Medical, Pharma, Engineering, Science, Technology and Business

Institut Élie Cartan de Nancy, Universite Nancy, CNRS, INRIA, Boulevard des Aiguillettes, B.P. 239, F-54506 Vanduvre-les-Nancy, France

- *Corresponding Author:
- Institut Élie Cartan de Nancy

UniversitÉe Nancy, CNRS, INRIA

Boulevard des Aiguillettes

B.P. 239, F-54506 Vandeauvre-lÉes-Nancy, France

**Received Date**: January 9, 2009; **Revised Date:** May 07, 2009

**Visit for more related articles at** Journal of Generalized Lie Theory and Applications

We dene symmetric bundles as vector bundles in the category of symmetric spaces; it is shown that this notion is the geometric analog of the one of a representation of a Lie triple system. A symmetric bundle has an underlying re ection space, and we investigate the corresponding forgetful functor both from the point of view of dierential geometry and from the point of view of representation theory. This functor is not injective, as is seen by constructing \unusual" symmetric bundle structures on the tangent bundles of certain symmetric spaces.

Although this is not common, linear representations of Lie groups may be defined as vector bun- dles in the category of Lie groups: if is a (say, finite-dimensional) representation of a Lie group in the usual sense, then the semidirect product of G and V is a Lie group and at the same time a vector bundle over G such that both structures are compatible in the following sense:

(R1) the projection is a Lie group homomorphism,

(R2) the group law is a morphism of vector bundles, i.e., it preserves fibers and, fiberwise, group multiplication is linear.

Conversely, given a vector bundle F over G with total space a Lie group and having such
properties, the representation of G can be recovered as the fiber F_{e} over the unit element e on
which G acts by conjugation. For instance, the tangent bundle TG corresponds to the *adjoint
representation*, and the cotangent bundle T*G to the coadjoint representation of G.

In this work we wish to promote the idea that this way of viewing representations is the good
point of view when looking for a notion of "representation" for other categories of spaces which,
like Lie groups, are defined by one or several "multiplication maps": somewhat simplified, a* rep-* *resentation of a given object M of such a category is a vector bundle over M in the given category,* where "vector bundle in the given category" essentially means that the analogs of (R1) and (R2)
hold. In fact, this simple notion came out as a result of our attempts to find a "global" or "geometric"
analog of the notion of *representation of general n-ary algebraic structures*: given a multilinear
algebraic structure defined by identities, such as Lie, Jordan or other algebras or triple systems,
Eilenberg [8] introduced a natural notion of *(general) representation* (which is widely used in Jordan theory, see [12,17,18]).1 Essentially, a representation V of such an n-ary algebra m is equivalent to defining on the direct summ an n-ary algebraic structure satisfying the same defining identities as m and such that some natural properties hold, which turn out to be exactly the "infinitesimal analogs" of (R1) and (R2): for instance, V will be an "abelian" ideal
in m corresponding to the role of the fiber in a vector bundle. The archetypical example
is given by the *adjoint representation* which is simply with ε^{2} = 0, the scalar extension of m by dual numbers, which will of course correspond to the tangent bundle in the geometric
picture. However, nothing guarantees in principle that there be a "coadjoint representation" and
a "cotangent bundle in the given category"!

This approach is very general and has a wide range of possible applications: at least locally,
any ane connection on a manifold gives rise to a smooth "multiplication map" (a *local loop*, see
[20]), which by deriving gives rise to n-ary algebras, and hence may be "represented" by vector
bundles. Concretely, we will show how all these ideas work for the most proeminent example of
such structures, namely for *symmetric spaces* (here, the approach to symmetric spaces by Loos
[16] turns out to be best suited; we recall some basic facts and the relation with homogeneous
spaces in Chapter 2). *Symmetric bundles* are defined as vector bundles in the category of *symmetric spaces* (Section 2.3) and their infinitesimal analogs, *representations of Lie triple* *systems* are introduced (Chapter 3). These have already been studied from a purely algebraic
point of view by Hodge and Parshall [11]. Another algebraic point of view (see [10]) features the
aspect of *representations of Lie algebras with involution* (cf. Section 4.1), which we use to prove
that, in the real finite-dimensional case, such representations are in one-to-one correspondence
with symmetric bundles (Theorem 4.2). This result implies that the cotangent bundle TM
of a real finite-dimensional symmetric space is again a symmetric bundle { this is much less
obvious than the corresponding fact for the tangent bundle, and it owes its validity to the fact
that, on the level of *representations of Lie triples systems*, every representation admits a dual
representation (Section 5.2). Whereas, among the algebraic constructions of new vector bundles
from old ones, the *dual* and the *direct sum constructions* survive in the category of symmetric
bundles, this is not the case for tensor *products and hom-bundles*: they have to replaced by other,
more complicated, constructions (Chapter 5).

Compared to the case of group representations, a new feature of symmetric bundles is that
they are "composed objects": for a Lie group, the group structure on and its structure of
a *homogeneous vector bundle* are entirely equivalent. For symmetric bundles, the structure of a *homogeneous vector bundle* carries strictly less information than that of the symmetric bundle:
let us assume F is a symmetric bundle over a homogeneous symmetric space M = G/H; then
F carries two structures: it is a homogeneous symmetric space F = L/K, and, under the action
of the smaller group G, it is a *homogeneous vector bundle* , with V = F_{o} being the fiber
over the base point o = eH. Basically, seeing F as a *homogeneous vector bundle* only retains the
representation of H on F_{o}, whereas seeing F as symmetric space L/K takes into account the
whole isotropy representation of the bigger group K. In other words, there is a forgetful functor *from symmetric bundles to homogeneous vector bundles*. Conversely, the following "extension
problem" arises: given a *homogeneous symmetric* space M = G/H, which *homogeneous vector* bundles (i.e., which H-representations) *admit a compatible structure of a symmetric bundle?* On the infinitesimal level of general representations of Lie triple systems, the forgetful functor
appears as follows: a general representation of a Lie triple system m consists of two trilinear maps
(r;m), and we simply forget the second component m (Section 3.4). The extension problem is
then: *when does r admit a compatible trilinear map m such that (r;m) is a representation of Lie* triple systems? For a geometric interpretation of this problem, one notes that the trilinear map
r is of the type of a *curvature tensor*, and indeed one can prove that every symmetric bundle
admits a canonical connection (Theorem 6.1) such that r becomes its curvature tensor (Theorem
6.2). It seems thus that the representations of H that admit an extension to a symmetric bundle
are those that can themselves be interpreted as holonomy representation of a connection on a
vector bundle. However, the situation is complicated by the fact that the canonical connection
on F does not determine completely the symmetric bundle structure on F.

