Lifting geometric objects to a cotangent
bundle, and the geometry of the cotangent
bundle of a tangent bundle
M. Crampin
Faculty of Mathematics
The Open University
Walton Hall, Milton Keynes MK7 6AA, U.K.
F. Cantrijn∗ and W. Sarlet
Instituut voor Theoretische Mechanika
Rijksuniversiteit Gent
Krijgslaan 281, B-9000 Gent, Belgium
Abstract
We show that the cotangent bundle T∗ TM of the tangent bundle of
any differentiable manifold M carries an integrable almost tangent
structure which is generated by a natural lifting procedure from the
canonical almost tangent structure (vertical endomorphism) of TM.
Using this almost tangent structure we show that T∗ TM is diffeomorphic to a tangent bundle, namely TT∗M. This provides a new and
geometrically instructive proof of a result of Tulczyjew, which has applications in Lagrangian and Hamiltonian dynamics and in field theory. The requisite general definitions and results concerning liftings of
geometric objects from a manifold to its cotangent bundle are given.
As an application, we shed new light on the meaning of so-called adjoint symmetries of second-order differential equations.
∗
Research Associate of the National Fund for Scientific Research, Belgium
1
Introduction
In a recent paper [1] we have investigated the curious fact that an arbitrary system of second-order ordinary differential equations may be given
both a Lagrangian and a Hamiltonian formulation, with essentially the same
function serving as both Lagrangian and Hamiltonian for the system, by the
introduction of additional variables. Our analysis began with the observation
that the system of differential equations defined by a vector field on a differential manifold N may be embedded in a system of Hamiltonian equations
by taking the complete lift of the vector field to the cotangent bundle T∗N .
We applied this construction in the case where N is already the tangent
bundle of a manifold M and the given vector field a second-order equation
field. In order to obtain the Lagrangian formulation it was necessary to find
an appropriate manifold with the structure of a tangent bundle related to
both TM and T∗ TM. Now T∗ TM is not itself a tangent bundle, of course;
but it happens to be globally diffeomorphic to TT∗M, and this fact played
a key role in our further investigations.
The existence of the diffeomorphism T∗ TM → TT∗M was demonstrated
by Tulczyjew in 1976 [16, 15], though it is not perhaps so well-known as it
might be. Our principal aim in this paper is to give a new proof of this
result using entirely different methods, ones which we believe are geometrically more instructive and appealing. The approach we take to the study
of the geometry of tangent and cotangent bundles depends very much on
exploiting the properties of the canonical geometric objects associated with
them. The obvious example of such an object is the canonical 1-form on the
cotangent bundle, from which its symplectic structure is derived. A somewhat analogous role is played in the case of a tangent bundle by its vertical
endomorphism or almost tangent structure. This is a type (1, 1) tensor field
whose kernel, as a linear map of tangent vectors, coincides with its image
and is just the vertical subspace at each point, and which is integrable in the
sense that its Nijenhuis tensor vanishes [2, 3, 4]. In addition each of these
manifolds carries a canonically defined vector field, as does any vector bundle: namely, its dilation field, which is the infinitesimal generator of dilations
of the fibres.
As a prerequisite for our analysis, we need various techniques for lifting
objects from a manifold N to its cotangent bundle T∗N . Most of these have
been described by Yano and Ishihara [18]. However, we find their approach
1
rather hard going and present therefore, in Sections 2 and 3, a new version of
this theory, with different methods of proof which do not rely on coordinate
calculations. In Section 4 we specialise to the case where N = TM and
arrive at the main result of this paper: a geometrically constuctive proof
of the diffeomorphism T∗ TM → TT∗M. As a by-product of the techniques
used in the proof, it becomes very simple to show that the fibration of T∗ TM
which gives rise to a tangent bundle structure, is actually Lagrangian and
has a Lagrangian cross-section. We therefore thought it instructive to make
a digression at this point; specifically, we construct in an appendix, relying
in part on a theorem of Weinstein [17], a global symplectomorphism from
T∗ TM to T∗ T∗M. The main part of the paper continues with a section on
dilation fields and coordinate expressions for most of the constructions we
have been dealing with.
In Section 7 we comment on applications. It is fair to say that Tulczyjew’s unification of Lagrangian and Hamiltonian mechanics [15, 16] relies for
a great deal on the diffeomorphism T∗ TM → TT∗M. Another application
is our description of the Lagrangian extension of an arbitrary second-order
system, referred to at the beginning. We believe, however, that there is more
to be gained from lifting objects to a tangent or cotangent bundle: such operations may help to understand the cohesion between different results or may
occasionally lead to the discovery of new properties. As an illustration of
this idea we show that some rather surprising results on adjoint symmetries
of second-order systems on TM, derived in [12], become perfectly plausible when one recognizes that they are manifestations of known properties
concerning the Lagrangian extension on T∗ TM.
Our notation is more-or-less standard; but we should warn the reader
that we do not distinguish notationally between the linear action of a type
(1, 1) tensor on covectors and its action on vectors.
2
Lifts to the cotangent bundle
We begin by describing various lifts from an arbitrary differentiable manifold
N to its cotangent bundle πN : T∗N → N .
We assume that the following lifting constructions are already known:
the vertical lift αv of a 1-form α on N
2
the complete lift X̃ of a vector field X on N
together with the following formulae:
[αv , β v ] = 0
[X̃, αv ] = (LX α)v
gY ].
[X̃, Ỹ ] = [X,
The vector fields obtained in this way span the vector fields on T∗N ; we use
this fact repeatedly in the sequel as a means of specifying geometric objects
on T∗N explicitly.
The vertical lifts of 1-forms and complete lifts of vector fields are related
to the canonical 1-form θN of T∗N as follows:
hαv , θN i = 0,
hX̃, θN i = hX
where hX denotes the fibre linear function on T∗N determined by X: hX (x, p)
= hXx , pi. We also have Lαv θN = πN ∗ α and LX̃ θN = 0, from which the
following formulae are easily derived:
dθN (αv , β v ) = 0
dθN (αv , X̃) = αv (hX ) = πN ∗ hX, αi
dθN (X̃, Ỹ ) = h[X,Y ] .
