 Research
 Open Access
 Published:
Limit properties of exceedance point processes of strongly dependent normal sequences
Journal of Inequalities and Applications volume 2015, Article number: 63 (2015)
Abstract
In this paper, we define an in plane Cox process and prove the timenormalized point process of exceedances by a dependent normal sequence converging to the Cox process in distribution under some mild conditions. As some applications of the convergence result, two important joint asymptotic distributions for the order statistics are derived.
Introduction
Let \(\{\xi_{i}, i\geq1\}\) be a standardized normal sequence with correlation coefficients \(r_{ij}=\operatorname{Cov}(\xi_{i},\xi_{j})\) and \(M_{n}^{(k)}\) be the kth largest maxima of \(\{\xi_{i},1\leq i\leq n\}\). A conventional assumption is that \(r_{ij}\rightarrow0\) as \(ji\rightarrow+\infty\) at different rates according to which dependent normal sequences are classified into two different types: ‘weakly dependent’ and ‘strongly dependent’, respectively. Leadbetter et al. [1] considered the case: \(r_{ij}=r_{ji}\) and \(r_{n}\log n\rightarrow0\) as \(n\rightarrow+\infty\), i.e. \(\{\xi_{i}, i\geq1\}\) is a weakly dependent stationary normal sequence. By using asymptotic independence, they focused on \(M_{n}^{(k)}\) and its location (which is written as \(L_{n}^{(k)}\)) and obtained the asymptotic behavior of the probabilities \(P(a_{n}(M_{n}^{(2)}b_{n})\leq x, L_{n}^{(2)}/n\leq t)\), and \(P(a_{n}(M_{n}^{(1)}b_{n})\leq x_{1}, a_{n}(M_{n}^{(2)}b_{n})\leq x_{2})\); here and in the sequel the standardized constants \(a_{n}\) and \(b_{n}\) are defined as
Mittal and Ylvisaker [2] showed that if \(r_{ij}=r_{ji}\) and \(r_{n}\log n\rightarrow\gamma>0\) as \(n\rightarrow+\infty\) (the strongly dependent stationary case) then \(a_{n}(M_{n}^{(1)}b_{n})\) tends in distribution to a convolution of \(\exp(e^{x})\) and a normal distribution function, and further if \(r_{n}\log n\rightarrow\infty\) then by a different normalization, the limiting distribution is normal. Recently, several important results for extremes of dependent normal sequences were established. Ho and Hsing [3] and Tan and Peng [4] investigated the joint asymptotic distributions of the maximum of \(\{\xi_{i}, 1\leq i\leq n\}\) and \(\sum^{n}_{i=1}\xi_{i}\) for dependent Gaussian sequences. Hashorva et al. [5] considered the joint limit distributions of maxima of complete and incomplete samples, respectively, i.e. the Piterbarg theorem under some conditions on convergence rate of the correlations. Leadbetter et al. [1] developed an important tool: the weak convergence of exceedance point processes which is crucial to study the joint asymptotic distributions of some extremes. Many authors further studied the asymptotic behavior of exceedance point processes under different conditions. We refer to Piterbarg [6], Hu et al. [7], Falk et al. [8], Peng et al. [9] and Hashorva et al. [10] for point processes of exceedances by weakly dependent stationary sequences including Gaussian ones and Wiśniewski [11], Lin et al. [12] for point processes of exceedances by strongly dependent Gaussian vector sequences.
Throughout this paper, let \(\{\xi_{i}, i\geq1\}\) be a standardized strongly dependent stationary normal sequence with correlation coefficients \(r_{ij}=\operatorname{Cov}(\xi_{i},\xi_{j})\). C stands for a constant which may vary from line to line and ‘→’ for the convergence as \(n\rightarrow\infty\). The remainder of the paper is organized as follows. In Section 2, we define an in plane Cox process and prove that the timenormalized point process \(N_{n}\) of exceedances of levels \(u^{(1)}_{n}, u^{(2)}_{n} ,\ldots, u^{(r)}_{n}\) by \(\{\xi_{i},1\leq i\leq n\}\) converges in distribution to the in plane Cox process. In Section 3, as the applications of our main result, the asymptotic results of the probabilities \(P(a_{n}(M_{n}^{(2)}b_{n})\leq x, L_{n}^{(2)}/n\leq t)\), and \(P(a_{n}(M_{n}^{(1)}b_{n})\leq x_{1}, a_{n}(M_{n}^{(2)}b_{n})\leq x_{2})\) are established.
