^{ij}=g^{ij\alpha}\partial_{\alpha}+b^{ij\alpha}_{k}u^{k}_{\alpha}+\omega^{ij}$$ (27)

as their Hamiltonian structures.

Theorem 8.

If a nonhomogeneous quasilinear system (26) admits a Hamiltonian formulation with a Hamiltonian structure (27), the following conditions need to be satisfied.

$$\omega^{ij}_{,s}W^{s}-\omega^{sj}W^{i}_{,s}-\omega^{is}W^{j}_{,s}=0$$ (28a)

$$\sum_{(\alpha,\beta)}\left(V^{j\alpha}_{s}g^{is\beta}-V^{i\alpha}_{s}g^{sj\beta}\right)=0$$ (28b)

$$g^{ij\beta}_{,s}V^{s\alpha}_{l}+g^{is\beta}\left(V^{j\alpha}_{s,l}-V^{j\alpha}_{l,s}\right)+g^{is\alpha}V^{j\beta}_{s,l}+b^{is\alpha}_{l}V^{j\beta}_{s}-g^{sj\beta}V^{i\alpha}_{l,s}-g^{sj\beta}_{,l}V^{i\alpha}_{s}-b^{sj\alpha}_{l}V^{i\beta}_{s}=0$$ (28c)

$$\sum_{(\alpha,\beta)}\left[g^{is\beta}\left(V^{j\alpha}_{s,l}-V^{j\alpha}_{l,s}\right)+b^{ij\beta}_{s}V^{s\alpha}_{l}-V^{i\alpha}_{s}b^{sj\beta}_{l}\right]=0$$ (28d)

$$g^{is\beta}\left(V^{j\alpha}_{s,lk}-V^{j\alpha}_{l,sk}\right)+g^{is\alpha}\left(V^{j\beta}_{s,kl}-V^{j\beta}_{k,sl}\right)+b^{ij\beta}_{k,h}V^{h\alpha}_{l}+b^{ij\alpha}_{l,h}V^{h\beta}_{k}$$

$$+b^{ij\beta}_{h}V^{h\alpha}_{l,k}+b^{ij\alpha}_{h}V^{h\beta}_{k,l}+b^{is\beta}_{k}\left(V^{j\alpha}_{s,l}-V^{j\alpha}_{l,s}\right)+b^{is\alpha}_{l}\left(V^{j\beta}_{s,k}-V^{j\beta}_{k,s}\right)$$

$$-b^{hj\beta}_{k}V^{i\alpha}_{l,h}-b^{hj\alpha}_{l}V^{i\beta}_{k,h}-b^{hj\beta}_{k,l}V^{i\alpha}_{h}-b^{hj\alpha}_{l,k}V^{i\beta}_{h}=0$$ (28e)

$$g^{ij\alpha}_{,s}W^{s}-g^{is\alpha}W^{j}_{,s}-W^{i}_{,s}g^{sj\alpha}+\omega^{is}V^{j\alpha}_{s}-V^{i\alpha}_{s}\omega^{sj}=0$$ (28f)

$$-g^{is\alpha}W^{j}_{,sl}+b^{ij\alpha}_{l,s}W^{s}+b^{ij\alpha}_{s}W^{s}_{,l}-b^{is\alpha}_{l}W^{j}_{,s}-W^{i}_{,s}b^{sj\alpha}_{l}$$

$$+\omega^{ij}_{,s}V^{s\alpha}_{l}+\omega^{is}\left(V^{j\alpha}_{s,l}-V^{j\alpha}_{l,s}\right)-V^{i\alpha}_{l,s}\omega^{sj}-V^{i\alpha}_{s}\omega^{sj}_{,l}=0,$$ (28g)

where $\sum$ represents the cyclic summation over indices.

Proof.

We note that the cotangent covering of the system (26) is (22). In order to obtain the necessary conditions, it is required to compute $l_{F}(P(p))=0$.

$$l_{F}(P(p))=(\delta_{ij}D_{t}-V^{i\alpha}_{l,j}-W^{i}_{,j}-V^{i\alpha}_{j}D_{\alpha})(g^{js\beta}p_{s,\beta}+b^{js\beta}_{k}u^{k}_{\beta}p_{s}+\omega^{js}p_{s}).$$ (29)

By setting the degree of $p_{i}$ to 0 and extracting the terms of degree 0, namely the term $p_{j}$, we can obtain (28a). Indeed, this is also the compatibility condition of the 0-order system $u^{i}_{t}=W^{i}(u)$ with respect to the 0-order ultralocal Hamiltonian operator. Continuing with the extraction of terms of degree 2, the terms $p_{j,\alpha\beta}$, $u^{l}_{\alpha}p_{j,\beta}$, $u^{l}_{\alpha\beta}p_{j}$, and $u^{l}_{\alpha}u^{k}_{\beta}p_{j}$ correspond to (28b), (28c), (28d), and (28e), respectively. Similarly, these conditions are the compatibility conditions for first-order homogeneous systems admitting hydrodynamic-type operators as their Hamiltonian structures. Finally, extracting the remaining terms $p_{j,\alpha},u^{l}_{\alpha}p_{j}$, we can obtain (28f) and (28g). ∎

3.4 Compatibility and Poisson cohomology

It is possible to offer a different interpretation of the property (23), which clearly highlights the relation between sufficiency and necessity of the compatibility condition (24). Note that in this paragraph we denote both the Poisson bivector and the Hamiltonian operator defining it with the same letter $P$, to keep a consistent notation with the previous formulae. Let us briefly recall the notion of Poisson cohomology for a Hamiltonian operator.

