Open Access

The strong deviation theorem for discrete-time and continuous-state nonhomogeneous Markov chains

Journal of Inequalities and Applications20132013:462

https://doi.org/10.1186/1029-242X-2013-462

Received: 22 May 2013

Accepted: 9 September 2013

Published: 7 November 2013

Abstract

In this paper, the notion of asymptotic average log-likelihood ratio, as a measure of the difference between the sequence of random variables and Markov chains, is introduced, and by constructing a nonnegative martingale, the strong deviation theorem for discrete-time and continuous-state nonhomogeneous Markov chains is established.

Keywords

discrete-time and continuous-state nonhomogeneous Markov chain asymptotic average log-likelihood ratio transition probability density strong deviation theorem

1 Introduction

Let ( Ω , F , P ) be the probability space, and let { X n , n 0 } be a sequence of continuous random variables taking values in R and with the joint density function f n ( X 0 , X 1 X n ) , n = 1 , 2 ,  . Let Q be another probability measure on ( Ω , F ) , and { X n , n 0 } be an independent random sequence on the measure Q, with the joint density function g n ( X 0 , X 1 X n ) , n = 1 , 2 ,  .

Let
r n ( ω ) = g n ( X 0 , X 1 X n ) f n ( X 0 , X 1 X n ) , r ( ω ) = lim inf n 1 n ln r n ( ω ) ( ln 0 = ) ,
(1)

where ω is a sample point. In statistical terms, r n ( ω ) and r ( ω ) are called the likelihood ratio and the asymptotic average log-likelihood ratio, respectively [1]. Obviously, if f n ( x 0 , x 1 x n ) = g n ( x 0 , x 1 x n ) , n 1 , then r n ( ω ) 0 , a.s. So r ( ω ) can be used as a measure of deviation between f n ( x 0 , x 1 x n ) and g n ( x 0 , x 1 x n ) when n tends to infinity. The smaller r ( ω ) is, the smaller the deviation is.

Definition 1 [2]

Let { X n , n 0 } be a nonhomogeneous Markov chain with the initial distribution u ( x ) , x R , and the transition probability density p n = p n ( x , y ) , x , y R , n 1 . If
P ( X 0 B ) = B u ( x ) d x , P ( X n + 1 B | X n = x ) = B p n ( x , y ) d y ,

this Markov chain is called a discrete-time and continuous-state nonhomogeneous Markov chain.

Let { X n , n 0 } be a discrete-time and continuous-state nonhomogeneous Markov chain on the measure Q with the initial distribution density u ( x ) , x R and the transition probability density p n = p n ( x , y ) , x , y R , n 1 . Then for any B o r e l , set B
Q ( X 0 B ) = B u ( x ) d x , Q ( X n + 1 B | X n = x ) = B p n ( x , y ) d y ,
then
g n ( x 0 , x 1 , , x n ) = u ( x 0 ) p 0 ( x 0 , x 1 ) p n 1 ( x n 1 , x n ) = u ( x 0 ) k = 1 n p k 1 ( x k 1 , x k ) ,
so
r n ( ω ) = u ( X 0 ) k = 1 n p k 1 ( X k 1 , X k ) f n ( X 0 , X 1 X n ) .
(2)

There have been some works on deviation theorem, a kind of strong limit theorem represented by inequalities. Liu and Yang [3] have studied the limit properties of a class of averages of functions of two variables of arbitrary information sources. Liu and Yang [4] investigated the strong deviation theorems for arbitrary information source relative to Markov information source. Liu [5] discussed a class of strong deviation theorems for an arbitrary stochastic sequence with respect to the marginal distribution by using generating function method, and also studied the problem above by means of Laplace transform [6]. Liu and Wang [7] have studied a strong limit theorem expressed by inequalities for the sequences of absolutely continuous random variables. Recently, Fan [8] has studied some strong deviation theorems for dependent continuous random sequence.

In this paper, by using the notion of asymptotic log-likehood and the martingale convergence theorem, and extending the analytic technique proposed by Liu [9], Liu and Yang [7] to the case of discrete-time and continuous-state nonhomogeneous Markov chains, we obtain the strong deviation theorem for discrete-time and continuous-state nonhomogeneous Markov chains.

