LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMSCONTENTS11.Geometry of Hamiltonian vector fields42.ThePoisson structure73.Completely integrableHamiltonian systems1.GEOMETRYOFHAMILTONIANVECTORFIELDS Symplectic vectorfield v.s.Hamiltonian vector field.Let (M,w)be a symplectic manifold. Then the non-degeneracy of w gives us alinear isomorphism between vector fields and 1-forms on M:W:Vect(M)→2(M),三→=w.Recall that a vector field E on M is called symplectic if t=w is a closed 1-form onM, and it is called Hamiltonian if tgw is an exact 1-form. So if we denote the set ofsymplectic vector fields by Vect(M,w) and the set of Hamiltonian vector fields byVectHam(M,w), then the restriction of gives us linear isomorphismsw : Vect(M,w) → Z'(M)andw : VectHam(M,w) → B'(M),where z(M) is the space of closed 1-forms on M, and B'(M) the space of exact1-forms.As a consequence, the quotient Vect(M,w)/VectHam(M,w) is just the firstdeRham cohomology group H'(M), and we have an exact sequence of vector spaces(1)0 → Vect(M,w) → VectHam(M,w) →H'(M) →0.In particular, we seeProposition 1.1. If H'(M) = [0), then every symplectic vector field on M isHamiltonian1
LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS Contents 1. Geometry of Hamiltonian vector fields 1 2. The Poisson structure 4 3. Completely integrable Hamiltonian systems 7 1. Geometry of Hamiltonian vector fields ¶ Symplectic vector field v.s. Hamiltonian vector field. Let (M, ω) be a symplectic manifold. Then the non-degeneracy of ω gives us a linear isomorphism between vector fields and 1-forms on M: ω˜ : Vect(M) → Ω 1 (M), Ξ 7→ ιΞω. Recall that a vector field Ξ on M is called symplectic if ιΞω is a closed 1-form on M, and it is called Hamiltonian if ιΞω is an exact 1-form. So if we denote the set of symplectic vector fields by Vect(M, ω) and the set of Hamiltonian vector fields by VectHam(M, ω), then the restriction of ˜ω gives us linear isomorphisms ω˜ : Vect(M, ω) → Z 1 (M) and ω˜ : VectHam(M, ω) → B 1 (M), where Z 1 (M) is the space of closed 1-forms on M, and B1 (M) the space of exact 1-forms. As a consequence, the quotient Vect(M, ω)/VectHam(M, ω) is just the first deRham cohomology group H1 (M), and we have an exact sequence of vector spaces (1) 0 → Vect(M, ω) → VectHam(M, ω) → H 1 (M) → 0. In particular, we see Proposition 1.1. If H1 (M) = {0}, then every symplectic vector field on M is Hamiltonian. 1
2LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMS Smooth function v.s. Hamiltonian vector field.Recallweshowedinlecture4thatifE,.三aretwosvmplecticvectorfieldsonM, then [三i, 三2] is a Hamiltonian vector field on M. In fact, one has(2)(31,=2)W = -d(w(三1,三2),This implies[Vect(M,w), Vect(M,w)] C VectHam(M,w),In other words, as a Lie algebra, Vect Ham(M,w) is an ideal of Vect(M,w). (This isof course related to the fact that Ham(M,w) is an ideal of the group Symp(M,w).)So the short exact sequence (1) is in fact a short exact sequence of Lie algebras, ifwe endowed with H'(M) the trivial Lie bracket.Modulo locally constant functions,Hamiltonian vector fields are in one-to-onecorrespondence with smooth functions on M,in the following sense:So if E ishamiltonian, then there exists a smooth function f e Co(M) so that tw =df:Conversely, since w is non-degenerate, for any f e Co(M), there is a unique vectorfield 三f on M so thatl=rw=df.The space of locally constant functions on M is the Oth de Rham cohomology groupH'(M).So we get another short exact sequence of vector spaces(3)0 -→ H°(M) →C(M) → VectHam(M,w) → 0.We shall see soon that this is again a short exact sequence