Clifford Algebras and Pin and Spin Groups

A Clifford algebra converts a quadratic form into multiplication. For a real vector space $(V,g)$, it is the universal associative algebra in which

$$v^2=g(v,v)\mathbf 1, \qquad vw+wv=2g(v,w)\mathbf 1.$$

This one relation does three jobs. It makes every nonisotropic vector invertible, realizes the reflection orthogonal to that vector by twisted conjugation, and packages products of reflections into the Pin and Spin groups. The resulting homomorphisms

$$\operatorname{Pin}(V,g)\longrightarrow O(V,g), \qquad \operatorname{Spin}(V,g)\longrightarrow SO(V,g)$$

are surjective double covers, with kernel $\{\pm1\}$ under the nondegeneracy and dimension hypotheses stated below. A Clifford module then restricts to a representation of the Spin group. This is how a quadratic form constructs both the cover and its spin representations; gamma matrices are one realization of the algebra, not the construction itself.

The distinctions matter. A Clifford algebra, a Pin or Spin group, a spinor module, and a spin structure on a manifold are four different objects. Likewise, a double cover need not be a universal cover, and changing the sign in the Clifford relation changes real-algebra and Pin-lift data.

Required background. Direct Sums, Tensor Products, and Index Structure supplies tensor-algebra quotients, gradings, and universal constructions.

Helpful background. Groups, Actions, Quotients, and Covers supplies covering homomorphisms, kernels, and descent.

Quadratic spaces, Clifford algebras, and double covers

Let $V$ be a finite-dimensional real vector space and let $g:V\times V\to\mathbb R$ be symmetric. The algebra exists even when $g$ is degenerate, but the Pin and Spin covering statements below assume that $g$ is nondegenerate and $\dim V\geq2$. Except where degeneracy is mentioned explicitly, assume those hypotheses from now on. Write the signature as $(p,q)$, with $p$ positive and $q$ negative directions, and set

$$\operatorname{Cl}_{p,q} = \operatorname{Cl}(V,g).$$

The nondegeneracy hypothesis cannot simply be dropped. If $0\ne r\in\operatorname{rad}g$, then

$$r^2=g(r,r)\mathbf1=0.$$

The basis theorem below keeps $r$ nonzero in the Clifford algebra, so $r$ is nilpotent rather than invertible. It cannot lift a reflection by conjugation, and the double-cover sequences below no longer follow.

Thus the site’s four-dimensional metric convention is

$$\eta_{\mu\nu} = \operatorname{diag}(1,-1,-1,-1), \qquad \operatorname{Cl}_{1,3}, \qquad \{\gamma^\mu,\gamma^\nu\} = 2\eta^{\mu\nu}\mathbf1.$$

No gamma-matrix basis is preferred. Real and complex Clifford algebras will be distinguished explicitly:

$$\operatorname{Cl}_{p,q}^{\mathbb C} = \operatorname{Cl}_{p,q}\otimes_{\mathbb R}\mathbb C.$$

Many geometry sources instead impose $v^2=-g(v,v)\mathbf1$. Translating such a source to this page means replacing its quadratic form by $-g$ before reading a signature label or a Pin-lift square. The invariant round-trip check is always the anticommutator $vw+wv=2g(v,w)\mathbf1$.

This page constructs the algebra, its grading, the two covering groups, and the induced group action on Clifford modules. It does not classify all real Clifford algebras, develop reality or chirality conditions, construct spinor bilinears, build a spin structure, or derive fermion dynamics.

The quotient and its universal property

The tensor algebra

$$T(V) = \bigoplus_{r=0}^{\infty}V^{\otimes r}$$

is the free unital associative algebra generated by $V$. The Clifford algebra is its quotient

$$\operatorname{Cl}(V,g) = \frac{T(V)} {\left\langle v\otimes v-g(v,v)\mathbf1:\ v\in V \right\rangle_{\mathrm{two\text{-}sided}}}.$$

Polarizing the defining relation gives

$$\begin{aligned} (v+w)^2-v^2-w^2 &= vw+wv,\\ g(v+w,v+w)-g(v,v)-g(w,w) &= 2g(v,w), \end{aligned}$$

and hence $vw+wv=2g(v,w)\mathbf1$. Conversely, setting $w=v$ in the anticommutator recovers $v^2=g(v,v)\mathbf1$, so the two presentations are equivalent over a field of characteristic different from two.

