The Three-Parameter Gauge and the Chiral Polar Decomposition of $M_4(\mathbb{C})$

Introduction

The frame carries a three-parameter complex torus, one parameter per simple nilpotent, and the mass block carries a chiral polar decomposition with two singular values; the two are different structures. The gauge is a complex rescaling of the frame; the polar decomposition is a factorization over the compact $*$-unitary group. The gauge has three complex parameters; the polar decomposition of the mass block has two real singular values and a $*$-unitary mixing, whose determinant contributes one phase. The article compares them and records the degeneration at $n=2$, where the gauge has a single parameter $\lambda$.

The article recalls the three-parameter gauge and its action on the mass block, proves the blockwise polar decomposition of a block-diagonal element and the antidiagonal case, states the general $U(4)$ polar factor, and compares the chiral polar decomposition with the single $\lambda$ of the biquaternion algebra.

The frame and the gauge are The Frame, the Simple Nilpotents and the Gauge of $M_4(\mathbb{C})$; the mass block is The Mass Block and the Chiral Polar Decomposition of $M_4(\mathbb{C})$; the full polar decomposition is The Polar Decomposition of $M_4(\mathbb{C})$; the exponential form is The Polar Decomposition of $M_4(\mathbb{C})$ in Exponential Coordinates.

The Three-Parameter Gauge

The gauge torus. Fix a frame $(v_1,v_2,v_3,v_4)$ and the matrix units

$$ E_{ij}=v_iv_j^{\vee}. $$

The other basis. The matrix units belong to the fixed frame $(v_1,v_2,v_3,v_4)$. The same frame is read on the biquaternion algebra $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ in the coefficient basis $e_t,\vec e_x,\vec e_y,\vec e_z$, where the four left and the four right multiplications are the signed permutation matrices of $M_4(\mathbb{C})$ Axis-Free Representation, and the sixteen products $L_{\mathrm{axf},e_\mu}R_{\mathrm{axf},e_\nu}$ are $M_4(\mathbb{C})$ Axis-Free Representation: Operators.

The gauge group is the torus

$$ T=(\mathbb{C}^\times)^4/\mathbb{C}^\times\cong(\mathbb{C}^\times)^3, $$

acting by

$$ v_i\longmapsto t_iv_i,\qquad v_j^{\vee}\longmapsto t_j^{-1}v_j^{\vee}, $$

so that

$$ E_{ij}\longmapsto\frac{t_i}{t_j}E_{ij}. $$

The overall scalar is divided out because it acts trivially on every $E_{ij}$.

One parameter per simple nilpotent. Write

$$ \lambda_1=\frac{t_1}{t_2},\qquad \lambda_2=\frac{t_2}{t_3},\qquad \lambda_3=\frac{t_3}{t_4}. $$

Then

$$ E_{12}\longmapsto\lambda_1E_{12},\qquad E_{23}\longmapsto\lambda_2E_{23},\qquad E_{34}\longmapsto\lambda_3E_{34}. $$

The parameters $\lambda_1,\lambda_2,\lambda_3$ are independent and coordinatise the torus. There are three, one per simple nilpotent, not four: the fourth ratio $t_4/t_1$ is the inverse of the product $\lambda_1\lambda_2\lambda_3$.

The composites are fixed by the cocycle. The composite units transform by the products:

$$ E_{13}=E_{12}E_{23}\longmapsto\lambda_1\lambda_2E_{13}, $$

$$ E_{24}=E_{23}E_{34}\longmapsto\lambda_2\lambda_3E_{24}, $$

$$ E_{14}=E_{12}E_{23}E_{34}\longmapsto\lambda_1\lambda_2\lambda_3E_{14}. $$

The weights of the composites are determined by the weights of the simple nilpotents and the product rule. There is no fourth independent scalar: the product $\lambda_1\lambda_2\lambda_3$ is not a new parameter, it is the weight of the top composite. The cocycle condition is

$$ \lambda_1\lambda_2\lambda_3\lambda_4=1, \qquad \lambda_4=\frac{t_4}{t_1}, $$

the statement that the product around the cycle is $1$.

