The Two-Parameter Gauge of $M_3(\mathbb{C})$

Introduction

A frame of $M_{3}(\mathbb{C})$ has three diagonal idempotents and six off-diagonal units. The idempotents are fixed by the root; the off-diagonal units are fixed only up to rescalings that preserve the matrix-unit table. The rescalings are

$$ E_{ij}\longmapsto\frac{t_{i}}{t_{j}}E_{ij},\qquad t_{i}\in\mathbb{C}^{\times}, $$

and the common scalar of the three $t_{i}$ acts trivially, so the effective group is $(\mathbb{C}^{\times})^{2}$. Its two parameters are $\lambda_{1}=t_{1}/t_{2}$ and $\lambda_{2}=t_{2}/t_{3}$; the third ratio is the product $\lambda_{1}\lambda_{2}$, and there is no third independent scalar. The idempotents do not depend on $\lambda_{1},\lambda_{2}$.

This article reads the gauge on the Peirce coordinates, identifies the gauge-invariant combinations of the off-diagonal entries, proves that the Peirce table is unchanged, compares the two-parameter gauge with the single $\lambda$ of the biquaternion algebra, and separates the gauge from the global phase of a quantum state. The frame and the simple nilpotents are The Frame, the Simple Nilpotents and the Gauge of $M_3(\mathbb{C})$; the Peirce decomposition is The Peirce Decomposition of $M_{3}(\mathbb{C})$; the corner is The $M_{2}$ Corner of $M_{3}(\mathbb{C})$.

The Gauge on the Peirce Coordinates

The frame change. Let $E_{ij}$ be a normalized frame, $E_{ij}E_{kl}=\delta_{jk}E_{il}$, $E_{ii}=P_{i}$. For $t_{1},t_{2},t_{3}\in\mathbb{C}^{\times}$ put

$$ E_{ij}'=\frac{t_{i}}{t_{j}}E_{ij}. $$

Then $E_{ij}'E_{kl}'=\delta_{jk}E_{il}'$ and $E_{ii}'=E_{ii}$, so the $E_{ij}'$ are again a normalized frame with the same diagonal idempotents. The map $E_{ij}\mapsto E_{ij}'$ is the gauge transformation of The Frame, the Simple Nilpotents and the Gauge of $M_3(\mathbb{C})$, with $E_{ij}'=D E_{ij}D^{-1}$ for $D=\operatorname{diag}(t_{1},t_{2},t_{3})$.

The coordinates. Write an element as

$$ A=\sum_{i,j}\alpha_{ij}E_{ij}=\sum_{i,j}\alpha_{ij}'E_{ij}'. $$

Since $E_{ij}'=(t_{i}/t_{j})E_{ij}$, the two coordinate arrays are related by

$$ \alpha_{ij}'=\frac{t_{j}}{t_{i}}\alpha_{ij}. $$

The diagonal coordinates are invariant,

$$ \alpha_{11}'=\alpha_{11},\qquad \alpha_{22}'=\alpha_{22},\qquad \alpha_{33}'=\alpha_{33}, $$

and the off-diagonal ones transform by the six weights

$$ \alpha_{12}'=\lambda_{1}^{-1}\alpha_{12},\quad \alpha_{23}'=\lambda_{2}^{-1}\alpha_{23},\quad \alpha_{13}'=(\lambda_{1}\lambda_{2})^{-1}\alpha_{13}, $$

$$ \alpha_{21}'=\lambda_{1}\alpha_{21},\quad \alpha_{32}'=\lambda_{2}\alpha_{32},\quad \alpha_{31}'=\lambda_{1}\lambda_{2}\alpha_{31}, $$

where $\lambda_{1}=t_{1}/t_{2}$ and $\lambda_{2}=t_{2}/t_{3}$. The pattern is the root system $A_{2}$: the units $E_{12},E_{23},E_{13}$ carry the positive weights $1,1,2$ and the units $E_{21},E_{32},E_{31}$ carry the negative weights $-1,-1,-2$, with the composite $E_{13}$ carrying the sum $1+1=2$. The coordinates transform by the inverse weights, as the formulas above show.