The quotient is characterized without choosing a basis. If $A$ is any unital associative algebra and a linear map $c:V\to A$ obeys

$$c(v)^2=g(v,v)\mathbf1_A,$$

then there is a unique unital algebra homomorphism

$$\widehat c:\operatorname{Cl}(V,g)\longrightarrow A$$

whose restriction to $V$ is $c$. This universal property is the cleanest way to prove that a proposed set of gamma matrices defines a Clifford representation. It also shows that an isometry $f:(V,g)\to(V',g')$ induces an algebra homomorphism $\operatorname{Cl}(f)$; the construction is independent of coordinates.

The quotient construction, universal property, grading, and basis theorem are developed in Lawson and Michelsohn 1989, Chapter I, §1 and in Morgan 2022, Lecture IV, §1, PDF. Morgan uses $v^2=-Q(v)$; his $Q$ must therefore be replaced by $-g(v,v)$ to match this page.

Basis, filtration, and parity

Choose a $g$-orthogonal basis $e_1,\ldots,e_n$. Distinct basis vectors anticommute, and repeated factors reduce to scalars:

$$e_i e_j=-e_j e_i \quad(i\ne j), \qquad e_i^2=g(e_i,e_i)\mathbf1.$$

The ordered monomials

$$\mathbf1, \quad e_{i_1}, \quad e_{i_1}e_{i_2}, \quad\ldots,\quad e_{i_1}\cdots e_{i_r}, \qquad i_1<\cdots<i_r,$$

form a basis. Therefore

$$\dim_{\mathbb R}\operatorname{Cl}(V,g)=2^n.$$

The relation mixes tensor degrees two and zero, so the tensor algebra’s full $\mathbb Z$-grading does not survive. Parity does:

$$\operatorname{Cl}(V,g) = \operatorname{Cl}^0(V,g) \oplus \operatorname{Cl}^1(V,g),$$

where even times even and odd times odd are even, while an even–odd product is odd. The grade involution is the algebra automorphism

$$\alpha(x)= \begin{cases} x,&x\in\operatorname{Cl}^0(V,g),\\ -x,&x\in\operatorname{Cl}^1(V,g). \end{cases}$$

Filtering by tensor degree does survive, and its associated graded algebra is

$$\operatorname{gr}\operatorname{Cl}(V,g) \cong \bigwedge V.$$

This does not make the Clifford algebra the exterior algebra as an algebra. For example, $v\wedge v=0$ in $\bigwedge V$, whereas $v^2=g(v,v)\mathbf1$ in $\operatorname{Cl}(V,g)$. The two have the same dimension and associated graded vector-space pattern, but different multiplication.

Signature is real algebraic data

The smallest examples already detect the sign convention. For a one-dimensional real space,

$$\begin{aligned} \operatorname{Cl}_{1,0} &\cong \mathbb R[e]/(e^2-1) \cong \mathbb R\oplus\mathbb R,\\ \operatorname{Cl}_{0,1} &\cong \mathbb R[e]/(e^2+1) \cong \mathbb C. \end{aligned}$$

These real algebras are not isomorphic. The pair $(p,q)$ therefore cannot be dropped from a real Clifford-algebra claim. More generally, real classification exhibits an eightfold pattern: the division-algebra or Morita type is governed by $p-q$ modulo eight, while the total dimension sets the matrix size; see Lawson and Michelsohn 1989, Chapter I, §§3–4 and Morgan 2022, §§1.2–1.4, PDF.

After complexification, the distinction between positive and negative basis squares can be removed by multiplying a generator by $i$. Thus $\operatorname{Cl}_{p,q}^{\mathbb C}$ depends, up to isomorphism, only on $n=p+q$, not on the split $(p,q)$. It still depends on the parity of $n$, and choosing a real structure on a complex module restores signature-sensitive information.

