Dirac operators and Spin structures

Paul Baum

notes taken by:

Pawel Witkowski

March 15, 2006

Contents

1 Dirac operators and Spin structures 2

1.1 The Dirac operator of Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Dirac operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Bott generator vector bundle . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Spin representation and Spinc . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Clifford algebras and spinor systems . . . . . . . . . . . . . . . . . . . 10

1
Chapter 1

Dirac operators and Spin structures

1.1 The Dirac operator of Rn

First we consider n even. We shall construct matrices

E1 , E2 , . . . , En , n = 2r

each Ej being 2r ×2r matrix of complex numbers. In fact each entry will be in {0, 1, −1, i, −i}.
Properties of Ej
1. Ej∗ = −Ej ,

2. each Ej is block anti-diagonal  

0 ∗
Ej =
∗ 0
and each block has size 2r−1 × 2r−1 ,

3. Ej2 = I2r ,

4. Ej Ek + Ek Ej = 0 for j 6= k,

5.  
r I2r−1 0
i E1 E2 . . . E n =
0 −I2r−1

We will proceed by induction on n even. For n = 2 we take

   
0 −1 0 i
E1 = , E2 =
1 0 i 0

Suppose we have E1 , E2 , . . . , En of size 2r ×2r . Then we put first n matrices of size 2r+1 ×2r+1
as      
0 E1 0 E2 0 En
, , ...,
E1 0 E2 0 En 0
and two additional matrices
 
  0 0 iI2r−1 0
0 −I2r  0 0 0 iI2r−1 
,  .
I2r 0  iI2r−1 0 0 0 
0 iI2r−1 0 0

2
Example 1.1. For n = 4 we have
   
0 0 0 −1 0 0 0 i
 0 0 1 0   0 0 i 0 
E1 = 
 0 −1
, E2 =  ,
0 0   0 i 0 0 
1 0 0 0 i 0 0 0
   
0 0 −1 0 0 0 i 0
 0 0 0 −1 
  0 0 0 −i 
E3 = 
 1 , E4 =  
0 0 0   i 0 0 0 
0 1 0 0 0 −i 0 0

For n odd, n = 2r + 1, we define matrices E1 , E2 , . . . , Er satisfying

1. Ej∗ = −Ej ,

2. Ej2 = I2r ,

3. Ej Ek + Ek Ej = 0 for j 6= k,

4. ir+1 E1 E2 . . . En = I2r .

First if n = 1 we set
E1 = [−i].
Then for n = 2r + 1 we use 2r matrices E1 , E2 , . . . , En−1 as for the even case and as the last
one we put  
−iI2r−1 0
.
0 iI2r−1
From E1 , E2 , . . . , En we obtain:

1. The Dirac operator of Rn (described above)

2. The Bott generator vector bundle on S n (n even)

3. The spin representation of Spinc (n)

1.1.1 Dirac operator

Now we can define Dirac operator of Rn . For each n we set
n
X ∂
D := Ej .
∂xj
j=1

Example 1.2. For n = 1 we have Dirac operator of R

∂
D = −i .
∂x
For n = 2    
0 −1 ∂ 0 i ∂
D= +
1 0 ∂x1 i 0 ∂x2
.

3
 For n = 2r and n = 2r + 1 D is an unbounded operator on the Hilbert space

L2 (Rn ) ⊕ L2 (Rn ) ⊕ . . . ⊕ L2 (Rn ) .

| {z }
2r

D is a first order elliptic differential operator on

Cc∞ (Rn ) ⊕ Cc∞ (Rn ) ⊕ . . . ⊕ Cc∞ (Rn )

| {z }
2r

With this domain D is symmetric (that is D is formally self-adjoint) and D is essentially

self-adjoint (that is D has unique self-adjoint extension). For n even
 
0 D−
D=
D+ 0

where D− is the formal adjoint of D+ .

