Spin 2010 (jmf) 11

Lecture 2: Clifford algebras: the classification

The small part of ignorance that we arrange and classify
we give the name of knowledge.
— Ambrose Bierce

In this section we will classify finite-dimensional real and complex Clifford algebras. Useful ref-
erences are [ABS64] (which treats only the positive- and negative-definite cases, albeit lucidly) and
[LM89, Har90].

2.1 A less-than-useful classification

We start with a result which is interesting in its own right, but perhaps not as useful as it may appear at
first.
Given two quadratic vector spaces (V, QV ) and (W, QW ), we can form their orthogonal direct sum
(V ⊕W, QV ⊕QW ). (Notice that although the direct sum is the coproduct in the category of vector spaces,
it is not the coproduct in QVec.) One natural question is whether the Clifford algebra C�(V⊕W, QV ⊕QW )
of the orthogonal direct sum is related to the Clifford algebras C�(V, QV ) and C�(W, QW ) of its summands.
The answer is very simple, which shows why one should take the Z2 -grading of the Clifford algebra very
seriously!

Proposition 2.1. Let (V, QV ) and (W, QW ) be quadratic vector spaces and let (V ⊕ W, QV ⊕ QW ) be their
orthogonal direct sum. Then there is an isomorphism of Z2 -graded associative algebras:

C�(V ⊕ W, QV ⊕ QW ) ∼ ˆ
= C�(V, QV )⊗C�(W, QW ) ,

ˆ denotes the Z2 -graded tensor product.

where ⊗

Tensor product of algebras

If A and B are associative algebras over the same ground field, their vector space tensor
product A ⊗ B becomes an algebra by

(a 1 ⊗ b 1 )(a 2 ⊗ b 2 ) = a 1 a 2 ⊗ b 1 b 2

and extending bilinearly. However when A = A0 ⊕ A1 and B = B0 ⊕ B1 are themselves Z2 -

graded, we can define on the Z2 -graded tensor product A⊗B,ˆ with (A⊗B)ˆ 0 = (A0 ⊗B0 )⊕(A1 ⊗B1 )
ˆ 1 = (A0 ⊗ B1 ) ⊕ (A1 ⊗ B0 ), the following associative multiplication
and (A⊗B)

(42) (a 1 ⊗ b 1 )(a 2 ⊗ b 2 ) = (−1)|a2 ||b1 | a 1 a 2 ⊗ b 1 b 2 ,

where a i and b i are homogeneous elements of parity |a i |, |b i |, respectively, and extending it

ˆ
bilinearly to all of A⊗B.

Proof. Define a linear map

(43) ˆ
φ : V ⊕ W −→ C�(V, QV )⊗C�(W, QW )

by φ(v + w) = v ⊗ 1 + 1 ⊗ w, where v ∈ V and w ∈ W and we are identifying V with its image in C�(V, QV )
and similarly for W. One checks that this map is Clifford precisely because of the sign in the definition
(42) of the multiplication in the Z2 -graded tensor product. Indeed,

φ(v + w)2 = (v ⊗ 1 + 1 ⊗ w)2

= v 2 ⊗ 1 + v ⊗ w + (−1)|v||w| v ⊗ w + 1 ⊗ w 2
= − (Q(v) + Q(w)) 1 ⊗ 1 .
Spin 2010 (jmf) 12

By universality it extends to a homomorphism of Z2 -graded associative algebras

ˆ
Φ : C�(V ⊕ W, QV ⊕ QW ) −→ C�(V, QV )⊗C�(W, QW ) ,

which is injective since it is one-to-one on generators and is surjective because the image contains
ˆ and 1⊗C�(W,
C�(V, QV )⊗1 ˆ ˆ
QW ), which together generate C�(V, QV )⊗C�(W, QW ).

