A stochastic version of the Hopfield-Ninio kinetic proofreading model

Eugenia Franco and Juan J. L. Velázquez

arXiv: 2405.10580v1 · PDF · 17 May 2024
Categories: math.PR
Licence: CC BY 4.0 The site shows the text.

Abstract

In this paper we study a simple stochastic version of the Hopfield-Ninio kinetic proofreading model. The model is characterized by means of two parameters, the unbinding time, which depends on the binding energy between a ligand and a receptor, and the number of times M \geq 1 that a ligand attaches to a receptor. We prove that, under suitable assumptions on M, our model has an extreme specificity, i.e. it is capable to discriminate between different ligands, and a high sensitivity, i.e. the response of the system does not change in a significant manner for ranges of ligands varying within several orders of magnitude. Additional quantities like the amount of energy used by the network or the time required to yield a response will be also computed. We also show that our results are robust, i.e., they do not depend on the specific choice of parameters that we make in this paper.

Items 62

Lemma

Lemma 2.1

DZ7I

Let \tau >0 and N \geq 1 and assume that n(0)=e_1 \in \mathbb R^N and R(0)=0. The solution of (?) is (n(t), R(t))=((n_1(t), \dots n_N(t)), R(t)) where

n_1(t)=e^{- t \left( \frac{1 + \tau }{ \tau} \right) }, \quad t >0

while

n_k(t)= \frac{t^{k-1}}{(k-1)!} e^{- t \left( \frac{1 + \tau}{\tau} \right) }, \quad k= 2, \dots N, \quad t >0

and

R(t) = \int_0^t n_N(s) ds \quad t >0.

Proof of Lemma 2.1

Since \frac{d n_1 }{ dt } = - \left( \frac{1+\tau}{ \tau} \right) n_1 and n(0)=e_1 we immediately deduce that n_1(t)=e^{- t \left( \frac{1+\tau}{ \tau} \right) }. We conclude the proof by induction. Indeed

\begin{aligned} n_2(t)= \int_0^t n_1(s) e^{(s-t) (\frac{1+\tau }{ \tau })} ds = te^{- t \left( \frac{1 + \tau }{ \tau} \right) }. \end{aligned}

Hence (2.1) holds for k=2. Assume now that for some k \in \{ 2, \dots, N \} it holds that n_k is given by (2.1). Then n_{k+1} (t) is given also by (2.1). Indeed

\begin{aligned} n_{k+1}(t)= \int_0^t n_k(s) e^{(s-t) ( \frac{1+\tau }{\tau })} ds =\frac{1}{(k-1)!} e^{- t \left( \frac{1+\tau }{ \tau} \right) } \int_0^t s^{k-1} ds=\frac{1}{k!} t^k e^{- t \left( \frac{1+\tau }{ \tau} \right) }. \end{aligned}

As a consequence (2.1) holds for every k \in \{2, \dots, N\}. Finally (2.1) follows just by the fact that \frac{d}{dt} R = n_N.

Proposition

Proposition 2.2

OC8Y

Assume \tau >0 and N \geq 1. Let (n_1(t) , n_2(t) , \dots, n_N(t) , R(t) ) be the solution of (?). Then the function p defined as (?) is given by (?).

Proof of Proposition 2.2

Notice that

n_N(t)=\left( \frac{\tau } {1+ \tau } \right)^N E\left(t; N , \frac{1+\tau }{ \tau }\right).

where E\left(t; N , \frac{1+\tau }{ \tau }\right) is the Erlang distribution (see [RE]) of parameters N and \frac{1+\tau}{ \tau }, where

E( t; n, \theta)= \frac{ \theta^n t^{n-1}e^{- \theta n }} {(n-1) ! } \quad x , \theta \geq 0.

Therefore

\begin{aligned} p(\tau) = \int_0^\infty n_N(t) dt = \left( \frac{\tau }{ 1+\tau } \right)^N. \end{aligned}

Definition

Definition 3.1

XJKZ

Let M >1, N > 1 and let p(\tau ) be given by (?). We define the critical time \tau_c= \tau_c(M) as the solution of

M p(\tau_c)=1.

Proposition

Proposition 3.2 (M \approx 1)

7G2R

Assume that M is a constant function of N and that M>1, then

\lim_{N \to \infty } \frac{ \tau_c \log(M) }{N} =1.

Proof of Proposition 3.2

Since M is a constant, then (?) implies the result.

Proposition

Proposition 3.3 (M\approx 1)

2S69

Assume that M>1 is a constant function of N. Then

\lim_{N \to \infty} p_M(\xi \tau_c) = 1- \left( 1-e^{ - \frac{\log(M)}{\xi} } \right)^{M},

where \tau_c is given by (?) and \xi >0.

Proof of Proposition 3.3

In this case we have that

\begin{aligned} q(\tau) :=1-p_M(\xi \tau_c) = (1-p(\xi \tau_c) )^{M} = \left( 1- \left(\frac{\xi \tau_c}{1 + \xi \tau_c} \right)^N \right)^{M}. \end{aligned}

We deduce that

\begin{aligned} \left(\frac{\tau_c \xi }{1 + \tau_c \xi } \right)^N= \left(\frac{1}{1 + \frac{1}{\tau_c \xi} } \right)^N = \exp \left( N \log \left(\frac{1}{1 + \frac{1}{\tau_c \xi} } \right)\right) \sim \exp \left( - \frac{N}{1 + \tau_c \xi } \right) \text{ as } N \rightarrow \infty. \end{aligned}

Notice that in the above computation we used the fact that \tau_c \rightarrow \infty as N \rightarrow \infty. Equality (Proof) implies

1-p_M(\xi \tau_c) = q( \xi \tau_c ) \sim \left( 1 - \exp \left( - \frac{N}{1 + \tau_c \xi } \right) \right)^{M} \text{ as } N \rightarrow \infty.

Using the fact that \tau_c \sim \frac{N}{\log(M)} as N \rightarrow \infty we obtain (3.3).

Proposition

Proposition 3.4 (M \gg 1)

JYR7

Assume that \lim_{N \to \infty } M= \infty.

If \lim_{N \to \infty} \frac{\log(M)}{N }= 0, then \lim_{N \to \infty} \frac{ \tau_c\log(M)}{N }=1.
If \lim_{N \to \infty} \frac{\log(M)}{N}= b >0, then \lim_{N \to \infty} \tau_c =\frac{1}{e^b-1}.
If \lim_{N \to \infty} \frac{\log(M)}{N}=\infty, then \lim_{N \to \infty} \tau_c M^{1/N} =1.

Used by Proof, Proof, Proof

Proof of Proposition 3.4

The result follows by (?).

Theorem

Theorem 3.5 (M \gg 1)

4HC5

Assume that \lim_{N \to \infty } M = \infty. Then the function p_M defined in (?) satisfies

\lim_{N \to \infty} p_M \left( \tau_c + \frac{\tau_c \xi (1+\tau_c)}{N} \right)= 1- \exp \left( - e^{ \xi} \right),

for \xi\in\mathbb R.

Used by Proof, Proof

Proof of Theorem 3.5

First we notice that if \tau = (\tau_c (1+\frac{ \xi}{N} (1+\tau_c )), then

\begin{aligned} \left( \frac{\tau }{1+\tau }\right)^N = \left( \frac{ \tau_c \left( 1+ \frac{(1+\tau_c) \xi}{N} \right) }{1 +\tau_c \left( 1+ \frac{(1+\tau_c) \xi}{N} \right) } \right)^N &= \left( \frac{\tau_c}{1+\tau_c} \right)^N \left( \frac{ (1+\tau_c) \left( 1+ \frac{(1+\tau_c) \xi}{N} \right) }{1 +\tau_c \left( 1+ \frac{(1+\tau_c) \xi}{N} \right) } \right)^N \\ &= \left( \frac{\tau_c}{1+\tau_c} \right)^N \left( 1+ \frac{\frac{ (1+\tau_c) \xi }{N}}{1+ \tau_c \left( 1+ \frac{ (1+\tau_c) \xi }{N} \right) } \right)^N . \end{aligned}

Therefore, since Proposition 3.4 guarantees that when \lim_{N \to \infty} M = \infty we have that \lim_{N \to \infty } \frac{\tau_c}{N } =0 we deduce that

\lim_{N \to \infty } \left(\frac{\tau }{1 + \tau }\right)^N = \lim_{N \to \infty } \left( \frac{\tau_c}{1+\tau_c} \right)^N \lim_{N \to \infty } \left( 1+\frac{ \xi }{N} \right)^N

Since

\lim_{N \to \infty } \left( 1+\frac{ \xi }{N} \right)^N = e^\xi.

we deduce that

\lim_{N \to \infty} \left( \frac{\tau}{1+\tau} \right)^N = e^\xi \lim_{N\to \infty } \left( \frac{\tau_c}{1+\tau_c} \right)^N.

