Spin representation

In mathematics, the spin representations are particular projective representations of the orthogonal or special orthogonal groups in arbitrary dimension and signature (i.e., including indefinite orthogonal groups). More precisely, they are two equivalent representations of the spin groups, which are double covers of the special orthogonal groups. They are usually studied over the real or complex numbers, but they can be defined over other fields.

Elements of a spin representation are called spinors. They play an important role in the physical description of fermions such as the electron.

The spin representations may be constructed in several ways, but typically the construction involves (perhaps only implicitly) the choice of a maximal isotropic subspace in the vector representation of the group. Over the real numbers, this usually requires using a complexification of the vector representation. For this reason, it is convenient to define the spin representations over the complex numbers first, and derive real representations by introducing real structures.

The properties of the spin representations depend, in a subtle way, on the dimension and signature of the orthogonal group. In particular, spin representations often admit invariant bilinear forms, which can be used to embed the spin groups into classical Lie groups. In low dimensions, these embeddings are surjective and determine special isomorphisms between the spin groups and more familiar Lie groups; this elucidates the properties of spinors in these dimensions.

Set-up
Let $V$ be a finite-dimensional real or complex vector space with a nondegenerate quadratic form $Q$. The (real or complex) linear maps preserving $Q$ form the orthogonal group $O(V, Q)$. The identity component of the group is called the special orthogonal group $SO(V, Q)$. (For $V$ real with an indefinite quadratic form, this terminology is not standard: the special orthogonal group is usually defined to be a subgroup with two components in this case.) Up to group isomorphism, $SO(V, Q)$ has a unique connected double cover, the spin group $Spin(V, Q)$. There is thus a group homomorphism $h: Spin(V, Q) → SO(V, Q)$ whose kernel has two elements denoted ${1, −1}$, where $1$ is the identity element. Thus, the group elements $g$ and $−g$ of $Spin(V, Q)$ are equivalent after the homomorphism to $SO(V, Q)$; that is, $h(g) = h(−g)$ for any $g$ in $Spin(V, Q)$.

The groups $O(V, Q), SO(V, Q)$ and $Spin(V, Q)$ are all Lie groups, and for fixed $(V, Q)$ they have the same Lie algebra, $so(V, Q)$. If $V$ is real, then $V$ is a real vector subspace of its complexification $V_{C} = V ⊗_{R} C$, and the quadratic form $Q$ extends naturally to a quadratic form $Q_{C}$ on $V_{C}$. This embeds $SO(V, Q)$ as a subgroup of $SO(V_{C}, Q_{C})$, and hence we may realise $Spin(V, Q)$ as a subgroup of $Spin(V_{C}, Q_{C})$. Furthermore, $so(V_{C}, Q_{C})$ is the complexification of $so(V, Q)$.

In the complex case, quadratic forms are determined uniquely up to isomorphism by the dimension $n$ of $V$. Concretely, we may assume $V = C^{n}$ and
 * $$Q(z_1,\ldots, z_n) = z_1^2+ z_2^2+\cdots+z_n^2.$$

The corresponding Lie groups are denoted $O(n, C), SO(n, C), Spin(n, C)$ and their Lie algebra as $so(n, C)$.

In the real case, quadratic forms are determined up to isomorphism by a pair of nonnegative integers $(p, q)$ where $n = p + q$ is the dimension of $V$, and $p − q$ is the signature. Concretely, we may assume $V = R^{n}$ and
 * $$Q(x_1,\ldots, x_n) = x_1^2+ x_2^2+\cdots+x_p^2-(x_{p+1}^2+\cdots +x_{p+q}^2).$$

The corresponding Lie groups and Lie algebra are denoted $O(p, q), SO(p, q), Spin(p, q)$ and $so(p, q)$. We write $R^{p,q}$ in place of $R^{n}$ to make the signature explicit.