We will descirbe these notions in a general context. Let H be Hilbert space. An un-
bounded operator on H is a pair (D, T ) such that

1. D ⊂ H is a vector subspace of H,

2. D is dense in H,

3. T : D → H is a C-linear map,

4. (D, T ) is closeable, i.e. the closure of graph(T ) in H ⊕ H is the graph of a C-linear map

P (graph(T )) → H

P (u, v) = u.

An unbounded operator (D, T ) is symmetric if and only if

hT u, vi = hu, T vi ∀ u, v ∈ D.

For an unbounded operator (D, T ) on H let

D(T ∗ ) := u ∈ H  v 7→ hu, T vi extends from D to H extends

 

to be a bounded linear functional on H}

For u ∈ D(T ∗ ) and v ∈ H there exists

T ∗ : D(T ∗ ) → H

such that
hu, T vi = hT ∗ u, vi.
Now (D, T ) is self-adjoint if and only if

(D, T ) = (D(T ∗ ), T ∗ ).

Remark 1.3. Symmetric operator needs not to be self-adjoint, but a self-adjoint operator is
symmetric.

4
Example 1.4. Take Cc∞ (R) ⊂ L2 (R) and

D = u ∈ L2 (R)  − i du
  2

dx ∈ L (R) in the distribution sense

= {u ∈ L2 (R)  xû ∈ L2 (R)},

where û is the Fourier transform of u and

x : R → R, x(t) = t, ∀ t ∈ R.

Then (Cc∞ (R), −i dx

d d
) has unique self-adjoint extension (D, −i dx ).
Let D be Dirac operator of Rn , n = 2r or 2r + 1.

Ω1 (Rn ) = {C ∞ 1-forms on Rn }

= {f1 dx1 + f2 dx2 + . . . + fn dxn  fj : Rn → C, j = 1, 2, . . . , n}

Ω1 (Rn ) acts on
Cc∞ (Rn ) ⊕ Cc∞ (Rn ) ⊕ . . . ⊕ Cc∞ (Rn )
| {z }
2r
in the following way. Let

ω = f1 dx1 + f2 dx2 + . . . + fn dxn ,

s = (s1 , s2 , . . . , s2r ), sl : Rn → C, l = 1, 2, . . . , 2r .
Then
n
X
ωs = fj Ej s.
j=1

There is following Leibniz rule for D

D(f s) = (df )s + f (Ds),

n
X ∂f
f : Rn → C, f ∈ C ∞ (Rn ), df = dxj .
∂xj
j=1

If M is C ∞ -manifold,
compact or non-compact, with or without boundary, dim M = M ,
then the Dirac operator of M is an elliptic operator which is locally like the Dirac operator
of Rn .

1.1.2 Bott generator vector bundle

Let W be finite dimensional C-vector space,

T ∈ HomC (W, W ), T 2 = −I.

Then eigenvalues of T are ±i and there is decomposition

W = Wi ⊕ W−i ,

Wi = {v ∈ W  T v = iv}

W−i = {v ∈ W  T v = −iv}

5
Assume that n is even, S n ⊂ Rn+1

S n = {(a1 , a2 , . . . , an+1 ) ∈ Rn  a21 + a22 + . . . + a2n+1 = 1}.



We have a map
S n → M (2r , C)
(a1 , a2 , . . . , an+1 ) 7→ a1 E1 + a2 E2 + . . . + an+1 En+1 =: F.
From the properties of Ej we obtain

F 2 = (a1 E1 + a2 E2 + . . . + an+1 En+1 )2

= (−a21 − a22 − . . . − a2n+1 )I

= −I

so the eigenvalues of F are ±i.

The Bott generator vector bundle β on S n is given by

β(a1 ,a2 ,...,an+1 ) := i-eigenspace of F

r 
= {v ∈ C2  F (v) = iv}

For n even and S n ⊂ Rn+1 there is an isomorphism

K0 (S n ) = Z ⊕ Z

1 β

where 1 = S n × C.

