Tangent Dirac structures and submanifolds

by Izu Vaisman

ABSTRACT. We write down the local equations that characterize the submanifolds N   of a Dirac manifold M   which have a normal bundle that is either a coisotropic or an isotropic submanifold of T M   endowed with the tangent Dirac structure. In the Poisson case, these formulas prove again a result of Xu: the submanifold N   has a normal bundle which is a coisotropic submanifold of T M   with the tangent Poisson structure iff N   is a Dirac submanifold. In the presymplectic case it is the isotropy of the normal bundle which characterizes the corresponding notion of a Dirac submanifold. On the way, we give a simple definition of the tangent Dirac structure, we make new remarks about it, and we establish characteristic, local formulas for various interesting classes of submanifolds of a Dirac manifold.

1 Introduction

The framework of the present paper is the C ∞   category. We will denote by Ω k   spaces of differential k   -forms, by χ k   spaces of k   -vector fields, by Γ   spaces of differentiable cross sections of vector bundles, and we will use the Einstein summation convention.
The Dirac structures were introduced in the study of constrained systems and unify Poisson and presymplectic geometry [1] . We will recall their definition later. A pair ( M , D )   that consists of an n   -dimensional manifold M   and a Dirac structure D   on M   is called a Dirac manifold. We started the study of submanifolds of a Dirac manifold in [11] , where we noticed the classes of properly normalized, totally Dirac and cosymplectic submanifolds. These classes extend the Poisson-Dirac with Dirac projection, Lie-Dirac (Dirac) and cosymplectic submanifolds of a Poisson manifold, respectively [13, 3] .
In [13] , Xu proved that the Dirac submanifolds of the Poisson manifold ( M , P )   are characterized by the nice property of having a normal bundle which is a coisotropic submanifold of the tangent manifold T M   endowed with the tangent Poisson structure. The latter is defined by the complete lift [14] of the bivector field P   . All the terms of Xu's result, including the notion of a tangent Dirac structure [2] , are also defined for Dirac manifolds, and Xu's result indicates that one may expect interesting connections between the geometry of a submanifold N   of a Dirac manifold M   and the geometry of a normal bundle of N   in the tangent manifold T M   . This is the motivation of the present paper.
We will discuss the geometric configuration of Xu's result in the general case of a Dirac manifold. The terms of the theorem are either new or not popular, and are based on either new or not popular geometric constructions.
Accordingly, it is an objective of the paper to explain these terms in detail.
Particularly, we recall the general construction of the vertical and complete lifts of tensor fields from a manifold M   to the total space of the tangent bundle T M   , and the main properties of these operations [14] . We use these lifts in order to give a simple definition of the tangent Dirac structure and make some new remarks about it.
Then, we turn to submanifolds. We define various classes of submanifolds of a Dirac manifold and characterize them via local coordinates and bases.
Furthermore, we obtain the local conditions that characterize submanifolds N   of ( M , D )   with a normal bundle ν N   which is either a coisotropic or an isotropic submanifold of T M   . These formulas imply the result proven by Xu in the case of Poisson manifolds. Another consequence of the established formulas is that the analogs of Dirac submanifolds of a presymplectic manifold M   are characterized by the existence of a normal bundle which is isotropic in T M   .

2 Complete and vertical lifts revisited

Let M   be an n   -dimensional, differentiable manifold and T M   be the total space of its tangent bundle. In the space of differentiable functions C ∞ ( T M )   one has the important linear subspace L ( T M )   of the fiberwise linear functions, the latter being functions of the form
l α ( x , v ) = α x ( v ) , x ∈ M , v ∈ T x M , α ∈ Ω 1 ( M ) . (2.1)
In particular, if we denote by x i   ( i = 1 , . . . , n )   local coordinates on M   and by v i   the corresponding natural coordinates on the fibers of T M   (i.e., coordinates of tangent vectors with respect to the bases ( ∂ / ∂ x i )   ), we have l d x i = v i   . Hence, locally, C ∞ ( T M )   is functionally spanned by the set of functions ( π * f = f ∘ π , l d g )   , where π : T M → M   is the natural projection and f , g ∈ C ∞ ( M )   . In what follows the function π * f   will be denoted again by f   .
Two other ingredients also are of great importance in the geometry of T M   .
The first is the Euler vector field E ∈ χ 1 ( T M )   of infinitesimal homotheties of the fibers, which is characterized by
E f = 0 , E l α = l α , f ∈ C ∞ ( M ) , α ∈ Ω 1 ( M ) , (2.2)
and has the local expression
E = v i ∂ ∂ v i . (2.3)
The second is the tangent structure tensor field S ∈ Γ ( E n d T ( T M ) )   , which is characterized by
( S X ) f = 0 , ( S X ) l α = α ( π * X ) , (2.4)
where X ∈ χ 1 ( T M ) , f ∈ C ∞ ( M ) , α ∈ Ω 1 ( M )   , and has the local expression
S ( ξ i ∂ ∂ x i + η i ∂ ∂ v i ) = ξ i ∂ ∂ v i . (2.5)
If, as usual, we denote by V ⊆ T ( T M )   the subbundle tangent to the fibers, called the vertical bundle, we have
S 2 = 0 , k e r S = i m S = V , (2.6)
and S   has a vanishing Nijenhuis tensor:
N S ( X , Y ) = [ S X , S Y ] − S [ S X , Y ] − S [ X , S Y ] + S 2 [ X , Y ] = 0 . (2.7)
Firstly, using the ingredients introduced above and denoting T q p ( M ) = ( ⊗ p T M ) ⊗ ( ⊗ q T * ( M ) )   , we get
Proposition 2.1. [14]  There exists a unique homomorphism of real tensor algebras that sends a tensor t ∈ T q , x p ( M )   to a tensor t V ∈ T q , v p ( T M )   , where v ∈ T M   and π ( v ) = x ∈ M   , called the vertical lift, such that
1 V = 1 , α V = π * α , X V = S X , (2.8)
where α ∈ T x * M   , X ∈ T x M   , and X ∈ T v ( T M )   is any vector such that π * X = X   . Moreover, the vertical lift of a differentiable tensor field is a differentiable tensor field and, for differential forms, the vertical lift commutes with the exterior differential.