Since \lim_{N \to \infty} \left( \frac{ \tau_c }{1 + \tau_c } \right)^{N}= \lim_{N \to \infty}\frac{1}{M} =0 we have that \lim_{N \to \infty} \left( \frac{\tau}{1+\tau} \right)^N =0.

As a consequence we arrive at the following asymptotics for q(\tau)=1-p_M(\tau)

\begin{aligned} q(\tau) &= (1-p(\tau) )^M = \exp \left( M \log \left( 1- \left(\frac{\tau}{1 + \tau} \right)^N \right) \right) \\ & = \exp \left( \left( \frac{1 + \tau_c }{ \tau_c } \right)^N \log \left( 1- \left(\frac{\tau}{1 + \tau} \right)^N \right) \right) \sim \exp \left( - \left( \frac{1 + \tau_c }{ \tau_c } \right)^N \left(\frac{\tau}{1 + \tau} \right)^N \right), \text{ as } N \to \infty. \end{aligned}

Proposition 3.4, together with the assumptions on \tau, imply

\begin{aligned} \frac{1}{ \frac{1 + \tau_c }{ \tau_c } \frac{\tau}{1 + \tau} } = 1 + \frac{1}{(1+\tau_c ) } \left( \frac{\tau_c}{\tau}-1 \right) \rightarrow 1 \text{ as } N \rightarrow \infty. \end{aligned}

Therefore we deduce from (?) that

\begin{aligned} q(\tau ) &\sim \exp \left( - \left( \frac{1 + \tau_c }{ \tau_c } \right)^N \left(\frac{\tau}{1 + \tau} \right)^N \right) = \exp \left( - \exp \left( N \log \left( \frac{1}{1 + \frac{1}{(1+\tau_c ) } \left( \frac{\tau_c}{\tau}-1 \right)} \right) \right) \right)\\ & \sim \exp \left( - \exp \left( \frac{N}{\tau_c} \frac{\tau_c-\tau}{ (1+\tau)} \right) \right) \sim \exp \left( - \exp \left( \frac{N}{\tau_c} \frac{\tau_c-\tau}{ (1+\tau_c)} \right) \right) \sim \exp \left( - e^{\xi} \right) \text{ as } N \to \infty. \end{aligned}

Then (3.5) follows.

Uses Proposition 3.4

Corollary

Corollary 3.6

KCBA

Assume that M is such that \lim_{N \to \infty } M= \infty. Let p_M be defined as in (?).

If \lim_{N \to \infty } \frac{\log ( M) }{N }=0, then
\lim_{N \to \infty} p_M \left( \tau_c + \frac{\tau_c \xi}{\log(M)} \right)= 1- \exp \left( - e^{\xi} \right)
for \xi\in\mathbb R.
If \lim_{N \to \infty } \frac{\log ( M) }{N }=\infty, then
\lim_{N \to \infty} p_M \left( \tau_c + \frac{\tau_c \xi}{N} \right)= 1- \exp \left( - e^{\xi} \right)
for \xi\in\mathbb R.

Proof of Corollary 3.6

This corollary follows by Proposition 3.4 and by Theorem 3.5.

Uses Proposition 3.4, Theorem 3.5

Proposition

Proposition 4.1 (M=1)

RSU5

Assume that M= 1. Then

\lim_{ N \to \infty } p(x N ) = e^{ - \frac{1}{x}},

for x >0.

Proof of Proposition 4.1

Using (?) we deduce that

\begin{aligned} p(x N )= & \left(\frac{ N x }{1 + Nx } \right)^N = \exp \left( N \log \left(\frac{1}{1 + \frac{1}{N x} } \right)\right) \sim \exp \left( - \frac{N}{1 + N x } \right) \sim e^{ - \frac{1}{ x } }\text{ as } N \rightarrow \infty. \end{aligned}

Hence (4.1) follows.

Proposition

Proposition 4.2 (M=1)

B5EJ

The function \overline p defined as (4.2) satisfies

\lim_{ N \to \infty } \overline p(\xi N ) =\frac{\xi }{1+\xi} ,

for \xi >0.

Proof of Proposition 4.2

Using (?) we deduce that \overline p(\xi N )= \frac{ N \xi }{N + N\xi } \sim \frac{\xi }{ 1 + \xi }\text{ as } N \rightarrow \infty.

Proposition

Proposition 4.3 (M large)

LW3X

If \lim_{N \to \infty } \frac{M}{N} =1, then \lim_{N \to \infty } \frac{M}{N}\overline \tau_c =1. If, instead \lim_{N \to \infty } \frac{\log(M)}{N} =b>0, then \lim_{N \to \infty } \frac{ (e^{bN} -1)\overline \tau_c }{N} = 1. Moreover, in both the two cases, the function \overline p_M defined as

\overline p_M(\tau):=1- \left( 1-\overline p (\tau) \right)^M

is such that

\lim_{N \to \infty } \overline p_M(x \overline \tau_c ) =1- e^{-x}

for x >0.

Proof of Proposition 4.3

The asymptotics of \tau_c as N \to \infty follows by (?). Moreover, in both cases we obtain that

\begin{aligned} \overline p_M( \tau)= 1 - \left( 1 - \frac{\tau}{\tau+ N }\right)^M = 1 - \left(\frac{N}{ \tau + N } \right)^M = 1- \left( \frac{1}{1+\frac{\tau}{N} }\right)^M \sim 1- e^{ - \frac{M}{N}\tau}, \end{aligned}

where \tau = x \overline \tau_c. In both cases we obtain (4.2).

Definition

Definition 5.1

K679

Let M> 1, N \geq 1 and L> 1. We define the critical unbinding time \tau_c(L) as the solution of

L M p(\tau_c(L))=1.

Proposition

Proposition 5.2 (M \approx 1and L\approx 1)

BJ49

Assume that L>1 and M >1 are both constant. Then

\lim_{N \to \infty } \frac{\tau_c(L)\left( \log(M) + \log(L)\right) }{N} =1

Proof of Proposition 5.2

The statement follows by (?).

Proposition

Proposition 5.3 (M \approx 1 and 1 \ll \log(L) \ll N)

8MNT

Assume that \lim_{N \to \infty } L =\infty and \lim_{N \to \infty} \frac{\log(L)}{N}=0 and M >1 and constant. Then

\lim_{N \to \infty } \frac{\tau_c(L)\left( \log(L)\right) }{N} =1

Proposition

Proposition 5.4 (1 \ll \log(M) \ll N and \log(L) \ll N)

MO4T

Assume that \lim_{N \to \infty } M = \infty and \lim_{N \to \infty}\frac{\log (L) }{N }= \lim_{N \to \infty}\frac{\log ( M) }{N }=0.

If \lim_{N \to \infty} \frac{\log(M)}{\log(L)}= 0, then \lim_{N \to \infty} \frac{ \tau_c(L) \log(L) }{N} =1.
If \lim_{N \to \infty} \frac{\log(M)}{\log(L)}= c >0, then \lim_{N \to \infty} \frac{(c+1) \log(L) \tau_c(L) } {N} =1.
If \lim_{N \to \infty} \frac{\log(M)}{\log(L)}=\infty, then \lim_{N \to \infty} \frac{\log(M) \tau_c(L) }{N} =1.

Proof of Proposition 5.4

The statement follows by (?).

Proposition

Proposition 5.5

0Q3S

Assume that \lim_{N \to \infty } M = \infty and \lim_{N \to \infty } \frac{\log (M) }{N } = b >0 and that \lim_{N \to \infty} \frac{\log(L) }{ N } =0. Then \lim_{N \to \infty} \tau_c(L ) = \frac{1}{e^b -1 }.

Proposition

Proposition 5.6

WJQM

Assume that \lim_{N \to \infty } \frac{\log ( M) }{N } = \infty.

