Introduction to the Operators on the Four Sectors in $M_2(\mathbb{C})$

Introduction

The question. Three operators carry the polar word on the algebra: the left multiplication $L_A$, the right multiplication $R_A$, and the two-sided operator $\mathrm{Sd}_A=L_AR_{A^{\dagger}}$. This article states what the three have in common — their arrays, their multiplicativity, the doubling rule and the table of the four sectors — and proves nothing about a single sector.

Why it is the same statement. This is the in-depth matrix reading of Introduction to the Operators on the Four Sectors. The source works frame-free on $\mathbb{B}$; the display $A=\mathsf M_2(\tilde Q)$ turns every operator into an explicit $4\times4$ array and every sector condition into a condition on $\operatorname{tr}A$ and $\det A$. The statements are the same statements, and only the writing is new.

What the article proves. In this order. The three operators as arrays: $L_A=\mathrm I_2\otimes A$ and $R_A=A^{\mathsf T}\otimes\mathrm I_2$ in the matrix-unit frame, and the coefficient-frame arrays $\sum_\mu Q_\mu M_\mu$ built from three signed-permutation multiplication tables. The three multiplicativity laws with their orders. The two projectors of the trace. The doubling rule: the two-sided operator squares every modulus and kills the central phase. The two sector indices $\det A$ and $B^2=\det A-\tfrac14(\operatorname{tr}A)^2$. The four-sector tables for $L_A$ and for $\mathrm{Sd}_A$.

What the article does not do. It does not develop any sector; that is One-Sided Operators on the N-Regular Sector in $M_2(\mathbb{C})$ and its three companions. It does not treat the two-sided operator as a family, which is The Hermitian Pair of Operators Acting on Two Sides in $M_2(\mathbb{C})$.

Conventions. The display is $\mathsf M_2(\tilde Q)=Q_t\mathrm I_2-i(Q_x\sigma_1+Q_y\sigma_2+Q_z\sigma_3)$, so that $$ A=\begin{pmatrix}Q_t-iQ_z & -iQ_x-Q_y\\ -iQ_x+Q_y & Q_t+iQ_z\end{pmatrix},\qquad \operatorname{tr}A=2Q_t,\qquad \det A=Q_t^2+Q_x^2+Q_y^2+Q_z^2=N(\tilde Q). $$ The vector form is $B^2=Q_x^2+Q_y^2+Q_z^2=\det A-\tfrac14(\operatorname{tr}A)^2$, and the discriminant is $\Delta=(\operatorname{tr}A)^2-4\det A=-4B^2$. The matrix adjoint is $A^{\dagger}=\overline{A^{\mathsf T}}$; it is the composite of the natural involution $\natural$ with the conjugation of the coefficients, verified below. Arrays are printed in the coefficient frame $(E_t,E_x,E_y,E_z)=(\mathrm I_2,-i\sigma_1,-i\sigma_2,-i\sigma_3)$ unless the matrix-unit frame is named.

Boundaries. The frame-free operator theory is Operators Acting on One Side in $M_2(\mathbb{C})$ and Operators Acting on Two Sides in $M_2(\mathbb{C})$. The polar decomposition of a matrix is The Adjoints and the Polar Decomposition of an Operator at $n=2$. The array privilege is The Axis-Free $4\times4$ Representation : Operators. The $n=2$ spine is The One-Sided Operators at $n=2$ and The Two-Sided Operators and the Segre Cone at $n=2$, and the display is $M_2(\mathbb{C})$ Pauli Representation : Operators.

Letters. $A=\mathsf M_2(\tilde Q)$ is the matrix of the element. $M_\mu$ are the four multiplication tables. $\mathrm{Sd}_A$ is the two-sided operator $Y\mapsto AYA^{\dagger}$. $\Sigma_{\varepsilon\eta}$ are the four sectors. $\lambda_\pm$ are the eigenvalues of $A$.

The Three Operators as Arrays

The definitions. For $Y\in M_2(\mathbb{C})$, $$ L_A(Y)=AY,\qquad R_A(Y)=YA,\qquad \mathrm{Sd}_A(Y)=AYA^{\dagger}. $$ All three are $\mathbb{C}$-linear endomorphisms of a space of complex dimension four, hence $4\times4$ arrays.

