EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED

SECOND EIGENVALUE
arXiv:2302.12230v1 [math.CO] 23 Feb 2023

CARL SCHILDKRAUT

Abstract. Answering a question of Jiang and Polyanskii as well as Jiang, Tidor, Yao,
Zhang, and Zhao, we show the existence of infinitely many angles θ for which the
maximum number of lines in Rn meeting at the origin with pairwise angles θ exceeds
n+Ω(log log n) but is at most n+o(n). To accomplish this, we construct, for various real
λ and integer d, d-regular graphs with second eigenvalue exactly λ and arbitrarily large
second eigenvalue multiplicity. Central to our construction is a distribution on factors
of bipartite graphs which possesses concentration properties.

1. Introduction
A set of lines in Rn passing through the origin is called equiangular if each pair meets
at the same angle θ. The question of the maximum number of equiangular lines in given
dimension is natural and well-studied. A short linear-algebraic

argument due to Gerzon (see
[LS73]) gives an upper bound in dimension n of n+1 2 , and a lower bound on the order of
n2 holds due to a construction of de Caen [dC00].
One may also consider the setting where the common angle is fixed. For an α ∈ (0, 1),
let Nα (n) denote the maximum number of lines in Rn each pair of which meets at an angle
arccos α. One can see that Nα (n) ≥ n for each α and n by constructing the lines spanned
by vectors generating the Gram matrix with ones on the diagonal and α off-diagonal. On
the other hand, Nα (n) ≤ n + 2 unless (1 − α)/(2α) is a totally real algebraic integer which
exceeds all of its Galois conjugates [JP20, Proposition 23].
Much of the literature on Nα (n) focuses on the limit limn→∞ Nα (n)/n. Due to a recent
result of Jiang, Tidor, Yao, Zhang, and Zhao [JTY+ 21] confirming a conjecture of Jiang and
Polyanskii [JP20, Conjecture A], this limit has (essentially) been completely determined. We
begin by stating their result, for which one definition is necessary. Whenever we speak of
the eigenvalues of a graph G, we refer to the eigenvalues of its adjacency matrix AG .
Definition 1.1. For a positive real λ, the spectral radius order k(λ) of λ is the minimum
number of vertices of a graph with largest eigenvalue exactly λ, or ∞ if no such graph exists.
Theorem 1.2 ([JTY+ 21, Theorem 1.2]). Fix α ∈ (0, 1), and let λ = (1 − α)/(2α).
(a) If k = k(λ) is finite, then Nα (n) = ⌊k(n − 1)/(k − 1)⌋ for all sufficiently large n.
(b) If k(λ) = ∞, then Nα (n) = n + o(n).
The o(n) term in case (b) given by [JTY+ 21] is on the order of n/ log log n, and arises
from the same bound on the multiplicity of the second largest eigenvalue of the adjacency
matrix of a bounded degree graph. This result inspired investigation into bounds on this
multiplicity. McKenzie, Rasmussen, and Srivastava [MRS21] improved the upper bound
to n/ log1/5−o(1) n in the case of regular graphs, and also provided sublinear upper bounds
1
2 CARL SCHILDKRAUT

in the unbounded degree case. On the other side, Haiman, Schildkraut, Zhang, and Zhao
[HSZZ22] gave a lower bound (i.e. a construction of bounded-degree graphs with large
second eigenvalue multiplicity) of n1/2−o(1) .
While Theorem 1.2 determines the main term of Nα (n) as n grows, it does not establish
the order of secondary terms in the k(λ) = ∞ case. The following bound is conjectured in
[JP20] and [JTY+ 21]:
Conjecture 1.3 ([JP20, Conjecture B], [JTY+ 21, Conjecture 6.1]). Fix α ∈ (0, 1), and let
λ = (1 − α)/(2α). If k(λ) = ∞, then Nα (n) = n + Oα (1).
Our main result is a disproof of Conjecture 1.3 for infinitely many α, demonstrating the
existence of “intermediate” behavior of Nα (n).
Theorem 1.4. For infinitely many α ∈ (0, 1), the function Nα (n) satisfies
n + Ω(log log n) ≤ Nα (n) ≤ n + O(n/ log log n).
We obtain this result via constructing a family of bounded-degree regular graphs with
second eigenvalue exactly λ = (1 − α)/(2α) for various λ. The second eigenvalue multiplicity
we construct is much smaller than that in [HSZZ22]. However, to our knowledge, this is the
first result giving unbounded second eigenvalue multiplicity when the eigenvalue is fixed.
Theorem 1.5. For infinitely many positive real λ, there exist infinitely many connected n-
vertex graphs with bounded degree and second eigenvalue exactly λ of multiplicity Ω(log log n).
In particular, this holds when
(a) λ is a√sufficiently large integer, or
(b) λ = 2 u2 + 1 − 1 for a sufficiently large integer u.
In fact, we can prove this result for any real λ which is the largest eigenvalue of a symmet-
ric integer matrix satisfying some technical conditions; see Definition 5.3 and Corollary 5.6.
A more precise version of part (b) — Theorem 5.8 — will be enough to give Theorem 1.4.

2. Proof overview
In this section, we give an overview of how the graphs satisfying Theorem 1.5 are con-
structed. Inspired by Marcus, Spielman, and Srivastava [MSS15], the construction will
proceed inductively via the 2-lift operation, first studied in conjunction with eigenvalues by
Bilu and Linial [BL06]. We construct a sequence G0 , G1 , . . . of graphs so that each Gi has
only one eigenvalue greater than λ and for which λ is an eigenvalue of Gi of multiplicity at
least i.
For any graph G, we denote by AG its adjacency matrix, the V (G) × V (G) matrix where
the entry vw for v, w ∈ V (G) is 1 if and only if vw ∈ E(G). For any square matrix M
and any positive integer i at most the dimension of M , we let λi (M ) denote the ith largest
eigenvalue of M , where eigenvalues are counted with multiplicity.

Integer second eigenvalue. When λ is an integer, i.e. to obtain Theorem 1.5(a), the
process is as follows:
(1) Begin with the bipartite graph G0 = Kd,d for some d slightly larger than λ.
(2) From Gi , repeatedly apply 2-lifts which (i) add no new large eigenvalues and (ii)
decrease the L∞ norm of the existing eigenvectors with large eigenvalue. To con-
struct such lifts, we use a result of Marcus, Spielman, and Srivastava [MSS15]; the
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 3

precise properties we need are stated here in Corollary 2.2. Call the graph resulting
from these successive lifts G′i .
(3) Once we have performed these lifts, select a subset of edges of G′i randomly from
an appropriate distribution (see Section 3) such that each vertex of G′i is incident
to exactly a := (d − λ)/2 edges in the subset. This subset informs a choice of 2-lift
Gi+1 of G′i . For this lift, the vector assigning 1 to all vertices in one copy of V (G′i )
and −1 to all vertices in the other copy is an eigenvector of AGi+1 with eigenvalue
d − 2a = λ. Assuming d is chosen suitably, this is with positive probability the
largest new eigenvalue added.
Non-integer second eigenvalue. To obtain Theorem 1.5(b), we follow the same general
strategy as above, but with some modifications. In step (1), instead of constructing d-regular
bipartite graphs, we use k-partite graphs for which the graph between each pair of parts
is a regular bipartite graph. In step (2), we perform the 2-lifts of Marcus, Spielman, and
Srivastava on each pair of parts. Finally, in (3), we choose a subset of edges from between
each pair of parts which forms a regular bipartite graph of specified degree. Varying the
degrees we choose between parts allows us to vary the eigenvalue which repeats. For a
more detailed overview of how the degrees between parts are chosen, see the beginning of
Section 5.
2-lifts. Given a graph G, a signing of G is a map s : E(G) → {±1}. Each signing possesses
an adjacency matrix As where entries of AG corresponding to an edge e are replaced with
s(e). To any signing s of G, one may associate a graph G̃ on the doubled vertex set
{v0 , v1 : v ∈ V (G)} wherein, for each edge uv ∈ E(G), edges u0 v0 and u1 v1 are drawn in G̃
if s(uv) = 1, and edges u0 v1 and u1 v0 are drawn in G̃ if s(uv) = −1. The adjacency matrix
of G̃ can be written in block form as
 