Given a Poisson bivector $P\in\Lambda^{2}$, the condition $[P,P]=0$ guarantees, by the properties of the Schouten bracket, that its adjoint action on poly-vector fields $d_{P}=[P,\cdot]\colon\Lambda^{k}\to\Lambda^{k+1}$ squares to 0, turning the space of poly-vector fields into a cochain complex known as the Poisson-Lichnerowicz complex. The linear operator $d_{P}$ is therefore called the Poisson differential.

$$0\to\Lambda^{0}\xrightarrow{d_{P}}\Lambda^{1}\xrightarrow{d_{P}}\Lambda^{2}\xrightarrow{d_{P}}\cdots\xrightarrow{d_{P}}\Lambda^{p}\xrightarrow{d_{P}}\cdots$$ (30)

Note that, in the infinite-dimensional case, the Poisson-Lichnerowicz complex does not terminate, while $0$-vector fields are identified with local functionals. The cohomology of such complex is referred to as the Poisson cohomology:

$$H(P)=\bigoplus\limits_{p\geq 0}H^{p}(P)=\bigoplus\limits_{p\geq 0}\frac{\mathrm{Ker}\,d_{P}\colon\Lambda^{p}\to\Lambda^{p+1}}{\mathrm{Im}\,d_{P}\colon\Lambda^{p-1}\to\Lambda^{p}}=\bigoplus\limits_{p\geq 0}\frac{Z^{p}}{B^{p}}$$ (31)

As in any cochain complex, the elements of $Z^{p}$ are called cocycles, while the elements of $B^{p}$ are called coboundaries. We also introduce a second grading on $H^{p}$ (resp. $Z^{p}$, $B^{p}$, $\Lambda^{p}$), corresponding to the differential order of the elements, by the rule $\deg u=0$, $\deg D_{\alpha}=1$, $\deg u^{i}_{Q}=q_{1}+\cdot+q_{D}$. The corresponding homogeneous components of $H^{p}$ (resp. for $Z,B,\ldots$) are denoted by $H^{p}_{d}$ ($Z^{p}_{d},\ldots$), $d\geq 0$.

A vector field that preserves the Hamiltonian structure (in the sense $\{X(F),G\}+\{F,X(G)\}=X(\{F,G\})$, where $\{\cdot,\cdot\}$ is the Poisson bracket defined in (6)) is called a symmetry of the structure and it is a $1$-cocycle. Indeed, the condition can equivalently be written as $\mathcal{L}_{X}(P)=0$ and the Lie derivative of $P$ along the vector field $X$ is, by the definition of Schouten bracket, $-d_{P}X$. On the other hand, a Hamiltonian vector field is obtained by the adjoint action of $P$ on a local functional (the Hamiltonian functional), $X_{H}=P(\delta H)=-[P,H]$, and it is therefore a $1$-coboundary. The elements of the first cohomology group $H^{1}(P)$ are the so-called non-Hamiltonian symmetries. They are vector fields that preserve the Hamiltonian structure without being Hamiltonian.

Proposition 9.

Let $X$ be the evolutionary vector field corresponding to the system $F^{i}=u^{i}_{t}-X^{i}=0$, where $\{X^{i}\}$ is the characteristics of $X$, and $P$ the Poisson bivector defined by a local Hamiltonian operator. Then condition (23) is equivalent to $X$ being a $1$-cocycle in the Poisson-Lichnerowicz complex of $P$ (namely, $X$ is a symmetry of $P$).

Proof.

This observation is essentially computational. First, let us explicitly write condition (23) in terms of $F^{i}=u^{i}_{t}-X^{i}$ and $P=P^{ij}_{S}\partial^{S}$. For any covector $\psi=\{\psi_{i}\}$, we have

$$\begin{split}\left(l_{F}(P\psi)\right)^{j}&=D_{t}\left(P^{ji}_{S}D^{S}\psi_{i}\right)-\frac{\partial X^{j}}{\partial u^{m}_{Q}}D^{Q}\left(P^{mi}_{S}D^{S}\psi_{i}\right)\\ &=\frac{\partial P^{ji}_{S}}{\partial u^{m}_{Q}}\left(D^{Q}X^{m}\right)D^{S}\psi_{i}+P^{ji}_{S}D_{t}D^{S}\psi_{i}-\frac{\partial X^{j}}{\partial u^{m}_{Q}}D^{Q}\left(P^{mi}_{S}D^{S}\psi_{i}\right)\end{split}$$ (32)