of Lie algebras, providedwedefineasuitableLiealgebrastructureonCoo(M)(whichis,however,differentfrom the Poisson bracket we will define below by a negative sign).Recall: The vector field Ef is called the Hamiltonian vector field associated totheHamiltonian function f.The flow generated by Eyis called the Hamiltonianflowassociated to f.Erample. Consider M =R2n with canonical symplectic form w =de;Adei. Thenafde(ofdridf =>OriDEItfollowsofaaf a(4)Ef=os;OciOriosSo the integral curve of 三, is a curve (r(t),s(t)) satisfyingaf(t) =f(t) =DOr,This set of equations is known as Hamiltonian equations
2 LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS ¶ Smooth function v.s. Hamiltonian vector field. Recall we showed in lecture 4 that if Ξ1, Ξ2 are two symplectic vector fields on M, then [Ξ1, Ξ2] is a Hamiltonian vector field on M. In fact, one has (2) ι[Ξ1,Ξ2]ω = −d(ω(Ξ1, Ξ2)). This implies [Vect(M, ω), Vect(M, ω)] ⊂ VectHam(M, ω). In other words, as a Lie algebra, VectHam(M, ω) is an ideal of Vect(M, ω). (This is of course related to the fact that Ham(M, ω) is an ideal of the group Symp(M, ω).) So the short exact sequence (1) is in fact a short exact sequence of Lie algebras, if we endowed with H1 (M) the trivial Lie bracket. Modulo locally constant functions, Hamiltonian vector fields are in one-to-one correspondence with smooth functions on M, in the following sense: So if Ξ is hamiltonian, then there exists a smooth function f ∈ C ∞(M) so that ιΞω = df; Conversely, since ω is non-degenerate, for any f ∈ C ∞(M), there is a unique vector field Ξf on M so that ιΞfω = df. The space of locally constant functions on M is the 0th de Rham cohomology group H0 (M). So we get another short exact sequence of vector spaces (3) 0 → H 0 (M) → C ∞(M) → VectHam(M, ω) → 0. We shall see soon that this is again a short exact sequence of Lie algebras, provided we define a suitable Lie algebra structure on C ∞(M) (which is, however, different from the Poisson bracket we will define below by a negative sign). Recall: The vector field Ξf is called the Hamiltonian vector field associated to the Hamiltonian function f. The flow generated by Ξf is called the Hamiltonian flow associated to f. Example. Consider M = R 2n with canonical symplectic form ω = Pdxi∧dξi . Then df = X ∂f ∂xi dxi + ∂f ∂ξi dξi . It follows (4) Ξf = X ∂f ∂ξi ∂ ∂xi − ∂f ∂xi ∂ ∂ξi So the integral curve of Ξf is a curve (x(t), ξ(t)) satisfying x˙(t) = ∂f ∂ξi , ˙ξ(t) = − ∂f ∂xi . This set of equations is known as Hamiltonian equations
3LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMSRemark. Obviously the formula (4) also gives the local expression of Ef on an arbitrary symplectic manifold if one uses theDarboux coordinates.Thevector field 三fis also called the symplectic gradient of f. Gradient vector field v.s. Hamiltonian vector field.Back to the R2n example. Note that the usual gradient vector field of f is(af aof a)vf =>orOrosa)So the symplectic gradient and the usual gradient of f are related byJ(E) = Vf.where J is theusual complex structure on R2n, i.e.aga0ariThis observation is easily extended to Kahler manifolds, or more generally anysymplectic manifold M with compatible triple (w,J,g).In this theusual gradientvectorfield of f is thevectorfieldf onM sothatg(Vf, ) = df().Using the fact g(, ) = w(, J.) it is easy to seeProposition 1.2. Let M be a symplectic manifold with compatible triple (w, J,g).Then Vf = JEf.Proof. We havedf = g(Vf, ) =w(Vf, J.) =w(-JVf, ).