Spectrahedral Shadows in Convex Geometry

• Spectrahedral shadows are convex sets defined as the linear projection of spectrahedra formed by linear matrix inequalities.
• They play a pivotal role in semidefinite programming and convex algebraic geometry by enabling SDP representations and sum‐of‐squares certificates.
• Efficient computational techniques leverage their structure for optimization, control, and state estimation in various applied settings.

A spectrahedral shadow is a convex subset of Euclidean space defined as the linear image (projection) of a spectrahedron—a set cut out by a linear matrix inequality (LMI) in the symmetric (or Hermitian) cone. The concept generalizes both spectrahedra and polyhedra, and appears fundamentally in convex algebraic geometry, semidefinite programming, and numerous areas where semidefinite relaxations or exact representations of convex sets are central.

1. Definition and Basic Structure

Let $S_N$ denote the space of real symmetric $N \times N$ matrices, and $S_N^+$ its cone of positive semidefinite matrices. A spectrahedron is the feasible region of a linear matrix inequality (LMI):

$$S = \left\{ x \in \mathbb{R}^n : A_0 + x_1A_1 + \cdots + x_nA_n \succeq 0 \right\},$$

where $A_0, A_1, \ldots, A_n \in S_N$. A spectrahedral shadow is any set $K \subseteq \mathbb{R}^n$ satisfying

$$K = \left\{ x \in \mathbb{R}^n : \exists y \in \mathbb{R}^m,\, A_0 + \sum_{i=1}^n x_i A_i + \sum_{j=1}^m y_j B_j \succeq 0 \right\}$$

for some $A_0, A_i, B_j \in S_k$ for suitable $k$; equivalently, $K$ is the linear image under projection of the spectrahedron $\widetilde S \subset \mathbb{R}^{n+m}$ defined by the same LMI. This class includes polyhedra, ellipsoids, second‐order cones, and many notable convex bodies (Harrison, 2015, Scheiderer, 2016, Wang et al., 26 Feb 2025).

2. Relationship to Semidefinite Programming and Convex Algebraic Geometry

Spectrahedral shadows are exactly the sets that can be described as feasible regions of semidefinite programs with linear objective functions, after possibly projecting onto variables of interest. Semidefinite representability (SDP‐representability) merges real algebraic geometry (polynomial equations/inequalities) with convexity, and the class of spectrahedral shadows is substantially broader than spectrahedra themselves (Wang et al., 26 Feb 2025). Projections can yield highly nontrivial convex sets with algebraic boundary of degree much higher than the original LMI, and are ubiquitous in convex algebraic geometry (Sinn et al., 2014).

The moment‐relaxation methodology, sum‐of‐squares certificates, and duality theory are intrinsic to deciding semidefinite representability. In particular, the spectrahedral shadow property is equivalent to a uniform sum‐of‐squares lift for all linear functionals nonnegative on the set. This is formalized in (Scheiderer, 2016), which gives necessary and sufficient conditions for a convex hull of a semialgebraic set to be a spectrahedral shadow.

3. Scope, Examples, and Closure Properties

Numerous natural convex sets are spectrahedral shadows, sometimes directly, sometimes only after projection:

Class	Spectrahedron	Spectrahedral Shadow	Reference
Polyhedron	Yes	Yes	(Scheiderer, 2016)
Ellipsoid	Yes	Yes	(Scheiderer, 2016)
Second-order cone	Yes	Yes	(Scheiderer, 2016)
Hyperbolicity cone (smooth)	Varies	Yes (if smooth)	(Netzer et al., 2012)
General convex semialgebraic set	Varies	Not always	(Scheiderer, 2016)
Spectral polyhedra	Yes	Yes	(Sanyal et al., 2020)

Spectrahedral shadows are closed under numerous operations: linear images, Minkowski sums, intersections, Cartesian products, and convex hulls of finitely many such sets (Wang et al., 26 Feb 2025). Efficient representations of the resulting sets can often be constructed by direct manipulations of the underlying LMIs and projection variables.

Not all convex semialgebraic sets are spectrahedral shadows: explicit counterexamples (e.g., certain closures of Veronese images, high-dimensional copositive cones, and cones of nonnegative polynomials not representable as sums of squares) demonstrate proper containment of the class among all convex semialgebraic sets (Scheiderer, 2016, Bodirsky et al., 2022).

