We define a Hofer-type norm for the Hamiltonian map on regular Poisson
manifold and prove that it is nondegenerate. We show that the L1,∞-norm and the L∞-norm
coincide for the Hamiltonian map on closed regular Poisson manifold and give some sufficient
conditions for a Hamiltonian path to be a geodesic. The norm between the Hamiltonian map
and the induced Hamiltonian map on the quotient of Poisson manifold (M,{·,·}) by a compact Lie group Hamiltonian action is also compared.
1. Introduction and Main Results
This paper is devoted to establishing an invariant norm for Hamiltonian maps on the Poisson manifold. When M is symplectic, a remarkable bi-invariant distance was defined on Ham(M). This bi-invariant distance was first discovered by Hofer on the group of compactly supported symplectic diffeomorphisms of (ℝ2n,ω0) (where ω0 is the standard symplectic form) [1]. Viterbo defined a bi-invariant metric by generating functions [2], Polterovich generalized Hofer’s metric to more symplectic manifold [3], and finally Lalonde and McDuff extended it to the group of compactly supported Hamiltonian diffeomorphisms on any symplectic manifold [4]. This norm plays an important role in studying symplectic topology and has close relationship with symplectic capacity and symplectic rigidity; many mathematicians have great work in this field, but there is few work on the Poisson case; this is because the lack of variational formulation in the Poisson case, it is not easy to prove the nondegenerate. In this paper, we define a Hofer-type norm on a class of Poisson manifolds, that is, regular Poisson manifolds; with the help of Casimir functions and the decomposition of Poisson manifold, we can prove the nondegenerate. Let (M,{·,·}) be a Poisson manifold; that is, there exists a Poisson bracket {·,·} on the smooth functions C∞(M). For any f,g,h∈C∞(M) it satisfies the following:
{f,g}=-{g,f},
{f,gh}=g{f,h}+h{f,g},
{f,{g,h}}+{h,{f,g}}+{g,{h,f}}=0.
Definition 1.
A smooth diffeomorphism φ:M→M is called a Poisson diffeomorphism if for all g,h∈C∞(M), one has φ*{g,h}={φ*(g),φ*(h)}.
Given h∈C∞(M), the Hamiltonian vector is defined by Xh={·,h}. Let Cas(M)={f∈C∞(M):{f,g}=0, ∀g∈C∞(M)} be the set of Casimir functions. In this paper, one considers the time-dependent Hamiltonian functions C∞([0,1]×M,R). If the manifold is compact, or the function is compactly supported, then the flow of the Hamiltonian vector globally exists. One denotes by 𝒫(M), Ham(M) the set of such Hamiltonian flows and set of time-1 map of such flows, respectively.
For f∈C∞(M), define
(1)∥f∥=inf{∥f1∥∞∣f=f1+f2,f2∈Cas(M)}.
If Ht is a Hamiltonian flow with some Hamiltonian function ht(x), one defines its length to be
(2)length(Ht)=∫01∥ht(x)∥dt;
one can generalize the bi-invariant metric on Ham(M) to the Poisson case.
Definition 2.
Now one can define the energy of ϕ∈Ham(M),
(3)E(ϕ)=inf{length(Ht)∣vHtisaHamiltonianflowendedwithϕ}.
So one can define d: Ham(M)×Ham(M)→[0,∞)(4)d(φ,ψ)=E(φ-1∘ψ),
for φ,ψ∈Ham(M).
Theorem 3.
Let (M,{·,·}) be a regular Poisson manifold; the function d is a bi-invariant metric; that is, for all φ,ψ,θ∈Ham(M) it satisfies the following:
d(φ,ψ)≥0 and d(φ,ψ)=0 if and only if φ=ψ,
d(φ,ψ)≤d(φ,θ)+d(θ,ψ) and d(φ,ψ)=d(ψ,φ),
d(θφ,θψ)=d(φ,ψ)=d(φθ,ψθ).
Here a Poisson manifold is called regular if the rank of the Poisson manifold is constant for all point. If one replaces the L1,∞-norm by the L∞-norm, one also gets a norm on Ham(M). One proves that they are equal on closed regular Poisson manifold.
Theorem 4.
For ϕ∈Ham(M) on a closed regular Poisson manifold, one has
(5)∥ϕ∥1,∞=∥ϕ∥∞.
Let (M,{·,·}) be a Poisson manifold and let G be a Lie group acting canonically, freely, and properly on M via the map Φ:G×M→M. Let 𝒥:M→M/AG′ be the corresponding optimal momentum map. Then the orbit space M/G is a Poisson manifold with Poisson bracket {·,·}M/G uniquely characterized by the relation
(6){f,g}M/G(π(m))={f∘π,g∘π}(m);
for any m∈M and f,g:M/G→R are arbitrary smooth functions.