口Itfollowsthat-JVf=-三f,i.e.Vf=J(E).So if one think of an almost complex structure on M as “rotation by 90 degreescouterclockwise", then the usual gradient vector field can be obtained from thesymplectic gradient vector fields via“rotation by 90degrees couterclockwise"!Symplectic form v.s.Hamiltonian vector field.Now suppose 三f is the Hamiltonian vector field associated with f.The followingproperties are easily seen from the definition. One should be aware of the use of thethree parts of the definition of a symplectic form.Proposition 1.3. Let (M,w) be a symplectic manifold, and f e Co(M)(1) C=,f = 0.(2) C=,w = 0.(3) IfE Symp(M,w), then 三* =p*三f
LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS 3 Remark. Obviously the formula (4) also gives the local expression of Ξf on an arbitrary symplectic manifold if one uses the Darboux coordinates. The vector field Ξf is also called the symplectic gradient of f. ¶ Gradient vector field v.s. Hamiltonian vector field. Back to the R 2n example. Note that the usual gradient vector field of f is ∇f = X ∂f ∂xi ∂ ∂xi + ∂f ∂ξi ∂ ∂ξi . So the symplectic gradient and the usual gradient of f are related by J(Ξf ) = ∇f. where J is the usual complex structure on R 2n , i.e. J( ∂ ∂xi ) = ∂ ∂ξi , J( ∂ ∂ξi ) = − ∂ ∂xi . This observation is easily extended to K¨ahler manifolds, or more generally any symplectic manifold M with compatible triple (ω, J, g). In this the usual gradient vector field of f is the vector field ∇f on M so that g(∇f, ·) = df(·). Using the fact g(·, ·) = ω(·, J·) it is easy to see Proposition 1.2. Let M be a symplectic manifold with compatible triple (ω, J, g). Then ∇f = JΞf . Proof. We have df = g(∇f, ·) = ω(∇f, J·) = ω(−J∇f, ·). It follows that −J∇f = Ξf , i.e. ∇f = J(Ξf ). So if one think of an almost complex structure on M as “rotation by 90 degrees couterclockwise”, then the usual gradient vector field can be obtained from the symplectic gradient vector fields via “rotation by 90 degrees couterclockwise”! ¶ Symplectic form v.s. Hamiltonian vector field. Now suppose Ξf is the Hamiltonian vector field associated with f. The following properties are easily seen from the definition. One should be aware of the use of the three parts of the definition of a symplectic form. Proposition 1.3. Let (M, ω) be a symplectic manifold, and f ∈ C ∞(M). (1) LΞf f = 0. (2) LΞfω = 0. (3) If ϕ ∈ Symp(M, ω), then Ξϕ∗f = ϕ ∗Ξf .
4LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMSProof. (1) follows from the skew-symmetry of w:Ca,f =ldf=w=0.(2) follows from the closeness of w:C=,w = di=jw = d(df) = 0.(3) follows from the non-degeneracy of w:l=W=d(0"f)=0df=0W=bp*,0'w=l0*w2.THEPOISSON STRUCTURE The Poisson bracket.Applying the identity (2) to Hamiltonian vector fields 三f and g, we get(5)[3f,三9] = =-w(=,3g)-Definition 2.1. For any f,g E Co(M), we call[f,g) =w(三f,三g)the Poisson bracket of f and gSo by definition,(f,g) = L=,w(Xg) =df(=g) = L=gf =三g(f)As a consequence we seeCorollary 2.2. (f,g] = 0 if and only if g is constant along the integral curves ofEf, i.e. the Hamiltonian vector field Ef is tangent to the level sets g = c.In particular, the hamiltonian vector field E, is always tangent to the level setsf=c.The poisson bracket also behaves well under symplectomorphisms:Proposition 2.3. If E Symp(M,w), then (*f,*g) =*[f,g).Proof. Using the fact 三*f = p*f we get[p*f, Φ*g) =w(三p*f,三p*g) =w(β*f,Φ*三g) = p*(w(三f,三g)) = p*[f,g)口
