Spin 2010 (jmf) 27

Lecture 4: Spin manifolds

Thus, the existence of a spinor structure appears, on phys-
ical grounds, to be a reasonable condition to impose on
any cosmological model in general relativity.
— Robert Geroch, 1969

In this lecture we will discuss the notion of a spin structure on a finite-dimensional smooth man-
ifold. We start with some basic notions, just in case the intended audience includes people with little
background in differential geometry.

4.1 What is a manifold?

We start with a familiar definition from topology.

Definition 4.1. A (topological) n-dimensional manifold is a Hausdorff topological space with a count-
able basis and which is locally homeomorphic to Rn ; that is, every point in M has a neighbourhood
which is homeomorphic to Rn .

We shall be interested in doing calculus on manifolds, and this requires introducing a differentiable
structure. We assume that we know how to do calculus on Rn and the point of a differentiable structure
is to enable us to do calculus on spaces which are locally “like” Rn in a way that it is as independent
as possible on the precise form of the local homeomorphisms. Please note that we will consider only
infinitely differentiable (or smooth) structures. This is not necessary, but it is certainly sufficient for our
purposes.

Definition
� � 4.2. A smooth structure on an n-dimensional manifold M is an atlas of coordinate charts
(Uα , φα ) α∈I , for I some indexing set, where {Uα } is an open cover of M and φα : Uα → Rn are homeo-
morphisms whose transition functions on nonempty overlaps Uαβ := Uα ∩ Uβ

g αβ = φα ◦ φ−1
β : φβ (Uαβ ) −→ φα (Uαβ )

are diffeomorphisms between open subsets of R�n ; that is,�g αβ are

� infinitely
� differentiable with infinitely
differentiable inverses. Two smooth structures (Uα , φα ) and (Vβ , ψβ ) are equivalent if their union
is also an atlas. A maximal atlas consists of the union of all atlases in one such equivalence class. A
topological manifold with a maximal atlas is called a smooth manifold.

Remark 4.3. Notice that not every topological manifold admits a smooth structure and that there are
topological manifolds admitting more than one inequivalent smooth structures. For example, it is
known that R4 admits an uncountably infinite number of smooth structures, but for us in this course
Rn will always have the standard smooth structure, unless otherwise explicitly stated.
n
Let M be a smooth manifold. A function f : M → R is smooth if for each α ∈ I, f ◦ φ−1 α : R → R is
p
smooth as a function of n real variables. Similarly a function f : M → R is smooth if each component
function f i : M → R, for i = 1, . . . , p, is smooth. A map f : Mn → Np between smooth manifolds of
dimensions n and p, respectively, is smooth if for every m ∈ M, ψβ ◦ f : Vβ → Rp is smooth for some (and
hence all) coordinate charts (Vβ , ψβ ) containing the point f (m). Smooth manifolds form the objects of
a category whose morphisms are the smooth maps between them. The isomorphisms in that category
are the called diffeomorphisms: namely, smooth maps f : M → N with a smooth inverse.
We could say a lot more about calculus on manifolds, but perhaps this suffices for now.

4.2 Fibre bundles

The definition of a spin structure on a smooth manifold is phrased in the language of fibre bundles and
we introduce this language in this section.
Let G be a Lie group and let F be a smooth manifold with a smooth G-action: G × F → F. We will
assume that G acts effectively so that if (g , f ) �→ f for all f ∈ F, then g = 1, the identity. It is often
Spin 2010 (jmf) 28

convenient to write the action as a map ρ : G → Aut(F), where Aut(F) is the automorphism group of
the fibre. In the most general case, Aut(F) = Diff(F) is the group of diffeomorphisms, but we will be
working mostly with vector bundles, for which F is a vector space and Aut(F) = GL(F), whence ρ is a
representation of G.