For a G-invariant smooth function on M, the Hamiltonian flow Ht of Xh induces a Hamiltonian flow HtM/G, so one has a well-defined homomorphism
(7)Ψ:Ham(M)G⟶Ham(MG),
where Ham(M)G denotes the G-invariant Hamiltonian maps. Now one can give a similar result as stated in [5].
Theorem 5.
For a G-invariant Hamiltonian path Ft with Hamiltonian function f(x), if infg∈Cas(M)∥f-g∥=infg∈CasG(M)∥f-g∥, one has
(8)∥Ψ(F1)∥≤length(Ft).
Moreover, if the path is length-minimizing, one has the following corollary.
Corollary 6.
If the G-invariant Hamiltonian path Ft is length-minimizing, then
(9)∥Ψ(F1)∥≤∥F1∥.
Organization of This Paper. The organization is as follows. First we will introduce the definition of the distance and give some properties. Next we will show the proofs of Theorems 3 and 4. Then we introduce the Poisson reduction. And last we give the proof of Theorem 5.
2. The Distance on Ham(M)
In this section, we recall the construction of Hofer-norm on Ham(M) and give our definition. For h∈C∞(M), define ∥h∥∞=maxx∈Mh(x)-minx∈Mh(x).
For f∈C∞(M), we now define(10)∥f∥=inf{∥f1∥∞∣f=f1+f2,f2∈Cas(M)}.
Proposition 7.
∥·∥ defined above is a pseudonorm.
Proof.
We just need to show the triangle inequality holds. Let f=f1+f2 and g=g1+g2 such that ∥f1∥∞≤∥f∥+ϵ, and ∥g1∥∞≤∥g∥+ϵ for given ϵ>0. Then
(11)∥f+g∥≤∥f1+g1∥∞≤∥f1∥∞+∥g1∥∞≤∥f∥+∥g∥+2ϵ.
From the definition we can see that ∥h∥=0 when h∈Cas(M), so together with the triangle inequality we get that if f,g∈C∞(M) and satisfies f-g∈Cas(M), then ∥f∥=∥g∥.
Proposition 8.
The new pseudonorm ∥·∥ is Ham(M)-invariant.
Proof.
First, note that ∥·∥∞ is Ham(M)-invariant. If f=f1+f2 then f1∘ϕ,f2∘ϕ is a decomposition of f∘ϕ. So
(12)∥f∘ϕ∥≤∥f1∘ϕ∥∞≤∥f1∥∞.
So ∥f∘ϕ∥≤∥f∥ for any ϕ∈Ham(M), and ∥f∥=∥f∘ϕ∘ϕ-1∥≤∥f∘ϕ∥.
We get that ∥f∘ϕ∥=∥f∥.
Now consider a Hamiltonian function ht∈C∞([0,1]×M,R); the length of the Hamiltonian path generated by ht can be defined as follows:
(13)length(H[0,1])=∫01∥ht(x)∥dt.
This length is well defined; that is, it is independent the choice of the Hamiltonian functions. This is because of our pseudonorm vanishing on Cas(M).
Definition 9.
Now one can define the energy of ϕ∈Ham(M),
(14)E(ϕ)=inf{length(Ht)∣llllllllllHtisaHamiltonianflowendedwithϕ}.
Proposition 10.
The energy function E:Ham(M)→R has the following properties:
E(φ)≥0 and E(φ)=0⇔φ=id,
E(φ)=E(φ-1),
E(ϑφϑ-1)=E(φ),
E(φ∘ψ)≤E(φ)+E(ψ),
where φ,ψ∈Ham(M) and ϑ is a Poisson diffeomorphism of M.
To prove this, we need to investigate the Hamiltonian functions of the Hamiltonian flows. Similar to the symplectic case, for the symplectic case, see page 144 of [6].
Definition 11.
If h, k are smooth functions in C∞([0,1]×M,R) and ϑ∈Ham(M) one defines the functions h¯,h#k and hϑ as follows:
(15)vvvivh¯(t,x)=-h(t,φht(x)),i(h#k)(t,x)=h(t,x)+k(t,(φht)-1(x)),vvivhϑ(t,x)=h(t,ϑ-1(x)).
Proposition 12.
If h, k are smooth functions the following formulate hold true:
(16)φh¯t=(φht)-1,φh#kt=φht∘φkt,ϑ∘φht∘ϑ-1=φhϑt,(φht)-1∘φkt=φgt,
where g=h¯#k=(k-h)(t,φht).
To prove this, we need the following fact.
Lemma 13 (see [7]).
If φ is a Poisson map, and f∈C∞([0,1]×M,ℝ), then
(17)dφ·Xf∘φ=Xf∘φ.
Proof.
For any function g∈C∞([0,1]×M,ℝ), we have
(18)Xf∘φ[g∘φ](z)=dg∘φ(z)·dφ(Xf∘φ(z)),Xf[g](φ(z))=dg(φ(z))·Xf(φ(z)).