4 LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS Proof. (1) follows from the skew-symmetry of ω: LΞf f = ιΞf df = ιΞf ιΞfω = 0. (2) follows from the closeness of ω: LΞfω = dιΞfω = d(df) = 0. (3) follows from the non-degeneracy of ω: ιΞϕ∗fω = d(ϕ ∗ f) = ϕ ∗ df = ϕ ∗ ιΞfω = ιϕ∗Ξfϕ ∗ω = ιϕ∗Ξfω. 2. The Poisson structure ¶ The Poisson bracket. Applying the identity (2) to Hamiltonian vector fields Ξf and Ξg, we get (5) [Ξf , Ξg] = Ξ−ω(Ξf ,Ξg) . Definition 2.1. For any f, g ∈ C ∞(M), we call {f, g} = ω(Ξf , Ξg) the Poisson bracket of f and g. So by definition, {f, g} = ιΞfω(Xg) = df(Ξg) = LΞg f = Ξg(f). As a consequence we see Corollary 2.2. {f, g} = 0 if and only if g is constant along the integral curves of Ξf , i.e. the Hamiltonian vector field Ξf is tangent to the level sets g = c. In particular, the hamiltonian vector field Ξf is always tangent to the level sets f = c. The poisson bracket also behaves well under symplectomorphisms: Proposition 2.3. If ϕ ∈ Symp(M, ω), then {ϕ ∗ f, ϕ∗ g} = ϕ ∗{f, g}. Proof. Using the fact Ξϕ∗f = ϕ ∗Ξf we get {ϕ ∗ f, ϕ∗ g} = ω(Ξϕ∗f , Ξϕ∗g) = ω(ϕ ∗Ξf , ϕ∗Ξg) = ϕ ∗ (ω(Ξf , Ξg)) = ϕ ∗ {f, g}.
5LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMS The Poisson bracket in local coordinates.In local Darboux coordinates, (4) gives(of ogof g)(f,g] =OroaoIn particular, the Poisson bracket of the Darboux coordinate functions are simple:[,c] =[,]=0,[]=uConversely, we haveProposition 2.4. A coordinate system [r1,, n,Si,...,En) on M is a Darbourcoordinate system if and only if they satisfies[r,,] =[,] =0,[c,] =oProof. We can rewrite these set of equations asd(三r)=0,d(三)=0,dr(三)=-de(三)=It followsaaEa =EeaOriand thusaa0000=0,=SiiwrEEOrOri口T The Poisson bracket as a Lie bracket.Using Poisson bracket we can rewrite the equation (5) as(6)E(f.g) = -[Ef,Eg].In particular, if (f, g] = o, then [Ef,=g] = o, thus the Hamiltonian flows of f andg commute.Theorem 2.5. Let (M,w) be a symplectic manifold.(1) The Poisson bracket [, ↓ is a Lie algebra structure on Co(M).(2) The mapCo(M)→ VectHam(M,w), f →Efis a Lie algebra anti-homomorphism.Proof. (1) Obviously (, J is bilinear and anti-symmetric. To show it is a Lie algebrastructure it remains to check the Jacobi identity:0 =(C=,)(三g,三h)=(dt=,)(三g,三h)=三g(w(三f,三h)) - 三h(w(三f,g)) -w(三f,[Eg,三hl)={{f,h),g) -{{(f,g),h) +w(三f,三(gh))={(f,h),g) +{(g, f],h) +{(h,g), f]
LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS 5 ¶ The Poisson bracket in local coordinates. In local Darboux coordinates, (4) gives {f, g} = X ∂f ∂xi ∂g ∂ξi − ∂f ∂ξi ∂g ∂xi . In particular, the Poisson bracket of the Darboux coordinate functions are simple: {xi , xj} = {ξi , ξj} = 0, {xi , ξj} = δij . Conversely, we have Proposition 2.4. A coordinate system {x1, · · · , xn, ξ1, · · · , ξn} on M is a Darboux coordinate system if and only if they satisfies {xi , xj} = {ξi , ξj} = 0, {xi , ξj} = δij . Proof. We can rewrite these set of equations as dxi(Ξxj ) = 0, dξi(Ξξj ) = 0, dxi(Ξξj ) = −dξi(Ξxi ) = δij . It follows Ξxi = ∂ ∂ξi , Ξξi = − ∂ ∂xi and thus ω( ∂ ∂xi , ∂ ∂xj ) = ω( ∂ ∂ξi , ∂ ∂ξj ) = 0, ω( ∂ ∂xi , ∂ ∂ξj ) = δij . ¶ The Poisson bracket as a Lie bracket. Using Poisson bracket we can rewrite the equation (5) as (6) Ξ{f,g} = −[Ξf , Ξg]. In particular, if {f, g} = 0, then [Ξf , Ξg] = 0, thus the Hamiltonian flows of f and g commute. Theorem 2.5. Let (M, ω) be a symplectic manifold. (1) The Poisson bracket {·, ·} is a Lie algebra structure on C ∞(M). (2) The map C ∞(M) → VectHam(M, ω), f 7→ Ξf is a Lie algebra anti-homomorphism. Proof. (1) Obviously {·, ·} is bilinear and anti-symmetric. To show it is a Lie algebra structure it remains to check the Jacobi identity: 0 = (LΞfω)(Ξg, Ξh) = (dιΞf )(Ξg, Ξh) = Ξg(ω(Ξf , Ξh)) − Ξh(ω(Ξf , Ξg)) − ω(Ξf , [Ξg, Ξh]) = {{f, h}, g} − {{f, g}, h} + ω(Ξf , Ξ{g,h}) = {{f, h}, g} + {{g, f}, h} + {{h, g}, f}.
6LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMS口(2)This is just another explanation of (6) Poisson manifolds.Another remarkablefact about thePoisson bracket on Coo(M)is that it satisfiesthe Leibniz law:Proposition2.6.Foranyf,g,hECo(M),(7)[fg, h] = (f,h]g + f(g,h]Proof. We have(fg,h)=三h(fg)=三h(f)g+f三h(g)=(f,hg+f(g,h)口Definition 2.7.(1)APoisson algebra (P,f,)is an associativealgebraPwitha Lie bracket f, such that the Leibniz law (7)holds.(2) A Poisson manifold is a manifold M equipped with a Poisson algebra struc-ture on Co(M).So any symplectic manifold M is a Poisson manifold. But we have more Poissonmanifolds thansymplecticmanifolds.Infact,anysmoothmanifold M is aPoissonmanifold if we equipped Co(M)with thetrivial bracket.Butthis is of coursenotinterestingErample. Let g be any Lie algebra of dimension n, and g* its dual. Then one cancanonically identify the double dual (g*)* with g.We thus get, for any f e Coo(g*)and any μ E g*, an identification of dμf with an element in g viadμf ETig*= (Tμg*)*= (g*)*= g.Definea bracket structure ong*via(f,g)(μ) =μ([dμf,dμg]),One can check that this is in fact a Poisson bracket structure.Let M be a Poisson manifold. According to the Leibniz law, for any h E Coo(M),the map(,h) : C(M) →Ris aderivative on Co(M),and thus defines auniquevector field 三h onM so thatfor any f,三r(f) = (f,h].As in the symplectic case, one calls 三n the Hamiltonian vector field of h.Poisson manfilds are closely related to symplectic manifolds. One can consult,for example, the book “Lectures on the Geometry of Poisson Manifolds (Progress inMathematics)"by I. Vaisman
6 LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS (2) This is just another explanation of (6). ¶ Poisson manifolds. Another remarkable fact about the Poisson bracket on C ∞(M) is that it satisfies the Leibniz law: Proposition 2.6. For any f, g, h ∈ C ∞(M), (7) {fg, h} = {f, h}g + f{g, h}. Proof. We have {fg, h} = Ξh(fg) = Ξh(f)g + fΞh(g) = {f, h}g + f{g, h}. Definition 2.7. (1) A Poisson algebra (P, {·, ·}) is an associative algebra P with a Lie bracket {·, ·} such that the Leibniz law (7) holds. (2) A Poisson manifold is a manifold M equipped with a Poisson algebra structure on C ∞(M). So any symplectic manifold M is a Poisson manifold. But we have more Poisson manifolds than symplectic manifolds. In fact, any smooth manifold M is a Poisson manifold if we equipped C ∞(M) with the trivial bracket. But this is of course not interesting. Example. Let g be any Lie algebra of dimension n, and g ∗ its dual. Then one can canonically identify the double dual (g ∗ ) ∗ with g. We thus get, for any f ∈ C ∞(g ∗ ) and any µ ∈ g ∗ , an identification of dµf with an element in g via dµf ∈ T ∗ µg ∗ = (Tµg ∗ ) ∗ = (g ∗ ) ∗ = g. Define a bracket structure on g ∗ via {f, g}(µ) = µ([dµf, dµg]). One can check that this is in fact a Poisson bracket structure. Let M be a Poisson manifold. According to the Leibniz law, for any h ∈ C ∞(M), the map {·, h} : C ∞(M) → R is a derivative on C ∞(M), and thus defines a unique vector field Ξh on M so that for any f, Ξh(f) = {f, h}. As in the symplectic case, one calls Ξh the Hamiltonian vector field of h. Poisson manfilds are closely related to symplectic manifolds. One can consult, for example, the book “Lectures on the Geometry of Poisson Manifolds (Progress in Mathematics)” by I. Vaisman.
LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMS73.COMPLETELYINTEGRABLEHAMILTONIANSYSTEMS Hamiltonian system.IntheHamiltonianfashionofclassicalmachanics,aclassicalmechanical systemis described via a symplectic manifold (M,w), called the phase space of the system.Any point in M is called a state of the system.Theevolution of the system isthen theHamiltonianflowon M of aHamiltonianfunction,and theevolution of aparticular state is then given by the integral curve of the Hamiltonian vector field.Any smoothfunction onM iscalled a classical observable.Erample. Consider a free particle moving under a force field in Rn (the configurationspace).For simplicity we assume the mass of the particle is 1. Denote by V(r) thepotential energy function of the force field, and denote by (ri(t),, n(t)) the positionvector of the particle at time t.According tothe Newton's law,the movement oftheparticleobeytheequationi(t) = -VV.Note that in this system the energy function E = /i(t)2 + V(r(t)) of the particleisa conserved quantity.To describe this system using the language of Hamiltonian mechanics, we intro-ducethephasespace R2n,with coordinates &i,..,&n,Si,..,Sn,where (ci(t),..*,an(t))is still the position vector of the particle in the configuration space,and(Si(t),*,Sn(t)) = (ti(t),.,n(t)is the momentum vector. So in particular the energy function becomes a Hamiltonianfunction on R2n:H(r,) =IsIP + V(2),As we have seen above, the integral curves of H satisfies the Hamiltonian equationaHaav=s(t), Et)=i(t) :(t)afOriOxiThis system of equations is of course equivalent to the Newtons's equation = -VV.Definition 3.1. A Hamiltonian system is a triple (M,w, H), where (M,w) is a sym-plectic manifold and H e Co(M) a smooth function (usually called the Hamiltonianof the system).Erample.Consider the1 dimensional Harmonic oscillar.The phase space is R? withcoordinates , s, where is the position (the signed distance from the stationaryposition) and is the momentum. For simplicity we again assume mass-1 and alsoassume the spring constant =1.The Hamiltonian is theenergy function.2H(r,s) =22
LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS 7 3. Completely integrable Hamiltonian systems ¶ Hamiltonian system. In the Hamiltonian fashion of classical machanics, a classical mechanical system is described via a symplectic manifold (M, ω), called the phase space of the system. Any point in M is called a state of the system. The evolution of the system is then the Hamiltonian flow on M of a Hamiltonian function, and the evolution of a particular state is then given by the integral curve of the Hamiltonian vector field. Any smooth function on M is called a classical observable. Example. Consider a free particle moving under a force field in R n (the configuration space). For simplicity we assume the mass of the particle is 1. Denote by V (x) the potential energy function of the force field, and denote by (x1(t), ·, xn(t)) the position vector of the particle at time t. According to the Newton’s law, the movement of the particle obey the equation x¨(t) = −∇V. Note that in this system the energy function E = 1 2 |x˙(t)| 2 + V (x(t)) of the particle is a conserved quantity. To describe this system using the language of Hamiltonian mechanics, we introduce the phase space R 2n , with coordinates x1, · · · , xn, ξ1, · · · , ξn, where (x1(t), · · · , xn(t)) is still the position vector of the particle in the configuration space, and (ξ1(t), · · · , ξn(t)) = ( ˙x1(t), · · · , x˙ n(t)) is the momentum vector. So in particular the energy function becomes a Hamiltonian function on R 2n : H(x, ξ) = 1 2 |ξ| 2 + V (x). As we have seen above, the integral curves of H satisfies the Hamiltonian equation x˙(t) = ∂H ∂ξi = ξi(t), ˙ξ(t) = − ∂H ∂xi = − ∂V ∂xi (t). This system of equations is of course equivalent to the Newtons’s equation ¨x = −∇V . Definition 3.1. A Hamiltonian system is a triple (M, ω, H), where (M, ω) is a symplectic manifold and H ∈ C ∞(M) a smooth function (usually called the Hamiltonian of the system). Example. Consider the 1 dimensional Harmonic oscillar. The phase space is R 2 with coordinates x, ξ, where x is the position (the signed distance from the stationary position) and ξ is the momentum. For simplicity we again assume mass=1 and also assume the spring constant =1. The Hamiltonian is the energy function H(x, ξ) = 1 2 ξ 2 + 1 2 x 2
8LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMSThe Hamilton's equations becomes(t) =s(t), E(t) = -r(t).Given initial condition r(O) = ro,E(O) = Eo, one can easily solve the system ofequations to getr(t) = cos(t)ro + sin(t)so, s(t) = cos(t)so -sin(t)ro.The Hamilton flow in the phase space is just the counterclockwise rotations on R?From the above example we see that the conserved quantity is important to aHamiltonian system.Erample.Thebilliard system in a planar elliptical region.Last time wehave seenthat the phase space of this system is C × (-1, 1), where C is the boundary ellipse:r?y?22+62More precisely, a state is a pair (p,t), where p is the position in C where thebilliard hit the boundary, and t the the length tangential projection of the unitvector describing the direction that the billiard moves after hitting the boundaryThe evolution of the system is a symplectomorphism which maps (p,t) to (p',t'),where (p',t') is the state describing the next hit of the billiard with the boundary.We have seen that the real trajectory of the billiard will always tangent to someconfocal ellipse (or confocal hyperbola) whose equation is2292a2-2+-=1.Then function Z is invariant under the symplectomorphism and plays the role ofthe Hamiltonian function for this system.Now suppose we have a Hamiltonian system (M,w, H). Let a E Co(M) be anyclasscial observable. Leta(t) = a(exp(t三)(po)be the evolution of the observable, where po is the initial state of the system.Proposition 3.2.a(t)=a,H)(exp(t=H)(po))Proof. Sinceexp(t三)(p)=三(exp(t三H)(p)),we havea(t)=da(三H)(exp(t三H)(po)) = (a,H)(exp(t三H)(po))口
8 LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS The Hamilton’s equations becomes x˙(t) = ξ(t), ˙ξ(t) = −x(t). Given initial condition x(0) = x0, ξ(0) = ξ0, one can easily solve the system of equations to get x(t) = cos(t)x0 + sin(t)ξ0, ξ(t) = cos(t)ξ0 − sin(t)x0. The Hamilton flow in the phase space is just the counterclockwise rotations on R 2 . From the above example we see that the conserved quantity is important to a Hamiltonian system. Example. The billiard system in a planar elliptical region. Last time we have seen that the phase space of this system is C × (−1, 1), where C is the boundary ellipse: x 2 a 2 + y 2 b 2 = 1. More precisely, a state is a pair (p, t), where p is the position in C where the billiard hit the boundary, and t the the length tangential projection of the unit vector describing the direction that the billiard moves after hitting the boundary. The evolution of the system is a symplectomorphism which maps (p, t) to (p 0 , t0 ), where (p 0 , t0 ) is the state describing the next hit of the billiard with the boundary. We have seen that the real trajectory of the billiard will always tangent to some confocal ellipse (or confocal hyperbola) whose equation is x 2 a 2 − Z2 + y 2 b 2 − Z2 = 1. Then function Z is invariant under the symplectomorphism and plays the role of the Hamiltonian function for this system. Now suppose we have a Hamiltonian system (M, ω, H). Let a ∈ C ∞(M) be any classcial observable. Let a(t) = a(exp(tΞH)(p0)) be the evolution of the observable, where p0 is the initial state of the system. Proposition 3.2. a˙(t) = {a, H}(exp(tΞH)(p0)). Proof. Since d dt exp(tΞH)(p) = ΞH(exp(tΞH)(p)), we have a˙(t) = da(ΞH)(exp(tΞH)(p0)) = {a, H}(exp(tΞH)(p0)).
