Equiangular Model: Unified Analysis

• Equiangular models are sets of lines through the origin that maintain a constant angle between any pair, serving as a key concept in discrete geometry.
• The framework employs algebraic representations like Seidel matrices and advanced semidefinite programming to derive tight bounds and existence conditions.
• Explicit constructions using lattices and strongly regular graphs illustrate applications in spherical codes, quasi-symmetric designs, and combinatorial optimization.

A set of lines through the origin in a Euclidean or complex vector space is called equiangular if any two distinct lines form the same angle. The equiangular model provides a unified algebraic, combinatorial, and geometric framework for studying maximal sets of equiangular lines, their cardinality as a function of dimension, and their deep connections to spherical codes and combinatorial design theory. In this context, the model subsumes both synthetic constructions (via lattices, special graphs, and algebraic structures) and analytic upper bounds (via semidefinite programming and spectral constraints), with definitive results in significant ranges of dimensions.

1. Definitions and Fundamental Bounds

Let $\ell_1, \ldots, \ell_m$ be lines through the origin in $\mathbb{R}^n$, and $u_i$ a unit vector on $\ell_i$. The set is equiangular if there exists $\alpha \in (0,1)$ such that $\langle u_i, u_j \rangle = \pm \alpha$ for all $i \neq j$, or equivalently, the angle between distinct lines is $\theta = \arccos \alpha$ (Barg et al., 2013). The Gram matrix $G$ of such a configuration has diagonal $1$, off-diagonal $|\langle u_i, u_j \rangle| = \alpha$. For $\mathbb{R}^n$, let $M(n)$ denote the maximal cardinality of an equiangular set, and $M_\alpha(n)$ the maximal cardinality for a fixed $\alpha$.

Two canonical upper bounds control the extremal size:

• Absolute (Gerzon) bound: $M(n) \leq n(n+1)/2$ (Barg et al., 2013).
• Relative bound: For $\alpha$ with $1 - n\alpha^2 > 0$, $M_\alpha(n) \leq n(1-\alpha^2)/(1-n\alpha^2)$ (Barg et al., 2013).

Neumann’s theorem asserts that if $M(n) > 2n$ for fixed $\alpha$, then $1/\alpha$ is an odd integer.

2. Algebraic Encodings: Seidel Matrices and Switching Classes

Every equiangular line system in $\mathbb{R}^n$ with representatives $u_i$ can be encoded by its Gram matrix $G$, or equivalently, by its Seidel matrix $S = (G - I)/\alpha$—a symmetric $\{0,\pm 1\}$ matrix with zero diagonal (Greaves, 2016, Greaves et al., 2014). The spectral properties of $S$ are tightly controlled: if $S$ arises from $N$ equiangular lines in $\mathbb{R}^d$ with common angle $\alpha$, then $S$ has smallest eigenvalue $\lambda_3 = -1/\alpha$ with multiplicity $m_3 = N-d$, and satisfies trace identities: $$\operatorname{tr} S = 0,\quad \operatorname{tr} S^2 = N(N-1).$$ Classification of such matrices, especially with three distinct eigenvalues, yields necessary and sufficient conditions for the existence of a regular graph in the switching class (Greaves, 2016), and enables enumeration via congruence conditions (mod $2$, $4$).

The switching class perspective is essential for both existence and non-existence results, particularly in organizing the extremal examples via strongly regular graphs and two-graphs.

3. Semidefinite Programming and Optimal Bounds

Contemporary progress in upper bounds for $M_\alpha(n)$ exploits semidefinite programming (SDP). The Bachoc–Vallentin framework encodes the spherical code problem as an SDP in auxiliary variables $x_1, \ldots, x_6$, seeking positivity of matrix-valued kernels built from Gegenbauer polynomials. The SDP constraints, after reduction, yield tight bounds for inner product configurations and enable numeric and symbolic determination of $M_\alpha(n)$ (Barg et al., 2013, Yu, 2016).