We do not attack in this work the problem of classifying representations of, say, finite dimensional simple symmetric spaces; but we give a large class of examples of "unusual" symmetric bundle structures on tangent bundles (Chapter 7), thus showing that the above mentioned forgetful functor is not injective. In fact, as observed in [3], many (but not all) symmetric spaces admit, besides their "usual" complexification another, "twisted" or "hermitian" one We show here that a similar construction works when one replaces "complexification" by "scalar extension by dual numbers" (replace the condition by and that in this way we obtain two difierent symmetric structures on the tangent bundle TM. A particularly pleasant example is the case of the general linear group in this case, the usual tangent bundle TM is the group (scalar extension by dual numbers), whereas the "unusual" symmetric structure on the tangent bundle is obtained by realizing TM as the homogeneous space where is some degenerate version of the quaternions (Theorem 7.1). We conjecture that, for real simple symmetric spaces, there are no other symmetric bundle structures on the tangent space than the ones just mentioned. In other words, we conjecture that the extension problem as formulated here is closely related to the "extension problem for the Jordan-Lie functor" from [3]; however, this remains a topic for future research.

The results presented in this paper partially extend results from the thesis [6], where a slightly dierent axiomatic definition of symmetric bundles was proposed in a purely algebraic setting, permitting to state the analog of Theorem 4.2 (equivalence of symmetric bundles and representations of Lie triple systems) in an algebraic framework (arbitrary dimension and arbitrary base field; see Theorem 2.1.2 in loc. cit.), based on results published in [7]. The present paper is independent from the results of [7], but nevertheless the framework still is quite general: our symmetric spaces are of arbitrary dimension and defined over very general topological base fields or rings K { for instance, the setting includes real or complex infinite dimensional (say, Banach) symmetric spaces or p-adic symmetric spaces (Section 2.1). We hope the reader will agree that, in the present case, this degree of generality does not complicate the theory, but rather simplifies it by forcing one to search for the very basic concepts.

After this work had been finished, we learned from Michael Kinyon that the question of defining "modules" for an object in a category had already been investigated by Beck in his thesis (see [2]; see also [1]): he defines a module to be an abelian group object in the slice category over the given object. It seems reasonable to conjecture that, in the cases considered here, this notion should agree with ours, but by lack of competence in category theory we have not been able to check this.

**Notation and general framework**

This work can be read on two dierent levels: the reader may tak to be the real base field and understand by "manifold" finite-dimensional real manifolds in the usual sense; then our symmetric spaces and Lie groups are the same as in [16] or [13], or one may consider a commutative topological field or ring having dense unit group and such that 2 is invertible in ;then we refer to [5] for the definition of manifolds and Lie groups over . Readers interested in the general case should just keep in mind that, in general,

symmetric spaces need no longer be homogeneous (cf. Example 2.2 below),

there is no exponential map and hence no general tool to "integrate" infinitesimal structures to local ones.

If we use such tools, it will be specifically mentioned that we are in the real (or complex)
finite-dimensional case. In the sequel, the word *linear space* means "(topological) -module"

**Symmetric spaces and re
ection spaces**

A re ection space ("Spiegelungsraum", introduced by Loos in [15]) is a smooth manifold M together with a smooth "product map" satisfying, for all

The reflection space is called a *symmetric space* if in addition

(S4) for all x ε M, the differential of the "symmetry" σ_{x} at x is the negative of the
identity of the tangent space .

In the real finite-dimensional case this is (via the implicit function theorem) equivalent to

(S4') for all x ε M, the fixed point x of σ_{x} is isolated.

Homomorphisms of such structures are smooth maps which commute with product maps. According
to (S3), all maps of the form are automorphisms; the subgroup G(M) of Aut(M) generated by these elements is called the *transvection group* of M. Often one considers
the category of re
ection spaces, respectively, symmetric spaces *with base point: a base point* is just a distinguished point, often denoted by x_{0} or o, and homomorphisms are then required
to preserve base points. If o ε M is a base point, one defines the *quadratic map* by

and the powers by

**Example 2.1**. The group case. Every Lie group with the new multiplication** ** is a symmetric space.

**Example 2.2. **Homogeneous symmetric spaces. We say that a symmetric space is homogeneous
if the group G := G(M) acts transitively on it and carries a Lie group structure such that this
action is smooth. Let o be a base point and H its stabilizer, so that Then the map is an involution of G, and the multiplication map on G/H is given
by

In finite dimension over every connected symmetric space is of this form, for a suitable involution σ of a Lie group G (see [15,16]).

**Example 2.3. **Linear symmetric spaces. Assume V is a *linear space* over ; we consider as a linear space and thus write Assume that V carries a symmetric space structure which is a -linear map. Because of (S4), the symmetry being a linear map, must agree with its tangent map -idV . Then it follows that

Conversely, every linear space equipped with the multiplication map is a symmetric space (group case G = V ). With respect to the zero vector as base point, is translation by 2x, and the powers are

**Example 2.4. **Polynomial symmetric spaces. In the same way as in the preceding example,
we can consider linear spaces together with a symmetric structure which is a *polynomial* map See [6] for a theory of such spaces.

**Symmetric bundles**

A *symmetric bundle* (or, longer but more precise, *symmetric vector bundle*) is a vector bundle

and are symmetric spaces such that is a homomorphism of symmetric spaces,

(SB2) for all the map induced by fiberwise,

(which is well defined according to (SB1)), is linear.