1.2 Spin representation and Spinc

Let G be a topological group, Hausdorff and paracompact, X topological space Hausdorff
and paracompact. A principal G-bundle on X is a pair (P, π) where
1. P is a Hausdorff and paracompact topological space with given continuous (right) action
of G
P ×G→P
(p, g) 7→ pg

2. π : P → X is a continuous map, mapping P onto X

such that given any x ∈ X, there exists an open subset U of X with x ∈ U and a homeomor-
phism
ϕ : U × G → π −1 (U )
with
πϕ(u, g) = u ∀ (u, g) ∈ U × G

ϕ(u, g1 g2 ) = ϕ(u, g1 )g2 ∀ (u, g1 , g2 ) ∈ U × G × G

Such ϕ : U × G → π −1 (U ) is referred to as a local trivialization.

6
 Two principal G-bundles (P, π) and (Q, θ) are isomorphic if there exists a G-equivariant
homeomorphism f : P → Q with commutativity in the diagram
f
P - Q

θ
-


X
Let G, H be two topological groups and let (P, π), (G, θ) be a principal G-bundle and H-
bundle on X. A homomorphism of principal bundles from (P, π) to (Q, θ) is a pair (η, ρ) such
that
1. ρ is a homomorphism of topological groups ρ : G → H

2. P → Q is a continuous map with commutativity in the diagrams

η - Q P ×G
η ×-
ρ
P Q×H
π

θ
-

? η ?
X P - Q

πp = θ(ηp) η(pg) = (ηp)(ρg)

A homomorphism of principal bundles on X will be denoted η : P → Q and ρ : G → H will

be referred to as homomorphism of topological groups underlying η.
Lemma 1.5. Let η : P → Q be a homomorphism of principal bundles on X with underlying
homomorphism of topological groups ρ : G → H. Then for any x ∈ X there exists an open
subset U of X with x ∈ U and local trivializations

ϕ : U × G → π −1 (U )

ψ : U × H → θ−1 (U )
such that the diagram
ϕ
U ×G - π −1 (U )
IdU ×η η
ψ
? ?
U ×H - θ−1 (U )
commutes.
Example 1.6. Let E be R-vector bundle on X, dimR (Ep ) = n for all p ∈ X. Denote

∆(E) := {(p, v1 , v2 , . . . , vn )  p ∈ X, v1 , v2 , . . . , vn form a vector space basis for Ep }

∆(E) is topologized by
∆(E) ⊂ |E ⊕ E ⊕
{z. . . ⊕ E} .
n
Define an action
∆(E) × GL(n, R) → ∆(E)

7
 ((p, v1 , v2 , . . . , vn ), [aij ]) 7→ (p, w1 , w2 , . . . , wn ),
X n
wj = aij vi , [aij ] ∈ GL(n, R)
i=1
and a map
θ : ∆(E) → X,
θ(p, v1 , v2 , . . . , vn ) = p.
Then (∆(E), θ) is a principal GL(n, R)-bundle on X.
For n > 3
π1 (SO(n)) = Z/2Z
and Spin(n) is the unique non-trivial 2-fold cover of SO(n). It is a compact connected Lie
group.
Spin(n)

?
SO(n) ⊂GL(n, R)
There is an exact sequence
1 → Z/2Z → Spin(n) → SO(n) → 1
The group Z/2Z embeds in the Spin(n) and S 1 as the {1, −1}. We define
Spinc (n) := S 1 ×Z/2Z Spin(n).
Then there is an exact sequence
1 → S 1 → Spinc (n) → SO(n) → 1
Spinc (n) is a compact connected Lie group
Spin(n)

?
Spinc (n)

?
SO(n) ⊂GL(n, R)
Example 1.7. For n = 1
Spin(1) = Z/2Z, SO(1) = 1
Spinc (1) = S 1
ρ : S 1 → pt .
For n = 2
Spin(2) = S 1 = SO(2)
Spin(2) → SO(2)
ζ 7→ ζ 2
and
Spinc (2) = S 1 ×Z/2Z Spin(2)
ρ(λ, ζ) = ζ 2 .