Definition 4.4. A fibre bundle (with structure group G and typical fibre F as above) over M is a smooth
surjection π : E → M together with a local triviality condition: every m ∈ M has a neighbourhood U and
a diffeomorphism φU : π−1 U −→ U × F such that the following triangle commutes:

φU
π−1 U� � U×F
�� � �
��
���
π ��� � pr1
� �� �
U

and such that on nonempty overlaps U ∩ V

�
φU ◦ φ−1 �
V {m}×F = ρ(g UV (m))

for some the transition functions g UV : U ∩ V → G. The manifold M is called the base of the fibre
bundle, whereas E is called the total space. For each m ∈ M, the fibre π−1 m = {e ∈ E|π(e) = m} over m is
a submanifold of E which is diffeomorphic to F.
pr1
The trivial bundle with typical fibre F is simply the Cartesian product M × F −→ M, in which case
we can take the φU to be the restriction to U of the identity diffeomorphism. Fibre bundles are of
course locally trivial, but can be twisted in the large. The maps φU in the general case are called local
trivialisations. Fibre bundles (with structure group G) over a fixed smooth manifold M are the objects
of a category, where a morphism between two fibre bundles π : E → M and π� : E� → M over M is a
G-equivariant fibre-preserving smooth map ϕ : E → E� such that the following triangle commutes:

ϕ
(69) E� � E�
� �� �
�� ��
π �� ���π�
� ��
M

The restriction of having the same structure group can be lifted and we can equally well consider
morphisms between fibre bundles with different structure groups where now the fibre-preserving map
ϕ : E → E� intertwines between the G and G� actions on the fibres. In any case, if ϕ is a diffeomorphism,
then the two bundles are said to be equivalent. A fibre bundle is said to be trivial if it is equivalent to
the trivial bundle M × F.
A fibre bundle gives rise to some local data from where it can then be reconstructed, up to equi-
valence. Let U = {Uα }α∈I be an open cover of M and let φα : π−1 Uα → � Uα × F be local trivialisations with
�
transition functions g αβ : Uαβ → G, defined by ρ(g αβ (m)) = φα ◦ φ−1
β
� as above. Notice that the {g αβ }
{m}×F
satisfy the following conditions:

1. g αα (m) = 1 for all m ∈ Uα ,

2. g αβ (m)g βα (m) = 1 for all m ∈ Uαβ , and

3. the cocycle condition

(70) g αβ (m)g βγ (m)g γα (m) = 1 for all m ∈ Uαβγ := Uα ∩ Uβ ∩ Uγ .

Notice that if we do not demand that α, β and γ be different, then the cocycle condition implies the
other two. We will refer to them collectively as the “cocycle conditions”. Notice as well that we are using
that G acts effectively on F, otherwise the right-hand sides of these equations would not necessarily
Spin 2010 (jmf) 29

be the identity in G, but anything in the kernel of the action ρ. Now the local trivialisations φα glue to
define a diffeomorphism
� ��
�
(71) E∼= Uα × F ∼ where (m, f ) ∼ (m, ρ(g αβ (m)) f ), for all m ∈ Uαβ and f ∈ F.
α∈I

From now on we shall drop ρ from the notation and simply write g f for g ∈ G acting on f ∈ F. Notice
that the cocycle conditions above are, respectively, the reflexive, symmetry and transitivity conditions
for the equivalence relation ∼.
This allows us to construct fibre bundles by gluing local data. Indeed, if we are given an open cover
U = {Uα }α∈I for M and functions g αβ : Uαβ → G on overlaps satisfying the cocycle conditions then we get
a fibre bundle by defining E by (71) and the surjection π : E → M by the projection pr1 : Uα × F → Uα ,
which is respected by the equivalence relation. The resulting bundle is a fibre bundle trivialised over U.

