Biorthogonal polynomials related to quantum transport theory of

disordered wires
Dong Wang∗

July 10, 2023

Abstract
We consider the Plancherel-Rotach type asymptotics of the biorthogonal polynomials
arXiv:2307.03720v1 [math-ph] 7 Jul 2023

associated to the biorthogonal ensemble with the joint probability density function
n
1 Y Y
(λj − λi )(f (λj ) − f (λi )) Wα(n) (x)dx,
C j=1
1≤i<j≤n

where
√
f (x) = sinh2 ( x), Wα(n) (x) = xα h(x)e−nV (x) .

In the special case that the potential function V is linear, this biorthogonal ensemble arises
in the quantum transport theory of disordered wires. We analyze the asymptotic problem
via 2-component vector-valued Riemann-Hilbert problems, and solve it under the one-cut
regular with a hard edge condition.

1 Introduction
1.1 Setup of the model
Let
1  2 √x √  √
f (x) =e − 2 + e−2 x = sinh2 ( x) (1.1)
4
On [0, ∞). Let V be a real analytic function on [0, ∞) and h be a positive valued real analytic
function h(x) on [0, ∞), satisfying the limit condition
V (x)
lim √ = +∞. (1.2)
x→+∞ max(log h(x), x + 1)

(n)
We then denote the weight function Wα (x), depending on the parameter α > −1 and , as

Wα(n) (x) = xα h(x)e−nV (x) . (1.3)

(n) (n)
We consider the monic polynomials pj (x) and qj (x), of degree j ≥ 0, determined by the
orthogonality conditions
Z Z
(n) k (n) (n)
pj (x)f (x) Wα (x)dx = 0, xk qj (f (x))Wα(n) (x)dx = 0, k = 0, 1, . . . , j − 1, (1.4)
R+ R+
∗
School of Mathematical Sciences, University of Chinese Academy of Sciences, Beijing, P. R. China 100049
email: wangdong@wangd-math.xyz

1
and define Z
(n) (n) (n)
hj = pj (x)qj (f (x))Wα(n) (x)dx. (1.5)
R+
(n) (n)
Because pj and qj satisfy the biorthogonal condition (1.4), they are called the biorthogonal
(n)
polynomials with respect to Wn .
In this paper, we are mainly concerned about the Plancherel-Rotach type asymptotics of
(n) (n)
pn+k (x) and qn+k (f (x)), as n → ∞ and k a fixed integer.
These biorthogonal polynomials are related to the point process consisting of n particles at
λ1 , . . . , λn ∈ [0, ∞), with the joint probability density function (pdf)
n
1 Y Y
(λj − λi )(f (λj ) − f (λi )) Wα(n) (x)dx. (1.6)
C
1≤i<j≤n j=1

Below we call the point process defined by (1.6) the biorthogonal ensemble.
(n) (n)
The first relation between pj , qj and the biorthogonal ensemble (1.6) is:

Proposition 1. p(n) (x) and q (n) (f (x)) have the representation

   
Yn n
Y
p(n)
n (z) = E
 (z − λj ) , qn(n) (f (z)) = E  (f (z) − f (λj )) , (1.7)
j=1 j=1

where E[·] is with respect to the joint pdf (1.6).

The proof of the proposition is analogous to the proofs of [5, Proposition 2.1] and [11,
Proposition 1], and we omit it. The most important consequence of this proposition in our
(n) (n) (n) (n)
paper is the existence and uniqueness of pn , qn , as well as pj , qj with general j, if we allow
(n)
the weight function Wα to be modified by a simple scaling transform.
(n) (n)
The next and more important relation between pj , qj and the biorthogonal ensemble
(1.6) is:

Proposition 2. The biorthogonal ensemble with joint pdf (1.6) is a determinantal point process,
and the following Kn (x, y) is its correlation kernel
q n−1
X 1 (n)
(n) (n) (n)
Kn (x, y) = Wα (x)Wα (y) p (x)qj (f (y)).
(n) j
(1.8)
j=0 hj

This result is based on the general theory of determinantal point processes, see [34] for the
framework and [6, Section 2] for a short discussion on the kernel formula of a similar model.
By Proposition 2, the Plancherel-Rotach type asymptotic result also implies the local lim-
iting distribution of particles in this biorthogonal ensemble.

1.2 Motivation
Our study of the biorthogonal ensemble defined by the general weight function (1.3) is motivated
by the concrete case with the weight function specialized by
 √ √ 1
x ′ 1 sinh( x) cosh( x) 2
V (x) = , α = 0, and h(x) = f (x) =
2 √ , (1.9)
M x

2
which is proposed in the study of the quantum transport theory of a disordered wire. See [3,
Formula (19)] and [4, Section III A, Formula (3.4)] for the physical derivation of the biorthogonal
ensemble. (In [4], M is denoted as s and n is denoted as N ). We also refer the interested readers
to the review article [2] for the physical theory that relates the biorthogonal ensemble with
specialization (1.9). Here we only mention that in the regime that n is large while 1 ≪ M ≪ n,
the distribution of particles λ1 , . . . , λn is (approximately) the distribution of the transmission
eigenvalues of the disordered wire, and the sum
n
X 1
Cn (M) := 2
p (1.10)
j=1
cosh ( λj )

is the conductance of the disordered wire. Results from experimental physics implies the fol-
lowing mathematical result:

• Ohm’s law
1
lim M lim E[Cn (M)] = 1, (1.11)
M→∞ n→∞ n

• Universal conductance fluctuation

1
lim lim Var[Cn (M)] = . (1.12)
M→∞ n→∞ 15

The framework of biorthogonal polynomials developed in our paper enables us to rigorously

analyze the limiting distribution of λ1 , . . . , λn , and prove (1.11) and (1.12). We remark that the
universal conductance fluctuation (1.12) has only been justified by physical argument [28], [26],
[29], while the counterpart of (1.12) for a simpler model, the quantum dot, has been rigorously
proved in [1], [20], [19], [33], and also see [21]. The Plancherel-Rotach type asymptotic result in
this paper establishes the foundation to rigorously prove the universal conductance fluctuation
(1.12).

1.3 Main results

1.3.1 Global results: Qualitative properties of equilibrium measure
Analogous to the determinantal point processes associated to orthogonal polynomials, (that is,
replacing f (λj ) − f (λi ) by (λj − λi ) in the joint pdf (1.6),) we define the equilibrium measure
supported on [0, ∞) as the minimizer of the functional
ZZ ZZ Z
1 −1 1 −1
IV (µ) := log|t − s| dµ(t)dµ(s) + log|f (t) − f (s)| dµ(t)dµ(s) + V (s)dµ(s).
2 2
(1.13)

Proposition 3. Let V be a continuous function on [0, ∞) that satisfies (1.2) with h(x) = 1.
Then there exists a unique measure µ = µV on R+ with compact support which minimizes the
functional among all Borel probability measures on R+ .

The basic idea of the proof is contained in [15], which gives a detailed proof of the result
if f (x) is changed into x and the integer domain is changed into R. If f (x) is changed into ex
and the integral domain is changed into R, an explanation is given in [11, Section 2]. Hence we
omit the proof here.

3
Remark 1. Proposition 3 is only for the existence and uniqueness of µ = µV , and it does not
tell us how to construct µ. Furthermore, it is only a potential theoretical result, and does not
have direct relation to either the n-particle biorthogonal ensemble (1.6) or the biorthogonal
(n) (n)
polynomials pn+k , qn+k . Although we can follow the argument given in [15, Chapter 6] to
establish the relations mentioned above, we do not pursue this approach.
We say that a potential function V satisfying (1.2) is “one-cut regular with a hard edge”
if there exists an absolutely continuous measure µ = µV such that dµ(x) = ψ(x) on R+ , that
satisfies:
Requirement 1.
R
1. supp µ = [0, b] for b > 0 that depends on V , and dµ(x) = 1.

2. dµ(x) = ψ(x)dx, where ψ(x) is continuous on (0, b) and ψ(x) > 0 for x ∈ (0, b).

3. For x ∈ [0, b], there exists a constant ℓ depending on V such that

Z Z
log|t − x|−1 dµ(t) + log|f (t) − f (x)|−1 dµ(t) + V (x) + ℓ = 0. (1.14)

4. For x > b,
Z Z
−1
log|t − x| dµ(t) + log|f (t) − f (x)|−1 dµ(t) + V (x) + ℓ > 0. (1.15)

5. The two limits

1 1
ψ0 := lim x 2 ψ(x) and ψb := lim (b − x)− 2 ψ(x) (1.16)
x→0+ x→b−

exist and both are positive.

It is clear that Item 1 implies that µ is a probability measure, and Item 2 further implies that
dµ(x) = ψ(x)dx is a “one-cut” probability measure in the sense that its support is a compact
interval and its probability density is positive everywhere in the interior of the support. Items 3
and 4 are slightly stronger than the Euler-Lagrange equation of the variational problem (1.13),
so if ψ(x) satisfies all Items 1, 2, 3 and 4, then dµ(x) = ψ(x)dx is the unique equilibrium
measure defined by the minimization of IV , as stated in Proposition 3. At last, Item 5 means
that the equilibrium measure is regular at 0, the “hard edge”, and b, the “soft edge”. (The
regularity of the equilibrium measure also includes Item 2 that means it is regular in the interior
of the support, and includes Item 4 that means it is regular out of the support.) Throughout
this paper, we only consider potential functions V that is one-cut regular with a hard edge.
Given a potential function V , generally it is hard to determine if V is one-cut regular with
a hard edge. In this paper, we are satisfied with the following partial result:

Theorem 4. If the potential function V is real analytic on [0, ∞), satisfies (1.2), and
√
U ′ (x) > 0 for all x ∈ (0, ∞), where U (x) = V ′ (x) x. (1.17)

then V is one-cut regular with a hard edge.

The proof of the theorem is given in Section 2.

4
1.3.2 Global results: quantitative properties of equilibrium measure
Given a one-cut regular with a hard edge potential V , it is still challenging to find the equilibrium
measure µV , even its right end point b. If V satisfies the conditions in Theorem 4, then b and
ψ(x) can be computed in principal, but we need to define a few functions to state the result.
First we collect some properties of function f defined in (1.1).

Lemma 5. f (x) defined in (1.1) has a natural extension into an analytic function on C, and
it satisfies the follows:

• f (x) ∈ R+ for all x ∈ R+ , and f (x) increases from 0 to +∞ as x runs from 0 to +∞.

• f (x) ∈ (−1, 0) for x ∈ (−π 2 /4, 0), and f (x) increases from −1 to 0 as x runs from −π 2 /4
to 0.

• As x ∈ R+ runs from 0 to +∞, f ( 41 (x2 − π 2 ) ± 12 πxi) ∈ R− and it decreases from −1 to

−∞.

Next, we define ρ the curve lying in C− with formula

t2 π2
ρ={ − − it | t ∈ (0, +∞)}. (1.18)
π2 4
Then ρ ∪ {0} ∪ ρ̄ is a parabola. We define P as a region of C to the right of the parabola. From
Lemma 5, we have that f : interior of P → C \ (−∞, −1] is conformal, while f maps both ρ and
ρ̄ to (−∞, −1]. Actually, if we glue ρ and ρ̄ by identifying z ∈ ρ with z̄ ∈ ρ̄ and view P as a
Riemann surface, then f is an conformal mapping between P and C.

2
− π4 0

Figure 1: Shape of P (the region the the right of the parabola).

We define, for all x ∈ R+ , the transformation

√
 
s+1 1
Jx (s) = x s + arcosh , and Jx (s) = (Jx (s))2 , s ∈ C+ ∪ (1, ∞), (1.19)
s−1 4
√
such that s takes the principal branch on arg s ∈ (−π, π), and arcosh takes the branch that
is the one-to-one mapping from C \ (−∞, 1] to {s ∈ C | ℜs > 0 and − π < ℑs < π}. Below
we extend the domain of Jx (s) and Jx (s). Naturally, They extend to s ∈ C− by Jx (s̄) = Jx (s)
and Jx (s̄) = Jx (s)). Jx (s) also extends to s ∈ (−∞, 0) by continuation. For s ∈ [0, 1], we leave
both Jx (s) and Jx (s) undefined, since they have a branch cut there.

Lemma 6. Jx (s) satisfies the following properties:

5
 1
p
1. Jx (s) ∈ R+ for s ∈ (1, +∞); Jx (s) decreases from +∞ to Jx (s2 (x)) = 2 (x + 1)2 − 1 +
arcosh(x + 1) as s runs from 1 to

s2 (x) = 1 + 2/x, (1.20)

and then it increases from Jx (s2 (x)) to +∞ as s runs from s2 (x) to ∞.

2. Jx (−s) ∈ iR for s ∈ (0, +∞). ℑJx (−s) increases monotonically from −π to +∞ as s
runs from 0 to ∞, and there is a unique s1 (x) ∈ (−∞, 0) such that Jx (s1 (x)) = 0. To be
precise, s1 (x) is the unique solution on (−∞, 0) of
√ −s − 1
 
x −s = arccos . (1.21)
−s + 1

3. Jx (s) + πi ∈ R+ for x ∈ (0, 1). ℜJx (x) increases from 0 to +∞ as s runs from 0 to 1.
4. There is a unique curve γ1 (x) ⊆ C+ connecting s1 (x) and s2 (x), such that Jx (z) ∈ R as
z ∈ γ1 (x), and Jx (z) increases from 0 to Jx (s2 (x)) as z moves from s1 (x) to s2 (x) along
γ1 (x).
Part 4 of this lemma will be proved in Appendix A. Parts 1, 2 and 3 can be verified by
direct computation, and we omit the detail.
Define
γ2 (x) = {z ∈ C− | z̄ ∈ γ1 (x)}, (1.22)
Then γ1 (x), γ2 (x) together enclose a region Dx ⊆ C. In this paper, we orient γ1 (x), γ2 (x) from
s1 (x) to s2 (x), unless otherwise stated.
Lemma 7. Jx maps C \ Dx conformally to C \ [0, b(x)] and maps Dx \ [0, 1] conformally to
P \ [0, b(x)].
This lemma will be proved in Appendix A.
For each x ∈ (0, ∞), we let
1 1 p 2
b(x) = Jx (s2 (x)) = Jx (s2 (x))2 = (x + 1)2 − 1 + arcosh(x + 1) . (1.23)
4 4
It is clear that b(x) is a continuous function of x, and it increases monotonically from 0 to ∞
as x runs from 0 to ∞.
Let the functions Ix,1 and Ix,2 be inverse functions of Jx , such that Ix,1 is the inverse map of
Jx from C \ [0, b(x)] to C \ Dx , and Ix,2 is the inverse map of Jx from P \ [0, b(x)] to Dx \ [0, 1]:

Ix,1 (Jx (s)) = s, s ∈ C \ Dx , (1.24)

Ix,2 (Jx (s)) = s, s ∈ Dx \ [0, 1]. (1.25)

We then denote for u ∈ (0, b(x))

Ix,+ (u) := lim Ix,1 (u + iϵ) = lim Ix,2 (u − iϵ), (1.26)

ϵ→0+ ϵ→0+

Ix,− (u) := lim Ix,1 (u − iϵ) = lim Ix,2 (u + iϵ). (1.27)

ϵ→0+ ϵ→0+

We have that Ix,+ (x) lies in C+ , Ix,− (x) lies in C− , and their loci are the upper and lower
boundaries of Dx , that is, γ1 (x) and γ2 (x) respectively. For later use, we define for ξ ∈ (0, b(x))
 p 
 ( I (u) + pI (ξ))(pI (u) − pI (ξ)) 
x,+ x,+ x,− x,+
Fx (u; ξ) = log  p . (1.28)
 
p p p
 ( Ix,+ (u) − Ix,+ (ξ))( Ix,− (u) + Ix,+ (ξ)) 

6
 γ1
w=Jc (s) (1) z=w2 /4
=⇒ =⇒
(1) (2) (3) (4) (5) Jc (γ1 ) (5) (1) Jc (γ1 ) (5)
s1 s2 √
0 1 0 = Jc (s1 ) 2 b =(4) π2 (2) 0 = Jc (s1 ) b = (4)
(2) − =
Jc (s2 ) 4 Jc (s2 )
(3) (3)
Jc (0)
−πi = Jc (0)

Figure 2: The schematic illustration of Jc and Jc on C+ . (The definition of Jc and Jc is

extended to C− naturally
p by complex conjugation.) If c is changed to a general x ∈ (0, ∞),
then Jx (s2 (x)) = 2 b(x) and Jx (s2 (x)) = b(x) will change, while Jx (s1 (x)) = Jx (s1 (x)) = 0,
Jx (0) = πi and Jx (0) = −π 2 /4 are unchanged.

Jc
=⇒
s1 D s2
0 b

Jc
=⇒
s1 0 D 1 s2
2
− π4 0 b

Figure 3: Jc maps C \ D to C \ [0, b] and maps D \ [0, 1] to P \ [0, b].

The graphic illustration is given in Figures 2 and 3. (To simplify the notation, we assume
x = c in Figure 3; see (1.30).)
The definition of Jx and Jx is independent of V . Suppose V is one-cut regular with a
hard edge, such that the equilibrium measure µV associated to V is supported on [0, b] and the
density function is ψ(x). We let c > 0 be the unique solution to
p √
b(x) = b, or equivalently, (x + 1)2 − 1 + arcosh(x + 1) = 2 b. (1.29)

Throughout this paper, we denote

si = si (c), γi = γi (c), D = Dc , Ii (u) = Ic,i (u), I± (u) = Ic,± (u), F (u; ξ) = Fc (u; ξ).
(1.30)

Below we state the result on the constructive description of the equilibrium measure µ
introduced in Proposition 3, under the assumption (1.17) in Theorem 4.
Theorem 8. Suppose V satisfies (1.17) in Theorem 4.
1. The parameter c is the unique solution to the equation in x ∈ (0, ∞)
J′x (ξ)V ′ (Jx (ξ))
I
1 1
dξ = , (1.31)
2πi γ(x) ξ − s2 (x) s2 (x) − 1

7
 where γ(x) = γ1 (x) ∪ γ2 (x) is the boundary of Dx , with positive orientation. Hence b, the
right-end point of the support of the equilibrium measure µ, is determined by b = b(c).

2. The density function ψ(x) of the equilibrium measure µ is determined by

Z b
1
ψ(x) = 2
√ U ′ (u)F (u; x)du, (1.32)
4π x 0

where U is defined in (1.17) and F (u; x) is defined by (1.30) and (1.28).

As a concrete example and a response to our motivation, we consider the case when V is a
linear function:

Corollary 9. When V (x) = x/M where M is a constant parameter, the parameter c = 2M,
and the end point
2
√

1p 1 p
b= (2M + 1)2 − 1 + arcosh(2M + 1) = ( M2 + M + arsinh M)2 . (1.33)
2 2

Furthermore, the density function ψ(x) is given by

r
1 I+ (x)
ψ(x) = ℑ . (1.34)
π x
In the limit that M → 0, we have that the support of the equilibrium measure is [0, 4M +
O(1)], and the the density function satisfies the limiting formula that for all ϵ > 0,
r
1 4M − x
ψ(x) = (1 + O(M)), x ∈ (ϵM, (4 − ϵ)M), (1.35)
2πM x
1 3
M− 2 M− 2 (1.36)
ψ0 = (1 + O(M)), ψb = (1 + O(M)).
π 4π
In the limit that M → ∞, we have that the support of the equilibrium measure is [0, M2 +O(M)],
and the density function satisfies the limiting formula that for all ϵ > 0,
1
ψ(x) = √ (1 + O(M−1 )), x ∈ (ϵM2 , (1 − ϵ)M2 ), (1.37)
2M x
1 1
ψ0 = (1 + O(M)), ψb = √ 5 (1 + O(M)). (1.38)
2M 2πM 2
Theorem 8 and Corollary 9 will be proved in Section 2.

1.3.3 Local results: Plancherel-Rotach asymptotics and universality of the hard

edge
For the local results of the biorthogonal polynomials, we assume that both V (x) and h(x) are
real analytic in a neighbourhood of R+ .
(n) (n) (n) (n) (n)
Since both pj (z) and qj (f (z)) are real analytic functions and pj (z̄) = pj (z), qj (f (z̄)) =
(n)
qj (f (z)), we only need to give their asymptotics in the upper half plane and the real line. To
be precise, we let δ > 0 be a small enough constant, and let Cδ = {z ∈ C+ ∪ R | |z| ≤ δ}
and Dδ = {z ∈ C+ ∪ R | |z − b| ≤ δ} be the two semicircles centred at 0 and b respectively,
Bδ = {z ∈ C ∪ R | ℑz ≤ δ/2 and |z| > d and |z − b| > d}, and Aδ = (C ∪ R) \ (Bδ ∪ Cδ ∪ Dδ }.

8
 Aδ
Cδ Bδ Dδ
0 b
(n)
Figure 4: The four regions in the upper complex plane where the asymptotics of pn+k (z) and
(n)
qn+k (f (z)) are given.