8
Remark 1.8. Since SO(n) ⊂ GL(n, R) we can view the standard map Spinc (n) → SO(n) as
Spinc (n) → GL(n, R).
Definition 1.9. A Spinc datum for an R-vector bundle E → X is a homomorphism of
principal bundles
η : P → ∆(E),
where P is a principal Spinc (n)-bundle on X (n = dimR (Ep )) and the homomorphism of
topological groups underlying η is the standard map
ρ : Spinc (n) → GL(n, R).
Two Spinc data η : P → ∆(E), η 0 : P 0 → ∆(E) are isomorphic if there exists an isomor-
phism f : P → P 0 of principal Spinc (n)-bundles on X with commutativity in the diagram
f - P0
P

η0 η = η 0 ◦ f.
η
-

∆(E)
Two Spinc data η : P → ∆(E), η 0 : P 0 → ∆(E) are homotopic if there exists a principal
Spinc (n)-bundle Q on X and a continuous map
Φ : Q × [0, 1] → ∆(E)
such that
1. For t ∈ [0, 1] each
Φt = Φ(−, t) : Q → ∆(E)
is a Spinc data.
2.
Φ0 : Q → ∆(E) is isomorphic to η : P → ∆(E)
Φ1 : Q → ∆(E) is isomorphic to η 0 : P → ∆(E)

Definition 1.10. A Spinc (n)-structure for E is an equivalence class of Spinc (n) data, where
the equivalence relation is homotopy.
A Spinc structure for an R-bundle E determines an orientation of E. Let w1 (E), w2 (E), . . . be
the Stiefel-Whitney classes of E, wj (E) Hj (X; Z/2Z)-Cech cohomology. Then E is orientable
if and only if w1 (E) = 0.
A spin manifold is a C ∞ manifold M , dim M = n, for which the structure group of the
tangent bundle T M has been lifted from GL(n, R) to Spin(n). Such lifting is possible if and
only if
w1 (M ) = 0, w1 (M ) ∈ H1 (M ; Z/2Z)
and
w2 (M ) = 0, w2 (M ) ∈ H2 (M ; Z/2Z).
A Spinc manifold is a C ∞ manifold M , dim M = n, for which the structure group of
the tangent bundle T M has been lifted from GL(n, R) to Spinc (n). Such lifting is possible if
and only if
w1 (M ) = 0, w1 (M ) ∈ H1 (M ; Z/2Z)
and
w2 (M ) is in the image of H2 (M ; Z) → H2 (M ; Z/2Z).

9
 Various well known structures on a manifold M make M into Spinc manifold
(complex analitic)
w
w
w

(symplectic) =====⇒ (almost complex)
w
w
w

(contact) ===⇒ (stably almost complex)
w
w
w

c
Spin ==============⇒ Spin
w
w
w
w

(oriented)
A Spinc manifold can be thought of as an oriented manifold with a slight extra bit of struc-
ture. Most of the oriented manifolds which occur in practice are Spinc manifolds. Spinc
structures behave very much like orientations. For example, an orientation on two of three R
vector bundles in a short exact sequence determine an orientation on the third vector bundle.
Analogous assertions are true for Spinc structures.
Lemma 1.11 (Two out of three lemma). Let

0 → E 0 → E → E 00 → 0

be an exact sequence of R vector bundles on X. If Spinc structures are given for any two of
E 0 , E, E 00 then a Spinc structure is determined for the third.
Corollary 1.12. If M is a Spinc manifold with boundary ∂M , then ∂M is in canonocal way
a Spinc manifold.
Proof. There is an exact sequence

0 → T ∂M → T M |∂M → ∂M × R → 0

Remark 1.13. If E is orientable (w1 (E) = 0), then the set of all possible orientations of E
is in 1-1 correspondence with H0 (X; Z/2Z). If E is Spinc -able (w1 (E) = 0 and w2 (E) ∈
im(H2 (X; Z) → H2 (X; Z/2Z))), then the set of all possible Spinc -structures for E is then in
1-1 correspondence with H0 (X; Z/2Z) × H2 (X; Z).

