Introduction
Degenerate Lagrangians and Hamiltonian constraints lead naturally to geometric descriptions in which the relevant closed two-form has a nontrivial kernel. The foundations of this approach were developed by Dirac and by Anderson and Bergmann [15 , 3 , 20 ] . Symmetries, their generators, and their relation to constraints were subsequently studied in detail within generalized Hamiltonian dynamics [28 , 26 ] . These results provide the background for the present work. Our purpose is to study the consequences of specifying additional geometric data on the kernel and on a complementary distribution, as occurs in q q q -cosymplectic and q q q -contact geometry.
On a q q q -cosymplectic manifold, the closed two-form Ω \Omega Ω defines a presymplectic structure with q q q -dimensional kernel R \mathcal{R} R . The closed one-forms λ 1 , … , λ q \lambda_1,\ldots,\lambda_q λ 1 , … , λ q determine a complementary distribution ξ = ⋂ i ker λ i \xi=\bigcap_i\ker\lambda_i ξ = ⋂ i ker λ i and a distinguished frame R 1 , … , R q R_1,\ldots,R_q R 1 , … , R q of R \mathcal{R} R , normalized by λ i ( R j ) = δ i j \lambda_i(R_j)=\delta_{ij} λ i ( R j ) = δ ij . Thus the structure includes a choice of transverse directions and a normalization of the characteristic directions. In the q q q -contact case, each λ i \lambda_i λ i is a precontact form, with characteristic distribution ker λ i ∩ ker d λ i \ker\lambda_i\cap\ker d\lambda_i ker λ i ∩ ker d λ i . The forms are related by the requirement that all d λ i d\lambda_i d λ i have the same kernel R \mathcal{R} R and are non-degenerate on the common complement ξ \xi ξ . Uniformity imposes the further condition d λ 1 = ⋯ = d λ q d\lambda_1=\cdots=d\lambda_q d λ 1 = ⋯ = d λ q .
These additional data affect the transformation theory. Preserving a presymplectic form preserves its kernel, but need not preserve a chosen complementary distribution or the individual Reeb fields. For a Hamiltonian constant along the kernel, the choice of ξ \xi ξ selects a unique horizontal solution of the presymplectic Hamiltonian equation. Similarly, preserving one precontact form does not in general preserve an entire q q q -contact structure. We therefore formulate canonical transformations, Hamiltonian vector fields, and canonoid transformations with explicit reference to the defining one-forms. The physical examples below explain how such data can arise and why their local and global properties require separate consideration.
Geometric and dynamical background
Two earlier constructions are directly relevant to the present work. Dubrovin, Giordano, Marmo and Simoni [16 ] construct Poisson brackets from a presymplectic form and a flat generalized connection. The q q q -cosymplectic bracket used here is obtained by taking the horizontal projector determined by the closed one-forms λ i \lambda_i λ i ; see Remark 2.3. Alekseevsky, Grabowski, Marmo and Michor [1 ] describe symplectic and Poisson structures on cotangent bundles of Lie groups and principal bundles in terms of momentum mappings, Maurer–Cartan forms, and connections. Their principal-bundle construction provides a geometric framework for the Cartan forms in our monopole examples; see Remark 3.3. These works supply the underlying bracket and bundle constructions. Our transformation theory keeps explicit the additional ordered one-forms and the normalized Reeb fields, and imposes compatibility with these data.
Canonical transformations are standard tools for comparing Hamiltonian descriptions [9 , 11 ] . Canonoid transformations preserve the Hamiltonian form of a specified dynamics while allowing its Hamiltonian description to change; their study in symplectic geometry includes [14 , 30 , 12 , 32 ] . Related constructions in cosymplectic, locally conformal symplectic, contact, and cocontact geometry connect transformation theory with symmetries, first integrals, and dissipation laws [4 , 5 , 6 , 31 , 36 ] .
Structures defined by several one-forms have been studied in connection with integrability, thermodynamical models, and Hamiltonian dynamics [17 , 23 ] . The q q q -cosymplectic structures considered here are those of [23 , 24 ] , while the q q q -contact setting is developed in [2 , 21 , 22 , 25 , 37 ] . Integrability, reduction, and Noether-type results are already available in these settings [23 , 24 , 25 , 37 , 18 ] . A broader framework allowing several two-forms with a common kernel is discussed in [18 ] . We use this framework when the forms arising from separate magnetic contributions must be distinguished from the form governing the total dynamics.
A related classical construction replaces the one-parameter group of time translations by a Lie group G G G . This idea was already considered by J. Schwinger in his lectures on quantum field theory [33 ] . Hamilton–Jacobi theory in this setting was studied by Mukunda [29 ] and by Marmo, Morandi, and Mukunda [27 ] . For a Hamiltonian action of the Abelian vector group R q \mathbb{R}^{q} R q with an equivariant momentum map, a regular momentum level is presymplectic, with kernel generated by the independent infinitesimal action fields. Equipping this level with closed one-forms whose restrictions to the kernel form a coframe gives a q q q -cosymplectic structure. The existence of such forms globally is an additional requirement. This construction relates the present geometry to the classical theory of Hamiltonian group actions. Our transformation results concern the Abelian setting: the Reeb fields commute and generate a local R q \mathbb{R}^{q} R q -action. Non-Abelian extensions require different structure equations and will not be dealt with in this paper.
Physical examples and their interpretation
Relativistic constraint dynamics. The no-interaction theorem [13 ] exhibits the restrictions imposed by the standard canonical world-line assumptions in a single-time relativistic particle description. Constraint formulations provide a way to accommodate interacting particles while retaining relativistic covariance [20 , 8 ] . We consider N N N independent generalized mass-shell constraints on T ∗ ( ( R 4 ) N ) T^{*}((\mathbb{R}^{4})^{N}) T ∗ (( R 4 ) N ) and restrict to a regular region in which these constraints Poisson-commute. This commutativity is a stronger assumption than the first-class property alone. The constraint surface has dimension 7 N 7N 7 N , and the restricted symplectic form has a kernel of rank N N N .
Functions χ 1 , … , χ N \chi_{1},\ldots,\chi_{N} χ 1 , … , χ N whose differentials are independent on this kernel determine an N N N -cosymplectic structure wherever the corresponding transversality matrix is invertible. Its Reeb fields are normalized linear combinations of the constraint fields. Their parameters describe directions along constraint orbits. In the free-particle example, our normalization identifies these parameters with constant multiples of the individual proper times. For interacting systems, an interpretation in terms of proper time requires a separate analysis of the chosen constraints and clock functions.
A charge in magnetic monopole fields.* Magnetic monopoles provide a setting in which the topology of the configuration space affects the existence of a global potential [19 , 7 , 35 ] . For a point monopole, the magnetic two-form on Q = R 3 ∖ { 0 } Q = \mathbb{R}^{3}\setminus\{0\} Q = R 3 ∖ { 0 } is closed but not exact. It therefore admits no global vector potential on Q Q Q and no global magnetic Lagrangian of the usual potential form on T Q TQ T Q . A global Lagrangian can instead be defined on T P TP T P , where P → Q P \to Q P → Q is a principal U ( 1 ) U(1) U ( 1 ) -bundle. Its Legendre image is a constraint submanifold of T ∗ P T^{*}P T ∗ P , on which a normalized Cartan form is contact and its Reeb field generates the gauge action [7 , 35 ] . The particle still evolves with one physical time parameter. Gauge-invariant observables descend to the magnetic phase space T Q TQ T Q , and the lifted Hamiltonian dynamics projects to the Lorentz force equation.
For two monopoles, the corresponding construction uses a principal U ( 1 ) × U ( 1 ) U(1)\times U(1) U ( 1 ) × U ( 1 ) -bundle and produces an eight-dimensional constraint manifold. The normalized forms associated with the separate monopoles define a non-uniform 2 2 2 -contact structure. The total Cartan form, which contains both magnetic contributions, determines the physical dynamics. There are two gauge directions and one physical time parameter.
Two different global issues occur in this example. If the magnetic charges have an irrational ratio, the one-parameter subgroup along which the total Cartan form would descend has dense orbits in the gauge tori. Its orbit space does not give the smooth seven-dimensional contact manifold obtained in the rational case [35 ] . The quotient by the full gauge torus remains the magnetic phase space. Independently of this irrationality condition, we prove that the topology of the monopole bundles prevents a uniform 2 2 2 -contact structure whose Reeb fields are the prescribed gauge generators. The example therefore distinguishes the obstruction to an intermediate contact quotient from the obstruction to uniformity.
Results and scope
We develop the transformation theory for q q q -cosymplectic and uniform q q q -contact structures, keeping the additional one-forms explicit. The local normal forms and Noether-type statements are presented as adaptations of established geometric methods to these specified data. The main results concern the compatibility conditions for transformations, their infinitesimal generators, and the geometric restrictions exhibited by the physical examples.
In the q q q -cosymplectic case, canonical transformations preserve both Ω \Omega Ω and each λ i \lambda_i λ i . We distinguish this requirement from preservation of the Reeb frame and the complementary distribution; see Proposition 4.2. We also characterize horizontal generators of generalized canonical transformations by locally Hamiltonian vector fields. For a Hamiltonian H H H , we consider the family E H a ⃗ = X H + ∑ i a i R i E_H^{\vec{a}}=X_H+\sum_i a_iR_i E H a = X H + ∑ i a i R i , where a ⃗ ∈ R q \vec{a}\in\mathbb{R}^q a ∈ R q . The Noether theorem relates horizontal infinitesimal symmetries to functions conserved by every member of this family. These functions are invariant under each Reeb field as well as under X H X_H X H . This simultaneous conservation condition does not require the different fields E H a ⃗ E_H^{\vec{a}} E H a to commute.
In the uniform q q q -contact case, a symmetry preserving H H H and all the defining one-forms gives a tuple ( λ 1 ( V ) , … , λ q ( V ) ) (\lambda_1(V),\ldots,\lambda_q(V)) ( λ 1 ( V ) , … , λ q ( V )) satisfying a common dissipation law. Uniformity also imposes a restriction: for such strict symmetries, the differences λ i ( V ) − λ j ( V ) \lambda_i(V)-\lambda_j(V) λ i ( V ) − λ j ( V ) are locally constant. We retain the tuple formulation to express compatibility with the whole family of one-forms.
For both structures, let X X X generate a one-parameter group of canonoid transformations with smoothly varying Hamiltonians K s K_s K s , normalized by K 0 = H K_0=H K 0 = H . Writing κ = d d s ∣ s = 0 K s \kappa=\left.\frac{d}{ds}\right|_{s=0}K_s κ = d s d s = 0 K s , we prove the necessary infinitesimal relation [ X , X H ] = X X ( H ) − κ [X,X_H]=X_{X(H)-\kappa} [ X , X H ] = X X ( H ) − κ . The equations determining κ \kappa κ depend on the geometric structure and, in the uniform q q q -contact case, must hold simultaneously for all defining one-forms. We also give an explicit class of canonoid flows valid for every Hamiltonian: those generated by vector fields satisfying L X λ i = λ i \mathcal{L}_X\lambda_i=\lambda_i L X λ i = λ i for all i i i .
The examples recover familiar conservation laws within these descriptions: the total angular momentum of free relativistic particles and the Poincaré angular momentum of a charge in a monopole field. The two-monopole example has a different role. Its non-uniform structure and the obstruction just described identify a physical situation beyond the uniform transformation theory developed here.
Organization of the paper
Section 2 recalls the geometric structures, establishes local normal forms, and describes the Jacobi structure associated with a uniform q q q -contact manifold. Section 3 develops the constraint and monopole constructions. It also gives a local passage from q q q -cosymplectic to uniform q q q -contact structures under a normalization of a primitive of Ω \Omega Ω , together with a global realization for free relativistic particles. Sections 4 and 5 contain the canonical and canonoid transformation results and their relations to symmetries and conservation or dissipation laws. Section 6 discusses the extension to non-uniform structures, the integration of infinitesimal conditions, and further questions suggested by the examples.
Preliminaries
All manifolds, differential forms, and vector fields are smooth and real. We use the contraction convention ( ι X Ω ) ( Y ) = Ω ( X , Y ) (\iota_X\Omega)(Y)=\Omega(X,Y) ( ι X Ω ) ( Y ) = Ω ( X , Y ) . The purpose of this section is to fix the geometric data and sign conventions used below, and to explain their relation to presymplectic and precontact geometry. We include proofs of the local normal forms and basic Hamiltonian identities for completeness.
A common geometric setting
Definition 2.1 (q q q -structured manifolds; cf. [18 ] ). Let n , q , m n,q,m n , q , m be positive integers and let M M M have dimension 2 n + q 2n+q 2 n + q . A q q q -structure on M M M consists of one-forms λ ⃗ = ( λ 1 , … , λ q ) \vec{\lambda}=(\lambda_1,\ldots,\lambda_q) λ = ( λ 1 , … , λ q ) , two-forms Ω ⃗ = ( Ω 1 , … , Ω m ) \vec{\Omega}=(\Omega_1,\ldots,\Omega_m) Ω = ( Ω 1 , … , Ω m ) , and a splitting T M = R ⊕ ξ TM=\mathcal{R}\oplus\xi T M = R ⊕ ξ such that:
(i) the λ i \lambda_i λ i are pointwise linearly independent and ξ = ⋂ i = 1 q ker λ i \xi=\bigcap_{i=1}^{q}\ker\lambda_i ξ = ⋂ i = 1 q ker λ i ;
(ii) each restriction Ω j ∣ ξ \Omega_j|_{\xi} Ω j ∣ ξ is non-degenerate;
(iii) ker Ω j = R \ker\Omega_j=\mathcal{R} ker Ω j = R for every j = 1 , … , m j=1,\ldots,m j = 1 , … , m .
A manifold endowed with these data is called a q q q -structured manifold. We use this terminology to collect the algebraic data shared by the structures considered below. No closedness condition is imposed at this stage. The case m = q m=q m = q is the framework of [18 ] ; here the number of two-forms is allowed to differ from the number of one-forms. This is useful in Proposition 3.3, where d λ 1 \mathrm{d}\lambda_1 d λ 1 , d λ 2 \mathrm{d}\lambda_2 d λ 2 , and the physically relevant form d θ L \mathrm{d}\theta_L d θ L occur together, so that q = 2 q=2 q = 2 and m = 3 m=3 m = 3 . We write ( M , λ ⃗ ⊕ Ω ⃗ , R ⊕ ξ ) (M,\vec{\lambda}\oplus\vec{\Omega},\mathcal{R}\oplus\xi) ( M , λ ⊕ Ω , R ⊕ ξ ) for these data. The bundles R \mathcal{R} R and ξ \xi ξ have ranks q q q and 2 n 2n 2 n , respectively, and are called the Reeb distribution and the horizontal distribution . The nowhere-vanishing q q q -form λ 1 ∧ ⋯ ∧ λ q \lambda_1\wedge\cdots\wedge\lambda_q λ 1 ∧ ⋯ ∧ λ q is the characteristic form .
Remark 2.1. Every Ω j \Omega_j Ω j has rank 2 n 2n 2 n . Its restriction to ξ \xi ξ makes ξ \xi ξ a symplectic vector bundle, a statement of fibrewise linear algebra that does not require d Ω j = 0 \mathrm{d}\Omega_j=0 d Ω j = 0 . Non-degeneracy of the individual restrictions does not imply non-vanishing of their mixed wedge products. For example, on R 4 \mathbb{R}^4 R 4 , let a = d x 1 ∧ d y 1 a=\mathrm{d}x_1\wedge\mathrm{d}y_1 a = d x 1 ∧ d y 1 and b = d x 2 ∧ d y 2 b=\mathrm{d}x_2\wedge\mathrm{d}y_2 b = d x 2 ∧ d y 2 . Both a + b a+b a + b and a − b a-b a − b are non-degenerate, whereas ( a + b ) ∧ ( a − b ) = 0 (a+b)\wedge(a-b)=0 ( a + b ) ∧ ( a − b ) = 0 .
Proposition 2.1. For each j = 1 , … , m j=1,\ldots,m j = 1 , … , m , the vector bundle map
♭ j : T M ⟶ T ∗ M , ♭ j ( X ) = ι X Ω j + ∑ k = 1 q λ k ( X ) λ k \flat^j:TM\longrightarrow T^*M,\qquad\flat^j(X)=\iota_X\Omega_j+\sum_{k=1}^{q}\lambda_k(X)\lambda_k ♭ j : T M ⟶ T ∗ M , ♭ j ( X ) = ι X Ω j + k = 1 ∑ q λ k ( X ) λ k is an isomorphism. It therefore induces an isomorphism of C ∞ ( M ) C^\infty(M) C ∞ ( M ) -modules between vector fields and one-forms.
Proof. Fix x ∈ M x\in M x ∈ M . If ♭ x j ( v ) = 0 \flat^j_x(v)=0 ♭ x j ( v ) = 0 , evaluation on v v v gives 0 = ∑ k = 1 q λ k ( v ) 2 0=\sum_{k=1}^{q}\lambda_k(v)^2 0 = ∑ k = 1 q λ k ( v ) 2 . Thus v ∈ ξ x v\in\xi_x v ∈ ξ x and ι v Ω j = 0 \iota_v\Omega_j=0 ι v Ω j = 0 . Non-degeneracy of Ω j ∣ ξ x \Omega_j|_{\xi_x} Ω j ∣ ξ x implies v = 0 v=0 v = 0 . The domain and codomain have the same dimension, so ♭ x j \flat^j_x ♭ x j is an isomorphism. These maps depend smoothly on x x x , and their inverses are smooth.
Corollary 2.1. There are unique vector fields R 1 , … , R q R_1,\ldots,R_q R 1 , … , R q satisfying
ι R k Ω j = 0 , λ l ( R k ) = δ l k , j = 1 , … , m , k , l = 1 , … , q . \iota_{R_k}\Omega_j=0,\qquad\lambda_l(R_k)=\delta_{lk},\qquad j=1,\ldots,m,\quad k,l=1,\ldots,q. ι R k Ω j = 0 , λ l ( R k ) = δ l k , j = 1 , … , m , k , l = 1 , … , q . They form a global frame of R \mathcal{R} R and are called the Reeb vector fields. Moreover, ( ♭ j ) − 1 ( λ k ) = R k (\flat^j)^{-1}(\lambda_k)=R_k ( ♭ j ) − 1 ( λ k ) = R k for every j j j .
Proof. At each point, consider the map
R x ⟶ R q , v ⟼ ( λ 1 ( v ) , … , λ q ( v ) ) . \mathcal{R}_x\longrightarrow\mathbb{R}^q,\qquad v\longmapsto\left(\lambda_1(v),\ldots,\lambda_q(v)\right). R x ⟶ R q , v ⟼ ( λ 1 ( v ) , … , λ q ( v ) ) . Its kernel is R x ∩ ξ x = { 0 } \mathcal{R}_x\cap\xi_x=\{0\} R x ∩ ξ x = { 0 } . Since both spaces have dimension q q q , this map is an isomorphism. The inverse images of the standard basis of R q \mathbb{R}^q R q define the required smooth fields. They belong to every ker Ω j \ker\Omega_j ker Ω j and satisfy the normalization conditions. These conditions also prove uniqueness. Finally, ♭ j ( R k ) = λ k \flat^j(R_k)=\lambda_k ♭ j ( R k ) = λ k .
In particular, the chosen forms determine the projections associated with the splitting:
pr R X = ∑ i = 1 q λ i ( X ) R i , pr ξ X = X − ∑ i = 1 q λ i ( X ) R i . \operatorname{pr}_{\mathcal{R}}X=\sum_{i=1}^{q}\lambda_i(X)R_i,\qquad\operatorname{pr}_{\xi}X=X-\sum_{i=1}^{q}\lambda_i(X)R_i. pr R X = i = 1 ∑ q λ i ( X ) R i , pr ξ X = X − i = 1 ∑ q λ i ( X ) R i .
At this level of generality, the Reeb fields need not commute. Their commutativity below follows from the additional differential conditions.
Remark 2.2. When all the Ω j \Omega_j Ω j coincide with a form Ω \Omega Ω , the data define an almost cosymplectic structure of order q q q in the terminology of [17 ] . If d Ω = 0 \mathrm{d}\Omega=0 d Ω = 0 , the structure is partially cosymplectic of order q q q . If the λ i \lambda_i λ i are also closed, we use the term q q q -cosymplectic, following [23 ] . When m = q m=q m = q and Ω i = d λ i \Omega_i=\mathrm{d}\lambda_i Ω i = d λ i , the data define a q q q -contact structure [2 , 21 ] . These cases have different integrability properties and will be treated separately.
q-cosymplectic manifolds
Definition 2.2 (q-cosymplectic manifolds [23 ] ). A q q q -cosymplectic structure on a manifold M 2 n + q M^{2n+q} M 2 n + q consists of a closed two-form Ω \Omega Ω of rank 2 n 2n 2 n and closed one-forms λ ⃗ = ( λ 1 , … , λ q ) \vec{\lambda}=(\lambda_1,\ldots,\lambda_q) λ = ( λ 1 , … , λ q ) such that
Ω n ∧ λ 1 ∧ ⋯ ∧ λ q ≠ 0 \Omega^n\wedge\lambda_1\wedge\cdots\wedge\lambda_q\ne0 Ω n ∧ λ 1 ∧ ⋯ ∧ λ q = 0 at every point. Equivalently, with R = ker Ω \mathcal{R}=\ker\Omega R = ker Ω and ξ = ⋂ i ker λ i \xi=\bigcap_i\ker\lambda_i ξ = ⋂ i ker λ i , one has T M = R ⊕ ξ TM=\mathcal{R}\oplus\xi T M = R ⊕ ξ and Ω ∣ ξ \Omega|_{\xi} Ω ∣ ξ is non-degenerate. We denote this structure by ( M , Ω , λ ⃗ ) (M,\Omega,\vec{\lambda}) ( M , Ω , λ ) .
The rank assumption in Definition 2.2 is part of the definition. Together with the volume condition, it says that the restrictions of the λ i \lambda_i λ i to ker Ω \ker\Omega ker Ω form a coframe. Thus ( M , Ω ) (M,\Omega) ( M , Ω ) is presymplectic, and the one-forms specify both a complementary distribution and a normalized frame of the kernel. The splitting and the Reeb fields follow from these data by Corollary 2.1.
Proposition 2.2. On a q q q -cosymplectic manifold, the Reeb fields satisfy
[ R i , R j ] = 0 , L R i Ω = 0 , L R i λ j = 0 [R_i,R_j]=0,\qquad\mathcal{L}_{R_i}\Omega=0,\qquad\mathcal{L}_{R_i}\lambda_j=0 [ R i , R j ] = 0 , L R i Ω = 0 , L R i λ j = 0 for all i , j i,j i , j . Moreover, ξ \xi ξ is integrable and Ω \Omega Ω restricts to a symplectic form on each leaf of ξ \xi ξ .
Proof. Cartan’s formula gives L R i Ω = 0 \mathcal{L}_{R_i}\Omega=0 L R i Ω = 0 and L R i λ j = d ( λ j ( R i ) ) = 0 \mathcal{L}_{R_i}\lambda_j=\mathrm{d}(\lambda_j(R_i))=0 L R i λ j = d ( λ j ( R i )) = 0 . Consequently,
ι [ R i , R j ] Ω = L R i ( ι R j Ω ) − ι R j ( L R i Ω ) = 0 , \iota_{[R_i,R_j]}\Omega=\mathcal{L}_{R_i}(\iota_{R_j}\Omega)-\iota_{R_j}(\mathcal{L}_{R_i}\Omega)=0, ι [ R i , R j ] Ω = L R i ( ι R j Ω ) − ι R j ( L R i Ω ) = 0 , while λ k ( [ R i , R j ] ) = R i ( λ k ( R j ) ) − ( L R i λ k ) ( R j ) = 0 \lambda_k([R_i,R_j])=R_i(\lambda_k(R_j))-(\mathcal{L}_{R_i}\lambda_k)(R_j)=0 λ k ([ R i , R j ]) = R i ( λ k ( R j )) − ( L R i λ k ) ( R j ) = 0 . Thus [ R i , R j ] [R_i,R_j] [ R i , R j ] belongs to R ∩ ξ = { 0 } \mathcal{R}\cap\xi=\{0\} R ∩ ξ = { 0 } .
If U , V U,V U , V take values in ξ \xi ξ , then λ i ( [ U , V ] ) = − d λ i ( U , V ) = 0 \lambda_i([U,V])=-\mathrm{d}\lambda_i(U,V)=0 λ i ([ U , V ]) = − d λ i ( U , V ) = 0 . Hence ξ \xi ξ is integrable. The restriction of Ω \Omega Ω to a leaf is closed and non-degenerate, so it is symplectic.
The Reeb fields generate a local R q \mathbb{R}^q R q -action. A global action requires their completeness. In particular, the closedness assumptions in Definition 2.2 force the Reeb algebra to be Abelian.
Theorem 2.1 (Darboux theorem). Let ( M , Ω , λ ⃗ ) (M,\Omega,\vec{\lambda}) ( M , Ω , λ ) be a q q q -cosymplectic manifold of dimension 2 n + q 2n+q 2 n + q . Around every point there are coordinates ( q 1 , … , q n , p 1 , … , p n , t 1 , … , t q ) (q^1,\ldots,q^n,p_1,\ldots,p_n,t_1,\ldots,t_q) ( q 1 , … , q n , p 1 , … , p n , t 1 , … , t q ) such that
Ω = ∑ i = 1 n d q i ∧ d p i , λ j = d t j , R j = ∂ ∂ t j . \Omega=\sum_{i=1}^{n}\mathrm{d}q^i\wedge\mathrm{d}p_i,\qquad\lambda_j=\mathrm{d}t_j,\qquad R_j=\frac{\partial}{\partial t_j}. Ω = i = 1 ∑ n d q i ∧ d p i , λ j = d t j , R j = ∂ t j ∂ .
