Compact Spectrahedron Fundamentals

• Compact spectrahedron is a convex set defined as the solution of a linear matrix inequality that is both bounded and closed.
• They underpin key applications in optimization, quantum information, and noncommutative geometry by providing well-structured duality and approximation frameworks.
• Their detailed structure, including free extreme points and facial properties, enables explicit LMI representations and efficient algorithmic volume computations.

A compact spectrahedron is a convex set defined as the solution set to a linear matrix inequality (LMI) with the additional property of compactness—i.e., boundedness and closedness. Such sets generalize classical convex compact bodies, and their operator-theoretic and algebraic structure underpins diverse areas such as optimization, quantum information, and noncommutative geometry. Compactness for spectrahedra is both a practical and theoretical constraint, crucial for understanding their extremal structure, duality, and approximation properties.

1. Definitions and Fundamental Structure

Let $\mathbb{F} = \mathbb{R}$ or $\mathbb{C}$. A classical (finite-dimensional) spectrahedron is

$$S = \{ x \in \mathbb{R}^n : L(x) = A_0 + x_1 A_1 + \cdots + x_n A_n \succeq 0 \}$$

where $A_i \in \mathrm{Sym}_r(\mathbb{R})$ (or Hermitian matrices in the complex case), and $\succeq 0$ denotes positive semidefiniteness. Such $S$ is always closed and convex. The set is called compact if and only if it is bounded, i.e., $S$ does not extend to infinity in any direction [$1611.05658$].

In the noncommutative (free or matrix) setting, a (free) spectrahedron is a sequence of sets $D_A(n)$ defined for each $n \geq 1$: $$D_A(n) = \{ X = (X_1, \ldots, X_g) \in SM_n(\mathbb{F})^g : L_A(X) = I_d \otimes I_n + \sum_{j=1}^g A_j \otimes X_j \succeq 0 \}$$ and the full free spectrahedron is $D_A = \bigsqcup_{n=1}^\infty D_A(n)$. Here, $A_j$ are $d \times d$ Hermitian, $SM_n(\mathbb{F})$ denotes self-adjoint matrices, and $L_A$ is the monic linear pencil [$1806.09053$].

A generalized free spectrahedron with compact coefficients is defined similarly, but with $A_j$ compact self-adjoint operators on (possibly infinite-dimensional) Hilbert space $\mathcal{H}$ [$2302.07382$]: $$D_A(n) = \{ X \in M_n(\mathbb{F})^g : L_A(X) := A_0 \otimes I_n + \sum_{j=1}^g A_j \otimes X_j \succeq 0 \}$$ where $A_j \in SA(\mathcal{H})$ are compact.

In both commutative and noncommutative cases, compactness requires that boundedness holds at all matrix levels.

2. Compactness Criteria and Spectrahedral Shadows

A spectrahedron is compact precisely when the corresponding LMI does not admit unbounded feasible directions, which can be concisely formulated as follows: for the homogeneous pencil $\Lambda_A(x) = A_1 x_1 + \cdots + A_g x_g$, there is no nonzero $x \in \mathbb{R}^g$ such that $\Lambda_A(x) \succeq 0$ [$1806.09053$]. Equivalently, the only solution to $A_1 x_1 + \cdots + A_g x_g \succeq 0$ is $x=0$.

More generally, compact convex basic closed semialgebraic sets $$S = \{x \in \mathbb{R}^n: g_i(x) \geq 0, i=1,\dots, m\}$$ that satisfy the Archimedean condition (i.e., some quadratic module $M(g)$ contains $N - \sum X_i^2$ for some $N$) admit semidefinite (SDP) representations as spectrahedral shadows; these representations are also compact by construction [$1704.07231$].

3. Extreme Points, Free-Minkowski Theorem, and Facial Structure

The extremal structure of compact spectrahedra is central to their convexity theory. In the finite-dimensional case, every element of a compact convex set can be decomposed as a convex combination of its extreme points (Minkowski theorem). In the matrix convex (free) setting, there are multiple notions of extreme points.

• Free extreme points are those $X$ that cannot be written as proper matrix convex combinations except in trivial unitary or dilation extensions (definitions in [$1806.09053$,$2302.07382$]). Free extreme points coincide with irreducible Arveson extreme points—those admitting no nontrivial dilation within the free spectrahedron.
• The free-Minkowski theorem for compact free spectrahedra states that every element is a matrix convex combination of free extreme points. Explicitly, for $X \in D_A(n)$ there exists a dilation to an Arveson extreme of size at most $n g$ (over $\mathbb{R}$) or $2 n g + n$ (over $\mathbb{C}$), and then compression yields a finite matrix convex combination. Hence,

$$\mathrm{co}^{\mathrm{mat}}(\mathrm{FreeExt}(D_A)) = D_A.$$

This result extends to generalized free spectrahedra with compact operator coefficients [$2302.07382$].