Worked example. Take

$$ A=\begin{pmatrix}1&2&3\\4&5&6\\7&8&9\end{pmatrix} $$

in the standard frame, and $\lambda_{1}=2$, $\lambda_{2}=3$, so $D=\operatorname{diag}(6,3,1)$. The coordinates become

$$ \alpha_{12}'=\frac{2}{2}=1,\quad \alpha_{23}'=\frac{6}{3}=2,\quad \alpha_{13}'=\frac{3}{6}=\frac12, $$

$$ \alpha_{21}'=4\cdot2=8,\quad \alpha_{32}'=8\cdot3=24,\quad \alpha_{31}'=7\cdot6=42, $$

while the diagonal $1,5,9$ is unchanged. The gauge is not a symmetry of the array of numbers; it is a change of the frame in which the array is read. The element $A$ itself is fixed.

The inverse reading. If instead one keeps the standard frame and conjugates the element, $A\mapsto DAD^{-1}$, the entries become

$$ (DAD^{-1})_{ij}=\frac{t_{i}}{t_{j}}a_{ij}=\lambda_{ij}a_{ij}, $$

which is the inverse of the transformation of the coordinates. The two readings are the two sides of the same change of frame: the frame is rescaled, or the element is conjugated. The gauge-invariant statements below are the same in either reading.

The Gauge-Invariant Combinations

The diagonal is invariant. The three numbers $\alpha_{11},\alpha_{22},\alpha_{33}$ are unchanged. They are the values of the three idempotents on $A$:

$$ P_{i}AP_{i}=\alpha_{ii}P_{i}. $$

The trace is $\alpha_{11}+\alpha_{22}+\alpha_{33}$, so it is invariant; the determinant is the determinant of the coordinate array, so it is invariant as well.

The opposite-pair products are invariant. The three products

$$ \alpha_{12}\alpha_{21},\qquad \alpha_{23}\alpha_{32},\qquad \alpha_{13}\alpha_{31} $$

are unchanged by the gauge:

$$ \alpha_{12}'\alpha_{21}'=\frac{\alpha_{12}}{\lambda_{1}}\cdot\lambda_{1}\alpha_{21}=\alpha_{12}\alpha_{21}, $$

and similarly for the other two pairs. These are the coherence entries of the Peirce coordinates: for each pair of opposite lines, the product of the two coordinates is a gauge-invariant number. The product is the coefficient of $P_{i}$ in the square of the off-diagonal part, and it is the natural pairing of the line with its dual.

The oriented three-cycles are invariant. The products

$$ \alpha_{12}\alpha_{23}\alpha_{31},\qquad \alpha_{21}\alpha_{32}\alpha_{13} $$

are also invariant:

$$ \alpha_{12}'\alpha_{23}'\alpha_{31}'=\frac{\alpha_{12}}{\lambda_{1}}\cdot\frac{\alpha_{23}}{\lambda_{2}}\cdot\lambda_{1}\lambda_{2}\alpha_{31}=\alpha_{12}\alpha_{23}\alpha_{31}, $$

and the other cycle is its opposite. The two cycles are the two orientations of the triangle $1\to2\to3\to1$; their product is the product of the three opposite-pair invariants.

The invariant ring. The gauge group $(\mathbb{C}^{\times})^{2}$ acts on the six off-diagonal coordinates with the six weights $\pm1,\pm1,\pm2$. The four invariant combinations

$$ \alpha_{12}\alpha_{21},\quad \alpha_{23}\alpha_{32},\quad \alpha_{13}\alpha_{31},\quad \alpha_{12}\alpha_{23}\alpha_{31} $$

generate the invariant ring; the second three-cycle is

$$ \alpha_{21}\alpha_{32}\alpha_{13} =\frac{(\alpha_{12}\alpha_{21})(\alpha_{23}\alpha_{32})(\alpha_{13}\alpha_{31})}{\alpha_{12}\alpha_{23}\alpha_{31}}. $$

The diagonal coordinates are additional invariants, independent of the gauge. The count of the off-diagonal invariants is $6-2=4$, matching the four generators; the ring has relations among them, but the four numbers are the gauge-invariant content of the six off-diagonal entries.

