Conformal Holonomy: a Classification, , 23 January 2005

1 Introduction

Conformal geometry is but one of the class of parabolic geometries, a group that includes, amongst others, projective, almost Grassmanian, almost quaternionic, and co-dimension one CR structures. Some of these geometries were studied as far back as E. Cartan [11] , whose technique of `moving frames' would ultimately develop into the concepts of principal bundles and Cartan connections.

Despite the work of T.Y. Thomas [26] , [27] who developed key ideas for Tractor calculus in the nineteen twenties and thirties, and Shiego Sasaki in 1943 [24] , [25] , the subject fell into abeyance until the work of N. Tanaka [23] in 1979. The subject was rediscovered and further developed by T.N. Bailey, M.G. Eastwood and R. Gover in 1994 [1] .

Since then, there have been a series of papers by A. Čap and R. Gover [9] , [8] , [14] , [10] , developing a lot of the techniques that will be used in the present paper.

Previous papers had focused on seeing the Cartan connection for conformal geometry as a property of a principal bundle

P

. More recently, the principal bundle is replaced by an associated vector bundle, the Tractor bundle

T

, and the Cartan connection by a connection form for

T

, the Tractor connection

\vec{\nabla}

With these tools, calculations are considerably simplified.

The purpose of this paper is to analyse one of the invariants of the Tractor connection, the holonomy group. There is an invariant metric of signature

(n + 1, 1)

T

, so this holonomy group must be a sub-group of

G = S O (n + 1, 1)

It is a well known fact that a parallel section of the Tractor bundle corresponds to the local existence of an Einstein metric in the conformal class of a manifold.

Beyond this, little was known about reductions of holonomy.

It is the purpose of this paper to classify all the possible local holonomy groups of

\vec{\nabla}

acting reducibly on

T

. In doing so, they must conserve a Lorentzian metric of signature

(n + 1, 1)

. Then a paper by A.J. Di Scala and C. Olmos [12] shows that we have the complete list: there exist no connected proper subgroups of

S O (n + 1, 1)

acting irreducibly on

R^{n + 1, 1}

. A very recent paper by Felipe Leitner, [20] , proves the same results as in this paper; but his methods, involving normal Killing Spinors, are different from those described here.

The classification comes in two main pieces; if a bundle of rank other than

1

n

is preserved, the manifold decomposes analogously to the De Rham decomposition:

Theorem 1.1. Let

(M, [g])

be a conformal,

n

-dimensional manifold, such that

T_{M}

has a holonomy preserved sub-bundle of rank

k

. Then there exists a metric

g \in [g]

such that

(M, g)

splits locally into the direct product of two Einstein manifolds

M_{U}

M_{V}

of dimensions

l = k - 1

and

n - l

. The Einstein constants

a

and

b

M_{U}

M_{V}

are related by

(n - l - 1) a = (1 - l) b

. Furthermore, the Tractor holonomy group of

M

is the direct product of those of

M_{U}

and

M_{V}

. This is possible as

M_{U}

and

M_{V}

are Einstein, hence their Tractor holonomies reduce.

The converse is also true. This decomposition, again, is a local result, and may become degenerate along some embedded sub-manifolds. The second result is to list all the possible Tractor holonomies for a conformally Einstein manifold. Using a metric cone construction, related to the Ambient Metric of [13] , [10] and [15] , the following list is established:

Theorem 1.2 (Einstein Classification). The Tractor holonomy of

(M^{n}, [g])

, for

M^{n}

conformal to an Einstein space of non-zero scalar curvature, is one of the following groups:

- $S O (n, 1), n \geq 4$ ,
- $S O (n + 1), n \geq 4$ ,
- $S U (m)$ for $2 m = n + 1, n \geq 4$ ,
- $S p (m)$ for $4 m = n + 1$ ,
- $G_{2}$ for $n = 6$ ,
- $S p i n (7)$ for $n = 7$ .

Moreover, all these actually occur as holonomy groups.

The Ricci-flat case must be treated differently; in fact, if

(M^{n}, g)

is Ricci-flat and conformally indecomposable, and

G

is the metric holonomy group of

\nabla^{g}

, then

(M^{n}, [g])

has Tractor holonomy

G ∖ R^{n}

. Thus:

Theorem 1.3. The possible indecomposable Tractor holonomy groups for the conformal manifold

(M^{n}, [g])

, conformally Ricci-flat, are:

- $S O (n) ∖ R^{n}, n \geq 4$ ,
- $S U (m) ∖ R^{2 m}, m \geq 2$ ,
- $S p (m) ∖ R^{4 m}, m \geq 1$ ,
- $G_{2} ∖ R^{7}$ ,
- $S p i n (7) ∖ R^{8}$ ,

and all of these groups do occur.

This paper begins with defining and laying out the groundwork for the conformal Tractor Bundle and connection, in Section

2

. Furthermore, it will prove the equivalence of this (second order) point of view with the standard equivalence class of metric structures.

The brief Section 3 presents some known results, showing how an Einstein structure in the conformal class is equivalent to a parallel section of the Tractor bundle.

Section 4, the longest and most full of technical results, proves the decomposition theorem previously mentioned.

Section 5 then establishes the list for the Einstein spaces via the metric cone construction, with Section 6 complementing it to list the possible holonomies for conformally Ricci-flat manifolds, using different methods. A brief note on symmetric spaces ends the paper.

Remark. In all the holonomy groups listed in this paper, the holonomy reduction corresponds to the existence of a particular metric in the conformal class. Hence we always have a canonical representative in the conformal class, whenever the holonomy reduces.

2 Cartan Connection: Theory

2.1 The Cartan Connection

With flat geometries, one deals with spaces

X = G / P

, for

G

a Lie group acting transitively and effectively on

X

and

P

a parabolic subgroup.

In the general case, we can use the Cartan connection to construct an infinitesimal analogue of this isomorphism, mapping the tangent space

T

into the quotient of Lie algebras

g / p

instead.

In order to do this, we will follow the exposition used in [9] . Let

G

be a semi-simple Lie group, with a

| k |

-grading of its Lie algebra, that is,

\begin{matrix} g = g_{- k} \oplus \dots \oplus g_{0} \oplus \dots \oplus g_{k}, \end{matrix}

with the additional conditions that no simple ideal of

g

is contained in

g_{0}

and the sub-algebra

p_{+} = g_{1} \oplus \dots \oplus g_{k}

is generated by

g_{1}

. In our case, we will in fact be dealing with

1

-gradings of a simple Lie algebra, which makes both of these conditions trivially true.

Then we define the groups

G_{0} \leq P

to be the subgroups of

G

that preserve the grading and the decreasing filtration respectively, i.e.

\begin{matrix} G_{0} & = & {g \in G : A d (g) g_{i} \subseteq g_{i}, \forall i}, \end{matrix}

\begin{matrix} P & = & {g \in G : A d (g) g_{i} \subseteq Σ_{j \geq i} g_{j}, \forall i} . \end{matrix}

Obviously the Lie Algebra of

G_{0}

g_{0}

and that of

P

p = g_{0} \oplus \dots \oplus g_{k}

. Given a manifold

M

, the group

G

and the grading of

g

, we can define the Normal Cartan Connection.

Definition 2.1 (Normal Cartan Connection). On

M

, given a principal

P

-bundle

P \to M

, a normal Cartan connection

ω

is a section of

T_{P}^{*} \otimes g

, with the following properties:

1. $ω$ is invariant under the $P$ -action ( $P$ acting by $A d$ on $g$ ),
2. $ω (σ_{A}) = A$ , where $σ_{A}$ is the fundamental vector field of $A \in p$ ,
3. $ω_{u} : T P_{u} \to g$ is a linear isomorphism for all $u \in P$ ,
4. The `curvature' $k (η, ξ) = d ω (η, ξ) + [ω (η), ω (ξ)]$ is such that $\partial^{*} k = 0$ where $\partial^{*}$ is the dual Spencer cohomology operator (see [3] for an exposition of Spencer cohomology).

Remark. The last condition is what makes the connection normal ; this is a uniqueness condition for the Cartan connection of a particular parabolic geometry, similar to the torsion-free condition for a Levi-Civita connection. See [9] for proof of the existence of a normal Cartan connection in all cases. In the conformal case, an equivalent condition is that

k (η, ξ) \in p

for all

η, ξ \in T P

The bundle

P

and the form

ω

together define the parabolic geometry.

Properties 1 and 2 make

ω

analogous to a standard connection; property 3, however, implying a point-wise isomorphism, is very different. However, the inclusion

P ↪ G

generates a principal bundle inclusion

i : P ↪ G

, and we get the following result.

Proposition 2.2. There is a unique

ω^{'} \in Ω^{1} (G, g)

such that

ω^{'}

is a standard connection form on

G

and

i^{*} ω^{'} = ω

Proof. At any point of $P ↪ G$ , define $ω^{'} (X) = ω (X)$ for $X \in Γ (T P)$ , and $ω^{'} (σ_{A}) = A$ for $σ_{A}$ the fundamental vector field of $A \in g$ . These two formulas correspond whenever they are both defined (property 2 from definition 2.1 ), and completely define $ω^{'}$ on $P$ . Then define $ω_{u}^{'} = g^{*} (ω_{g (u)}^{'})$ in the general case, for $g (u) \in P$ .
Property 1 for $ω$ ensures this is well defined. It is then easy to see that $ω^{'}$ is a connection, and is unique. $□$

Remark. This

ω^{'}

is hence a connection in the usual sense, equivalent with the Cartan connection that generated it; it is the so-called Tractor connection.

The Tractor connection

ω^{'}

obviously generates a Cartan connection by pull-back to

T P

. From now on, we shall use Cartan and Tractor connections interchangeably.

It is not the case that any

G

connection

μ

will correspond to a Cartan connection via pull-back to

P

, as the isomorphism condition might be violated.

2.2 Conformal Geometry

A conformal manifold, in the standard sense, is a manifold

(M, [g])

, where

[g]

is an equivalence class of metrics related by multiplication by a nowhere zero function. Alternatively, it is a reduction of the principal bundle of the tangent space of

M

to a principal

C O (n)

bundle which we shall call

G_{0}

. In this case,

C O (n) = {λ s | λ \in R^{+} and s \in S O (n)}

From the Cartan connection point of view, however, the situation seems very different; later we shall see that both are equivalent.

In the classical, flat, case, Conformal Geometry is modelled on the sphere

S^{n}

. Taking the sphere as the collection of null-lines in

R^{n + 1, 1}

, the group

G

of conformal transformations is

S O (n + 1, 1)

. Then its Lie algebra has a 1-grading,

\begin{matrix} g = R^{n} \oplus c o (n) \oplus R^{n *}, \end{matrix}

where the conformal group

c o (n)

decomposes into the semisimple part

s o (n)

and the centre

R

, which is responsible for the conformal weight in representations of

c o (n)

Thus the data are

G = S O (n + 1, 1)

G_{0} = C O (n)

and

P = C O (n) ∖ R^{n *}

, on an

n

-dimensional space. Note that we have a natural action of

G_{0}

g

and hence an associated bundle to the

G_{0}

bundle mentioned previously:

\begin{matrix} G_{0} \times_{G_{0}} g . \end{matrix}

Moreover, the action of

G_{0}

splits

g

, giving a corresponding splitting:

\begin{matrix} G_{0} \times_{G_{0}} g = T \oplus c o (T) \oplus T^{*} . \end{matrix}

This decomposition will be used extensively.