The spin representations are, in a sense, the simplest representations of $Spin(n, C)$ and $Spin(p, q)$ that do not come from representations of $SO(n, C)$ and $SO(p, q)$. A spin representation is, therefore, a real or complex vector space $S$ together with a group homomorphism $ρ$ from $Spin(n, C)$ or $Spin(p, q)$ to the general linear group $GL(S)$ such that the element $−1$ is not in the kernel of $ρ$.

If $S$ is such a representation, then according to the relation between Lie groups and Lie algebras, it induces a Lie algebra representation, i.e., a Lie algebra homomorphism from $so(n, C)$ or $so(p, q)$ to the Lie algebra $gl(S)$ of endomorphisms of $S$ with the commutator bracket.

Spin representations can be analysed according to the following strategy: if $S$ is a real spin representation of $Spin(p, q)$, then its complexification is a complex spin representation of $Spin(p, q)$; as a representation of $so(p, q)$, it therefore extends to a complex representation of $so(n, C)$. Proceeding in reverse, we therefore first construct complex spin representations of $Spin(n, C)$ and $so(n, C)$, then restrict them to complex spin representations of $so(p, q)$ and $Spin(p, q)$, then finally analyse possible reductions to real spin representations.

Complex spin representations
Let $V = C^{n}$ with the standard quadratic form $Q$ so that
 * $$\mathfrak{so}(V,Q) = \mathfrak{so}(n,\mathbb C).$$

The symmetric bilinear form on $V$ associated to $Q$ by polarization is denoted $⟨.,.⟩$.

Isotropic subspaces and root systems
A standard construction of the spin representations of $so(n, C)$ begins with a choice of a pair $(W, W^{∗})$ of maximal totally isotropic subspaces (with respect to $Q$) of $V$ with $W ∩ W^{∗} = 0$. Let us make such a choice. If $n = 2m$ or $n = 2m + 1$, then $W$ and $W^{∗}$ both have dimension $m$. If $n = 2m$, then $V = W ⊕ W^{∗}$, whereas if $n = 2m + 1$, then $V = W ⊕ U ⊕ W^{∗}$, where $U$ is the 1-dimensional orthogonal complement to $W ⊕ W^{∗}$. The bilinear form $⟨.,.⟩$ associated to $Q$ induces a pairing between $W$ and $W^{∗}$, which must be nondegenerate, because $W$ and $W^{∗}$ are totally isotropic subspaces and $Q$ is nondegenerate. Hence $W$ and $W^{∗}$ are dual vector spaces.

More concretely, let $a_{1}, ... a_{m}$ be a basis for $W$. Then there is a unique basis $&alpha;_{1}, ... &alpha;_{m}$ of $W^{∗}$ such that
 * $$ \langle \alpha_i,a_j\rangle = \delta_{ij}.$$

If $A$ is an $m &times; m$ matrix, then $A$ induces an endomorphism of $W$ with respect to this basis and the transpose $A^{T}$ induces a transformation of $W^{∗}$ with
 * $$ \langle Aw, w^* \rangle = \langle w,A^\mathrm{T} w^*\rangle$$

for all $w$ in $W$ and $w^{∗}$ in $W^{∗}$. It follows that the endomorphism $&rho;_{A}$ of $V$, equal to $A$ on $W$, $&minus;A^{T}$ on $W^{∗}$ and zero on $U$ (if $n$ is odd), is skew,
 * $$ \langle \rho_A u, v \rangle = -\langle u,\rho_A v\rangle$$

for all $u, v$ in $V$, and hence (see classical group) an element of $so(n, C) ⊂ End(V)$.

Using the diagonal matrices in this construction defines a Cartan subalgebra $h$ of $so(n, C)$: the rank of $so(n, C)$ is $m$, and the diagonal $n &times; n$ matrices determine an $m$-dimensional abelian subalgebra.

