The Clebsch–Gordan Decomposition of $M_4(\mathbb{C})$: $\tfrac32\otimes\tfrac32=\bigoplus_{k=0}^{3}V_k$

Introduction

The subject. The algebra of order four is the tensor square of its own spin module, and the tensor square of the module of label $\tfrac32$ decomposes into four irreducible pieces of labels $3,2,1,0$. The article proves the identification, computes the four summands with their dimensions, and expresses the projectors as polynomials in the Casimir.

The order of the article. The first section identifies the algebra with $V\otimes V^{*}$ and then with $V\otimes V$. The section after it writes the action of the rotor on the algebra as the Hermitian pair $\rho(U)M\rho(U)^{\dagger}$, one operator on each side. The next applies the Clebsch–Gordan rule and counts dimensions. The third checks the result on weights. The fourth records the Casimir spectrum and the projectors. The fifth states the general rule at order $n$ and draws the boundary with the articles on the individual carriers.

It assumes The Four Weights of the Module and the Half Angle at $n=4$ for the weight basis and The Spin Group at $n=4$: the Principal Copy, $\operatorname{Sym}^3$ and the Kernel $\{\pm\mathrm I\}$ for the copy. The row at order two is The Clebsch–Gordan Decomposition of the Biquaternion Algebra.

The result in one line. $M_4(\mathbb{C})\cong V_{3/2}\otimes V_{3/2}\cong V_3\oplus V_2\oplus V_1\oplus V_0$ with dimensions $7+5+3+1=16$, each summand occurring once.

Boundaries. The article proves the decomposition and computes the Casimir; it does not describe the individual carriers in detail, which is the business of the four articles on $V_3,V_2,V_1,V_0$. It does not treat the product of two carriers, which is not a carrier and is the subject of The Truncated Rule.

The Algebra as a Tensor Square

Proposition. For every finite-dimensional $V$ there is a canonical isomorphism $\operatorname{End}(V)\cong V\otimes V^{*}$, and under it the conjugation action of $G$ on $\operatorname{End}(V)$ becomes the action on $V\otimes V^{*}$.

Proof. The map sends $u\otimes\lambda$ to the operator $x\mapsto\lambda(x)u$; it is bilinear and has the right dimension. Conjugation by $\rho(g)$ sends $\lambda(x)u$ to $(\lambda\circ\rho(g)^{-1})(x)\,\rho(g)u$, which is the tensor action.

Corollary. At $n=4$ one has $M_4(\mathbb{C})\cong V\otimes V^{*}$ with $V=V_{3/2}$.

Proof. $V$ is the defining module of the copy and $\operatorname{End}(V)=M_4(\mathbb{C})$.

Proposition. $V^{*}\cong V$ as modules of the copy, through the invariant Hermitian form, so that $M_4(\mathbb{C})\cong V\otimes V$.

Proof. The module carries a positive definite invariant form, since the copy is compact and the module is irreducible; the form identifies $V$ with its dual and is invariant, hence equivariant.

Corollary. The conjugation action on the algebra is the tensor product of the module with itself.

Proof. The two propositions composed.

The Hermitian Pair of Operators Acting on Two Sides

Definition. Let $U$ be an element of the spin group and let $\rho(U)$ be its matrix of label $\tfrac32$ on $V_{3/2}=\mathbb{C}^4$. The Hermitian pair of $\rho(U)$ is the pair $\bigl(\rho(U),\rho(U)^{\dagger}\bigr)$, read as two operators, one acting on the left of a matrix and one on the right. The pair acts on the algebra by $$ \Phi_U:\ M\longmapsto \rho(U)\,M\,\rho(U)^{\dagger},\qquad M\in M_4(\mathbb{C}). $$ The two members of the pair are Hermitian conjugates of each other, and each is unitary, so the pair is the operator and its inverse.

Proposition. $\Phi_U$ is a linear map of the sixteen-dimensional space into itself, and it is invertible, with inverse $\Phi_{U^{\dagger}}$.

Proof. Linearity in $M$ is clear. The product $\rho(U)\rho(U)^{\dagger}=I$ gives $\Phi_{U}\Phi_{U^{\dagger}}=\Phi_{U^{\dagger}}\Phi_U=\mathrm{id}$, because $\rho(U^{\dagger})=\rho(U)^{-1}=\rho(U)^{\dagger}$.

Proposition (one side is not enough). The left multiplication $M\mapsto\rho(U)M$ alone is not a map of algebras, and the same for the right; the pair together is.