1 AG + As AG − As
AG̃ = .
2 AG − As AG + As
Such a graph G̃ is called a 2-lift of G, on account of the 2-covering map G̃ → G mapping
v0 and v1 to v. We make the following observations:
(1) The multiset of the eigenvalues of AG̃ is the disjoint union of that of AG and that
of As . If x is an eigenvector of AG (resp. As ) with eigenvalue µ, then x ⊕ x (resp.
x ⊕ −x) is an eigenvector of AG̃ with the same eigenvalue µ.
(2) If a vertex v ∈ V (G) has degree dv , the vertices v0 , v1 ∈ V (G̃) above v also have
degree dv .
(3) If G is k-partite, with V (G) = V1 ⊔ V2 ⊔ · · · ⊔ Vk , then G̃ is also k-partite; there are
no edges between vertices corresponding to elements of Vi for each i.
Ramanujan lifts. A special case of a theorem of Marcus, Spielman, and Srivastava gives
the existence of lifts which, in a very strong sense, add no large eigenvalues.
Theorem 2.1 ([MSS15, Theorem 5.3]). Every d-regular√ graph possesses a signing whose
adjacency matrix As has second eigenvalue at most 2 d − 1.
We will make use of the following restatement in the case of bipartite graphs.
Corollary 2.2. For every d-regular bipartite graph √ G, there exists a 2-lift G̃ of G for which,
for every eigenvalue µ of AG̃ satisfying |µ| > 2 d − 1, the µ-eigenspace of AG̃ has the same
dimension as that of AG . Moreover, if {x1 , . . . , xk } is a basis of the µ-eigenspace of AG ,
then {x1 ⊕ x1 , . . . , xk ⊕ xk } is a basis of the µ-eigenspace of AG .
4 CARL SCHILDKRAUT

√
Proof. Let s be a signing of G as in Theorem 2.1 so that λ1 (As ) ≤ 2 d − 1. Since √ As is
a signing of a bipartite graph, its spectrum is symmetric about 0, and so kAs k ≤ 2 d − 1.
The statement then follows from observation (1) on 2-lifts. 
Multiplicity-incrementing lifts and graph factors. While we need Ramanujan lifts to
ensure we don’t have too many large eigenvalues, we will construct more specialized 2-lifts
for the step in which second eigenvalue multiplicity is incremented. Such lifts come from
choosing appropriate a-factors.
Definition 2.3. Given a d-regular graph G and an integer 0 ≤ a ≤ d, an a-factor of G is
an a-regular subgraph of G on the vertex set V (G). Equivalently, a-factors are determined
by selection of a subset S ⊂ E(G) in which every vertex of G is incident to exactly a edges
in S.
As stated in the outline of our approach (specialized to integer eigenvalue), we shall lift
by a signing in which each vertex is incident to the same number of edges signed −1. Letting
H denote the graph consisting of the edges between two parts which are signed −1, this is
equivalent to H being a factor of G. Consider choosing an a-factor H of a d-regular graph
G at random from a distribution in which each edge appears in H with equal probability,
and let s be the signing determined by letting s(e) = −1 if and only if e ∈ E(H). Then
 
2a
E[As ] = E[AG − 2AH ] = 1 − AG .
d
Given any H, As will have an eigenvector (the all-ones vector) with eigenvalue d − 2a. On
the other hand, if we can choose H so that As is close in spectral norm to its expectation,
then the eigenvalues of AG should “shrink” by a factor of 1 − 2a/d up to some small error,
and d − 2a should be the largest eigenvalue of As . We are not able to show that H can be
chosen in this way. However, in accordance with this goal, we give the following proposition,
which gives the existence of a distribution on a-factors of a d-regular bipartite graph G that,
up to a matrix of small spectral norm, displays concentration for Lipschitz functions of the
form u⊺ AH v. Once we have performed enough Ramanujan lifts, this concentration result
will suffices to show the existence of a signing s with λ1 (As ) = d − 2a.
Proposition 2.4. Let a and d be positive integers with 0 ≤ a ≤ d. Let G be a bipartite
d-regular graph on n vertices. There exists a probability distribution on the√ set of pairs
(H, M ) where H is an a-factor of G and M is an n × n matrix with kM k ≤ 6 d such that,
for any vectors u, v ∈ Rn and any real t ≥ 0,
h  a  √ p i 2
Pr u⊺ AH − AG + M v > t(2 + 2) dn⌈log2 n⌉kuk∞ kvk∞ ≤ 2(log2 d)e−t /2 .
d
The matrix M is more an artifact of the proof than an essential component of the distribu-
tion, and one should think of the distribution described in the proposition as a distribution
solely on a-factors with some “extra data” attached which is useful for analysis.
We conclude this section with an outline of the remainder of the paper. In Section 3,
we show Proposition 2.4 and give a corollary which allows us to construct a-factors whose
adjacency matrices are, on particular subspaces of RV (G) , close to (a/d)AG in spectral norm.
In Section 4, we apply this construction to show case (a) of Theorem 1.5, giving large mul-
tiplicity of large integer second eigenvalues from regular bipartite graphs. In Section 5, we
generalize this construction to k-partite graphs and show case (b) of Theorem 1.5. Then, in
Section 6, we apply the graphs constructed in Theorem 1.5(b) to the problem of equiangular
lines, giving Theorem 1.4.
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 5

3. Random factors of graphs

In this section, we show Proposition 2.4, which gives a distribution on a-factors H of any
d-regular bipartite graph G which, up to a matrix of small spectral norm, displays relatively
tight concentration. The distribution is constructed as follows:
(1) In the special case when d is even and a = d/2:
(a) Partition G into edge-disjoint cycles, at each step adding the smallest remaining
cycle to the partition. Lemma 3.1 shows that the union of the large cycles has
small spectral norm.
(b) Note that all cycles in the partition have even length. From each cycle with
2ℓ edges, choose an alternating set of ℓ edges uniformly at random from the
two possible choices and add this set to H. For large cycles, one may make the
choices deterministically. For example, if one such cycle is v1 v2 v3 v4 v5 v6 , add
either the edges {v1 v2 , v3 v4 , v5 v6 } or the edges {v2 v3 , v4 v5 , v6 v1 } to H.
(2) Using (1) as a subroutine, proceed recursively:
(a) In the base case of d = 1, there is nothing to choose: G has a unique 0-factor
and a unique 1-factor.
(b) If a > d/2, construct a random (d − a)-factor and take its complement in G.
(c) If a ≤ d/2 and d is even, construct a random (d/2)-factor G′ of G using (1),
and then construct a random a-factor of the d/2-regular bipartite graph G′ .
(d) If d is odd, remove a 1-factor (i.e. a perfect matching) from G arbitrarily to
form a (d − 1)-regular graph G′ , and then find a random a-factor of G′ .
The terms arising from the large cycles in step 1(b) and from the arbitrary perfect matchings
in 2(d) are absorbed into the auxiliary matrix M . What is left is a sum of many “small”
random choices, and so concentration arises from Chernoff-style arguments.
To prove our concentration result, we will first require the following technical lemma,
which assists in bounding the spectral radius of M .
Lemma 3.1. Let G be a graph on n vertices √ with√maximum degree at most d and no cycles
of length at most 2⌈log2 n⌉. Then kAG k ≤ (2 2) d.
Proof. Let r = ⌈log2 n⌉. On one hand, we have
n
X
tr A2r
G = λi (AG )2r ≥ λ1 (AG )2r .
i=1

On the other hand, tr A2rG counts the number of closed walks of length 2r in G. Each such
walk is also a walk on the universal cover of G, since otherwise would imply the existence
of a cycle of G of length at most 2r. Since every vertex has maximum degree d, the number
of such walks starting at a given vertex is at most dr Cr ≤ (4d)r , where Cr is a Catalan
number. This gives
n(4d)r ≥ λ1 (AG )2r ,
and so
√ 1 √ √ √
λ1 (AG ) ≤ n1/(2r) 4d ≤ n 2 log2 n 4d = (2 2) d. 