*Homomorphisms of symmetric* bundles are vector bundle homomorphisms that are also homomorphisms
of the symmetric spaces in question. Clearly, the concept of symmetric bundle could
be adapted to other classes of bundles whenever the fibers belong to a category that admits
direct products (e.g., *multilinear bundles* in the sense of [5]): it suffices to replace (SB2) by the
requirement that the map be a morphism in that category. Also, it is clear
that such concepts exist for any category of manifolds equipped with binary, ternary or other
"multiplication maps", such as g*eneralized projective geometries* (cf. [4]). However, in the sequel,
we will stick to the case of vector bundles and symmetric spaces.

A symmetric bundle is called trivial if it is trivial as a bundle, and if, as a symmetric space, it is simply the direct product of M with a vector space. The first nontrivial example of a symmetric bundle is the tangent bundle F := TM of a symmetric space : as to (SB1), it is well known that TM with product map is a symmetric space such that the canonical projection is a homomorphism and the fibers are at subspaces (see [16] for the real finite-dimensional case and [5] for the general case). Property (SB2) follows immediately from the linearity of the tangent map

**Some elementary properties of symmetric bundles**

For a symmetric bundle F over M, the following holds:

(SB3) The symmetric space structure on the fiber over x ε M coincides with the canonical symmetric space structure of the vector space

(SB4) The zero-section is a homomorphism of symmetric spaces. (Hence in the sequel we may identifyM with z(M), and the use of the same letter μ for the multiplication maps of M and F does not lead to confusion.)

(SB5) For all the fiberwise dilation map

is an endomorphism of the symmetric space F; for it is an automorphism.

In fact, for p = q, (SB2) says that the fiber F_{p} is a symmetric subspace of F such that its
structure map is linear, and (SB3) now follows in view of Example 2.3. Since a linear map sends zero vector to zero vector, we have proving (SB4), and to prove (SB5), just note that for

In particular, note that (0)_{F} is the projection onto the zero section, and that (-1)F can be seen
as a "horizontal re
ection with respect to the zero section".

**Horizontal and vertical symmetries**

Let a symmetric bundle and We define the horizontal (resp., vertical) symmetry (with respect to u) by

For the maps commute with each other because of (SB5):

Therefore v_{u} then is also is of order 2. Conjugating by we see that for all we get three pairwise commuting automorphisms of order 2 and fixing the point

**Lemma 2.5. **The vertical symmetry depends only on the fiber F_{p}, that is, for all we
have

**Proof.** Let us show that with i.e.,

But this follows from and the fact that is an automorphism.

**Proposition 2.6. **The space F together with the binary map is
a reflection space.

**Proof.** The defining properties (S1) and (S2) say that v_{v} is of order 2 and fixes v, and this has
already been proved above. In order to establish (S3), let and Then, using the preceding lemma,

We say that (F;μ) is the reflection space associated to the symmetric bundle (F; μ). Thus from symmetric vector bundles to re
ection spaces; it will be a recurrent theme in this work to interpret this functor as a forgetful functor. Note that the
dierential of v_{u} has 1-eigenspace tangent to the fiber through u and -1-eigenspace complementary
to it; thus the distribution of the "vertical" 1-eigenspaces is integrable, whereas the
distribution of the "horizontal" -1-eigenspaces is in general not (see Chapter 6: the curvature
of the corresponding Ehresmann connection does in general not vanish).

**Automorphisms downstairs and upstairs**

The canonical projection does in general not admit a cross-section; we cannot even guarantee that it is surjective. However, it is easily seen that the projection of transvection groups is surjective: write as a composition of symmetries at points of M; identifying M with the zero section in F we see that g gives rise to an element with In particular, if M is homogeneous, then so is F: in fact, if then there exists with g:o = x; then In the real finite-dimensional case, we may replace G(M) by its universal covering; then the zero section z : M ! F induces a homomorphism of this universal covering into G(F), having discrete kernel. Hence, if we write F = L/K, it is not misleading to think of G as a subgroup of L and of H as a subgroup of K (possibly up to a discrete subgroup).

**Homogeneous bundles over symmetric spaces**

Assume that M = G/H is a homogeneous symmetric space (Example 2.2). To any smooth action on a manifold U one can associate the homogeneous bundle

When the base M = G/H is a symmetric space, we define

and one can show that becomes a reflection space such that the projection onto M
becomes a homomorphism of reflection spaces (cf. [15, Theorem 1.5]). Let us say that then F is a reflection space over the symmetric space M. The preceding formula shows that the
reflection v_{v} does not depend on the choice of v ε U, i.e., it depends only on the base. Loos
has shown [15] that, conversely, every real finite-dimensional and connected reflection space can
be written in this way as a homogeneous bundle over a symmetric space. In particular, linear
representations of H and reflection spaces over M with linear fibers ("reflection vector bundles
over M") correspond to each other.

Having this in mind, we now consider a symmetric bundle over a homogeneous
symmetric space M = G/H. As we have just seen, F is then also homogeneous, say, F = L/K. Looking at H as a subgroup of K (see Section 2.6), we get a linear representation of H on the fiber F_{o}, and we can write as a homogeneous bundle over the base M. In this way, the functor from symmetric bundles to reflection spaces corresponds in the homogeneous case to the
functor from symmetric bundles F = L/K overM = G/H to the associated homogeneous bundle Conversely, we can formulate an *extension problem: For which representations* *does the homogeneous bundle* *carry a symmetric bundle structure? If
it does, how many such structures are there?*

**Derivations of symmetric bundles and vertical automorphisms**

A derivation of a symmetric bundle F is a homomorphism of symmetric spaces which at the same time is a smooth section of π (see [16] for this terminology in case of the tangent bundle). A *vertical automorphism* of a symmetric bundle is an automorphism f of the symmetric bundle F preserving fibers, i.e., then is a
derivation of F. Conversely, if X is a derivation, define

Then f is a vertical automorphism: it is clearly smooth, preserves fibers, and is bijective. It is an automorphism: using (SB2),

Summing up, vertical automorphisms are the same as derivations. Moreover, they clearly form a normal subgroup Vaut(F) in the group Aut(F), where composition corresponds to addition of sections. It follows that the space of derivations is stable under addition; it is also stable under multiplication by scalars, hence forms a vector group. The same kind of arguments shows that in fact we have an exact sequence

(which essentially splits if we take transvection groups; cf. the discussion in Section 2.6). Now
fix a base point o ε M; then the involution given by conjugation with σ_{0o} restricts to Vaut(F)
and thus defines a linear map. Let us write

for the corresponding eigenspace decomposition.

