
RANDOM FLAT BUNDLES AND EQUIDISTRIBUTION 3
Corollary 1.4. Suppose r≥1and a= (a1,...,ar)and b= (b1,...,br)are r-tuples of non-negative integers. Then
for Xany cusped hyperbolic surface, asymptotically almost surely as n→ ∞,χ:π1(X)→U(n)satisfies
1
πX(T)X
ℓ(γ)≤T
r
Y
j=1
tr(χ(γj))ajtr(χ(γj))bj=δa,bZCr|z1|2a1...|zr|2ar
r
Y
j=1
1
je−π|zj|2/j dxjdyj+O(e−T /4)
as T→ ∞, where δa,bis the Kronecker delta.
This says essentially that the (tr(χ(γ)),tr(χ(γ2)),...,tr(χ(γr))) distribute according to the law of rindependent com-
plex Gaussians with a good error term. The main term on the right hand side is the expectation of Qr
j=1 tr(gj)ajtr(gj)bj
over U(n). Explicitly, we have
ZCr|z1|2a1...|zr|2ar
r
Y
j=1
1
je−π|zj|2/j dxjdyj=
r
Y
j=1
jajaj!.
A more difficult problem is the study of the rate at which equidistribution happens. This could be made precise by
bounding the lengths of the geodesic in terms of n. We could ask the following question.
Question 1.5.Given signature λ6= 0, is there a constant C(X, λ)>0 such that
P
χ:π1(X)→U(n) :
1
πX(T(n)) X
ℓ(γ)<T (n)
χλ(χ(γ))≤C(X, λ)e−c(n)
n→∞
−−−−→ 1
for some functions T(n), c(n)>0 going to infinity as n→ ∞? What are the optimal such functions?
Note that given a signature λof length m, we may extend it by zeros to obtain signatures of lengths nfor every
n≥m. This is how a fixed signature λgives rise to an infinite family of irreducible unitary representations of U(n)
for large enough n.
Definition 1.6. A signature λis unbalanced if the sum of its entries is nonzero. Otherwise, λis balanced.
In relation to Question 1.5, we prove the following theorem whose proof occupies the bulk of this paper.
Theorem 1.7. Suppose Xis a cusped hyperbolic surface, and suppose λ(n)is sequence of unbalanced signatures such
that |λ(n)| ≤ clog n
log log nfor large enough n, where c > 0is the universal constant in Theorem 1.2. Suppose T(n)is a
function of nsuch that T(n)→ ∞ as n→ ∞. For every ε > 0,
(1) P
χ:
1
πX(log T(n)) X
ℓ(γ)≤log T(n)
χλ(n)(χ(γ))≤n2dim Vλ(n)T(n)−1
4+ε
n→∞
−−−−→ 1.
A reason we impose the condition that the signatures λ(n) are unbalanced is the following. For unbalanced signatures
λ(n), we show that almost surely, πλ(n)◦χgives a flat bundle Eχ,λ(n)that is non-singular, that is, whose spectrum has
no continuous part. For balanced signatures λ(n) whose sum of entries is zero, the irreducible representations πλ(n)
are such that for every A∈U(n), πλ(n)(A) has 1 as an eigenvalue. Since Xis cusped hyperbolic, this means that
for such representations, we always have a continuous part in the spectrum of Eχ,λ(n). The existence of a continuous
spectrum requires a study of the main term of the hyperbolic scattering determinants of such flat bundles. The study
of the hyperbolic scattering determinant in this setting is closely related to currently intractable problems in random
matrix theory. Furthermore, in order to obtain Theorem 1.7 for balanced λ(n), we also need a bound in terms of
Xand λ(n) on the real parts of the resonances of Eχ,λ(n), at least asymptotically almost surely as n→ ∞. This
is also related to understanding the main term of the hyperbolic scattering determinants. We briefly discuss these
complications in Section 8.
1.1. Relation to other works. In Hide–Magee [HM21], the authors show that a cusped hyperbolic surface is such
that, for any ε > 0, if one takes n-sheeted covers of it, as n→ ∞, asymptotically almost surely they do not introduce
new eigenvalues less than 1
4−ε. In particular, if one starts with a cusped hyperbolic surface with spectral gap at
least 1
4, then one obtains that n-sheeted covers asymptotically almost surely have spectral gaps at least 1
4−ε. A
compactification argument allows them to compactify their sequence of hyperbolic surfaces to show the existence of
an infinite family of connected closed hyperbolic surfaces {Xn}nwith increasing genera and spectral gaps λ1(Xn)
satisfying
lim sup
n→∞
λ1(Xn) = 1
4.