21 Spin, Clifford algebras, and angular mo-

mentum
21.1 Historical motivation from physics
Arguably the best motivation for this chapter, is figuring out the mystery of
spin, which first appeared experimentally in non-relativistic quantum physics
as a mysterious kind of angular momentum without a satisfactory classical
mechanics correspondence. This is explained in Section 21.6, after establishing
some terminology.
Nowadays, the motivation is often illustrated by a puzzle in special relativis-
tic quantum mechanics. For C-valued “scalar” wavefunctions on Rd , the stan-
ℏ2
dard Schrödinger equation has kinetic energy term being 2m times the Lapla-
29
P d 2
cian ∆ = − j=1 ∂j . This has no chance of being strictly compatible with
ℏ2
relativity, since the Schrödinger equation iℏ∂t = 2m ∆ has time-derivative and
spatial-derivatives of different orders. Considering the massless case for sim-
plicity (and taking the speed of light c = 1), the relativistic energy-momentum
relation is E 2 = |p|2 (see (4.3)). So under the quantum mechanical correspon-
∂
dence pj ∼ −iℏ∂j and E ∼ H, theP abstract Schrödinger equation iℏ ∂t =H
2 2 2 2 2 2
should have H ∼ E = |p| ∼ −ℏ j ∂j = ℏ ∆. Thus we seek a self-adjoint
2
“first-order square root of ℏ ∆” to serve as the “relativistic Hamiltonian” H.
At first, this seems like an impossible task: ( j −iℏ∂j )2 contains mixed
P
derivatives ∂j ∂k which are not present in the Laplacian. But in 1928, P.A.M.
Dirac found a beautiful trick to get rid of ∂j ∂k . It should be stressed that Dirac
was inspired by a matrix-angular-momentum construction of Pauli in 1927,
which was itself motivated by puzzling experimental results in the properties
of quantum electrons. For (3 + 1)-dimensional Minkowski spacetime, Dirac
formulated his famous first-order Dirac operator D, / which acted on C4 -valued
“functions” ψ. The use of quotation marks is deliberate. As we will learn, ψ
is a “spinor” object, fundamentally distinct from a naı̈ve collection of scalar
functions, much like how a tangent vector field is not merely a collection of d
scalar functions.
The key algebraic idea behind Dirac’s construction can be illustrated from
29
In view of Section 3.5, we should rather view ∆ as a connection Laplacian (Section 23.2)
for a flat connection on a trivializable Hermitian line bundle over Rd .

218
the case of Euclidean R2 , where the Dirac operator is
   
0 ∂x − i∂y 0 ∂
D/ = −i ≡ −2i ¯ , (21.1)
∂x + i∂y 0 ∂ 0

and satisfies D/ 2 = ∆ ⊗ 12 as desired. What makes D / work is that a ma-

trix ej is attached to each orthonormal tangent direction ∂j , such that ej , ek
anticommute. Then the “matrix-enriched” mixed partial derivatives cancel,
ej ∂j ek ∂k + ek ∂k ej ∂j = (ej ek + ek ej )∂j ∂k = 0.
| {z }
0

As we will learn, the matrix realization of the algebraic condition ej ek = −ek ej

is precisely encoding the hidden geometry of “spin rotations”. This is why
Dirac’s operator (for the relativistic electron) is intimately related to Pauli’s
analysis of spin (angular momentum of not-necessarily-relativistic electron).
The fundamental concept is not really special relativity per se, but gauge
theory for Riemannian or Lorentzian spin-geometry, as the case may be.

21.2 Clifford algebra

Let V be a real vector space. The tensor algebra of V is
∞
M
T (V ) = V ⊗k = R ⊕ V ⊕ (V ⊗ V ) ⊕ . . . ,
k=0

with multiplication given by tensor product. So 1 ∈ R serves as the multi-

plicative unit, and we are allowed to consider the “product” of two vectors
u, v ∈ V , written as
uv ≡ u ⊗ v ∈ V ⊗ V ⊂ T (V ).
Now suppose V has an inner product, with associated quadratic form
q(v) := ⟨v|v⟩, v ∈ V.
Definition 92. Let (V, q) be a real inner product space. The Clifford algebra
is the quotient algebra
Cl(V, q) := T (V )/Iq .
where Iq is the ideal generated by the relation
v 2 + q(v)1, v ∈ V.

219
 By design, v 2 = −q(v) holds in Cl(V, q), so we have the anticommutation
relations in Cl(V, q),

uv + vu = (u + v)2 − u2 − v 2 = −q(u + v) + q(u) + q(v)

= −2⟨u|v⟩, u, v ∈ V. (21.2)

In particular, if {ei }i=1,...,n is an orthonormal basis for V , then

{ei , ej } ≡ ei ej + ej ei = −2δij . (21.3)

Proposition 21.1. (Optional.) There is an injective map ιV : V ,→ T (V ) →

Cl(V, q). Furthermore, Cl(V, q) is the unique K-algebra with the property:
given a linear map f : V → A into an associative K-algebra A satisfying
f (v)2 = −q(v)1A for all v ∈ V , there exists a unique extension of f to an
algebra homomorphism, fˆ : Cl(V, q) → A . Diagrammatically:
ιV
V Cl(V, q)
∃! fˆ
.
f
A

Sketch. The map

F : T (V ) → A , v1 ⊗ · · · ⊗ vn 7→ f (v1 ) ⊗ · · · ⊗ f (vn ).

annihilates the ideal Iq , so it descends to a homomorphism

fˆ : T (V )/Iq ≡ Cl(V, q) → A .