This page needs no full periodicity table. Its invariant lesson is:

Data required in convention-sensitive Clifford and spinor claims

Object	Data that must remain visible
Real Clifford algebra	Scalar field, dimension, and signature (p,q)
Complex Clifford algebra	Complex scalar field and dimension
Pin lift	Clifford sign and the squares of reflection lifts
Spinor reality or chirality claim	Dimension, signature, and chosen real or complex structure

The last row is developed on Spinors, Conjugations, Bilinears, Chirality, and Fierz Identities.

One vector gives one reflection

Let $u\in V$ be nonisotropic. The Clifford relation makes it a unit:

$$u^{-1} = \frac{u}{g(u,u)}.$$

Not every Clifford unit preserves the generating vector space. The homogeneous Clifford group is

$$\Gamma(V,g) = \left\{ x\in\operatorname{Cl}(V,g)^\times: x\text{ is homogeneous and }xVx^{-1}=V \right\}.$$

For $x\in\Gamma(V,g)$, define its twisted adjoint action on $V$ by

$$\rho(x)(v) = \alpha(x)\,v\,x^{-1}.$$

Homogeneity gives $\rho(x)(v)^2=g(v,v)\mathbf1$, and the grade involution makes $\rho(xy)=\rho(x)\rho(y)$. Thus $\rho:\Gamma(V,g)\to O(V,g)$ is a group homomorphism.

For one vector $u$, use $uv=-vu+2g(u,v)\mathbf1$ to obtain

$$\begin{aligned} \rho(u)(v) &= -uvu^{-1}\\ &= v -2\frac{g(u,v)}{g(u,u)}u. \end{aligned}$$

This is precisely the orthogonal reflection $r_u$ in the hyperplane $u^\perp$: it sends $u$ to $-u$ and fixes every vector orthogonal to $u$. It depends only on the line through $u$, since $r_{\lambda u}=r_u$ for $\lambda\ne0$.

The twist is essential. Ordinary conjugation gives

$$uvu^{-1}=-r_u(v)$$

for the odd element $u$. On even products, $\alpha(x)=x$, so twisted and ordinary conjugation agree. This is why a source can appear to use ordinary conjugation for Spin while a Pin construction still requires the grade involution.

Every real orthogonal transformation of a nondegenerate finite-dimensional quadratic space is a product of reflections by the Cartan–Dieudonné theorem. The calculation above is therefore the local mechanism from which the covering groups arise; see Morgan 2022, §§2.1–2.3, PDF and Lawson and Michelsohn 1989, Chapter I, §2.

Pin and Spin as double covers

Normalize nonisotropic vectors so that $g(u,u)=\pm1$. Define

$$\operatorname{Pin}(V,g) = \left\langle u\in V:\ g(u,u)=\pm1 \right\rangle \subset \operatorname{Cl}(V,g)^\times$$

and define the full even subgroup

$$\operatorname{Spin}(V,g) = \operatorname{Pin}(V,g)\cap\operatorname{Cl}^0(V,g).$$

An element of Pin is a product of normalized vectors. Its image under $\rho$ is the corresponding product of reflections. An even product has determinant $+1$, so it lands in

$$SO(V,g) = O(V,g)\cap SL(V).$$

For nondegenerate $g$ and $\dim V\geq2$, Cartan–Dieudonné gives surjectivity, while the kernel calculation gives the exact sequences

$$\begin{aligned} 1&\longrightarrow\{\pm1\} \longrightarrow\operatorname{Pin}(V,g) \overset{\rho}{\longrightarrow}O(V,g) \longrightarrow1,\\ 1&\longrightarrow\{\pm1\} \longrightarrow\operatorname{Spin}(V,g) \overset{\rho}{\longrightarrow}SO(V,g) \longrightarrow1. \end{aligned}$$

The two elements $x$ and $-x$ therefore induce the same orthogonal transformation. That is the double cover. The generated-group construction, surjectivity, and kernel calculation are treated in Lawson and Michelsohn 1989, Chapter I, §2 and Gallier 2020, §§1.5 and 1.8–1.9, PDF.

If $\tau$ denotes reversal of the order of Clifford factors, then an element $x=u_1\cdots u_{2r}$ satisfies

$$\tau(x)x = \prod_{j=1}^{2r}g(u_j,u_j) \in \{\pm1\}$$

and the value can be $-1$ in indefinite signature. A definition that silently imposes $\tau(x)x=+1$ can therefore discard a legitimate component of the full even group. The generated-group definition above avoids that ambiguity.