If \lim_{N \to \infty} \frac{\log(M)}{\log(L)}= 0, then \lim_{N \to \infty} \tau_c(L) L^{1/N} =1 .
If \lim_{N \to \infty} \frac{\log(M)}{\log(L)}= d >0, then \lim_{N \to \infty} (d+1) L^{1/N}\tau_c (L) =1.
If \lim_{N \to \infty} \frac{\log(M)}{\log(L)}=\infty, then \lim_{N \to \infty} {M}^{1/N}\tau_c(L) =1.

Proof of Proposition 5.6

Since \lim_{N \to \infty } \frac{\log (L M) }{N } =\infty, then (?) implies

\tau_c\sim \frac{1} {\exp\left(\frac{ \log({M})+ \log(L)}{N}\right)} \text{ as } N \to \infty.

The desired conclusion follows by considering the dominant term in the three cases.

Lemma

Lemma 6.1

5WHG

Let \tau >0, N \geq 1. Let \{ m_k (t) \}_{k=1}^N be the solution of (?). Then

m_k(\infty ) := \lim_{ t \rightarrow \infty } m_k(t)= \frac{1}{1+\tau } \left( \frac{\tau }{1+\tau }\right)^k, \text{ for every } k \in \{0, \dots, N -1 \}.

Moreover

m_N(\infty):= \lim_{ t \rightarrow \infty } R (t)= \left( \frac{\tau }{1+\tau } \right)^N.

Used by Theorem 6.2, Theorem 6.3, Theorem 6.4

Proof of Lemma 6.1

For every k \in \{ 0, \dots, N-1\} we have that

\hat{m}_k(z) = \frac{1}{\tau} \hat{n}_{k+1}(z)= \frac{1}{\tau z } \left( \frac{1}{1+z +\frac{1}{\tau}}\right)^{k+1} \text{ for } z \in \mathbb C,

where \hat{m}_k is the Laplace transform (see [feller1991introduction] for the definition of Laplace transform) of m_k and \hat{n}_k is the Laplace transform of n_k.

Using the final value theorem for Laplace transforms ([feller1991introduction]) we deduce that for every k \in \{0, \dots, N-1 \}

\begin{aligned} m_k(\infty)= \lim_{ t \rightarrow \infty } m_k(t)= \lim_{ z \rightarrow 0 } z \hat{m}_k(z) = \frac{1}{1+\tau } \left( \frac{\tau }{1+\tau }\right)^k. \end{aligned}

Moreover,

R(\infty)= \lim_{ t \rightarrow \infty } R (t)=\int_0^\infty n_N(t) dt = \left( \frac{\tau }{1+\tau } \right)^N.

Theorem

Theorem 6.2

IKKS

Let \{ m_k (\infty) \}_{k=0}^N be given by (6.1) as a function of \tau= N x for x >0. Then

\mu_N \rightharpoonup \mu \text{ as } N \rightarrow \infty

where \mu_N and \mu are the probability measures on [0,1]

\mu_N (d \ell ) := \sum_{k=0}^N m_k (\infty) \delta\left(\ell - \frac{k}{N} \right)

for every N \geq 1 and

\mu (d \ell) := \frac{e^{- \frac{\ell }{x} }}{x} \chi_{[0,1) } ( \ell) d \ell + e^{- \frac{\ell }{x} } \delta (\ell -1)

Uses Lemma 6.1

Proof (Proof of Theorem 6.2) of Theorem 6.2

Notice that

\begin{aligned} \mu_N (d \ell ) &= \sum_{k=0}^N m_k (\infty) \delta\left(\ell - \frac{k}{N} \right) = \sum_{k=0}^{N-1} m_k (\infty) \delta\left(\ell - \frac{k}{N} \right) + m_N (\infty) \delta\left(\ell - 1 \right) \\ &= \frac{1}{\tau} \sum_{k=0}^{N-1} \left( \frac{\tau }{1+\tau} \right)^{k+1 } \delta\left(\ell - \frac{k}{N} \right) + \left( \frac{\tau }{1+\tau} \right)^N \delta\left(\ell - 1 \right) . \end{aligned}

where \tau= x N. Moreover, since

\left( \frac{\tau }{1+\tau} \right)^N = \left( \frac{ x N }{1+ x N } \right)^N \sim e^{- \frac{1}{x} } \text{ as } N \rightarrow \infty,

we deduce that

\left( \frac{\tau }{1+\tau} \right)^N \delta\left(\ell - 1 \right) \rightharpoonup e^{-\frac{1}{x}} \delta(\ell -1 ) \text{ as } N \to \infty .

Moreover,

\frac{1}{\tau} \sum_{k=0}^{N-1} \left( \frac{\tau }{1+\tau} \right)^{k+1 } \delta\left(\ell - \frac{k}{N} \right) = \frac{1}{xN} \sum_{k=0}^{N-1} \left( \frac{x N }{1+xN}\right)^{k+1} \delta \left(\ell - \frac{k }{N} \right) \rightharpoonup \frac{e^{- \frac{ \ell }{x }}}{x} \chi_{[0, 1) } ( d \ell ) \text{ as } N \to \infty.

Indeed for every f \in C_b ([0, \infty)]) we have that

\frac{1}{N }\sum_{k=0}^{N-1} \left( \frac{x N }{1+xN}\right)^{k+1} f\left( \frac{k}{N} \right) \rightarrow \int_0^1 e^{- \frac{ \ell }{x }} f(\ell) d\ell

as N \to \infty. The desired conclusion then follows.

Theorem

Theorem 6.3

CLU1

Let \{ m_k (\infty) \}_{k=0}^{N} be given by (6.1) as a function of \tau= N x for x >0. Assume that M >1 is constant. Then

\nu_N \rightharpoonup \nu \text{ as } N \rightarrow \infty

where \nu_N is given by

\nu_N (d \ell ) := \sum_{k=0}^N Q_k \delta\left(\ell - \frac{k}{N} \right).

where \{ Q_k\}_{k=1}^{MN} are given by (?) as a function of \{ m_k (\infty) \}_{k=0}^{N} and

\nu (d \ell) := e^{- \frac{M}{x}} \delta(\ell - M )+ \frac{ e^{- \frac{\ell}{x}}}{x} \sum_{k=1}^M \binom{M}{k} \sum_{ j=0}^k \binom{k }{j} (-1)^{k-j+1} \frac{1}{(k-1)!} \left( \frac{\ell - M + j }{x} \right)^{k-1} \chi_{[0, M-j ] } ( \ell) d \ell .

Uses Lemma 6.1

Proof (Proof of Theorem 6.3) of Theorem 6.3

Since Q_k is defined by (?) we deduce that \nu_N = \overbrace{\mu_N * \mu_N * \dots * \mu_N}^{M \text{ times} }. As a consequence we have that

\hat{\nu}_N(z) = \left( \hat{\mu}_N (z)\right)^M.

Since \mu_N \rightharpoonup \mu as N \rightarrow \infty, by the duality between convergence of Laplace transforms on compact sets and weak convergence of measures (see [feller1991introduction]), we have that \hat{\mu}_N \rightarrow \hat{\mu } as N \rightarrow \infty on compact sets. Therefore using (Proof) we deduce that \hat{\nu}_N=\left( \hat{\mu}_N \right)^M \rightarrow \left( \hat{\mu} \right)^M on compact sets as N \to \infty. This implies that \nu_N \rightharpoonup \nu, as N \to \infty, where

\hat{\nu} (z)= \left( \hat{\mu}(z) \right)^M =\left( \frac{1}{1+ x z } \right)^M \left( 1+ x z e^{-\frac{1}{x}(1+ z x ) }\right)^M, \quad z \in \mathbb C, \ z \neq -\frac{1}{x} .

Indeed,

\hat{\mu } (z)=\int_0^\infty e^{ - z \ell } \mu(d \ell)=\int_0^1\frac{e^{-{\ell \left( z + \frac{1}{x}\right) } }}{x} d \ell + e^{- \left( z+ \frac{1}{x}\right)} = \frac{1}{1+xz} + \left( \frac{xz}{1+ xz}\right) e^{-\frac{1}{x} \left( 1+ xz\right)}, \quad z \in \mathbb C, \ z \neq - \frac{1}{x} .