Proof. Fix x 0 ∈ M x_0 \in M x 0 ∈ M . After restricting to a neighbourhood of x 0 x_0 x 0 , write λ j = d t j \lambda_j = \mathrm{d}t_j λ j = d t j with t j ( x 0 ) = 0 t_j(x_0) = 0 t j ( x 0 ) = 0 . The differentials d t j \mathrm{d}t_j d t j are independent, so Σ 0 = { t 1 = ⋯ = t q = 0 } \Sigma_0 = \{t_1 = \cdots= t_q = 0\} Σ 0 = { t 1 = ⋯ = t q = 0 } is locally a 2 n 2n 2 n -dimensional submanifold with T Σ 0 = ξ ∣ Σ 0 T\Sigma_0 = \left.\xi\right|_{\Sigma_0} T Σ 0 = ξ ∣ Σ 0 . If ι : Σ 0 ↪ M \iota: \Sigma_0 \hookrightarrow M ι : Σ 0 ↪ M is the inclusion, then ι ∗ Ω \iota^*\Omega ι ∗ Ω is symplectic. Choose symplectic Darboux coordinates ( q , p ) (q,p) ( q , p ) on a neighbourhood in Σ 0 \Sigma_0 Σ 0 .
Let ϕ s j \phi_s^j ϕ s j denote the local flow of R j R_j R j . For x ∈ Σ 0 x \in\Sigma_0 x ∈ Σ 0 and sufficiently small s = ( s 1 , … , s q ) s = (s_1,\ldots,s_q) s = ( s 1 , … , s q ) , define
Φ ( x , s ) = ϕ s 1 1 ∘ ⋯ ∘ ϕ s q q ( x ) . \Phi(x,s) = \phi_{s_1}^1 \circ\cdots\circ\phi_{s_q}^q(x). Φ ( x , s ) = ϕ s 1 1 ∘ ⋯ ∘ ϕ s q q ( x ) .
At ( x 0 , 0 ) (x_0,0) ( x 0 , 0 ) its differential is ( v , a ) ↦ v + ∑ j a j R j ( x 0 ) (v,a) \mapsto v + \sum_j a_jR_j(x_0) ( v , a ) ↦ v + ∑ j a j R j ( x 0 ) , which is an isomorphism because T x 0 M = T x 0 Σ 0 ⊕ R x 0 T_{x_0}M = T_{x_0}\Sigma_0 \oplus R_{x_0} T x 0 M = T x 0 Σ 0 ⊕ R x 0 . Thus, after shrinking the domain, Φ \Phi Φ is a diffeomorphism onto a neighbourhood of x 0 x_0 x 0 .
By Proposition 2.2, the flows commute and Φ ∗ ∂ s j = R j \Phi_*\partial_{s_j} = R_j Φ ∗ ∂ s j = R j . Since R j ( t k ) = δ j k R_j(t_k) = \delta_{jk} R j ( t k ) = δ j k and the t k t_k t k vanish on Σ 0 \Sigma_0 Σ 0 , one has t k ∘ Φ = s k t_k \circ\Phi= s_k t k ∘ Φ = s k . Extend the functions ( q , p ) (q,p) ( q , p ) from Σ 0 \Sigma_0 Σ 0 by making them constant along the Reeb flows. Then ( q , p , t ) (q,p,t) ( q , p , t ) are coordinates with R j = ∂ t j R_j = \partial_{t_j} R j = ∂ t j and λ j = d t j \lambda_j = \mathrm{d}t_j λ j = d t j .
Finally, L ∂ t j Ω = 0 \mathcal{L}_{\partial_{t_j}}\Omega= 0 L ∂ t j Ω = 0 excludes all d t j \mathrm{d}t_j d t j -components of Ω \Omega Ω , and L ∂ t j Ω = 0 \mathcal{L}_{\partial_{t_j}}\Omega= 0 L ∂ t j Ω = 0 makes the remaining coefficients independent of t t t . Its expression is therefore the pull-back of its Darboux expression on Σ 0 \Sigma_0 Σ 0 .
Definition 2.3. For H ∈ C ∞ ( M ) H \in C^\infty(M) H ∈ C ∞ ( M ) , its Hamiltonian vector field is the unique field X H X_H X H satisfying
ι X H Ω = d H − ∑ i = 1 q R i ( H ) λ i , λ i ( X H ) = 0 , i = 1 , … , q . \iota_{X_H}\Omega= \mathrm{d}H - \sum_{i=1}^{q}R_i(H)\lambda_i,\qquad\lambda_i(X_H) = 0,\qquad i=1,\ldots,q. ι X H Ω = d H − i = 1 ∑ q R i ( H ) λ i , λ i ( X H ) = 0 , i = 1 , … , q . The associated bracket is { f , g } : = Ω ( X f , X g ) \{f,g\} := \Omega(X_f,X_g) { f , g } := Ω ( X f , X g ) .
The right-hand side of the contraction equation annihilates R \mathcal{R} R . Existence and uniqueness of X H ∈ ξ X_H \in\xi X H ∈ ξ therefore follow from non-degeneracy of Ω ∣ ξ \left.\Omega\right|_\xi Ω ∣ ξ . In the coordinates of Theorem 2.1,
X f = ∑ i = 1 n ( ∂ f ∂ p i ∂ ∂ q i − ∂ f ∂ q i ∂ ∂ p i ) , { f , g } = ∑ i = 1 n ( ∂ f ∂ q i ∂ g ∂ p i − ∂ f ∂ p i ∂ g ∂ q i ) . X_f = \sum_{i=1}^{n}\left(\frac{\partial f}{\partial p_i}\frac{\partial}{\partial q^i}-\frac{\partial f}{\partial q^i}\frac{\partial}{\partial p_i}\right),\qquad\{f,g\} = \sum_{i=1}^{n}\left(\frac{\partial f}{\partial q^i}\frac{\partial g}{\partial p_i}-\frac{\partial f}{\partial p_i}\frac{\partial g}{\partial q^i}\right). X f = i = 1 ∑ n ( ∂ p i ∂ f ∂ q i ∂ − ∂ q i ∂ f ∂ p i ∂ ) , { f , g } = i = 1 ∑ n ( ∂ q i ∂ f ∂ p i ∂ g − ∂ p i ∂ f ∂ q i ∂ g ) .
In particular, f ↦ X f f \mapsto X_f f ↦ X f is R \mathbb{R} R -linear.
Proposition 2.3. The bracket { ⋅ , ⋅ } \{\cdot,\cdot\} { ⋅ , ⋅ } is Poisson, with symplectic leaves given by the leaves of ξ \xi ξ . For all f , g ∈ C ∞ ( M ) f,g \in C^\infty(M) f , g ∈ C ∞ ( M ) ,
{ f , g } = X g ( f ) , [ X f , X g ] = − X { f , g } , [ R i , X f ] = X R i ( f ) . \{f,g\} = X_g(f),\qquad[X_f,X_g] = -X_{\{f,g\}},\qquad[R_i,X_f] = X_{R_i(f)}. { f , g } = X g ( f ) , [ X f , X g ] = − X { f , g } , [ R i , X f ] = X R i ( f ) .
Proof. By the defining equations and the horizontality of X g X_g X g ,
Ω ( X f , X g ) = d f ( X g ) − ∑ i R i ( f ) λ i ( X g ) = X g ( f ) . \Omega(X_f,X_g) = \mathrm{d}f(X_g) - \sum_i R_i(f)\lambda_i(X_g) = X_g(f). Ω ( X f , X g ) = d f ( X g ) − i ∑ R i ( f ) λ i ( X g ) = X g ( f ) .
In Darboux coordinates, this is the canonical Poisson bracket in ( q , p ) (q,p) ( q , p ) , with t t t as parameters. It satisfies the Leibniz and Jacobi identities, and its characteristic distribution is ξ \xi ξ .
For any smooth function h h h , the Jacobi identity gives
[ X f , X g ] ( h ) = { { h , g } , f } − { { h , f } , g } = { h , { g , f } } = − X { f , g } ( h ) . [X_f,X_g](h) = \{\{h,g\},f\} - \{\{h,f\},g\} = \{h,\{g,f\}\} = -X_{\{f,g\}}(h). [ X f , X g ] ( h ) = {{ h , g } , f } − {{ h , f } , g } = { h , { g , f }} = − X { f , g } ( h ) .
The last identity follows by differentiating the coordinate expression of X f X_f X f with respect to t i t_i t i , since R i = ∂ t i R_i = \partial_{t_i} R i = ∂ t i .
Remark 2.3 (Relation to presymplectic Poisson brackets). The bracket above is a special case of the construction in [16 Sec. 4] . Indeed, the additional one-forms determine the projector
Q = id T M − ∑ i = 1 q R i ⊗ λ i Q = \operatorname{id}_{TM} - \sum_{i=1}^{q} R_i \otimes\lambda_i Q = id T M − i = 1 ∑ q R i ⊗ λ i
with Q 2 = Q Q^2 = Q Q 2 = Q , ker Q = ker Ω = R \ker Q = \ker\Omega= \mathcal{R} ker Q = ker Ω = R , and Im Q = ξ \operatorname{Im} Q = \xi Im Q = ξ . This is the generalized connection denoted by A A A in that reference.
Their Eqs. (4.4)–(4.5) associate to a one-form α \alpha α the unique vector field satisfying ι X α Ω = Q ∗ α \iota_{X_\alpha}\Omega= Q^*\alpha ι X α Ω = Q ∗ α and Q X α = X α QX_\alpha= X_\alpha Q X α = X α , where Q ∗ α = α ∘ Q Q^*\alpha= \alpha\circ Q Q ∗ α = α ∘ Q . Taking α = d f \alpha= df α = df gives
ι X f Ω = d f − ∑ i = 1 q R i ( f ) λ i , X f ∈ Γ ( ξ ) . \iota_{X_f}\Omega= df - \sum_{i=1}^{q} R_i(f)\lambda_i, \qquad X_f \in\Gamma(\xi). ι X f Ω = df − i = 1 ∑ q R i ( f ) λ i , X f ∈ Γ ( ξ ) .
Thus, with our bracket convention, their construction gives exactly { f , g } = Ω ( X f , X g ) \{f,g\} = \Omega(X_f,X_g) { f , g } = Ω ( X f , X g ) . In [16 ] , the authors also state that this bracket satisfies the Jacobi identity if and only if the generalized connection is flat. The vertical distribution is involutive because d Ω = 0 d\Omega= 0 d Ω = 0 , while the horizontal curvature vanishes:
( id T M − Q ) [ Q U , Q V ] = − ∑ i = 1 q d λ i ( Q U , Q V ) R i = 0. (\operatorname{id}_{TM}-Q)[QU,QV] = -\sum_{i=1}^{q}d\lambda_i(QU,QV)R_i = 0. ( id T M − Q ) [ Q U , Q V ] = − i = 1 ∑ q d λ i ( Q U , Q V ) R i = 0.
Hence the closedness of the λ i \lambda_i λ i supplies precisely the required flatness.
On the algebra A Ω = { f ∈ C ∞ ( M ) : d f ∣ R = 0 } \mathcal{A}_{\Omega} = \{f \in C^\infty(M) : df|_{\mathcal{R}} = 0\} A Ω = { f ∈ C ∞ ( M ) : df ∣ R = 0 } , the bracket agrees with the intrinsic presymplectic bracket. For such f f f , the solutions of ι Y f Ω = d f \iota_{Y_f}\Omega= df ι Y f Ω = df are Y f = X f + ∑ i a i R i Y_f = X_f + \sum_i a_iR_i Y f = X f + ∑ i a i R i , and their pairings under Ω \Omega Ω are independent of the functions a i a_i a i . The extension to all smooth functions depends on the chosen complement ξ \xi ξ . No global smooth quotient by R \mathcal{R} R is required.
The full q q q -cosymplectic structure also specifies the ordered closed one-forms and their dual Reeb frame. Preservation of these data together with Ω \Omega Ω is stronger than preservation of the induced Poisson bracket. In particular, Cartan’s formula gives
L X f λ i = 0 , L X f Ω = − ∑ i = 1 q d ( R i ( f ) ) ∧ λ i . \mathcal{L}_{X_f}\lambda_i = 0, \qquad\mathcal{L}_{X_f}\Omega= -\sum_{i=1}^{q}d(R_i(f)) \wedge\lambda_i. L X f λ i = 0 , L X f Ω = − i = 1 ∑ q d ( R i ( f )) ∧ λ i .
These identities will be used in the transformation theory.
q-contact manifolds
Definition 2.4 (q q q -contact manifolds [2 , 21 , 22 , 37 ] ). A q q q -contact structure on M 2 n + q M^{2n+q} M 2 n + q consists of pointwise linearly independent one-forms λ ⃗ = ( λ 1 , … , λ q ) \vec{\lambda} = (\lambda_1,\ldots,\lambda_q) λ = ( λ 1 , … , λ q ) and a splitting T M = R ⊕ ξ TM = \mathcal{R} \oplus\xi T M = R ⊕ ξ such that ξ = ⋂ i ker λ i \xi= \bigcap_i \ker\lambda_i ξ = ⋂ i ker λ i , ker d λ i = R \ker d\lambda_i = \mathcal{R} ker d λ i = R , and d λ i ∣ ξ d\lambda_i|_{\xi} d λ i ∣ ξ is non-degenerate for every i i i . It is uniform if d λ 1 = ⋯ = d λ q = : Ω d\lambda_1 = \cdots= d\lambda_q =: \Omega d λ 1 = ⋯ = d λ q =: Ω .
These data fit the common setting above with m = q m=q m = q and Ω i = d λ i \Omega_i = d\lambda_i Ω i = d λ i . Thus there are unique Reeb fields R 1 , … , R q R_1,\ldots,R_q R 1 , … , R q such that λ i ( R j ) = δ i j \lambda_i(R_j) = \delta_{ij} λ i ( R j ) = δ ij and ι R j d λ i = 0 \iota_{R_j}d\lambda_i = 0 ι R j d λ i = 0 for every i , j i,j i , j . The next statement does not require uniformity.
Proposition 2.4 [2 , 21 ] . On every q q q -contact manifold, the Reeb fields form a commuting frame of R \mathcal{R} R and satisfy
L R i λ j = 0 , L R i d λ j = 0. \mathcal{L}_{R_i}\lambda_j=0,\qquad\mathcal{L}_{R_i}d\lambda_j=0. L R i λ j = 0 , L R i d λ j = 0. In the uniform case, in particular, L R i Ω = 0 \mathcal{L}_{R_i}\Omega=0 L R i Ω = 0 .
Proof. Cartan’s formula gives L R i λ j = d ( δ j i ) + ι R i d λ j = 0 \mathcal{L}_{R_i}\lambda_j=d(\delta^i_j)+\iota_{R_i}d\lambda_j=0 L R i λ j = d ( δ j i ) + ι R i d λ j = 0 , and exterior differentiation gives the second identity. For every k k k , λ k ( [ R i , R j ] ) = 0 \lambda_k([R_i,R_j])=0 λ k ([ R i , R j ]) = 0 . Also,
ι [ R i , R j ] d λ k = L R i ( ι R j d λ k ) − ι R j ( L R i d λ k ) = 0. \iota_{[R_i,R_j]}d\lambda_k=\mathcal{L}_{R_i}(\iota_{R_j}d\lambda_k)-\iota_{R_j}(\mathcal{L}_{R_i}d\lambda_k)=0. ι [ R i , R j ] d λ k = L R i ( ι R j d λ k ) − ι R j ( L R i d λ k ) = 0. Hence [ R i , R j ] [R_i,R_j] [ R i , R j ] belongs to both ξ \xi ξ and R \mathcal{R} R , and therefore vanishes. □
We now assume uniformity. The following normal form is recalled from [21 ] ; we include a proof adapted to our conventions.
Proposition 2.5 (Darboux theorem). Let ( M , λ ⃗ ) (M,\vec{\lambda}) ( M , λ ) be a uniform q q q -contact manifold of dimension 2 n + q 2n+q 2 n + q . Around every point there are coordinates ( q 1 , … , q n , p 1 , … , p n , z 1 , … , z q ) (q^1,\ldots,q^n,p_1,\ldots,p_n,z_1,\ldots,z_q) ( q 1 , … , q n , p 1 , … , p n , z 1 , … , z q ) such that
λ j = d z j − ∑ i = 1 n p i d q i , Ω = ∑ i = 1 n d q i ∧ d p i , R j = ∂ ∂ z j . \lambda_j=dz_j-\sum_{i=1}^{n}p_i\,dq^i,\qquad\Omega=\sum_{i=1}^{n}dq^i\wedge dp_i,\qquad R_j=\frac{\partial}{\partial z_j}. λ j = d z j − i = 1 ∑ n p i d q i , Ω = i = 1 ∑ n d q i ∧ d p i , R j = ∂ z j ∂ . Proof. By Proposition 2.4, R \mathcal{R} R is integrable. On a sufficiently small foliation chart U U U , let π : U → B \pi:U\to B π : U → B be a submersion onto a 2 n 2n 2 n -dimensional manifold with ker T π = R ∣ U \ker T\pi=\mathcal{R}|_U ker T π = R ∣ U . The form Ω \Omega Ω annihilates R \mathcal{R} R and is invariant under the Reeb fields. It therefore descends to a form ω \omega ω on B B B , with Ω = π ∗ ω \Omega=\pi^*\omega Ω = π ∗ ω .
The form ω \omega ω is closed. If a vector in T B TB T B is in its kernel, any lift of that vector lies in ker Ω = ker T π \ker\Omega=\ker T\pi ker Ω = ker T π . Thus ω \omega ω is non-degenerate. Choose Darboux coordinates ( q , p ) (q,p) ( q , p ) on B B B and set θ = − ∑ i p i d q i \theta=-\sum_i p_i\,dq^i θ = − ∑ i p i d q i , so that d θ = ω d\theta=\omega d θ = ω . Since d ( λ j − π ∗ θ ) = 0 d(\lambda_j-\pi^*\theta)=0 d ( λ j − π ∗ θ ) = 0 , after shrinking U U U there are functions z j z_j z j such that λ j − π ∗ θ = d z j \lambda_j-\pi^*\theta=dz_j λ j − π ∗ θ = d z j .
The differentials of the pulled-back functions ( q , p ) (q,p) ( q , p ) annihilate R \mathcal{R} R , whereas d z j ( R k ) = δ j k dz_j(R_k)=\delta_{jk} d z j ( R k ) = δ j k . Consequently, the differentials of ( q , p , z ) (q,p,z) ( q , p , z ) are linearly independent and these functions form a coordinate system. The stated expressions for λ j \lambda_j λ j , Ω \Omega Ω , and R j R_j R j follow. Only a local quotient has been used in this argument. □
Proposition 2.6 (Contact structures on leaves). Set E = ∑ i = 1 q R i E=\sum_{i=1}^{q}R_i E = ∑ i = 1 q R i . For j = 2 , … , q j=2,\ldots,q j = 2 , … , q , the forms β j = λ j − λ 1 \beta_j=\lambda_j-\lambda_1 β j = λ j − λ 1 are closed and pointwise linearly independent. Their common kernel is the integrable distribution
D : = ⋂ j = 2 q ker β j = ξ ⊕ span { E } . D:=\bigcap_{j=2}^{q}\ker\beta_j=\xi\oplus\operatorname{span}\{E\}. D := j = 2 ⋂ q ker β j = ξ ⊕ span { E } . Every leaf L L L of D D D carries the contact form α L = λ 1 ∣ L = ⋯ = λ q ∣ L \alpha_L=\lambda_1|_L=\cdots=\lambda_q|_L α L = λ 1 ∣ L = ⋯ = λ q ∣ L , whose Reeb field is E ∣ L E|_L E ∣ L . For q = 1 q=1 q = 1 , the intersection is understood as T M TM T M .
Proof. Uniformity gives d β j = 0 d\beta_j=0 d β j = 0 . If ∑ j = 2 q c j β j = 0 \sum_{j=2}^{q}c_j\beta_j=0 ∑ j = 2 q c j β j = 0 , evaluation on R k R_k R k , for k ≥ 2 k\ge2 k ≥ 2 , gives c k = 0 c_k=0 c k = 0 . Thus the β j \beta_j β j are independent. A vector belongs to their common kernel exactly when all its values under the λ i \lambda_i λ i coincide. Using the splitting T M = ξ ⊕ R TM=\xi\oplus R T M = ξ ⊕ R , this is equivalent to being of the form Y + a E Y+aE Y + a E with Y ∈ ξ Y\in\xi Y ∈ ξ . Since the β j \beta_j β j are closed, D \mathcal{D} D is integrable.
On a leaf L L L , the forms λ i ∣ L \lambda_i|_L λ i ∣ L coincide. Moreover, α L ( E ) = 1 \alpha_L(E)=1 α L ( E ) = 1 , ι E d α L = 0 \iota_E d\alpha_L=0 ι E d α L = 0 , and d α L ∣ ξ = Ω ∣ ξ d\alpha_L|_{\xi}=\Omega|_{\xi} d α L ∣ ξ = Ω ∣ ξ is non-degenerate. Hence α L ∧ ( d α L ) n \alpha_L\wedge(d\alpha_L)^n α L ∧ ( d α L ) n is nowhere zero, and E ∣ L E|_L E ∣ L is its Reeb field. □
In the coordinates of Proposition 2.5, put w = z 1 w=z_1 w = z 1 and u j = z j − z 1 u_j=z_j-z_1 u j = z j − z 1 for j = 2 , … , q j=2,\ldots,q j = 2 , … , q . Then
λ 1 = d w − ∑ i p i d q i , λ j = λ 1 + d u j , E = ∂ w . \lambda_1=dw-\sum_i p_i\,dq^i,\qquad\lambda_j=\lambda_1+du_j,\qquad E=\partial_w. λ 1 = d w − i ∑ p i d q i , λ j = λ 1 + d u j , E = ∂ w .
The leaves of D \mathcal{D} D are locally the sets u = u= u = constant. This gives a local description by ordinary contact manifolds together with transverse closed forms. The local functions u j u_j u j need not exist globally, and the description does not imply a global product decomposition.
Definition 2.5. Let ( M , λ ) (M,\lambda) ( M , λ ) be uniform and let H ∈ C ∞ ( M ) H\in C^\infty(M) H ∈ C ∞ ( M ) . The Hamiltonian vector field X H X_H X H is the unique field satisfying
ι X H Ω = d H − ∑ i = 1 q R i ( H ) λ i , λ i ( X H ) = − H , i = 1 , … , q . \iota_{X_H}\Omega=dH-\sum_{i=1}^{q}R_i(H)\lambda_i,\qquad\lambda_i(X_H)=-H,\quad i=1,\ldots,q. ι X H Ω = d H − i = 1 ∑ q R i ( H ) λ i , λ i ( X H ) = − H , i = 1 , … , q . The associated bracket is
{ f , g } : = Ω ( X f , X g ) + f E ( g ) − g E ( f ) , E = ∑ i = 1 q R i . \{f,g\}:=\Omega(X_f,X_g)+fE(g)-gE(f),\qquad E=\sum_{i=1}^{q}R_i. { f , g } := Ω ( X f , X g ) + f E ( g ) − g E ( f ) , E = i = 1 ∑ q R i . The Hamiltonian system is called good if R i ( H ) = 0 R_i(H)=0 R i ( H ) = 0 for every i i i .
To see existence and uniqueness, let Y H ∈ ξ Y_H\in\xi Y H ∈ ξ be the unique solution of ι Y H Ω = d H − ∑ i R i ( H ) λ i \iota_{Y_H}\Omega=dH-\sum_iR_i(H)\lambda_i ι Y H Ω = d H − ∑ i R i ( H ) λ i . Then X H = Y H − H E X_H=Y_H-HE X H = Y H − H E satisfies both equations. Any two solutions differ by a vector in R R R annihilated by all the λ i \lambda_i λ i , and hence coincide. In particular, X H X_H X H is tangent to D \mathcal{D} D . On each leaf L L L , its restriction is the usual contact Hamiltonian field of H ∣ L H|_L H ∣ L , with convention α L ( X H ) = − H ∣ L \alpha_L(X_H)=-H|_L α L ( X H ) = − H ∣ L .
In Darboux coordinates, the expression is
X H = ∑ i = 1 n [ ∂ H ∂ p i ∂ q i − ( ∂ H ∂ q i + p i ∑ j = 1 q ∂ H ∂ z j ) ∂ p i ] + ( ∑ i = 1 n p i ∂ H ∂ p i − H ) ∑ j = 1 q ∂ z j . \begin{aligned}
X_H=\sum_{i=1}^{n}\left[\frac{\partial H}{\partial p_i}\partial_{q^i}-\left(\frac{\partial H}{\partial q^i}+p_i\sum_{j=1}^{q}\frac{\partial H}{\partial z_j}\right)\partial_{p_i}\right] \\
&+\left(\sum_{i=1}^{n}p_i\frac{\partial H}{\partial p_i}-H\right)\sum_{j=1}^{q}\partial_{z_j}.
\end{aligned} X H = i = 1 ∑ n [ ∂ p i ∂ H ∂ q i − ( ∂ q i ∂ H + p i j = 1 ∑ q ∂ z j ∂ H ) ∂ p i ] + ( i = 1 ∑ n p i ∂ p i ∂ H − H ) j = 1 ∑ q ∂ z j .