**Lemma 2.7. **We have and the map

*is a bijection with inverse*

Proof. See [16] or [5, Proposition 5.9] for the proof in the case of the tangent bundle; the same arguments apply here.

**Definition 3.1**. *A Lie triple system* (Lts) is a linear space m over together with a trilinear map such that, writing also R(X; Y ) for the endomorphism

(skew-symmetry),

(the Jacobi identity),

is a derivation of the trilinear product on m, i.e.,

For instance, if is a Lie algebra with involution, then the -1-eigenspace m of σ with is an Lts. Every Lts arises in this way (see Section 4.1 below). For later use we introduce also the "middle multiplication operators" then, in presence of (LT1), property (LT2) can be written in operator form

and similarly, reading (LT3) as an identity of operators, applied to the variable W; V or Y , we get the following equivalent conditions, respectively:

**The Lie triple system of a symmetric space**

Let (M;μ) be a symmetric space with base point o. Consider the tangent bundle TM and
write for the derivations of the symmetric bundle TM, and for the
eigenspace decomposition from Lemma 2.7. One shows that g, seen as a space of vector fields
on M, is stable under the Lie bracket and that σ_{0} induces an involution of this Lie algebra
structure. Hence the -1-eigenspace m is an Lts. Via the bijection from
Lemma 2.7 this Lts structure can be transferred to the tangent space T_{o}M, which, by definition,
is the *Lts associated to the pointed symmetric space* (M,o) (cf. [16] or [5, Chap 5]). The Lts
depends functorially on M and plays a similar role for symmetric spaces as the Lie algebra for a
Lie group. (In particular, in the real finite-dimensional case there is an equivalence of categories
between Lts and connected simply connected spaces with base point, cf. [16].)

**The Lts of a symmetric bundle**

Now assume that is a symmetric bundle over M and fix a base point o ε M and let Since F is a symmetric space, f is a Lie triple system. We wish to describe its structure in more detail. The dierentials of the three involutions and from Section 2.5 act by automorphisms on the Lts f (where we write 0 instead of 0_{o}). The +1-eigenspace of # is the tangent space of the fiber which we identify with V (in the notation from [5] we could also write "εV for this "vertical space").

**Lemma 3.2.*** The decomposition* *has the following properties*:

(1) V* is an ideal of* as* soon as one of the x; y; z belongs to V* ,

**Proof. **(1) is clear since V is the kernel of a homomorphism (the differential of at
0); (2) follows from the fact that m is the fixed point space of the horizontal automorphism (moreover, we see that m is isomorphically mapped by onto the Lts of M), and (3) holds since is an automorphism and hence

hence belongs to m and thus to since V is an ideal. Finally, the fiber F_{o} carries the "flat" symmetric space structure of a linear space and hence [V; V; V ] = 0.

A side-remark: if we adapt the whole set-up to the case of *bilinear bundles* (in the sense of
[5]) instead of vector bundles, we get essentially the same properties, the only dierence being
that (3) does no longer hold: e.g., for the bilinear bundle TTM over M, the tangent model is where the term in brackets is still an abelian Lts, but (3) does no
longer hold: in fact, is nonzero in general.

**Representations and modules**

Let m be an Lts. An m-module is a vector space V such that the direct sum carries the structure of an Lts satisfying the properties from the preceding lemma. More explicitly, this means, by decomposing

that we are given two trilinear maps r and m:

satisfying the properties given by the following lemma.

**Lemma 3.3. **For any Lts the space *with a triple bracket given by (3.1) is an L*ts *if and only if r and m satisfy the following relations*:

**Proof.** We have to show that (LT1){(LT3) for are equivalent to (R1)-{(R4): first of all, we note that a bracket is zero if more than one of the three arguments belongs to V . Now, (LT1)
is equivalent to (R1) if both arguments belong to m and holds by (3.1) if one is in V and the
other in m. Next, (LT2) is an identity in three variables. We may assume that two variables, say
X and Z, belong to m, and write (LT2) in its operator form (LT2a). Thus we see that (LT2)
is equivalent to (R2). Finally, (LT3) is an identity in 5 variables. In order to get a nontrivial
identity, we can assume that at least four of them belong to m. We then write (LT3) in operator
form (identities (LT3a,b,c) from Definition 3.1), and see that (LT3) is equivalent to (R3) and
(R4), thus proving our claim.

Note that, in view of (R3), identity (R4) is equivalent to the following identity:

Condition (R4) can be rephrased by saying that the operator defined by belongs to the space of *derivations from m into* V ,

**Definition 3.4. **A *general representation of a Lie triple system *m in a unital associative algebra
A is given by two bilinear maps

such that (R1)-{(R4) hold (where o has to be interpreted as the product in A and the bracket is
the Lie bracket in A). If A = End(V ) is the endomorphism algebra of a vector space, we say that
V is an *m-module. Homomorphisms of m-modules* are defined in the obvious way, thus turning
m-modules into a category. Given an m-module V , the Lts with bracket defined by
(3.1) is called the *split null extension* of m by the module V . It is fairly obvious that the split
null extension depends functorially on the m-module V .

**Example 3.5** (regular representation). For any Lts m, consider its "extension by dual numbers", i.e., let (ring of dual numbers over ), and

with the ε-trilinear extension of the bracket from m:

This is nothing but the split null extension of m by the regular representation, which by definition is given by V = m and

If M is a symmetric space, then the Lts of the tangent bundle TM is precisely (cf. [5]).Hence the regular representation corresponds to the tangent bundle of M.

**The extension problem revisited**

The forgetful functor associating to a symmetric vector bundle its underlying re ection space corresponds to the forgetful functor . Namely, if h is the image of the skew-symmetric map

then (LT3) implies that is a Lie algebra, and for any representationwe may define then the first relation of (R3) is equivalent to *p* being a
representation. Thus we get the infinitesimal version of a homogeneous vector bundle. Now the
problem of finding a compatible symmetric vector bundle structure corresponds to finding the
second component m such that (r;m) defines a representation of m.