The count. The root has $12$ complex parameters and the gauge has $3$, for

$$ 12+3=15=\dim_\mathbb{C}PGL_4. $$

The gauge is the fibre of the frame bundle over the root; the three parameters are the rescalings of the frame that leave the root's spectral data unchanged. The count is the orbit–stabiliser count of The Adjoint Action, the Orbits and the Flag Variety of $M_4(\mathbb{C})$.

The Gauge on the Mass Block

The mass block in the frame basis. The mass block is

$$ \mathcal{B}=P_+M_4P_-=\operatorname{span}\{E_{12},E_{14},E_{32},E_{34}\}, $$

with the ordered basis $(E_{12},E_{14},E_{32},E_{34})$. A mass operator is

$$ M=aE_{12}+bE_{14}+cE_{32}+dE_{34} =\begin{pmatrix}a&b\\c&d\end{pmatrix}, $$

where the rows are the chirality $+$ indices $1,3$ and the columns are the chirality $-$ indices $2,4$.

Proposition. Under the gauge,

$$ E_{12}\longmapsto\lambda_1E_{12},\qquad E_{14}\longmapsto\lambda_1\lambda_2\lambda_3E_{14}, $$

$$ E_{32}\longmapsto\lambda_2^{-1}E_{32},\qquad E_{34}\longmapsto\lambda_3E_{34}. $$

Equivalently,

$$ \begin{pmatrix}a&b\\c&d\end{pmatrix} \longmapsto \begin{pmatrix} \lambda_1a&\lambda_1\lambda_2\lambda_3b\\ \lambda_2^{-1}c&\lambda_3d \end{pmatrix} = \begin{pmatrix}\lambda_1&0\\0&\lambda_2^{-1}\end{pmatrix} \begin{pmatrix}a&b\\c&d\end{pmatrix} \begin{pmatrix}1&0\\0&\lambda_2\lambda_3\end{pmatrix}. $$

The action is $M\mapsto D_+MD_-^{-1}$ with $D_+=\operatorname{diag}(\lambda_1,\lambda_2^{-1})$ and $D_-=\operatorname{diag}(1,(\lambda_2\lambda_3)^{-1})$; the determinant product $\det D_+\det D_-^{-1}$ is $\lambda_1\lambda_3$.

Proof. The weights are the ratios $t_i/t_j$ for $i\in\{1,3\}$ and $j\in\{2,4\}$, expressed through $\lambda_1=t_1/t_2$, $\lambda_2=t_2/t_3$, $\lambda_3=t_3/t_4$. The matrix form is the entrywise computation.

The determinant weight. The determinant of the mass matrix is $ad-bc$. The two terms have the same weight:

$$ ad\longmapsto\lambda_1\lambda_3\,ad,\qquad bc\longmapsto\lambda_1\lambda_2\lambda_3\cdot\lambda_2^{-1}\,bc=\lambda_1\lambda_3\,bc. $$

Hence

$$ \det M\longmapsto\lambda_1\lambda_3\det M. $$

The weight of the determinant is $\lambda_1\lambda_3$, not the product $\lambda_1\lambda_2\lambda_3$ of all three. The middle parameter $\lambda_2$ cancels between the two terms. This is the mass-block form of the cocycle.

The gauge is complex, not $*$-unitary. The parameters $\lambda_k$ are nonzero complex numbers. The gauge changes the singular values of $M$: for a diagonal mass matrix the two singular values are multiplied by $\lvert\lambda_1\rvert$ and $\lvert\lambda_3\rvert$. The compact subgroup

$$ U(1)^3=\{(\lambda_1,\lambda_2,\lambda_3):\lvert\lambda_k\rvert=1\} $$

preserves the singular values and only changes the phases. The full compact chiral group $U(2)_L\times U(2)_R$ is larger; the compact three-parameter torus is a three-dimensional subtorus of it, and the polar decomposition is adapted to the full compact group, not to the complex torus.

