1. Trang chủ
  2. » Giáo án - Bài giảng

a note on the strong law of large numbers for markov chains indexed by irregular trees

8 4 0

Đang tải... (xem toàn văn)

THÔNG TIN TÀI LIỆU

Thông tin cơ bản

Định dạng
Số trang 8
Dung lượng 200,25 KB

Các công cụ chuyển đổi và chỉnh sửa cho tài liệu này

Nội dung

The strong law of large numbers and the Shannon-McMillan theorem for Markov chains indexed by an infinite irregular tree are established.. Yang and Liu [] studied the strong law of large

Trang 1

R E S E A R C H Open Access

A note on the strong law of large numbers for Markov chains indexed by irregular trees

Wei-cai Peng*

* Correspondence:

weicaipeng@126.com

Department of Mathematics,

Chaohu University, Chaohu, 238000,

P.R China

Abstract

In this paper, a kind of an infinite irregular tree is introduced The strong law of large numbers and the Shannon-McMillan theorem for Markov chains indexed by an infinite irregular tree are established The outcomes generalize some known results

on regular trees and uniformly bounded degree trees

Keywords: Markov chain; tree; strong law of large numbers; AEP

1 Introduction

By a tree T we mean an infinite, locally finite, connected graph with a distinguished vertex

o called the root and without loops or cycles We only consider trees without leaves That

is, the degree (the number of neighboring vertices) of each vertex (except o) is required to

be at least 

Let T be an infinite tree with root o, the set of all vertices with distance n from the root

is called the nth generation (or nth level) of T We denote by T (n)the union of the first

n generations of T , by T (m) (n) the union from the mth to nth generations of T , by L n the

subgraph of T containing the vertices in the nth generation For each vertex t, there is a unique path from o to t, and |t| for the number of edges on this path We denote the first predecessor of t by  t , the second predecessor of t by  t , and by n t the nth predecessor

of t We also call t one of  t ’s sons For any two vertices s and t, denote by s ≤ t, if s is on the unique path from the root o to t, denote by s ∧ t the vertex farthest from o satisfying

s ∧ t ≤ s and s ∧ t ≤ t X A={X t , t ∈ A} and denote by |A| the number of vertices of A.

If each vertex on a tree T has m +  neighboring vertices, we call it a Bethe tree T B,m;

if the root has m neighbors and the other vertices have m +  neighbors on a tree T , we call it a Cayley tree T C,m Both the Bethe tree and the Cayley tree are called regular (or

homogeneous) trees If the degrees of all vertices on a tree T are uniformly bounded, then

we call T a uniformly bounded degree tree (see [] and []).

Definition  (see []) Let T be a locally finite, infinite tree, S be a finite state-space, {X t , t

T } be a collection of S-valued random variables defined on the probability space (, F, P).

Let



© 2014 Peng; licensee Springer This is an Open Access article distributed under the terms of the Creative Commons Attribu-tion License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribuAttribu-tion, and reproducAttribu-tion in any

Trang 2

be a distribution on S, and



be a stochastic matrix on S If for any vertex t,

P(X t = y|Xt = x and X s for t ∧ s ≤  t)

=P(X t = y|Xt = x) = P(y|x) ∀x, y ∈ S () and

P(X= x) = p(x) ∀x ∈ S,

then{X t , t ∈ T} will be called S-valued Markov chains indexed by an infinite tree T with

initial distribution () and transition matrix (), or called T -indexed Markov chains with

state-space S.

Benjamini and Peres [] gave the notion of tree-indexed Markov chains and studied the recurrence and ray-recurrence for them Berger and Ye [] studied the existence of

en-tropy rate for some stationary random fields on a homogeneous tree Ye and Berger [], by

using Pemantle’s result [] and a combinatorial approach, studied the Shannon-McMillan

theorem with convergence in probability for a PPG-invariant and ergodic random field

on a homogeneous tree Yang and Liu [] studied the strong law of large numbers and