We shall be concerned with two different ways of lifting a type (1, 1)
tensor field from N to T∗N the first of which, now to be described, results
in a vertical vector field. Let R be a type (1, 1) tensor field on N . For each
x ∈ N , R determines a linear endomorphism Rx of the cotangent space T∗x N .
Let ρt be the one-parameter group of transformations of T∗N given by
ρt (x, p) = x, etRx p .
The generator of ρt is a vertical vector field on T∗N which we call the vertical
lift of R and denote Rv .
The vertical lifts of 1-forms and type (1, 1) tensor fields to T∗N are derived from the action of the affine group in the fibres, each of which is of
3
course a vector space. The bracket relations between vertical lifts essentially
reproduce the Lie algebra structure of the affine group:
[αv , β v ] = 0
[αv , Rv ] = R(α)v
[Qv , Rv ] = [Q, R]v
where [Q, R] represents the commutator of Q and R derived from their actions
on vectors. The bracket of a complete lift of a vector field and the vertical
lift of a type (1, 1) tensor field may be computed from consideration of the
flows they generate, and is given by
[X̃, Rv ] = (LX R)v .
By considering the action of ρt it is easy to show that
Rv (hX ) = hR(X) .
Furthermore
dθN (Rv , αv ) = 0
dθN (Rv , X̃) = Rv hX̃, θN i − h[Rv , X̃], θN i
= Rv (hX ) + h(LX R)v , θN i
= hR(X) .
3
Complete lifts of type (1, 1) tensor fields
We next define another lift of a type (1, 1) tensor field on N to T∗N , which
leads this time to a type (1, 1) tensor field again, rather than to a vector field.
The construction is based on the non-degeneracy of dθN and the possibility
of using it, as a consequence, to convert 2-forms (or, in general, type (0, 2)
tensor fields) into type (1, 1) tensor fields, in the manner of raising an index
with a metric.
Any type (1, 1) tensor field R on N determines a fibre linear map τR : T∗N →
∗
T N , fibred over the identity, by
τR (x, p) = (x, Rx p);
4
the complete lift R̃ of R to T∗N is the type (1, 1) tensor field defined by
ιR̃(ξ) dθN = ιξ (τR∗ dθN )
where ι denotes the interior product.
The map τR is just the map whose exponential was used to define the
vertical lift of R. It follows that R̃ satisfies
ιR̃(ξ) dθN = ιξ (LRv dθN ).
The tensor R̃ may be specified explicitly by evaluating it on the vertical
lift of a 1-form and on the complete lift of a vector field, as follows.
Theorem 1 For any 1-form α and vector field X on N
R̃(αv ) = R(α)v
g + (L R)v .
R̃(X̃) = R(X)
X
Proof The proof consists essentially of repeated applications of the preceeding formula. To obtain the second result we use this formula with ξ = X̃.
dθN (R̃(X̃), β v ) = LRv (dθN (X̃, β v ))
+ dθN ((LX R)v , β v ) + dθN (X̃, R(β)v )
= dθN (X̃, R(β)v )
= −πN ∗ hX, R(β)i
= −πN ∗ hR(X), βi
On the other hand
g β v ).
= dθN (R(X),
dθN (R̃(X̃), Ỹ ) = LRv (dθN (X̃, Ỹ ))
+ dθN ((LX R)v , Ỹ ) + dθN (X̃, (LY R)v )
= LRv (h[X,Y ] ) + dθN ((LX R)v , Ỹ ) − hLY R(X)
= hR([X,Y ]) + dθN ((LX R)v , Ỹ ) − h[Y,R(X)] + hR([Y,X])
= dθN ((LX R)v , Ỹ ) + h[R(X),Y ]
g Ỹ ).
= dθN ((LX R)v , Ỹ ) + dθN (R(X),
The second assertion of the theorem now follows.
The first assertion is easily verified by similar considerations with ξ = αv .
5
Corollary 1 The tensor fields R̃ and R are πN -related, in the sense that
πN ∗ ◦ R̃ = R ◦ πN ∗ .
Corollary 2 If Q is another type (1, 1) tensor field on N then
R̃(Qv ) = (Q ◦ R)v .
Proof
dθN (R̃(Qv ), αv ) = LRv dθN (Qv , αv ) = 0
dθN (R̃(Qv ), X̃) = LRv (dθN (Qv , X̃))
− dθN ([R, Q]v , X̃) + dθN (Qv , (LX R)v )
= hRQ(X) − h[R,Q](X)
= hQR(X)
= dθN ((Q ◦ R)v , X̃),
from which the result follows.
Finally, it is easy to prove the following result, by evaluating both sides on
a vertical and a complete lift, using the formulae obtained in Theorem 1.
Corollary 3 For any vector field X on N
LX̃ R̃ = Lg
X R.
We now derive some relationship between the Nijenhuis tensor of a type
(1, 1) tensor on N and the Nijenhuis tensor of its complete lift to T∗N .
We first remind the reader that if R is any type (1, 1) tensor, its Nijenhuis
tensor NR is the type (1, 2) tensor given by
NR (X, Y )
= R2 ([X, Y ]) + [R(X), R(Y )] − R([R(X), Y ]) − R([X, R(Y )]).
This may be rewritten as (ιX NR )(Y ) = NR (X, Y ), where ιX NR denotes the
type (1, 1) tensor LR(X) R − R ◦ LX R.
Lemma 1 For any type (1, 1) tensor R on N
R̃2 (αv ) = R2 (α)v
R̃2 (X̃) = R2g
(X) + (LX R2 )v + (ιX NR )v .
6
Proof These formulae follow directly from those in Theorem 1 and the definition of ιX NR .
Lemma 2 For any type (1, 1) tensor R on N
NR̃ (αv , β v ) = 0
NR̃ (X̃, αv ) = (ιX NR (α))v
g Y)+ ι
NR̃ (X̃, Ỹ ) = NR (X,
[X,Y ] NR + LY (ιX NR ) − LX (ιY NR )
v
.
Proof Again, the proof consists of calculations using the formulae given in
Theorem 1, and in this case also the second formula in Lemma 1. We shall
derive the last formula as an example.