We now state and prove the concentration result for step (1) of the above outline.
Lemma 3.2. Let d be an even positive integer. Let G be a bipartite d-regular√ graph on
n vertices. There exists a (deterministic) n × n matrix M satisfying kM k ≤ 2d and a
6 CARL SCHILDKRAUT

probability distribution on the set of d/2-factors H of G such that, for any vectors u, v ∈ Rn
and any real t ≥ 0,
   
1 p 2
Pr u AH − AG + M v ≥ t dn⌈log2 n⌉kuk∞ kvk∞ ≤ 2e−t /2 .
⊺
2
Proof. First, partition the edges E = E(G) of G into sets E0 and E1 as follows: while there
is a cycle in G of length at most 2⌈log2 n⌉, add the edges of this cycle to E1 and remove them
from G; once no such cycle exists, add all remaining edges to E0 . Let G0 and G1 be the
graphs on the
√ same vertex set as G with edge sets E0 and E1 , respectively. By Lemma 3.1,
kAG0 k ≤ 2 2d.
We construct H by partitioning E0 deterministically and E1 randomly. Since E1 is a
union of cycles, every vertex has even degree in G1 and thus in G0 , and so the edges of E0
can be partitioned into cycles. For each such cycle, pick an arbitrary alternating set of edges
of the cycle as in step 1(b) above; let the resulting edges be E0′ and let H0 be the graph with
edge set E0′ . Each vertex has degree in H0 half of that in G0 . Let M = 1/2 · A √G0 − AH0 ;
since M is half the adjacency matrix of a signing of G0 , kM k ≤ (1/2)kAG0 k ≤ 2d.
Now, G1 is a union of cycles of length at most 2⌈log2 n⌉. For each such cycle, choose an
alternating set of half the edges randomly from the two possible choices, and add its edges
to E1′ ; let H1 be the graph with edge set E1′ . Let H be the graph with edge set E0′ ∪ E1′ ; H
is an a-factor of G, and
a 1 1 1 1
AH − AG + M = AH0 + AH1 − AG0 − AG1 + AG0 − AH0 = AH1 − AG1 .
d 2 2 2 2
Consider the random variable
 
1
X = u AH1 − AG1 v.
⊺
2
Letting C be the set of cycles the union of which is E1 ,
 
X 1
X= XC u ⊺
ÃC v,
2
C∈C

where XC ∼ Unif({±1}) and ÃC is the adjacency matrix of a signing of C in which incident
edges are signed alternately. Since u⊺ ÃC v is a sum of 2 len(C) terms of the form ±ui vj for
integers i and j,  
1
u⊺ ÃC v ≤ len(C)kuk∞ kvk∞ .
2
So, the bounded differences inequality implies, for any µ ≥ 0,
!
−2µ2
Pr [|X| ≥ µ] ≤ 2 exp P 2 .
C∈C (2 len(C)kuk∞ kvk∞ )
We have
X  X
2
len(C) ≤ max len(C) len(C) ≤ 2⌈log2 n⌉E(G) = dn⌈log2 n⌉,
C∈C
C∈C C∈C
so  
−µ2
Pr [|X| ≥ µ] ≤ 2 exp .
2dn⌈log2 n⌉kuk2∞ kvk2∞
p
This implies the desired concentration upon setting µ = t dn⌈log2 n⌉kuk∞ kvk∞ . 
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 7

We use this lemma as a black box, as described in step (2) of the outline, to show
Proposition 2.4.
Proof of Proposition 2.4. Let cd be
√ the sequence defined recursively by c1 = 0, cd = cd−1 + 1
if d > 1 is odd, and cd = cd/2 + 2d if d is even. We proceed by strong induction on d to
show that there exists a distribution on pairs (H, M ) of a-factors of G and n × n matrices
wherein kM k ≤ cd and, for any t ≥ 0 and u, v ∈ Rn ,
h  a  √ p i 2
Pr u⊺ AH − AG + M v > t(2 + 2) dn⌈log2 n⌉kuk∞ kvk∞ ≤ 2(log2 d)e−t /2 .
d
In the base case of d = 1, there is not much choice; if a = 0 then H is empty, while if a = 1
then H = G. In either case, we can take M = 0, and u⊺ (AH − a/d · AG + M )v = 0 always.
For the inductive step, we first treat the case where d is even and a ≤ d/2. In this case,
we proceed in two steps:
(1) Construct a d/2-factor G1 of G via Lemma 3.2, and let M1 be the associated matrix.
(2) Construct an a-factor H of G1 via the inductive hypothesis, and let M2 be the
associated matrix.
√
We set M = M2 + 2a/d · M1. We have kM1 k ≤ 2d by Lemma 3.2 and kM2 k ≤ cd/2 by the
inductive hypothesis. This gives
√
kM k ≤ kM2 k + (2a/d)kM1 k ≤ cd/2 + 1 · 2d = cd
and    
a a 2a 1
AH − AG + M = AH − AG1 + M2 + AG1 − AG + M1 ,
d d/2 d 2
as well as the concentration inequalities
   
1 p 2
Pr u AG1 − AG + M1 v ≥ t dn⌈log2 n⌉kuk∞ kvk∞ ≤ 2e−t /2
⊺
2
"   #
a
u⊺ AH − d/2 AG1 + M2 v 2
Pr √ p ≤ 2(log2 (d/2))e−t /2 .
≥ t(2 + 2) (d/2)n⌈log2 n⌉kuk∞ kvk∞
This implies by the union bound that the probability that u⊺ (AH − a/d · AG + M )v exceeds
√ p 2a  p 
t(2 + 2) (d/2)n⌈log2 n⌉kuk∞kvk∞ + t dn⌈log2 n⌉kuk∞ kvk∞
d
2
in magnitude is at most 2(log2 d)e−t /2 , as this would imply that one of the two above events
occurs. Since
√ p 2a √ √ √ √ √ √
(2 + 2) d/2 + d ≤ ( 2 + 1) d + d = (2 + 2) d,
d
this is sufficient.
In the case where d is even and a > d/2, we first construct a pair (H ′ , M ′ ) where H ′ is
a (d − a)-factor of G. Let H be the complement of H ′ in G, and let M = −M ′ . Since
 
a a d−a
AH − AG + M = (AG − AH ′ ) − AG − M ′ = − AH ′ − AG + M ′ ,
d d d
the hypothesis for (d − a)-factors of d-regular graphs suffices.
Finally, we treat the case where d is odd. If a > d/2 then we can proceed as in the
previous paragraph from the a < d/2 case, so assume a < d/2. We proceed as follows:
8 CARL SCHILDKRAUT

(1) Select an arbitrary perfect matching H ′ ⊂ G, and let G1 be the complement of H ′

in G, a (d − 1)-factor of G.
(2) Construct an a-factor H of G1 via the inductive hypothesis, and let M1 be the
associated matrix.
Define M = ad AH ′ − a
d(d−1) AG1 + M1 . We have

a a a a
AH − AG + M = AH − AG + AH ′ − AG + M1
d d d d(d − 1) 1
a a
= AH − AG1 − AG + M1
d d(d − 1) 1
a
= AH − AG + M1 ,
d−1 1
as well as
a a a a
kM k ≤ kAH ′ k + kAG1 k + kM1 k ≤ + + cd−1 ≤ cd−1 + 1
d d(d − 1) d d

so the inductive hypothesis at d − 1 suffices.

√
All that remains is to show that cd ≤ 6 d for all d. We can show
√ √
cd ≤ (2 + 2)( 2d − 1) + (# of ones in the binary representation of d)
√ √
≤ (2 + 2)( 2d − 1) + log2 d

by induction on d, so the result follows from

√ √ √
log2 d − (2 + 2) ≤ (4 − 2 2) d,

which is easy to check. 

We conclude this section with a corollary of Proposition 2.4 which shows a concentration
result for factors of regular bipartite graphs on subspaces with an orthonormal basis with
small L∞ norm. This is what we will use when constructing our multiplicity-incrementing
lifts.