See Figure 4 to see the shapes of the regions. We assume that V (z) and h(z) are analytic in
B δ ∪ Cδ ∪ D δ .
Let ψ(x) be the density function of µV on (0, b). We then define the functions
Z b Z b
g(z) := log(z − x)ψ(x)dx, g̃(z) := log(f (z) − f (x))ψ(x)dx, (1.39)
0 0

with the branch cut of the logarithms for z ∈ (−∞, x) and f (z) ∈ (−∞, f (x)) respectively. Let

ϕ(z) = g(z) + g̃(z) − V (z) − ℓ (1.40)

for z ∈ P \ (0, b), where ℓ is a constant to make ϕ(0) = ϕ(b) = 0. (See (1.14) and (1.15).) Then
we will see later on, see Section 3.6, that
 2
3 3 2
fb (z) := − ϕ(z) , with fb (b) = 0 and fb′ (b) = (πψb ) 3 > 0 (1.41)
4

is a well defined analytic function in a certain neighbourhood of b. Similarly,

1
f0 (z) = (ϕ(z) ∓ πi)2 , with f0 (0) = 0 and f0′ (0) = −(πψ0 )2 < 0 (1.42)
16
(where the sign is − in C+ and − in C− ), is a well defined analytic function in a certain
neighbourhood of 0.
Recall the contours γ1 , γ2 defined in (1.30), γ = γ1 ∪ γ2 with positive orientation, and h(z)
is a real analytic function on [0, ∞). Let γ ′ and γ ′′ be positively oriented contours such that
γ ′ encloses γ, and γ ′′ is enclosed by γ, such that h(Jc (s)) is well defined and analytic in the
annular region between γ ′ and γ ′′ . We also assume that {I1 (z), I2 (z) : z ∈ Bδ ∪ Cδ ∪ Dδ } lies
inside the annular region. Then define
 I 
1 dζ
D(s) = exp log h(Jc (ζ)) , z is outside γ ′′ , (1.43)
2πi γ ′′ ζ −s
−1
 I 
dζ
D̃(s) = exp log h(Jc (ζ)) , z is inside γ ′ . (1.44)
2πi γ ′ ζ −s

We have that between γ ′ and γ ′′ , both Dk (s) and D̃k (s) are defined, and

D(s)D̃(s) = h(Jc (s))−1 , z is between γ ′ and γ ′′ . (1.45)

We define
2 1 1
( c )α+ 2 +k (s − s1 )α+1 s 2 (s − 1)k
Gk (s) = 4 1p D(s), s is outside γ ′′ and not on γ1 , (1.46)
Jc (s)α+ 2 (s − s1 )(s − s2 )

9
 1 1
where (s − s1 )(s − s2 ) is analytic in C \ γ1 and is ∼ s as s → ∞, and (s − s1 )α+1 s 2 /Jc (s)α+ 2
p

is analytic in {z ∈ C : outside of γ ′′ and not on γ1 }, and is ∼ (4/x2 )α+1/2 as s → ∞. We also

define
1√
(1 − s1 )−α− 2 s2 − 1iD̃(1)−1
G̃k (s) = p D̃(s). s is inside γ ′ and not on γ2 , (1.47)
(s − s1 )α (s − 1)k (s − s1 )(s − s2 )

where (s − s1 )(s − s2 ) is analytic in C \ γ2 and is ∼ s as s → ∞, and (s − s1 )α is analytic in

p

C \ (−∞, s1 ], and takes the principal branch.

p
in the special case α = 0 and h(z) = f ′ (z) as in (1.9), we have the explicit formulas
1  2 k
√
r
s − 1 s4 c
Gk (s) = c (s − 1) , z ∈ C \ D̄, (1.48)
s − s1 Jc (s) 21 4
1
2i s4 (s − 1)−k
G̃k (s) = √ p , z ∈ D \ [0, 1], (1.49)
c sinh(Jc (s)) (s − 1)(s − s2 )

where all power functions take the principal branch, and Gk (s) and G̃k (s) are then extended to
their domains in (1.46) and (1.47) respectively.
Based on Gk and G̃k , we then define

rk (x) = 2|Gk,+ (I+ (x))|, θk (x) = arg(Gk,+ (I+ (x))), (1.50)

r̃k (x) = 2|G̃k,+ (I− (x))|, θ̃k (x) = arg(G̃k,+ (I− (x))), (1.51)

where Gk,+ (I+ (x)) is the limit of Gk (z) as z approaches I+ (x) ∈ γ1 in C \ D, and G̃k,+ (I− (x))
is the limit of G̃k (z) as z approaches I− (x) ∈ γ2 in D.

Theorem 10. Let V be one-cut regular with a hard edge and assume V (z) is analytic in an
(n)
open set containing [0, +∞). As n → ∞, we have the following asymptotics of pn+k (z) and
(n)
qn+k (f (z)), k ∈ Z, uniformly for z in regions Aδ , Bδ , Cδ and Dδ , if δ > 0 is small enough.

1. In region Aδ we have
(n)
pn+k (z) = (1 + O(n−1 )Gk (I1 (z))eng(z) , z ∈ Aδ , (1.52)
(n)
qn+k (f (z)) = (1 + O(n−1 )G̃k (I2 (z))eng̃(z) , z ∈ Aδ ∩ P. (1.53)

2. In region Bδ we have
(n)
pn+k (z) = (1 + O(n−1 ))Gk (I1 (z))eng(z) + (1 + O(n−1 ))Gk (I2 (z))en(V (z)−g̃(z)+ℓ) ,
(1.54)
(n)
qn+k (f (z)) = (1 + O(n−1 ))G̃k (I2 (z))eng̃(z) + (1 + O(n−1 ))G̃k (I1 (z))en(V (z)−g(z)+ℓ) .
(1.55)

Especially, if x ∈ (δ, b − δ), we have

(n)
R
pn+k (x) = rk (x)en cos (nπµV ([x, b]) + θk (x)) + O(n−1 ) ,
log|x−y|dµV (y)
 
(1.56)
h   i
(n)
R
qn+k (f (x)) = r̃k (x)en log|f (x)−f (y)|dµV (y) cos nπµV ([x, b]) + θ̃k (x) + O(n−1 ) . (1.57)

10
3. In region Dδ we have
n (n)
e− 2 (g(z)−g̃(z)+V (z)+ℓ) pn+k (z) =
√
 1
1 2
π n 6 fb4 (z) (1 + O(n−1 ))Gk (I1 (z)) − (1 + O(1))iGk (I2 (z)) Ai(n 3 fb (z))


− 61 −1 −1

′ 2
−n fb 4 (z) (1 + O(n ))Gk (I1 (z)) + (1 + O(1))iGk (I2 (z)) Ai (n 3 fb (z)) ,
(1.58)

n (n)
e− 2 (g̃(z)−g(z)+V (z)+ℓ) qn+k (f (z)) =
√
 1  
1 2
π n 6 fb4 (z) (1 + O(n−1 ))G̃k (I2 (z)) − (1 + O(1))iG̃k (I1 (z)) Ai(n 3 fb (z))

−1
 
− 16 −1 ′ 2
−n fb 4 (z) (1 + O(n ))G̃k (I2 (z)) + (1 + O(1))iG̃k (I1 (z)) Ai (n 3 fb (z)) .
(1.59)

In particular, if z = b + fb′ (b)−1 n−2/3 t with t bounded, then

1 n (n)
n− 6 e− 2 (g(z)−g̃(z)+V (z)+ℓ) pn+k (z) =
 2 α+ 12  
1√ 1 3 1 c (s2 − s1 ) c k ′ 1  1

8
2 4 πb 8 s2 c 2 fb (b) 4 D(s2 ) Ai(t) + O(n− 3 ) , (1.60)
4b 2

1 n (n)
n− 6 e− 2 (g̃(z)−g(z)+V (z)+ℓ) qn+k (f (z)) =
3√ 1
 c k  b  81 1 D̃(s2 )
 1

−α−
2 4 π[(1 − s1 )(s2 − s1 )] 2 fb′ (b) 4 Ai(t) + O(n− 3 ) . (1.61)
2 s2 D̃(1)

Here Ai is the Airy function.

4. In region Cδ we have
n (n)
e− 2 (g(z)−g̃(z)+V (z)+ℓ) pn+k (z) =
√
 1
1
π n 2 f04 (z) (1 + O(n−1 ))Gk (I1 (z)) − (1 + O(1))ieαπi Gk (I2 (z)) Iα (2n f0 (z))

p


− 12 −1 −1 απi

p
+n f0 4 (z) (1 + O(n ))Gk (I1 (z)) + (1 + O(1))ie Gk (I2 (z)) Iα (2n f0 (z)) ,
(1.62)

n (n)
e− 2 (g̃(z)−g(z)+V (z)+ℓ) qn+k (f (z)) =
√
 1  
1
π n 2 f04 (z) (1 + O(n−1 ))G̃k (I2 (z)) − (1 + O(1))ie−απi G̃k (I1 (z)) Iα (2n f0 (z))
p


−1
 
− 21 −1 −απi
p
+n f0 4 (z) (1 + O(n ))G̃k (I2 (z)) + (1 + O(1))ie G̃k (I1 (z)) Iα (2n f0 (z)) .
(1.63)

11
 In particular, if z = −f0′ (0)−1 n−2 t, with t bounded, then

√
r
− 21 − n (g(z)−g̃(z)+V (z)+ℓ) α (n) −s1
n e 2 z 2 pn+k (z) =2 π
s2 − s1
√ α+ 1  2 k
c(1 − s1 ) −s1 √

2 c 1
 
× (s1 − 1) D(s1 )(−f0 (0)) 4 Jα (2 t) + O(n−1 ) , (1.64)
s2 − s1 4

√
r
− 21 − n (g̃(z)−g(z)+V (z)+ℓ) α (n) s2 − 1
n e 2 z 2 qn+k (f (z)) =2 π
s2 − s1
α+ 1
c(s2 − s1 ) √

2 D̃(s1 ) 1
 
× √ (s1 − 1)−k (−f0 (z)) 4 Jα (2 t) + O(n−1 ) . (1.65)
4(1 − s1 )2 −s1 D̃(1)

Here Jα is the Bessel function and Iα is the modified Bessel function.

5.
α+ 12  k+ 14
c2 c2

(n) 2π
hn+k = + O(n−1 ). (1.66)
D̃(1) 4(1 − s1 ) 4

The proof of the theorem will be given in Section 5 based on the Riemann-Hilbert analysis
in Sections 3 and 4.
Based on Theorem 10, we can use the method in [12] to show that the correlation kernel
Kn (x, y) in (1.8) has the limit as the Airy kernel around b, and the sine kernel in (0, b), upon
proper scaling transform and conjugation. Also we can show that Kn (x, y) has the limit as the
Bessel kernel, using the method in [35]. (In [35] the limit of the Muttalib-Borodin correlation
kernel is shown to converge to a Meijer-G kernel with a parameter θ. If θ = 1, the Meijer G
kernel specializes into the Bessel kernel.) Hence, the biorthogonal ensemble defined in (1.6) has
the desired limiting universal behaviour at the bulk (0, b), the soft edge b, and the hard edge 0.
We put off the proof of the above claims to a subsequent paper.

1.4 Related models and previous results

Orthogonal polynomials were related to Riemann-Hilbert problems by [16], [17], and then
the powerful (Deift-Zhou) nonlinear steepest-descent method was successfully applied to such
Riemann-Hilbert problems [14], [13], and opened the door to manifold of limiting results of
orthogonal polynomials and their generalizations, see [24] for a review. These various Riemann-
Hilbert problems are all matrix valued, some 2 × 2 and some of larger size.
In the study of a special kind of biorthogonal polynomials related to the random matrix
model with equispaced external source [11], Claeys and the author related the biorthogonal
polynomials to 2-component vector-valued Riemann-Hilbert problems, and applied the Deift-
Zhou method on them to find the limiting results. The biorthogonal polynomials in [11] has, in
our notation, f (x) = ex in (1.4) where the integral domain is replaced by R.
After [11] and before the current paper, the method of vector-valued Riemann-Hilbert prob-
lems was, to the author’s limited knowledge, only applied to the Muttalib-Borodin biorthogonal
polynomials [10], [35], [8]. The Muttalib-Borodin biorthogonal polynomials is characterized
by f (x) = xθ in (1.4), and the particle model given in (1.6) with f (x) = xθ is called the
Muttalib-Borodin ensemble.
We remark that Muttalib-Borodin ensemble was proposed by physicist Muttalib [31] as
a simplification of the biorthogonal ensemble considered in our paper, and the study of the

12
Muttalib-Borodin ensemble [6], [9], [18], [36] is partially motivated by its indirect relation to
the quantum transport theory of disordered wires.
We also remark that technically the Muttalib-Borodin ensemble is more challenging, for its
limit behaviour at the hard edge is the more complicated Meijer G kernel, rather than the Bessel
kernel.
At last, we note that biorthogonal ensembles are also investigated from other aspects, for
instance, [7], [27], [25], [23], [30].

Acknowledgements
The author was partially supported by the National Natural Science Foundation of China under
grant numbers 12271502 and 11871425, and the University of Chinese Academy of Sciences start-
up grant 118900M043. The author thanks K. A. Muttalib for discussion in the early stage of
this project, and thanks Tiefeng Jiang for the help in literature review.

2 Construction of the equilibrium measure

In this section we assume that V is a potential function that satisfies the conditions in Theorem
4, and show that V is one-cut regular with a hard edge at 0, by an explicit construction of its
equilibrium measure. Like in [11], we first give the support of the equilibrium measure as an
ansatz, then compute the density within the support of the equilibrium measure, and at last
verify that the measure constructed satisfies the criteria of one-cut regularity, and conclude that
it is the unique equilibrium measure.
At every step, we analyze the V = x/M case and get explicit formulas for this special case.

2.1 A technical lemma

Recall the mapping Jx defined in (1.19) and γ(x) defined in Part 1 of Theorem 8.
Lemma 11. Suppose V satisfies the condition required in Theorem 4. There is a unique x ∈
(0, ∞) such that the equation with unknown x (1.31) holds.
Proof. Using the formulas (1.19) and (1.20), we can rewrite (1.31) as
Jx (ξ)V ′ (Jx (ξ))
I
1
F (x) = 2, where F (x) = √ dξ. (2.1)
2πi γ ′ ξ(ξ − 1)

Here γ ′ can be γ(x), but can also be a slightly bigger contour circling γ(x), as long as V ′ (Jc (ξ))
√ √
is well defined there. Let U (z) = V ′ (z) z where z takes the principal branch. By direct
computation, we have I
′ 1 dξ
F (x) = U ′ (Jx (ξ))Jx (ξ) . (2.2)
4πi γ ′ ξ−1
Then taking γ ′ = γ(x), and change the variable y = Jx (ξ), we have
Z b(x) p p !
′ 1 ′ Ix,− (y) Ix,+ (y)
F (x) = U (y) − dy
xπi 0 Ix,− (y) − s2 (x)) Ix,+ (y) − s2 (x))
Z b(x) p (2.3)
2 ′ Ix,+ (y)
= U (y)ℑ .
xπ 0 Ix,+ (y) − s2 (x))

Suppose y ∈ (0, b(x)). We note that U ′ (y) = V ′′ (y)x1/2 + 12 V ′ (y)x−1/2 > 1. By the definition
(1.26) of Ix,+ , we know that Ix,+ (y) ∈ γ1 (x) for all y ∈ (0, b(x)). In the proof of part 4 of

13
Lemma 6 in Appendix A, we have the parametrization of γ1 (x) in (A.3).p(It is proved there
that γ1′ (x) in (A.3) is γ1 (x).) It is clear that arg Ix,+ (y)p
∈ (0, π) and arg Ix,+ (y) ∈ (0, π/2).
Also we have arg(Ip x,+ (y) − s2 (x)) ∈ (π/2, π). Hence arg( Ix,+ (y)/(Ix,+ (y) − s2 (x))) ∈ (−π, 0),
and similarly arg( Ix,+ (y)/(Ix,+ (y) − s2 (x))) ∈ (−π, 0). We conclude that the integrand on
the right-hand side of (2.3) is positive, and so F (x) is a monotonically increasing function.
In the special case V ′ (Jc (s)) = C > 0 is a constant, which is equivalent to U ′ (y) =
(C/2)y −1/2 , then we have

C
I
Jx (ξ) C
I
x C
I arcosh ξ+1
ξ−1
F (x)|V ′ (Jc (s))=C = √ dξ = dξ + √ dξ = Cx, (2.4)
2πi γ′ ξ(ξ − 1) 2πi γ′ ξ−1 2πi γ′ ξ(ξ − 1)
since the first term in the parentheses is Cx, and the second term vanishes, which can be
verified easily by deforming γ ′ into a very large circular contour. Then for general V that
satisfies assumption (1.17), by a comparison argument of U ′ (y) and (C/2)y −1/2 and the integral
formula (2.3), we also have that F (x) → 0 as x → 0 and F (x) → ∞ as x → ∞. So F (x) = 2
has a unique solution on (0, ∞).

Now we name this solution c′ . Later in Section 2.2 we will see that the unique solution to
(1.31) is equal to c, the parameter in (1.29).
In the special case V (x) = x/M, we have that V ′ (Jc (s)) = M−1 , and then by (2.4), we have
that the solution to (1.31) is
c′ = 2M. (2.5)

2.2 The g̃-functions and the density of the equilibrium measure

In Section 1.3.3, we define the functions g(z) and g̃(z) in (1.39) under the assumption that the
end point b and the density function ψ(x) are known. In this subsection, we first assume the
existence of b and ψ, and derive a Riemann-Hilbert problem satisfied by g′ (z) and g̃′ (z). Then
we solve the Riemann-Hilbert problem by direct calculation, and then confirm the value of b
and express ψ in a computable way. Thus we prove Theorems 4 and 8.
We recall that g(z) and g̃(z) defined in (1.39) are analytic on C\(−∞, b] and on P\(−π 2 /4, b]
respectively.
Lemma 12. Suppose V is one-cut regular with a hard edge. Then
1. For x ∈ (−∞, 0), g± (x) are continuous, and
g+ (x) = g− (x) + 2πi; (2.6)
for x ∈ (−π 2 /4, 0), g̃± (x) are continuous, and
g̃+ (x) = g̃− (x) + 2πi; (2.7)
and for z ∈ ρ, the lower parabola defined in (1.18), g̃(z) and g̃(z) are continuous, and
g̃− (z̄) = g̃+ (z) + 2πi, (2.8)
where z̄ is the complex conjugate of z lying on the upper parabola, with both ρ and ρ̄
oriented from left to right.
2. For x ∈ (0, b), we have
1 1
− (g′ (x)+ − g−
′
(x)) = − (g̃′ (x)+ − g̃−
′
(x)) > 0, (2.9)
2πi 2πi
and the left-hand side is equal to ψ(x).

14
 3. As z → b, the limits of g(z), g̃(z) and g′ (z), g̃′ (z) exist, and as x → b, g′ (z) − g′ (b) =
O(|z − b|1/2 ), g̃′ (z) − g̃′ (b) = O(|z − b|1/2 ), and
′ (x) − g′ (x))
i(g+ ′ (x) − g̃′ (x))
i(g̃+
− −
lim √ = lim √ ∈ (0, +∞), (2.10)
x→b− b−x x→b− b−x
and as z → 0 in C+ or C− , the limits of g(z), g̃(z) exist, and as x → 0, g′ (z) = O(|z|−1/2 ),
g̃′ (z) = O(|z|−1/2 ), and
′ ′ √ ′ ′ √
lim i(g+ (x) − g− (x)) x = lim i(g̃+ (x) − g̃− (x)) x ∈ (0, +∞). (2.11)
x→0+ x→b−

The two limits in (2.10) and (2.11) are 2πψb and 2πψ0 respectively.
4. As z → ∞ in C, g′ (z) = z −1 + O(z −2 ), and as f (z) → ∞ (i.e., ℜz → ∞), g̃′ (z) =
z −1/2 (1 + O(f (z)−1 ).
5. For x ∈ [0, b], there exists a constant ℓ such that
g± (x) + g̃∓ (x) − V (x) − ℓ = 0. (2.12)
For x ∈ (b, ∞), we have
g± (x) + g̃∓ (x) − V (x) − ℓ < 0. (2.13)
Conversely, if functions g(x) and g̃(x) which are analytic on C \ (−∞, b] and on P \ (−π 2 /4, b]
respectively satisfies all the properties listed above, then V is one-cut regular with a hard edge,
and its equilibrium measure is dµ(x) = ψ(x)dx supported on [0, b] with
1 1
ψ(x) = − (G(x)+ − G− (x)) = − (G̃(x)+ − G̃− (x)), (2.14)
2πi 2πi
where
b b
f ′ (z)ψ(x)dx
Z Z
′ ψ(x)dx ′
G(z) = g (z) = , G̃(z) = g̃ (z) = . (2.15)
0 z−x 0 f (z) − f (x)
The proof of Lemma 12 is straightforward and we omit it. Below we construct g(z) and
g̃(z) that satisfies the properties, under the condition of V given in Theorem 4. We note that
in the construction procedure, the value of b is unknown and needs to be determined.
To construct g(z) and g̃(z), it is equivalent to construct their the derivatives G(z) and
G̃(z). We recall the function b(x) defined in (1.23) and s1 (x), s2 (x) defined in Lemma 6. We
define the following two RH (Riemann-Hilbert) problem for (H (x) (z), H̃ (x) (z)) and N (x) (s) with
a parameter x > 0:
RH Problem 13.
1. H (x) (z) is analytic in C \ [0, b(x)] and H̃ (x) (z) is analytic in P \ [0, b(x)].
(x) (x)
2. We have the boundary conditions that H± (z) and H̃± (z) are continuous functions on
(0, b(x)) 1 and
1
H (x) (z) = + O(z −2 ), as z → ∞, (2.16)
z
1
H̃ (x) (z) = z − 2 (1 + O(f (z)−1 ), as f (z) → ∞ (i.e., ℜz → +∞), (2.17)
H (x) (z) = O(1), H̃ (x) (z) = O(1), as z → b(x), (2.18)
− 21 − 12
H (x) (z) = O(z ), H̃ (x) (z) = O(z ), as z → 0. (2.19)
1
In all RH problems in this paper, the boundary values of functions are continuous on the two sides of the
jump curves, unless otherwise stated. In later RH problems we omit the statement of continuity.