Since φ is a Poisson map, we have
(19)Xg∘φφ*f={f∘φ,g∘φ}={f,g}∘φ=φ*Xgf.
So we have
(20)dφ(Xf∘φ(z))=Xf∘φ(z).
Proof of Lemma 13.
The third formula is just the transition law of Hamiltonian vector. We know that if φ is a Poisson map then
(21)φ*Xf∘φ=Xf(φ),ddtϑ∘φht∘ϑ-1=ϑ*Xht∘ϑ-1=Xh∘ϑ-1.
Now we prove the second formula; we abbreviate the notation and observe that
(22)ddtφt=Xh∘φt,φ0=id,ddtψt=Xh∘ψt,ψ0=id.
We need to show that φt∘ψt is the flow of Xh#k,
(23)ddtφt∘ψt=ddtφt∘ψt+(dφt∘ψt)·ddtψt=Xh(φt∘ψt)+[dφt∘(φt)-1∘φt∘ψt]·Xk∘[(φt)-1∘φt∘ψt]=Xh(φt∘ψt)+[d(φt)-1]-1·Xk∘[(φt)-1∘φt∘ψt]=Xh(φt∘ψt)+Xk∘(φt)-1(φt∘ψt)=Xh#k(φt∘ψt).
By the property of the Poisson diffeomorphism, we get that the second term is Xk∘(φt)-1. This finishes the proof of the second formula. We can obtain the first formula from the second. From the first two we can get the last one.
We are now ready to prove Proposition 10.
Proof of Proposition 10.
From Proposition 12 and the Ham-invariant of our pseudonorm, we get
(24)length(φh¯t)=length((φht)-1),
and thus E(φ)=E(φ-1). From
(25)ϑ∘φht∘ϑ-1=φhϑt,
we find
(26)length(ϑ∘φh∘ϑ-1)[0,1]=length(φhϑ)[0,1],
so the third equality holds.
To prove the last one, we note that φh#kt=φht∘φkt, and
(27)length(φh#k[0,1])=∫01∥h#k∥dt≤∫01∥ht∥+∥kt∥dt=length((φh)[0,1])+length((φk)[0,1]),
so we have E(φh∘φk)≤E(φh)+E(φk), and this implies the last one. Now we prove the first one; that is, E(φ)=0 implies φ=id. By definition, E(φ)=inf{length(Ht)∣Htgeneratesϕ}. Note that regular Poisson manifold is essentially a union of symplectic manifolds which fit together in a smooth way, so we denote by Pα the symplectic leaf of (M,{·,·}). The Hamiltonian vector field restricted to each leaf is just the Hamiltonian vector field generated by the restriction of the Hamiltonian function to the leaf. And the Hamiltonian flow keeps the symplectic leaf; that is, φhtPα0=Pα0, so we can consider the restriction of φ to each leaf. For each Hamiltonian function h generating φ, we denote by h~ the restriction of h on Pα0; for any Casimir function g, the restriction of g on each leaf is constant. Let ∥·∥Pα0 be the Hofer norm on the symplectic leaf,
(28)∥h~∥Pα0=∥h~-g~∥Pα0=∫01[maxx∈Pα0h(t,x)-g(t,x)]-minx∈Pα0[h(t,x)-g(t,x)]dt≤∫01[maxx∈Mh(t,x)-g(t,x)]-minx∈M[h(t,x)-g(t,x)]dt.
Taking the infimum of g, we get
(29)∥h~∥Pα0≤∥h∥
and hence
(30)infhgeneratsφ∥h~∥Pα0≤infhgeneratsφ∥h∥.
By the assumption of E(φ)=0, we get that infhgeneratsφ∥h~∥Pα0=0, by the definition of Hofer metric, φ|Pα0=id and so φ=id.
Now we can define d: Ham(M)×Ham(M)→[0,∞)(31)d(φ,ψ)=E(φ-1∘ψ),
for φ,ψ∈Ham(M). This distance has the following properties.
Theorem 3'. Let (M,{·,·}) be a Poisson manifold; the function d is a bi-invariant metric; that is, for all φ,ψ,θ∈Ham(M) it satisfies the following:
d(φ,ψ)≥0 and d(φ,ψ)=0 if and only if φ=ψ,
d(φ,ψ)≤d(φ,θ)+d(θ,ψ) and d(φ,ψ)=d(ψ,φ),
d(θφ,θψ)=d(φ,ψ)=d(φθ,ψθ).
Proof.
The proof of this theorem is a consequence of Proposition 10.
Example 14.
We consider the trivial Poisson manifold M; the Poisson bracket {·,·} is always zero. In this situation, any function on M is Casimir function, the Hamiltonian vector is always zero and the Hamiltonian diffeomorphism is only id; the Hofer-norm of the Hamiltonian map is 0. When the Poisson manifold is symplectic, that is, there is only one leaf, in this case the Hofer norm is just the one defined by Hofer.