The matrix-unit frame. With the basis ordered $(E_{11},E_{12},E_{21},E_{22})$ and $Y$ read as a column, the two one-sided operators are the Kronecker products $$ L_A=\mathrm I_2\otimes A,\qquad R_A=A^{\mathsf T}\otimes\mathrm I_2 . $$ For the displayed $A$ this gives $$ L_A=\begin{pmatrix} Q_t-iQ_z & -iQ_x-Q_y & 0 & 0\\ -iQ_x+Q_y & Q_t+iQ_z & 0 & 0\\ 0 & 0 & Q_t-iQ_z & -iQ_x-Q_y\\ 0 & 0 & -iQ_x+Q_y & Q_t+iQ_z \end{pmatrix}, $$ verified against the action $Y\mapsto AY$ to $4\cdot10^{-16}$, and $R_A$ is the same array with the two diagonal $2\times2$ blocks exchanged. The two commute, since they act on different tensor slots.

The coefficient frame. In the basis $(E_t,E_x,E_y,E_z)$ the same operator is $$ L_A=Q_tM_t+Q_xM_x+Q_yM_y+Q_zM_z,\qquad M_t=\mathrm I_4, $$ with rows and columns indexed $t,x,y,z$, $$ M_x=\begin{pmatrix}0&-1&0&0\\1&0&0&0\\0&0&0&-1\\0&0&1&0\end{pmatrix},\quad M_y=\begin{pmatrix}0&0&-1&0\\0&0&0&1\\1&0&0&0\\0&-1&0&0\end{pmatrix},\quad M_z=\begin{pmatrix}0&0&0&-1\\0&0&-1&0\\0&1&0&0\\1&0&0&0\end{pmatrix}. $$ Every row has one nonzero entry of modulus one, so the three are signed permutation matrices: they are the multiplication table read aloud. The identities $M_x^2=M_y^2=M_z^2=-\mathrm I_4$ and $M_xM_y=M_z$ hold exactly, with residual $0$, and reproduce $\vec e_k^2=-e_t$ and $\vec e_x\vec e_y=\vec e_z$.

Why the coefficient array is not the Kronecker array. The two frames differ by the change of basis from the matrix units to the $E_\mu$, and that change is not a permutation: $E_x=-i\sigma_1$ is not a matrix unit. The Kronecker form is the one that shows the tensor slots; the coefficient form is the one in which $\sum_\mu Q_\mu M_\mu$ is linear in the coefficients with integer structure constants.

Rank and invertibility. $L_A$ is invertible exactly when $\det A\neq0$. When $A\neq0$ has $\det A=0$ the rank is $2$; when $A=0$ it is $0$. The trichotomy $4,2,0$ holds for $R_A$ as well, and $\mathrm{Sd}_A$ is invertible exactly when $A$ is.

Multiplicativity

The three laws. For all $A,A'\in M_2(\mathbb{C})$, $$ L_{AA'}=L_A L_{A'},\qquad R_{AA'}=R_{A'}R_A,\qquad \mathrm{Sd}_{AA'}=\mathrm{Sd}_A\,\mathrm{Sd}_{A'}. $$ Each is associativity read at one position, and the third uses $(AA')^{\dagger}=A'^{\dagger}A^{\dagger}$. Verified on a hundred random pairs, with residuals $1.5\cdot10^{-14}$, $1.6\cdot10^{-14}$ and $2\cdot10^{-13}$.

The orders. The left multiplication is a representation, the right multiplication an anti-representation, and the two-sided operator a representation again because the two reversals cancel. In arrays, $L$ acts on the first Kronecker slot and $R$ on the second.

The image on the invertibles. $A\mapsto L_A$ is faithful with image $GL(2,\mathbb{C})$, of real dimension eight. $A\mapsto\mathrm{Sd}_A$ kills the central circle: $\mathrm{Sd}_{\omega A}=\mathrm{Sd}_A$ for every $\omega\in\mathbb{C}$ with $|\omega|=1$, because $\omega$ is central and $\omega\omega^{\dagger}=1$. Its image is the conformal group $CO^{+}(1,3)$, of real dimension seven.