where we use the multi-index notation $D^{S}$ to denote $D_{1}^{s_{1}}D_{2}^{s_{2}}\cdots D^{s_{D}}_{D}$. Similarly,

$$\begin{split}\left(P^{*}(l_{F}^{*}\psi)\right)^{j}&=(-D)^{S}\left(P^{ij}_{S}\left(-D_{t}\psi_{i}-(-D)^{Q}\frac{\partial X^{m}}{\partial u^{i}_{Q}}\psi_{m}\right)\right)\\ &=-\left[(-D)^{S}P^{ij}D_{t}\psi_{i}\right]-(-D)^{S}\left[P^{mj}_{S}(-D)^{Q}\left(\frac{\partial X^{i}}{\partial u^{m}_{Q}}\psi_{i}\right)\right].\end{split}$$ (33)

Now we subtract (32) from (33), utilize the commutativity between spatial and time derivatives $[D_{\alpha},D_{t}]=0$ and the skewsymmetry property for $P$, namely $P^{ji}_{S}D^{S}=-(-D^{S})P^{ij}_{S}$, to obtain that condition (23) is, explicitly,

$$\left[P^{jm}_{S}D^{S}(-D)^{Q}\frac{\partial X^{i}}{\partial u^{m}_{Q}}+\frac{\partial X^{j}}{\partial u^{m}_{Q}}D^{Q}P^{mi}_{S}D^{S}-\left(D^{Q}X^{m}\right)\frac{\partial P^{ji}_{S}}{\partial u^{m}_{Q}}D^{S}\right]\psi_{i}=0$$ (34)

for any $\psi=\{\psi_{i}\}$.

On the other hand, the cocycle condition $d_{P}X=0$ can be expressed using the formalism of PVAs [5] as

$$\left\{X(u^{i})_{\lambda}u^{j}\right\}-\left\{u^{i}{}_{\lambda}X(u^{j})\right\}-X\left(\left\{u^{i}{}_{\lambda}u^{j}\right\}\right)=0$$ (35)

for all the pairs $(i,j)$ and a $\lambda$ bracket $\{u^{i}{}_{\lambda}u^{j}\}=P^{ji}_{S}\lambda^{S}$. This can be explicitly computed with the master formula (7), obtaining

$$P^{jm}_{S}(\lambda+\partial)^{S}(-\lambda-\partial)^{Q}\frac{\partial X^{i}}{\partial u^{m}_{Q}}+\frac{\partial X^{j}}{\partial u^{m}_{Q}}(\lambda+\partial)^{Q}P^{mi}_{S}\lambda^{S}-\left(D^{Q}X^{m}\right)\frac{\partial P^{ji}_{S}}{\partial u^{m}_{Q}}\lambda^{S}=0.$$ (36)

Now it is enough to observe that the total derivatives $D$ in (34) may act on both $\psi_{i}$ and differential expressions with $u$ variables. Denoting a total derivative $D^{S}$ acting on $\psi_{i}$ by $\lambda^{S}$ and a total derivative acting on a differential expression of the $u$ variables by $\partial^{S}$ (this second case, indeed, coincides with the definition of $\partial$ we used in presenting multidimensional PVAs), by the Leibniz rule we have $D^{S}=(\partial+\lambda)^{S}$. This replacement gives exactly (36). ∎

It is now clear that the sufficiency of Condition (23) (and of its equivalent form (24)) is closely related to the triviality of $H^{1}(P)$. Indeed, the existence of any nontrivial cohomology class denotes the existence of compatible evolutionary systems without Hamiltonian formulation.

The study of the Poisson cohomology for multidimensional Hamiltonian operators has shown that, in general, such an object is highly non-trivial [5, 3, 4]; we can therefore expect that we will need to discard some solutions of system (24) to identify bona fide Hamiltonian systems. An independent knowledge of the Poisson cohomology of the operator can help us in this task. Conversely, finding the solutions of (24) that fail to be Hamiltonian provides an explicit method to compute the (first) Poisson cohomology of $P$.

While the Poisson cohomology for homogeneous operators, and in particular for structures of hydrodynamic type, has been previously studied in the aforementioned references, the one for nonhomogeneous ones has not been computed yet, at best of our knowledge. We will hereinafter demonstrate how $H^{1}$ for a nonhomogeneous operator of order $1+0$ is closely related to the bi-Hamiltonian cohomology [16]. Finally, in the following section we will show that in the $N=D=2$ case (extensively discussed in [5]) it is isomorphic to the cohomology of the leading order operator.

Proposition 10.

Let us consider a $1+0$ order Poisson bivector $P=P_{1}+P_{0}$ with nondegenerate $P_{0}$ (the underscripts, here and in the following, denote the differential degree of the objects). Then $H^{1}_{\leq 1}(P)=BH^{1}_{1}(P_{1},P_{0})$.

Proof.