The invariants are the data of the element. Two coordinate arrays related by the gauge describe the same element $A$ in two frames. The invariants are the functions of $A$ that do not depend on the frame: the diagonal values, the three opposite-pair products and the two three-cycles. The off-diagonal entries themselves are not invariant, and no individual off-diagonal entry has a gauge-independent meaning.

The Peirce Table Is Unchanged

Theorem. The gauge transformation preserves the Peirce lines, the Peirce table, the multiplication rule and the idempotents. It changes only the choice of the normalized matrix units inside the lines.

Proof. The lines are $A_{ij}=P_{i}M_{3}P_{j}$, defined by the idempotents, and the idempotents are fixed. The matrix units $E_{ij}$ and $E_{ij}'$ both lie in $A_{ij}$ and are both nonzero, so they differ by a scalar, which is $t_{i}/t_{j}$. The multiplication rule

$$ E_{ij}E_{kl}=\delta_{jk}E_{il} $$

is preserved by the rescaling:

$$ E_{ij}'E_{kl}'=\frac{t_{i}}{t_{j}}\frac{t_{k}}{t_{l}}E_{ij}E_{kl} =\delta_{jk}\frac{t_{i}}{t_{l}}E_{il}=\delta_{jk}E_{il}'. $$

The structure constants of the algebra in the frame are the numbers $1$ and $0$ of this table, and they are unchanged. The gauge is exactly the freedom of choosing the normalized frame; it is not a change of the multiplication.

The table in coordinates. The product of two elements is

$$ AB=\sum_{i,j}\Bigl(\sum_{m}\alpha_{im}\beta_{mj}\Bigr)E_{ij} =\sum_{i,j}\Bigl(\sum_{m}\alpha_{im}'\beta_{mj}'\Bigr)E_{ij}', $$

because the coordinate arrays multiply by the same matrix product in either frame. The gauge is a change of basis of the coordinate space that preserves the matrix multiplication; it is the conjugation $\alpha'=T^{-1}\alpha T$ with $T=\operatorname{diag}(t_{1},t_{2},t_{3})$. Indeed $\alpha'_{ij}=(t_{j}/t_{i})\alpha_{ij}$. The product is preserved because conjugation by $T$ is an algebra automorphism of the array algebra. This is the coordinate form of the frame change.

The gauge is an automorphism of the coordinate algebra. The map $\alpha\mapsto T^{-1}\alpha T$ is an automorphism of $M_{3}(\mathbb{C})$; it is inner, given by the diagonal matrix $T$. The gauge group is the image of the diagonal torus in $PGL_{3}$, and its action on the coordinates is the adjoint action. The frame change and the coordinate conjugation are the same inner automorphism read in two ways.

The Count and the Cocycle

The gauge group. The parameters $t_{1},t_{2},t_{3}$ and $ct_{1},ct_{2},ct_{3}$ give the same rescaling, so the effective group is

$$ (\mathbb{C}^{\times})^{3}/\mathbb{C}^{\times}\cong(\mathbb{C}^{\times})^{2}, $$

of complex dimension two. The two parameters are

$$ \lambda_{1}=\frac{t_{1}}{t_{2}},\qquad \lambda_{2}=\frac{t_{2}}{t_{3}}, \qquad \frac{t_{1}}{t_{3}}=\lambda_{1}\lambda_{2}. $$

The third ratio is not independent. If a rescaling $E_{ij}\mapsto\lambda_{ij}E_{ij}$ preserves the frame, the cocycle

$$ \lambda_{ij}\lambda_{jk}=\lambda_{ik},\qquad \lambda_{ii}=1,\qquad \lambda_{ji}=\lambda_{ij}^{-1} $$

forces $\lambda_{ij}=t_{i}/t_{j}$ for $t_{i}=\lambda_{i0}$, so the solution space is exactly the two-parameter torus. In particular $\lambda_{13}=\lambda_{12}\lambda_{23}$: the composite carries the product.

