Research Contribution DOI:10.56994/ARMJ.011.004.007
Received: 27 Dec 2024; Accepted: 24 Aug 2025


On symplectic linearizable actions

Eva Miranda Laboratory of Geometry and Dynamical Systems
and
SYMCREA research unit,
Department of Mathematics,
Universitat Politècnica de Catalunya,
Barcelona, Spain
and
Centre de Recerca Matemática, CRM
email: eva.miranda@upc.edu×
Abstact.
We prove that linearizable actions are also symplectically linearizable (either smoothly or analytically) in a neighborhood of a fixed point. Specifically, the fundamental vector fields associated with the action can be simultaneously linearized in Darboux coordinates. This result extends equivariant symplectic local normal forms to non-compact group actions.

In both formal and analytic frameworks, the existence of linearizing coordinates is tied to a cohomological equation, which admits a solution for semisimple actions [9, 8]. Consequently, an analytic symplectic action of a semisimple Lie algebra can be locally linearized in Darboux coordinates, enabling the simultaneous analytic linearization of Hamiltonian vector fields near a shared zero. However, in the smooth setting, this result is restricted to semisimple Lie algebras of compact type. We construct an explicit example of a smooth, non-linearizable Hamiltonian action with a semisimple linear part, thereby answering in the negative a question posed by Eliasson [5].

1 Introduction

A classical result due to Bochner [1] establishes that a compact Lie group action on a smooth manifold is locally equivalent, in the neighbourhood of a fixed point, to its linearization. This result holds in the Ck category. It is worth exploring if similar results hold in the non-compact case.

As observed in [8], if the Lie group is connected, the linearization problem can be formulated in the following terms: find a linear system of coordinates for the vector fields corresponding to the one-parameter subgroups of G; or more generally, consider the representation of a Lie algebra and find coordinates on the manifold that simultaneously linearize the vector fields in the image of the representation vanishing at a point. This is the perspective we adopt in this note when referring to linearization.

In the formal and analytic cases, the existence of coordinates that linearize the action is related to a cohomological equation that can always be solved when the Lie group under consideration is semisimple [9], [8]. Guillemin and Sternberg also studied the problem in the C setting. At the end of [8], they presented the celebrated example of a non-linearizable action of 𝔰𝔩(2,) on 3, constructed via a perturbation involving the radial vector field with flat coefficients. This example has been foundational in the literature, inspiring the construction of other examples with profound geometric implications, such as Weinstein’s non-stable Poisson structure example [19].

When the semisimple Lie algebras are of compact type, the linearization of the action can be achieved by combining the local integration of the Lie group action with Bochner’s theorem, leading to the linearization of the associated Lie algebra action [6].

Linearization techniques also play a significant role in Hamiltonian systems. When a Hamiltonian system arises from a symplectic action of a compact Lie group fixing a point, the equivariant version of Darboux’s theorem ([18], [3]) ensures that the group action can be linearized in Darboux coordinates near the fixed point. It is worth exploring if similar results apply beyond the compact case.

For complete integrable systems, an associated abelian symplectic action emerges. When the integrable system in local coordinates has a “linear part” linked to a Cartan subalgebra, this leads to non-degenerate singularities [4]. As shown in [5], [4], [10], [11] and [12], complete integrable systems near non-degenerate singular points are equivalent to their linear models. Consequently, the Hamiltonian system itself is equivalent to the linear one. This result provides normal forms for integrable systems near singular non-degenerate points and, specifically, ensures the simultaneous linearization of Hamiltonian vector fields near a common zero.

The next challenge involves Hamiltonian systems with a semisimple linear part, as proposed by Eliasson in [5]. In the formal or analytic setting, results by Guillemin and Sternberg [8] and Kushnirenko [9] demonstrate that such systems are equivalent to the linear model when the symplectic form is disregarded. In this note, we establish that not only can the Hamiltonian vector fields be linearized, but they can also be linearized in Darboux coordinates.