Let $d_{P}$ (resp. $d_{0}$ and $d_{1}$) the Poisson differentials associated to the bivectors $P$, $P_{0}$ and $P_{1}$, and note that $\deg P_{0}=0$, $\deg P_{1}=1$. The first observation is that $d_{0}$ and $d_{1}$ form a differential bi-complex, namely $d_{0}^{2}=0$, $d_{1}^{2}=0$, $d_{0}d_{1}=-d_{1}d_{0}$. The first two identities are obvious because $P_{0}$ and $P_{1}$ are Poisson bivectors, while the third one corresponds to their compatibility in the bi-Hamiltonian sense. All the three identities can be obtained by considering the homogeneous parts of $d_{P}^{2}=(d_{0}+d_{1})^{2}=0$. Now let us consider elements of $Z^{1}_{\leq 1}(P)$. They must be vector fields $X=X_{0}+X_{1}$ such that $d_{P}X=0$; expanding the condition by differential degree we have

$$\left\{\begin{array}[]{l}d_{0}X_{0}=0\\ d_{0}X_{1}+d_{1}X_{0}=0\\ d_{1}X_{1}=0\end{array}\right.$$ (37)

In particular, then, $X_{0}\in Z^{1}_{0}(P_{0})$ and $X_{1}\in Z^{1}_{1}(P_{1})$. It has been known for long time [11] that, for invertible bivectors $P_{0}$ (and trivial De Rham cohomology of the target space $M$, which we always assume is a ball) $H^{1}(P_{0})=0$, so there exists a local functional $h_{0}$ such that $X_{0}=d_{0}h_{0}$; on the other hand, $H^{1}_{1}(P_{1})$ is in general non trivial: therefore $X_{1}=d_{1}k_{0}+\xi_{1}$, where $k_{0}$ is a local functional (that, for homogeneity reasons, must be of differential order 0) and $\xi_{1}\in H^{1}_{1}(P_{1})$. We can now replace the expressions for $X_{0}$ and $X_{1}$ in the second equation of (37). This gives us the identity

$$d_{0}d_{1}k_{0}+d_{0}\xi_{1}+d_{1}d_{0}h_{0}=d_{0}\left(d_{1}(k_{0}-h_{0})+\xi_{1}\right)=0$$ (38)

which, thanks to the triviality of $H^{1}(P_{0})$, is

$$d_{1}(k_{0}-h_{0})+\xi_{1}=d_{0}h_{1}.$$ (39)

This means that, for $X=X_{0}+X_{1}$, we can write

$$X=\left(d_{0}+d_{1}\right)h_{0}+d_{1}(k_{0}-h_{0})+\xi_{1}=d_{P}h_{0}+d_{0}h_{1}.$$ (40)

However, this expression shows that the elements of $H^{1}_{\leq 1}(P)$ are of the form $d_{0}h_{1}=d_{1}g_{0}+\xi_{1}$, where we simply denote $k_{0}-h_{0}=g_{0}$. In particular we have $d_{0}^{2}h_{1}=0$ and, thanks to (39), $d_{1}d_{0}h_{1}=0$, so $H^{1}_{\leq 1}(P)=\left(\mathrm{Ker}\,d_{0}\cap\mathrm{Ker}\,d_{1}\right)^{1}_{1}$. This is exactly the definition of $BH^{1}_{1}(P_{0},P_{1})$, see for instance [9, 16]. ∎

4 Examples

4.1 N=D=2 systems

4.1.1 The nonhomogeneous Hamiltonian structures

In this section, we focus on the study of the Poisson brackets of 1+0 hydrodynamic type in the case $D=N=2$, along with their compatible systems. In this case we consider the three possible normal forms for the Hamiltonian structures of hydrodynamic type [10, 19]

$$P_{1}=\begin{pmatrix}1&0\\ 0&0\\ \end{pmatrix}\frac{d}{dx}+\begin{pmatrix}0&0\\ 0&1\\ \end{pmatrix}\frac{d}{dy}$$ (41a)

$$P_{2}=\begin{pmatrix}0&1\\ 1&0\\ \end{pmatrix}\frac{d}{dx}+\begin{pmatrix}0&0\\ 0&1\\ \end{pmatrix}\frac{d}{dy}$$ (41b)

$$P_{3}=\begin{pmatrix}2u&v\\ v&0\\ \end{pmatrix}\frac{d}{dx}+\begin{pmatrix}0&u\\ u&2v\\ \end{pmatrix}\frac{d}{dy}+\begin{pmatrix}u_{x}&u_{y}\\ v_{x}&v_{y}\\ \end{pmatrix}$$ (41c)

and the ultralocal structure

$$\omega=\begin{pmatrix}0&f(u,v)\\ -f(u,v)&0\\ \end{pmatrix}$$ (42)

which is Hamiltonian for any choice of $f$.

Now one should investigate which kind of nonhomogeneous structures are allowed starting with these pairs of Hamiltonian structures $(P_{i},\omega)$. The research method involves applying Theorem 5 to obtain compatibility conditions, leading to the following results:

Proposition 11.

The operators $P=P_{1}+\omega$ and $P_{2}+\omega$ are Hamiltonian if and only if $f=\eta$ ($\eta$ is a constant). The operator $P_{3}+\omega$ is Hamiltonian if and only if $f=0$.