Corollary 3.3. Let a and d be positive integers with 0 ≤ a ≤ d. Let G be a bipartite d-regular
graph on n vertices, and let 1 ≤ K, m ≤ n. Let S be a finite set containing V (G), and let
U ⊂ RS be a subspace of dimension p at most m spanned by pairwise orthogonal unit vectors
u1 , u2 , . . . , um′ satisfying kui k∞ ≤ K/n for each 1 ≤ i ≤ m′ . If m ≥ max(2 log2 d, 15)
and n ≥ K 3 m6 , then there exists an a-regular subgraph H ⊂ G on V (G) satisfying
 a  √
v ⊺ AH − AG v ≤ 7 d
d
for every unit vector v ∈ U , where AH and AG are extended by zeros to a matrix in RS×S .

Proof. For each√1 ≤ i ≤ m′ , let ui be the restriction of ui to RV (G) ; note that kui k∞ ≤
kui k∞ . Set t = 6.5 ln m, so that
2
/2
2(log2 d)e−t ≤ me−3.25 ln m < m−2 .
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 9

By Proposition 2.4 and the union bound, there exists

√ an a-regular subgraph H ⊂ G on
V (G) and an n × n matrix M satisfying kM k ≤ 4 d for which
 a   a 
u⊺i AH − AG + M uj = ui ⊺ AH − AG + M uj
d d
√ p
≤ t(2 + 2) dn⌈log2 n⌉kui k∞ kuj k∞
r
d ln m⌈log2 n⌉
< 9K
n
for all 1 ≤ i, j ≤ m′ . Now, take any unit vector v ∈ U . Since {ui }i is an orthonormal basis
Pm′ P 2
of U , we can write v = i=1 ci ui for some real c1 , . . . , cm′ with ci = 1. This gives
  m′ Xm′  
a X a
v ⊺
AH − AG + M v ≤ |ci ||cj | u⊺i AH − AG + M uj
d i=1 j=1
d
r
d ln m⌈log2 n⌉
< 9K(|c1 | + · · · + |cm′ |)2
n
r
d ln m⌈log2 n⌉
≤ 9Km ,
n
where we have √used the Cauchy–Schwarz inequality. Since n ≥ 2, n/⌈log2 n⌉ ≥ n2/3 , and
since m ≥ 15, 9 ln m ≤ m, so this bound gives
 a  √ √
v ⊺ AH − AG + M v ≤ Km2 n−1/3 d ≤ d
d
√
by our bound on n. The fact that kM k ≤ 6 d finishes the proof. 

4. Integer second eigenvalue

In this section, we use Corollary 3.3 to construct fixed-degree regular graphs with a fixed
second eigenvalue of multiplicity Ω(log log n); the fixed eigenvalue can be any sufficiently
large integer. The main result of this section will be the following “incrementing” proposi-
tion. Iterating this will allow us to obtain Theorem 1.5(a).
√
Proposition 4.1. Let d ≥ 22000, and let a = ⌈12 d⌉. Suppose G is a bipartite d-regular
graph on n vertices for which λ2 (AG ) ≤ d − 2a, and let ℓ be the multiplicity of d − 2a as an
eigenvalue of AG . Then there exists a bipartite d-regular graph G̃ on at most 4n9 vertices
with second eigenvalue d − 2a of multiplicity at least ℓ + 1.
The process will be, as described at the beginning of Section 2, to first perform many
lifts which add no new large eigenvalue, and then to apply a multiplicity-incrementing lift
determined by a random a-factor of G using Corollary 3.3. The properties of this lift are
described in the following lemma.
√
Lemma 4.2. Let d ≥ 22000, and let a = ⌈12 d⌉. Let G be a d-regular bipartite graph
with n vertices satisfying
√ λ2 (AG ) ≤ d − 2a. Suppose that there are at most m eigenvalues
of G greater than 2 d − 1, and AG possesses an orthonormal
p eigenbasis in which each
component of these m′ ≤ m top eigenvectors is at most m/n in magnitude. Then, as long
as m ≥ 2 log2 d and n > m9 , there exists an a-factor H of G with λ1 (AG − 2AH ) = d − 2a.
We first prove the proposition given this lemma, and then prove the lemma.
10 CARL SCHILDKRAUT

Proof of Proposition 4.1. Let t be a positive integer. By applying Corollary 2.2 t times, we
can find a d-regular bipartite
√ graph G′ on n′ := n2t vertices for which, for each eigenvalue
µ of AG′ with |µ| ≥ 2 d − 1, the µ-eigenspace of AG′ is exactly
 
 
x ⊕ x ⊕ · · · ⊕ x : AG x = µx .
| {z } 
2t copies of x

In particular,
√
λ2 (AG′ ) ≤ max(λ2 (AG ), 2 d − 1) ≤ d − 2a,

and AG′ has eigenvalue d− 2a√with multiplicity ℓ. Moreover, A′G possesses at most |V (G)| =
n eigenvalues greater than 2 d − 1, and an orthonormal eigenbasis of AG′ containing
 
1
(x ⊕ · · · ⊕ x) : x is in an orthonormal eigenbasis of AG
2t/2
√
possessespno eigenvector with an eigenvalue larger than 2 d − 1 and an entry larger than
2−t/2 = n/n′ in magnitude. So, as long as n ≥ 2 log2 d (which holds since there exists a
d-regular graph G on n vertices) and n2t > n9 , we can apply Lemma 4.2 to find an a-factor
H of G′ with λ1 (AG′ − 2AH ) = d − 2a. Letting s be the associated signing of G′ and G̃
be the associated 2-lift, the spectrum of G̃ contains the eigenvalue d − 2a with multiplicity
at least ℓ + 1. Selecting t to be the smallest positive integer satisfying 2t > n8 finishes the
proof. 

Proof
√ of Lemma 4.2. Let U be the span of all eigenvectors of AG with eigenvalue exceeding
′
2 d − 1 and strictly less than d; note that dim U ≤ mp ≤ m, and U possesses an orthonormal
basis in which each vector has L∞ norm at most m/n. So, by Corollary 3.3 (which we
can apply since 2 log2 d > 15), there exists an a-factor H of G with
 a  √
v ⊺ AH − AG v ≤ 7 d
d

for every unit vector v ∈ U . In particular,

a ⊺ √
(⋆) v ⊺ AH v ≥ v AG v − 7 dkvk2 for all v ∈ U.
d

Now, since H is a-regular, the matrix AG − 2AH has eigenvalue d − 2a with the all-ones
vector as an eigenvalue. We need to show that there are no larger eigenvalues. To this end,
consider any vector w orthogonal to the all-ones vector; it suffices to show w⊺ AH w ≤ d − 2a.
Write w =√x + y, where x ∈ U and y ∈ U ′ , the span of all eigenvectors of AG with eigenvalue
at most 2 d − 1. We compute

w⊺ AH w = x⊺ AH x + 2x⊺ AH y + 2y ⊺ AH y
a √
≥ x⊺ AG x − (7 d)kxk2 − 2kxk(ay) − akyk2
d
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 11

where we have used (⋆) and the fact that kAH k = a. Now, this allows us to bound
w⊺ (AG − 2AH )w = x⊺ AG x + y ⊺ AG y − 2w⊺ AH w
√ 2a
≤ x⊺ AG x + (2 d − 1)kyk2 − x⊺ AG x
√ d 2
+ (14 d)kxk + 2akyk(2kxk + kyk)
 