21.2.1 Linear basis for Cl(V, q)

Let {ei }i=1,...,n be an orthonormal basis for (V, q). Let I denote an increasing
sequence of indices 1 ≤ i1 < . . . < ik ≤ n with length |I| = k. Then we use
the multi-index notation,

eI := ei1 . . . eik ∈ Cl(V, q), e∅ = 1.

Note that the maximum length of a multi-index is n.

220
Proposition 21.2. Cl(V, q) has dimension nk=0 nk = 2n , with a basis given
P

by eI with I running over all multi-indices. Each of Cl0 (V, q) and Cl1 (V, q) has
dimension 2n−1 , except for the case n = 0, where Cl(V, q) = Cl0 (V, q) = K.

Proof. Exercise. (This is algebraic, but not that easy.)

Thus Cl(V, q) is actually linearly isomorphic to the exterior algebra Λ∗ (V )
of differential forms. (Consider what happens when q = 0).

21.2.2 Even/odd elements

The “tensor degree” is ill-defined on Cl(V, q) — v ⊗ v and −q(v)1 are tensors
with different degrees, but represent the same element in Cl(V, q). What makes
is whether an element of Cl(V, q) is even or odd, and accordingly, we split

Cl(V, q) = Cl0 (V, q) ⊕ Cl1 (V, q).

Note that Cl0 (V, q) is a subalgebra of Cl(V, q), while Cl1 (V, q) is a comple-
mentary linear subspace.

21.3 Clifford algebras Cln , Cln

21.3.1 Real Clifford algebras
We write
Cln := Cl(Rn , qstandard ).
By Prop. 21.1, for a general (V, q), we obtain an isomorphism Cl(V, q) ∼
= Cln
∼
=
upon choosing a basis Rn → V . So we shall just be considering V = Rn below,
and {ei }i=1,...,n will denote the standard orthonormal basis of Rn .

Cl0 . We have Cl0 = R as a real algebra.

Cl1 . A linear basis for Cl1 is {1, e1 }, where e21 = −1. So e1 is a square root
of −1, and Cl1 ∼
=C∼ = C as R-algebras.

221
Cl2 . A linear basis for Cl2 is {1, e1 , e2 , e1 e2 }. The elements e1 , e2 , e1 e2 are
anticommuting square roots of −1. So Cl2 is isomorphic to the quaternions
H = spanR {1, I, J, K} as R-algebras, which have multiplication rule
I 2 = J 2 = K 2 = IJK = −1.

Exercise 21.1. By considering the map

f : Rn → Cln+1
0
, ei 7→ ei en+1 , i = 1, . . . , n,
show that there is an algebra isomorphism Cln ∼ 0
= Cln+1 .

21.3.2 Complex Clifford algebras

The complexification of Cln is
Cln := Cln ⊗R C.
This is a tensor product over R, meaning that
aλ ⊗ µ = a ⊗ λµ, λ ∈ R, a ∈ Cln , µ ∈ C.
As a result, we can scalar multiply (on the right) by complex numbers,
(a ⊗ µ)ν = a ⊗ µν, µ, ν ∈ C.
We usually just write a ⊗ µ ≡ aµ = µa. The algebra product is extended from
Cln to Cln complex bilinearly, e.g.,
(a ⊗ µ)(b ⊗ ν) = ab ⊗ µν.

Volume/chirality element. The volume/chirality element of Cln is

n+1
ω C ≡ ωnC := i⌊ 2
⌋
e1 · · · en .
The inversion map v 7→ −v on Rn induces the parity automorphism α of Cln .
Each ei acquires a minus sign, so for n odd, α exchanges ω C with −ω C .
Exercise 21.2. Check that ω C has the following properties:
• If {e′i }i=1,...,n is another positively oriented orthonormal basis for Rn , then
n+1
i⌊ 2 ⌋ e′1 · · · e′n = ω C .
• (ω C )2 = 1.
• When n is odd, ω C commutes with all elements of Cln .
When n is even, ω C commutes with Cln0 .

222
Cl0 . This is C as a C-algebra.

Cl1 . There is an isomorphism of C-algebras,

(
1+e1 i
(1, 0) 7→ 2
,
C ⊕ C → Cl1 , 1−e1 i
(0, 1) 7→ 2
.