4.1.2 Compatibility conditions and Hamiltonian equations

We can build upon the results from the previous subsection and search for compatible quasilinear systems $u^{i}_{t}=V^{i\alpha}_{l}u^{l}_{\alpha}+W^{i}$. The corresponding method involves computing the seven conditions outlined in Theorem 8.

Proposition 12.

The solutions of the compatibility system corresponding to the Hamiltonian operator

$$P_{1}+\omega=\begin{pmatrix}1&0\\ 0&0\\ \end{pmatrix}\frac{d}{dx}+\begin{pmatrix}0&0\\ 0&1\\ \end{pmatrix}\frac{d}{dy}+\begin{pmatrix}0&\eta\\ -\eta&0\\ \end{pmatrix}$$ (43)

are as follows:

$$V^{2x}_{1}=V^{1y}_{2}=0 V^{2x}_{2}=\alpha V^{1y}_{1}=\beta$$ (44)

$$V^{1x}_{2}=V^{2y}_{1}=\frac{\partial^{2}M}{\partial u\partial v} V^{1x}_{1}=\frac{\partial^{2}M}{\partial u^{2}}+\alpha V^{2y}_{2}=\frac{\partial^{2}M}{\partial v^{2}}+\beta$$

$$W^{1}=\eta\frac{\partial M}{\partial v} W^{2}=-\eta\frac{\partial M}{\partial u}$$

where $M$ is an arbitrary function of $u$ and $v$, and $\alpha,\beta$ are arbitrary constants.

Proposition 13.

The solutions of the compatibility system corresponding to the Hamiltonian operator

$$P_{2}+\omega=\begin{pmatrix}0&1\\ 1&0\\ \end{pmatrix}\frac{d}{dx}+\begin{pmatrix}0&0\\ 0&1\\ \end{pmatrix}\frac{d}{dy}+\begin{pmatrix}0&\eta\\ -\eta&0\\ \end{pmatrix}$$ (45)

are as follows:

$$V^{1y}_{2}=0 V^{1y}_{1}=\alpha$$ (46)

$$V^{2x}_{1}=\frac{\partial^{2}M}{\partial u^{2}} V^{2y}_{2}=\frac{\partial^{2}M}{\partial v^{2}}+\alpha$$

$$V^{2x}_{2}=\frac{\partial^{2}M}{\partial u\partial v}+\beta V^{1x}_{1}=\frac{\partial^{2}M}{\partial u\partial v}+\beta$$

$$V^{2y}_{1}=\frac{\partial^{2}M}{\partial u\partial v} V^{1x}_{2}=\frac{\partial^{2}M}{\partial v^{2}}$$

$$W^{1}=\eta\frac{\partial M}{\partial v} W^{2}=-\eta\frac{\partial M}{\partial u}$$

where $M$ is an arbitrary function of $u$ and $v$, and $\alpha,\beta$ are arbitrary constants.

Proposition 14.

The solutions of the compatibility system corresponding to the Hamiltonian operator

$$P_{3}=\begin{pmatrix}2u&v\\ v&0\\ \end{pmatrix}\frac{d}{dx}+\begin{pmatrix}0&u\\ u&2v\\ \end{pmatrix}\frac{d}{dy}+\begin{pmatrix}u_{x}&u_{y}\\ v_{x}&v_{y}\\ \end{pmatrix}$$ (47)

are as follows:

$$V^{1x}_{1}=2u\frac{\partial^{2}M}{\partial u^{2}}+\frac{\partial M}{\partial u}+v\frac{\partial^{2}M}{\partial u\partial v} V^{1x}_{2}=2u\frac{\partial^{2}M}{\partial u\partial v}+v\frac{\partial^{2}M}{\partial u\partial v}$$ (48)

$$V^{2x}_{1}=v\frac{\partial^{2}M}{\partial u^{2}} V^{2x}_{2}=\frac{\partial M}{\partial u}+v\frac{\partial^{2}M}{\partial u\partial v}$$

$$V^{1y}_{1}=\frac{\partial M}{\partial v}+u\frac{\partial^{2}M}{\partial u\partial v} V^{1y}_{2}=u\frac{\partial^{2}M}{\partial v^{2}}$$

$$V^{2y}_{1}=u\frac{\partial^{2}M}{\partial u^{2}}+2v\frac{\partial^{2}M}{\partial u\partial v} V^{2y}_{2}=\frac{\partial M}{\partial v}+u\frac{\partial^{2}M}{\partial u\partial v}+2v\frac{\partial^{2}M}{\partial v^{2}}$$

where $M$ is an arbitrary function of $u$ and $v$.

To demonstrate our results, let us briefly present as an example the proof of Proposition 12.

Proof.