2a √ √
= 1− x⊺ AG x + (14 d)kxk2 + 4akxkkyk + (2a + 2 d − 1)kyk2
d
   
2a √ √
≤ (d − 2a) 1 − + 14 d kxk2 + 4akxkkyk + (2a + 2 d − 1)kyk2 .
d
This is a quadratic form in kxk and kyk. The maximum of a quadratic
p form Rx2 +2Sxy+T y 2
over the unit circle can be easily computed to be (R + T + (R − T )2 + 4S 2 )/2. Since
kxk2 + kyk2 = kwk2 = 1, all that remains is to show that this is at most d − 2a for the
quadratic form above in kxk and kyk as long as d ≥ 22000. This is a simple computation. 
We conclude this section by establishing Theorem 1.5(a).
Proof of Theorem 1.5(a). Let ℓ ≥ 20000 be a positive integer. We will show that, for
infinitely many n, there exist bounded-degree regular graphs on n vertices with second
eigenvalue exactly ℓ, of multiplicity Ω(log log n).
√ √
Let d be a positive integer exceeding 22000 for which d−2⌈12 d⌉ = ℓ and let a = ⌈12 d⌉,
so that ℓ = d − 2a. Set G0 = Kd,d and n0 = 2d, and define a sequence (ni ) recursively by
ni+1 = 4n9i . Note that λ2 (G0 ) = 0 < d − 2a. Repeated applications of Proposition 4.1 show
that, for each i ≥ 1, there exists a bipartite d-regular graph Gi such that
(i) Gi has at most ni vertices, and
(ii) d − 2a is the second eigenvalue of Gi and has multiplicity at least i.
Θ(i)
The fact that ni = 22 finishes the proof, since i = Θ(log log ni ). 
Remark 4.3. If we were able to find, for any bipartite d-regular graph G, an a-factor H
with
a
AH − AG = o(a),
d
this would allow us to remove the Ramanujan lift step and attain second eigenvalue multi-
plicity on the order of log n. This aim is reminiscent of a result of Bilu and Linial [BL06,
Theorem 3.1] that there exists a signing s of such a graph with kAs k = O(d1/2 log3/2 d),
but in our distribution there seems not to be enough randomness to follow their spectral
radius-bounding framework.

5. Non-integer second eigenvalue

In this section we give a generalization of the framework from the previous section, which
we will use to prove our main result. We begin by elaborating on the sketch of the k-partite
graphs-based strategy given in Section 2. To do this, we first need some terminology.
Definition 5.1. Recall that a k×k symmetric matrix M is irreducible if the (not necessarily
simple) graph with adjacency matrix M is connected. (Irreducibility ensures that we can
apply the Perron–Frobenius theorem to M .) Given a symmetric k × k irreducible matrix
M with zeros on the diagonal and nonnegative integer entries:
12 CARL SCHILDKRAUT

• A graph lift of M is a lift of the non-simple graph with adjacency matrix M . In

other words, it is a k-partite graph G on vertex set V1 ⊔ V2 ⊔ · · · ⊔ Vk such that, for
each 1 ≤ i, j ≤ k, G[Vi ⊔ Vj ] is a Mij -regular bipartite graph.
• A sign matrix of M is a symmetric integer matrix M ′ for which Mij′ ≡ Mij (mod 2)
and |Mij′ | ≤ Mij for each 1 ≤ i, j ≤ k.
• An M ′ -signing of a graph lift G of M is an assignment of an element of {±1} to
each edge of G such that, for each 1 ≤ i, j ≤ k, the labels of the edges of G[Vi ⊔ Vj ]
incident to each vertex of Vi ⊔ Vj sum to Mij′ .
For example, take    
0 d 0 d − 2a
M= , M′ =
d 0 d − 2a 0
√
for a = ⌈12 d⌉. Then M ′ is a sign matrix of M , and the graph lifts of M are exactly
the d-regular bipartite graphs. Lemma 4.2 demonstrates the existence of an M ′ -signing
of such a graph with top eigenvalue exactly d − 2a, under certain conditions. We now
prove the following lemma, which gives information about the eigendata of graph lifts and
matrix-signings.
Lemma 5.2. Let (M, M ′ , G) be as in Definition 5.1.
(a) The top eigenvalue of AG is λ1 (M ); if v is a top eigenvector of M , then the vec-
tor which assigns each vertex in Vi the value vi is an eigenvector of AG with this
eigenvalue.
(b) Consider an M ′ -signing s : E(G) → {±1}. The eigenvalues of M ′ are eigenvalues
of the adjacency matrix As of this signing, and moreover their eigenspaces span the
space of vectors v in RV (G) which are constant on each part of the k-partition.
Proof. For (a), it is easy to check that λ1 (M ) is an eigenvalue of AG with the eigenvector
described as such. The fact that this is the top eigenvalue follows from the observation that,
since both AG and M possess nonnegative entries, v is a Perron eigenvector, and so λ1 (M )
is the Perron eigenvalue of AG . For (b), it is easy to check that if w is an eigenvector of M ′
with eigenvalue µ, the vector which assigns a vertex in Vi the value wi is an eigenvector of
As with eigenvalue µ as well. 
Due to this lemma, the eigenvalue for which we are able to guarantee large multiplicity is
λ1 (M ′ ). We know that it will appear in any M ′ -signing; all that remains is to show that,
under certain conditions, no eigenvalue is larger. We first make these conditions explicit.
Definition 5.3. Denote by M the set of triples (M, M ′ , β) for which
(a) M is an irreducible k × k nonnegative integer symmetric matrix with zeros on the
diagonal, and M ′ is a sign matrix of M ,
(b) β ∈ (0, 1/2), and
′
(c) if D = (MP−Mp)/2, γ = λ1 (|D −βM |) (where the absolute value is taken entrywise),
and σ = i,j Mij , then
 
(1 − 2β)λ1 (M ′ ) + 2γ + 7σ 2λ1 (D)
λ1 ≤ λ1 (M ′ ).
2λ1 (D) 2λ1 (D) + σ
We may now state the following proposition which generalizes Proposition 4.1.
Proposition 5.4. Suppose (M, M ′ , β) ∈ M. Let G be a graph lift of M on n vertices for
which λ2 (AG ) ≤ λ1 (M ′ ), and let ℓ be the multiplicity of λ1 (M ′ ) as an eigenvalue of AG .
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 13

Then there exists a graph lift G̃ of G on at most 4n9 vertices with second eigenvalue λ1 (M ′ )
of multiplicity at least ℓ + 1.
As in the proof of Proposition 4.1, the main technical piece will be a lemma which gives
the existence of a multiplicity-incrementing lift. We apply such a lift after applying enough
Ramanujan lifts between each pair of parts.
Lemma 5.5. Suppose (M, M ′ , β) ∈P M, pand let G be a graph lift of M on n vertices. Let
m be a positive integer. Define σ = i,j Mij , as in Definition 5.3. Suppose
(a) λ2 (AG ) ≤ λ1 (M ′ ),
(b) at most m eigenvalues of G exceed σ,
(c) AG possesses an orthonormal eigenbasis, a subset of which is a basis for those vectors
constant on each part of the k-partition of V (G) afforded by M , in which p each
component of each eigenvector with eigenvalue greater than σ is at most m/n in
magnitude, and
(d) n > m9 .
Then there exists a M ′ -signing s : E(G) → {±1} with adjacency matrix As satisfying
λ1 (As ) = λ1 (M ′ ).
Each of the conditions in this lemma parallels a condition in Lemma 4.2. In particular,
′
the condition
√ that (M, M , β) ∈ M analogizes the degree bound d ≥ 22000 and the selection
of a = ⌈12 d⌉. The proof will be quite similar to that of Lemma 4.2.
Proof. As in Definition 5.3, define D = (M − M ′ )/2 and γ = λ1 (|D − βM |). Let V1 ⊔
V2 ⊔ · · · ⊔ Vk be the k-partition of V (G) afforded by M . Let U0 ⊂ RV (G) be the space of
vectors which are constant on each Vi . By Lemma 5.2, the top eigenvalue of G is λ1 (M ),
and its eigenvector lies in U0 . Also by Lemma 5.2, we know that for any M ′ -signing s of
G there exists a basis of U0 which is contained within an eigenbasis of As , one vector in
this basis contributes an eigenvalue λ1 (M ′ ) to As , and no vector in this basis contributes a
larger eigenvalue. So, it suffices to find an M ′ -signing s for which, for any vector w ∈ RV (G)
orthogonal to U0 , w⊺ As w ≤ λ1 (M ′ ).
We construct this signing separately on the edges between Vi and Vj for each pair (i, j).
Let Gij be the graph G[Vi ⊔ Vj ] on vertex set V (G), so that
X
AGij = AG .
1≤i<j≤k

