Abstract
Pointvortex dynamics describe idealized, nonsmooth solutions to the incompressible Euler equations on twodimensional manifolds. Integrability results for few pointvortices on various domains is a vivid topic, with many results and techniques scattered in the literature. Here, we give a unified framework for proving integrability results for \(N=2\), 3, or 4 pointvortices (and also more general Hamiltonian systems), based on symplectic reduction theory. The approach works on any twodimensional manifold with a symmetry group; we illustrate it on the sphere, the plane, the hyperbolic plane, and the flat torus. A systematic study of integrability is prompted by advances in twodimensional turbulence, bridging the longtime behaviour of 2D Euler equations with questions of pointvortex integrability. A gallery of solutions is given in the appendix.
1 Introduction
The governing equations for an incompressible inviscid fluid were formulated by Euler in 1757 [11] and have since been cardinal in the ever growing field of hydrodynamics. On a general orientable Riemannian manifold M, Euler’s equations are
where \({\mathbf {v}}\) is a vector field on M describing the motion of infinitesimal fluid particles, p is the pressure function, and \(\nabla _{\mathbf {v}}\) denotes the covariant derivative along \({\mathbf {v}}\). Arnold [2] discovered a rich and beautiful geometry underlying these equations; they can be interpreted as the geodesic equation for a rightinvariant Riemannian metric on the infinitedimensional Lie group of volume preserving diffeomorphism of M. For an entrylevel introduction to Arnold’s discovery see the lecture notes by Modin [32]. More detailed expositions are given by Arnold and Khesin [3] and by Khesin and Wendt [18].
101 years after Euler’s influential paper on incompressible fluids, Helmholtz [14] showed that the 2D Euler equations exhibit special solutions consisting of a finite number of pointvortices. These are nonsmooth solutions where \({\text {curl}}{\mathbf {v}}\), called vorticity, is a finite sum of Dirac distributions
Here, \(\Gamma _i \in \mathbb {R}\backslash \{0\}\) is the strength of vortex i and \({\mathbf {r}}_i \in M\) is its position (M can be any twodimensional orientable Riemannian manifold). If \(G:M\times M \rightarrow \mathbb {R}\) denotes the Green’s function on M for the Laplacian, i.e., for \({\mathbf {r}}\in M\) we have
then the equations of motion for N pointvortices are given by
Here, \(\nabla ^\bot _{{\mathbf {r}}_i}\) is the skew gradient with respect to \({\mathbf {r}}_i\), i.e., the operation which first takes the gradient with respect to \({\mathbf {r}}_i\) and then rotates this vector by \(\pi /2\) in positive orientation. The pointvortex equations stated like this are rather abstract; in section 2 below we present these equations explicitly for different manifolds M.
Kirchhoff [21] was the first to point out that the pointvortex system has a Hamiltonian structure; in the planar case \({\mathbf {r}}_i = (x_i,y_i)\) the canonical coordinates are given by pairs of Euclidean coordinates \(x_i\) and \(y_i\) scaled by the strengths \(\Gamma _i\). Thus, pointvortex dynamics fits within the realm of Hamiltonian mechanics, a central observation for understanding its phase space geometry.
The classical way of thinking is that pointvortices constitute a formal ansatz for solutions of the Euler equations. By following Arnold’s approach, however, Marsden and Weinstein [28] showed that pointvortex solutions automatically appear from a systematic study of the symmetries of the system.
The interest in pointvortex dynamics has been growing steadily since the work of Helmholtz. Historically, it is fair to say that the field has emerged in two branches. One branch, originating from Onsager [38], is to study a large number \(N\gg 1\) of pointvortices via the tools provided by statistical mechanics for Hamiltonian dynamics. Onsager’s work laid out a statistical theory of hydrodynamics for predicting the formation of coherent structures in 2D turbulence. Since then, mathematicians and physicists have followed up on his approach, which has led to many deep and beautiful results; see [25] and references therein. However, the question concerning the longtime behaviour of classical solutions to the 2D Euler equations is still unanswered; numerical simulations suggest a more complicated generic behaviour than predicted by theories based on statistical mechanics (see Sect. 7 below).
The other branch is to study a few number of pointvortices, and ask whether the dynamics is integrable. Early contributors are Gröbli [12] and Poincaré [39], who explicitly (Gröbli) and implicitly (Poincaré) proved integrability of the planar \(N=3\) case. Since then, many results were obtained, on various domains, on integrability, equilibrium solutions, and relative equilibrium solutions; for an overview see the survey by Aref [1] and the monograph by Newton [37].
The purpose of this paper is to advocate symplectic reduction theory—an underused tool in dynamical systems—as a universal framework for proving integrability results of pointvortex dynamics (and variations thereof) on twodimensional manifolds. That symplectic reduction theory can be used to obtain integrability results was already stressed in the paper by Marsden and Weinstein [28]. Indeed, concerning the planar case they say: “For \(N=3\) one can check that the motion is (completely) integrable in the sense that the (nonabelian) reduced phase spaces are points”. They then go on to say: “However one can also see that the dynamics of 3 pointvortices is (completely) integrable by exhibiting 3 independent integrals in involution...” In the literature since, mostly the second method—to find enough first integrals in involution—has been used. Kimura [20], in particular, used this method to prove integrability of the \(N=3\) case on any surface of constant curvature. However, in many ways, the symplectic reduction approach is more fundamental, as it starts from only the Hamiltonian structure and the symmetries.
Concerning \(N=4\), Ziglin [46] gave the earliest results on nonintegrability. A few years later Marsden and Weinstein [28] said: “The motion of four vortices is generally believed to be chaotic.” Today, we know that \(N=4\) is a borderliner: depending on the geometry of the domain, and on other conditions such as total circulation or momentum, the \(N=4\) case may be integrable. For example, Eckhardt [10] showed that the planar case yields integrable dynamics when restricted to the subset of solutions with vanishing total circulation and momentum. More recently, Sakajo [42] showed that the \(N=4\) case on the sphere with vanishing total linear momentum is integrable. The proofs of these results are based on explicit calculations in specific coordinates. A main point of our paper is to show that the same results quickly fall out from the symplectic reduction framework essentially without any calculations.
We remark that there is also another approach to integrability for pointvortex dynamics, based on noncommutative integrability where the integrals of motion form a Lie algebra [5, 6]. It is possible that the symplectic reduction approach considered here and the noncommutative integrability approach are two viewpoints of the same theory. However, such a connection is not investigated here.
Our motivation for questions of integrability of pointvortex dynamics originates from recent numerical results for 2D Euler equations indicating that integrability of pointvortex dynamics, rather than prevailing statistical mechanics based theories, is central for predicting the longtime behaviour of solutions [34].