The count of the frame. The frame has dimension $8$, the root orbit dimension $6$, and the gauge dimension $2$:

$$ 8=6+2. $$

The frame is a point of $PGL_{3}$, the root is a point of $PGL_{3}/T$, and the gauge is the torus $T\cong(\mathbb{C}^{\times})^{2}$. The count is the orbit-stabiliser count of The Frame, the Simple Nilpotents and the Gauge of $M_3(\mathbb{C})$.

The composite is forced. The simple nilpotents $E_{12},E_{23}$ carry $\lambda_{1},\lambda_{2}$; the composite $E_{13}=E_{12}E_{23}$ carries $\lambda_{1}\lambda_{2}$. There is no independent scalar $\lambda_{3}$ on $E_{13}$: the cocycle would fail unless $\lambda_{3}=\lambda_{1}\lambda_{2}$. The same holds for the opposite composite $E_{31}=E_{32}E_{21}$, which carries $\lambda_{1}^{-1}\lambda_{2}^{-1}$.

The two parameters are the simple roots. In the character lattice of the torus, the weights of $E_{12}$ and $E_{23}$ are the two simple roots $\alpha_{1},\alpha_{2}$ of $A_{2}$, and the weight of $E_{13}$ is the highest root $\alpha_{1}+\alpha_{2}$. The gauge parameters $\lambda_{1},\lambda_{2}$ are the values of the two simple roots; the composite value is their product. The gauge is the maximal torus of $PGL_{3}$, and the root system is the $A_{2}$ of The Frame, the Simple Nilpotents and the Gauge of $M_3(\mathbb{C})$.

The Comparison with the Biquaternion Single $\lambda$

In $M_{2}(\mathbb{C})$ the frame has one gauge parameter, not two.

For the biquaternion algebra $\mathbb{B}\cong M_{2}(\mathbb{C})$, the frame is $E_{11},E_{22},E_{12},E_{21}$, and the gauge is

$$ E_{12}\longmapsto\mu E_{12},\qquad E_{21}\longmapsto\mu^{-1}E_{21},\qquad \mu\in\mathbb{C}^{\times}. $$

The effective group is $(\mathbb{C}^{\times})^{2}/\mathbb{C}^{\times}\cong\mathbb{C}^{\times}$, of dimension one. The single parameter $\mu$ is the analogue of the product $\lambda_{1}\lambda_{2}$ of the $M_{3}$ gauge, but in $M_{2}$ there is no simple-root decomposition into two steps: the root system is $A_{1}$, with one node, and the one off-diagonal pair carries the one weight $\mu$.

The comparison.

  • $M_{2}$: one idempotent pair, one off-diagonal pair, one gauge parameter $\mu$, one invariant $\alpha_{12}\alpha_{21}$.
  • $M_{3}$: three idempotents, three off-diagonal pairs, two gauge parameters $\lambda_{1},\lambda_{2}$, and the composite pair carries $\lambda_{1}\lambda_{2}$.
  • The pattern: for $M_{n}(\mathbb{C})$ the gauge is $(\mathbb{C}^{\times})^{n-1}$; the count is the number of simple roots of $A_{n-1}$, which is $n-1$. The biquaternion case is $n=2$, one parameter; the $M_{3}$ case is $n=3$, two parameters.

The single $\lambda$ of $\mathbb{B}$ is not a third parameter of $M_{3}$. The biquaternion $\mu$ and the $M_{3}$ pair $(\lambda_{1},\lambda_{2})$ are the gauge groups of two different frames. In $M_{3}$ the composite carries the product; in $M_{2}$ there is no composite. The two-parameter gauge of $M_{3}$ reduces to the one-parameter gauge of $M_{2}$ only after a degeneration in which one of the two simple nilpotents is set to zero; the remaining parameter is then the single $\lambda$ of the biquaternion corner. This degeneration is the subject of The $M_{2}$ Corner of $M_{3}(\mathbb{C})$.

The Gauge Is Not a Physical Phase

The gauge is an algebraic redundancy of the frame; the global phase is a redundancy of a vector.