Cl2 . We can represent Cl2 as

     
0 −i 0 −1 1 0
e1 ↔ , e2 ↔ , ω ↔ ie1 e2 =
C
−i 0 1 0 0 −1

Thus Cl2 = spanC {1, e1 , e2 , e1 e2 } ∼

= M2 (C) as C-algebras.
The Cln , n > 2, are obtained by “algebraic Bott periodicity”.
Proposition 21.3. For each n ≥ 0, there is an isomorphism Cln+2 ∼
= Cln ⊗Cl2 .
Proof. In Cln , we have odd, anticommuting elements ei , i = 1, . . . , n, each
squaring to −1. Similarly, in Cl2 , we have ẽ1 , ẽ2 . In Cln+2 we have odd, anti-
commuting εi , i = i, . . . , n + 2, each squaring to −1. Then, the correspondence
ωC
z }|2 {
√
εi ↔ ei ⊗ −1ẽ1 ẽ2 , i = 1, . . . , n,
εn+j ↔ 1⊗ẽj j = 1, 2,

gives the desired identification Cln+2 ∼

= Cln ⊗Cl2 .

21.4 Spin groups

On Rn , the basic orthogonal transformation is a reflection in some hyperplane.
Such a reflection is labelled by the unit vector y ∈ S n−1 orthogonal to the
hyperplane, via the formula

Oy : v 7→ v − 2⟨y|v⟩y, v ∈ Rn .

Note the ambiguity in the y-label, since O−y = Oy .

Theorem 21.4 (Cartan–Dieudonné). Any element of O(n) is the product of
at most n reflections.

223
(Optional.) Let R ∈ O(n) and d = dim ker(R − 1n ). By induction on n − d,
we will show that R is a product of n − d reflections.
• For n − d = 0, R is the identity transformation.

• Assume the result for all R′ ∈ O(n) with

0 ≤ n − dim ker(R′ − 1n ) < n − d.

Choose a unit vector v ∈

/ ker(R − 1n ), and let S be the reflection in the
hyperplane orthogonal to R(v) − v. (Explicitly, S = O R(v)−v .) So
||R(v)−v||

S(v) = R(v), S(R(v)) = v.

Consider R′ = S ◦ R. Then

u ∈ ker(R −1n ) ⇒ (R′ −1n )(u) = S(R(u))−u = 0 ⇒ u ∈ ker(R′ −1n ),

and also
R′ (v) = S(R(v)) = v ⇒ v ∈ ker(R′ − 1n ).
Thus
n − dim ker(R′ − 1n ) ≤ n − (d + 1),
and the induction hypothesis applies to R′ : it is the product of at most

n − dim ker(R′ − 1n ) ≤ n − 1 − d

reflections. Thus R = SR′ is the product of at most n − d reflections.

The Cartan–Dieudonné result, Theorem 21.4, shows that any SO(n) ele-
ment is a product of an even number of reflections, Oy1 . . . Oy2k .
Now, observe that in the Clifford algebra, the anticommutation relation,
(21.2), leads to

(±y)v(±y) = (yv)y = (−vy − 2⟨y|v⟩)y = v − 2⟨y|v⟩y = Oy (v). (21.4)

Thus the reflection Oy can be realized as “Clifford-sandwiching” of v ∈ Rn ⊂

Cln with either y or −y. The following definition is motivated by thinking of
unit vectors y ∈ Cln as abstract reflections.

224
Definition 93. The spin group is the group in Cln generated by products of
even numbers of unit vectors,

Spin(n) := {y1 · · · y2k : k ∈ N, yi ∈ S n−1 ⊂ Rn }.

Note the important fact that −1 = y 2 lies in Spin(n).

Proposition 21.5. For n ≥ 1, there is a short exact sequence of Lie groups

χ
1 → {±1} → Spin(n) → SO(n) → 1,

where for g ∈ Spin(n), χ(g) is the orthogonal transformation

χ(g) : Rn → Rn , v 7→ gvg −1 . (21.5)

Proof. For y1 , y2 ∈ S n−1 , we have y2 y1 = (−y2 )(−y1 ) = y2−1 y1−1 = (y1 y2 )−1 .
Thus, by (21.4),

(y1 y2 )v(y1 y2 )−1 = y1 y2 vy2 y1 = y1 (Oy2 (v))y1 = Oy1 Oy2 (v), ∀ v ∈ Rn .

Let g ∈ Spin(n), so g = (y1 y2 ) . . . (y2k−1 y2k ) with yi ∈ S n−1 . Then

χ(g)(v) = gvg −1 = Oy1 . . . Oy2k (v), v ∈ Rn ,

verifying that χ(g) ∈ SO(n). This also shows that χ is a homomorphism,

whose surjectivity follows from the Cartan–Dieudonné result.
Now let g ∈ ker χ, so

gv = vg, ∀ v ∈ Rn . (21.6)

Set v = e1 above. As g lies in the even subalgebra Cln0 , we may write

X X
g= g I eI + g J e1 eJ =: g0 + e1 g1 ,
|I| even, ik ̸=1 |J| odd, jk ̸=1

where neither g0 nor g1 has e1 factors. So e1 commutes with g0 and anticom-

mutes with e1 , and it follows that
(21.6)
e1 g0 − g1 = e1 g = ge1 = g0 e1 + e1 g1 e1 = e1 g0 + g1 .