The paper is organized as follows. In Sect. 2, we describe, in more detail, the pointvortex equations and their Hamiltonian structures on the sphere, the plane, the hyperbolic plane, and the flat torus. Integrability results are given in Sect. 3. A brief review of symplectic reduction theory is given in Sect. 4. In Sect. 5, we prove the integrability results using symplectic reduction. A summary of the known nonintegrability results are given in Sect. 6. Thereafter, Sect. 7 contains an outline of the connection between longtime behaviour of 2D Euler equations and pointvortex integrability. Finally, in Appendix A, we provide a gallery of pointvortex solutions—a sort of ‘visual summary’ of the paper.
2 PointVortex Equations and Their Conservation Laws
In this section, we give more detailed presentations of pointvortex dynamics in four specific cases: the sphere \(\mathbb {S}^2\), the plane \(\mathbb {R}^2\), the hyperbolic plane \(\mathbb {H}^2\), and the flat torus \(\mathbb {T}^2=\mathbb {R}^2/\mathbb {Z}^2\). From the point of view of symplectic reduction, these cases correspond to different structures of the symmetry group: compact nonAbelian, semidirect product, noncompact but semisimple, and Abelian. These structures illustrate well the different scenarios that can occur in symplectic reduction, as we shall see in Sect. 4 below.
2.1 The Sphere
Consider the sphere \(\mathbb {S}^2\) embedded in Euclidean 3space as the subset of unitary vectors. The standard area form equips \(\mathbb {S}^2\) with a symplectic structure, denoted \(\Omega _{\mathbb {S}^2}\). The corresponding Poisson bracket is
We may think of \(\nabla F\) and \(\nabla G\) as the intrinsic Riemannian gradients on \(\mathbb {S}^2\), but equally well as Euclidean gradients for extensions of F and G to a neighbourhood of \(\mathbb {S}^2\) in \(\mathbb {R}^3\); the parts of \(\nabla F\) and \(\nabla G\) not orthogonal to \(\mathbb {S}^2\) are cancelled out in the triple product.
The phase space of N pointvortices on \(\mathbb {S}^2\) is given by
equipped with the direct product symplectic structure
where as before, \(\Gamma _i\) should be interpreted as the vortex strengths and \({\mathbf {u}}\) and \({\mathbf {w}}\) are tangent vectors of \(P^N_{\mathbb {S}^2}\). The equations of motion for a Hamiltonian \(H = H({\mathbf {r}}_1,\ldots ,{\mathbf {r}}_N)\) are then given by
where \(X_H\) is called the Hamiltonian vector field. Notice that, for any choice of Hamiltonian, the right hand side is tangent to \(\mathbb {S}^2\) so the dynamics evolves on \((\mathbb {S}^2)^N\). Since the Green’s function on \(\mathbb {S}^2\) is given by \(G({\mathbf {r}},{\mathbf {r}}') = \frac{1}{4\pi }\log (1{\mathbf {r}}\cdot {\mathbf {r}}')\), the specific Hamiltonian corresponding to pointvortex dynamics is
which leads to the pointvortex equations on the sphere
Let us now turn to the symmetries for the system (4). Clearly, the Hamiltonian (3) is invariant under the diagonal action of \(\mathrm {SO}(3)\) on \(P^N_{\mathbb {S}^2}\). That is, if \(R\in \mathrm {SO}(3)\) then
Furthermore, this action preserves the symplectic form \(\Omega _{N}\) (since it is isometric and, therefore, preserves the area form on each sphere). By the Hamiltonian version of Nöther’s theorem (cf. [27, Thm. 11.4.1]) the symmetry, together with the fact that the action is symplectic, leads to a conservation law stated in terms of the momentum map associated with the action of \(\mathrm {SO}(3)\) on \(P^N_{\mathbb {S}^2}\). Recall that the momentum map \({\mathbf {J}}:P^N_{\mathbb {S}^2}\rightarrow \mathfrak {so}(3)^*\) is defined by the condition that, for any \(\varvec{\xi }\in \mathfrak {so}(3)\), the Hamiltonian vector field
corresponds to the infinitesimal action of \(\xi \) on \(P^N_{\mathbb {S}^2}\). If we identify \(\mathfrak {so}(3)\) with \(\mathbb {R}^3\) via the usual hat map (cf. [27, Eq. 9.2.7]), then the infinitesimal action is given by
i.e., infinitesimal rotation of each \({\mathbf {r}}_i\) about the axis \(\varvec{\xi }\). From (2) it follows that if
then the right hand side of (5) equal that of (6). Thus, identifying \(\mathfrak {so}(3)^*\) with \(\mathbb {R}^3\) via the Euclidean inner product, we get
From Nöther’s theorem, it follows that the three components of \({\mathbf {J}}\), called total linear momentum, are conserved for the pointvortex flow (4), or, more generally, for any \(\mathrm {SO}(3)\)invariant Hamiltonian flow on \(P^N_{\mathbb {S}^2}\).
We now come to a property of momentum maps that is central in the symplectic reduction framework, namely equivariance. A momentum map is called equivariant if it commutes with the symplectic action of the underlying symmetry group. Thus, the momentum map \({\mathbf {J}}:P^N_{\mathbb {S}^2}\rightarrow \mathfrak {so}(3)^*\) is equivariant if
where \({{\,\mathrm{Ad}\,}}_R^*\) is the coadjoint action, here defined as matrixvector multiplication. It is easy to check that this condition is fulfilled, so the momentum map \({\mathbf {J}}\) on \(P^N_{\mathbb {S}^2}\) is indeed equivariant. A general result states that if the symmetry group is semisimple, then an equivariant momentum map always exists. In the next section, however, we encounter a symplectic action that does not have an equivariant momentum map, which has consequences for the symplectic reduction.
2.2 The Plane
Consider now the Euclidean plane \(\mathbb {R}^2\), with standard coordinates \({\mathbf {r}} = (x,y)\). The canonical symplectic structure is \(\Omega _{\mathbb {R}^2} \equiv dx\wedge d y\), which gives the Poisson bracket
where \(\nabla \) is the standard gradient operator and \(\nabla ^\perp \) is the skewgradient: \(\nabla ^\perp G = (\partial _y G,\partial _x G)\). The phase space of N pointvortices on \(\mathbb {R}^2\) is given by
equipped with the scaled direct product symplectic structure
The equations of motion for a Hamiltonian \(H = H(\mathbf{r}_1,\ldots ,{\mathbf {r}}_N)\) are the set of scaled canonical Hamiltonian equations
Since the Green’s function for the Laplacian on \(\mathbb {R}^2\) is \(G({\mathbf {r}},{\mathbf {r}}') = \frac{1}{4\pi }\log (\mathbf{r}{\mathbf {r}}' ^2)\), the Hamiltonian for pointvortex dynamics is
From (8), it follows that the corresponding pointvortex equations on the plane are
Concerning the symmetries of the equations (10), it is clear that the Hamiltonian (9) is invariant to the diagonal action of the Euclidean group \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\). We recall that the semidirect group product is defined as:
for \(R_1,R_2\in \mathrm {SO}(2)\) and \({\mathbf {u}}_1,\mathbf{u}_2\in \mathbb {R}^2\) and the action of \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\) on \(\mathbb {R}^2\) is:
for \(R\in \mathrm {SO}(2)\) and \({\mathbf {u}},{\mathbf {r}}\in \mathbb {R}^2\). The action of the Euclidean group is clearly symplectic, being areapreserving. Therefore, we get the following momentum map \(\mathbf{J }:P^N_{\mathbb {R}^2}\rightarrow (\mathfrak {so}(2)\ltimes \mathbb {R}^2)^*\):
whose first component is the total angular momentum, and the second component contains the total linear momenta. However, the momentum map (11) is not, in general, equivariant; the condition for equivariance is that the total circulation vanishes, i.e., \(\sum _{i}\Gamma _i = 0\). In fact, the coadjoint action of \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\) on \((\mathfrak {so}(2)\ltimes \mathbb {R}^2)^*\) is:
for \((R,{\mathbf {u}})\in \mathrm {SO}(2)\ltimes \mathbb {R}^2\) and \((\xi ,{\mathbf {w}})\in (\mathfrak {so}(2)\ltimes \mathbb {R}^2)^*\) (see [26, ch. 4.2]). Therefore,
and
Hence, to satisfy the equivariance condition:
we must have that the total circulation is zero, i.e., \(\sum _{i}\Gamma _i = 0\).
To construct a momentum maps that is always equivariant, we shall explore a less obvious symmetry of equations (10). Consider the group \(G\subset \mathrm {GL}(2N,\mathbb {R})\) generated by the infinitesimal generators \(\xi ,\eta \) defined by
We notice that \(\eta \) is the infinitesimal generator of \(\mathrm {SO}(2)\) and \(\xi \) is the infinitesimal generator of a 1dimensional group K. The symplectic form \(\Omega _N\) on \(\mathbb {R}^{2N}\) has the following representation in matrix form:
It is straightforward to check that \([\xi ,\eta ]=0\), \(\Omega _N\xi + \xi ^\top \Omega _N=0\) and \(\Omega _N\eta + \eta ^\top \Omega _N=0\). Therefore, \(G=\mathrm {SO}(2)\times K\) is Abelian and the action of G on \(\mathbb {R}^{2N}\) is symplectic. The action of G has momentum map \({\mathbf {L}}:P^N_{\mathbb {R}^2}\rightarrow \mathfrak {so}(2)^*\oplus \mathfrak {k}^*\):
It is straightforward to check that \({\mathbf {L}}\) is an equivariant momentum map. Indeed, on one hand, SO(2) acts as a diagonal isometry of \(P^N_{\mathbb {R}^2}\), and therefore, it preserves the Euclidean norms. On the other hand, for the action of K on \({\mathbf {L}}\), we get:
Hence, since \(SO(2)\times K\) is Abelian, the momentum map \(\mathbf{L}\) is equivariant.
Remark 1
The geometric interpretation of the oneparameter group K acting on \(({\mathbf {r}}_1,\ldots ,{\mathbf {r}}_N)\) is the following. First, if the total linear momentum \({\mathbf {L}}\) vanishes, then K acts as the identity. Second, denote by \(\Gamma =\sum _i \Gamma _i\) the total circulation. If \({\mathbf {L}}\ne 0\), then K rotates synchronously the point vortices \(({\mathbf {r}}_1,\ldots ,\mathbf{r}_N)\) in circles of radius \(r = \frac{\left {\mathbf {L}} \right }{\Gamma }\) and with centres \({\mathbf {C}}_i\) which satisfy \({\mathbf {C}}_i + \frac{{\mathbf {L}}}{\Gamma }= {\mathbf {r}}_i\). If the total circulation vanishes, i.e., \(\Gamma =0\), then the radius becomes infinite and the motion is along straight lines.
2.3 The Hyperbolic Plane
Consider the hyperbolic plane model \(\mathbb {H}^2=\lbrace (x,y,z)\in \mathbb {R}^3z^2x^2y^2=1,z>0\rbrace \), with Riemannian structure induced by the bilinear form on \(\mathbb {R}^3\):
for any \({\mathbf {a}},{\mathbf {b}}\in \mathbb {R}^3\) and \(L=\text{ diag }(1,1,1)\). As on the sphere, the volume form on \(\mathbb {H}^2\) defines a symplectic structure \(\Omega _{\mathbb {H}^2}\). Defining \({\mathbf {a}}\times _L {\mathbf {b}} :=L ({\mathbf {a}}\times {\mathbf {b}})\), the corresponding symplectic form is given by
The phase space of N pointvortices on \(\mathbb {H}^2\) is given by
equipped with the socalled direct product symplectic structure
with \(\Gamma _i\) as before and \({\mathbf {u}}\), \({\mathbf {w}}\) tangent vectors of \(P^N_{\mathbb {H}^2}\). The equations of motion for a Hamiltonian \(H = H({\mathbf {r}}_1,\ldots ,{\mathbf {r}}_N)\) are
For pointvortices, the Hamiltonian is
which gives the pointvortex equations on the hyperbolic plane [15, 16, 36]
These equations constitute a Lie–Poisson system on \((\mathfrak {sl}(2,\mathbb {R})^*)^N\simeq (\mathbb {R}^3,\times _L)^N\). Equations (13) constrain the vortices to the hyperboloid \(x^2+y^2z^2=1\). Furthermore, the \(\mathrm {SL}(2,\mathbb {R})\) symmetry of (13) gives, analogously to \(\mathbb {S}^2\), the equivariant momentum map
For further background on pointvortices on the hyperbolic plane, see [15, 16, 36]. For hydrodynamics on the hyperbolic plane, see [17].
2.4 The Flat Torus
The easiest way to work with the flat torus \(\mathbb {R}^2/\mathbb {Z}^2\) is to extend everything to \(\mathbb {R}^2\) and then assure that all operations and equations are \(2\pi \)periodic in both directions. Thus, with definitions as in subsection 2.2, the Hamiltonian for pointvortex dynamics on the flat torus, expanded to a Hamiltonian on the plane, is given by
where \(h(x,y) = \dfrac{x^2}{2\pi }+\sum _{m=\infty }^{+\infty }\log \left( \dfrac{\cosh (x2\pi m)\cos (y)}{\cosh (2\pi m)}\right) \). The corresponding equations of motion (cf. [45]) are
These equations are (equivalent to) the pointvortex equation on a flat torus, as explained in the following remark.
Remark 2
The equations (15) can be derived in two ways. The first, proposed in [45], starts from a \(2\pi \)periodic distribution of vortices in the pointvortex equation on \(\mathbb {R}^2\) and sums up the infinite number of contributions given by the different vortices, obtaining in the limit a well defined vector field for the right hand side in (15). As shown in [45], this vector field is \(2\pi \)periodic. Therefore, it can be seen as a vector field on \(\mathbb {T}^2\). The second way consist of explicitly calculate the Green’s function of the Laplacian on a flat twotorus. The calculations given in [24] confirm the equivalence of the two approaches, and the description of the Green’s function in terms of the First Jacobi Theta function guarantees the smoothness of the vector field on \(P^N_{\mathbb {R}^2}\). We stress that this equivalence is crucial in the analysis of the equations (15). In fact, on one hand, we have that the twotorus has nontrivial first de Rham cohomology, whereas on the other hand for \(\mathbb {R}^2\) the cohomology is trivial, and so for any infinitesimal symmetry, we get a conservation law. In particular, we get two more first integrals as stated below.
The Hamiltonian (14) is invariant with respect to the diagonal action of \(\mathbb {R}^2\). Since this action is symplectic, we get the (linear) momentum map
As noticed in remark 2, it is convenient to consider the pointvortex equations in \(\mathbb {R}^2\) according to the Hamiltonian (14). In this setting, the \(\mathbb {R}^2\) action on \(\mathbb {R}^2\) is free and proper, being the map R in Definition 5 invertible. However, the momentum map (16) is in general not equivariant. In fact, since \(\mathbb {R}^2\) is an Abelian group, the \({{\,\mathrm{Ad}\,}}\) operator is the identity. Hence, the equivariance equation \({{\,\mathrm{Ad}\,}}^*_{g^{1}}\circ \mu =\mu \circ \rho _g\) implies that \(\mu =\mu \circ \rho _g\). We notice that this is true only if the circulation is zero, i.e., \(\Gamma _1+\dots +\Gamma _N=0\). Moreover, since \(\mathbb {R}^2\) is neither compact nor semisimple, Theorem [27, Thm 11.5.2] cannot be applied to get a modified equivariant momentum map. We notice that considering the pointvortex equation directly on the flat twotorus and the action of \(U(1)^2\) instead of \(\mathbb {R}^2\) would have in principle guarantee the equivariance of the momentum map, but the nontrivial first group of De Rham cohomology of the twotorus would have prevent the existence of the momentum map itself.
3 Integrability Results
In this section, we state results on integrability for the pointvortex equations presented in the previous section. As we shall see in the sections below, all of these results, which are valid not just for the pointvortex equations but for all invariant Hamiltonian systems on the respective phase spaces, follow directly from the symplectic reduction framework. Most of the results are known since before; the point here is to demonstrate how the results naturally fall out from symplectic reduction in a purely geometric way. In particular, symplectic reduction gives a clear geometric understanding for why the vanishing momentum and circulation cases are structurally different from the nonvanishing cases.
Before proceeding to the specific manifolds \(\mathbb {S}^2\), \(\mathbb {R}^2\), \(\mathbb {H}^2\), and \(\mathbb {T}^2\), we recall the following general definition:
Definition 1
Let P be a phase space manifold for a dynamical system and assume that P is acted upon by a Lie group G. A solution \(t\mapsto {\mathbf {r}}(t) \in P\) is called a Grelative equilibrium if there exists an \({\mathbf {r}}_0\in P\) and \(\varvec{\xi }\in \mathfrak {g}\) such that \({\mathbf {r}}(t) = \exp (t\varvec{\xi })\cdot {\mathbf {r}}_0\).
3.1 The Sphere
Integrability of the pointvortex equations (4) for \(N=3\) is given by Sakajo [41]. The proof is based on specific coordinates for which one finds three first integrals in involution. Eight years later Sakajo [42] also proved integrability of the vanishing momentum \(N=4\) case, essentially by reducing it to the \(N=3\) case. See also the work by Borisov et al. [7].
Theorem 1
(\(\mathbb {S}^2\) with nonvanishing momentum) Consider Hamilton’s equations (2) on \(P_{\mathbb {S}^2}^N\) for an \(\mathrm {SO}(3)\) invariant Hamiltonian, restricted to the part of phase space with nonvanishing total linear momentum