Remark 15.
Theorem 3 holds not only for regular manifold, but also for many other manifolds, for example, when the rank of the Poisson manifold is not zero, or the symplectic leaves are always open or always closed.
Next we can get an estimate for the commutators in Ham(M), denoted by
(32)[φ,ψ]:=φψφ-1ψ-1.
Claim 16.
E([φ,ψ])≤2min{E(φ),E(ψ)} for φ,ψ∈Ham(M).
Proof.
From Proposition 10 we get
(33)E([φ,ψ])=E(φψφ-1ψ-1)≤E(φψφ-1)+E(ψ-1)≤2E(ψ).
Similarly,
(34)E([φ,ψ])=E(φψφ-1ψ-1)≤E(φ)+E(ψφ-1ψ-1)≤2E(φ),
so the claim is proved.
Proposition 17.
IfU⊂M is open and bounded, ϑ∈Ham(M) satisfies ϑ(U)∩U=∅, then
(35)E([φ,ψ])≤4E(ϑ),
for all φ,ψ∈Ham(M) with supports contained in U.
Proof.
Define
(36)γ:=φϑ-1φ-1ϑ∈Ham(M);
since ϑ(U)∩U=∅ and supp(φ)⊂U, we get that γ|U=φ|U. Consequently
(37)φψφ-1ψ-1=γψγ-1ψ-1.
The above maps equal M. According to Claim 16 and Proposition 10 we get
(38)E([φ,ψ])=E([γ,ψ])≤2E(γ)≤2E(ϑ-1)+2E(ϑ)=4E(ϑ).
Proposition 18.
For a subset A⊆M, define
(39)e(A)=inf{d(id,ϕ)ϕ(A)∩A=∅,ϕ∈Ham(M)},
the displacement energy of A. Then for any open bounded nonempty subset A, one has
e(A)≠0,
if A⊂B, then e(A)≤e(B);
e(ϕ(A))=e(A) for ϕ∈Ham(M).
Proof.
e(A)≠0 is a consequence of above statements and the monotonicity is by the definition. We just prove the second one. For ϕ,θ∈Ham(M), if θ(A)∩A=∅, then ϕθϕ-1ϕ(A)∩ϕ(A)=∅; if θϕ(A)∩ϕ(A)=∅, then ϕ-1θϕ(A)∩A=∅. According to Proposition 10, we have E(ϑφϑ-1)=E(φ); from the above identities, we get the conclusion.
Following [8], we now define a new function for a Poisson map. For a Poisson map ϕ on a closed Poisson manifold (M,{·,·}), define
(40)r(ϕ)=sup{E([ϕ,f])f∈Ham(M)},
where [ϕ,f]=ϕf(ϕ)-1f-1. Similarly, we can define rα(ϕ) if we restrict f in the closed ball of radius of α of Ham(M) centered at id.
Definition 19.
A map ϕ on a closed regular Poisson manifold (M,{·,·}) is bounded if r(ϕ)<+∞ and unbounded otherwise.
Proposition 20.
For any α∈(0,+∞], the function rα(ϕ) is bi-invariant, assumes the value 0 only at id, and satisfies the triangle inequality
(41)rα(ϕφ)≤rα(ϕ)+rα(φ),rα(ϕ)≤2α,
for any Poisson map ϕ and α∈(0,+∞].
Proof.
The proof is a consequence of the definitions of rα and the metric.
Proposition 21.
Let (M,{·,·}) be a closed regular Poisson manifold; if there exists some ϕ unbounded, then the Hofer metric d does not extend to a bi-invariant metric on the groups of Poisson maps.
Proof.
Assume that we can extend the Hofer metric to the groups of Poisson maps; we still denote the metric by d; then E([ϕ,f])=E(ϕf(ϕ)-1f-1)≤2E(ϕ) for any f∈Ham(M). So we have
(42)r(ϕ)≤2E(ϕ),
for any ϕ, and this is impossible; hence we finish the proof.
Now we consider the geodesic under the above norm in Ham(M). For the standard symplectic manifold, Hofer proved that, the Hamiltonian flow generated by the time-independent compactly supported function is a geodesic [6]. Later Bialy and Polterovich gave a sufficient and necessary condition for a path to be geodesic [9]; last Lalonde and McDuff extended it to all symplectic manifolds [4]. Now we consider similar questions on regular Poisson manifold.
Theorem 22.
Let φht be a Hamiltonian flow with compactly supported time-independent Hamiltonian function h on (M,{·,·}), and if the maximal and minimal point of h lie in the same symplectic leaf, then φht is a geodesic; that is, d(φht,φhs)=(t-s)∥h∥ for |s-t| sufficiently small.
Proof.
First, by the definition we have
(43)d(φht,φhs)=d(φht-s,id)≤|s-t|∥h∥.