  • Proof. We notice that ∀ λ ∈ T v * ( T M )   there exists a unique α λ ∈ T x * M   such that α λ ( X ) = λ ( S X )   ( π * X = X )   . Then, we define
    t V ( X 1 , . . . , X q , λ 1 , . . . , λ p ) = t ( π * X 1 , . . . , π * X q , α λ 1 , . . . , α λ p ) . (2.9)
    The assertions about tensor fields and differential forms follows from the fact that in the case of a differential form Φ   (functions included) one has Φ V = π * Φ   , and in the case of a vector field X = ξ i ( ∂ / ∂ x i )   one has X V = ξ i ( ∂ / ∂ v i )   .
Secondly, we define an operation on tensor fields known as the complete lift [14] . For any vector field X ∈ χ 1 ( M )   , the flow e x p t X   lifts to a local 1   -parameter Lie group ( e x p t X ) *   on the manifold T M   , which is defined by a vector field X C ∈ χ 1 ( T M )   called the complete lift of X   . The local expression of X C   is
X C = ξ i ∂ ∂ x i + v i ∂ ξ j ∂ x i ∂ ∂ v j , (2.10)
and it is easy to check that [14] 
X C f V = ( X f ) V , X C l d f = l d ( X f ) , (2.11)
( X + Y ) C = X C + Y C , ( f X ) C = f V X C + ( l d f ) X V , (2.12)
α V ( X C ) = ( α ( X ) ) V ,  
[ X C , Y C ] = [ X , Y ] C , [ X V , Y C ] = [ X , Y ] V , [ X V , Y V ] = 0 . (2.13)
Furthermore, one has
Proposition 2.2. [14]  There exists a unique homomorphism of real linear spaces which sends a tensor field P ∈ Γ T q p ( M )   to a field P C ∈ Γ T q p ( T M )   , called the complete lift of P   , such that, ∀ f ∈ C ∞ ( M )   , f C = l d f   , ∀ X ∈ χ 1 ( M )   , X C   is given by ( 2.11 ), and
( P ⊗ Q ) C = P C ⊗ Q V + P V ⊗ Q C . (2.14)
  • Proof. Notice that the lift f C   of a function was chosen such that for any vector field seen as X : M → T M   the pull back f C ∘ X = X f   . The definition of f C   and condition ( 2.14 ) compel us to define the complete lift of a 1   -form α ∈ Ω 1 ( M )   by
    α C ( X V ) = ( α ( X ) ) V , α C ( X C ) = ( α ( X ) ) C = l d ( α ( X ) ) . (2.15)
    The corresponding local coordinate expression is
    α C = v j ∂ α i ∂ x j d x i + α i d v i , (2.16)
    and for any vector field X : M → T M   the pull back of α C   is given by X * α C = L X α   , where L   denotes the Lie derivative. Finally, condition ( 2.14 ) uniquely defines the complete lift of an arbitrary tensor field because this condition is compatible with associativity.
We also indicate the following properties of the complete lift:
a) [14] The complete lift of a k   -form Φ ∈ Ω k ( M )   is a k   -form Φ C ∈ Ω k ( T M )   and d Φ C = ( d Φ ) C   . Indeed, a straightforward calculation shows that this condition holds for functions and 1   -forms. Then, the condition for an arbitrary form follows by expressing the latter locally as a real linear combination of exterior products of 1   -forms and using ( 2.14 ).
b) [14] The Lie derivative of a tensor field Φ ∈ Γ T q p ( M )   has the following lifts:
( L X Φ ) V = L X C Φ V = L X V Φ C , ( L X Φ ) C = L X C Φ C , L X V Φ V = 0 . (2.17)
It is enough to check ( 2.17 ) for functions, vector fields and 1   -forms, and this can be done with the already explained formulas ( 2.8 ) ( 2.13 ). Then, the general result follows from ( 2.14 ).
c) The complete lift of a k   -vector field P ∈ χ k ( M )   is a k   -vector field P C ∈ χ k ( T M )   and the Schouten-Nijenhuis bracket satisfies the condition [ P C , Q C ] = [ P , Q ] C   . This follows by expressing P , Q   locally as real linear combinations of exterior products of vector fields, using the expression of the Schouten-Nijenhuis bracket of such exterior products (e.g., [9] ) and ( 2.13 ), ( 2.14 ).
As an application of property c), if P ∈ χ 2 ( M )   is a Poisson bivector field on M   , i.e. [ P , P ] = 0   , then P C   is a Poisson bivector field on the manifold T M   . The Poisson structure defined by P C   is called the tangent Poisson structure and it was used by many authors ([2, 12] , etc.).
d) The complete lifts X C   of all the vector fields X ∈ χ 1 ( M )   span a generalized foliation of T M   such that, if we identify M   with the zero section of T M   , the leaves through points of M   are the connected components of M   and the leaf through v ∈ T M   , v ≠ 0   , is the connected component of v   in T M \ M   . Indeed, by ( 2.10 ), at a point v = 0   we have s p a n { X C / X ∈ χ 1 ( M ) } = s p a n { ∂ ∂ x i = ( ∂ ∂ x i ) C } ,   and at a point v = λ i ( ∂ / ∂ v i )   such that, for instance, λ 1 ≠ 0   , we have s p a n { X C } = s p a n { ∂ ∂ x i = ( ∂ ∂ x i ) C , ∂ ∂ v i = ( x 1 λ 1 ∂ ∂ x i ) C − x 1 λ 1 ∂ ∂ x i } .   Finally, we refer the reader to [6] and the references therein for generalizations of the lift operations discussed above.