Therefore, in order to obtain the probability measure \nu we need to compute the inverse Laplace of \hat{\nu }= \left( \hat{\mu} \right)^M. First of all we rewrite \hat{\nu} as follows

\begin{aligned} \hat{\nu}(z) &= \left( \frac{1}{1+ x z } \right)^M \left( e^{ z + \frac{1}{x}}+ x z \right)^M e^{- \frac{M}{x}} e^{ - M z} \\ &= \left[ \left( \frac{1}{1+ x z } \right)^M \left( e^{ z + \frac{1}{x}}+ x z \right)^M - 1 \right] e^{- \frac{M}{x}} e^{ - M z} + e^{- \frac{M}{x}} e^{ -zM} . \end{aligned}

Now notice that

e^{- \frac{M}{x}} e^{ - M z} = \int_0^\infty e^{- \frac{M}{x}} \delta(\ell - M ) e^{-z \ell} d\ell = e^{- \frac{M}{x}} \widehat{\delta (\cdot - M )} (z).

Therefore since the function z \mapsto \left[ \left( \frac{1}{1+ x z } \right)^M \left( e^{ z + \frac{1}{x}}+ x z \right)^M - 1 \right] e^{- \frac{M}{x}} e^{ - M z} is analytic we have that

\begin{aligned} \nu ( d \ell) &= \frac{1}{2 \pi i } \lim_{T \rightarrow \infty} \int_{a-i T }^{a+ i T } \left[ \left( \frac{1}{1+ x z } \right)^M \left( e^{ z + \frac{1}{x}}+ x z \right)^M - 1 \right] e^{- \frac{M}{x}} e^{ z (\ell - M)} dz + e^{- \frac{M}{x}} \delta(\ell - M ), \end{aligned}

where we consider a < - \frac{1}{x}.

Now notice that if \ell > M, then deforming the contour as \mathop{}\!\mathrm{Re}\,(z) \rightarrow -\infty, we obtain

Indeed as \mathop{}\!\mathrm{Re}\,(z) \rightarrow - \infty we have that

e^{ z (\ell - M)} \left[ \left( \frac{1}{1+ x z } \right)^M \left( e^{ z + \frac{1}{x}}+ x z \right)^M - 1 \right] \leq e^{z (\ell - M) } \rightarrow 0 \text{ as } \mathop{}\!\mathrm{Re}\, (z) \rightarrow - \infty .

Therefore, this together, with the binomial expansion of (e^{z+\frac{1}{x}} -1 + 1-xz )^M and of (e^{z+\frac{1}{x}}-1)^k we deduce that

\begin{aligned} & \frac{ e^{- \frac{M}{x}} }{2 \pi i } \lim_{T \rightarrow \infty} \int_{a-i T }^{a+ i T } e^{ z (\ell - M)} \left[ \left( \frac{1}{1+ x z } \right)^M \left( e^{ z + \frac{1}{x}}+ x z \right)^M - 1 \right] dz \\ &= \chi_{[0, M )} ( \ell) \frac{ e^{- \frac{M}{x}} }{2 \pi i } \lim_{T \rightarrow \infty} \int_{a-i T }^{a+ i T } e^{ z (\ell - M)} \left[ \left( \frac{1}{1+ x z } \right)^M \left( e^{ z + \frac{1}{x}} -1 + 1 + x z \right)^M - 1 \right] dz \\ &= \chi_{(0, M )} ( \ell) \frac{ e^{- \frac{M}{x}} }{2 \pi i } \lim_{T \rightarrow \infty}\sum_{k=1}^M \binom{M}{k} \int_{a-i T }^{a+ i T } e^{ z (\ell - M)} \left( \frac{e^{ z + \frac{1}{x}} -1 }{1+ x z }\right)^k dz \\ &= \chi_{(0, M )} ( \ell) \frac{ e^{- \frac{M}{x}} }{2 \pi i } \lim_{T \rightarrow \infty}\sum_{k=1}^M \binom{M}{k} \sum_{j=0}^k \binom{k }{j} (-1)^{k-j} \int_{a-i T }^{a+ i T } e^{ z (\ell - M)} e^{ j\left(z + \frac{1}{x}\right)} \left( \frac{ 1 }{1+ x z }\right)^k dz \\ &= e^{- \frac{M}{x}} \sum_{k=1}^M \binom{M}{k} \sum_{j=0}^k I_{k, j } (x,z; \ell ) \end{aligned}

where

I_{k,j} (x,z; \ell):= \chi_{(0, M )} ( \ell) \frac{ 1 }{2 \pi i } \binom{k }{j} (-1)^{k-j} e^{ \frac{j}{x}} \lim_{T \rightarrow \infty} \int_{a-i T }^{a+ i T } e^{ z (\ell - M+ j )} \left( \frac{ 1 }{1+ x z }\right)^k dz.

Now notice that if \ell - M + j =: r<0, then

\begin{aligned} \frac{ 1 }{2 \pi i } \lim_{T \rightarrow \infty} \int_{a-i T }^{a+ i T } e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz & = \frac{ 1 }{2 \pi i } e^{- \frac{r}{x}} \lim_{T \rightarrow \infty} \int_{a-i T }^{a+ i T } e^{ r \left( z+ \frac{1}{x} \right)} \left( \frac{ 1 }{1+ x z }\right)^k dz \\ &= \frac{1}{(k-1)!} e^{- \frac{r}{x}} \left( \frac{r}{x} \right)^{k-1} \frac{1}{x} \end{aligned}

The last equality follows by contour integration as \mathop{}\!\mathrm{Re}\,(z) \rightarrow \infty around the pole z=- \frac{1}{x} of the function

z \mapsto e^{ r \left( z+ \frac{1}{x} \right)} \left( \frac{ 1 }{1+ x z } \right)^k

and by Cauchy’s Residue Theorem. Indeed

\begin{aligned} \int_{a-i T }^{a+ i T } e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz &= \oint_{C} e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz- \int_{\Gamma_1 } e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz - \int_{\Gamma_2} e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz \\ & - \int_{\Gamma_3} e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz \end{aligned}

where \Gamma_1 is the line in the complex plan connecting a+i T to v + i T (where v > -1/x), \Gamma_2 is the line connecting v+ iT to v- i T while \Gamma_3 is the line connecting v- i T to a- iT, finally C us the cycle obtain by concatenating the line from a-i T to a+ i T with \Gamma_1, \Gamma_2 and \Gamma_3. Now notice that

\lim_{v \rightarrow \infty } \lim_{ T \to \infty } \int_{\Gamma_2 } e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz =0\ \text{ and } \lim_{ T \to \infty } \int_{\Gamma_i } e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz =0 \text{ for } i= 1,3.

Finally notice that by Cauchy Residue Theorem we have that

\frac{1}{2 \pi i } \oint_{C} e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz = - \operatorname{Res}_{z=-1/x} \left( e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k \right) = - \frac{1}{(k-1)!} e^{- \frac{r}{x}} \left( \frac{r}{x} \right)^{k-1} \frac{1}{x}

Therefore (?) follows.

On the other hand, with the same type of arguments used to derive (?) we have that if r= \ell - M + j =0, then

\begin{aligned} \frac{ 1 }{2 \pi i } \lim_{T \rightarrow \infty} \int_{a-i T }^{a+ i T } \left( \frac{ 1 }{1+ x z }\right)^k dz = \begin{cases} & 0 \text{ if } k >1, \\ & - \frac{1}{x} \text{ if } k=1 \end{cases} \end{aligned}

Finally when r > 0 we have that contour deformations as \mathop{}\!\mathrm{Re}\,(z) \rightarrow - \infty imply that

\begin{aligned} \frac{ 1 }{2 \pi i } \lim_{T \rightarrow \infty} \int_{a-i T }^{a+ i T } e^{ z r } \left( \frac{ 1 }{1+ x z }\right)^k dz=0. \end{aligned}

As a consequence we deduce that for 1 \leq k \leq M, 0 \leq j \leq k

\begin{aligned} I_{k,j} (z,x; \ell )= \binom{k }{j} (-1)^{k-j+1} \frac{1}{(k-1)!} e^{- \frac{\ell - M }{x}} \left( \frac{\ell - M + j }{x} \right)^{k-1} \frac{1}{x} \chi_{[0, M-j ] } ( \ell) d \ell . \end{aligned}

As a consequence

\begin{aligned} \nu ( d \ell) &= e^{- \frac{M}{x}} \delta(\ell - M )+ e^{- \frac{M}{x}} \sum_{k=1}^M \binom{M}{k} \sum_{j=0}^k I_{k, j } (x,z; \ell ) \end{aligned}

and the statement follows.