• In the context of polar orbitopes, compactness and explicit spectrahedral (LMI) representations are established. The facial structure is linked via momentum polytopes—faces correspond bijectively to those of the momentum polytope, and all faces are exposed [$1611.05658$,$2010.02045$].

4. Explicit Representations and Size Lower Bounds

The explicit LMI defining a compact spectrahedron depends on the structure of the data matrices. For polar orbitopes—convex hulls of orbits under compact Lie group actions—one constructs a block-diagonal matrix built from fundamental representation data: $$\mathscr{O}_x = \{ H \in \mathfrak{p}: \pi_j(H) \preceq c_j I_{V_j} \ \forall j \}$$ where each $\pi_j$ is a fundamental highest-weight representation suited to the action, and $c_j$ depends on $x$.

The total LMI size is given by $D = \sum_j \dim V_j$ and is independent of the point $x$ aside from its Weyl orbit. For special cases, block diagonalization leads directly to inequalities involving singular values or sums thereof (Ky Fan norms) [$2010.02045$].

Lower bounds on the LMI size for compact spectrahedra are established for sets defined by quadratic or cubic polynomials:

• The $n$-dimensional unit ball as a spectrahedron requires $r \geq n/2$; when $n = 2^k+1$, $r = n$ is necessary and sufficient.
• For convex bodies defined by a cubic polynomial in $\mathbb{R}^3$, the LMI must have size at least $r=5$ if the boundary is smooth [$1506.07699$].

5. Volume, Algorithmic Approximation, and Applications

The volume computation of compact spectrahedra is tractable by convex optimization, relying on determinant maximization: $\max_{P \in S} \frac{N+1}{2} \log \det P$ for $S$ an affine section of the PSD cone. The unique maximizer is the analytic center. Under well-conditioned assumptions, the volume admits a closed-form asymptotic expression: $$\mathrm{vol}(S) \approx \left(\frac{N+1}{4\pi}\right)^{m/2} e^{\phi(P^\star)}$$ with $m$ the number of affine constraints. For sequences of spectrahedra where $m_n^3 \log N_n / N_n \to 0$, the approximation becomes exact [$2211.12541$].

Specific applications include:

• Multi-way Birkhoff spectrahedra, encoding maximally entangled quantum states with uniform marginals;
• Central sections of spectraplexes (density matrices with trace and moment constraints) possessing asymptotically identical volumes.

Deterministic algorithms for such volume computations are based solely on convex optimization and are guaranteed to be asymptotically exact in high dimensions under mild growth conditions.

6. Relations to Lasserre Relaxations and Semidefinite Shadows

For general compact convex semialgebraic sets, the Lasserre relaxation hierarchy provides semidefinite representations: the $d$\textsuperscript{th} level moments and localizing matrices yield an explicit lifted LMI. Under strict quasiconcavity or SOS-concavity conditions on the defining polynomials and the Archimedean property, these relaxations are exact at some finite $d^\ast$: $$S(g) = \left\{ x \in \mathbb{R}^n \mid \exists y:\ M_d(y)\succeq 0,\ M_{d-d_i}(g_i y)\succeq 0, y_{\mathbf{e}_j}=x_j \right\}$$ This gives a practical route for constructing explicit compact spectrahedral shadows for broad classes of convex bodies [$1704.07231$].

7. Illustrative Examples and Special Constructions

Notable classes of compact spectrahedra include:

• Rank-one coefficient pencils in infinite dimension produce free half-spaces that, despite infinite-dimensionality, are spanned by extreme points at minimal levels [$2302.07382$].
• Diagonal compact operators lead to intersections of classical spectrahedra along eigendirections.
• Polar orbitopes from group actions include sets defined by majorization or Ky Fan norm inequalities, with direct LMI realizations.
• Spectrahedra arising from single quadratic inequalities (unit balls, ellipsoids) highlight minimal size constraints and the general tightness of lower bounds.

Such examples clarify the structure, algebraic constraints, and function-analytical richness of compact spectrahedra, emphasizing their foundational role in matrix convexity, optimization, and representation theory.