3 Tangent Dirac structures

Now, we will use the complete and vertical lifts in order to define the notion of a tangent Dirac structure, first introduced by Courant [2] , and make some new remarks about it.
The Dirac structures are defined as a class of subbundles of the vector bundle E ( M ) = T M ⊕ T * M   . The bundle E ( M )   has several interesting geometric objects. The first is the non degenerate metric of zero signature
g ( ( X , α ) , ( Y , β ) ) = 1 2 ( α ( Y ) + β ( X ) ) , (3.1)
where X , Y   are tangent vectors and α , β   are tangent covectors at x ∈ M   .
The second is F ∈ Γ E n d ( E ( M ) )   given by
F ( X , α ) = ( X , − α ) , (3.2)
which is a so-called para-Hermitian structure [4] . The third object is the non degenerate 2   -form
ω ( ( X , α ) , ( Y , β ) ) = g ( ( X , α ) , F ( Y , β ) ) = 1 2 ( α ( Y ) − β ( X ) ) . (3.3)
Definition 3.1. A maximal g   -isotropic subbundle A ⊆ E ( M )   is called an almost Dirac structure on M   .
The almost Dirac structure may be interpreted in terms of T M   alone.
Namely [1] , A   yields the pair ( A , ϖ )   , where A   is the generalized distribution defined as the natural projection of A   on T M   and, ∀ x ∈ M   , ϖ x ∈ ∧ 2 A x *   is the 2   -form induced by ω   of ( 3.3 ) ( ∀ X , Y ∈ A x   , the value produced by ( 3.3 ) does not depend on the choice of α , β   ). Conversely, the pair ( A , ϖ )   allows us to reconstruct A   as follows
A = { ( X , α ) / X ∈ A & α | A = i ( X ) ϖ } . (3.4)
The next important thing for the bundle E ( M )   is the Courant bracket, which is the operation defined on Γ E ( M )   by [1] 
[ ( X , α ) , ( Y , β ) ] = ( [ X , Y ] , L X β − L Y α + d ( ω ( ( X , α ) , ( Y , β ) ) ) ) (3.5)
= ( [ X , Y ] , i ( X ) d β − i ( Y ) d α + 1 2 d ( β ( X ) − α ( Y ) ) ) ,   where X , Y   are vector fields and α , β   are differential 1   -forms on M   , [ X , Y ]   is the usual Lie bracket and L   denotes the Lie derivative. The Courant bracket is skew-symmetric but satisfies a more complicated than the Jacobi identity.
Definition 3.2. An almost Dirac structure D ⊆ E ( M )   is called a Dirac structure on M   if Γ D   is closed by Courant brackets.
In [1] , it was proven that the almost Dirac structure is Dirac iff the equivalent pair ( A , ϖ )   satisfies the following conditions: i) A   is a generalized foliation, ii) the form ϖ   is closed along the leaves of A   . This means that the leaves of A   are presymplectic manifolds (of a non constant rank!) and are called the presymplectic leaves of D   . If the leaves are symplectic D   is equivalent with a Poisson structure. Namely, if P   is the corresponding Poisson bivector field, the Dirac structure is
D P = { ( i ( α ) P , α ) / α ∈ T * M } . (3.6)
If the leaves are the connected components of M   , D   is a presymplectic structure on M   with the presymplectic form ϖ   such that
D = D ϖ = { ( X , i ( X ) ϖ ) / X ∈ T M } . (3.7)
Another fundamental property of a Dirac structure is that the restriction of the Courant bracket ( 3.5 ) to Γ D   makes D   into a Lie algebroid where the anchor is projection on T M   (e.g., [9] ).
In [2] , a Dirac structure of M   was lifted to the manifold T M   . In what follows, we give a simpler definition of this lift. For this purpose we look at the locally free C ∞ ( M ) ̲   -module sheaf D ̲   of rank n   of the germs of cross sections of D   , where C ∞ ( M ) ̲   is the sheaf of germs of differentiable functions on M   (e.g., [8] ). Then, we get
Proposition 3.1. The C ∞ ( T M ) ̲   -module sheaf D ̲ t g   spanned by the germs ( X C , α C )   , ( X V , α V )   , ∀ ( X , α ) ∈ D ̲   is locally free of rank 2 n   and it is isomorphic with the sheaf of germs of cross sections of a Dirac structure D t g   on T M   .
  • Proof. Firstly, we show that the sheaf D ̲ t g   is locally free of rank 2 n   . If ( B i , ε i )   ( i = 1 , . . . , n )   is a local basis for the sheaf D ̲   on M   , an arbitrary germ ( X , α ) ∈ D ̲   is of the form ( X , α ) = ∑ i = 1 n λ i ( B i , ε i ) ,   whence
    ( X C , α C ) = ∑ i = 1 n [ λ i V ( B i C , ε i C ) + λ i C ( B i V , ε i V ) ] ,
    ( X V , α V ) = ∑ i = 1 n λ i V ( B i V , ε i V ) .
    This shows that ( B i C , ε i C ) , ( B i V , ε i V )   is a local basis of D ̲ t g   . (If we assume that B i = b i j ∂ ∂ x j , ε i = ε i j d x j   and use formulas ( 2.10 ), ( 2.16 ), linear independence follows from that of ( B i , ε i )   .) Thus [8] , D ̲ t g   is isomorphic with the sheaf of germs of cross sections of the vector bundle with local trivialization bases ( B i C , ε i C ) , ( B i V , ε i V )   , which may be identified with a vector subbundle D t g   of T ( T M )   .
    Furthermore, if we denote by indices M   and T M   , respectively, objects on the two manifolds, formulas ( 3.1 ), ( 3.3 ) and ( 2.12 ), ( 2.15 ) give
    g T M ( ( X C , α C ) , ( Y C , β C ) ) = ( g M ( ( X , α ) , ( Y , β ) ) ) C , g T M ( ( X C , α C ) , ( Y V , β V ) ) = ( g M ( ( X , α ) , ( Y , β ) ) ) V , g T M ( ( X V , α V ) , ( Y V , β V ) ) = 0 , (3.8)
    and similar formulas relate ω T M   to ω M   . These formulas ensure the isotropy property for D t g   .