We have shown that a representation of an Lts is the derived version of a symmetric bundle. Conversely, can one reconstruct a symmetric bundle from a representation of an Lts? As a first step, it is always possible to recover Lie algebras from Lie triple systems, and certain Lie algebra representations from Lie triple representations. The second step is then to lift these constructions to the space level: here we have to make assumptions on the base field and on the topological nature of M.

**From Lie triple systems to Lie algebras with involution**

Lie triple systems and graded Lie algebras: the standard imbedding. Every Lie algebra together with an involution σ gives rise to an Lts equipped with the triple Lie bracket [[X; Y ];Z]. Conversely, every Lts m can be obtained in this way: let h be the subalgebra of the algebra of derivations of m generated by the endomorphisms R(x; y), x; y ε m. Then the space carries a Lie bracket given by

This Lie algebra, called the standard imbedding of the Lts m, does in general not depend functorially on m; see [11,21] for a detailed study of functorial properties related to this and other constructions. Note that, in terms of the Lie algebra g, we can write

-modules with involution. Assume is a Lie algebra with involution. A representation is called amodule with involution if W is equipped with a direct sum decomposition which is compatible with σ in the sense that

commutes, where is the identity on W^{+} and -1 on W^{-}, and

**Lemma 4.1. **Let m be an Lts and g its standard imbedding. There exists a bijection between
--modules with involution and m-modules.

**Proof. **Given a -module with involution (W; T ), we first form the semidirect product This is a Lie algebra carrying an involution given by Its -1-eigenspace is an
Lts satisfying the relations from Lemma 3.2, and hence is the split null extension corresponding
to an m-module W^{-}.

Given an m-module V , we construct first the split null extension and then its standard imbedding Then

is a g-module with involution.

Again, the correspondence set up by the lemma is functorial in one direction but not in the other; see [11] for this issue.

**From modules to bundles**

**Theorem 4.2**. Let and M be a finite-dimensional connected simply connected symmetric
space with base point o. Let be its associated Lts and g its standard imbedding, with involution
σ. Then the following objects are in one-to-one correspondence:

(1) (finite-dimensional) symmetric vector bundles over M,

(2) (finite-dimensional) -modules with involution,

(3) (finite-dimensional) -modules.

The bijection between (1) and (3) is an equivalence of categories.

**Proof. **We have already seen how to go from (1) to (3), and that (2) and (3) are in bijection.
Let us give a construction from (2) and (3) to (1): let be the split null extension coming from an -module V and the standard imbedding of and let W be
the corresponding g-module with involution. Let G be the simply connected covering of the transvection group G(M) and write M = G/H. Then the representation of g on W integrates
to a representation of G on W. Let

We claim that

(i) F is a vector bundle over M, isomorphic to the homogeneous bundle

(ii) F carries the structure of a symmetric bundle over M.

Proof of (i): first of all,

is a well-defined bijection. Since ,this proves the first claim.

Proof of (ii): the Lie algebra is the fixed ed point space of an involution of b, and hence is a symmetric space. Its Lts is . The projection map has as dierential the projection from to m and hence is a homomorphism of symmetric spaces.

Let us show that the structure map is linear. Since we already know that F is a homogeneous G-bundle, we may assume that p = o is the base point. Now we proceed in two steps:

(a) we show that is linear. In fact, here we use that W is a G-module with involution, i.e.,

Thus this map is described by and thus is linear.

(b) Since is a linear bijection, is linear (and well defined) if and only if so is the map But the last map is the same as (the point stands for the action of W^{-} on F; recall that in every symmetric space Summing up, it suffices now to show that the map

is well defined and linear. Proof of this: let and Then

Thus our map is described by with a linear map that depends on g, and hence is linear, proving claim (ii).

Finally, the fact that homomorphisms in the categories defined by (1) and (3) correspond to each other follows from the corresponding fact for (connected simply connected) symmetric spaces and Lie triple systems, see [16].

So far we do not know any representations other than the regular one and the trivial ones. In the following we discuss the standard linear algebra constructions producing new representations from old ones.

**Direct sums**

Clearly, if and are m-modules, then is again a general representation. Correspondingly, the direct sum of symmetric bundles can be turned into a symmetric bundle.

**The dual representation**

If (V; r;m) is an m-module, then the dual space V* can be turned into an m-module by putting

(5.1)

where is the dual operator of an operator In fact, the properties (R1)-{(R3) for are easily verified; for (R4) note that (R4') written out for the dual is precisely (R4).

Equivalently: if is a)-module with involution, one verifies that the dual module, also is a module with involution It follows that

leading to Formula (5.1).

In particular, in the finite-dimensional real case, invoking Theorem 3.4, the dual of the regular representation corresponds to the cotangent bundle which thus again carries a symmetric bundle structure. (If M = G/H, then we may also write where It remains intriguing that there seems to be no really intrinsic construction of this symmetric space structure on For this reason we cannot affirm that (in cases where a reasonable topological dual m* of m exists), in the infinite-dimensional case or over other base fields than is again a symmetric space.

**A duality principle**

Note that, for finite-dimensional modules over a field, V is the dual of its dual module V*. More generally, we can define for any general representation of q in an algebra A its opposite representation in the algebra Aopp by putting

as above it is seen that this is again a representation. As an application of these remarks we get a duality principle similar as the o ne for Jordan pairs formulated by Loos (cf. [18]):

**Proposition 5.1**. If I is an identity in R(X; Y ) and M(U; V ) valid for all Lie triples over then its dual identity I*, obtained by replacing R(X; Y ) by R(Y;X) and M(U; V ) by M(V;U)
and reversing the order of all factors, is also valid for all Lie triples over

**Proof. **If I is valid for all Lts, then it is also valid for all split null extensions obtained from
representations and hence the corresponding identity, with R(X; Y ) replaced by r(X; Y ) and
M(X; Y ) by m(X; Y ), is valid for all representations. Since the set of all representations is the
same as the set of all opposite representations, and since the opposite functor changes order of
factors and order of arguments, we see that I* is valid for all representations. In particular, it is
valid for the regular representation and hence holds in q.