Let v2 , . . . , vm′ be the eigenvectors of AG in the partition described in (c) with eigenvalues
exceeding σ and which do not lie in U0 , and let U1 be the subspace of RV (G) they span.
By (b), dim U1 ≤ m′ < m. The space U1 satisfies the condition of Corollary 3.3 with
K = m. So, since n ≥ m9 (condition (d)), we can for each 1 ≤ i < j ≤ k find a Dij -factor
Hij of Gij which satisfies
 
Dij p
v ⊺ AHij − AGij v ≤ 7 Mij kvk2
Mij
for every v ∈ U1 . We conclude, letting H be the graph formed by the union of the edge sets
P Dij
Hij and letting B = Mij AGij , that
 
X p 7
|v ⊺ (AH − B)v| ≤ 7kvk2  Mij  = σkvk2
2
1≤i<j≤k
14 CARL SCHILDKRAUT

for every v ∈ U1 . Let s be the signing for which edges in Hij for any (i, j) are assigned
−1 and all other edges are assigned 1. By definition, s is an M ′ -signing; we also have
As = AG − 2AH .
Now, let w ∈ RV (G) be any unit vector orthogonal to U0 ; write w = x + y, where x ∈ U1
and y lies in the orthogonal complement of U0 + U1 ; note that kxk2 + kyk2 = 1. We first see
w⊺ AH w = x⊺ AH x + 2x⊺ AH y + y ⊺ AH y
≥ x⊺ AH x − 2kxkkykkAH k − kyk2 kAH k
7
≥ x⊺ Bx − σkxk2 − (2kxkkyk + kyk2 )kAH k
2  
7
≥ βx AG x − kB − βAG k + σ kxk2 − (2kxkkyk + kyk2 )kAH k.
⊺
2
Since H is a graph lift of D, and the entries of D are nonnegative integers, Lemma 5.2(a)
implies that kAH k = λ1 (AH ) = λ1 (D). We now upper-bound kB − βAG k. For any unit
vector u ∈ RV (G) , write u = u1 + · · · + uk , where ui is supported on Vi . We have, similarly
to the proof of Lemma 5.2,
X  Dij 
u (B − βAG )u =
⊺
− β u⊺ AGij u
Mij
1≤i<j≤k
X  Dij 
= − β 2Mij kui kkuj k
Mij
1≤i<j≤k
k X
X k
= |Dij − βMij |kui kkuj k ≤ λ1 (|D − βM |) = γ,
i=1 j=1

since ku1 k2 + · · · + kuk k2 = 1. So, kB − βAG k ≤ γ. This implies that

w⊺ As w = w⊺ (AG − 2AH )w
= x⊺ AG x + y ⊺ AG y − 2w⊺ AH w
≤ x⊺ AG x + σkyk2 − 2βx⊺ AG x + (2γ + 7σ) kxk2 + 4λ1 (D)kxkkyk + 2λ1 (D)kyk2


= (1 − 2β)x⊺ AG x + (2γ + 7σ) kxk2 + 4λ1 (D)kxkkyk + 2λ1 (D) + σ kyk2


≤ ((1 − 2β)λ1 (M ′ ) + 2γ + 7σ) kxk2 + 4λ1 (D)kxkkyk + 2λ1 (D) + σ kyk2 ,
where we have used condition (a) in the last bound, since x ∈ U0 . Since we have for any
R, S, T and any kxk, kyk ∈ R with kxk2 + kyk2 = 1,
    
2 2

R S kxk R S
Rkxk + 2Skxkkyk + T kyk = kxk kyk ≤ λ1 ,
S T kyk S T
the result follows from the definition of M. 
Proof of Proposition 5.4. Armed with the above lemma, the proof will be very similar to
that of Proposition 4.1. Let t be a nonnegative integer. We will first construct a graph
lift Gt of M on nt := n2t vertices for which, for each eigenvalue µ of AGt with µ ≥ σ, the
µ-eigenspace of AGt is exactly
 
 
x ⊕ x ⊕ · · · ⊕ x : AG x = µx .
| {z } 
2t copies of x
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 15

Indeed, starting with G0 = G when t = 0, we can construct these graphs by successively

applying Ramanujan 2-lifts between each pair of parts. Let V1 ⊔V2 ⊔· · ·⊔Vk be the k-partition
of V (Gt ) afforded by M . For each 1 ≤ i < j ≤ k, let sij : E(Gt [Vi ⊔Vj ]) → {±1} be a signing
of the Mijp -regular bipartite
p graph Gt [Vi ⊔ Vj ] whose adjacency matrix has second eigenvalue
at most 2 Mij − 1 < 2 Mij , guaranteed to exist by Theorem 2.1 (if Mij = 0, then sij is
p
trivial and its adjacency matrix has second eigenvalue 0 = 2 Mij ). Construct P a signing
s : E(Gt ) → {±1} by combining the partial signings sij , so that As = 1≤i<j≤k Asij has
top eigenvalue at most
X X k X
X k
p p
λ1 (Asij ) ≤ 2 Mij = Mij = σ.
1≤i<j≤k 1≤i<j≤k i=1 j=1

Then the 2-lift Gt+1 of Gt associated to s has no eigenvalues exceeding σ which do not come
from AGt , and so Gt+1 satisfies the desired properties. In particular, for each t,

λ2 (AGt ) ≤ max(λ2 (AG ), σ) ≤ λ1 (M ′ ),

(we know σ ≤ λ1 (M ′ ) by condition (c) in the definition of M), and AGt has eigenvalue
λ1 (M ′ ) with multiplicity k. Identically to the proof of Proposition 4.1, conditions (b) and
(c) of Lemma 5.5 are satisfied by Gt , since we have described the eigenspaces of AGt with
large eigenvalue. So, we can apply this lemma to get an M ′ -signing s : E(G) → {±1}
with λ1 (As ) = λ1 (M ′ ) as long as 2j > n8 . The spectrum of the associated 2-lift contains
the eigenvalue λ1 (M ′ ) with multiplicity at least ℓ + 1, and so selecting t to be the smallest
positive integer satisfying 2t > n8 finishes the proof. 

These multiplicity-incrementing steps give us graphs with large multiplicity of fixed sec-
ond eigenvalue.

Corollary 5.6. Suppose (M, M ′ , β) ∈ M and that there exists a connected graph lift G0 of
M on at most r vertices with λ2 (AG0 ) ≤ λ1 (M ′ ). Then there exists an infinite sequence r =
n0 < n1 < · · · of positive integers, each satisfying ni+1 ≤ 4n9i , for which there exists a graph
on ni vertices with top eigenvalue λ1 (M ) and second eigenvalue λ1 (M ′ ) with multiplicity
at least i = Ω(log log ni ). Furthermore, if the all-ones vector is an eigenvector of M , then
these graphs can be chosen to be regular.

Proof. This parallels the proof of Theorem 1.5(a). In the previous section. Let n0 = r, and
define a sequence (ni ) recursively so that ni+1 is the unique integer in the interval (2n9i , 4n9i ]
for which ni+1 /ni is a power of 2. Repeated applications of Proposition 5.4 show that, for
each i ≥ 1, there exists a graph lift Gi of M such that
(i) Gi has ni vertices, and
(ii) λ1 (M ′ ) is the second eigenvalue of Gi and has multiplicity at least i.
Since Gi is a graph lift of M , its top eigenvalue is λ1 (M ) by Lemma 5.2(a). If the all-
ones vector is an eigenvector of M , then it is the Perron eigenvector of AGi (again by
Lemma 5.2(a)) for each i, and so (since each Gi will be connected) each Gi is regular with
degree λ1 (M ). 