Theorem

Theorem 6.4

0KBJ

Let \{ m_k (\infty) \}_{k=0}^N be given by (6.1) as a function of

\tau= \tau_c \left( 1+ \frac{x}{N} (1+\tau_c) \right),

where x\in \mathbb R and \tau_c is given by (?). Assume that \lim_{N \to \infty} M = \infty. Then the random variable \mathcal E_N defined in (?) is such that

\frac{\mathcal E_N - \tau M }{ \sqrt{ \tau (1+\tau) M} } \overset{d}{\rightarrow}\mathcal N (0,1) \text{ as } N \rightarrow \infty,

where \mathcal N (0,1) is the normal distribution with mean 0 and standard deviation 1.

Uses Lemma 6.1 · Used by Lemma A.2, Proof

Proof (Proof of Theorem 6.4) of Theorem 6.4

Since G(z)^M is the generating function of Q_k to prove the theorem we want to compute

\begin{aligned} Q_k &= \frac{1}{2 \pi i } \oint_{\partial B_\delta(0)} \frac{G(z)^M } {z^{k+1} } dz \text{ as } N \rightarrow \infty \text{ for } k = \frac{\tau M }{\sqrt{(1+\tau) \tau M }}, \end{aligned}

where \delta < z_0. First of all we notice that the main contribution to this integral is due to the region of values of z around the value \hat{z} at which \left| \frac{G(z)^M }{z^k }\right| reaches the maximum. Notice that when k \sim \tau M as N \to \infty we have that \hat{z} \rightarrow 1, hence \left| \frac{G(z)^M }{z^{1+k} }\right| \leq \frac{1}{z} for every z \in \mathbb C. Therefore deforming the contour as \mathop{}\!\mathrm{Re}\,(z) \rightarrow - \infty and Lemma A.2 we deduce that, if \delta \leq 1+ \frac{\log(M)}{N \sqrt{M}}, then

\begin{aligned} Q_k &= \frac{1}{2 \pi i } \oint_{\partial B_\delta(0)} \frac{G(z)^M } {z^{k+1} } dz \sim \frac{1}{2 \pi i } \oint_{\partial B_\delta(0)} \frac{ \left( G_\infty (z)\right)^M } {z^{k+1} } dz \text{ as } N \rightarrow \infty, \end{aligned}

where G_\infty (z)= \frac{z_0-1}{z_0-z}. By Cauchy Residue Theorem we deduce that

\begin{aligned} \frac{1}{2 \pi i } \oint_{\partial B_\delta(0)} \frac{\left( G_\infty (z)\right)^M } {z^{k+1} } dz &= Res \left( \frac{\left( G_\infty (z)\right)^M } {z^{k+1} }, 0 \right) = \frac{1}{k !} {\frac{d^{k} }{d z^k} (G_\infty (z))^M }_{| z=0} \\ &= \frac{M (M+1) \dots (M+k-1)}{k !} (z_0-1)^M \left( \frac{1}{z_0}\right)^{M+k} \\ &= \binom{ M+k -1 }{ k } \left(1 -\frac{1}{z_0} \right)^M \left( \frac{1}{z_0}\right)^{k}. \end{aligned}

Since \lim_{N \to \infty} M = \infty as well as \lim_{N \to \infty} k = \infty we apply Stirling formula to deduce that

\begin{aligned} \binom{ M+k -1 }{ k } \left(1 -\frac{1}{z_0} \right)^M \left( \frac{1}{z_0}\right)^{k} & = \frac{M }{ M + k } \binom{ M+k }{ k } \left(1 -\frac{1}{z_0} \right)^M \left( \frac{1}{z_0}\right)^{k} \\ & \sim \frac{M }{ M + k } \sqrt{\frac{M+k }{ 2 \pi k M}} \left( \frac{k + M }{ k z_0 } \right)^k \left( \frac{k + M }{ M } \left( 1- \frac{1}{z_0}\right) \right)^M \text{ as } N \rightarrow \infty. \end{aligned}

We now perform the change of variables \ell = \frac{k - \tau M }{\sqrt{M\tau (1+\tau) }} and use the fact that z_0 is given by (A.1) to deduce that

\left( \frac{k + M }{ M } \left( 1- \frac{1}{z_0}\right) \right)^M = \exp \left( M \ln \left( 1+ \ell \sqrt{ \frac{\tau}{M(1+\tau )}} \right) \right) \sim \exp \left( \ell \sqrt{ \frac{M\tau}{1+\tau }} \right), \text{ as } N \to \infty.

On the other hand,

\begin{aligned} \left( \frac{k + M }{ k } \frac{1}{z_0} \right)^k &= \exp \left( - (M \tau + \ell\sqrt{M\tau (1+\tau)}) \ln \left( \frac{k z_0} {M+k } \right) \right) \\ &\sim \exp \left( - (M \tau + \ell\sqrt{M\tau (1+\tau)})\ln \left( \frac{k z_0} {M+k } \right) \right) \\ & \sim \exp \left( - (M \tau + \ell\sqrt{M\tau (1+\tau)})\frac{ \ell \sqrt{M (1+\tau) \tau } } { M \tau (1+\tau)} \right) \\ & \sim \exp \left( -\left( \ell^2 + \ell \sqrt{\frac{ M\tau} { 1+ \tau }} \right) \right) \text{ as } N \rightarrow \infty. \end{aligned}

Combining the two asymptotics we deduce that for \ell = \frac{k - \tau M }{\sqrt{M\tau (1+\tau) }} we have that

\begin{aligned} \binom{ M+k -1 }{ k } \left(1 -\frac{1}{z_0} \right)^M \left( \frac{1}{z_0}\right)^{k} \sim \sqrt{\frac{ 1 }{ 2 \pi }} e^{ - \ell^2 }\text{ as } N \rightarrow \infty. \end{aligned}

Therefore the desire result follows.

Uses Lemma A.2, Lemma A.1

Lemma

Lemma 7.1

UDMC

Let N > 1 and \tau >0. The conditional probability density \Psi_R(t) defined as (?) is an Erlang distribution of parameters N and \frac{1+\tau}{\tau }. Namely \Psi_R(t)= E(t; N, \frac{1 + \tau }{\tau }) where E is the Erlang probability distribution defined in (Proof) and

\int_0^\infty t \Psi_R(t) dt= \frac{N \tau }{ 1+\tau }, \text{ and } \int_0^\infty \Psi_R(t) dt= 1

Proof of Lemma 7.1

Since the solution (n_1(t), \dots n_N(t), R(t) ) of (?) is such that n_N is given by (Proof). As a consequence then we have that

\int_0^\infty n_N(t) dt = \left( \frac{\tau }{1+\tau }\right)^N .

As a consequence

\Psi_R(t) = E\left(t; N , \frac{1+\tau }{ \tau }\right).

Equality (7.1) follows by the fact that the Erlang distribution of parameters k and \lambda has mean k / \lambda.

Uses Proof

Proposition

Proposition 7.2

W8J3

Assume that \lim_{N \to \infty} M=\infty and that \lim_{N \to \infty} \frac{\log(M) }{N}= b >0. Assume that

\tau = \tau_c \left( 1+ \frac{\xi (1+\tau_c)}{N }\right)

for \xi \in \mathbb R. Then

\lim_{N \to \infty } \hat{\Psi}_M \left( \frac{\zeta }{ \tau M } \right) = \frac{1- e^{- \zeta} e^{- e^{\xi }}}{ 1 - e^{- e^{\xi }} } \cdot \frac{e^{ \xi }}{\zeta + e^{\xi }},\ \zeta \in \mathbb C.