To prove the converse, let Pα be the symplectic leaves of the Poisson manifold; we still denote by h~ the restriction of h on Pα0; for any compactly supported Hamiltonian function g generating φht-s, we have
(44)infggeneratsφht-s∥g~∥Pα0≤infggeneratsφht-s∥g∥.
So
(45)∥φht-s∥≥∥φht-s∥|Pα0,
on each leaf, but on each leaf, according to the results on symplectic manifold mentioned above, the flow φh~t is a geodesic; we thus have ∥φht-s∥|Pα0≥|s-t|∥h~∥ for |s-t| sufficiently small. By the assumption, we have
(46)maxx∈Pα0h(x)-minx∈Pα0h(x)≥maxx∈Mh(x)-minx∈Mh(x),
for some leaf Pα0. Hence
(47)maxx∈Pα0h(x)-minx∈Pα0h(x)≥infg∈Cas(M)maxx∈Mh(x)-g(x)-minx∈Mh(x)-g(x).
Since h is compactly supported, we have the C2-norm of h is bounded above by some number T, and moreover,
(48)∥h∥C2≥∥h∥C2|Pα0,
so by Theorem 1.2 of [3], we can choose a constant T′>T such that ∥φht-s∥≥|s-t|∥h∥ when |s-t|≤1/T′ and this finishes the proof.
For the time-dependent case, we have a similar result. First we recall the definition of quasiautonomous function.
Definition 23.
A function h(t,x) on [0,1]×M is called quasiautonomous if there exist two points x+,x-∈M such that maxxh(t,x)=h(t,x+), minxh(t,x)=h(t,x-) for all t∈[0,1].
Theorem 24.
Let φht be a Hamiltonian flow with compactly supported Hamiltonian function h on (M,{·,·}); if h is quasiautonomous on each symplectic leaf, and the fixed maximum x+ and fixed minimum x- of h lie in the same symplectic leaf, then φht is a geodesic; that is, d(φha,φhb)=Length(φh[a,b]) for |a-b| sufficiently small.
Proof.
We first adopt the transformation in [10] to simplify the problem; for interval [a,b], define δ(t)=a+t(b-a) and k(t,x)=(b-a)h(δ(t),x) for t∈[0,1], x∈M. Then the Hamiltonian flow of k(t,x) satisfies(49)φkt=φhδ(t)∘(φha)-1fort∈[0,1],φk0=id,φk1=φhb∘(φha)-1,∫01infg∈Cas(M)[maxxk(t,x)-g(x)-minxk(t,x)-g(x)]dt=∫abinfg∈Cas(M){maxxh(t,x)-1b-ag(x)vvvvvvvvvvv-minxh(t,x)-1b-ag(x)}dt=∫abinfg∈Cas(M){maxxh(t,x)-g(x)vvvvvvvvvvvvv-minxh(t,x)-g(x)}dt.
This implies that ∥k∥=∥h∥; by the assumption, h(t,x) is quasiautonomous on each symplectic leaf and so is k; by Theorem 1.2 of [11, 12], we know that φkt is a minimum geodesic on each leaf provided that |a-b| sufficiently small; that is,
(50)d(φk0,φk1)|Pα0=∥k∥|Pα0.
Because the fixed maximum x+ and fixed minimum x- of h lie in the same symplectic leaf, we may assume that this leaf is Pα0. We have
(51)∫01maxx∈Pα0h(t,x)-minx∈Pα0h(t,x)dt≥∫01maxx∈Mh(t,x)-minx∈Mh(t,x)dt.
So,
(52)∫01maxx∈Pα0h(t,x)-minx∈Pα0h(t,x)dt≥∫01infg∈Cas(M){maxx∈Mh(t,x)-g(x)vvvvvvvvvvvvvvvvvv-minx∈Mh(t,x)-g(x)}dt=∥h∥.
By Theorem 3, we have ∥h∥=∥k∥=d(φk0,φk1)=d(φha,φhb).
Remark 25.
If the Poisson manifold is symplectic, then Theorems 22–24 reduce to the results in [4, 6, 9]. If we make more assumptions on the Hamiltonian functions, we can get similar results about the minimizing geodesics.
Theorem 26.
Let gi,g be a sequence of smooth Hamiltonians on a closed Poisson manifold M; suppose that
gi→g in the C0-topology,
ϕgit→ϕgt in the C0-topology, t∈[0,1].
If all ϕgit are minimizing, then ϕgt is also minimizing.
Proof.
We employ the method of Oh in [13] to prove it; for reader's convenience, we write it here. Suppose that ϕgt is not minimizing, then we choose a function f, such that ∥f∥<∥g∥; choose δ>0 such that
(53)∥f∥<∥g∥-δ;
then
(54)∥f∥<∥gi∥-δ2,
when i is sufficiently large. Define
(55)fi=gi-g(ϕgit)+f(ϕgt∘(ϕgit)-1).