15
 3. For z ∈ (0, b(x)), we have
(x) (x)
H± (z) + H̃∓ (z) − V ′ (z) = 0. (2.20)

4. H̃ (x) (z) is continuous up to the boundary ρ ∪ {0} ∪ ρ̄ of P. For z ∈ ρ ∈ C− such that

z = limϵ→0+ J′x (y + ϵi) with y ∈ (0, 1), we have

lim H̃ (x) (z + ϵi)J′x (y + ϵi) = lim H̃ (x) (z̄ − ϵi)J′x (y − ϵi). (2.21)
ϵ→0+ ϵ→0+

Our RH problem 13 is motivated by the properties satisfied by (G(z), G̃(z)) defined by

(2.15). Analogous to (2.14), we define for y ∈ (0, b(x))
1 (x) 1 (x)
ψ (x) (y) = − (H (x) (y)+ − H− (y)) = − (H̃ (x) (y)+ − H̃− (y)), (2.22)
2πi 2πi
and then have, analogous to (2.15),
b(x) b(x)
f ′ (z)ψ(y)dy
Z Z
ψ(y)dy
H (x) (z) = , H̃ (x) (z) = . (2.23)
0 z−y 0 f (z) − f (y)
RH Problem 14.
1. N (x) (s) is analytic in C \ γ(x), where the contour γ(x) = γ1 (x) ∪ γ2 (x) defined in Theorem
8.
(x)
2. N± (s) is bounded on γ1 (x) and γ2 (x) and N (x) (s) is bounded if s → s1 (x) or s → s2 (x)
or s → 0. N (x) (s) has the behaviour

N (x) (s) = O(s−1 ), as s → ∞. (2.24)

3. N (x) (s) satisfies the jump condition

(x) (x) s−1

N+ (s) + N− (s) = J′ (s)V ′ (Jx (s)), s ∈ γ(x) \ {s1 (x), s2 (x)}, (2.25)
s − s2 (x) x

where the mapping Jx (s) is defined in (1.19), constant s2 (x) is defined in (1.20).
If (H (x) (z), H̃ (x) (z)) is a solution to RH problem 13, then the function Ñ (x) (s) defined by
(
(x) s−1 J′x (s)H (x) (Jx (s)), s ∈ C \ Dx ,
Ñ (s) = × (2.26)
s − s2 (x) J′x (s)H̃ (x) (Jx (s)), s ∈ Dx \ [0, 1],

is a solution to RH problem 14. (Although by (2.26) Ñ (x) (s) is undefined on (0, 1), it can
be naturally extended to (0, 1) by continuation, due to Item 4 of RH problem 13.) Moreover,
N (x) (s) given by (2.26) satisfies a condition stronger than (2.24)

Ñ (x) (1) = (s2 − 1)−1 , Ñ (x) (s) = s−1 + O(s−2 ), s → ∞. (2.27)

We also have that for each x > 0, RH problem 14 may have at most one solution. Suppose
both N (x),1 (s) and N (x),2 (s) are solutions to RH problem 14, then the function
(
(x) N (x),1 (s) − N (x),2 (s), s ∈ C \ D̄,
M (s) = (x),1 (x),2
(2.28)
−N (s) + N (s), s ∈ D.

16
We have that M (x) (s) is analytic in C \ {s1 (x), s2 (x)} after analytic continuation on γ(x) \
{s1 (x), s2 (x)}, and it is bounded as s approaches s1 (x), s2 (x) and M (x) (s) → 0 as s → ∞. By
Liouville’s theorem, M (x) (s) = 0, and then Ñ (x),1 (s) = N (x),2 (s).
The unique solution to RH problem 14 has an explicit formula
(ξ−1)J′x (ξ)V ′ (Jx (ξ))
(
1
R
(x)
− 2πi γ(x) (ξ−s2 (x))(ξ−s) dξ, s ∈ C \ Dx ,
N (s) = 1
R (ξ−1)J′x (ξ)V ′ (Jx (ξ)) (2.29)
2πi γ(x) (ξ−s2 (x))(ξ−s) dξ, s ∈ Dx \ [0, 1].
By direct computation, we have that
N (x) (1) = (s2 (x) − 1)−1 Cx , lim N (x) (s) = Cx s−1 + O(s−2 ), (2.30)
s→∞
where
(ξ − 1)J′x (ξ)V ′ (Jx (ξ)) s2 (x) − 1 J′x (ξ)V ′ (Jx (ξ))
Z Z
1
Cx = dξ = dξ (2.31)
2πi γ(x) ξ − s2 (x) 2πi γ(x) ξ − s2 (x)
We conclude that RH problem 13 has a solution, which implies that RH problem 14 has a
solution that in addition satisfies (2.27), only if Cx = 1, which is equivalent to equation (1.31)
in Theorem 8. Hence RH problem 14 has a solution only if x = c′ , the unique solution of
equation (1.31) as in Lemma 11. We then find that under the condition required in Theorem
4, and assuming the one-cut regular with a hard edge property of V , the only possible value of
the right-end point of the support of the equilibrium measure is b′ = b(c′ ), the only candidates
of functions G, G̃ defined by (1.39) and (2.15) are
r r
(c′ ) 2 Ic′ ,1 (z) (c′ ) (c′ ) 2 Ic′ ,2 (z) (c′ )
H (z) = ′ N (Ic′ ,1 (z)), H̃ (z) = ′ N (Ic′ ,2 (z)), (2.32)
c z c z
′
and the only candidate of the density function is ψ (c ) (x) defined in (2.22).
′
The remaining part of the proof is to show that the ψ (c ) (x)dx on [0, b′ ] does satisfy Require-
ment 1. Identities (2.16) and (2.17) in Item 2 of RH problem 13 implies that the total mass of
′
the (possibly signed) measure ψ (c ) (x)dx is 1, and identity (1.14) in Part 3 of Requirement 1 is
implied by (2.20) in Item 3 of RH problem 13. Hence, we only need to verify Parts 2, 4 and 5
of Requirement 1.
To this end, for s ∈ C \ D, we express (U (u) is defined in (1.17))
−1 c′ V ′ (Jc′ (ξ)) Jc′ (ξ)
I p
(c′ )
N (s) = √ dξ
2πi 2 γ(c′ ) ξ(ξ − s)
−1 c′ V ′ (Jc′ (ξ)) Jc′ (ξ) 1 c′ V ′ (Jc′ (ξ)) Jc′ (ξ)
Z p Z p
= √ dξ + √ dξ
2πi 2 γ1 (c′ ) ξ(ξ − s) 2πi 2 γ2 (c′ ) ξ(ξ − s)
U (u)I′c′ ,+ (u) U (u)I′c′ ,− (u)
"Z ′ Z b′ #
b
−1 c′
= p du − p du
2πi 2 0 Ic′ ,+ (u)(Ic′ ,+ (u) − s) 0 Ic′ ,− (u)(Ic′ ,− (u) − s)
′
"Z ′
b
p
−1 ′ U (u) Ic′ ,+ (u)
= c p √ p √ du (2.33)
2πi 0 ( Ic′ ,+ (u) + s)( Ic′ ,+ (u) − s)
Z b′ ′
p #
U (u) Ic′ ,− (u)
− p √ p √ du
0 ( Ic′ ,− (u) + s)( Ic′ ,− (u) − s)
Z b′ √ √ !
−1 c′
p p
d Ic′ ,+ (u) − s Ic′ ,− (u) − s
= √ U (u) log p √ − log p √ du
2πi 2 s 0 du Ic′ ,+ (u) + s Ic′ ,− (u) + s
Z b′ √ √ !
1 c′
p p
′ Ic′ ,+ (u) − s Ic′ ,− (u) − s
= √ U (u) log p √ − log p √ du,
2πi 2 s 0 Ic′ ,+ (u) + s Ic′ ,− (u) + s

17
 √
where s takes the principal branch. Hence, for x ∈ (0, b),
′
ψ (c ) (x)
1 Z b′
x− 2
p p p p
′ ( Ic′ ,+ (u) − Ic′ ,1 (x + ϵi))( Ic′ ,− (u) + Ic′ ,1 (x + ϵi))
= − ℜ lim U (u) log p du
4π 2 ϵ→0+ 0
p p p
( Ic′ ,+ (u) + Ic′ ,1 (x + ϵi))( Ic′ ,− (u) − Ic′ ,1 (x + ϵi))
1 Z ′
x− 2 b ′
= U (u)Fc′ (u; x)du,
4π 2 0
(2.34)
where Fc′ (u; x) is defined in (1.28). Since Fc′ (u; x) is a continuous function on u ∈ [0, x) ∪ (x, b′ )
and blows up at u = x as O(log(|u − x|), the integral (2.34) is well defined.

Positivity, also regularity in the bulk Using the fact that for all y ∈ (0, b′ ), ℜIc′ ,+ (y) =
ℜIc′ ,− (y) > 0 and ℑIc′ ,+ (y) = −ℑIc′ ,− (y) > 0, it is clear that for all x, u ∈ (0, b′ ),
p  p 
 I ′ (u) + pI ′ (x)   I ′ (u) − pI ′ (x) 
c ,+ c ,+ c ,− c ,+
 > 1,  > 1. (2.35)
   
p p p p
 Ic′ ,− (u) + Ic′ ,+ (x)   Ic′ ,+ (u) − Ic′ ,+ (x) 
Hence, as a function in u,
Fc′ (u; x) > 0, for all u ∈ (0, x) ∪ (x, b′ ). (2.36)
Together with the assumption U ′ (x) > 0 in (1.17), we conclude that ψ (c′ ) (x) > 0 for all x ∈
(0, b′ ).
′ ′ √
Regularity at the edges We need to show that ψ (c ) (x) satisfies that limx→b′− ψ (c ) (x)/ b′ − x =
′ √
c1 (1 + O(x − b′ )) and limx→0+ ψ (c ) (x) x = c2 (1 + O(x)) for some c1 > 0 and c2 > 0. By (2.34),
this is equivalent to that the integral
Z b′
U ′ (u)Fc′ (u; x)du (2.37)
0
decreases like a square root as x moves to b′ and approaches a nonzero limit as x moves to 0.
The desired property of (2.37) relies on the properties of U ′ (u) and Fc′ (u; x). They are both
continuous functions as u ∈ (0, b′ ), and we have the positivity results (1.17) for U ′ (u) on (0, b′ )
and (2.36) for Fc′ (u; x) as u, x ∈ (0, b′ ).
Moreover, since V (z) is analytic at 0, U (x) either converges to 0 (when V ′ (0) = 0) or blows
up like x−1/2 as u → 0+ (when V ′ (0) > 0).
For x ∈ (0, b′ ) in the vicinity of 0, from the properties of Ic′ ,± (x) given in Appendix A, we
have that the function
Fc′ (u; x)
q (2.38)
log2 (|u − x|) + 1
is uniformly bounded, and it converges uniformly to a non-vanishing limit as x → 0. Hence
(2.37) converges to a non-zero limit as x → 0. For x ∈ (0, b′ ) in the vicinity of b′ , we have (d1 (x)
is defined in (A.16))
√ √ √ √
√
 ′
 2b + d1 (c′ )( b′ − u + b′ − x)i b′ − u + b′ − x 

Fc′ (u; x) = log  ′ √ √ √ √  (1 + b′ − uf (u; x)),
2b + d1 (c′ )( b′ − u − b′ − x)i b′ − u − b′ − x 
(2.39)
such that f (u; x) is continuous on [0, b ] and converge uniformly to a limit function as x → b′ .
′

Hence the product of the integral in (2.37) and (b − x)−1/2 converges to a non-zero limit as
x → b− .

18
 ′
Regularity away from the support We want to show that if ψ(x) is defined as ψ (c ) (x) in
(2.22) and b = b′ , then Item 4 of Requirement 1 is satisfied. To this end, we denote
′ ′
g(x) = H (c ) (x) + H̃ (c ) (x) − V ′ (x), x ∈ (b′ , +∞), (2.40)

and it suffices to show that g(x) < 0 for all x > b′ . Since by (2.20) and the the continuity
′ ′
of H (c ) (z) and H̃ (c ) (z) at z = b′ , we have limx→b+ g(x) = 0. Hence it suffices to show that
d √ ′ d √
dx (g(x) x) < 0 on (b , ∞). By (2.23), we can express dx (g(x) x) as

b′ √ √ √ 
sinh2 ( x) + cosh(2 x) sinh2 ( t)
Z 
x+t ′
− √ 2
+ √ 2 √ 2
√ ψ (c ) (t)dt − U ′ (x), (2.41)
0 2 x(x − t) 2 x(sinh ( x) − sinh ( t)
′
and it is negative for all x ∈ (b′ , +∞), due to the positivity of ψ (c ) (t) and U ′ (x).

Proof of Theorems 4 and 8. We only need to prove Theorem 8 that is a quantitative version of
Theorem 4.
By the computation above in this subsection, we find that with c′ being the unique solution
′
of (1.31), the explicitly constructed density function ψ (c ) (x) and the measure it defines satisfy
all the items in Requirement 1, so it is the desired equilibrium measure. (The uniqueness is
guaranteed by Proposition 3.) Hence we also verify that c = c′ and b = b(c′ ) = b′ .

In the special case V (x) = x/M, we have that for s ∈ D,

I M−1 c (c√ξ + arcosh ξ+1 )
1 4 ξ−1
N (s) = √ dξ (2.42a)
2πi γ ξ(ξ − s)
I arcosh ξ+1
c2 1
I
dξ c 1 ξ−1
= + √ dξ, (2.42b)
4M 2πi γ ξ − s 4M 2πi γ ξ(ξ − s)

and we find that the first term in (2.42b) is c2 (4M)−1 s and the second term vanishes, by
deforming γ into a large circle. Similarly we can evaluate N (s) with s ∈ C \ D, and have
(recalling c = 2M in this case)
(
M, s ∈ D,
N (s) = 1 s+1 (2.43)
√
2 s
arcosh s−1 , s ∈ C \ D.

Hence, we have that

r r
1 I1 (z) I2 (z)
G(z) = − , G̃(z) = , (2.44)
M z z
and then as z = I2 (x − iϵ) and ϵ → 0+ ,
r
1 1 I+ (x)
ψ(x) = lim ℑG̃(z) = ℑ , (2.45)
ϵ→0+ π π x

which yields (1.34).

19
 (n)
3 Asymptotic analysis for pn+k (x)
3.1 RH problem of the polynomials
Consider the following modified Cauchy transform of pj :
Z
1 pj (x)
Cpj (z) := W (n) (x)dx, (3.1)
2πi R+ f (x) − f (z) α
(n)
which is well defined for z ∈ P \ R+ . Since Wα (x) is real analytic and vanishes rapidly as
x → +∞, for any polynomial p(x), we have the following asymptotic expansion for Cp(z) as
z ∈ P \ R+ and ℜz → +∞:
−1
Z
p(x)
Cp(z) = W (n) (x)dx
2πif (z) R+ 1 − f (x)/f (z) α
M Z (3.2)
−1 X

k (n) −(k+1) −(M +2)
= p(x)f (x)Wα (x)dx f (z) + O(f (z)),
2πif (z) R+
k=0

for any M ∈ N and uniformly in ℑz. Thus due to the orthogonality,

−h(n) −(j+1)
Cpj (z) = f (z) + O(f −(j+2) (z)), (3.3)
2πi
(n)
where hj is given in (1.5).
Hence we conclude that if we define the array
Y (z) = Y (j,n) (z) := (pj (z), Cpj (z)), (3.4)
then they satisfy the following conditions:
RH Problem 15.
1. Y = (Y1 , Y2 ), where Y1 is analytic on C, and Y2 is analytic on P \ R+ .
2. With the standard orientation of R+ ,
!
(n)
1 Wα (x)/f ′ (x)
Y+ (x) = Y− (x) , for x ∈ R+ . (3.5)
0 1

3. As z → ∞ in C, Y1 (z) = z j + O(z j−1 ).

4. As f (z) → ∞ in P (i.e., ℜz → +∞), Y2 (z) = O(f −(j+1) (z)).
5. As z → 0 in C or P,

O(1),
 α > 0,
Y1 (z) = O(1), Y2 (z) = O(log z), α = 0, (3.6)

O(z α ), α ∈ (−1, 0).


6. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, the limit Y2 (z) := limw→z in P Y2 (w) exists and is continuous, and
Y2 (z) = Y2 (z̄). (3.7)

Below we take j = n + k where k is a constant integer, and our goal is to obtain the
asymptotics for Y = Y (n+k,n) as n → ∞.

20
3.1.1 Uniqueness of RH problem 15
For later use in the proof of Lemma 29, we consider the uniqueness of a weaker form of RH
problem 15 such that Item 5 of RH problem 15 is replace by

5’. As z → 0 in C or P,

O(z ),
 −α α > 0 and arg z ∈ [− π3 , π3 ],
Y1 (z) = O(log z), α = 0, (3.8)
α ∈ (−1, 0) or α > 0 and arg(−z) ∈ (− 2π 2π

O(1), 3 , 3 ),


O(1),
 α > 0,
Y2 (z) = O(log z), α = 0, (3.9)

O(z α ), α ∈ (−1, 0).


and Y1 (z) and Y2 (z) are allowed to have a mild blowup at b, such that
1 1
Y1 (z) = O((z − b)− 2 ), Y2 (z) = O((z − b)− 2 ), as z → b, (3.10)

Proof of the uniqueness of RH problem 15 with Item 5 weakened by 5’. Despite nominally Y1 may
have blowups at 0 and b, since Y1 has pole singularities at 1, b, and it may only blow up at b like
an inverse square root and may only blow up at 0 like an inverse logarithm at 0 in the sector
arg(−z) ∈ (− 2π 2π
3 , 3 ), we find that Y1 (z) actually has no singularities at 1, b. Hence, by Item 3
of RH problem 15, we find that Y1 is a polynomial of degree j. Next, define
Z
1 pj (x)
Z2 (z) = Y2 (z) − W (n) (x)dx. (3.11)
2πi R+ f (x) − f (z) α

Then Z2 (z) has only a trivial jump on R+ , so its can be defined analytically on P \ {0, b},
and by an argument similar to that applied to Y1 above, b is not a singular point of Z2 (z),
and Z2 (z) can be defined analytically on P \ {0}. Now consider the function Z2 (f (z)) where
z ∈ C \ (−∞, −1]. By Item 6 of RH problem 15, Z2 (f (z)) can be extended analytically to
C \ {0}. Then by (3.6), Z2 (f (z)) = o(z −1 ) as z → 0. Hence Z2 (f (z)) is analytic on C, and we
conclude that Z2 (z) = 0 by Item 4 of RH problem 15. At last, Item 4 of RH problem 15 entails
the orthogonality of Y1 (z), so the uniqueness of biorthogonal polynomials given in Proposition
1 concludes the proof.

3.2 First transformation Y 7→ T

Recall g(z) and g̃(z)defined in (1.39) on C \ [0, b] and P ⊆ [0, b]. Denote Y = Y (n+k,n) and
define T as  −ng(z) 
− nℓ e 0 nℓ
T (z) = e 2 Y (z) ng̃z e 2 σ3 , (3.12)
0 e
where ℓ is the constant appearing in (1.14), and σ3 = ( 10 −1
0 ). Then T satisfies a RH problem

with the same domain of analyticity as Y , but with a different asymptotic behaviour and a
different jump relation.

RH Problem 16.

1. T = (T1 , T2 ), where T1 is analytic in C \ R+ , and T2 is analytic in P \ R+ .

21
 2. T satisfies the jump relation

T+ (x) = T− (x)JT (x), for x ∈ R+ , (3.13)

where
xα h(x) n(g− (x)+g̃+ (x)−V (z)−ℓ)
!
en(g− (x)−g+ (x)) f ′ (x) e
JT (x) = . (3.14)
0 en(g̃+ (x)−g̃− (x))

3.

T1 (z) = z k + O(z k−1 ) as z → ∞ in C, T2 (z) = O(f −(k+1) (z)) as f (z) → ∞ in P.

(3.15)

4. As z → 0 in C or in P, T (z) has the same limit behaviour as Y (z) in (3.6).

5.

T1 (z) = O(1), T2 (z) = O(1), as z → b. (3.16)

6. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, T2 (z) satisfies the same boundary condition as Y2 (z) in (3.7).

