The Clebsch–Gordan Decomposition of the Biquaternion Algebra in $M_2(\mathbb{C})$
Introduction
The question. The algebra carries four complex dimensions. Under conjugation by unitary matrices, how does it split, and why is the split $1\oplus3$ and nothing else?
Why it is the same statement. This is the in-depth matrix reading of The Clebsch–Gordan Decomposition of the Biquaternion Algebra. The source's projectors $P_0=\tfrac12(\mathrm{id}+\natural)$ and $P_1=\tfrac12(\mathrm{id}-\natural)$ become, under $A=\mathsf M_2(\tilde Q)$, the projectors onto the scalar line and onto the traceless part, $$ P_0(Y)=\frac{\operatorname{tr}Y}{2}\,\mathrm I_2,\qquad P_1(Y)=Y-\frac{\operatorname{tr}Y}{2}\,\mathrm I_2, $$ and the two-sided action becomes $Y\mapsto UYU^{\dagger}$. The statements are the same statements, and only the writing is new.
What the article proves. In this order. The involution $\natural$ as the map $Y\mapsto(\operatorname{tr}Y)\mathrm I_2-Y$, and the two projectors. Their images: $\mathbb{C}\mathrm I_2$ and $\mathfrak{sl}_2$. The equivariance under unitary conjugation. The weights $\{-1,0,+1,0\}$ of the diagonal generator. The Casimir operator $\mathbf H^2_{\mathrm{tot}}=2P_1=\mathrm{id}-\natural$ as the map $Y\mapsto2Y-(\operatorname{tr}Y)\mathrm I_2$. The single-slot value $\tfrac34$. The reason no half-integer summand occurs, and why the slot is a factor and not a summand.
What the article does not do. It does not prove the addition rule; it does not resolve the summand into weight vectors, which is the axis-dependent step; and it does not treat the minimal left ideals that realise the slot, which are Introduction to the Spin-Half Bispinors Spaces in $M_2(\mathbb{C})$.
Conventions. $\mathsf M_2(\tilde Q)=Q_t\mathrm I_2-i(Q_x\sigma_1+Q_y\sigma_2+Q_z\sigma_3)$, so $\operatorname{tr}A=2Q_t$. The coefficient basis is $(\mathrm I_2,-i\sigma_1,-i\sigma_2,-i\sigma_3)$; the coefficient-frame array of an operator is written in that basis. The involution $\natural$ acts on matrices by $A\mapsto\overline{A}$ and on operators by $T\mapsto\natural\circ T\circ\natural$. The inner product is $\langle X,Y\rangle=\tfrac12\operatorname{tr}(X^{\dagger}Y)$.
Boundaries. The dictionary is The Pauli Representation : Operators. The two sides separately are Operators Acting on One Side in $M_2(\mathbb{C})$ and Operators Acting on Two Sides in $M_2(\mathbb{C})$. The carriers built on the split are From the Spin to the Carriers and the Closures in $M_2(\mathbb{C})$. The $n=2$ spine is The One-Sided Operators at $n=2$ and The Two-Sided Operators and the Segre Cone at $n=2$.
Letters. $Y$ is a general matrix, $\natural$ the matrix conjugation, $P_0,P_1$ the two projectors, $H_k$ the Hermitian generators, $\mathbf H^2_{\mathrm{tot}}$ the Casimir operator.
The Involution in Matrices
The map. On the coefficient basis $\natural$ fixes $E_t=\mathrm I_2$ and negates $E_x,E_y,E_z=-i\sigma_k$. So its array is $$ [\natural]=\operatorname{diag}(1,-1,-1,-1), $$ and on a general matrix $$ A^{\natural}=Q_t\mathrm I_2+(\text{vector part reversed})=\overline{A}. $$ Equivalently, since $\overline{A}=(\operatorname{tr}A)\mathrm I_2-A$ for a $2\times2$ matrix, $$ \natural(Y)=\left(\operatorname{tr}Y\right)\mathrm I_2-Y . $$ Verified on random matrices with residual $6\cdot10^{-17}$.
Linearity. $\natural$ is complex-linear, $\natural(iY)=i\natural(Y)$, and involutive, $\natural^2=\mathrm{id}$. The Hermitian adjoint is the other composite, $Y^{\dagger}=\natural(Y)^{\mathsf T}$, and it is conjugate-linear. A conjugate-linear involution has no $\pm1$ eigenspaces inside a complex space, so the split must be built from $\natural$; this is the source's reason, read on the array.
The trace form. The map $Y\mapsto\operatorname{tr}(Y)\mathrm I_2$ is the projection onto the scalar line, and its kernel is the traceless part. Composing, one gets the three-term identity $$ \natural = 2P_0-\mathrm{id},\qquad P_0(Y)=\frac{\operatorname{tr}Y}{2}\mathrm I_2 . $$
The Two Projectors
The definition. $P_0=\tfrac12(\mathrm{id}+\natural)$ and $P_1=\tfrac12(\mathrm{id}-\natural)$. On matrices, $$ P_0(Y)=\frac{\operatorname{tr}Y}{2}\mathrm I_2,\qquad P_1(Y)=Y-\frac{\operatorname{tr}Y}{2}\mathrm I_2 . $$
The algebra. $P_0^2=P_0$, $P_1^2=P_1$, $P_0P_1=P_1P_0=0$, $P_0+P_1=\mathrm{id}$; verified exactly on random matrices.