    Finally, from ( 3.8 ) for ω   , ( 2.11 ) ( 2.13 ) and property b) of the complete lift given at the end of Section 1, we get the following formulas for Courant brackets:
    [ ( X C , α C ) , ( Y C , β C ) ] = [ ( X , α ) , ( Y , β ) ] C , [ ( X C , α C ) , ( Y V , β V ) ] = [ ( X , α ) , ( Y , β ) ] V , [ ( X V , α V ) , ( Y V , β V ) ] = 0 . (3.9)
    These formulas, and the fact that a Dirac structure is a Lie algebroid, ensure that D t g   is closed by Courant brackets.
Definition 3.3. The Dirac structure D t g   of T M   provided by Proposition  3.1 is called the tangent Dirac structure of the Dirac structure D   of M   .
The comparison of the generating pairs of D t g   with the bases produced by the computations of [2] or with the alternative definition of the reviewer of that paper [5] shows that the tangent Dirac structure of Definition  3.3 is the same as that of [2] .
Remark 3.1. The tangent Dirac structure is invariant by the operator S   of the tangent structure of the manifold T M   . Indeed, the action of S   is defined by S ( X , Ξ ) = ( S X , Ξ ∘ S )   ( X , Ξ ) ∈ D t g )   , and the definitions of Section 1 yield S X C = X V , S X V = 0 , α C ∘ S = α V , α V ∘ S = 0 .  
Example 3.1. A Poisson structure of M   defined by the bivector field P   with [ P , P ] = 0   is equivalent with the Dirac structure D P   given by ( 3.6 ).
From ( 2.12 ) and ( 2.15 ), it follows easily:
i ( α V ) P C = ( i ( α ) P ) V , i ( α C ) P C = ( i ( α ) P ) C .   Hence, the tangent Dirac structure of D P   is the Poisson structure defined on T M   by the bivector field P C   , which is the usual definition of a tangent Poisson structure. Similarly, if M   has a presymplectic structure defined by the closed 2   -form ϖ   , this structure may be seen as the Dirac structure D ϖ   given by ( 3.7 ) and the tangent Dirac structure of D ϖ   is the presymplectic structure defined on T M   by ϖ C   .
Example 3.2. The construction of the tangent Dirac structure extends to complex Dirac structures L ⊆ E ( M ) ⊗ R C   . Such a structure L   is a generalized complex structure of M   if L ∩ L ¯ = 0   [7] . By looking at a complex basis ( B i , ε i )   of L   , it follows easily that if L   is a generalized complex structure the same is true for its tangent structure L t g   . Therefore, the tangent manifold of a generalized complex manifold is a generalized complex manifold, again, in a canonical way. If M   has a usual complex structure, L   is the direct sum of the holomorphic tangent bundle of M   and the anti-holomorphic cotangent bundle [7] , and L t g   has the similar structure for the usual complex structure of T M   . On the other hand, if the generalized complex structure is defined by a symplectic form ω   on M   , L   is the complexification of the Dirac structure D ω   of ( 3.7 ) [7] , and the generalized complex structure of T M   is defined by the symplectic form ω C   .
Now, we will give some more results about the tangent Dirac structure.
Proposition 3.2. If S   is a presymplectic leaf of D   on M   with the presymplectic form ϖ   and if v ∈ T S   , the presymplectic leaf of D t g   through v   is the tangent manifold T S ⊆ T M   , and its presymplectic form is ϖ C   , where the complete lift is from S   to T S   .
  • Proof. Obviously, the tangent space of the presymplectic leaf S ( x , v ) ( D t g )   of D t g   at a point ( x , v ) ∈ T M   , π ( v ) = x   , is spanned by the vectors X C ( x , v )   , X V ( x , v )   , where X   is a vector field tangent to the presymplectic leaf S x ( D )   .
    In particular, if v ∈ T S   , we get the first part of the proposition. Furthermore, if ϖ   is the presymplectic form of the leaf S   we have [1] 
    ϖ ( X , Y ) = ω ( ( X , α ) , ( Y , β ) ) , (3.10)
    where X , Y ∈ T S   , ( X , α ) , ( Y , β ) ∈ D   and the form ω   is defined by formula ( 3.3 ). Then, the definitions and properties of the complete and vertical lifts yield
    ϖ C ( X C , Y C ) = ω T M ( ( X C , α C ) , ( Y C , β C ) ) , ϖ C ( X C , Y V ) = ω T M ( ( X C , α C ) , ( Y V , β V ) ) , ϖ C ( X V , Y V ) = ω T M ( ( X V , α V ) , ( Y V , β V ) ) = 0 , (3.11)
    and we are done.
We recall that a Poisson structure P   is called homogeneous if there exists a vector field Z   such that
L Z P + P = 0 . (3.12)
It is well known that the tangent Poisson structure P C   of any Poisson structure P   of M   is homogeneous with Z = E   , where E   is the Euler vector field ( 2.3 ). The generalization of homogeneity to Dirac structures D   is the condition
( X , α ) ∈ D ⇒ ( [ Z , X ] + X , L Z α ) ∈ D , (3.13)
which reduces to ( 3.12 ) in the Poisson case [11] . Now, we get
Proposition 3.3. For an arbitrary Dirac structure D   of M   , the tangent Dirac structure D t g   is homogeneous with Z = E   .
  • Proof. The Euler field E   satisfies ( 2.2 ) and also
    [ E , X C ] = 0 , [ E , X V ] = − X V , X ∈ χ 1 ( M ) , (3.14)
    which follows by easy, local coordinates calculations. Thus, the result is proven if we show that ∀ ( X , α ) , ( Y , β ) ∈ Γ D   the pairs ( X C , L E α C ) , ( 0 , L E α V )   are g T M   -orthogonal to the pairs ( Y C , β C ) , ( Y V , β V )   . The examination of the corresponding scalar products shows that this is the case indeed.