4.2.1 Vector and principal bundles

As mentioned above, a general fibre bundle will have as structure group the diffeomorphism group of
the typical fibre, but there are important examples where G is much smaller. For example, if we take
F to be a vector space and G to act linearly, then we have a vector bundle. Similarly if F ∼ = G itself and
G acts on G by left multiplication, then we have a principal G-bundle. In this latter case, there is a
well-defined right action of G on the total space E of the principal bundle: (m, g ) �→ (m, g g � ) which is
fibre-preserving and clearly G-equivariant, since associativity of the group multiplication says that left
and right multiplications commute. Since right multiplication of G on itself is simply transitive, we have
that M ∼= E/G is the quotient of E by this right action of G. One often sees this as the starting point for a
definition of a principal bundle.
We can go back and forth between vector and principal bundles via two natural constructions:
bundle of frames
vector −−−−−−−−−−−−−→ principal
bundles ←−−−−−−−−−−−−− bundles
associated bundle
π
Given a vector bundle E −
→ M with typical fibre a vector space F, let us define a principal bundle
Π
GL(E) − → M by declaring the fibre GL(E)m to be the set of frames of the vector space Em . This is a
principal homogeneous space (or torsor) of the general linear group in that any two frames are related
by a unique invertible linear transformation. This means that as a set GL(E)m ∼ = GL(Em ), but the iso-
morphism is �not natural:� it depends on choosing a reference frame. Nevertheless, in terms of a local
trivialisation (Uα , φα ) α∈I for E, with transition functions g αβ : Uαβ → GL(F) we define
� ��
�
GL(E) = Uα × GL(F) ∼ where (m, g ) ∼ (m, g αβ (m)g ), for all m ∈ Uαβ and g ∈ GL(F).
α∈I

This is then a principal GL(F)-bundle.

Π
Conversely, given a principal G-bundle P − → M and a finite-dimensional representation ρ : G →
GL(F) on a vector space F, we have a right G-action on P × F given by (p, f )g = (pg , ρ(g −1 ) f ) for p ∈ P,
f ∈ F and g ∈ G. This action is free because G acts freely on P, and hence the quotient E = (P × F)/G can
be given a smooth structure. Since P/G ∼ = M, we see that this is a fibre bundle with typical fibre F. The
surjection π : E → M is induced from the surjection Π : P → M which is preserved � by the� equivalence
relation by virtue of Π(pg ) = Π(p). Alternatively, in terms of a local trivialisation Uα , φα α∈I for P with
transition functions g αβ : Uαβ → G, we define
� ��
�
E= Uα × F ∼ with (m, f ) ∼ (m, g αβ (m) f ) for m ∈ Uαβ and f ∈ F
α∈I

and π : E → M defined by π[(m, f )] = m.

In summary, the two constructions above relate fibre bundles which are locally trivialisable over
the same cover and the corresponding transition functions are simply related. It is largely a matter of
choice whether one decides to work with principal bundles and their associated bundles or with vector
bundles and their bundles of frames. For the most part we will choose the former.
Spin 2010 (jmf) 30

4.2.2 Equivalence classes of principal bundles

From the above discussion, the emerging picture is one of principal G-bundles defined by data con-
sisting of a trivialising cover U = {Uα }α∈I and functions g αβ : Uαβ → G on double overlaps satisfying the
cocycle conditions (70). Different choices of U and of cocycles {g αβ } can still give rise to equivalent
bundles.
From the definition of the g αβ in terms of local trivialisations g αβ = φα ◦ φ−1
β
one can see that it is
still possible to compose φα with functions g α : Uα → G to give rise to a new trivialisation φ�α = g α ◦ φα
�
and hence to new transition functions g αβ = g α ◦ g αβ ◦ g β−1 which still satisfy the cocycle conditions. We
� �
say that two cocycles {g αβ } and {g αβ } are equivalent if g αβ = g α ◦ g αβ ◦ g β−1 for some “cochain” {g α }. We
will let H1 (U, G) denote the set of equivalence classes of cocycles. It classifies the principal G-bundles
trivialised on U up to equivalence.

Remark 4.5. Those familiar with sheaf cohomology will recognise H1 (U, G) as the first Čech cohomo-
logy set of the sheaf of germs of smooth functions M → G relative to the open cover U. For G nonabelian,
this will fail to be a group and be only a pointed set, with distinguished element the isomorphism class
of the trivial bundle.