The Blockwise Polar Decomposition

The block-diagonal case. Let

$$ A=\begin{pmatrix}A_+&0\\0&A_-\end{pmatrix} $$

be block-diagonal, with $A_+\in M_2(\mathbb{C})$ on the chirality $+$ and $A_-\in M_2(\mathbb{C})$ on the chirality $-$. The polar decompositions of the two blocks are

$$ A_+=U_+P_+,\qquad A_-=U_-P_-, $$

with $U_\pm\in U(2)$ and $P_\pm$ positive semidefinite. Then

$$ A=UP,\qquad U=\begin{pmatrix}U_+&0\\0&U_-\end{pmatrix}\in U(2)\times U(2), \qquad P=\begin{pmatrix}P_+&0\\0&P_-\end{pmatrix}. $$

The polar factor is block-diagonal, and the singular values of $A$ are the four singular values of $A_+$ and $A_-$, two from each block. The chiralities decouple: the polar decomposition is the direct sum of the two chiral polar decompositions.

The antidiagonal case. Let

$$ A=\begin{pmatrix}0&M\\N&0\end{pmatrix}, \qquad M:\mathbb{C}^2_-\to\mathbb{C}^2_+,\quad N:\mathbb{C}^2_+\to\mathbb{C}^2_- . $$

Then

$$ A^*A=\begin{pmatrix}N^*N&0\\0&M^*M\end{pmatrix}, $$

so the positive factor is

$$ P=\begin{pmatrix}\lvert N\rvert&0\\0&\lvert M\rvert\end{pmatrix}, \qquad \lvert N\rvert=\sqrt{N^*N},\quad \lvert M\rvert=\sqrt{M^*M}. $$

On the invertible blocks, the polar factor is

$$ U=AP^{-1}= \begin{pmatrix} 0&M\lvert M\rvert^{-1}\\ N\lvert N\rvert^{-1}&0 \end{pmatrix}. $$

The blocks $M\lvert M\rvert^{-1}$ and $N\lvert N\rvert^{-1}$ are the $*$-unitary chiral phases of $M$ and $N$. The full polar decomposition of an antidiagonal matrix is the direct sum of the two chiral polar decompositions, one for each off-diagonal block.

The general case. For a general block matrix

$$ A=\begin{pmatrix}A_+&M\\N&A_-\end{pmatrix}, $$

the polar factor $U=A(A^*A)^{-1/2}$ is not block-diagonal and not block-antidiagonal. It mixes the chiralities, and the blockwise decomposition is only the special case in which $M=N=0$. The general $U(4)$ polar factor is the subject of the next section.

The General Element and the $U(4)$ Polar Factor

Theorem. Every $A\in M_4(\mathbb{C})$ has a polar decomposition

$$ A=UP, $$

with $U\in U(4)$ and $P$ positive semidefinite. If $A$ is invertible, the decomposition is unique, with

$$ P=(A^*A)^{1/2},\qquad U=A(A^*A)^{-1/2}. $$

If $A$ is singular, $P$ is unique and $U$ is unique on the range of $P$ and arbitrary on its kernel.

Proof. The operator $A^*A$ is positive semidefinite, so it has a unique positive semidefinite square root $P$. On the range of $P$, define $U(Pv)=Av$; this is well defined and isometric because

$$ \lVert Av\rVert^2=\langle Av,Av\rangle=\langle v,A^*Av\rangle=\langle v,P^2v\rangle=\lVert Pv\rVert^2. $$

Extend $U$ isometrically from the range of $P$ to its orthogonal complement, which is the kernel of $P$, with values in the kernel of $A^*$. The dimensions agree, because $\operatorname{rank}P=\operatorname{rank}A$, so the extension exists and $U$ is $*$-unitary. Then $A=UP$. Uniqueness on the invertible elements follows because $P$ is determined by $A^*A$ and $U=AP^{-1}$.

The chiral polar decomposition is different. The decomposition $M=U\Sigma V^*$ of the mass block uses two unitaries, one on each chirality, and a diagonal map between them. It is the polar decomposition of a map between two different Hilbert spaces, not of an endomorphism of one space. The full $U(4)$ polar factor of the block matrix is an endomorphism of $\mathbb{C}^4$; it mixes the chiralities in general, and it is not the pair $(U_L,U_R)$ of the mass block.