The images. $P_0M_2(\mathbb{C})=\mathbb{C}\mathrm I_2$, of complex dimension one; $P_1M_2(\mathbb{C})=\mathfrak{sl}_2(\mathbb{C})$, the traceless matrices, of complex dimension three. As arrays, $$ [P_0]=\operatorname{diag}(1,0,0,0),\qquad [P_1]=\operatorname{diag}(0,1,1,1), $$ and $1+3=4$.
The equivariance. For unitary $U$, $$ P_0(UYU^{\dagger})=U P_0(Y)U^{\dagger},\qquad P_1(UYU^{\dagger})=U P_1(Y)U^{\dagger}, $$ because $\operatorname{tr}(UYU^{\dagger})=\operatorname{tr}Y$. So the split is invariant under the two-sided action, and it is the decomposition of the algebra as a module for that action.
The one-sided failure. The split does not commute with $Y\mapsto AY$: left multiplication moves the scalar line into the traceless part, $\operatorname{tr}(AY)=2Q_t\operatorname{tr}Y/2+\dots$. The split belongs to the diagonal action and not to either factor.
The Casimir Is the Projector
The adjoint arrays. $H_k=\tfrac i2(L_k-R_k)$ is the Hermitian generator of the diagonal action, and $L_k-R_k$ is the coefficient-frame array of $Y\mapsto[E_k,Y]$. Each $L_k-R_k$ is real antisymmetric, verified exactly.
The identity. In arrays, $$ \sum_k(L_k-R_k)^2=-8\,[P_1],\qquad \mathbf H^2_{\mathrm{tot}}=\sum_kH_k^2=2\,[P_1]=\mathrm{id}-\natural, $$ verified exactly. On matrices the Casimir is the operator $$ Y\mapsto 2Y-\left(\operatorname{tr}Y\right)\mathrm I_2, $$ with eigenvalues $2$ on $\mathfrak{sl}_2$ and $0$ on $\mathbb{C}\mathrm I_2$.
The weights. The eigenvalues of $H_{\mathrm{tot},z}=\tfrac i2[E_z,\cdot]$ are $+1,0,0,-1$; verified. The weight set is $\{-1,0,+1,0\}$, the union of $\{-1,0,+1\}$ on the traceless part and $\{0\}$ on the scalar line.
The two values. $2=1\cdot(1+1)$ and $0=0\cdot(0+1)$, so the traceless part is spin one and the scalar line is spin zero. The label is not bookkeeping: the Casimir is the projector, and the split is the difference of two eigenvalues of one operator.
The cross term. Writing the total Casimir as $\tfrac32\mathrm{id}-2\sum_kH_kH'_k$ and comparing with $2P_1$ gives $$ \sum_kH_kH'_k=\tfrac34\mathrm{id}-P_1, $$ with values $\tfrac34$ on the traceless part and $-\tfrac14$ on the scalar line. The difference of the two values is the split, and the average is the single-slot value $\tfrac34$.
The Single Slot
The value. For one side alone, $\sum_kH_k^2=\tfrac34\mathrm I_4$ (the previous article in this category), and $\tfrac34=\tfrac12\cdot\tfrac32=s(s+1)$ at $s=\tfrac12$.
The two numbers. $\tfrac34$ on a factor, $2$ on the summand of top label; the difference is the difference between a factor and a summand, and the algebra has both.
What the Split Contains and What It Does Not
The integer labels. The summands are $0$ and $1$, both integers. No half-integer occurs, and the full turn is the identity: $\exp(2\pi i\,H_{\mathrm{tot},z})$ acts as $e^{\pm2\pi i}=1$ on the weight $\pm1$ directions and as $1$ on the weight $0$ directions. Verified numerically to $7.5\cdot10^{-15}$.
The half turn. $\exp(\pi i\,H_{\mathrm{tot},z})$ acts as $-1$ on the two weight-$\pm1$ directions and $+1$ on the two weight-$0$ directions; verified, with array $\operatorname{diag}(1,-1,-1,1)$. It is not the identity. This is the difference between the rotation group and its cover, read on the algebra.
The slot is not a summand. The spin-$\tfrac12$ carrier is $\mathbb{C}^2$, the defining carrier, and it is not a subspace of $M_2(\mathbb{C})$: it is one of the two factors of $M_2(\mathbb{C})=\mathbb{C}^2\otimes(\mathbb{C}^2)^{*}$. To see it inside the algebra one chooses a rank-one projection $\Pi=uu^{\dagger}$, and then $$ M_2(\mathbb{C})\,\Pi=\{Y\Pi\} $$ is a minimal left ideal, of complex dimension two, carrying the defining action of $SU(2)$ by left multiplication. The choice of $u$ is a choice of axis, and different choices give different ideals.
