On accuracy of simulation of gaussian stationary processes in L2([0, T])

A theorem about simulation of a Gaussian stochastic process with given accuracy and reliability in L2([0, T ]) using wavelets has been proved.

Збережено в:
Бібліографічні деталі
Дата:2006
Автор: Turchyn, Y.
Формат: Стаття
Мова:English
Опубліковано: Інститут математики НАН України 2006
Онлайн доступ:https://nasplib.isofts.kiev.ua/handle/123456789/4469
Теги: Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
Назва журналу:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Цитувати:On accuracy of simulation of gaussian stationary processes in L2([0, T]) / Y. Turchyn // Theory of Stochastic Processes. — 2006. — Т. 12 (28), № 3-4. — С. 255–260. — Бібліогр.: 5 назв.— англ.

Репозитарії

Digital Library of Periodicals of National Academy of Sciences of Ukraine
id nasplib_isofts_kiev_ua-123456789-4469
record_format dspace
spelling Turchyn, Y.
2009-11-11T15:29:55Z
2009-11-11T15:29:55Z
2006
On accuracy of simulation of gaussian stationary processes in L2([0, T]) / Y. Turchyn // Theory of Stochastic Processes. — 2006. — Т. 12 (28), № 3-4. — С. 255–260. — Бібліогр.: 5 назв.— англ.
0321-3900
https://nasplib.isofts.kiev.ua/handle/123456789/4469
A theorem about simulation of a Gaussian stochastic process with given accuracy and reliability in L2([0, T ]) using wavelets has been proved.
en
Інститут математики НАН України
On accuracy of simulation of gaussian stationary processes in L2([0, T])
Article
published earlier
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
collection DSpace DC
title On accuracy of simulation of gaussian stationary processes in L2([0, T])
spellingShingle On accuracy of simulation of gaussian stationary processes in L2([0, T])
Turchyn, Y.
title_short On accuracy of simulation of gaussian stationary processes in L2([0, T])
title_full On accuracy of simulation of gaussian stationary processes in L2([0, T])
title_fullStr On accuracy of simulation of gaussian stationary processes in L2([0, T])
title_full_unstemmed On accuracy of simulation of gaussian stationary processes in L2([0, T])
title_sort on accuracy of simulation of gaussian stationary processes in l2([0, t])
author Turchyn, Y.
author_facet Turchyn, Y.
publishDate 2006
language English
publisher Інститут математики НАН України
format Article
description A theorem about simulation of a Gaussian stochastic process with given accuracy and reliability in L2([0, T ]) using wavelets has been proved.
issn 0321-3900
url https://nasplib.isofts.kiev.ua/handle/123456789/4469
citation_txt On accuracy of simulation of gaussian stationary processes in L2([0, T]) / Y. Turchyn // Theory of Stochastic Processes. — 2006. — Т. 12 (28), № 3-4. — С. 255–260. — Бібліогр.: 5 назв.— англ.
work_keys_str_mv AT turchyny onaccuracyofsimulationofgaussianstationaryprocessesinl20t
first_indexed 2025-11-25T13:48:46Z
last_indexed 2025-11-25T13:48:46Z
_version_ 1850513606900187136