Now suppose that we are given two principal G-bundles defined by local data (U = {Uα }α∈I , {g αβ })
�
and (V = {Vα }α∈J , {g αβ }). In order to compare them we would like to define the two bundles relative to
the same trivialising cover. This is done by passing to a common refinement of U and V. More precisely
let U = {Uα }α∈I be an open cover for M. We say that an open cover V = {Vβ }β∈J refines U if there is a
reindexing map j : J → I such that for every β ∈ J, Vβ ⊆ U j (β) . Now any two open covers U = {Uα }α∈I and
V = {Vβ }β∈J have a common refinement. For example, we can take W = {Uα ∩ Vβ }(α,β)∈I×J . It is clearly
again an open cover and it is clear that it refines both U and W: the reindexing functions I × J → I and
I × J → J are the cartesian projections. (This makes the set of open covers into a directed set.) We can
�
then restrict the cocycles {g αβ } defined on U and {g αβ } defined on V to W and in effect consider them
as cocycles on W where they can be compared as above. So that two principal bundles defined by local
�
data (U = {Uα }α∈I , {g αβ }) and (V = {Vα }α∈J , {g αβ }) are equivalent if the restriction of their cocycles to some
refinement W are equivalent, whence they define the same class in H1 (W, G). The way to formalise this
is to define define H1 (M, G) = lim H1 (U, G), the direct limit of the restriction maps H1 (U, G) → H1 (V, G)
−−→U
for V a refinement of U. Then we see that two principal G-bundles are equivalent if they define the
1
same class in H (M, G), which then becomes the set of equivalence classes of principal G-bundles on
M. It is a pointed set with the trivial bundle as distinguished element.

4.3 Some vector bundles on riemannian manifolds

We will now specialise to riemannian manifolds.

4.3.1 Orientability and the orthonormal frame bundle

Tangent bundle

Let M be a smooth manifold and m ∈ M a point. Then by a curve through m we mean a

smooth function t �→ c(t ), with c(0) = m. Its velocity at m is the derivative with respect to
t evaluated at t = 0: c � (0). The space of the velocities at m of all curves though m defines
�
the tangent space Tm M of M at m. It is a vector space. The union TM := m∈M Tm M can be
given the structure of a smooth manifold in such a way that the map π : TM → M which sends
v ∈ Tm M to m is a surjection making it into a vector bundle over M.

A riemannian manifold (M, g ) is a manifold M together with a metric g , which is a smoothly varying
family of nondegenerate symmetric bilinear forms on the tangent spaces of M. Notice that we do not
demand that g be positive-definite.
Spin 2010 (jmf) 31

Remark 4.6. Although every (paracompact) smooth manifold admits a positive-definite metric, the ex-
istence of indefinite metrics often imposes topological restrictions on M. For example, if M is compact
(and orientable?) then it admits a lorentzian metric (i.e., one of signature (1, n − 1) or (n − 1, 1) for n > 1)
if and only if its Euler characteristic vanishes – a result due to Geroch.

The tangent bundle of a smooth n-dimensional manifold has structure group GL(n, R), but for a
riemannian manifold, the existence of orthonormal frames implies that it is equivalent to a vector
bundle with structure group O(s, t ) if the metric has signature (s, t ). If U = {Uα }α∈I is a trivialising cover
for the tangent bundle, we let g αβ : Uαβ → O(s, t ) be the transition functions for the bundle O(M) → M
of orthonormal frames.

Example 4.7. Let S n ⊂ Rn+1 be the unit sphere. For x ∈ S n , the tangent space Tx S n is given by those
vectors in Rn+1 which are perpendicular to x. The orthonormal frame bundle is O(n + 1) → S n . Indeed,
given x ∈ S n and an orthonormal frame e 1 , . . . , e n for Tx S n , the (n + 1) × (n + 1)-matrix whose first n
columns are given by the e i and whose last column is given by x is orthogonal. Conversely, given a ∈
O(n + 1), the map π : O(n + 1) → S n defined by setting π(a) to be the last column of a is such that the
fibre at π(a) is the set of orthonormal frames for the perpendicular subspace to π(a) in Rn+1 .