3.3 Second transformation T 7→ S

For x ∈ (b, +∞), it follows from the analyticity of g(z) there and (2.13) that the jump matrix
JT (x) tends to the identity matrix exponentially fast in the limit n → ∞. For x ∈ (0, b), we
decompose the jump matrix JT (x) as

en(g− (x)+g̃+ (x)−V (x)−ℓ)

! ! !
1 0 0 f ′ (x)x−α h(x)−1
1 0
f ′ (x) −nϕ− (x) f ′ (x)x−α h(x)−1 f ′ (x) −nϕ+ (x) .
xα h(x) e 1 − 0 xα h(x) e 1
en(g− (x)+g̃+ (x)−V (x)−ℓ)
(3.17)
Here ϕ(z) = g(z) + g̃(z) − V (z) − ℓ is defined as in (1.40). The function ϕ(z) has discontinuity
on R− and (0, b), such that

ϕ+ (x) = ϕ− (x) + 4πi, x < 0, (3.18)

ϕ+ (x) = − ϕ− (x), x ∈ (0, b). (3.19)

Then we “open the lens”, where the lens ΣS is a contour consisting of R+ and two arcs from
0 to b. We assume that one of the two arcs lies in the upper half plane and denote it by Σ1 , the
other lies in the lower half plane and denote it by Σ2 , see Figure 5. (We may take Σ1 and Σ2
symmetric about the real axis.) We do not fix the shape of ΣS at this state, but only require
that ΣS is in P and V is analytic in a simply-connected region containing ΣS . The exact shape
of Σ1 and Σ2 will be given in Sections 3.6, 3.7 and 3.8.

Σ1
0 b

Σ2

Figure 5: The lens ΣS .

22
 Define


T (z), ! outside of the lens,

1 0



T (z)
−α −1 ′ −nϕ(z)
, in the lower part of the lens,
S(z) := z h(z) f (z)e 1 ! (3.20)

1 0



T (z)

−α −1 ′ −nϕ(z)
, in the upper part of the lens.
−z h(z) f (z)e 1


From the definition of S, identity (2.12) and decomposition of JT (x) in (3.17), we have that S
satisfies the following:
RH Problem 17.
1. S = (S1 , S2 ), where S1 is analytic in C \ ΣS , and S2 is analytic in P \ ΣS .
2. We have
S+ (z) = S− (z)JS (z), for z ∈ ΣS , (3.21)
where
 !
 1 0
, for z ∈ Σ1 ∪ Σ2 ,


z −α h(z)−1 f ′ (z)e−nϕ(z) 1




 !
z α h(z)f ′ (z)−1

 0
JS (z) = , for z ∈ (0, b), (3.22)


 −z −α h(z)−1 f ′ (z) ! 0
α ′ −1 nϕ(z)


 1 z h(z)f (z) e
, for z ∈ (b, ∞).



 1

3. As z → ∞ in C or P, S(z) has the same limit behaviour as T (z) in (3.15).

4. As z → 0 in C \ Σ, we have

−α
O(z ),
 α > 0 and z inside the lens,
S1 (z) = O(log z), α = 0 and z inside the lens, (3.23)

O(1), z outside the lens or −1 < α < 0.


5. As z → 0 in P, S2 has the same limit behaviour as Y2 (z) in (3.6).

6. As z → b, S(z) has the same limit behaviour as T (z) as in (3.16).
7. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, S2 (z) satisfies the same boundary condition as Y2 (z) in (3.7).
By (2.12), for x ∈ (0, b), we have

ϕ′± (x) = g±
′ ′
(x) + g̃± (x) − V ′ (x) = g±
′ ′
(x) − g̃∓ (x) = ∓2πiψ(x). (3.24)

Since ψ(x) > 0 for all x ∈ (0, b), we have, by the Cauchy-Riemann condition, ℜϕ(z) > 0
on both the upper arc Σ1 and lower arc Σ2 , if these arcs are sufficiently close to (0, b). As
a consequence, the jump matrix for S on the lenses tend to the identity matrix as n → ∞.
Uniform convergence breaks down when x approaches the end points 0 and b, so we need to use
special local parametrices near these points.
Remark 2. Here and in subsequent RH problems, we may deform the jump contour (b, +∞)
locally, as long as ℜϕ(z) < 0 there, so that the entry z α h(z)f ′ (z)−1 enϕ(z) remains exponentially
small.

23
3.4 Construction of the global parametrix
Since JS (z) converges to I on Σ1 ∪ Σ2 ∪ (b, ∞), we construct the following

RH Problem 18.
(∞) (∞) (∞) (∞)
1. P (∞) = (P1 , P2 ), where P1 is analytic in C \ [0, b], and P2 is analytic in P \ [0, b].

2. For x ∈ (0, b), we have

xα h(x)f ′ (x)−1
 
(∞) (∞) 0
P+ (x) = P− (x) .
−x−α h(x)−1 f ′ (x) 0

3. As z → ∞ in C or P, P (∞) (z) has the same limit behaviour as T (z) in (3.15).

(∞)
4. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, P2 (z) satisfies the same boundary condition as Y2 (z) in (3.7).

To construct a solution to the above RH problem, we follow the idea in [11] to map the RH
problem for P (∞) to a scalar RH problem which can be solved explicitly. More precisely, using
the function Jc (s) defined in (1.19), we set
( (∞)
P1 (Jc (s)), s ∈ C \ D,
P(s) := (∞) (3.25)
P2 (Jc (s)), s ∈ D \ [0, 1],

where D is the region bounded by the curves γ1 and γ2 , as shown in Figure 2. Due to Item
4 of RH Problem 18, the function P is then well defined onto [0, 1) by continuation. It is
straightforward to check that P satisfies the following:

RH Problem 19.

1. P is analytic in C \ (γ1 ∪ γ2 ∪ {1}).

2. For s ∈ γ1 ∪ γ2 , P+ (s) = P− (s)JP (s), where, with γ1 and γ2 oriented from s1 to s2 ,

sinh(Jc (s))
(
− α (s)h(J , s ∈ γ1 ,
JP (s) = JαcJ(s)h(J
c c (s))Jc (s)
c (s))Jc (s)
(3.26)
sinh(Jc (s)) , s ∈ γ2 .

2
3. As s → ∞, P(s) = ( c4 )k sk + O(sk−1 ).

4. As s → 1, P(s) = O((s − 1)k+1 ).

The solution to RH problem 19 may not be unique. One solution is


Gk (s), s ∈ C \ D̄,
P(s) = c2 α+ 1 +k α+1
1
k (3.27)
2 ( 4 ) (s−s1 ) s 2 (s−1)
2
√ D̃(s)−1 , s ∈ D.
sinh(Jc (s)) (s−s1 )(s−s2 )

where the power function of (s − s1 )α+1 takes the principal branch and (s − s1 )(s − s2 ) ∼ s
p

in C \ γ1 . Later in this paper we take (3.27) as the definition of P(s). We note that
1
(
O((s − s1 )−α− 2 ), s → s1 in C \ D̄, 1
P(s) = α− 12
P(s) = O((s − s2 )− 2 ), s → s2 . (3.28)
O((s − s1 ) ), s → s1 in D,

24
 Based on this solution, we construct the solution to RH problem 18 as
(∞)
P1 (z) = P(I1 (z)), z ∈ C \ [0, b], (3.29)
(∞)
P2 (z) = P(I2 (z)), z ∈ P \ [0, b]. (3.30)
By direct calculation in Section A, we have
(∞) α 1 (∞) α 1
P1 (z) = O(z − 2 − 4 ), P2 (z) = O(z 2 − 4 ), as z → 0, (3.31)
(∞) − 41 (∞) − 14
P1 (z) = O(z ), P2 (z) = O(z ), as z → b. (3.32)

3.5 Third transformation S 7→ Q

(∞) (∞)
Noting that P1 (z) ̸= 0 for all z ∈ C \ [0, b] and P2 (z) ̸= 0 for all z ∈ P \ [0, b], we define the
third transformation by
 
S1 (z) S2 (z)
Q(z) = (Q1 (z), Q2 (z)) = , . (3.33)
P1∞ (z) P2∞ (z)
In view of the RH problems 17 and 18, and the properties of P (∞) (z) in (3.31) and (3.32), it is
then easily seen that Q satisfies the following RH problem:
RH Problem 20.
1. Q = (Q1 , Q2 ), where Q1 is analytic in C \ Σ, and Q2 is analytic in P \ Σ.
2. For z ∈ Σ, we have
Q+ (z) = Q− (z)JQ (z), (3.34)
where  

 1 0
(∞) , z ∈ Σ1 ∪ Σ2 ,

P (z)

z −α h(z)−1 f ′ (z) 2(∞) e−nϕ(z) 1



P1 (z)



 !
 0 1
JQ (z) = , z ∈ (0, b), (3.35)


 1 0
(∞)
 
α h(z)f ′ (z)−1 P1 (z) e−nϕ(z)



 1 z (∞)
P2 (z) , z ∈ (b, ∞).




0 1


3.
Q1 (z) = 1 + O(z −1 ), as z → ∞ in C, Q2 (z) = O(f −1 (z)), as f (z) → ∞ in P.
(3.36)

4. As z → 0 in C \ Σ, we have
−α+1

O(z 2 4 ),
 α > 0 and z inside the lens,
1
Q1 (z) = O(z 4 log z), α = 0 and z inside the lens, (3.37)
α 1
O(z 2 + 4 ),

z outside the lens or −1 < α < 0.


5. As z → 0 in P, we have
−α+1

O(z 2 4 ),
 α > 0,
1
Q2 (z) = O(z 4 log z), α = 0, (3.38)
α 1
O(z 2 + 4 ),

α ∈ (−1, 0).


25
 6.
1 1
Q1 (z) = O((z − b) 4 ), Q2 (z) = O((z − b) 4 ), as z → b. (3.39)

7. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, Q2 (z) satisfies the same boundary condition as Y2 (z) in (3.7).

3.6 Construction of local parametrix near b

First we consider the local parametrix near b. Using Part 5 of Lemma 12, we have that
limz→b ϕ(z) = 0 where ϕ is defined in (1.40). Then by Part 3 of Lemma 12 and Part 5 of
Requirement 1, we obtain the local behaviour for ϕ in the vicinity of b that (ψb is defined in
(1.16))
4π 3 5
ϕ(z) = − ψb (z − b) 2 + O(|z − b| 2 ), (3.40)
3
ϕ(z)/(z − b)3/2 is analytic at b, and then
 2
3 3
fb (z) = − ϕ(z) (3.41)
4
is a conformal mapping in a neighbourhood D(b, ϵ) around b satisfying (1.41), where ϵ > 0 is a
small enough constant. Moreover, we also choose the shape of the contour Σ so that the image
of Σ ∩ D(b, ϵ) under the mapping fb coincides with the jump contour ΓAi defined in (B.3) that
is the jump contour of the RH problem 47 for the Airy parametrix.
Let
(b) f ′ (z)/h(z) (b) zα
g1 (z) = (∞)
, g2 (z) = (∞)
, (3.42)
P1 (z) P2 (z)
and define
n
!
(b)
(b) (Ai) 2 e− 2 ϕ(z) g1 (z) 0
P (z) := Ψ (n fb (z))
3 n (b) , z ∈ D(b, ϵ) \ Σ. (3.43)
0 e 2 ϕ(z) g2 (z)

From (3.42) and RH problem 18 satisfied by P (∞) (z), we have

(b) (b) (b) (b)
g1,+ (x) = − g2,− (x), g2,+ (x) = g1,− (x), for x ∈ (b − ϵ, b), (3.44)
(b) 1 (b) 1
g1 (z) = O((z − b) ), 4 g2 (z) = O((z − b) ), 4 for z → b. (3.45)

Then we have the following RH problem satisfied by P(b) (z):

RH Problem 21.
1. P(b) (z) is a 2 × 2 matrix-valued function analytic for z ∈ D(b, ϵ) \ Σ.

2. For z ∈ Σ ∩ D(b, ϵ), we have

(b) (b)
P+ (z) = P− (z)JQ (z), (3.46)

where JQ (z) is defined in (3.35).

3.
1 1
(P(b) )ij (z) = O((z − b) 4 ), ((P(b) )−1 )ij (z) = O((z − b)− 4 ), as z → b, i, j = 1, 2.
(3.47)

26
 4. For z ∈ ∂D(b, ϵ), we have, as n → ∞,
E (b) (z)P(b) (z) = I + O(n−1 ), (3.48)
where
!−1 1 1
!
(b)  
(b) 1 g1 (z) 0 πi
σ 1 −1 n 6 fb (z) 4 0
E (z) = √ (b) e 4 3
1 1 . (3.49)
2 0 g2 (z) 1 1 0 n− 6 fb (z)− 4

It is straightforward to see that E (b) (z) defined in (3.49) is analytic on D(b, ϵ) \ (b − ϵ, b],
and for x ∈ (b − ϵ, b)  
(b) (b) −1 0 1
E+ (x)E− (x) = , (3.50)
1 0
and as z → b,
1
!
O(1) O(z − 2 ) O(1) O(1)
 
(b) (b) −1
E (z) = 1 , E (z) = 1 1 . (3.51)
O(1) O(z − 2 ) O(z 2 ) O(z 2 )
Then we define a 2 × 2 matrix-valued function
P (b) (z) = E (b) (z)P(b) (z), z ∈ D(b, ϵ) \ Σ, (3.52)
where P(b) is given in (3.43). Then we have the following RH problem satisfied by P (b) (z):
RH Problem 22.
1. P (b) (z) is analytic in D(b, ϵ) \ Σ.
2. For z ∈ Σ ∩ D(b, ϵ), we have
 (b)
P− (z)J

! Q (z), z ∈ Σ ∩ D(b, ϵ) \ (b − ϵ, b],
(b)
P+ (z) = 0 1 (b) (3.53)
 P− (z)JQ (z), z ∈ (b − ϵ, b).
1 0


3.
1 1
(P (b) )ij (z) = O((z − b)− 4 ), ((P (b) )−1 )ij (z) = O((z − b)− 4 ), as z → b, i, j = 1, 2.
(3.54)

4. For z on the boundary ∂D(b, ϵ), we have, as n → ∞, P (b) (z) = I + O(n−1 ).

At last, we define a vector-valued function V (b) by
V (b) (z) = Q(z)P (b) (z)−1 , z ∈ D(b, ϵ) \ Σ, (3.55)
where Q(z) is defined in (3.33). We find that V (b) (z) has only the trivial jump on (Σ1 ∪ Σ2 ∪
[b, b + ϵ)) ∩ D(b, ϵ), so V (b) (z) can be defined by continuation on D(b, ϵ) \ (b − ϵ, b]. It satisfies
the following RH problem:
RH Problem 23.
(b) (b)
1. V (b) = (V1 , V2 ) is analytic in D(b, ϵ) \ (b − ϵ, b].
2. For x ∈ (b − ϵ, b), we have  
(b) (b) 0 1
V+ (x) = V− (x) . (3.56)
1 0
3.
(b) (b)
V1 (z) = O(1), V2 (z) = O(1), as z → b. (3.57)

4. For z ∈ ∂D(b, ϵ), we have, as n → ∞, V (b) (z) = Q(z)(I + O(n−1 )).

27
3.7 Construction of local parametrix near 0
Using Part 1 of Lemma 12, we have that limz→0 in C± ϕ(z) = ±πi. We define in the vicinity of
0 that (
ϕ(z) − πi, z ∈ C+ ,
ϕL (z) = (3.58)
ϕ(z) + πi, z ∈ C− ,
Then by Part 3 of Lemma 12 and Part 5 of Requirement 1, we obtain the local behaviour for ϕ
(or equivalently ϕL ) at 0 that (ψ0 is the positive constant defined in Part 5 of Requirement 1)

ϕL (z) = 4πψ0 (−z)1/2 + O((−z)3/2 ), (3.59)

ϕ0 (z)/(−z)1/2 is analytic at 0, and then

ϕL (z)2
f0 (z) := (3.60)
16
is a conformal mapping in a neighbourhood D(0, ϵ) around 0 satisfying (1.42), where ϵ > 0 is a
small enough constant. Moreover, we also choose the shape of the contour Σ so that the image
of Σ ∩ D(0, ϵ) under the mapping f0 coincides with the jump contour ΓBe defined in (B.9) that
is the jump contour of the RH problem 48 for the Bessel parametrix.
Let
(0) (−z)−α/2 f ′ (z)/h(z) (0) (−z)α/2
g1 (z) = (∞)
, g2 (z) = (∞)
, (3.61)
P1 (z) P2 (z)

where the (−z)±α/2 is taken the principal branch, and define

n
!
(0)
(0) (Be) 2 e− 2 ϕ(z) g1 (z) 0
P (z) = Φ (n f0 (z)) n (0) , z ∈ D(0, ϵ) \ Σ. (3.62)
0 e 2 ϕ(z) g2 (z)

From (3.61) and RH problem 18 satisfied by P (∞) (z), we have

(0) (0) (0) (0)
g1,+ (x) = − g2,− (x), g2,+ (x) = g1,− (x), for x ∈ (0, ϵ), (3.63)
(0) 1 (0) 1
g1 (z) = O(z ), 4 g2 (z) = O(z ), 4 for z → 0. (3.64)

Then we have the following RH problem satisfied by P(0) (z):

RH Problem 24.
1. P(0) (z) is a 2 × 2 matrix-valued function analytic for z ∈ D(0, ϵ) \ Σ.
2. For z ∈ Σ ∩ D(0, ϵ), we have
(0) (0)
P+ (z) = P− (z)JQ (z), (3.65)

where JQ (z) is defined in (3.35).

3. As z ∈ ∂D(0, ϵ), we have

E (0) (z)P(0) (z) = (I + O(n−1 )), (3.66)

where
!−1  1 1
!
(0) 
(0) 1 g1 (z) 0 1 i n 2 f0 (z) 4 0 1
E (z) = √ (0) − 12 1 (2π) 2 σ3 . (3.67)
2 0 g2 (z) i 1 0 n f0 (z)− 4

28
 4. As z → 0, if α ∈ (−1, 0), then
α 1 α 1
! α 1 α 1
!
(0) O(z 2 + 4 ) O(z 2 + 4 ) (0) −1 O(z 2 − 4 ) O(z 2 − 4 )
P (z) = α 1 α 1 , P (z) = α 1 α 1 , (3.68)
O(z 2 + 4 ) O(z 2 + 4 ) O(z 2 − 4 ) O(z 2 − 4 )

if α = 0, then
1 1
! 1 1
!
O(z 4 log z) O(z 4 log z) O(z − 4 log z) O(z − 4 log z)
P (0)
(z) = 1 1 , P(0) (z)−1 = 1 1 ,
O(z log z) O(z log z)
4 4 O(z − 4 log z) O(z − 4 log z)
(3.69)

and if α > 0, then outside the lens

α 1 α 1
! α 1 α 1
!
O(z 2 + 4 ) O(z − 2 + 4 ) O(z − 2 − 4 ) O(z − 2 − 4 )
P(0) (z) = α 1 α 1 , P(0) (z)−1 = α 1 α 1 , (3.70)
O(z 2 + 4 ) O(z − 2 + 4 ) O(z 2 − 4 ) O(z 2 − 4 )

and inside the lens

α 1 α 1
! α 1 α 1
!
(0) O(z − 2 + 4 ) O(z − 2 + 4 ) (0) −1 O(z − 2 − 4 ) O(z − 2 − 4 )
P (z) = α 1 α 1 , P (z) = α 1 α 1 ,
O(z − 2 + 4 ) O(z − 2 + 4 ) O(z − 2 − 4 ) O(z − 2 − 4 )
(3.71)

It is straightforward to see that E (0) (z) is analytic on D(0, ϵ) \ [0, ϵ), and for x ∈ (0, ϵ),
 
(0) (0) −1 0 1
E+ (x)E− (x) = , (3.72)
1 0

and as z → 0,
1
!
O(1) O(z − 2 ) O(1) O(1)
 
E (0)
(z) = 1 , E (0) (z)−1 = 1 1 . (3.73)
O(1) O(z − 2 ) O(z 2 ) O(z 2 )

Then we define a 2 × 2 matrix-valued function

P (0) (z) = E (0) (z)P(0) (z), z ∈ D(0, ϵ) \ Σ, (3.74)

where P(0) is given in (3.62) and Hence, we have the following RH problem satisfied by P (0) (z):
RH Problem 25.
1. P (0) (z) is analytic in D(0, ϵ) \ Σ.