The gauge group $(\mathbb{C}^{\times})^{2}$ acts on the frame by $E_{ij}\mapsto(t_{i}/t_{j})E_{ij}$. It fixes the idempotents, the root, the spectrum, the trace, the determinant and the characteristic polynomial. It is a change of the matrix units, not a change of the element. A physical global phase $e^{i\theta}$ acts on a vector $\psi\in\mathbb{C}^{3}$ by $\psi\mapsto e^{i\theta}\psi$; it is a one-parameter group $U(1)$, and it does not act on the frame in the same way. The two groups act on different spaces and have different dimensions.

The gauge is complex. The parameters $t_{i}$ are nonzero complex numbers, not phases. The gauge rescales the off-diagonal entries by complex numbers, including their moduli. The compact real form of the gauge is $U(1)^{2}$, the group of relative phases of the three lines; the diagonal factor $U(1)$ of $U(1)^{3}$ acts trivially on the frame ratios $t_{i}/t_{j}$ and is divided out. This diagonal factor is the algebraic analogue of the global phase of a vector, but it is not the same action: the global phase multiplies a vector of $\mathbb{C}^{3}$, while the diagonal frame rescaling multiplies the matrix units. The two are distinct one-parameter groups acting on different spaces, and the article does not identify them.

The gauge does not change the state. A density matrix $\rho$ written in the frame has off-diagonal entries that rescale under the gauge; the diagonal entries and the products of opposite coherences are invariant. The physical state is the element $\rho$, not its coordinate array; the gauge is the freedom of the frame in which the array is written. The global phase of a pure state is the freedom of the vector, not of the frame. The two freedoms are distinct, and the article keeps them distinct.

Summary

The gauge of the $M_{3}$ frame is

$$ \boxed{\;E_{ij}\longmapsto\frac{t_{i}}{t_{j}}E_{ij},\qquad (t_{1},t_{2},t_{3})\in(\mathbb{C}^{\times})^{3}/\mathbb{C}^{\times}\cong(\mathbb{C}^{\times})^{2},\;} $$

with the two parameters

$$ \lambda_{1}=\frac{t_{1}}{t_{2}},\qquad \lambda_{2}=\frac{t_{2}}{t_{3}},\qquad \lambda_{3}=\lambda_{1}\lambda_{2}. $$

It fixes the idempotents, the root, the trace, the determinant and the spectrum. On the Peirce coordinates it acts by $\alpha_{ij}\mapsto(t_{j}/t_{i})\alpha_{ij}$; the diagonal entries and the products $\alpha_{12}\alpha_{21}$, $\alpha_{23}\alpha_{32}$, $\alpha_{13}\alpha_{31}$ and the three-cycles are invariant. The Peirce table and the multiplication rule are unchanged: the gauge is the freedom of the normalized frame, not a change of the algebra.

Quantity Value for $M_{3}(\mathbb{C})$
Gauge group $(\mathbb{C}^{\times})^{2}$
Parameters $\lambda_{1},\lambda_{2}$
Composite rescaling $\lambda_{1}\lambda_{2}$
Third independent scalar none
Idempotents depending on the gauge none
Frame dimension $8$
Root orbit dimension $6$
Gauge dimension $2$
Biquaternion gauge one parameter $\mu$

The single $\lambda$ of the biquaternion algebra is the $n=2$ case of the pattern $(\mathbb{C}^{\times})^{n-1}$; the $M_{3}$ case has two parameters because the root system $A_{2}$ has two simple roots. The gauge is not the global phase of a quantum state: it is a complex two-parameter redundancy of the frame, whose compact form is $U(1)^{2}$.

Summary of Notation

Symbol Meaning
$t_{i}$ the three gauge parameters, modulo a common scalar
$\lambda_{1}=t_{1}/t_{2}$ the gauge parameter on $E_{12}$
$\lambda_{2}=t_{2}/t_{3}$ the gauge parameter on $E_{23}$
$\lambda_{3}=\lambda_{1}\lambda_{2}$ the forced parameter on $E_{13}$
$\alpha_{ij}$ the Peirce coordinates
$\alpha_{ij}'$ the coordinates after the gauge
$\mu$ the single gauge parameter of the biquaternion frame

Further Reading