For later on, it will be important to explicate the Lie bracket of this algebra [9] . In fact,

[T, T] = [T^{*}, T^{*}] = 0

, the Lie bracket on

c o (T)

is the natural commutator of endomorphisms, and

[c, t^{*}] = - c (t^{*})

[c, t] = c (t)

, for

t, c, t^{*}

sections of

T, c o (T)

and

T^{*}

respectively. The bracket between

T

and

T^{*}

is more complicated, and in fact

\begin{matrix} [t, s^{*}] = t \otimes s^{*} - (t \otimes s^{*})^{τ} + s^{*} (t) δ, \end{matrix}

with

τ

the transpose operator, and

δ

the identity element in (the centre of )

c o (T)

In their papers [9] and [8] , the authors demonstrate that the Cartan connection is equivalent to the standard conformal structure on a manifold

(M, [g])

. This is an alternative treatment.

By definition, on a conformal manifold

(M, [g])

, the structure group reduces to the principal

G_{0}

bundle

G_{0}

. This allows us to define the bundles

ℰ [w]

, the weighted line-bundles coming from the centre of

C O (n)

, i.e.

\begin{matrix} ℰ [w] = G_{0} \times_{ρ} R, ρ (c) (y) = - w \frac{d e t (c)}{2 n} y . \end{matrix}

From now on, we'll use the notation

B [w]

for

B \otimes ℰ [w]

. Then the conformal structure

[g]

can be more properly thought of as a section

g

(⊙^{2} T) [- 2]

. We will use

[g]

and

g

interchangeably from now on.

Then given a non-vanishing section

ξ

ℰ [1]

a conformal scale there is a corresponding metric in the conformal class

\begin{matrix} g^{ξ} = ξ^{- 2} g, \end{matrix}

with a corresponding Levi-Civita connection

\nabla

Another equivalent way of defining the conformal structure is to use the class of preferred connections:

Definition 2.3. Given a conformal manifold

G_{0} \to M

, a preferred connection

\nabla

is simply a torsion-free

G_{0}

connection.

These preferred connections offer the first hint of how the Cartan connection ties in with the more conventional definitions of conformal structures, via the change of connection formula:

Theorem 2.4. Let

s

be a section of any bundle associated to

G_{0}

, and let

X

be any vector field. Then if

\nabla

and

\hat{\nabla}

are two preferred connections, there exists a one-form

ϒ

such that

\begin{matrix} {\hat{\nabla}}_{X} s = \nabla_{X} s + [ϒ, X] . s, \end{matrix}

where

[,]

is the Lie bracket for

g

previously described.

Proof. For X ∈ Γ ( T ) and ϒ ∈ Ω 1 ( M ) , [ ϒ , X ] is a section of c o ( T )
, so this definition makes sense.
We know that ∇ ^ X = ∇ X + q ( X ) , where q is a one-form with values in c o ( T )
.
However preferred connections are torsion-free, so $q (X) . Y$ must be symmetric in $X$ and $Y$ , implying that $q$ lies in the bundle $Q$ , the symmetrisation of $T^{*} \otimes c o (T)$ around the first two factors.
Then since $[ϒ, X] . Y = [[ϒ, X], Y]$ , the Jacobi identity implies
$\begin{matrix} [[ϒ, X], Y] - [[ϒ, Y], X] = - [[X, Y], ϒ] = 0, \end{matrix}$
giving us a (clearly injective) map $μ : T^{*} \to Q$ .
Choose a metric $g \in [g]$ to split the bundle $T^{*} \otimes c o (T)$ as $(T^{*} \otimes s o (T)) \oplus (T^{*} \otimes I d)$ .
Lemma 2.5. $(T^{*} \otimes s o (T)) \cap Q = 0$

Proof. Let $\tilde{\nabla}$ be the Levi-Civita connection of $g$ . If $a \in Γ ((T^{*} \otimes s o (T)) \cap Q)$ , then $\tilde{\nabla} + a$ is another torsion-free connection annihilating $g$ . Hence, since the Levi-Civita connection is unique,
$\begin{matrix} \tilde{\nabla} = \tilde{\nabla} + a, \end{matrix}$
meaning $a = 0$ . $□$

The rank of $T^{*} \otimes s o (T)$ is $\frac{n (n^{2} - n)}{2}$ , while the rank of $T^{*} \otimes c o (T)$ is $\frac{n (n^{2} - n)}{2} + n$ ; hence $Q$ is of rank at most $n$ . Thus $μ$ must actually be a bijection, proving the result. $□$

Note that if

\nabla

were a metric connection, then

\hat{\nabla}

would be metric if and only if

ϒ

were a closed form. In fact:

Proposition 2.6. Let

\nabla

and

\hat{\nabla} = \nabla + ϒ

be two metric, preferred connections, with

ξ, \hat{ξ}

the corresponding conformal scales. Defining the function

f

f = ξ \otimes {\hat{ξ}}^{- 1} \in Γ (ℰ [0]) = C^{\infty} (M)

, we have

\begin{matrix} ϒ = f^{- 1} d f = d (l o g f) . \end{matrix}

Proof. By direct calculation, using the fact that $\nabla$ annihilates $g^{ξ}$ while $\hat{\nabla}$ annihilates $g^{\hat{ξ}} = f^{2} (g^{ξ})$ . $□$

A variety of tensors connected with these preferred connections will be needed in subsequent chapters. To define them, we will use Penrose's abstract index notation, where

Q^{i}

is understood as a section of the tangent bundle,

Q_{i}

a section of the co-tangent bundle, and symmetrisation and anti-symmetrisation of indexes to be denoted by

(i j)

and

[i j]

respectively. This notation will be used intermittently throughout the paper.

Then if

R_{i j k l}

is the curvature tensor of

\nabla

, recall [14] :

\begin{matrix} R_{i j k l} = W_{i j k l} + 2 g_{k [i} P_{j] l} - 2 g_{l [i} P_{j] k} - 2 P_{[i j]} g_{k l} \end{matrix}

(1)

with

W_{i j k l}

the conformally invariant Weyl tensor, and the rho-tensor

P

\begin{matrix} P_{i j} = - \frac{1}{n - 2} (\frac{1}{n} R i c_{i j} + \frac{n - 1}{n} R i c_{j i} - \frac{1}{2 n - 2} R g_{i j}) \end{matrix}

a particularly important tensor for the rest of the paper. In this case,

R i c_{i j}

is the Ricci curvature, and

R

the scalar curvature

R i c_{i j} g^{i j}

a section of

ℰ [2]

This is in the general case for a conformal connection; in the metric case, the picture is the same, except that

P

follows the simpler symmetric formula

\begin{matrix} P_{i j} = - \frac{1}{n - 2} (R i c_{i j} - \frac{1}{2 n - 2} R g_{i j}) . \end{matrix}

The last relevant tensor for

\nabla

is the Cotton-York tensor:

\begin{matrix} C Y_{i j k} = 2 \nabla_{[i} P_{j] k} . \end{matrix}

(2)

It will be important to understand how the tensor

P

varies under a change of conformal structure, as this formula is the key to defining the Tractor bundle.

Letting

P

be the rho-tensor for

\nabla

and

\hat{P}

be that of

\hat{\nabla} = \nabla + ϒ

\begin{matrix} \hat{P} (ξ) = P (ξ) - \nabla_{ξ} ϒ + \frac{1}{2} [ϒ, [ϒ, ξ]], \end{matrix}

(3)

for

ξ

any vector field.

2.2.1 Equivalences

So far there are three ways of envisaging the conformal structure on a manifold

M

- via a conformal metric $g$ (a zero order structure),
- via a class of preferred connections $\nabla$ (a first order structure) or
- via a Cartan/Tractor Connection $ω$ / $\vec{\nabla}$ (a second order structure),

and we will be using all three of them. This subsection will be dedicated to proving the important fact that all three viewpoints are equivalent. Though we will draw heavily on [8] for this exposition, we will use a slightly unconventional approach, which has the advantage of constructing the vital `Tractor Bundle' directly.

Remark. For a variety of reasons to do mainly with conventional notation and ease of calculations, we will be working with the Tractor bundle

T

in the rest of the paper. However, to get a better understanding of what this bundle actually is, we need to start by defining the dual bundle

T^{*}

Consider the two-jet prolongation of

J^{2} (ℰ [1])

of the weighted bundle

ℰ [1]

By definition, we have the short exact sequences

\begin{matrix} 0 ⟶ ⊙^{2} T^{*} [1] ⟶ & J^{2} (ℰ [1]) & ⟶ J^{1} (ℰ [1]) ⟶ 0, \end{matrix}

\begin{matrix} 0 ⟶ T^{*} [1] ⟶ & J^{1} (ℰ [1]) & ⟶ ℰ [1] ⟶ 0 . \end{matrix}

The conformal structure

g

contracts

⊙^{2} T^{*}

ℰ [- 2]

. Hence

⊙^{2} T^{*}

splits as

(⊙^{2} T^{*})_{0} \oplus ℰ [- 2]

, where the first space is the kernel of the contraction. Then we define the dual Tractor bundle

T^{*}

as the quotient:

\begin{matrix} 0 ⟶ (⊙^{2} T^{*})_{0} [1] ⟶ & J^{2} (ℰ [1]) & ⟶ T^{*} ⟶ 0 . \end{matrix}

It is actually possible to realise

T^{*}

as a sub-bundle of

J^{2} (ℰ [1])

rather than a quotient bundle; we shall not be needing this result, though. Let

D

be the second order operator

Γ (ℰ [1]) \to Γ (T^{*})

given by composing the projection

J^{2} (ℰ [1]) \to T^{*}

with the two-jet operator

j^{2}

Proposition 2.7. Given a preferred connection

\nabla

s

any section of

ℰ [1]

and

b

any point on the manifold, the map

\begin{matrix} D s (b) \to (s (b), \nabla_{i} s (b), \frac{1}{n} g^{i j} (- \nabla_{i} \nabla_{j} s (b) + P_{i j} s (b))) \end{matrix}

generates an isomorphism

T^{*} \to ℰ [1] \oplus T^{*} [1] \oplus ℰ [- 1]

Proof. This formula clearly generates a bundle map $J^{2} (ℰ [1]) \to ℰ [1] \oplus T^{*} [1] \oplus ℰ [- 1]$ . All that remains is to prove that $(⊙^{2} T^{*})_{0} [1]$ is the kernel of this map.
Assume $D s (b) = 0$ .
Then obviously $j^{1} (s) = 0$ at $b$ , implying that $\nabla_{i} \nabla_{j} s (b)$ is the (well-defined) section of $⊙^{2} T^{*}$ that corresponds to the second derivative of $s$ at $b$ . Thus $- g^{i j} \nabla_{i} \nabla_{j} s (b) = 0$ , or equivalently $j^{2} (s) (b) = \nabla_{i} \nabla_{j} s (b) \in (⊙^{2} T^{*})_{0}$ . $□$

Notice that we have not used the tensor

P

yet; the next proposition shows what we need it for.