There is a notation warning in indefinite signature. The determinant-one group $SO(p,q)$ can be disconnected, and the full even Pin subgroup above can be disconnected as well. Some references use $\operatorname{Spin}(p,q)$ only for the component covering the identity component $SO^+(p,q)$. On this page, the component is written explicitly. For the indefinite exact sequence, this page follows Gallier’s convention: $SO(p,q)$ is the full determinant-one subgroup and $\operatorname{Spin}(p,q)$ is the full even group. Morgan is used as a teaching source for the quotient and reflection mechanism; its opposite Clifford sign and its indefinite connectedness statement are not imported. In four-dimensional Lorentz signature, define the preimage inside the full even group:

$$\operatorname{Spin}^+(1,3) := \left\{ x\in\operatorname{Spin}(1,3): \rho(x)\in SO^+(1,3) \right\}.$$

Then

$$1\longrightarrow\{\pm1\} \longrightarrow\operatorname{Spin}^+(1,3) \longrightarrow SO^+(1,3) \longrightarrow1, \qquad \operatorname{Spin}^+(1,3)\cong SL(2,\mathbb C).$$

The four-dimensional identification and its action on spinors are treated in Lawson and Michelsohn 1989, Chapter I, §§2 and 4 and Tong 2006, §4.1. This particular cover is universal. A double cover is not automatically universal: for example, $\operatorname{Spin}(2)\to SO(2)$ is a twofold circle cover, while the universal cover of $SO(2)$ is $\mathbb R$.

The superscript in $\operatorname{Spin}^+(1,3)$ names the proper-orthochronous component. It is not the reflection-lift-square label in $\operatorname{Pin}^\pm(n)$ below.

What the superscripts on Pin mean

For a normalized reflection vector, the chosen lift satisfies

$$u^2=g(u,u)=\pm1.$$

In positive-definite Euclidean signature, the convention $v^2=g(v,v)$ gives reflection lifts whose square is $+1$. This realizes the cover conventionally called $\operatorname{Pin}^+(n)$. Replacing $g$ by $-g$ gives lifts whose square is $-1$ and realizes $\operatorname{Pin}^-(n)$.

This reflection-square naming is stated explicitly in Tachikawa 2024, §10.4, PDF. If a source starts from $v^2=-g(v,v)$, the association reverses. The labels $+$ and $-$ are therefore not decoration and must not be inferred from a bare word “Pin.” In indefinite signature, state $(p,q)$ and record separately which timelike and spacelike reflection lifts square to $+1$ or $-1$.

This distinction affects the extension from orientation-preserving transformations to reflections. It does not change the fact that the even subgroup supplies the Spin cover of the appropriate determinant-one orthogonal group.

Bivectors are the infinitesimal spin generators

The even algebra contains a canonical copy of the orthogonal Lie algebra. For $u,v\in V$, set

$$\iota(u\wedge v) = \frac14(uv-vu) \in \operatorname{Cl}^0(V,g).$$

A direct use of the Clifford relation gives

$$\left[ \iota(u\wedge v),z \right] = g(v,z)u-g(u,z)v, \qquad z\in V.$$

The right-hand side is the standard $g$-skew endomorphism associated with $u\wedge v$. Thus

$$\mathfrak{spin}(V,g) \cong \bigwedge\nolimits^2 V \cong \mathfrak{so}(V,g)$$

as Lie algebras, with the Clifford commutator on the left and the endomorphism commutator on the right. The Lie algebras agree locally even though the groups differ globally by a two-element kernel; see Lawson and Michelsohn 1989, Chapter I, §6.

For orthogonal $u$ and $v$ the embedding simplifies to $\iota(u\wedge v)=\tfrac12uv$. Exponentiating a bivector produces a rotor, an even Clifford unit whose conjugation action is an orthogonal transformation. The factor of one half in the exponent is the algebraic source of the doubled angular period of spinors.