2. For z ∈ Σ ∩ D(0, ϵ), we have

 (0)
P− (z)J

! Q (z), z ∈ Σ ∩ D(0, ϵ) \ [0, ϵ),
(0)
P+ (z) = 0 1 (0) (3.75)
 P− (z)JQ (z), z ∈ (0, ϵ).
1 0


3. As z → 0, if α ∈ (−1, 0), then

α 1 α 1
! α 1 α 1
!
O(z 2 − 4 ) O(z 2 − 4 ) O(z 2 − 4 ) O(z 2 − 4 )
P (0) (z) = α 1 α 1 , P (0) (z)−1 = α 1 α 1 , (3.76)
O(z 2 − 4 ) O(z 2 − 4 ) O(z 2 − 4 ) O(z 2 − 4 )

29
 if α = 0, then
1 1
! 1 1
!
O(z − 4 log z) O(z − 4 log z) O(z − 4 log z) O(z − 4 log z)
P (0)
(z) = − 14 − 41
, P (0) (z)−1 = 1 1 ,
O(z log z) O(z log z) O(z − 4 log z) O(z − 4 log z)
(3.77)

and if α > 0, then outside the lens

α 1 α 1
! α 1 α 1
!
O(z 2 − 4 ) O(z − 2 − 4 ) O(z − 2 − 4 ) O(z − 2 − 4 )
P (0) (z) = α 1 α 1 , P (0) (z)−1 = α 1 α 1 . (3.78)
O(z 2 − 4 ) O(z − 2 − 4 ) O(z 2 − 4 ) O(z 2 − 4 )

and inside the lens

α 1 α 1
! α 1 α 1
!
O(z − 2 − 4 ) O(z − 2 − 4 ) O(z − 2 − 4 ) O(z − 2 − 4 )
P (0) (z) = α 1 α 1 , P (0) (z)−1 = α 1 α 1 ,
O(z − 2 − 4 ) O(z − 2 − 4 ) O(z − 2 − 4 ) O(z − 2 − 4 )
(3.79)

4. For z on the boundary ∂D(0, ϵ), we have, as n → ∞, P (0) (z) = I + O(n−1 ).

Consider the vector-valued function

U (z) = (U1 (z), U2 (z)) := Q(z)P(0) (z)−1 , ∈ D(0, ϵ) \ Σ, (3.80)

where Q(z) is defined in (3.33), and then define the vector-valued function V (0) on D(0, ϵ) \ Σ
by
(0) (0)
V (0) (z) = (V1 (z), V2 (z)) := Q(z)P (0) (z)−1 = U (z)E (0) (z)−1 . (3.81)
Due to the jump conditions (3.34) and (3.46), we have that U (z) can be extended analytically
to D(0, ϵ) \ {0}. Furthermore, as z → 0, from part 4 of RH problem 20 satisfied by Q(z) and
part 4 of RHP 24 satisfied by P(0) (z), we have that



(O(1), O(1)), α > 0 and z is outside the lens,
(O(z −α ), O(z −α )),

α > 0 and z is inside the lens,
(U1 (z), U2 (z)) = 2 2
(3.82)
(O((log z) ), O((log z) )), α = 0,


(O(z α ), O(z α )),

α ∈ (−1, 0).

Since 0 is an isolated singular point of U1 (z) and U2 (z), the estimates above implies that 0
is a removable singular point of U1 (z) and U2 (z), or equivalently, these two functions can be
extended analytically to D(0, ϵ). Hence, V (0) (z) satisfies the following RH problem:

RH Problem 26.
(0) (0)
1. V (0) (z) = (V1 (z), V2 (z)) is analytic in D(0, ϵ) \ [0, ϵ).

2. For x ∈ (0, ϵ), we have  

(0) (0) 0 1
V+ (x) = V− (x) . (3.83)
1 0

3. As z → 0, we have V (z) = (O(1), O(1)).

4. As z ∈ ∂D(0, ϵ), V (0) (z) = Q(z)(I + O(n−1 )).

30
3.8 Final transformation
We define R(z) = (R1 (z), R2 (z)) as
 (b)
V1 (z), z ∈ D(b, ϵ) \ (b − ϵ, b],

R1 (z) = V1(0) (z), z ∈ D(0, ϵ) \ [0, ϵ), (3.84)

Q1 (z), z ∈ C \ (D(b, ϵ) ∪ D(0, ϵ) ∪ Σ),

 (b)
V2 (z), z ∈ D(b, ϵ) \ (b − ϵ, b],

R2 (z) = V2(0) (z), z ∈ D(0, ϵ) \ [0, ϵ), (3.85)

Q2 (z), z ∈ P \ (D(b, ϵ) ∪ D(0, ϵ) ∪ Σ).


We set
ΣR := [0, b] ∪ [b + ϵ, ∞) ∪ ∂D(0, ϵ) ∪ ∂D(b, ϵ) ∪ ΣR R
1 ∪ Σ2 , (3.86)
where
ΣR
i := Σi \ {D(0, ϵ) ∪ D(b, ϵ)}, i = 1, 2. (3.87)
See Figure 6 for an illustration and the orientation of the arcs. Here we can fix the shape of ΣRi
(and so finally fix the shape of Σi ) by letting Σi be a continuous arc and ℜϕ(z) < 0 on ΣR i . It
is straightforward to check that R satisfies the following RH problem:

ΣR
1

0 b

ΣR
2

Figure 6: Contour ΣR .

RH Problem 27.
1. R(z) = (R1 (z), R2 (z)), where R1 (z) is analytic in C\ΣR , and R2 (z) is analytic in P\ΣR ).

2. R(z) satisfies the following jump conditions:



 JQ (z), z ∈ ΣR R
1 ∪ Σ2 ∪ (b + ϵ, +∞),

P (b) (z), z ∈ ∂D(b, ϵ),



R+ (z) = R− (z) P (0) (z), z ∈ ∂D(0, ϵ), (3.88)
 !
0 1


z ∈ (0, b) \ {ϵ, b − ϵ}.


 ,
1 0


3.

R1 (z) = 1 + O(z −1 ) as z → ∞ in C, R2 (z) = O(1) as f (z) → ∞ in P. (3.89)

4.

R1 (z) = O(1), R2 (z) = O(1), as z → 0, (3.90)

R1 (z) = O(1), R2 (z) = O(1), as z → b. (3.91)

31
 5. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, R2 (z) satisfies the same boundary condition as Y2 (z) in (3.7).
Similar to the idea used in the construction of global parametrix, to estimate R for large n,
we now transform the RH problem for R to a scalar one on the complex s-plane by defining
(
R1 (Jc (s)), s ∈ C \ D and s ∈ / I1 (ΣR ),
R(s) = (3.92)
/ I2 (ΣR ),
R2 (Jc (s)), s ∈ D \ [0, 1] and s ∈

where we recall that D is the region bounded by the curves γ1 and γ2 , I1 : C \ [0, b] → C \ D
and I2 : P \ [0, b] are defined in (1.24) and (1.25), respectively.
We are now at the stage of describing the RH problem for R. For this purpose, we define
′
Σ (1) := I1 (ΣR R
1 ∪ Σ2 ) ⊆ C \ D, Σ (1 ) := I2 (ΣR R
1 ∪ Σ2 ) ⊆ D,
′
Σ (2) := I1 ((b + ϵ, +∞)) ⊆ C \ D, Σ (2 ) := I2 ((b + ϵ, +∞)) ⊆ D,
′ (3.93)
Σ (3) := I1 (∂D(b, ϵ)) ⊆ C \ D, Σ (3 ) := I2 (∂D(b, ϵ)) ⊆ D,
′
Σ (4) := I1 (∂D(0, ϵ)) ⊆ C \ D, Σ (4 ) := I2 (∂D(0, ϵ)) ⊆ D,

and set
′ ′ ′ ′
Σ := Σ (1) ∪ Σ (1 ) ∪ Σ (2) ∪ Σ (2 ) ∪ Σ (3) ∪ Σ (3 ) ∪ Σ (4) ∪ Σ (4 ) . (3.94)
See Figure 7 for an illustration. We also define the following functions on each curve constituting
Σ : (Below z = Jc (s))

I1 (ΣR
1)

I2 (ΣR
2)

s1 1 s2

I2 (ΣR
1)

I1 (ΣR
2)

Σ ).) The solid and the dotted curves are the non-trivial
Figure 7: Contour Σ . (It is also ω(Σ̃
jump contour for the RH problem for R. (The solid and dashed curves, upon the mapping ω,
are the non-trivial jump contour for the RH problem for R̃.)

JΣ (1) (s) = (JQ )21 (z), s ∈ Σ (1) , (3.95)

(2′ )
JΣ (2′ ) (s) = (JQ )12 (z), s∈Σ , (3.96)
JΣ1 (3) (s) = (P (b)
)11 (z) − 1, JΣ2 (3) (s) = (P (b) )21 (z), s∈Σ (3)
, (3.97)
(3′ )
JΣ1 (3′ ) (s) = (P (b) )22 (z) − 1, JΣ2 (3′ ) (s) = (P (b) )12 (z), s∈Σ , (3.98)
JΣ1 (4) (s) = (P (0) )11 (z) − 1, JΣ2 (4) (s) = (P (0) )21 (z), s ∈ Σ (4) , (3.99)
(4′ )
JΣ1 (4′ ) (s) = (P (0)
)22 (z) − 1, JΣ2 (4′ ) (s) = (P (0) )12 (z), s∈Σ , (3.100)

32
where z = Jc (s) is in ΣR R
1 ∪ Σ2 in (3.95); in (b + ϵ, +∞) in (3.96); in ∂D(b, ϵ) in (3.97) and (3.98);
in ∂D(0, ϵ) in (3.99) and (3.100). With the aid of these functions, we further define an operator
∆Σ such that for any complex-valued function f (s) defined on Σ , ∆Σ transforms it linearly
into a function ∆Σ f that is also a complex-valued function defined on Σ , with expression
′


 JΣ (1) (s)f (s̃), s ∈ Σ (1) and s̃ = I2 (Jc (s)) ∈ Σ (1 ) ,
 ′
JΣ (2′ ) (s)f (s̃), s ∈ Σ (2 ) and s̃ = I1 (Jc (s)) ∈ Σ (2) ,




 ′
0, s ∈ Σ (1 ) ∪ Σ (2) ,



′
(∆Σ f )(s) = JΣ1 (3) (s)f (s) + JΣ2 (3) (s)f (s̃), s ∈ Σ (3) and s̃ = I2 (Jc (s)) ∈ Σ (3 ) , (3.101)
 1 2 ′
(3 ) and s̃ = I (J (s)) ∈ Σ (3) ,
JΣ (3′ ) (s)f (s) + JΣ (3′ ) (s)f (s̃), s ∈ Σ

1 c



1 2 (4) ′
JΣ (4) (s)f (s) + JΣ (4) (s)f (s̃), s ∈ Σ and s̃ = I2 (Jc (s)) ∈ Σ (4 ) ,



 1 ′
JΣ (4′ ) (s)f (s) + JΣ2 (4′ ) (s)f (s̃), s ∈ Σ (4 ) and s̃ = I1 (Jc (s)) ∈ Σ (4) .

Σ Σ
We note that all the functions JΣ (1) (s), . . . , JΣ2 (4′ ) (s) that define ∆Σ in (3.101) are uniformly
O(n−1 ). If we view ∆Σ as an operator from L2 (Σ Σ ) to L2 (Σ Σ ), then we have the estimate that
for all large enough n, there is a constant MΣ > 0 such that
−1
∥∆Σ ∥L2 (Σ
Σ ) ≤ MΣ n . (3.102)
RH problem 27 entails a scalar shifted RH problem for R:
RH Problem 28.
1. R(s) is analytic in C \ Σ , where the contour Σ is defined in (3.94).
2. For s ∈ Σ , we have
R+ (s) − R− (s) = (∆Σ R− )(s), (3.103)
where ∆Σ s the operator defined in (3.94).
3. As s → ∞, we have
R(s) = 1 + O(s−1 ). (3.104)
4. As s → 0, we have R(s) = O(1).
We have the following uniqueness result about the solution of the above RH problem.
Lemma 29. The function R(s) defined in (3.92) is the unique solution of RH problem 28.
Proof. Suppose Rsol (s) is one solution to RH problem 28, then using (3.92) backwardly, we
have a solution Rsol (z) = (R1sol (z), R2sol (z)) to RH problem 27. From Rsol (s), we define the
vector-valued function U sol (z) on D(0, ϵ) \ [0, ϵ) as (using (3.81) backwardly)
(U1sol (z), U2sol (z)) = (R1sol (z), R2sol (z))E (0) (z), (3.105)
where E (0) (z) is defined in (3.73). Then U1sol (z) and U2sol (z) can be defined analytically in
D(0, ϵ) \ {0}, and at the isolated singularity 0 they may only blow up like inverse square
root. Hence, U1sol (z) and U2sol (z) are actually analytic in D(0, ϵ). Next, we define Qsol (z) =
(Qsol sol
1 (z), Q2 (z)) by (using (3.84), (3.85) and (3.80) backwardly)

sol (b) sol (b)
R1 (z)(P )11 (z) + R2 (z)(P )21 (z), z ∈ D(b, ϵ) \ Σ,

Qsol
1 (z) = U1sol (z)(P(0) )11 (z) + U2sol (z)(P(0) )21 (z), z ∈ D(0, ϵ) \ Σ, (3.106)

 sol
R1 (z), z ∈ C \ (D(b, ϵ) ∪ D(0, ϵ) ∪ Σ),

sol (b) sol (b)
R1 (z)(P )12 (z) + R2 (z)(P )22 (z), z ∈ D(b, ϵ) \ Σ,

Qsol
2 (z) = U1sol (z)(P(0) )12 (z) + U2sol (z)(P(0) )22 (z), z ∈ D(0, ϵ) \ Σ, (3.107)

 sol
R2 (z), z ∈ P \ (D(b, ϵ) ∪ D(0, ϵ) ∪ Σ),

33
We find that Qsol sol
1 (z) and Q1 (z) can be defined analytically on C \ Σ and P \ Σ, respectively,
and find that Qsol (z) satisfies the variation of RH problem 20, such that in Item 4, the limit
behaviour of Q1 (z) as z → 0 from outside of the lens is changed to Q1 (z) = O(z 1/4 log z), and
1 1
in Item 6, the occurrences of (z − b) 4 in (3.39) are replaced by those of (z − b)− 4 .
Furthermore, we do the transforms Y → T → S → Q backwardly, and find that from Qsol
we can construct Y sol (z) = (Y1sol (z), Y2sol (z)) that satisfies the variation of RH problem 15 such
that Item 5 is replaced in Item 5’. in Section 3.1.1. By the argument in Section 3.1.1, we have
that Y sol (z) is unique, and then Rsol (s) is unique.

Finally, we show that

Lemma 30. For all s ∈ C \ Σ , we have the uniform convergence

R(s) = 1 + O(n−1 ). (3.108)

This lemma immediately yields that

R1 (z) = 1 + O(n−1 ) uniformly in C \ ΣR , R2 (z) = 1 + O(n−1 ) uniformly in P \ ΣR .

(3.109)

Proof of Lemma 30. We use the strategy proposed in [11], and start with the claim that R
satisfies the integral equation
R(s) = 1 + C(∆Σ R− )(s), (3.110)
where C is the Cauchy transform on Σ , such that for any g(s) defined on Σ ,
Z
1 g(ξ)
Cg(s) = dξ, s ∈ C \ Σ . (3.111)
2πi Σ ξ − s
To verify (3.110), by the uniqueness of RH problem 28, it suffices to show that the right-hand
side of (3.110) satisfies RH problem 28, and it is straightforward.
(3.110) can be written as
∆Σ (R− − 1)(ξ)
Z Z
1 1 ∆Σ (1)(ξ)
R(s) − 1 = dξ + dξ, s ∈ C \ Σ . (3.112)
2πi Σ ξ−s 2πi Σ ξ − s
Below we estimate the two terms on the right-hand side of the above formula.
By taking the limit where s approaches the minus side of Σ , we obtain from (3.112) that

R− (s) − 1 = C∆Σ (R− − 1)(s) + C− (∆Σ (1))(s), (3.113)

where C− is the Hilbert-like transform defined on Σ ,

Z
1 g(ξ)
C− g(s) = lim dξ, and C∆Σ f (s) = C− (∆Σ (f ))(s), (3.114)
2πi s′ →s− Σ ξ − s′
such that the limit s′ → s− is taken when approaching the contour from the minus side. Since
the Hilbert-like operator C− is bounded on L2 (ΣΣ ), we see from estimate (3.102) that the operator
1
− 2θ+1
norm of C∆Σ is also uniformly O(n ) as n → ∞. Hence, if n is large enough, the operator
1 − C∆Σ is invertible, and we could rewrite (3.113) as

R− (s) − 1 = (1 − C∆Σ )−1 (C− (∆Σ (1)))(s). (3.115)

As one can check directly that

−1
∥∆Σ (1)∥L2 (Σ
Σ ) = O(n ), (3.116)

34
combining the above two formulas gives us
−1
∥R− − 1∥L2 (Σ
Σ ) = O(n ). (3.117)

By (3.112), we have that for any fixed δ > 0, if dist(s, Σ ) > δ, then
1 1
∥ 2 Σ ) = O(n−1 ).


|R(s) − 1| ≤ ∥∆Σ (R− − 1)∥L2 (Σ
Σ ) + ∥∆Σ (1)∥L2 (Σ
Σ) · ∥ (3.118)
2π ξ − s L (Σ
As a consequence, we conclude (3.108) holds uniformly in {s ∈ C : dist(s, Σ ) > δ}. Since we
can deform the contour Σ outside a neighbourhood of 1, say D(1, ϵ′ ), by varying the value of
ϵ in (3.84) and (3.85), choosing different shapes of ΣR R
1 and Σ2 in (3.87), and deforming the
jump contour (b, ∞) as in Remark 2, and we can then show that (3.108) holds uniformly in
/ Σ and |s − 1| ≥ ϵ′ }. (We cannot freely deform Σ freely around 1, because Σ needs
{s ∈ C : s ∈
to connect to 1 as a vertex.) At last, in D(1, ϵ′ ), R(s) satisfies a simple RH problem: its value
on ∂D(1, ϵ′ ) is uniformly 1 + O(n−1 ), its limit at 1 is 1 and it has a jump along [1, 1 + ϵ′ ), where
the jump is given by JΣ (2′ ) (s) that is exponentially small. Hence we also conclude that (3.108)
holds uniformly in {s ∈ D(1, ϵ′ ) : s ∈/ Σ }.

(n)
4 Asymptotic analysis for qn+k (f (z))
(n) (n)
In this section we analyze qn+k (f (z)) in the same way as we do pn+k (z) in Section 3. Since
the method is parallel, we omit some detail. It is worth noting that the jump contours in this
section can be taken the same as those in Section 3.

4.1 RH problem of the polynomials

Consider the following Cauchy transform of qj :
Z
1 qj (f (x)) (n)
C̃qj (z) := Wα (x)dx, (4.1)
2πi R+ x − z
(n)
which is well defined for z ∈ P \ R+ . Since Wα (x) is real analytic and vanishes rapidly as
x → +∞, for any polynomial p(x), we have the following asymptotic expansion for C̃q(z) as
z ∈ P \ R+ and ℜz → +∞:
−1
Z
q(f (x)) (n)
C̃q(z) = W (x)dx
2πiz R+ 1 − x/z α
M Z (4.2)
−1 X

k (n) −(k+1) −(M +2)
= q(f (x))x Wα (x)dx z + O(z ),
2πiz R+
k=0

for any M ∈ N and uniformly in ℑz. Thus due to the orthogonality,

−h(n) −(j+1)
C̃qj (z) = z + O(z −(j+2) ), (4.3)
2πi
(n)
where hj is given in (1.5).
Hence we conclude that if we define the array

Ỹ (z) = Ỹ (j,n) (z) := (qj (f (z)), Cqj (z)), (4.4)

then they satisfy the following conditions

35
RH Problem 31.
1. Ỹ = (Ỹ1 , Ỹ2 ), where Ỹ1 is analytic on P, and Ỹ2 is analytic on C \ R+ .

2. With the standard orientation of R+ ,

!
(n)
1 Wα (x)
Ỹ+ (x) = Ỹ− (x) , for x ∈ R+ . (4.5)
0 1

3. As f (z) → ∞ in P (i.e., ℜz → +∞), Ỹ1 (z) = f (z)j + O(f (z)j−1 ).

4. As z → ∞ in C, Ỹ2 (z) = O(z −(j+1) ).

5. As z → 0 in P or C,

O(1),
 α > 0,
Ỹ1 (z) = O(1), Ỹ2 (z) = O(log z), α = 0, (4.6)

O(z α ), α ∈ (−1, 0).


6. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, the limit Ỹ1 (z) := limw→z in P Ỹ1 (w) exists and is continuous, and

Ỹ1 (z) = Ỹ1 (z̄). (4.7)

Conversely, the RH problem for Ỹ has a unique solution given by (3.4). We omit the proof,
since it is analogous to that of RH problem 15 given in Section 3.1.1.
Below we take j = n + k where k is a constant integer, and our goal is to obtain the
asymptotics for Ỹ = Ỹ (n+k,n) as n → ∞.

4.2 First transformation Ỹ 7→ T̃

Analogous to (3.12), we denote Ỹ = Ỹ (n+k,n) and define T̃ as
 −ng̃(z) 
− nℓ e 0 nℓ
T̃ (z) = e 2 Ỹ (z) e 2 σ3 . (4.8)
0 engz

Then T̃ satisfies a RH problem with the same domain of analyticity as Ỹ , but with a different
asymptotic behaviour and a different jump relation.
RH Problem 32.
1. T̃ = (T̃1 , T̃2 ), where T̃1 is analytic in P \ R+ , and T̃2 is analytic in C \ R+ .