By computing the conditions in Theorem 8, we can obtain the following results.

$$W^{1}_{,1}+W^{2}_{,2}=0$$ (49a)

$$V^{2x}_{1}=V^{1y}_{2}=0$$ (49b)

$$V^{2x}_{2,1}=V^{2x}_{2,2}=V^{1y}_{1,1}=V^{1y}_{1,2}=0$$ (49c)

$$V^{1x}_{2}=V^{2y}_{1}$$ (49d)

$$V^{1x}_{1,2}=V^{1x}_{2,1}$$ (49e)

$$V^{2y}_{1,2}=V^{2y}_{2,1}$$ (49f)

$$W^{1}_{,1}=\eta V^{1x}_{2}$$ (49g)

$$W^{2}_{,2}+\eta V^{2y}_{1}=0$$ (49h)

$$W^{1}_{,2}+\eta V^{1y}_{1}=\eta V^{2y}_{2}$$ (49i)

$$W^{2}_{,1}+\eta V^{1x}_{1}=\eta V^{2x}_{2}$$ (49j)

From (49a), it can be deduced that

$$W^{1}=\frac{\partial P}{\partial v}$$ (50)

$$W^{2}=-\frac{\partial P}{\partial u},$$

where $P=P(u,v)$ is an arbitrary function.

Upon observing (49c), we deduce that

$$V^{2x}_{2}=\alpha$$ (51)

$$V^{1y}_{1}=\beta,$$

where $\alpha$ and $\beta$ are arbitrary constants.

Continuing with (49d), (49e), and (49f), we can obtain

$$V^{1x}_{1}=\frac{\partial^{2}R}{\partial u^{2}}$$ (52)

$$V^{2y}_{2}=\frac{\partial^{2}R}{\partial v^{2}}$$

$$V^{1x}_{2}=V^{2y}_{1}=\frac{\partial^{2}R}{\partial u\partial v},$$

where $R=R(u,v)$ is an arbitrary function.

Subsequently, substituting these obtained results into the remaining equations (49g), (49h), (49i), and (49j), we can establish the following relationships:

$$\frac{\partial^{2}R}{\partial u\partial v}=\frac{1}{\eta}\frac{\partial^{2}P}{\partial u\partial v}$$ (53)

$$\frac{\partial^{2}R}{\partial u^{2}}=\frac{1}{\eta}\frac{\partial^{2}P}{\partial u^{2}}+\alpha$$

$$\frac{\partial^{2}R}{\partial v^{2}}=\frac{1}{\eta}\frac{\partial^{2}P}{\partial v^{2}}+\beta.$$

Finally, by rescaling $P$ to write $M=\frac{1}{\eta}P$, we obtain (44). ∎

The Hamiltonian equations of the operator $P_{1}+\omega$ can be immediately compared with (44). On the other hand, given a Hamiltonian function $h(u,v)$, the associated equation has the form

$$u_{t}=\frac{\partial^{2}h}{\partial u^{2}}u_{x}+\frac{\partial^{2}h}{\partial u\partial v}v_{x}+\eta\frac{\partial h}{\partial v}$$ (54a)

$$v_{t}=-\eta\frac{\partial h}{\partial u}+\frac{\partial^{2}h}{\partial u\partial v}u_{y}+\frac{\partial^{2}h}{\partial v^{2}}v_{y}.$$ (54b)

Our result implies the existence of certain non-Hamiltonian solutions depending on arbitrary constants $\alpha$ and $\beta$ in the compatibility conditions.

Similarly, upon examining the remaining two operators, it can be observed that for both $P_{1}+\omega$ and $P_{2}+\omega$ exist non-Hamiltonian solutions, while all the solutions of the compatibility condition for $P_{3}$ are Hamiltonian.

4.1.3 Agreement with the Poisson cohomology

Let us recall from Section 3.4 that the non-Hamiltonian systems selected by the compatibility conditions are in one-to-one correspondence to elements of $BH^{1}_{1}(P_{i}+\omega)$, and such elements are of the form $d_{i}g_{0}+\xi_{1}$ ($d_{i}:=d_{P_{i}}$), with $\xi_{1}\in H^{1}_{1}(P_{i})$. The relevant results about the first cohomology group for the Poisson bracket defined by $P_{1},P_{2},P_{3}$ have been presented in [5]. We have

$$H^{1}_{0}(P_{1})\cong\mathbb{R}^{2}\qquad H^{1}_{0}(P_{2})\cong\mathbb{R}^{2}\qquad H^{1}_{0}(P_{3})\cong 0$$ (55)

$$H^{1}_{1}(P_{1})\cong\mathbb{R}^{2}\qquad H^{1}_{1}(P_{2})\cong\mathbb{R}^{2}\qquad H^{1}_{1}(P_{3})\cong 0$$