Proof. The left map sends $MN$ to $\rho(U)MN$, while the product of the images is $\rho(U)M\rho(U)N$, and the two differ unless $\rho(U)=\mathrm I$. The pair sends $MN$ to $\rho(U)MN\rho(U)^{\dagger}$, which is the product of $\rho(U)M\rho(U)^{\dagger}$ and $\rho(U)N\rho(U)^{\dagger}$.

Theorem. Under the identification $\operatorname{End}(V_{3/2})\cong V_{3/2}\otimes V_{3/2}^{*}$ of the first section, the pair is the tensor product of the matrix and the contragredient matrix, $$ \Phi_U=\rho(U)\otimes\rho(U)^{*} . $$ For $\mathrm{SU}(2)$ the contragredient is equivalent to the matrix itself, because the module carries an invariant alternating form; the pair is therefore, up to that equivalence, the tensor square $\rho(U)\otimes\rho(U)$ acting on $V_{3/2}\otimes V_{3/2}$.

Proof. Write $M$ as the sum of its sixteen matrix units; the map sends a unit to the outer product of a column read through $\rho(U)$ and a row read through $\rho(U)^{\dagger}$, which is the tensor product of the two actions, one on each factor. The second factor is the complex conjugate matrix, which is the contragredient read in a fixed basis, and for $\mathrm{SU}(2)$ it is equivalent to the matrix itself.

Corollary (the four carriers are invariant). Each of the four summands $V_3,V_2,V_1,V_0$ is carried into itself by the pair.

Proof. The pair is a tensor product of two copies of the same irreducible module, so it commutes with the copy acting diagonally; a map commuting with the copy preserves the isotypic decomposition, and here each label occurs once.

Corollary (the pair is unitary on sixteen dimensions). For the Hilbert–Schmidt form $\langle A,B\rangle=\operatorname{tr}(A^{\dagger}B)$ the map $\Phi_U$ is unitary; the sixteen-by-sixteen matrix of the pair is therefore an element of $U(16)$, and of $SU(16)$ because its determinant is one.

Proof. $\langle\Phi_U A,\Phi_U B\rangle=\operatorname{tr}(\rho(U)A^{\dagger}\rho(U)^{\dagger}\rho(U)B\rho(U)^{\dagger})=\operatorname{tr}(A^{\dagger}B)$. The determinant is one because the pair is a tensor product of two matrices of determinant one.

Proposition (the pair has period $2\pi$ in the angle). If $U$ is replaced by $-U$, the pair is unchanged, while $\rho(-U)=-\rho(U)$. So the pair has the period $2\pi$ of the rotation group and the matrix $\rho$ has the period $4\pi$ of the sphere.

Proof. $\Phi_{-U}(M)=(-1)M(-1)=M$ for the factor in front and the factor behind, while $\rho(-U)=-\rho(U)$ because the exponential of a shift by $2\pi$ in the angle multiplies the matrix by $-1$.

Proposition (the infinitesimal pair). At $U=e^{-i\vec\theta\cdot\vec J}$ the derivative of the pair at $\vec\theta=0$ is $M\mapsto-i[\vec\theta\cdot\vec J,M]$, that is $-i\operatorname{ad}_{\vec\theta\cdot\vec J}$. The Lie algebra of the pair is spanned by the three $\operatorname{ad}_{J_i}$, and the Casimir of the pair is $\sum_i\operatorname{ad}_{J_i}^{2}$.

Proof. Differentiate $\rho(U)M\rho(U)^{\dagger}$ at the identity, using $\rho(U)^{\dagger}=\rho(U)^{-1}$; the two terms are $[\vec\theta\cdot\vec J,M]$ and its negative is absorbed by the $i$.

Corollary (the two routes to the four carriers coincide). The Casimir of the pair has the spectrum $0,2,6,12$ with multiplicities $1,3,5,7$, the same four values the Clebsch–Gordan rule gives, and the pair acts on the summand of label $k$ as the matrix of label $k$.

Proof. The Casimir of the tensor product is the sum of the two Casimirs plus twice the mixed term; on $V_{3/2}\otimes V_{3/2}$ the sum is $\tfrac{15}{4}+\tfrac{15}{4}+2\sum_i J_i\otimes J_i$, which on the summand of label $k$ takes the value $k(k+1)$. The second statement is the corollary on the invariance of the summands.