225
We conclude that g1 = 0, so g = g0 has no e1 factor. The same argument with
v = e2 , . . . , en shows that g does not contain any e2 , . . . , en−1 , en factor, so g
can only be a scalar in Cln . Together with
g 2 = (y1 · · · y2k )2 = ±y12 . . . y2k
2
= ±1,
we conclude that g ∈ ker χ has to be either +1 or −1. Clearly −1 = e21 and
χ(−1) = 1n , so −1 ∈ ker χ. Thus ker χ = {±1}.
Finally, we sketch why Spin(n) is a Lie group. Let Cln× denote the group of
invertible elements of Cln ; it may be checked to be an open submanifold, thus
a Lie group (this is a general feature of finite-dimensional algebras regarded
as vector space manifolds). The adjoint representation,
Ad : Cln× → GL(Cln ), Ad(a) : b 7→ aba−1 ,
is a continuous map. We can regard χ as the restriction of Ad to the subgroup
Spin(n); here, χ(g) ∈ SO(n) is promoted to an automorphism of Cln . Since
SO(n) is closed, the preimage Spin(n) is a closed subgroup of Cln× , thus a Lie
group (by a general closed subgroup theorem in Lie theory).
Proposition 21.6. For n ≥ 2, the spin group Spin(n) is connected.
Proof. Write g ∈ Spin(n) as g = y1 , . . . y2k with yi ∈ S n−1 . Let γi (t) be a path
in S n−1 connecting yi to y1 , so Π2k
i=1 γi (t) is a path in Spin(n) connecting g to
y12k = (−1)k . This shows that g is connected to either +1 or −1 in Spin(n).
It remains to check that +1 and −1 are connected in Spin(n). For or-
thonormal e1 , e2 , the path
γ(t) = − cos(t) − sin(t)e1 e2 , t ∈ [0, π].
joins −1 to +1 in Cln . But we can factorize
γ(t) = (cos(t/2)e1 + sin(t/2)e2 ) (cos(t/2)e1 − sin(t/2)e2 ) ,
into the product of two paths of unit vectors. So γ is actually a path within
Spin(n).
Remark 6. For n ≥ 3, it may be shown that SO(n) has fundamental group
π1 (SO(n)) = Z2 . For instance,
 
cos t − sin t 0
γ : t 7→  sin t cos t 0 , t ∈ [0, 2π] (21.7)
0 0 1n−2

226
is a non-trivial loop in SO(n), but the doubled loop with t ∈ [0, 4π] is de-
formable to a constant loop. On the other hand, Spin(n) is simply-connected
(for n ≥ 3). So Spin(n) is also the universal cover of SO(n).
The existence of the spin double-cover has great physical significance. After
applying a full rotation of SO(n) reference frames, a vector in Rn is certainly
returned to itself. We know that quantum mechanics encodes physical config-
urations in complex√Hilbert spaces, so it is conceivable that nature uses some
intricate complex “ vector” which can distinguish between the application of
one full rotation versus two full ones.
Mathematically, we are seeking a quantity

spinor = [spin frame, components]

= [spin frame · g −1 , ρ(g) · components], g ∈ Spin(n). (21.8)

The set of “spin frames” would be a Spin(n)-torsor, and the components of a

spinor would come from some complex vector space S on which Spin(n) acts
faithfully. The loop in SO(n) given in (21.7) only lifts to a path in Spin(n) with
endpoints differing by −1 (Exercise). Consequently, at the end of the path γ̃,
the spin frame would have been rotated to −1 of the initial spin frame, and
the components of a spinor will acquire a corresponding minus sign.

21.4.1 Spin groups in low dimensions

Spin(1). This is the two element group {±1}.

Spin(2). It is not hard to deduce that

Spin(2) = {a + b e1 e2 : a2 + b2 = 1},

with group inversion given by (a, b) 7→ (a, −b). It may be parametrized by the
1-parameter subgroup generated by e1 e2 ∈ spin(2),

θ 7→ exp(θe1 e2 ) = cos θ + sin θe1 e2 ,

where we used (e1 e2 )2 = −1. For g = cos θ + sin θe1 e2 , we calculate

χ(g)(e1 ) = ge1 g −1 = cos 2θe1 + sin 2θ e2

χ(g)(e2 ) = ge2 g −1 = − sin 2θe1 + cos 2θe2 .

227
So the double-covering map is
    
χ cos 2θ − sin 2θ 0 −1
Spin(2) ∋ exp(θe1 e2 ) 7→ = exp 2θ ∈ SO(2).
sin 2θ cos 2θ 1 0
(21.9)

Spin(3). By a slight modification of Exercise 21.1, Cl30 can be identified with

the Clifford algebra Cl2 by setting

ẽ1 = e2 e3 , ẽ2 = e3 e1 .

Thus

Cl30 ∼
= Cl2 = spanR {1, ẽ1 , ẽ2 , ẽ1 ẽ2 } = spanR {e ∼
2 e3 , e3 e1 , e1 e2 } = H. (21.10)
|{z} |{z} |{z}
I J K

Now, H is a 1-dimensional “quaternionic vector space” with right scalar

multiplication by H itself. In this regard, we write H1 . The conjugate of
q = a + bI + cJ + dK ∈ H1 is defined as q = a − bI − cJ − dK. The “inner
product” on H1 is (q1 , q2 ) = q1 q2 , so the “norm” is

||q||2 = a2 + b2 + c2 + d2 .

