A (Somewhat Brief) Introduction to Dirac

Operators

Mariano Echeverria
Dirac Operators and Mathematics

Mathematical
Physics
. " -

Clifford Dirac Operator ! Elliptic

Algebras Operators
& # %
Differential
Geometry
The Dirac Equation

In special relativity we have the famous relationship

q
E = px2 + py2 + pz2 + m2 c 4 (1)

In Quantum Mechanics one “quantizes” the previous

equation by turning, E , px , py , pz into differential
operators
@ @ @ @
E ! i @t px ! i @x py ! i @y pz ! i @z

Equation 1 becomes
s
@ @2 @2 @2
i = + m2 c 4
@t @x 2 @y 2 @z 2
Dirac Operator in 1d

Goal: find an operator D such that

d2
D2 =
dx 2
Dirac Operator in 1d

If you know complex numbers the previous problem is not too

difficult, just take
d
D=i
dx
Formal Self-Adjointness

Assume, that f , g 2 C 1 (R, C) have compact support; then

integration by parts says that
´ ⇣ df (x) ⌘
< Df , g >L2 = R
i dx g (x)dx

= (if (x)ḡ (x))|x=1 if (x) d ḡdx(x) dx

´
x= 1 R

f (x) i dg
´
= R dx
dx

= < f , Dg >L2

so the Dirac operator in 1d is formally self-adjoint!

Dirac Operator in 2d

Goal: find an operator D such that

@2 @2
D2 =
@x 2 @y 2
Dirac Operator in 2d

Inspired in the 1d case, we use the following “ansatz”

@ @
D=a +b
@x @y

where a, b are some constants. Since

✓ ◆✓ ◆
2 @ @ @ @ @2 @2 @2
D = a +b a +b = a2 2 +(ab + ba) +b 2 2
@x @y @x @y @x @x@y @y

so we need 8
2
<a = 1
>
ab + ba = 0
>
: 2
b = 1
In particular, a and b must anticommute!
Dirac Operator in 2d

It is not hard to check that

✓ ◆ ✓ ◆
0 i 0 1
a= b=
i 0 1 0

satisfy a2 = b 2 = Id and ab = ba. Therefore, our Dirac

Operator D = a @x @ @
+ b @y in 2d is
✓ ◆
0 @
i @x + @
@y
D= @
i @x @
@y
0

and so
! ✓ ◆
@2 @2
2 @x 2 @y 2
0 @2 @2
D = @2 @2 = Id
0 @x 2 @y 2
@x 2 @y 2
The Cauchy Riemann Operator

We can rewrite the Dirac operator as

0 ⇣ ⌘ 1
1 ✓ ◆
0 2 @x
@
i @
@y 0 @
D = 2i @ ⇣ ⌘ A = 2i @z
1 @ @
+ i @y 0
@
@ z̄
0
2 @x

Therefore
✓ ◆
2 ker D ! is holomorphic and is antiholomorphic
Formal Self-Adjointness

If we write
! ✓ ◆
@ @
0 i @x + @y 0 D̃
D= @ @ =
i @x @y 0 D 0

using Green’s Theorem it can be checked that for f , g 2 C 1 (R2 , C)

< Df , g >L2 =< f , D̃g >L2
so that D̃ is the formal adjoint of D. So we can write
✓ ◆
0 D⇤
D= (2)
D 0
which proves again that the Dirac operator is formally self-adjoint!
Spoiler Alert: in even dimensions we will always have a decomposition
for the Dirac operator like previous one!
Dirac Operator in 3d

Just as we did for one and two dimensions, if we take

@ @ @
D = c1 + c2 + c3
@x @y @z
⇣ 2 ⌘
@2 @2
then D2 = @
@x 2
+ @y 2 + @z 2 if and only if
(
c12 = c22 = c32 = 1
c1 c2 + c2 c 1 = c1 c 3 + c 3 c 1 = c 2 c 3 + c 3 c 2 = 0
The previous system of equations is satisfied by Hamilton’s
Quaternions, i.e, we can take
@ @ @
D=i +j +k
@x @y @z
Dirac Operator on Rn and Clifford Algebras

For Rn , the Dirac Operator (for the standard inner product)

will be
@ @ @
D= 1 + 2 + ··· + n
@x1 @x2 @xn
where the i are the generators for the Clifford Algebra of Rn ,
i.e, (
2
i = 1
{ i , j } = 0 if i =
6 j
What About the Dirac Equation?