The map H ↦ X H H\mapsto X_H H ↦ X H is R \mathbb{R} R -linear and injective, since λ i ( X H ) = − H \lambda_i(X_H)=-H λ i ( X H ) = − H . Cartan’s formula and the contraction equation give
L X H λ i = − ∑ j = 1 q R j ( H ) λ j , X H ( H ) = − H E ( H ) . \mathcal{L}_{X_H}\lambda_i=-\sum_{j=1}^{q}R_j(H)\lambda_j,\qquad X_H(H)=-H E(H). L X H λ i = − j = 1 ∑ q R j ( H ) λ j , X H ( H ) = − H E ( H ) .
Thus a good Hamiltonian is conserved along its flow, and its Hamiltonian field preserves every λ i \lambda_i λ i . For all H H H , one also has X H ( u j ) = 0 X_H(u_j)=0 X H ( u j ) = 0 in the local coordinates above.
The motivating examples
We now explain how the geometric data of § 2 arise in constrained mechanics. We first describe the passage from a regular constraint surface to a q q q -cosymplectic structure, including an explicit construction for the Abelian vector group. We then consider relativistic particle constraints and global Lagrangian descriptions of magnetic monopoles. The underlying presymplectic constructions belong to the classical theory of constrained systems [26 , 35 ] . Our purpose is to identify the additional one-forms, their normalization, and the resulting restrictions on the transformation theory.
The interpretation of the characteristic directions depends on the system. In relativistic constraint dynamics they parametrize constraint orbits; in the free-particle example these parameters are proportional to the individual proper times. For a charge in a magnetic field, the physical motion has a single time parameter, while the Reeb fields of the lifted description generate internal gauge transformations. In particular, the integer q q q does not by itself count physical times or gauge directions that cannot be quotiented out.
On a symplectic manifold we use ι X f ω = d f \iota_{X_f}\omega= d f ι X f ω = df and { f , g } ω = ω ( X f , X g ) = X g ( f ) \{f,g\}_{\omega} = \omega(X_f,X_g) = X_g(f) { f , g } ω = ω ( X f , X g ) = X g ( f ) , consistently with § 2 . When θ \theta θ is a Liouville or Cartan form, we retain the convention ω = d θ \omega= d\theta ω = d θ used in the mechanical examples below. The variational equations then have the form ι Γ d θ = − d E \iota_{\Gamma}d\theta= -dE ι Γ d θ = − d E .
Constraint surfaces and Abelian group actions
The following construction separates the presymplectic structure determined by the constraints from the additional choice of clock functions.
Proposition 3.1. Let ( P , ω ) (P,\omega) ( P , ω ) be symplectic of dimension 2 n + 2 q 2n+2q 2 n + 2 q . Let C : P → R q C:P\to\mathbb{R}^{q} C : P → R q have components C 1 , … , C q C_1,\ldots,C_q C 1 , … , C q . Suppose that c c c is a regular value, that M = C − 1 ( c ) M=C^{-1}(c) M = C − 1 ( c ) is nonempty, and that { C a , C b } ω ∣ M = 0 \{C_a,C_b\}_{\omega}|_M=0 { C a , C b } ω ∣ M = 0 for all a , b a,b a , b . Write ι : M ↪ P \iota:M\hookrightarrow P ι : M ↪ P and Ω = ι ∗ ω \Omega=\iota^*\omega Ω = ι ∗ ω . Then Ω \Omega Ω has rank 2 n 2n 2 n , and
R : = ker Ω = span { X C 1 ∣ M , … , X C q ∣ M } . \mathcal{R}:=\ker\Omega=\operatorname{span}\{X_{C_1}|_M,\ldots,X_{C_q}|_M\}. R := ker Ω = span { X C 1 ∣ M , … , X C q ∣ M } . Let χ 1 , … , χ q \chi_1,\ldots,\chi_q χ 1 , … , χ q be functions on an open subset U ⊂ M U\subset M U ⊂ M such that the matrix A a b = d χ a ( X C b ∣ M ) A_{ab}=d\chi_a(X_{C_b}|_M) A ab = d χ a ( X C b ∣ M ) is invertible. The functions χ a \chi_a χ a thus provide local gauge-fixing conditions for the first-class constraints C a − c a = 0 C_a-c_a=0 C a − c a = 0 . More precisely, if χ ~ a \widetilde{\chi}_a χ a are local extensions of χ a \chi_a χ a to P P P , then the constraints C a − c a = 0 C_a-c_a=0 C a − c a = 0 and χ ~ a − τ a = 0 \widetilde{\chi}_a-\tau_a=0 χ a − τ a = 0 , a = 1 , … , q a=1,\ldots,q a = 1 , … , q , form a second-class system along each nonempty slice Σ τ = { x ∈ U : χ ( x ) = τ } \Sigma_\tau=\{x\in U:\chi(x)=\tau\} Σ τ = { x ∈ U : χ ( x ) = τ } , where χ = ( χ 1 , … , χ q ) \chi=(\chi_1,\ldots,\chi_q) χ = ( χ 1 , … , χ q ) . Then ( U , Ω ∣ U , d χ 1 , … , d χ q ) (U,\Omega|_U,d\chi_1,\ldots,d\chi_q) ( U , Ω ∣ U , d χ 1 , … , d χ q ) is q q q -cosymplectic, with Reeb fields
R a = ∑ b = 1 q ( A − 1 ) b a X C b ∣ U . R_a=\sum_{b=1}^{q}(A^{-1})_{ba}X_{C_b}|_U. R a = b = 1 ∑ q ( A − 1 ) ba X C b ∣ U . Such functions exist locally around every point of M M M .
Proof. Regularity gives dim M = 2 n + q \dim M=2n+q dim M = 2 n + q and linear independence of the constraint fields along M M M . At each x ∈ M x\in M x ∈ M , the symplectic orthogonal of T x M T_xM T x M is spanned by X C a ( x ) X_{C_a}(x) X C a ( x ) . The assumption on the brackets makes these fields tangent to M M M . Thus M M M is coisotropic, and ker Ω x = ( T x M ) ω \ker\Omega_x=(T_xM)^\omega ker Ω x = ( T x M ) ω has dimension q q q .
Invertibility of A A A means that the restrictions of d χ 1 , … , d χ q d\chi_1,\ldots,d\chi_q d χ 1 , … , d χ q to R \mathcal{R} R form a coframe. Since X C b X_{C_b} X C b is tangent to M M M , we have { χ ~ a , C b } ω ∣ U = A a b \{\widetilde{\chi}_a,C_b\}_{\omega}|_U=A_{ab} { χ a , C b } ω ∣ U = A ab for any such extensions. Ordering the constraints as ( C − c , χ ~ − τ ) (C-c,\widetilde{\chi}-\tau) ( C − c , χ − τ ) , their Poisson bracket matrix along Σ τ \Sigma_\tau Σ τ is
( 0 − A T A B ) , B a b = { χ ~ a , χ ~ b } ω ∣ Σ τ . \begin{pmatrix}
0 & -A^{\mathsf T}\\
A & B
\end{pmatrix},
\qquad B_{ab}=\{\widetilde{\chi}_a,\widetilde{\chi}_b\}_{\omega}|_{\Sigma_\tau}. ( 0 A − A T B ) , B ab = { χ a , χ b } ω ∣ Σ τ .
Its determinant is ( det A ) 2 (\det A)^2 ( det A ) 2 , so the combined constraints are second class. Consequently, T U = R ∣ U ⊕ ⋂ a ker d χ a TU=\mathcal{R}|_U\oplus\bigcap_a\ker d\chi_a T U = R ∣ U ⊕ ⋂ a ker d χ a . The restriction of Ω \Omega Ω to this complement is non-degenerate, and all the defining forms are closed. The displayed formula for R a R_a R a follows from d χ b ( R a ) = δ b a d\chi_b(R_a)=\delta_{ba} d χ b ( R a ) = δ ba .
Finally, at any point choose q q q covectors whose restrictions to R \mathcal{R} R are independent, and realize them as differentials of local functions. The corresponding matrix remains invertible after shrinking the neighbourhood. ∎
The hypothesis on the brackets is the first-class condition for these regular constraints on M M M . It does not assert strong Poisson commutativity in a neighbourhood of M M M . For example, brackets of the form { C a , C b } ω = ∑ d f a b d ( C d − c d ) \{C_a,C_b\}_{\omega}=\sum_d f_{ab}^{d}(C_d-c_d) { C a , C b } ω = ∑ d f ab d ( C d − c d ) satisfy the hypothesis, although the restricted constraint fields need not commute. The normalized Reeb fields in Proposition 3.1 do commute, by Proposition 2.2. Thus the choice of clock functions can change the frame of the characteristic distribution; it does not preserve a prescribed non-Abelian frame.
In particular, let R q \mathbb{R}^{q} R q act in a Hamiltonian fashion with an equivariant momentum map J : P → ( R q ) ∗ J:P\to(\mathbb{R}^{q})^{*} J : P → ( R q ) ∗ . Its components Poisson-commute, and a regular momentum level has the presymplectic structure described above. Closed one-forms restricting to a coframe on its kernel complete it to a q q q -cosymplectic structure. This completion always exists locally; its global existence requires additional data. The regular momentum level alone determines the presymplectic form, but not the clock forms or the horizontal distribution.
An explicit global model exhibits the relation with Hamilton–Jacobi theory for an Abelian group [29 , 27 ] . Let ( S , ω S ) (S,\omega_{S}) ( S , ω S ) be a symplectic manifold of dimension 2 n 2n 2 n and let H 1 , … , H q ∈ C ∞ ( S ) H_{1},\ldots,H_{q}\in C^{\infty}(S) H 1 , … , H q ∈ C ∞ ( S ) satisfy { H a , H b } ω S = 0 \{H_{a},H_{b}\}_{\omega_{S}}=0 { H a , H b } ω S = 0 . Set Y a = X H a Y_{a}=X_{H_{a}} Y a = X H a and consider
P ^ = S × T ∗ R q , ω ^ = ω S + ∑ a = 1 q d t a ∧ d π a , C a = π a + H a . \widehat{P}=S\times T^{*}\mathbb{R}^{q},\qquad\widehat{\omega}=\omega_{S}+\sum_{a=1}^{q}dt_{a}\wedge d\pi_{a},\qquad C_{a}=\pi_{a}+H_{a}. P = S × T ∗ R q , ω = ω S + a = 1 ∑ q d t a ∧ d π a , C a = π a + H a .
The C a C_{a} C a Poisson-commute and have independent differentials. Their common level C = c C=c C = c is the graph π a = c a − H a \pi_{a}=c_{a}-H_{a} π a = c a − H a , hence is diffeomorphic to S × R q S\times\mathbb{R}^{q} S × R q . On this graph the induced structure is
Ω = ω S + ∑ a = 1 q d H a ∧ d t a , λ a = d t a , R a = ∂ t a + Y a . \Omega=\omega_{S}+\sum_{a=1}^{q}dH_{a}\wedge dt_{a},\qquad\lambda_{a}=dt_{a},\qquad R_{a}=\partial_{t_{a}}+Y_{a}. Ω = ω S + a = 1 ∑ q d H a ∧ d t a , λ a = d t a , R a = ∂ t a + Y a .
Indeed, ι R a Ω = 0 \iota_{R_{a}}\Omega=0 ι R a Ω = 0 and d t b ( R a ) = δ a b dt_{b}(R_{a})=\delta_{ab} d t b ( R a ) = δ ab . The characteristic equations are ∂ x / ∂ t a = Y a ( x ) \partial x/\partial t_{a}=Y_{a}(x) ∂ x / ∂ t a = Y a ( x ) , and their compatibility follows from [ Y a , Y b ] = 0 [Y_{a},Y_{b}]=0 [ Y a , Y b ] = 0 . The R a R_{a} R a generate a local R q \mathbb{R}^{q} R q -action, and a global one when the Y a Y_{a} Y a are complete. Here the multi-parameter evolution is already encoded in the Reeb fields; the functions H a H_{a} H a should not be confused with an additional Hamiltonian H H H in the formalism of § 4 .
Remark 3.1. The Abelian hypothesis has a precise structural role. The cotangent-bundle description in [1 Sec. III] expresses the Liouville form through momentum mappings and Maurer–Cartan forms. For a left Maurer–Cartan form κ = ∑ a κ a e a \kappa=\sum_{a}\kappa^{a}e_{a} κ = ∑ a κ a e a , with [ e a , e b ] = ∑ c c a b c e c [e_{a},e_{b}]=\sum_{c}c^{c}_{ab}e_{c} [ e a , e b ] = ∑ c c ab c e c , one has
d κ c = − 1 2 ∑ a , b c a b c κ a ∧ κ b . d\kappa^{c}=-\frac{1}{2}\sum_{a,b}c^{c}_{ab}\kappa^{a}\wedge\kappa^{b}. d κ c = − 2 1 a , b ∑ c ab c κ a ∧ κ b .
For G = R q G=\mathbb{R}^{q} G = R q , these components are the closed forms d t a dt_{a} d t a used in the preceding construction. More generally, if independent fields V a V_{a} V a satisfy [ V a , V b ] = ∑ c c a b c V c [V_{a},V_{b}]=\sum_{c}c^{c}_{ab}V_{c} [ V a , V b ] = ∑ c c ab c V c and λ c ( V a ) = δ c a \lambda_{c}(V_{a})=\delta_{ca} λ c ( V a ) = δ c a , then d λ c ( V a , V b ) = − c a b c d\lambda_{c}(V_{a},V_{b})=-c^{c}_{ab} d λ c ( V a , V b ) = − c ab c . Closed clock forms therefore cannot be dual to a prescribed non-Abelian frame. Retaining such a frame requires modified structure equations of Maurer–Cartan type along the group orbits. This is a separate extension from the non-uniform q q q -contact case, whose Reeb fields still commute.
From q-cosymplectic to uniform q-contact structures
The next observation relates the two kinds of additional data on an exact presymplectic manifold. The horizontal distribution generally changes, although the Reeb fields remain the same.
Proposition 3.2. Let ( M , Ω , λ ⃗ ) (M,\Omega,\vec{\lambda}) ( M , Ω , λ ) be q q q -cosymplectic, with Reeb fields R 1 , … , R q R_1,\ldots,R_q R 1 , … , R q . Suppose that Ω = d θ \Omega=d\theta Ω = d θ and θ ( R b ) = 1 \theta(R_b)=1 θ ( R b ) = 1 for every b b b . Then
η a = θ − ∑ c ≠ a λ c , a = 1 , … , q , \eta_a=\theta-\sum_{c\ne a}\lambda_c,\qquad a=1,\ldots,q, η a = θ − c = a ∑ λ c , a = 1 , … , q , define a uniform q q q -contact structure with d η a = Ω d\eta_a=\Omega d η a = Ω and Reeb fields R a R_a R a . The normalization of the primitive can always be achieved locally. More precisely, if Ω = d θ 0 \Omega=d\theta_0 Ω = d θ 0 on a neighbourhood, then, after shrinking it, there is a function g g g such that θ = θ 0 + d g \theta=\theta_0+dg θ = θ 0 + d g satisfies θ ( R b ) = 1 \theta(R_b)=1 θ ( R b ) = 1 for all b b b .
Proof. We have η a ( R b ) = 1 − ∑ c ≠ a δ c b = δ a b \eta_a(R_b)=1-\sum_{c\ne a}\delta_{cb}=\delta_{ab} η a ( R b ) = 1 − ∑ c = a δ c b = δ ab . Thus the η a \eta_a η a are independent and their common kernel ξ ′ \xi' ξ ′ is a complement of R \mathcal{R} R . Since the λ c \lambda_c λ c are closed, d η a = d θ = Ω d\eta_a=d\theta=\Omega d η a = d θ = Ω . A two-form is non-degenerate on any complement of its kernel, so Ω ∣ ξ ′ \Omega|_{\xi'} Ω ∣ ξ ′ is non-degenerate. These statements give the asserted uniform q q q -contact structure and its Reeb fields.
For the local normalization, solve R b ( g ) = 1 − θ 0 ( R b ) R_b(g)=1-\theta_0(R_b) R b ( g ) = 1 − θ 0 ( R b ) . The compatibility conditions follow from
R c ( θ 0 ( R b ) ) − R b ( θ 0 ( R c ) ) = d θ 0 ( R c , R b ) + θ 0 ( [ R c , R b ] ) = 0. R_c(\theta_0(R_b))-R_b(\theta_0(R_c))=d\theta_0(R_c,R_b)+\theta_0([R_c,R_b])=0. R c ( θ 0 ( R b )) − R b ( θ 0 ( R c )) = d θ 0 ( R c , R b ) + θ 0 ([ R c , R b ]) = 0.
In the coordinates of Theorem 2.1, R b = ∂ t b R_b=\partial_{t_b} R b = ∂ t b , and the Poincaré lemma in the t t t variables, with the remaining coordinates as parameters, gives a local solution. □
Exactness of the clock forms is not needed in this proposition. If λ a = d τ a \lambda_a=d\tau_a λ a = d τ a , its formula becomes η a = θ − ∑ c ≠ a d τ c \eta_a=\theta-\sum_{c\ne a}d\tau_c η a = θ − ∑ c = a d τ c . Global existence of a normalized primitive is an additional requirement. A global solution of the displayed equations for g g g is sufficient; more generally, a primitive may be modified by a closed one-form that need not be exact. Thus local solvability alone gives no global existence statement.
Relativistic interacting particles
We consider the constraint formulation of [8 Model I] . Let Γ = T ∗ ( R 4 ) N \Gamma=T^*(\mathbb{R}^4)^N Γ = T ∗ ( R 4 ) N , with coordinates ( q a μ , p a μ ) (q_a^\mu,p_{a\mu}) ( q a μ , p a μ ) , where a = 1 , … , N a=1,\ldots,N a = 1 , … , N and μ = 0 , … , 3 \mu=0,\ldots,3 μ = 0 , … , 3 . Write θ = ∑ a p a μ d q a μ \theta=\sum_a p_{a\mu}\,dq_a^\mu θ = ∑ a p a μ d q a μ and ω = d θ \omega=d\theta ω = d θ , and use the Minkowski metric of signature ( + , − , − , − ) (+,-,-,-) ( + , − , − , − ) . Repeated spacetime indices are summed. For masses m a > 0 m_a>0 m a > 0 , consider the generalized mass-shell constraints
K a = p a 2 − m a 2 − v a , a = 1 , … , N . K_a=p_a^2-m_a^2-v_a,\qquad a=1,\ldots,N. K a = p a 2 − m a 2 − v a , a = 1 , … , N .
We assume that the potentials are chosen so that { K a , K b } ω = 0 \{K_a,K_b\}_\omega=0 { K a , K b } ω = 0 and restrict to a regular part of M = K − 1 ( 0 ) M=K^{-1}(0) M = K − 1 ( 0 ) . Strong commutativity is an assumption of this Abelian model, not a consequence of the first-class property alone. The manifold M M M has dimension 7 N 7N 7 N , and Ω = ι ∗ ω \Omega=\iota^*\omega Ω = ι ∗ ω has rank 6 N 6N 6 N with
R = ker Ω = span { X K 1 ∣ M , … , X K N ∣ M } . \mathcal{R}=\ker\Omega=\operatorname{span}\{X_{K_1}|_M,\ldots,X_{K_N}|_M\}. R = ker Ω = span { X K 1 ∣ M , … , X K N ∣ M } .
Choose clock functions χ 1 , … , χ N \chi_1,\ldots,\chi_N χ 1 , … , χ N on a region of M M M such that A a b = d χ a ( X K b ∣ M ) A_{ab}=d\chi_a(X_{K_b}|_M) A ab = d χ a ( X K b ∣ M ) is invertible. When the clocks extend to Γ \Gamma Γ , our convention gives A a b = { χ a , K b } ω ∣ M A_{ab}=\{\chi_a,K_b\}_\omega|_M A ab = { χ a , K b } ω ∣ M . Proposition 3.1 yields an N N N -cosymplectic structure with λ a = d χ a \lambda_a=d\chi_a λ a = d χ a and R a = ∑ b ( A − 1 ) b a X K b ∣ M R_a=\sum_b(A^{-1})_{ba}X_{K_b}|_M R a = ∑ b ( A − 1 ) ba X K b ∣ M . These forms specify both a normalization of the constraint actions and a complementary distribution.
Locally, the characteristic quotient has dimension 6 N 6N 6 N and carries the reduced symplectic form. A global reduced phase space requires the characteristic foliation to have a smooth quotient. Functions constant along R \mathcal{R} R descend to observables on that quotient when it exists. The Reeb parameters describe displacements along constraint orbits; interpreting them as individual proper times requires information about the chosen constraints and clocks.
Example 3.1 (Free particles). Set v a = 0 v_a=0 v a = 0 and restrict to the future mass shells p a 0 > 0 p_a^0>0 p a 0 > 0 . With ω = d θ \omega=d\theta ω = d θ and ι X K a ω = d K a \iota_{X_{K_a}}\omega=dK_a ι X K a ω = d K a , one has X K a = − 2 p a μ ∂ ∂ q a μ X_{K_a}=-2p_a^\mu\frac{\partial}{\partial q_a^\mu} X K a = − 2 p a μ ∂ q a μ ∂ . Thus the fields
R a = p a μ m a 2 ∂ ∂ q a μ R_a=\frac{p_a^\mu}{m_a^2}\frac{\partial}{\partial q_a^\mu} R a = m a 2 p a μ ∂ q a μ ∂ span R \mathcal{R} R . Define τ a = p a μ q a μ \tau_a=p_{a\mu}q_a^\mu τ a = p a μ q a μ on M M M . The mass-shell equations give R b ( τ a ) = δ a b R_b(\tau_a)=\delta_{ab} R b ( τ a ) = δ ab and ( ι ∗ θ ) ( R b ) = 1 (\iota^*\theta)(R_b)=1 ( ι ∗ θ ) ( R b ) = 1 . Consequently ( M , Ω , d τ 1 , … , d τ N ) (M,\Omega,d\tau_1,\ldots,d\tau_N) ( M , Ω , d τ 1 , … , d τ N ) is globally N N N -cosymplectic, and Proposition 3.2 gives the global uniform N N N -contact structure
η a = ι ∗ θ − ∑ c ≠ a d τ c . \eta_a=\iota^*\theta-\sum_{c\ne a}d\tau_c. η a = ι ∗ θ − c = a ∑ d τ c . Along an integral curve of R a R_a R a , q a ( s ) = q a ( 0 ) + s p a / m a 2 q_a(s)=q_a(0)+s\,p_a/m_a^2 q a ( s ) = q a ( 0 ) + s p a / m a 2 , while the other particles remain fixed. Since p a 2 = m a 2 p_a^2=m_a^2 p a 2 = m a 2 , the proper-time increment is d s / m a ds/m_a d s / m a , whereas τ a ( s ) − τ a ( 0 ) = s \tau_a(s)-\tau_a(0)=s τ a ( s ) − τ a ( 0 ) = s . Thus τ a \tau_a τ a is m a m_a m a times the proper time, up to an additive constant. The independent Reeb flows describe the reparametrizations of the free world lines.
The individual Liouville forms θ a = ι ∗ ( p a μ d q a μ ) \theta_a=\iota^*(p_{a\mu}dq_a^\mu) θ a = ι ∗ ( p a μ d q a μ ) already satisfy θ a ( R b ) = δ a b \theta_a(R_b)=\delta_{ab} θ a ( R b ) = δ ab . For N > 1 N>1 N > 1 , however, ( θ 1 , … , θ N ) (\theta_1,\ldots,\theta_N) ( θ 1 , … , θ N ) is not N N N -contact: d θ a d\theta_a d θ a vanishes on all tangent directions belonging to particles c ≠ a c\ne a c = a , so its kernel is larger than R \mathcal{R} R . The forms η a = θ a + ∑ c ≠ a ( θ c − d τ c ) \eta_a=\theta_a+\sum_{c\ne a}(\theta_c-d\tau_c) η a = θ a + ∑ c = a ( θ c − d τ c ) supply the required common curvature.
The cotangent lift of a proper orthochronous Lorentz transformation preserves the future mass shells, θ \theta θ , and every τ a \tau_a τ a . It therefore preserves both structures. The corresponding conservation laws are discussed in Examples 4.3 and 5.4.
For interacting particles, on any region with admissible clocks, the restricted Liouville form is a primitive of Ω \Omega Ω . Proposition 3.2 therefore provides a uniform N N N -contact structure locally, with the normalized constraint fields R a R_a R a as Reeb fields. This does not identify their parameters with proper time. Global existence requires both the clock data and a globally normalized primitive, and is not asserted here.
A charge in the field of a magnetic monopole
Let Q = R 3 ∖ { 0 } Q=\mathbb{R}^3\setminus\{0\} Q = R 3 ∖ { 0 } , let m > 0 m>0 m > 0 be the particle mass, and assume e g ≠ 0 eg\ne0 e g = 0 . Fix the magnetic two-form B = − e g sin ϑ d ϑ ∧ d ϕ B=-eg\sin\vartheta\,d\vartheta\wedge d\phi B = − e g sin ϑ d ϑ ∧ d ϕ . Its flux through a sphere about the origin is − 4 π e g -4\pi eg − 4 π e g , so it is closed but not exact. There is consequently no global vector potential on Q Q Q , and no global magnetic Lagrangian of the usual kinetic-plus-potential form on T Q TQ T Q . A global description is obtained by enlarging the configuration space [7 , 35 ] .