To conclude the proof of Theorem 1.5(b), we need to choose our matrices M and M ′ and
show that they satisfy the necessary properties. We do this in the following lemma.
16 CARL SCHILDKRAUT

Lemma 5.7. Let t > u be positive integers of the same parity. Define the matrices
   
t t 3 u u 1
t 3 t  , M ′ = u −3 u

M = t 3
,
t u −3 u
3 t t 1 u u
√
and let β = (t − u)/(2t). Suppose (5/6)t ≤ u ≤ t − 56 t − 62. Then
√
(a) λ1 (M ′ ) = 2 u2 + 1 − 1,
(b) there exists a (connected) graph lift G0 of M on 4t vertices with λ2 (AG0 ) ≤ λ1 (M ′ ),
and
(c) (M, M ′ , β) ∈ M.
Proof. We first note that the bounds on u imply that β ≤ 1/12 and t, u ≥ 104 . To prove
part (a), it suffices to compute the eigenvalues of M ′ explicitly to be
n p p o
2 u2 + 1 − 1, 2 u2 + 1 − 1, 3, −1 .
√
Since u ≥ 2, the largest of these is 2 u2 + 1 − 1.
To prove part (b), we construct such a graph. Between the parts to be connected by
(1)
t-regular bipartite graphs, we place complete bipartite graphs, forming a graph G0 . Then
we place 3-regular bipartite graphs between the other two pairs of parts arbitrarily (this is
(2)
possible since t ≥ 3); let the union of these two 3-regular bipartite graphs be G0 . Since
(2)
G0 is 3-regular, its adjacency matrix has spectral norm at most 3. Weyl’s inequality thus
tells us  
λ2 (AG0 ) ≤ λ2 AG(1) + 3.
0
(1)
The spectrum of G0 consists of one copy each of ±2t and 4t − 2 copies of 0. This shows
λ2 (AG0 ) ≤ 3 ≤ λ1 (M ′ ), as desired.
Finally, we show (c). It is clear that M ′ is a sign matrix of M ; the rest requires some
computation. We have the matrices
   
tβ tβ 1 1 − 3β
M − M′  tβ 3 tβ   3 − 3β 
D := =  , D − βM =  ;
2  tβ 3 tβ   3 − 3β 
1 tβ tβ 1 − 3β
√ √ √
this gives λ1 (D) ≤ 2tβ + 3 and γ = λ1 (|D − βM |) ≤ 3. Also, σ = 8 t + 4 3 < 8 t + 7.
Since the top eigenvalue of positive matrices may only increase if entries are increased, it
suffices to show, using that 2u − 1 ≤ λ1 (M ′ ) ≤ 2u,
 √ 
(1 − 2β)(2u) + 56 t + 55 4tβ √
+6
L := λ1 ≤ 2u − 1.
4tβ + 6 4tβ + 8 t + 13
By sub-additivity of top eigenvalue, we have
 √   
56 t + 55 √ 6 2(1 − 2β)2 4β √
L ≤ λ1 + t · λ1 ≤ 56 t + 61 + t(2 − 6β),
6 8 t + 13 4β 4β
where we have used the computation that
 
2(1 − 2β)2 4β
λ1 ≤ 2 − 6β
4β 4β
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 17

for β ≤ 1/12. We thus need √

56 t + 62 ≤ 2βt = t − u,
which holds by assumption. 
We conclude with the proof of Theorem 1.5(b). In fact, we following prove the (ever so
slightly) stronger statement; we will need its precision to show Theorem 1.4.
Theorem 5.8. Let u ≥ 105 be an integer. There exists some integer d ≤ 5u/2 for which,
for an infinite sequence n0 < n1 < · · · of positive integers, each satisfying√ ni+1 ≤ 4n9i ,
there exists a d-regular graph on ni vertices with second eigenvalue exactly 2 u2 + 1 − 1 of
multiplicity i = Ω(log log ni ).
√
Proof. Let t be a positive integer of the same parity as u satisfying 5t/6 ≤ u ≤ t− 56 t− 62,
which exists since u ≥ 105 . Define
   
t t 3 u u 1
t 3 t  , M ′ = u −3 u

M = t 3
,
t   u −3 u
3 t t 1 u u
√
and β = (t − u)/(2t). By Lemma 5.7, λ1 (M ′ ) = 2 u2 + 1 − 1, there exists a connected
graph lift G0 of M on 4t vertices with λ2 (AG0 ) ≤ λ1 (M ′ ), and (M, M ′ , β) ∈ M. The fact
that the all-ones vector is an eigenvector of M with eigenvalue 2t + 3 ≤ 12u/5 + 3 ≤ 5u/2
is enough to finish the proof by appealing to Corollary 5.6. 

6. Application: Equiangular lines

We apply Theorem 5.8 to the problem of equiangular lines with fixed angle. Recall that,
if the spectral radius order k(λ) of λ = (1 − α)/(2α) is finite, then Nα (n) ≥ (1 + ε)n for
some ε > 0 and large n. So, to prove Theorem 1.4, we need to find λ which is not the top
eigenvalue of any graph, but which may still be the second eigenvalue. To this end, we begin
by giving a necessary condition for k(λ) < ∞. This condition is stated in [JP20].
Lemma 6.1. Suppose λ is such that k(λ) < ∞. Then λ is an algebraic integer all of whose
Galois conjugates are real and at most λ in magnitude.
Proof. Let G be any graph, and consider the characteristic polynomial pG (t) = det(tI −AG ).
This polynomial is monic of degree |V (G)| and has integer coefficients; this implies (1) that
every root of pG is an algebraic integer and (2) that if λ is a root of pG , then every Galois
conjugate of λ is as well (the minimal polynomial of λ must divide pG ). Since AG is
symmetric, every eigenvalue of AG is real. The remainder of the lemma follows from the
fact that, since every entry of AG is nonnegative, every eigenvalue of AG is at most λ1 (AG )
in magnitude, since λ1 (AG ) is a Perron eigenvalue. 
Remark 6.2. This condition is not sufficient. For any interval I ⊂ R of length strictly
exceeding 4 (resp. strictly less than 4), there exist infinitely many (resp. finitely many)
totally real algebraic integers all of whose Galois conjugates lie within I [Rob62]. In partic-
ular, there are infinitely many totally real algebraic integers all of whose conjugates lie in
[−1.995, 2.006], and, for all but finitely many of them, the largest conjugate strictly exceeds
2. On the other hand, due to a result of [CDG82] (see [CR90, Theorem 2.3] and surrounding
remarks) no undirected graph has largest eigenvalue in the interval (2, 2.007].
Our choice of λ is described in the following corollary.
18 CARL SCHILDKRAUT

√
Corollary 6.3. If t is a positive integer which is not a square, then k(2 t − 1) = ∞.
√ √ √
Proof. For non-square t, −2 t − 1 is a Galois conjugate of 2 t − 1 and exceeds 2 t − 1 in
magnitude. 

Now, we give a result which allows us to derive sets of equiangular lines in every dimension
from the graphs we have constructed.

Lemma 6.4. Suppose there exists an n-vertex d-regular graph G with second eigenvalue
λ > 0 of multiplicity k, and let α = 1/(2λ + 1).
(a) If n ≥ 2d, then Nα√
(n − k) ≥ n.
(b) If moreover λ ≥ 2 d − 1 and λ ≥ 2d/3, then Nα (m) ≥ m + k for every integer
m ≥ n − k.

Proof. Let AG be the adjacency

√ matrix of G; note that, since G is d-regular, the vector v1
with each component 1/ n is the top eigenvector of AG with eigenvalue d. Now, let J be
the n × n all-ones matrix, and consider the n × n symmetric matrix

M = (1 − α)I + αJ − 2αAG .