Now every (finite-dimensional) real quadratic vector space (V, Q) is isomorphic to an orthogonal
direct sum

(44) R(0) ⊕ · · · ⊕ R(0) ⊕ R(+) ⊕ · · · ⊕ R(+) ⊕ R(−) ⊕ · · · ⊕ R(−)

� �� � � �� � � �� �
r s t

where R(0) is the one-dimensional vector space with zero quadratic form, whereas R(±) is the one di-
mensional vector space with quadratic form Q(1) = ±1. It then follows from the above proposition that

(45) C�(V, Q) ∼ r ˆ
= ΛR ⊗ ˆ · · · ⊗C�(1,
C�(1, 0)⊗ ˆ ˆ C�(0, 1)⊗
0) ⊗ ˆ · · · ⊗C�(0,
ˆ 1) ,
� �� � � �� �
s t

where we have used that the Clifford algebra associated to the zero quadratic form is the exterior al-
gebra. Using that C�(1, 0) ∼
= C and C�(0, 1) ∼
= R ⊕ R, the above result determines in principle all the finite-
dimensional real Clifford algebras as Z2 -graded associative algebras. This is nice, but one can do much
better and actually identify the Clifford algebras in terms of the matrix algebras K(n), for K = R, C, H as
we started doing in the first lecture.

2.2 Complex Clifford algebras

Before presenting the classification, let us consider the complex Clifford algebras. Since a complex
quadratic form has no signature, every (finite-dimensional) complex quadratic vector space is iso-
morphic to an orthogonal direct sum

(46) C(0) ⊕ · · · ⊕ C(0) ⊕ C(+) ⊕ · · · ⊕ C(+) ,

� �� � � �� �
r n

where C(0) is the one-dimensional complex vector spaces with zero quadratic form and C(+) the one-
dimensional complex vector space with Q(1) = 1. Proposition 2.1 then says that the corresponding
complex Clifford algebra is isomorphic to

(47) ΛCr ⊗
ˆ C�(1)⊗
ˆ · · · ⊗C�(1)
ˆ ,
� �� �
n

where C�(1) denotes the Clifford algebra C�(C(+)). To identify it, notice that C�(1) is the complex asso-
ciative algebra generated by e obeying e 2 = −1. This means that (i e)2 = 1 and we define complementary
projectors p ± = 12 (1 ± i e), which induce an isomorphism C�(1) ∼ = C ⊕ C with zp + + w p − ↔ (z, w).
One can complexify real quadratic vector spaces as follows. Let (V, Q) be a real quadratic vector
space and let VC = V ⊗R C be its complexification. We extend Q complex linearly to a quadratic form QC
defined by QC (v ⊗ z) = z 2 Q(v). This turns the pair (VC , QC ) into a complex quadratic vector space. It is
natural to ask whether C�(V, Q) and C�(VC , QC ) are related and the answer could not be nicer.

Proposition 2.2. The Clifford functor C� commutes with complexification; that is,

C�(VC , QC ) ∼
= C�(V, Q) ⊗R C .

Proof. The map V × C → C�(V, Q) ⊗R C defined by φ(v, z) = v ⊗ z is real bilinear, whence it defines an
R-linear map
φ
V ⊗R C −−−−−→ C�(V, Q) ⊗R C .
Spin 2010 (jmf) 13

Since φ(v ⊗z) = v ⊗z = (v ⊗1)z = φ(v ⊗1)z, we see that it is also C-linear. Also since φ(v ⊗1)2 = (v ⊗1)2 =
−Q(v)(1 ⊗ 1), we see that φ is Clifford, whence it extends uniquely to a homomorphism of complex
algebras
Φ : C�(VC , QC ) −→ C�(V, Q) ⊗R C .
It is clearly injective on generators and either by counting dimension or by observing that the image
of Φ contains C�(V, Q) ⊗ 1 and 1 ⊗ C and these generate the right-hand side, we conclude that Φ is an
isomorphism.

As a corollary we have that upon complexification we lose the information about the signature:

(48) C�(s, t ) ⊗R C ∼
= C�(s + t ) .

The absence of signature for a complex quadratic vector space means that the complex Clifford algebras
have a much simpler structure than the real Clifford algebras.