If \(N=2\) all solution are relative equilibria for the \(\mathrm {SO}(2)\)action, the isotropy group of the momentum \({\mathbf {J}}\).

If \(N=3\) the system is completely integrable (but solutions are typically not relative equilibria).
In particular, these results are valid for the pointvortex equations (4) on \(P_{\mathbb {S}^2}^N\).
Theorem 2
(\(\mathbb {S}^2\) with vanishing momentum)] Consider Hamilton’s equations (2) on \(P_{\mathbb {S}^2}^N\) for an \(\mathrm {SO}(3)\) invariant Hamiltonian, restricted to the part of phase space with vanishing total linear momentum

If \(N=2\) all solutions are equilibria of antipodal points.

If \(N=3\) all solution are relative equilibria for the \(\mathrm {SO}(3)\)action.

If \(N=4\) the system is completely integrable (but solutions are typically not relative equilibria).
In particular, these results are valid for the pointvortex equations (4) on \(P_{\mathbb {S}^2}^N\).
3.2 The Plane
Integrability of pointvortices in the plane were the first to be studied. Early results for \(N=3\) were given by Gröbli [12] and by Poincaré [39]. The \(N=4\) result was given by Eckhardt [10].
Theorem 3
(\(\mathbb {R}^2\) with nonvanishing circulation or momentum) Consider Hamilton’s equations (8) on \(P_{\mathbb {R}^2}^N\) for an \(\mathrm {SO}(2)\times K\) invariant Hamiltonian, restricted to the case of nonvanishing circulation
or restricted to the part of phase space with nonvanishing total linear momentum

If \(N=2\) and \(\sum _i \Gamma _i \ne 0\) (nonvanishing circulation), then the solutions are \(\mathrm {SO}(2)\times K\) relative equilibria.

If \(N=2\) and \(\sum _i \Gamma _i = 0\) (vanishing circulation), then the solutions are \(\mathbb {R}^2\)relative equilibria (travelling vortex dipoles).

If \(N=3\) the system is completely integrable (but solutions are typically not relative equilibria).
In particular, these results are valid for the pointvortex equations (10) on \(P_{\mathbb {R}^2}^N\).
Theorem 4
(\(\mathbb {R}^2\) with vanishing circulation and momentum) Consider Hamilton’s equations (8) on \(P_{\mathbb {R}^2}^N\) for an \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\) invariant Hamiltonian, restricted to case of vanishing circulation and total linear momentum:

If \(N=2\), the pointvortex equation (10) is not defined.

If \(N=3\) all solutions are \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\) relative equilibria.

If \(N=4\) the system is completely integrable (but solutions are typically not relative equilibria).
In particular, these results are valid for the pointvortex equations (10) on \(P_{\mathbb {R}^2}^N\).
3.3 The Hyperbolic Plane
Integrability results on the hyperbolic plane reflect the results on the sphere, as from a symplectic reduction point of view, the two settings are almost the same, with an equivariant momentum map for a semisimple threedimensional symmetry group.
Theorem 5
(\(\mathbb {H}^2\) with nonvanishing momentum) Consider Hamilton’s equations (12) on \(P_{\mathbb {S}^2}^N\) for an \(\mathrm {SL}(2)\) invariant Hamiltonian, restricted to the part of phase space with nonvanishing total linear momentum

If \(N=2\) all solution are relative equilibria for the action of the isotropy subgroup of the momentum \({\mathbf {J}}\).

If \(N=3\) the system is completely integrable (but solutions are typically not relative equilibria).
In particular, these results are valid for the pointvortex equations (13) on \(P_{\mathbb {H}^2}^N\).
Theorem 6
(\(\mathbb {H}^2\) with vanishing momentum) Consider the pointvortex equation (13) restricted to the part of phase space with vanishing total linear momentum

If \(N=2\) the pointvortex equation (13) is not defined.

If \(N=3\) all solution are relative equilibria for the \(\mathrm {SL}(2)\)action.

If \(N=4\) the system is completely integrable (but solutions are typically not relative equilibria).
In particular, these results are valid for the pointvortex equations (13) on \(P_{\mathbb {H}^2}^N\).
3.4 The Flat Torus
Integrability results for pointvortex dynamics on the flat torus were given by Stremler and Aref [43], in the case of zero total circulation, and Kilin and Artemova [19]. As they do, we take here the viewpoint that the phase space is embedded as periodic solutions in \(P_{\mathbb {R}^2}^N\).
Theorem 7
(\(\mathbb {T}^2\) with nonvanishing circulation) Consider Hamilton’s equations (8) on \(P_{\mathbb {R}^2}^N\) for an \(\mathbb {R}^2\) invariant Hamiltonian, restricted to the case of nonvanishing circulation:

If \(N=2\) the system is completely integrable (but solutions are typically not relative equilibria).
In particular, these results are valid for the pointvortex equations (15) on \(P_{\mathbb {T}^2}^N\).
Theorem 8
(\(\mathbb {T}^2\) with vanishing circulation) Consider Hamilton’s equations (8) on \(P_{\mathbb {R}^2}^N\) for an \(\mathbb {R}^2\) invariant Hamiltonian, restricted to the case of vanishing circulation

If \(N=2\) all solutions are \(\mathbb {R}^2\)relative equilibria.