The relation. For the antidiagonal block matrix $A=\begin{pmatrix}0&M\\N&0\end{pmatrix}$, the two decompositions agree: the $U(4)$ polar factor is the block-antidiagonal matrix built from the chiral phases of $M$ and $N$, and the positive factor is the block-diagonal matrix of the two mass moduli. For a general block matrix, the two decompositions are different.

The Chiral Polar Decomposition of the Mass Block

The decomposition. Every $M\in\mathcal{B}$ has

$$ M=U\Sigma V^*, \qquad U\in U(2)_L,\quad V\in U(2)_R,\quad \Sigma=\operatorname{diag}(\sigma_1,\sigma_2), $$

with $\sigma_1\ge\sigma_2\ge0$. The chiral polar decomposition is

$$ M=WP,\qquad W=UV^*:\mathbb{C}^2_-\to\mathbb{C}^2_+,\quad P=V\Sigma V^*:\mathbb{C}^2_-\to\mathbb{C}^2_- . $$

The factor $W$ is the $*$-unitary phase, and $P$ is the mass modulus. The two singular values $\sigma_1,\sigma_2$ are the invariants of the compact chiral gauge; the phase of $\det M$ is the remaining invariant under the special unitary gauge.

The gauge weights of the polar data. Under the three-parameter gauge, the mass block transforms by $M\mapsto D_+MD_-^{-1}$ with $D_+=\operatorname{diag}(\lambda_1,\lambda_2^{-1})$ and $D_-=\operatorname{diag}(1,(\lambda_2\lambda_3)^{-1})$. The singular values are invariant exactly when the gauge lies in the compact subgroup $U(1)^3$, that is, when $\lvert\lambda_1\rvert=\lvert\lambda_2\rvert=\lvert\lambda_3\rvert=1$; a complex rescaling changes the moduli. The determinant transforms by $\lambda_1\lambda_3$. The chiral phase transforms by the phase of $\lambda_1\lambda_3$. The two singular values and the phase are therefore not the three gauge parameters: the gauge is complex, and it changes the moduli.

The invariant count. The mass block has complex dimension $4$, hence real dimension $8$. The three-parameter gauge has complex dimension $3$, hence real dimension $6$. The quotient has real dimension $2$. On the open dense locus where $bc\ne0$, the quotient is parametrised by one complex invariant, the cross ratio $ad/(bc)$ of the four entries; as two real invariants, it is the modulus and the argument of the cross ratio. The two singular values are two real invariants of the compact gauge, but they are not invariant under the complex gauge; the quotient invariants are the combinations that survive the complex rescalings. The phase of $\det M$ is not a quotient invariant.

The Single $\lambda$ of $\mathbb{B}$

At $n=2$ the gauge is one-dimensional. The biquaternion algebra is $\mathbb{B}=M_2(\mathbb{C})$. The frame is $(v_1,v_2)$, the gauge is

$$ T=(\mathbb{C}^\times)^2/\mathbb{C}^\times\cong\mathbb{C}^\times, \qquad \lambda=\frac{t_1}{t_2}, $$

and the only simple nilpotent is $E_{12}$, with

$$ E_{12}\longmapsto\lambda E_{12}. $$

The mass block is the line $\mathbb{C}E_{12}$, and the mass operator is a complex scalar $z$. The gauge acts by

$$ z\longmapsto\lambda z. $$

The polar decomposition of the scalar. The polar decomposition is

$$ z=e^{i\theta}\lvert z\rvert . $$

The modulus $\lvert z\rvert$ is invariant under the compact gauge $U(1)\times U(1)$, and the phase $e^{i\theta}$ is removed by the compact gauge. The complex gauge $\lambda\in\mathbb{C}^\times$ is transitive on the nonzero masses: for $z\ne0$, choose $\lambda=1/z$ to send $z$ to $1$. The only invariant of the complex gauge is whether $z$ is zero.