The algebra H acts by left-multiplication on H1 . This is quaternion-linear since

left and right multiplications commute. The subgroup in H which preserves
the inner product is called the symplectic group Sp(1). It is easily checked
that Sp(1) comprises the unit-norm quaternions, thus Sp(1) is a Lie group
diffeomorphic to S 3 .
Let us check that the algebra isomorphism Cl30 ∼ = H of (21.10) restricts
∼
to a Lie group isomorphism Spin(3) = Sp(1). By definition, Spin(3) elements
P3 be written as even Clifford-products y1 · · · y2k of unit 3-vectors. If y =
can
i=1 ai ei is a unit 3-vector, then

Cl30 ∋ e3 y = a1 e3 e1 + a2 e3 e2 − a3 ←→ a1 J − a2 I − a3 1 ∈ H,

so e3 y is identified as a unit quaternion, i.e., Sp(1) element. Similarly for

ye3 ∈ Sp(1). So

Spin(3) ∋ y1 · · · y2k = (y1 e3 )(e3 y2 ) · · · (y2k−1 e3 )(e3 y2k ) ∈ Sp(1).

228
Thus we have an injective homomorphism Spin(3) ,→ Sp(1) of Lie groups.
Since Sp(1) ∼
= S 3 is a connected Lie group of the same dimension30 as Spin(3),
general Lie theory (omitted) shows that Spin(3) ,→ Sp(1) is in fact an isomor-
phism.
There is a further accidental isomorphism Sp(1) ∼ = SU(2). One such iso-
morphism is given below. Represent the R-algebra H as an R-subalgebra of
M2 (C), via  
a − id −c − ib
H ∋ a + bI + cJ + dK = . (21.11)
c − ib a + id
It is easy to check that the matrix on the right side belongs to SU(2) iff
a2 + b2 + c2 + d2 = 1 iff the left side is a unit quaternion, i.e., Sp(1) element.
Thus we have (non-canonical) identifications
(21.10) (21.11)
  
w −z̄
Spin(3) ∼ = Sp(1) ∼ = SU(2) = 2 2
: |w| + |z| = 1 . (21.12)
z w̄

21.5 Spin Lie algebra

Recall that so(n) is identified with the space of skew-symmetric n×n matrices.
A convenient basis comprises the matrices Eij , 1 ≤ i < j ≤ n, with entries

1,
 column i, row j,
−1, column j, row i,

0, otherwise.


For example,  
0 −1 0 · · · 0
1 0
 0 ··· 0 
E12
0 0
= 0 ··· 0 .
 .. .. .. .. .. 
. . . . .
0 ··· ··· ··· 0
Next, let us analyze spin(n). The group Cln× is open in the vector space Cln
(exercise), so it is a Lie group whose Lie algebra cl×
n is identified with Cln , at
least as a vector space. Let us check that the Lie bracket on cl×n coincides with
the commutator on Cln . Note that Cln ⊂ End(Cln ) via left-multiplication.
30
SO(3) and Spin(3) are 3-dimensional.

229
So Cln× ⊂ GL(Cln ) is a matrix Lie group. Thus the Lie bracket on cl×
n = Cln
coincides with the commutator Lie bracket on End(Cln ) (restricted to the
subalgebra Cln ). By a further restriction, Spin(n) ⊂ Cln× has spin(n) ⊂ cl×
n
being a Lie subalgebra of Cln with commutator as Lie bracket.
Proposition 21.7. The double-covering homomorphism χ : Spin(n) → SO(n)
induces the Lie algebra isomorphism
χ∗ : spin(n) → so(n)
ei ej 7→ 2Eij , 1 ≤ i < j ≤ n.
Proof. For 1 ≤ i < j ≤ n, consider the curve
γ : t 7→ cos(t) + sin(t)ei ej , 1 ≤ i < j ≤ n,
which lies in Spin(n) by the proof of Prop. 21.6. It has
γ̇(0) = ei ej ∈ Te Spin(n) = spin(n).
The surjective Lie group homomorphism χ induces, by Exercise 18.1, a Lie
algebra map
χ∗ : spin(n) → so(n)
d(χ ◦ γ)
ei ej 7→ .
dt t=0

To understand the right side as a skew-symmetric matrix, we apply it to an

arbitrary v ∈ Rn . A preliminary observation is
d(γ(t)−1 ) dγ(t)
=− = −ei ej .
dt t=0 dt t=0
Now compute
d(χ ◦ γ)
χ∗ (ei ej )(v) = (v)
dt t=0
d
γ(t)vγ(t)−1


=
dt t=0
= γ̇(0)vγ(0)−1 + γ(0)v (γ −1 ˙ )(0)
= ei ej v − vei ej
= ei (−vej − 2⟨ej |v⟩) + (ei v + 2⟨ei |v⟩)ej (Eq. (21.2))
= −2ei ⟨ej |v⟩ + 2ej ⟨ei |v⟩
= 2Eij (v).