The involution, verified. The natural involution $\natural$ of the source is the $\mathbb{C}$-linear quaternion conjugation $q\mapsto(Q_t,-Q_x,-Q_y,-Q_z)$, which on the display is $A\mapsto(\operatorname{tr}A)\mathrm I_2-A$. Composed with the conjugation of the coefficients it is the matrix adjoint. On the coefficient vector $q=(Q_t,Q_x,Q_y,Q_z)$ the composite acts by $$ q\mapsto\left(\bar Q_t,-\bar Q_x,-\bar Q_y,-\bar Q_z\right), $$ and the display of the resulting vector is the matrix adjoint: $$ A^{\dagger}=\overline{A^{\mathsf T}}. $$ Verified on a hundred random elements with residual exactly $0$.

The trap in the array form. Entrywise conjugation of the array is not the conjugation of the coefficients, because $\sigma_2$ is imaginary: entrywise conjugation agrees with $A^{\dagger}$ only when $A$ is symmetric, and a random $A$ fails by $O(1)$. The safest reading is on the coefficient vector, where the composite is a signed conjugation.

The two projectors. The $\mathbb{C}$-linear quaternion conjugation, without the conjugation of the coefficients, is $$ A\longmapsto \operatorname{tr}(A)\mathrm I_2-A, $$ verified to $6\cdot10^{-17}$; equivalently $q\mapsto(Q_t,-Q_x,-Q_y,-Q_z)$. Its two projectors are $$ P_0(A)=\frac{\operatorname{tr}A}{2}\mathrm I_2,\qquad P_1(A)=A-\frac{\operatorname{tr}A}{2}\mathrm I_2, $$ with $P_0^2=P_0$, $P_1^2=P_1$, $P_0P_1=0$, $P_0+P_1=\mathrm{id}$. Their images are the centre $\mathbb{C}\mathrm I_2$ and the traceless part $\mathfrak{sl}_2$, of complex dimensions $1$ and $3$. They are canonical, built from the trace alone.

The Doubling Rule

The polar word. Write $A=A_HA_0$ with $A_H$ positive semidefinite Hermitian and $A_0$ a partial isometry. Multiplicativity gives $$ \mathrm{Sd}_A=\mathrm{Sd}_{A_H}\circ\mathrm{Sd}_{A_0}. $$ The modulus of the operator is $\lvert\mathrm{Sd}_A\rvert=\mathrm{Sd}_{A_H}$, and the singular values of $\mathrm{Sd}_A$ are the squares of those of $A$.

The table.

element two-sided operator
scale $r$ $r^2$
singular value $2a$ $4a^2$
phase $e^{i\theta}$ $1$

Why the phase cancels. $A$ and $\omega A$ with $|\omega|=1$ give the same $\mathrm{Sd}$, since $\mathrm{Sd}_{\omega A}(Y)=\omega AY\bar\omega A^{\dagger}=AYA^{\dagger}$.

The collapse on the singular locus. When $A$ has rank one, $A=2aA_0$ with $A_0$ a rank-one partial isometry, and $$ \mathrm{Sd}_A=4a^2\,\mathrm{Sd}_{A_0}. $$ Two-sided operator, scalar, times a partial isometry of rank one.

The Four Sectors

The two indices. The sectors are indexed by $$ N=\det A,\qquad B^2=\det A-\tfrac14(\operatorname{tr}A)^2, $$ so that $$ \Sigma_{11}=\{\det A\neq0,\ B^2\neq0\},\quad \Sigma_{10}=\{\det A\neq0,\ B^2=0\},\quad \Sigma_{01}=\{\det A=0,\ A\neq0,\ B^2\neq0\},\quad \Sigma_{00}=\{\det A=0,\ A\neq0,\ B^2=0\}. $$

The matrix forms. On $\Sigma_{11}$ the matrix has two distinct eigenvalues $\lambda_\pm=Q_t\pm iB$ and is diagonalizable. On $\Sigma_{10}$ it has the single eigenvalue $Q_t$; it is the scalar $Q_t\mathrm I_2$ exactly on the scalar line, and otherwise similar to $\begin{pmatrix}Q_t&1\\0&Q_t\end{pmatrix}$. On $\Sigma_{01}$ it is a rank-one matrix with eigenvalues $2Q_t$ and $0$. On $\Sigma_{00}$ it is a non-zero nilpotent, $A^2=0$, similar to $\begin{pmatrix}0&1\\0&0\end{pmatrix}$.