Convergence of point processes of exceedances
Let \(\{\xi_{i}, i\geq1\}\) be a standardized normal sequence with correlation coefficients \(r_{ij}=\operatorname{Cov}(\xi_{i},\xi_{j})\) satisfying the following assumptions:
We concentrate on deriving the convergence of the timenormalized exceedance point process \(N_{n}\) of the levels \(u^{(1)}_{n}, u^{(2)}_{n},\ldots,u^{(r)}_{n}\) by \(\{\xi_{i},1\leq i\leq n\}\) where \(u^{(k)}_{n}=x_{k}/a_{n}+b_{n}\), \(k=1,2,\ldots,r\). In order to prove the main result in this section, we shall use the famous Berman’s inequality which is based on the early work of Slepian [13], Berman [14] and is polished up in Li and Shao [15]. The latest results related to Berman’s inequality are Hashorva and Weng [16] and Lu and Wang [17]. The former gave a detailed introduction to Berman’s inequality and derived the inequality for some general scaling random variable and thus obtained a Berman inequality for nonnormal random vector. The upper bound of Berman’s inequality gives an estimate of the difference between two standardized ndimensional distribution functions by a convenient function of their covariances. According to Hashorva and Weng [16], some results for normal sequences may be extended to nonnormal cases. The following lemmas are also needed in the proof of our result.
Lemma 2.1
Let \(d>0\) and \(\gamma\geq0\) be constants, put \(\rho_{n}=\gamma/\log n\) and suppose that \(r_{n}\log n\rightarrow\gamma\) as \(n\rightarrow\infty\). Then, for any sequence \(\{u_{n}\}\) such that \(n(1\Phi(u_{n}))\) is bounded, we have
where \(w_{k}=\max\{r_{k},\rho_{n}\}\).
Proof
The proof can be found on p.134 in Leadbetter et al. [1]. □
Leadbetter et al. [1] defined a Cox process N with intensity \(\exp(x\gamma+\sqrt{2\gamma}\zeta)\), where ζ is a standard normal random variable, i.e. the process has the distribution determined by the following probability:
where \(m(\cdot)\) is the Lebesgue measure and proved that a point process \(N_{n}\) of timenormalized exceedances converges in distribution to N on \((0,+\infty)\) which is summarized as Lemma 2.2 below.
Lemma 2.2
Suppose \(\{\xi_{i},i\geq1\}\) is a standard stationary normal sequence with covariances satisfying (2.1). Then the point process \(N_{n}\) of timenormalized exceedances of the level \(u_{n}(u_{n}=x/a_{n}+b_{n})\) converges in distribution to N on \((0,+\infty)\), where N is the Cox process defined by (2.2).
Proof
The proof can be found on p.136 in Leadbetter et al. [1]. □
In Theorem 2.1 below, we extend Lemma 2.2 to the case of exceedances of several levels and study a vector of point processes \(N_{n}=(N_{n}^{(1)},N_{n}^{(2)},\ldots,N_{n}^{(r)})\) which arises when \(\{\xi_{i},1\leq i\leq n\}\) exceeds the levels \(u_{n}^{(1)},u_{n}^{(2)},\ldots,u_{n}^{(r)}\), where \(u_{n}^{(k)}=x_{k}/a_{n}+b_{n}\), \(1\leq k\leq r\). For clarity, we record the locations of \(u_{n}^{(1)},u_{n}^{(2)},\ldots,u_{n}^{(r)}\) along fixed horizontal lines \(L_{1},L_{2},\ldots,L_{r}\) in the plane. The structure of the process vector is the same as that of the exceedances process on pp.111112 in Leadbetter et al. [1] where the authors presented a detailed and visualized introduction. According to Lemma 2.2, each onedimensional point process, on a given \(L_{k}\), i.e. \(N_{n}^{(k)}\) converges to a Cox process in distribution under appropriate conditions. Before presenting Theorem 2.1, we first give two definitions, one of which concerns a twodimension Cox process, i.e. an in plane Cox process.