Let P 1 = R + × S U ( 2 ) P_1=\mathbb{R}_+\times SU(2) P 1 = R + × S U ( 2 ) , and define π : P 1 → Q \pi:P_1\to Q π : P 1 → Q by π ( r , s ) = r x ^ \pi(r,s)=r\hat{x} π ( r , s ) = r x ^ , where x ^ ⋅ σ = s σ 3 s − 1 \hat{x}\cdot\sigma=s\sigma_3s^{-1} x ^ ⋅ σ = s σ 3 s − 1 and σ 1 , σ 2 , σ 3 \sigma_1,\sigma_2,\sigma_3 σ 1 , σ 2 , σ 3 are the Pauli matrices. The right action s ↦ s e i σ 3 χ s\mapsto se^{i\sigma_3\chi} s ↦ s e i σ 3 χ makes P 1 P_1 P 1 a principal U ( 1 ) U(1) U ( 1 ) -bundle, with generator Z Z Z of period 2 π 2\pi 2 π . The globally defined one-form
A = i e g Tr ( σ 3 s − 1 d s ) \mathcal{A}=ieg\,\operatorname{Tr}(\sigma_3s^{-1}ds) A = i e g Tr ( σ 3 s − 1 d s )
satisfies A ( Z ) = − 2 e g \mathcal{A}(Z)=-2eg A ( Z ) = − 2 e g and d A = π ∗ B d\mathcal{A}=\pi^*B d A = π ∗ B .
The Lagrangian description. For v u ∈ T u P 1 v_u\in T_uP_1 v u ∈ T u P 1 , set x = π ( u ) x=\pi(u) x = π ( u ) and x ˙ = T u π ( v u ) \dot{x}=T_u\pi(v_u) x ˙ = T u π ( v u ) . The Lagrangian is L ( v u ) = 1 2 m ∣ x ˙ ∣ 2 + A u ( v u ) L(v_u)=\frac{1}{2}m|\dot{x}|^2+\mathcal{A}_u(v_u) L ( v u ) = 2 1 m ∣ x ˙ ∣ 2 + A u ( v u ) [7 , 35 ] . It is a function on the full tangent bundle T P 1 TP_1 T P 1 . Writing τ : T P 1 → P 1 \tau:TP_1\to P_1 τ : T P 1 → P 1 and τ Q : T Q → Q \tau_Q:TQ\to Q τ Q : T Q → Q for the projections, its Cartan form and energy are
θ L = m ∑ i = 1 3 x ˙ i d x i + τ ∗ A , E L = 1 2 m ∣ x ˙ ∣ 2 . (1) \theta_L=m\sum_{i=1}^{3}\dot{x}^i dx^i+\tau^*\mathcal{A},\qquad E_L=\frac{1}{2}m|\dot{x}|^2. \tag*{(1)} θ L = m i = 1 ∑ 3 x ˙ i d x i + τ ∗ A , E L = 2 1 m ∣ x ˙ ∣ 2 . (1)
Moreover, ω L = d θ L = ( T π ) ∗ ω B \omega_L=d\theta_L=(T\pi)^*\omega_B ω L = d θ L = ( T π ) ∗ ω B , where
ω B = m ∑ i = 1 3 d x ˙ i ∧ d x i + τ Q ∗ B \omega_B=m\sum_{i=1}^{3}d\dot{x}^{i}\wedge dx^{i}+\tau_Q^*B ω B = m i = 1 ∑ 3 d x ˙ i ∧ d x i + τ Q ∗ B
is symplectic on T Q TQ T Q . Hence ker ω L = ker T ( T π ) \ker\omega_L=\ker T(T\pi) ker ω L = ker T ( T π ) has rank two and is spanned by the vertical lift Z V Z^V Z V and the tangent lift Z T Z^T Z T of Z Z Z . For both lifts we refer to [34 ] , where the tangent lift is called the complete lift . If ϕ t \phi_t ϕ t is the flow of Z Z Z on P 1 P_1 P 1 , then Z T Z^T Z T generates the flow T ϕ t T\phi_t T ϕ t on T P 1 TP_1 T P 1 , while Z V Z^V Z V generates translation in each tangent fibre, v u ↦ v u + t Z ( u ) v_u\mapsto v_u+tZ(u) v u ↦ v u + tZ ( u ) for v u ∈ T u P 1 v_u\in T_uP_1 v u ∈ T u P 1 . The Cartan form satisfies θ L ( Z V ) = 0 \theta_L(Z^V)=0 θ L ( Z V ) = 0 and θ L ( Z T ) = − 2 e g \theta_L(Z^T)=-2eg θ L ( Z T ) = − 2 e g , and the energy is invariant under both lifts.
The Euler–Lagrange equation ı Γ ω L = − d E L \imath_{\Gamma}\omega_L=-dE_L Γ ω L = − d E L , together with the second-order condition, determines Γ \Gamma Γ up to an arbitrary multiple of Z V Z^V Z V . Its projection to T Q TQ T Q is the unique vector field Y Y Y satisfying ı Y ω B = − d E \imath_Y\omega_B=-dE Y ω B = − d E , with E = 1 2 m ∣ x ˙ ∣ 2 E=\frac{1}{2}m|\dot{x}|^2 E = 2 1 m ∣ x ˙ ∣ 2 . Thus the degeneracy represents gauge freedom in the lift, while the physical position and velocity have a well-defined evolution.
The constraint submanifold. Let θ 0 \theta_0 θ 0 be the Liouville form on T ∗ P 1 T^*P_1 T ∗ P 1 and put ω 0 = d θ 0 \omega_0=d\theta_0 ω 0 = d θ 0 . The Legendre map is
F L ( v u ) = ( T u π ) ∗ ( m x ˙ ♭ ) + A u , FL(v_u)=(T_u\pi)^*(m\dot{x}^{\flat})+\mathcal{A}_u, F L ( v u ) = ( T u π ) ∗ ( m x ˙ ♭ ) + A u ,
where x ˙ ♭ \dot{x}^{\flat} x ˙ ♭ is the Euclidean covector associated with x ˙ \dot{x} x ˙ . For p Z ( α u ) = α u ( Z u ) p_Z(\alpha_u)=\alpha_u(Z_u) p Z ( α u ) = α u ( Z u ) , its image is the regular hypersurface
M 1 = { α u ∈ T ∗ P 1 : p Z ( α u ) = − 2 e g } . M_1=\{\alpha_u\in T^*P_1:p_Z(\alpha_u)=-2eg\}. M 1 = { α u ∈ T ∗ P 1 : p Z ( α u ) = − 2 e g } .
The map F L : T P 1 → M 1 FL:TP_1\to M_1 F L : T P 1 → M 1 is a surjective submersion whose fibres are the orbits of Z V Z^V Z V . Conversely, α u − A u \alpha_u-\mathcal{A}_u α u − A u annihilates Z u Z_u Z u , so every α u ∈ M 1 \alpha_u\in M_1 α u ∈ M 1 uniquely determines x ˙ \dot{x} x ˙ by the displayed formula. Thus M 1 ≅ T Q × Q P 1 M_1\cong TQ\times_QP_1 M 1 ≅ T Q × Q P 1 and dim M 1 = 7 \dim M_1=7 dim M 1 = 7 .
Let ρ : M 1 → T Q \rho:M_1\to TQ ρ : M 1 → T Q be the natural projection. We use θ L \theta_L θ L also for the restricted Liouville form on M 1 M_1 M 1 , so that θ L = m ∑ i x ˙ i d x i + A \theta_L=m\sum_i\dot{x}^idx^i+\mathcal{A} θ L = m ∑ i x ˙ i d x i + A and d θ L = ρ ∗ ω B d\theta_L=\rho^*\omega_B d θ L = ρ ∗ ω B . Here and below forms on the bundle are pulled back to M 1 M_1 M 1 without changing notation. The cotangent lift Z ^ \widehat{Z} Z satisfies
L Z ^ ω 0 = − d p Z . \mathcal{L}_{\widehat{Z}}\omega_0=-dp_Z. L Z ω 0 = − d p Z .
In our Hamiltonian convention, Z ^ = − X p Z \widehat{Z}=-X_{p_Z} Z = − X p Z . It is tangent to M 1 M_1 M 1 , is F L FL F L -related to Z T Z^T Z T , and spans ker d θ L \ker d\theta_L ker d θ L . We write Z Z Z also for its restriction to M 1 M_1 M 1 .
Since θ L ( Z ) = − 2 e g \theta_L(Z)=-2eg θ L ( Z ) = − 2 e g , the form λ = − ( 2 e g ) − 1 θ L \lambda=-(2eg)^{-1}\theta_L λ = − ( 2 e g ) − 1 θ L is contact and has Reeb field Z Z Z . Indeed, d λ d\lambda d λ has rank six and λ ( Z ) = 1 \lambda(Z)=1 λ ( Z ) = 1 . The quotient M 1 / U ( 1 ) M_1/U(1) M 1 / U ( 1 ) is the smooth symplectic manifold ( T Q , ω B ) (TQ,\omega_B) ( T Q , ω B ) . The contact description retains a global potential on the enlarged space; the nonzero flux of B B B prevents a global potential for ω B \omega_B ω B on T Q TQ T Q [35 ] .
The energy E = 1 2 m ∣ x ˙ ∣ 2 E=\frac{1}{2}m|\dot{x}|^2 E = 2 1 m ∣ x ˙ ∣ 2 is invariant under Z Z Z , and H = E / ( 2 e g ) H=E/(2eg) H = E / ( 2 e g ) is a good contact Hamiltonian. Its defining equations are equivalent to
ı X H d θ L = − d E , θ L ( X H ) = E . \imath_{X_H}d\theta_L=-dE,\qquad\theta_L(X_H)=E. X H d θ L = − d E , θ L ( X H ) = E .
The first equation projects to the physical magnetic dynamics, while the second fixes the gauge component. With the standard orientation of R 3 \mathbb{R}^3 R 3 , the projected equation is
m x ¨ = e g ∣ x ∣ 3 x × x ˙ . m\ddot{x}=\frac{eg}{|x|^3}x\times\dot{x}. m x ¨ = ∣ x ∣ 3 e g x × x ˙ .
Thus our convention for B B B corresponds to the magnetic field − g x / ∣ x ∣ 3 -g\,x/|x|^3 − g x /∣ x ∣ 3 . There is one physical time parameter, and the Reeb flow changes only the gauge variable. Rotations and their conserved quantities are considered in Example 5.3.
Two monopoles with irrationally related charges
Let x 1 ≠ x 2 x_1 \ne x_2 x 1 = x 2 , put Q = R 3 ∖ { x 1 , x 2 } Q = \mathbb{R}^3 \setminus\{x_1,x_2\} Q = R 3 ∖ { x 1 , x 2 } , and assume m > 0 m > 0 m > 0 and e g 1 g 2 ≠ 0 eg_1g_2 \ne0 e g 1 g 2 = 0 . For j = 1 , 2 j = 1,2 j = 1 , 2 , let P ( j ) → Q P^{(j)} \to Q P ( j ) → Q be the pullback of the preceding monopole bundle under x ↦ x − x j x \mapsto x-x_j x ↦ x − x j . Its points are represented by ( x , s j ) (x,s_j) ( x , s j ) with s j σ 3 s j − 1 = r ^ j ⋅ σ s_j\sigma_3s_j^{-1} = \hat{r}_j \cdot\sigma s j σ 3 s j − 1 = r ^ j ⋅ σ , where r ^ j = ( x − x j ) / ∣ x − x j ∣ \hat{r}_j = (x-x_j)/|x-x_j| r ^ j = ( x − x j ) /∣ x − x j ∣ . The fibre product P = P ( 1 ) × Q P ( 2 ) P = P^{(1)} \times_Q P^{(2)} P = P ( 1 ) × Q P ( 2 ) is a five-dimensional principal T 2 = U ( 1 ) × U ( 1 ) T^2 = U(1) \times U(1) T 2 = U ( 1 ) × U ( 1 ) -bundle. Write Z 1 , Z 2 Z_1,Z_2 Z 1 , Z 2 for its generators, each of period 2 π 2\pi 2 π , and set A j = i e g j Tr ( σ 3 s j − 1 d s j ) \mathcal{A}_j = ie g_j\operatorname{Tr}(\sigma_3s_j^{-1}ds_j) A j = i e g j Tr ( σ 3 s j − 1 d s j ) . Then A j ( Z k ) = − 2 e g j δ j k \mathcal{A}_j(Z_k) = -2eg_j\delta_{jk} A j ( Z k ) = − 2 e g j δ j k and d A j = π ∗ B j d\mathcal{A}_j = \pi^*B_j d A j = π ∗ B j , where B j B_j B j is the magnetic form of the j j j -th monopole.
The global Lagrangian on T P TP T P is L = 1 2 m ∣ x ˙ ∣ 2 + A 1 ( v ) + A 2 ( v ) L = \frac{1}{2}m|\dot{x}|^2 + \mathcal{A}_1(v) + \mathcal{A}_2(v) L = 2 1 m ∣ x ˙ ∣ 2 + A 1 ( v ) + A 2 ( v ) , with x ˙ = T π ( v ) \dot{x} = T\pi(v) x ˙ = T π ( v ) [35 ] . Its Cartan two-form is the pullback of
ω B = m ∑ i = 1 3 d x i ∧ d x i + τ Q ∗ ( B 1 + B 2 ) . \omega_B = m\sum_{i=1}^{3}dx^i \wedge dx^i + \tau_Q^*(B_1+B_2). ω B = m i = 1 ∑ 3 d x i ∧ d x i + τ Q ∗ ( B 1 + B 2 ) .
Consequently its kernel on T P TP T P has rank four, spanned by Z 1 V , Z 2 V , Z 1 T , Z 2 T Z_1^V,Z_2^V,Z_1^T,Z_2^T Z 1 V , Z 2 V , Z 1 T , Z 2 T . The Legendre image is
M 2 = { α u ∈ T ∗ P : α u ( Z 1 ) = − 2 e g 1 , α u ( Z 2 ) = − 2 e g 2 } . M_2 = \{\alpha_u \in T^*P : \alpha_u(Z_1) = -2eg_1,\ \alpha_u(Z_2) = -2eg_2\}. M 2 = { α u ∈ T ∗ P : α u ( Z 1 ) = − 2 e g 1 , α u ( Z 2 ) = − 2 e g 2 } .
The two constraints are independent and Poisson-commute. The Legendre fibres are generated by Z 1 V , Z 2 V Z_1^V,Z_2^V Z 1 V , Z 2 V , and dim M 2 = 8 = 2 ⋅ 3 + 2 \dim M_2 = 8 = 2 \cdot3 + 2 dim M 2 = 8 = 2 ⋅ 3 + 2 .
As before, every point of M 2 M_2 M 2 has the unique representation α u = ( T u π ) ∗ ( m x ˙ b ) + A 1 , u + A 2 , u \alpha_u = (T_u\pi)^*(m\dot{x}^{\,b}) + \mathcal{A}_{1,u} + \mathcal{A}_{2,u} α u = ( T u π ) ∗ ( m x ˙ b ) + A 1 , u + A 2 , u . Thus M 2 ≅ T Q × Q P M_2 \cong TQ \times_Q P M 2 ≅ T Q × Q P , and the projection ρ : M 2 → T Q \rho: M_2 \to TQ ρ : M 2 → T Q is a principal T 2 T^2 T 2 -bundle. For the restricted Liouville form,
θ L = m ∑ i x ˙ i d x i + A 1 + A 2 , d θ L = ρ ∗ ω B . \theta_L = m\sum_i \dot{x}^{\,i}dx^i + \mathcal{A}_1+\mathcal{A}_2,\qquad d\theta_L = \rho^*\omega_B. θ L = m i ∑ x ˙ i d x i + A 1 + A 2 , d θ L = ρ ∗ ω B .
Writing Z j Z_j Z j also for the restricted cotangent lifts, we have ker d θ L = R = span { Z 1 , Z 2 } \ker d\theta_L = \mathcal{R} = \operatorname{span}\{Z_1,Z_2\} ker d θ L = R = span { Z 1 , Z 2 } and θ L ( Z j ) = − 2 e g j \theta_L(Z_j) = -2eg_j θ L ( Z j ) = − 2 e g j .
Physical reduction and the intermediate contact quotient. The full torus quotient is always smooth: M 2 / T 2 ≅ T Q M_2/T^2 \cong TQ M 2 / T 2 ≅ T Q . It carries the physical dynamics ι Y ω B = − d E \iota_Y\omega_B = -dE ι Y ω B = − d E , with E = 1 2 m ∣ x ˙ ∣ 2 E = \frac{1}{2}m|\dot{x}|^2 E = 2 1 m ∣ x ˙ ∣ 2 . In particular,
m x ¨ = e ∑ j = 1 2 g j ∣ x − x j ∣ 3 ( x − x j ) × x ˙ . m\ddot{x} = e\sum_{j=1}^{2}\frac{g_j}{|x-x_j|^3}(x-x_j)\times\dot{x}. m x ¨ = e j = 1 ∑ 2 ∣ x − x j ∣ 3 g j ( x − x j ) × x ˙ .
There is a single physical time parameter, and functions of ( x , x ˙ ) (x,\dot{x}) ( x , x ˙ ) are gauge-invariant observables. A different question is whether the lifted potential θ L \theta_L θ L descends to a seven-dimensional contact manifold. Its characteristic distribution is
ker θ L ∩ ker d θ L = span { W } , W = g 2 Z 1 − g 1 Z 2 . \ker\theta_L \cap\ker d\theta_L = \operatorname{span}\{W\},\qquad W = g_2Z_1-g_1Z_2. ker θ L ∩ ker d θ L = span { W } , W = g 2 Z 1 − g 1 Z 2 .
We have θ L ( W ) = 0 \theta_L(W) = 0 θ L ( W ) = 0 and L W θ L = 0 \mathcal{L}_W\theta_L = 0 L W θ L = 0 . If g 1 / g 2 ∈ Q g_1/g_2 \in\mathbb{Q} g 1 / g 2 ∈ Q , this line generates a closed circle subgroup K ⊂ T 2 K \subset T^2 K ⊂ T 2 . The quotient M 2 / K M_2/K M 2 / K is a smooth principal circle bundle over T Q TQ T Q , and θ L \theta_L θ L descends to a contact form: its exterior derivative has rank six, and the induced one-form is nonzero on the remaining vertical direction.
If g 1 / g 2 ∉ Q g_1/g_2 \notin\mathbb{Q} g 1 / g 2 ∈ / Q , the W W W -orbits are dense in the fibre tori. The orbit space is not a smooth Hausdorff manifold, so this intermediate contact quotient is unavailable [35 ] . Equivalently, the magnetic periods − 4 π e g 1 -4\pi eg_1 − 4 π e g 1 and − 4 π e g 2 -4\pi eg_2 − 4 π e g 2 cannot both be integral multiples of one nonzero common scale, as would be required for a single principal circle bundle with connection curvature proportional to ω B \omega_B ω B . This obstruction concerns a global description retaining the potential. It does not obstruct the full symplectic quotient M 2 / T 2 M_2/T^2 M 2 / T 2 .
The non-uniform 2-contact structure. The separate magnetic contributions define additional forms on the unreduced space.
Proposition 3.3. The one-forms
λ j = − 1 2 e g j ( m ∑ i x ˙ i d x i + A j ) , j = 1 , 2 , \lambda_j=-\frac{1}{2eg_j}\left(m\sum_i \dot{x}^i\,dx^i+\mathcal{A}_j\right),\qquad j=1,2, λ j = − 2 e g j 1 ( m i ∑ x ˙ i d x i + A j ) , j = 1 , 2 , define a non-uniform 2-contact structure on M 2 M_2 M 2 with Reeb fields Z 1 , Z 2 Z_1,Z_2 Z 1 , Z 2 . Moreover, with ξ = ker λ 1 ∩ ker λ 2 \xi=\ker\lambda_1\cap\ker\lambda_2 ξ = ker λ 1 ∩ ker λ 2 , the data
( M 2 , ( λ 1 , λ 2 ) ⊕ ( d λ 1 , d λ 2 , d θ L ) , R ⊕ ξ ) \left(M_2,(\lambda_1,\lambda_2)\oplus(d\lambda_1,d\lambda_2,d\theta_L),\mathcal{R}\oplus\xi\right) ( M 2 , ( λ 1 , λ 2 ) ⊕ ( d λ 1 , d λ 2 , d θ L ) , R ⊕ ξ ) equip M 2 M_2 M 2 with a 2-structure involving three two-forms.
Proof. The kinetic term vanishes on the gauge directions, so λ j ( Z k ) = δ j k \lambda_j(Z_k)=\delta_{jk} λ j ( Z k ) = δ j k . Thus the λ j \lambda_j λ j are independent and ξ \xi ξ is a complement of R = ker T ρ \mathcal{R}=\ker T\rho R = ker T ρ . Write
ω j = m ∑ i d x ˙ i ∧ d x i + τ Q ∗ B j . \omega_j=m\sum_i d\dot{x}^i\wedge dx^i+\tau_Q^*B_j. ω j = m i ∑ d x ˙ i ∧ d x i + τ Q ∗ B j .
Each ω j \omega_j ω j is symplectic: pairing a vector in its kernel first with all velocity directions forces its position component to vanish, and pairing with all position directions then forces its velocity component to vanish. Since d λ j = − ( 2 e g j ) − 1 ρ ∗ ω j d\lambda_j=-(2eg_j)^{-1}\rho^*\omega_j d λ j = − ( 2 e g j ) − 1 ρ ∗ ω j , we obtain ker d λ j = R \ker d\lambda_j=\mathcal{R} ker d λ j = R and non-degeneracy on ξ \xi ξ . The same argument applies to d θ L = ρ ∗ ω B d\theta_L=\rho^*\omega_B d θ L = ρ ∗ ω B .
To verify non-uniformity, suppose that d λ 1 = d λ 2 d\lambda_1=d\lambda_2 d λ 1 = d λ 2 . Pullback by a surjective submersion is injective, so g 1 − 1 ω 1 = g 2 − 1 ω 2 g_1^{-1}\omega_1=g_2^{-1}\omega_2 g 1 − 1 ω 1 = g 2 − 1 ω 2 . Comparison of the velocity–position terms gives g 1 = g 2 g_1=g_2 g 1 = g 2 , hence B 1 = B 2 B_1=B_2 B 1 = B 2 . This is impossible: on a small sphere about x 1 x_1 x 1 not enclosing x 2 x_2 x 2 , their fluxes are − 4 π e g 1 -4\pi eg_1 − 4 π e g 1 and zero, respectively.
Up to normalization, λ j \lambda_j λ j is the Cartan form for the kinetic energy and the j j j -th magnetic contribution alone. The physical Cartan form instead contains the kinetic term once and both magnetic contributions. Its derivative d θ L d\theta_L d θ L governs the total motion; it is not to be replaced by either d λ j d\lambda_j d λ j . This is the reason for retaining three two-forms in the common framework of Section 2. The physical dynamics on M 2 M_2 M 2 is specified presymplectically, up to the two gauge directions, by ı Γ d θ L = − d E \imath_\Gamma d\theta_L=-dE Γ d θ L = − d E . We do not apply the uniform Hamiltonian definition to this non-uniform structure.
The next obstruction concerns all uniform structures having the prescribed gauge fields as Reeb fields, not only the particular forms above.
Proposition 3.4. There is no uniform 2-contact structure ( η 1 , η 2 ) (\eta_1,\eta_2) ( η 1 , η 2 ) on M 2 M_2 M 2 whose Reeb fields are Z 1 , Z 2 Z_1,Z_2 Z 1 , Z 2 .
Proof. Suppose such a structure exists. Then β = η 1 − η 2 \beta=\eta_1-\eta_2 β = η 1 − η 2 is closed and β ( Z 1 ) = 1 \beta(Z_1)=1 β ( Z 1 ) = 1 . Choose a small sphere S ⊂ Q S\subset Q S ⊂ Q about x 1 x_1 x 1 not enclosing x 2 x_2 x 2 , and let
N = { ( x , x ˙ , u ) ∈ T Q × Q P : x ∈ S , x ˙ = 0 } ⊂ M 2 . N=\{(x,\dot{x},u)\in TQ\times_QP:x\in S,\ \dot{x}=0\}\subset M_2. N = {( x , x ˙ , u ) ∈ T Q × Q P : x ∈ S , x ˙ = 0 } ⊂ M 2 .
Over S S S , the first circle bundle is the Hopf bundle, whereas the second is trivial because it extends over the ball bounded by S S S . Hence N ≅ S 3 × S 1 N\cong S^3\times S^1 N ≅ S 3 × S 1 . An orbit C C C of Z 1 Z_1 Z 1 is a Hopf fibre in S 3 × { p t } S^3\times\{\mathrm{pt}\} S 3 × { pt } and is null-homologous in N N N . Closedness therefore implies ∮ C β = 0 \oint_C\beta=0 ∮ C β = 0 . On the other hand, the period of Z 1 Z_1 Z 1 is 2 π 2\pi 2 π , so
∮ C β = ∫ 0 2 π β ( Z 1 ) d x = 2 π , \oint_C\beta=\int_0^{2\pi}\beta(Z_1)\,dx=2\pi, ∮ C β = ∫ 0 2 π β ( Z 1 ) d x = 2 π ,
a contradiction.
Remark 3.2. The two obstructions have different meanings. Irrationality of g 1 / g 2 g_1/g_2 g 1 / g 2 prevents the intermediate contact quotient associated with θ L \theta_L θ L . The topological obstruction in Proposition 3.4 is independent of that ratio and prevents a uniform structure with the prescribed Reeb fields. It does not assert nonexistence of every possible uniform structure on the underlying manifold. The constructed non-uniform structure therefore lies outside the transformation theory of Section 5 and motivates the extension discussed in Section 6.