For both $P_{1}$ and $P_{2}$, we have that a compatible (but not Hamiltonian) vector field must be of the form

$$X=d_{1}g_{0}+\xi_{1}$$ (56)

for $\xi^{1}\in H^{1}_{1}$ and a function $g_{0}=g_{0}(u,v)$, subject to the condition $d_{0}(d_{1}g_{0}+\xi^{1})=0$. The explicit form for a representative in the cohomology classes $H^{1}_{1}(P_{1})$ and $H^{1}_{1}(P_{2})$ are, respectively,

$$\xi_{1}^{(1)} =\begin{pmatrix}\beta u_{y}\\ \alpha v_{x}\end{pmatrix} \xi_{1}^{(2)} =\begin{pmatrix}\alpha(u_{y}-v_{x})\\ -\beta u_{y}\end{pmatrix}$$ (57)

for $(\alpha,\beta)\in\mathbb{R}^{2}$. This choice of names for the constants matches the results of Section 4.1.2. Explicitly solving $d_{0}\left(d_{1}g_{0}^{(i)}+\xi^{(i)}_{1}\right)=0$ we find $g_{0}$ and

$$X^{(1)} =\begin{pmatrix}\alpha u_{x}+\beta u_{y}\\ \alpha v_{x}+\beta v_{y}\end{pmatrix} X^{(2)} =\begin{pmatrix}\alpha u_{y}+\beta u_{x}\\ \alpha v_{y}+\beta v_{x}\end{pmatrix},$$ (58)

a 2-dimensional space of solutions exactly corresponding to the non-Hamiltonian solutions in (44) and (46).

On the other hand, $H^{1}_{1}(P_{3})$ is trivial, so the only possible non-Hamiltonian vector field must be of the form $X=d_{1}g_{0}$ and such that $d_{0}d_{1}g_{0}=0$. This also means that $d_{0}g_{0}\in H^{1}_{0}(P_{3})$, which is trivial too, implying $d_{0}g_{0}\in B^{1}_{0}(P_{1})$. However, $d_{1}$ is an operator of differential order $1$, therefore $B^{1}_{0}(P_{1})=0$ and $g_{0}\in\ker d_{0}$, i.e. $g_{0}=\mathrm{const}$. Finally, constants are in the kernel of $d_{1}$, too, which means that non-Hamiltonian solutions of the compatibility equation do not exist.

4.2 3-waves system

Let us consider a real reduction of the 3-waves system ($N=3$ in (17)), i.e. $u_{12}=-u_{21}=u^{3}$, etc., as follows.

$$\left\{\begin{aligned} &u^{1}_{t}=au^{1}_{x}+du^{1}_{y}+(b-c)u^{2}u^{3}\\ &u^{2}_{t}=bu^{2}_{x}+eu^{2}_{y}+(c-a)u^{3}u^{1}\\ &u^{3}_{t}=cu^{3}_{x}+fu^{3}_{y}+(a-b)u^{1}u^{2}\end{aligned}\right.$$ (59)

Where $a,b,c,d,e,f$ are parameters and distinct from each other.

We need to search for the Hamiltonian structure and the Hamiltonian function of the system (59). According to the conditions in Theorem 8, the compatibility solutions of the Hamiltonian operator (27) can be deduced.

Proposition 15.

The system (59) is compatible with a Hamiltonian operator $P$ if and only if

$$f=\frac{d(c-b)+e(a-c)}{a-b}$$ (60)

and

$$P=\begin{pmatrix}S&0&0\\ 0&S&0\\ 0&0&S\\ \end{pmatrix}\partial_{x}+\begin{pmatrix}\frac{d-e}{a-b}S&0&0\\ 0&\frac{d-e}{a-b}S&0\\ 0&0&\frac{d-e}{a-b}S\\ \end{pmatrix}\partial_{y}+\begin{pmatrix}0&Su^{3}&-Su^{2}\\ -Su^{3}&0&Su^{1}\\ Su^{2}&-Su^{1}&0\\ \end{pmatrix},$$ (61)

where $S$ is an arbitrary constant.

Proof.

By performing calculations (28b), (28c), (28d), and (28e), we can deduce that $g^{ij\alpha}=0$ ($i\neq j$), and $g^{ii\alpha}$ depends only on $u^{i}$. Additionally, we obtain $b^{ij\alpha}_{l}=0$ ($i\neq j$ or $i\neq l$), and $b^{ii\alpha}_{i}$ depends only on $u^{i}$. Continuing with the calculation (28f), we can further deduce that

$$g^{iix}=C_{i}\hskip 54.00002ptg^{iiy}=D_{i}$$ (62)

$$\omega^{12}=\frac{C_{1}(c-a)+C_{2}(b-c)}{b-a}u^{3}$$

$$\omega^{13}=\frac{C_{1}(a-b)+C_{3}(b-c)}{c-a}u^{2}$$

$$\omega^{23}=\frac{C_{2}(a-b)+C_{3}(c-a)}{c-b}u^{1},$$

where $C_{i}$ and $D_{i}$ are arbitrary constants. Finally, utilizing equations (28a) and (28g), it can be deduced that

$$b^{ii\alpha}_{i}=0\hskip 85.00012ptC_{1}=C_{2}=C_{3}=S$$ (63)

$$D_{1}=D_{2}=D_{3}=S\frac{f-e}{c-b}=S\frac{f-d}{c-a}=S\frac{d-e}{a-b},$$

where $S$ is an arbitrary constant.