Definition 2.1
The locations of order statistics are the places where order statistics appear in the index set, for example the location of the maxima of \(\{\xi_{i},1\leq i\leq n\}\) varying among \(1,\ldots,n\).
Definition 2.2
Let \(\{\sigma_{1j},j=1,2,\ldots\}\) be the points of a Cox process \(N^{(k)}\) on \(L_{r}\) with (stochastic) intensity \(\exp(x_{r}\gamma+\sqrt{2\gamma}\zeta)\), where ζ is a standard normal random variable, i.e. \(N^{(k)}\) has the distribution characterized in (2.2). Let \(\beta_{j}\), \(j=1,2,\ldots\) be independent and identically distributed (i.i.d.) random variables, independent also of the Cox process on \(L_{r}\), taking values \(1,2,\ldots,r\) with conditional probabilities
i.e. \(P(\beta_{j}\geq s\zeta=z)=\tau_{rs+1}/\tau_{r}\) for \(s=1,2,\ldots,r\) where \(\tau_{i}=e^{x_{i}\gamma+\sqrt{2\gamma}z}\), \(i=1, 2,\ldots,r\). For each j, place points \(\sigma_{2j}, \sigma_{3j},\ldots, \sigma_{\beta_{j}j}\) on \(\beta_{j}1\) lines \(L_{r1}, L_{r2},\ldots, L_{r\beta_{j}+1}\), vertically above \(\sigma_{1j}\), we can obtain an in plane Cox process N. Obviously the probability that a point appears \(L_{r1}\) above \(\sigma_{1j}\) is \(P(\beta_{j}\geq2\zeta=z)=\tau_{r1}/\tau_{r}\) and the deletions are conditionally independent, so that \(N^{(r1)}\) is obtained as a conditionally independent thinning of the Cox process \(N^{(r)}\).
The structure of the in plane Cox process N is very similar to that of the Poisson process on p.112 in Leadbetter et al. [1], but the independent thinning is replaced with the conditionally independent thinning here.
Theorem 2.1
Suppose \(\{\xi_{i},i\geq1\}\) is a standardized normal sequence satisfying the conditions in Lemma 2.2. Let \(u^{(k)}_{n}=x_{k}/a_{n}+b_{n}\) satisfy \(u^{(1)}_{n}\geq u^{(2)}_{n}\geq\cdots\geq u^{(r)}_{n}\) (\(1\leq k\leq r\)) where \(a_{n}\) and \(b_{n}\) are defined in (1.1). Then the timenormalized point process \(N_{n}\) of exceedances of levels \(u^{(1)}_{n}, u^{(2)}_{n} ,\ldots, u^{(r)}_{n}\) by \(\{\xi_{i},1\leq i\leq n\}\) converges in distribution to the abovementioned in plane Cox process.
Proof
It is sufficient to show that when n goes to ∞:

(a)
\(E(N_{n}(B))\rightarrow E(N(B))\) for all sets B of the form \((c,d]\times(r,\delta]\), \(r<\delta\), \(0< c<d\), where \(d\leq1\) and \(E(\cdot)\) is the expectation,

(b)
\(P(N_{n}(B)=0)\rightarrow P(N(B)=0)\) for all sets B which are finite unions of disjoint sets of this form.