If \(N=3\) the system is completely integrable (but solutions are typically not relative equilibria).
In particular, these results are valid for the pointvortex equations (15) on \(P_{\mathbb {T}^2}^N\).
4 Symplectic Reduction Theory
To prove the theorems stated in Sect. 3, we need to recall some definitions and notations. We will denote in the following the smooth action of a Lie group G on a manifold M with:
for each \(g\in G,p\in M\).
Definition 2
(Symplectic action) Let G be a Lie group acting on a smooth symplectic manifold \((M,\omega )\). Then the action of G is said to be symplectic if \(\rho ^*_g\omega = \omega \), for every \(g\in G\).
Definition 3
(Momentum map) Let G be a Lie group and \(\mathfrak {g}\) its Lie algebra. Assume that G acts symplectically on a smooth symplectic manifold \((M,\omega )\) and that \(\xi _p:=d_e\rho _{\exp (\xi )}(p)\), for every \(p\in M\), is its inifinitesimal action. Then a map \(\mu :M\rightarrow \mathfrak {g}^*\), fulfilling
is called a momentum map.
When G is compact, and a momentum map exists, it can always be chosen to satisfy the equivariance identity:
for \(\xi ,\eta \in \mathfrak {g}\) and \(p\in M\), where the bracket \(\lbrace \cdot ,\cdot \rbrace \) is the Poisson bracket on M. In this case the Lie group action will be said to be Hamiltonian. A crucial fact on momentum maps is that they are conserved quantities for Hamiltonian systems with symmetries.
Theorem 9
[27] Let M be a smooth symplectic manifold and let G be a Lie group with Hamiltonian action on it and momentum map \(\mu :M\rightarrow \mathfrak {g}^*\). Let \(H:M\rightarrow \mathbb {R}\) be a smooth function such that, for each \(g\in G,x\in M\), \(H(g\cdot x)=H(x)\). Then, if \(\phi ^H_t\) is the flow of the Hamiltonian vector field \(X_H\), we have that \((\phi ^H_t)^*\mu =\mu \), for each \(t\ge 0\).
Let us now continue defining some properties of the group actions.
Definition 4
(Free action) Let G be a Lie group acting on a smooth manifold M. Then the action of G is said to be free if, for each \(x\in M\), \(g\cdot x=x\) implies \(g=e\).
Definition 5
(Proper action) Let G be a Lie group acting on a smooth manifold M. Then the action of G is said to be proper if the map \(R :G\times M\rightarrow M\times M\), \(R(g,p)=(\rho _g(p),p)\) is a proper map of topological spaces.
Definition 6
(Isotropy subgroup) Let G be a Lie group acting on a smooth manifold M. Then the set:
is a Lie subgroup of G called isotropy subgroup.
Let us now recall the symplectic reduction theorem:
Theorem 10
(Symplectic reduction theorem, cf. [26]) Let G be a Lie group with Hamiltonian free and proper action on a smooth symplectic manifold \((M,\omega )\) and let \(\mu :M\rightarrow \mathfrak {g}^*\) be an equivariant momentum map with respect to this action. Then, for each \(\phi \in \mathfrak {g}^*\), the quotient:
is a symplectic manifold of dimension \(d:=\dim (M)2\dim (G_\phi )\), with symplectic form \(\omega _\phi \) uniquely characterized by
where \(\pi :\mu ^{1}(\phi )\rightarrow \mu ^{1}(\phi )/G_\phi \) is the projection and \(\iota : \mu ^{1}(\phi )\rightarrow M\) is the inclusion.
5 Proofs by Symplectic Reduction
In this section we prove the results stated in Sect. 3. All of our proofs are based on the symplectic reduction Theorem 10, which makes them very streamlined. Essentially, the proofs consist in checking that the action is free and proper, that the momentum map is equivariant, and then counting the dimensions of the reduced phase space.
5.1 The Sphere
For the sphere, the \(\mathrm {SO}(3)\) action on \(P_{\mathbb {S}^2}^N\) is free for any \(N\ge 3\) and for \(N=2\) if the points are not antipodal. It is also proper since \(\mathrm {SO}(3)\) is compact. Furthermore, the momentum map (7) is equivariant, as one can directly check.
Proof of Theorem 1
For \(N=2\), the \(\mathrm {SO}(3)\) action is free, unless the two vortices are antipodal, which from (4) implies that solutions are equilibria. Hence, Theorem 10 tells us that systems evolves on a zerodimensional manifold, since \(G_\phi \simeq \mathrm {SO}(2)\). Thus, the reconstructed motion is an \(\mathrm {SO}(3)\) relative equilibrium, i.e., a steady rotation of the initial positions.
For \(N=3\), the \(\mathrm {SO}(3)\) action is always free and Theorem 10 tells us that the system evolves on a twodimensional manifold. Any Hamiltonian system on a twodimensional manifold is integrable, so the reconstructed system is also integrable, as follows, for example, from the standard Floquet theory (cf. [33, Proof of thm 3.1]). \(\square \)
Proof of Theorem 2
For \(N=2\), the vanishing momentum condition (17) implies that \({\mathbf {r}}_1,{\mathbf {r}}_2\) are antipodal points with equal strength. From (4) it follows directly that solutions are equilibria.
For \(N=3\), the \(\mathrm {SO}(3)\) action is free and Theorem 10 tells us that the reduced system evolves on a zerodimensional manifold, so solutions are relative equilibria.
For \(N=4\), the \(\mathrm {SO}(3)\) action is free and by Theorem 10 the system evolves on a twodimensional reduced manifold, so the dynamics is integrable. \(\square \)
5.2 The Plane
The action of \(\mathrm {SO}(2)\) on \(P_{\mathbb {R}^2}^N\) is free for \(N=1\) unless \({\mathbf {r}}_1 ={\mathbf {0}}\). The action is always free for \(N\ge 2\). The action of K is free and proper only under certain conditions, as stated in the following lemmas.
Lemma 1
If \(\sum \Gamma _i{\mathbf {r}}_i\ne 0\), the action \(\rho :K\times P_{\mathbb {R}^2}^N\rightarrow P_{\mathbb {R}^2}^N\) is free.
Proof
Since K is onedimensional, the action of K is nonfree if and only if K has fixed points. Let \(\xi \) be the infinitesimal generator of K, as defined in Sect. 2.2. Then, the action \(\rho \) of K is free if and only if the kernel of \(\xi \) is trivial. It is straightforward to check that the kernel of \(\xi \) is given by:
Hence, under the nonvanishing linear momentum condition the action of K is free at any time. \(\square \)
Since the linear momentum is a first integral of the Eq. (9), it follows from Lemma 1 that the K action is free provided that the linear momentum is nonvanishing for the initial condition.
Lemma 2
If \(\sum \Gamma _i\ne 0\), the action \(\rho :K\times P_{\mathbb {R}^2}^N\rightarrow P_{\mathbb {R}^2}^N\) is proper.
Proof
Let \(\xi \) be the infinitesimal generator of K, as defined in Sect. 2.2. Then, it is straightforward to check that \(\xi \) has rank 2, for any \(N\ge 1\) and its nonzero eigenvalues are purely imaginary and equal to \(\pm i\sum \Gamma _i\ne 0\). Hence, the group K is bounded and closed in the operator norm topology. Therefore, it is compact and so its action is proper. \(\square \)
For \(N\ge 2\), the action of \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\) on \(P_{\mathbb {R}^2}^N\) is free and proper. Indeed, if \(R\in SO(2)\) is not the identity, \(v\in \mathbb {R}^2\) and \((p,q)\in P_{\mathbb {R}^2}^2\), then \(Rp+v=p\) and \(Rq+v=q\) imply \(p=q\), since R does not have real eigenvectors, and so the action is free. Moreover, SO(2) is compact and the map R in Definition 5 for \(\mathbb {R}^2\)action has continuous inverse, therefore, the semidirect product action is a composition of proper maps and so it is proper.
Proof of Theorem 3
We divide the proof in two cases.
Case 1: \(\sum _i\Gamma _i\ne 0\). Without loss of generality due to the translational invariance of equations (9) and to the nonvanishing total circulation, we can assume that the linear momentum is nonzero. Then, Lemmas 12 ensure the \(\mathrm {SO}(2)\times K\) action is free and proper, for any \(N\ge 2\).
For \(N=2\), Theorem 10 tells us that the reduced Hamiltonian system has dimension 0, and therefore the motion is a \(\mathrm {SO}(2)\times K\)relative equilibrium.
For \(N=3\), Theorem 10 with respect to the \(\mathrm {SO}(2)\times K\) action tells us that the reduced Hamiltonian system has dimension 2, which implies that it is integrable.
Case 2: \(\sum _i\Gamma _i = 0\) If the circulation is zero, the linear momentum map due to the action of \(\mathbb {R}^2\) is equivariant. Moreover, the \(\mathbb {R}^2\) action is always free and proper for \(N\ge 1\).
For \(N=2\), Theorem 10 tells us that the reduced Hamiltonian system has dimension 0 when \(N=2\). Therefore the motion can only be up to translations; this explains the vortex dipole solutions.
For \(N=3\), Theorem 10 tells us that the reduced Hamiltonian system has dimension 2, which implies that it is integrable. \(\square \)
Proof of Theorem 4
For \(N=2\), the zerocirculation and zeromomentum conditions imply that \({\mathbf {r}}_1={\mathbf {r}}_2\), so the pointvortex equation 10 is not defined.
For \(N=3\), the \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\) the momentum map is equivariant, so Theorem 10 tells us that the reduced systems evolves on a zerodimensional manifold. Therefore, the motion of the pointvortices is a \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\) relative equilibrium.
For \(N=4\), the \(\mathrm {SO}(2)\ltimes \mathbb {R}^2\) the momentum map is equivariant, so Theorem 10 tells us that the reduced systems evolves on a on a twodimensional manifold. Integrability then follows. \(\square \)
5.3 The Hyperbolic Plane
To get a free action for \(N\ge 2\), we need to restrict to \(\mathrm {PSL}(2)\cong \mathrm {SL}(2)/\lbrace \pm 1\rbrace \). Furthermore, it is known that the action is proper [36].
Proof of Theorem 5
For \(N=2\), Theorem 10 tells us that systems evolves on a zerodimensional manifold, since \(G_\phi \) is a onedimensional Lie group (see [36]). Thus, solutions are relative equilibria with respect to Möbius transformations.
For \(N=3\), Theorem 10 tells us that systems evolves on a twodimensional manifold, so the dynamics is integrable. \(\square \)
Proof of Theorem 6
For \(N=2\), the zero momentum condition (17) implies that \({\mathbf {r}}_1,{\mathbf {r}}_2\) lie on common line through the origin. Hence, \({\mathbf {r}}_1={\mathbf {r}}_2\) and so the equations (13) are not defined.
For \(N=3\), the \(\mathrm {PSL}(2)\) action is free and Theorem 10 tells us that systems evolves on a zerodimensional manifold. Thus, solutions are relative equilibria with respect to Möbius transformations.
For \(N=4\), the \(\mathrm {PSL}(2)\) action is free and Theorem 10 tells us that systems evolves on a twodimensional manifold, so the dynamics is integrable. \(\square \)
5.4 The Flat Torus
The action of \(\mathbb {R}^2\) on \(\mathbb {T}^2\) is not free, but as we have seen in Sect. 2, we understand the pointvortex equations on \(\mathbb {T}^2\) as a special case of the pointvortex dynamics on \(\mathbb {R}^2\). Therefore, the action of \(\mathbb {R}^2\) on itself via translations is free and proper. We can also consider the action of the group K, as defined in Sect. 2.2. Since its momentum map only depends on the conservation of the \(\mathbb {R}^2\) momentum map, K is a symmetry also of any translational invariant Hamiltonian on the torus.
Proof of Theorem 7
For \(N=2\), without loss of generality due to the translational invariance of equations (14) and to the nonvanishing total circulation, we can assume that the linear momentum is nonzero. Then, Lemmas 12 ensure the K action is free and proper. Therefore, Theorem 10 tells us that systems evolves on a twodimensional manifold, so the dynamics is integrable. \(\square \)
Proof of Theorem 8
For \(N=2\), the \(\mathbb {R}^2\) action is free and Theorem 10 tells us that systems evolves on a zerodimensional manifold. This means that the initial condition can only be transformed via translations, which conserve the Hamiltonian.
For \(N=3\), the \(\mathbb {R}^2\) action is free and Theorem 10 tells us that systems evolves on a twodimensional manifold. By definition, any Hamiltonian system on a twodimensional manifold is integrable. \(\square \)
6 Nonintegrability Results
In this section, we briefly summarize the results in literature on nonintegrability of pointvortex dynamics. Unlike the integrability results, which have been addressed extensively, only the planar case has been fully analysed and completed. Some results concern nonintegrability for a restricted model of the pointvortex dynamics, defined as follows:
Definition 7
The restricted problem of N pointvortices consists of a partition of the vortices in two sets A and B, such that the vortices in A have nonzero strength and do not interact with vortices in B, and the vortices in B have vorticity \(0\le \varepsilon \ll 1\) and interact with all the other vortices.
Theorem 11
(\(\mathbb {S}^2\) with nonvanishing momentum, [4]) Consider the pointvortex equations (4) on \(P_{\mathbb {S}^2}^N\), restricted to the part of phase space with nonvanishing total linear momentum