For instance, identities (R4) and (R4') (cf. Lemma 3.3 and remark following it) are dual in the sense of the proposition. We don't know about any application of Proposition 5.1; however, one may note that the original definition of Lie triple systems by Jacobson [12] as well as the exposition by Lister [14] are based on a set of five identities, among which two identities are equivalent to each other by the duality principle; they correspond to (LT3c) and its dual identity.

**Tensor products**

The tensor product of two symmetric vector bundles is in general no longer a symmetric vector bundle: let be the two fibers in question, regarded as m-modules. Extends A and B to modules with involution, It is easily verified that then is again a module with involution. Now, the minus-part in

is

which therefore is another m-module, replacing the ordinary tensor product

and there is a similar expression for the m-components and for It is obvious that the operation fi is compatible with direct sums, and it also associative (in the same sense as the usual tensor product): the minus-part both in and in is

where and with etc.

**Hom-bundles**

If A and B are m-modules, then Hom(A;B) is in general not an m-module, but we can use the same construction as in Section 5.4 to see that

is again an m-module.

**Universal bundles**

Various definitions of universal or enveloping algebras (resp., universal representations), attached to triple systems or algebras with involution, can be given; see [10,17,18,19]. These objects should correspond to certain "universal bundles" over a given symmetric space M. We intend to investigate such questions elsewhere.

We now study in more detail the dierential geometric aspects of symmetric bundles. It is immediately
clear from Section 2.5 that a symmetric bundle F carries a fiber bundle connection in
the general sense of Ehresmann, i.e., there is a distribution of horizontal subspaces, complementary
to the vertical subspaces (tangent spaces of the fiber); namely, as horizontal
subspace take the fixed point spaces of the dierentials of the horizontal symmetry v_{u}

In the sequel, we show that this Ehresmann connection is indeed a linear connection (in the general sense defined in [5], which in the real case amounts to the usual definitions), and that in general it has nonvanishing curvature, so that the distribution is in general not integrable.

**Theorem 6.1**. Let be a symmetric bundle over M. Then there exists a unique linear connection on the vector bundle F which is invariant under all symmetries

**Proof. **The proof is similar to the one of [5, Theorem 26.3], and therefore will not be spelled
out here in full detail. The uniqueness statement is proved by observing that the dierence of two linear connections on F is a tensor field such that is bilinear; if both L_{1} and L_{2} are invariant under σ_{p}, then and hence It follows that The main argument for the proof of existence consists in
proving that the fibers of the bundle TF over M are abelian symmetric spaces (for this one has
to analyze the map in the fiber over a point in the same way as the
corresponding map was analyzed in [5, Lemma 26.4]); then one
concludes by general arguments that the fibers of TF over M carry canonically the structure of
a linear space which is bilinearly related to all linear structures induced by bundle charts. By
definition, this is what we call a linear connection on F.

**Extension problem: on uniqueness**

When is the tangent bundle, with its canonical symmetric bundle structure, the connection defined by the preceding theorem is precisely the canonical connection of the symmetric space M (cf. [5, 16]). We will see in the next chapter that the abstract bundle may carry several dierent (nonequivalent) symmetric bundle structures over M. The uniqueness statement of the theorem shows that they all lead to the same linear connection on over M. Therefore the symmetric bundle structure is not uniquely determined by the linear connection from Theorem 6.1. Only at second order, by considering connections on over one is able to distinguish two symmetric bundle structures on

Theorem 6.2. Let be the curvature tensor of the linear connection on the symmetric bun- dle F over M defined in the preceding theorem. Then is given by the r-component of the corresponding Lts representation: for all sections X; Y of TM and sections

(i.e., with respect to an arbitrary base point for all

**Proof. **In case where Ω is the curvature of the canonical connection of M, it is well
known that
is (possibly up to a sign, which is a matter of convention) given by the Lie triple system of M, i.e., see [16,13,3,5] for three dierent proofs.
It is inevitable to go into third-order calculations, and therefore none of these proofs is really
short. Theorem 6.2 generalizes this result, and it can also be proved in several dierent ways.
We will here just present the basic ideas and refer the reader to the above references for details.
(a) Approaches using sections. Recall that the curvature
may be defined by

where is the horizontal lift of a vector field According to the
definition of the horizontal space H_{u} given above, the horizontal lift of a vector field X is given,for by

By homogenity, it suces to calculate
Ω_{o}, the value at the base point, and then it is enough to
plug in the vector fields having value for This already implies that and so we are left with calculating This can be
done by a calculation in a chart, corresponding essentially to [5, Lemma 26.4]; the outcome is [v;w; u], as expected.

(b) Approaches using higher-order tangent bundles. We analyze the structure of the bundles TF and TTF in exactly the same way as done in [5, Chapter 27], for the case of the tangent bundle: as mentioned in the proof of the preceding theorem, TF has abelian fibers defining the canonical connection, and TTF is non-abelian, leading to an intrinsic description of the curvature in terms of the symmetric space structure of the fibers. The conclusion is the same as with for u; v;w belonging to the three "axes" of TTF.

**Extension problem: on existence**

Via Theorem 5.3, we can relate the extension problem to a problem on holonomy representations: assume M = G/H is a homogeneous symmetric space and assume given a representation then the associated bundle carries a tensor field of curvature type coming from the derived representation Can we find a connection (coming from a symmetric bundle structure on F) such that r is its curvature, i.e., such that this representation becomes a holonomy representation? In case of the tangent bundle, F = TM, the answer clearly is positive, since H is the holonomy group of the canonical connection on the tangent bundle.

In this chapter we will show that, for a rather big class of symmetric spacesM, the tangent bundle TM carries (at least) two dierent symmetric structures with the same underlying re ection space structure. For instance, this is the case for the general linear group, seen as symmetric space.

**Example: the general linear group**

The tangent bundle of the group can be identified with the general linear group over the dual numbers

with ε-bilinear multiplication The canonical symmetric space structures on and on its tangent bundle are given by the product map (see Example 2.1).