Focus on (a) firstly. If \(B=(c,d]\times(r,\delta]\) intersects any of the lines, suppose these are \(L_{s},L_{s+1},\ldots,L_{t}\) (\(1\leq s\leq t\leq r\)). Then
and the number of points \(j/n\) in \((c,d]\) is \([nd][nc]\). As in the proof of Theorem 5.5.1 on p.113 in Leadbetter et al. [1], we have \(E(N_{n}(B))=([nd][nc])\sum^{t}_{k=s}(1F(u_{n}^{(k)}))\), where
Obviously
Thus, we have \(E(N_{n}(B))\sim n(dc)\sum^{t}_{k=s}(\frac{e^{x_{k}}}{n}+o(\frac{1}{n}))\rightarrow (dc)\sum^{t}_{k=s}e^{x_{k}}\). Since
(a) follows. In order to prove (b), we must show that \(P(N_{n}(B)=0)\rightarrow P(N(B)=0)\), where \(B=\bigcup^{m}_{1}C_{k}\) with disjoint \(C_{k}=(c_{k}, d_{k}]\times(r_{k}, s_{k}]\). It is convenient to discard any set \(C_{k}\) which does not intersect any of the lines \(L_{1}, L_{2}, \ldots, L_{r}\). Because there are intersections and differences of the intervals \((c_{k}, d_{k}]\), we may write B in the form \(\bigcup^{s}_{k=1}(c_{k}, d_{k}]\times E_{k}\), where \((c_{k}, d_{k}]\) are disjoint and \(E_{k}\) is a finite union of semiclosed intervals. It therefore follows that
where \(F_{k}=(c_{k},d_{k}]\times E_{k}\). Denote the lowest \(L_{j}\) intersecting \(F_{k}\) by \(L_{l_{k}}\). By the above thinning property, obviously
where \(M_{n}(c_{k},d_{k})\) stands for the maximum of \(\{\eta_{i}, i\geq1\}\) with index k (\([cn]< k\leq[dn]\)). Consider the probabilities of (2.4) and (2.5) and obtain
It is convenient to firstly prove the following result. Let \(\{\bar{\xi}_{i}, i\geq1\}\) be a standardized normal sequence with the correlation coefficient ρ. \(M_{n}(c,d;\rho)\) stands for the maximum of \(\{\bar{\xi}_{k}\}\) with index k (\([cn]< k\leq[dn]\)). It is well known that \(M_{n}(c_{1}, d_{1}; \rho),\ldots,M_{n}(c_{k}, d_{k}; \rho)\) have the same distribution as \((1\rho)^{1/2}M_{n}(c_{1}, d_{1}; 0)+\rho^{1/2}\zeta,\ldots, (1\rho)^{1/2}M_{n}(c_{k}, d_{k}; 0)+\rho^{1/2}\zeta\), where \(c=c_{1}< d_{1}<\cdots<c_{k}<d_{k}=d\) and ζ is a standard normal variable; see Leadbetter et al. [1]. Next we must estimate the bound of
where \(\rho_{n}=\gamma/\log n\).
Using Berman’s inequality, the bound of (2.7) does not exceed
where the sum is carried out over \(i< j\) and \(i,j\in\bigcup^{s}_{k=1}([c_{k}n],[d_{k}n]]\), \(u_{n}^{(i)}\) or \(u_{n}^{(j)}\) stands for \(x_{i}/a_{n}+b_{n}\) or \(x_{j}/a_{n}+b_{n}\), and \(\omega_{ij}=\max\{r_{ij},\rho_{n}\}\). Furthermore, (2.8) does not exceed
Noting \(n(1\Phi((\min_{1\leq i\leq n}x_{i})/a_{n}+b_{n}))\) is bounded, the last ‘→’ attributes to Lemma 2.1. So it suffices to prove
Noting the definition of \(M_{n}(c_{k}, d_{k}, \rho_{n})\), clearly, it follows that
where the proof of the last ‘=’ can be completed by using the argument on the first line from the bottom on p.136 in Leadbetter et al. [1]. Since \(a_{n}=(2\log n)^{\frac{1}{2}}\), \(b_{n}=a_{n}+O(a_{n}^{1}\log\log n)\), and \(\rho_{n}=\gamma/\log n\), it is easy to show
see also the proof of Theorem 6.5.1 on p.137 in Leadbetter et al. [1]. Furthermore, we may obtain the following result:
where \(\tilde{\zeta}_{k}\) is a sequence of independent standard normal variables and we used the same arguments as (2.3) for the last step. Using the dominated convergence theorem, it follows that
The proof of (b) is completed. □
Corollary 2.1
Suppose \(\{\xi_{i}, i\geq1\}\) satisfies the conditions of Theorem 2.1. Let \(B_{1},\ldots,B_{s}\) be Borel subsets of the unit interval, whose boundaries have zero Lebesgue measure. Then for integers \(m^{(k)}_{j}\),
Proof
Combining Theorem 2.1 and the proof of Corollary 5.5.2 in Leadbetter et al. [1], we can complete the proof. □
Theorem 2.2
Let the levels \(u^{(k)}_{n}\) (\(1\leq k\leq r\)) satisfy
with \(u^{(1)}_{n}\geq u^{(2)}_{n}\geq\cdots\geq u^{(r)}_{n}\). Let \(S^{(k)}_{n}\) be the number of exceedances of \(u^{(k)}_{n}\) by \(\{\xi_{i},1\leq i\leq n\}\). Then for \(k_{1}\geq0,k_{2}\geq0,\ldots, k_{r}\geq0\),
Proof
By Corollary 2.1, the lefthand side of (2.9) converges to
where \(S^{(i)}=N^{(i)}([0, 1])\). In our paper, the definition of the Cox process is similar to that of the in plane Poisson process in Leadbetter et al. [1]. So we can refer to the proof of Theorem 5.6.1 in Leadbetter et al. [1] and find that (2.10) equals
The proof is completed since
□
The joint distributions of some order statistics
This section contains two important results which concerns the joint distributions of order statistics of \(\{\xi_{i},i\geq1\}\).
Theorem 3.1
Suppose \(\{\xi_{i}, i\geq1\}\) is a standard normal sequence satisfying the conditions of Theorem 2.1. Let \(u^{(k)}_{n}=x_{k}/a_{n}+b_{n}\) and \(M_{n}^{(2)}\), \(L_{n}^{(2)}\) be the second largest maxima of \(\xi_{1}, \xi_{2},\ldots,\xi_{n}\) and its location. Then for \(x_{1}>x_{2}\), as \(n\rightarrow\infty\),
and
where
Proof
Clearly the lefthand side of (3.1) is equal to
where \(S^{(i)}_{n}\) is the number of exceedances of \(u^{(i)}_{n}\) by \(\eta_{1}, \eta_{2}, \ldots, \eta_{n}\). Using Theorem 2.2 in the special case yields (3.1). In order to prove (3.2), write I and J for intervals \(\{1, 2,\ldots, [nt]\}\) and \(\{[nt] + 1,\ldots, n\}\), respectively. \(M^{(1)}(I)\), \(M^{(2)}(I)\), \(M^{(1)}(J)\), \(M^{(2)}(J)\) stand for the maxima and second largest maxima of \(\xi_{1}, \xi_{2},\ldots, \xi_{n}\) in the intervals I, J, respectively. Let \(H_{n}(x_{1}, x_{2}, x_{3}, x_{4})\) be the joint d.f. of the normalized r.v.
Generally let \(x_{1} > x_{2}\) and \(x_{3} > x_{4}\), that is,
where \(I'=(0,t]\) and \(J'=(t,1]\). By using Corollary 2.1 with \(B_{1}=I'\) and \(B_{2} = J'\), we obtain
Now the lefthand side of (3.2) is equal to
Obviously H is absolutely continuous and the boundaries of sets in \(R^{4}\) such as \(\{(w_{1},w_{2},w_{3}, w_{4}): w_{2}\leq x_{2},w_{2}>w_{3}\}\) and \(\{(w_{1},w_{2},w_{3},w_{4}) : w_{1}\leq x_{2},w_{3} > w_{1}\geq w_{4}\}\) have zero Lebesgue measure. Thus using Corollary 2.1, it follows that (3.3) converges to
According to the joint distribution \(H(x_{1}, x_{2}, x_{3}, x_{4})\) of \(X_{1}\), \(X_{2}\), \(Y_{1}\), and \(Y_{2}\), a simple evaluation completes the proof. □
Remark 3.1
We may obtain the joint asymptotic distribution of \(M^{(1)}_{n} ,M^{(2)}_{n} , \ldots, M^{(k)}_{n}\) by using the same method as in Theorem 3.1.
