Hyperbolicity Cone: Structure & Applications

• Hyperbolicity cones are convex algebraic sets defined by hyperbolic polynomials with respect to a given direction, ensuring all perturbation roots are real and nonnegative.
• They generalize spectrahedral cones and are central to semidefinite programming research, exemplified by the Generalized Lax Conjecture and Renegar’s derivative framework.
• Advanced representations like spectrahedral shadows and second-order cone formulations make hyperbolicity cones computationally tractable for modern optimization methods.

A hyperbolicity cone is a prominent class of convex algebraic cones arising from the theory of hyperbolic polynomials, structuring deep connections between convex optimization, real algebraic geometry, and combinatorial analysis. For a homogeneous real polynomial $p$ of degree $d$ in $n$ variables and a direction $e$ with $p(e) > 0$, the hyperbolicity cone $\Lambda_+(p, e)$ is defined as the set of $x \in \mathbb{R}^n$ such that all zeros of $t \mapsto p(x + t e)$ are real and nonnegative. This class encompasses all spectrahedral cones, polynomial-time solvable via semidefinite programming (SDP), but is significantly richer, prompting major conjectures regarding their semidefinite representability and facial structure.

1. Definition and Foundational Properties

Given a homogeneous polynomial $p \in \mathbb{R}[x_1, \dots, x_n]$ of degree $d$ and a direction $d$0 with $d$1, $d$2 is hyperbolic with respect to $d$3 if, for all $d$4, $d$5 has only real zeros. The associated hyperbolicity cone is

$d$6

Alternately, it can be described as

$d$7

By Gårding’s theorem, $d$8 is always a closed convex cone and coincides with the closure of the connected component of $d$9 containing $n$0 (Lourenço et al., 2021).

The boundary of $n$1, denoted $n$2, encodes rich geometric and algebraic structure, often stratified by the multiplicity of root $n$3 in univariate slices of $n$4. The algebraic boundary is specified by vanishing of $n$5 and its Renegar derivatives $n$6, making the cone a basic semialgebraic set with a finite polynomial description (Naldi et al., 2016).

2. Spectrahedrality and the Generalized Lax Conjecture

Spectrahedral cones are sets of the form $n$7 for symmetric $n$8, and are precisely the feasible regions of SDPs. Every spectrahedral cone is a hyperbolicity cone of the determinant polynomial; conversely, the Generalized Lax Conjecture posits that every hyperbolicity cone is spectrahedral (Raghavendra et al., 2017). This is proved for $n$9 (Helton–Vinnikov theorem) and for specific families (e.g., elementary symmetric polynomials (Brändén, 2012, Amini, 2016), multivariate matching/interlacing polynomials).

However, no general proof is known for arbitrary $e$0 and $e$1. Notably, the specialized Vámos polynomial illustrates subtlety: it is hyperbolic, yet does not admit a definite determinantal representation in any positive power, but its hyperbolicity cone is still spectrahedral via augmentation by auxiliary factors (Kummer, 2013).

For smooth hyperbolicity cones (i.e., those whose boundary contains only nonsingular points outside the origin), Netzer and Sanyal proved representability as spectrahedral shadows—linear projections of spectrahedral cones—via local strict quasi-concavity and the Helton–Nie criterion (Netzer et al., 2012, Brändén, 2012). This result was recently sharpened to show that every Nash-smooth hyperbolicity cone is second-order cone representable (SOCR), i.e., it admits an explicit lift to block-diagonal LMIs with blocks of size $e$2 (Scheiderer, 21 Sep 2025).

Result Class	Property	Reference
All hyperbolicity cones	Closed, convex, semialgebraic	(Lourenço et al., 2021)
Smooth hyperbolicity cones	Spectrahedral shadow	(Netzer et al., 2012)
Nash-smooth boundary, pos. curv.	Second-order cone representable (SOCR)	(Scheiderer, 21 Sep 2025)
Determinantal/MMP cones	Spectrahedral	(Brändén, 2012, Amini, 2016)
General	Spectrahedrality open	(Raghavendra et al., 2017)

A key complexity-theoretic limitation is that certain explicit families of hyperbolicity cones require semidefinite representations of exponential size in the degree $e$3 even for approximate representations (Raghavendra et al., 2017).