NR̃ (X̃, Ỹ )
= R̃2 ([X̃, Ỹ ]) + [R̃(X̃), R̃(Ỹ ] − R̃([R̃(X̃), Ỹ ]) − R̃([X̃, R̃(Ỹ )])
gY ]) + [R(X)
g + (L R)v , R(Y
g ) + (L R)v ]
= R̃2 ([X,
X
Y
g + (L R)v , Ỹ ]) − R̃([X̃, R(Y
g ) + (L R))v ])
− R̃([R(X)
X
Y
gY ]) + [R(X),
g R(Y
g )] − R̃([R(X),
g Ỹ ]) − R̃([X̃, R(Y
g )])
= R̃2 ([X,
g )] + [R(X),
g (L R)v ] + [(L R)v , (L R)v ]
+ [(LX R)v , R(Y
Y
X
Y
v
v
− R̃([(LX R) , Ỹ ]) − R̃([X̃, (LY R) ])
v
v
g Y ) + (ι
= NR (X,
[X,Y ] NR ) + (Q(X,Y ) )
where the type (1, 1) tensor Q(X,Y ) on N is given by
Q(X,Y ) = L[X,Y ] R2 − L[R(X),Y ] R − L[X,R(Y )] R
− LR(Y ) LX R + LR(X) LY R
+ [LX R, LY R] + LY LX R ◦ R − LX LY R ◦ R
= R ◦ L[X,Y ] R + LY LR(X) R − LX LR(Y ) R
+ LX R ◦ LY R − LY R ◦ LX R
= LY (LR(X) R − R ◦ LX R) − LX (LR(Y ) R − R ◦ LY R).
Thus
g Y)+ ι
NR̃ (X̃, Ỹ ) = NR (X,
[X,Y ] NR + LY (ιX NR ) − LX (ιY NR )
7
v
as asserted.
From these two lemmas there follows
Theorem 2 If a type (1, 1) tensor R on N satisfies R2 = 0 and NR = 0
then its complete lift has similar properties:
R̃2 = 0
4
NR̃ = 0.
The case where the base is a tangent bundle
We now specialise to the case where the base manifold N is already the
tangent bundle τM : TM → M of another manifold M. We shall be concerned therefore with lifting geometrical objects from TM to T∗ TM, and in
particular with the complete lift of the canonical integrable almost tangent
structure on TM. The almost tangent structure on TM is a type (1, 1) tensor, which is often called the vertical endomorphism. We shall denote it S. It
satisfies the conditions ker S = im S, whence S 2 = 0, (the condition for being
an almost tangent structure); and NS = 0 (the condition of integrability).
Theorem 3 The complete lift S̃ of S to T∗ TM is also an integrable almost
tangent structure. With its aid, T∗ TM may be given the structure of a
tangent bundle, and is in fact diffeomorphic to TT∗M.
Proof By Theorem 2, S̃ 2 = 0. It follows that im S̃ ⊆ ker S̃. But
dim ker S̃ + dim im S̃ = 2 dim TM, so that dim im S̃ ≤ dim TM, with equality implying that im S̃ = ker S̃. We show that dim im S̃ ≥ dim TM.
For any 1-form α and vector field X on TM
S̃(αv ) = S(α)v
g + (L S)v .
S̃(X̃) = S(X)
X
Now let β be any basic 1-form on TM: there is a 1-form α on TM such that
β = S(α), and therefore β v ∈ im S̃. Again, let Y be any vertical vector field
on TM which is the vertical lift of a vector field Z on M. Then if Z C denotes
the complete lift of Z to TM, we have Y = S(Z C ); moreover, LZ C S = 0,
g
C ) ∈ im S̃. Thus if {E } is a local basis of vector fields on
and so Ỹ = S̃(Z
a
a
∗
g
ν
M, and {ω } the dual local basis of 1-forms, the vector fields {(τM
ω a )v , E
a}
8
on T∗ TM, where the superscript ν denotes the vertical lift to the tangent
bundle, belong to im S̃. They are clearly linearly independent and dim TM
in number, which proves that dim im S̃ ≥ dim TM, as required.
Thus S̃ is an almost tangent structure. The fact that it is integrable
follows from Theorem 2.
It is true for any integrable almost tangent structure that its image, or
equivalently kernel, distribution is integrable in the sense of Frobenius’s Theorem. In this case the image distribution has, over a suitable open subset
of M, a local basis consisting of complete, pairwise commuting vector fields:
the vector fields given above have these properties, since vertical lifts to either a tangent or a cotangent bundle are necessarily complete as they are
effectively affine vector fields in the fibres, and the complete lift of a complete vector field is necessarily complete; furthermore these vector fields do
commute pairwise as a consequence of the bracket relations for complete and
vertical lifts. Thus each leaf of the image distribution is diffeomorphic to
R2m , where m = dim M, or at worst to a quotient space of it by a discrete
group of translations. But the latter possibility is ruled out by the fact that
the leaf projects onto a fibre of TM, and is itself fibered by vector subspaces
of the fibres of T∗ TM → TM. Thus the leaves of the image distribution are
each diffeomorphic to R2m .
We next define an imbedding of T∗M into T∗ TM. Consider the zero
section M0 of TM, which we may identify with M itself. The tangent space
to TM at a point (x, 0) in M0 has a direct sum decomposition T(x,0) TM =
Z ⊕ V where Z is the subspace consisting of vectors tangent to the zero
section and V the subspace consisting of vectors tangent to the fibre, that
is, the vertical subspace. Each of these is a copy of Tx M. Define a map
T∗M → T∗ TM by mapping (x, p) to the covector p̂ at (x, 0) ∈ TM defined
in terms of the direct sum decomposition by h(z, v), p̂i = hv, pi. Thus T∗M is
identified by this map with TM⊥
0 , the annihilator, along the zero section M0
of TM, of the tangent spaces to M0 . Under the projection πTM : T∗ TM →
TM the submanifold TM⊥
0 maps onto M0 .
We now show that TM⊥
0 is a cross-section of the distribution im S̃, that
is to say, that it intersects each of its leaves in exactly one point. The
projection of any leaf onto TM is a fibre of τM , and therefore intersects the
projection M0 of TM⊥
0 in exactly one point, say (x, 0). The leaf therefore
intersects the restriction of T∗ TM to TM⊥
0 in a subset of the fibre over
9
(x, 0), that is, T∗(x,0) TM. Now the restriction of the distribution to this
vector space is the subspace generated by the vertical lifts to T∗(x,0) TM of
basic covectors at (x, 0). In terms of the direct sum decomposition used to
construct TM⊥
0 this is the annihilator of the tangent space to the fibre of τM ,
which is complementary to the annihilator of the tangent space to the zero
section. Thus the leaf of the distribution im S̃ intersects T∗(x,0) TM in exactly
one point, the zero covector; and so the leaf intersects TM⊥
0 in exactly one
point.