Controlled cover example: one plane

Let $e_1,e_2$ span a definite two-plane with

$$e_1^2=e_2^2=\varepsilon, \qquad \varepsilon\in\{+1,-1\}, \qquad e_1e_2=-e_2e_1.$$

For $B=e_1e_2$,

$$B^2=-1, \qquad [B,e_1]=-2\varepsilon e_2, \qquad [B,e_2]=2\varepsilon e_1.$$

Define

$$s(\theta) = \exp\left(-\frac{\varepsilon\theta}{2}B\right) = \cos\frac{\theta}{2} -\varepsilon B\sin\frac{\theta}{2}.$$

This is a Spin element rather than merely an even unit: if $u(\theta)=\cos(\theta/2)e_1+\sin(\theta/2)e_2$, then

$$s(\theta)=\varepsilon\,u(\theta)e_1, \qquad u(\theta)^2=e_1^2=\varepsilon,$$

so it is a product of normalized reflection lifts, up to the kernel element $\varepsilon=\pm1$.

Because $s(\theta)$ is even, its action is ordinary conjugation. Expanding the exponential or differentiating with respect to $\theta$ gives

$$\begin{aligned} s(\theta)e_1s(\theta)^{-1} &= \cos\theta\,e_1+\sin\theta\,e_2,\\ s(\theta)e_2s(\theta)^{-1} &= -\sin\theta\,e_1+\cos\theta\,e_2. \end{aligned}$$

The image is an ordinary rotation through $\theta$, but

$$s(2\pi)=-1, \qquad s(4\pi)=+1.$$

Thus a $2\pi$ rotation is the identity in $SO(2)$ but reaches the nontrivial kernel element in $\operatorname{Spin}(2)$. Only after $4\pi$ does the lift return to the identity. The same calculation works for a positive or negative definite plane; the factor $\varepsilon$ translates the generator sign.

Clifford modules furnish spin representations

A left Clifford module is a vector space $S$ together with a unital algebra homomorphism

$$c:\operatorname{Cl}(V,g) \longrightarrow \operatorname{End}(S).$$

Since every Pin element is a unit, restriction gives group representations

$$c:\operatorname{Pin}(V,g)\longrightarrow GL(S), \qquad c:\operatorname{Spin}(V,g)\longrightarrow GL(S).$$

For $x\in\operatorname{Spin}(V,g)$ and $v\in V$,

$$c(x)c(v)c(x)^{-1} = c\!\left(\rho(x)v\right).$$

This is the compatibility that makes $c(v)$ transform as a vector while vectors in $S$ transform as spinors. Moreover,

$$c(-1)=-\mathbf1_S.$$

The nontrivial kernel element therefore acts nontrivially on every nonzero unital Clifford module. The resulting Spin representation does not descend to an ordinary representation of $SO(V,g)$.

The terminology has a controlled ambiguity. An irreducible Clifford module is often called a spinor module, and representations obtained by restriction are spin representations. Some Spin representations do not extend to modules of the full Clifford algebra without additional choices. Irreducibility can also change on restriction to the even algebra. Those classification, chirality, conjugation, reality, and bilinear questions belong to the next page. The module-to-group construction here is supported by Lawson and Michelsohn 1989, Chapter I, §5.

Gamma matrices are simply the images

$$\gamma(v)=c(v), \qquad \gamma(v)\gamma(w)+\gamma(w)\gamma(v) = 2g(v,w)\mathbf1_S$$

in a chosen module and basis. A matrix list satisfying this relation is a representation of the Clifford algebra; it is not the abstract algebra itself.

Controlled QFT example: a relativistic spinor transformation

Now complexify the four-dimensional Lorentzian algebra and choose a complex Clifford module $S$ with

$$\{\gamma^\mu,\gamma^\nu\} = 2\eta^{\mu\nu}\mathbf1_S.$$

No explicit matrices are needed. With the site’s $i$-weighted Lorentz generator convention, define

$$M_{\mathrm{spin}}^{\mu\nu} = \frac{i}{4} \left[\gamma^\mu,\gamma^\nu\right].$$

The Clifford relation alone yields

$$\left[ M_{\mathrm{spin}}^{\mu\nu},\gamma^\rho \right] = i\left( \eta^{\nu\rho}\gamma^\mu -\eta^{\mu\rho}\gamma^\nu \right),$$

so these matrices implement the infinitesimal Lorentz action on gamma vectors and satisfy the Lorentz commutators. This convention translation is the algebraic core of the finite spinor transformation in Tong 2006, §4.1.