2. T̃ satisfies the jump relation

T̃+ (x) = T̃− (x)JT̃ (x), for x ∈ R+ , (4.9)

where  n(g̃ (x)−g̃ (x))

xα h(x)en(g− (x)+g̃+ (x)−V (z)−ℓ)

e − +
JT̃ (x) = . (4.10)
0 en(g+ (x)−g− (x))

3.

T̃1 (z) = f (z)k + O(f (z)k−1 ) as f (z) → ∞ in P, T̃2 (z) = O(z −(k+1) ) as z ∈ ∞ in C.
(4.11)

36
 4. As z → 0 in P or in C, T̃ (z) has the same limit behaviour as Ỹ (z) in (4.6).

5.

T̃1 (z) = O(1), T̃2 (z) = O(1), as z → b. (4.12)

6. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, T̃1 (z) satisfies the same boundary condition as Ỹ1 (z) in (4.7).

4.3 Second transformation T̃ 7→ S̃

Analogous to (3.17), for x ∈ (0, b), we decompose the jump matrix JT̃ (x) as
! 
1 0 0 xα h(x)en(g̃− (x)+g+ (x)−V (x)−ℓ)
1 −nϕ− (x)
 1 
xα h(x) e 1 − 0
xα h(x)en(g̃− (x)+g+ (x)−V (x)−ℓ)
!
1 0
× 1 −nϕ+ (x) 1 . (4.13)
xα h(x) e

Then analogous to (3.20), define


T̃ (z),

 ! outside of the lens,
1 0



T̃ (z)
−α −1 −nϕ(z)
, in the lower part of the lens,
S̃(z) := z h(z) e 1 ! (4.14)

1 0



T̃ (z) , in the upper part of the lens.

−α −1 −nϕ(z)

−z h(z) e 1

From the definition of S̃, and decomposition of JT̃ (x) in (3.17), we have analogous to RH
problem 17 that
RH Problem 33.
1. S̃ = (S̃1 , S̃2 ), where S̃1 is analytic in P \ ΣS , and S̃2 is analytic in C \ ΣS .

2. We have
S̃+ (z) = S̃− (z)JS̃ (z), for z ∈ ΣS , (4.15)
where  !
 1 0
, for z ∈ Σ1 ∪ Σ2 ,


z −α h(z)−1 e−nϕ(z) 1 !





z α h(z)

 0
JS̃ (z) = −α h(z)−1
, for z ∈ (0, b), (4.16)


 −z ! 0
α h(z)enϕ(z)


 1 z
, for z ∈ (b, ∞).



 1

3. As z → ∞ in P or C, S̃(z) has the same limit behaviour as T̃ (z) in (4.11).

4. As z → 0 in P \ Σ, we have

−α
O(z ),
 α > 0 and z inside the lens,
S̃1 (z) = O(log z), α = 0 and z inside the lens, (4.17)

O(1), z outside the lens or −1 < α < 0.


37
 5. As z → 0 in C, S̃2 has the same behaviour as Ỹ2 (z) in (4.6).

6. As z → b, S̃(z) has the same limit behaviour as T̃ (z) in (4.12).

7. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, S̃1 (z) satisfies the same boundary condition as Ỹ1 (z) in (4.7).

4.4 Construction of the global parametrix

Analogous to RH problem 18, we construct the following

RH Problem 34.
(∞) (∞) (∞) (∞)
1. P̃ (∞) = (P̃1 , P̃2 ), where P̃1 is analytic in P \ [0, b], and P̃2 is analytic in C \ [0, b].

2. For x ∈ (0, b), we have

z α h(z)
 
(∞) (∞) 0
P̃+ (x) = P̃− (x) .
−z −α h(z)−1 0

3. As z → ∞ in P or C, P̃ (∞) (z) has the same limit behaviour as T̃ (z) in (4.11).

(∞)
4. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, P̃1 (z) satisfies the same boundary condition as Ỹ1 (z) in (4.7).

To construct a solution to the above RH problem, we set, analogous to (3.25),

( (∞)
P̃2 (Jc (s)), s ∈ C \ D,
P̃(s) := (∞) (4.18)
P̃1 (Jc (s)), s ∈ D \ [0, 1].

Like P(s) in (3.25), P̃ is well defined onto [0, 1) by continuation. Analogous to RH problem 19
for P, P̃ satisfies the following:

RH Problem 35.

1. P̃ is analytic in C \ (γ1 ∪ γ2 ∪ {1}).

2. For s ∈ γ1 ∪ γ2 , P̃+ (s) = P̃− (s)JP̃ (s), where

(
Jαc (s)h(Jc (s)), s ∈ γ1 ,
JP̃ (s) = (4.19)
−J−α −1
c (s)h(Jc (s)) , s ∈ γ2 .

3. As s → ∞, P̃(s) = O(s−(k+1) ).

4. As s → 1, P̃(s) = (s − 1)−k + O((s − 1)−(k+1) ).

A solution P̃ to the above RH problem is explicitly given by


G̃k (s), s ∈ D \ {1},
P̃(s) = (1−s1 )−α− 21 √s2 −1iD̃(1)−1 Jc (s)α (4.20)
 α k
√ D(s)−1 , ∈ C \ D,
(s−s1 ) (s−1) (s−s1 )(s−s2 )
p
where the square root is taken in C \ γ2 with (s − s1 )(s − s2 ) ∼ s as s → ∞. Later in this
paper we take (4.20) as the definition of P̃(s).

38
 Based on this solution, we construct the solution to RH problem 18 as
(∞)
P̃2 (z) = P̃(I1 (z)), z ∈ C \ [0, b], (4.21)
(∞)
P̃1 (z) = P̃(I2 (z)), z ∈ P \ [0, b]. (4.22)

By direct calculation in Section A, we have, analogous to (3.31) and (3.32),

(∞) α 1 (∞) α 1
P̃1 (z) = O(z − 2 − 4 ), P̃2 (z) = O(z 2 − 4 ), as z → 0, (4.23)
(∞) − 41 (∞) − 14
P̃1 (z) = O(z ), P̃2 (z) = O(z ), as z → b. (4.24)

4.5 Third transformation S̃ 7→ Q̃

(∞) (∞)
Noting that P̃1 (z) ̸= 0 for all z ∈ P \ [0, b] and P̃2 (z) ̸= 0 for all z ∈ C \ [0, b], we define
analogous to (3.33) !
S̃1 (z) S̃2 (z)
Q̃(z) = (Q̃1 (z), Q̃2 (z)) = , . (4.25)
P̃1∞ (z) P̃2∞ (z)
Analogous to RH problem 20, Q̃ satisfies the following RH problem:
RH Problem 36.
1. Q̃ = (Q̃1 , Q̃2 ), where Q̃1 is analytic in P \ Σ, and Q̃2 is analytic in C \ Σ.
2. For z ∈ Σ, we have
Q̃+ (z) = Q̃− (z)JQ̃ (z), (4.26)
where  

 1 0
(∞) , z ∈ Σ1 ∪ Σ2 ,

P̃ (z)

z −α h(z)−1 2(∞) e−nϕ(z) 1



P̃1 (z)



 !
 0 1
JQ̃ (z) = , z ∈ (0, b), (4.27)


 1 0
(∞)
 
P̃1 (z) −nϕ(z)

α


 1 z h(z) (∞) e
P̃2 (z) , z ∈ (b, ∞).




0 1


3.

Q̃1 (z) = 1 + O(f (z)−1 ), as f (z) → ∞ in P, Q̃2 (z) = O(z −1 ), as z → ∞ in C.

(4.28)

4. As z → 0 in P \ Σ, we have
−α+1

O(z 2 4 ),
 α > 0 and z inside the lens,
1
Q̃1 (z) = O(z 4 log z), α = 0 and z inside the lens, (4.29)
α 1
O(z 2 + 4 ),

z outside the lens or −1 < α < 0.


5. As z → 0 in C, we have
−α+1

O(z 2 4 ),
 α > 0,
1
Q̃2 (z) = O(z 4 log z), α = 0, (4.30)
α 1
O(z 2 + 4 ),

α ∈ (−1, 0).


39
 6.
1 1
Q̃1 (z) = O((z − b) 4 ), Q̃2 (z) = O((z − b) 4 ), as z → b. (4.31)

7. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, Q̃1 (z) satisfies the same boundary condition as Ỹ1 (z) in (4.7).

4.6 Construction of local parametrix near b

Let, analogous to (3.42),

(b) h(z)−1 (b) zα

g̃1 (z) = (∞)
, g̃2 (z) = (∞)
, (4.32)
P̃1 (z) P̃2 (z)

and define analogous to (3.43)

!
−n ϕ(z) (b)
2 e 2 g̃ (z) 0
P̃(b) (z) := Ψ(Ai) (n fb (z))
3 1
n (b) , z ∈ D(b, ϵ) \ Σ. (4.33)
0 e 2 ϕ(z) g̃2 (z)

From (4.32) and RH problem 34 satisfied by P̃ (∞) (z), we have, analogous to (3.44) and (3.45),
(b) (b) (b) (b)
g̃1,+ (x) = − g̃2,− (x), g̃2,+ (x) = g̃1,− (x), for x ∈ (b − ϵ, b), (4.34)
(b) 1 (b) 1
g̃1 (z) = O((z − b) ), 4 g̃2 (z) = O((z − b) ), 4 for z → b. (4.35)

Then analogous to RH problem 21, we have the following RH problem satisfied by P̃(b) (z):

RH Problem 37.

1. P̃(b) (z) is a 2 × 2 matrix-valued function analytic for z ∈ D(b, ϵ) \ Σ.

2. For z ∈ Σ ∩ D(b, ϵ), we have

(b) (b)
P̃+ (z) = P̃− (z)JQ̃ (z), (4.36)

where JQ̃ (z) is defined in (4.27).

3. As z → b, the limit behaviour of P̃(b) (z) and (P̃(b) )−1 (z) is the same as that of P(b) (z) and
(P(b) )−1 (z) in (3.47).

4. For z ∈ ∂D(b, ϵ), we have, as n → ∞,

Ẽ (b) (z)P̃(b) (z) = I + O(n−1 ), (4.37)

where
!−1 1 1
!
(b)  
1 g̃1 (z) 0 πi
σ3 1 −1 n 6 fb (z) 4 0
Ẽ (b) (z) = √ (b) e 4
− 61 1 P̃(b) (z)
2 0 g̃2 (z) 1 1 0 n fb (z)− 4
= I + O(n−1 ).
(4.38)

40
 We now define a 2 × 2 matrix-valued function

P̃ (b) (z) = Ẽ (b) (z)P̃(b) (z) z ∈ D(b, ϵ) \ Σ, (4.39)

where P̃(b) is given in (4.33) and Ẽ (b) (z) is defined in (4.38), like E (b) (z) defined in (3.49), it is
straightforward to see that Ẽ (b) (z) is analytic on D(b, ϵ) \ (b − ϵ, b], and for x ∈ (b − ϵ, b)
 
(b) (b) −1 0 1
Ẽ+ (x)Ẽ− (x) = , (4.40)
1 0

and as z → b,
1
!
O(1) O(z − 2 ) O(1) O(1)
 
(b) (b) −1
Ẽ (z) = 1 , Ẽ (z) = 1 1 . (4.41)
O(1) O(z − 2 ) O(z 2 ) O(z 2 )

Hence, analogous to RH problem 22, we have the following RH problem satisfied by P̃ (b) (z):

RH Problem 38.

1. P̃ (b) (z) is analytic in D(b, ϵ) \ Σ.

2. For z ∈ Σ ∩ D(b, ϵ), we have

 (b)
P̃− (z)J (z), z ∈ Σ ∩ D(b, ϵ) \ (b − ϵ, b],
! Q̃

(b)
P̃+ (z) = 0 1 (b) (4.42)
 P̃− (z)JQ̃ (z), z ∈ (b − ϵ, b).
1 0


3. As z → b, the limit behaviour of P̃ (b) (z) and (P̃ (b) )−1 (z) is the same as that of P (b) (z)
and (P (b) )−1 (z) in (3.54).

4. For z on the boundary ∂D(b, ϵ), we have, as n → ∞, P̃ (b) (z) = I + O(n−1 ).

At last, analogous to (3.55), we define a vector-valued function Ṽ (b) by

Ṽ (b) (z) = Q̃(z)P̃ (b) (z)−1 , z ∈ D(b, ϵ) \ Σ, (4.43)

where Q̃(z) is defined in (4.25). It satisfies the following RH problem that is analogous to RH
problem 23:

RH Problem 39.
(b) (b)
1. Ṽ (b) = (Ṽ1 , Ṽ2 ) is analytic in D(b, ϵ) \ (b − ϵ, b].

2. For x ∈ (b − ϵ, b), we have  

(b) (b) 0 1
Ṽ+ (x) = Ṽ− (x) . (4.44)
1 0

3.
(b) (b)
Ṽ1 (z) = O(1), Ṽ2 (z) = O(1), as z → b. (4.45)

4. For z ∈ ∂D(b, ϵ), we have, as n → ∞, Ṽ (b) (z) = Q(z)(I + O(n−1 )).

41
4.7 Construction of local parametrix near 0
Let, analogous to (3.61)

(0) (−z)−α/2 /h(z) (0) (−z)α/2

g̃1 (z) = (∞)
, g̃2 (z) = (∞)
, (4.46)
P̃1 (z) P̃2 (z)

where the (−z)±α/2 is taken the principal branch, and define, analogous to (3.62),
!
−n ϕ(z) (0)
e 2 g̃ (z) 0
P(0) (z) = Φ(Be) (n2 f0 (z)) 1
n (0) , z ∈ D(0, ϵ) \ Σ. (4.47)
0 e 2 ϕ(z) g̃2 (z)

Analogous to (3.63) and (3.64), we have

(0) (0) (0) (0)
g̃1,+ (x) = − g̃2,− (x), g̃2,+ (x) = g̃1,− (x), for x ∈ (0, ϵ), (4.48)
(0) 1 (0) 1
g̃1 (z) = O(z ), 4 g̃2 (z) = O(z ), 4 for z → 0. (4.49)

Analogous to RH problem 24, we have the following RH problem satisfied by P̃(0) (z):
RH Problem 40.
1. P̃(0) (z) is a 2 × 2 matrix-valued function analytic for z ∈ D(0, ϵ) \ Σ.
2. For z ∈ Σ ∩ D(0, ϵ), we have
(0) (0)
P̃+ (z) = P̃− (z)JQ̃ (z), (4.50)

where JQ̃ (z) is defined in (4.27).

3. As z ∈ ∂D(0, ϵ), we have

Ẽ (0) (z)P̃(0) (z) = (I + O(n−1 )), (4.51)

where
!−1  1 1
!
(0) 
(0) 1 g̃1 (z) 0 1 i n 2 f0 (z) 4 0 1
Ẽ (z) = √ (0) 1 1 (2π) 2 σ3 . (4.52)
2 0 g̃2 (z) i 1 0 n− 2 f0 (z)− 4

4. As z → 0, the limit behaviour of P̃(0) (z) and (P̃(0) )−1 (z) is the same as that of P(0) (z) and
(P(0) )−1 (z) in (3.68)–(3.71).
Like E (0) (z), it is straightforward to see that Ẽ (0) (z) is analytic on D(0, ϵ) \ [0, ϵ), and for
x ∈ (0, ϵ), like (3.72)  
(0) (0) −1 0 1
Ẽ+ (x)Ẽ− (x) = , (4.53)
1 0
and as z → 0, like (3.73)
1
!
O(1) O(z − 2 ) O(1) O(1)
 
(0) (0) −1
Ẽ (z) = 1 , Ẽ (z) = 1 1 . (4.54)
O(1) O(z − 2 ) O(z 2 ) O(z 2 )

Then we define a 2 × 2 matrix-valued function

P̃ (0) (z) = Ẽ (0) (z)P̃(0) (z), z ∈ D(0, ϵ) \ Σ, (4.55)

where P̃(0) is given in (4.47) and Hence, we have the following RH problem satisfied by P̃ (0) (z):

42
RH Problem 41.

1. P̃ (0) (z) is analytic in D(0, ϵ) \ Σ.

2. For z ∈ Σ ∩ D(0, ϵ), we have

 (0)
P̃− (z)J

! Q (z), z ∈ Σ ∩ D(0, ϵ) \ [0, ϵ),
(0)
P̃+ (z) = 0 1 (0) (4.56)
 P̃− (z)JQ (z), z ∈ (0, ϵ).
1 0


3. As z → 0, the limit behaviour of P̃ (0) (z) and (P̃ (0) )−1 (z) is the same as that of P (0) (z)
and (P (0) )−1 (z) in (3.76)–(3.79).

4. For z on the boundary ∂D(0, ϵ), we have, as n → ∞, P̃ (0) (z) = I + O(n−1 ).

Consider the vector-valued function

Ũ (z) = (Ũ1 (z), Ũ2 (z)) := Q̃(z)P̃(0) (z)−1 , ∈ D(0, ϵ) \ Σ, (4.57)

where Q̃(z) is defined in (4.25), and then define the vector-valued function Ṽ (0) on D(0, ϵ) \ Σ
by
(0) (0)
Ṽ (0) (z) = (Ṽ1 (z), Ṽ2 (z)) := Q̃(z)P (0) (z)−1 = Ũ (z)Ẽ (0) (z)−1 . (4.58)
Like U (z) defined in (3.80), we can show that 0 is a removable singular point of Ũ1 (z) and
Ũ2 (z). Hence, like RH problem 26, Ṽ (0) (z) satisfies the following RH problem:

RH Problem 42.
(0) (0)
1. Ṽ (0) (z) = (Ṽ1 (z), Ṽ2 (z)) is analytic in D(0, ϵ) \ [0, ϵ).

2. For x ∈ (0, ϵ), we have  

(0) (0) 0 1
Ṽ+ (x) = Ṽ− (x) . (4.59)
1 0

3. As z → 0, we have Ṽ (z) = (O(1), O(1)).

4. As z ∈ ∂D(0, ϵ), Ṽ (0) (z) = Q̃(z)(I + O(n−1 )).

4.8 Final transformation

Analogous to (3.84) and (3.85), we define R̃(z) = (R̃1 (z), R̃2 (z)) as
 (b)
Ṽ1 (z),
 z ∈ D(b, ϵ) \ (b − ϵ, b],
R̃1 (z) = Ṽ1(0) (z), z ∈ D(0, ϵ) \ [0, ϵ), (4.60)

Q̃1 (z), z ∈ P \ (D(b, ϵ) ∪ D(0, ϵ) ∪ Σ),

 (b)
Ṽ2 (z),
 z ∈ D(b, ϵ) \ (b − ϵ, b],
R̃2 (z) = Ṽ2(0) (z), z ∈ D(0, ϵ) \ [0, ϵ), (4.61)

Q̃2 (z), z ∈ C \ (D(b, ϵ) ∪ D(0, ϵ) ∪ Σ).


Then R̃ satisfies the following RH problem that is analogous to RH problem 27:

RH Problem 43.

43
 1. R̃(z) = (R̃1 (z), R̃2 (z)), where R̃1 (z) is analytic in P\ΣR , and R̃2 (z) is analytic in C\ΣR ).
2. R̃(z) satisfies the following jump conditions:


JQ̃ (z), z ∈ ΣR R
1 ∪ Σ2 ∪ (b + ϵ, +∞),

P̃ (b) (z), z ∈ ∂D(b, ϵ),



R̃+ (z) = R̃− (z) P̃ (0) (z), z ∈ ∂D(0, ϵ), (4.62)
 !
0 1


z ∈ (0, b) \ {ϵ, b − ϵ}.