Used by Proof

Proof (Proof of Proposition 7.2) of Proposition 7.2

First of all we prove that

\lim_{N \to \infty } e^{ \zeta } \left( \frac{ \hat{s} \left(\frac{\zeta }{\tau M } \right) }{ \hat{s} (0)} \right)^M =1

Indeed

\begin{aligned} \left( \frac{\hat{s} \left(\frac{\zeta }{\tau M } \right) }{ \hat{s} (0)} \right)^M &= \left( \frac{1}{1+ \frac{\zeta }{M }} \right)^M \left( \frac{ 1- \left( \frac{1}{1+\frac{\zeta}{\tau M } + \frac{1}{\tau}} \right)^N }{1- \left( \frac{1}{1+\frac{1}{\tau} } \right)^N } \right)^M \sim e^{- \zeta } \left( \frac{ 1- \left( \frac{1}{1+ \frac{1}{\tau}}\right)^N \left( \frac{1+ \frac{1}{\tau}}{1+\frac{\zeta}{\tau M } + \frac{1}{\tau}} \right)^N }{1- \left( \frac{1}{1+\frac{1}{\tau} } \right)^N } \right)^M \\ & \sim e^{- \zeta } \left( \frac{ 1- \left( \frac{1}{1+ \frac{1}{\tau}}\right)^N \exp \left( - \frac{N \zeta }{M (1+\tau) }\right) }{1- \left( \frac{1}{1+\frac{1}{\tau} } \right)^N } \right)^M \sim e^{- \zeta } \left( \frac{ 1- \left( \frac{1}{1+ \frac{1}{\tau}}\right)^N\left( 1- \frac{N \zeta }{M (1+\tau) }\right) }{1- \left( \frac{1}{1+\frac{1}{\tau} } \right)^N } \right)^M \\ & \sim e^{- \zeta } \left( 1- \left( \frac{1}{1+ \frac{1}{\tau}}\right)^N+ \left( \frac{1}{1+ \frac{1}{\tau}}\right)^N \frac{N \zeta }{M (1+\tau) } \right)^M \\ & \sim e^{- \zeta } \left( 1+ \left( \frac{1}{1+ \frac{1}{\tau}}\right)^N \frac{N \zeta }{M (1+\tau) } \right)^M \sim e^{- \zeta } \exp \left[ \left( \frac{\tau }{1+\tau }\right)^N \frac{N \zeta }{ (1+\tau) } \right] \text{ as } N \to \infty. \end{aligned}

Now notice that, as in the proof of Theorem 3.5 we have that

0 \leq \left( \frac{\tau }{1+\tau }\right)^N \frac{N \zeta }{ (1+\tau) } \leq \frac{1}{M } \left( 1+ \frac{\xi }{N } \right)^N \frac{ N \zeta }{ 1+ \tau } \leq e^\xi \frac{ N \zeta }{ M ( 1+ \tau ) } \rightarrow 0 \text{ as } N \to \infty

Hence \left( \frac{\hat{s} \left(\frac{\zeta }{\tau M } \right) }{ \hat{s} (0)} \right)^M \sim e^{- \zeta } as N \to \infty. Therefore (Proof) follows.

Now recall that

\left( \frac{\tau }{1+\tau }\right)^N \sim \frac{1}{M } e^{ \xi } \text{ as } N \to \infty.

Hence

\begin{aligned} 1- \hat{s}\left( \frac{\zeta }{ \tau M }\right) = \frac{1}{M } \left( \frac{\zeta}{ 1+ \frac{\zeta }{M }} + \frac{M }{ 1+ \frac{\zeta }{M }} \left( \frac{\tau }{1+\tau }\right)^N \right) \sim \frac{1}{M } \left( \zeta+ e^{ \xi } \right) \text{ as } N \to \infty \end{aligned}

Therefore using (B.1) we deduce that

\begin{aligned} \hat{ \Psi}_M \left( \frac{\zeta }{ \tau M } \right) &\sim \frac{M }{ \zeta+ e^{ \xi } } \left( \frac{1}{1+\frac{\zeta }{\tau M } + \frac{1 }{\tau }} \right)^N \frac{1- (\hat{s}(0))^M e^{- \zeta } }{1- (\hat{s}(0))^M } \\ & \sim \frac{1}{ \zeta+ e^{ \xi } } M\left( \frac{\tau}{1+\tau} \right)^N \frac{1- (\hat{s}(0))^M e^{- \zeta } }{1- (\hat{s}(0))^M } \sim \frac{e^{ \xi}}{ \zeta+ e^{ \xi }} \frac{1- (\hat{s}(0))^M e^{- \zeta } }{1- (\hat{s}(0))^M } \\ & \sim \frac{e^{ \xi}}{ \zeta+ e^{ \xi }} \frac{1- e^{- e^{ \xi}} e^{- \zeta } }{1- e^{- e^{ \xi}} } \text{ as } N \to \infty \end{aligned}

where in the last step we used the fact that

\hat{s }(0)^M = \left( 1- \left(\frac{\tau }{1+\tau }\right)^N \right)^M \sim \exp \left( - M \left( \frac{\tau }{1+\tau }\right)^N \right) \sim \exp \left( - \exp ( \xi ) \right) \text{ as } N \to \infty.

Uses Lemma B.1

Theorem

Theorem 7.3

EWG9

Assume that \lim_{N \to \infty} M=\infty and that \lim_{N \to \infty} \frac{\log(M) }{N}= b >0. Assume that

\tau = \tau_c \left( 1+ \frac{\xi (1+\tau_c)}{N }\right)

for a fixed \xi \in \mathbb R. Let t \mapsto F(t) := \tau M \Psi_M (\tau M t ). Then

\lim_{\xi \to - \infty } \lim_{N \to \infty } F(t) = \chi_{[0, 1 ] } (t) , \ \text{ a.e. } t \geq 0

and t \mapsto G (t) := \frac{\tau M }{e^{\xi} } \Psi_M( \frac{\tau M t }{e^\xi} )

\lim_{ \xi \to \infty } \lim_{N \to \infty} G (t) = e^{- t } \ \text{ a.e. } t \geq 0.

Proof (Proof of Theorem 7.3) of Theorem 7.3

By Proposition 7.2 we know that taking the limits as | \xi | \rightarrow \infty in (7.2) we obtain that

\lim_{\xi \to - \infty } \lim_{N \to \infty } \hat{\Psi}_M\left( \frac{ \zeta }{ \tau M }\right)= \lim_{\xi \to - \infty } \frac{1- e^{- \zeta} e^{- e^{ \xi }}}{ 1 - e^{- e^{ \xi }} } \cdot \frac{e^{\xi }}{\zeta + e^{\xi }} = \frac{1- e^{- \zeta} }{\zeta }

and

\lim_{\xi \to \infty } \lim_{N \to \infty } \hat{\Psi}_M\left( \frac{ \zeta e^{\xi} }{ \tau M }\right) = \lim_{\xi \to \infty } \frac{1- e^{- \zeta e^\xi } e^{- e^{ \xi }}}{ 1 - e^{- e^{ \xi }} } \cdot \frac{e^{ \xi }}{\zeta e^\xi + e^{\xi }} = \frac{1}{1+\zeta }.

Now notice that \widehat{ e^{-t} } (\zeta) = \frac{1}{1+\zeta} and \widehat{\chi_{[0, 1]} } (\zeta )= \frac{1- e^{- \zeta }}{\zeta }. Since the convergence of the Laplace transforms of measures on compact sets implies the weak convergence of the measures (see [feller1991introduction]) the desired conclusion follows.

Uses Proposition 7.2

Theorem

Theorem 8.1

2MP9

Assume that \lim_{N \to \infty } M = \infty and assume that \tau is such that \lim_{N \to \infty } M p(\tau)= \infty, where p(\tau) is given by (?). Then

\frac{X_M - M p (\tau ) }{ \sqrt{M p (\tau) \left[ 1-p(\tau)\right] }} \overset{d}{\rightarrow} \mathcal N(0,1) \text{ as } N \rightarrow \infty.

Used by Proof

Proof of Theorem 8.1

Notice that the random variable X_M follows a Binomial distribution of mean \mathbb E[X_M]=M p (\tau) and variance \operatorname{Var}[X_M ]= Mp (\tau) \left[ 1- p (\tau) \right]. By assumption we have that M p(\tau) \rightarrow \infty as N \to \infty (and hence as M \to \infty). Then the result follows by an application of the central limit theorem ([feller1991introduction]).

Corollary

Corollary 8.2

K7E2

Assume that \lim_{N \to \infty} M=\infty and \lim_{N \to \infty } \frac{\log( M)}{N} = b where b >0. Moreover assume that \tau is such that

\lim_{N \to \infty } N \frac{\tau- \tau_c }{\tau_c (1 + \tau) } = \infty,

and

\tau (1-\tau_c)\leq 2 \tau_c \text{ when } \tau_c \neq 1 \text{ and } \tau \leq 1 \text{ when } \tau_c =1, \text{ for every } N \geq 1

where \tau_c is given by (?). Then

\frac{X_M - M p (\tau ) }{ \sqrt{M p (\tau) }} \overset{d}{\rightarrow} \mathcal N(0,1) \text{ as } N \rightarrow \infty.