Let $ε_{1}, ... ε_{m}$ be the basis of $h^{∗}$ such that, for a diagonal matrix $A, ε_{k}(&rho;_{A})$ is the $k$th diagonal entry of $A$. Clearly this is a basis for $h^{∗}$. Since the bilinear form identifies $so(n, C)$ with $$\wedge^2 V$$, explicitly,
 * $$x \wedge y \mapsto \varphi_{x \wedge y}, \quad \varphi_{x \wedge y}(v) = 2(\langle y, v\rangle x - \langle x, v\rangle y),\quad x \wedge y \in \wedge^2V,\quad x,y,v \in V, \quad \varphi_{x \wedge y} \in \mathfrak{so}(n, \mathbb{C}),$$

it is now easy to construct the root system associated to $h$. The root spaces (simultaneous eigenspaces for the action of $h$) are spanned by the following elements:
 * $$ a_i\wedge a_j,\; i\neq j,$$ with root (simultaneous eigenvalue) $$\varepsilon_i + \varepsilon_j$$
 * $$ a_i\wedge \alpha_j$$ (which is in $h$ if $i = j)$ with root $$ \varepsilon_i - \varepsilon_j$$
 * $$ \alpha_i\wedge \alpha_j,\; i\neq j,$$ with root $$ -\varepsilon_i - \varepsilon_j,$$

and, if $n$ is odd, and $u$ is a nonzero element of $U$,
 * $$ a_i\wedge u,$$ with root $$ \varepsilon_i $$
 * $$ \alpha_i\wedge u,$$ with root $$ -\varepsilon_i.$$

Thus, with respect to the basis $ε_{1}, ... ε_{m}$, the roots are the vectors in $h^{∗}$ that are permutations of
 * $$(\pm 1,\pm 1, 0, 0, \dots, 0)$$

together with the permutations of
 * $$(\pm 1, 0, 0, \dots, 0)$$

if $n = 2m + 1$ is odd.

A system of positive roots is given by $ε_{i} + ε_{j} (i ≠ j), ε_{i} &minus; ε_{j} (i < j)$ and (for $n$ odd) $ε_{i}$. The corresponding simple roots are
 * $$\varepsilon_1-\varepsilon_2, \varepsilon_2-\varepsilon_3, \ldots, \varepsilon_{m-1}-\varepsilon_m, \left\{\begin{matrix}

\varepsilon_{m-1}+\varepsilon_m& n=2m\\ \varepsilon_m & n=2m+1. \end{matrix}\right.$$ The positive roots are nonnegative integer linear combinations of the simple roots.

Spin representations and their weights
One construction of the spin representations of $so(n, C)$ uses the exterior algebra(s)
 * $$S=\wedge^\bullet W$$ and/or $$S'=\wedge^\bullet W^*.$$

There is an action of $V$ on $S$ such that for any element $v = w + w^{∗}$ in $W ⊕ W^{∗}$ and any $&psi;$ in $S$ the action is given by:
 * $$ v\cdot \psi = 2^{\frac{1}{2}}(w\wedge\psi+\iota(w^*)\psi), $$

where the second term is a contraction (interior multiplication) defined using the bilinear form, which pairs $W$ and $W^{∗}$. This action respects the Clifford relations $v^{2} = Q(v)1$, and so induces a homomorphism from the Clifford algebra $Cl_{n}C$ of $V$ to $End(S)$. A similar action can be defined on $S&prime;$, so that both $S$ and $S&prime;$ are Clifford modules.

The Lie algebra $so(n, C)$ is isomorphic to the complexified Lie algebra $spin_{n}^{C}$ in $Cl_{n}C$ via the mapping induced by the covering $Spin(n) → SO(n)$
 * $$ v \wedge w \mapsto \tfrac14[v,w].$$

It follows that both $S$ and $S&prime;$ are representations of $so(n, C)$. They are actually equivalent representations, so we focus on S.