Remark 3.2. A Dirac structure D   of M   yields a pointwisely defined pull back π * ( D )   to T M   by the natural projection π : T M → M   , which is defined by
π * ( D ) = { ( X , π * α ) , / X ∈ T ( T M ) , α ∈ T * M , ( π * X , α ) ∈ D } . (3.15)
From ( 3.15 ), it follows that the equivalent, locally free sheaf of rank 2 n   is spanned by ( X C , α V ) , ( Z , 0 )   where ( X , α ) ∈ D ̲   and Z   is vertical on T M   .
If ( B i , ε i )   is a local basis of D ̲   , ( B i C , ε i V ) , ( ∂ / ∂ v i , 0 )   is a local basis of π * ( D ) ̲   , and we see that π * ( D )   is a differentiable Dirac structure on T M   , which is different from the tangent Dirac structure. In particular, if D = D P   where P   is a Poisson bivector field, π * ( D P )   is not a Poisson structure. We might say that D t g   is the complete lift of D   and π * ( D )   is the vertical lift. The presymplectic leaves of π * ( D )   are the restrictions T S M   of the tangent bundle T M   to the presymplectic leaves S   of D   and the corresponding presymplectic form is the vertical lift of the presymplectic form of S   .

4 Submanifolds of a Dirac manifold

We begin by defining various classes of submanifolds of a Dirac manifold.
More details and motivation on that may be found in [11] . For simplicity, all the submanifolds are assumed to be embedded submanifolds.
Definition 4.1. A submanifold N ↪ M   is properly normalizable if there exists a normal bundle ν N   of N   such that
( X , α ) ∈ D | N ⇒ ( p r T N X , p r T * N α ) ∈ D | N , (4.1)
where the projections are defined by the decomposition T N M = ν N ⊕ T N   .
If ( 4.1 ) holds, the pair ( N , ν N )   is a properly normalized submanifold of M   .
It follows easily that condition ( 4.1 ) is equivalent with
D | N = ( D | N ∩ ( T N ⊕ T * N ) ⊕ D | N ∩ ( ν N ⊕ ν * N ) ) (4.2)
and D | N ∩ ( T N ⊕ T * N )   is a differentiable Dirac structure of N   equal to the pull back of D   by the embedding of N   in M   [11] . Therefore, a properly normalizable submanifold has a well defined induced Dirac structure. If D   comes from a Poisson structure a properly normalizable submanifold is a Poisson-Dirac submanifold which admits a Dirac projection in the sense of [3] . If D   comes from a presymplectic form σ   the submanifold N   is properly normalizable iff there exists a normal bundle ν N   of N   which is σ   -orthogonal to N   .
In [11] we have defined an interesting invariant of a properly normalized submanifold called the second fundamental form of ( N , ν N ) )   in ( M , D )   . This invariant associates with every pair ( X , α ) , ( Y , β ) ∈ D | N ∩ ( T N ⊕ T * N )   a 1   -form B ( ( X , α ) , ( Y , β ) ) ∈ ν * N   with the value on Z ∈ ν N   given by
B ( ( X , α ) , ( Y , β ) ) ( Z ) = Z ( α ~ ( Y ~ ) ) − α ( [ Z ~ , Y ~ ] ) + β ( [ Z ~ , X ~ ] ) , (4.3)
where Z ~ , ( X ~ , α ~ ) , ( Y ~ , β ~ )   extend Z , ( X , α ) , ( Y , β )   from N   to M   and ( X ~ , α ~ ) , ( Y ~ , β ~ )   ∈ Γ D   . The result is independent of the choice of the extensions because the right hand side of ( 4.3 ) is C ∞ ( M )   -linear in all arguments.
In order to define another class of submanifolds we notice the existence of the field of subspaces along N  
H x ( N , D ) = { Z ∈ T x M / ∃ α ∈ a n n T x N & ( Z , α ) ∈ D x } ( x ∈ N ) (4.4)
( a n n   denotes the annihilator space). This field may not be differentiable and the subspaces may have various dimensions and may intersect T x N   . For these reasons we will say that H ( N , D )   is the pseudo-normal field of N   with respect to D   .
Definition 4.2. The submanifold N ↪ M   is a cosymplectic submanifold if the pseudo-normal field H ( N , D )   is a differentiable normal bundle ν N   of N   in M   . H ( N , D )   will be called the natural normal bundle of the cosymplectic submanifold N   .
In [11] , it was proven that, if N   is a cosymplectic submanifold, ( N , ν N = H ( N , D ) )   is a properly normalized submanifold, that N   is cosymplectic iff
D | N ∩ ( T N ⊕ a n n T N ) = { 0 } , (4.5)
that the induced Dirac structure of a cosymplectic submanifold is Poisson and, along N   , one has
D | N = { ( X , i ( X ) σ ) + ( i ( λ ) Π , λ ) / X ∈ H ( N , D ) , λ ∈ T * N } , (4.6)
where Π ∈ χ 2 ( N )   is the bivector field of the induced Poisson structure and σ ∈ Γ ( ∧ 2 a n n T N )   is a 2   -form the graph of which is D | N ∩ ( H ( N , D ) ⊕ H * ( N , D ) )   . If D   comes from a Poisson structure the notion of a cosymplectic submanifold is the known one [13] . If D   comes from a presymplectic form σ   the submanifold N   is cosymplectic iff the pull back of σ   to N   is non degenerate.
Because of ( 4.5 ) we give the following definition:
Definition 4.3. The function d : N → Z   defined by d ( x ) = d i m [ D | N ∩ ( T N ⊕ a n n T N ) ]   ( x ∈ N )   is called the cosymplecticity default of the submanifold N   of ( M , D )   .