A riemannian manifold (M, g ) is oriented if we can restrict consistently to oriented orthonormal

frames or, in other words, whether we can reduce the structure group of TM from O(s, t ) to SO(s, t ). Con-
cretely, this means being able to choose transition functions for the orthonormal frame bundle which lie
in SO(s, t ), perhaps relative to a refinement of the trivialising cover. So given {g αβ } taking values in O(s, t )
�
we ask whether we can find {g αβ } taking values in SO(s, t ). Let f αβ (m) = det g αβ (m) for m ∈ Uαβ . Since an
orthogonal matrix, independently of the signature, has determinant ±1, the f αβ (m) take values in the
group {±1} of order 2. The cocycle condition for {g αβ } imply the cocycle condition for { f αβ }, whence this
defines a principal fibre bundle with structure group Z2 . Orientability of M is equivalent to the triviality
of this bundle. Indeed, if (and only if ) f αβ (m) = f α (m) f β (m) for some Z2 -valued “cochain” f α : Uα → Z2 ,
�
then we can define g αβ (m) = g α (m)g αβ (m)g β−1 (m) for some g α : Uα → O(s, t ) with det g α (m) = f α (m),
�
whence g αβ : Uαβ → SO(s, t ) and still satisfies the cocycle conditions. The cohomology class defined by
{ f αβ } in H1 (M, Z2 ) which measures the failure of M to being orientable is called the first Stiefel–Whitney
class of M. Its vanishing is tantamount to orientability. Since H1 (M, Z2 ) ∼= Hom(π1 (M), Z2 ), if M is simply
connected then it is automatically orientable. Even if M is not orientable, there is a double cover, namely
the total space of the principal Z2 -bundle defined by the class of { f αβ } in H1 (M, Z2 ), which is oriented
and locally isometric to (M, g ).

Remark 4.8. Readers familiar with Čech cohomology will recognise the obstruction of orientability as
image of the class in H1 (M, O(s, t )) corresponding to the orthonormal frame bundle under the last map
in the long exact cohomology sequence

H1 (M, SO(s, t )) � H1 (M, O(s, t )) � H1 (M, Z2 )

coming from the exact sheaf sequence which is induced from the exact sequence of groups

� SO(s, t ) � O(s, t ) det � Z2 �1

1

Notice that since O(s, t ) and SO(s, t ) are nonabelian groups, there are no Hp>1 , whence the exact co-
homology sequence ends there.

Example 4.9. For n ≥ 2, the sphere S n is simply connected, whence it is orientable. In fact, the oriented
orthonormal frame bundle is SO(n +1) → S n , with the map given again by the last column of the matrix.

4.3.2 The Clifford bundle and the obstruction to defining a pinor bundle

Any functorial construction on vector spaces — e.g., ⊕, ⊗, Hom,... — gives rise to a similar construction
on vector bundles, and in particular any such construction on representations of G gives rise to sim-
ilar constructions on associated vector bundles to any principal G-bundle. On a riemannian manifold
Spin 2010 (jmf) 32

(M, g ) each tangent space becomes a quadratic vector space, relative to the quadratic form induced
from the inner product defined by the metric. Hence one should expect that any functorial construc-
tion on quadratic vector spaces should globalise to a similar construction on a riemannian manifold.
One such construction is the Clifford algebra, which gives rise to a Clifford bundle C�(TM). As a vec-
tor bundle, C�(TM) ∼ = ΛTM, but C�(TM) is actually a bundle of Clifford algebras. Alternatively we can
define it from a local trivialisation of the orthonormal frame bundle O(M):
� ��
�
C�(TM) = Uα × C�(s, t ) ∼ with (m, c) ∼ (m, C�(g αβ (m))c) for m ∈ Uαβ and c ∈ C�(s, t ),
α∈I

where C�(g αβ (m)) is the Clifford algebra automorphism derived functorially from the orthogonal trans-
formation g αβ (m). Since C�(g αβ (m)) are automorphisms of the Clifford algebra, the Clifford product on
C�(TM) is well-defined.
A natural question, given the existence of the Clifford bundle, is whether there is a vector bundle as-
sociated to the pinor representation of the Clifford algebra. If P(s, t ) is a pinor representation of C�(s, t )
one could try to build such a bundle from local data as follows
� ��
? �
P= Uα × P(s, t ) ∼ with (m, p) ∼ (m, g αβ (m)p) for m ∈ Uαβ and p ∈ C�(s, t ),
α∈I

except that O(s, t ) does not act on P(s, t ) and hence we don’t know what g αβ (m)p is.
Since Pin(s, t ) does act on P(s, t ), we could try to define
� ��
? �
P= Uα × P(s, t ) ∼ with (m, p) ∼ (m, g�
αβ (m)p) for m ∈ Uαβ and p ∈ C�(s, t ),
α∈I