If \(N=4\), the restricted model with A of three identical vortices and B of one single vortex with strength \(\varepsilon =0\) is nonintegrable.
Theorem 12
(\(\mathbb {R}^2\) with nonvanishing circulation or momentum, [22, 46, 47]) Consider Hamilton’s equations (10) on \(P_{\mathbb {R}^2}^N\), either restricted to the case nonvanishing circulation
or restricted to the part of phase space with nonvanishing total momentum

If \(N=4\), there exist configurations of pointvortices for which the motion is nonintegrable.

If \(N\ge 5\), there exist configurations of pointvortices for which the Arnold diffusion occurs.
Regarding the flat torus, Kilin and Artemova [19] gave numerical evidence of nonintegrability for three pointvortices on \(\mathbb {T}^2\) with nonzero total circulation. Our pointvortex gallery in Appendix A also indicate chaotic behaviour in this case. For hyperbolic space, we have also conducted numerical simulations in Appendix A indicating the same behaviour as for the sphere: chaotic behaviour for four pointvortices with nonvanishing momentum. To the best of our knowledge there are yet no rigorous nonintegrability results available for the torus or the hyperbolic space.
7 Outlook: LongTime Predictions for 2D Euler Equations
The prevailing theories for the longtime behaviour of the Euler equations (1) on a twodimensional manifold are those given by Miller [31] and by Robert and Sommeria [40], here referred to as MRS. The MRS approach is a generalization of Onsager’s [38] ideas, from discrete to continuous vorticity fields. These theories state that, in the longtime limit, the vorticity field evolves towards a state where the entropy of a coursegrain probability distribution of macroscopic states is maximized under the constraint of conservation of energy and Casimirs. Consequently, this leads to a coursegrain steady vorticity state, characterized by a functional dependence between vorticity and stream function. For a survey of MRS and the statistical approach to 2D turbulence, see the article by Bouchet and Venaille [8].
However, in a numerical study for 2D Euler equations on the sphere, Dritschel et al. [9] gave results that contradict MRS theory, yielding, for randomly generated initial conditions, a seemingly nonsteady vortex blob configuration. The numerical method used in [9] did not conserve Casimir functions, which raised questions of the reliability since MRS theory is based on conservation of Casimirs. As a remedy, we develop in [34] a Casimir preserving numerical method for Euler equations on the sphere that captures all the features of 2D Euler equations: conservation of Casimirs, energy, and the Lie–Poisson structure. Using this method we obtain, again, strong evidence against the MRS predictions, but now with a more reliable, structure preserving method. Furthermore, we found a new mechanism that connects the longtime behaviour with pointvortex integrability results. Here is a brief outline of how it works:

(1)
As is well known in 2D turbulence, the inverse energy cascade, discovered by Kraichnan [23], forces smaller vortex formations of the same sign to merge into larger ones by vortex stretching, forming positive and negative vortex blobs.