Following Guillemin and Sternberg’s approach, we prove that if a symplectic Lie algebra action of semisimple type fixes a point, there exist analytic Darboux coordinates in which the analytic vector fields generating the Lie algebra action are linear. This result also extends to complex analytic Lie algebra actions on complex analytic manifolds. Additionally, we construct an example of a Hamiltonian system with a semisimple linear part that is not C-linearizable.

Organization of this article: In Section 2, we prove that linearizable actions on symplectic manifolds can be locally linearized in Darboux coordinates. In Section 3, we apply this to show that any real analytic symplectic action of a semisimple Lie algebra can be linearized in real analytic Darboux coordinates in a neighborhood of a fixed point. Furthermore, this result extends to analytic complex manifolds and complex analytic actions of semisimple Lie algebras. In Section 4, we present a counterexample proving that the linearization result does not hold in general for smooth Hamiltonian actions of semisimple Lie algebras.

Acknowledgements: I would like to thank Häkan Eliasson, Ghani Zeghib, David Martínez-Torres and Marco Castrillón for their valuable feedback on an earlier version of this note.

2 Linearizable actions in Darboux coordinates

Let 𝔤 be a Lie algebra and let ρ:𝔤Lanalytic stand for a representation of 𝔤 in the algebra of real (or complex) analytic vector fields on a real (or complex) analytic manifold M.

We say that pM is a fixed point for ρ if the vector fields in ρ(𝔤) vanish at p. We say that ρ can be linearized in a neighborhood of a fixed point if there exist local coordinates in a neighbourhood of p such that the vector fields in the image of ρ can be simultaneously linearized (i.e, ρ is equivalent to a linear representation).

Assume that the Lie algebra action is (analytically/smoothly) linearizable and assume that M is endowed with a symplectic structure (smooth, analytic). We first prove that it is then symplectically linearizable.

Theorem 2.1.

Let 𝔤 be a Lie algebra and let (M,ω) be a (real or complex) analytic symplectic manifold. Let ρ be a representation by analytic symplectic vector fields. Let p be a fixed point for ρ and assume that ρ can be linearized. Then there exist local analytic coordinates (x1,y1,,xn,yn) in a neighborhood of p for ρ such that ρ is a linear representation and ω can be written as,

ω=i=1ndxidyi.
Proof.

Let ρ be an analytic symplectic action of a Lie algebra on a manifold M, with a fixed point pM. Choose analytic coordinates (x1,y1,,xn,yn) centered at p in which the action ρ is linear. Let ω1 denote the symplectic form in these coordinates. Although ρ is now linear, ω1 need not be of Darboux type.

We denote by ω0 the constant (degree-zero) term in the Taylor expansion of ω1 at the origin. Since ω1 is preserved by ρ and ρ is linear, it follows that ω0 is preserved by the linearized action ρ(1)=ρ. In particular, ω0 is a constant symplectic form invariant under the action. Our goal is to construct a local analytic diffeomorphism ϕ, fixing the origin, such that ϕ(ω1)=ω0 and ϕ commutes with ρ. That is, we seek an equivariant analytic Darboux theorem for ω1, linearizing the form while preserving the linear action ρ.

To this end, we apply the path method [16] for analytic symplectic structures. By the Poincaré lemma, there exists an analytic 1-form α such that

ω1=ω0+dα.

Define a path of symplectic forms:

ωt=tω1+(1t)ω0,t[0,1].

Each ωt is an analytic symplectic form in a neighbourhood of the origin. Moreover, the action ρ preserves both ω0 and ω1, hence it preserves the entire path ωt.

We now define the time-dependent analytic vector field Xt by Moser’s equation:

iXtωt=α. (2.1)

In order to ensure that Xt is invariant under ρ, it suffices to construct α invariant under ρ. For this purpose, we apply the standard homotopy operator used in the proof of the Poincaré lemma, adapted to our equivariant setting.