References
 1.
Leadbetter, MR, Lindgren, G, Rootzen, H: Extremes and Related Properties of Sequences and Processes. Springer, New York (1983)
 2.
Mittal, Y, Ylvisaker, D: Limit distributions for the maxima of stationary Gaussian processes. Stoch. Process. Appl. 3, 118 (1975)
 3.
Ho, HC, Hsing, T: On the asymptotic joint distribution of the sum and maximum of stationary normal random variables. J. Appl. Probab. 33, 138145 (1996)
 4.
Tan, Z, Peng, Z: Joint asymptotic distribution of exceedances point process and partial sum of stationary Gaussian sequence. Appl. Math. J. Chin. Univ. Ser. B 26, 319326 (2011)
 5.
Hashorva, E, Peng, Z, Weng, Z: On Piterbarg theorem for maxima of stationary Gaussian sequences. Lith. Math. J. 53, 280292 (2013)
 6.
Piterbarg, VI: Asymptotic Methods in the Theory of Gaussian Processes and Fields. Translations of Mathematical Monographs, vol. 148. Am. Math. Soc., Providence (1996)
 7.
Hu, A, Peng, Z, Qi, Y: Limit laws for maxima of contracted stationary Gaussian sequences. Metrika 70, 279295 (2009)
 8.
Falk, M, Hüsler, J, Reiss, RD: Laws of Small Numbers: Extremes and Rare Events, 3rd edn. DMV Seminar, vol. 23. Birkhäuser, Basel (2010)
 9.
Peng, Z, Tong, J, Weng, Z: Joint limit distributions of exceedances point processes and partial sums of Gaussian vector sequence. Acta Math. Sin. Engl. Ser. 28, 16471662 (2012)
 10.
Hashorva, E, Peng, Z, Weng, Z: Limit properties of exceedances point processes of scaled stationary Gaussian sequences. Probab. Math. Stat. 34, 4559 (2014)
 11.
Wiśniewski, M: Multidimensional point processes of extreme order statistics. Demonstr. Math. 27, 475485 (1994)
 12.
Lin, F, Shi, D, Jiang, Y: Some distributional limit theorems for the maxima of Gaussian vector sequences. Comput. Math. Appl. 64, 24972506 (2012)
 13.
Slepian, D: The onesided barrier problem for Gaussian noise. Bell Syst. Tech. J. 41, 463501 (1962)
 14.
Berman, SM: Limit theorems for the maximum term in stationary sequences. Ann. Math. Stat. 35, 502516 (1964)
 15.
Li, WV, Shao, QM: A normal comparison inequality and its applications. Probab. Theory Relat. Fields 122, 494508 (2002)
 16.
Hashorva, E, Weng, Z: Berman’s inequality under random scaling. Stat. Interface 7, 339349 (2014)
 17.
Lu, D, Wang, X: Some new normal comparison inequalities related to Gordon’s inequality. Stat. Probab. Lett. 88, 133140 (2014)
Acknowledgements
The research of the second author was supported by the National Natural Science Foundation of China, Grant No. 71171166. The research of other authors was supported by the Scientific Research Fund of Sichuan Provincial Education Department under Grant 12ZB082, the Scientific research cultivation project of Sichuan University of Science and Engineering under Grant 2013PY07, the Scientific Research Fund of Sichuan University of Science and Engineering under Grant 2013KY03, and the Science Research Programs for Doctors in Southwestern University of Finance and Economics, Grant No. JBK1207085.
Author information
Affiliations
Corresponding author
Additional information
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
All authors contributed equally to this work. All authors read and approved the final manuscript.
Rights and permissions
Open Access This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.
About this article
Cite this article
Lin, F., Shi, D. & Jiang, Y. Limit properties of exceedance point processes of strongly dependent normal sequences. J Inequal Appl 2015, 63 (2015). https://doi.org/10.1186/s1366001505858
Received:
Accepted:
Published:
MSC
 60F05
 62E20
Keywords
 Cox process
 exceedance point process
 strongly dependent normal sequences
 kth maxima