Remark 3.3 (Relation to the principal-bundle construction). The Cartan forms of the monopole examples can be related explicitly to the construction of [1 Sec. V] . Let π : P → Q \pi:P\to Q π : P → Q be a principal G G G -bundle, let γ ∈ Ω 1 ( P ; g ) \gamma\in\Omega^1(P;\mathfrak{g}) γ ∈ Ω 1 ( P ; g ) be a principal connection, and write τ P : T ∗ P → P \tau_P:T^*P\to P τ P : T ∗ P → P . For an equivariant map f : T ∗ P → g ∗ f:T^*P\to\mathfrak{g}^* f : T ∗ P → g ∗ , their Eqs. (5.5)–(5.6) use the generalized Liouville form
Θ f = ⟨ f , τ P ∗ γ ⟩ , d Θ f = ⟨ d f ∧ τ P ∗ γ ⟩ + ⟨ f , τ P ∗ d γ ⟩ . \Theta_f=\langle f,\tau_P^*\gamma\rangle,\qquad d\Theta_f=\langle df\wedge\tau_P^*\gamma\rangle+\langle f,\tau_P^*d\gamma\rangle. Θ f = ⟨ f , τ P ∗ γ ⟩ , d Θ f = ⟨ df ∧ τ P ∗ γ ⟩ + ⟨ f , τ P ∗ d γ ⟩ .
Here the brackets denote the natural pairing and its extension to differential forms. Writing F γ = d γ + 1 2 [ γ , γ ] F_\gamma=d\gamma+\frac{1}{2}[\gamma,\gamma] F γ = d γ + 2 1 [ γ , γ ] expresses the second term in terms of the curvature and the Lie-algebra bracket. Adding a horizontal Liouville term leads to the coupled cotangent-bundle structures studied in [1 Sec. V.C] .
For our examples, G = T q G=T^q G = T q with q = 1 q=1 q = 1 or 2 2 2 . Set μ a = − 2 e g a \mu_a=-2eg_a μ a = − 2 e g a and γ a = A a / μ a \gamma^a=\mathcal{A}_a/\mu_a γ a = A a / μ a , taking g 1 = g g_1=g g 1 = g and A 1 = A \mathcal{A}_1=\mathcal{A} A 1 = A when q = 1 q=1 q = 1 . Then γ a ( Z b ) = δ a b \gamma^a(Z_b)=\delta_{ab} γ a ( Z b ) = δ ab , so the γ a \gamma^a γ a are the components of a principal torus connection, and d γ a = π ∗ ( B a / μ a ) d\gamma^a=\pi^*(B_a/\mu_a) d γ a = π ∗ ( B a / μ a ) . The cotangent momenta f a ( α u ) = α u ( Z a ) f_a(\alpha_u)=\alpha_u(Z_a) f a ( α u ) = α u ( Z a ) have the fixed values μ a \mu_a μ a on M 1 M_1 M 1 or M 2 M_2 M 2 . With forms pulled back to the constraint manifold and with θ k i n = m ∑ i x ˙ i d x i \theta_{\rm kin}=m\sum_i\dot{x}^i\,dx^i θ kin = m ∑ i x ˙ i d x i , the restricted Liouville form is
θ L = θ k i n + ∑ a = 1 q μ a γ a , d θ L = d θ k i n + ∑ a = 1 q μ a d γ a . \theta_L=\theta_{\rm kin}+\sum_{a=1}^{q}\mu_a\gamma^a,\qquad d\theta_L=d\theta_{\rm kin}+\sum_{a=1}^{q}\mu_a\,d\gamma^a. θ L = θ kin + a = 1 ∑ q μ a γ a , d θ L = d θ kin + a = 1 ∑ q μ a d γ a .
Thus the magnetic contribution is precisely the pairing of the fixed momentum with the curvature. The additional forms used here are
λ a = γ a + μ a − 1 θ k i n , d λ a = μ a − 1 ( d θ k i n + d A a ) . \lambda_a=\gamma^a+\mu_a^{-1}\theta_{\rm kin},\qquad d\lambda_a=\mu_a^{-1}(d\theta_{\rm kin}+d\mathcal{A}_a). λ a = γ a + μ a − 1 θ kin , d λ a = μ a − 1 ( d θ kin + d A a ) .
For one monopole this is the contact normalization λ = − ( 2 e g ) − 1 θ L \lambda=-(2eg)^{-1}\theta_L λ = − ( 2 e g ) − 1 θ L . For two monopoles it gives the separate forms of Proposition 3.3, whereas the physical form θ L \theta_L θ L contains both magnetic contributions. This identifies the principal-bundle origin of the physical presymplectic form and the additional data retained by the q q q -contact description.
The bundle connection γ \gamma γ generally has nonzero curvature. It must be distinguished from the flat horizontal projector in Remark 2.3. The principal-bundle construction therefore does not by itself supply the closed one-forms required for a q q q -cosymplectic structure.
Canonical and canonoid transformations on q-cosymplectic manifolds
Throughout this section, ( M , Ω , λ ⃗ ) (M,\Omega,\vec{\lambda}) ( M , Ω , λ ) is a q q q -cosymplectic of dimension 2 n + q 2n+q 2 n + q . We retain the conventions of Section 2, in particular { f , g } = X g ( f ) \{f,g\}=X_g(f) { f , g } = X g ( f ) . All statements about flows are local unless completeness is explicitly assumed. We distinguish preservation of the full data ( Ω , λ ⃗ ) (\Omega,\vec{\lambda}) ( Ω , λ ) from preservation of the symplectic geometry on the leaves of ξ \xi ξ . For canonoid transformations we also specify which parts of the Reeb data are fixed.
Directional evolution and constants of motion
For H ∈ C ∞ ( M ) H \in C^{\infty}(M) H ∈ C ∞ ( M ) and a constant vector a ⃗ = ( a 1 , … , a q ) ∈ R q \vec{a}=(a_1,\ldots,a_q)\in\mathbb{R}^q a = ( a 1 , … , a q ) ∈ R q , define
E H a ⃗ = X H + ∑ i = 1 q a i R i . E^{\vec{a}}_H=X_H+\sum_{i=1}^{q}a_iR_i. E H a = X H + i = 1 ∑ q a i R i .
When a ⃗ \vec{a} a is fixed we also write E H E_H E H . This is the unique field satisfying ι E H a ⃗ Ω = d H − ∑ i R i ( H ) λ i \iota_{E^{\vec{a}}_H}\Omega=dH-\sum_iR_i(H)\lambda_i ι E H a Ω = d H − ∑ i R i ( H ) λ i and λ i ( E H a ⃗ ) = a i \lambda_i(E^{\vec{a}}_H)=a_i λ i ( E H a ) = a i . In Darboux coordinates its integral curves obey
q ˙ i = ∂ H ∂ p i , p ˙ i = − ∂ H ∂ q i , t ˙ j = a j . \dot{q}^{\,i}=\frac{\partial H}{\partial p_i},\qquad\dot{p}_i=-\frac{\partial H}{\partial q^i},\qquad\dot{t}_j=a_j. q ˙ i = ∂ p i ∂ H , p ˙ i = − ∂ q i ∂ H , t ˙ j = a j .
For q = 1 q=1 q = 1 and a 1 = 1 a_1=1 a 1 = 1 , these are the usual non-autonomous Hamilton equations. For several clock variables, this construction selects a one-parameter evolution with prescribed clock rates, within the q q q -cosymplectic framework [23 ] .
Remark 4.1. The fields E H a ⃗ E^{\vec{a}}_H E H a do not generally form a commuting family. Proposition 2.3 gives
[ E H a ⃗ , E H b ⃗ ] = ∑ i = 1 q ( a i − b i ) X R i ( H ) . [E^{\vec{a}}_H,E^{\vec{b}}_H]=\sum_{i=1}^{q}(a_i-b_i)X_{R_i(H)}. [ E H a , E H b ] = i = 1 ∑ q ( a i − b i ) X R i ( H ) .
For example, if Ω = d x ∧ d p \Omega=dx\wedge dp Ω = d x ∧ d p , λ i = d t i \lambda_i=dt_i λ i = d t i , q = 2 q=2 q = 2 , and H = t 1 p H=t_1p H = t 1 p , then [ E H e ⃗ 1 , E H e ⃗ 2 ] = ∂ x [E^{\vec{e}_1}_H,E^{\vec{e}_2}_H]=\partial_x [ E H e 1 , E H e 2 ] = ∂ x . Thus these directional evolutions should be distinguished from a compatible multi-parameter evolution generated by commuting fields, such as the Reeb fields constructed in § 3.1 . The entire family E H a ⃗ E^{\vec{a}}_H E H a commutes precisely when X R i ( H ) = 0 X_{R_i(H)}=0 X R i ( H ) = 0 for all i i i .
A function f f f is a constant of motion along the a ⃗ \vec{a} a -direction if E H a ⃗ ( f ) = 0 E^{\vec{a}}_H(f)=0 E H a ( f ) = 0 . Equivalently, { f , H } + ∑ i a i R i ( f ) = 0 \{f,H\}+\sum_i a_iR_i(f)=0 { f , H } + ∑ i a i R i ( f ) = 0 .
Proposition 4.1. A function f f f is a constant of motion along every a ⃗ \vec{a} a -direction if and only if X H ( f ) = 0 X_H(f)=0 X H ( f ) = 0 and R i ( f ) = 0 R_i(f)=0 R i ( f ) = 0 for every i i i .
Proof. Taking a ⃗ = 0 \vec{a}=0 a = 0 gives X H ( f ) = 0 X_H(f)=0 X H ( f ) = 0 . Taking a ⃗ = e ⃗ i \vec{a}=\vec{e}_i a = e i then gives R i ( f ) = 0 R_i(f)=0 R i ( f ) = 0 . Conversely, these conditions imply E H a ⃗ ( f ) = 0 E^{\vec{a}}_H(f)=0 E H a ( f ) = 0 for every a ⃗ \vec{a} a .
This condition also makes f f f invariant under every field X H + ∑ i a i R i X_H+\sum_i a_iR_i X H + ∑ i a i R i with smooth, rather than constant, coefficients a i a_i a i . It is stronger than conservation for one chosen direction.
Example 4.1 (Relativistic particles: observables). On a region of the constraint surface equipped with the clocks of § 3.3 , take H = 0 H=0 H = 0 . Then E 0 a ⃗ = ∑ a a a R a E^{\vec{a}}_0=\sum_a a_aR_a E 0 a = ∑ a a a R a is tangent to the characteristic foliation. By Proposition 4.1, conservation for every direction is equivalent to R a ( f ) = 0 R_a(f)=0 R a ( f ) = 0 for all a a a . Such functions are constant along characteristic leaves and descend to observables on a smooth reduced phase space whenever that quotient exists [8 ] . For the free particles of Example 3.1, the clock parameters are mass multiples of proper time. In that case these observables are independent of the chosen synchronization of the world lines. No proper-time interpretation is assumed for general interacting constraints.
Canonical transformations and Noether symmetries
Definition 4.1. A canonical transformation is a diffeomorphism ψ \psi ψ satisfying ψ ∗ Ω = Ω \psi^{*}\Omega= \Omega ψ ∗ Ω = Ω and ψ ∗ λ i = λ i \psi^{*}\lambda_i = \lambda_i ψ ∗ λ i = λ i for every i i i . A vector field is canonical if its local flow consists of canonical transformations.
The following observation isolates preservation of the Reeb frame and the horizontal distribution from preservation of Ω \Omega Ω .
Proposition 4.2. For a vector field X X X , the following are equivalent:
L X λ i = 0 \mathcal{L}_X\lambda_i = 0 L X λ i = 0 for every i i i and ι R i L X Ω = 0 \iota_{R_i}\mathcal{L}_X\Omega= 0 ι R i L X Ω = 0 for every i i i ;
[ X , R i ] = 0 [X,R_i] = 0 [ X , R i ] = 0 for every i i i and [ X , Γ ( ξ ) ] ⊂ Γ ( ξ ) [X,\Gamma(\xi)] \subset\Gamma(\xi) [ X , Γ ( ξ )] ⊂ Γ ( ξ ) .
Proof. Assume the first condition. From λ j ( R i ) = δ j i \lambda_j(R_i) = \delta_{ji} λ j ( R i ) = δ j i and L X λ j = 0 \mathcal{L}_X\lambda_j = 0 L X λ j = 0 we obtain λ j ( [ X , R i ] ) = 0 \lambda_j([X,R_i]) = 0 λ j ([ X , R i ]) = 0 . Also,
ι [ X , R i ] Ω = L X ( ι R i Ω ) − ι R i L X Ω = 0 \iota_{[X,R_i]}\Omega= \mathcal{L}_X(\iota_{R_i}\Omega) - \iota_{R_i}\mathcal{L}_X\Omega= 0 ι [ X , R i ] Ω = L X ( ι R i Ω ) − ι R i L X Ω = 0
Thus [ X , R i ] ∈ ξ ∩ R = { 0 } [X,R_i] \in\xi\cap\mathcal{R} = \{0\} [ X , R i ] ∈ ξ ∩ R = { 0 } . If Y ∈ Γ ( ξ ) Y \in\Gamma(\xi) Y ∈ Γ ( ξ ) , then λ j ( [ X , Y ] ) = − ( L X λ j ) ( Y ) = 0 \lambda_j([X,Y]) = -(\mathcal{L}_X\lambda_j)(Y) = 0 λ j ([ X , Y ]) = − ( L X λ j ) ( Y ) = 0 , so [ X , Y ] ∈ Γ ( ξ ) [X,Y] \in\Gamma(\xi) [ X , Y ] ∈ Γ ( ξ ) .
Conversely, assume the second condition. The form L X λ j \mathcal{L}_X\lambda_j L X λ j vanishes on each R i R_i R i because [ X , R i ] = 0 [X,R_i] = 0 [ X , R i ] = 0 , and on ξ \xi ξ because [ X , Γ ( ξ ) ] ⊂ Γ ( ξ ) [X,\Gamma(\xi)] \subset\Gamma(\xi) [ X , Γ ( ξ )] ⊂ Γ ( ξ ) . Hence L X λ j = 0 \mathcal{L}_X\lambda_j = 0 L X λ j = 0 . The same contraction identity gives ι R i L X Ω = 0 \iota_{R_i}\mathcal{L}_X\Omega= 0 ι R i L X Ω = 0 .
These equivalent conditions are weaker than canonicity. For example, with Ω = d x ∧ d p \Omega= dx \wedge dp Ω = d x ∧ d p and λ i = d t i \lambda_i = dt_i λ i = d t i , the field X = x ∂ x X = x\partial_x X = x ∂ x satisfies both conditions but L X Ω = Ω \mathcal{L}_X\Omega= \Omega L X Ω = Ω . In Darboux coordinates the conditions mean that X = Y ( q , p ) + ∑ i c i ∂ t i X = Y(q,p) + \sum_i c_i\partial_{t_i} X = Y ( q , p ) + ∑ i c i ∂ t i , where Y Y Y is horizontal and the c i c_i c i are constants; no symplectic condition on Y Y Y is imposed.
Lemma 4.1. A vector field X X X is canonical if and only if L X Ω = 0 \mathcal{L}_X\Omega= 0 L X Ω = 0 and L X λ i = 0 \mathcal{L}_X\lambda_i = 0 L X λ i = 0 for every i i i . Locally, every such field has the form
X = X f + ∑ i = 1 q c i R i , R i ( f ) = 0 , c i ∈ R . X = X_f + \sum_{i=1}^{q} c_iR_i, \qquad R_i(f) = 0, \qquad c_i \in\mathbb{R}. X = X f + i = 1 ∑ q c i R i , R i ( f ) = 0 , c i ∈ R . Moreover, for every H H H , [ X , X H ] = X X ( H ) [X,X_H] = X_{X(H)} [ X , X H ] = X X ( H ) .
Proof. For the local flow ψ s \psi_s ψ s of X X X , d d s ψ s ∗ α = ψ s ∗ L X α \frac{d}{ds}\psi_s^*\alpha=\psi_s^*\mathcal{L}_X\alpha d s d ψ s ∗ α = ψ s ∗ L X α . Applying this identity to Ω \Omega Ω and the λ i \lambda_i λ i proves the first assertion.
Since the λ i \lambda_i λ i are closed, L X λ i = d ( λ i ( X ) ) = 0 \mathcal{L}_X\lambda_i=d(\lambda_i(X))=0 L X λ i = d ( λ i ( X )) = 0 . Thus c i = λ i ( X ) c_i=\lambda_i(X) c i = λ i ( X ) is locally constant. The one-form α = ι X Ω \alpha=\iota_X\Omega α = ι X Ω is closed and annihilates every R i R_i R i . Locally α = d f \alpha=df α = df , and therefore R i ( f ) = 0 R_i(f)=0 R i ( f ) = 0 . The horizontal field X − ∑ i c i R i X-\sum_i c_iR_i X − ∑ i c i R i is then X f X_f X f .
By Proposition 4.2, [ X , R i ] = 0 [X,R_i]=0 [ X , R i ] = 0 . Consequently,
ι [ X , X H ] Ω = L X ( ι X H Ω ) − ι X H L X Ω = d ( X ( H ) ) − ∑ i R i ( X ( H ) ) λ i . \iota_{[X,X_H]}\Omega=\mathcal{L}_X(\iota_{X_H}\Omega)-\iota_{X_H}\mathcal{L}_X\Omega
=\mathrm{d}(X(H))-\sum_iR_i(X(H))\lambda_i. ι [ X , X H ] Ω = L X ( ι X H Ω ) − ι X H L X Ω = d ( X ( H )) − i ∑ R i ( X ( H )) λ i .
Also λ i ( [ X , X H ] ) = 0 \lambda_i([X,X_H])=0 λ i ([ X , X H ]) = 0 , so uniqueness of the Hamiltonian field proves the last assertion.
The local normal form displays the effect of the additional data: arbitrary components along the presymplectic kernel are replaced by constant Reeb translations. Globally, the existence of a function f f f with d f = ι X Ω df=\iota_X\Omega df = ι X Ω requires this closed one-form to be exact.
Definition 4.2. An infinitesimal symmetry of ( M , Ω , λ ⃗ , H ) (M,\Omega,\vec{\lambda},H) ( M , Ω , λ , H ) is a field V V V satisfying L V Ω = 0 \mathcal{L}_V\Omega=0 L V Ω = 0 , L V λ i = 0 \mathcal{L}_V\lambda_i=0 L V λ i = 0 for every i i i , and V ( H ) = 0 V(H)=0 V ( H ) = 0 . It is special if also λ i ( V ) = 0 \lambda_i(V)=0 λ i ( V ) = 0 for every $i.
Proposition 4.3. A field V V V is an infinitesimal symmetry if and only if its local flow is canonical and preserves H H H . Every such field satisfies [ V , E H a ⃗ ] = 0 [V,E_H^{\vec{a}}]=0 [ V , E H a ] = 0 for every constant a ⃗ \vec{a} a .
Proof. The equivalence follows from Lemma 4.1 and d d s ϕ s ∗ H = ϕ s ∗ ( V ( H ) ) \frac{d}{ds}\phi_s^*H=\phi_s^*(V(H)) d s d ϕ s ∗ H = ϕ s ∗ ( V ( H )) . The same lemma gives [ V , X H ] = X V ( H ) = 0 [V,X_H]=X_{V(H)}=0 [ V , X H ] = X V ( H ) = 0 , and Proposition 4.2 gives [ V , R i ] = 0 [V,R_i]=0 [ V , R i ] = 0 . Thus [ V , E H a ⃗ ] = 0 [V,E_H^{\vec{a}}]=0 [ V , E H a ] = 0 .
Geometric symmetry and Noether-type results in related settings are discussed in [10 , 4 , 5 , 31 ] . Here the symmetry preserves the fixed forms and the Hamiltonian separately, and the conserved function is required to be independent of every Reeb direction. The following statement is the corresponding symplectic Noether argument with these additional compatibility conditions.
Theorem 4.1 (Noether theorem). 1. If f f f is a constant of motion along every a ⃗ \vec{a} a -direction, then X f X_f X f is a special infinitesimal symmetry.
Conversely, every special infinitesimal symmetry V V V locally has the form V = X f V=X_f V = X f , where R i ( f ) = 0 R_i(f)=0 R i ( f ) = 0 for every i i i and f f f is a constant of motion along every a ⃗ \vec{a} a -direction. Such an f f f can be chosen globally if and only if the closed one-form ι V Ω \iota_V\Omega ι V Ω is exact.
Proof. For the first assertion, Proposition 4.1 gives R i ( f ) = 0 R_i(f)=0 R i ( f ) = 0 and X H ( f ) = 0 X_H(f)=0 X H ( f ) = 0 . Hence ι X f Ω = d f \iota_{X_f}\Omega=df ι X f Ω = df and L X f Ω = 0 \mathcal{L}_{X_f}\Omega=0 L X f Ω = 0 . Horizontality and closedness of the λ i \lambda_i λ i give L X f λ i = 0 \mathcal{L}_{X_f}\lambda_i=0 L X f λ i = 0 . Finally, X f ( H ) = { H , f } = − { f , H } = − X H ( f ) = 0 X_f(H)=\{H,f\}=-\{f,H\}=-X_H(f)=0 X f ( H ) = { H , f } = − { f , H } = − X H ( f ) = 0 .
Conversely, L V Ω = 0 \mathcal{L}_V\Omega=0 L V Ω = 0 implies d ( ι V Ω ) = 0 d(\iota_V\Omega)=0 d ( ι V Ω ) = 0 . Choose a local primitive f f f , so ι V Ω = d f \iota_V\Omega=df ι V Ω = df . Evaluating on R i R_i R i gives R i ( f ) = 0 R_i(f)=0 R i ( f ) = 0 . Together with λ i ( V ) = 0 \lambda_i(V)=0 λ i ( V ) = 0 , this yields V = X f V=X_f V = X f . Furthermore,
X H ( f ) = d f ( X H ) = Ω ( V , X H ) = − d H ( V ) + ∑ i R i ( H ) λ i ( V ) = 0. X_H(f)=df(X_H)=\Omega(V,X_H)=-dH(V)+\sum_iR_i(H)\lambda_i(V)=0. X H ( f ) = df ( X H ) = Ω ( V , X H ) = − d H ( V ) + i ∑ R i ( H ) λ i ( V ) = 0.
Proposition 4.1 proves the conservation statement. The same argument works globally exactly when ι V Ω \iota_V\Omega ι V Ω has a global primitive.
Example 4.2. On R 6 \mathbb{R}^6 R 6 with coordinates ( q 1 , q 2 , p 1 , p 2 , t 1 , t 2 ) (q^1,q^2,p_1,p_2,t_1,t_2) ( q 1 , q 2 , p 1 , p 2 , t 1 , t 2 ) , let Ω = d q 1 ∧ d p 1 + d q 2 ∧ d p 2 \Omega=dq^1\wedge dp_1+dq^2\wedge dp_2 Ω = d q 1 ∧ d p 1 + d q 2 ∧ d p 2 , λ i = d t i \lambda_i=dt_i λ i = d t i , and H = 1 2 ( p 1 2 + p 2 2 ) + t 1 q 1 H=\frac{1}{2}(p_1^2+p_2^2)+t_1q^1 H = 2 1 ( p 1 2 + p 2 2 ) + t 1 q 1 . Then f = p 2 f=p_2 f = p 2 is Hamiltonian by X H , R 1 , R 2 X_H,R_1,R_2 X H , R 1 , R 2 . Its Hamiltonian field X f = ∂ q 2 X_f=\partial_{q^2} X f = ∂ q 2 is therefore a special infinitesimal symmetry.
Example 4.3 (Relativistic particles: Lorentz symmetry and angular momentum). Consider the free-particle structure of Example 3.1, with H = 0 H=0 H = 0 . Write θ M = ι ∗ θ \theta_M=\iota^*\theta θ M = ι ∗ θ . Let V V V be the restriction of the cotangent lift of an infinitesimal Lorentz transformation, represented by a matrix ε \varepsilon ε satisfying ε μ ν = − ε ν μ \varepsilon_{\mu\nu}=-\varepsilon_{\nu\mu} ε μν = − ε ν μ . It is tangent to the mass shells and satisfies L V θ M = 0 \mathcal{L}_V\theta_M=0 L V θ M = 0 and V ( τ a ) = 0 V(\tau_a)=0 V ( τ a ) = 0 . Thus V V V is a special infinitesimal symmetry.
Cartan’s formula gives ι V Ω = − d ( θ M ( V ) ) \iota_V\Omega=-d(\theta_M(V)) ι V Ω = − d ( θ M ( V )) , so the corresponding global Hamiltonian is
f = − θ M ( V ) = − ∑ a p a μ ε μ ν q a ν . f=-\theta_M(V)=-\sum_a p_{a\mu}\varepsilon^\mu{}_\nu q_a^\nu. f = − θ M ( V ) = − a ∑ p a μ ε μ ν q a ν . For J μ ν = ∑ a ( q a μ p a ν − q a ν p a μ ) J^{\mu\nu}=\sum_a(q_a^\mu p_a^\nu-q_a^\nu p_a^\mu) J μν = ∑ a ( q a μ p a ν − q a ν p a μ ) , this is f = 1 2 ε μ ν J μ ν f=\frac{1}{2}\varepsilon_{\mu\nu}J^{\mu\nu} f = 2 1 ε μν J μν with the stated conventions. Since ε \varepsilon ε is arbitrary, every component of J J J is annihilated by each R a R_a R a . The total relativistic angular momentum is therefore conserved independently of the synchronization of the free world lines.
Locally Hamiltonian vector fields and generalized canonical transformations
The Poisson structure of Proposition 2.3 gives a leafwise notion of Hamiltonian vector field. This notion need not preserve the full q q q -cosymplectic structure.
Definition 4.3. A field V V V is locally Hamiltonian if λ i ( V ) = 0 \lambda_i(V)=0 λ i ( V ) = 0 for every i i i and, near every point, there is a function f f f such that ι V Ω = d f − ∑ i R i ( f ) λ i \iota_V\Omega=df-\sum_iR_i(f)\lambda_i ι V Ω = df − ∑ i R i ( f ) λ i . It is globally Hamiltonian if a single globally defined f f f has this property.