Let R=xixi+yiyi be the radial vector field, and ht the homothety xtx. Then, we define

α:=011tht(iRβ)𝑑t,where β=ω1ω0.

Because β is ρ-invariant and ρ commutes with R, it follows that α is also ρ-invariant. Thus, the vector field Xt is invariant under ρ.

Let ϕt denote the flow of Xt, satisfying the differential equation

ϕtt(q)=Xt(ϕt(q)),ϕ0=id. (2.2)

Since Xt is ρ-invariant, the flow ϕt commutes with the action ρ. Moreover, because α vanishes at the origin, so does Xt, ensuring that each ϕt fixes the origin.

By construction, ϕt(ωt)=ω0, and in particular ϕ1(ω1)=ω0. The diffeomorphism ϕ:=ϕ1 is then the desired equivariant analytic transformation taking ω1 to ω0 while preserving the linear action ρ.

This completes the proof. ∎

Remark 2.2.

The theorem above is stated in the analytic category; however, if the linearization is assumed to hold in the smooth category, the symplectic diffeomorphism obtained from the proof is also smooth.

3 The case of analytic semisimple Lie algebra actions

Guillemin and Sternberg provided in [8] a complete characterization of analytically linearizable actions. They demonstrated that a necessary and sufficient condition for the representation

ρ:gLanalytic

to be locally analytically linearizable is the existence of an analytic vector field X, defined in a neighborhood of p, vanishing at p, with the identity matrix as its Jacobian at p, and commuting with all the vector fields in g~.

This condition was elegantly recast in cohomological terms in [8]. They proved that the first cohomology group H1(g,V) acts as an obstruction to analytic linearization. For semisimple g, H1(g,V) vanishes for all representation spaces V, ensuring the possibility of analytic linearization. On the other hand, for non-semisimple g, one can construct a representation space V such that H1(g,V)0, which precludes analytic linearization.

This result establishes the semisimple case as a natural candidate for analytic linearization.

Guillemin and Sternberg [8] and Kushnirenko [9] proved the following.

Theorem 3.1 (Guillemin-Sternberg, Kushnirenko).

The representation ρ:𝔤Lanalytic with 𝔤 semisimple is locally equivalent, via an analytic diffeomorphism, to a linear representation of 𝔤 in a neighbourhood of a fixed point for ρ.

As an application of theorem 2.1: When the representation is done by Hamiltonian vector fields (locally symplectic), the analytic diffeomorphism that gives the equivalence of the initial representation to the linear representation can be chosen to take the initial symplectic form to the Darboux one. Namely,

Corollary 3.2.

Let 𝔤 be a semisimple Lie algebra and let (M,ω) be a (real or complex) analytic symplectic manifold. Let ρ:𝔤Lanalytic be a representation by analytic symplectic vector fields. Then there exist local analytic coordinates (x1,y1,,xn,yn) in a neighbourhood of a fixed point p for ρ such that ρ is a linear representation and ω can be written as,

ω=i=1ndxidyi.

4 Non-linearizable semisimple smooth actions

4.1 The counterexample of Cairns and Ghys

In this section we recall the results of Cairns and Ghys concerning a C-action of SL(2,) on 3 which is not linearizable. All results mentioned in this section are contained in section 8 of [2].

Consider the basis {X,Y,Z} of 𝔰𝔩(2,) satisfying the relations:

[X,Y]=Z,[Z,X]=Y,[Z,Y]=X

Now consider the representation on 3 defined on this basis as:

ρ(X)=yz+zyρ(Y)=xz+zxρ(Z)=xyyx (4.1)

The orbits of this action are the level sets of the quadratic form Q=x2+y2z2=r2z2 (where r2=x2+y2). These level sets are non-degenerate quadrics: one-sheeted hyperboloids for Q>0, two-sheeted hyperboloids for Q<0, and a quadratic cone for Q=0.