The reading. The pair is the reason the algebra carries four carriers and not one. A single matrix $\rho(U)$ of label $\tfrac32$ generates, through its action on the two sides, the representations of labels $3,2,1,0$ at once: the label of the carrier is the label of the symmetric or alternating piece of the tensor square, and the pair sees only those pieces. The pair is also the reason the labels are integers. The matrix changes sign when the rotor goes once around, and the pair does not, so the pair is a representation of the rotation group and the four carriers are its representations; the matrix is a representation of the sphere over it.

The Rule

Theorem. $V_{3/2}\otimes V_{3/2}\cong V_3\oplus V_2\oplus V_1\oplus V_0$, each summand occurring exactly once, with complex dimensions $7,5,3,1$.

Proof. The Clebsch–Gordan rule for $SU(2)$ states that $V_j\otimes V_k=\bigoplus_{m=|j-k|}^{j+k}V_m$ with each $m$ occurring once. With $j=k=\tfrac32$ the labels run over $3,2,1,0$. The dimensions are $2m+1$ for $m=3,2,1,0$, and their sum is $7+5+3+1=16=\dim M_4(\mathbb{C})$, which checks the count.

Corollary. The algebra splits into four irreducible invariant subspaces of complex dimensions $7,5,3,1$ and real dimensions $14,10,6,2$.

Proof. The theorem, and the real dimension of a complex space of dimension $d$ is $2d$.

Corollary. The traceless part is $V_3\oplus V_2\oplus V_1$, of dimension $15$, and the centre is $V_0$.

Proof. The trace is invariant and vanishes on the three summands of dimension $7,5,3$; the scalars are the one-dimensional invariant subspace on which the trace does not vanish.

Corollary. The four labels are $0,1,2,3$: three integers below the label of the module and one equal to nothing in the module.

Proof. The labels displayed in the theorem; the module has label $\tfrac32$, which is not among them.

The Check on Weights

Proposition. The weights of $V\otimes V$ are the sixteen sums $m+m'$ of two weights of $V$, and their multiset is

$$ 3,\;2,\;1,\;0,\;0,\;-1,\;-2,\;-3 $$

together with $2,\;1,\;0,\;-1,\;-2$ and $1,\;0,\;-1$ and $0$.

Proof. The weights of $V$ are $\tfrac32,\tfrac12,-\tfrac12,-\tfrac32$; the sums of two of them are $3$ (once), $2$ (twice), $1$ (three times), $0$ (four times), and symmetrically. Reading the multiset $3,2,2,1,1,1,0,0,0,0,-1,-1,-1,-2,-2,-3$ as the strings $3,2,1,0,-1,-2,-3$; $2,1,0,-1,-2$; $1,0,-1$; $0$ gives the four carriers.

Corollary. The multiset of weights of the algebra is $0^{4},\pm1^{3},\pm2^{2},\pm3$, and the four strings read from the top are exactly the four labels $3,2,1,0$.

Proof. The proposition, counted by multiplicity.

Corollary. The decomposition is visible in the weights alone: the highest weight $3$ occurs once, so exactly one summand of label $3$ is present; after removing its string the next highest weight $2$ occurs twice, one of which belongs to the same summand, and so on.

Proof. The strings of $V_3,V_2,V_1,V_0$ contain $3$ once, $2$ once, $1$ once, $0$ once at their tops, and the remaining copies are accounted for by the lower members.

The Casimir and the Projectors

Definition. The Casimir element of the copy acting on the algebra is $C=\sum_{i}\operatorname{ad}_{J_i}^{2}$, where $J_1,J_2,J_3$ are the three Hermitian generators. The sign is the one that makes the Casimir positive: the operator $\operatorname{ad}_{J}$ is self-adjoint for the Hilbert–Schmidt form, so $\operatorname{ad}_{J}^{2}$ is positive semidefinite and the displayed sum is the positive operator whose eigenvalues are the values $k(k+1)$.

Proposition. $C$ acts on the summand of label $k$ by the scalar $k(k+1)$, so its spectrum on the algebra is $0,2,6,12$ with multiplicities $1,3,5,7$, computed to a residual of order $10^{-15}$.

Proof. The Casimir of $\mathfrak{su}_2$ acts on the irreducible module of highest weight $k$ by $k(k+1)$; the multiplicities are the dimensions $2k+1$. The numerical diagonalisation of the sixteen-by-sixteen matrix $C$ gives the four eigenvalues with the stated multiplicities.