Thus T∗ TM is the total space of a vector bundle τ̂TM : T∗ TM → TM⊥
0
whose fibres are the leaves of the distribution im S̃. Let ψ denote the imbedding T∗M → T∗ TM. Then at each point (x, p) ∈ T∗M the map ψ∗ is a
linear isomorphism of T(x,p) T∗M with the tangent space to the image TM⊥
0
⊥
at ψ(x, p). But TM0 is a cross-section of the distribution im S̃, and so its
tangent space at any point is complementary to the image space of S̃ at that
point. Thus S̃ψ(x,p) maps the tangent space to TM⊥
0 at ψ(x, p) linearly and
isomorphically onto im S̃ψ(x,p) . Now im S̃ψ(x,p) is the tangent space to the leaf
of the image distribution through ψ(x, p) , and may be identified with the
leaf itself, since it is a vector space. The map Ψ = I ◦ S̃ ◦ ψ∗ defined in this
way, where I represents the identification of the tangent space to a vector
space at its origin with the vector space itself, is a diffeomorphism of TT∗M
with T∗ TM which is a linear bundle map with respect to the vector bundle
structures τT∗M : TT∗M → T∗M and τ̂TM : T∗ TM → TM⊥
0 and matches the
integrable almost tangent structures on the two manifolds.
The maps introduced in the proof of Theorem 3 may be conveniently
incorporated into a commutative diagram:
10
Ψ
TT∗M
✲ T∗ TM
τT∗M
τ̂TM
❄
❄
✲
T∗M
TM⊥
0
ψ
It is further interesting to observe the following property of the fibration
τ̂TM .
Theorem 4 The fibration τ̂TM : T∗ TM → TM⊥
0 is Lagrangian with respect
∗
to the canonical symplectic structure of T TM, and TM⊥
0 is a Lagrangian
cross-section.
Proof For any basic 1-forms α and β on TM and any vertical lifts V and
W to TM of vector fields on M,
dθTM (αv , β v ) = 0;
dθTM (αv , Ṽ ) = πTM ∗ hV, αi = 0
because V is vertical and α basic; and
dθTM (Ṽ , W̃ ) = h[V,W ] = 0
because the bracket of two vertical lifts to TM is zero. Since these vector
fields span the distribution whose leaves are the fibres of τ̂TM it follows that
the fibration is Lagrangian.
A vertical lift αv is tangent to TM⊥
0 if and only if the 1-form α on TM
annihilates the tangent spaces to the zero section M0 . A complete lift X̃, on
v
the other hand, is tangent to TM⊥
0 if and only X is tangent to M0 . If α ,
⊥
β v , X̃ and Ỹ are tangent to TM⊥
0 then, on TM0 ,
dθTM (αv , β v ) = 0;
11
dθTM (αv , X̃) = πTM ∗ hX, αi = 0
because X is tangent to M0 and α annihilates vectors tangent to it; and
dθTM (X̃, Ỹ ) = h[X,Y ] = 0
since [X, Y ] is tangent to M0 , while the value of h[X,Y ] at any point of TM⊥
0
involves the pairing of [X, Y ] with a covector which annihilates vectors tan∗
gent to M0 . Thus TM⊥
0 is a Lagrangian submanifold of T TM.
According to a theorem of Weinstein [17], if a symplectic manifold has a
Lagrangian foliation which admits a cross-section which is also a Lagrangian
submanifold then a neighbourhood of the cross-section is symplectomorphic
to a neighbourhood of the zero section in the cotangent bundle of the crosssection, in such a way that the fibres of the two manifolds correspond. Thus
in the present case we are assured of the existence of a symplectic bundle
diffeomorphism T∗ TM → T∗ T∗M, at least in neighbourhoods of the corresponding cross-sections. The fact that each of these manifolds is a vector
bundle means that the diffeomorphism is global and indeed fibre linear. Actually, we are able to describe an explicit construction of this diffeomorphism,
which is presented in the appendix.
5
Dilation fields
As well as the (almost) tangent structure of T∗ TM we must consider the
associated dilation field.
The dilation field ∆ on TM, the infinitesimal generator of dilations
(x, u) 7→ (x, et u), satisfies
S(∆) = 0
L∆ S = −S
∆ vanishes on the zero section
and these properties determine it uniquely. From them follow certain prop˜ of ∆ to T∗ TM.
erties of the complete lift ∆
12
Lemma 3 The complete lift of ∆ satisfies
˜ = −S v
S̃(∆)
L∆˜ S̃ = −S̃.
Proof
g + (L S)v = −S v
˜ = S(∆)
S̃(∆)
∆
L∆˜ S̃ = Lg
∆ S = −S̃
as required.
On the other hand, T∗ TM carries a dilation field ∆∗ by virtue of the
fact that it is a cotangent bundle. When one takes the Lie derivative with
respect to ∆∗ of a geometric object on T∗ TM which is homogeneous in the
fibre coordinates one obtains a result which incorporates the homogeneity
degree in the manner of Euler’s Theorem on homogeneous functions. The
vector field ∆∗ has the following properties in relation to S̃:
Lemma 4 The dilation field ∆∗ satisfies
S̃(∆∗ ) = S v
L∆∗ S̃ = 0.
Proof For any 1-form α and vector field X on TM
dθTM (S̃(∆∗ ), αv ) = LS v dθTM (∆∗ , αv ) = 0
dθTM (S̃(∆∗ ), X̃) = LS v (dθTM (∆∗ , X̃))
− dθTM (LS v ∆∗ , X̃) + dθTM (∆∗ , (LX S)v )
= LS v (dθTM (∆∗ , X̃))
since LS v ∆∗ = −L∆∗ S v = 0, S v being homogeneous of degree 0. Now
dθTM (∆∗ , X̃) = ∆∗ hX̃, θTM i = ∆∗ hX = hX
in view of the homogeneity of hX . Consequently,
LS v (dθTM (∆∗ , X̃)) = LS v hX = hS(X) = dθTM (S v , X̃).
