Eigenvalue multiplicity
and equiangular lines

Zilin Jiang
Arizona State University
March 27, 2026

Equiangular lines in $\mathbb{R}^n$ are lines through origin
pairwise separated by the same angle

What's the maximum number $E(n)$ of
equiangular lines in $\mathbb{R}^n$?

$n$
22
3-43-4
55
66
77
8-148-14
1515
1616
1717
1818
max
33
66
1010
1616
2828
28-3628-3628-3628-3628
3636
40-12640-12640-12640-4940
48-12648-12648-12648-4948
48-12648-12648-12648-12657-7457-59
1919
2020
2121
2222
2323
24-4124-41
4242
4343
72-12672-12672-12672-12672-74
90-12690-12690-12690-12690-94
126126
176176
276276
276-861276-861276
276-903276-903276-288
344-946344-946344
  • Classically, $E(2)=3$.
  • In 1948, Haantjes determined $E(3)=6$ and $E(4)=6$.
  • In 1966, van Lint and Seidel determined $E(5)=10$, $E(6)=16$, and $E(7)=28$.
  • In 1973, Gerzon's absolute bound gave $E(n)\le n(n+1)/2$; Lemmens and Seidel determined $E(15)=36$, $E(21)=126$, $E(22)=176$, and $E(23)=276$, which by monotonicity also sharpened upper bounds below dimensions 15 and 21; they also reported a number of lower bounds.
  • From 2014 to 2016, Barg and Yu proved $E(n)=276$ for $24 \le n \le 41$ and $E(43)=344$, leaving $276 \le E(42) \le 288$; Greaves, Koolen, Munemasa, and Szöllősi also improved the upper bounds on $E(14)$ and $E(16)$.
  • In 2019, Greaves and Yatsyna improved the upper bound to $E(17)\le 49$, hence also $E(16)\le 49$.
  • In 2021 and 2023, Greaves, Syatriadi, and Yatsyna proved $E(14)=28$, $E(16)=40$, and $E(17)=48$, and they improved the bounds to $E(18)\ge 57$, $E(19)\le 74$, and $E(20)\le 94$; in particular, $E(18)\le 74$ by monotonicity.
  • In 2024, Greaves and Syatriadi improved the upper bound to $E(18)\le 59$.

What's the maximum number $E(n)$ of equiangular lines in $\mathbb{R}^n$?

Gerzon 1973

At most $\frac{1}{2}n(n+1)$

de Caen 2000

At least $cn^2$

Angles $\to$ 90° as $n\to\infty$

What happens if the angles are held fixed?

What's the maximum number $E_\alpha(n)$ of
equiangular lines in $\mathbb{R}^n$ with a fixed angle $\arccos\alpha$?

Lemmens and
Seidel 1973
$E_{1/3}(n) = 2,4,6,10,16$
for $n \in \{2, \dots, 6\}$, and
$\max(28, 2(n-1))$ for $n \ge 7$.
Neumann 1973$E_{\alpha}(n) \le 2n$, unless $1/\alpha$ is odd
Neumaier 1989$E_{1/5}(n) = \lfloor \frac{3}{2}(n-1) \rfloor$ for $n \ge n_0$
Cao, Koolen, Lin, and Yu 2022$E_{1/5}(n) = \max(276, \lfloor 3(n-1)/2\rfloor)$ for $n \ge 23$

What's the maximum number $E_{1/5}(n)$ of
equiangular lines in $\mathbb{R}^n$ with a fixed angle $\arccos(1/5)$?

$n$2345678910111213
max234679101216182026
14151617181920212223-41
2836404857-5972-7490-94126176276

Known results on $E(d)$ and $E_{1/5}(n)$ agree for $n \in \{14, \dots , 41\}$

Lemmens–Seidel 1973 for $d \le 48$
$E(n) \le \max(E_{1/3}(d), N_{1/5}(n), \lfloor \frac{48d}{49-d}\rfloor)$

What's the maximum number $E_\alpha(n)$ of equiangular lines
in $\mathbb{R}^n$ with a fixed angle $\arccos\alpha$ in high dimensions?

Neumaier 1989$E_{1/5}(n) = \lfloor \frac{3}{2}(n-1) \rfloor$ for $n \ge n_0$
Bukh 2016$E_{\alpha}(n) \le c_\alpha n$
Balla, Dräxler,
Sudakov, Keevash 2018
$E_{\alpha}(n) \le 1.93n$
for $n \ge n_0(\alpha)$ if $\alpha \neq 1/3$

Bukh's conjecture on equiangular lines with fixed angle

$E_{1/(2k-1)}(n) \approx \frac{kn}{k-1}$. $E_{1/7}(n) \approx \frac{4}{3}n$.