Proposition 2.8. Under a change of preferred connection

\nabla \to \hat{\nabla} = \nabla + ϒ

, the isomorphism of Proposition 2.7 changes as

\begin{matrix} (\begin{matrix} x \\ ω_{i} \\ z \end{matrix}) \to (\begin{matrix} x \\ ω_{i} - ϒ_{i} x \\ z + g^{i j} ω_{i} ϒ_{j} - \frac{1}{2} g^{i j} ϒ_{i} ϒ_{j} x \end{matrix}) . \end{matrix}

Proof. Direct computation from the transformation properties of

\nabla

and

P

The first component obviously stays the same. As

\hat{\nabla} = \nabla + ϒ

, we have the transformation law for the second component. For the third component:

\begin{matrix} {\hat{\nabla}}_{i} {\hat{\nabla}}_{j} = (\nabla_{i} + ϒ_{i}) (\nabla_{j} + ϒ_{j}) = \nabla_{i} \nabla_{j} + \nabla_{i} ϒ_{j} + ϒ_{i} \nabla_{j} + ϒ_{i j}^{2} . \end{matrix}

We need to remember that

ϒ_{i}

acts on a section

x

ℰ [1]

ϒ_{i} . x = - ϒ_{i} x

Then recall the transformation law for

P

in equation 3

()

\begin{matrix} {\hat{P}}_{i j} = P_{i j} - \nabla_{j} ϒ_{i} + \frac{1}{2} ϒ_{i j}^{2} . \end{matrix}

Thus in total:

\begin{matrix} (- {\hat{\nabla}}_{i} {\hat{\nabla}}_{j} + {\hat{P}}_{i j}) . x & = & (- \nabla_{i} \nabla_{j} + P_{i j}) . x + ϒ_{i} \nabla_{j} x - \frac{1}{2} ϒ_{i j}^{2} x \end{matrix}

\begin{matrix} - (\nabla_{j} ϒ_{i} + \nabla_{i} ϒ_{j}) x \end{matrix}

The last term in brackets is anti-symmetric, so disappears upon taking the contraction with the symmetric

g^{i j}

. Hence, for

D (s) = (x, ω_{i}, z)

b

\begin{matrix} \frac{1}{n} g^{i j} (- {\hat{\nabla}}_{i} {\hat{\nabla}}_{j} + {\hat{P}}_{i j}) s & = & \frac{1}{n} g^{i j} (- \nabla_{i} \nabla_{j} + P_{i j} + ϒ_{i} \circ \nabla_{j} - \frac{1}{2} ϒ_{i j}^{2}) s \end{matrix}

\begin{matrix} = & z + g^{i j} ω_{i} ϒ_{j} - \frac{1}{2} g^{i j} ϒ_{i} ϒ_{j} x . \end{matrix}

□

However, for reasons of convenience and notation, we will be working not with the bundle

T^{*}

but with its dual. Define the Tractor Bundle as

T = (T^{*})^{*}

The previous results carry through to the dual of

T^{*}

; any preferred connection

\nabla

defines a splitting

T = ℰ [1] \oplus T [- 1] \oplus ℰ [- 1]

, and under a change of connection, this splitting changes via

\begin{matrix} (\begin{matrix} x \\ Y \\ z \end{matrix}) \to (\begin{matrix} x \\ Y + ϒ^{*} x \\ z - ϒ (Y) - \frac{1}{2} g (ϒ, ϒ) x \end{matrix}), \end{matrix}

where

ϒ^{*} \in T [- 2]

is the dual to

ϒ

using the conformal metric

g

This particularly nice change of splitting formula implies the next vital lemma:

Lemma 2.9. There is a natural metric

〈, 〉

, of type

(n + 1, 1)

, on

T

Proof. Given a preferred connection

\nabla

and two sections of

T

(x, Y, z)

and

(x^{'}, Y^{'}, z^{'})

, we define the metric by

\begin{matrix} 〈 (\begin{matrix} x \\ Y \\ z \end{matrix}), (\begin{matrix} x^{'} \\ Y^{'} \\ z^{'} \end{matrix}) 〉 = x z^{'} + x^{'} z + g (Y, Y^{'}) . \end{matrix}

Direct calculation then shows this formula is invariant under a change of splitting.

□

Since

T^{*}

came about as a quotient bundle of a jet-bundle, we have invariant subspaces of

T

\begin{matrix} ℰ [- 1] ↪ T \oplus ℰ [- 1] ↪ T, \end{matrix}

and invariant projections:

\begin{matrix} π^{1} : & T \to & ℰ [1] \oplus T, \end{matrix}

\begin{matrix} π^{2} : & T \to & ℰ [1] . \end{matrix}

Call

E

the sub-bundle of

T

that is the inclusion of

ℰ [- 1]

. Note that

E

is null under

〈, 〉

Lemma 2.10. There is a

P

-bundle

P

which is a principal bundle for

T

Proof. The metric $〈, 〉$ shows that the structure group of $T$ reduces to $G = S O (n + 1, 1)$ . The invariant null sub-bundle $E$ further reduces the structure group to ${g \in G | g (E) = E}$ , i.e. to $P$ .
Then we define $P$ to be the bundle of orthonormal frames of $T$ preserving $E$ . $□$

Let us review what we have so far. Starting from the conformal metric

g

and the class of preferred connections, we have constructed, via a tensor

P

dependent on the connections, a bundle

T

. And this bundle generates a principal bundle

P

, where it is natural to suppose the Cartan connection living.

Define the Lie Algebra bundle

A = P \times_{P} g

. Then given a preferred connection

\nabla

we have a splitting of

T = ℰ [1] \oplus T [- 1] \oplus ℰ [- 1]

, and hence a corresponding splitting:

\begin{matrix} A = T \oplus c o (T) \oplus T^{*} . \end{matrix}

In order to finish the construction of this Cartan connection, we will start by building a

G = S O (n + 1, 1)

connection on

A

and then prove that it is a Tractor connection.

Definition 2.11. Given a preferred connection

\nabla

, we have a splitting of

A = T \oplus c o (T) \oplus T^{*}

. Each of these bundles is a

G_{0}

bundle, so

\nabla

ascends to a connection on

A

. Then we define the Tractor connection

\vec{\nabla}

\begin{matrix} {\vec{\nabla}}_{X} = \nabla_{X} + a d (X) + a d P (X), \end{matrix}

with the vector

X

and the one-form

P (X)

seen as sections of the Lie algebra bundle

A

Then since

\nabla

is a

g_{0}

connection,

\vec{\nabla}

is a

g

connection. Of course, this definition makes no sense without:

Proposition 2.12. This definition is independent of the choice of

\nabla

Proof. The formula for the change of splitting of

A

(deduced directly from that of

T

) is:

\begin{matrix} (\begin{matrix} X \\ Ψ \\ ω \end{matrix}) \to (\begin{matrix} X \\ Ψ + [ϒ, X] \\ ω + [ϒ, Ψ] + \frac{1}{2} [ϒ, [ϒ, X]] \end{matrix}), \end{matrix}

Then a direct calculation proves the result.

□

Thus for any bundle

B

associated to

G

, we have an invariant connection form:

\begin{matrix} \vec{\nabla} = \nabla + ρ (X) + ρ P (X) . \end{matrix}

In the case of the Tractor bundle

T

, the detailed expression is:

\begin{matrix} {\vec{\nabla}}_{X} (\begin{matrix} x \\ Y \\ z \end{matrix}) = (\begin{matrix} \nabla_{X} x - g (X, Y) \\ \nabla_{X} Y + z X - x P (X) \\ \nabla_{X} z + P (X, Y) \end{matrix}), \end{matrix}

(4)

Now we get to the result that ties all the structures together:

Theorem 2.13. The connection

\vec{\nabla}

is a normal Tractor connection.

Proof. Let

i

be the inclusion

i : P ↪ G

π

projection

π : P \to M

, and let

ω \in Ω^{1} (G, g)

be the one-form associated with the connection

\vec{\nabla}

We need to prove that

μ = i^{*} (ω)

is an isomorphism

T P_{u} \to g

for all points

u \in P

; then

μ

will be the Cartan connection generating the Tractor connection

ω

Reasoning by contradiction, assume that

μ

is not an isomorphism at some point

u

, so there exists a vector

ξ \in T P_{u}

such that

μ (ξ) = 0

. As vertical vectors in

P

are mapped isomorphically onto

p

X = π_{*} (ξ)

is a non zero vector in

T_{π (u)}

Then define a local section

j

M

P

, such that

j_{*} X = ξ

. This also gives us a section

i \circ j

G

. Then in the frame bundle determined by this section, the connection

\vec{\nabla}

is of the form

\begin{matrix} d + j^{*} i^{*} ω = d + j^{*} μ \end{matrix}

This shows that the bundle

P

is infinitesimally conserved at

π (u)

in the

X

direction, or, switching to the associated bundle

E = P \times_{P} e

, that there is a section

s

E

, non-zero at

π (u)

, such that

{\vec{\nabla}}_{X} s = 0

π (u)

However, the connection on the tractor bundle is given by equation ( 4 ):

\begin{matrix} {\vec{\nabla}}_{X} (\begin{matrix} 0 \\ 0 \\ z \end{matrix}) = (\begin{matrix} 0 \\ z X \\ \nabla_{X} z \end{matrix}), \end{matrix}

which is a contradiction as

s

(hence

z

) is non-zero at

π (u)

. So

\vec{\nabla}

is indeed a Tractor connection.

And finally, to complete the circle:

Lemma 2.14. The Cartan connection generated by

\vec{\nabla}

is normal.

Proof. In order to demonstrate this result, it suffices to show that the curvature of

\vec{\nabla}

lies in the Lie algebra bundle of

p

or, equivalently, that it preserves the canonical bundle

E

In abstract index notation, the expression for

{\vec{\nabla}}_{i} {\vec{\nabla}}_{j}

is:

\begin{matrix} {\vec{\nabla}}_{i} {\vec{\nabla}}_{j} & = & \nabla_{i} \nabla_{j} + (δ_{j}^{l} \nabla_{i} + δ_{i}^{l} \nabla_{j}) + (P_{j k} \nabla_{i} + P_{i k} \nabla_{j}) + \nabla_{i} (P_{j k}) \end{matrix}

\begin{matrix} + δ_{i}^{l} \circ δ_{j}^{m} + δ_{i}^{l} \circ P_{j n} + P_{i k} \circ δ_{j}^{m} + P_{i k} \circ P_{j n} . \end{matrix}

Here we have used the connection

\nabla

T

to define the second covariant derivative; however, we could have used any other connection, as we are about to anti-symmetrise

i

and

j

. Upon doing this, the terms in brackets vanish.