The eigenvalues of the operators. Because $A$ acts on the left and on the right on two tensor slots, $$ \operatorname{spec}L_A=\operatorname{spec}R_A=\{\lambda_+,\lambda_+,\lambda_-,\lambda_-\},\qquad \operatorname{spec}\mathrm{Sd}_A=\{\lambda_i\bar\lambda_j\}. $$ The four products were computed on random elements of $\Sigma_{11}$ and agree with the spectrum of the array to $10^{-15}$.

The one-sided table.

sector rank $L_A$ spectrum Jordan type image
$\Sigma_{11}$ $4$ $\lambda_+,\lambda_-$, each twice $\operatorname{diag}(\lambda_+,\lambda_+,\lambda_-,\lambda_-)$ $M_2(\mathbb{C})$
$\Sigma_{10}$, scalar line $4$ $Q_t$ four times, geometric $4$ $Q_t\mathrm I_4$ $M_2(\mathbb{C})$
$\Sigma_{10}$, Jordan part $4$ $Q_t$ four times, geometric $2$ $J_2(Q_t)\oplus J_2(Q_t)$ $M_2(\mathbb{C})$
$\Sigma_{01}$ $2$ $c,0$ with $c=2Q_t$, multiplicities $2,2$ $\operatorname{diag}(c,c,0,0)$ the rank-one ideal $AM_2(\mathbb{C})$
$\Sigma_{00}$ $2$ $0$ four times, geometric $2$ $J_2(0)\oplus J_2(0)$ $\mathbb{C}A\oplus\mathbb{C}P_+=\ker L_A$

The two image columns. On $\Sigma_{01}$ the images of $L_A$ and $R_A$ are the distinct rank-one ideals $AM_2(\mathbb{C})$ and $M_2(\mathbb{C})A$, meeting in $\mathbb{C}A$. On $\Sigma_{00}$ they are $\mathbb{C}A\oplus\mathbb{C}P_+$ and $\mathbb{C}A\oplus\mathbb{C}P_-$, where $P_\pm$ are the canonical rank-one projections with $A=P_+AP_-$. On $\Sigma_{11}$ and $\Sigma_{10}$ both images are the whole algebra.

The two-sided table.

sector rank spectrum Jordan type image
$\Sigma_{11}$ $4$ $\lambda_i\bar\lambda_j$ semisimple $M_2(\mathbb{C})$
$\Sigma_{10}$, scalar line $4$ $\lvert Q_t\rvert^2$ four times, geometric $4$ $\lvert Q_t\rvert^2\mathrm I_4$ $M_2(\mathbb{C})$
$\Sigma_{10}$, Jordan part $4$ $\lvert Q_t\rvert^2$ four times, geometric $2$ $J_3(\lvert Q_t\rvert^2)\oplus J_1(\lvert Q_t\rvert^2)$ $M_2(\mathbb{C})$
$\Sigma_{01}$ $1$ $\lvert c\rvert^2,0$, multiplicities $1,3$ $\operatorname{diag}(\lvert c\rvert^2,0,0,0)$ the line $\mathbb{C}P$, $P=AA^{\dagger}/\operatorname{tr}(AA^{\dagger})$
$\Sigma_{00}$ $1$ $0$ four times, geometric $3$ $J_2(0)\oplus0\oplus0$ the line $\mathbb{C}P_+$

Reading the two tables. Along a row the rank drops when $\det A=0$, and the Jordan type leaves the semisimple case when $B^2=0$. Along a column the two-sided operator replaces $\lambda_i$ by $\lambda_i\bar\lambda_j$, so it loses one factor of the spectrum and gains one degree of defect: the block $2+2$ becomes $3+1$, and $J_2(0)\oplus J_2(0)$ becomes $J_2(0)\oplus0\oplus0$.

