The Two Halves and Their Topology at $n=2$
Introduction
The two halves over the real field are a three-dimensional and a one-dimensional vector space, both contractible; the two halves over the complex field are two copies of a four-dimensional space; and the only nontrivial topology in the halves comes from the sign of the determinant, which cuts the symmetric half into three pieces and the Hermitian half into two.
The article computes the dimensions and the contractibility of the four halves, studies the quadratic form given by the determinant on the symmetric and the Hermitian halves, describes the positive semidefinite cone and its boundary, and lists the strata of the halves by rank. It assumes The Topology of $M_2(\mathbb{C})$, The Two Halves over $\mathbb{R}$: Symmetric and Antisymmetric at $n=2$ and The Two Halves over $\mathbb{C}$: Hermitian and Anti-Hermitian at $n=2$.
It follows the biquaternion article Subspaces and Decompositions of the Biquaternion Algebra for the same subspaces in the eight real coordinates.
The Four Halves as Vector Spaces
The real halves have dimensions three and one, and the complex halves have complex dimension two each, hence real dimension four each.
Proposition. The four halves are $\operatorname{Sym}_2\cong\mathbb{R}^{3}$, $\operatorname{Skew}_2\cong\mathbb{R}$, $\operatorname{Herm}_2\cong\mathbb{R}^{4}$ and its conjugate $i\operatorname{Herm}_2$, and each is a linear subspace of the matrix algebra.
Proof. The symmetric half has three free real entries, the antisymmetric half one, and the Hermitian half two real diagonal entries and one complex off-diagonal entry.
Corollary. Each half is contractible, connected and simply connected, since it is a vector space.
Proof. Contracting along the rays to the origin gives the homotopy.
Corollary. The anti-Hermitian half is the image of the Hermitian half under multiplication by $i$, and it is a real subspace but not a complex one.
Proof. Multiplying a Hermitian matrix by $i$ gives an anti-Hermitian one, and the product of a complex scalar with an anti-Hermitian matrix is Hermitian again only for the real scalars.
The Trace Form on the Halves
The trace form is positive definite on the Hermitian half and on the symmetric half, and negative definite on the antisymmetric half.
Proposition. On the Hermitian half the trace form is two times the Euclidean form of the four real coordinates, and on the symmetric half its Gram matrix in the basis of the three matrix units has the eigenvalues one, one and two.
Proof. The trace of the product of two Hermitian matrices is the sum of the products of the entries with their conjugates, which is twice the sum of squares of the four real coordinates; on the symmetric half the off-diagonal basis element contributes twice.
Corollary. Each half is a Hilbert space with the trace form as inner product, up to a sign on the antisymmetric half.
Proof. The form is positive definite on three of the halves and negative definite on the fourth, and a sign change makes it an inner product there too.
Corollary. The orthogonal projection of the matrix algebra onto each half is taken with the trace form and is continuous.
Proof. The trace form is nondegenerate on each half, so the projection is a bounded operator.
The Symmetric Half and the Determinant
Theorem. On the symmetric half the determinant is a quadratic form of inertia $(1,2)$, so its positive set has two components and its negative set one.**
Proof. In the coordinates of the entries the determinant of $\begin{pmatrix}a&b\\b&c\end{pmatrix}$ is $ac-b^{2}$, whose Gram matrix has the eigenvalues $\tfrac12,-\tfrac12,-\tfrac12$.
Corollary. The definite matrices form two open convex cones, one positive definite and one negative definite, and no segment joining the two cones stays definite.
Proof. The set where the determinant is positive is two cones by the inertia, each is convex because a convex combination of two positive definite matrices is positive definite, and a segment between the two cones meets the singular set.
Corollary. The indefinite matrices form one connected set, which separates the two cones of definite matrices.
Proof. The set where the determinant is negative is connected by the same inertia computation, and its closure contains the origin and the rank-one cone.
The Positive Semidefinite Cone
Proposition. The positive semidefinite cone is closed and convex with nonempty interior, it is generated by the rank-one matrices $vv^{\mathsf T}$, and its boundary is the cone of rank-one positive semidefinite matrices.