13
The first result follows.
For the second we use the formula
(L∆∗ S̃)(ξ) = [∆∗ , S̃(ξ)] − S̃([∆∗ , ξ]).
With ξ = αv the right hand side becomes
[∆∗ , S(α)v ] − S̃([∆∗ , αv ]) = −S(α)v + S̃(αv ) = 0
since a vertical lift of a 1-form is, in effect, homogeneous of degree −1. With
ξ = X̃ we obtain for the right hand side
g + (L S)v ] = 0
[∆∗ , S̃(X̃)] = [∆∗ , S(X)
X
because complete lifts of vector fields and vertical lifts of type (1, 1) tensor
fields are both homogeneous of degree 0. This completes the proof.
Using these results we can obtain the dilation field associated with S̃.
Theorem 5 The dilation field D associated with S̃ is given by
˜ + ∆∗ .
D=∆
Proof
From Lemmas 3 and 4 we have
˜ + ∆∗ ) = 0
S̃(∆
L∆+∆
∗ S̃ = −S̃.
˜
˜ + ∆∗ vanishes on the “zero” section
It remains to be shown that D = ∆
⊥
∗
TM0 of the fibration of T TM determined by S̃. Now ∆ generates the
dilations (x, u) 7→ (x, et u) of TM, which leave M0 invariant. It follows that
˜ maps TM⊥
the one-parameter group generated by ∆
0 to itself. In constructing the complete lift of a vector field to a cotangent bundle one takes the
inverse of the induced map of cotangent vectors. Bearing this in mind, as
˜ generates on TM⊥ a onewell as the linearity of the action, one sees that ∆
0
parameter group of transformations which may be written (x, p) 7→ (x, e−t p)
when that space is identified with T∗M. On the other hand, ∆∗ generates
the one-parameter group (x, p) 7→ (x, et p). Thus the one-parameter groups
˜ and ∆∗ , both of which leave TM⊥ invariant, are inverses of
generated by ∆
0
each other when restricted to that submanifold; so their generators satisfy
14
˜ = −∆∗ there as required.
∆
In the proof of the theorem in the appendix we shall use the following
result.
Lemma 5 The dilation field D satisfies
LD θTM = θTM .
Proof We have
L∆˜ θTM = 0
because the Lie derivative of θTM along any complete lift vanishes, and
L∆∗ θTM = θTM
by homogeneity.
6
Coordinate formulae
Before proceeding to an application, we collect together coordinate formulae
for some of the quantities defined in the preceeding sections of the paper.
In the first place, given a manifold N with local coordinates (xi ), and
coordinates (xi , pi ) on T∗N adapted to the cotangent bundle structure, we
have the expressions
α v = αi
∂
∂pi
for the vertical lift of a 1-form α = αi dxi , and
X̃ = X i
∂X i ∂
∂
−
p
i
∂xi
∂xj ∂pj
for the complete lift of a vector field X = X i ∂/∂xi . The vertical lift of a
type (1, 1) tensor field R = Rji (∂/∂xi ) ⊗ dxj is
Rv = pi Rji
∂
∂pj
15
and its complete lift is
R̃ =
Rji
!
∂
∂Rik ∂Rjk
∂
j
⊗
dx
+
⊗
dp
−
+
p
i
k
∂xi
∂pj
∂xj
∂xi
!
∂
⊗ dxj .
∂pi
The following relations, similar in nature to the ones of Theorem 1, will
be useful in the next section.
Lemma 6 For a non-singular type (1, 1) tensor field R on N we have
τR∗ αv = R(α)v
τR∗ X̃ = X̃ + (R−1 ◦ LX R)v .
Proof It is straightforward to verify these formulae in coordinates.
We specialise now to the case N = TM. We take tangent bundle coordinates (xi , ui ) on TM with corresponding coordinates (xi , ui , yi , vi ) on T∗ TM.
The vertical endomorphism on TM is given by
S=
∂
⊗ dxi
i
∂u
and its complete lift by
S̃ =
∂
∂
⊗ dxi +
⊗ dvi .
i
∂u
∂yi
Its image distribution is spanned by the coordinate vector fields ∂/∂ui and
∂/∂yi . The imbedding ψ: T∗M → T∗ TM has the coordinate representation
(xi , pi ) 7→ (xi , 0, 0, pi ).
The point with coordinates (xi , pi , ri , si ) in TT∗M corresponds to the vector
ri
∂
∂
+ si
i
∂x
∂pi
at the point (xi , pi ) in T∗M; its image under ψ∗ is the vector
ri
∂
∂
+
s
i
∂xi
∂vi
16
at the point (xi , 0, 0, pi ) in T∗ TM. The result of applying S̃ to this is the
vector
ri
∂
∂
+ si
i
∂u
∂yi
tangent to the fibre of τ̂TM , which determines the point with coordinates
(xi , ri , si , pi ) in T∗ TM. Thus the diffeomorphism Ψ: TT∗M → T∗ TM derived in Theorem 3 has the coordinate representation
(xi , pi , ri , si ) 7→ (xi , ri , si , pi ).
(The simplicity of this coordinate representation makes a coordinate definition seem appealing: but it is worth pointing out that the confirmation that
the map is actually well-defined, by consideration of the effects of coordinate
transformations, is neither straightforward nor informative.)
The three dilation-related vector fields discussed in Section 5 are given
by
˜ = ui ∂ − v i ∂
∆
∂ui
∂vi
∆ ∗ = yi
D = ui
7
∂
∂
+ vi
∂yi
∂vi
∂
∂
+ yi
.
i
∂u
∂yi
Application: adjoint symmetries and the
Lagrangian extension of second-order systems
Let us first make some general comments about the possible relevance of lifting objects to a tangent or cotangent bundle of a manifold. When a manifold
N is the natural carrier space for some dynamical system, it is perhaps not
to be expected that essentially new features will be discovered by looking at
lifted events, taking place say on TN or on T∗N . Nevertheless, one is constantly forced to keep an eye on these spaces, if only as image spaces of vector
17
fields and 1-forms on N . It even occasionally can advance our understanding
if we look at such images rather than at the objects on N themselves. A
nice example in this respect is Tulczyjew’s description of mechanics in terms
of special symplectic structures [15], in which Lagrangian and Hamiltonian
mechanica appear—loosely speaking—as two different manifestations of the
same Lagrangian submanifold on TT∗M. A key role in this description was
played by the diffeomorphism Ψ: TT∗M → T∗ TM, the geometry of which
we hope to have fully unravelled above. Note further that a similar approach
later proved to be useful in field theory, in particular with respect to the
energy-momentum tensor (see [7]).