Moreover,

ρ (X) \circ ρ (Y) = ρ (Y) \circ ρ (X)

and similarly for one-forms, meaning that:

\begin{matrix} R_{i j}^{\vec{\nabla}} & = & R_{i j}^{\nabla} + \nabla_{[i} (P_{j] k}) + δ_{[i}^{l} \circ P_{j] n} + P_{k [i} \circ δ_{j]}^{m} . \end{matrix}

Looking back at equations 1

and 2

, we see that this expression is the sum of the Weyl tensor and the Cotton York one. Or, expressed in more conventional notation, in the splitting of

A

determined by

\nabla

\begin{matrix} R_{X, Y}^{\vec{\nabla}} = (\begin{matrix} 0 \\ W (X, Y) \\ C Y (X, Y) \end{matrix}) . \end{matrix}

This curvature expression obviously preserves the canonical bundle

E

□

□

3 Conformally Einstein Manifolds

3.1 Important Note

In most of the proofs in the remainder of this paper, it will be assumed that for a certain holonomy preserved sub-bundle

U \subset T

used in the proof, one has

π^{2} (U) \neq 0

. This will not be the case everywhere, of course; however:

Proposition 3.1. Let

U

be a holonomy preserved sub-bundle. Then

π^{2} (U) \neq 0

Σ

, an open, dense subset of

M

Proof. $Σ$ is open because of the $π^{2} (U) \neq 0$ condition.
Let $b \in M \ Σ$ , and $u (b) = (0, Y (b), z (b))$ be a non-zero element of $U_{b}$ . Then extend $u (b)$ locally to a section $u = (x, Y, z)$ of $U$ by parallel transport along `rays' from $b$ . This implies that $\vec{\nabla} u = 0$ at $b$ . Then picking any nowhere-zero section $τ$ of $ℰ [- 1]$ , we can define the function $f : M \to R^{n}$ by
$\begin{matrix} f (c) = π^{2} (u) τ . \end{matrix}$
The derivative of $f$ is $(\nabla x) τ + x (\nabla τ)$ . At $b$ , this is just $(\nabla x) τ$ , and, since $\vec{\nabla} u = 0$ at $b$ :
$\begin{matrix} X . f (b) = (\nabla_{X} x) τ = - g (X, Y) τ . \end{matrix}$
If $Y \neq 0$ , this is non-zero for some $X$ , so $f$ is non-zero arbitrarily close to $b$ .
If $Y = 0$ , then the first derivative is zero, and the second derivative is thus:
$\begin{matrix} Z . (X . f) (b) & = & (\nabla_{Z} \nabla_{X} x) τ \end{matrix}$

$\begin{matrix} = & - g (X, \nabla_{Z} Y) τ \end{matrix}$

$\begin{matrix} = & - g (X, - z Z) τ . \end{matrix}$
with $z (b) \neq 0$ as $u (b) \neq 0$ . Then the second derivative is non-zero for $X = Z \neq 0$ , for instance, forcing $f$ to be non-zero arbitrarily close to $b$ .
This implies that $π^{2} (u) \neq 0$ arbitrarily close to $b$ , proving the result. $□$

In fact, if the first derivative vanishes,

b

must be an isolated point.

The classic examples of this are the various conformally Einstein metrics on the sphere

S^{n}

. The sphere is conformally flat, so there are many holonomy preserved sections of its Tractor bundle.

A preserved section

u

of negative norm corresponds to the Spherical metric

g = π^{2} (u)^{- 2} g

on the whole space. In this case,

π^{2} (u)

is never zero.

A preserved section

u

of zero norm corresponds to the Euclidean metric

g = π^{2} (u)^{- 2} g

R^{n} \sim = S^{n} \ {\infty}

. In this case,

π^{2} (u) (b) \neq 0

for

b \neq \infty

A preserved section

u

of positive norm corresponds to the Hyperbolic metric

g = π^{2} (u)^{- 2} g

on two half spheres of

S^{n}

. In this case

π^{2} (u)

is zero only on the boundary

S^{n - 1}

cutting

S^{n}

into two.

3.2 Einstein Spaces

Though it is well known in general that any conformally Einstein space corresponds to a parallel section of the tractor bundle

T

, what follows is a direct proof of this fact using the Tractor connection approach.

Remark. This is a first instance of a holonomy reduction of

\vec{\nabla}

Theorem 3.2. If

(M, g)

has an Einstein metric

g

in its conformal class then there exists a parallel section

s

of its tractor bundle

T

Proof. Let

g

be the Einstein metric,

R i c^{g} = λ g

. Then the

P

-tensor is

\begin{matrix} P = - \frac{λ}{2 n - 2} g . \end{matrix}

Hence

\begin{matrix} \vec{\nabla} (\begin{matrix} 1 \\ 0 \\ - \frac{λ}{2 n - 2} \end{matrix}) = (\begin{matrix} 0 \\ \frac{λ}{2 n - 2} - \frac{λ}{2 n - 2} \\ 0 \end{matrix}) = 0, \end{matrix}

where

1

is the section of

ℰ [1]

corresponding to

g

□

To prove the converse of this theorem, we need the following lemma:

Lemma 3.3. If a conformal connection

\nabla

has a symmetric Ricci tensor, then

\nabla

is actually a metric connection.

Proof. Let

R_{j k l}^{i}

be the curvature of

\nabla

. Then

R_{j k l}^{i}

acts on the determinant bundle

ℰ [n]

via its trace

R_{i k l}^{i}

. However, by the first Bianci identity,

\begin{matrix} R_{i k l}^{i} & = & - R_{k l i}^{i} - R_{l i k}^{i} \end{matrix}

\begin{matrix} = & R i c_{k l} - R i c_{l k}, \end{matrix}

the anti-symmetric part of the Ricci tensor. So if

\nabla

has a symmetric Ricci tensor, its curvature must vanish on

ℰ [n]

, so locally

\nabla

must preserve a section

η

of the determinant bundle. Then

η^{1 / n}

is a preserved conformal scale and

\begin{matrix} g = η^{- 2 / n} g \end{matrix}

a metric preserved by

\nabla

□

Theorem 3.4. If a line bundle

L

T

is holonomy preserved, then a section

s

L

is preserved, and

(M, [g])

has an Einstein metric

g = (π^{2} (s))^{- 2} g

in its conformal class, wherever

π^{2} (s) \neq 0

Proof. The line bundle $L$ defines a connection on $ℰ [1]$ , and hence a torsion free connection on $T$ , in the following way. Let $e$ be any section of $ℰ [1]$ , and let $l$ be the section of $L$ such that $π^{2} (l) = e$ . Then define $\nabla e = π^{2} (\vec{\nabla} l)$ ; it is easy to see that this is indeed a connection. A connection on $ℰ [1]$ naturally defines a torsion-free $G_{0}$ connection on $T$ using the conformal metric $g$ and the same formula as for the Levi-Civita connection.
Using $\nabla$ , we split $T = ℰ [1] \oplus T [- 1] \oplus ℰ [- 1]$ in the usual way. Then equation ( 4 ) implies that $π^{2} (\vec{\nabla} (x, Y, z)) = \nabla x - g (Y, -)$ . Thus, by the very definition of $\nabla$ , $ξ = 0$ for any section of $L$ .
If $L$ is not null, then a section $s = (x, 0, x μ)$ of constant norm, is preserved.
This generates a metric $g = (π^{2} (s))^{- 2} g$ .
If $L$ is null, then $z = 0$ .
But ${\vec{\nabla}}_{X} (x, 0, z) = (\nabla_{X} x, z (X) - x P (X), \nabla_{X} z)$ .
Consequently, if $L$ is not null, then $P = μ g$ , so $R i c^{g} = λ g$ for $λ = (2 - 2 n) μ$ .
Alternatively, if $L$ is null, then $P = 0$ . Hence $\nabla$ has a symmetric Ricci tensor, implying that it is actually a metric connection. Let $x \in Γ (ℰ [1])$ be the section corresponding to this metric. Then the section $s = (x, 0, 0)$ , is parallel.
Finally, the metric $g = (π^{2} (s))^{- 2} g$ is Ricci-flat. $□$

Remark. Note that the sign of

〈 s, s 〉

is the opposite of the sign of the Einstein constant

λ

4 Decomposition Theorem

This section presents the decomposition theorem for Tractor connections, similar to the De Rham decomposition for Riemannian connections.

Remark. Related terminology may be found in [18 ]

The theorem to be proved is the following:

Theorem 4.1. Let

(M, [g])

be a conformal,

n

-dimensional manifold, such that

T_{M}

has a holonomy-preserved sub-bundle of rank

k

. Then there exists a metric

g \in [g]

such that

(M, g)

splits locally into the direct product of two Einstein manifolds

M_{U}

M_{V}

of dimensions

l = k - 1

and

n - l

. The Einstein constants

a

and

b

M_{U}

M_{V}

are related by

(n - l - 1) a = (1 - l) b

. Furthermore, the Tractor holonomy group of

M

is the direct product of those of

M_{U}

and

M_{V}

This theorem also has a converse.

4.1 Preparatory Results

The contents of this section deal with various geometric results for submanifolds that will be needed in the second part of this chapter. The proofs are technical geometric manipulations.

Let

(M, g)

be a Riemannian manifold, with submanifold

N

, and let

\nabla^{g}

be the Levi-Civita connection of

g

The second fundamental form on

N

is a section of

T N^{*} \otimes T N^{*} \otimes T N^{⊥}

, defined as

S (X, Y) = (\nabla_{X} Y)^{⊥}

for

X

and

Y

sections of

T N

. As

\nabla^{g}

is torsion free,

S

is symmetric.

The submanifold

N

is called minimal if

S

is trace free. It is called umbilical if

S = g H

, for

H

a section of

T N^{⊥}

. It is totally geodesic if

S = 0

. Of these three, only umbilical is a conformally invariant concept.

Proposition 4.2. Given a submanifold

N

, there is a canonical inclusion of the tractor bundle of

N

i : T_{N} ↪ T |_{N}

Proof. First of all, we will need the following lemma:

Lemma 4.3. Given any point

b \in N

, there exists a metric

g^{'} \in [g]

such that

N

is minimal at

b

Proof. Let

H =

Trace

S \in Γ (T N^{⊥})

. Then remember that for

f

a change of conformal weight,

g^{'} = f^{2} g

, the Levi-Civita connection changes as:

\begin{matrix} \nabla_{X}^{g^{'}} Y & = & \nabla_{X}^{g} Y + [[ϒ, X], Y] \end{matrix}

\begin{matrix} = & \nabla_{X}^{g} Y + ϒ (X) Y + ϒ (Y) X - g (X, Y) ϒ^{*}, \end{matrix}

where

ϒ = f^{- 1} d f

and

ϒ^{*}

is the dual to

ϒ

under

g

. This implies that the second fundamental form changes as

\begin{matrix} S^{g^{'}} = S^{g} - (ϒ^{*})^{⊥} g . \end{matrix}

Hence it suffices to find an

f

such that

ϒ^{*} = H

b

to make

N

minimal at

b

To construct such an

f

, pick local coordinates

(y^{i}, x^{j})

such that

N = {c \in M | x^{j} (c) = 0}

, and the submanifold

{c \in M | y^{i} (c) = 0}

is orthogonal to

N

b

Then

H_{b}^{*} = e_{j} d x^{j}

for some real numbers

e_{j}

. Then define

\begin{matrix} f = 1 + σ e_{j} x^{j}, \end{matrix}

for

σ

a bump function around

b

. This gives

ϒ_{b} = f^{- 1} d f = 1 . e_{j} d x^{j} = H_{b}^{*}

, proving the lemma.

□

Back to the proof of the main proposition. Given any

g \in [g]

that makes

N

minimal at

b

, the connection

\nabla^{g}

splits

T |_{b} = ℰ [- 1] \oplus T [- 1] \oplus ℰ [1]

Similarly, for

h = g |_{T N}

, the connection

\nabla^{h}

N

splits

T_{N} |_{b} = ℰ [- 1] \oplus T N [- 1] \oplus ℰ [1]

. Then the inclusion

T N \subset T

defines an inclusion

i_{b} : T_{N} ↪ T |_{N}

. It remains to verify that this inclusion is well defined.