By simply computations as we know that ϕfi1=ϕgi1, ∥f∥<∥fi∥-δ/2 and fi→f in the C0-topology, so for any Casmir function h, we have fi-h→f-h, and thus
(56)limi→∞∥fi∥=∥f∥.
This is a contraction and we finish the proof.
Theorem 27.
Assume that ϕi and ϕ∈Ham(M), and ψ is a homeomorphism of M. If
d(ϕi,ϕ)→0,
limi→∞ϕi=ψ, locally uniformly,
then ϕ=ψ.
Proof.
We assume that ϕ=id, and by the assumption we have E(ϕ)→0 and limi→∞ϕi=ψ. We restrict them to each leaf and adapt the same notations and arguments in Proposition 10; we have E(ϕ|Pα0)→0 and limi→∞ϕi|Pα0=ψ|Pα0. By Theorem 6 page 169 in [6], we know that ψ|Pα0=id; this holds on each leaf, so ϕ=ψ.
Corollary 28.
If ϕhi∈Ham(M) and ψ is a homeomorphism of M satisfying
hi→h uniformly, ht∈Cas(M)
limi→∞ϕi=ψ, locally uniformly
then ψ=id.
If one replaces the L1,∞-norm by the L∞-norm, one also gets a pseudonorm on Ham(M).
For a Hamiltonian function h∈C∞([0,1]×M,R), define the pseudolength of the Hamiltonian path generated by h as follows:
(57)length(H[0,1])=maxt∈[0,1]∥ht(x)∥.
Similarly, one can define the energy and the pseudometric.
Definition 29.
The energy of ϕ∈Ham(M),
(58)E(ϕ)=inf{length(Ht)∣llllllllllHtisaHamiltonianflowendedwithϕ}.
Define d: Ham(M)×Ham(M)→[0,∞)(59)d(φ,ψ)=E(φ-1∘ψ),
for φ,ψ∈Ham(M). This d is also a bi-invariant metric. One denotes by ∥·∥1,∞, ∥·∥∞ the induced L1,∞-norm and the L∞-norm, respectively.
Recall that in the symplectic case, Polterovich proved that the L1,∞-norm and the L∞-norm coincide on closed symplectic manifolds [14]. We now give a similar result in the Poisson case.
Theorem 4'. For ϕ∈Ham(M) on a closed Poisson manifold, one has(60)∥ϕ∥1,∞=∥ϕ∥∞.
We first show the following results which will be useful in the proof.
Proposition 30.
Cas(M) is closed in the C0-topology.
Proof.
We just show that this is true on each symplectic leaf, but in the symplectic case, the Casimir functions are constants; this finishes the proof.
Proposition 31.
Let Ft be a flow generated by a Hamiltonian function f on a closed Poisson manifold. Then there exists an arbitrary small loop Ht such that the Hamiltonian function kt of the flow Ht-1Ft satisfies {k(t),·}≠0 for every t.
Proof.
First, by Proposition 12 we know that the Hamiltonian function of Ht-1Ft can be given by k(t,x)=f(t,Htx)-h(t,Htx). We now just show that {k(t),·}≠0 for all t. Take a function g∈C∞([0,1]×M,R) such that ∫01g(t,x)dt=0 for every x∈M. Then define Ht(ϵ)∈Ham(M) as the time-ϵ map of the Hamiltonian flow generated by ∫0tg(s,x)ds. Let h(t,ϵ,x) be the Hamiltonian function of the loop Ht(ϵ), then we need the following proposition of Banyaga.
Let hs,t be a 2-family smooth parameters of diffeomorphisms on a smooth manifold M such that h0,0=id. Let Xs,t,Ys,t be the families of vectors on M defined by
(61)Xs,t(x)=ddths,t(hs,t-1(x)),Xs,t(x)=ddshs,t(hs,t-1(x));
then
(62)∂Xs,t∂s=∂Ys,t∂t+[Xs,t,Ys,t].
By the above proposition, we now compute (∂/∂ϵ)|ϵ=0h(t,ϵ,x). By definition
(63)X·,ϵ={·,h(·,ϵ)},Yt,·={·,h(t,·)},∂Xt,ϵ∂ϵ|ϵ=0={·,∂h(t,ϵ,x)∂ϵ|ϵ=0}={·,∂h(t,ϵ,x)∂t|ϵ=0}+[Xt,ϵ,Yt,ϵ]={·,g(t,x)}.
So (∂/∂ϵ)|ϵ=0h(t,ϵ,x)=g(t,x) up to a Casimir function.
Now we define
(64)Ht(ϵ1,…,ϵk)=Ht1(ϵ1)∘⋯∘Htk(ϵk).
Here every Hj is constructed with the help of gj as above; we know that the partial derivative of the Hamiltonian h(t,ϵ,x) with respect to ϵj at ϵ=0 equals gj.