1.2.1 Clifford algebras and spinor systems

Let V be a finite dimensional R-vector space, h−, −i a positive defninite, symmetric, bilinear
R-valued inner product on V . We can form a tensor algebra

T V := R ⊕ V ⊕ (V ⊗ V ) ⊕ (V ⊗ V ⊗ V ) ⊕ . . .

with multiplication given by composing the tensors, and then define Clifford algebra

Cliff(V ) := T V /(v ⊗ v + hv, vi · 1)

10
where (v ⊗ v + hv, vi · 1) denotes the two-sided ideal in T V generated by all elements of the
form
v ⊗ v + hv, vi · 1, v ∈ V, 1 ∈ R.
As a vector space over R Cliff(V ) is canonically isomorphic to the exterior algebra

Λ∗ V = R ⊕ V ⊕ Λ2 V ⊕ . . . Λn V, n = dimR V.

Let e1 , e2 , . . . , en be an orthonormal basis of V . The monomials

e11 e22 . . . enn , j ∈ {0, 1}

form a vector space basis of Cliff(V ). The canonical isomorphism of R-vector spaces

Cliff(V ) → Λ∗ V

is given by
e11 e22 . . . enn 7→ e11 ∧ e22 ∧ . . . ∧ enn .
This isomorphism does not depend on the choice of orthonormal basis of V .

dimR (Cliff(V )) = 2n , n = dimR V.

In Cliff(V ) we have following identities

e2j = −1, j = 1, 2, . . . , n,

ei ej + ej ei = 0, i 6= j.
We can introduce Z/2Z-grading on Cliff(V ) in the following way

Cliff(V ) = (Cliff(V ))0 ⊕ (Cliff(V ))1 ,

where (Cliff(V ))0 is an R-vector space spanned by e11 e22 . . . enn with 1 + 2 + . . . + n even,
and (Cliff(V ))1 is an R-vector space spanned by e11 e22 . . . enn with 1 + 2 + . . . + n odd. This
Z/2Z-grading does not depend on the choice of orthonormal basis of V .
Take Rn with the usual inner product

S n−1 ⊂ Rn ⊂ Cliff(Rn ).

The elements of S n−1 are invertible in Cliff(Rn ). Let Pin(n) be the subgroup of the invertible
elements of Cliff(Rn ) generated by S n−1 . Then

Spin(n) = Pin(n) ∩ (Cliff(Rn ))0

ρ : Spin(n) → SO(n)
(ρg)(x) = gxg −1 , g ∈ S n−1 , x ∈ Rn .
For n > 3 this is the unique non-trivial 2-fold covering space of SO(n).
Consider complexification

Cliff C (V ) := C ⊗R Cliff(V ).

Then Cliff C (V ) is a C ∗ -algebra with

v ∗ = −v

11
for
v ∈ V ⊂ Cliff(V ) ⊂ Cliff C (V ).
Let
Cliff C (Rn ) := CR Cliff(Rn ),
Spinc (n) = S 1 ×Z/2Z Spin(n) ⊂ Cliff C (Rn ).
Then Spinc (n) is a subgroup of the group of unitary elements of the C ∗ -algebra Cliff C (Rn ).
Let us now choose an orthogonal basis e1 , e2 , . . . , en for even-dimensional R-vector space
V , n = 2n = dimR (V ). Recall 2r × 2r matrices E1 , E2 , . . . , En defined in the beginning of the
chapter and then define a mapping

Cliff C (V ) → M (2r , C)

ej 7→ Ej , j = 1, 2, . . . , n.
This gives an isomorphism of C ∗ -algebras Cliff C (V ) and M (2r , C). For an odd dimension
n = 2r + 1 recall 2r × 2r matrices E1 , E2 , . . . , En and define two mappings

