The Spinor Representation

Math G4344, Spring 2012

As we have seen, the groups Spin(n) have a representation on Rn given

by identifying v ∈ Rn as an element of the Clifford algebra C(n) and having
g̃ ∈ Spin(n) ⊂ C(n) act by
v → g̃vg̃ −1
This is also a SO(n) representation, the fundamental representation on vectors.
Replacing v in this formula by an arbitrary element of C(n) we get a represen-
tation of Spin(n) (and also SO(n)) on C(n) which can be identified with its
representation on Λ∗ (Rn ). This representation is reducible, the decomposition
into irreducibles is just the decomposition of Λ∗ (Rn ) into the various Λk (Rn )
for k = 0, · · · , n. The k = 0 and k = n cases give the trivial representation, but
we get fundamental irreducible representations for k = 1, · · · , n − 1.
The fundamental irreducible representations that are missed by this con-
struction are called the spinor representations. They are true representations
of Spin(n), but only representations up to sign (projective representations) of
SO(n). For the even case of Spin(2n), we will see that there are two different
irreducible half-spinor representations of dimension 2n−1 each, for the odd di-
mensional case there is just one irreducible spinor representation, of dimension
2n .
In terms of the Dynkin diagrams of these groups, for SO(2n + 1), the spinor
representation corresponds to the node at one end connected to the rest by a
double bond. For SO(2n) the two half-spin representations correspond to the
two nodes that branch off one end of the diagram.

1 Spinors
We will be constructing spinor representations on complex vector spaces using
the Clifford algebra and our first step is to consider what happens when one
complexifies the Clifford algebra. The complex Clifford algebras will turn out to
have a much simpler structure than the real ones, with a periodicity of degree
2 rather than degree 8 as in the real case.
Definition 1 (Complex Clifford Algebra). The complex Clifford algebra C C (V, Q)
is the Clifford algebra constructed by starting with the complexified vector space
V ⊗R C, extending Q to this by complex-linearity, then using the same definition
as in the real case. If we start with a real vector space V of dimension n, this
will be denoted C C (n).
One can easily see that C C (n) = C(n) ⊗ C. The construction of the spin
representation as invertible elements in C(n) can also be complexified, producing
a construction of Spin(n, C) (the complexification of Spin(n)) as invertible
elements in C C (n). Note that the the various real forms Spin(p, q) all are
inside C C (n), but to recover them one needs to keep track of an extra piece of

1
structure. When the complexified Clifford algebra comes is the Clifford algebra
of V ⊗R C, the complexification of a real vector space V , one has a conjugation
operation
C : V ⊗R C → V ⊗R C
that acts by complex conjugation of the C factor and leaves invariant the real
subspace V ⊗ 1.
We will study the structure of the algebras C C (n) by an inductive argument.
To begin the induction, recall that

C(1) = C, C(2) = H

so
C C (1) = C(1) ⊗R C = C ⊕ C
and
C C (2) = C(2) ⊗R C = H ⊗R C = M (2, C)
Theorem 1.

C C (n + 2) = C C (n) ⊗C C C (2) = C C (n) ⊗C M (2, C)

Using the cases n = 1, 2 to start the induction, one finds

Corollary 1. If n = 2k

C C (2k) = M (2, C) ⊗ · · · ⊗ M (2, C) = M (2k , C)

where the product has k factors, and if n = 2k + 1,

C C (2k + 1) = C C (1) ⊗ M (2k , C) = M (2k , C) ⊕ M (2k , C)

Proof of Theorem: Choose generators h1 , h2 of C(2), f1 , · · · , fn of C(n) and

e1 , · · · en+2 of C(n + 2). Then the isomorphism of the theorem is given by the
following map of generators
e1 → 1 ⊗ h1
e2 → 1 ⊗ h2
e3 → if1 ⊗ h1 h2
···
en+2 → ifn ⊗ h1 h2
One can check that this map preserves the Clifford algebra relations and is
surjective, thus an isomorphism of algebras.
From now we’ll concentrate on the even case n = 2k. In this case we have
seen that the complexified Clifford algebra is the algebra of 2k by 2k complex
matrices. A spinor space S will be a vector space that these matrices act on:

2
Definition 2 (Spinors). A spinor module S for the Clifford algebra C C (2k) is
given by a choice of a 2k dimensional complex vector space S, together with an
identification C C (2k) = End(S) of the Clifford algebra with the algebra of linear
endomorphisms of S.
So a spinor space is a complex dimensional vector space S, together with a
choice of how the 2k generators ei of the Clifford algebra act as linear operators
on S.
To actually construct such an S, together with appropriate operators on
it, we will use exterior algebra techniques. We’ll begin by considering the real
exterior algebra Λ∗ (Rn ). Associated to any vector v there is an operator on
Λ∗ (Rn ) given by exterior multiplication by v

v∧ : α → v ∧ α

If one has chosen an inner product on Rn there is an induced inner product

on Λ∗ (Rn ) (the one where the wedge products of orthonormal basis vectors
are orthonormal). With respect to this inner product, the operation v∧ has an
adjoint operator, vb.
Physicists have a useful notation for these operations, in which

a† (v) = v∧, a(v) = vb

and
a† (ei ) = a†i , a(ei ) = ai
The algebra satisfied by these operators is called the algebra of Canonical An-
ticommutation Relations (CAR)
Definition 3 (CAR). For each positive integer n, there is an algebra called
the algebra of Canonical Anticommutation Relations, with 2n generators ai , a†i ,
i = 1, · · · , n satisfying the relations

{ai , aj } = {a†i , a†j } = 0

{ai , a†j } = δij

As we have seen, this algebra can be represented as operators on Λ∗ (Rn ).
One can identify this algebra with the Clifford algebra C(n) as follows

v ∈ C(n) → a† (v) − a(v)

since one can check that

v 2 = (a† (v) − a(v))2 = (a† (v))2 + (a(v))2 − {a† (v), a(v)} = −||v||2 1

Specializing for simplicity to the case n = 2k, one can complexify Λ∗ (Rn )
and get a complex representation of C C (2k) on this space. This representation
is of dimension 22k , so it is not the 2k dimensional irreducible representation

3
on a spinor space S. Some way must be found to pick out a single irreducible
representation S from the reducible representation on Λ∗ (Rn )⊗C. This problem
is a bit like the one faced in the Borel-Weil approach to the representations of
compact Lie groups, where we used a complex structure to pick out a single
irreducible representation. Here we will have to use a similar trick.
If we pick a complex structure J on V = R2k , we have a decomposition

V ⊗ C = WJ ⊕ W̄J

Where WJ is the +i eigenspace of J, W̄J is the −i eigenspace. We will pick an

orthogonal complex structure J, i.e. one satisfying

< Jv, Jw >=< v, w >

Definition 4 (Isotropic Subspace). A subspace W of a vector space V with

inner product < ·, · > is an isotropic subspace of V if

< w1 , w2 >= 0 ∀ w1 , w2 ∈ W

and one has

Claim 1. The subspaces WJ and W̄J are isotropic subspaces of V ⊗ C.
This is true since, for w1 , w2 ∈ WJ ,

< w1 , w2 >=< Jw1 , Jw2 >=< iw1 , iw2 >= − < w1 , w2 >

Since WJ is isotropic, the Clifford subalgebra C(WJ ) ⊂ C C (2k) generated

by elements of WJ is actually the exterior algebra Λ∗ (WJ ) since on WJ the
quadratic form coming from the inner product is zero, and the Clifford algebra
for zero quadratic form is just the exterior algebra. In a similar fashion C(W̄J )
can be identified with Λ∗ (W̄J ). There are various ways of setting this up, but
what we plan to do is to construct a spinor space S, as, say Λ∗ (WJ ). This has
the right dimension (2k ) and we just need to show that C C (2k) can be identified
with the algebra of endomorphisms of this space.
Later on we’ll examine the dependence of this whole set-up on the choice
of complex structure J, but for now we will simply pick the standard choice of
complex structure, identifying a complex basis wj = e2j−1 +ie2j for j = 1, · · · , k,
with these vectors orthogonal with respect to the standard Hermitian structure.
Claim 2.
C C (2k) = End(Λ∗ (Ck ))
Using the creation and annihilation operator notation for operators on Λ∗ (Ck ),
we can identify Clifford algebra generators with generators of the CAR algebra
as follows (for j = 1, · · · , k)

e2j−1 = a†j − aj

e2j = −i(a†j + aj )