LECTURE6:GEOMETRYOFHAMILTONIANSYSTEMS9Completelyintegrable systems.Let (M,w, H) be a Hamiltonain system.Recall that (f,H) = o if and only if fis constant along the integral curves of EH. Such a function is called an integral ofmotion (or a first integral). In general one cannot hope that a Hamiltonian systemadmits any integral of motion that is independent of the Hamiltonian function itself.Definition 3.3. A set of functions fi,..., f, on M are called independent if thedifferentials dfi,..,df, are linarly independent in an open dense subset of M.Weare interested inHamiltonian systems with manyPoissoncommuting integralof motions.Now suppose fi =H,.:,fare independent integral of motions of(M,w,H) that are Poisson commuting, i.e.fi,fi}=0,Vl≤i,j,≤k.It follows that for any i,j,w(三f,三f,)= 0.So at a point p where df's are linearly independent, the vectors 三(p),...,三 (p)span an isotropic subspace of T,M. In particular, we getProposition 3.4.If fi,...,fu are independent Poisson commuting integral of mo-tions for(M,w,H),thenk≤n.Definition 3.5. A Hamiltonian system (M,w, H) is called completely integrable ifitadmitsn=dimM independent Poisson commuting integral of motions fiH,f2,...,fn.Of course any 2 dimsensional Hamiltonian system (M,w,H) (with dH ± 0almost everywhere)is completelyintegrabel.Inparticular,the1 dimensionalHar-monic oscillator,the billiard system in theplanar ellipse are completely integrableHamiltonian systems.There exists more complicated examples that we shall discusslater.Now let (M,w,H)be a completely integrable Hamiltonian system, and fiH, f2, ..., fn be independent Poisson commuting integral of motions. Let c e Rn bea regular value of the map F = (fi,..-, fn) : M → IRn. Then the above argumentshows that F-1(c) is a Lagrangian submanifold of M. Ation-angle coordinates.Student presentation
LECTURE 6: GEOMETRY OF HAMILTONIAN SYSTEMS 9 ¶ Completely integrable systems. Let (M, ω, H) be a Hamiltonain system. Recall that {f, H} = 0 if and only if f is constant along the integral curves of ΞH. Such a function is called an integral of motion (or a first integral). In general one cannot hope that a Hamiltonian system admits any integral of motion that is independent of the Hamiltonian function itself. Definition 3.3. A set of functions f1, · · · , fk on M are called independent if the differentials df1, · · · , dfp are linarly independent in an open dense subset of M. We are interested in Hamiltonian systems with many Poisson commuting integral of motions. Now suppose f1 = H, · · · , fk are independent integral of motions of (M, ω, H) that are Poisson commuting, i.e. {fi , fj} = 0, ∀1 ≤ i, j, ≤ k. It follows that for any i, j, ω(Ξfi , Ξfj ) = 0. So at a point p where dfi ’s are linearly independent, the vectors Ξf1 (p), · · · , Ξfk (p) span an isotropic subspace of TpM. In particular, we get Proposition 3.4. If f1, · · · , fk are independent Poisson commuting integral of motions for (M, ω, H), then k ≤ n. Definition 3.5. A Hamiltonian system (M, ω, H) is called completely integrable if it admits n = 1 2 dim M independent Poisson commuting integral of motions f1 = H, f2, · · · , fn. Of course any 2 dimsensional Hamiltonian system (M, ω, H) (with dH 6= 0 almost everywhere) is completely integrabel. In particular, the 1 dimensional Harmonic oscillator, the billiard system in the planar ellipse are completely integrable Hamiltonian systems. There exists more complicated examples that we shall discuss later. Now let (M, ω, H) be a completely integrable Hamiltonian system, and f1 = H, f2, · · · , fn be independent Poisson commuting integral of motions. Let c ∈ R n be a regular value of the map F = (f1, · · · , fn) : M → R n . Then the above argument shows that F −1 (c) is a Lagrangian submanifold of M. ¶ Ation-angle coordinates. Student presentation