Our present application is intended to shed new light on the meaning
of so-called adjoint symmetries of an arbitrary second-order equation field
Γ on TM. Among other things, it was shown in [12] that adjoint symmetries, which are related to invariant 1-forms, can give rise to first integrals or
can generate, under appropriate circumstances, a Lagrangian for the system.
Such occurrencies are much better understood when it concerns symmetry
vector fields of a system which is a priori known to be Lagrangian. New
insights in the role of adjoint symmetries can therefore be expected if we lift
the relevant objects to T∗ TM and relate them to the Lagrangian extension
of Γ, referred to in the Introduction.
Let us first recall some definitions and results from [11, 12]. To every
second-order equation field Γ on TM we associate the following two sets, of
1-forms and vector fields respectively:
XΓ∗ = {α ∈ X ∗ (TM) | LΓ (S(α)) = α},
XΓ = {X ∈ X (TM) | S([Γ, X]) = 0}.
We have at our disposal the following projection operators:
πΓ : X ∗ (TM) → XΓ∗ , α → πΓ (α) = LΓ (S(α)),
πΓ : X (TM) → XΓ , X → πΓ (X) = X + S([Γ, X]).
A 1-form α on TM is said to be an adjoint symmetry of Γ if α ∈ XΓ∗ and
LΓ α ∈ XΓ∗ . An equivalent formulation is that α is an adjoint symmetry
if and only if LΓ S(α) is an invariant form for Γ. In coordinates, an adjoint
symmetry is of the form α = αi dui +Γ(αi ) dxi , where the functions αi satisfy
the adjoint linear variational equations of Γ.
18
Concerning the Lagrangian extension of Γ to T∗ TM, this is obtained as
follows. The vector field Γ on TM induces a function on T∗ TM (as does
every vector field), namely the function hΓ = hΓ̃, θTM i. Choosing L ≡ hΓ ,
and using the tangent bundle structure of T∗ TM we proceed to construct
a Lagrangian vector field ΓL in the usual way, that is, we introduce the
Poincaré–Cartan 1-form θL = S̃(dL) and denote by ΓL the vector field determined by
ιΓL dθL = −dEL ,
where EL is the energy function associated to L. The function hΓ is fibre
linear with respect to the cotangent bundle structure, but not with respect
to the tangent bundle structure; in fact dθL is a symplectic form, as will
become clear from the next argument. The type (1, 1) tensor field LΓ S on
TM is non-singular: in fact (LΓ S)2 = 1. Therefore, τLΓ S , which henceforth
will be abbreviated to τΓ , is a diffeomorphism of T∗ TM. The main result,
reported in [1] states that
τΓ ∗ dθTM = −dθL , τΓ ∗ L = −EL ,
and
τΓ ∗ Γ̃ = τΓ−1 ∗ Γ̃ = ΓL .
Since τΓ (which incidentally is equal to τΓ−1 ) is fibred over the identity map of
TM, it follows from the last relation that ΓL projects onto Γ. We call ΓL the
Lagrangian extension of Γ. Naturally, ΓL being a second-order equation field,
we can also introduce the sets XΓ∗L and XΓL on T∗ TM, with the corresponding
projection operators πΓL . The main point we wish to make now is that adjoint
symmetries of Γ become symmetries of the Lagrangian extension ΓL , when
vertically lifted to T∗ TM.
Before proceeding, we should issue a little warning here. In those formulae
of the previous section where a composition of (1, 1) tensors occurred, the
tensors were regarded as linear maps on the module of vector fields. In some
of the subsequent calculations, we will be dealing with the dual picture,
where (1, 1) tensors which act linearly on 1-forms are composed. The order
of composition is then of course the opposite of the order of composition in
the action on vector fields. As an example, let us recall that, for the action
on 1-forms on TM we have the properties
S = S ◦ LΓ S = −LΓ S ◦ S.
19
Lemma 7 For any 1-form α on TM, we have
πΓL (αv ) = πΓ (α)v .
Proof Using various results of the previous sections and the fact that τΓ =
τΓ−1 , we have
[ΓL , αv ] = [τΓ−1 ∗ Γ̃, αv ] = τΓ−1 ∗ [Γ̃, τΓ∗ αv ]
= τΓ∗ [Γ̃, (LΓ S(α))v ] = τΓ∗ (LΓ (LΓ S(α)))v
=
It follows that
v
LΓ S(LΓ (LΓ S(α))) .
S̃([ΓL , αv ]) =
=
(S ◦ LΓ S)(LΓ (LΓ S(α)))
S(LΓ (LΓ S(α)))
v
v
= (LΓ (S ◦ LΓ S(α)) − (LΓ S)2 (α))v
= (LΓ (S(α)))v − αv
whence
πΓL (αv ) = αv + S̃([ΓL , αv ]) = LΓ (S(α))v = πΓ (α)v ,
as asserted.
Theorem 6 Let α be a 1-form on TM, then
(i) α ∈ XΓ∗ ⇐⇒ αv ∈ XΓL ,
(ii) α is an adjoint symmetry of Γ ⇐⇒ αv is a symmetry of ΓL .
Proof (i) This is an immediate consequence of Lemma 7.
(ii) It follows from the first calculation in the proof of Lemma 7 that
[αv , ΓL ] = 0 ⇐⇒ LΓ (LΓ S(α)) = 0,
which is exactly what we wished to show.
Through the above stated equivalence, we are now in a position to reinterprete properties of adjoint symmetries of arbitrary second-order equations in terms of their more familiar counterparts for the Lagrangian system
20
ΓL . Alternatively—and perhaps more importantly—one could arrive at new
properties of adjoint symmetries by searching for those known properties of
Lagrangian systems which project down to TM with respect to the lifting
process under consideration.