4
One can check that the CAR algebra relations imply the Clifford algebra rela-
tions
{ei , ej } = −2δij
The complex basis elements wj ∈ WJ of the complexified Clifford algebra have
adjoints wj ∈ WJ .
With this explicit model S = Λ∗ (Ck ) and the explicit Clifford algebra action
on it, one can see how elements of spin(2k) act on S, and thus compute the char-
acter of S as a Spin(2k) representation. The k commuting elements of spin(2k)
that generate the maximal torus are the 12 e2j−1 e2j . In the representation on
Λ∗ (Ck ) they are given by
1 1 1
e2j−1 e2j = −i (a†j − aj )(a†j + aj ) = i [aj , a†j ]
2 2 2
The eigenvalues of [aj , a†j ] on Λ∗ (Ck ) are ±1, depending on whether or not
the basis vector ej is in the string of wedge products that makes up the eigen-
vector. The weights of the the representation S are sets of k choices of ± 21
1 1
(± , · · · , ± )
2 2
In this normalization, representations of Spin(2k) that are actually representa-
tions of SO(2k) have integral weights, the ones that are just representations of
Spin(2k) have half-integral weights.
The decomposition into half-spin representations S + and S − corresponds
to the decomposition into weights with an even or odd number of minus signs.
With a standard choice of positive roots, the highest weight of one half-spin
representation is
1 1 1 1
(+ , + , · · · , + , + )
2 2 2 2
and for the other it is
1 1 1 1
(+ , + , · · · , + , − )
2 2 2 2
Note that these weights are not the weights of the representation of U (k) ⊂
SO(2k) on Λ∗ (Ck ). If one looks at how the Lie algebra of the maximal torus
there acts on Λ∗ (Ck ), one finds that the weights are all 0 or 1. So the weights
of Λ∗ (Ck ) are the same as the weights of S, but shifted by an overall factor
1 1 1
(− , − , · · · , − )
2 2 2
In other words, as a representation it of the maximal torus of Spin(2k) it is best
to think of S as not Λ∗ (Ck ), but as
1
S = Λ∗ (Ck ) ⊗ (Λk (Ck ))− 2

This expression makes sense not for U (k) ⊂ SO(2k), but for its double-cover
]
U (k) and its maximal torus Te.

5
 This overall factor disappears if you just consider the projective representa-
tion on P (S) (complex lines in S), since then

P (S) = P (Λ∗ (Ck ))

This is a reflection of the general point that by trying to construct S only

knowing End(S) = C C (2k), so we can only canonically construct P (S), not S
itself. Multiplying S by a scalar doesn’t affect End(S).

2 Complex Structures and Borel-Weil for Spinors

The construction of S given above used a specific choice of complex structure
J. In general, a choice of J allows us to write

R2k ⊗ C = WJ ⊕ W̄J

and construct S as Λ∗ (WJ ). Recall that to understand its transformation prop-

erties under the maximal torus of Spin(n), it is better to think of
1
S = Λ∗ (WJ ) ⊗ (Λk (WJ ))− 2

In the language of the CAR algebra, one can think of S as being generated
by the operators a†j and aj acting on a “vacuum vector” ΩJ . ΩJ depends on
the choice of complex structure, and transforms under the maximal torus of
1
Spin(2k), transforming as (Λk (W̄J )) 2 .
Given the spinor representation S, one can think of ΩJ as a vector in S
satisfying
w̄ · ΩJ = 0 ∀w̄ ∈ W̄J
This relation defines ΩJ up to a scalar factor and sets up an identification be-
tween orthogonal complex structures J and lines in S. Not all elements of P (S)
correspond to possible J’s. The ones that do are said to be the lines generated
by “pure spinors”. This correspondance between pure spinors and orthogonal
complex structures J could equivalently be thought of as a correspondance be-
twen pure spinors and isotropic subspaces of R2k ⊗ C of maximal dimension
(the W̄J ).
Once one has made a specific choice of J, say J0 , this defines a model for
the spinor space Λ∗ (WJ0 ) and in this model ΩJ0 = 1. If one changes complex
structure from J0 to a J such that

WJ = graph(S : WJ0 → W̄J0 )

then one has identified J with a map S ∈ WJ0 × W̄J∗0 . Using the inner product
and the fact that WJ is isotropic one can show that S ∈ Λ2 (WJ0 ). One can
then show that taking
1
ΩJ = e 2 S
solves the equation
w̄ · ΩJ = 0 ∀w̄ ∈ W̄J

6
for this model of S. For more details about this construction, see [3] Chapter
12.
The space of all orthogonal complex structures on R2k can be identified
with the homogeneous space O(2k)/U (k). An orthogonal complex structure J
is explicitly represented by an orthogonal map J ∈ O(2k). This space has two
components, corresponding to a choice of orientation. We’ll mainly consider
the component J ∈ SO(2k). For any such J, the subgroup that preserves the
decompostion of R2k ⊗C into eigenspaces of J (and thus preserves that complex
structure) is a subgroup isomorphic to U (k).
The space of orientation-preserving orthogonal complex structures can be
thought of as
SO(2k)/U (k) = Spin(2k)/Ũ (k)
There is a complex line bundle L over this space whose fiber above J is the com-
1
plex line in P (S) generated by ΩJ . The fact that ΩJ transforms as (Λk (WJ ))− 2
corresponds to the global fact that as line bundles

L ⊗ L = det−1

where det is the line bundle whose fiber over J is Λk (WJ ).