Equivalently, V V V is locally Hamiltonian if it is horizontal and d ( ι V Ω ) ∣ ξ × ξ = 0 d(\iota_V\Omega)|_{\xi\times\xi}=0 d ( ι V Ω ) ∣ ξ × ξ = 0 . Indeed, in a product Darboux chart ( x , t ) (x,t) ( x , t ) , a horizontal one-form has the form β = ∑ a β a ( x , t ) d x a \beta=\sum_a\beta_a(x,t)\,dx^a β = ∑ a β a ( x , t ) d x a . When its restriction to each leaf is closed, the Poincaré lemma with parameters gives a smooth f f f with ∂ f / ∂ x a = β a \partial f/\partial x^a=\beta_a ∂ f / ∂ x a = β a . Thus β = d f − ∑ i R i ( f ) λ i \beta=df-\sum_iR_i(f)\lambda_i β = df − ∑ i R i ( f ) λ i . This condition is weaker than d ( ι V Ω ) = 0 d(\iota_V\Omega)=0 d ( ι V Ω ) = 0 .
Definition 4.4. A generalized canonical transformation is a diffeomorphism ψ \psi ψ such that ψ ∗ λ i = λ i \psi^*\lambda_i=\lambda_i ψ ∗ λ i = λ i for every i i i and, locally on its domain,
ψ ∗ Ω − Ω = ∑ i = 1 q d J i ∧ λ i (2) \psi^*\Omega-\Omega=\sum_{i=1}^{q}dJ_i\wedge\lambda_i \tag*{(2)} ψ ∗ Ω − Ω = i = 1 ∑ q d J i ∧ λ i (2) for smooth functions J i J_i J i . Global existence of these functions is an additional requirement when explicitly stated.
Proposition 4.4. For a diffeomorphism ψ \psi ψ satisfying ψ ∗ λ i = λ i \psi^*\lambda_i=\lambda_i ψ ∗ λ i = λ i , the following conditions are equivalent:
ψ \psi ψ is generalized canonical;
( ψ ∗ Ω ) ∣ ξ × ξ = Ω ∣ ξ × ξ (\psi^*\Omega)|_{\xi\times\xi}=\Omega|_{\xi\times\xi} ( ψ ∗ Ω ) ∣ ξ × ξ = Ω ∣ ξ × ξ ;
ψ \psi ψ preserves the Poisson bracket.
Locally, these maps have the form ψ ( q , p , t ) = ( Φ t ( q , p ) , t + c ) \psi(q,p,t)=(\Phi_t(q,p),t+c) ψ ( q , p , t ) = ( Φ t ( q , p ) , t + c ) , where every Φ t \Phi_t Φ t is symplectic and c c c is constant.
Proof. The first condition immediately implies the second. Conversely, let Δ = ψ ∗ Ω − Ω \Delta=\psi^*\Omega-\Omega Δ = ψ ∗ Ω − Ω . This form is closed and vanishes on ξ × ξ \xi\times\xi ξ × ξ . On a sufficiently small product Darboux chart choose a one-form α \alpha α with d α = Δ d\alpha=\Delta d α = Δ . The restriction of α \alpha α to each plaque t = c o n s t t=\mathrm{const} t = const is closed. The Poincaré lemma with parameters therefore gives f f f whose differential in the leaf variables is that restriction. Consequently α − d f = ∑ i J i d t i \alpha-df=\sum_i J_i\,dt_i α − df = ∑ i J i d t i , and d α = ∑ i d J i ∧ d t i d\alpha=\sum_i dJ_i\wedge dt_i d α = ∑ i d J i ∧ d t i , proving (4.1).
Preservation of the λ i \lambda_i λ i implies preservation of ξ \xi ξ and locally gives t ∘ ψ = t + c t\circ\psi=t+c t ∘ ψ = t + c . The second condition says precisely that the induced maps between its symplectic leaves are symplectic. Since the Poisson tensor is the inverse of the symplectic form on these leaves, this is equivalent to the third condition and gives the local form.
Remark 4.3. Every canonical transformation is generalized canonical, but a generalized canonical map need not preserve the Reeb distribution. In fact, (4.1) gives ( ψ ∗ Ω ) ( R k , R l ) = R k ( J l ) − R l ( J k ) (\psi^*\Omega)(R_k,R_l)=R_k(J_l)-R_l(J_k) ( ψ ∗ Ω ) ( R k , R l ) = R k ( J l ) − R l ( J k ) . For q = 2 q=2 q = 2 and Ω = d x ∧ d p \Omega=dx\wedge dp Ω = d x ∧ d p , the map
ψ ( x , p , t 1 , t 2 ) = ( x + t 2 , p + t 1 , t 1 , t 2 ) \psi(x,p,t_1,t_2)=(x+t_2,p+t_1,t_1,t_2) ψ ( x , p , t 1 , t 2 ) = ( x + t 2 , p + t 1 , t 1 , t 2 )
satisfies (4.1) with J 1 = x + t 2 J_1=x+t_2 J 1 = x + t 2 , J 2 = − p J_2=-p J 2 = − p . Nevertheless, ( ψ ∗ Ω ) ( R 1 , R 2 ) = − 1 (\psi^*\Omega)(R_1,R_2)=-1 ( ψ ∗ Ω ) ( R 1 , R 2 ) = − 1 . Thus preservation of the Poisson bracket does not recover the full presymplectic form or its characteristic distribution.
Example 4.4. For the same standard structure, let ψ ( x , p , t 1 , t 2 ) = ( x , p + t 1 , t 1 , t 2 ) \psi(x,p,t_1,t_2)=(x,p+t_1,t_1,t_2) ψ ( x , p , t 1 , t 2 ) = ( x , p + t 1 , t 1 , t 2 ) . Then { x , p + t 1 } = 1 \{x,p+t_1\}=1 { x , p + t 1 } = 1 , but ψ ∗ Ω = Ω + d x ∧ d t 1 ≠ Ω \psi^*\Omega=\Omega+dx\wedge dt_1\ne\Omega ψ ∗ Ω = Ω + d x ∧ d t 1 = Ω . It is generalized canonical with J 1 = x J_1=x J 1 = x , $J_2=0. The map is the time-one flow of V = t 1 ∂ p = X − t 1 x V=t_1\partial_p=X_{-t_1x} V = t 1 ∂ p = X − t 1 x . Thus a globally Hamiltonian field can generate generalized canonical transformations without being canonical. Indeed, L V Ω = d x ∧ d t 1 ≠ 0 \mathcal{L}_V\Omega=dx\wedge dt_1\ne0 L V Ω = d x ∧ d t 1 = 0 .
Theorem 4.2. Let V V V satisfy λ i ( V ) = 0 \lambda_i(V)=0 λ i ( V ) = 0 for every i i i , and let Ψ s \Psi_s Ψ s be its local flow. The maps Ψ s \Psi_s Ψ s are generalized canonical if and only if V V V is locally Hamiltonian. If V = X f V=X_f V = X f globally, one may take
J i ( s ) = − ∫ 0 s Ψ u ∗ ( R i ( f ) ) d u J_i^{(s)}=-\int_0^s\Psi_u^*(R_i(f))\,du J i ( s ) = − ∫ 0 s Ψ u ∗ ( R i ( f )) d u on the flow domain.
Proof. Horizontality and d λ i = 0 d\lambda_i=0 d λ i = 0 give L V λ i = 0 \mathcal{L}_V\lambda_i=0 L V λ i = 0 , so Ψ s ∗ λ i = λ i \Psi_s^*\lambda_i=\lambda_i Ψ s ∗ λ i = λ i . If V = X f V=X_f V = X f on a neighbourhood, then L V Ω = − ∑ i d ( R i ( f ) ) ∧ λ i \mathcal{L}_V\Omega=-\sum_i d(R_i(f))\wedge\lambda_i L V Ω = − ∑ i d ( R i ( f )) ∧ λ i . On a smaller neighbourhood where the relevant flow segments remain in that domain,
Ψ s ∗ Ω − Ω = ∫ 0 s Ψ u ∗ L V Ω d u = ∑ i d ( − ∫ 0 s Ψ u ∗ ( R i ( f ) ) d u ) ∧ λ i . \begin{aligned}
\Psi_s^*\Omega-\Omega
&=\int_0^s\Psi_u^*\mathcal{L}_V\Omega\,du\\
&=\sum_i d\left(-\int_0^s\Psi_u^*(R_i(f))\,du\right)\wedge\lambda_i.
\end{aligned} Ψ s ∗ Ω − Ω = ∫ 0 s Ψ u ∗ L V Ω d u = i ∑ d ( − ∫ 0 s Ψ u ∗ ( R i ( f )) d u ) ∧ λ i .
This proves the required local statement. For longer flow segments the same conclusion follows by composing such maps, which preserve the symplectic structures on the leaves. If f f f is global, the displayed primitives are global on the flow domain.
Conversely, Proposition 4.4 gives ( Ψ s ∗ Ω ) ∣ ξ × ξ = Ω ∣ ξ × ξ (\Psi_s^*\Omega)|_{\xi\times\xi}=\Omega|_{\xi\times\xi} ( Ψ s ∗ Ω ) ∣ ξ × ξ = Ω ∣ ξ × ξ . Differentiating this identity at s = 0 s=0 s = 0 yields d ( ι V Ω ) ∣ ξ × ξ = 0 d(\iota_V\Omega)|_{\xi\times\xi}=0 d ( ι V Ω ) ∣ ξ × ξ = 0 . The leafwise Poincaré argument above proves that V V V is locally Hamiltonian. This proof does not require a preliminary smooth choice of the functions J i ( s ) J_i^{(s)} J i ( s ) .
The horizontality assumption is essential: R 1 R_1 R 1 generates canonical transformations but is not locally Hamiltonian in the stated sense.
Canonoid transformations
We now fix H H H . The canonoid transformations considered here preserve the clock forms and the Reeb frame, while allowing the symplectic form on the horizontal distribution to change. This specifies a restricted class of transformations relative to the given q q q -cosymplectic data.
Definition 4.5. A possibly local diffeomorphism F F F is canonoid for ( M , Ω , λ ⃗ , H ) (M,\Omega,\vec{\lambda},H) ( M , Ω , λ , H ) if F ∗ λ i = λ i F^*\lambda_i=\lambda_i F ∗ λ i = λ i for every i i i , R ⊂ ker F ∗ Ω \mathcal{R}\subset\ker F^*\Omega R ⊂ ker F ∗ Ω , and there is a smooth function K K K on its domain such that
ι X H F ∗ Ω = d K − ∑ i = 1 q R i ( K ) λ i . \iota_{X_H}F^*\Omega=dK-\sum_{i=1}^{q}R_i(K)\lambda_i. ι X H F ∗ Ω = d K − i = 1 ∑ q R i ( K ) λ i . If this last function exists only locally, we explicitly speak of a locally canonoid transformation.
Since F ∗ Ω F^*\Omega F ∗ Ω has rank 2 n 2n 2 n , the kernel inclusion is an equality. Together with F ∗ λ i = λ i F^*\lambda_i=\lambda_i F ∗ λ i = λ i , it implies F ∗ R i = R i F_*R_i=R_i F ∗ R i = R i . Conversely, preservation of the clock forms and each Reeb field implies the kernel condition. In Darboux charts such maps have the form F ( q , p , t ) = ( Φ ( q , p ) , t + c ) F(q,p,t)=(\Phi(q,p),t+c) F ( q , p , t ) = ( Φ ( q , p ) , t + c ) , with no clock dependence in Φ \Phi Φ ; Φ \Phi Φ need not be symplectic.
Thus ( F ∗ Ω , λ ⃗ ) (F^*\Omega,\vec{\lambda}) ( F ∗ Ω , λ ) is q q q -cosymplectic with the original Reeb fields, and the definition says that X H X_H X H is Hamiltonian for this new structure. Equivalently, in the original structure on the target domain,
F ∗ X H = X K ~ , F ∗ E ⃗ H a ⃗ = X K ~ + ∑ i a i R i , K ~ = K ∘ F − 1 . F_*X_H=X_{\widetilde K},\qquad F_*\vec E_H^{\vec a}=X_{\widetilde K}+\sum_i a_iR_i,\qquad\widetilde K=K\circ F^{-1}. F ∗ X H = X K , F ∗ E H a = X K + i ∑ a i R i , K = K ∘ F − 1 .
These identities follow by pushing forward the defining contraction equations. They explain the dynamical meaning of the definition.
Every canonical map is canonoid, with K = H K=H K = H . Also, evaluating the defining equation on X H X_H X H gives X H ( K ) = 0 X_H(K)=0 X H ( K ) = 0 . This does not imply conservation of K K K along every E ⃗ H a ⃗ \vec E_H^{\vec a} E H a unless also R i ( K ) = 0 R_i(K)=0 R i ( K ) = 0 .
Lemma 4.2. Let X X X generate a local flow φ s \varphi_s φ s of canonoid transformations, with new Hamiltonians K s K_s K s smooth in ( s , x ) (s,x) ( s , x ) . Then L X λ i = 0 \mathcal{L}_X\lambda_i=0 L X λ i = 0 , ι R i L X Ω = 0 \iota_{R_i}\mathcal{L}_X\Omega=0 ι R i L X Ω = 0 , and
ι X H L X Ω = d κ − ∑ i = 1 q R i ( κ ) λ i , κ = ∂ K s ∂ s ∣ s = 0 . (3) \iota_{X_H}\mathcal{L}_X\Omega=d\kappa-\sum_{i=1}^{q}R_i(\kappa)\lambda_i,\qquad\kappa=\left.\frac{\partial K_s}{\partial s}\right|_{s=0}. \tag*{(3)} ι X H L X Ω = d κ − i = 1 ∑ q R i ( κ ) λ i , κ = ∂ s ∂ K s s = 0 . (3) In particular, X H ( κ ) = 0 X_H(\kappa)=0 X H ( κ ) = 0 .
Proof. Differentiate φ s ∗ λ i = λ i \varphi_s^*\lambda_i=\lambda_i φ s ∗ λ i = λ i , ι R i φ s ∗ Ω = 0 \iota_{R_i}\varphi_s^*\Omega=0 ι R i φ s ∗ Ω = 0 , and
ι X H φ s ∗ Ω = d K s − ∑ i R i ( K s ) λ i \iota_{X_H}\varphi_s^*\Omega=dK_s-\sum_iR_i(K_s)\lambda_i ι X H φ s ∗ Ω = d K s − i ∑ R i ( K s ) λ i
at s = 0 s=0 s = 0 . The fields X H X_H X H , R i R_i R i and the forms λ i \lambda_i λ i are fixed in these equations, giving the stated identities. Evaluating (4.2) on X H X_H X H gives X H ( κ ) = 0 X_H(\kappa)=0 X H ( κ ) = 0 .
At s = 0 s=0 s = 0 , the difference K 0 − H K_0-H K 0 − H is a Casimir of the Poisson bracket, namely it is constant along ξ \xi ξ . Locally it is a function of t t t alone. The same Casimir ambiguity applies to κ \kappa κ in (4.2); it does not change the Hamiltonian vector fields below.
Theorem 4.3. Under the hypotheses of Lemma 4.2, [ X , R i ] = 0 [X,R_i]=0 [ X , R i ] = 0 for every i i i , and
[ X , X H ] = X X ( H ) − κ . [X,X_H]=X_{X(H)-\kappa}. [ X , X H ] = X X ( H ) − κ . More generally, suppose that L X λ i = 0 \mathcal{L}_X\lambda_i=0 L X λ i = 0 and ι R i L X Ω = 0 \iota_{R_i}\mathcal{L}_X\Omega=0 ι R i L X Ω = 0 for every i i i . Then (4.2) holds for some κ \kappa κ if and only if [ X , X H ] [X,X_H] [ X , X H ] is Hamiltonian. The same equivalence holds locally.
Proof. The structural assumptions imply [ X , R i ] = 0 [X,R_i]=0 [ X , R i ] = 0 by Proposition 4.2. First,
λ i ( [ X , X H ] ) = X ( λ i ( X H ) ) − ( L X λ i ) ( X H ) = 0. (4) \lambda_i([X,X_H])=X(\lambda_i(X_H))-(\mathcal{L}_X\lambda_i)(X_H)=0. \tag*{(4)} λ i ([ X , X H ]) = X ( λ i ( X H )) − ( L X λ i ) ( X H ) = 0. (4)
Second,
ι [ X , X H ] Ω = L X ( ι X H Ω ) − ι X H L X Ω = d ( X ( H ) ) − ∑ i R i ( X ( H ) ) λ i − ι X H L X Ω = d ( X ( H ) − κ ) − ∑ i R i ( X ( H ) − κ ) λ i . (5) \begin{aligned}
\iota_{[X,X_H]}\Omega&=\mathcal{L}_X(\iota_{X_H}\Omega)-\iota_{X_H}\mathcal{L}_X\Omega\\
&=d(X(H))-\sum_i R_i(X(H))\lambda_i-\iota_{X_H}\mathcal{L}_X\Omega\\
&=d(X(H)-\kappa)-\sum_i R_i(X(H)-\kappa)\lambda_i. \tag*{(5)}
\end{aligned} ι [ X , X H ] Ω = L X ( ι X H Ω ) − ι X H L X Ω = d ( X ( H )) − i ∑ R i ( X ( H )) λ i − ι X H L X Ω = d ( X ( H ) − κ ) − i ∑ R i ( X ( H ) − κ ) λ i . (5)
This proves the first implication and the Hamiltonian formula. Conversely, if [ X , X H ] = X G [X,X_H]=X_G [ X , X H ] = X G , the first two lines of (5) give
ι X H L X Ω = d ( X ( H ) − G ) − ∑ i R i ( X ( H ) − G ) λ i . \iota_{X_H}\mathcal{L}_X\Omega=d(X(H)-G)-\sum_i R_i(X(H)-G)\lambda_i. ι X H L X Ω = d ( X ( H ) − G ) − i ∑ R i ( X ( H ) − G ) λ i .
Thus (4.2) holds with κ = X ( H ) − G \kappa=X(H)-G κ = X ( H ) − G .
Remark 4.4. Call X X X infinitesimally canonoid if it satisfies the structural assumptions of the theorem and (4.2). The theorem characterizes this infinitesimal condition; it does not assert that the entire flow is canonoid.
Indeed, the structural conditions do integrate: φ s ∗ λ i = λ i \varphi_s^*\lambda_i=\lambda_i φ s ∗ λ i = λ i and ( φ s ) ∗ R i = R i (\varphi_s)_*R_i=R_i ( φ s ) ∗ R i = R i , hence ker φ s ∗ Ω = R \ker\varphi_s^*\Omega=\mathcal{R} ker φ s ∗ Ω = R . Put Ω s = φ s ∗ Ω \Omega_s=\varphi_s^*\Omega Ω s = φ s ∗ Ω . On a sufficiently small product Darboux chart, the remaining canonoid condition is equivalent to
d ( ι X H Ω s ) ∣ ξ × ξ = 0 for every sufficiently small s . d(\iota_{X_H}\Omega_s)|_{\xi\times\xi}=0\quad\text{for every sufficiently small }s. d ( ι X H Ω s ) ∣ ξ × ξ = 0 for every sufficiently small s .
The equivalence follows from the Poincaré lemma on the leaves, with the clock variables and s s s as parameters. Global Hamiltonians require global primitives of the same horizontal equations. The infinitesimal condition controls only the first derivative at s = 0 s=0 s = 0 of this family of equations. The next example shows that the second derivative can already obstruct integration.
Example 4.5 (An infinitesimal condition that does not integrate). Let M = R 2 + q M=\mathbb{R}^{2+q} M = R 2 + q with Ω = d x ∧ d p \Omega=dx\wedge dp Ω = d x ∧ d p , λ i = d t i \lambda_i=dt_i λ i = d t i , and H = p H=p H = p . Then X H = ∂ x X_H=\partial_x X H = ∂ x . Consider
X = x p ∂ x + x ∂ p . X=xp\,\partial_x+x\,\partial_p. X = x p ∂ x + x ∂ p . It preserves the clock forms and commutes with every R i = ∂ t i R_i=\partial_{t_i} R i = ∂ t i . Moreover, L X Ω = p Ω \mathcal{L}_X\Omega=p\,\Omega L X Ω = p Ω , so
ι X H L X Ω = p d p = d ( p 2 / 2 ) . \iota_{X_H}\mathcal{L}_X\Omega=p\,dp=d(p^2/2). ι X H L X Ω = p d p = d ( p 2 /2 ) . Thus X X X is infinitesimally canonoid with κ = p 2 / 2 \kappa=p^2/2 κ = p 2 /2 . Correspondingly,
[ X , X H ] = − p ∂ x − ∂ p = X x − p 2 / 2 , X ( H ) − κ = x − p 2 / 2. [X,X_H]=-p\,\partial_x-\partial_p=X_{x-p^2/2},\qquad X(H)-\kappa=x-p^2/2. [ X , X H ] = − p ∂ x − ∂ p = X x − p 2 /2 , X ( H ) − κ = x − p 2 /2. Nevertheless, its local flow is not a flow of canonoid transformations. Since X X X does not involve the clock variables, write φ s ∗ Ω = A s ( x , p ) d x ∧ d p \varphi_s^*\Omega=A_s(x,p)\,dx\wedge dp φ s ∗ Ω = A s ( x , p ) d x ∧ d p . We have L X 2 Ω = ( x + p 2 ) Ω \mathcal{L}_X^2\Omega=(x+p^2)\Omega L X 2 Ω = ( x + p 2 ) Ω , and hence, locally with the remainder also controlled after one spatial derivative,
A s = 1 + s p + s 2 2 ( x + p 2 ) + O ( s 3 ) . A_s=1+sp+\frac{s^2}{2}(x+p^2)+O(s^3). A s = 1 + s p + 2 s 2 ( x + p 2 ) + O ( s 3 ) .
Therefore
ι X H φ s ∗ Ω = A s d p , \iota_{X_H}\varphi_s^*\Omega=A_s\,dp, ι X H φ s ∗ Ω = A s d p ,
d ( ι X H φ s ∗ Ω ) ∣ ξ × ξ = ∂ x A s d x ∧ d p = ( s 2 2 + O ( s 3 ) ) d x ∧ d p . d(\iota_{X_H}\varphi_s^*\Omega)|_{\xi\times\xi}=\partial_xA_s\,dx\wedge dp
=\left(\frac{s^2}{2}+O(s^3)\right)dx\wedge dp. d ( ι X H φ s ∗ Ω ) ∣ ξ × ξ = ∂ x A s d x ∧ d p = ( 2 s 2 + O ( s 3 ) ) d x ∧ d p .
At ( x , p ) = ( 0 , 0 ) (x,p)=(0,0) ( x , p ) = ( 0 , 0 ) this is nonzero for sufficiently small s ≠ 0 s\ne0 s = 0 . Thus no local function K s K_s K s can satisfy the canonoid equation near that point. The counterexample works for every q ≥ 1 q\ge1 q ≥ 1 .
There are, however, useful sufficient conditions that can be checked for the whole flow. They include the following versions of dynamical symmetries and constant rescalings [5 ] .
Proposition 4.5. Suppose that L X λ i = 0 \mathcal{L}_X\lambda_i=0 L X λ i = 0 and ι R i L X Ω = 0 \iota_{R_i}\mathcal{L}_X\Omega=0 ι R i L X Ω = 0 for every i i i . Either of the following conditions ensures that the local flow φ s \varphi_s φ s is canonoid:
[ X , X H ] = 0 [X,X_H]=0 [ X , X H ] = 0 , in which case K s = φ s ∗ H K_s=\varphi_s^*H K s = φ s ∗ H ;
L X Ω = c Ω \mathcal{L}_X\Omega=c\Omega L X Ω = c Ω for a constant c c c , in which case K s = e c s H K_s=e^{cs}H K s = e cs H .
The second condition works for every Hamiltonian H H H .
Proof. The structural conditions preserve the clock forms and the Reeb fields along the flow. In the first case, X H X_H X H is also preserved, so
ι X H φ s ∗ Ω = φ s ∗ ( ι X H Ω ) = d ( φ s ∗ H ) − ∑ i R i ( φ s ∗ H ) λ i . \iota_{X_H}\varphi_s^*\Omega=\varphi_s^*(\iota_{X_H}\Omega)
=d(\varphi_s^*H)-\sum_iR_i(\varphi_s^*H)\lambda_i. ι X H φ s ∗ Ω = φ s ∗ ( ι X H Ω ) = d ( φ s ∗ H ) − i ∑ R i ( φ s ∗ H ) λ i .
In the second case, φ s ∗ Ω = e c s Ω \varphi_s^*\Omega=e^{cs}\Omega φ s ∗ Ω = e cs Ω , and therefore
ι X H φ s ∗ Ω = d ( e c s H ) − ∑ i R i ( e c s H ) λ i . \iota_{X_H}\varphi_s^*\Omega=d(e^{cs}H)-\sum_iR_i(e^{cs}H)\lambda_i. ι X H φ s ∗ Ω = d ( e cs H ) − i ∑ R i ( e cs H ) λ i .
These explicit families prove canonoidness.
Example 4.6 (A canonoid, non-canonical flow with explicit clock dependence in the Hamiltonian). Let Ω = d x ∧ d p \Omega=dx\wedge dp Ω = d x ∧ d p , λ i = d t i \lambda_i=dt_i λ i = d t i for i = 1 , 2 i=1,2 i = 1 , 2 , and H = 1 2 p 2 + ( t 1 + t 2 ) x H=\frac{1}{2}p^2+(t_1+t_2)x H = 2 1 p 2 + ( t 1 + t 2 ) x . Then X H = p ∂ x − ( t 1 + t 2 ) ∂ p X_H=p\partial_x-(t_1+t_2)\partial_p X H = p ∂ x − ( t 1 + t 2 ) ∂ p .