Fixed a regular point y∈M, let N be the symplectic leaf though y, consider the linear space E=TyN⊆TyM. Choose 2n smooth closed curves α1(t),…,α2n(t) (where t∈S1) satisfying the following conditions:
∫01αj(t)dt=0 for all j=1,…,2n;
the vectors α1(t),…,α2n(t) are linearly independent for every t. The existence of such system of curves is shown in [14]; for example, choose a basis u1,v1,…,un,vn in E and take the curves of the form ujcos2πt+vjsin2πt and -ujsin2πt+vjcos2πt.
Now choose αj(t),gj(t) such that
gj(t)∉Cas(M) for every t;
∫01gj(t,x)dt=0 for every x∈M;
{·,gj(t)}(y)=αj(t).
Take the corresponding 2n-parameter variation Ht(ϵ) of the constant loop as above. Consider the map Γ:S1×R2n(ϵ1,…,ϵ2n)→E defined by
(65)(t,ϵ)⟶{·,ft-ht(ϵ)}(y).
It follows that Γ is a submersion in some neighbourhood U of the circle {ϵ=0}. Indeed from our construction we have
(66)∂∂ϵj|ϵ=0Γ(t,ϵ)=αj(t).
But these vectors generates the whole E. Denote by Γ~ the restriction of Γ to S1×U. Since Γ~ is a submersion, the set Γ~-1(0) is a one-dimensional submanifold of S1×U, so there exist arbitrary small values of the parameter ϵ such that {·,ft-ht(ϵ)}(y)≠0 for all t. This completes the argument.
Proof of Theorem 4.
For ϕ∈Ham(M), clearly we have ∥ϕ∥1,∞≤∥ϕ∥∞. Now we prove the converse. Fix a positive number ϵ, choose a path Ft∈Ham(M) such that F0=id, F1=ϕ with Hamiltonian functions ft and ∫01∥ft∥dt≤∥ϕ∥1,∞+ϵ. By Proposition 31 we can assume that ∥ft∥>0 for all t since the manifold is good manifold. Here this pseudonorm is defined in Proposition 7. Define a(t) as the inverse of
(67)b(t)=∫0t∥ft∥dt∫01∥ft∥dt,
where t∈[0,1]. Note that the Hamiltonian of Fa(t) can be generated by fa=a′(t)f(a(t),x), where a′(t) denotes the derivative with respect to t. Note that
(68)maxt∈[0,1]∥a′(t)f(a(t),x)∥=maxt∈[0,1]∥ft∥b′(t)=∫01∥ft∥dt;
we get that maxt∈[0,1]∥a′(t)f(a(t),x)∥≤∥ϕ∥1,∞+ϵ. Approximating a in the C1-topology by a smooth one, we get a Hamiltonian denoted by f~; we have maxt∈[0,1]∥f~∥+2ϵ. Since ϵ is arbitrary, we conclude that ∥ϕ∥∞≤∥ϕ∥1,∞. This completes the proof.
Corollary 33.
For f∈C∞([0,1]×M,R), one has
(69)∥ϕf1∥1,∞=∥ϕf1∥∞=inf{∥f-h∥∞∣ϕh1=id}.
Proof.
Let ϕgt be another Hamiltonian flow with Hamiltonian g(t,x) and satisfy ϕg1=ϕf1; then there exists a Hamiltonian function h(t,x) such that ϕgt=ϕht∘ϕft and ϕh0=ϕh1=id. On the other hand, for each loop ϕht, the time one map of the flow ϕht∘ϕft is also ϕf1; by Proposition 12 its Hamiltonian can be
(70)g(t,x)=h(t,x)+f(t,(ϕht)-1x).
Note that h-=-h(t,ϕhtx) generates loop (ϕht)-1, so
(71)∥ϕf1∥1,∞=inf{∥f-h∥∞∣ϕh1=id}.
3. Poisson Reduction
In this section, we briefly introduce the Poisson reduction. Let G be a Lie group acting canonically on M; if the action is free and proper, we know that the orbit space M/G is a smooth manifold and the canonical projection π:M→M/G is a smooth surjective submersion. Let 𝒥:M→M/AG′ be the corresponding optimal momentum map. The orbit space M/G is a Poisson manifold with Poisson bracket {·,·}M/G uniquely characterized by the relation
(72){f,g}M/G(π(m))={f∘π,g∘π}(m),
for any m∈M and f,g:M/G→R are arbitrary smooth functions.
The Poisson structure induced by the bracket {·,·}M/G on M/G is the only one for which the projection π:(M,{·,·})→(M/G,{·,·}M/G) is a Poisson map. Let h∈C∞(M)G be a G-invariant Hamiltonian flow Ft of Xh commutes with the G-action, so it induces a flow FtM/G on M/G characterized by
(73)π∘Ft=FtM/G∘π.