Introduce the radial vector field

R=xx+yy+zz,

and perturb ρ by setting

X~ =ρ(X)+fR, (4.2)
Y~ =ρ(Y)+gR,
Z~ =ρ(Z),

where

f(x,y,z)=xA(z,x2+y2),g(x,y,z)=yA(z,x2+y2),

and

A(z,r)=a(r2z2)r2,

with a: any C–function which vanishes for r2z20 and is bounded.

By [2], the fields X~,Y~,Z~ still close under the Lie bracket to an 𝔰𝔩(2,)–algebra and are complete. Hence they integrate to an action ρ^ of the universal cover of SL(2,), which descends to SL(2,) itself since Z~=ρ(Z) is unchanged. Moreover:

  • On the “hyperbolic region” {x2+y2>z2}, one has a(r2z2)0 and {X~,Y~,Z~} are linearly independent, so ρ^–orbits are 3‐dimensional.

  • On and inside the “cone” {x2+y2z2}, one has a(r2z2)=0 so X~=ρ(X), Y~=ρ(Y), and ρ^ coincides with the linear action.

Since the original linear action never has 3‐dimensional orbits, ρ^ cannot be conjugate to it. Therefore, the deformed action is not linearizable.

4.2 The counterexample of Guillemin and Sternberg

The construction of Guillemin and Sternberg [8] follows the guidelines outlined below. It is quite similar to the counterexample of Grant and Cairns; however, the key difference is that the vector field the perturbation does not preserve Z, so it cannot be guaranteed that it lifts to SL(2,).

If we perturb the initial action of 𝔰𝔩(2,) to the non-linear action:

ρ^(X) =ρ(X)+xzr2g(r2z2)R,
ρ^(Y) =ρ(Y)yzr2g(r2z2)R,
ρ^(Z) =ρ(Z)+g(r2z2)R,

where R=xx+yy+zz is the radial vector field, and gC() is such that g(x)>0 if x>0, and g(x)=0 if x0.

Inside the cone r2z2, the two sets of vector fields are identical. However, if we choose g(u)=e1/u2,u>0, and g(u)=0,u0, for example, then outside the cone r2z2=0, the vector field ρ(X)=xyyx has closed circular orbits, while the corresponding deformed vector field ρ^(X) has orbits that spiral outward.

Therefore, it is impossible to find a C-mapping defined in a neighborhood of the origin. Hence, ρ is not linearizable.

4.3 A Hamiltonian counterexample

We construct a counterexample to C–linearization under the hypothesis that the action is Hamiltonian, thereby giving a negative answer to a question of Eliasson [5]. We keep the notation of the Cairns–Ghys construction from Section 4.1.

Proposition 4.1.

Let α be the SL(2,)–action on 3 generated by

X¯=ρ(X)+fR,Y¯=ρ(Y)+gR,Z¯=ρ(Z),

where R=xx+yy+zz and

f=xA(z,x2+y2),g=yA(z,x2+y2),A(z,r)=a(r2z2)r2,

with a: smooth, bounded, and vanishing on . Let α^ be the cotangent lift of α to T(3). Then α^ is Hamiltonian and not C–linearizable in a neighbourhood of the origin.

Proof.

For a diffeomorphism g:MM, the cotangent lift is

g^(q,p)=(g(q),(dgq1)p),

so

πg^=gπandg^(q,0)=(g(q),0). (4.3)

Thus fibres map to fibres and the zero section is preserved by every cotangent lift. In particular, with

F:=π1(0)={x=y=z=0},Z:={a=b=c=0},O:=(0,0,0;0,0,0),

both F and Z are invariant for the lifted linear action and for the lifted Cairns–Ghys action (note that 0 is fixed in the base, since ρ is linear and fR,gR vanish at 0).

Let α(1) denote the linear part of α and α^(1) its cotangent lift. The latter is Hamiltonian for ω=dθ (θ=adx+bdy+cdz), with moment map

μ=(zb+cy,az+cx,ay+bx)𝔰𝔩(2,),

whose Jacobian is

Dμ=(0cb0zyc0az0xba0yx0).