2 Main result

Theorem 1 Let { X n , n 0 } be a discrete-time and continuous-state nonhomogeneous Markov chain on the measure Q, r ( ω ) and r n ( ω ) be defined by (1) and (2), respectively. Let { B n , n 1 } be a sequence of B o r e l set of the real line, and I B n be the indicative function of B n . Let
a ( ω ) = lim sup n 1 n k = 1 n B k p k 1 ( X k 1 , x k ) d x k b , ω Ω ,
(3)
and
D 1 = { ω : r ( ω ) b } , D 2 = { ω : r ( ω ) b } .
Then
(a) lim sup n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] 2 b r ( ω ) + r ( ω ) a.s. ;
(4)
(b) lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] 2 b r ( ω ) a.s. on  D 1 ,
(5)
and
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] b r ( ω ) a.s. on  D 2 .
(6)
Proof Let λ be a nonnegative constant, and let
h k ( x k 1 , x k ) = { λ p k 1 ( x k 1 , x k ) 1 + ( λ 1 ) B k p k 1 ( x k 1 , x k ) d x k , x k B k ; p k 1 ( x k 1 , x k ) 1 + ( λ 1 ) B k p k 1 ( x k 1 , x k ) d x k , x k B k .
(7)
It is easy to see that u ( x 0 ) k = 1 n h ( x k 1 , x k ) is a density function of n + 1 variables. Let
t n ( λ , ω ) = u ( X 0 ) k = 1 n h k ( X k 1 , X k ) f n ( X 0 , X 1 X n ) ,
(8)
then t n ( λ , ω ) is a nonnegative supermartingale that converges a.s. Hence there exists A ( λ ) F , P ( A ( λ ) ) = 1 such that
lim sup n 1 n ln t n ( λ , ω ) 0 , ω A ( λ ) .
(9)
Letting λ = 1 in (9), we obtain
lim sup n 1 n ln r n ( ω ) 0 , ω A ( 1 ) .
(10)
This implies that
r ( ω ) 0 , ω A ( 1 ) .
(11)
We have by (7)
k = 1 n h k ( X k 1 , X k ) = k = 1 n λ I B k ( X k ) p k 1 ( X k 1 , X k ) 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k = λ k = 1 n I B k ( X k ) k = 1 n p k 1 ( X k 1 , X k ) 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k .
(12)
It follows from (2), (8), and (12) that
ln t n ( λ , ω ) = k = 1 n I B k ( X k ) ln λ k = 1 n ln [ 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ] + ln r n ( ω ) .
(13)
By (9) and (13), we have
lim sup n 1 n ( k = 1 n I B k ( X k ) ln λ + ln r n ( ω ) k = 1 n ln [ 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ] ) 0 , ω A ( λ ) .
(14)
(a) Let λ > 1 . Dividing the two sides of (14) by lnλ, we obtain
lim sup n 1 n ( k = 1 n I B k ( X k ) + ln r n ( ω ) ln λ k = 1 n ln [ 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ] ln λ ) 0 , ω A ( λ ) .
(15)
By (1) and (15), we have
lim sup n 1 n ( k = 1 n I B k ( X k ) k = 1 n ln [ 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ] ln λ ) r ( ω ) ln λ , ω A ( λ ) .
(16)
By (3), (16), the property of the superior limit
lim sup n ( a n b n ) d lim sup n ( a n c n ) lim sup n ( b n c n ) + d ,
and the inequality 0 ln ( 1 + x ) x ( x 0 ), we have
lim sup n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] lim sup n 1 n k = 1 n [ ln [ 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ] ln λ B k p k 1 ( X k 1 , x k ) d x k ] + r ( ω ) ln λ lim sup n 1 n k = 1 n [ ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ln λ B k p k 1 ( X k 1 , x k ) d x k ] + r ( ω ) ln λ b ( λ 1 ln λ 1 ) + r ( ω ) ln λ , ω A ( λ ) .
(17)
By using the inequality 1 1 λ < ln λ ( λ > 1 ), we have by (17)
lim sup n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] b ( λ 1 ) + λ r ( ω ) λ 1 , ω A ( λ ) .
(18)
Let Q be the set of rational numbers in the interval ( 1 , + ) , and let
A = λ Q A ( λ ) , g ( λ , r ) = b ( λ 1 ) + λ r λ 1 .
(19)
Then we have by (18),
lim sup n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] g ( λ , r ( ω ) ) , ω A , λ Q .
(20)
Let b > 0 . It is easy to see if r > 0 , g ( λ , r ) as a function of λ attains its smallest value g ( 1 + r b , r ) = 2 b r + r on the interval ( 1 , + ) , and g ( λ , 0 ) is increasing on the interval ( 1 , + ) and lim λ 1 + 0 g ( λ , 0 ) = 0 . For each ω A A ( 1 ) if r ( ω ) , take λ n ( ω ) Q , n = 1 , 2 , such that λ n ( ω ) 1 + r ( ω ) b , we have
lim n g ( λ n ( ω ) , r ( ω ) ) = 2 b r ( ω ) + r ( ω ) .
(21)
By (20), we have
lim sup n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] g ( λ n ( ω ) , r ( ω ) ) , n = 1 , 2 , .
(22)
By (21) and (22), we have
lim sup n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] 2 b r ( ω ) + r ( ω ) , ω A A ( 1 ) .
(23)