It turns out that the second fundamental form of a cosymplectic manifold vanishes [11] . This property is the source of
Definition 4.4. A submanifold N   of a Dirac manifold ( M , D )   is a totally Dirac submanifold if it is properly normalizable by a certain normal bundle ν N   and the second fundamental form of ( N , ν N )   is zero.
One can see [11] that, if D   comes from a Poisson structure, a totally Dirac submanifold is just a Dirac submanifold in the sense of Xu [13] (in [3] these were called Lie-Dirac submanifolds). Indeed, if P   is the Poisson bivector field, the second fundamental form of N   becomes B ( ( i ( α ) P , α ) , ( i ( β ) P , β ) ) ( Z ) = ( L Z ~ P ) ( α , β )   ∀ α , β ∈ T * N   (then i ( α ) P , i ( β ) P ∈ T N   because of the proper normalization property), and, if B = 0   , we are in the case where P   is soldered to ( N , ν N )   [10] . Similarly, if D = D σ   where σ   is a presymplectic form on M   , for the properly normalized submanifold ( N , ν N )   we get B ( ( X , i ( X ) σ ) , ( Y , i ( Y ) σ ) ) ( Z ) = ( L Z ~ σ ) ( X , Y ) ( X , Y ∈ χ 1 ( N ) ) ,   and the vanishing of this form, together with ν N ⊆ T ⊥ σ N   means that σ   is soldered to ( N , ν N )   in the sense of [10] .
Finally, in accordance with the Poisson case, we also define
Definition 4.5. A submanifold N   of a Dirac manifold ( M , D )   is coisotropic if the conditions α ∈ a n n T N   and ( X , α ) ∈ D   imply X ∈ T N   .
Dually, N   is an isotropic submanifold of ( M , D )   if the conditions X ∈ T N   and ( X , α ) ∈ D   imply α ∈ a n n T N   .
In the presymplectic case the coisotropy and isotropy properties are the classical ones (i.e., T ⊥ σ N ⊆ T N   and T ⊥ σ N ⊇ T N   , respectively, where σ   is the presymplectic form). In the Poisson case, N   is coisotropic iff, ∀ x ∈ N   , T x N ∩ T S   is a coisotropic subspace of T S   , respectively, N   is an isotropic submanifold of S   , S   being the symplectic leaf through x   . Obviously, N   is a coisotropic submanifold of ( M , D )   iff its pseudo-normal field satisfies the condition H ( N , D ) ⊆ T N   .
Now, we shall explain how to represent a Dirac structure D   of M   in the neighborhood of a point x 0   of a submanifold N   of M   by means of local bases.
Using a tubular neighborhood of N   with fibers tangent to a chosen normal bundle ν N   , we get local coordinates ( x u , y a )   ( u = 1 , . . . , d i m N   ; a = 1 , . . . , c o d i m N )   around x 0   such that x u   are coordinates along N   and y a   are coordinates along the tubular fibers. Then the local equations of N   are y a = 0   , and
T N = s p a n { ∂ ∂ x u } y = 0 , ν N = s p a n { ∂ ∂ y a } y = 0 , (4.7)
T * N = s p a n { d x u } y = 0 , ν * N = s p a n { d y a } y = 0 . (4.8)
On the coordinate neighborhood obtained above (shrunken if necessary), we may consider a basis of D   that consists of n   independent pairs ( B u , ε u ) , ( C a ,   τ a )   where
B u = b u s ∂ ∂ x s + b u ′ h ∂ ∂ y h , C a = c a s ∂ ∂ x s + c a ′ h ∂ ∂ y h , ε u = e u s d x s + e u h ′ d y h , τ a = t a s d x s + t a h ′ d y h . (4.9)
In these, and in all the formulas that follow, our convention is that any index of coordinates x   takes the same values as the index u   and any index of coordinates y   takes the same values as the index a   .
Of course, these bases must satisfy the conditions implied by the definition of a Dirac structure namely, isotropy:
ε u ( B s ) + ε s ( B u ) = 0 , ε u ( C a ) + τ a ( B u ) = 0 , τ a ( C h ) + τ h ( C a ) = 0 , (4.10)
and integrability:
[ ( B u , ε u ) , ( B s , ε s ) ] ∈ D , [ ( B u , ε u ) , ( C a , τ a ) ] ∈ D , (4.11)
[ ( C a , τ a ) , ( C h , τ h ) ] ∈ D .   We also add that the tangent distribution of the presymplectic foliation of D   is
A ( D ) = s p a n { B u , C a } (4.12)
and the presymplectic form ϖ   is determined by
ϖ ( B u , B s ) = ε u ( B s ) , ϖ ( B u , C a ) = ε u ( C a ) , ϖ ( C a , C h ) = τ a ( C h ) . (4.13)
In what follows we write down the characteristic conditions for the various classes of submanifolds. Definition  4.1 shows that N   is a properly normalizable submanifold of ( M , D )   iff it has a normal bundle ν N   such that there exist local bases ( 4.9 ) of D   which satisfy the conditions
b u ′ h ( x , 0 ) = 0 , e u h ′ ( x , 0 ) = 0 , c a s ( x , 0 ) = 0 , t a s ( x , 0 ) = 0 . (4.14)
We call them adapted bases. Furthermore, N   is cosymplectic in ( M , D )   iff there exist ν N   and bases that satisfy ( 4.14 ) and the supplementary conditions
e u s ( x , 0 ) = δ u s , c a ′ h ( x , 0 ) = δ a h . (4.15)
This is an immediate consequence of formula ( 4.6 ).
Back to a properly normalized submanifold ( N , ν N )   , with ( 4.3 ) we can compute the components of the second fundamental form with respect to adapted bases, and the result is
B ( ( B u , ε u ) , ( B v , ε v ) ) ( ∂ ∂ y a | y = 0 ) = ( b v s ∂ e u s ∂ y a + e v s ∂ b u s ∂ y a ) y = 0 . (4.16)
Therefore, following Definition  4.4 , the characterization of a totally Dirac submanifold consists of conditions ( 4.14 ) and the annulation of the components ( 4.16 ).