Let

g^{'} = f^{2} g

be another metric that makes

N

minimal at

b

. This implies that at

b

ϒ_{f} \in T N_{b}^{*}

. Then remembering the formula for the change of splitting:

\begin{matrix} (\begin{matrix} x \\ Y \\ z \end{matrix}) \to (\begin{matrix} x \\ Y - ϒ_{f} x \\ z + ϒ_{f} (Y) - \frac{1}{2} g (ϒ_{f}, ϒ_{f}) x \end{matrix}) \end{matrix}

and as

h^{'} = f^{2} h

, this implies that the two splittings change in exactly the same way, hence the inclusion

i_{b}

is natural. As

i_{b}

is obviously smooth for varying

b

, we have a natural inclusion

i : T_{N} ↪ T |_{N}

□

Note that it is still important to distinguish

T_{N}

and

i (T_{N})

, as, for instance, they generally behave very differently under their various tractor connections.

To get any further, we need to place some restrictions on

N

Definition 4.4. A manifold

M

admits a crossfoliation if it admits two separate foliations, with tangent bundles

E_{1}

and

E_{2}

, such that

- $E_{1} \oplus E_{2} = T,$
- $E_{1} \cap E_{2} = 0 .$

If the two sets of leaves are umbilical, we call this crossfoliation umbilical. If

E_{1}

and

E_{2}

are orthogonal, they form an orthogonal crossfoliation.

Definition 4.5. A metric

g

on a manifold

M

with an orthogonal crossfoliation is said to be split at a point

b

, if, for

N

and

L

the two leaves intersecting orthogonally at

b

and

h = g |_{T N}

h^{'} = g |_{T L}

we have:

\begin{matrix} \nabla^{g} h = \nabla^{g} h^{'} = 0 . \end{matrix}

This last definition is equivalent with the fact that

\begin{matrix} \nabla^{g} (T N) \subset T^{*} \otimes T N and \nabla^{g} (T L) \subset T^{*} \otimes T L . \end{matrix}

This gives us the main result of this section:

Lemma 4.6 (Metric Splitting). Let the submanifold

N

be a leaf in an umbilical orthogonal crossfoliation, and let

b

be a point in

N

. Then there exists a metric

g \in [g]

such that

N

is totally geodesic in

(M, g)

, and

g

splits at

b

Proof. Pick local coordinates $(y^{i}, x^{j})$ so that $y^{i} = c o n s t$ and $x^{j} = c o n s t$ are the leaves of the foliation, with $N$ being the leaf for $x^{j} = 0$ . Then as $N$ is totally umbilical, $S^{g} = g H$ , for some vector field $H \in Γ (T N^{⊥})$ . Thus $H^{*} = e_{j} d x^{j}$ , with $e_{j}$ functions of $y_{j}$ only.
Then, as in the previous theorem, defining $g^{'} = f^{2} g$ for $f = 1 + e_{j} x^{j}$ gives us a new second fundamental form $S^{g^{'}} = S^{g} - g ϒ^{*} = g H - g H = 0$ . Hence $N$ is totally geodesic.
Then given any point $b \in N$ , shift the $y^{i}$ by a constant so that $(y^{i}, x^{j}) = 0$ at $b$ . Then let $L = {c \in M | y^{i} (c) = 0}$ be (locally) the orthogonal submanifold to $N$ at $b$ . Then, as $L$ is itself umbilical by definition, $S^{L} = g K$ for $K = e_{i} d y^{i}$ as $K$ is in $T L^{⊥} \sim = T N$ at $b$ . Thus we may change $g^{'}$ by a function $f = 1 + e_{i} (b) y^{i}$ ; this makes $L$ totally geodesic at $b$ , and since $ϒ (f) = f^{- 1} e_{i} (b) d y^{i}$ is a section of $T N^{*}$ , $N$ remains totally geodesic.
Then with $N$ totally geodesic at $b$ , for any section $X$ of $T N$ , $\nabla_{X}^{g} (T N) \subset T N$ and $\nabla_{X}^{g} (T L) \subset T L$ , as $\nabla^{g}$ is metric. As $L$ is also totally geodesic at $b$ , the previous formulas are also true for $X$ a section of $T L$ , hence for any $X$ at $b$ . $□$

Remark. It is possible to get much stronger results from the conditions of the lemma; however, they will not be needed, and will drop out quite naturally from later considerations.

Theorem 4.7. If

N

is a leaf of an umbilical orthogonal crossfoliation, then the sub-bundle

i (T_{N}) \subset T |_{N}

is conserved under parallel transport along

N

Proof. Using the previous lemma ( 4.6 ), let us pick

g

so that

N

is totally geodesic and

g

splits at

b

. Then

N

is a fortiori minimal at each point, so

T_{N} \subset T

is given in the splitting

\nabla^{g}

{(x, Y, z) | Y \in Γ (T N)}

Then for

X \in Γ (T N)

\begin{matrix} {\vec{\nabla}}_{X} (\begin{matrix} x \\ Y \\ z \end{matrix}) = (\begin{matrix} \nabla_{X} x - g (X, Y) \\ \nabla_{X} Y + z X - x P (X) \\ \nabla_{X} z + P (X, Y) \end{matrix}), \end{matrix}

where the splitting of

g

ensures that that

\nabla_{X} Y \in T N_{b}

b

. This metric splitting further makes the curvature and Ricci tensors split at

b

R i c_{M} = R i c_{N} + R i c_{N^{⊥}}

. Then the

P

-tensor also splits, implying that

P (X)_{b} \in T N_{b}

In consequence,

{\vec{\nabla}}_{X} (T_{N}) \subset T_{N}

b

, and as

b

is arbitrary, on all points of

N

□

Remark. Note that this does not imply that the Tractor connection

{\vec{\nabla}}^{M}

acts on

i (T_{N})

the same way as

{\vec{\nabla}}^{N}

acts on

T_{N}

. Indeed, though

P

splits into

T N

and

T N^{⊥}

components, it is not generally equal to

P^{N} + P^{N^{⊥}}

, the various coefficients being different, dependent as they are on the dimensions of the different (sub)manifolds.

4.2 Decomposition Theorem

Let

(M, [g])

be a conformal manifold of dimension

n

, with tractor bundle

T

, and let

K

of rank

k

be a holonomy preserved sub-bundle.

Let

E = {(0, 0, a)} \subset T

be the canonical null-line, and assume furthermore that

K \cap E = 0

and

K^{⊥} \cap E = 0

(this will always be the case on an open, dense subset of

M

, by Proposition 3.1 ).

Let

S = E^{⊥}

be the kernel of

π^{2} : T \to ℰ [1]

. So

S = {(0, Y^{i}, z)} \subset T

, in any splitting we can choose.

Then define

\tilde{U} = S \cap K

, and

\tilde{V} = S \cap K^{⊥}

The conditions on

K

and

K^{⊥}

imply that neither of these two bundles is contained in

S

, since

S = E^{⊥}

. As rank

S = (

rank

T) - 1

, we get rank

\tilde{U} = k - 1

and rank

\tilde{V} = n + 2 - k - 1 = n - (k - 1)

Lemma 4.8.

\tilde{U} \cap \tilde{V} = 0

Proof. As $\tilde{U} ⊥ \tilde{V}$ , then any vector in $\tilde{U} \cap \tilde{V} = 0$ is null.
The tractor metric on $S$ is given by $h ((0, Y, z), (0, Y^{'}, z^{'})) = g (Y, Y^{'})$ . Hence the only null vectors in $S$ lie in $E$ . Thus as $K \cap E = 0$ , we must have $\tilde{U} \cap \tilde{V} = 0$ . $□$

Lemma 4.9.

(\tilde{U} \oplus \tilde{V}) \cap E = 0

Proof. Let $a + b \in Γ (E)$ for $a \in Γ (\tilde{U})$ and $b \in Γ (\tilde{V})$ . Then $a = (0, Y, z)$ and $b = (0, - Y, z^{'})$ . Then as $0 = h (a, b) = - g (Y, Y)$ , implying that $Y = 0$ , hence $a, b \in Γ (E)$ , implying that $a, b = 0$ and proving the lemma. $□$

Remark. This lemma only works in the case of definite signature, and is the reason that the decomposition theorem cannot be extended to the indefinite case.

These two results together prove:

Proposition 4.10. As

E

is the kernel of

π : S \to T [- 1]

π

maps

\tilde{U} \oplus \tilde{V}

injectively onto

T [- 1]

. As scale is irrelevant, let

T [- 1] \sim = T

, via any

g

Let

U = π \tilde{U}

and

V = π \tilde{V}

. Then

U^{⊥} = V

, and

U \oplus V = T

Theorem 4.11.

U

and

V

are both integrable, giving an orthogonal foliation. Moreover, this foliation is umbilical.

Proof. Let the

(X_{i})

be an orthogonal frame of

U

. Pick any metric

g \in g

Let

π^{- 1} X_{j} = {\tilde{X}}_{j} = (0, X_{j}, z) \in Γ (\tilde{U})

Then

T = {\vec{\nabla}}_{X_{i}} {\tilde{X}}_{j} = (0, \nabla_{X_{i}}^{g} X_{j} + z X_{i}, q)

for some

q

. As

K

is conserved under holonomy,

T \in Γ (K)

. As the first component is zero,

T

is a section of

S

, hence

T \in Γ (\tilde{U})

. Thus

\nabla_{X_{i}}^{g} X_{j} + z X_{i}

is a section of

U

, and hence so is

\nabla_{X_{i}}^{g} X_{j}

So for

g (X_{i}, X_{j}) = 0

\nabla_{X_{i}}^{g} X_{j} \in Γ (U)

, and, a fortiori,

[X_{i}, X_{j}] \in Γ (U)

. Since

(X_{i})

are an orthogonal frame, both these properties generalize to arbitrary sections of

U

. Thus

U

is integrable.

Lemma 4.12. The corresponding leaves of

U

are umbilical.

Proof. Let

(X_{i})

be an orthonormal frame in the metric

g

. We know that

\nabla_{X} Y \in Γ (U)

for

g (X, Y) = 0

, so

S^{g} (X, Y) = 0

where

S^{g}

is the second fundamental form of

g

Now

g (X_{i} + X_{j}, X_{i} - X_{j}) = 0

, implying that

\begin{matrix} 0 = S^{g} (X_{i} + X_{j}, X_{i} - X_{j}) = S^{g} (X_{i}, X_{i}) - S^{g} (X_{j}, X_{j}) . \end{matrix}

Hence

S^{g} (X_{i}, X_{j}) = g (X_{i}, X_{j}) H

for

H = S^{g} (X_{1}, X_{1})

. As both

S^{g}

and

g

are bilinear, this equality extends to any sections of

U

, making the leaves umbilical.

□

The same proof works for

V

□

Let

M_{U}

and

M_{V}

be these leaves. As we are in the orthogonal umbilical crossfoliation case, Theorem 4.7 shows that there are sub-tractor bundles

T_{U}

and

T_{V}

that are conserved along

M_{U}

and

M_{V}

respectively.

The rank of

T_{U}

k - 1 + 2 = k + 1

, hence has a non-zero intersection with

K^{⊥}

. This sub-bundle is hence conserved along

M_{U}

. The same is true for

T_{V}

and

K

along

M_{V}

Proposition 4.13.

T_{U} \cap K^{⊥}

is a line bundle.