2.3 Filling in the Clifford chessboard

From now on we will restrict attention to the case of nondegenerate quadratic forms, whence any real
quadratic vector space is isomorphic to Rs,t for some s, t . We would like to identify the Clifford algebras
C�(s, t ) for all s, t ≥ 0. In the last lecture we already filled in a corner of the table of Clifford algebras:

R(2)
R⊕R R(2)
R C H
−−−−−−−→
s

This corner of the table is enough to fill in the rest of the table, thanks to the following isomorphisms.

Theorem 2.3. For all n, s, t ≥ 0 we have the following isomorphisms

C�(n, 0) ⊗ C�(0, 2) ∼
= C�(0, n + 2)
(49) C�(0, n) ⊗ C�(2, 0) ∼
= C�(n + 2, 0)
C�(s, t ) ⊗ C�(1, 1) ∼
= C�(s + 1, t + 1)

with ⊗ the ungraded tensor product.

Proof. As the three cases are very similar, we shall prove the second equation in (49) and leave the other
two as exercises for the reader.
Let us write Rn+2,0 = Rn,0 ⊕ R2,0 . Let e 1 , e 2 be an orthonormal basis for R2,0 and let us denote by the
same symbols their image in C�(2, 0). This means that e 12 = e 22 = −1 and e 1 e 2 = −e 2 e 1 . The element
e 1 e 2 ∈ C�(2, 0) satisfies the following easily verifiable identities: (e 1 e 2 )2 = −1, e 1 e 2 e i = −e i e 1 e 2 for i =
1, 2. Let us define a linear map
φ : Rn+2,0 −→ C�(0, n) ⊗ C�(2, 0)
by
φ(x) = x ⊗ e 1 e 2 and φ(e i ) = 1 ⊗ e i ,
Spin 2010 (jmf) 14

for x ∈ Rn,0 . This map is Clifford by virtue of the identities satisfied by e 1 e 2 ; indeed,

φ(x + λe 1 + µe 2 )2 = (x ⊗ e 1 e 2 + λ1 ⊗ e 1 + µ1 ⊗ e 2 )2
= −x 2 ⊗ 1 − λ2 1 ⊗ 1 − µ2 1 ⊗ 1
+ λx ⊗ (e 1 e 1 e 2 + e 1 e 2 e 1 ) + µx ⊗ (e 2 e 1 e 2 + e 1 e 2 e 2 ) + λµ1 ⊗ (e 1 e 2 + e 2 e 1 )
= −(x 2 + λ2 + µ2 )1 ⊗ 1
= −Q(x + λe 1 + µe 2 )1 ⊗ 1 .

Hence φ extends uniquely to an algebra homomorphism Φ : C�(n + 2, 0) → C�(0, n) ⊗ C�(2, 0) which is

injective on generators and by dimension must be an isomorphism.

Notice that the first two isomorphisms in (49) allows us to fill the left column and the bottom row in
the table, whereas the last isomorphism allows us to move diagonally. Since any square in the table lies
on some diagonal, it can in principle be determined by using the isomorphisms. Moving diagonally (one
step) is the same as tensoring with C�(1, 1) ∼
= R(2). To apply this we need to make use of the following
standard isomorphism of matrix algebras.

Lemma 2.4. Let K stand for any of R, C and H and let K(n) denote the real algebra of n ×n matrices with
entries in K. Then we have the following isomorphisms of real associative algebras:

(50) K(m) ⊗R R(n) ∼

= K(mn) .