 ,
1 0


3.
R̃1 (z) = 1 + O(f (z)−1 ) as f (z) → ∞ in P, R̃2 (z) = O(1) as z → ∞ in C. (4.63)

4.
R̃1 (z) = O(1), R̃2 (z) = O(1), as z → 0, (4.64)
R̃1 (z) = O(1), R̃2 (z) = O(1), as z → b. (4.65)

5. At z ∈ ρ ∪ {−π 2 /4} ∪ ρ̄, R̃1 (z) satisfies the same boundary condition as Ỹ1 (z) in (4.7).
Similar to the idea used in the construction of global parametrix, to estimate R for large
n, we now transform the RH problem for R to a scalar one on the complex s-plane. Let
ω : z → ω(z) be a mapping from C \ {1} to C \ {1} as ω(z) = 1 + 1/(z − 1). Then let J̃c (s) be
the mapping
J̃c (s) = Jc (ω(s)). (4.66)
by defining (
R̃1 (J̃c (s)), s ∈ ω(C \ D) and s ∈ / ω(I1 (ΣR )),
R̃(s) = (4.67)
/ ω(I2 (ΣR )),
R̃2 (J̃c (s)), s ∈ ω(D \ [0, 1]) and s ∈
where we recall that D is the region bounded by the curves γ1 and γ2 , I1 : C \ [0, b] → C \ D
and I2 : P \ [0, b] are defined in (1.24) and (1.25), respectively.
We are now at the stage of describing the RH problem for R. For this purpose, we define,
analogous to (3.94) and set
(1) (1′ ) (2) (2′ ) (3) (3′ ) (4) (4′ ) (∗)
Σ := Σ̃
Σ̃ Σ ∪ Σ̃
Σ ∪ Σ̃
Σ ∪ Σ̃
Σ ∪ Σ̃
Σ ∪ Σ̃
Σ ∪ Σ̃
Σ ∪ Σ̃
Σ , Σ
where Σ̃ Σ (∗) ), (4.68)
= ω(Σ
for ∗ = 1, 1′ , 2, 2′ , 3, 3′ , 4, 4′ respectively.
Σ : (Below z = J̃c (s))
We also define the following functions on each curve constituting Σ̃

(1′ )
J (1′ ) (s) = (JQ̃ )21 (z), s ∈ Σ̃
Σ , (4.69)
Σ
Σ̃
(2)
J (2) (s) = (JQ̃ )12 (z), s ∈ Σ̃
Σ , (4.70)
Σ
Σ̃
(3)
J 1 (3) (s) = (P̃ (b) )22 (z) − 1, J 2 (3) (s) = (P̃ (b) )12 (z), s ∈ Σ̃
Σ , (4.71)
Σ
Σ̃ Σ
Σ̃
(3′ )
J 1 (3′ ) (s) = (P̃ (b) )11 (z) − 1, J 2 (3′ ) (s) = (P̃ (b) )21 (z), s ∈ Σ̃
Σ , (4.72)
Σ
Σ̃ Σ
Σ̃
(4)
J 1 (4) (s) = (P̃ (0) )22 (z) − 1, J 2 (4) (s) = (P̃ (0) )12 (z), s ∈ Σ̃
Σ , (4.73)
Σ
Σ̃ Σ
Σ̃
(4′ )
J 1 (4′ ) (s) = (P̃ (0) )11 (z) − 1, J 2 (4′ ) (s) = (P̃ (0) )21 (z), s ∈ Σ̃
Σ , (4.74)
Σ
Σ̃ Σ
Σ̃

44
where z = J̃c (s) is in ΣR R
1 ∪ Σ2 in (4.69); in (b + ϵ, +∞) in (4.70); in ∂D(b, ϵ) in (4.71) and (4.72);
in ∂D(0, ϵ) in (4.73) and (4.74). With the aid of these functions, we further define an operator
∆Σ̃
Σ such that for any complex-valued function f (s) defined on Σ̃ Σ , ∆Σ̃Σ transforms it linearly
into a function ∆Σ̃Σ f that is also a complex-valued function defined on Σ̃ Σ , with expression
(1) (1′ )

 J (1′ ) (s)f (s̃), s ∈ Σ
Σ̃ and s̃ = I 1 (J c (s)) ∈ Σ
Σ̃ ,
Σ
Σ̃


(2) (2)



 J (2) (s)f (s̃), s ∈ Σ̃
Σ and s̃ = I2 (Jc (s)) ∈ Σ̃ Σ ,
 Σ̃Σ


(1) ′
(2 )
s ∈ Σ̃
Σ ∪ Σ̃



 0, Σ ,
(3′ )


1 2 (3)
(∆Σ̃ J (s)f (s) + J (s)f (s̃), s ∈ Σ
Σ̃ and s̃ = I (J (s)) ∈ Σ
Σ̃ ,
Σ f )(s) =  Σ̃ (4.75)
(3) (3) 2 c
Σ Σ
Σ̃ ′)
1 2 (3 (3)
J (3′ ) (s)f (s) + J (3′ ) (s)f (s̃), s ∈ Σ̃ and s̃ = I1 (Jc (s)) ∈ Σ̃



 Σ Σ ,
 Σ
Σ̃ Σ
Σ̃
(4) (4′ )

J 1 (4) (s)f (s) + J 2 (4) (s)f (s̃), s ∈ Σ̃

Σ and s̃ = I2 (Jc (s)) ∈ Σ̃ Σ ,


 Σ
Σ̃ Σ
Σ̃
(4′ )


J 1 ′ (s)f (s) + J 2 ′ (s)f (s̃), s ∈ Σ̃ (4)
Σ and s̃ = I1 (Jc (s)) ∈ Σ̃ Σ .

(4 ) (4 )
Σ
Σ̃ Σ
Σ̃

We note that all the functions J (1′ ) (s), . . . , J 2 (4′ ) (s) that define ∆Σ̃
Σ in (4.75) are uniformly
Σ
Σ̃ Σ
Σ̃
O(n−1 ). If we view ∆Σ̃ 2 Σ ) to L2 (Σ̃
Σ as an operator from L (Σ̃ Σ ), then we have the estimate that
for all large enough n, there is a constant MΣ̃
Σ > 0 such that
−1
∥∆Σ̃
Σ ∥L2 (Σ̃
Σ ) ≤ MΣ̃
Σn . (4.76)

RH problem 43 entails a scalar shifted RH problem for R:

RH Problem 44.
1. R̃(s) is analytic in C \ Σ̃
Σ , where the contour Σ̃
Σ is defined in (4.68).

2. For s ∈ Σ̃
Σ , we have
R̃+ (s) − R̃− (s) = (∆Σ̃
Σ R̃− )(s), (4.77)
where ∆Σ̃
Σ s the operator defined in (4.75).

3. As s → ∞, we have
R̃(s) = 1 + O(s−1 ). (4.78)

4. As s → 0, we have R̃(s) = O(1).

Like Lemma 29, we have
Lemma 45. The function R̃(s) defined in (4.67) is the unique solution of RH problem 44 after
trivial analytical extension.
The proof is the same as that of Lemma 29, and we omit it.
Finally, we have analogous to Lemma 30 that
Lemma 46. For all s ∈ C \ Σ̃
Σ , we have the uniform convergence

R̃(s) = 1 + O(n−1 ). (4.79)

The proof of this lemma is analogous to that of Lemma 30, and we omit it. This lemma
immediately yields that

R̃1 (z) = 1 + O(n−1 ) uniformly in P \ ΣR , R̃2 (z) = 1 + O(n−1 ) uniformly in C \ ΣR .

(4.80)

45
5 Proof of main results
In this section we prove Theorem 10. In the statement of Theorem 10, C+ ∪R is divided into Aδ ,
(n) (n)
Bδ , Cδ and Dδ . We can consider the limit of pn+k (z) and and qn+k (z) in the hard edge region
D(0, ϵ), the soft edge region D(b, ϵ), the bulk region enclosed by ΣR R
1 , Σ2 , ∂D(0, ϵ), ∂D(b, ϵ), and
the outside region that is the complement of these three, since by deforming the shape of contour
ΣR , these four regions cover Cδ , Dδ , Bδ , Aδ respectively.

5.1 Outside region

For z ∈ C that is out of the lens and out of D(0, ϵ) and D(b, ϵ), we have, by RH problems 15,
16, 17, 20 and 27
(n) (n+k,n) (∞)
pn+k (z) = Y1 (z) = T1 (z)eng(z) = S1 (z)eng(z) = Q1 (z)P1 (z)eng(z)
(∞)
(5.1)
= R1 (z)P1 (z)eng(z) .
(∞)
Since R1 (z) = 1 + O(n−1 ) uniformly by (3.109) and P1 (z) = Gk (I1 (z)) by (3.29) and (3.27),
we prove (1.52).
Similarly, for z ∈ P that is out of the lens and out of D(0, ϵ) and D(b, ϵ), we consider the
(n) (n)
counterpart of (5.1) for qn+k (f (z)). For later use in Section 5.5, we also consider C̃qn+k (f (z))
for z ∈ C that is out of the lens and out of D(0, ϵ) and D(b, ϵ). We have, by RH problems 31,
32, 33, 36 and 43, like (5.1),
(n) (∞) (n) (∞)
qn+k (f (z)) = R̃1 (z)P̃1 (z)eng̃(z) , C̃qn+k (z) = enℓ R̃2 (z)P̃2 (z)e−ng(z) . (5.2)
(∞)
Since R̃1 (z) = 1 + O(n−1 ) and R̃2 (z) = 1 + O(n−1 ) uniformly by (4.80), P̃1 (z) = G̃k (I2 (z))
(∞)
by (4.22) and (4.20), and P̃2 (z) has the expression by (4.21) and (4.20) we prove (1.53) and
get

1√
(n) (1 − s1 )−α− 2 s2 − 1iD̃(1)−1 z α enℓ e−ng(z)
C̃qn+k (z) = p (1 + O(n−1 ). (5.3)
α k
(I1 (z) − s1 ) (I1 (z) − 1) (I1 (z) − s1 )(I1 (z) − s2 )D(I1 (z))

5.2 Bulk region

Similar to (5.1), we have that for z in the upper lens and out of D(0, ϵ) and D(b, ϵ),
(n) (n+k,n)
pn+k (z) = Y1 (z) = T1 (z)eng(z) = S1 (z)eng(z) + z −α h(z)−1 f ′ (z)S2 (z)en(V (z)−g̃(z)+ℓ)
(∞) (∞)
= Q1 (z)P1 (z)eng(z) + z −α h(z)−1 f ′ (z)Q2 (z)P2 (z)en(V (z)−g̃(z)+ℓ) (5.4)
(∞) (∞)
= R1 (z)P1 (z)eng(z) + z −α h(z)−1 f ′ (z)R2 (z)P2 (z)en(V (z)−g̃(z)+ℓ) .

Like in (5.1), R1 (z) = 1 + O(n−1 ) and R2 (z) = 1 + O(n−1 ) uniformly by (3.109). By (3.29),
(3.30), (3.27), (1.46), (1.47) and (1.45), we have
(∞) (∞)
P1 (z) = Gk (I1 (z)), z −α h(z)−1 f ′ (z)P2 (z) = Gk (I2 (z)). (5.5)

Hence we prove (1.54).

In particular, if x ∈ (ϵ, b − ϵ) and z → x from above, we have by (2.12) that limz→x g(z) =
g+ (x) and limz→x V (z) − g̃(z) + ℓ = RV (x) − g̃+ (x) + ℓ = g− (x) = g+ (x). From the defini-
tion (1.39) of g(z), we have g± (x) = log|x − y|dµV (y) ± µV ([x, b])i. On the other hand, as

46
z → x from above, by (1.26) and (1.27), I1 (z) and I2 (z) converges to I+ (x) and I− (x) respec-
tively. Noting that I− (x) = I+ (x), and then limz→x Gk (I2 (z)) = Gk,+ (I− (x)) = Gk,+ (I− (x)) =
limz→x Gk (I2 (z)) and limz→x R1 (z) = R(I+ (x)) = R(I+ (x)) = limz→x R2 (z), we have
h i
(n) (n)
pn+k (x) = lim pn+k (z) = 2ℜ R(I+ (x))Gk,+ (I+ (x))eng+ (x)
z→x in C+
  (5.6)
= 2ℜ (1 + O(n−1 ))Gk,+ (I+ (x))eng+ (x) ,

which implies (1.56).

Analogous to (5.4), we have that for z in the upper lens and out of D(0, ϵ) and D(b, ϵ),
(n) (∞) (∞)
qn+k (f (z)) = R̃1 (z)P̃1 (z)eng̃(z) + z −α h(z)−1 R̃2 (z)P̃2 (z)en(V (z)−g(z)+ℓ) . (5.7)

Like in (5.2), R̃1 (z) = 1 + O(n−1 ) and R̃2 (z) = 1 + O(n−1 ) uniformly by (4.80). By (4.22),
(4.21) and (4.20), (1.47) and (1.45), we have
(∞) (∞)
P̃1 (z) = G̃k (I2 (z)), z −α h(z)−1 P̃2 (z) = G̃k (I1 (z)). (5.8)
Hence we prove (1.55).
In particular, if x ∈ (ϵ, b − ϵ) and z → x from above, we have by (2.12) that limz→x g̃(z) =
g̃+ (x) and limz→x V (z) − g(z) +Rℓ = V (x) − g+ (x) + ℓ = g̃− (x) = g̃+ (x). From the definition
(1.39) of g̃(z), we have g̃± (x) = log|f (x) − f (y)|dµV (y) ± µV ([x, b])i. Like (5.6), we have
 
(n) (n)
qn+k (f (x)) = lim qn+k (f (z)) = 2ℜ (1 + O(n−1 ))G̃k,− (I− (x))eng̃+ (x) , (5.9)
z→x in C+

which implies (1.56).

5.3 Soft edge region

Like (5.1) and (5.4), we have, for z ∈ D(b, ϵ) ∩ C+ ,
(n) (n+k,n) (∞)
pn+k (z) = Y1 (z) = T1 (z)eng(z) = S1 (z)eng(z) = P1 (z)Q1 (z)eng(z)
(
0, outside the lens,
+ (∞) (5.10)
z −α h(z)−1 f ′ (z)P2 (z)Q2 (z)en(V (z)−g̃(z)+ℓ , inside the upper lens.

and then by (3.55), (3.52), (3.49), (3.43), (3.42) and (1.40) (also (B.6) if z is inside the upper
lens)

√ f ′ (z) (∞)
 1
 
(n) 1 (∞) (b) (b) 2
pn+k (z) = π n 6 fb (z) P1 (z)V1 (z) − i α
4
P2 (z)V2 (z) Ai(n 3 fb (z))
z h(z)
′
  
1
− 61 − 4 (∞) (b) f (z) (∞) (b) ′ 2 n
−n fb (z) P1 (z)V1 (z) + i α P2 (z)V2 (z) Ai (n fb (z)) e 2 (g(z)−g̃(z)+V (z)+ℓ) ,
3
z h(z)
(5.11)
By (3.84) and (3.85), we have that V1 (z) = R1 (z) and V2 (z) = R2 (z), so they are both uniformly
(∞) (∞)
1 + O(n−1 ) by (3.109). Also we still have (5.5) for P1 , P2 like in the bulk region. Hence
we have (1.58).
For z ∈ C+ in the vicinity of b, we have the limits of I1 (z) and I2 (z) by (A.14) and (A.15)
respectively, and for s in the vicinity of s2 , we have by (1.46),
α+ 12   1
c2 (s2 − s1 )

c k 2 1
Gk (s) = s2 D(s2 )(s − s2 )− 2 (1 + O(s − s2 )). (5.12)
4b 2

47
where (s − s2 )−1/2 is positive on (s2 , +∞) and has the branch along γ1 . Hence we derive (1.60).
Like (5.10), we have, for z ∈ D(b, ϵ) ∩ C+ ,

(n) (∞)
qn+k (f (z)) = P̃1 (z)Q̃1 (z)eng̃(z)
(
0, outside the lens,
+ (∞) (5.13)
z −α h(z)−1 P̃2 (z)Q̃2 (z)en(V (z)−g(z)+ℓ , inside the upper lens.

and then by (4.43), (4.39), (4.38), (4.33), (4.32) and (1.40) (also (B.6) if z is inside the upper
lens)

√
  
(n) 1 1
(∞) (b) 1 (∞) (b) 2
= π n fb (z)
qn+k (f (z)) 6 4
P̃1 (z)Ṽ1 (z) −i α P̃2 (z)Ṽ2 (z) Ai(n 3 fb (z))
z h(z)
  
1
− 16 − 4 (∞) (b) 1 (∞) (b) ′ 2 n
−n fb (z) P̃1 (z)Ṽ1 (z) + i P̃2 (z)Ṽ2 (z) Ai (n fb (z)) e 2 (g̃(z)−g(z)+V (z)+ℓ) ,
3
z α h(z)
(5.14)

By (4.60) and (4.61), we have that Ṽ1 (z) = R̃1 (z) and Ṽ2 (z) = R̃2 (z), so they are both uniformly
(∞) (∞)
1 + O(n−1 ) by (4.80). Also we still have (5.8) for P̃1 , P̃2 like in the bulk region. Hence we
have (1.59).
Similar to (5.12), we use the limits of I1 (z) and I2 (z), and have for s in the vicinity of s2
1
(c/2)k− 2 D̃(s2 )i 1
G̃k (s) = (s − s2 )− 2 (1 + O(s − s2 )), (5.15)
α+ 21
[(1 − s1 )(s2 − s1 )] D̃(1)

where (s − s2 )−1/2 is positive on (s2 , +∞) and has the branch along γ2 . Hence we prove (1.61)
in the same way as (1.60).

5.4 Hard edge region

For z ∈ D(0, ϵ), (5.10), still holds, and then by (3.81), (3.74), (3.67), (3.62), (3.61), and (1.40)
(also (B.16) if z is inside the upper lens)

√ f ′ (z)
  1

(n) 1 (∞) (0) (∞) (0)
p
4
pn+k (z) = π n f0 (z) P1 (z)V1 (z) + i
2 P 2 (z)V 2 (z) I α (2n f0 (z))
(−z)α h(z)
f ′ (z)
1
  
− 12 − 4 (∞) (0) (∞) (0) ′
p
+ n f0 (z) P1 (z)V1 (z) − i P (z)V2 (z) Iα (2n f0 (z))
(−z)α h(z) 2
n
× e 2 (g(z)−g̃(z)+V (z)+ℓ) , (5.16)

where the (−z)α factor takes the principal branch as arg(−z) ∈ (−π, π). Like in the soft edge
(0) (0) (∞) (∞)
region, V1 (z) and V2 (z) are both uniformly 1 + O(n−1 ) by (3.109), and Π1 (z) and P2 (z)
are still given by (3.29), (3.30), (3.27), (1.46) and (1.45). Then we have (1.62).
For z ∈ C+ in the vicinity of 0, we have the limits of I1 (z) and I2 (z) by (A.9) and (A.10)
respectively, and for s in the vicinity of s1 , we have by (1.46),
α+ 12  k r
4(1 − s1 )2 (−s1 ) c2 −s1

1
Gk (s) = (s1 − 1) D(s1 )(s1 − s)−α− 2 (1 + O(s − s1 )),
(s2 − s1 )2 4 s2 − s1
(5.17)

48
where (s1 − s)−α−1/2 is positive on (−∞, s1 ) and has the branch cut along γ1 . Hence we derive,
uniformly for z ∈ D(0, ϵ) ∩ C+ ,
1 n α (n)
n− 2 e− 2 (g(z)−g̃(z)+V (z)+ℓ) (−z) 2 pn+k (z) =
√ α+ 1  2 k
√ √
r
−s1 c(1 − s1 ) −s1

2 c 1
 
2 π (s1 − 1) D(s1 )(−f0 (0)) 4 Iα (2 t) + O(n−1 ) ,
s2 − s1 s2 − s1 4
(5.18)
which implies (1.64) on D(0, ϵ) ∩ (C+ ∪ R) by changing Iα into Jα as [32, 10.27.6].
Like (5.16), we have, for z ∈ D(0, ϵ) ∩ C+ ,
√
  
(n) 1 1
4 (∞) (0) 1 (∞) (0)
p
qn+k (f (z)) = π n 2 f0 (z) P̃1 (z)Ṽ1 (z) + i α
P̃2 (z)Ṽ2 (z) Iα (2n f0 (z))
(−z) h(z)
  
1
− 12 − 4 (∞) (0) 1 (∞) (0) ′
p
+n f0 (z) P̃1 (z)Ṽ1 (z) − i P̃ (z)Ṽ2 (z) Iα (2n f0 (z))
(−z)α h(z) 2
n
× e 2 (g̃(z)−g(z)+V (z)+ℓ) . (5.19)

By (4.60) and (4.61), we have that Ṽ1 (z) = R̃1 (z) and Ṽ2 (z) = R̃2 (z), so they are both uniformly
(∞) (∞)
1 + O(n−1 ) by (4.80). Also we still have (5.8) for P̃1 , P̃2 like in the bulk region and the
soft edge region. Hence we have (1.63).
Similar to (5.17), we also have for s in the vicinity of s1
r
−α− 12 −k s2 − 1 D̃(s1 ) 1
G̃k (s) = (1 − s1 ) (s1 − 1) (s − s1 )−α− 2 (1 + O(s − s1 )). (5.20)
s2 − s1 D̃(1)
Hence we prove (1.65) in the same way as (1.64).

(n)
5.5 Computation of hn+k
(n) (n) (n)
hn+k is defined in (1.5). By the limiting formulas (1.52) for pn+k (z) and (1.53) for qn+k (f (z)),
and the regularity condition (2.13), we have that if ϵ > 0 is a small constant, then there is
δ = δ(ϵ) > 0, such that for any R > b + δ,
  Z R
 (n) (n) (n) (n) (n)
hn+k − hn+k (R) = oe(1−ϵ)nℓ , where hn+k (R) = pn+k (x)qn+k (f (x))Wα(n) (x)dx. (5.21)

0

Hence to prove (1.66), we only need to compute

I
(n) (n) (n)
hn+k (R) = − pn+k (z)C̃qn+k (z)dz, (5.22)
CR

where CR is the circular contour with positive orientation, centred at 0 with radius R. With
the help of (1.52) and (5.3), we have
−α− 21 √
!
I
G (I (z))(1 − s ) s − 1i D̃(1) −1 z α D(I (z))−1
(n) k 1 1 2 1
hn+k (R) = enℓ − p dz + O(n−1 )
α k (I (z) − s )(I (z) − s )
CR (I1 (z) − s1 ) (I1 (z) − 1) 1 1 1 2
√ α+ 12  2 k I !
s2 − 1i c2 I1 (z)1/2 D(I1 (z))−1

nℓ c −1
=e − 1/2 (I (z) − s )
dz + O(n )
D̃(1) 4(1 − s1 ) 4 CR z 1 2
√ 2
α+ 12  2 k !
−

s2 1i c c
= enℓ − (cπi) + O(n−1 ) ,
D̃(1) 4(1 − s1 ) 4
(5.23)

49
where in the last step we take R → ∞ and use I1 (z) = 4c−2 z + O(1) and D(I1 (z)) = 1 + O(z −1 )
as z → ∞. Hence we derive (1.66).