The space Spin(2k)/Ũ (k) is one of the generalized flag manifolds that occurs
if one develops the Borel-Weil picture for the compact group Spin(2k). It is a
complex manifold, with a description as a quotient of complex Lie groups

SO(2k)/U (k) = Spin(2k)/Ũ (k) = Spin(2k, C)/P

where P is a parabolic subgroup of Spin(2k, C). The line bundle L is a holo-

morphic line bundle, and corresponds to the Borel-Weil construction of a line
bundle with highest weight
1 1 1 1
(+ , + , · · · , + , + )
2 2 2 2
This is the holomorphic line bundle corresponding to one of the half-spinor
representations S + and the Borel-Weil theorem gives a construction of S + as

S + = Γhol (L)

Unlike the construction in terms of the complex exterior algebra, this construc-
tion doesn’t depend upon a fixed choice of complex structure, but uses the global
geometry of the space of all complex structures.

3 Other Constructions of Spinors and References

There are several other possible approaches to constructing spinor representa-
tions, These include

7
 • One could of course just use the general theory for compact Lie groups,
which includes constructions of representations in terms of Verma modules
or geometrically as in Borel-Weil-Bott. An example of a textbook that
works out properties of spinors just using roots, weights, the Weyl group,
etc. is [1].
• One can work with a split real form of Q of signature (k, k), and then
complexify. This has the advantage that the real Clifford algebra is just
C(k, k) = M (2k , R)
and the vector space V can be written as W ⊕ W ∗ , with W a real k-
dimensional isotropic subspace and W ∗ its dual, also isotropic. One then
has
C(W ) = Λ∗ (W ), C(W ∗ ) = Λ∗ (W ∗ )
and can realize the spinor representation on Λ∗ (W ) with C(W ) acting by
exterior multiplication, C(W ∗ ) by the adjoint operation. For examples of
where this construction is carried out, see [2] and [4], as well as [5].
The problem with this is that we are actually interested not in the split
real form Spin(k, k) of the spin group, but the compact form Spin(2k),
so need to somehow recover that structure using a conjugation operator.
We will soon turn to an analog of the spin story using antisymmetric
forms and symplectic groups. In that case, the analog construction for
the split form is just the so-called Schrödinger representation, and there
it is representations of the split form that we will be interested in. The
construction we have used here is the analog of the Fock representation
we will describe in the symplectic case.
• One can think of what we are doing is trying to construct minimal left
ideals of the algebra C C (n), and do this by finding an element I so that
our spinors gets constructed as the left action of C C (n) on C C (n)/I. An
example of a book treating the subject from this point of view is [6].
• By analogy with what we did in the Peter-Weyl/Borel-Weil case, trying
to identify a representation V when what we had was End(V ) = V ⊗
V ∗ , one can use the right Clifford action to impose conditions that pick
out a unique one-dimensional space in V ∗ , say the highest-weight space.
Knowing the highest weight of the representation and how the maximal
torus acts in terms of quadratic Clifford elements, this can be done by
constucting k such elements
Qj = e2j−1 ej ⊗ i ∈ C C (2k)
which satisfy Q2j = 1 and imposing the k conditions picking out the
eigenspace where they act by +1.
Besides the references given in the notes on Clifford algebras and spin groups,
another good reference for this material, from a point of view close to the one
taken here, is [7]. For another point of view, including some discussion of the
relation to the Heisenberg algebra, see [?].

8
References
[1] Bump, D., Lie Groups, Springer-Verlag, 2004.
[2] Fulton, W., Harris, J., Representation Theory: A First Course, Springer-
Verlag, 1991.
[3] Segal, G., and Pressley, A., Loop Groups, Oxford University Press, 1986.

[4] Symmetry, Representations and Invariants, GTM 255 Springer-Verlag,

2009.
[5] Meinrenken, E., Clifford Algebras and Lie Groups Course notes at
http://www.math.toronto.edu/mein/teaching/clif main.pdf
[6] Lounesto, P., Clifford Algebras and Spinors Cambridge University Press,
1997.
[7] Freed, D., Geometry of Dirac Operators Course notes at
http://www.ma.utexas.edu/users/dafr/DiracNotes.pdf
[8] Deligne, P., Notes on Spinors, in Quantum Fields and Strings: A Course
for Mathematicians, AMS, 1999.