ϕ+ : Cliff C (V ) → M (2r , C)

ϕ+ (ej ) = Ej , j = 1, 2, . . . , n,
ϕ− : Cliff C (V ) → M (2r , C)
ϕ− (ej ) = −Ej , j = 1, 2, . . . , n.
Then
ϕ+ ⊕ ϕ− : Cliff C (V ) → M (2r , C) ⊕ M (2r , C)
is an isomorphism of C ∗ -algebras.
Remark 1.14. This isomorphisms are non-canonical since they depend on the choice of an
orthonormal basis for V .
Let E be an R-vector bundle on X. Assume given an inner product h−, −i for E. Then
define Cliff C (E) as a bundle of C ∗ -algebras over X whose fiber at p ∈ X is Cliff C (Ep ).

Definition 1.15. An Hermitian module over Cliff C (E) is a complex vector bundle F on
X with a C-valued inner product (−, −) and a module structure

Cliff C (E) ⊗ F → F

such that

1. (−, −) makes Fp into a finite dimensional Hilbert space,

2. for each p ∈ X, the module map

Cliff C (Ep ) → L(Fp )

is a unital homomorphism of C ∗ -algebras.

Remark 1.16. Of course all structures here are assumed to be continuous. If X is a C ∞

manifold then we could take everything to be C ∞ .

12
 If E is oriented define a section ω of Cliff C (E) as follows. Given p ∈ X, choose a positively
oriented orthonormal basis e1 , e2 , . . . , en of Ep . For n even, n = 2r, set

ω(p) = ir e1 e2 . . . e2r .

For n = 2r + 1 odd
ω(p) = ir+1 e1 e2 . . . e2r+1 .
Then ω(p) does not depend on the choice of positively oriented orthonormal basis. In
Cliff C (Ep ) we have
(ω(p))2 = 1.
If n is odd, then ω(p) is in the center of Cliff C (Ep ). Note that to define ω, E must be oriented.
Reversing the orientation will change ω to −ω.

Definition 1.17. Let E be an R-vector bundle on X. A Spinor system for E is a triple

(, h−, −i, F ) such that

1.  is an orientation of E,

2. h−, −i is an inner product for E,

3. F is an Hermitian module over Cliff C (E) with each Fp an irreducible module over
Cliff C (Ep ),

4. if n = dimR (Ep ) is odd, then ω(p) acts identically on Fp .

Remark 1.18. The irreducibillity of Fp in (3) is equivalent to dimC (Fp ) = 2r , where n = 2r or

n = 2r + 1. In (4) note that ω(p)2 = 1 so for n odd ω(p) is in the center of Cliff C (Ep ). Hence
irreducibility of Fp implies that ω(p) acts either by I or −I on Fp . Thus (4) normalizes the
matter by requiring that ω(p) acts as I. When n = dimR (Ep ) is even no such normalization
is made.
If (, h−, −i, F ) is a Spinor system for E, then F is referred to as the Spinor bundle.
Suppose that n = dimR (Ep ) is even. Let Fp+ (Fp+ ) be the +1 (−1) eigenspace of ω(p).
We have a direct sum decomposition

F = F + ⊕ F −,

where F + , F − are 21 − Spin bundles. F + (F − ) is a vector bundle of positive (negative)

spinors.
Assume we have right and left actions of the group G on topological spaces X, Y

X ×G→X

G×Y →Y
Then
X ×G Y := X × Y / ∼, (xg, y) ∼ (x, gy).
Example 1.19. Let E be an R-vector bundle on X. Then

∆(E)×GL(n,R) ' E

((p, v1 , v2 , . . . , vn ), (a1 , a2 , . . . , an )) 7→ a1 v1 + a2 v2 + . . . + an vn .