The explicit description shows that the elements $&alpha;_{i} ∧ a_{i}$ of the Cartan subalgebra $h$ act on $S$ by
 * $$ (\alpha_i\wedge a_i) \cdot \psi = \tfrac14 (2^{\tfrac12})^{2} ( \iota(\alpha_i)(a_i\wedge\psi)-a_i\wedge(\iota(\alpha_i)\psi))

= \tfrac12 \psi - a_i\wedge(\iota(\alpha_i)\psi).$$ A basis for $S$ is given by elements of the form
 * $$ a_{i_1}\wedge a_{i_2}\wedge\cdots\wedge a_{i_k}$$

for $0 ≤ k ≤ m$ and $i_{1} < ... < i_{k}$. These clearly span weight spaces for the action of $h$: $&alpha;_{i} ∧ a_{i}$ has eigenvalue −1/2 on the given basis vector if $i = i_{j}$ for some $j$, and has eigenvalue $1/2$ otherwise.

It follows that the weights of $S$ are all possible combinations of
 * $$\bigl(\pm \tfrac12,\pm \tfrac12, \ldots \pm\tfrac12\bigr)$$

and each weight space is one-dimensional. Elements of $S$ are called Dirac spinors.

When $n$ is even, $S$ is not an irreducible representation: $$S_+=\wedge^{\mathrm{even}} W$$ and $$S_-=\wedge^{\mathrm{odd}} W$$ are invariant subspaces. The weights divide into those with an even number of minus signs, and those with an odd number of minus signs. Both S+ and S&minus; are irreducible representations of dimension 2m&minus;1 whose elements are called Weyl spinors. They are also known as chiral spin representations or half-spin representations. With respect to the positive root system above, the highest weights of S+ and S&minus; are
 * $$\bigl(\tfrac12,\tfrac12, \ldots\tfrac12, \tfrac12\bigr)$$ and $$\bigl(\tfrac12,\tfrac12, \ldots\tfrac12, -\tfrac12\bigr)$$

respectively. The Clifford action identifies ClnC with End(S) and the even subalgebra is identified with the endomorphisms preserving S+ and S&minus;. The other Clifford module S&prime; is isomorphic to S in this case.

When n is odd, S is an irreducible representation of so(n,C) of dimension 2m: the Clifford action of a unit vector u ∈ U is given by
 * $$ u\cdot \psi = \left\{\begin{matrix}

\psi&\hbox{if } \psi\in \wedge^{\mathrm{even}} W\\ -\psi&\hbox{if } \psi\in \wedge^{\mathrm{odd}} W \end{matrix}\right.$$ and so elements of so(n,C) of the form u∧w or u∧w∗ do not preserve the even and odd parts of the exterior algebra of W. The highest weight of S is
 * $$\bigl(\tfrac12,\tfrac12, \ldots \tfrac12\bigr).$$

The Clifford action is not faithful on S: ClnC can be identified with End(S) ⊕ End(S&prime;), where u acts with the opposite sign on S&prime;. More precisely, the two representations are related by the parity involution &alpha; of ClnC (also known as the principal automorphism), which is the identity on the even subalgebra, and minus the identity on the odd part of ClnC. In other words, there is a linear isomorphism from S to S&prime;, which identifies the action of A in ClnC on S with the action of &alpha;(A) on S&prime;.

Bilinear forms
if &lambda; is a weight of S, so is &minus;&lambda;. It follows that S is isomorphic to the dual representation S∗.

When n = 2m + 1 is odd, the isomorphism B: S → S∗ is unique up to scale by Schur's lemma, since S is irreducible, and it defines a nondegenerate invariant bilinear form &beta; on S via
 * $$\beta(\varphi,\psi) = B(\varphi)(\psi).$$