What's the maximum number $E_\alpha(n)$ of equiangular lines
in $\mathbb{R}^n$ with a fixed angle $\arccos\alpha$ in high dimensions?

Conjecture J.–Polyanskii

$E_{\alpha}(n) \approx \frac{kn}{k-1}$, where $k = k(\lambda)$, $\lambda = \frac{1-\alpha}{2\alpha}$.

Spectral radius order $k(\lambda) := $ smallest $k$ such that
$\exists$ $k$-vertex graph $G$ whose adjacency matrix has spectral radius $\lambda$

$\alpha$$\lambda$$G$$k$$E_\alpha(n)$
$\tfrac{1}{3}$$1$$2$$2n$
$\tfrac{1}{1+2\sqrt2}$$\sqrt 2$$3$$\tfrac{3n}{2}$
$\tfrac{1}{5}$$2$$3$$\tfrac{3n}{2}$
$\frac{1}{7}$$3$$4$$\tfrac{4n}{3}$

J.–Polyanskii 2018

True for all $\lambda \le \sqrt{2 + \sqrt{5}}$.

2.058 barrier

What's the maximum number $E_\alpha(n)$ of equiangular lines
in $\mathbb{R}^n$ with a fixed angle $\arccos\alpha$ in high dimensions?

me
Jonathan Tidor
Yufei Zhao
Yuan Yao
Shengtong Zhang

J., Tidor, Yao, Zhang, Zhao 2021 Annals of Mathematics

$E_\alpha(n) = \lfloor \frac{k}{k-1}(n-1) \rfloor$ for $n \ge n_0(\alpha)$ if $k(\lambda) < \infty$;
$E_\alpha(n) = n+o(n)$ otherwise.

J., Tidor, Yao, Zhang, Zhao 2021 Annals of Mathematics

$E_\alpha(n) = \lfloor \frac{k}{k-1}(n-1) \rfloor$ for $n \ge n_0(\alpha)$ if $k(\lambda) < \infty$;
$E_\alpha(n) = n+o(n)$ otherwise.

Spectral radius order $k(\lambda) := $ smallest $k$ such that
$\exists$ $k$-vertex graph $G$ whose adjacency matrix has spectral radius $\lambda$

  1. When $k < \infty$, proof needs $n \ge 2^{2^{C\lambda k}}$. Balla showed $n \ge 2^{k^{Ck}}$; Balla and Bucić showed $n \ge 2^{k^C}$ for $\alpha = 1/(2k-1)$.
  2. When $k = \infty$, $o(n) = O_{\alpha}(n/\log\log n)$.
    Schildkraut constructed $\Omega_{\alpha}(\log\log n)$.

Connection to Spectral Graph Theory

Equiangular lines in $\mathbb{R}^n$ with angle $\arccos\alpha$
$\Leftrightarrow$ Unit vectors in $\mathbb{R}^n$ with inner product $\pm \alpha$
$\Leftarrow$$\Rightarrow$ Graph $G$, where $u \sim v$ in $G$ iff $\langle u, v\rangle = -\alpha$
s.t. $(1-\alpha) I - 2\alpha A_G + \alpha J \succeq 0$ and rank $\le n$

Equivalently, given $\alpha$ and $n$, find $G$ with max order s.t.
$(1-\alpha) I - 2\alpha A_G + \alpha J \succeq 0$ and rank $\le n$

Example when $\alpha = 1/5$, take $\lfloor (n-1)/2 \rfloor$ copies of

Upper bound on $E_\alpha(n)$

Given $\alpha$ and $n$, find $G$ with max order s.t.
$M := (1-\alpha) I - 2\alpha A_G + \alpha J \succeq 0$ and rank $\le n$

Rank-nullity
$E_\alpha(n) \le n + \mathrm{null}(M)$$\le \mathrm{null}((1-\alpha) I - 2\alpha A_G) + n + 1$

Since $M \succeq 0$, $\lambda := \frac{1-\alpha}{2\alpha}$ could be

  1. largest eigenvalue of $G$ (equality case)
  2. second largest eigenvalue of connected $G$ (need to rule out)

$\lambda := \frac{1-\alpha}{2\alpha}$ could be second largest eigenvalue of connected $G$

Switching $v$ to $-v$ in $\mathbb{R}^n$

Balla, Dräxler, Keevash, Sudakov 2018

Can switch so that max degree of $G$ $\le \Delta$


J., Tidor, Yao, Zhang, Zhao 2021

Multiplicity of second largest eigenvalue of an $n$-vertex connected graph of bounded degree is at most $Cn/\log\log n$.

Sublinear Second Eigenvalue Multiplicity

J., Tidor, Yao, Zhang, Zhao 2021

Multiplicity of second largest eigenvalue of an $n$-vertex connected graph of bounded degree is at most $Cn/\log\log n$.