Proof. In this case, we need to use a specific metric. Given a leaf $M_{U}$ , let $g \in [g]$ be a metric such that $M_{U}$ is totally geodesic in $(M, g)$ .
Then along $M_{U}$ , in the splitting given by $g$ , $T_{U} = {(x, Y, z) | Y \in Γ (U)}$ .
Reasoning by contradiction, assume there exist $a, b$ , sections of $T_{U} \cap K^{⊥}$ , linearly independent. Then in the span of $a$ and $b$ there exists $c \in Γ (S)$ . Hence $c \in Γ (\tilde{V})$ , so $c = (0, A, z)$ , with $A \in Γ (V)$ . But since $c$ is in $T_{U}$ as well, this gives $A = Y = 0$ . So $c \in Γ (E)$ , which, by the definition of $K^{⊥}$ , is a contradiction. $□$

The same proof shows that

T_{V} \cap K

is a line bundle as well.

4.2.1 Split Metrics in the Conformal Class

Let

I = T_{U} \cap K^{⊥}

and

J = T_{V} \cap K

. Then

I

and

J

are orthogonal line bundles, and we are going to use them to construct a special metric on

(M, [g])

. From now on, let

X, Y, Z

be sections of

U

A, B, C

sections of

V

For

b \in M

, let

M_{U}, M_{V}

be the leaves that cross at

b

. Then choose a metric

g

that splits at

b

(Lemma 4.6 ).

Hence at

b

, in the splitting of

T

defined by

g

T_{U} = {(x, Y, z) | Y \in U_{b}}

and

T_{V} = {(a, B, c) | B \in V_{b}}

Lemma 4.14. In this splitting,

\tilde{U} = (0, U, 0)

and

\tilde{V} = (0, V, 0)

b

Proof. Let $s = (0, X, z)$ be a section of $\tilde{U}$ . Then for some $c$ , ${\vec{\nabla}}_{A} s = (0, \nabla_{A}^{g} X + z A, c)$ is a section of $\tilde{U}$ .
So $\nabla_{A}^{g} X + z A$ is a section of $U$ . However, at $b$ , $\nabla_{A}^{g} X$ is a section of $U$ as well; hence $z = 0$ at $b$ .
The same proof implies the same result for $\tilde{V}$ . $□$

Lemma 4.15. In this splitting, at

b

I = {(a, 0, λ a)}

for some

λ

, and

J = {(a, 0, - λ a)}

Proof. At $b$ , $I = T_{U} \cap K^{⊥} = (x, U, z) \cap (0, V, 0)$ . Hence, as $I$ is a line bundle, $I = {(a, 0, λ_{1} a)}$ . The same proof shows that $J = {(a, 0, λ_{2} a)}$ . Moreover, $I$ and $J$ are orthogonal, hence $λ_{1} = - λ_{2}$ . $□$

We notice two things at

b

: firstly that

π (I) = π (J)

. Secondly for sections

i, j

of these bundles,

π^{2} (i) = π^{2} (j)

implies that

h (i, i) = - h (j, j)

I

is non-null, then the second implication is an equivalence.

However,

b

is a generic point on the manifold, and these two properties are conformally invariant; hence they are true in general, over the whole manifold.

I

is null, then

I = J

and

I

is holonomy preserved over the whole manifold.

Then by Theorem 3.4 , there exists a Ricci-flat metric

\hat{g} \in [g]

Conversely, let

I

(and hence

J

) be non-null. Without loss of generality, let

I

be the positive definite one of the two. Then for any constant

c > 0

, the section

i

such that

h (i, i) = 2 c

, defines a metric

\hat{g} \in [g]

on the manifold via

π^{2} (i)

Lemma 4.16. The metric

\hat{g}

splits at every point of the manifold.

Proof. In the splitting defined by $\hat{g}$ , $i = (1, H, z)$ . Then there is a corresponding section $j = (1, H, y)$ of $J$ , with $i$ covariantly constant along $U$ , and $j$ covariantly constant along $V$ . However, the first component of ${\vec{\nabla}}_{Y} i$ is $- \hat{g} (Y, H)$ , and the first component of ${\vec{\nabla}}_{A} j$ is $- \hat{g} (A, H)$ . Hence $H$ must be zero, and thus $z = c$ , and $y = - c$ if $I$ is non-null, and $z = 0$ in the null-case.
Letting $\tilde{X} = (0, X, α)$ be a section of $\tilde{U}$ , then ${\vec{\nabla}}_{Y} \tilde{X} = (- \hat{g} (Y, X), \nabla_{Y}^{\hat{g}} X + α Y, β)$ is a section of $K$ . Hence ${\vec{\nabla}}_{Y} \tilde{X} + j \hat{g} (Y, X) = (0, \nabla_{Y}^{\hat{g}} X + α Y, γ)$ is a section of $\tilde{U}$ , making $\nabla_{Y}^{\hat{g}} X$ into a section of $U$ . The metric nature of $\nabla^{\hat{g}}$ then implies that $\nabla_{Y}^{\hat{g}} A$ is a section of $V$ .
The same relationship applied to sections of $\tilde{V}$ completes the proof. $□$

Theorem 4.17. In

(M, \hat{g})

, the leaves

M_{U}

and

M_{V}

are totally geodesic, and hence,

(M, \hat{g})

is a direct product.

Proof. By the previous lemma, $S_{U}^{\hat{g}} (X, Y)$ and $S_{V}^{\hat{g}} (A, B)$ are both zero. $□$

Theorem 4.18. If

h_{U} = \hat{g} |_{U}

, then the leaf

(M_{U}, h_{U})

is Einstein. The same result holds for the

(M_{V}, h_{V})

leaves. Furthermore, the two Einstein constants

a

and

b

are related by

(n - l - 1) a = (1 - l) b

, where

l = k - 1

Proof. Along

M_{U}

0 = {\vec{\nabla}}_{X} i = (0, c X - P (X), 0)

along

U

. Similarly,

0 = {\vec{\nabla}}_{A} j = (0, - c Y - P (Y), 0)

along

V

, so:

\begin{matrix} P = c (h_{U} - h_{V}) . \end{matrix}

This result is also valid in the null case, with

c = 0

But

\begin{matrix} P = - \frac{1}{n - 2} (R i c - \frac{1}{2 n - 2} R g), \end{matrix}

and

R i c

splits as

R i c^{U} + R i c^{V}

, and

R = R^{U} + R^{V}

Let

l = k - 1 =

dim

M_{U}

. Then there exist constants

a

and

b

such that

R i c^{U} = a (h_{U})

and

R i c^{V} = b (h_{V})

, and

R = a l + b (n - l)

This gives the simultaneous equations:

\begin{matrix} c + \frac{1}{n - 2} (a - \frac{a l + b (n - l)}{2 n - 2}) = 0 \end{matrix}

\begin{matrix} - c + \frac{1}{n - 2} (b - \frac{a l + b (n - l)}{2 n - 2}) = 0 \end{matrix}

Solving these, and remembering that we can freely rescale

c

\begin{matrix} a & = & (1 - l) c \end{matrix}

\begin{matrix} b & = & (n - 1 - l) c . \end{matrix}

M_{U}

is non-positive Einstein, and

M_{V}

is non-negative Einstein.

□

Finally, to complete the decomposition result, one needs,

Theorem 4.19. The tractor holonomy of

(M, [g])

is the direct product of those of

(M_{U}, [h_{U}])

and

(M_{V}, [h_{V}])

Proof. The tractor expression on

M

in the split metric is

\begin{matrix} {\vec{\nabla}}_{X} = \nabla_{X} + ρ (X) + ρ (P^{M} (X)) . \end{matrix}

Since

M_{U}

is totally geodesic,

\nabla^{h_{U}} = \nabla |_{U}

. Thus the tractor expression on

M_{U}

is the same as above, with

P^{M}

replaced by

P^{U}

. Then the result is proved along

M_{U}

by the following lemma:

Lemma 4.20. In the circumstances as above,

P^{U}

P^{M} |_{U}

Proof. By direct calculation. As

a

is the Einstein constant on

M_{U}

, a manifold of dimension

l

\begin{matrix} P^{U} = - \frac{a}{2 l - 2} h_{U} . \end{matrix}

On the other hand, the scalar curvature of

M

R^{M} = l a + (n - l) b

, so

\begin{matrix} P^{M} |_{U} & = & - \frac{1}{n - 2} (a - \frac{l a + (n - l) b}{2 n - 2}) h_{U} \end{matrix}

\begin{matrix} = & - \frac{a}{2 l - 2} h_{U} = P^{U}, \end{matrix}

by the relation between

a

and

b

□

If any of the leaves are oneor two-dimensional, the situation is different, as there is no intrinsic tractor connection on the tractor bundle

T_{U}

. The restriction of

\vec{\nabla}

i (T_{U})

, along

U

, does define a restricted Tractor connection however. If the leaf

M_{U}

is one dimensional, then this connection must be flat. If not . . .

Lemma 4.21. If the leaf

M_{U}

is two dimensional, then the restricted connection on it is flat.

Proof. By the relation between the Einstein structures,

P^{M} |_{U} = - \frac{a}{2} h_{U}

. Then the curvature of

\vec{\nabla}

along

U

is:

\begin{matrix} {\vec{R}}_{X Y} = (\begin{matrix} 0 \\ R_{X Y}^{U} - a (X \circ h_{U} (Y) - Y \circ h_{U} (X)) \\ 0 \end{matrix}), \end{matrix}

where

R^{U}

is the metric curvature of

M_{U}

But we know that

R i c^{U} = a h_{U}

, and since

R^{U}

takes values in

Λ^{2} T \otimes Λ^{2} T

, a rank one bundle, for the trace of

R^{U}

to be what we have found,

R^{U}

itself must be

\begin{matrix} R_{X Y}^{U} = a (X \circ h_{U} (Y) - Y \circ h_{U} (X)) \end{matrix}

So the curvature of

\vec{\nabla}

restricted to

U

, is zero.

□

Then to prove the whole theorem, one needs only to note the next obvious result:

Lemma 4.22. The tractor connection on

M

\vec{\nabla}

, makes

i (T_{U})

parallel along

V

Proof. Let

s = (x, Y, z)

and

A

be sections of

i (T_{U})

and

M_{V}

respectively, extended to

M

trivially. Then

\nabla_{A} s = 0

s

is trivially extended, and since the eigenspaces of

P^{M}

are orthogonal,

g (A, Y) = P (A, Y) = 0

. So

\begin{matrix} {\vec{\nabla}}_{A} s = (\nabla_{A} + ρ (A) + ρ P (A)) s = 0 \end{matrix}

□

Obviously, the same result holds for

M_{V}

as well, so the holonomy splits.

□

Remark. As is well known, if an Einstein manifold is three dimensional, it is conformally flat. Adding this to the flatness results along one or two dimensional leaves, it is revealed that groups of the form

S O (2, 1)

S O (3, 1)

and

S O (4, 1)

, can never appear as elements of a group product in the holonomy of any conformal manifold.