where g� �
αβ (m) ∈ Pin(s, t ) is a lift of g αβ (m) ∈ O(s, t ). In other words, Ad�
�
g (m) = g αβ (m), where Ad :
αβ
Pin(s, t ) → O(s, t ) is the surjection in Proposition 3.3. For the above definition to make sense, ∼ must be
an equivalence relation and this is tantamount to the cocycle condition for g� αβ (m) : Uαβ → Pin(s, t ):

g� � �
αβ (m)g βγ (m)g γα (m) = 1 for all m ∈ Uαβγ .

� to the cocycle conditions, we obtain the cocycle conditions for the g αβ (m) : Uαβ → O(s, t ),
Applying Ad
which are satisfied, hence

f αβγ (m) := g� � � �
αβ (m)g βγ (m)g γα (m) ∈ ker Ad = Z2

and hence defines maps f αβγ : Uαβγ → Z2 . Moreover f αβγ is itself a cocycle, in that in quadruple overlaps

f αβγ (m) f αβδ (m) f αγδ (m) f βγδ (m) = 1 for all m ∈ Uαβγδ .

� has nontrivial kernel, the lift g�

Since Ad αβ (m) is not unique and any other lift is related to this by some
cochain f αβ : Uαβ → Z2 . This changes the cocycle f αβγ by a coboundary
�
f αβγ (m) �→ f αβγ (m) := f αβγ (m) f αβ (m) f βγ (m) f αγ (m) .

�
In particular, f αβγ still satisfies the cocycle condition on quadruple overlaps and its class in H2 (U, Z2 ),
and hence in H2 (M, Z2 ), is unchanged. If (and only if ) this class vanishes, will we be able to lift the g αβ
to Pin(s, t ) in such a way that the cocycle conditions are satisfied. Indeed, the class of f αβγ in H2 (M, Z2 )
vanishes if on some “good” cover U, f αβγ (m) = f αβ (m) f βγ (m) f αγ (m) for some f αβ : Uαβ → Z2 . This being
� 2
the case, then g αβ (m) = g�
αβ (m) f αβ (m) is our desired Pin(s, t )-valued cocycle. The class in H (M, Z2 )
defined by the f αβγ is essentially the second Stiefel–Whitney class of M, and the pinor bundle can be
defined if and only if this class vanishes.
We can view the same class appearing in the more traditional approach to defining a spin structure,
to which we now turn.
Spin 2010 (jmf) 33

4.3.3 Spin structures

Let (M, g ) be an orientable riemannian manifold of signature (s, t ) and let SO(M) → M denote the bundle
of oriented orthonormal frames.

Definition 4.10. A spin structure on (M, g ) is a principal Spin(s, t )-bundle Spin(M) → M together with
a bundle morphism
ϕ
Spin(M) � SO(M)
�� �
�� ��
��
�� ���
�
�� �� �
M
� : Spin(s, t ) → SO(s, t ) of Proposition 3.5.
which restricts fibrewise to the covering homomorphism Ad

Spin structures need not exist and even if they do they need not be unique. To understand the ob-
struction let us try to build a spin bundle starting with a trivialisation (U, {g αβ }) of SO(M). We choose
g� � �
αβ (m) ∈ Spin(s, t ) such that under Ad : Spin(s, t ) → SO(s, t ), g αβ (m) �→ g αβ (m). This choice is not
�
�
unique, of course: any other choice g (m) is related to g�
αβ αβ (m) by multiplication with some f αβ (m) ∈
� = Z2 : g�
ker Ad �
(m) = g�
αβ (m) f αβ (m). We would build the spin bundle Spin(M) as usual by
αβ
� ��
? �
Spin(M) = Uα × Spin(s, t ) ∼ with (m, s) ∼ (m, g�
αβ (m)s) for m ∈ Uαβ and s ∈ Spin(s, t ),
α∈I

