Spin 2010 (jmf) 44

Lecture 6: The spin connection

On the tangent bundle of a riemannian manifold (M, g ) there is a privileged connection called the Levi-
Civita connection. Thinking of the tangent bundle as an associated vector bundle to the bundle O(M)
of orthonormal frames, we will see that this connection is induced from a connection on O(M), which
restricts to a connection on SO(M) when (M, g ) is orientable and lifts to a connection on any spin bundle
Spin(M) if (M, g ) is spin. That being the case, it defines a connection on the spinor bundles which is
usually called the spin connection.

6.1 The Levi-Civita connection

Let (M, g ) be a riemannian manifold. We summarise here the basic definitions and results of the rieman-
nian geometry of (M, g ).

Theorem 6.1 (The fundamental theorem of riemannian geometry). There is a unique connection on the
tangent bundle TM which is

1. metric-compatible:

∇X g = 0 equivalently Xg (Y, Z) = g (∇X Y, Z) + g (Y, ∇X Z) ,

2. and torsion-free:
∇X Y − ∇Y X = [X, Y] ,

where X, Y, Z are vector fields on M and [X, Y] denotes the Lie bracket of vector fields.

Proof. The proof consists in finding an explicit formula for the connection in terms of the metric. Let
X, Y, Z ∈ X (M). The metric compatibility condition says that

Xg (Y, Z) = g (∇X Y, Z) + g (Y, ∇X Z)

Yg (Z, X) = g (∇Y Z, X) + g (Z, ∇Y X)
Zg (X, Y) = g (∇Z X, Y) + g (X, ∇Z Y) ,

whereas the vanishing of the torsion allows to rewrite the middle equation as

Yg (Z, X) = g (∇Y Z, X) + g (Z, ∇X Y) + g (Z, [X, Y]) .

We now compute

Xg (Y, Z) + Yg (Z, X) − Zg (X, Y) = 2g (∇X Y, Z) + g (Y, ∇X Z − ∇Z X) + g (∇Y Z − ∇Z Y, X) + g (Z, [X, Y])

and use the torsionless condition once again to arrive at the Koszul formula

(79) 2g (∇X Y, Z) = Xg (Y, Z) + Yg (Z, X) − Zg (X, Y) − g (Y, [X, Z]) − g ([Y, Z], X) − g (Z, [X, Y])

which determines ∇X Y uniquely.

The connection so defined is called the Levi-Civita connection. Its curvature, defined by

(80) R(X, Y)Z = ∇[X,Y] Z − ∇X ∇Y Z − ∇Y ∇X Z ,

gives rise to the Riemann curvature tensor

R(X, Y, Z, W) := g (R(X, Y)Z, W) .

Proposition 6.2. The curvature satisfies the following identities

Spin 2010 (jmf) 45

1. symmetry conditions:

R(X, Y)Z = −R(Y, X)Z and R(X, Y, Z, W) = −R(X, Y, W, Z) ,

2. algebraic Bianchi identity:

R(X, Y)Z + R(Y, Z)X + R(Z, X)Y = 0 ,

3. differential Bianchi identity:

∇X R(Y, Z) + ∇Y R(Z, X) + ∇Z R(X, Y) = 0 .

A tensor satisfying the symmetry conditions and the algebraic Bianchi identity is called an algebraic
curvature tensor.
If we fix X, Y ∈ X (M), the curvature defines a linear map Z �→ R(X, Z)Y, whose trace is the Ricci
(curvature) tensor r (X, Y).

Proposition 6.3. The Ricci tensor is symmetric: r (X, Y) = r (Y, X).

The trace (relative to the metric g ) of the Ricci tensor is called the scalar curvature of (M, g ) and
denoted s.

Definition 6.4. A riemannian manifold (M, g ) is said to be Einstein if r (X, Y) = λg (X, Y) for some λ ∈ R.
Clearly λ = s/n where n is the dimension of M. It is said to be Ricci-flat if r = 0 and flat if R = 0.

If h, k ∈ C∞ (M, S 2 T ∗ M) are two symmetric tensors, their Kulkarni–Nomizu product h ⊙ k is the al-
gebraic curvature tensor defined by

(81) (h ⊙ k)(X, Y, Z, W) = h(X, Z)k(Y, W) + h(Y, W)k(X, Z) − h(X, W)k(Y, Z) − h(Y, Z)k(X, W) ,

for all X, Y, Z, W ∈ X (M).

Proposition 6.5. The Riemann curvature tensor can be decomposed as

s 1 s
R= g ⊙g + (r − g ) ⊙ g + W
2n(n − 1) n −2 n

where W is the Weyl (curvature) tensor.

The Weyl tensor is the “traceless” part of the Riemann tensor. It is conformally invariant and if it
vanishes, (M, g ) is said to be conformally flat. If (M, g ) is Einstein, then the middle term in R is absent.
If only the first term is present then (M, g ) is said to have constant sectional curvature.