Proof of Corollary 8.2

First of all notice that the assumptions on \tau imply that M p (\tau) \rightarrow \infty as N \to \infty. Indeed

\begin{aligned} M p (\tau )= \left( e^b \left(\frac{\tau}{1+\tau} \right) \right)^N = \exp \left[ N \log \left( \frac{1+\tau_c}{\tau_c } \cdot \frac{\tau }{ 1 + \tau } \right) \right] \end{aligned}

Using the fact that (1-\tau_c) \tau < 2 \tau_c when \tau_c \neq 1 and that \tau \leq 1 when \tau_c = 1, we deduce that

\begin{aligned} M p (\tau) \sim \exp \left[ N \frac{\tau- \tau_c }{ \tau_c (1+\tau) } \right] \text{ as } N \to \infty. \end{aligned}

Hence \lim_{N \to \infty } M p (\tau ) = \infty. Therefore Theorem 8.1 implies that (8.1) holds. Moreover notice that under the assumptions that we have on \tau we always have that p (\tau ) \rightarrow 0 as N \to \infty. As a consequence (8.1) reduces to (8.2).

Uses Theorem 8.1

Theorem

Theorem 8.3

60JV

Assume that \lim_{N \to \infty} M=\infty and that \lim_{N \to \infty } \frac{\log( M)}{N} = b where b >0.

Assume that \tau = \tau_c \left( 1 + \frac{x}{N} (1+\tau_c) \right) where x \in \mathbb R and where \tau_c is given by (?). Then

X_M \overset{d}{\rightarrow} \operatorname{Pois} (x) \text{ as } N \rightarrow \infty.

Proof of Theorem 8.3

First of all notice that

\begin{aligned} M p(\tau) = \left[\frac{1+\tau_c }{\tau_c } \cdot \frac{\tau}{1+\tau } \right]^N \sim \left( 1+\frac{ x }{N}\right)^N \text{ as } N \to \infty. \end{aligned}

Hence

\lim_{N \to \infty } M p (\tau) =\lim_{N \to \infty } \left( 1+ \frac{x}{N} \right)^N =e^x.

Then we can approximate the binomial distribution with a Poisson distribution of parameter \lim_{N \to \infty } M p (\tau) = x, see [feller1991introduction]. Hence in this case we obtain (8.3).

Lemma

Lemma 9.1

IL8K

Let N \geq 1 and let Let \{ \tau_k \}_{k=1}^N and \{ \varphi_k \}_{k=1}^N be such that \tau_k >0 and \varphi_k >0 . Assume that n(0)=e_1 \in \mathbb R^N and R(0)=0. Then the solution ((n_1, \dots, n_N), R ) of equation (?) is such that

P := \int_0^\infty R(t) dt = \prod_{k =1 }^N \left( \frac{ \varphi_k }{ \varphi_k + \frac{1 }{ \tau_k }} \right).

Lemma

Lemma 9.2

PPGB

Let N \geq 1, \{ b_k \}_{k=1}^N such that b_k >0 for every k \in \{ 1, \dots, N \}. The function G_N is monotone and such that

G_N(0)=0 \text{ and } \lim_{\tau\to\infty}G_N(\tau)=1.

Lemma

Lemma 9.3

U095

Assume that \lim_{N \to \infty } M = \infty and \lim_{N \to \infty } \frac{ \log(M) }{N } = b >0. Assume that \{ b_k \}_{k=1}^N is such that there exists a \mu \in \mathcal M_+ such that

\frac{1}{ N } \sum_{k =1 }^N \delta (x- b_k ) \rightharpoonup \mu (dx) \text{ as } N \to \infty.

Then \lim_{N \to \infty} \tilde{\tau_c } = \overline T \in (0, \infty ) and is the solution of

\int_{0}^\infty \log \left( \frac{x \overline T }{1 + x \overline T } \right) \mu(dx) = - b.

Proof of Lemma 9.3

First of all notice that a solution to (?) exists for every N \geq 1 due to the monotonicity of \tau \mapsto G_N(\tau) and the fact that G_N(0)=0 and G_N(\infty )=1. We now want to prove that \lim_{N \to \infty } \tilde{\tau}_c is finite.

Notice that (?) can be written as

\exp \left( \frac{1}{N} \sum_{k=1}^N \log\left(1- \frac{1 }{1+ b_k \tilde{\tau}_c } \right) \right) =\left( G_N ( \tilde{\tau}_c ) \right)^{\frac{1}{N}}=\left( \frac{1}{M } \right)^{\frac{1}{N}} = \exp \left( - \frac{ \log(M)}{N} \right).

Since we assume that \lim_{N \to \infty } \frac{\log(M) }{N } = b we deduce that

\begin{aligned} -b &= \lim_{N \to \infty } \left( \frac{1}{N} \sum_{k=1}^N \log\left(1- \frac{1 }{1+ b_k \tilde{\tau}_c } \right) \right) = \lim_{N \to \infty } \frac{1}{N} \sum_{k=1}^N \int_0^\infty \log\left(1- \frac{1 }{1+ x \tilde{\tau}_c } \right) \delta (x- b_k ) dx \\ &= \lim_{N \to \infty } \int_0^\infty \log\left(1- \frac{1 }{1+ x \tilde{\tau}_c } \right) \frac{1}{N} \sum_{k=1}^N \delta (x- b_k ) dx= \lim_{N \to \infty } \int_0^\infty \log\left(1- \frac{1 }{1+ x \lim_{N \to \infty} \tilde{\tau}_c } \right) \mu(dx). \end{aligned}

Now notice that since the function

\tau \mapsto \int_0^\infty \log\left(1- \frac{1 }{1+ x \tau } \right) \mu(dx).

is monotonically increasing there exists a unique solution to

- b = \int_0^\infty \log\left(1- \frac{1 }{1+ x \tau } \right) \mu(dx)

which is 0 < \tau = \lim_{N \to \infty } \tilde {\tau}_c < \infty.

Proposition

Proposition 9.4

KPZH

Assume that \lim_{N \to \infty } M = \infty and \lim_{N \to \infty } \frac{ \log(M) }{N } = b >0. Assume that \{ b_k \}_{k=1}^N is as in Lemma 9.3. Then

\lim_{N \to \infty } P_M \left( \tilde{\tau}_c \left( 1 + \frac{\xi }{N D} \right) \right) = 1- \exp ( - \exp (\xi)) \text{ as } N \to \infty,

where \xi \in \mathbb R and where D:= \lim_{N \to \infty } \frac{1}{N} \sum_{k=1}^N \frac{ 1}{1 + \tilde{\tau}_c b_k } = \mu([0, \infty)).

Proof of Proposition 9.4

The proof is similar to the one of Theorem 3.5. Indeed, let \tau = \tilde{\tau}_c \left( 1 + \frac{x }{N } \right), then

\begin{aligned} P_M \left( \tau \right) \sim 1- \left( 1- G_N(\tau )\right)^M \sim 1-\exp \left( M \log \left( 1- G_N(\tau )\right) \right) \sim 1- \exp (- M G_N (\tau ) ) \text{ as } N \to \infty. \end{aligned}

Now notice that

\begin{aligned} M G_N (\tau )&= \prod_{k=1}^N \left( \frac{ 1 + \tilde{\tau}_c b_k }{\tilde{\tau}_c b_k } \right) \prod_{k=1}^N \left( \frac{ \tau b_k }{ 1+ \tau b_k } \right) = \prod_{k=1}^N \left( \frac{ 1 + \tilde{\tau}_c b_k }{\tilde{\tau}_c b_k } \right)\prod_{j=1}^N \frac{ \tilde{\tau}_c b_j \left( 1 + \frac{x }{N }\right) }{ \left( 1+ \tilde{\tau}_c b_j \left( 1 + \frac{x}{N }\right) \right) } \\ &= \left( 1+ \frac{x }{N }\right)^N \prod_{k=1}^N \frac{ \left(1 + \tilde{\tau}_c b_k \right) }{\left( 1 + \tilde{\tau}_c b_k + \frac{x}{N} \tilde{\tau}_c b_k \right) } \\ & \sim e^{x} \exp\left( \log \left( \prod_{k=1}^N \frac{ \left(1 + \tilde{\tau}_c b_k \right) }{\left( 1 + \tilde{\tau}_c b_k + \frac{x}{N} \tilde{\tau}_c b_k \right) } \right) \right) = e^{x} \exp\left( \sum_{k=1}^N \log \left( \frac{ 1 + \tilde{\tau}_c b_k }{1 + \tilde{\tau}_c b_k + \frac{x}{N} \tilde{\tau}_c b_k } \right) \right) \\ & = e^{x } \exp\left( \sum_{k=1}^N \log \left( \frac{ 1 }{1 + \frac{x}{N} \frac{ \tilde{\tau}_c b_k}{1 + \tilde{\tau}_c b_k } } \right) \right) \sim \exp\left(x \left(1- \frac{1 }{N} \sum_{k=1}^N \frac{ \tilde{\tau}_c b_k}{1 + \tilde{\tau}_c b_k } \right) \right) \text{ as } N \to \infty. \end{aligned}

Hence the desired conclusion follows taking x = \frac{\xi }{1 - \frac{1}{N }\sum_{k=1}^N \frac{ \tilde{\tau}_c b_k}{1 + \tilde{\tau}_c b_k } }.