For a start, let us discuss the case where an adjoint symmetry α of Γ
satisfies the additional requirement LΓ S(α) = dF for some function F . Then,
clearly, Γ(F ) is a constant c (on connected components of TM) and for c = 0,
F is a first integral. We have shown in [12] that if Γ itself happens to be
Lagrangian, this is nothing but a dual version of Noether’s theorem. The
very simple relationship between firts integrals F and adjoint symmetries
α, expressed by LΓ S(α) = dF , may therefore come forward as something
more fundamental than Noether’s theorem, because it equally applies when
Γ is not Lagrangian. This now should no longer surprise us because we are
about to show that it is actually a manifestation of Noether’s theorem for the
Lagrangian extension ΓL on T∗ TM. For the sake of clarity, we recall that a
Noether symmetry of a Lagrangian system is a vector field which leaves both
the Poincaré–Cartan 2-form and the energy function invariant.
Lemma 8 For any 1-form α on TM, we have
∗
LΓ S(α) = dF ⇐⇒ ιαv dθL = −πTM
(dF )
∗
hΓ, LΓ S(α)i.
αv (EL ) = −πTM
Proof We have
ιαv dθL = −ιαv τΓ ∗ dθTM = −τΓ ∗ ιτΓ∗ αv dθTM
∗
(LΓ S(α)))
= −τΓ ∗ ιLΓ S(α)v dθTM = −τΓ ∗ (πTM
∗
= −πTM
(LΓ S(α)),
from which the result readily follows.
For the second part
αv (EL ) =
=
=
=
−αv (τΓ ∗ L) = −τΓ ∗ (τΓ∗ αv (L))
−τΓ ∗ (LΓ S(α)v (hΓ ))
∗
−τΓ ∗ (πTM
hΓ, LΓ S(α)i)
∗
−πTM hΓ, LΓ S(α)i,
21
as asserted.
The content of the next statement is now obvious.
Theorem 7 If α is an adjoint symmetry of Γ on TM then LΓ S(α) = dF if
and only if αv is a symmetry of the Lagrangian extension ΓL for which ιαv dθL
is exact. Under these circumstances F is a first integral of Γ on TM if and
only if αv is a Noether symmetry of ΓL on T∗ TM.
Secondly, we wish to illustrate briefly how the reverse procedure of “projecting” down known results for a Lagrangian system such as ΓL can, in
principle, lead to the discovery of new properties concerning adjoint symmetries of Γ.
It is well known that a point symmetry of a Lagrangian system (which
is not of Noether type) produces an alternative (or subordinate) Lagrangian
(see [9]). The same thing may still happen for symmetries Y depending on
the “velocities”, provided an extra condition is satisfied, guaranteeing that
the Lie derivative with respect to Y of the original Poincaré–Cartan 2-form
is again a Poincaré–Cartan form (see form example [10]). To be specific,
for the Lagrangian system ΓL at hand, the extra condition amounts to the
existence of functions L′ and f such that
LY θL = θL′ + df.
The general idea now is to require that Y be a symmetry vector field of the
form αv for some 1-form α on TM and that both L′ and f are the pull back
of functions on TM. Under these circumstances one can readily verify that
the above requirement will be satisfied, provided α is an adjoint symmetry
of Γ with the property
α = πΓ (dF )
for some function F . The expectation then is that the latter restriction on
adjoint symmetries will lead to an interesting conclusion. And it surely does,
because we know from [12] that any second-order equation field Γ (a priori
not of Lagrangian type), for which an adjoint symmetry α exists of the form
α = πΓ (dF ), turns out to be Lagrangian afterall with Γ(F ) as Lagrangian
function.
22
There is a reasonable chance that the kind of techniques employed in this
application would also be fruitful in field theory. For example, our notion of
adjoint symmetry of a second-order equation was inspired by work of Gordon
[6] which originates from field theory [5], where the adjoint linear equation
of a partial differential equation appears to be more popular in defining the
notion of symmetry. Of course, to even think of a further analogy, one
would need a notion of almost tangent structure on jet bundles and some
analogue of a second-order equation field. These are, however, exactly the
kind of questions which have been tackled in recent work on jet fields by
Saunders [13, 14], whereby it must be stipulated that Saunders’s concept of
jet field is directly related to the Cartan–Ehresmann connections in the work
of Mangiarotti and Modugno [8].
Appendix
The construction below of a linear bundle diffeomorphism Φ: T∗ TM →
T∗ T∗M follows the first stage of the general construction given by Weinstein. At one stage in the proof we use the fact that, given a vector at a
point of T∗ TM which is tangent to the fibre of τ̂TM , one may always find
a local vector field which generates fibre translations, leaves dθTM invariant
and agrees with the given vector at its point of definition. To see this, observe
first that translations in the fibre of τ̂TM leave dθTM invariant, provided, in
the case of those coming from the vertical lifts of basic 1-forms, that these
1-forms are closed. This follows from the general fomulae Lαv θN = πN ∗ α and
LX̃ θN = 0 quoted in Section 2. The remainder of the argument relies on the
fact that a given covector at a point can always be regarded as originating
from a closed 1-form defined on a neighbourhood of that point.
Theorem 8 T∗ TM is diffeomorphic to T∗ T∗M by a map which is a linear
bundle map with respect to the vector bundle structures defined by τ̂TM and
∗
πT∗M respectively, which is fibred over ψ −1 : TM⊥
0 → T M, and which is a
symplectomorphism with respect to the canonical symplectic structures of the
two cotangent bundles.
Proof We shall define a map Φ: T∗ TM → T∗ T∗M by using the vector space
∗ ∗
∗
structure of the fibres of τ̂TM : T∗ TM → TM⊥
0 and πT∗M : T T M → T M.
∗
To any point q ∈ T TM there corresponds a vector ζ at τ̂TM (q) ∈ TM⊥
0
23
which is tangent to the fibre of τ̂TM there, by the identification of a vector
space with its tangent space at the origin. The covector ιζ dθTM annihilates
the tangent space to the fibre, since the fibration is Lagrangian, and may
therefore be regarded as an element of T∗τ̂TM (q) TM⊥
0 ; in fact taking the interior product with dθTM at τ̂TM (q) is an isomorphism between the tangent
space to the fibre and the cotangent space to the base, since the tangent
spaces to the fibre and the base are complementary Lagrangian subspaces.