The Jordan check. For $A=\begin{pmatrix}1&2\\0&1\end{pmatrix}$ the ranks of $(\mathrm{Sd}_A-\mathrm I_4)^k$ are $4,2,1,0$, that is nullities $0,2,3,4$, which is $J_3(1)\oplus J_1(1)$. For $A=\begin{pmatrix}0&1\\0&0\end{pmatrix}$ the ranks of $\mathrm{Sd}_A^k$ are $4,1,0$, which is $J_2(0)\oplus0\oplus0$. For $A=\operatorname{diag}(1,0)$ they are $4,1,1$, a rank-one projection.

Where $n=2$ Differs

The size of the operator. At $n=2$ the arrays are $4\times4$ and the operator algebra is $M_4(\mathbb{C})$, of complex dimension sixteen; there is no larger space, since $\dim_{\mathbb{C}}\mathbb{B}=4$.

The sector count. At $n=2$ the two indices $\det A$ and $B^2$ are single complex numbers, so the discriminant $\Delta=-4B^2$ is a single number and there are exactly four sectors. At order $n$ the second index is a form on a space of dimension $n^2-1$ and the sector count is not four.

The eigenvalue count. A $2\times2$ matrix has two eigenvalues, so $L_A$ has at most two distinct eigenvalues each of multiplicity two, and $\mathrm{Sd}_A$ at most four distinct ones.

What does not differ. The multiplicativity laws, the killing of the central phase and the doubling of the modulus are identities at every order.

Summary

The three operators are $L_A(Y)=AY$, $R_A(Y)=YA$ and $\mathrm{Sd}_A(Y)=AYA^{\dagger}$, with $\mathrm{Sd}_A=L_AR_{A^{\dagger}}$. In the matrix-unit frame $L_A=\mathrm I_2\otimes A$ and $R_A=A^{\mathsf T}\otimes\mathrm I_2$; in the coefficient frame they are $\sum_\mu Q_\mu M_\mu$ with the signed-permutation tables $M_x,M_y,M_z$ satisfying $M_k^2=-\mathrm I_4$ exactly.

The three are multiplicative, with the order reversed on the right and restored on the two-sided operator, verified to $2\cdot10^{-13}$. The composite of the natural involution with the conjugation of the coefficients is the matrix adjoint $A^{\dagger}$, the display of $q\mapsto(\bar Q_t,-\bar Q_x,-\bar Q_y,-\bar Q_z)$ with residual $0$; entrywise conjugation of the array is a different operation, because $\sigma_2$ is imaginary. The $\mathbb{C}$-linear quaternion conjugation is $A\mapsto\operatorname{tr}(A)\mathrm I_2-A$, whose projectors are the scalar and traceless parts.

The two-sided operator squares the scale and the singular value and kills the central phase. The four sectors are indexed by $\det A$ and $B^2$, with the forms $\operatorname{diag}(\lambda_+,\lambda_+,\lambda_-,\lambda_-)$, $J_2\oplus J_2$, $\operatorname{diag}(c,c,0,0)$ and $J_2(0)\oplus J_2(0)$ for $L_A$, and spectra $\lambda_i\bar\lambda_j$ with the Jordan types $J_3\oplus J_1$ and $J_2\oplus0\oplus0$ for $\mathrm{Sd}_A$.

Summary of Notation

Symbol Meaning
$A=\mathsf M_2(\tilde Q)$ the $2\times2$ matrix of the element
$L_A,R_A$ $Y\mapsto AY$ and $Y\mapsto YA$
$\mathrm{Sd}_A$ $Y\mapsto AYA^{\dagger}$
$M_\mu$ the four multiplication tables in the coefficient frame
$N=\det A$ the norm
$B^2=\det A-\tfrac14(\operatorname{tr}A)^2$ the vector form
$\Delta=-4B^2$ the discriminant
$\lambda_\pm=Q_t\pm iB$ the eigenvalues of $A$
$\Sigma_{\varepsilon\eta}$ the four sectors
$P_\pm$ the canonical rank-one projections of a nilpotent

Further Reading