(2)
As long as the vortex blobs are not too close to each other, so they are not ripped apart and merged, their dynamics are well described by pointvortex dynamics. Theoretical results in this direction are given, for example, in the monograph by Marchioro and Pulvirenti [25].

(3)
The vortex merging continue, into fewer and larger vortex blobs, until the blob dynamics become integrable, with well separated vortex blob trajectories. Because of quasiperiodicity, the vortex blobs are then ‘stuck’ in this part of phase space and no further merging occurs.

(4)
A prediction for the final number of vortex blobs N is thus given by integrability results for pointvortex dynamics: for the given fluid configuration (in terms of circulation, energy, momentum, etc.), find the largest N such that the dynamics is integrable for N vortices, but nonintegrable for \(N+1\) vortices.
For the Euler equations on the sphere, Casimir preserving numerical simulations with randomly generated initial conditions, presented in [34], perfectly align with this mechanism: if the total linear momentum is zero (or very small) we see the formation of four vortex blobs interacting in a quasiperiodic, nonsteady fashion reflecting the results in Theorem 2 above. If the linear momentum is nonzero, we see the formation of three vortex blobs, reflecting Theorem 1. We anticipate that the longtime behaviour on other domains also shall be reflected in the corresponding pointvortex integrability results, at least for the plane and the hyperbolic plane. (As we can see in the proofs above, the flat torus is, from the symplectic reduction viewpoint, more complicated than the other cases; the connection to the corresponding Euler equations is not direct.)
But of course, numerical simulations alone, even if they preserve all the underlying structures, are not enough and must be accompanied with rigorous mathematical analysis, attempting to prove the connection between integrability and the longtime behaviour. We consider the paper at hand the first step in this direction.
References
Aref, H.: Point vortex dynamics: a classical mathematics playground. J. Math. Phys. 48(065401), 23 (2007)
Arnold, V.I.: Sur la géométrie différentielle des groupes de Lie de dimension infinie et ses applications à l’hydrodynamique des fluides parfaits. Ann. Inst. Fourier (Grenoble) 16, 319–361 (1966)
Arnold, V.I., Khesin, B.: Topological Methods in Hydrodynamics. Springer, New York (1998)
Bagrets, A., Bagrets, D.: Nonintegrability of two problems in vortex dynamics. Chaos 7, 368–375 (1997)
Bolsinov, A.V., Borisov, A.V., Mamaev, I.S.: Lie algebras in vortex dynamics and celestial mechanics  IV. Regul. Chaotic Dyn. 4, 23–50 (1999)
Bolsinov, A.V., Jovanović, B.: Noncommutative integrability, moment map and geodesic flows. Ann. Glob. Anal. Geom 23, 305–322 (2003)
Borisov, A.V., Kilin, A.A., Mamaev, I.S.: Reduction and chaotic behavior of point vortices on a plane and a sphere. Nelineinaya Dinamika 233–246 (2005)
Bouchet, F., Venaille, A.: Statistical mechanics of twodimensional and geophysical flows. Phys. Rep. 515, 227–295 (2012)
Dritschel, D.G., Qi, W., Marston, J.B.: On the latetime behavior of a bounded, inviscid twodimensional flow. J. Fluid Mech. 783, 1–22 (2015)
Eckhardt, B.: Integrable four vortex motion. Phys. Fluid 31, 2796–2801 (1988)
Euler, L.: Principes généraux de l’état d’équilibre d’un fluide. Académie Royale des Sciences et des BellesLettres de Berlin, MémoiresAcadémie Royale des Sciences et des BellesLettres de Berlin, Mémoires 11, 217–273 (1757)
Gröbli, W.: Specielle Probleme über die Bewegung geradliniger paralleler Wirbelfäden. Druck von Zürcher und Furrer, Zürich (1877)
Hairer, E., Lubich, C., Wanner, G.: Geometric Numerical Integration. Springer Series in Computational Mathematics. Springer, Berlin (2006)
Helmholtz, H.: Über integrale der hydrodynamischen gleichungen, welcher der wirbelbewegungen entsprechen. J. Reine Angew. Math. 55, 25–55 (1858)
Hwang, S., Kim, S.C.: Point vortices on hyperbolic sphere. J. Geom. Phys. 59, 475–488 (2009)
Hwang, S., Kim, S.C.: Relative equilibria of point vortices on the hyperbolic sphere. J. Math. Phys. 54, 063101 (2013)
Khesin, B., Misiołek, G.: Euler and Navier–Stokes equations on the hyperbolic plane. Proc. Natl. Acad. Sci. (USA) 109, 18324–18326 (2012)
Khesin, B., Wendt, R.: The Geometry of InfiniteDimensional Groups. Springer, Berlin (2009)
Kilin, A. A., Artemova, E. M.: Integrability and chaos in vortex lattice dynamics. Regul. Chaotic Dyn. 101–113 (2019)
Kimura, Y.: Vortex motion on surfaces with constant curvature. Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 455, 245–259 (1999)
Kirchhoff, G.: Vorlesungen über mathematische physik: mechanik. BG Teubner, Leipzig (1876)
Koiller, J., Carvalho, S.: Nonintegrability of the 4vortex system: analytical proof. Commun. Math. Phys. 120, 643–652 (1989)
Kraichnan, R.H.: Inertial ranges in twodimensional turbulence. Phys. Fluid 10, 1417–1423 (1967)
Mamode, M.: Fundamental solution of the Laplacian on flat tori and boundary value problems for the planar Poisson equation in rectangles. Bound. Value Probl. 221 (2014)
Marchioro, C., Pulvirenti, M.: Mathematical Theory of Incompressible Nonviscous Fluids. Springer, Berlin (2012)
Marsden, J.E., Misiołek, G., Ortega, J.P., Perlmutter, M., Ratiu, T.S.: Hamiltonian Reduction by Stages. Springer, Berlin (2007)
Marsden, J.E., Ratiu, T.S.: Introduction to Mechanics and Symmetry. Springer, Berlin (1998)
Marsden, J.E., Weinstein, A.: Coadjoint orbits, vortices, and Clebsch variables for incompressible fluids. Phys. D 7, 305–323 (1983)
McLachlan, R.I., Modin, K., Verdier, O.: Symplectic integrators for spin systems. Phys. Rev. E 89, 061301 (2014)
McLachlan, R.I., Modin, K., Verdier, O.: A minimalvariable symplectic integrator on spheres. Math. Comp. 86, 2325–2344 (2016)
Miller, J.: Statistical mechanics of Euler equations in two dimensions. Phys. Rev. Lett. 65, 2137–2140 (1990)
Modin, K.: Geometric hydrodynamics: from Euler, to Poincaré, to Arnold, 13th Young Researchers Workshop on Geometry, Mechanics and Control: Three Minicourses, vol. 48 of Textos de Matematica, pp. 69–92, Departamento de Matematica da Universidade de Coimbra, Portugal (2019)
Modin, K., Verdier, O.: Integrability of nonholonomically coupled oscillators. Discr. Contin. Dyn. Syst. 34, 1121–1130 (2014)
Modin, K., Viviani, M.: A Casimir preserving scheme for longtime simulation of spherical ideal hydrodynamics. J. Fluid Mech. 884, A22 (2020a)
Modin, K., Viviani, M.: LiePoisson numerical schemes for isospectral flows. Found. Comput. Math. 20, 889–921 (2020b)
Montaldi, J., NavaGaxiola, C.: Point vortices on the hyperbolic plane. J. Math. Phys. 55, 102702 (2014)
Newton, P.K.: The NVortex Problem: Analytical Techniques. Springer, Berlin (2013)
Onsager, L.: Statistical hydrodynamics. Il Nuovo Cimento (1943–1954) 6(6), 279–287 (1949)