fulltext Theory of Stochastic Processes Vol. 12 (28), no. 3–4, 2006, pp. 255–260 YEVGEN TURCHYN ON ACCURACY OF SIMULATION OF GAUSSIAN STATIONARY PROCESSES IN L2([0, T ]) A theorem about simulation of a Gaussian stochastic process with given accuracy and reliability in L2([0, T ]) using wavelets has been proved. 1. Introduction Problems of expansion of random processes in series over uncorrelated random variables using wavelets were considered in [3], [4], [5] and other papers. There has been proved in [3] a theorem about accuracy and reli- ability of such expansions in Lp([0, T ]) for Gaussian wide-sense stationary processes. We’ll give a refinement of this theorem for L2([0, T ]). Since Gaussian processes are widely used in financial and actuarial mathematics results of the article may be used in these areas. First of all we need to list some necessary facts. Let ϕ ∈ L2(R) be such a function that the following assumptions hold: i) ∑ k∈Z |ϕ̂(y+2πk)|2 = 1 almost everywhere, where ϕ̂(y) is the Fourier transform of ϕ, ϕ̂(y) = ∫ R exp{−iyx}ϕ(x)dx; ii) There exists a function m0 ∈ L2([0, 2π]) such that m0(x) has pe- riod 2π and almost everywhere ϕ̂(y) = m0 (y 2 ) ϕ̂ (y 2 ) ; iii) ϕ̂(0) �= 0 and the function ϕ̂(y) is continuos at 0. Function ϕ(x) is called f-wavelet. Let ψ(x) be the inverse Fourier trans- form of the function ψ̂(y) = m0 (y 2 + π ) exp { − i y 2 } ϕ̂ (y 2 ) . Function ψ(x) is called m-wavelet. Let ϕjk(x) = 2 j 2 ϕ(2jx − k), ψjk(x) = 2 j 2 ψ(2jx−k), k ∈ Z, j = 0, 1, 2, . . . , It is known that the family of functions {ϕ0k, ψjk, j = 0, 1, 2, . . . , k ∈ Z} is an orthonormal basis in L2(R) (see, for 2000 Mathematics Subject Classification. Primary 60G10, 60G15, 42C40. Key words and phrases. Stochastic processes, simulation, wavelets. 255 256 YEVGEN TURCHYN example, the book [1]). Such a basis is called wavelet basis. Any function f ∈ L2(R) can be represented in the form (1) f(x) = ∑ k∈Z α0kϕ0k(x) + ∞∑ j=0 ∑ k∈Z βjkψjk(x), where α0k = ∫ R f(x)ϕ0k(x)dx, βjk = ∫ R f(x)ψjk(x)dx, ∑ k∈Z |α0k|2 + ∞∑ j=0 ∑ k∈Z |βjk|2 < ∞. That is, series (1) converges in the norm of the space L2(R). Representation (1) is called wavelet representation. Let X = {X(t), t ∈ R} be a centered wide-sense stationary random process (from this moment we will refer to wide-sense stationary processes simply as stationary processes), X(t, ω) ∈ L2(Ω, F, P ) for all t ∈ R (where (Ω, F, P ) is the standard probability space to which belong all random vari- ables X(t, ω) ), R(τ) = EX(t + τ)X(t). There has been proved the following theorem in [3]. Theorem 1. Let X = {X(t), t ∈ R} be centered stationary random process, R(τ) = EX(t+τ)X(t). Suppose that R(τ) is a continuous function and process X(t) has spectral density, i.e. R(τ) = ∫ R exp{−iτλ}f(λ)dλ, where f is real-valued, f(λ) ≥ 0, ∫∞ −∞ f(λ)dλ = R(0) < ∞. Let {ϕ0k(x), ψjk(x), k ∈ Z, j = 0, 1, 2, . . .} be a wavelet basis. Then (2) X(t) = ∑ k∈Z ξ0kα0k(t) + ∞∑ j=0 ∑ k∈Z ηjkβjk(t), where (3) α0k(t) = 1√ 2π ∫ R (f(y))1/2 exp{−iy(t − k)}ϕ̂(y)dy, (4) βjk(t) = 1√ 2π2j/2 ∫ R (f(y))1/2 exp { −iy ( t − k 2j )} ψ̂ ( y 2j ) dy, ξ0k and ηjk are centered random variables such that Eξ0kξ0l = δkl, Eηmkηnl = δmnδkl, Eξ0kηnl = 0, δkl is Kronecker delta and series (2) converges in mean square. The expansion (2) has been used in [3] for modeling of Gaussian stochas- tic processes and obtaining inequality for given