The dimension bookkeeping. $1+3=4$ complex dimensions for the two summands; $2+2=4$ for the factor and its dual; and the algebra is four complex dimensional. The half-integer carrier contributes to the second count only, which is why the list of three carriers is not a direct sum.
The refinement. The Clebsch–Gordan coefficients appear when the traceless part is resolved into weight vectors, and that resolution needs an axis: with $\sigma_3$ chosen, the weight vectors are $\sigma_1\pm i\sigma_2$ and $\sigma_3$. The series itself is coefficient-free.
Worked Example
The three Pauli directions. $\sigma_1,\sigma_2,\sigma_3$ are traceless, so $P_1\sigma_k=\sigma_k$ and $P_0\sigma_k=0$; $\mathrm I_2$ is scalar, so $P_0\mathrm I_2=\mathrm I_2$ and $P_1\mathrm I_2=0$. The array of $P_1$ is $\operatorname{diag}(0,1,1,1)$ in the coefficient basis.
A general matrix. $Y=\begin{pmatrix}1&2\\3&4\end{pmatrix}$ has $\operatorname{tr}Y=5$, so $P_0Y=\tfrac52\mathrm I_2$ and $P_1Y=\begin{pmatrix}-1.5&2\\3&1.5\end{pmatrix}$, with trace zero.
The Casimir on it. $2Y-(\operatorname{tr}Y)\mathrm I_2=\begin{pmatrix}-3&4\\6&3\end{pmatrix}$, which is $2P_1Y$; the scalar part is killed.
Where $n=2$ Differs
The split. At $n=2$ the split is $1\oplus3$: the scalar line plus the traceless part. At order $n$ the same two projectors exist, and the split is $1\oplus(n^2-1)$; the traceless part is $n^2-1$ complex dimensional, and it is irreducible only at $n=2$.
The Casimir. At $n=2$ the traceless part is irreducible with Casimir $2$. At order $n$ the adjoint carrier is irreducible with Casimir $n$, and the value $2$ is the value at $n=2$.
The labels. At $n=2$ the labels are $0$ and $1$ and the chain stops. At order $n$ the labels run from $0$ to $n-1$, and the top label is $n-1$.
What does not differ. The two projectors, their algebra, the equivariance and the identity "Casimir equals twice the projector" hold at every order.
Summary
The involution $\natural$ is the matrix conjugation, and on a $2\times2$ matrix it is $Y\mapsto(\operatorname{tr}Y)\mathrm I_2-Y$; it is complex-linear and involutive. Its averaging operators are $P_0(Y)=\tfrac12\operatorname{tr}(Y)\mathrm I_2$ and $P_1(Y)=Y-\tfrac12\operatorname{tr}(Y)\mathrm I_2$, with images the scalar line and $\mathfrak{sl}_2$, of complex dimensions one and three.
Both are equivariant for $Y\mapsto UYU^{\dagger}$, and neither commutes with one-sided multiplication, so the split belongs to the diagonal action. The diagonal weights are $\{-1,0,+1,0\}$ and the Casimir operator is $\mathbf H^2_{\mathrm{tot}}=2P_1=\mathrm{id}-\natural$, that is $Y\mapsto2Y-(\operatorname{tr}Y)\mathrm I_2$, with values $2$ and $0$: spin one and spin zero. The single-slot Casimir is $\tfrac34$, and the cross term is $\tfrac34\mathrm{id}-P_1$.
No half-integer summand occurs, because the split has only integer labels and the full turn is the identity. The half-integer carrier is the defining factor $\mathbb{C}^2$, realised inside the algebra only as a minimal left ideal $M_2\Pi$ after a rank-one projection is chosen.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\natural$ | the matrix conjugation $Y\mapsto(\operatorname{tr}Y)\mathrm I_2-Y$ |
| $P_0,P_1$ | the projectors onto scalars and traceless matrices |
| $H_k$ | $\tfrac i2(L_k-R_k)$, the diagonal generators |
| $\mathbf H^2_{\mathrm{tot}}$ | the Casimir, $2P_1$ |
| $\Pi$ | a rank-one projection, for the minimal left ideal |
Further Reading
- The Clebsch–Gordan Decomposition of the Biquaternion Algebra, for the source's canonical construction and its remarks on the coefficients.
- From the Spin to the Carriers and the Closures in $M_2(\mathbb{C})$, for the three carriers built on this split.
- From the Spin Group to the Spin of a Single Rotor in $M_2(\mathbb{C})$, for the single-slot Casimir $\tfrac34$.
- Operators Acting on Two Sides in $M_2(\mathbb{C})$, for the diagonal action whose Casimir this is.
- Introduction to the Spin-Half Bispinors Spaces in $M_2(\mathbb{C})$, for the minimal left ideals that realise the factor.