230
By dimension considerations,
spin(n) = spanR {ei ej : 1 ≤ i < j ≤ n},
so we have completely specified χ∗ , which is thus also an isomorphism.

Example 21.1. The isomorphism χ∗ : spin(2) → so(2), e1 e2 7→ 2E12 can be

seen directly from (21.9).

21.6 (Spin) angular momentum

On Euclidean R3 , a classical particle at position x and momentum p has orbital
angular momentum “vector” L = x ∧ p with respect to the origin. To make
sense of this in d ̸= 3, it is better to regard x, p regarded as cotangent vectors
at the origin via the inner product, and x ∧ p as a 2-form31 L at the origin.
Note that implicitly, we need to parallel transport p back to the origin.
Generally speaking, it makes sense to ask how much angular momentum
there is with respect to an oriented tangent 2-plane. This does not depend on
the embedding in a d ≥ 3 space or a choice of orientation on such an ambient
space. Using the orthogonal tangent basis ∂1 , . . . , ∂d , there is a corresponding
basis dxj ∧ dxk , 1 ≤ j < k ≤ d of 2-forms, and we can decompose
X
L= Ljk dxj ∧ dxk
1≤j<k≤d

d


into 2 independent components.
A typical example is a particle with mass m rotating in a circle of radius
r in the j-k plane with constant angular velocity ωjk , its angular momentum
2-form is
L = Ljk dxj ∧ dxk , Ljk = mr2 ωjk .
If the rotating particle has electric charge q (so it is a “current loop”), then it
possess a magnetic moment 2-form P µ, in the sense that when subjected to a
background magnetic 2-form B = j<k Bjk dxj ∧dxk , it has magnetic potential
energy X
−⟨µ, B⟩ = − µjk Bjk .
j<k

31
In physics, a 3D angular momentum is sometimes called a pseudovector, because it
changes sign simply by an unphysical choice of orientation.

231
Here, it is assumed that Bjk are constant over the region of motion32 . Classical
electromagnetic theory gives
q
µ= L, (21.13)
2m
q
with the proportionality constant 2m called the gyromagnetic ratio. Thus
q
magnetic potential energy = − ⟨L, B⟩. (21.14)
2m
For a positively-charged particle, this energy is minimized/maximized when
its L is aligned/antialigned with B.
As one might anticipate, orbital angular momentum becomes tricky to
define for curved Riemannian manifolds, and we shall not attempt to do so.
What is more important is that nature turns out to have another “intrinsic”
source of magnetic moment, which does not arise from an orbiting charge
picture. It is supposed to come from a different corresponding intrinsic notion
of angular momentum, called spin angular momentum, carried by quantum
mechanical point particles (with electric charge). The role of electromagnetism
in this story will be deferred to Section 23.4.
Let us nail down mathematically, the concept of spin angular momentum.
First, we reexamine how orbital angular momentum in Euclidean R3 is pre-
sented as quantum mechanical operators. These are the L1 , L2 , L3 operators
given in Example 11.4. (A better labelling would be L23 , L31 , L12 .) They are
easily checked to satisfy the angular momentum commutation relations

[L1 , L2 ] = iℏL3 ⇔ [L23 , L31 ] = iℏL12 . (21.15)

In so(3), we also have the Lie bracket

[E23 , E31 ] = E12 ,

so actually,
1
Ljk ↔ Ejk (21.16)
ih
is giving a representation of so(3) on L2 (R3 ). This generalizes to orbital an-
gular momentum operators representing so(d) on L2 (Rd ).
32
In practice, one assumes the current loop to be very small, e.g., an electron orbiting a
nucleus (which is, strictly speaking, a fictitious classical analogy). In fact, sometimes one
speaks of a magnetic dipole as an informal r → 0 limit of a current loop, keeping µ constant.

232
 Orbital angular momentum operators are significant when studying Hamil-
tonians on L2 (Rd ) with SO(d) rotation symmetry where SO(d) acts on the
manifold Rd fixing a rotation centre. Basic examples are H = −∇2 + V with
V = V (r) a potential depending only on the radial coordinate. For such H,
any component of angular momentum, Ljk , commutes with H, and leads to a
conserved quantity in the quantum mechanics sense of Section 11.2. Namely,
there are simultaneous eigenfunctions of H and Ljk , describing describe quan-
tum states with sharply-defined and time-independent expectation values for
angular momentum in the j-k plane. (But beware that the Ljk do not commute
amongst themselves, so there are no quantum states with all components of
angular momentum being simultaneously sharply-defined!) Significantly, one
may deduce from the commutation relations themselves, that the eigenvalues
of Ljk are constrained to be integer multiples of ℏ (Exercise; see any physics
quantum mechanics textbook). Consequently, for an “orbiting” electron with
q = −e and mass me , the orbital angular momentum, thus also magnetic
moment, must be quantized,
eℏ
µjk ∈ integer × . (21.17)
2m
|{z}e
Bohr magneton