Here invariance means that
 * $$\beta(\xi\cdot\varphi,\psi) + \beta(\varphi,\xi\cdot\psi) = 0$$

for all &xi; in so(n,C) and &phi;, &psi; in S &mdash; in other words the action of &xi; is skew with respect to &beta;. In fact, more is true: S∗ is a representation of the opposite Clifford algebra, and therefore, since ClnC only has two nontrivial simple modules S and S&prime;, related by the parity involution &alpha;, there is an antiautomorphism &tau; of ClnC such that
 * $$\quad\beta(A\cdot\varphi,\psi) = \beta(\varphi,\tau(A)\cdot\psi)\qquad (1)$$

for any A in ClnC. In fact &tau; is reversion (the antiautomorphism induced by the identity on V) for m even, and conjugation (the antiautomorphism induced by minus the identity on V) for m odd. These two antiautomorphisms are related by parity involution &alpha;, which is the automorphism induced by minus the identity on V. Both satisfy &tau;(&xi;) = &minus;&xi; for &xi; in so(n,C).

When n = 2m, the situation depends more sensitively upon the parity of m. For m even, a weight &lambda; has an even number of minus signs if and only if &minus;&lambda; does; it follows that there are separate isomorphisms B±: S± → S±∗ of each half-spin representation with its dual, each determined uniquely up to scale. These may be combined into an isomorphism B: S → S∗. For m odd, &lambda; is a weight of S+ if and only if &minus;&lambda; is a weight of S&minus;; thus there is an isomorphism from S+ to S&minus;∗, again unique up to scale, and its transpose provides an isomorphism from S&minus; to S+∗. These may again be combined into an isomorphism B: S → S∗.

For both m even and m odd, the freedom in the choice of B may be restricted to an overall scale by insisting that the bilinear form &beta; corresponding to B satisfies (1), where &tau; is a fixed antiautomorphism (either reversion or conjugation).

Symmetry and the tensor square
The symmetry properties of &beta;: S ⊗ S → C can be determined using Clifford algebras or representation theory. In fact much more can be said: the tensor square S ⊗ S must decompose into a direct sum of k-forms on V for various k, because its weights are all elements in h∗ whose components belong to {&minus;1,0,1}. Now equivariant linear maps S ⊗ S → ∧kV∗ correspond bijectively to invariant maps ∧kV ⊗ S ⊗ S → C and nonzero such maps can be constructed via the inclusion of ∧kV into the Clifford algebra. Furthermore, if &beta;(&phi;,&psi;) = &epsilon; &beta;(&psi;,&phi;) and &tau; has sign &epsilon;k on ∧kV then
 * $$\beta(A\cdot\varphi,\psi) = \varepsilon\varepsilon_k \beta(A\cdot\psi,\varphi)$$

for A in ∧kV.

If n = 2m+1 is odd then it follows from Schur's Lemma that
 * $$ S\otimes S \cong \bigoplus_{j=0}^{m} \wedge^{2j} V^*$$

(both sides have dimension 22m and the representations on the right are inequivalent). Because the symmetries are governed by an involution &tau; that is either conjugation or reversion, the symmetry of the ∧2jV∗ component alternates with j. Elementary combinatorics gives
 * $$ \sum_{j=0}^m (-1)^j \dim \wedge^{2j} \Complex^{2m+1} = (-1)^{\frac12 m(m+1)} 2^m = (-1)^{\frac12 m(m+1)}(\dim \mathrm S^2S-\dim \wedge^2 S)$$

and the sign determines which representations occur in S2S and which occur in ∧2S. In particular
 * $$ \beta(\phi,\psi)=(-1)^{\frac12 m(m+1)}\beta(\psi,\phi),$$ and
 * $$ \beta(v\cdot\phi,\psi) = (-1)^m(-1)^{\frac12 m(m+1)}\beta(v\cdot\psi,\phi) = (-1)^m \beta(\phi,v\cdot\psi)$$

for v ∈ V (which is isomorphic to ∧2mV), confirming that &tau; is reversion for m even, and conjugation for m odd.

If n = 2m is even, then the analysis is more involved, but the result is a more refined decomposition: S2S±, &and;2S± and S+ ⊗ S&minus; can each be decomposed as a direct sum of k-forms (where for k = m there is a further decomposition into selfdual and antiselfdual m-forms).