For a rotation through $\theta$ about the third spatial axis,

$$J_3 = M_{\mathrm{spin}}^{12} = \frac{i}{2}\gamma^1\gamma^2, \qquad (\gamma^1\gamma^2)^2=-1.$$

Hence

$$\begin{aligned} S_z(\theta) &= \exp(-i\theta J_3)\\ &= \exp\left( \frac{\theta}{2}\gamma^1\gamma^2 \right)\\ &= \cos\frac{\theta}{2} +\gamma^1\gamma^2\sin\frac{\theta}{2}. \end{aligned}$$

Conjugation gives

$$\begin{aligned} S_z(\theta)\gamma^1S_z(\theta)^{-1} &= \cos\theta\,\gamma^1+\sin\theta\,\gamma^2,\\ S_z(\theta)\gamma^2S_z(\theta)^{-1} &= -\sin\theta\,\gamma^1+\cos\theta\,\gamma^2, \end{aligned}$$

while $\gamma^0$ and $\gamma^3$ are unchanged. At one full turn,

$$S_z(2\pi)=-\mathbf1_S, \qquad \Lambda_z(2\pi)=\mathbf1_{\mathbb R^{1,3}}.$$

For $A\in\operatorname{Spin}^+(1,3)$ projecting to $\Lambda(A)\in SO^+(1,3)$, write $S(A)=c(A)$ for the restricted module representation. The corresponding kinematic field law is

$$\bigl(A\cdot\psi\bigr)(x) = S(A)\, \psi\!\left(\Lambda(A)^{-1}x\right),$$

with the covariance identity

$$S(A)\gamma(v)S(A)^{-1} = \gamma\!\left(\Lambda(A)v\right).$$

This answers the first application question: the Lorentz transformation of a relativistic fermion uses the Spin lift, not merely the projected orthogonal matrix. It does not yet choose a Dirac action, impose an equation of motion, quantize the field, identify one-particle states, or develop bilinears. Those physical steps belong to The Dirac Field, while spinor conjugations and bilinears belong to the next mathematical leaf.

What is not constructed here

The following distinctions are structural, not terminological:

Distinct algebraic, geometric, and physical objects

Object	What it is	What additional data it needs
Cl(V,g)	An associative algebra	Scalar field and quadratic form
Pin(V,g) or Spin(V,g)	A subgroup of Clifford units	Nondegenerate form and declared component convention
Spinor module S	A module or group representation	Real or complex representation choice
Spin structure	A lift of an oriented orthonormal frame bundle	A manifold, metric, bundle topology, and existence choice
Dirac field	A physical field theory	Spacetime assumptions, action, dynamics, and quantization data

A vector space with gamma matrices need not be a spin structure. A manifold whose tangent spaces admit local Clifford algebras need not admit a global spin structure. A Spin representation by itself does not specify a fermion theory or its particle content. These are the exact boundaries enforced by the later geometry and Foundations pages.

Sign and cover checks

When Clifford or spin notation appears in a calculation:

1. State the scalar field, dimension, signature, and whether $v^2=+g(v,v)$ or $v^2=-g(v,v)$.
2. Recover the anticommutator and test one positive- and one negative-norm basis vector.
3. For a reflection, use the twisted adjoint and verify its action on the normal vector and its orthogonal complement.
4. State whether the full orthogonal group, determinant-one subgroup, or identity component is being covered.
5. Test the central element $-1$ on the proposed module before claiming descent to an orthogonal group.
6. Separate the algebraic module from any later spin structure or physical fermion field.

This sequence catches most convention errors before they propagate into a gamma-matrix or fermion calculation.

Common pitfalls

Calling gamma matrices “the Clifford algebra.” Gamma matrices are the images of generators in one representation. Different matrix representations can realize the same abstract algebra, and a representation can fail to be faithful.

Using ordinary conjugation for an odd reflection lift. For a vector $u$, ordinary conjugation gives minus the desired hyperplane reflection. The twisted adjoint $\alpha(u)vu^{-1}$ supplies the reflection; the twist disappears only for even Spin elements.