For Hamiltonian actions one has dim(Gm)=rankdμm (see, for instance, [7, §26]). A direct computation shows that

rankDμ={0,p=O,2,(x,y,z)×(a,b,c)=0,pO,3,otherwise.

Equivalently, the rank drops to 2 precisely when (x,y,z) is collinear with (a,b,c), a locus containing F={x=y=z=0} and Z={a=b=c=0} but strictly larger. Consequently,

dim𝒪α^(1)(p)={0,p=O,2,(x,y,z)×(a,b,c)=0,pO,3,otherwise, (4.4)

and, in particular, dim𝒪α^(1)(q,0)=2 for all q0.

For any lifted action, the fundamental vector fields satisfy

ξ^X(q,p)=(ξX(q),pdξX(q)). (4.5)

Hence along Z one has ξ^X(q,0)=(ξX(q),0), so every orbit starting in Z stays in Z. Conversely, no orbit through a point (q,p)Z can be contained in Z, since it already contains (q,p)Z. Thus, near O, the orbits contained in Z are exactly those starting in Z. Moreover, from (4.5) we read off

dim𝒪α^(q,0)=dim𝒪α(q).

By [2, §8] there exist points q0 in the base with 3–dimensional α–orbits. For such q, put p=(q,0)Z{O}. Then

dim𝒪α^(q,0)=dim𝒪α(q)=3. (4.6)

Assume, for contradiction, that there exists a germ Φ:(TM,O)(TM,O) with Φα^(1)=α^Φ. Conjugacy carries orbits diffeomorphically to orbits, preserving their dimension. Since, near O, the orbits contained in Z are precisely those starting in Z for both lifted actions, necessarily Φ(Z)=Z and Φ(F)=F. Hence Φ1(p)Z{O}, so by (4.4)

dim𝒪α^(1)(Φ1(p))=2,

whereas by (4.6) dim𝒪α^(p)=3, a contradiction. Therefore, α^ is not C–linearizable near the zero section. ∎

Remark 4.2.

We can employ the same strategy, adopting the counterexample by Guillemin and Sternberg in the process.

Consider the Lie algebra action of 𝔰𝔩(2,), denoted by ρ¯, on 3, generated by the vector fields:

ρ^(X) =ρ(X)+xzr2g(r2z2)R,
ρ^(Y) =ρ(Y)yzr2g(r2z2)R,
ρ^(Z) =ρ(Z)+g(r2z2)R,

where R=xx+yy+zz is the radial vector field, and gC() satisfies g(u)=e1/u2 for u>0 and g(u)=0 for u0.

Using similar guidelines to those of Guillemin and Sternberg in [8], we can verify that the lifted action to T(3) is not C-linearizable.

The lift of the action can be computed using the Liouville one-form. Let θ=adx+bdy+cdz. Then, the lift of the non-perturbed vector field is a Hamiltonian vector field with the Hamiltonian function

f=ay+bx,

and the lifted vector field of the perturbed system is the Hamiltonian vector field with respect to the function

f=ay+bx+g(r2z2)(ax+by).

The Hamiltonian vector field of f is given by:

xyyx+abba,

and it exhibits periodic orbits. In contrast, the corresponding deformed vector field, the Hamiltonian vector field of f, has orbits that spiral outward.

4.4 The case of semisimple Lie algebras of compact type

When the Lie algebra action is of compact type, it can be integrated into an action of a compact Lie group G (see [6] for a proof, which is based on the use of algebroids).

Given a fixed point for the action p, we can associate a linear action of the group in a neighbourhood of p, with the group action preserving the symplectic structure (which we can assume to be in Darboux coordinates). Applying the equivariant Darboux theorem [3], we obtain a diffeomorphism ϕ that linearizes the group action G in Darboux coordinates. By differentiation, this provides the linearization of the Lie algebra action ρ.