We were trying to find D such that

@2 @2 @2 @2
D2 =
@t 2 @x 2 @y 2 @z 2

In fact we can take

0 1
@t 0 @z @x i@y
B 0 @t @x + i@y @z C
D=B @
C
A
@z @x + i@y @t 0
@x i@y @z 0 @t

The Dirac equation was used to predict the existence of

antiparticles!
What About Other Spaces?

What do we mean by the Laplacian on an arbitrary

manifold M? Does it always exist?

What do we mean by the Dirac Operator on an arbitrary

manifold M? Does it always exist?
Ingredient 1: Fourier Transform

Recall that for a sufficiently well behaved function u : R ! R

we can define its Fourier transform Fu : R ! R by
ˆ
Fu(x) = û(p) = e 2⇡ipx u(x)dx
R

Modulo some constants, differentiation becomes multiplication

in that
d û
ˆ
= \
( 2⇡ix) e 2⇡ipx û(p)dp = [ 2⇡ixu(x)] = F ( 2⇡ixu(x))
dp R

2⇡ix·
u ! 2⇡iu
F # #F
d
dp d û
û ! dp
Ingredient 2: Local Coordinates

On a general manifold M we don’t have global coordinates

(think of a sphere), however, we can use local coordinates
x1 , · · · , xn to describe it. In particular, a linear differential
operator L of order m can be described locally as an operator

L : C 1 (M; Rq ) ! C 1 (M; Rp )

given by
X @ ↵1 @ ↵n
Lu = a↵ (x) ↵1 · · · u
@x1 @xn↵n
|↵|m | {z }
D↵

where u = (u1 , · · · , uq ) : M ! Rq and the a↵ (x) are p ⇥ q

matrices
Symbol of a Differential Operator

For our differential operator

X @ ↵1 @ ↵n
Lu = a↵ (x) · · · u
@x1↵1 @xn↵n
|↵|m

we define its (leading) symbol by replacing the partial

derivatives by “momenta”
X
L (x, p) ⌘ a↵ (x)p1↵1 · · · pn↵n
|↵|=m
Example on M = R3

3 2
L : C 1 (M, R
⇣ ) ! C (M, R )
1
⌘
2
@ 2 u2 @ 2 u3
(u1 (x), u2 (x), u3 (x)) ! x1 x2 @@xu21 + @u3
@x1 + @x1 @x3 , @x1 @x3
1

For the (leading) symbol keep the highest order terms so we work with
✓ ◆ ✓ ◆ ✓ 2 ◆
@ 2 u1 @ 2 u2 @ 2 u3 @ 2 u1 @ u2 @ 2 u3
x1 x2 + , = x1 x2 ,0 + ,
@x12 @x1 @x3 @x1 @x3 @x12 @x1 @x3 @x1 @x3

0 1
✓ ◆ ✓ ◆ u1
x1 x2 0 0 @2 0 1 0 @2 @ u2 A
L(u1 , u2 , u3 ) = +
0 0 0 @x12 0 0 1 @x1 @x3
u3
✓ ◆
x1 x2 p12 p1 p3 0
L (x1 , x2 , x3 , p1 , p2 , p3 ) =
0 0 p1 p3
Gradient on M = R3

3
r : C 1 (M, R) 1
⌘ R)
⇣ ! C (M,
@u @u @u
ru = , ,
@x @y @z

20 1 0 1 0 1 3
1 0 0
u(x1 , x2 , x3 ) ! 4@ 0 A @x
@
+ @ 1 A @y
@
+ @ 0 A @z
@ 5
u
0 0 1

Its symbol is
01 0 1 0 1 0 1
1 0 0 p1
r (x, p) =
@ 0 A p1 + @ 1 A p2 + @ 0 A p3 = @ p2 A
0 0 1 p3
Curl on M = R3

(M, R3 ) ! C 1 (M, R3 )
curl : C 1⇣ ⌘
@u3 @u2 @u1 @u3 @u2 @u1
r ⇥ (u1 , u2 , u3 ) = @y @z , @z @x , @x @y

20 1 0 u3 ) ! 0
(u1 , u2 ,1 1 30 1
0 0 0 0 0 1 0 1 0 u1
4@ 0 0 1 A @x
@
+@ 0 0 0 A @y @
+@ 1 0 0 A @z
@ 5@
u2 A
0 1 0 1 0 0 0 0 0 u3

Its symbol is
0 1
0 p3 p2
curl (x, p) =
@ p3 0 p1 A
p2 p1 0
Divergence on M = R3

div : C 1 (M, R3 ) ! C 1 (M, R)

r · (u1 , u2 , u3 ) = @u 1 @u2 @u3
@x + @y + @z

(u1 , u2 , u3 ) ! 0 1
h i u1
1 0 0 @
+ 0 1 0 @
+ 0 0 1 @ @ u2 A
@x @y @z
u3

Its symbol is
div (x, p) = p1 p2 p3
Generalized Laplacian and Dirac Type Operator