Remark 4.1. The skew-symmetry of the second fundamental form B   is an immediate consequence of the first condition ( 4.10 ). The vanishing of B   for a cosymplectic submanifold follows from the first condition ( 4.11 ). Indeed, if N   is a cosymplectic submanifold, the T N   -component of D | N   behaves like a Poisson structure (see ( 4.6 )), and the fact that
[ ( B u , ε u ) , ( B s , ε s ) ] = ( [ B u , B s ] , L B u ε s − L B s ε u + d ( ε u ( B s ) ) ) (4.17)
belongs to D   along N   implies the annulation of the 1   -form component of ( 4.17 ) when calculated on a vector field of the form p a s ∂ ∂ x s + q a h ∂ ∂ y h , p a u ( x , 0 ) = 0 , q a h ( x , 0 ) = δ a h ,   and evaluated at y = 0   . This exactly yields B = 0   .
Finally, Definition  4.5 shows that a submanifold N   is coisotropic if the coefficients of the formulas ( 4.9 ) are such that ∀ ( λ u , λ ′ a )   one has
λ u e u s ( x , 0 ) + λ ′ a t a s ( x , 0 ) = 0 ⇒ λ u b u ′ h ( x , 0 ) + λ ′ a c a ′ h ( x , 0 ) = 0 . (4.18)
The condition for an isotropic submanifold is obtained by reversing the sense of the implication in ( 4.18 ).
Now, we have all the ingredients required to discuss the result of Xu quoted in Introduction in the framework of Dirac manifolds. We will deduce the conditions for a submanifold N   of a Dirac manifold ( M , D )   to have a normal bundle ν N   which is either a coisotropic or an isotropic submanifold of ( T M , D t g )   , and obtain some geometric conclusions of these conditions.
We consider a point x 0 ∈ N   , a normal bundle ν N   of N   in M   , and the local coordinates and bases of formulas ( 4.9 ) around x 0   . Then, if we denote by ( v u , w a )   the corresponding natural coordinates on the fibers of T M   , the submanifold ν N ⊆ T M   has the local equations y a = 0 , v u = 0   , and
T ( ν N ) = s p a n { ∂ ∂ x u , ∂ ∂ w a } y = 0 , v = 0 , (4.19)
a n n ( T ν N ) = s p a n { d y a , d v u } y = 0 , v = 0 . (4.20)
With the bases ( 4.9 ) the tangent Dirac structure D t g   is locally spanned by ( B u C , ε u C ) , ( C a C , τ a C ) , ( B u V , ε u V ) , ( C a V , τ a V )   , where
B u C = l d b u s ∂ ∂ v s + l d b u ′ h ∂ ∂ w h + b u s ∂ ∂ x s + b u ′ h ∂ ∂ y h , C a C = l d c a s ∂ ∂ v s + l d c a ′ h ∂ ∂ w h + c a s ∂ ∂ x s + c a ′ h ∂ ∂ y h , B u V = b u s ∂ ∂ v s + b u ′ h ∂ ∂ w h , C a V = c a s ∂ ∂ v s + c a ′ h ∂ ∂ w h , (4.21)
ε u C = l d e u s d x s + l d e u h ′ d y h + e u s d v s + e u h ′ d w h , τ a C = l d t a s d x s + l d t a h ′ d y h + t a s d v s + t a h ′ d w h , ε u V = e u s d x s + e u h ′ d y h , τ a V = t a s d x s + t a h ′ d y h . (4.22)
By writing down a linear combination of these pairs with coefficients λ u , ν a , μ u , ξ a   we get a local cross section ( Ξ , Ψ )   of D t g   which has the property Ψ ∈ a n n ( T ν N )   iff
( μ u e u s + ξ a t a s ) y = 0 = 0 , ( λ u e u h ′ + ν a t a h ′ ) y = 0 = 0 , ( λ u ∂ e u s ∂ y h + ν a ∂ t a s ∂ y h ) y = 0 = 0 . (4.23)
The same cross section ( Ξ , Ψ )   satisfies the condition Ξ ∈ T ( ν N )   iff
( μ u b u s + ξ a c a s ) y = 0 = 0 , ( λ u b u ′ h + ν a c a ′ h ) y = 0 = 0 , ( λ u ∂ b u s ∂ y h + ν a ∂ c a s ∂ y h ) y = 0 = 0 . (4.24)
Therefore, we have proven
Proposition 4.1. A submanifold N   of a Dirac manifold ( M , D )   has a normal bundle ν N   which is coisotropic in ( T M , D t g )   iff, around the points of N   , D   has bases ( 4.9 ) such that the equations ( 4.24 ) are a consequence of the equations ( 4.23 ). Similarly, ν N   is isotropic in ( T M , D t g )   iff the equations ( 4.23 ) are a consequence of the equations ( 4.24 ).
As consequences of Proposition  4.1 we get
Proposition 4.2. [13]  A submanifold N   of a Poisson manifold M   with the Poisson bivector field P   is totally Dirac iff it has a normal bundle ν N   which is a coisotropic submanifold of ( T M , P C )   .
  • Proof. Using coordinates as in ( 4.9 ) we may write
    P = 1 2 P u s ∂ ∂ x u ∧ ∂ ∂ x s + Q u a ∂ ∂ x u ∧ ∂ ∂ y a + 1 2 S a b ∂ ∂ y a ∧ ∂ ∂ y b . (4.25)
    Accordingly, the bases ( 4.9 ) may be taken under the form B u = P u s ∂ ∂ x s + Q u a ∂ ∂ y a , ε u = d x u ,   C a = − Q u a ∂ ∂ x u + S a b ∂ ∂ y b , τ a = d y a .   Then, the equations ( 4.23 ) become μ u = 0 , ν a = 0   , and the equations ( 4.24 ) become ∑ u μ u P u s − ∑ h ξ h Q s h = 0 , ∑ u λ u Q u a + ∑ h ν h S h a = 0 ,   ∑ u λ u ∂ P u s ∂ y h − ∑ a ν a ∂ Q s a ∂ y h = 0 ,   for y = 0   . Obviously, the first system implies the second iff Q u a = 0 , ∂ P u s / ∂ y h   = 0   for y = 0   . These exactly are the conditions for N   to be a (totally) Dirac submanifold [13] .