We may therefore define by this procedure a fibre linear diffeomorphism
∗
⊥
∗ ∗
T∗ TM → T∗ TM⊥
0 . Composing this with the map T TM0 → T T M
induced by ψ gives the required map Φ. It is clear from the construction
that Φ is a diffeomorphism, is fibre linear, and is fibred over ψ −1 .
It remains to be shown that Φ is symplectic. The proof is based on the observation that the vector ζ at τ̂TM (q) corresponding to the point q ∈ T∗ TM
is the translate to τ̂TM (q) of Dq (where D is the dilation field associated with
S̃).
We show first that Φ∗ θT∗M = ιD dθTM . For any vector w at q ∈ T∗ TM
we have
hw, Φ∗ θT∗M iq = hΦ∗ w, θT∗M iΦ(q) = hπT∗M∗ Φ∗ w, Φ(q)i
= hψ∗−1 τ̂TM∗ w, Φ(q)i = hτ̂TM∗ w, ιζ dθTM i
= dθTM (ζ, τ̂TM∗ w)τ̂TM (q) .
By the remarks before the statement of the theorem, this final expression is
equal to its translate back to q, along the vector ζ; moreover, the vector w
and its translate to τ̂TM (q) have the same projection under τ̂TM∗ . Thus
hw, Φ∗ θT∗M iq = dθTM (Dq , w),
from which the result follows.
It follows from Lemma 5 that
Φ∗ dθT∗M = d(ιD dθTM ) = LD dθTM = dθTM
and so Φ is symplectic, as required.
Notice that, although Φ is symplectic, it is not the case that Φ∗ θT∗ M =
θTM . In fact these two 1-forms differ by an exact form:
Φ∗ θT∗ M − θTM = ιD dθTM − LD θTM = −dhD, θTM i.
24
As well as the diffeomorphism Φ: T∗ TM → T∗ T∗M defined above, there
is a standard diffeomorphism TT∗M → T∗ T∗M constructed using the symplectic structure of T∗M. It is not difficult to show that the two constructions
are consistent, in the sense that the following diagram commutes:
Ψ
✲ T∗ TM
∗
TT M
❏
❏
❏
✡
✢
✡
❏
❫
❏
✡
✡
✡
Φ
T∗ T∗M
In coordinates, the construction of the symplectomorphism Φ: T∗ TM →
T T∗M proceeds as follows. The canonical 2-form on T∗ TM is
∗
dθTM = dyi ∧ dxi + dvi ∧ dui .
The point (xi , ui , yi , vi ) in T∗ TM determines the vector
ui
∂
∂
+ yi
i
∂u
∂yi
at (xi , 0, 0, vi ) in TM⊥
0 . Its interior product with dθTM is the covector
yi dxi − ui dvi .
The map Φ therefore has the coordinate representation
(xi , ui , yi , vi ) 7→ (xi , vi , yi , −ui ).
The map TT∗M → T∗ T∗M based on the canonical 2-form dθM is given by
(xi , pi , ri , si ) 7→ (xi , pi , si , −ri ),
which is the composition of Φ with the diffeomorphism TT∗M → T∗ TM.
25
References
[1] F. Cantrijn, M. Crampin and W. Sarlet. Evading the Inverse Problem
for Second Order Ordinary Differential Equations by Using Additional
Variables. Inverse Problems, 3:51–63, 1987.
[2] R. S. Clark and M. R. Bruckheimer. Sur les Structures Presque Tangents.
Comptes Rendus de l’Academie des Sciences, Paris, 251:627–629, 1960.
[3] M. Crampin. Tangent Bundle Geometry for Lagrangian Dynamics. Journal of Physics A: Mathematical and General, 16:3755–3772, 1983.
[4] M. Crampin and G. Thompson. Affine Bundles and Integrable Almost
Tangent Structures. Mathematical Proceedings of the Cambridge Philosophical Society, 98:61–71, 1985.
[5] T. J. Gordon. Equivalent conserved currents and generalized Noether’s
theorem. Annals of Physics, 115:85–107, 1984.
[6] T. J. Gordon. On the symmetries and invariants of the harmonic oscillator. Journal of Physics A: Mathematical and General, 19:183–189,
1986.
[7] J. Kijowski and W. M. Tulczyjew. A symplectic framework for field theories. Lecture Notes in Physics 107, Springer–Verlag, Berlin, 1979.
[8] L. Mangiarotti and M. Modugno. Fibred spaces, jet spaces and connections. Proceedings of the meeting “Geometry and Physics”,(October
1982). Pitagora Editrice, Bologna:135–165, 1983.
[9] G. Marmo and E. J. Saletan. Ambiguities in the Lagrangian and Hamiltonian formalism: transformation properties. Nuovo Cimento, 40B:67–
89, 1977.
[10] G. E. Prince. Toward a classification of dynamical symmetries in classical
mechanics. Bulletin of the Australian Mathematical Society, 27:53–71,
1983.
[11] W. Sarlet, F. Cantrijn and M. Crampin. A new look at second-order
equations and Lagrangian mechanics. Journal of Physics A: Mathematical and General, 17:1999–2009, 1984.
26
[12] W. Sarlet, F. Cantrijn and M. Crampin. Pseudo-symmetries, Noether’s
theorem and the adjoint equation. Journal of Physics A: Mathematical
and General, 20:1365–1376, 1987.
[13] D. J. Saunders. An alternative approach to the Cartan form in Lagrangian field theories. Journal of Physics A: Mathematical and General,
20:339–349, 1987.
[14] D. J. Saunders. Jet fields, connections and second-order differential
equations. Journal of Physics A: Mathematical and General, 20:3261–
3270, 1987.
[15] W. M. Tulczyjew. Hamiltonian systems, Lagrangian systems and the
Legendre transformation. Symposia Mathematica, 16:247–258, 1974.
[16] W. M. Tulczyjew. Les Sous-variétés Lagrangiennes et la Dynamique
Lagrangienne. Comptes Rendus de l’Academie des Sciences, Paris,
283:675–678, 1976.
[17] A. Weinstein. Symplectic Manifolds and their Lagrangian Submanifolds.
Advances in Mathematics, 6:329–346, 1976.
[18] K. Yano and S. Ishihara. Tangent and Cotangent Bundles. Marcel
Dekker, New York, 1973.
27