This already allows us to fill in five of the diagonals in the table:

R(64) R(64)⊕ R(64) R(128)

R(32) R(32)⊕ R(32) R(64) C(64)
R(16) R(16)⊕ R(16) R(32) C(32) H(32)
R(8) R(8) ⊕ R(8) R(16) C(16) H(16)
R(4) R(4) ⊕ R(4) R(8) C(8) H(8)
R(2) R(2) ⊕ R(2) R(4) C(4) H(4)
R⊕R R(2) C(2) H(2)
R C H
−−−−−−−−→
s

To continue it is necessary to extend the bottom row and the left column. For example, let us con-
tinue with the bottom row. From the second of the isomorphisms in (49), we have

C�(3, 0) ∼
= C�(0, 1) ⊗ C�(2, 0) ∼
= (R ⊕ R) ⊗ H = H ⊕ H ,

where we have used the distributivity of ⊗ over ⊕. In the same way we obtain

C�(4, 0) ∼
= C�(0, 2) ⊗ C�(2, 0) ∼
= R(2) ⊗ H = H(2) .

We cannot continue along the bottom row without first extending the left column. Using the first of the
isomorphisms in (49), we find

C�(0, 3) ∼
= C�(1, 0) ⊗ C�(0, 2) ∼
= C ⊗ R(2) ∼
= C(2),

whereas
C�(0, 4) ∼
= C�(2, 0) ⊗ C�(0, 2) ∼
= H ⊗ R(2) ∼
= H(2),
C�(0, 5) ∼
= C�(3, 0) ⊗ C�(0, 2) ∼
= (H ⊕ H) ⊗ R(2) ∼
= H(2) ⊕ H(2),
and
C�(0, 6) ∼
= C�(4, 0) ⊗ C�(0, 2) ∼
= H(2) ⊗ R(2) ∼
= H(4) .
This allows us to fill in six more diagonals in the table!
Spin 2010 (jmf) 15

H(8) H(8) ⊕ H(8) H(16) C(32) R(64) R(64) ⊕ R(64) R(128)

H(4) H(4) ⊕ H(4) H(8) C(16) R(32) R(32) ⊕ R(32) R(64) C(64)
H(2) ⊕ H(2) H(4) C(8) R(16) R(16) ⊕ R(16) R(32) C(32) H(32)
H(2) C(4) R(8) R(8) ⊕ R(8) R(16) C(16) H(16) H(16)⊕H(16)
C(2) R(4) R(4) ⊕ R(4) R(8) C(8) H(8) H(8) ⊕ H(8) H(16)
R(2) R(2) ⊕ R(2) R(4) C(4) H(4) H(4) ⊕ H(4) H(8)
R⊕R R(2) C(2) H(2) H(2) ⊕ H(2) H(4)
R C H H⊕H H(2)
−−−−−−−−→
s

To continue we have to determine C�(5, 0). From the second of the isomorphisms in (49), we find

C�(5, 0) ∼
= C�(0, 3) ⊗ C�(2, 0) ∼
= C(2) ⊗ H ∼
=?

To answer the question we need the following result.

Lemma 2.5. The following are isomorphisms of real associative algebras:

1. C ⊗R H ∼
= C(2)

2. H ⊗R H ∼
= R(4)

3. C ⊗R C ∼
= C⊕C

Proof. 1. To prove the first isomorphism, let us give H the structure of a complex vector space by
left multiplication by the complex subalgebra of H generated by i , say. Then we construct a real
bilinear map
φ : C × H −→ EndC (H) by φ(z, q)x = zxq,
for all z ∈ C and x, q ∈ H. By universality of the tensor product, it defines a real linear map

Φ : C ⊗R H −→ EndC (H) by Φ(z ⊗ q) = φ(z, q).

We check that Φ is a homomorphism of real algebras:

Φ(z 1 ⊗ q 1 )Φ(z 2 ⊗ q 1 )x = z 1 (z 2 xq 2 )q 1 = (z 1 z 2 )xq 1 q 2 = Φ(z 1 z 2 ⊗ q 1 q 2 )x .

It is clearly injective because C and H are division algebras and counting dimension (dimR = 8) we
see that Φ must be an isomorphism. Now H ∼ = C2 as a complex vector space, whence EndC (H) ∼
=
C(2).

2. This is proved in a very similar manner to the first isomorphism. Namely we define a real bilinear
map
φ : H × H −→ EndR (H) by φ(q 1 , q 2 )x = q 1 xq 2 ,
for all q i , x ∈ H, which by universality of the tensor product induces a real linear map

Φ : H ⊗R H −→ EndR (H) by Φ(q 1 ⊗ q 2 ) = φ(q 1 , q 2 ).