Similarly, we have
Proposition 4.3. A submanifold N   of a presymplectic manifold M   with the closed 2   -form σ   is totally Dirac iff it has a normal bundle ν N   which is an isotropic submanifold of ( T M , σ C )   .
  • Proof. With the same notation, we have σ = 1 2 σ u s d x u ∧ d x s + φ u a d x u ∧ d y a + 1 2 θ a h d y a ∧ d y h ,   and the bases B u = ∂ ∂ x u , C a = ∂ ∂ y a , ε u = σ u s d x s + φ u a d y a , τ a = − φ u a d x u + θ a h d y h .   Then, the system ( 4.23 ) becomes μ u σ u s − ξ a φ s a = 0 , λ u φ u h + ν a θ a h = 0 , λ u ∂ σ u s ∂ y h − ν a ∂ φ s a ∂ y h = 0   for y = 0   , and the system ( 4.24 ) becomes μ u = 0 , ν a = 0   . The latter conditions imply the former iff φ u a = 0   and ∂ σ u s / ∂ y h = 0   for y = 0   .
    These are the conditions that characterize a totally Dirac submanifold of a presymplectic manifold.
Proposition 4.4. If N   is a cosymplectic submanifold of the Dirac manifold ( M , D )   , the cosymplecticity default of the natural normal bundle H ( N , D )   seen as a submanifold of ( T M , D t g )   satisfies the inequalities c o d i m N ≤ d ( x ) ≤ d i m M   ( x ∈ N )   .
  • Proof. The local cross sections of D t g ∩ [ T ( H ( N , D ) ) ⊕ a n n ( T ( H ( N ,   D ) ) ) ]   must satisfy both ( 4.23 ) and ( 4.24 ), which, modulo ( 4.14 ) and ( 4.15 ), include the conditions μ u = 0 , ν a = 0   and do not restrict the coefficients ξ a   . Therefore, D t g ∩ [ T ( H ( N , D ) ) ⊕ a n n ( T ( H ( N , D ) ) ) ]   has a basis which consists of the pairs ( C a V , τ a V )   and of linear combinations of ( B u C , ε u C )   .
Remark 4.2. The questions discussed above may also be considered for the vertical lift π * ( D )   defined in Remark  3.2 , instead of the tangent structure D t g   . The local bases of π * ( D )   are ( B u C , ε u V ) , ( C a C , τ a V ) , ( ∂ / ∂ v u , 0 ) ( ∂ / ∂ w a , 0 )   .
The conditions ( 4.23 ) are to be replaced by
λ u e u s ( x , 0 ) + ν a t a s ( x , 0 ) = 0 . (4.26)
The conditions ( 4.24 ) are to be replaced by
λ u b u ′ h ( x , 0 ) + ν a c a ′ h ( x , 0 ) = 0 , μ u = 0 , λ u ∂ b u s ∂ y h | y = 0 + ν a ∂ c a s ∂ y h | y = 0 = 0 . (4.27)
From these formulas, we see that ν N   is never coisotropic in ( T M , π * ( D ) )   .
References

  1. T. J. Courant, Dirac Manifolds, Transactions Amer. Math. Soc., 319 (1990), 631-661.
  2. T. J. Courant, Tangent Dirac structures, J. Phys. A, 23 (1990), no. 22, 5153 5168.
  3. M. Crainic and R. L. Fernandez, Integrability of Poisson brackets, J. Differential Geom., 66 (2004), 71-137.
  4. V. Cruceanu, P. Fortuny and P. M. Gadea, A survey on paracomplex geometry, Rocky Mountain J. of Math., 26 (1996), 1-33.
  5. M. de Leon, Math. Reviews 1085863 (92d:58064).
  6. J. Grabowski and P. Urbański, Tangent and cotangent lifts and graded Lie algebras associated with Lie algebroids, Ann. Global Analysis Geom., 15 (1997), 447-486.
  7. M. Gualtieri, Generalized complex geometry, Oxford University DPhil thesis, 107 pages, arXiv:math.DG/0401221.
  8. Tennison, B. R., Sheaf theory, London Math. Soc. Lecture Notes Series 20, Cambridge Univ. Press, Cambridge, 1975.
  9. I. Vaisman, Lectures on the geometry of Poisson manifolds, Progress in Math., vol. 118, Birkhäuser, Basel, 1994.
  10. I. Vaisman, Dirac submanifolds of Jacobi manifolds. In: The Breadth of Symplectic and Poisson Geometry, Festschrift in Honor of Alan Weinstein (J. E. Marsden and T. Ratiu, eds.), Progress in Math., vol. 232, p. 603-622, Birkhäuser, Boston, 2005.
  11. I. Vaisman, Foliation coupling Dirac structures, arXiv:math.SG/0412318.
  12. A. Weinstein, Lagrangian mechanics and grupoids, Mechanics day (Waterloo ON, 1992), Fields Inst. Commun., 7, AMS, Providence, RI, 1996, 207-231.
  13. P. Xu, Dirac submanifolds and Poisson involutions, Ann. Sci. Ec. Norm. Sup., 36 (2003), 403-430.
  14. K. Yano and S. Ishihara, Tangent and Cotangent Bundles, M. Dekker, Inc., New York, 1973.

Department of Mathematics
University of Haifa, Israel
E-mail: vaisman@math.haifa.ac.il

*   2000 Mathematics Subject Classification: 53D17.

Key words and phrases: Complete Lift; Vertical Lift; Dirac Structure;Isotropic Submanifolds; Coisotropic Submanifolds.