except that, for this to make sense, the g� αβ (m) should satisfy the cocycle condition. As in the case of
the construction of the pinor bundle, the obstruction is the class of f αβγ (m) = g� � �
αβ (m)g βγ (m)g γα (m) in
2
H (M, Z2 ), which is again the second Stiefel–Whitney class of M. If and only if this class vanishes does
(M, g ) admit a spin structure. Assuming the class vanishes, then one can ask whether the spin structure
is unique. Spin structures are in bijective correspondence with the inequivalent lifts g� αβ of g αβ . As
mentioned above, any two lifts are related by multiplication by f αβ : Uαβ ∈ Z2 . The cocycle conditions of
the two lifts implies the cocycle condition of f αβ , whence it defines a class in H1 (M, Z2 ). If (and only if)
this class is trivial, so that f αβ (m) = f α (m) f β (m) for some f α : Uα → Z2 , do the two lifts yield equivalent
spin bundles. In summary, spin structures are classified by H1 (M, Z2 ) ∼ = Hom(π1 (M), Z2 ), whence it
usually comes down to assigning signs to noncontractible loops consistently.

Remark 4.11. Readers familiar with Čech cohomology will recognise the obstruction of the existence
of a spin structure as the image of the class of SO(M) in H1 (M, SO(s, t )) under the connecting map in the
long exact cohomology sequence

H1 (M, Z2 ) � H1 (M, Spin(s, t )) � H1 (M, SO(s, t )) � H2 (M, Z2 )

coming from the exact sheaf sequence which is induced from the exact sequence of groups

�
Ad
1 � Z2 � Spin(s, t ) � SO(s, t ) �1

Again notice that since SO(s, t ) and Spin(s, t ) are (in general) nonabelian, there are no Hp>1 , whence the
exact cohomology sequence ends there. Indeed a principal SO(s, t )-bundle admits a Spin lift if and only
its image in H2 (M, Z2 ) under the connecting homomorphism vanishes and the “difference” of any two
lifts lives in H1 (M, Z2 ).

Example 4.12. For n ≥ 2, the sphere S n admits a unique spin structure, and indeed Spin(S n ) = Spin(n +
1) and the bundle morphism Spin(S n ) → SO(S n ) is the covering homomorphism Spin(n+1) → SO(n+1).

Example 4.13. The circle S 1 has two inequivalent spin structures which, in some quarters at least, go
by the names of Ramond and Neveu–Schwarz. (This is not a joke.)
Spin 2010 (jmf) 34

Example 4.14. A compact Riemann surface Σ of genus g admits 22g inequivalent spin structures. The
second Stiefel–Whitney class vanishes because if it the reduction mod 2 of the Euler class and the
Euler characteristic is even (= 2 − 2g ). The inequivalent spin structures are classified by homomorph-
isms Hom(π1 (Σ), Z2 ). Now the fundamental group of Σ is generated by 2g elements A1 , . . . , A g , B1 , . . . , Bg
subject to the relation [A1 , B1 ][A2 , B2 ] · · · [A g , Bg ] = 1, where [A, B] = ABA−1 B−1 is the (group-theoretical)
commutator of A, B. Every homomorphism is determined by what it does on generators, subject to the
relation being satisfied. Clearly, though, since Z2 is abelian, any homomorphism from the free group
generated by the Ai and the Bi automatically preserves the relation. Thus every spin structure is spe-
cified by associating a sign to every generator. For the case of genus 1, there are four spin structures
which, in some quarters at least, are called Neveu–Schwarz/Neveu–Schwarz, Neveu–Schwarz/Ramond,
Ramond/Neveu–Schwarz and Ramond/Ramond. (This is not a joke either and moreover illustrates the
multiplicative nature of the spin structures.)

Given a spin structure Spin(M) → M we can now construct spinor bundles as associated vector
bundles. Let S(s, t ) denote a spinor representation of Spin(s, t ) and define S(M) → M to be the vector
bundle with total space
� ��
S(M) = Spin(M) × S(s, t ) Spin(s, t ) .

Depending on signature we might also have half-spinor bundles S ± (M) associated to the half-spinor
representations S(s, t )± .
FIXME: I am not very happy with this lecture. I will eventually update this to include a small
discussion of Čech cohomology with coefficients in a sheaf, to allow me at the very least to use the
language freely.