It is clear that it is injective and counting dimension (dimR = 16), it is an isomorphism of real
vector spaces, but again one checks that Φ is an algebra morphism:

Φ(q 1 ⊗ q 2 )Φ(q 1� ⊗ q 2� )x = q 1 (q 1� xq 2� )q 2 = (q 1 q 1� )xq 2 q 2� = Φ(q 1 q 1� ⊗ q 2 q 2� )x .

3. This is even easier. Notice that the element i ⊗ i ∈ C ⊗R C squares to the identity, whence we can
form complementary projectors p ± = 12 (1 ⊗ 1 ± i ⊗ i ) whose images are commuting subalgebras
isomorphic to C. Explicitly, the isomorphism C ⊕ C → C ⊗R C is given by (z 1 , z 2 ) �→ z 1 p + + z 2 p − .
Spin 2010 (jmf) 16

It follows at once from the first of these isomorphisms, that

C(2) ⊗ H ∼
= R(2) ⊗ C ⊗ H ∼
= R(2) ⊗ C(2) ∼
= C(4) ,

whence C�(5, 0) ∼
= C(4), and hence

C�(0, 7) ∼
= C�(5, 0) ⊗ C�(0, 2) ∼
= C(4) ⊗ R(2) ∼
= C(8) .

In the same way one can show that C�(6, 0) ∼

= R(8) and C�(7, 0) ∼
= R(8)⊕R(8), which allows us to complete
the first 8 × 8 corner of the table:

C(8) H(8) H(8) ⊕ H(8) H(16) C(32) R(64) R(64) ⊕ R(64) R(128)
H(4) H(4) ⊕ H(4) H(8) C(16) R(32) R(32) ⊕ R(32) R(64) C(64)
H(2) ⊕ H(2) H(4) C(8) R(16) R(16) ⊕ R(16) R(32) C(32) H(32)
H(2) C(4) R(8) R(8) ⊕ R(8) R(16) C(16) H(16) H(16)⊕H(16)
C(2) R(4) R(4) ⊕ R(4) R(8) C(8) H(8) H(8) ⊕ H(8) H(16)
R(2) R(2) ⊕ R(2) R(4) C(4) H(4) H(4) ⊕ H(4) H(8) C(16)
R⊕R R(2) C(2) H(2) H(2) ⊕ H(2) H(4) C(8) R(16)
R C H H⊕H H(2) C(4) R(8) R(8) ⊕ R(8)
−−−−−−−−→
s

As nice as this is, it might seem that in order to classify real Clifford algebras in arbitrary (albeit finite)
dimension we have to do lots of work. Luckily this is not the case, which explains a posteriori why I have
restricted myself to an 8 × 8 corner, the so-called Clifford chessboard. It turns out that the real Clifford
algebras have simpler periodicities, which are an easy consequence of Theorem 2.3. We call them the
Bott periodicities.

Corollary 2.6. For all n, s, t ≥ 0, the following are isomorphisms of real algebras:

1. C�(n + 8, 0) ∼
= C�(n, 0) ⊗R R(16),

2. C�(0, n + 8) ∼
= C�(0, n) ⊗R R(16), and

3. C�(s + 4, t + 4) ∼
= C�(s, t ) ⊗R R(16).

Proof. This follows directly from repeated application of Theorem 2.3 and the following isomorphisms:

C�(1, 1)⊗4 ∼
= R(16) and C�(2, 0)⊗2 ⊗ C�(0, 2)⊗2 ∼
= R(16) .