The field X = p ∂ p X=p\partial_p X = p ∂ p has flow φ s ( x , p , t ) = ( x , e s p , t ) \varphi_s(x,p,t)=(x,e^sp,t) φ s ( x , p , t ) = ( x , e s p , t ) and satisfies L X Ω = Ω \mathcal{L}_X\Omega=\Omega L X Ω = Ω . Thus φ s \varphi_s φ s is canonoid with K s = e s H K_s=e^sH K s = e s H , although it is not canonical for s ≠ 0 s\ne0 s = 0 .
Here κ = H \kappa=H κ = H , so X ( H ) − κ = 1 2 p 2 − ( t 1 + t 2 ) x X(H)-\kappa=\frac{1}{2}p^2-(t_1+t_2)x X ( H ) − κ = 2 1 p 2 − ( t 1 + t 2 ) x . Directly,
[ X , X H ] = p ∂ x + ( t 1 + t 2 ) ∂ p = X 1 2 p 2 − ( t 1 + t 2 ) x . [X,X_H]=p\partial_x+(t_1+t_2)\partial_p=X_{\frac{1}{2}p^2-(t_1+t_2)x}. [ X , X H ] = p ∂ x + ( t 1 + t 2 ) ∂ p = X 2 1 p 2 − ( t 1 + t 2 ) x . More generally, on the standard product model, X = ∑ i p i ∂ p i X=\sum_i p_i\partial_{p_i} X = ∑ i p i ∂ p i generates such a canonoid flow for every H H H , with [ X , X H ] = X X ( H ) − H [X,X_H]=X_{X(H)-H} [ X , X H ] = X X ( H ) − H . The rescaling field need not extend globally to an arbitrary q q q -cosymplectic manifold.
Canonical and canonoid transformations on uniform q q q -contact manifolds
Throughout this section, ( M , λ ⃗ ) (M,\vec{\lambda}) ( M , λ ) is a uniform q q q -contact manifold, with Ω = d λ 1 = ⋯ = d λ q \Omega=d\lambda_1=\cdots=d\lambda_q Ω = d λ 1 = ⋯ = d λ q . Its Reeb fields are denoted by R 1 , … , R q R_1,\ldots,R_q R 1 , … , R q , and E = ∑ i R i E=\sum_iR_i E = ∑ i R i . We use the Hamiltonian fields and Jacobi bracket of Section 2. All flow statements are local unless completeness is assumed.
We study transformations relative to the full ordered family λ ⃗ \vec{\lambda} λ . The distinction from the underlying Jacobi structure is relevant here: strict preservation of every λ i \lambda_i λ i imposes conditions on the transverse data as well as on the contact structures of the leaves.
Canonical transformations and symmetries
Definition 5.1. A canonical transformation is a possibly local diffeomorphism F F F such that F ∗ λ i = λ i F^*\lambda_i=\lambda_i F ∗ λ i = λ i for every i i i . A field X X X is canonical if its local flow consists of canonical transformations.
The flow identity d d s ϕ s ∗ λ i = ϕ s ∗ L X λ i \frac{d}{ds}\phi_s^*\lambda_i=\phi_s^*\mathcal{L}_X\lambda_i d s d ϕ s ∗ λ i = ϕ s ∗ L X λ i shows that X X X is canonical precisely when L X λ i = 0 \mathcal{L}_X\lambda_i=0 L X λ i = 0 for every i i i . These fields form a Lie subalgebra, since L [ X , Y ] λ i = L X L Y λ i − L Y L X λ i \mathcal{L}_{[X,Y]}\lambda_i=\mathcal{L}_X\mathcal{L}_Y\lambda_i-\mathcal{L}_Y\mathcal{L}_X\lambda_i L [ X , Y ] λ i = L X L Y λ i − L Y L X λ i . They do not form a C ∞ ( M ) C^\infty(M) C ∞ ( M ) -module: L f X λ i = f L X λ i + λ i ( X ) d f \mathcal{L}_{fX}\lambda_i=f\mathcal{L}_X\lambda_i+\lambda_i(X)\,df L f X λ i = f L X λ i + λ i ( X ) df . For example, R i R_i R i is canonical, whereas L f R i λ i = d f \mathcal{L}_{fR_i}\lambda_i=df L f R i λ i = df .
Proposition 5.1. For a vector field X X X , the following are equivalent:
L X λ i = 0 \mathcal{L}_X\lambda_i=0 L X λ i = 0 for every i i i ;
[ X , R i ] = 0 [X,R_i]=0 [ X , R i ] = 0 for every i i i and [ X , Γ ( ξ ) ] ⊂ Γ ( ξ ) [X,\Gamma(\xi)]\subset\Gamma(\xi) [ X , Γ ( ξ )] ⊂ Γ ( ξ ) .
Proof. Assume the first condition. Differentiating λ j ( R i ) = δ j i \lambda_j(R_i)=\delta^i_j λ j ( R i ) = δ j i gives λ j ( [ X , R i ] ) = 0 \lambda_j([X,R_i])=0 λ j ([ X , R i ]) = 0 . Also L X Ω = d L X λ 1 = 0 \mathcal{L}_X\Omega=d\mathcal{L}_X\lambda_1=0 L X Ω = d L X λ 1 = 0 , so
ι [ X , R i ] Ω = L X ( ι R i Ω ) − ι R i L X Ω = 0. \iota_{[X,R_i]}\Omega=\mathcal{L}_X(\iota_{R_i}\Omega)-\iota_{R_i}\mathcal{L}_X\Omega=0. ι [ X , R i ] Ω = L X ( ι R i Ω ) − ι R i L X Ω = 0. Hence [ X , R i ] ∈ ξ ∩ R = { 0 } [X,R_i]\in\xi\cap\mathcal{R}=\{0\} [ X , R i ] ∈ ξ ∩ R = { 0 } . For Y ∈ Γ ( ξ ) Y\in\Gamma(\xi) Y ∈ Γ ( ξ ) , λ i ( [ X , Y ] ) = − ( L X λ i ) ( Y ) = 0 \lambda_i([X,Y])=-(\mathcal{L}_X\lambda_i)(Y)=0 λ i ([ X , Y ]) = − ( L X λ i ) ( Y ) = 0 , which proves preservation of ξ \xi ξ .
Conversely, the second condition implies that L X λ j \mathcal{L}_X\lambda_j L X λ j vanishes on every R i R_i R i and on every field with values in ξ \xi ξ . Since T M = R ⊕ ξ TM=\mathcal{R}\oplus\xi T M = R ⊕ ξ , it follows that L X λ j = 0 \mathcal{L}_X\lambda_j=0 L X λ j = 0 .
Unlike the analogous conditions in Proposition 4.2, these conditions already imply preservation of the two-form, because Ω = d λ i \Omega=d\lambda_i Ω = d λ i .
The following description makes the restrictions on canonical fields more explicit.
Proposition 5.2. On each connected component, every canonical field V V V admits the unique decomposition
V = X h + ∑ i = 2 q c i R i , h = − λ 1 ( V ) , R j ( h ) = 0 , c i ∈ R . V=X_h+\sum_{i=2}^{q}c_iR_i,\qquad h=-\lambda_1(V),\qquad R_j(h)=0,\qquad c_i\in\mathbb{R}. V = X h + i = 2 ∑ q c i R i , h = − λ 1 ( V ) , R j ( h ) = 0 , c i ∈ R . The sum is empty when q = 1 q=1 q = 1 . Conversely, every field of this form is canonical. For every canonical field V V V and every Hamiltonian H H H , [ V , X H ] = X V ( H ) [V,X_H]=X_{V(H)} [ V , X H ] = X V ( H ) .
Proof. Put f i = λ i ( V ) f_i=\lambda_i(V) f i = λ i ( V ) . Cartan’s formula gives d f i = − ι V Ω df_i=-\iota_V\Omega d f i = − ι V Ω for every i i i . Therefore R j ( f i ) = 0 R_j(f_i)=0 R j ( f i ) = 0 and d ( f i − f 1 ) = 0 d(f_i-f_1)=0 d ( f i − f 1 ) = 0 . On a connected component, write c i = f i − f 1 c_i=f_i-f_1 c i = f i − f 1 , with c 1 = 0 c_1=0 c 1 = 0 , and set h = − f 1 h=-f_1 h = − f 1 . Then R j ( h ) = 0 R_j(h)=0 R j ( h ) = 0 and ι V Ω = d h \iota_V\Omega=dh ι V Ω = d h .
The field X h X_h X h satisfies ι X h Ω = d h \iota_{X_h}\Omega=dh ι X h Ω = d h and λ i ( X h ) = − h = f 1 \lambda_i(X_h)=-h=f_1 λ i ( X h ) = − h = f 1 . Thus V − X h V-X_h V − X h belongs to R \mathcal{R} R , and its contractions with the λ i \lambda_i λ i are the constants c i c_i c i . This proves the decomposition and uniqueness. Conversely, R j ( h ) = 0 R_j(h)=0 R j ( h ) = 0 implies L X h λ i = 0 \mathcal{L}_{X_h}\lambda_i=0 L X h λ i = 0 , and each R i R_i R i is canonical.
Finally, Proposition 5.1 gives [ V , R i ] = 0 [V,R_i]=0 [ V , R i ] = 0 , while L V Ω = 0 \mathcal{L}_V\Omega=0 L V Ω = 0 . Consequently,
λ i ( [ V , X H ] ) = − V ( H ) , ι [ V , X H ] Ω = d ( V ( H ) ) − ∑ j R j ( V ( H ) ) λ j . \lambda_i([V,X_H])=-V(H),
\qquad
\iota_{[V,X_H]}\Omega=d(V(H))-\sum_jR_j(V(H))\lambda_j. λ i ([ V , X H ]) = − V ( H ) , ι [ V , X H ] Ω = d ( V ( H )) − j ∑ R j ( V ( H )) λ j . These are the defining equations for X V ( H ) X_{V(H)} X V ( H ) .
In particular, a canonical field is Hamiltonian in the sense of Section 2 precisely when all its contractions λ i ( V ) \lambda_i(V) λ i ( V ) coincide. For q > 1 q>1 q > 1 , the remaining constant Reeb components must also be retained.
Definition 5.2. An infinitesimal symmetry of ( M , λ ⃗ , H ) (M,\vec{\lambda},H) ( M , λ , H ) is a canonical field V V V such that V ( H ) = 0 V(H)=0 V ( H ) = 0 . We use dynamical symmetry for a field satisfying [ V , X H ] = 0 [V,X_H]=0 [ V , X H ] = 0 , without requiring preservation of the defining forms.
Proposition 5.3. A field V V V is an infinitesimal symmetry if and only if its local flow is canonical and preserves H H H .
Proof. Apply d d s ϕ s ∗ α = ϕ s ∗ L V α \frac{d}{ds}\phi_s^*\alpha=\phi_s^*\mathcal{L}_V\alpha d s d ϕ s ∗ α = ϕ s ∗ L V α to each λ i \lambda_i λ i and to H H H . Vanishing of the Lie derivatives makes the pullbacks constant in s s s . Conversely, differentiation of the invariance identities at s = 0 s=0 s = 0 gives the defining conditions.
Dissipated q-tuples
Dissipated quantities and their relation to symmetries in ordinary contact mechanics are discussed in [31 , 5 ] . Here the identity L X H λ i = − ∑ j R j ( H ) λ j \mathcal{L}_{X_H}\lambda_i=-\sum_jR_j(H)\lambda_j L X H λ i = − ∑ j R j ( H ) λ j leads to a coupled law for the contractions with the full family of one-forms.
Definition 5.3. A dissipated q q q -tuple is a tuple ( f 1 , … , f q ) (f_1,\ldots,f_q) ( f 1 , … , f q ) satisfying
X H ( f i ) = − ∑ j = 1 q R j ( H ) f j , i = 1 , … , q . X_H(f_i)=-\sum_{j=1}^{q}R_j(H)f_j,\qquad i=1,\ldots,q. X H ( f i ) = − j = 1 ∑ q R j ( H ) f j , i = 1 , … , q . For any such tuple, X H ( f i − f j ) = 0 X_H(f_i-f_j)=0 X H ( f i − f j ) = 0 . Thus the differences are first integrals, although they need not be constant functions for an arbitrary dissipated tuple.
More generally, for any vector field V V V ,
X H ( λ i ( V ) ) = − ∑ j R j ( H ) λ j ( V ) + λ i ( [ X H , V ] ) . X_H(\lambda_i(V))=-\sum_jR_j(H)\lambda_j(V)+\lambda_i([X_H,V]). X H ( λ i ( V )) = − j ∑ R j ( H ) λ j ( V ) + λ i ([ X H , V ]) . This follows from the formula for L X H λ i \mathcal{L}_{X_H}\lambda_i L X H λ i . In particular, every dynamical symmetry gives a dissipated tuple. Strict infinitesimal symmetries impose the additional restrictions in the next theorem.
Theorem 5.1. Let V V V be an infinitesimal symmetry and set f i = λ i ( V ) f_i=\lambda_i(V) f i = λ i ( V ) . Then, for every i , k i,k i , k ,
X H ( f i ) = − ∑ j = 1 q R j ( H ) f j , R k ( f i ) = 0 , d ( f i − f k ) = 0. X_H(f_i)=-\sum_{j=1}^{q}R_j(H)f_j,\qquad R_k(f_i)=0,\qquad d(f_i-f_k)=0. X H ( f i ) = − j = 1 ∑ q R j ( H ) f j , R k ( f i ) = 0 , d ( f i − f k ) = 0. Consequently:
( f 1 , … , f q ) (f_1,\ldots,f_q) ( f 1 , … , f q ) is a dissipated tuple;
its differences are locally constant;
every f i f_i f i is a first integral if and only if ∑ j R j ( H ) f j = 0 \sum_jR_j(H)f_j=0 ∑ j R j ( H ) f j = 0 .
Proof. Since V V V is a dynamical symmetry, the general contraction identity above gives the dissipation law. Canonicity gives d f i = − ι V Ω df_i=-\iota_V\Omega d f i = − ι V Ω for every i i i . Evaluation on R k R_k R k yields R k ( f i ) = 0 R_k(f_i)=0 R k ( f i ) = 0 , and subtraction gives d ( f i − f k ) = 0 d(f_i-f_k)=0 d ( f i − f k ) = 0 . The last assertion follows directly from the common right-hand side of the dissipation equations.
Thus a fixed strict symmetry contributes at most one functionally independent nonconstant function among its q q q contractions. Its constant differences should not be counted as additional independent conservation laws. In the decomposition of Proposition 5.2, f i = − h + c i f_i=-h+c_i f i = − h + c i , with c 1 = 0 c_1=0 c 1 = 0 , and the common dissipation equation is equivalently
X H ( h ) = − h E ( H ) + ∑ i = 2 q c i R i ( H ) . X_H(h)=-hE(H)+\sum_{i=2}^{q}c_iR_i(H). X H ( h ) = − h E ( H ) + i = 2 ∑ q c i R i ( H ) .
Remark 5.2. The infinitesimal symmetries for which all λ i ( V ) \lambda_i(V) λ i ( V ) are first integrals form a Lie ideal in the algebra of infinitesimal symmetries. Indeed, write S V = ∑ i R i ( H ) λ i ( V ) S_V=\sum_iR_i(H)\lambda_i(V) S V = ∑ i R i ( H ) λ i ( V ) . For another infinitesimal symmetry W W W , [ W , R i ] = 0 [W,R_i]=0 [ W , R i ] = 0 , W ( H ) = 0 W(H)=0 W ( H ) = 0 , and L W λ i = 0 \mathcal{L}_W\lambda_i=0 L W λ i = 0 give W ( S V ) = S [ W , V ] W(S_V)=S_{[W,V]} W ( S V ) = S [ W , V ] . Hence S V = 0 S_V=0 S V = 0 implies S [ W , V ] = 0 S_{[W,V]}=0 S [ W , V ] = 0 .
Corollary 5.1. For a good Hamiltonian system, every infinitesimal symmetry V V V gives first integrals λ 1 ( V ) , … , λ q ( V ) \lambda_1(V),\ldots,\lambda_q(V) λ 1 ( V ) , … , λ q ( V ) . They are invariant under all Reeb flows and differ only by locally constant functions.
For comparison, the scalar dissipation law associated with the Jacobi structure is X H ( f ) = − E ( H ) f X_H(f)=-E(H)f X H ( f ) = − E ( H ) f . It satisfies the familiar Hamiltonian symmetry correspondence:
X H ( f ) = − E ( H ) f ⟺ { f , H } = 0 ⟺ [ X f , X H ] = 0. X_H(f)=-E(H)f\quad\Longleftrightarrow\quad\{f,H\}=0\quad\Longleftrightarrow\quad[X_f,X_H]=0. X H ( f ) = − E ( H ) f ⟺ { f , H } = 0 ⟺ [ X f , X H ] = 0.
The first equivalence uses { f , H } = X H ( f ) + f E ( H ) \{f,H\}=X_H(f)+fE(H) { f , H } = X H ( f ) + f E ( H ) . The second follows from [ X f , X H ] = − X { f , H } [X_f,X_H]=-X_{\{f,H\}} [ X f , X H ] = − X { f , H } and injectivity of f ↦ X f f\mapsto X_f f ↦ X f . Thus scalar dissipated quantities correspond to Hamiltonian dynamical symmetries, which need not preserve each λ i \lambda_i λ i strictly.
Example 5.1 (Linear damping). Let n = 1 n=1 n = 1 , q = 2 q=2 q = 2 , and λ i = d z i − p d x \lambda_i=dz_i-p\,dx λ i = d z i − p d x . For H = 1 2 p 2 + γ z 1 H=\frac{1}{2}p^2+\gamma z_1 H = 2 1 p 2 + γ z 1 , with γ > 0 \gamma>0 γ > 0 , the equation for p p p is p ˙ = − γ p \dot p=-\gamma p p ˙ = − γ p . The field V = ∂ x V=\partial_x V = ∂ x is an infinitesimal symmetry, and λ 1 ( V ) = λ 2 ( V ) = − p \lambda_1(V)=\lambda_2(V)=-p λ 1 ( V ) = λ 2 ( V ) = − p . Theorem 5.1 gives X H ( − p ) = γ p X_H(-p)=\gamma p X H ( − p ) = γ p , hence p ( t ) = p ( 0 ) e − γ t p(t)=p(0)e^{-\gamma t} p ( t ) = p ( 0 ) e − γ t . The two contractions express the same damping law, rather than two independent dissipated quantities.
Example 5.2 (A dynamical symmetry with a nonconstant difference). On the same structure, let H = p 2 / 2 H=p^2/2 H = p 2 /2 and u = z 2 − z 1 u=z_2-z_1 u = z 2 − z 1 . Then X H = p ∂ x + 1 2 p 2 ( R 1 + R 2 ) X_H=p\partial_x+\frac{1}{2}p^2(R_1+R_2) X H = p ∂ x + 2 1 p 2 ( R 1 + R 2 ) , X H ( u ) = 0 X_H(u)=0 X H ( u ) = 0 , and [ X H , R 2 ] = 0 [X_H,R_2]=0 [ X H , R 2 ] = 0 . Consequently V = u R 2 V=uR_2 V = u R 2 is a dynamical symmetry. Its contractions are ( 0 , u ) (0,u) ( 0 , u ) , a dissipated tuple with nonconstant difference. However, L V λ 2 = d u ≠ 0 \mathcal{L}_V\lambda_2=du\ne0 L V λ 2 = d u = 0 . This illustrates why local constancy of the differences in Theorem 5.1 uses strict preservation of the defining forms.
Example 5.3 (Charge–monopole system: rotations). Consider the contact manifold ( M 1 , λ ) (M_1,\lambda) ( M 1 , λ ) of § 3.4 , with λ = − θ L / ( 2 e g ) \lambda=-\theta_L/(2eg) λ = − θ L / ( 2 e g ) and good Hamiltonian H = E / ( 2 e g ) H=E/(2eg) H = E / ( 2 e g ) . For α = 1 , 2 , 3 \alpha=1,2,3 α = 1 , 2 , 3 , let V α V_\alpha V α be the restriction of the cotangent lift of the left-translation for s ↦ e i ϵ α σ / 2 s\mapsto e^{i\epsilon_\alpha\sigma/2} s ↦ e i ϵ α σ /2 . This action commutes with the gauge action and preserves A \mathcal{A} A . Its lift therefore preserves M 1 M_1 M 1 and θ L \theta_L θ L . Since the kinetic energy is rotation invariant, V α V_\alpha V α is an infinitesimal symmetry [7 ] .
With the conventions of § 3.4 , the Pauli identity [ σ α , σ β ] = 2 i ϵ α β γ σ γ [\sigma_\alpha,\sigma_\beta]=2i\epsilon_{\alpha\beta\gamma}\sigma_\gamma [ σ α , σ β ] = 2 i ϵ α β γ σ γ gives V α ( x ) = − ϵ α × x V_\alpha(x)=-\epsilon_\alpha\times x V α ( x ) = − ϵ α × x . Also,
A ( V α ) = − e g x ^ α , m x ˙ ⋅ V α ( x ) = − ( x × m x ˙ ) α . \mathcal{A}(V_\alpha)=-eg\,\hat{x}_\alpha,\qquad m\dot{x}\cdot V_\alpha(x)=-(x\times m\dot{x})_\alpha. A ( V α ) = − e g x ^ α , m x ˙ ⋅ V α ( x ) = − ( x × m x ˙ ) α .
Thus
λ ( V α ) = J α 2 e g , J = x × m x ˙ + e g x ^ . \lambda(V_\alpha)=\frac{J_\alpha}{2eg}, \qquad J=x\times m\dot{x}+eg\,\hat{x}. λ ( V α ) = 2 e g J α , J = x × m x ˙ + e g x ^ .
By Theorem 5.1, these functions are conserved. They give the usual angular momentum with its Poincaré monopole term [7 ] .
The sign is consistent with the equation of motion in § 3.4 :
d d t ( x × m x ˙ ) = e g ∣ x ∣ 3 x × ( x × x ˙ ) = − e g d x ^ d t . \frac{d}{dt}(x\times m\dot{x})=\frac{eg}{|x|^3}x\times(x\times\dot{x})=-eg\frac{d\hat{x}}{dt}. d t d ( x × m x ˙ ) = ∣ x ∣ 3 e g x × ( x × x ˙ ) = − e g d t d x ^ .
Hence d J / d t = 0 dJ/dt=0 dJ / d t = 0 directly.
Example 5.4 (Relativistic particles: conservation along the Reeb flows). Consider the free uniform N N N -contact structure of Example 3.1. Write θ M = ι ∗ θ \theta_M=\iota^*\theta θ M = ι ∗ θ , and let V V V be an infinitesimal Lorentz transformation. Since L V θ M = 0 \mathcal{L}_V\theta_M=0 L V θ M = 0 and V ( τ c ) = 0 V(\tau_c)=0 V ( τ c ) = 0 , each η a \eta_a η a is preserved and η a ( V ) = θ M ( V ) \eta_a(V)=\theta_M(V) η a ( V ) = θ M ( V ) . For the conventions of Example 4.3,
η a ( V ) = − 1 2 ε μ ν J μ ν , a = 1 , … , N . \eta_a(V)=-\frac{1}{2}\varepsilon_{\mu\nu}J^{\mu\nu}, \qquad a=1,\ldots,N. η a ( V ) = − 2 1 ε μν J μν , a = 1 , … , N .
Canonicity alone gives R k ( η a ( V ) ) = 0 R_k(\eta_a(V))=0 R k ( η a ( V )) = 0 for every k k k , as in Proposition 5.2. Thus the angular momentum is conserved along each independent free-particle proper-time flow.
This statement concerns the Reeb dynamics. If one additionally specifies a Lorentz-invariant good Hamiltonian, for example H = P μ P μ H=P_\mu P^\mu H = P μ P μ with P = ∑ a p a P=\sum_a p_a P = ∑ a p a , then V V V is an infinitesimal symmetry of that Hamiltonian system. The same functions are then also first integrals of X H X_H X H . The choice H = 0 H=0 H = 0 gives X H = 0 X_H=0 X H = 0 and is not needed to establish conservation along the Reeb fields.
Canonoid transformations
Every diffeomorphism F F F pulls the given structure back to a uniform q q q -contact structure. Its Reeb fields are F ∗ R j = ( F − 1 ) ∗ R j F^*R_j=(F^{-1})_*R_j F ∗ R j = ( F − 1 ) ∗ R j , with the vector-field pullback understood on the domain of F F F . In contrast to the class used in § 4 , we do not require these fields to equal the original R j R_j R j .
Definition 5.4. A possibly local diffeomorphism F F F is canonoid for ( M , λ ⃗ , H ) (M,\vec{\lambda},H) ( M , λ , H ) if there is a smooth function K K K on its domain such that
ι X H F ∗ λ i = − K , ι X H d F ∗ λ i = d K − ∑ j = 1 q ( F ∗ R j ) ( K ) F ∗ λ j \iota_{X_H}F^*\lambda_i=-K,\qquad\iota_{X_H}dF^*\lambda_i=dK-\sum_{j=1}^{q}(F^*R_j)(K)F^*\lambda_j ι X H F ∗ λ i = − K , ι X H d F ∗ λ i = d K − j = 1 ∑ q ( F ∗ R j ) ( K ) F ∗ λ j
for every i = 1 , … , q i=1,\ldots,q i = 1 , … , q .
The same function K K K must satisfy all the contraction equations. In particular, ( F ∗ ( λ i − λ k ) ) ( X H ) = 0 (F^*(\lambda_i-\lambda_k))(X_H)=0 ( F ∗ ( λ i − λ k )) ( X H ) = 0 for every i , k i,k i , k , and K K K is uniquely determined whenever these equations hold. Pushing forward the defining equations gives the equivalent statement F ∗ X H = X K ~ F_*X_H=X_{\widetilde{K}} F ∗ X H = X K , where K ~ = K ∘ F − 1 \widetilde{K}=K\circ F^{-1} K = K ∘ F − 1 .