Suppose that E is a vector bundle over a Riemannian manifold

M.
A Generalized Laplacian on E is a second order
differential operator 4 such that

4 (x, p) = kpk2

An Operator of Dirac Type on E is a first order

differential operator D such that

D ⇤ D (x, p) = kpk2

i.e, the symbol of its “square” acts in the same way as the
symbol of a generalized Laplacian
An Old Friend

Recall that in R2 our Dirac operator was

✓ @ ◆ ✓ ◆
0 @
i @x + @y 0 D⇤
D= =
@
i @x @
@y
0 D 0

Since

D (x, p) = ip1 p2 D ⇤ (x, p) = ip1 + p2

then
D ⇤ D (x, p) = D ⇤ (x, p) D (x, p) = (ip1 + p2 ) (ip1 p2 ) = p12 p22 = kpk2

so D is an operator of Dirac type.

Vector Operations on M = R3 and de Rham
Cohomology

r r⇥ r·
0 ! CR13 (R) ! CR13 (R3 ) ! CR13 (R) ! CR13 (R) ! 0

From vector calculus we know that

8
<r ⇥ (r ) = 0
> curl(grad)=0

>
:
r · (r ⇥ E) = 0 div(curl)=0

therefore we can define a cohomology HDR

⇤
(R3 ) !
 0
Computing HDR (R3)

r r⇥ r·
0 ! CR13 (R) ! CR13 (R3 ) ! CR13 (R) ! CR13 (R) ! 0

0
HDR (R3 ) = ker r = : R3 ! R | r ⌘ 0
Now, r ⌘ 0 if and only if is a constant scalar field. Since
there is a constant scalar field for each real number we have
that
0
HDR (R3 ) ' R
 1
Computing HDR (R3)

r r⇥ r·
0 ! CR13 (R) ! CR13 (R3 ) ! CR13 (R) ! CR13 (R) ! 0

1 ker(curl)
HDR (R3 ) = = [E] | r ⇥ E = 0 and E ⇠ E0 iff E0 = E + r
im (grad)

From Advanced Calculus we know that every irrotational

vector field E is conservative, i.e, E = r so E 2 [0] and
hence
1
HDR (R3 ) = 0
 2
Computing HDR (R3)

r r⇥ r·
0 ! CR13 (R) ! CR13 (R3 ) ! CR13 (R) ! CR13 (R) ! 0

2 ker(div)
HDR (R3 ) = = {[B] | r · B = 0 and B ⇠ B0 iff B0 = B + r ⇥ A}
im(curl)

From Electromagnetism/Vector Calculus we know that every

solenoidal vector field B has a vector potential A, i.e,
B = r ⇥ A so B 2 [0] and hence
2
HDR (R3 ) = 0
 3
Computing HDR (R3)

r r⇥ r·
0 ! CR13 (R) ! CR13 (R3 ) ! CR13 (R) ! CR13 (R) ! 0

3
HDR (R3 ) = coker(div) = [f ] | f : R3 ! R and f ⇠ g iff f = g + r · B
´x
Since f (x, y , z) = r · 0
f (t, y , z)dt, 0, 0 we have f 2 [0] so
3
HDR (R3 ) = 0
Differential Equations and Topology: de Rham’s
Theorem

De Rham’s Theorem: the de Rham cohomology is isomorphic

to the cohomology of the manifold, i.e,
⇤
HDR (M) ' H ⇤ (M)

In particular we can compute the Euler characteristic of the

manifold as
dim
XM
(M) = ( 1)i dim HDR
i
(M)
i=0

which means that the topology of the space restricts the

dimensions of the space of solutions of certain differential
equations (Laws of Physics) on our manifold!
The Hodge Laplacian

r r⇥ r·
0 CR13 (R) CR13 (R3 ) CR13 (R) CR13 (R) 0
r· r⇥ r

We can define the Hodge Laplacians

40 : CR13 (R) ! CR13 (R) 41 : CR13 (R3 ) ! CR13 (R3 )

40 (f ) = r · (rf ) 41 (u) = r ( r · u) + r ⇥ (r ⇥ u)
⇣ ⌘
@2f @2f @2f
= @x 2
+ @y 2
+ @z 2
= (40 (u1 ), 40 (u2 ), 40 (u3 ))
The Hodge Operator