Lemma

Lemma A.1

Z4TA

Let \tau >0. The generating function G of \{ m_k (\infty) \}_{k=0}^{N} is given by

G(z)=\frac{z_0-1 }{z_0-z } \left( 1 + \left( \frac{z}{z_0} \right)^N \tau (1-z) \right), \quad z \in \mathbb C

where

z_0:= 1+ \frac{1}{\tau}.

Used by Lemma A.2, Proof, Proof

Proof of Lemma A.1

The definition of G, implies that

\begin{aligned} G(z)= \left( \frac{ z \tau }{1+\tau } \right)^N + \sum_{k=0}^{N-1} \left(\frac{1}{1+\tau } \right) \left( \frac{\tau z }{1+\tau }\right)^k = a^N + \frac{S}{1+\tau }, \end{aligned}

where

a:=\frac{ z \tau }{1+\tau }

and S:= \sum_{k=0}^{N-1} a^k. Since S= \frac{1-a^N }{1-a }. we have that (Proof) can be rewritten as

\begin{aligned} G(z)= a^N + \frac{1}{1+\tau } \left(\frac{1-a^N} {1-a} \right) = \frac{1}{(1+\tau) (1-a) } \left( 1+ a^N \left( \tau -\tau a -a \right) \right). \end{aligned}

Using the fact that a is given by (Proof) we deduce that

G(z)= \frac{ 1+ a^N \tau (1-z) }{1+\tau(1-z) }.

Equality (A.1) follows by substituting (A.1) and (Proof) in (Proof).

Lemma

Lemma A.2

4ACS

Assume that \lim_{N \to \infty} M = \infty, that \tau is given by (6.4) and that |z-1 | \leq \frac{\log(M)}{ N \sqrt{M} }. Then

\lim_{N \to \infty} \left( 1 + \left( \frac{z}{z_0} \right)^N \tau (1-z) \right)^M =1.

where z_0 is given by (A.1).

Uses Theorem 6.4, Lemma A.1 · Used by Proof

Proof of Lemma A.2

First of all we prove that

\lim_{N \to \infty} \left|M \left( \frac{z}{z_0} \right)^N \tau (1-z)\right| =0.

Notice that by the definition of z_0, (A.1), we have that \frac{1}{z_0} = \frac{\tau}{1+\tau }. Therefore, as in (Proof) in the proof of Theorem 3.5, using that \tau satisfies (6.4), we deduce that

\left(\frac{\tau}{1+\tau } \right)^N \leq e^x\frac{1}{M}.

As a consequence

\begin{aligned} \left| M \left( \frac{z}{z_0} \right)^N \tau (1-z) \right| = \left| M z^N \left(\frac{\tau}{1+\tau } \right)^N \tau (1-z) \right| \leq e^x \left| z^N \tau (1-z) \right| \leq e^x \tau e^{ \frac{\log(M)}{\sqrt{M}}} \frac{\log(M) } { N \sqrt{M} } . \end{aligned}

Since \tau is given by (6.4), the asymptotic behaviour of \tau_c in Proposition 3.4 implies that

\lim_{N \to \infty} \left| M \left( \frac{z}{z_0} \right)^N \tau (1-z) \right| =0.

As a consequence

\begin{aligned} \lim_{N \to \infty} \left( 1 + \left( \frac{z}{z_0} \right)^N \tau (1-z) \right)^M &= \lim_{N \to \infty} \exp \left[ M \log \left( 1 + \left( \frac{z}{z_0}\right)^N \tau (1-z) \right) \right] \\ &= \lim_{N \to \infty } \exp \left[ M \left( \frac{z}{z_0}\right)^N \tau (1-z) \right] = 1 . \end{aligned}

Uses Lemma A.1, Theorem 3.5, Theorem 6.4, Proposition 3.4

Lemma

Lemma B.1

PI7K

Let N \geq 1, M \geq 1 and \tau >0. Let \Psi_M be given by (?). Then the Laplace transform of \psi_M is given by

\hat{ \Psi}_M (z) = \frac{1}{1- \hat{s}(z)} \left( \frac{1}{1+ z + \frac{1}{\tau} } \right)^N \frac{1- (\hat{s }( z ))^M }{1- (\hat{s }( 0 ))^M}, \quad z \in \mathbb C

where \hat{s} is the Laplace transform of (?) and is given by

\hat{s}(z)= \frac{1}{1+\tau z } \left( 1- \left( \frac{1}{1+z + \frac{1}{\tau}}\right)^N \right),\ z \in \mathbb C.

Used by Proof

Proof of Lemma B.1

First of all notice that (?) implies that

\hat{\Psi}_M(z) = \frac{1}{p_M(\tau) } \hat{r}(z) \sum_{j =1}^M \hat{s}^{j-1}(z)

where p_M(\tau ) = 1 - \left( 1- \left( \frac{\tau }{1+\tau }\right)^N \right)^M= 1- (\hat{s}(0))^M and

\hat{s }(z)= \frac{1}{\tau} \sum_{k =1 }^{N} \hat{n}_k(z) = \frac{1}{\tau} \sum_{k =1 }^{N} \left(\frac{1}{1+z+\frac{1}{\tau}} \right)^k= \frac{1}{\tau}\left(\frac{1}{1+z+\frac{1}{\tau}} \right) \sum_{k =0 }^{N-1} \left(\frac{1}{1+z+\frac{1}{\tau}} \right)^k=\frac{ \left( 1+z + \frac{1}{\tau} \right)^N -1 }{(1+ \tau z )\left( 1+ z + \frac{1}{\tau} \right)^N }

and

\hat{r}(z)= \hat{n}_N (z)=\left(\frac{1}{1+z+\frac{1}{\tau}} \right)^N = 1- (1+ \tau z ) \hat{s } (z)

Therefore

\begin{aligned} \hat{\Psi}_M(z) &= \frac{1}{1- (\hat{s} (0))^M } ( 1- (1+ \tau z ) \hat{s } (z) ) \sum_{j =1}^M \hat{s}^{j-1}(z) = \frac{1- (\hat{s}(z))^M - \tau z \sum_{j =1}^M \hat{s}^{j}(z)}{1- (\hat{s} (0))^M } \\ &= \frac{1}{1- (\hat{s} (0))^M }\left( 1- (\hat{s}(z))^M - \tau z \hat{s}(z) \frac{1- (\hat{s}(z))^M}{ 1 - \hat{s}(z)} \right) = \frac{1- (\hat{s}(z))^M}{1- (\hat{s} (0))^M }\left( 1 - \frac{\tau z \hat{s}(z)}{ 1 - \hat{s}(z)} \right) \\ &= \frac{1- (\hat{s}(z))^M}{1- (\hat{s} (0))^M } \frac{ 1- (1+\tau z ) \hat{s}(z)}{ 1 - \hat{s}(z)} = \frac{1}{1- \hat{s}(z)} \left( \frac{1}{1+ z + \frac{1}{\tau} } \right)^N \frac{1- (\hat{s }( z ))^M }{1- (\hat{s }( 0 ))^M} \end{aligned}

where in the last equality we have used (Proof).

Source and licence

The text on this page is from arXiv:2405.10580v1 by Eugenia Franco and Juan J. L. Velázquez, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2405/2405.10580.jsonl.