**Theorem 7.1. **The vector bundle admits a second symmetric bundle structure iso-
morphic to the homogeneous symmetric space

where is the noncommutative ring of "degenerate quaternions over ",

with group involution of L induced by conjugation of with respect to its subalgebra of diagonal matrices.

**Proof. **Before coming to details of the calculation, let us give a heuristic argument: the "Hermitian
complexification" of the group is the symmetric space of complex structures on (see [3, Chapter IV]). In principle, in the present context we should replace complex structures by "infinitesimal structures" but this attempt
fails since E is not invertible. However, changing the point of view by considering a complex
structure on rather as a-form of the algebra and viewing rather as the suitably reformulated arguments carry over from the Cayley-
Dickson extension to the "degenerate Cayley-Dickson extension" (and
in fact to any extension Á la Cayley-Dickson of a commutative ring with nontrivial involution,
see below). In the following calculations we need the matrices

Let us abbreviate Then

where we call map an this
is justified by the fact that the fixed ring is the ring of diagonal matrices in D, which is
isomorphic to A, and that *T* anticommutes with the "structure map"For
the corresponding K-linear maps we will again write and f instead of and

The group

acts on the space by conjugation. The stabilizer of is given by b = 0, i.e., it is the group Thus the orbit of is a homogeneous symmetric space:

We will show now that is a vector bundle over and that this vector bundle is isomorphic to the homogeneous bundl over To this end, observe that the group contains a subgroup isomorphic to namely the group of matrices of the form

with Now let this subgroup act on

The stabilizer of *T* is gotten by taking g = h, and so the orbit of under this group is isomorphic
to the symmetric space (group case). On the other hand, contains the abelian normal subgroup of matrices of the form

with and is a semidirect product of these two subgroups. It follows that O is a vector bundle over the orbit Let us determine fiber over the base point F_{n}. Since

the stabilizer is gotten by X = 0, whereas for Y = 0 we get the fiber of over the base point:

Thus the fiber is isomorphic to and hence O is isomorphic as a homogeneous bundle to The remaining task of calculating the Lts of the symmetric space O becomes easier
by transforming everything with the "Cayley transform" R_{n}: the Cayley transformed version of

The Lie algebra of is imbedded as the subalgebra given by the conditions a = d and b = c. The tangent spaces of our two special orbits inside are complementary to this subalgebra, namely

so that is the Lts of O, where the triple product is the usual triple Lie bracket of matrices since the group action is by ordinary conjugation of matrices. The formula shows that m clearly is isomorphic to the usual Lts of that V is abelian and that acts on V by conjugation, i.e., the r-component of the corresponding Lts representation of m is the usual one (corresponding to the fact that O is isomorphic to as a homogeneous bundle). However, the whole Lts representation just defined is not equivalent to the adjoint Lts representation of In fact, the corresponding involutive Lie algebras are not isomorphic: already for n = 1, they are not isomorphic since is commutative, whereas is not.

As mentioned above, instead of dual numbers we could have taken for A another ring extension of the form with arbitrary instead of Then the symmetric space has two dierent scalar extensions from the "straight one", simply gotten
by taking the group case and another, "twisted" one, given by the homogeneous symmetric spacewhere is the split Cayley-Dickson extension of (see [6] for details). One could even replace here the algebra of square matrices by any other associative K-algebra. For we are back in the example of the "twisted complexification of It is interesting that the interpretation of O as the "space of complex structures"
(i.e., endomorphisms with E^{2} = ) works only for invertible scalars , whereas the interpretation
given here works uniformly for all scalars

**ordan-extensions**

Besides general linear groups, for all other "classical" symmetric spaces, there exist similar descriptions of symmetric bundle structures on the tangent bundle, see [6] for the case of Grassmannians, Lagrangians and orthogonal groups. The latter example gives rise to the "D-unitary groups", analogs of Sp(n) with H replaced by D. The general construction principle behind these examples uses Jordan theory:

**Definition 7.2**. A Jordan-extension of an Lts (m;R) is given by a Jordan triple product T : such that

Recall that a Jordan triple system (Jts) is a linear space m together with a trilinear map such that

is symmetric in the outer variables:

For any Jts T, the trilinear map defined by is an Lts: we call the correspondence the Jordan-Lie functor (cf. [3]).

**Theorem 7.3 **(the twisted regular representation defined by a Jordan extension). Assume is an Lts having a Jordan extension T. Let be the regular representation of m (recall that the corresponding split null extension is just the scalar extension by dual numbers). Then there exists another representation in general, not isomorphic to the regular representation, but having the same r-component

Proof. We follow the lines of the proof of the corresponding statement for complexifications in [3, Chapter III] LET and be the ε-trilinear scalar extension of T by dual numbers. Then the conjugation

is a K-automorphism of T[ε]. But for any involutive K-automorphism, the "-isotope"

is again a Jts (cf. [3, Lemma III.4.5] for the easy proof). Moreover, this new Jts is "-linear in the
outer variables and "-antilinear in the inner variable, and since acts trivially on m, restriction
of this new Jts to m^{3} gives us back T again. Now we let

This is an Lts which coincides with R on m since T was chosen to be a Jordan extension of R. Next, "m is an ideal of by (anti-)linearity it is an ideal of and hence it is one of Finally, if two terms belong to then application of gives zero, and therefore also application of gives zero. Thus is an Lts having the properties from Lemma 3.2, and hence is the split null extension corresponding to a representation on εm.

Now we show that for

In order to prove that (r;m) and are in general not isomorphic, observe that the split null extension of (r;m), being just scalar extension by has the property that On the other hand,

Thus R and cannot belong to isomorphic representations unless they vanish both. (This does not exclude that, as Lie triple systems over K, they may be isomorphic in special cases.)

**Final comments**

Essentially, all classical Lie triple systems (and about half of the exceptional ones) admit Jordan extensions (cf. [3, Chapter XII]). For instance, it is easily checked that with the triple product is a Jts, and then

so that we have a Jordan extension of Correspondingly, and essentially all classical symmetric spaces admit on their tangent bundle a symmetric bundle structure that is dierent from the usual one. We conjecture that (at least for simple finite-dimensional Lts over all symmetric bundle structures on the tangent bundle are exactly of two types:

(1) "straight": given by the canonical symmetric structure on TM, corresponding to the regular representation of the Lts m,

(2) "twisted": given by the construction from the preceding theorem.

