The Positive Cone in $\mathrm{Herm}_2$ and Its Boundary
Introduction
The positive semidefinite Hermitian matrices form a closed convex cone of real dimension four with no lines, self-dual for the trace form, whose trace hyperplane section is the Bloch ball and whose boundary is the cone over the two-sphere of pure states.
The article defines the cone, identifies it with the Lorentz cone, computes its section by the trace hyperplane, proves that the boundary consists exactly of the rank-one matrices, describes the extreme rays and the faces, and shows self-duality. It assumes The Two Halves and Their Topology at $n=2$, The Spectral Theorem: the Hermitian Case and the Sphere of Idempotents and The Trace Form and Its Inertia on the Two Halves of $M_2(\mathbb{R})$.
It follows the biquaternion article The Four Forms and the Four Algebraic Norms over the Biquaternion Algebra for the norm whose square is positive on the cone.
The Cone and Its Properties
The positive semidefinite matrices form a closed pointed convex cone with nonempty interior in the Hermitian half.
Definition. Let $C=\{A\in\operatorname{Herm}_2 : \langle v,Av\rangle\ge 0 \text{ for every } v\}$ be the positive semidefinite cone.
Proposition. The cone $C$ is closed, convex, pointed and has nonempty interior, and it is invariant under conjugation by unitary matrices.
Proof. The condition $\langle v,Av\rangle\ge 0$ is a closed condition for each $v$; a convex combination of two positive semidefinite forms is positive semidefinite; the cone contains no line because a matrix and its negative cannot both be nonzero and semidefinite; and the identity lies in the interior.
Corollary. The cone is generated by its rank-one elements, which are the matrices $vv^{*}$.
Proof. A positive semidefinite matrix is a sum of rank-one terms by the spectral theorem, and each term is a multiple of $vv^{*}$ for an eigenvector $v$.
Corollary. The cone has real dimension four, and its interior is the set of positive definite matrices.
Proof. The Hermitian half has dimension four, and the positive definite matrices form a nonempty open subset.
The Lorentz Form
Proposition. The determinant is a quadratic form of inertia $(1,3)$ on the Hermitian half, and the cone is the set where the determinant is positive and the trace is positive.
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 $\tfrac12,-\tfrac12,-\tfrac12,-\tfrac12$; the positivity of the form plus a positive trace is exactly the positive semidefinite condition.
Corollary. The cone is a Lorentz cone, isometric to the cone of future-directed timelike vectors of four-dimensional Minkowski space.
Proof. The map sending a Hermitian matrix to its four coordinates turns the determinant into the Minkowski form and the trace into the time coordinate.
Corollary. The cone is not polyhedral, because its extreme rays form a two-sphere rather than a finite set.
Proof. A polyhedral cone has finitely many extreme rays, and the extreme rays of the present cone are parametrised by a two-sphere, which is infinite.
The Trace Hyperplane and the Bloch Ball
Theorem. The section of the cone by the hyperplane of trace one is a closed ball of radius one in the three-dimensional space of traceless Hermitian matrices, and its boundary is the sphere of rank-one matrices.
Proof. A trace-one Hermitian matrix is $\tfrac12(I+\vec s\cdot\vec\sigma)$ with $\vec s$ in three-space, and the eigenvalues are $\tfrac12(1\pm|\vec s|)$; both are nonnegative exactly when $|\vec s|\le 1$. The boundary $|\vec s|=1$ is the set of matrices with one eigenvalue zero, which is the rank-one condition.
Corollary. The extreme points of the ball are the rank-one matrices, and they correspond to the points of the two-sphere.
Proof. The extreme points of a ball are its boundary sphere, and the boundary is the set of rank-one matrices.
Corollary. The centre of the ball is the maximally mixed matrix $\tfrac12 I$, whose two eigenvalues are equal.
Proof. The centre has $\vec s=0$, and the matrix is one half of the identity.
The Boundary
Proposition. The boundary of the cone is exactly the set of nonzero positive semidefinite matrices of rank one, which is the cone over the two-sphere.