Poincaré, H.: Théorie des Tourbillons. Carré, Paris (1893)
Robert, R., Sommeria, J.: Statistical equilibrium states for twodimensional flows. J. Fluid Mech. 229, 291–310 (1991)
Sakajo, T.: The motion of three point vortices on a sphere. Jpn. J. Ind. Appl. Math. 16, 321–347 (1999)
Sakajo, T.: Integrable fourvortex motion on sphere with zero moment of vorticity. Phys. Fluid 19 (2007)
Stremler, M.A., Aref, H.: Motion of three point vortices in a periodic parallelogram. J. Fluid Mech. 392, 101–128 (1999)
Viviani, M.: A minimalvariable symplectic method for isospectral flows. BIT Numer. Math. 60, 741–758 (2020)
Weiss, J.B., McWilliams, J.C.: Nonergodicity of point vortices. Phys. Fluid. A 3, 835–844 (1991)
Ziglin, S.L.: Nonintegrability of the problem of the motion of four point vortices. Dokl. Akad. Nauk SSSR 250, 1296–1300 (1980)
Ziglin, S.L.: Quasiperiodic motions of vortex systems. Phys. D 4, 261–269 (1982). (appendix of the article)
Acknowledgements
The authors acknowledge support from the European Union Horizon 2020 research and innovation programmes under the Marie SkodowskaCurie grant agreement No. 691070, by the Swedish Foundation for International Cooperation in Research and Higher Eduction (STINT) grant No. PT20145823, by the Swedish Research Council (VR) grant No. 201705040, and by the Knut and Alice Wallenberg Foundation (KAW) grant No. WAF2019.0201.
Funding
Open access funding provided by Chalmers University of Technology.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A: Gallery of pointvortex solutions
Appendix A: Gallery of pointvortex solutions
In this appendix we showcase typical pointvortex phase space trajectories for all of the investigated geometries: the sphere, the plane, the hyperbolic plane, and the flat torus. The results verify the integrability results in the theorems above: if conditions for integrability (or relative equilibria) are fulfilled, then beautiful, quasiperiodic patterns emerge. Interestingly, our gallery of pointvortex trajectories also indicate that the symplectic reduction approach to integrability is sharp in the following sense: in all tested cases where symplectic reduction could not be used to prove integrability, the resulting trajectories appear to be chaotic. This leads us to a conjecture about nonintegrability. Before that, we explain how to compute the trajectories and we give some specific comments on the results.
Initial conditions are generated randomly; for the compact domains they are drawn from the uniform distribution (with respect to the underlying Riemannian volume form). For the plane, we use the multivariate Gaussian distribution with zero mean and unit variance. The vortex strengths are randomly generated from independent normal distributions. For initial conditions with constraints, such as vanishing momentum or circulation, we use orthogonal projection.
To compute the trajectories we use numerical integration methods. For the sphere we use the spherical midpoint method [29, 30]. For the flat geometries, we use the implicit midpoint method (cf. [13]). For the hyperbolic plane, we use the hyperbolic midpoint method [44] (see also [35]). All these methods are second order approximations of the exact flow map.
Of course, the numerical discretizations introduce errors, so the computed trajectories are not true pointvortex trajectories. One can argue that they therefore might not display the correct qualitative behaviour (such as quasiperiodic versus chaotic behaviour). For general numerical methods this is a sound consideration. Here, however, the numerical methods we use are specifically chosen to uphold the qualitative feature. Indeed, all the methods we use are symplectic (with respect to the underlying symplectic structure) and furthermore equivariant (with respect to the underlying invariance group). Since the methods are symplectic, one can utilize the framework of backward error analysis (cf. [13]) to prove that the discretized flow corresponds, for exponentially longtime intervals, to the exact flow of a slightly modified Hamiltonian system. Furthermore, since the methods are equivariant, the modified Hamiltonian function is going to be invariant. Thus, since the integrability Theorems 1–8 above are valid not just for the Hamiltonian corresponding to pointvortex dynamics, but for any invariant Hamiltonian, we conclude that the theorems are valid also for the modified Hamiltonian systems yielding the computed trajectories, independent of the magnitude of the numerical errors.
For each of the settings in the theorems (vanishing momentum, nonvanishing momentum, etc.), we generated several trajectories for each geometry. To check the sharpness of our integrability results, we also computed solutions for the simplest possible setting for which the appropriate theorem do not apply. The typical behaviour is illustrated in Figs. 1, 2, 3 and 4; this is our gallery of point vortex solutions.
For the sphere, in Fig. 1, we obtain quasiperiodic or relative equilibria solution in accordance with Theorems 1–2. For 4 pointvortices with nonvanishing momentum the behaviour is chaotic; a typical example is shown in Fig. 1d. Also for the plane, in Fig. 2, we see quasiperiodic solutions whenever Theorems 4–3 apply. Recall that integrability for the planar 4 pointvortex solutions depend on two conditions: both momentum and circulation have to vanish. If either of these conditions are not fulfilled, the trajectories become chaotic; typical examples are shown in Fig. 2c, d. Interestingly, from the pointofview of symplectic reduction, the nonvanishing of momentum and circulation corresponds to two different violations of the conditions needed: nonvanishing momentum implies that the symmetry group is too small to obtain integrability, whereas nonvanishing circulation implies that the momentum map is not equivariant (which is a condition in Theorem 10). Similarly, in Fig. 3 for the hyperbolic plane, and in Fig. 4 for the flat torus, we observe the behaviour predicted by the corresponding integrability theorem, and whenever the conditions are not fulfilled we observe chaotic trajectories. This leads us to formulate the following:
Conjecture 1
Let M be a twodimensional Riemannian manifold with symmetry group G. Then a generic Ginvariant Hamiltonian system on \(P^N_M\) is integrable only when the geometric conditions (free and proper action, and equivariant momentum map) are such that Theorem 10 directly implies integrability.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Modin, K., Viviani, M. Integrability of PointVortex Dynamics via Symplectic Reduction: A Survey. Arnold Math J. 7, 357–385 (2021). https://doi.org/10.1007/s40598020001628
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40598020001628
Keywords
 Pointvortex dynamics
 Integrable systems
 Euler equations
 Symplectic reduction
Mathematics Subject Classification
 37J15
 53D20
 70H06
 35Q31
 76B47