This conjecture is of course supported by the corresponding fact for complexifications of symmetric spaces, which (for simple finite-dimensional Lts over R) are either straight or twisted (see [3, Corollary V.1.12]). However, the proof given in loc. cit. for the complex and para-complex cases does not carry over to the tangent case (the invertibility of i, resp. j, in K[i], resp. K[j], is used at a crucial point, and " clearly is not invertible in K["ε"]). A proof covering all three cases at the same time would be of great value for a better understanding of the "Jordan-Lie functor" (see [3]), and it should relate the "extension problem for the Jordan-Lie functor" with the extension problem for Lts representations as discussed here. One might conjecture that an interpretation in terms of the Cayley-Dickson process, which turned out to be useful in the special case of GL(n,K), could be the key for proving the conjecture, but this is not clear at present.

The first author would like to thank the Hausdor Institute (Bonn) for hospitality when part of this work was carried out.

- Barr M (1996) Cartan-Eilenberg cohomology and triples. J Pure Applied Algebra 112: 219-238.
- Beck JM (1967) Triples, Algebras and Cohomology. Ph.D.Thesis Columbia University.
- Bertram W (1754) The Geometry of Jordan and Lie Structures. Lecture Notes in Math, Springer, Berlin.
- Bertram W (2002) Generalized projective geometries: General theory and equivalence with Jordan structures. Adv. Geom 2:329-369.
- Bertram W (2008) Differential geometry, Lie groups and symmetric spaces over general base fields and
- rings. Mem Amer. Math. Soc 192.
- Didry M (2006) Structures alg´ebriques associ´ees aux espaces sym´etriques. Thesis, Institut Elie Cartan,Nancy.
- Didry M (2007) Construction of groups associated to Lie- and Leibniz algebras. J. Lie theory 17:399-426.
- Eilenberg S (1948) Extensions of general algebras. Ann. Soc. Pol. Math 21:125-134.
- Faraut J, Kor`anyi A (1994) Analysis on Symmetric Cones. Clarendon Press, Oxford.
- Harris B (1961) Cohomology of Lie triple systems and Lie algebras with involution. Trans. Amer. Math.Soc 98:148-162.
- Hodge TL, Parshall BJ (2002) On the representation theory of Lie triple systems. Trans. Amer.Math. Soc.354:4359-4391.
- Jacobson N (1951) General representation theory of Jordan algebras. Trans. Amer. Math Soc 70: 509-530.
- Kobayashi S, Nomizu K (1969) Foundations of Differential Geometry. Wiley, New York,2.
- Lister WG (1952) A structure theory for Lie triple systems. Trans. Amer. Math. Soc. 72: 217-242.
- Loos O (1967) Spiegelungsr¨aume und homogene symmetrische R¨aume. Math Z. 99: 141–170.
- Loos O (1969) Symmetric Spaces I. Benjamin, New York.
- Loos O (1973) Representations of Jordan triples. Trans. Amer. Math. Soc 185:199-211.
- Loos O (1975) Jordan Pairs. Lecture Notes in Math, Springer, Berlin.
- Mostovoy J, P´erez-Izquierdo JM (2005) Ideals in non-associative universal enveloping algebras of Lie triple systems. Preprint arXiv:math.
- Sabinin LV (1999) Smooth quasigroups and loops. Mathematics and its Applications, Kluwer Academic Publishers, Dordrecht.
- Smirnov O (2009) Imbedding of Lie triple systemss into Lie algebras. Preprint arXiv:math.

Select your language of interest to view the total content in your interested language

- Adomian Decomposition Method
- Algebra
- Algebraic Geometry
- Analytical Geometry
- Applied Mathematics
- Axioms
- Balance Law
- Behaviometrics
- Big Data Analytics
- Binary and Non-normal Continuous Data
- Binomial Regression
- Biometrics
- Biostatistics methods
- Clinical Trail
- Combinatorics
- Complex Analysis
- Computational Model
- Convection Diffusion Equations
- Cross-Covariance and Cross-Correlation
- Deformations Theory
- Differential Equations
- Differential Transform Method
- Fourier Analysis
- Fuzzy Boundary Value
- Fuzzy Environments
- Fuzzy Quasi-Metric Space
- Genetic Linkage
- Geometry
- Hamilton Mechanics
- Harmonic Analysis
- Homological Algebra
- Homotopical Algebra
- Hypothesis Testing
- Integrated Analysis
- Integration
- Large-scale Survey Data
- Latin Squares
- Lie Algebra
- Lie Superalgebra
- Lie Theory
- Lie Triple Systems
- Loop Algebra
- Matrix
- Microarray Studies
- Mixed Initial-boundary Value
- Molecular Modelling
- Multivariate-Normal Model
- Noether's theorem
- Non rigid Image Registration
- Nonlinear Differential Equations
- Number Theory
- Numerical Solutions
- Operad Theory
- Physical Mathematics
- Quantum Group
- Quantum Mechanics
- Quantum electrodynamics
- Quasi-Group
- Quasilinear Hyperbolic Systems
- Regressions
- Relativity
- Representation theory
- Riemannian Geometry
- Robust Method
- Semi Analytical-Solution
- Sensitivity Analysis
- Smooth Complexities
- Soft biometrics
- Spatial Gaussian Markov Random Fields
- Statistical Methods
- Super Algebras
- Symmetric Spaces
- Theoretical Physics
- Theory of Mathematical Modeling
- Three Dimensional Steady State
- Topologies
- Topology
- mirror symmetry
- vector bundle

- 7th International Conference on Biostatistics and Bioinformatics

September 26-27, 2018 Chicago, USA - Conference on Biostatistics and Informatics

December 05-06-2018 Dubai, UAE - Mathematics Congress - From Applied to Derivatives

December 5-6, 2018 Dubai, UAE

- Total views:
**11792** - [From(publication date):

December-2009 - Jan 18, 2018] - Breakdown by view type
- HTML page views :
**7996** - PDF downloads :
**3796**

Peer Reviewed Journals

International Conferences
2018-19