Proof. A positive semidefinite matrix is on the boundary exactly when it is not positive definite, which for a semidefinite matrix of order two means determinant zero, hence rank one; and the rank-one matrices are $vv^{*}$, parametrised by a ray of the two-sphere.
Corollary. The boundary is a cone of real dimension three, and its section by the trace hyperplane is the two-sphere.
Proof. The cone over a two-dimensional sphere has dimension three, and the trace-one section is the sphere itself.
Corollary. The interior of the cone is the positive definite set, which is the open cone over the open ball, and it retracts onto the ball.
Proof. Every interior point can be scaled to trace one, and the scaling gives the retraction.
The Faces
Proposition. The faces of the cone are the cone itself, the boundary cone of rank-one matrices, the single ray of a rank-one matrix, and the origin.
Proof. A face is the intersection of the cone with a supporting hyperplane; the supporting hyperplanes at the boundary touch along the extreme rays, and the faces of a Lorentz cone are the cone, the boundary, the extreme rays and the apex.
Corollary. The extreme rays are the rays through the rank-one matrices, and each of them is a face of dimension one.
Proof. A ray is extreme when it cannot be written as a sum of two other rays, which holds exactly for the rank-one matrices.
Corollary. The cone has infinitely many extreme rays, in bijection with the points of the two-sphere.
Proof. The rank-one matrices are parametrised by a line in the plane up to phase, which is the two-sphere.
Self-Duality
Proposition. The cone is self-dual for the trace form: the matrices whose pairing with every element of the cone is nonnegative are exactly the elements of the cone.
Proof. The trace pairing of two Hermitian matrices is the Minkowski pairing of their four coordinates, and the Lorentz cone is self-dual for the Minkowski form.
Corollary. The dual cone of the positive semidefinite cone is itself, so the cone is the polar of itself with respect to the trace form.
Proof. Self-duality means that the polar cone equals the cone.
Corollary. The trace form is positive definite on the cone, and the pairing of two positive definite matrices is positive.
Proof. The trace of the product of two positive definite matrices is the sum of the eigenvalues of a positive definite matrix, which are positive.
Corollary. The automorphism group of the cone is the Lorentz group of the four coordinates, of real dimension six, and it contains the unitary conjugations as a subgroup of real dimension four.
Proof. A linear automorphism of the Lorentz cone preserves the Minkowski form up to the time orientation, which is the Lorentz group; the unitary conjugations preserve the determinant and act transitively on the boundary sphere.
Corollary. The cone is homogeneous: for any two interior points there is an automorphism carrying one to the other.
Proof. The Lorentz group acts transitively on the interior of the cone, and it preserves the cone by definition.
Summary
The positive semidefinite Hermitian matrices form a closed pointed convex cone of real dimension four, self-dual for the trace form and linearly isomorphic to the Lorentz cone of Minkowski space. Its trace-one section is the Bloch ball, a closed ball of radius one, centred at one half of the identity.
The boundary of the cone is the cone over the two-sphere of rank-one matrices, its extreme rays are those rank-one rays, and its faces are the cone, the boundary, the extreme rays and the apex. The interior is the positive definite set and retracts onto the open ball.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $C$ | the positive semidefinite cone in $\operatorname{Herm}_2$ |
| $vv^{*}$ | a rank-one positive semidefinite matrix |
| $\tfrac12(I+\vec s\cdot\vec\sigma)$ | a trace-one Hermitian matrix |
| $|\vec s|\le 1$ | the Bloch ball condition |
| $(1,3)$ | the inertia of the determinant on the half |
Further Reading
- Rajendra Bhatia, Positive Definite Matrices (Princeton University Press, 2007), for cones of positive matrices and their faces.
- Michael A. Nielsen and Isaac L. Chuang, Quantum Computation and Quantum Information (Cambridge University Press, 2010), for the Bloch ball of density matrices.
- The Two Halves and Their Topology at $n=2$, for the cone in its topological setting.
- The Spectral Theorem: the Hermitian Case and the Sphere of Idempotents, for the rank-one Hermitian matrices.
- The Trace Form and Its Inertia on the Two Halves of $M_2(\mathbb{R})$, for the pairing used in self-duality.