Proof. A matrix is positive semidefinite exactly when it is a sum of rank-one terms, the semidefinite condition is closed under convex combinations, and a semidefinite matrix that is not definite has rank one.
Corollary. The cone is the cone over the closed disc in the two-plane of the traceless part, and its base is a disc rather than a sphere.
Proof. Subtracting the trace leaves a two-dimensional traceless part, and the semidefinite condition bounds it by the radius of a disc.
Corollary. The boundary of the cone is a single cone over a circle, and the interior is the union of the two definite cones.
Proof. The boundary consists of the rank-one matrices, parametrised by a direction and a scale, and the interior is the set of definite matrices.
The Antisymmetric Half
Proposition. The antisymmetric half is one-dimensional and contractible, and its nonzero elements are all of the form $a\,J$ with $J$ the rotation by a quarter turn.
Proof. An antisymmetric matrix is determined by one entry, and that entry multiplies the standard quarter turn.
Corollary. The set of nonzero elements of the antisymmetric half has two connected components, distinguished by the sign of the parameter.
Proof. The real line minus the origin has two components.
Corollary. The exponential maps the antisymmetric half onto the rotations, and the two signs of the parameter give the two signs of the angle.
Proof. The exponential of $aJ$ is the rotation by the angle $a$, computed in The Exponential of $\operatorname{Skew}_2$: the Single Rotation Plane.
The Hermitian Half and Its Cone
Proposition. On the Hermitian half the determinant is a quadratic form of inertia $(1,3)$, so its positive set is one cone and its negative set is connected.
Proof. In the coordinates of the two real diagonal entries and the two real components of the off-diagonal entry the determinant is $ac-|b|^{2}$, whose Gram matrix has the eigenvalues $\tfrac12,-\tfrac12,-\tfrac12,-\tfrac12$.
Corollary. The positive definite Hermitian matrices form a single convex cone of real dimension four, with boundary the rank-one Hermitian matrices.
Proof. The inertia gives a single positive cone, and the boundary consists of the semidefinite matrices of rank one.
Corollary. The cone is the cone over a three-dimensional ball, and its boundary is the cone over a two-sphere.
Proof. The trace normalisation cuts the cone in a ball, whose boundary is a sphere.
Corollary. The Hermitian cone is linearly isomorphic to the Lorentz cone of the four-dimensional space of inertia $(1,3)$.
Proof. The determinant is a form of inertia $(1,3)$ on the Hermitian half, and the cone of positive definiteness is the set where that form is positive and the trace is positive, which is the Lorentz cone of special relativity.
Summary
The four halves are vector spaces of real dimensions three, one, four and four, so all four are contractible. The trace form is positive definite on the symmetric and the Hermitian halves and negative definite on the antisymmetric one.
The determinant restricted to the symmetric half has inertia $(1,2)$ and cuts it into two definite cones and one indefinite region; restricted to the Hermitian half it has inertia $(1,3)$ and gives a single positive cone over a ball. The positive semidefinite cone is closed and convex, generated by the rank-one matrices, and its boundary is the rank-one cone.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\operatorname{Sym}_2\cong\mathbb{R}^{3}$ | the symmetric half |
| $\operatorname{Skew}_2\cong\mathbb{R}$ | the antisymmetric half |
| $\operatorname{Herm}_2\cong\mathbb{R}^{4}$ | the Hermitian half |
| $(1,2)$, $(1,3)$ | the inertias of the determinant on the two halves |
| $\det\ge 0$ | the semidefinite cone |
Further Reading
- Rajendra Bhatia, Positive Definite Matrices (Princeton University Press, 2007), for the cone of positive semidefinite matrices.
- James R. Munkres, Topology (Pearson, 2000), for the topology of the cones and their components.
- The Topology of $M_2(\mathbb{C})$, for the ambient topology.
- The Two Halves over $\mathbb{R}$: Symmetric and Antisymmetric at $n=2$, for the halves studied here.
- The Trace Form and Its Inertia on the Two Halves of $M_2(\mathbb{R})$, for the inertia of the trace form used here.