6.2 The connection one-forms on O(M), SO(M) and Spin(M)

The Levi-Civita connection of a riemannian manifold induces a connection one-form ω on the or-
thonormal frame bundle and, if orientable, also on the oriented orthonormal frame bundle. Indeed,
let us assume that M is orientable and let E : U ⊂ M → SO(M) be local orthonormal frame, i.e., a local
section of SO(M). Then we may pull ω back to a gauge field E ∗ ω on U with values in so(s, t ), for (M, g )
of signature (s, t ). We can describe the gauge field explicitly as follows. Let (e i ) denote the elements in
the frame E . Being orthonormal, their inner products are given by g (e i , e j ) = εi δi j , where εi = ±1. Then
we have �
E ∗ ω = 12 ωi j εi ε j e i � e j ,
i,j

where ωi j ∈ Ω1 (U) is defined by

(82) ωi j (X) = g (∇X e i , e j )

Spin 2010 (jmf) 46

for all X ∈ X (M) and e i � e j ∈ so(s, t ) are the skewsymmetric endomorphisms defined by (53). It is
convenient in calculations to introduce the dual frame e i = εi e i , where now g (e i , e j ) = δi j , and in terms
of which �
E ∗ ω = 12 ωi j e i � e j .
i,j

If E is another local frame E : U → SO(M), so that on U ∩ U � , E � = E h for some h : U ∩ U � → SO(s, t ),

� � �

then on U ∩ U � ,
E �∗ ω = h E ∗ ωh −1 − d hh −1 ,
whence it does indeed give rise to a gauge field.
Now let
ϕ
Spin(M) � SO(M)
�� �
�� ��
��
�� ���
�
�� �� �
M
denote a spin bundle. The connection 1-form ω on SO(M) pulls back to a connection 1-form ϕ∗ ω on
Spin(M), called the spin connection. Now given a local section E of SO(M), let E� denote a local section
of Spin(M) such that ϕ ◦ E� = E . Then the gauge field associated to ϕ∗ ω via E� coincides with the one
associated to ω via E :

(83) E�∗ ϕ∗ ω = (ϕ ◦ E�)∗ ω = E ∗ ω .

If � : Spin(s, t ) → GL(F) is any representation, then on sections of the associated vector bundle
Spin(M) ×Spin(s,t ) F we have a covariant derivative
�
(84) d ∇ = d + 12 ωi j �(e i � e j ) ,
i,j

where we also denote by � : so(s, t ) → gl(F) the representation of the Lie algebra.
We shall be interested primarily in the spinor representations of Spin(s, t ), which are induced by re-
striction from pinor representations of C�(s, t ). This means that the associated bundle Spin(M)×Spin(s,t )
F is (perhaps a subbundle of) a bundle C�(TM)×C�(s,t ) P of Clifford modules. In this case, it is convenient
to think of the gauge field as taking values in the Clifford algebra. If we let ρ : so(s, t ) → C�(s, t ) denote
the embedding defined in (55), then
�
(85) ρ(E ∗ ω) = 14 ωi j e i e j ,
i,j

where e i e j ∈ C�(s, t ). The curvature two-form of this connection is given by

�
(86) ρ(E ∗ Ω) = 14 Ωi j e i e j ,
i,j

where Ωi j (X, Y) = g (R(X, Y)e i , e j ) for all X, Y ∈ X (M), with R(X, Y) defined by (80).
The Clifford algebra-valued covariant derivative is compatible with Clifford action in the following
sense. Suppose that θ ∈ C�(TM) and ψ is a section of a bundle of Clifford modules associated to C�(TM).
Then for all vector fields X ∈ X (M), we have that

(87) ∇X (θ · ψ) = ∇X θ · ψ + θ · ∇X ψ ,

where ∇X θ agrees with the action of the Levi-Civita connection on θ viewed as a section of ΛTM.
Spin 2010 (jmf) 47

6.3 Parallel spinor fields

We can now define the notion of a parallel spinor field as a (nonzero) section of a spinor bundle which
is covariantly constant. On a trivialising neighbourhood U of M, where Spin(M) is trivialised by a local
section E� lifing a local orthonormal frame E , a spinor field is given by a function ψ : U → S(s, t ) taking
values in the spinor representation, which we think of as the restriction to Spin(s, t ) of an irreducible
C�(s, t )-module. Depending on (s, t ), it may very well be the case that the S(s, t ) so defined is not irre-
ducible, in which case S(s, t ) = S(s, t )+ ⊕S(s, t )− decomposes into two half-spinor irreducible represent-
ations of Spin(s, t ). The covariant derivative of ψ is given by
�
(88) d ∇ ψ = d ψ + 14 ωi j e i e j ψ ,
i,j

and we say that ψ is covariantly constant (or parallel) if d ∇ ψ = 0. The fact (78) that d ∇ is covariant
means that this equation is well-defined on global section of the spinor bundle.
Differentiating d ∇ ψ again we obtain an integrability condition for the existence of parallel spinor
fields, namely
�
(89) d ∇ d ∇ ψ = 14 Ωi j e i e j ψ = 0 .
i,j

This equation is equivalent to

(90) R(X, Y)ψ = 0 ,

where R(X, Y) ∈ C�(TM) acts on ψ via Clifford multiplication. Relative to the local orthonormal frame
E = (e i ), we have
�
(91) R(e i , e j ) · ψ = 0 =⇒ Ri j k� e k e � ψ = 0 .
k,�

If we multiply the above equation with e j and sum over j , we obtain the following:
�
0= Ri j k� e j e k e � ψ
j ,k,�
� � �
= Ri j k� e j k� − g j k e � + g j � e k ψ
j ,k,�
� � �
= Ri j k� e j k� + 2g j � e k ψ .
j ,k,�

The first term vanishes by the algebraic Bianchi identity and the second term yields the Ricci tensor,
whence the integrability condition becomes
�
(92) Ri j e j ψ = 0 .
j

More invariantly, this says the following. The Ricci tensor defines an endomorphism R of the tangent
bundle called the Ricci operator, by g (R(X), Y) = r (X, Y). Then the above integrability condition says
that R(X)ψ = 0 for all X ∈ X (M). Hitting this equation again with R(X), we see that g (R(X), R(X)) = 0 for
all X. If g is positive-definite, then R(X) = 0 and (M, g ) is Ricci-flat. In indefinite signature, the image of
the Ricci operator consists of null vectors, whence we could call such manifolds Ricci-null.
In the next lecture we will reformulate the question of which spin manifolds admit parallel spinor
fields in terms of holonomy.