4.2.2 Converse to the Decomposition

Theorem 4.23. Let

M

be locally a direct product of the Einstein manifolds

(M_{U}, h_{U})

and

(M_{V}, h_{V})

Furthermore, let the Einstein constants

a, b

M_{U}

and

M_{V}

obey the relations

(n - l - 1) a = (1 - l) b

. Then the tractor bundle

T

(M, [h_{U} + h_{V}])

has a holonomy-conserved sub-bundle

K

Proof. Let U = T U and V = T V , and let's work in the h U + h V metric. Then since the foliation is totally geodesic, T U ⊂ T is of the form { ( x , Y , z ) | Y ∈ Γ ( U ) } , and the converse holds for T V
.
Theorem 4.18 reverses, giving P = c ( h U − h V ) for some c . This gives a section i = ( 1 , 0 , c ) ∈ T U , covariantly constant along U and j = ( 1 , 0 , − c ) ∈ T V , covariantly constant along V . Note that these definitions work in the Ricci-flat case as well, for c = 0 and using theorem 3.4 . Define I = span ( i ) , J = span ( j )
.
Moreover, let $\tilde{V} = (T_{U})^{⊥} \subset T$ , and $\tilde{U} = (T_{V})^{⊥}$ ; then in this metric, $\tilde{U} = {(0, Y, 0) | Y \in Γ (U)}$ , and similarly for $\tilde{V}$ .
Finally, let $K = \tilde{U} \oplus J$ and $L = \tilde{V} \oplus I$ .
Lemma 4.24. $L = K^{⊥}$ .

Proof. $\tilde{U}$ and $\tilde{V}$ are certainly orthogonal. $J$ is orthogonal to $\tilde{V}$ and $I$ to $\tilde{U}$ by components, and they are mutually orthogonal as well. Thus $L \subset K^{⊥}$ and a consideration of rank proves equality. $□$

Now to the main theorem. $K^{⊥}$ is conserved along $U$ , as both $I$ and $\tilde{V} = T_{U}^{⊥}$ are. As $\vec{\nabla}$ is pseudo-metric, $K$ is thus conserved along $U$ as well. Reversing the argument, we see that $K$ , and hence $K^{⊥}$ are conserved along $V$ , consequently along any path. $□$

A final technicality exists, for the case of oneor two-dimensional manifolds, where the Tractor connection is not defined. We may still define

i : T_{N} ↪ T

formally for dim

N \leq 2

, as before. This implies Theorem 4.23 , meaning that the converse still applies, with the caveat that the Einstein coefficients in the two dimensional case must be constant.

Example. To illustrate these proofs, we can see that

S^{4} \times R^{4}

does not have any holonomy-conserved sub-bundles in its tractor connection (in fact it has full holonomy), while

S^{4} \times H^{4}

is conformally flat, for

H^{4}

the hyperbolic 4-space.

Remark. Some old results of H. W. Brinkmann [5] , [6] can be proved directly using this decomposition theorem. For instance, the fact that any 4-manifold with two distinct Einstein structures in the conformal class is conformally flat. In our setting, the preserved sub-bundle spanned by the two Einstein vectors decomposes the manifold into a direct product of 3and 1-dimensional Einstein spaces. But both these spaces are conformally flat, so our original manifold has trivial holonomy; in other words, it is conformally flat.

Remark. Analogously to the previous remark, we can see that if not conformally flat, a five dimensional manifold can have up to two linearly independent Einstein structures, a six dimensional manifold can have three, an

n

dimensional manifold

n - 3

5 Einstein Spaces: Metric Cones

In this section we will give a full classification of the possible Tractor holonomies of the non Ricci-flat Einstein spaces, using to this effect the construction of a metric cone, whose Levi-Civita holonomy corresponds to the Tractor holonomy of the original manifold.

Remark. As we mentioned in the introduction, this metric cone construction is related to the Ambient Metric construction of [13] and [10] , for conformally Einstein manifolds. The actual relation is slightly subtle. This also provides a direct proof of a result of [15] , namely that the Ambient Metric construction always exists if the manifold is conformally Einstein.

Definition 5.1. A conformal manifold

M

is said to be indecomposable if they cannot be decomposed into Einstein spaces as in the previous section. In other words,

\vec{\nabla}

may preserve a single line bundle (and its orthogonal complement), but nothing else.

Remark 5.1. In the non Ricci-flat Einstein case, indecomposable implies that the tractor holonomy acts irreducibly on

R^{n + 1}

R^{n, 1}

(since the only preserved line bundle is positive or negative definite).

As all Einstein manifolds of dimension 3 are conformally flat, we shall assume

n > 3

Let

(M, g)

be an Einstein manifold,

R i c = λ g

λ \neq 0

Theorem 5.2 (Einstein Classification). The Tractor holonomy of

M^{n}

is one of the following,

n \geq 4

- $S O (n, 1)$ ,
- $S O (n + 1)$ ,
- $S U (m)$ for $2 m = n + 1$ ,
- $S p (m)$ for $4 m = n + 1$ ,
- $G_{2}$ for $n = 6$ ,
- $S p i n (7)$ for $n = 7$ .

Moreover, all these holonomy groups actually occur.

Remark. It is interesting to note that there is only a single holonomy possible for an indecomposable Einstein manifold with negative constant.

The remainder of this chapter will be dedicated to proving Theorem 5.2 .

Definition 5.3. Given an Einstein manifold

(M, g)

, we define the metric cone on

M

(N = R^{+} \times M, h)

with

\begin{matrix} h = \frac{1}{μ} d t^{2} + t^{2} g, \end{matrix}

and

μ = \frac{λ}{n - 1}

Note that

h

is of definite signature if and only if

M

has positive Einstein constant. In the negative case, we call

(N, h)

a Lorentzian cone.

Then defining

\nabla

as the Levi-Civita connection of

N

, and remembering the formula:

\begin{matrix} 2 〈 \nabla_{X} Y, Z 〉 & = & X . 〈 Y, Z 〉 + Y . 〈 X, Z 〉 - Z . 〈 X, Y 〉 \end{matrix}

\begin{matrix} + 〈 [X, Y], Z 〉 + 〈 [Z, X], Y 〉 + 〈 [Z, Y], X 〉, \end{matrix}

we can calculate the following equalities. For

T = \frac{\partial}{\partial t}

, and

X_{i}

a local basis of vector fields of

M

, extended trivially to

N

\begin{matrix} \nabla_{T} T & = & 0, \end{matrix}

\begin{matrix} \nabla_{X_{i}} T & = & \frac{1}{t} X_{i}, \end{matrix}

\begin{matrix} \nabla_{T} X_{i} & = & \frac{1}{t} X_{i}, \end{matrix}

\begin{matrix} \nabla_{X_{i}} X_{j} & = & {\tilde{\nabla}}_{X_{i}} X_{j} - t μ g (X_{i}, X_{j}) T, \end{matrix}

with

\tilde{\nabla}

the Levi-Civita connection of

g

Given a path

τ

{1} \times M

, with tangent vector field

Z

, let

Y

be the parallel transport of a vector along the path, thus

\nabla_{Z} Y = 0

. Split

Y

Y^{⊥} + a T

, with

Y^{⊥} \in Γ (T)

. Then we get the following result:

Lemma 5.4. Extend

Z

and

Y

in the

T

direction, with

Z (t, x) = Z (x)

, and

\begin{matrix} Y (t, x) = \frac{1}{t} Y^{⊥} (x) + a (x) T . \end{matrix}

Then

\nabla_{T} Y = 0

and

\nabla_{Z} Y = 0

τ \times R^{+}

Proof. The function

a

is independent of

t

, so

T (a) = 0

. Hence

\nabla_{T} (a T) = 0

Furthermore,

\begin{matrix} \nabla_{T} (\frac{1}{t} Y^{⊥}) & = & T (\frac{1}{t}) Y^{⊥} + \frac{1}{t^{2}} Y^{⊥} \end{matrix}

\begin{matrix} = & (- \frac{1}{t^{2}} + \frac{1}{t^{2}}) Y^{⊥} \end{matrix}

\begin{matrix} = & 0, \end{matrix}

\nabla_{T} Y = 0

We can expand out the original equation

\nabla_{Z} Y = 0

t = 1

, giving:

\begin{matrix} 0 = {\tilde{\nabla}}_{Z} Y^{⊥} - μ g (Z, Y^{⊥}) T + a Z + Z (a) T . \end{matrix}

By linearity, this is equivalent to the two equations

0 = {\tilde{\nabla}}_{Z} Y^{⊥} + a Z

and

0 = (- μ g (Z, Y^{⊥}) + Z (a)) T

Then similarly expanding

\nabla_{Z} Y

for varying

t

\begin{matrix} \nabla_{Z} Y & = & \frac{1}{t} {\tilde{\nabla}}_{Z} Y^{⊥} - \frac{t}{t} μ g (Z, Y^{⊥}) T + \frac{a}{t} Z + Z (a) T \end{matrix}

\begin{matrix} = & \frac{1}{t} ({\tilde{\nabla}}_{Z} Y^{⊥} + a Z) + (- μ g (Z, Y^{⊥}) + Z (a)) T \end{matrix}

\begin{matrix} = & 0 . \end{matrix}

□

The previous result shows that when we're computing the holonomy of

\nabla

, we only need to consider paths in

{1} \times M \sim = M

We can now turn to the tractor connection

\vec{\nabla}

T

, for the conformal structure

g ≃ [g]

. Using the splitting given by the metric

g

, we can see the formal similarities with

\nabla

t = 1

g

is Einstein, with coefficient

λ

, then

\vec{\nabla} (\frac{n - 1}{λ}, 0, - \frac{1}{2}) = 0

. Furthermore, for

R = (\frac{n - 1}{λ}, 0, \frac{1}{2})

, then:

\begin{matrix} {\vec{\nabla}}_{X_{i}} R = (\begin{matrix} 0 \\ X_{i} \\ 0 \end{matrix}) a n d {\vec{\nabla}}_{X_{i}} (\begin{matrix} 0 \\ X_{j} \\ 0 \end{matrix}) = (\begin{matrix} 0 \\ {\tilde{\nabla}}_{X_{i}} X_{j} \\ 0 \end{matrix}) - μ g (X_{i}, X_{j}) R . \end{matrix}

Hence under the formal identification of

R

with

T

and

(0, X_{i}, 0)

with

X_{i}

, we get

\nabla_{Z} \sim = {\vec{\nabla}}_{Z}

for

Z \in Γ (T)

t = 1

. Then by the previous lemma and its implication for the holonomy of

\nabla

, the next theorem is proved:

Theorem 5.5. The holonomy groups of

(T, \vec{\nabla}, (M, g))

and

(T N, \nabla, (N, h))

are isomorphic.

Hence the holonomy of

\vec{\nabla}

is metric, and irreducible by Remark 5.1 , and must be one of those classified by Merkulov and Schwachhöfer in [22] . In the negative Einstein case, a look at the table shows that the only possible holonomy is the full

S O (n, 1)

group itself. For the positive Einstein, we need the following result:

Proposition 5.6. The metric cone

(N, h)

is Ricci-flat.

Proof. By the definition of

h

and the corresponding

\nabla

, the curvature

R = R^{\nabla}

\begin{matrix} R_{T, -} & = & 0, \end{matrix}

\begin{matrix} R_{X_{i} X_{j}} X_{k} & = & {\tilde{R}}_{X_{i} X_{j}} X_{k} - μ g (X_{j}, X_{k}) X_{i} + μ g (X_{i}, X_{k}) X_{j}, \end{matrix}

with

\tilde{R}

the curvature of

\tilde{\nabla}

Then taking traces,

\begin{matrix} R i c (T, -) & = & 0, \end{matrix}

\begin{matrix} R i c (X_{j}, X_{k}) & = & \tilde{R i c} (X_{j}, X_{k}) + (1 - n) μ g (X_{j}, X_{k}) \end{matrix}

\begin{matrix} = & λ g (X_{j}, X_{k}) - λ g (X_{j}, X_{k}) \end{matrix}

\begin{matrix} = & 0 . \end{matrix}

□

So the possible holonomies reduce to those corresponding to metrics which are Ricci-flat, namely

S O (n + 1), S U (m), S p (m), G_{2}

and

S p i n (7)

The

S O (n + 1)

case is generic. The

S U (m)

holonomy on the cone corresponds to Sasaki-Einstein manifolds, the

S p (m)

to 3-Sasakian ones,

G_{2}

and

S p i n (7)

to weak holonomy manifolds [2] ; all of which can be realised on compact manifolds.