accuracy and reliability of appropriate model in Lp([0, T ]). We’ll give refinement of these theorems for L2([0, T ]). ON ACCURACY OF SIMULATION OF GAUSSIAN PROCESSES 257 2. Simulation of stationary Gaussian processes. If we have a Gaussian process X(t) which satisfies conditions of theorem 1, then we can consider as a model of X(t) a process X̂(t) = N0−1∑ k=−(N0−1) ξ0kα0k(t) + N−1∑ j=0 Mj−1∑ k=−(Mj−1) ηjkβjk(t), where ξ0k, ηjk are independent random variables with distribution N(0, 1), α0k(t) and βjk(t) are calculated using formulae (3) and (4), N0 > 1, N > 1, Mj > 1 (j = 0, . . . , N − 1). Definition. Model X̂(t) approximates process X(t) with given reliability 1 − δ, 0 < δ < 1, and accuracy ε > 0 in Lp([0, T ]) if P {(∫ T 0 |X(t) − X̂(t)|pdt )1/p > ε } ≤ δ. There has been proved the following theorem in [3]. Theorem 2. Stochastic process X̂(t) approximates process X(t) with reli- ability 1 − δ and accuracy ε in Lp([0, T ]) (p ≥ 1, 0 < δ < min{1, 2e−p/2}), (5) P {(∫ T 0 |X(t) − X̂(t)|pdt )1/p > ε } ≤ δ if sup t∈[0,T ] E|X(t) − X̂(t)|2 = sup t∈[0,T ] ( ∑ k:|k|≥N0 |α0k(t)|2 + N−1∑ j=0 ∑ k:|k|≥Mj |βjk(t)|2+ ∞∑ j=N ∑ k∈Z |βjk(t)|2 ) ≤ ε2 2T 2/p ln 2 δ . This result can be made more exact for p = 2. We’ll give first some auxiliary facts. There has been proved the following assertion in [2], which we give here in simplified form. Theorem 3. Let {T,U, μ} be a measurable space. Consider a random series (6) S(t) = ∞∑ k=1 ξkfk(t), t ∈ T, where ξ = {ξk, k = 1, 2, . . .} is a family of Gaussian random variables and f = {fk(t), k = 1, 2, . . .} is a family of real-valued L2(T) functions. Let the 258 YEVGEN TURCHYN random variables in (6) be either uncorrelated with Eξ2 k = σ2 k or the system of functions fk(t) be orthogonal, that is,∫ T fk(t)fl(t)dμ(t) = δkla 2 k, where δkl is the Kronecker symbol. If the series ∞∑ k=1 σ2 ka 2 k < ∞ converges, then the series (6) is mean square convergent in L2(T) and for all x > √ An and n=1,2, . . . we have (7) P ⎧⎨ ⎩ ∥∥∥∥∥ ∞∑ k=n ξkfk(t) ∥∥∥∥∥ L2(T) > x ⎫⎬ ⎭ ≤ e1/2 x√ An exp { − x2 2An } , where An = ∑∞ k=n σ2 ka 2 k. The following statement about bounds for coefficients α0k(t) and βjk(t) was proved in [3]. Lemma 1. Let R(τ) be a covariance function which satisfies conditions of theorem 1, R(τ) = ∫ R exp{−iτλ}f(λ)dλ. Let ϕ̂(y) be the Fourier transform of a f-wavelet ϕ(x). Let ϕ̂(y) be a continuous function and assertions i) – iii) hold true for all y ∈ R. Let ψ̂(y) be the Fourier transform of m-wavelet ψ(x) corresponding to ϕ(x). Let g(y) = √ f(y) and there exist g′(y), ψ̂′(y), ϕ̂′(y); |ψ̂(y)| < C1, |ψ̂′(y)| < C2, |ϕ̂(y)| is bounded,∫ R g(y)dy < ∞, ∫ R |g′(y)||y|dy < ∞, ∫ R g(y)|y|dy < ∞, ∫ R |g′(y)||ϕ̂(y)|dy < ∞, ∫ R g(y)|ϕ̂′(y)|dy < ∞, α0k(t) and βjk(t) are given in (3) and (4). If k �= 0 then for all t ∈ R |βjk(t)| ≤ A + B|t| |k|2j/2 , where A = C2√ 2π ∫ R (|g′(y)||y|+ g(y))dy, B = C2√ 2π ∫ R g(y)|y|dy; |α0k(t)| ≤ A1 + B1|t| |k| , where A1 = 1√ 2π ∫ R (|g′(y)||ϕ̂(y)|+ g(y)|ϕ̂′(y)|)dy, B1 = 1√ 2π ∫ R g(y)|y||ϕ̂(y)|dy. ON ACCURACY OF SIMULATION OF GAUSSIAN PROCESSES 259 For all t ∈ R and j = 0, 1, 2, . . . |βj0(t)| ≤ C2√ 2π23j/2 ∫ R g(y)|y|dy, |α00(t)| ≤ 1√ 2π ∫ R