A Properties of Jx (s) and related functions

Proof of Part 4 of Lemma 6 Below we give a constructive description of γ1 (x). Any s ∈ C+
can be represented as
cosh(u + iv) + 1
s= , u ∈ (0, ∞) and v ∈ (−π, 0). (A.1)
cosh(u + iv) − 1
Then the condition s ∈ γ1 (x) ⊆ C+ is equivalent to ℑJx (s) = 0, which can be expressed as
x sin v/(cosh u − cos v) − v = 0, or equivalently,
sin v
cosh u = x + cos v. (A.2)
v
By direct computation, we see that the right-hand side of (A.2) is an increasing on (−π, 0),
and its limits at −π and 0 are −1 and x + 1 respectively. Hence, (A.2) has a solution only if
v ∈ (v ∗ , 0), where v ∗ ∈ (−π, 0) is the solution to 1 = x sin(v)/v + cos v. We then construct the
curve
 
′ cosh(u(v) + iv) + 1 ∗ sin v
γ1 (x) = { : v ∈ [v , 0]}, where u(v) = arcosh x + cos v . (A.3)
cosh(u(v) + iv) − 1 v
which lies in C+ and connects γ1′ (v ∗ ) = (cosh(iv ∗ ) + 1)/(cosh(iv ∗ ) − 1) = s1 (x) and γ1′ (0) =
1 + 2/x = s2 (x).
By expressing s in u, v in (A.1) and parametrizing u by v as in (A.3), we have that
Jx (s) |s∈γ1′ (x) is parametrized by v ∈ (v ∗ , 0). Then we can compute
d x + 1 − cosh(u(v) + iv) ′
Jx (s(u(v), v)) = − (u (v) + i). (A.4)
dv cosh(u(v) + iv) − 1
Since by the construction, we know that the Jx (s) ∈ R if s ∈ γ1′ (x), we know that the derivative
above is real valued wherever it is well defined on (v ∗ , 0). On the other hand, we have that
x + 1 − cosh(u(v) + iv) ̸= 0 and u′ (v) + i ̸= 0 on (v ∗ , 0). Hence the derivative in (A.4) is
real, non-vanishing, and continuous on (v ∗ , 0), and then it has to be always positive or always
negative there. By comparing Jx (s1 (x)) and Jx (s2 (x)), we conclude that the derivative is always
positive on (v ∗ , 0).
Now we see that the curve γ1′ (x) satisfies the properties of γ1 (x) described in Part 4 of
Lemma 6, and it is the only candidate of γ1 (x). So we let γ1 (x) be γ1′ (x) constructed in (A.3),
and prove constructively Part 4 of Lemma 6.

Proof of Lemma 7 Both C+ and C+ \ D are simply connected regions. From Parts 1, 2
and 4 of Lemma 6, we have that Jx (s) maps (−∞, s1 (x)] ∪ γ1 (x) ∪ [s2 (x), ∞), the boundary of
C+ \ D, homeomorphically to R, the boundary of C+ . Also we have J[x](s) = (x2 /4)s + O(1).
For any large enough R ∈ R+ , we let CR be the region enclosed by the semicircle SR := {Reiθ :
θ ∈ (0, π)}J[x](s) and the lower boundary (−R, s1 (s)] ∪ γ1 (x) ∪ [s2 (x), R). We then have that
Jx (s) maps the boundary of CR homeomorphically to its image lying in C+ ∪ R. Hence by
standard argument in complex analysis, Jx (s) maps CR analytically and bijectively to its image
lying in C+ . Letting R → ∞, we have that Jx (s) maps C+ \ analytically and bijectively to C+ .
Similarly, we can prove that Jx (s) maps C+ ∩ D bijectively to C− ∩ P.
At last, since Jx (s̄) = Jx (s), the results above can be extended to C \ D̄ and C ∩ D. Hence
we finish the proof of Lemma 7.

50
Limiting shape of γ1 (x) as x → ∞ and x → 0 With the help of expression (A.3) of
γ1 (x) = γ1′ (x), we have the following results by direct calculation.
x2
1. As x → ∞, s1 (x) = −( πx )2 (1 + O(x−1 )) and b(x) = −1
4 (1 + O(x )). Given ϵ > 0, we have

ϵx2 (1 − ϵ)x2
 
4 √
Ix,+ (y) = 2 y(1 + O(x−1 )) + iπ y(1 + O(x−1 )) ,


y∈ , , (A.5)
x 4 4

where the two O(x−1 ) terms are uniform for y ∈ (ϵx2 /4, (1 − ϵ)x2 /4).

2. As x → 0+ , s1 (x) = − x2 (1 + O(x)) and b(x) = 2x(1 + O(x)). Given ϵ > 0, we have

2  p 
Ix,+ (y) = (y − x) + y(2x − y)i (1 + O(x)), y ∈ (ϵx, (2 − ϵ)x), (A.6)
x2
where the 1 + O(x) term is uniform for y ∈ (ϵx, (2 − ϵ)x).

Limit behaviour of Jx (s) around s1 (x) and s2 (x), and its inverse functions around 0
and b(x) By direct computation, we have the follows.

1. In the vicinity of s1 (x)

x2 (s2 − s1 )2
Jx (s) = (s − s1 (x))2 + O((s − s1 (x))3 ), (A.7)
16s1 (1 − s1 )2

and around 0, for a small enough ϵ > 0,

√
Ix,± (y) = s1 (x) ± e1 (x) yi + O(y), y ∈ (0, ϵ), (A.8)
1
Ix,1 (z) = s1 (x) − e1 (x)(−z) + O(z),
2 z ∈ N (0, ϵ) \ [0, ϵ), (A.9)
1
Ix,2 (z) = s1 (x) + e1 (x)(−z) + O(z),
2 z ∈ N (0, ϵ) \ [0, ϵ). (A.10)

where p
4−s1 (x)(1 − s1 (x))
e1 (x) = . (A.11)
x(s2 (x) − s1 (x))

2. In the vicinity of s2 (x)

x2 b(x)1/2
Jx (s) = b(x) + (s − s2 (x))2 + O((s − s2 (x))3 ), (A.12)
8s2 (x)1/2

and around b(x), for a small enough ϵ > 0,

p
Ix,± (y) = s2 (x) ± d1 (x) b(x) − yi + O(b(x) − y), y ∈ (b(x) − ϵ, b(x)), (A.13)
p
Ix,1 (z) = s2 (x) + d1 (x) z − b(x) + O(b(x) − z), z ∈ N (b(x), ϵ) \ [b(x) − ϵ, b(x)),
(A.14)
p
Ix,2 (z) = s2 (x) − d1 (x) z − b(x) + O(b(x) − z), z ∈ N (b(x), ϵ) \ [b(x) − ϵ, b(x)).
(A.15)

where √
2 2s2 (x)1/4
d1 (x) = . (A.16)
xb(x)1/4

51
B Local universal parametrices
B.1 The Airy parametrix
In this subsection, let y0 , y1 and y2 be the functions defined by
√ πi √ πi √ πi
y0 (ζ) = 2πe− 4 Ai(ζ), y1 (ζ) = 2πe− 4 ω Ai(ωζ), y2 (ζ) = 2πe− 4 ω 2 Ai(ω 2 ζ), (B.1)

where Ai is the usual Airy function (cf. [32, Chapter 9]) and ω = e2πi/3 . We then define a 2 × 2
matrix-valued function Ψ(Ai) by
  
y 0 (ζ) −y 2 (ζ)
 y ′ (ζ) −y ′ (ζ) ,

 arg ζ ∈ (0, 2π
3 ),
0 2


  

 −y1 (ζ) −y2 (ζ) 2π
′ (ζ) −y ′ (ζ) , arg ζ ∈ ( 3 , π),


−y

(Ai) 1 2
Ψ (ζ) =   (B.2)
−y2 (ζ) y1 (ζ) 2π
 −y ′ (ζ) y ′ (ζ) , arg ζ ∈ (−π, − 3 ),





  2 1 
 y0 (ζ) y1 (ζ)
arg ζ ∈ (− 2π

 , 3 , 0).
y0′ (ζ) y1′ (ζ)


It is well-known that det(Ψ(Ai) (z)) = 1 and Ψ(Ai) (ζ) is the unique solution of the following 2 × 2
RH problem; cf. [15, Section 7.6].

RH Problem 47.

1. Ψ(ζ) is analytic in C \ ΓAi , where the contour ΓAi is defined in

2πi 2πi
ΓAi := e− 3 [0, +∞) ∪ R ∪ e 3 [0, +∞) (B.3)

with the orientation shown in Figure 8.

2. For z ∈ ΓAi , we have

 !
 1 1
, arg ζ = 0,






 0 1!

 1 0
Ψ+ (ζ) = Ψ− (ζ) , arg ζ = ± 2π
3 , (B.4)


 1 1 !


 0 1
 −1 0 , arg ζ = π.




3. As ζ → ∞, we have
 
1 1 1 − πi4 σ3 2 3
− 14 σ3 3
Ψ(ζ) = ζ √ e (I + O(ζ − 2 ))e− 3 ζ 2 σ3 . (B.5)
2 −1 1

4. As ζ → 0, we have Ψi,j (ζ) = O(1), where i, j = 1, 2.

We note that the jump condition (B.4) can be derived from the identity [15, Equation
(7.116)]
Ai(ζ) + ω Ai(ωζ) + ω 2 Ai(ω 2 ζ) = 0. (B.6)

52
 + −
− + 1 0
( eπiα
( 11 01 ) 0 1 1)
− ( −1 0 )
0 1
+ ( −1 0 ) ( 10 11 ) +
− − +
1 0
( 11 01 ) − ( e−πiα 1)
+
− +

Figure 8: The jump contour ΓAi for the RH problem Figure 9: The jump contour ΓBe for
(Be)
47 for Ψ(Ai) . the RH problem 47 for Ψα .

B.2 The Bessel parametrix

The Bessel kernel used in our paper is essentially the Ψ̃ constructed in [22, Equation (6.51)]. In
this subsection, let w0 , w1 , w2 and w3 be the functions defined by
1 −i 1 1 1 1 1
w0 (ζ) = Iα (2ζ 2 ), w1 (ζ) = Kα (2ζ 2 ), w2 (ζ) = Hα(1) (2(−ζ) 2 ), w3 (ζ) = Hα(2) (2(−ζ) 2 ),
π 2 2
(B.7)

where all ζ 1/2 are defined by the principal branch on the sector arg ζ ∈ (−π, π), Iα and Kα are
(1) (2)
modified Bessel functions of order α, and Hα and Hα are Hankel functions of order α of the
(Be)
first and second kind, respectively. We then define a 2 × 2 matrix-valued function Ψα by
 !
w0 (ζ) w1 (ζ)
arg ζ ∈ (− 2π 2π

, 3 , 3 ),


′ (ζ) w ′ (ζ)




 w 0 1 !
    w (ζ) −w (ζ)
(Be) 1 0 2 3 α
πiσ3
Ψα (ζ) = × e 2 , arg ζ ∈ ( 2π
3 , π), (B.8)
0 −2πiζ 
 w ′
2 (ζ) −w ′
3 (ζ)

 !
 w 3 (ζ) w 2 (ζ) − α πiσ3
, arg ζ ∈ (−π, − 2π

 w′ (ζ) w′ (ζ) e 2 3 ).



3 2

(Be) (Be)
It is well known that det(Ψα ) = 1 and Ψα is the unique solution of the following RH
problem:

RH Problem 48.

1. Ψ(ζ) is analytic in C \ ΓBe , where the contour ΓBe is defined in

2πi 2πi
ΓBe := e− 3 [0, +∞) ∪ R− ∪ e 3 [0, +∞) (B.9)

with the orientation shown in Figure 9.

2. For z ∈ ΓBe , we have

 !
 1 0 2π
, arg ζ = 3 ,


πiα 1




 e !

 1 0
Ψ+ (ζ) = Ψ− (ζ) −πiα 1
, arg ζ = − 2π
3 , (B.10)


 e !


 0 1
 −1 0 , arg ζ = π.




53
 3. As ζ → ∞, we have
 
− 12 σ3 − 14 σ3 1 1 −i 1 1
Ψ(ζ) = (2π) ζ √ (I + O(ζ − 2 ))e2ζ 2 σ3 . (B.11)
2 −i 1

4. As ζ → 0, if α ∈ (−1, 0), then

O(ζ α/2 ) O(ζ α/2 ) O(ζ α/2 ) O(ζ α/2 )

   
−1
Ψ(ζ) = , Ψ(ζ) = , (B.12)
O(ζ α/2 ) O(ζ α/2 ) O(ζ α/2 ) O(ζ α/2 )

if α = 0, then
   
O(log ζ) O(log ζ) −1 O(log ζ) O(log ζ)
Ψ(ζ) = , Ψ(ζ) = , (B.13)
O(log ζ) O(log ζ) O(log ζ) O(log ζ)

and if α > 0, then in the sector arg ζ ∈ (−2π/3, 2π/3),

O(ζ α/2 ) O(ζ −α/2 ) O(ζ −α/2 ) O(ζ −α/2 )

   
−1
Ψ(ζ) = , Ψ(ζ) = , (B.14)
O(ζ α/2 ) O(ζ −α/2 ) O(ζ α/2 ) O(ζ α/2 )

and in the sector arg ζ ∈ (2π/3, π) or in the sector arg ζ ∈ (−π, −2π/3),

O(ζ −α/2 ) O(ζ −α/2 ) O(ζ −α/2 ) O(ζ −α/2 )

   
−1
Ψ(ζ) = , Ψ(ζ) = . (B.15)
O(ζ −α/2 ) O(ζ −α/2 ) O(ζ −α/2 ) O(ζ −α/2 )

Like (B.6), the jump condition (B.10) can be derived from the identity [22, Proof of Theorem
6.3]
w2 (ζ) + w3 (ζ) = w0 (ζ). (B.16)

References
[1] Harold U Baranger and Pier A Mello. Mesoscopic transport through chaotic cavities: A
random S-matrix theory approach. Physical review letters, 73(1):142, 1994.

[2] C. W. J. Beenakker. Random-matrix theory of quantum transport. Rev. Mod. Phys.,

69:731–808, Jul 1997.

[3] C. W. J. Beenakker and B. Rejaei. Nonlogarithmic repulsion of transmission eigenvalues

in a disordered wire. Phys. Rev. Lett., 71:3689–3692, Nov 1993.

[4] C. W. J. Beenakker and B. Rejaei. Exact solution for the distribution of transmission
eigenvalues in a disordered wire and comparison with random-matrix theory. Phys. Rev.
B, 49(11):7499, 1994.

[5] P. M. Bleher and A. B. J. Kuijlaars. Random matrices with external source and multiple
orthogonal polynomials. Int. Math. Res. Not., (3):109–129, 2004.

[6] Alexei Borodin. Biorthogonal ensembles. Nuclear Phys. B, 536(3):704–732, 1999.

[7] Gaëtan Borot, Alice Guionnet, and Karol K. Kozlowski. Large-N asymptotic expansion for
mean field models with Coulomb gas interaction. Int. Math. Res. Not. IMRN, (20):10451–
10524, 2015.

54
 [8] Christophe Charlier. Asymptotics of Muttalib-Borodin determinants with Fisher-Hartwig
singularities. Selecta Math. (N.S.), 28(3):Paper No. 50, 60, 2022.

[9] Dimitris Cheliotis. Triangular random matrices and biorthogonal ensembles. Statist.
Probab. Lett., 134:36–44, 2018.

[10] Tom Claeys and Stefano Romano. Biorthogonal ensembles with two-particle interactions.
Nonlinearity, 27(10):2419–2444, 2014.

[11] Tom Claeys and Dong Wang. Random matrices with equispaced external source. Comm.
Math. Phys., 328(3):1023–1077, 2014.

[12] Tom Claeys and Dong Wang. Universality for random matrices with equi-spaced external
source: a case study of a biorthogonal ensemble. J. Stat. Phys., 188(2):Paper No. 11, 28,
2022.

[13] P. Deift, T. Kriecherbauer, K. T-R McLaughlin, S. Venakides, and X. Zhou. Strong asymp-
totics of orthogonal polynomials with respect to exponential weights. Comm. Pure Appl.
Math., 52(12):1491–1552, 1999.

[14] P. Deift, T. Kriecherbauer, K. T.-R. McLaughlin, S. Venakides, and X. Zhou. Uniform

asymptotics for polynomials orthogonal with respect to varying exponential weights and
applications to universality questions in random matrix theory. Comm. Pure Appl. Math.,
52(11):1335–1425, 1999.

[15] P. A. Deift. Orthogonal polynomials and random matrices: a Riemann-Hilbert approach,

volume 3 of Courant Lecture Notes in Mathematics. New York University Courant Institute
of Mathematical Sciences, New York, 1999.

[16] A. S. Fokas, A. R. Its, and A. V. Kitaev. Discrete Painlevé equations and their appearance
in quantum gravity. Comm. Math. Phys., 142(2):313–344, 1991.

[17] A. S. Fokas, A. R. Its, and A. V. Kitaev. The isomonodromy approach to matrix models
in 2D quantum gravity. Comm. Math. Phys., 147(2):395–430, 1992.

[18] Peter J. Forrester and Dong Wang. Muttalib-Borodin ensembles in random matrix theory—
realisations and correlation functions. Electron. J. Probab., 22:Paper No. 54, 43, 2017.

[19] S Iida, HA Weidenmüller, and JA Zuk. Statistical scattering theory, the supersymmetry
method and universal conductance fluctuations. Annals of Physics, 200(2):219–270, 1990.

[20] RA Jalabert, J-L Pichard, and CWJ Beenakker. Universal quantum signatures of chaos in
ballistic transport. Europhysics Letters, 27(4):255, 1994.

[21] Tiefeng Jiang. A variance formula related to a quantum conductance problem. Physics
Letters A, 373(25):2117–2121, 2009.

[22] A. B. J. Kuijlaars, K. T.-R. McLaughlin, W. Van Assche, and M. Vanlessen. The Riemann-
Hilbert approach to strong asymptotics for orthogonal polynomials on [−1, 1]. Adv. Math.,
188(2):337–398, 2004.

[23] A. B. J. Kuijlaars and L. D. Molag. The local universality of Muttalib-Borodin biorthogonal

ensembles with parameter θ = 21 . Nonlinearity, 32(8):3023–3081, 2019.

55
[24] Arno B. J. Kuijlaars. Multiple orthogonal polynomials in random matrix theory. In Pro-
ceedings of the International Congress of Mathematicians. Volume III, pages 1417–1432,
New Delhi, 2010. Hindustan Book Agency.

[25] Arno B. J. Kuijlaars. A vector equilibrium problem for Muttalib-Borodin biorthogonal

ensembles. SIGMA Symmetry Integrability Geom. Methods Appl., 12:Paper No. 065, 15,
2016.

[26] P. A. Lee and A. Douglas Stone. Universal conductance fluctuations in metals. Phys. Rev.
Lett., 55:1622–1625, Oct 1985.

[27] D. S. Lubinsky, A. Sidi, and H. Stahl. Asymptotic zero distribution of biorthogonal poly-
nomials. J. Approx. Theory, 190:26–49, 2015.

[28] Pier A. Mello. Macroscopic approach to universal conductance fluctuations in disordered

metals. Phys. Rev. Lett., 60:1089–1092, Mar 1988.

[29] Pier A. Mello and A. Douglas Stone. Maximum-entropy model for quantum-mechanical
interference effects in metallic conductors. Phys. Rev. B, 44:3559–3576, Aug 1991.

[30] L. D. Molag. The local universality of Muttalib-Borodin ensembles when the parameter θ
is the reciprocal of an integer. Nonlinearity, 34(5):3485–3564, 2021.

[31] K. A. Muttalib. Random matrix models with additional interactions. J. Phys. A,

28(5):L159–L164, 1995.

[32] Frank W. J. Olver, Daniel W. Lozier, Ronald F. Boisvert, and Charles W. Clark, editors.
NIST handbook of mathematical functions. U.S. Department of Commerce, National In-
stitute of Standards and Technology, Washington, DC; Cambridge University Press, Cam-
bridge, 2010. With 1 CD-ROM (Windows, Macintosh and UNIX).

[33] DV Savin and H-J Sommers. Shot noise in chaotic cavities with an arbitrary number of
open channels. Physical Review B, 73(8):081307, 2006.

[34] Craig A. Tracy and Harold Widom. Correlation functions, cluster functions, and spacing
distributions for random matrices. J. Statist. Phys., 92(5-6):809–835, 1998.

[35] Dong Wang and Lun Zhang. A vector Riemann-Hilbert approach to the Muttalib-Borodin
ensembles. J. Funct. Anal., 282(7):Paper No. 109380, 84, 2022.

[36] Lun Zhang. Local universality in biorthogonal Laguerre ensembles. J. Stat. Phys.,
161(3):688–711, 2015.

56