Theorem 2.7 (Classification theorem). The Clifford algebra C�(s, t ) is isomorphic to the real associative
algebras in the following table, where d = s + t :

s − t mod 8 C�(s, t )
� d /2 �
0, 6 R 2
� � � �
7 R 2(d −1)/2 ⊕ R 2(d −1)/2
� (d −1)/2 �
1, 5 C 2
� �
2, 4 H 2(d −2)/2
� � � �
3 H 2(d −3)/2 ⊕ H 2(d −3)/2

The proof follows from the Clifford chessboard and the periodicities and it is simply a matter of
bookkeeping.
Spin 2010 (jmf) 17

2.3.1 The even subalgebra of the Clifford algebra

In the study of spinor representations it is important to identify the even subalgebra C�(s, t )0 of the Clif-
ford algebras C�(s, t ) as an ungraded real associative algebra. Recall that C�(s, t )0 is the fixed subalgebra
under the automorphism induced by the orthogonal transformation in O(s, t ) which sends x �→ −x for
all x ∈ Rs,t . This means that every element in C�(s, t )0 can be written as a linear combination of products
of an even number of elements in the image in C�(s, t ) of Rs,t .
Luckily, C�(s, t )0 can be determined from the Clifford algebra one dimension lower.

Proposition 2.8. For all s, t ≥ 0, we have the following isomorphisms of ungraded real associative algeb-
ras:
C�(s, t ) ∼
= C�(s + 1, t )0 ∼
= C�(t , s + 1)0 .
In particular, we have that C�(s, t )0 ∼
= C�(t , s)0 .

Proof. We will prove one of the isomorphisms and leave the other as an exercise. Let us define φ : Rs,t →
C�(s + 1, t )0 by φ(x) = xe s+1 , where we write Rs+1,t = Rs,t ⊕ Re s+1 . We check that φ is a Clifford map:

φ(x)2 = xe s+1 xe s+1 = −x 2 e s+1

2
= x 2 = −Q(x)1 ,

whence it extends uniquely to an algebra homomorphism Φ : C�(s, t ) → C�(s+1, t )0 . It is clearly injective

on generators and counting dimension, we conclude Φ is an isomorphism.

As a corollary of the classification theorem 2.7, we immediately have a classification of the C�(s, t )0 .

Corollary 2.9. The even Clifford algebra C�(s, t )0 is isomorphic to the real associative algebras in the
following table, where d = s + t :

s − t mod 8 C�(s, t )0
� (d −1)/2 �
1, 7 R 2
� � � �
0 R 2(d −2)/2 ⊕ R 2(d −2)/2
� �
2, 6 C 2(d −2)/2
� �
3, 5 H 2(d −3)/2
� � � �
4 H 2(d −4)/2 ⊕ H 2(d −4)/2

2.4 Classification of complex Clifford algebras

Having determined the real Clifford algebras, it is a simple matter to use Proposition 2.2 and determine
the complex Clifford algebras. It is however easier to derive the complex Bott periodicity directly.

Proposition 2.10. For all n ≥ 0 there is an isomorphism of complex associative algebras

C�(n + 2) ∼
= C�(n) ⊗C C(2) .

Proof. Write Cn+2 = Cn ⊕ Ce 1 ⊕ Ce 2 and define a complex linear map

φ : Cn+2 −→ C�(n) ⊗C C(2)

by � � � � � �
0 i 0 i i 0
φ(x) = x ⊗ φ(e 1 ) = 1 ⊗ and φ(e 2 ) = 1 ⊗
−i 0 i 0 0 −i
for all x ∈ Cn . One checks that φ is Clifford and that the induced map Φ : C�(n +2) → C�(n)⊗C C(2), being
injective on generators and mapping between equidimensional spaces, is an isomorphism.

The classification of complex Clifford algebras is then an easy corollary.

Spin 2010 (jmf) 18

Corollary 2.11. For every n ≥ 0, the complex Clifford algebra C�(n) is isomorphic to
� � �
∼ C 2n/2 if n is even,
C�(n) = � (n−1)/2 � � (n−1)/2 �
C 2 ⊕C 2 if n is odd.

Proof. This follows easily from complex Bott periodicity and the “initial conditions” C�(0) ∼
= C and
C�(1) ∼
= C ⊕ C.