Since J = nv1 v1⊺ and AG are simultaneously diagonalizable, the spectrum of M consists
of one copy of 1 − α + αn − 2αd > 0 corresponding to the all-ones vector and a copy of
(1 − α) − 2αµ for every eigenvalue µ < d of AG , with multiplicity equal to that of µ. Since
λ2 (AG ) = λ = (1 − α)/(2α), this means that M is positive semidefinite and has eigenvalue
0 with multiplicity k. So, M is a Gram matrix of n vectors w1 , . . . , wn ∈ Rn−k . By its
definition, every diagonal entry of M is 1 and every off-diagonal entry is ±α. So, each wi is
a unit vector, and the lines spanned by distinct wi each meet at an angle of arccos α. This
finishes the proof of part (a).
We now show part (b). By the above argument, it suffices to show, for each integer
ℓ ≥ n, the existence of a positive semidefinite ℓ × ℓ matrix whose diagonal consists only of
ones, whose off-diagonal entries are each ±α, and whose rank is at most ℓ − k. We will
show that, if an n-vertex graph G exists as in the lemma statement, such a matrix exists
for each ℓ ∈ [n, 2n). From here, it suffices to note that, if such an n-vertex graph G = G0
exists, there is a 2i n-vertex graph Gi also
√ satisfying the necessary properties for all i ≥ 0:
by Theorem 2.1 and the fact that λ ≥ 2 d − 1, we can pick Gi+1 to be a suitable 2-lift of
Gi .
Now, consider the n × n matrix M = (1 − α)I + αJ − 2αAG described above; it is positive
semidefinite, and the all-ones vector is an eigenvector of M with eigenvalue αn−(2αd−1+α).
For each integer ℓ ∈ (n, 2n), define the ℓ × ℓ matrix Mℓ by appending ℓ − n rows and columns
to M , filling the diagonal with ones and all off-diagonal entries with α. If M v = 0, then
v ⊕ (0, . . . , 0) is an eigenvector of Mℓ with eigenvalue 0, so Mℓ has rank at most ℓ − k.
Moreover, the diagonal entries of Mℓ are all 1 and the off-diagonal entries are ±α. So, it
suffices to prove that Mℓ is positive semidefinite. For each r, let 1r be the all-ones vector of
length r, and let t = ℓ − n; we need to show
  

M α1n 1⊺t u
u v
⊺ ⊺
α1t 1⊺n α1t 1⊺t + (1 − α)It v
= u⊺ M u + 2α(u⊺ 1n )(v ⊺ 1t ) + α(v ⊺ 1t )2 + (1 − α)v ⊺ v ≥ 0
 EQUIANGULAR LINES AND LARGE MULTIPLICITY OF FIXED SECOND EIGENVALUE 19

for each u ∈ Rn and v ∈ Rt . Indeed, since 1n is an eigenvector of M with eigenvalue

1 − α + αn − 2αd,
(u ⊕ v)⊺ Mℓ (u ⊕ v) = u⊺ M u + 2α(u⊺ 1n )(v ⊺ 1t ) + α(v ⊺ 1t )2 + (1 − α)v ⊺ v
   
2αd − (1 − α) 2 2 1−α
≥ α− (u 1n ) + 2αxy + αy +
⊺
(v ⊺ 1t )2
n t
   
2αd − (1 − α) 1−α
= α− x2 + 2αxy + α + y2,
n t
where we have set x = u⊺ 1n and y = v ⊺ 1t , and used that M is positive semidefinite. It thus
suffices that   
2αd − (1 − α) 1−α
α− α+ ≥ α2 .
n t
The left side is decreasing in t, so we need only verify the inequality when t = n. Upon
substituting (1 − α)/α = 2λ, the sufficient condition becomes
  
2(d − λ) 2λ
1− 1+ ≥ 1.
n n
λ 2λ
This simplifies to d−λ −1 ≥ n ; the facts that λ ≥ 2d/3 and n ≥ 2d ≥ 2λ suffice to verify
this inequality. 

Remark 6.5. It is not generally true that Nα (n + 1) ≥ Nα (n) + 1. In fact, N1/3 (7) = · · · =
N1/3 (14) = 28 [GSY21].
Finally, we prove Theorem 1.4.

Proof of Theorem 1.4. Let u ≥ 105 be an integer, and let

1
α= √ .
2
4 u +1−1
We will show that n + Ω(log log n) ≤ Nα (n) ≤ n + O(n/ log log n).
√
Let λ = (1 − α)/(2α) = 2 u2 + 1 − 1. The upper bound on Nα (n) follows from Theo-
rem 1.2 and Corollary 6.3: since u2 + 1 is not a square, Corollary 6.3 gives that k(λ) = ∞,
so Theorem 1.2(b) implies Nα (n) = n + O(n/ log log n). For the lower bound, we use the
graphs from Theorem 5.8. For some d ≤ 5u/2 ≤ 3λ/2 and some sequence n0 < n1 < · · ·
satisfying ni ≤ 4n9i−1 , there exist d-regular graphs on ni vertices with second eigenvalue λ
of multiplicity Ω(log log ni ). Defining f : N → R by f (n) = 0 for n < n0 and f (n) = i
whenever ni ≤ n < ni+1 , we have f (n) = Ω(log log n). Since λ ≥ 2d/3, Lemma 6.4 gives
that Nα (n) ≥ n + i as long as n ≥ ni − i; this implies
Nα (n) ≥ n + f (n) ≥ n + Ω(log log n),
as desired. 

Acknowledgment
The author would like to thank Yufei Zhao for his mentorship, for suggesting the problem,
and for providing helpful comments and suggestions.
20 CARL SCHILDKRAUT

References
[BL06] Yonatan Bilu and Nathan Linial. Lifts, Discrepancy and Nearly Optimal Spectral Gap. Combi-
natorica, 26(5):495–519, Oct. 2006. doi:10.1007/s00493-006-0029-7.
[CDG82] Dragoš Cvetković,
√ Michael Doob, and Ivan Gutman. On graphs whose spectral radius does not
exceed (2 + 5)1/2 . Ars Combinatoria, 14:225–239, 1982.
[CR90] D. Cvetković and P. Rowlinson. The largest eigenvalue of a graph: A survey. Linear and Multi-
linear Algebra, 28(1-2):3–33, Oct. 1990. doi:10.1080/03081089008818026.
[dC00] D. de Caen. Large Equiangular Sets of Lines in Euclidean Space. The Electronic Journal of
Combinatorics, 7:R55, Nov. 2000. doi:10.37236/1533.
[GSY21] Gary R. W. Greaves, Jeven Syatriadi, and Pavlo Yatsyna. Equiangular Lines in Low Dimensional
Euclidean Spaces. Combinatorica, 41(6):839–872, Dec. 2021. doi:10.1007/s00493-020-4523-0.
[HSZZ22] Milan Haiman, Carl Schildkraut, Shengtong Zhang, and Yufei Zhao. Graphs with high second
eigenvalue multiplicity. Bulletin of the London Mathematical Society, 54(5):1630–1652, 2022.
doi:10.1112/blms.12647.
[JP20] Zilin Jiang and Alexandr Polyanskii. Forbidden Subgraphs for Graphs of Bounded Spectral Ra-
dius, with Applications to Equiangular Lines. Israel Journal of Mathematics, 236(1):393–421,
Mar. 2020. doi:10.1007/s11856-020-1983-2.
[JTY+ 21] Zilin Jiang, Jonathan Tidor, Yuan Yao, Shengtong Zhang, and Yufei Zhao. Equian-
gular lines with a fixed angle. Annals of Mathematics, 194(3):729–743, Nov. 2021.
doi:10.4007/annals.2021.194.3.3.
[LS73] P. W. H Lemmens and J. J Seidel. Equiangular lines. Journal of Algebra, 24(3):494–512, Mar.
1973. doi:10.1016/0021-8693(73)90123-3.
[MRS21] Theo McKenzie, Peter Michael Reichstein Rasmussen, and Nikhil Srivastava. Support of closed
walks and second eigenvalue multiplicity of graphs. In Proceedings of the 53rd Annual ACM
SIGACT Symposium on Theory of Computing, STOC 2021, pages 396–407, New York, NY,
USA, June 2021. Association for Computing Machinery. doi:10.1145/3406325.3451129.
[MSS15] Adam Marcus, Daniel Spielman, and Nikhil Srivastava. Interlacing families I: Bipartite
Ramanujan graphs of all degrees. Annals of Mathematics, pages 307–325, July 2015.
doi:10.4007/annals.2015.182.1.7.
[Rob62] Raphael M. Robinson. Intervals containing infinitely many sets of conjugate algebraic integers.
In Studies in mathematical analysis and related topics, pages 305–315. Stanford Univ. Press,
Stanford, Calif., 1962.

Department of Mathematics, Massachusetts Institute of Technology, Cambridge, MA 02139,

USA
Email address: [email protected]