Thus, if ( Q , P , Z ) (Q,P,Z) ( Q , P , Z ) are Darboux coordinates on the target, the transformed equations are
Q ˙ a = ∂ K ~ ∂ P a \dot{Q}^{a}=\frac{\partial\widetilde{K}}{\partial P_a} Q ˙ a = ∂ P a ∂ K
P ˙ a = − ∂ K ~ ∂ Q a − P a ∑ j = 1 q ∂ K ~ ∂ Z j \dot{P}_a=-\frac{\partial\widetilde{K}}{\partial Q^a}-P_a\sum_{j=1}^{q}\frac{\partial\widetilde{K}}{\partial Z_j} P ˙ a = − ∂ Q a ∂ K − P a j = 1 ∑ q ∂ Z j ∂ K
Z ˙ i = ∑ a = 1 n P a ∂ K ~ ∂ P a − K ~ . \dot{Z}_i=\sum_{a=1}^{n}P_a\frac{\partial\widetilde{K}}{\partial P_a}-\widetilde{K}. Z ˙ i = a = 1 ∑ n P a ∂ P a ∂ K − K .
The distinction between K K K on the source and K ~ \widetilde{K} K on the target avoids identifying two different coordinate expressions. A canonical transformation is canonoid with K = H K=H K = H .
Theorem 5.2. Let X X X generate a local flow φ s \varphi_s φ s of canonoid transformations, and let K s K_s K s be the corresponding Hamiltonians. Then K 0 = H K_0=H K 0 = H , the functions K s K_s K s depend smoothly on ( s , x ) (s,x) ( s , x ) , and, with κ = ∂ s K s ∣ s = 0 \kappa=\left.\partial_sK_s\right|_{s=0} κ = ∂ s K s ∣ s = 0 ,
κ = − ( L X λ i ) ( X H ) for every i , [ X , X H ] = X X ( H ) − κ . \kappa=-(\mathcal{L}_X\lambda_i)(X_H)\quad\text{for every }i,\qquad[X,X_H]=X_{X(H)-\kappa}. κ = − ( L X λ i ) ( X H ) for every i , [ X , X H ] = X X ( H ) − κ .
Proof. Set Y = X H Y=X_H Y = X H . The canonoid equations are
( φ s ∗ λ i ) ( Y ) = − K s , (\varphi_s^*\lambda_i)(Y)=-K_s, ( φ s ∗ λ i ) ( Y ) = − K s ,
ι Y d φ s ∗ λ i = d K s − ∑ j ( φ s ∗ R j ) ( K s ) φ s ∗ λ j . (6) \iota_Yd\varphi_s^*\lambda_i=dK_s-\sum_j(\varphi_s^*R_j)(K_s)\varphi_s^*\lambda_j. \tag*{(6)} ι Y d φ s ∗ λ i = d K s − j ∑ ( φ s ∗ R j ) ( K s ) φ s ∗ λ j . (6)
The first identity determines K s K_s K s uniquely and smoothly, and gives K 0 = H K_0=H K 0 = H . Its derivative at zero yields
κ = − ( L X λ i ) ( Y ) , i = 1 , … , q . (7) \kappa=-(\mathcal{L}_X\lambda_i)(Y),\qquad i=1,\ldots,q. \tag*{(7)} κ = − ( L X λ i ) ( Y ) , i = 1 , … , q . (7)
Since ∂ s ( φ s ∗ R j ) ∣ s = 0 = [ X , R j ] \left.\partial_s(\varphi_s^*R_j)\right|_{s=0}=[X,R_j] ∂ s ( φ s ∗ R j ) ∣ s = 0 = [ X , R j ] , differentiating the second identity gives
ι Y d L X λ i = d κ − ∑ j ( [ X , R j ] ( H ) + R j ( κ ) ) λ j − ∑ j R j ( H ) L X λ j . (8) \iota_Yd\mathcal{L}_X\lambda_i=d\kappa-\sum_j\bigl([X,R_j](H)+R_j(\kappa)\bigr)\lambda_j-\sum_jR_j(H)\mathcal{L}_X\lambda_j. \tag*{(8)} ι Y d L X λ i = d κ − j ∑ ( [ X , R j ] ( H ) + R j ( κ ) ) λ j − j ∑ R j ( H ) L X λ j . (8)
Together with (5.6), Cartan’s formula equivalently gives
L Y L X λ i = − ∑ j ( [ X , R j ] ( H ) + R j ( κ ) ) λ j − ∑ j R j ( H ) L X λ j . (9) \mathcal{L}_Y\mathcal{L}_X\lambda_i=-\sum_j\bigl([X,R_j](H)+R_j(\kappa)\bigr)\lambda_j-\sum_jR_j(H)\mathcal{L}_X\lambda_j. \tag*{(9)} L Y L X λ i = − j ∑ ( [ X , R j ] ( H ) + R j ( κ ) ) λ j − j ∑ R j ( H ) L X λ j . (9)
Now λ i ( [ X , Y ] ) = X ( λ i ( Y ) ) − ( L X λ i ) ( Y ) = − X ( H ) + κ \lambda_i([X,Y])=X(\lambda_i(Y))-(\mathcal{L}_X\lambda_i)(Y)=-X(H)+\kappa λ i ([ X , Y ]) = X ( λ i ( Y )) − ( L X λ i ) ( Y ) = − X ( H ) + κ . Moreover, using (5.7) and d λ i = Ω d\lambda_i=\Omega d λ i = Ω ,
ι [ X , Y ] Ω = L X ( ι Y Ω ) − ι Y L X Ω = d ( X ( H ) ) − ∑ j X ( R j ( H ) ) λ j − ∑ j R j ( H ) L X λ j − ι Y d L X λ i = d ( X ( H ) − κ ) − ∑ j R j ( X ( H ) − κ ) λ j . \begin{aligned}
\iota_{[X,Y]}\Omega
&=\mathcal{L}_X(\iota_Y\Omega)-\iota_Y\mathcal{L}_X\Omega\\
&=d(X(H))-\sum_jX(R_j(H))\lambda_j-\sum_jR_j(H)\mathcal{L}_X\lambda_j-\iota_Yd\mathcal{L}_X\lambda_i\\
&=d(X(H)-\kappa)-\sum_jR_j(X(H)-\kappa)\lambda_j.
\end{aligned} ι [ X , Y ] Ω = L X ( ι Y Ω ) − ι Y L X Ω = d ( X ( H )) − j ∑ X ( R j ( H )) λ j − j ∑ R j ( H ) L X λ j − ι Y d L X λ i = d ( X ( H ) − κ ) − j ∑ R j ( X ( H ) − κ ) λ j .
Here X ( R j ( H ) ) − [ X , R j ] ( H ) = R j ( X ( H ) ) X(R_j(H))-[X,R_j](H)=R_j(X(H)) X ( R j ( H )) − [ X , R j ] ( H ) = R j ( X ( H )) . The contraction equations therefore identify [ X , Y ] [X,Y] [ X , Y ] with X X ( H ) − κ X_{X(H)-\kappa} X X ( H ) − κ .
The Hamiltonian formula has the same form as in Section 4. Here, however, κ \kappa κ is fixed by the contractions with the one-forms, and the Hamiltonian assignment is injective. There is no Casimir ambiguity.
Remark 5.3. Call X X X infinitesimally canonoid if there is a function κ \kappa κ satisfying (5.6) and (5.7). These first-order conditions are equivalent to [ X , X H ] [X,X_H] [ X , X H ] being Hamiltonian. Indeed, the theorem’s calculation proves one implication. Conversely, suppose [ X , X H ] = X G [X,X_H]=X_G [ X , X H ] = X G and put κ = X ( H ) − G \kappa=X(H)-G κ = X ( H ) − G . Contraction with λ i \lambda_i λ i gives − ( L X λ i ) ( X H ) = κ -(\mathcal{L}_X\lambda_i)(X_H)=\kappa − ( L X λ i ) ( X H ) = κ for every i i i . Also,
ι X H d L X λ i = L X ( ι X H Ω ) − ι X G Ω \iota_{X_H}d\mathcal{L}_X\lambda_i=\mathcal{L}_X(\iota_{X_H}\Omega)-\iota_{X_G}\Omega ι X H d L X λ i = L X ( ι X H Ω ) − ι X G Ω = d κ − ∑ j ( [ X , R j ] ( H ) + R j ( κ ) ) λ j − ∑ j R j ( H ) L X λ j . =d\kappa-\sum_j\bigl([X,R_j](H)+R_j(\kappa)\bigr)\lambda_j-\sum_jR_j(H)\mathcal{L}_X\lambda_j. = d κ − j ∑ ( [ X , R j ] ( H ) + R j ( κ ) ) λ j − j ∑ R j ( H ) L X λ j .
which is (5.7). This characterizes the infinitesimal condition, not the canonoid property of the entire flow.
Example 5.5 (An infinitesimal condition that does not integrate). Let M = R 2 + q M = \mathbb{R}^{2+q} M = R 2 + q , with λ i = d z i − p d x \lambda_i = dz_i - p\,dx λ i = d z i − p d x , and take H = p H = p H = p . Then X H = ∂ x X_H = \partial_x X H = ∂ x . Consider
X = z 1 ∂ p + x ∑ j = 1 q ∂ z j . X = z_1\partial_p + x\sum_{j=1}^{q}\partial_{z_j}. X = z 1 ∂ p + x j = 1 ∑ q ∂ z j . We have L X λ i = ( 1 − z 1 ) d x \mathcal{L}_X\lambda_i = (1-z_1)\,dx L X λ i = ( 1 − z 1 ) d x for every i i i , κ = z 1 − 1 \kappa= z_1-1 κ = z 1 − 1 , and
[ X , X H ] = − ∑ j ∂ z j = X 1 , X ( H ) − κ = 1. [X,X_H] = -\sum_j\partial_{z_j}=X_1,\qquad X(H)-\kappa=1. [ X , X H ] = − j ∑ ∂ z j = X 1 , X ( H ) − κ = 1. Thus X X X is infinitesimally canonoid.
Its flow is globally explicit:
φ s ( x , p , z 1 , … , z q ) = ( x , p + s z 1 + 1 2 s 2 x , z 1 + s x , … , z q + s x ) . \varphi_s(x,p,z_1,\ldots,z_q)=\left(x,p+sz_1+\frac{1}{2}s^2x,z_1+sx,\ldots,z_q+sx\right). φ s ( x , p , z 1 , … , z q ) = ( x , p + s z 1 + 2 1 s 2 x , z 1 + s x , … , z q + s x ) . Consequently,
φ s ∗ λ i = λ i + ( s − s z 1 − 1 2 s 2 x ) d x . \varphi_s^*\lambda_i=\lambda_i+\left(s-sz_1-\frac{1}{2}s^2x\right)dx. φ s ∗ λ i = λ i + ( s − s z 1 − 2 1 s 2 x ) d x . The first canonoid equation forces the unique candidate K s = p − s + s z 1 + 1 2 s 2 x K_s=p-s+sz_1+\frac{1}{2}s^2x K s = p − s + s z 1 + 2 1 s 2 x . The pulled-back Reeb fields are φ s ∗ R 1 = ∂ z 1 − s ∂ p \varphi_s^*R_1=\partial_{z_1}-s\partial_p φ s ∗ R 1 = ∂ z 1 − s ∂ p and φ s ∗ R j = ∂ z j \varphi_s^*R_j=\partial_{z_j} φ s ∗ R j = ∂ z j for j ≥ 2 j\geq2 j ≥ 2 . All of them annihilate K s K_s K s . However,
ι X H d φ s ∗ λ i = d p + s d z 1 , d K s = d p + s d z 1 + 1 2 s 2 d x . \iota_{X_H}d\varphi_s^*\lambda_i=dp+s\,dz_1,\qquad dK_s=dp+s\,dz_1+\frac{1}{2}s^2dx. ι X H d φ s ∗ λ i = d p + s d z 1 , d K s = d p + s d z 1 + 2 1 s 2 d x . The second canonoid equation therefore fails for every s ≠ 0 s\ne0 s = 0 . This provides a finite-flow obstruction for every q ≥ 1 q\geq1 q ≥ 1 , even though the original Hamiltonian is good.
Sufficient conditions for canonoid flows
The preceding example requires a distinction between first-order tests and hypotheses that control the whole flow. Two useful sufficient conditions follow.
Proposition 5.4. If [ X , X H ] = 0 [X,X_H]=0 [ X , X H ] = 0 , the local flow φ s \varphi_s φ s of X X X is canonoid, with K s = φ s ∗ H K_s=\varphi_s^*H K s = φ s ∗ H .
Proof. The flow preserves Y = X H Y=X_H Y = X H . Hence ( φ s ∗ λ i ) ( Y ) = φ s ∗ ( λ i ( Y ) ) = − φ s ∗ H (\varphi_s^*\lambda_i)(Y)=\varphi_s^*(\lambda_i(Y))=-\varphi_s^*H ( φ s ∗ λ i ) ( Y ) = φ s ∗ ( λ i ( Y )) = − φ s ∗ H . Naturality of contraction and differentiation also gives
ι Y d φ s ∗ λ i = φ s ∗ ( ι Y Ω ) = d ( φ s ∗ H ) − ∑ j ( φ s ∗ R j ) ( φ s ∗ H ) φ s ∗ λ j . \iota_Yd\varphi_s^*\lambda_i=\varphi_s^*(\iota_Y\Omega)
=d(\varphi_s^*H)-\sum_j(\varphi_s^*R_j)(\varphi_s^*H)\,\varphi_s^*\lambda_j. ι Y d φ s ∗ λ i = φ s ∗ ( ι Y Ω ) = d ( φ s ∗ H ) − j ∑ ( φ s ∗ R j ) ( φ s ∗ H ) φ s ∗ λ j . These are the canonoid equations.
Proposition 5.5 (Common constant rescalings). Suppose L X λ i = c λ i \mathcal{L}_X\lambda_i=c\lambda_i L X λ i = c λ i for every i i i , where c ∈ R c\in\mathbb{R} c ∈ R is constant. Then the local flow of X X X is canonoid for every Hamiltonian H H H , with
K s = e c s H , [ X , X H ] = X X ( H ) − c H . K_s=e^{cs}H,\qquad[X,X_H]=X_{X(H)-cH}. K s = e cs H , [ X , X H ] = X X ( H ) − cH . Moreover, when q ≥ 2 q\geq2 q ≥ 2 , if L X λ i = μ λ i \mathcal{L}_X\lambda_i=\mu\lambda_i L X λ i = μ λ i for every i i i with a common smooth function μ \mu μ , then μ \mu μ is constant on each connected component.
Proof. For constant c c c , the flow satisfies φ s ∗ λ i = e c s λ i \varphi_s^*\lambda_i=e^{cs}\lambda_i φ s ∗ λ i = e cs λ i and d φ s ∗ λ i = e c s Ω d\varphi_s^*\lambda_i=e^{cs}\Omega d φ s ∗ λ i = e cs Ω . The defining Reeb equations give φ s ∗ R j = e − c s R j \varphi_{s*}R_j=e^{-cs}R_j φ s ∗ R j = e − cs R j . Therefore ( φ s ∗ λ i ) ( X H ) = − e c s H (\varphi_s^*\lambda_i)(X_H)=-e^{cs}H ( φ s ∗ λ i ) ( X H ) = − e cs H , and
ι X H d φ s ∗ λ i = d ( e c s H ) − ∑ j ( e − c s R j ) ( e c s H ) e c s λ j . \iota_{X_H}d\varphi_s^*\lambda_i=d(e^{cs}H)-\sum_j(e^{-cs}R_j)(e^{cs}H)e^{cs}\lambda_j. ι X H d φ s ∗ λ i = d ( e cs H ) − j ∑ ( e − cs R j ) ( e cs H ) e cs λ j .
This proves canonoidness with K s = e c s H K_s=e^{cs}H K s = e cs H . Theorem 5.2, with κ = c H \kappa=cH κ = cH , gives the commutator formula.
Now suppose q ≥ 2 q\geq2 q ≥ 2 and L X λ i = μ λ i \mathcal{L}_X\lambda_i=\mu\lambda_i L X λ i = μ λ i . Set f i = λ i ( X ) f_i=\lambda_i(X) f i = λ i ( X ) . Cartan’s formula, evaluated on R k R_k R k , gives R k ( f i ) = μ δ i k R_k(f_i)=\mu\delta_{ik} R k ( f i ) = μ δ ik . For k ≠ i k\neq i k = i , commutativity of the Reeb fields therefore yields R k ( μ ) = R k R i ( f i ) = R i R k ( f i ) = 0 R_k(\mu)=R_kR_i(f_i)=R_iR_k(f_i)=0 R k ( μ ) = R k R i ( f i ) = R i R k ( f i ) = 0 . Choosing i ≠ k i\neq k i = k proves R k ( μ ) = 0 R_k(\mu)=0 R k ( μ ) = 0 for every k k k .
Differentiating the common rescaling equations gives
L X Ω = d μ ∧ λ i + μ Ω for every i . \mathcal{L}_X\Omega=d\mu\wedge\lambda_i+\mu\Omega\qquad\text{for every }i. L X Ω = d μ ∧ λ i + μ Ω for every i .
Thus d μ ∧ ( λ i − λ j ) = 0 d\mu\wedge(\lambda_i-\lambda_j)=0 d μ ∧ ( λ i − λ j ) = 0 . For Y ∈ ξ Y\in\xi Y ∈ ξ and i ≠ j i\neq j i = j , evaluation on ( Y , R i ) (Y,R_i) ( Y , R i ) gives d μ ( Y ) = 0 d\mu(Y)=0 d μ ( Y ) = 0 . Since T M = ξ ⊕ R TM=\xi\oplus\mathcal{R} T M = ξ ⊕ R , we conclude that d μ = 0 d\mu=0 d μ = 0 . □ \square □
The last assertion is a restriction imposed by simultaneously rescaling at least two defining forms of a uniform structure. For q = 1 q=1 q = 1 , a common conformal rate is simply the ordinary contact conformal rate and need not be constant. The constant-rescaling case above is the direct analogue of the scaling constructions in [5 ] .
Example 5.6. In the Darboux coordinates of Proposition 2.5, the field
X = ∑ a p a ∂ p a + ∑ j z j ∂ z j X=\sum_a p_a\partial_{p_a}+\sum_j z_j\partial_{z_j} X = a ∑ p a ∂ p a + j ∑ z j ∂ z j
satisfies L X λ i = λ i \mathcal{L}_X\lambda_i=\lambda_i L X λ i = λ i . Its flow is φ s ( q , p , z ) = ( q , e s p , e s z ) \varphi_s(q,p,z)=(q,e^sp,e^sz) φ s ( q , p , z ) = ( q , e s p , e s z ) , so it is canonoid for every H H H , with K s = e s H K_s=e^sH K s = e s H .
For n = 1 n=1 n = 1 and H = p 2 / 2 H=p^2/2 H = p 2 /2 , X H = p ∂ q + 1 2 p 2 ∑ j ∂ z j X_H=p\partial_q+\frac{1}{2}p^2\sum_j\partial_{z_j} X H = p ∂ q + 2 1 p 2 ∑ j ∂ z j . Since X ( H ) − H = H X(H)-H=H X ( H ) − H = H , direct computation gives [ X , X H ] = X H [X,X_H]=X_H [ X , X H ] = X H , as predicted. This construction is local on a general manifold; a global rescaling field is an additional geometric requirement.
The results of this section use uniformity throughout. In particular, the common contraction equations, the constant differences for strict symmetries, and the rescaling restriction depend on d λ 1 = ⋯ = d λ q d\lambda_1=\cdots=d\lambda_q d λ 1 = ⋯ = d λ q . The two-monopole structure of Proposition 3.3 requires a separate non-uniform theory.
Conclusions and open problems
We have studied canonical and canonoid transformations on q q q -cosymplectic and uniform q q q -contact manifolds, with particular attention to their origin in constrained Hamiltonian mechanics. The underlying presymplectic geometry determines a characteristic distribution, while the additional one-forms specify a complementary distribution and a distinguished frame of characteristic fields. Our transformation theory is formulated relative to these additional data. Regular first-class constraint surfaces admit local q q q -cosymplectic completions after a choice of transverse clock functions; the first-class property alone does not make the original constraint fields commute or provide a global closed dual coframe. Regular momentum levels for Hamiltonian vector-group actions give an Abelian realization of this construction.
In the q q q -cosymplectic setting, we characterized canonical fields and related generalized canonical transformations to the leafwise symplectic geometry and locally Hamiltonian fields. The Noether theorem identifies quantities conserved both by X H X_H X H and by every Reeb field, hence along every direction X H + ∑ i a i R i X_H + \sum_i a_iR_i X H + ∑ i a i R i . This conservation statement does not require the directional evolution fields themselves to commute.
In the uniform q q q -contact setting, we described canonical fields in terms of Reeb-invariant Hamiltonians and constant Reeb components. We also related dynamical symmetries to dissipated q q q -tuples. For strict infinitesimal symmetries, the differences λ i ( V ) − λ j ( V ) \lambda_i(V)-\lambda_j(V) λ i ( V ) − λ j ( V ) are locally constant, so a single such symmetry does not yield q q q functionally independent conservation laws. More general dynamical symmetries can produce dissipated tuples with nonconstant conserved differences. Common constant rescalings of the defining one-forms generate canonoid transformations for every Hamiltonian; for q ≥ 2 q \ge2 q ≥ 2 , a common infinitesimal conformal factor is necessarily locally constant.
For a one-parameter group of canonoid transformations in either setting, the infinitesimal relation is [ X , X H ] = X L X H − κ [X,X_H] = X_{\mathcal{L}_XH-\kappa} [ X , X H ] = X L X H − κ , with the structural conditions on κ \kappa κ appropriate to each geometry. The counterexamples in § 4 and § 5 show that these infinitesimal conditions do not, by themselves, ensure that the flow consists of canonoid transformations. We also gave sufficient conditions under which the finite-flow property holds.
The physical examples clarify the different roles of the characteristic flows. For free relativistic particles, the chosen clocks are proportional to the individual proper times, and the Lorentz momentum functions are invariant under each corresponding Reeb flow. For interacting constraint models, the existence and physical interpretation of suitable clocks require additional assumptions. In the monopole systems, the Reeb fields generate gauge transformations, while the particle has a single physical time. The rotational symmetry of the one-monopole model recovers the Poincaré angular momentum.
The two-monopole model separates two global issues. Irrationally related charges obstruct the intermediate quotient by the characteristic flow of the total Cartan form, whose orbits are dense in the gauge tori; the quotient by the full gauge torus remains smooth. Independently of this irrationality, the model carries a non-uniform 2-contact structure, and the topology of the bundle obstructs a uniform structure with the prescribed gauge generators as Reeb fields. This obstruction concerns the fixed Reeb frame, rather than all possible uniform structures on the underlying manifold.
These results suggest the following directions.
Non-Abelian characteristic symmetries. The Lie-group formulation of Hamilton–Jacobi theory in [29 , 27 ] suggests retaining a prescribed frame V 1 , … , V q V_1,\ldots,V_q V 1 , … , V q with [ V a , V b ] = ∑ c c a b c V c [V_a,V_b] = \sum_c c^c_{ab}V_c [ V a , V b ] = ∑ c c ab c V c . A dual family necessarily satisfies d λ c ( V a , V b ) = − c a b c \mathrm{d}\lambda_c(V_a,V_b) = -c^c_{ab} d λ c ( V a , V b ) = − c ab c . Thus non-Abelian extensions require modifying the structural conditions themselves: allowing distinct d λ i \mathrm{d}\lambda_i d λ i within the present definition of a q q q -contact structure still gives commuting Reeb fields. A suitable extension should specify the compatible two-forms, the relevant group actions and momentum levels, and the corresponding notions of canonical and canonoid transformation.
Non-uniform structures and their dynamics. Develop transformation theory for non-uniform q q q -contact structures and for q q q -structured manifolds with distinct two-forms, in conjunction with the Noether framework of [18 ] . The two-monopole example requires keeping track of the physical presymplectic form as well as the individual d λ i \mathrm{d}\lambda_i d λ i . A central question is which
transformations preserve the compatibility between these forms and the chosen dynamics. 3. From infinitesimal conditions to canonoid flows. Identify geometric hypotheses that make the infinitesimal canonoid conditions sufficient along the entire flow, and describe the higher-order obstructions exhibited by our counterexamples. The distinction between local Hamiltonian functions and globally defined transformed Hamiltonians should also be included in this analysis.
Global clocks and relativistic covariance. Determine when interacting constraint systems admit global closed clock forms and primitives with the normalization required by Proposition 3.2. It is also necessary to understand the action of the full Poincaré group on these choices. In the free model, translations do not preserve the particular clock forms used here; their action between different choices of geometric data deserves a separate treatment.
Reduction, integrability and dissipative applications. Determine when canonical and canonoid transformations descend through reduction, and how the additional one-forms and Hamiltonians descend with them. Relations with integrability should include the compatibility conditions needed for commuting evolution fields. Thermodynamical models such as those in [17 ] provide a further setting in which to investigate these questions and the interpretation of dissipated tuples.
In a forthcoming paper, we shall investigate more closely the relation between the present geometric approach and Schwinger’s quantum action principle.
Acknowledgment
The research of X. Zhao is supported by NSFC (Grant No. 12401234).
Conflict of interest statement. On behalf of all authors, the corresponding author states that there is no conflict of interest.
Data availability. Data sharing is not applicable to this article as no new data were created or analyzed in this study.