Shannon-McMillan theorems for Markov chains field on the Cayley tree Yang [] studied

some strong limit theorems for homogeneous Markov chains indexed by a homogeneous

tree and the strong law of large numbers and the asymptotic equipartition property (AEP)

for finite homogeneous Markov chains indexed by a homogeneous tree Yang and Ye []

studied strong theorems for countable nonhomogeneous Markov chains indexed by a

ho-mogeneous tree and the strong law of large numbers and the AEP for finite

nonhomoge-neous Markov chains indexed by a homogenonhomoge-neous tree Bao and Ye [] studied the strong

law of large numbers and asymptotic equipartition property for nonsymmetric Markov

chain fields on Cayley trees Takacs [] studied the strong law of large numbers for the

univariate functions of finite Markov chains indexed by an infinite tree with uniformly

bounded degree Huang and Yang [] studied the strong law of large numbers for Markov

chains indexed by uniformly bounded degree trees

However, the degrees of the vertices in the tree models are uniformly bounded What if the degrees of the vertices are not uniformly bounded? In this paper, we drop the uniformly

bounded restriction We mainly study the strong law of large numbers and AEP with a.e

convergence for finite Markov chains indexed by trees under the following assumption

For any integer N ≥ , let d(t) :=  and denote

d N (t) := |σ ∈ T : N σ = t| ()

by the amount of t’s N th descendants Denote

O(n) =



c n:  < lim sup

n→∞

c n

n ≤ c, c is a constant



Trang 3

We assume that for enough large n and any given integer N≥ ,

max

d N (t) : t ∈ T (n)

≤ O



ln|T (n+N)|

|T (n)|

The following examples are used to explain assumption ()

Example  Both the Bethe tree T B,m and the Cayley tree T C,msatisfy assumption ()

Ac-tually, max{d N (t) : t ∈ T (n–N) } is a constant m N, and ln(|T (n) |/|T (n–N) |) = N ln m.

Example  A uniformly bounded degree tree satisfies assumption () In fact, if the tree T

is a uniformly bounded degree tree, then max{d N (t) : t ∈ T (n–N)} is no more than a constant

a N, and

ln |T (n)|

|T (n–N)|≤ ln

|T (n–N) | × a N

is also a constant

Example  Define the lower growth rate of the tree to be gr T = lim inf n→∞|T (n)|

n and

the upper growth rate of the tree to be Gr T = lim sup n→∞|T (n)|

n

If both the gr T and Gr T are finite, then

ln|T (n+N)|

|T (n)| ≤ ln

(Gr T) n+N

(gr T) n = n lnGrT

grT + N ln Gr T,

hence () implies that

max

d N (t) : t ∈ T (n)

≤ O



ln|T (n+N)|

|T (n)|

≤ O(n).

2 Some notations and lemmas

In the following, letδ k(·) be a Kronecker δ-function For any given integer N ≥ , denote

S N k

T (n) :=

t ∈T (n–N)

which can be considered as the number of k’s among the variables in T (n–N), weighted

according to the number of N th descendants By (), we have

k ∈S

S n N (k) = T (n–N) – . () Define

H n(ω) =

t ∈T (n) \{o}

and

G n(ω) =

t ∈T (n) \{o}

E

Trang 4

Lemma  (see []) Let T be an infinite tree with assumption () holds Let (X t)t ∈T be a

T -indexed Markov chain with state-space S defined as before, {g t (x, y), t ∈ T} be functions

defined on S Let L o={o}, F n=σ (X T (n)

),

t n(λ, ω) = e

λt ∈T(n)\{o} g t (X t ,X t)



t ∈T (n) \{o} E[e λg t (X t ,X t)|Xt], ()

where λ is a real number Then {t n(λ, ω), F n , n ≥ } is a nonnegative martingale.

Lemma  (see []) Under the assumption of Lemma , let {a n , n ≥ } be a sequence of

nonnegative random variables, α >  Set

B = lim

and

D(α) =

 lim sup

n→∞

a n

t ∈T (n) \{o}

E

g t(Xt , X t )e α|g t (X t ,X t)||Xt = M( ω) < ∞



∩ B. ()

Then

lim

n→∞

H n(ω) – G n(ω)

3 Strong law of large numbers and Shannon-McMillan theorem

In this section, we study the strong law of large numbers and the Shannon-McMillan

the-orem for finite Markov chains indexed by an infinite tree with assumption () holds

Theorem  Let T be an infinite tree with assumption () holds Then under the assumption

of Lemma , for all k ∈ S and N ≥ , we have

lim

n→∞

|T (n+N)|



S N k

T (n) –

l ∈S

S N+ l 

T (n–)

P(k |l)



=  a.e. ()