g(y)|ϕ̂(y)|dy. Now we can formulate the main result of our paper. Theorem 4. Let Gaussian process X(t) satisfy restrictions of theorem 1, f-wavelet ϕ and m-wavelet ψ satisfy conditions of lemma 1. Model X̂(t) approximates process X(t) with accuracy ε and reliability 1− δ (0 ≤ δ ≤ 1) in L2([0, T ]), (8) P {(∫ T 0 |X(t) − X̂(t)|2dt )1/2 > ε } ≤ δ, if An < ε2 x2 δ , where xδ (xδ ≥ 1) is a root of equation e1/2xe−x2/2 = δ, An = ∑ k:|k|≥N0 ∫ T 0 |α0k(t)|2dt+ + N−1∑ j=0 ∑ k:|k|≥Mj ∫ T 0 |βjk(t)|2dt + ∞∑ j=N ∑ k∈Z ∫ T 0 |βjk(t)|2dt, α0k(t) and βjk(t) are defined by formulae (3) and (4). Proof. Let’s consider stochastic process (9) X̃(t) = ∑ k∈Z ξ∗0kα0k(t) + ∞∑ j=0 ∑ k∈Z η∗ jkβjk(t), where ξ∗0k, η ∗ jl (j = 0, 1, 2, . . . ; k, l ∈ Z) are i.i.d. random variables with distribution N(0, 1). It’s easy to see that correlation functions of processes X(t) and X̃(t) are the same. Therefore we may consider process X̃(t) instead of process X(t). If we apply theorem 3 to series (9) we obtain inequality (8). Corollary. Model X̂(t) approximates process X(t) with accuracy ε and reliability 1 − δ in L2([0, T ]) if the following inequalities are satisfied: N0 > 1 + 6D2 ε1 , N > log2 ( 3 ε1 ( 8D1 + 8D2T 7 )) , Mj > 12D1 ε1 ( 1 − 1 2N ) + 1, (j = 0, 1, . . . , N − 1), where ε1 = ε2 x2 δ , D = C2√ 2π ∫ R g(y)|y| dy, D1 = A2T + ABT 2 + 1 3 B2T 3, D2 = A2 1T + A1B1T 2 + 1 3 B2 1T 3; g(y), A, B, A1, B1 C2 are defined in lemma 1, xδ is defined in theorem 4. 260 YEVGEN TURCHYN Proof. If we apply lemma 1 it’s easy to see that under conditions of the corollary the following inequalities are true: An < 2D2 N0 − 1 + 4D1 2N−1 + D2T 1 7 · 8N−1 + D1 N−1∑ j=0 1 2j−1(Mj − 1) , 2D2 N0 − 1 < ε1 3 , 4D1 2N−1 + D2T 1 7 · 8N−1 < ε1 3 , D1 N−1∑ j=0 1 2j−1(Mj − 1) < ε1 3 . Statement of the corollary immediately follows from these inequalities and theorem 4. � 3. Conclusions There has been obtained a theorem about accuracy and reliability of simulation in L2([0, T ]) for a certain class of stationary Gaussian processes. This theorem is more exact than previous result for Lp([0, T ]) when p = 2. Author expresses his thanks to professor Yuriy V. Kozachenko for valu- able discussions. References 1. W.Härdle, G. Kerkacharian, D. Picard, A. Tsybakov, Wavelets, Approximation and Statistical Applications, Springer, New York, (1998). 2. Yu.V. Kozachenko, A.O. Pashko, Accuracy of simulation of stochastic processes in norms of Orlicz spaces. I, Theory Probab. Math. Statist., 58, (1999), 51–92. 3. Yu. Kozachenko, E. Turchyn, On a generalization of Walter-Zhang’s wavelet- based KL-like expansion for stationary random processes, submitted to Applied and Computational Harmonic Analysis. 4. G. Walter, J. Zhang, A wavelet-based KL-like expansion for wide-sense stationary random processes, IEEE Trans. Signal Process., 42, (1994), no.7, 1737–1745. 5. G. Walter, J. Zhang, Wavelets based on band-limited processes with a KL-type property, Proc. SPIE Conf. Visual Inform. Processing II (Orlando, FL), Apr. 1993, 336–343. Department of Higher Mathematics, Dnipropetrovsk State Agricultural University, Voroshylova str., 25, Dnipropetrovsk, Ukraine E-mail address: turchyn@a-teleport.com