Dropping the sign convention from a signature table. The relation $v^2=g(v,v)$ labels real Clifford algebras oppositely from the common geometry convention $v^2=-g(v,v)$. Translate the defining relation first, then read the table.

Treating every double cover as universal. The kernel $\{\pm1\}$ establishes a twofold cover, not simple connectedness. $\operatorname{Spin}(2)\to SO(2)$ is the standard counterexample.

Conflating the even subgroup with an identity component. In indefinite signature, $SO(p,q)$ and the full even Pin subgroup can be disconnected. Define the intended component explicitly. This page uses the superscript $+$ only for the declared four-dimensional preimage of $SO^+(1,3)$; it does not impose that notation in every signature.

Calling every even Clifford unit a Spin element. Spin consists of even products of normalized nonisotropic vectors, not all of $\operatorname{Cl}^0(V,g)^\times$. In indefinite signature, replacing that generated group by an undeclared reversion-norm-one subgroup can also lose components.

Equating the Clifford and exterior products. Their associated graded structures agree, but $v^2=g(v,v)$ in the Clifford algebra and $v\wedge v=0$ in the exterior algebra.

Inferring a spin structure or a fermion theory from a Spin group. A spin structure is global bundle data, and a fermion theory additionally needs dynamics and quantization. Neither follows from the group construction alone.

Exercises

1. Compare the one-dimensional real algebras and Pin groups

Let $e^2=+1$ in one case and $f^2=-1$ in the other. Construct explicit isomorphisms

$$\mathbb R[e]/(e^2-1)\cong\mathbb R\oplus\mathbb R, \qquad \mathbb R[f]/(f^2+1)\cong\mathbb C.$$

Why does this rule out an isomorphism between the two real algebras? Then identify the groups generated by the normalized reflection lift in each case.

Solution

Send

$$a+be\longmapsto(a+b,a-b).$$

Multiplication is preserved because the two coordinates evaluate the polynomial at $e=+1$ and $e=-1$. Send

$$a+bf\longmapsto a+bi$$

in the second case. The first algebra has the nonzero zero divisors $(1,0)$ and $(0,1)$, whereas $\mathbb C$ is a field. Therefore the real algebras are not isomorphic.

In the positive-square case,

$$\operatorname{Pin}^+(1) = \{1,-1,e,-e\} \cong \mathbb Z_2\times\mathbb Z_2,$$

because every nonidentity generator has square $1$. In the negative-square case,

$$\operatorname{Pin}^-(1) = \{1,-1,f,-f\} \cong \mathbb Z_4,$$

because $f^2=-1$ and $f$ has order four. The same orthogonal reflection therefore has lifts with different group-theoretic squares.

2. Derive the reflection and diagnose the missing twist

For nonisotropic $u$, derive both

$$-uvu^{-1} = v-2\frac{g(u,v)}{g(u,u)}u$$

and the action of ordinary conjugation $uvu^{-1}$ on $u$ and on $u^\perp$.

Solution

From $uv=-vu+2g(u,v)$ and $u^{-1}=u/g(u,u)$,

$$\begin{aligned} -uvu^{-1} &= -\left(-vu+2g(u,v)\right)u^{-1}\\ &= v-2\frac{g(u,v)}{g(u,u)}u. \end{aligned}$$

The twisted action sends $u$ to $-u$ and fixes $u^\perp$, so it is the reflection in $u^\perp$. Ordinary conjugation does the opposite: it fixes $u$ and sends every $w\in u^\perp$ to $-w$. It is minus the hyperplane reflection.

3. Verify the doubled angle in a definite plane

For $e_1^2=e_2^2=\varepsilon$ and $B=e_1e_2$, verify the formula for $s(\theta)$ above and determine its values at $2\pi$ and $4\pi$.

Solution

Anticommutation gives

$$B^2 = e_1e_2e_1e_2 = -e_1^2e_2^2 = -1.$$

Therefore

$$\exp\left(-\frac{\varepsilon\theta}{2}B\right) = \cos\frac{\theta}{2} -\varepsilon B\sin\frac{\theta}{2}.$$

At $\theta=2\pi$ this is $-1$; at $\theta=4\pi$ it is $+1$. Differentiating $s(\theta)e_is(\theta)^{-1}$ and using the displayed commutators shows that its image rotates through the full angle $\theta$.