If r ( ω ) = , (23) holds obviously. Since P ( A A ( 1 ) ) = 1 , (4) holds by (23) when b > 0 .

When b = 0 , letting λ = e in (20), we have
lim sup n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] r ( ω ) , ω A ( e ) .
(24)

Since P ( A ( e ) ) = 1 , (4) also holds by (24) when b = 0 .

(b) Let 0 < λ < 1 . Dividing the two sides of (14) by lnλ, we have
lim inf n 1 n ( k = 1 n I B k ( X k ) k = 1 n ln [ 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ] ln λ + ln r n ( ω ) ln λ ) 0 , ω A ( λ ) .
(25)
By (1) and (25), we have
lim inf n 1 n ( k = 1 n I B k ( X k ) k = 1 n ln [ 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ] ln λ ) r ( ω ) ln λ , ω A ( λ ) .
(26)
By (26), (3), the property of the inferior limit
lim inf n ( a n b n ) d lim inf n ( a n c n ) lim inf n ( b n c n ) + d ,
and the inequality ln ( 1 + x ) x ( 1 < x 0 ), we have
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] lim inf n 1 n k = 1 n [ ln [ 1 + ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ] ln λ B k p k 1 ( X k 1 , x k ) d x k ] + r ( ω ) ln λ lim inf n 1 n k = 1 n [ ( λ 1 ) B k p k 1 ( X k 1 , x k ) d x k ln λ B k p k 1 ( X k 1 , x k ) d x k ] + r ( ω ) ln λ b ( λ 1 ln λ 1 ) + r ( ω ) ln λ , ω A ( λ ) .
(27)
By using the inequality 1 1 λ < ln λ < 0 and ln λ < λ 1 < 0 ( 0 < λ < 1 ), we have by (27)
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] b ( λ 1 ) + r ( ω ) λ 1 , ω A ( λ ) A ( 1 ) .
(28)
Let Q be the set of rational numbers in the interval ( 0 , 1 ) , and let
A = λ Q A ( λ ) , h ( λ , r ) = b ( λ 1 ) + r λ 1 .
(29)
Then we have by (28)
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] h ( λ , r ( ω ) ) , ω A A ( 1 ) , λ Q .
(30)
Let b > 0 . It is easy to see that if 0 < r < b , then h ( λ , r ) as a function of λ attains its largest value h ( 1 r b , r ) = 2 b r on the interval ( 0 , 1 ) , and h ( λ , 0 ) is increasing on the interval ( 0 , 1 ) and lim λ 1 0 h ( λ , 0 ) = 0 , and h ( λ , b ) = b ( λ 1 + 1 λ 1 ) is decreasing on the interval ( 0 , 1 ) and lim λ 0 + h ( λ , b ) = 2 b . For each ω A A ( 1 ) D 1 , take τ n ( ω ) Q , n = 1 , 2 , such that τ n ( ω ) 1 r ( ω ) b . Then we have
lim n h ( τ n ( ω ) , r ( ω ) ) = 2 b r ( ω ) .
(31)
By (30), we have
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] h ( τ n ( ω ) , r ( ω ) ) , n = 1 , 2 , .
(32)
By (31) and (32),
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] 2 b r ( ω ) , ω A A ( 1 ) D 1 .
(33)

Since P ( A A ( 1 ) ) = 1 , (5) holds by (33) when b > 0 .