The comparison. In $\mathbb{B}$ the gauge has one complex parameter $\lambda$, the mass block has one complex entry $z$, and the gauge is transitive on the nonzero locus. The polar decomposition has one modulus and one phase; the modulus is the compact invariant, and the phase is a gauge artefact. In $M_4$ the gauge has three complex parameters $\lambda_1,\lambda_2,\lambda_3$, the mass block has four complex entries, and the gauge is not transitive: its orbits have complex dimension $3$ and the quotient has complex dimension $1$. The polar decomposition has two singular values and one phase; the two singular values are compact invariants, and the phase is a compact artefact. The coincidence of counts at $n=2$ — one complex gauge parameter, one complex mass — does not persist: at $n=4$ the three gauge parameters are not the two singular values plus the phase.

The single $\lambda$ is the degeneration. The three parameters $\lambda_1,\lambda_2,\lambda_3$ degenerate to the single $\lambda$ when the chain has one step. The mass block degenerates from a $2\times2$ matrix to a scalar. The polar decomposition degenerates from two singular values to one modulus. The pattern is the $m=1$ case of the general $M_{2m}$:

$$ \underbrace{2m-1}_{\text{gauge}}\ \text{complex parameters},\qquad \underbrace{m}_{\text{singular values}}\ \text{real invariants},\qquad \underbrace{1}_{\text{phase}}\ \text{compact phase}. $$

For $m=1$: one gauge parameter, one singular value, one phase. For $m=2$: three gauge parameters, two singular values, one phase.

Summary

The frame carries the three-parameter complex torus $(\mathbb{C}^\times)^3$, one parameter $\lambda_k$ per simple nilpotent $E_{k,k+1}$. The mass block $\mathcal{B}$ is spanned by $E_{12},E_{14},E_{32},E_{34}$ with gauge weights $\lambda_1,\lambda_1\lambda_2\lambda_3,\lambda_2^{-1},\lambda_3$; the determinant has weight $\lambda_1\lambda_3$. The blockwise polar decomposition of a block-diagonal element is the direct sum of the two chiral polar decompositions; the antidiagonal case reduces to the chiral phases and moduli of the two off-diagonal blocks; the general element has a $U(4)$ polar factor that mixes the chiralities. The mass block has the chiral polar decomposition $M=U\Sigma V^*=WP$, with two singular values and a chiral phase. The three complex gauge parameters are not the two real singular values and the phase: the gauge quotient of the mass block has complex dimension one, while the polar data have three real parameters. At $n=2$ the gauge has a single $\lambda$, the mass block is a line, and the polar decomposition is the polar form of a scalar; the coincidence of counts is the degeneration of the three-parameter case.

$$ \boxed{\ T=(\mathbb{C}^\times)^3,\quad \det M\mapsto\lambda_1\lambda_3\det M,\quad M=U\Sigma V^*,\quad \sigma_1\ge\sigma_2\ge0,\quad \dim_\mathbb{C}\mathcal{B}=4. } $$

Object Complex dimension Real dimension
Gauge torus $T$ $3$ $6$
Mass block $\mathcal{B}$ $4$ $8$
Gauge orbit (generic) $3$ $6$
Gauge quotient (generic) $1$ $2$
Chiral polar data — $2$ singular values $+1$ phase
Biquaternion gauge $1$ $2$
Biquaternion mass block $1$ $2$

Summary of Notation

Symbol Meaning
$\lambda_1,\lambda_2,\lambda_3$ the gauge parameters of the simple nilpotents
$T=(\mathbb{C}^\times)^3$ the gauge torus
$D_+=\operatorname{diag}(\lambda_1,\lambda_2^{-1})$ the left gauge matrix on the mass block
$D_-=\operatorname{diag}(1,(\lambda_2\lambda_3)^{-1})$ the right gauge matrix, entering as $D_-^{-1}$
$M=U\Sigma V^*$ the chiral singular value decomposition
$W=UV^*$ the chiral phase
$P=V\Sigma V^*$ the mass modulus
$\lambda$ the single gauge parameter of $\mathbb{B}$

Further Reading