The main lesson is that the abstract concept of angular momentum is an

infinitesimal rotation, and is quantum mechanically represented as a Hilbert
space operator. Orbital angular momentum is an example, but it relies on
SO(d)-symmetry on the manifold X and the Hamiltonian to be a good con-
served quantity. In this sense, it is not geometrically “universal” or “intrinsic”.
Now, we have learned that an arbitrary oriented Riemannian X automati-
cally has gauge symmetry under the SO(d)-action on the oriented orthonormal
frames at each point x ∈ X. The tangent spaces intrinsically carry the funda-
mental representation of SO(d), thus also so(d). With parallel transport, we
can even compare the “intrinsic angular momenta” at different points. The
catch is that tangent spaces only provide a real representation of so(d) — these
are not quantum mechanical. For example, the representative antisymmetric
matrices will have no non-zero eigenvalues (over R). In any case, we need to
introduce i (see Eq. (21.16)) to get self-adjoint operators.
Instead, in view of Remark 6, what can be considered in quantum me-
chanics are the Spin(d)-torsors of spin frames at each manifold point. An
infinitesimal spin rotation is an element of spin(d), and we define the spin

233
angular momenta to be the “infinitesimal spin rotation”
iℏ
Σjk := ej ek ∈ spin(d), 1 ≤ j < k ≤ d. (21.18)
2
There is an important extra factor of 21 compared to (21.16), inherited from the
Lie algebra isomorphism χ−1∗ : so(d) → spin(d), which is needed to maintain
the Lie bracket relations of angular momenta, Eq. (21.15). Soon, we will
provide the concrete “spin-representation” of Spin(d), which will realize these
Σij as concrete Hermitian spin angular momentum matrices.
A key point is that each manifold point carries its own intrinsic spin-
representation space, generalizing the tangent space, so spin orbital momen-
tum is a gauge-theoretic concept. The spin-representation is an intrinsic “con-
served” label for the quantum particle in question, independent of whatever
Hamiltonian is specifying the dynamics.

21.7 Spin representations

Proposition 21.3, the initial cases Cl0 , Cl1 , and Cl2 ∼
= M2 (C) imply that
(
M2m (C), n = 2m even,
Cln ∼
=
M2m (C) ⊕ M2m (C), n = 2m + 1 odd.

Consequently, the representation theory of Cln is rather straightforward.

Even n = 2m. The algebra Cl2m ∼ = M2m (C) has only one irreducible repre-
m
sentation (over C) up to equivalence, namely, the defining one on C2 .

Odd n = 2m + 1. The chirality element splits in Cln into two chiral subal-
gebras,
1 ± ωC 1 ± ωC
Cln = Cln+ ⊕ Cln− , Cln± := Cln = Cln .
2 2
(Exercise.) Recall the parity automorphism α induced by v 7→ −v, which
+ −
swaps ω C for −ω C , thus Cl2m+1 for Cl2m+1 . In fact, chasing through the
periodicity isomorphism in Prop. 21.3, and the initial case Cl1 ∼
= C⊕C =
+ −
Cl1 ⊕ Cl1 , we obtain
Cl2m+1 ∼ +
= M2m (C) ⊕ M2m (C) = Cl2m+1 −
⊕ Cl2m+1 . (21.19)

234
Thus, there are two inequivalent representations of Cl2m+1 , according to whether
the second or first copy of M2m (C) is represented as 0; in turn, they are dis-
tinguished by whether the chirality element ω C is represented as +1 or as
−1.
We shall write S ± for the irreducible representation of Cl2m+1 which has
ω C acting as ±1, and refer to them as right-handed (+) and left-handed (−)
respectively33 .

Proposition 21.8. Let n ≥ 2 be even, and let c : Cln → End(S) be an

irreducible representation. Write S = S + ⊕ S − according to the ± eigenspaces
of c(ω C ). When restricted to the even subalgebra, c : Cln0 → End(S + ⊕ S − )
reduces to c+ : Cln0 → End(S + ) or c− : Cln0 → End(S − ). Furthermore, c+ and
c− are faithful but inequivalent irreducible representations of Cln0 .

Proof. In Cln , the even subalgebra Cln0 commutes with ω C . So the reduction
of c : Cln0 → End(S + ⊕ S − ) to S ± makes sense. Choose an isomorphism
Cln0 ∼
= Cln−1 , e.g., ẽi = ei en , i = 1, . . . , n − 1 (Exercise 21.1). Then the
chirality element for Cln0 ∼
= Cln−1 would be
n
C
ωn−1 = i 2 ẽ1 · · · ẽn−1 = (e1 en ) · · · (en−1 en )
n (n−1)(n−2)
= i 2 (−1) 2 (e1 · · · en−1 en )en−2
n
n
= i 2 e1 · · · en = ωnC ,

Now work in the irreducible representation c : Cln → End(S), which we

n
know has has S ∼ = C2 2 . Note that c(en ) is an invertible operator which an-
ticommutes with c(ωnC ), thus both S + and S − are non-trivial representation
n −1
spaces of Cln−1 of dimension 22 2 . Thus, they are the irreducible (and faith-
ful) representations of Cln−1 . Furthermore, c± (ωn−1
C
) = c± (ωnC ) = ±1, so c+
and c are inequivalent representations of Cln−1 ∼
−
= Cln0 .
Definition 94. Let c : Cln → End(S) be an irreducible complex representa-
tion of Cln . Restricting this to Spin(n) ⊂ Cln0 gives a homomorphism

δn : Spin(n) → GL(S),

called a spin representation of Spin(n).