Therefore, a constraint must be imposed, namely

$$f=\frac{d(c-b)+e(a-c)}{a-b}.$$ (64)

By synthesizing the aforementioned results, we obtain the operator (61), which satisfies the Hamiltonian conditions. ∎

As prescribed by the general theory, the compatibility of the system (59) with

$$f=\frac{d(c-b)+e(a-c)}{a-b}$$ (65)

with the Hamiltonian structure (61) is only a necessary condition for (59) to be a Hamiltonian system. In fact, (59) is Hamiltonian with a 1+0-order operator if and only if

$$\frac{f}{c}=\frac{e}{b}=\frac{d}{a}$$ (66)

and

$$P=\begin{pmatrix}S&0&0\\ 0&S&0\\ 0&0&S\\ \end{pmatrix}\partial_{x}+\begin{pmatrix}\frac{d}{a}S&0&0\\ 0&\frac{d}{a}S&0\\ 0&0&\frac{d}{a}S\\ \end{pmatrix}\partial_{y}+\begin{pmatrix}0&Su^{3}&-Su^{2}\\ -Su^{3}&0&Su^{1}\\ Su^{2}&-Su^{1}&0\\ \end{pmatrix},$$ (67)

with the corresponding Hamiltonian functional

$$H=\int\frac{a}{2S}{(u^{1})}^{2}+\frac{b}{2S}{(u^{2})}^{2}+\frac{c}{2S}{(u^{3})}^{2},$$ (68)

where $S$ is an arbitrary constant.

4.2.1 Relation with the Poisson cohomology

Note that in this example the results of Proposition 10 cannot be applied, and we cannot directly relate the Poisson cohomology of $P$ with the biHamiltonian cohomology of the two homogeneous Poisson structures. Indeed, for $N=3$ we cannot rely either on the isomorphism between $H(P)$ and $H_{DR}(M)$ established by Lichnerowicz [15], because $P_{0}$ is not invertible, or on the triviality result for the first and second smooth Poisson cohomology group of $\mathfrak{so}^{*}(3)$ proved by Conn [6] and Ginzburg-Weinstein [12], because we consider a broader class of vector fields: indeed, we can explicitly obtain non-Hamiltonian symmetries of $P_{0}$, invalidating the assumption $X_{0}=d_{0}h_{0}$. Moreover, a linear change of independent variables as the one discussed in [3] simplifies the computations for the first order operator and quickly leads to $H^{1}_{1}(P_{1})\cong\mathbb{R}^{6}$.

However, the results of the previous paragraph show that there only exist a one-real parameter family of real 3-wave systems compatible with a Hamiltonian operators but not Hamiltonian – indeed, by rescaling (67) and (68) we have

$$\begin{pmatrix}u^{1}_{t}\\ u^{2}_{t}\\ u^{3}_{t}\end{pmatrix}=\begin{pmatrix}a\left(u^{1}_{x}+\kappa u^{1}_{y}\right)+(b-c)u_{2}u_{3}\\ b\left(u^{2}_{x}+\kappa u^{2}_{y}\right)+(c-a)u_{3}u_{2}\\ c\left(u^{3}_{x}+\kappa u^{3}_{y}\right)+(a-b)u_{1}u_{2}\end{pmatrix}-\lambda\begin{pmatrix}au^{1}_{y}\\ bu^{2}_{y}\\ cu^{3}_{y}\end{pmatrix}$$ (69)

where $\kappa=d/a$ and the system is not Hamiltonian for $\lambda\neq 0$.

5 Conclusion

In this paper, the necessary conditions for a multidimensional first-order quasilinear system to admit a Hamiltonian structure are discussed. Firstly, we computed the conditions under which the 1+0 order nonhomogeneous hydrodynamic operator satisfies the property of skewadjointness and $[P,P]=0$ using the PVA theory. Next, leveraging the theory of differential coverings, we have computed the compatibility conditions of multidimensional first-order quasilinear systems which admit 1+0 order operators as their Hamiltonian structures. Finally, two examples for the $N=D=2$ system and a real reduction of the 3-waves system are presented. For the first case, we consider three Hamiltonian structures of hydrodynamic type along with the ultralocal structure. Based on the previous conclusions, we investigate which pairs of Hamiltonian structures possess Hamiltonian properties and search for compatible systems with such structures. The appearance of certain arbitrary constants in the compatible quasilinear systems indicates the existence of non-Hamiltonian systems. This consistency aligns with the results of the first Poisson cohomology group and highlights the essentially cohomological nature of the compatibility conditions. For the second case, we have obtained the Hamiltonian operator compatible with the 3-waves system and the constraints on the parameters in the system by computing compatibility conditions. Under specific conditions, we have derived the Hamiltonian structure and Hamiltonian functional of the 3-waves system.

Acknowledgement

This work was sponsored by the National Science Foundation of China (Grant no. 12101341), Ningbo City Yongjiang Innovative Talent Program and Ningbo University Talent Introduction and Resarch Initiation Fund. M. C. wishes to thank P. Vergallo and I. Marcut for the useful insights.