When b = 0 , r ( ω ) = 0 for ω D 1 A ( 1 ) , hence we have by (30)
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] 0 , ω A ( λ ) A ( 1 ) D 1 ,
(34)

since P ( A ( λ ) A ( 1 ) ) = 1 , (5) also holds by (34) when b = 0 .

It is easy to see that when 0 b < r , h ( λ , r ) as a function of λ is decreasing on the interval ( 0 , 1 ) and lim λ 0 + h ( λ , r ) = ( r + b ) . For each ω A A ( 1 ) D 2 , when r ( ω ) , take λ n ( ω ) Q , n = 1 , 2 ,  , such that λ n ( ω ) 0 . We have
lim n h ( λ n ( ω ) , r ( ω ) ) = r ( ω ) b .
(35)
By (30), we have
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] h ( λ n ( ω ) , r ( ω ) ) , n = 1 , 2 , .
(36)
It follows from (35) and (36) that
lim inf n 1 n k = 1 n [ I B k ( X k ) B k p k 1 ( X k 1 , x k ) d x k ] r ( ω ) b , ω A A ( 1 ) D 2
(37)

when r ( ω ) = , (37) also holds obviously. Since P ( A A ( 1 ) ) = 1 , (6) follows from (37) directly. □

Declarations

Acknowledgements

This work is supported by the National Natural Science Foundations of China (11071104, 11226210), and the Research Foundation for Advanced Talents of Jiangsu University (11JDG116).

Authors’ Affiliations

(1)
Faculty of Science, Jiangsu University

References

  1. Liu W: Strong Deviation and Analytic Method. Science Press, Beijing; 2003.Google Scholar
  2. Wang B: Strong law of large numbers for discrete-time and continuous-state nonhomogeneous Markov chains. J. Jiangsu Univ. Nat. Sci. 2008, 29(1):86–88.MATHGoogle Scholar
  3. Liu W, Yang WG: The limit properties of a class of averages of functions of two variables of arbitrary information sources. Chinese J. Appl. Probab. Statist. 1995, 11(2):195–203.MathSciNetMATHGoogle Scholar
  4. Liu W, Yang WG: A comparison between arbitrary information sources and nonhomogeneous Markov information sources and some small deviations theorems. Acta Math. Sin. (Chin. Ser.) 1997, 40(1):22–36.MATHGoogle Scholar
  5. Liu W, Yang WG: The Markov approximation of the sequences of N -valued random variables and a class of small deviation theorems. Stoch. Process. Appl. 2000, 89: 117–130. 10.1016/S0304-4149(00)00016-8View ArticleMathSciNetMATHGoogle Scholar
  6. Liu W: A class of strong deviation theorems and Laplace transform methods. Chin. Sci. Bull. 1998, 43(10):1036–1041.MathSciNetGoogle Scholar
  7. Liu W, Wang YJ: A strong limit theorem expressed by inequalities for the sequences of absolutely continuous random variables. Hiroshima Math. J. 2002, 32: 379–387.MathSciNetMATHGoogle Scholar
  8. Fan AH: Some strong deviation theorems for dependent continuous random sequence. Chin. Q. J. Math. 2010, 25(4):572–577.MATHGoogle Scholar
  9. Liu W: Relative entropy densities and a class of limit theorems of the sequences of m -valued random variables. Ann. Probab. 1990, 18: 829–839. 10.1214/aop/1176990860MathSciNetView ArticleMATHGoogle Scholar

Copyright

© Wang and Shi; licensee Springer. 2013

This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.