4. Recover the orthogonal generator from a bivector

Show directly that

$$\left[ \frac14(uv-vu),z \right] = g(v,z)u-g(u,z)v.$$

Check that the resulting linear map is $g$-skew.

Solution

Repeatedly use $vz=-zv+2g(v,z)$:

$$[uv,z] = 2g(v,z)u-2g(u,z)v.$$

Similarly,

$$[vu,z] = 2g(u,z)v-2g(v,z)u.$$

Subtracting and dividing by four gives the required formula. If $A(z)=g(v,z)u-g(u,z)v$, then

$$g(Az,w)+g(z,Aw)=0$$

after expanding the four scalar products, so $A\in\mathfrak{so}(V,g)$.

5. Check a 2π fermion rotation without choosing matrices

Using only

$$(\gamma^1)^2=(\gamma^2)^2=-1, \qquad \gamma^1\gamma^2=-\gamma^2\gamma^1,$$

derive $S_z(\theta)$ and compare its action at $2\pi$ on a Lorentz vector and on a spinor.

Solution

The product obeys $(\gamma^1\gamma^2)^2=-1$, so

$$S_z(\theta) = \cos\frac{\theta}{2} +\gamma^1\gamma^2\sin\frac{\theta}{2}.$$

Its conjugation action rotates $\gamma^1,\gamma^2$ by $\theta$, hence the projected vector transformation at $2\pi$ is the identity. But $S_z(2\pi)=-\mathbf1_S$, so the spinor changes sign. A second turn gives $S_z(4\pi)=+\mathbf1_S$.

Where to continue

• For the first physical application to relativistic fermion transformation laws, continue to The Dirac Field, after its action-principle and one-particle-state prerequisites.
• Spinors, Conjugations, Bilinears, Chirality, and Fierz Identities hard-requires this page and the representation page. It develops dimension- and signature-dependent module classifications, conjugations, reality, chirality, bilinears, and Fierz identities.
• Spin Structures and Dirac Operators hard-requires this page together with independent curvature and bundle inputs. It is the destination for global geometric lifts and Dirac operators.
• The page also provides required input for later treatments of CPT hypotheses, tangential structures in extended TQFT, and dimension-dependent supersymmetry reality conditions. Those destinations retain all of their other prerequisites.

References

• Jean Gallier, Clifford Algebras, Clifford Groups, and a Generalization of the Quaternions: The Pin and Spin Groups, PDF, §§1.2–1.5 and 1.8–1.9, University of Pennsylvania, 2020. This open structural reference uses $\operatorname{Cl}_{p,q}$ with positive generators squaring to $+1$, defines $SO(p,q)$ as the full determinant-one group, and proves that the corresponding full Pin and Spin groups are double covers. Those are the component and signature conventions used here.
• H. Blaine Lawson Jr. and Marie-Louise Michelsohn, Spin Geometry, Princeton Mathematical Series 38, Princeton University Press, 1989, Chapter I, §§1–6. These sections establish Clifford algebras, Pin and Spin groups, Clifford modules, spin representations, and the relevant Lie structures. Its later bundle and Dirac-operator material is reserved for the geometric continuation.
• John W. Morgan, Lie Groups: Fall 2022, Lecture IV—Clifford Algebras and the Spin Groups, PDF, §§1 and 2.1–2.4, Columbia University, 2022. These open lecture notes develop the quotient, grading, reflection, and covering constructions. They use $v^2=-Q(v)$; every cited identity above has been translated by $Q=-g$ and checked against $vw+wv=2g(v,w)$.
• Yuji Tachikawa (2024), Algebraic Topology for Physicists, PDF, §10.4, “Pin groups and pinors,” Kavli IPMU lecture notes. This is the convention source for the names $\operatorname{Pin}^+(n)$ and $\operatorname{Pin}^-(n)$ as the covers whose reflection lifts square to $+1$ and $-1$, respectively.
• David Tong (2006), Quantum Field Theory, §4, “The Dirac Equation”, especially §4.1, Cambridge Part III lecture notes. This section derives gamma matrices, Lorentz spin generators, and the bounded relativistic spinor-transformation example. Generator and metric signs have been translated to the conventions declared above.