Proof Let g t (x, y) = d N (t) δ k (y), a n=|T (n+N)| Since

G n(ω) =

t ∈T (n) \{o}

E

g t (Xt , X t)|Xt

=

t ∈T (n) \{o}

d N (t)

x t ∈S

δ k (x t )P(x t |Xt)

=

t ∈T (n) \{o}

d N (t)P(k |Xt)

=

l ∈S t ∈T (n) \{o}

δ l (Xt )d N (t)P(k |l)

=

l ∈S t ∈T (n–)

δ l (X t )d N+ (t)P(k|l)

=

l ∈S

S N+

l



T (n–)

Trang 5

H n(ω) =

t ∈T (n) \{o}

g t (Xt , X t) =

t ∈T (n) \{o}

d N (t) δ k (X t ) = S N k

T (n) –δ k (X o )d N (o). ()

By Lemma , we know that{t n(λ, ω), F n , n≥ } is a nonnegative martingale According

to the Doob martingale convergence theorem, we have

lim

We have by ()

lim sup

n→∞

lnt n(λ, ω)

By (), () and (), we get

lim sup

n→∞

|T (n+N)|



λH n(ω) –

t ∈T (n) \{o}

ln

E

e λg(X t ,X t)|Xt ≤  a.e ω ∈ B. ()

Letλ >  Dividing two sides of () by λ, we have

lim sup

n→∞

a n



H n(ω) –

t ∈T (n) \{o}

ln[E[eλg(X t ,X t)|Xt]]

λ



≤  a.e ω ∈ B. ()

The case{d N (t) : t ∈ T (n)} is uniformly bounded was considered in [], we only consider

the case{d N (t) : t ∈ T (n) } is not uniformly bounded By () and inequalities ln x ≤ x – 

(x > ),  ≤ e x –  – x≤ –xe |x|, as  <λ ≤ α, we have

lim sup

n→∞

|T (n+N)|



H n(ω) –

t ∈T (n) \{o}

E

g t (Xt , X t)|Xt

≤ lim sup

n→∞

|T (n+N)|

t ∈T (n)

 ln[E[eλg t (X t ,X t)|Xt]]

g t (Xt , X t)|Xt

≤ lim sup

n→∞

|T (n+N)|

t ∈T (n)



E[e λg t (X t ,X t)|Xt] – 

g t (Xt , X t)|Xt

λ

lim supn→∞

|T (n+N)|

t ∈T (n)

E

g t(Xt , X t )e λ|g t (X t ,X t)||Xt

=λ

lim supn→∞

|T (n+N)|

t ∈T (n)

()

E 

d N (t) δ k (X t)

e λ|d N (t) δ k (X t)||Xt

λ

lim supn→∞

|T (n+N)|

t ∈T (n)

()



d N (t)

e λd N (t) P(k |Xt)

λ

lim supn→∞

|T (n+N)|

t ∈T (n)



d N (t)

e λd N (t)

Trang 6

lim supn→∞

|T (n+N)|

t ∈T (n)

()

eλd N (t) 

for enough large d N (t)

λ

lim supn→∞

|T (n)| – 

|T (n+N)| max



eλd N (t) , t ∈ T (n)

()



λ

lim supn→∞

|T (n)| – 

|T (n+N)|



emax{dN (t),t∈T

(n)

() }λ

By (), there exists a constantβ >  such that

max

d N (t), t ∈ T (n)

()



≤ β ln |T |T (n+N) (n)||, hence,



emax{dN (t),t∈T

(n)

() }λ

<



|T (n+N)|

|T (n)|

λβ

Set  <λ <

β, by () and () we have

lim sup

n→∞

H n(ω) – G n(ω)

λ

lim supn→∞

|T (n)| – 

|T (n+N)| ×

|T (n+N)|

|T (n)|

λβ

() Letλ → +in (), by () and () we have

lim

n→∞

|T (n+N)|



S N k

T (n) –

l ∈S

S N+ l 

T (n–)

P(k |l)



≤  a.e ()

Let –β ≤ λ → – By (), we similarly get

lim

n→∞

|T (n+N)|



S N k

T (n) –

l ∈S

S N+ l 

T (n–)

P(k |l)



≥  a.e ()

Combining () and (), we obtain () directly 

Let T be a tree, (X t)t ∈T be a stochastic process indexed by the tree T with state-space S.

