The Geometry of the Two Halves at $n=2$, and the Return of the Two-Sided Operators

Introduction

The trace form gives the symmetric half a positive definite metric of inertia $(3,0)$ and the antisymmetric half a negative definite metric of inertia $(0,1)$, the determinant gives the symmetric half a Lorentz metric of inertia $(1,2)$, and the two-sided operators act on the matrix algebra as the group $\mathrm{GL}_2\times\mathrm{GL}_2$ whose orbits are the three rank strata.

The article collects the geometric data of the halves, computes the isometry groups of the two metrics, describes the orbits of the two-sided action, relates the cone of the symmetric half to its Lorentz metric, describes the geodesics, and interprets the two-sided operators as geometric transformations preserving the rank. It assumes The Two Halves and the Analysis: Smooth Structures at $n=2$, The Trace Form and Its Inertia on the Two Halves of $M_2(\mathbb{R})$ and The Two-Sided Operators and the Segre Cone at $n=2$.

It follows the biquaternion article The Four Forms and the Four Algebraic Norms over the Biquaternion Algebra for the same metric data in the biquaternion coordinates.

The Metric on the Halves

The trace form is a positive definite metric on the symmetric half and a negative definite metric on the antisymmetric half.

Proposition. On the symmetric half the trace form has the eigenvalues one, one and two, and on the antisymmetric half it has the single eigenvalue minus two; both are nondegenerate.

Proof. The Gram matrix of the trace form on the symmetric half in the basis of the two diagonal units and the off-diagonal unit is diagonal with entries one, one and two; on the antisymmetric half the single basis element has squared trace minus two.

Corollary. The matrix algebra carries a metric of inertia $(4,4)$ split as three plus one in the real halves and as four plus four in the complex halves.

Proof. The trace form is positive definite on the Hermitian half and negative definite on the anti-Hermitian half, and the real halves split the same way by the symmetric and antisymmetric parts.

Corollary. The orthogonal complement of a half is the other half, so the halves are the eigenspaces of the transposition with the two signs.

Proof. The trace of the product of a symmetric with an antisymmetric matrix is zero, and the dimensions add to the total.

The Isometry Groups

Proposition. The isometries of the trace metric on the matrix algebra form the unitary group of the four-dimensional complex space of matrices, which contains the conjugation by unitary matrices and the two-sided action with unitary parameters.

Proof. The trace metric is the standard Hermitian form on the four complex coordinates, whose linear isometries are the unitary matrices of order four; the conjugation and the two-sided action with unitary parameters are unitary for that form.

Corollary. The isometries preserving each half are the conjugation by orthogonal matrices on the real halves and by unitary matrices on the complex ones.

Proof. An isometry that preserves the symmetric half also preserves its orthogonal complement, which is the antisymmetric half.

Corollary. The cone of the symmetric half is preserved by the Lorentz group of inertia $(1,2)$, which is larger than the orthogonal group.

Proof. The determinant on the symmetric half is a form of inertia $(1,2)$, and its automorphism group is the Lorentz group of that inertia.

The Orbits of the Two-Sided Action

Theorem. The group $\mathrm{GL}_2(\mathbb{C})\times\mathrm{GL}_2(\mathbb{C})$ acts on the matrix algebra by $(X,Y)\cdot Z=XZY$, and its orbits are the origin, the rank-one matrices and the invertible matrices.

Proof. The action preserves the rank by the product formula, and conversely two matrices of the same rank are carried to each other by choosing bases in their images and kernels.

Corollary. The orbit of the identity is the group itself, of real dimension eight, and it is open and dense.

Proof. The orbit of the identity is the set of invertible matrices, which is open and dense.

Corollary. The orbit of a rank-one matrix is the cone of decomposable elements, of real dimension six, with the origin deleted.

Proof. A rank-one matrix is carried to any other by the action, and the set of rank-one matrices is the decomposable cone.

The Cone and the Metric

Proposition. The positive semidefinite cone of the symmetric half is a cone of the Lorentz metric of inertia $(1,2)$, and its boundary is the set of null directions of the rank-one form.

Proof. The determinant vanishes exactly on the rank-one matrices, which are the null directions of the Lorentz form, and the cone is one of its two time cones.

Corollary. The geodesics of the Lorentz metric that stay inside the cone are the rays through the origin and the curves of constant Lorentz length.

Proof. The rays are the null geodesics of the cone, and the curves of constant length are the geodesics through the interior points.

Corollary. The cone is homogeneous for the Lorentz group of the half, which acts transitively on its interior.

Proof. The Lorentz group acts transitively on the interior of its time cone, and the stabiliser of a point is compact.

Geodesics and the Exponential

Proposition. The trace metric on the symmetric half is flat, its geodesics are the straight lines, and its Riemannian exponential at a point $A$ is $S\mapsto A+S$.

Proof. The metric is constant in the coordinates of the entries, so the Levi-Civita connection vanishes and the geodesics are the straight lines.

Proposition. The matrix exponential is a diffeomorphism from the symmetric half onto the positive definite cone, and it is not the Riemannian exponential of the trace metric.

Proof. The exponential of a symmetric matrix is positive definite, its inverse is the logarithm, and the two maps are smooth; it is a different map from the affine displacement of the flat metric.

Corollary. The positive definite cone is convex in the flat metric, and every pair of its points is joined by a unique geodesic inside it.

Proof. The cone is convex as a set, and the geodesics of the flat metric are the straight segments.

Corollary. The exponential map is a diffeomorphism from the symmetric half onto the positive definite cone.

Proof. The exponential has the identity as derivative at the origin, is injective on symmetric matrices, and its image is the cone.

The Return of the Two-Sided Operators

Proposition. The two-sided operators that preserve the trace metric are exactly those with unitary parameters, and they form the group of isometries preserving the rank strata.

Proof. The two-sided operator multiplies the trace of $Z^{*}Z$ by the product of the squared norms of the parameters, and the isometry condition forces both parameters to be unitary.

Corollary. The two-sided operators act by similarities of the trace metric, with ratio the product of the two lengths, so the conformal group of the matrix space contains the group of two-sided operators.

Proof. The metric is scaled by the product of the squared norms, which is a single scalar.

Corollary. The two-sided operators preserve the Segre cone and act transitively on each rank stratum, so the geometry of the strata is homogeneous.

Proof. The orbits of the action are the rank strata, and each is a single orbit.

Summary

The trace form gives the symmetric half a positive definite metric with the eigenvalues one, one and two and the antisymmetric half a negative definite metric; the matrix algebra carries a metric of inertia $(4,4)$. The isometry group is the unitary group of the Hilbert–Schmidt form, acting by conjugation and by the two-sided action.

The group of two-sided operators has the origin, the rank-one cone and the invertible matrices as its orbits, so its geometry is the geometry of the rank strata. The positive cone of the symmetric half is a Lorentz cone whose null directions are the rank-one matrices, and the two-sided operators act by conformal similarities of the trace metric.

Summary of Notation

Symbol Meaning
$\langle Z,W\rangle=\operatorname{tr}(Z^{*}W)$ the trace metric on the algebra
$(4,4)$ its inertia on the whole algebra
$(1,2)$ the inertia of the determinant on the symmetric half
$\mathrm{GL}_2\times\mathrm{GL}_2$ the group of two-sided operators
$Z\mapsto XZY$ the two-sided action

Further Reading