Acknowledgments

Eva Miranda is supported by the Catalan Institution for Research and Advanced Studies via an ICREA Academia Prize 2021 and by the Alexander Von Humboldt Foundation via a Friedrich Wilhelm Bessel Research Award. The author is also supported by the Spanish State Research Agency, through the Severo Ochoa and María de Maeztu Program for Centers and Units of Excellence in R&D (project CEX2020-001084-M) and the project PID2023-146936NB-I00 by the Spanish State Research Agency MCIU/AEI / 10.13039/501100011033/ FEDER, UE. She was also supported by the AGAUR project 2021 SGR 00603 Geometry of Manifolds and Applications, GEOMVAP.

References

  • [1] S.  Bochner, Compact groups of differentiable transformations. Ann. of Math. (2) 46, (1945). 372–381.
  • [2] G.  Cairns and E.  Ghys, The local linearization problem for smooth SL(n)-actions, Enseign. Math. (2) 43 (1997), no. 1-2, 133-171.
  • [3] M.  Chaperon, Quelques outils de la théorie des actions différentiables Third Schnepfenried geometry conference, Vol. 1 (Schnepfenried, 1982), 259–275, Astérisque, 107-108, Soc. Math. France, Paris, 1983.
  • [4] L. H.  Eliasson, Normal forms for Hamiltonian systems with Poisson commuting integrals, Ph.D. Thesis (1984).
  • [5] L.H.  Eliasson, Normal forms for Hamiltonian systems with Poisson commuting integrals—elliptic case, Comment. Math. Helv. 65 (1990), no. 1, 4–35.
  • [6] R.  L. Fernandes and Ph.  Monnier, Linearization of Poisson brackets, Lett. Math. Phys. 69, (2004) 89-114.
  • [7] V. Guillemin and S. Sternberg, Symplectic Techniques in Physics, Cambridge University Press, Cambridge, 1990.
  • [8] V.  Guillemin and S.  Sternberg, Remarks on a paper of Hermann, Trans. Amer. Math. Soc. 130 (1968) 110–116.
  • [9] A.G.  Kushnirenko,Linear-equivalent action of a semisimple Lie group in the neighbourhood of a stationary point, Functional Anal. Appll, 1, (1967) 89-90.
  • [10] E. Miranda, On symplectic linearization of singular Lagrangian foliations, Ph.D. thesis, Universitat de Barcelona, ISBN: 9788469412374, 2003.
  • [11] E. Miranda, Integrable systems and group actions. Cent. Eur. J. Math. 12 (2014), no. 2, 240–270.
  • [12] E. Miranda and N.T. Zung, Equivariant normal forms for nondegenerate singular orbits of integrable Hamiltonian systems, Ann. Sci. Ecole Norm. Sup., 37, no. 6, 2004, pp. 819–839.
  • [13] E.  Miranda and N. T.  Zung, A note on equivariant normal forms of Poisson structures, Math. Research Notes, 2006, vol 13-6, 1001–1012.
  • [14] E.  Miranda, Some rigidity results for Symplectic and Poisson group actions, XV International Workshop on Geometry and Physics, 177–183, Publ. R. Soc. Mat. Esp., 11, R. Soc. Mat. Esp., Madrid, 2007.
  • [15] E.  Miranda, P.  Monnier and N.T.  Zung, Rigidity of Hamiltonian actions on Poisson manifolds. Adv. Math. 229, No. 2, 1136-1179 (2012).
  • [16] J.  Moser, On the volume elements on a manifold, Trans. Am. Math. Soc. 120, 286-294 (1965).
  • [17] R.  Palais, Equivalence of nearby differentiable actions of a compact group. Bull. Amer. Math. Soc. 67 1961 362–364.
  • [18] A.  Weinstein, Lectures on symplectic manifolds. Regional Conference Series in Mathematics, No.29. American Mathematical Society, Providence, R.I., 1977.
  • [19] A.  Weinstein, The local structure of Poisson manifolds., J. Differential Geom. 18 (1983), no. 3, 523–557.