Remark. In [3] R.J. Baston presents a local twistor theory, which, in the case of conformal manifolds, is just given by the spin representation of

G_{0}

and the extension of

\vec{\nabla}

to this new context. A parallel section of this bundle is equivalent with the existence of a spinor

ψ

solving the twistor equation for all vector field

X

\begin{matrix} \nabla_{X} ψ + \frac{1}{n} X . D ψ = 0, \end{matrix}

with

D

the Dirac operator. Paper [16] by Katharina Habermann analyses solutions to this twistor equation; she shows that these imply that the manifold is conformally Einstein, of non-negative scalar curvature.

So the Tractor holonomy groups

G_{2}

and

S p i n (7)

actually correspond to the existence of twistor-spinors on the manifold.

Remark. The concept of a twistor-spinor is a generalisation of that of a Killing spinor. A Killing spinor is a spinor

ψ

solving the equation

\begin{matrix} \nabla_{X} ψ = λ X . ψ \end{matrix}

for all vector fields

X

and some constant

λ

. In [2] , C. Bär showed that having a Killing spinor is equivalent with having a parallel spinor on the metric cone. So the cases of weak holonomy SU(3) and nearly Kählerian structures are covered by the Tractor connection; in fact in his paper [7] constructing manifolds of exceptional holonomy, R.L. Bryant produces manifolds of holonomy

G_{2}

and

S p i n (7)

as metric cones on

S U (3) / T^{2}

and

S O (5) / S O (3)

respectively. Thus all the holonomies listed actually occur.

We can now turn to the Ricci-flat case, which is actually simpler than the general Einstein case, but with an added subtlety.

6 Ricci-Flat Spaces

Let

(M^{n}, g)

be a Ricci-flat space of indecomposable tractor holonomy. As

M^{n}

is Ricci-flat, its Tractor holonomy is contained within

S O (n) ∖ R^{n}

. Fix a point

b \in M

for calculating the holonomy groups, and let

H

be the metric holonomy of

M

D

its Tractor holonomy,

h

l

their Lie algebras.

Then:

Lemma 6.1.

H \subset D

, or, equivalently,

h \subset l

Proof. Let

Y

be the parallel transport of a vector along a path

τ

with tangent field

X

; in other words

\nabla_{X} Y = 0

, for

\nabla

the metric connection on

M

. Then

\begin{matrix} {\vec{\nabla}}_{X} (\begin{matrix} x \\ Y \\ 0 \end{matrix}) = (\begin{matrix} \nabla_{X} x - g (X, Y) \\ \nabla_{X} Y = 0 \\ 0 \end{matrix}), \end{matrix}

which is zero for

x = \int_{τ} g (X, Y)

, proving that every metric holonomy element is a tractor holonomy element. This argument also works in reverse, showing that

π (l) = h

, where

π

is the projection of

c o (n)_{b} \oplus T_{b}

onto its first component.

□

This demonstrates that

l \subset b = h \oplus T_{b}

. But first:

Lemma 6.2. The representation of

h

T_{b}

is irreducible.

Proof. If a bundle $S \subset T$ is preserved by $\nabla$ , then the bundle $ℰ [- 1] \oplus S$ is preserved by $\vec{\nabla}$ . Thus, since we assume our Tractor holonomy to be indecomposable, then $h$ must act irreducibly on $T_{b}$ . $□$

Then since the Lie bracket on

T_{b}

is trivial, the adjoint representation of

h \subset l \subset b

on the second component of

b

is the usual, irreducible one. Accordingly this adjoint representation splits

b

into two irreducible representations, isomorphic to

h

and

T_{b}

As a consequence,

l \sim = h

l \sim = b

. We now claim that

Lemma 6.3.

l \sim = b

Proof. Reasoning by contradiction, we assume that $l \sim = h$ , and go on to show that this violates our indecomposability assumption.
Express $b$ as $h_{0} \oplus R_{0}^{n}$ , the sum of the irreducible representations of $h$ . Then, as $h_{0} \sim = h$ acts irreducibly on $T_{b}$ , there is, at $b$ , a new splitting of $T$ corresponding to the splitting $R^{n *} \oplus h_{0} \oplus R_{0}^{n}$ . This splitting is
$\begin{matrix} ℰ [- 1]_{0} \oplus T [- 1]_{0} \oplus ℰ [- 1] \end{matrix}$
Then $h \subset l$ preserves the new vectors $(1, 0, 0)$ and $(0, 0, 1)$ . This shows that $l$ preserves a rank two sub-bundle, contradicting indecomposability. $□$

Putting this together, we can now claim the following theorem:

Theorem 6.4. The possible indecomposable Tractor holonomy groups for the conformal manifold

(M, g)

, conformally Ricci-flat, are:

- $S O (n) ∖ R^{n}, n \geq 4$ (generic),
- $S U (m) ∖ R^{2 m}, m \geq 2$ (Calabi-Yau),
- $S p (m) ∖ R^{4 m}, m \geq 1$ (Hyper-Kähler),
- $G_{2} ∖ R^{7}$ (see [7] ),
- $S p i n (7) ∖ R^{8}$ (see [7] ),

and all of these groups do occur.

Remark. This result offers an alternative proof for the theorems in Mario Listing's paper [21] , in the special case of conformally Ricci-flat manifolds.

6.1 Addendum: Symmetric Spaces

A symmetric space

(S, g)

is a manifold such that

\nabla^{g} R^{g} = 0

for

R^{g}

the full curvature tensor. It is quite easily to show, using the infinitesimal holonomy developed by S. Kobayashi and K. Nomizu [17] , that any indecomposable conformal manifold that is conformal to a symmetric space has the maximal holonomy in its category -

S O (n + 1, 1)

if the symmetric space is not Einstein, and

S O (n + 1)

S O (n, 1)

if it is (Ricci-flat symmetric spaces are flat).

Example. These results give an independent proof to the results of F. Leitner [19] , that the conformal holonomy of

S O (4)

, locally isomorphic to

S^{3} \times S^{3}

, is

S O (7)

. The group

S O (4)

is a positive Einstein Symmetric space, not conformally flat (consider the fate of the tractor vector

(1, 0, 0)

under parallel translation), so the result follows.

Example. Note that the same argument shows that the manifold

S^{n} (a) \times S^{n} (b)

where

a \neq b

are the radii of the spheres, has full holonomy

S O (2 n + 1, 1)

Two very similar results also are implied:

Example. If a manifold

(M, g)

is conformal to an Einstein Symmetric Space, then it cannot be conformal to any other Einstein space, or any other Symmetric Space, unless it is conformally flat.

And:

Example. If

(M, g)

is conformal to a Symmetric space in two different ways, then its tractor holonomy is full or null.

References

T.N. Bailey, M.G. Eastwood and R. Gover: Thomas's Structure Bundle for Conformal, Projective and Related Structures, Rocky Mountain J., 24 (1994), 1191-1217. MR 95h:53064
C. Bär: Real Killing Spinors and Holonomy, Comm. Math. Phys., 154 (1993), No. 3, 509-521.
R.J. Baston: Almost Hermitian Symmetric Manifolds I Local Twistor Theory, Duke Mathematical Journal, 63 (1991), No. 4, 81-112.
T. Branson, A. Čap, M. Eastwood and R. Gover: Prolongations of Geometric Overdetermined Systems, arXiv:math.DG/ 0402100 v2 (2004).
H.W. Brinkmann: Riemmann Spaces Conformal to Einstein Spaces, Math. Ann., 91 (1924), 269-278.
H.W. Brinkmann: Einstein Spaces which are Mapped Conformally on Each Other, Math. Ann., 94 (1925), 119-145.
R.L. Bryant: Metrics with Exceptional Holonomy, Annals of Mathematics, 126 (1987), No. 3, 525-576.
A. Čap and A.R. Gover: Tractor Calculi For Irreducible Parabolic Geometries, Global analysis and harmonic analysis (Marseille-Luminy, 1999), Sémin. Congr., 4 (2000), Soc. Math. France, Paris, 129-154.
A. Čap and A.R. Gover: Tractor Calculi For Parabolic Geometries, Transactions of the American Mathematical Society, 354 (2001), No. 4, 1511-1548.
A. Čap and R. Gover: Standard Tractors and the Conformal Ambient Metric Construction, Annals of Global Analysis and Geometry, 24 (2003), No. 3, 231-259.
E. Cartan: Les Espaces à Connexion Conform, Ann. Soc. Pol. Math., 2 (1923), 171-202.
A.J. Di Scala, C. Olmos: The Geometry of Homogeneous Submanifolds of Hyperbolic Space, Mathematische Zeitschrift, 237 (2001), 199-209.
C. Fefferman and C.R. Graham: Conformal Invariants, Astérisque, hors série: `Elie Cartan et les Mathématiques d'Aujourd'hui' (1985), 95-116.
A.R. Gover: Invariant Theory and Calculus for Conformal Geometries, Advances in Mathematics, 163 (2001), 206-257.
R. Gover and L.J. Peterson: The Ambient Obstruction Tensor and the Conformal Deformation Complex, arXiv:math.DG/0408229 v1 (2004).
K. Habermann: The Twistor Equation on Riemannian Manifolds, Journal of Geometry and Physics, 7 (1990), No. 4, 469-488.
S. Kobayashi and K. Nomizu: Foundations of Differential Geometry, 1 (1996), Wiley-Interscience Publication.
N. Koike: Totally Umbilic Orthogonal Nets and Decomposition Theorems, Saitama Mathematical Journal, 10 (1992), 1-19.
F. Leitner: Conformal Holonomy of Bi-Invariant Metrics, arXiv:math.DG/ 0406299 v1 (2004).
F. Leitner: Normal Conformal Killing Forms, arXiv:math.DG/ 0406316 (2004).
M. Listing: Conformal Einstein Spaces in $N$ -Dimensions, Annals of Global Analysis and Geometry, 20 (2001), 183-197.
S. Merkulov, L. Schwachhöfer: Classification of Irreducible Holonomies of Torsion-free Affine Connections, Annals of Mathematics, 150 (1999), 77-150.
N. Tanaka: On the Equivalence Problem Associated with Simple Graded Lie Algebras, Hokkaido Math. J., 8 (1979), 23-84. MR 80h:53034
S. Sasaki: On the Spaces with Normal Conformal Connexions whose Groups of Holonomy fixes a Point or a Hypersphere. I. II. III., Jap. Journ. Math., 18 (1943), 615-622, 623-633, 634-795.
S. Sasaki and K. Yano: On the Structure of Spaces with Normal Conformal Connection whose Holonomy Group leaves Invariant a Sphere of Arbitrary Dimension, Sugaku (Mathematics), 1 (1947), 18-28.
T.Y. Thomas: On Conformal Geometry, Proc. N.A.S., 12 (1926), 352-359.
T.Y. Thomas: Conformal Tensors, Proc. N.A.S., 18 (1931), 103-189.

Conformal Holonomy: a Classification

Stuart Armstrong

23 January 2005