3. Facial Structure, Amenability, and Automorphisms

Hyperbolicity cones are amenable: every face is exposed with a uniform error bound for projections, a property strictly stronger than being facially dual complete or facially exposed (Lourenço et al., 2021). Any face $e$4 of $e$5 is itself a hyperbolicity cone for a suitable Renegar derivative restricted to the span of $e$6, and intersections of hyperbolicity cones are again hyperbolicity cones.

The group of linear automorphisms of a hyperbolicity cone (those mappings $e$7 with $e$8) is critical in optimization and convex geometry. For rank-one generated (ROG) hyperbolicity cones—all extreme rays generated by rank-one elements—the automorphism groups of derivative relaxations can be described exactly: only those automorphisms fixing the hyperbolicity direction up to scale are admitted (Ito et al., 2022). For derivative relaxations of classic cones (e.g., positive orthant, PSD cone) the continuous automorphism group is extremely restricted, consisting only of scaling (plus orthogonal transformations in the PSD case).

4. Representability: Second-Order and Spectrahedral Shadows

The smoothness and curvature properties of the boundary play a decisive role in representability. If the boundary of a convex semialgebraic set $e$9 is Nash-smooth with everywhere strictly positive curvature, then $p(e) > 0$0 is SOCR (Scheiderer, 21 Sep 2025). This is established using tensor evaluation, showing that the semidefinite extension degree is at most two: for each extreme ray, positive tangents at boundary points result in sum-of-squares representations with at most two rank-one terms. The same techniques yield SOCP formulations for classes previously accessible only via general (and larger) SDP lifts, with concrete computational benefits.

Prominent examples include the PSD cone and the nonnegative orthant, which are hyperbolicity cones of $p(e) > 0$1 and $p(e) > 0$2, respectively, and are both SOCR (Scheiderer, 21 Sep 2025). Factorizable hyperbolic polynomials with smooth, irreducible factors yield SOCR cones by applying these results factorwise.

5. Relaxations: Derivative Cones and Central Swaths

Derivative cones—so-called Renegar relaxations—arise by taking successive directional derivatives of a hyperbolic polynomial along the hyperbolicity direction. Each $p(e) > 0$3-th derivative $p(e) > 0$4 remains hyperbolic, and the associated cones $p(e) > 0$5 satisfy the nested chain

$p(e) > 0$6

This hierarchy yields increasingly relaxed convex sets. In convex programming, these derivative cones underpin the generalized "central swaths" paradigm, a family of interior-point methods for optimization over $p(e) > 0$7, interpolating between classical central paths for LP/SDP and more general hyperbolic programs (Renegar, 2010). For ROG cones, all faces and automorphisms of these relaxations are explicitly described (Ito et al., 2022).

6. Algorithmic and Analytical Aspects

Hyperbolicity cones support efficient first-order and second-order optimization techniques. The barrier functions induced by hyperbolic polynomials, including logarithmic barriers, enjoy self-concordance and strict convexity in the cone interior for many $p(e) > 0$8 (notably determinants and elementary symmetric polynomials) (Nagano et al., 2024). Projection onto general hyperbolicity cones and optimization over them can be performed numerically using dual Frank–Wolfe schemes, leveraging the generalized eigenvalue structure for closed-form subproblem solutions (Nagano et al., 2024).

Symbolic computation approaches enable exact solution of hyperbolic programs by exploiting the stratification of the algebraic boundary and reducing to real algebraic equation solving within multiplicity strata. This approach is independent of determinantal representability (Naldi et al., 2016).

7. Applications, Extensions, and Open Questions

Hyperbolicity cones are instrumental in hyperbolic programming, a strict generalization of semidefinite programming, enabling interior-point methods for a broader array of combinatorial and real algebraic optimization problems (Raghavendra et al., 2017). Key applications include combinatorics (matching and independence polynomials), matrix theory, and convex geometry. The framework of $p(e) > 0$9-Lorentzian polynomials further extends hyperbolicity cones to dynamics and semipositivity theory (Dey, 24 Dec 2025).

Central open problems include:

• Full resolution of the Generalized Lax Conjecture in all dimensions and for singular cones.
• Constructive, dimension-efficient spectrahedral or SOCR representations for larger subclasses of hyperbolicity cones.
• Understanding the expressive gap between hyperbolic programming and SDP in high-degree regimes.

The study of hyperbolicity cones continues to drive advances in optimization theory, convex geometry, and real algebraic geometry, with current research closing representation gaps and sharpening computational techniques (Scheiderer, 21 Sep 2025, Netzer et al., 2012, Raghavendra et al., 2017, Nagano et al., 2024).