13
 Let E be an R-vector bundle on X. A Spinc datum
η : P → ∆(E)
determines a Spinor system (, h−, −i, F ) for E. For p ∈ X, given orientation , and inner
product h−, −i, an R-basis v1 , v2 , . . . , vn of Ep is positively oriented and orthonormal if and
only if
(v1 , v2 , . . . , vn ) ∈ im(η).
The Spinor bundle for n = 2r or n = 2r + 1
r
F = P ×Spinc (n) C2 .
r
We have to describe how Spinc (n) acts on C2 . For n odd Spinc (n) has an irreducible
representation known as its spin representation
Spinc (n) → GL(2r , C), n = 2r + 1.
For n even Spinc (n) has two irreducible representations known as its 1
2 − Spin representations
Spinc (n) → GL(2r−1 , C),
Spinc (n) → GL(2r−1 , C), n = 2r.
The direct sum
Spinc (n) → GL(2r−1 , C) ⊕ GL(2r−1 , C) ⊂ GL(2r , C),
of these representations is the spin representation of Spinc (n).
Consider Rn with its usual inner product and usual orthonormal basis e1 , e2 , . . . , en
ϕ : Cliff C (Rn ) → M (2r , C)
ϕ(ej ) = Ej , j = 1, 2, . . . , n.
There is a canonical inclusion
Spinc (n) ⊂ Cliff C (Rn )
and ϕ restricted to Spinc (n) maps Spinc (n) to 2r × 2r unitary matrices
Spinc (n) → U(2r ) ⊂ GL(n, C).
r
This is Spin representation of Spinc (n) and Spinc (n) acts on GL(2r , C) acts on C2 via
this representation.
Let M be C ∞ manifold, possibly ∂M non-empty, T M the tangent bundle of M . Then
Spinc datum for T M
 

η : P → ∆(T M )
 ↓ 
Spinor system for T M
(, h−, −i, F )
 ↓ 
Dirac operator
D : Cc∞ (M, F ) → Cc∞ (M, F )
where F is the Spinor bundle on M and Cc∞ (M, F ) are its C ∞ sections with compact support.
The Dirac operator
D : Cc∞ (M, F ) → Cc∞ (M, F )
is such that

14
 1. D is C-linear
D(s1 + s2 ) = Ds1 + Ds2 ,
D(λs) = λDs, s1 , s2 , s ∈ Cc∞ (M, F ), λ ∈ C.

2. If f : M → C is a C ∞ function, then

D(f s) = (df )s + f (Ds).

3. If s1 , s2 ∈ Cc∞ (M, F ) then

Z Z
(Ds1 (x), s2 (x))dx = (s1 (x), Ds2 (x))dx
M M

4. If dim M is even, then D is off-diagonal

F = F+ ⊕ F−
0 D−
 
D=
D+ 0

D : Cc∞ (M, F ) → Cc∞ (M, F ) is an elliptic first-order differential operator. It can be viewed
as an unbounded operator on the Hilbert space L2 (M, F ) with the scalar product
Z
(s1 , s2 ) := (s1 (x), s2 (x))dx.
M

Moreover it is a symmetric operator.

One proves existence of D by constructing it locally and patching together with a C ∞
partition of unity. The uniqueness of D is obtained by the fact that if D0 , D1 satisfy conditions
(1)-(4) above, then
D0 − D1 : F → F
is a vector bundle map, hence D0 , D1 differ by lower order terms.
Example 1.20. Let n be even, S n ⊂ Rn+1 , D-Dirac operator of S n , F -Spinor bundle of S n ,
F = F + ⊕ F −.
D : Cc∞ (S n , F ) → Cc∞ (S n , F )
0 D−
 
D=
D+ 0
D+ : Cc∞ (S n , F + ) → Cc∞ (S n , F − )
Then
Index(D+ ) := dimC (ker D+ ) − dimC (coker D+ ).
Theorem 1.21.
Index(D+ ) = 0.
We can tensor D+ with the Bott generator vector bundle β from section (1.1.2)

Dβ+ : Cc∞ (S n , F + ⊗ β) → Cc∞ (S n , F − ⊗ β).

Then we have
Theorem 1.22.
Index(Dβ+ ) = 1.

15