Denote

P

x T (n)

= P

X T (n) = x T (n)

Let

f n(ω) = – |T(n)|lnP

X T (n)

f n(ω) will be called the entropy density of X T (n) If (X t)t ∈T is a T -indexed Markov chain

with state-space S defined by Definition , we have by ()

f n(ω) = – |T(n)|



lnP(X) +

t ∈T (n) \{o}

lnP(Xt , X t)



Trang 7

The convergence of f n(ω) to a constant in a sense (Lconvergence, convergence in prob-ability, a.e convergence) is called the Shannon-McMillan theorem or the entropy theorem

or the AEP in information theory

Theorem  Let T be an infinite tree with assumption () holds Let k ∈ S, and P be an

ergodic stochastic matrix Denote the unique stationary distribution of P by π Let (X t)t ∈T

be a T -indexed Markov chain with state-space S generated by P Then, for given integer

N≥ ,

lim

n→∞

S N

k (T (n))

Let S l,k (T (n)) :=|{t ∈ T (n) : (Xt , X t ) = (l, k) }|, then

lim

n→∞

S l,k (T (n))

Let f n(ω) be defined as (), then

lim

n→∞f n(ω) = –

l ∈S k ∈S

Proof The proofs of () and () are similar to those of Huang and Yang ([], Theorem 

and Corollary ) Letting g t (x, y) = – ln P(y |x) in Lemma , then

lim

n→∞f n(ω) = lim

n→∞

H n(ω)

|T (n)| = – limn→∞

|T (n)|

t ∈T (n) \{o}

lnP(Xt , X t)

= – lim

n→∞

|T (n)|

t ∈T (n) l ∈S k ∈S

δ l (Xt)δ k (X t ) ln P(k |l)

= – lim

n→∞

l ∈S k ∈S

lnP(k |l) S l,k (T (n))

|T (n)|

Competing interests

The author declares that they have no competing interests.

Acknowledgements

This work is supported by the Foundation of Anhui Educational Committee (No KJ2014A174).

Received: 9 February 2014 Accepted: 4 June 2014 Published: 23 June 2014

References

1 Takacs, C: Strong law of large numbers for branching Markov chains Markov Process Relat Fields 8, 107-116 (2001)

2 Huang, HL, Yang, WG: Strong law of large numbers for Markov chains indexed by an infinite tree with uniformly

bounded degree Sci China Ser A 51(2), 195-202 (2008)

3 Benjamini, I, Peres, Y: Markov chains indexed by trees Ann Probab 22, 219-243 (1994)

4 Berger, T, Ye, Z: Entropic aspects of random fields on trees IEEE Trans Inf Theory 36, 1006-1018 (1990)

5 Ye, Z, Berger, T: Ergodic, regular and asymptotic equipartition property of random fields on trees J Comb Inf Syst.

Sci 21, 157-184 (1996)

Trang 8

7 Yang, WG, Liu, W: Strong law of large numbers and Shannon-McMillan theorem for Markov chains field on Cayley

tree Acta Math Sci Ser B 21(4), 495-502 (2001)

8 Yang, WG: Some limit properties for Markov chains indexed by a homogeneous tree Stat Probab Lett 65, 241-250

(2003)

9 Yang, WG, Ye, Z: The asymptotic equipartition property for nonhomogeneous Markov chains indexed by a

homogeneous tree IEEE Trans Inf Theory 53(9), 3275-3280 (2007)

10 Bao, ZH, Ye, Z: Strong law of large numbers and asymptotic equipartition property for nonsymmetric Markov chain

fields on Cayley trees Acta Math Sci Ser B 27(4), 829-837 (2007)

doi:10.1186/1029-242X-2014-244

Cite this article as: Peng: A note on the strong law of large numbers for Markov chains indexed by irregular trees.

Journal of Inequalities and Applications 2014 2014:244.

...

Cite this article as: Peng: A note on the strong law of large numbers for Markov chains indexed by irregular trees.

Journal of Inequalities and Applications... Shannon-McMillan theorem

In this section, we study the strong law of large numbers and the Shannon-McMillan

the- orem for finite Markov chains indexed by an infinite tree with assumption... (2001)

2 Huang, HL, Yang, WG: Strong law of large numbers for Markov chains indexed by an infinite tree with uniformly

bounded degree Sci China Ser A 51(2), 195-202

Ngày đăng: 02/11/2022, 08:46

TÀI LIỆU CÙNG NGƯỜI DÙNG

TÀI LIỆU LIÊN QUAN

🧩 Sản phẩm bạn có thể quan tâm