For dimensions $24 \leq n \leq 41$, numerical solution at $p=5$ shows $M(n)=276$; for $n=43$, $M(43)=344$. For $n$ up to $136$, the SDP strictly improves the Gerzon bound. Symbolic relaxations further yield

$M_{1/a}(n) \leq \frac{1}{2}(a^2-2)(a^2-1)$

for $n \in (a^2-2,\, 3a^2-16)$, $a \geq 3$ (Yu, 2016, Kao et al., 2022). Four-point SDP constraints refine these results and establish uniqueness for $M(n)$ in certain parameter regimes.

4. Explicit Constructions and Saturation Phenomena

Maximal and near-maximal equiangular line sets arise from a variety of constructions:

• Lattices: The Leech lattice in $\mathbb{R}^{24}$ yields $276$ lines in $\mathbb{R}^{23}$, corresponding to a regular two-graph stabilized by Conway’s group $\mathrm{Co}_3$ (Gillespie, 2018).
• Strongly Regular Graphs: Block and two-graph constructions produce $344$ lines in $\mathbb{R}^{43}$ at angle $1/7$ (Barg et al., 2013).
• Graph saturation methods: For $14 \leq d \leq 20$ (excluding $d=15$), no larger equiangular set can be formed by extending the current known constructions; saturation is certified via clique analysis in the switching graph (Lin et al., 2018).
• Lower bounds via design theory: Infinite families with $N(d) \geq c d^2$ for $c>0$ are constructed by design-theoretic methods (cf. de Caen’s method) (Greaves et al., 2014).

Examples with three distinct eigenvalues in the Seidel matrix yield all feasible regular graphs, with arithmetic conditions on eigenvalues and multiplicities (Greaves, 2016).

5. Connections to Geometry, Designs, and Spherical Codes

Equiangular line systems are deeply linked to:

• Spherical $t$-designs: Tight spherical 5-designs only exist for $d=2,3,7,23$, corresponding exactly to observed maximal equiangular line sets saturating Gerzon’s bound (Gillespie, 2018);
• Quasi-symmetric block designs: The structure of maximal incoherent subsets (those whose triple products are positive) ties equiangular sets to quasi-symmetric designs, and saturation of relative and incoherence bounds is equivalent to particular combinatorial designs (Gillespie, 2018);
• Root lattices and exceptional curves: Intersection patterns in root systems ($E_8$, Leech lattice) correspond to equiangular line sets and their hierarchical blowup in del Pezzo surfaces (Gillespie, 2018).
• Spherical codes: Generalization to spherical $L$-codes with $k$ positive angles yields tight $O(n^k)$ bounds for maximal size, with Ramsey-type reductions and projections (Balla et al., 2016).

6. Key Open Problems, Asymptotics, and Uniqueness

Fundamental questions in the equiangular model remain open:

• Determining the exact asymptotic growth $\limsup M(n)/n^2$ for large $n$—while de Caen’s construction provides $M(n) \geq 2/9 n^2$ for infinitely many $n$, SDP and linear programming bounds yield lower than quadratic rates for certain restricted angles (Barg et al., 2013).
• The “stable ranges” for fixed angles $\alpha=1/(2k-1)$, where $M_\alpha(n)$ appears constant for intervals of $n$, present intriguing parametric phenomena (Barg et al., 2013).
• Uniqueness of maximal sets: Four-point SDP shows uniqueness of the $28$-line system in $7 \leq d \leq 14$ for $\alpha=1/3$ and of the $276$-line system in $23 \leq d \leq 64$ for $\alpha=1/5$ (Kao et al., 2022).
• Exact determination for dimensions where Gerzon’s bound is tight but constructions are elusive (e.g., $n=42$, certain $n=47,79,119$ related to tight spherical 4-designs) (Barg et al., 2013).

Additional future directions include extending SDP hierarchies, refining combinatorial design approaches, and clarifying the deep algebraic structure underlying equiangular configurations.

---

The equiangular model synthesizes analytic bounds (via SDP and spectral theory), combinatorial design constructions (via graphs and lattices), and geometric characterizations (via spherical codes and designs) to transitively organize maximal equiangular sets and chart the frontier of discrete geometry concerning line configurations in high dimensions.