We can combine them to define a Laplacian on

⇥ ⇤ ⇥ 1 ⇤
⇤
⌦M = CR13 (R) CR13 (R3 ) CR3 (R) CR13 (R3 )

Define the Hodge- Laplacian

⇤ ⇤
✓ 4H :◆⌦M ✓ ! ⌦M ◆
(f , u) (40 f , 41 u)
4H =
(g , v) (40 g , 41 v)
Does it have a Dirac operator?
Observe that
rank⌦M⇤
= 8 = 23 = 2dim M
The Hodge-Dirac Operator

r r⇥ r·
0 CR13 (R) CR13 (R3 ) CR13 (R) CR13 (R) 0
r· r⇥ r
f v u g
0 1 2 3
⇤
D : ⌦M ! ⌦M⇤
✓ ◆ ✓ ◆ ✓ ◆✓
(f , u) ( r · v, rg + r ⇥ v) 0 D⇤ (f , u)
D = =
(g , v) (r · u, rf + r ⇥ u) D 0 (g , v)
(
D(f , u) = (r · u, rf + r ⇥ u)
D ⇤ (g , v) = ( r · v, rg + r ⇥ v)
✓ ◆ ✓ ◆
2 (f , u) (f , u)
D = 4H
(g , v) (g , v)
Therefore we found a square root for the Laplacian!
“Index” Hodge-Dirac Operator on M = R3

We have that
ker D = {(f , u) | r · u = 0 and rf = r ⇥ u}

ker D ⇤ = {(g , v) | r · v = 0 and rg = r ⇥ v}

and (f , u) ! ( f , u) gives a bijection between ker D and
ker D ⇤ .
?
indexD ⌘ dim ker D dim ker D ⇤ = 0
The previous calculation fails because ker D and ker D ⇤ are
infinite dimensional! However, if we run the same argument on
a compact, oriented, Riemannian manifold it can be show that
indexD = (M)
so an analytical quantity (the index) is determined by a
topological quantity (Euler Characteristic)!
Fixing the problem: Fredholm Operators

A bounded linear operator T : E ! F between Banach spaces is

called Fredholm if it has finite dimensional kernel and cokernel. We
can define its index by

indexT ⌘ dim ker T dim cokerT

::::: If E , F are finite dimensional vector spaces then by the

rank-nullity theorem (dim E = dim ker T + dim imT ) the index
is independent of the operator since
indexT = dim E dim F
::::: In infinite dimensions the index can depend on the
operator. For example, in l 2 we have
shift+ (c0 , c1 , c2 , · · · ) = (0, c0 , c1 , c2 , · · · , ) ind shift+ = 1

shift (c0 , c1 , c2 , · · · ) = (c1 , c2 , · · · , ) ind(shift ) = +1

Fixing the problem: Elliptic Operators

An operator D is elliptic if

p 6= 0 ! D (x, p) is invertible

For example, the symbol for D(f , u) = (r · u, rf + r ⇥ u)

is 0 1
0 p1 p2 p3
B p1 0 p3 p2 C
B
D (x, p) = @
C
p2 p3 0 p1 A
p3 p2 p1 0
and since det D (x, p) = (p12 + p22 + p32 )2 we see that
whenever (p1 , p2 , p3 ) 6= (0, 0, 0) the matrix D (x, p) is
invertible, i.e, D is an elliptic operator!
:::: The “wave operator”

@2 @2 @2 @2
@t 2 @x 2 @y 2 @z 2

has symbol

⇤ (x, p) = p02 p12 p22 p32

and it clearly vanishes on the “light cones”

p02 p12 p22 p32 = 0
::::The “heat kernel operator”
@ @2 @2 @2
@t @x 2 @y 2 @z 2

has symbol
H (x, p) = p12 p22 p32
and it clearly vanishes whenever p1 = p2 = p3 = 0 and p0 2 R.
Why do we care about Elliptic Operators on a
compact manifold?

1) The operators of Dirac type are always elliptic

2) Over a compact manifold M, being elliptic implies being
Fredholm
3) Fredholm operators are very stable under perturbations,
which suggests that the index of a Fredholm operator might
be computed via topological quantities
The Atiyah-Singer Theorem

AS
indexD = topological stuff!
where by “topological stuff” we mean certain characteristic
classes associated to the manifold and the vector bundle on
which the Dirac operator acts.
What I Left Out

Harmonic Analysis

"

Lie Groups Dirac Operators ! Noncommutative

and Algebras Geometry
#
K-Theory
Thank you!

Source:
https://commons.wikimedia.org/wiki/File:Dirac%27s_commemorative_marker.jpg