4. Helton–Nie Conjecture and Obstructions

The Helton–Nie Conjecture (HNC) posited that every convex semialgebraic set is a spectrahedral shadow. This was shown to be equivalent to the Pseudospectrahedron Conjecture (PSC), which states that the convex hull of the rank-one locus of any spectrahedron is itself a spectrahedral shadow (Harrison, 2015). However, Scheiderer (Scheiderer, 2016) and subsequent work (Bodirsky et al., 2022, Bettiol et al., 2019) produced explicit counterexamples, establishing that the class of spectrahedral shadows does not universally capture all convex semialgebraic geometry. These obstructions are fundamentally tied to the failure of sum-of-squares certificates in real algebraic geometry and can be traced to classical results (e.g., non-SOS positive polynomials and copositive cones for $n\geq 5$).

The table below summarizes critical distinctions:

Property	Spectrahedral Shadow	General Convex Semialgebraic Set
Projection of spectrahedron	Yes	No
Always admits sum-of-squares certificate	Yes	No
SDP-exact modeling possible	Yes	No

5. Algebraic and Boundary Structure

The algebraic boundary of a generic spectrahedral shadow is a union of irreducible real hypersurfaces corresponding to loci of fixed corank in the lifted matrix pencil. The degrees of these hypersurfaces are given by explicit formulas involving the algebraic degrees of semidefinite programming, as in the Pataki inequalities and the classifications in (Sinn et al., 2014). For instance, the boundary defining polynomial $\Phi_S(x)$ factors as a product $\prod_r f_r(x)$, with $f_r$ irreducible of prescribed degree.

These algebraic properties have practical significance in optimization and certificate design—explicit knowledge of the boundary equations can be used to analyze extreme points or develop rational separation or membership tests.

6. Computational and Structural Algorithms

Spectrahedral shadows support efficient algorithms for many set operations required in optimization and control, including exact analytical forms for linear maps, inverses, Minkowski sums, intersections, convex hulls, and cartesian products. Order reduction (size reduction of the underlying LMIs) and sparse representation are essential for scalability in state estimation and reachability computations (Wang et al., 26 Feb 2025). Approximations (polyhedral inner/outer) can be efficiently constructed using homogenization and recession cone analysis, transforming spectrahedral shadow approximation into finite sequences of small semidefinite programs (Dörfler et al., 2023, Dörfler et al., 2022).

Validation (emptiness, point containment, boundedness) is handled via carefully tailored SDPs based on the shadow representation, yielding computational tractability in many high-dimensional applications.

7. Applications, Limitations, and Open Questions

Spectrahedral shadows unify and generalize many constructs in semidefinite programming, convex optimization, and computational geometry. Notable applications include robust control, state estimation, nonlinear reachability, algebraic characterization of convex bodies, and relaxations of polynomial optimization problems (Wang et al., 26 Feb 2025, Sanyal et al., 2020, Netzer et al., 2012). They provide tractable frameworks for representing, manipulating, and validating complex convex sets, provided that the set in question admits a semidefinite representation.

Fundamental limitations remain: not all convex semialgebraic sets, hyperbolicity cones, or cones of nonnegative polynomials are spectrahedral shadows, and determining SDP-representability is generally nontrivial and depends on deep algebraic properties (e.g., SOS certificates, symmetry). The precise boundary between SDP-representable convex geometry and the strictly larger class of semialgebraic convex sets continues to be delineated through explicit obstructions and advanced duality technology (Scheiderer, 2016, Bodirsky et al., 2022, Bettiol et al., 2019).

Key open questions address minimal-dimensional counterexamples, improved geometric or effective SDP-representability criteria, boundary regularity conditions ensuring shadow character, and the full resolution of the generalized Lax conjecture for hyperbolicity cones (Scheiderer, 2016, Netzer et al., 2012).

---

For further technical development and comprehensive proofs, see (Harrison, 2015, Scheiderer, 2016, Bodirsky et al., 2022, Wang et al., 26 Feb 2025, Dörfler et al., 2023).