Haiman, Schildkraut, Zhang, Zhao 2022: constructed second eigenvalue multiplicity $\ge C\sqrt{n/\log_2 n}$; Cayley $\ge n^{2/5}$

Proof Sketch

Multiplicity of second largest eigenvalue $\lambda$ of an $n$-vertex connected graph $G$ of bounded degree is at most $Cn/\log\log n$.

$r = c\log\log n$ and $s = c \log n$

  • $H$ is $G$ with small $r$-net removed
  • Radius $s$ balls in $H$ have spectral radius $\le (\lambda^{2r} - 1)^{1/2r}$
    Vertex removal lowers spectral radius
  • Bound $\mathrm{mult}(\lambda, H)$ via moments
  • $\mathrm{mult}(\lambda, G) \le \mathrm{mult}(\lambda, H) + \lvert r\text{-net} \rvert = o(n)$
    by Cauchy interlacing

What is the maximum second eigenvalue multiplicity of
an $n$-vertex connected graph of bounded degree?

McKenzie, Rasmussen, Srivastava 2021
For regular graphs, $O(n/\log^c n)$

What is the maximum second eigenvalue multiplicity of
an $n$-vertex Cayley graph of bounded degree?

Gromov, Colding–Minicozzi, Kleiner
Lee–Makarychev: Constant for abelian groups

What's the maximum number $E_\alpha(n)$ of
equiangular lines in $\mathbb{R}^n$ with a fixed angle $\arccos\alpha$?

$\alpha$$\lambda$$G$$k$$E_\alpha(n)$
$\tfrac{1}{3}$$1$$2$$2n$
$\tfrac{1}{1+2\sqrt2}$$\sqrt 2$$3$$\tfrac{3n}{2}$
$\tfrac{1}{5}$$2$$3$$\tfrac{3n}{2}$

Gossett, J., Teets, Wellner 2026 for $\alpha^* = \tfrac{1}{1+2\sqrt2}$

$n$234681014151617182022
max1347101319242424252831

$E_{\alpha^*}(n) = \max(24, \lfloor 3(n-1)/2 \rfloor)$ for $n \ge 23$

Strongest possible sense

Classify equiangular line systems with fixed angle
$\arccos(1/(1+2\sqrt2))$ in $\mathbb{R}^n$ for all $n$,
up to orthogonal transformations

Generic construction: take vertex-disjoint union of

Classification: when $n \ge 29$, only generic equiangular line systems

For which $\alpha$, are there only finitely many non-generic equiangular line systems?

Spherical Two-distance Sets

$E_{\alpha, \beta}(n) = $ maximum number of unit vectors in $\mathbb{R}^n$
such that pairwise inner products are $\beta < 0 < \alpha$

Bukh 2005: $E_{\alpha, \beta}(n) = O_{\alpha,\beta}(n)$

Problem Determine $\lim_{n\to\infty} E_{\alpha, \beta}(n) /n$

Conjecture Depend on $k_p(\lambda)$ in terms of signed graphs
where $p = \lfloor -\alpha/\beta\rfloor+1$ and $\lambda = (1-\alpha)/(\alpha - \beta)$

Solved in special cases

  • J., Tidor, Yao, Zhang, Zhao 2023: $p \le 2$ or $\lambda = 1,\sqrt{2},\sqrt{3}$
  • J., Polyanskii 2025: $\lambda < \lambda^* \approx 2.0198$
    $\lambda^* := \sqrt{\rho} + 1/\sqrt{\rho}$, where $\rho$ is root of $x^3 = x + 1$

Solution Framework

  1. Local forbidden subgraphs: leverage Gram matrix $\succeq 0$
  2. Global graph structure: Ramsey theory
    • Equiangular lines: bounded degree graph
    • Spherical two-distance sets:
      complete $p$-partite XOR bounded degree

Need new insights in spectral graph theory:

  • Sublinear eigenvalue multiplicity is false for signed graphs
  • J., Polyanskii 2025: local forbidden subgraphs alone can never break $\lambda^*$ barrier
  • J., Wang 2026: local forbidden subgraphs + global graph structure can never break $\lambda^*$ barrier as soon as $p \ge 3$

Other "Equiangular Lines"

Zauner's conjecture 1999

Maximum number of equiangular lines in $\mathbb{C}^n$ is $n^2$

Fixed angle: determine $\lim_{n\to\infty} E_\alpha^\mathbb{C}(n)/n$

Equiangular subspaces: configurations of $k$-dimensional subspaces in $\mathbb{R}^d$ with given pairwise angles

Zilin Jiang
Arizona State University
[email protected]