Corollary. The four summands are the eigenspaces of $C$, and each is the range of the Lagrange projector

$$ \Pi_k=\prod_{l\neq k}\frac{C-l(l+1)}{k(k+1)-l(l+1)},\qquad k=0,1,2,3, $$

which is idempotent, Hermitian, and of trace $2k+1$.

Proof. A polynomial in $C$ acts on the eigenspace of eigenvalue $k(k+1)$ by the value at that point; the Lagrange polynomial with value one there and zero at the other three eigenvalues gives the projector. The trace is the dimension of the range.

Corollary. The four projectors sum to the identity and are mutually orthogonal, and their traces sum to $1+3+5+7=16$.

Proof. The Lagrange polynomials sum to one on the spectrum, and the ranges are the eigenspaces.

The numerical check. Applying $\Pi_k$ to a random element of the algebra and then applying the Casimir reproduces the element multiplied by $k(k+1)$ to a residual of order $10^{-15}$, and the composition $\Pi_k\Pi_l$ vanishes for $k\neq l$ to the same order.

Proof. The two matrices are polynomials in one matrix, so they commute and act diagonally in the same basis.

The General Rule

Proposition. At order $n$ the module is $V_{(n-1)/2}$ and the algebra is $\bigoplus_{k=0}^{n-1}V_k$, with $\sum_{k=0}^{n-1}(2k+1)=n^{2}$.

Proof. The tensor square of $V_j$ with $2j=n-1$ decomposes into the labels $0,1,\dots,2j=n-1$, and the dimensions sum to $n^{2}$, which is the dimension of the algebra.

Corollary. At $n=2$ the decomposition is $V_1\oplus V_0$ of dimensions $3+1=4$; at $n=3$ it is $V_2\oplus V_1\oplus V_0$ of dimensions $5+3+1=9$; at $n=4$ it is the four summands of the theorem.

Proof. The proposition at the three orders.

Corollary. The number of summands is $n$ and the labels are $0,\dots,n-1$; the module's label is $\tfrac{n-1}{2}$, which is a summand label exactly when $n$ is odd.

Proof. The labels displayed; $\tfrac{n-1}{2}$ is an integer exactly for odd $n$.

The Boundary

Against the carriers. The four summands are the four carriers, and each has its own article: The Spin-Three Carrier, Introduction to the Spin-Two Carrier, The Spin-One Carrier and The Spin-Zero Carrier. The present article proves that the list is complete and that each occurs once.

Against the product of carriers. The product of two summands is not a summand; it is a subspace whose projection on the summands is computed in The Truncated Rule: the Product of Two Carriers Is Not a Carrier. The decomposition is a decomposition of the space and not of the algebra as an algebra.

Against the module. The module has label $\tfrac32$ and appears in the algebra only as a minimal left ideal, not as a summand, as Where the Half-Integers Go in $M_4(\mathbb{C})$ explains.

Summary

The algebra is $\operatorname{End}(V)\cong V\otimes V^{*}\cong V\otimes V$ with $V=V_{3/2}$, and the Clebsch–Gordan rule for the tensor square of the label $\tfrac32$ gives $V_3\oplus V_2\oplus V_1\oplus V_0$ with dimensions $7+5+3+1=16$, each summand once. The traceless part is the sum of the first three, of dimension fifteen, and the centre is the last.

The weights of the algebra are the sixteen sums of two weights of the module, with multiset $0^{4},\pm1^{3},\pm2^{2},\pm3$, and the four strings read from the top are the four labels. The Casimir acts by $k(k+1)$, with spectrum $0,2,6,12$ and multiplicities $1,3,5,7$; the projectors are the Lagrange polynomials in the Casimir, and they are idempotent, Hermitian, orthogonal and of traces $1,3,5,7$.

At order $n$ the decomposition is $\bigoplus_{k=0}^{n-1}V_k$ with $\sum(2k+1)=n^{2}$, and the module's label is a summand label exactly when $n$ is odd.

Summary of Notation

Symbol Meaning
$V=V_{3/2}$ the module, complex dimension four
$V_k$ the carrier of label $k$, dimension $2k+1$
$C=\sum_i\operatorname{ad}_{J_i}^{2}$ the Casimir of the conjugation action
$\rho(U)$ the matrix of the rotor of label $\tfrac32$, of order four
$\rho(U)M\rho(U)^{\dagger}$ the Hermitian pair of the rotor acting on the two sides
$\Pi_k$ the projector onto the carrier of label $k$
$0^{4},\pm1^{3},\pm2^{2},\pm3$ the weight multiset of the algebra
$n^{2}$ the sum of the dimensions

Further Reading