The flow FtM/G is Hamiltonian on (M/G,{·,·}M/G) for the reduced Hamiltonian function h~∈C∞(M/G) defined by
(74)h~∘π=h.
The vector fields Xh and Xh~ are π-related.
So we have a well-defined homomorphism
(75)Ψ:Ham(M)G⟶Ham(MG),
where Ham(M)G denotes the G-invariant Hamiltonian maps.
More details can be found in [16].
4. Proof of Theorem 5
Now we can give the proof of Theorem 5.
Theorem 5'. For a G-invariant Hamiltonian path Ft with Hamiltonian function f, if infg∈Cas(M)∥f-g∥=infg∈CasG(M)∥f-g∥, one has(76)∥Ψ(F1)∥≤length(Ft).
Proof.
From the above discussion, we know that For any G-invariant Hamiltonian path Ft, it induces a Hamiltonian path Ψ(Ft) on (M/G,{·,·}M/G).
Let f∈C∞(M)G,f~∈C∞(M/G) be the Hamiltonian of the Hamiltonian path Ft, and the induced path Ψ(Ft). We have
(77)f~∘π=f.
By the definition of the norm, we have
(78)supx∈M/Gf(x)~-infx∈M/Gf(x)~≤supx∈Mf(x)-infx∈Mf(x),
and note that
(79)infg∈Cas(M)∥f-g∥=infg∈CasG(M)∥f-g∥.
According to the above discussions, we have
(80)∥Ψ(F1)∥≤length(Ψ(Ft))=infg∈Cas(M/G)∥f~-g∥≤infg∈CasG(M)∥f-g∥≤∥f(x)∥,
so
(81)∥Ψ(F1)∥≤length(Ft).
Moreover, if the path is length-minimizing, that is, Length(Ft)=∥F1∥ then we have Corollary 6.
Remark 34.
If the Poisson manifold is symplectic, then the pseudonorm is the Hofer norm. Give a Hamiltonian G-action; then we can get the results in the symplectic case as in [5].
Conflict of Interests
The authors declare that they have no conflicts of interest regarding this work.
Acknowledgments
The authors would like to express their deep gratitude to Professor Yiming Long for many valuable discussions. The research was supported by TianYuan Program of National Natural Science Foundation of China (11226158), Natural Science Foundation of Henan (2011B110011), and Doctor Fund of Henan University of Technology.
HoferH.On the topological properties of symplectic maps19901151-2253810.1017/S0308210500024549MR1059642ZBL0713.58004ViterboC.Symplectic topology as the geometry of generating functions1992292468571010.1007/BF01444643MR1157321ZBL0735.58019PolterovichL.Symplectic displacement energy for Lagrangian submanifolds199313235736710.1017/S0143385700007410MR1235478ZBL0781.58007LalondeF.McDuffD.The geometry of symplectic energy1995141234937110.2307/2118524MR1324138ZBL0829.53025PedrozaA.Induced Hamiltonian maps on the symplectic quotient200826550350710.1016/j.difgeo.2008.04.005MR2458276ZBL1151.53071HoferH.ZehnderE.1994Basel, SwitzerlandBirkhäuserxiv+34110.1007/978-3-0348-8540-9MR1306732MarsdenJ. E.RatiuT. S.199417New York, NY, USASpringerxvi+500Texts in Applied Mathematics10.1007/978-1-4612-2682-6MR1304682LalondeF.PolterovichL.Symplectic diffeomorphisms as isometries of Hofer's norm199736371172710.1016/S0040-9383(96)00024-9MR1422431ZBL0874.53028BialyM.PolterovichL.Geodesics of Hofer's metric on the group of Hamiltonian diffeomorphisms199476127329210.1215/S0012-7094-94-07609-6MR1301192ZBL0819.58006LongY. M.Geodesics in the compactly supported Hamiltonian diffeomorphism group1995220227929410.1007/BF02572615MR1355031ZBL0843.58022LalondeF.McDuffD.Hofer's L∞-geometry: energy and stability of Hamiltonian flows, part I1995122113310.1007/BF01231437LalondeF.McDuffD.Hofer's L∞-geometry: energy and stability of Hamiltonian flows, part II1995122135692-s2.0-3424976100710.1007/BF01231438OhY.-G.Chain level Floer theory and Hofer's geometry of the Hamiltonian diffeomorphism group200264579624MR1958084ZBL1038.53084PolterovichL.2001Basel, SwitzerlandBirkhäuserxii+132Lectures in Mathematics ETH Zürich10.1007/978-3-0348-8299-6MR1826128BanyagaA.1997400Dordrecht, The NetherlandsKluwer Academic Publishersxii+197Mathematics and Its ApplicationsMR1445290OrtegaJ.-P.RatiuT. S.2004222Boston, Mass, USABirkhäuserxxxiv+497Progress in MathematicsMR2021152