The main outcome is a realisation of so(n,C) as a subalgebra of a classical Lie algebra on S, depending upon n modulo 8, according to the following table:

For n ≤ 6, these embeddings are isomorphisms (onto sl rather than gl for n = 6):
 * $$ \mathfrak{so}(2,\mathbb C) \cong \mathfrak{gl}(1,\mathbb C)\qquad(=\mathbb C)$$
 * $$ \mathfrak{so}(3,\mathbb C) \cong \mathfrak{sp}(2,\mathbb C)\qquad(=\mathfrak{sl}(2,\mathbb C))$$
 * $$ \mathfrak{so}(4,\mathbb C) \cong \mathfrak{sp}(2,\mathbb C)\oplus\mathfrak{sp}(2,\mathbb C)$$
 * $$ \mathfrak{so}(5,\mathbb C) \cong \mathfrak{sp}(4,\mathbb C)$$
 * $$ \mathfrak{so}(6,\mathbb C) \cong \mathfrak{sl}(4,\mathbb C).$$

Real representations
The complex spin representations of so(n,C) yield real representations S of so(p,q) by restricting the action to the real subalgebras. However, there are additional "reality" structures that are invariant under the action of the real Lie algebras. These come in three types.
 * 1) There is an invariant complex antilinear map r: S → S with r2 = idS. The fixed point set of r is then a real vector subspace SR of S with SR ⊗ C = S. This is called a real structure.
 * 2) There is an invariant complex antilinear map j: S → S with j2 = &minus;idS. It follows that the triple i, j and k:=ij make S into a quaternionic vector space SH. This is called a quaternionic structure.
 * 3) There is an invariant complex antilinear map b: S → S∗ that is invertible. This defines a pseudohermitian bilinear form on S and is called a hermitian structure.

The type of structure invariant under so(p,q) depends only on the signature p &minus; q modulo 8, and is given by the following table.

Here R, C and H denote real, hermitian and quaternionic structures respectively, and R + R and H + H indicate that the half-spin representations both admit real or quaternionic structures respectively.

Description and tables
To complete the description of real representation, we must describe how these structures interact with the invariant bilinear forms. Since n = p + q ≅ p − q mod 2, there are two cases: the dimension and signature are both even, and the dimension and signature are both odd.

The odd case is simpler, there is only one complex spin representation S, and hermitian structures do not occur. Apart from the trivial case n = 1, S is always even-dimensional, say dim S = 2N. The real forms of so(2N,C) are so(K,L) with K + L = 2N and so∗(N,H), while the real forms of sp(2N,C) are sp(2N,R) and sp(K,L) with K + L = N. The presence of a Clifford action of V on S forces K = L in both cases unless pq = 0, in which case KL=0, which is denoted simply so(2N) or sp(N). Hence the odd spin representations may be summarized in the following table. (†) $N$ is even for $n > 3$ and for $n = 3$, this is $sp(1)$.

The even-dimensional case is similar. For $n > 2$, the complex half-spin representations are even-dimensional. We have additionally to deal with hermitian structures and the real forms of $sl(2N, C)$, which are $sl(2N, R)$, $su(K, L)$ with $K + L = 2N$, and $sl(N, H)$. The resulting even spin representations are summarized as follows. (*) For $pq = 0$, we have instead $so(2N) + so(2N)$

(†) $N$ is even for $n > 4$ and for $pq = 0$ (which includes $n = 4$ with $N = 1$), we have instead $sp(N) + sp(N)$

The low-dimensional isomorphisms in the complex case have the following real forms. The only special isomorphisms of real Lie algebras missing from this table are $$\mathfrak{so}^*(3,\mathbb H) \cong \mathfrak{su}(3,1)$$ and $$\mathfrak{so}^*(4,\mathbb H)\cong\mathfrak{so}(6,2).$$