33
Different authors have different conventions.

235
Proposition 21.9. For odd n, δn is an irreducible representation of Spin(n),
and is independent of which irreducible representation of Cln is used.
For even n, δn splits into δn+ ⊕ δn− acting on S + ⊕ S − , with δn+ and δn− being
inequivalent irreducible representations of Spin(n).
Proof. (Optional.) Odd n = 2m + 1. Recall the splitting of Cl2m+1 into chiral
subalgebras, Eq. (21.19), exchanged by the parity automorphism α. Since
0
Cl2m+1 is fixed under α, we have
0 + −
Cl2m+1 = {(φ, α(φ)) ∈ Cl2m+1 ⊕ Cl2m+1 }.
±
So the (inequivalent) irreducible representations c± : Cl2m+1 → Cl2m+1 →
End(S ± ) actually lead to equivalent irreducible representations
∼
=
c± : Cl2m+1
0 ±
→ Cl2m+1 → End(S ± ),

intertwined by α. In particular, the restrictions c± : Spin(2m + 1) → End(S ± )

are equivalent, and we call either one δ2m+1 . We observe that Spin(2m + 1)
0
contains the linear basis {eI }|I| even for Cl2m+1 , so δ2m+1 is also irreducible.
Even n = 2m. From Prop. 21.8, there are two inequivalent irreducible
representations c± : Cl2m 0
→ End(S ± ) labelled by whether ω2m C
acts as ±1.
±
These correspond to irreducible representations δ2m : Spin(2m) → End(S ± ).
Since (−i)m ω2m
C
= e1 . . . e2m ∈ Spin(2m) acts as different scalars ±(−i)m in the
two representations, they are inequivalent representations of Spin(2m).
Example 21.2. For Cl2 ∼
= M2 (C), we have S = S + ⊕ S − = C ⊕ C. Explicitly,
     
0 −i 0 −1 C 1 0
c(e1 ) = , c(e2 ) = , c(ω ) = ie1 e2 = .
−i 0 1 0 0 −1
Accordingly, the spin representation is
e−iθ 0
 
δ
Spin(2) ∋ cos θ + sin θe1 e2 7→ , (21.20)
0 eiθ

and clearly splits into inequivalent subrepresentations δ ± on S ± .

Example 21.3. The inequivalent irreducible representations of Cl3 on S ± ∼
= C2
are given by
     
0 −i 0 −1 −i 0
c(e1 ) = ± , c(e2 ) = ± , c(e3 ) = ± , (21.21)
−i 0 1 0 0 i

236
distinguished by whether c(ω C ) = c(−e1 e2 e3 ) = ±1. Either choice restricts to
the same representation for Cl30 = spanC {1, e1 e2 , e2 e3 , e3 e1 }, thus of Spin(3).
The standard basis for spin(3) is represented as
     
0 −i 0 −1 −i 0
c(e2 e3 ) = , c(e3 e1 ) = , c(e1 e2 ) = . (21.22)
−i 0 1 0 0 i
We had found this representation earlier, in Section 21.4.1, Eq. (21.10)–(21.11).
Recall (21.18), that the spin angular momenta are the spin(3) elements
iℏ iℏ iℏ
Σ23 = e2 e3 , Σ31 = e3 e1 , Σ12 = e1 e2 .
2 2 2
±
They are concretely represented on S as the spin angular momentum matrices
     
ℏ 0 1 ℏ 0 −i ℏ 1 0
S1 := c(Σ23 ) = , S2 := c(Σ31 ) = , S3 := c(Σ12 ) = ,
2 1 0 2 i 0 2 0 −1
(21.23)
first written down by Pauli in 1927.

Unitarity and orthogonal splitting of spin representations. For the

n = 0, 1, 2 cases, the concrete spin/Clifford algebra representations given above
have the operators c(ei ) being unitary, therefore also skew-adjoint (since they
square to −1), with respect to the standard inner product. This unitarity then
carries over to the general n case, when applying the periodicity isomorphisms.
Consequently, every unit vector y ∈ Rn has c(y) being a skew-adjoint unitary.
In particular, Spin(n) is unitarily represented in the spin representations.
In the even n case, the chirality element c(ω C ) is self-adjoint, orthogonally
splits S into S + ⊕ S − , and c(ei ) are odd operators exchanging S + with S − .

Clifford multiplication
Proposition 21.10. The spin representation δn is compatible with Clifford
multiplication c(v), v ∈ Rn , in the sense that for all g ∈ Spin(n), v ∈ Rn ,
δn (g)c(v)δn (g −1 ) = c χ(g)(v) .


(21.24)
Proof. Since δn (·) is the restriction of c : Cln → End(S), we have
(21.5)
δn (g)c(v)δn (g −1 ) = c(g)c(v)c(g −1 ) = c(gvg −1 ) = c χ(g)(v) .

237