Stochastic impulsive processes on superposition of two renewal processes

Stochastic impulsive processes given by a sum of random variables on superposition of two renewal processes are considered on increasing time intervals. Algorithms of average, diffusion approximation and large deviation generators are realized in the series scheme with a small series parameter under...

Full description

Saved in:
Bibliographic Details
Published in:Український математичний вісник
Date:2014
Main Authors: Koroliuk, V.S., Manca, R., D'Amico, G.
Format: Article
Language:English
Published: Інститут прикладної математики і механіки НАН України 2014
Online Access:https://nasplib.isofts.kiev.ua/handle/123456789/124466
Tags: Add Tag
No Tags, Be the first to tag this record!
Journal Title:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Cite this:Stochastic impulsive processes on superposition of two renewal processes / V.S. Koroliuk, R. Manca, G. D'Amico // Український математичний вісник. — 2014. — Т. 11, № 3. — С. 366-379. — Бібліогр.: 7 назв. — англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
id nasplib_isofts_kiev_ua-123456789-124466
record_format dspace
spelling Koroliuk, V.S.
Manca, R.
D'Amico, G.
2017-09-26T16:20:04Z
2017-09-26T16:20:04Z
2014
Stochastic impulsive processes on superposition of two renewal processes / V.S. Koroliuk, R. Manca, G. D'Amico // Український математичний вісник. — 2014. — Т. 11, № 3. — С. 366-379. — Бібліогр.: 7 назв. — англ.
1810-3200
2010 MSC. 60J45, 60K05.
https://nasplib.isofts.kiev.ua/handle/123456789/124466
Stochastic impulsive processes given by a sum of random variables on superposition of two renewal processes are considered on increasing time intervals. Algorithms of average, diffusion approximation and large deviation generators are realized in the series scheme with a small series parameter under suitable scalings.
en
Інститут прикладної математики і механіки НАН України
Український математичний вісник
Stochastic impulsive processes on superposition of two renewal processes
Article
published earlier
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
collection DSpace DC
title Stochastic impulsive processes on superposition of two renewal processes
spellingShingle Stochastic impulsive processes on superposition of two renewal processes
Koroliuk, V.S.
Manca, R.
D'Amico, G.
title_short Stochastic impulsive processes on superposition of two renewal processes
title_full Stochastic impulsive processes on superposition of two renewal processes
title_fullStr Stochastic impulsive processes on superposition of two renewal processes
title_full_unstemmed Stochastic impulsive processes on superposition of two renewal processes
title_sort stochastic impulsive processes on superposition of two renewal processes
author Koroliuk, V.S.
Manca, R.
D'Amico, G.
author_facet Koroliuk, V.S.
Manca, R.
D'Amico, G.
publishDate 2014
language English
container_title Український математичний вісник
publisher Інститут прикладної математики і механіки НАН України
format Article
description Stochastic impulsive processes given by a sum of random variables on superposition of two renewal processes are considered on increasing time intervals. Algorithms of average, diffusion approximation and large deviation generators are realized in the series scheme with a small series parameter under suitable scalings.
issn 1810-3200
url https://nasplib.isofts.kiev.ua/handle/123456789/124466
citation_txt Stochastic impulsive processes on superposition of two renewal processes / V.S. Koroliuk, R. Manca, G. D'Amico // Український математичний вісник. — 2014. — Т. 11, № 3. — С. 366-379. — Бібліогр.: 7 назв. — англ.
work_keys_str_mv AT koroliukvs stochasticimpulsiveprocessesonsuperpositionoftworenewalprocesses
AT mancar stochasticimpulsiveprocessesonsuperpositionoftworenewalprocesses
AT damicog stochasticimpulsiveprocessesonsuperpositionoftworenewalprocesses
first_indexed 2025-11-24T02:27:47Z
last_indexed 2025-11-24T02:27:47Z
_version_ 1850838142108565504
fulltext Український математичний вiсник Том 11 (2014), № 3, 366 – 379 Stochastic impulsive processes on superposition of two renewal processes Vladimir S. Koroliuk, Raimondo Manca, and Guglielmo D’Amico Abstract. Stochastic impulsive processes given by a sum of random variables on superposition of two renewal processes are considered on increasing time intervals. Algorithms of average, diffusion approximation and large deviation generators are realized in the series scheme with a small series parameter under suitable scalings. 2010 MSC. 60J45, 60K05. Key words and phrases. Impulsive process, average, diffusion ap- proximation, large deviation problem. 1. Introduction The Stochastic Impulsive Process (SIP) given by a sum of random variables on the Markov chains is described by the superposition on two Renewal Processes (RP) Sn = u+ n∑ k=1 αk(xk), n ≥ 0, S0 = u ∈ Rd (1.1) The Markov chain xk, k ≥ 0 is defined by the Markov Renewal Process (MRP) [1, ch. 1] on the space E = {±x; x > 0} with the sojourn times θ±n (x) := θ±n ∧ x, x ∈ R+ = (0,+∞). Each of the two RP is defined by a sum of positive i.i.d. random variables [2, S. 8.3]: τ±n = n∑ k=1 θ±k , n ≥ 0, τ±0 = 0, P±(t) = P{θ±k ≤ t}, t ≥ 0. (1.2) Received 3.03.2014 ISSN 1810 – 3200. c⃝ Iнститут математики НАН України V. S. Koroliuk, R. Manca, and G. D’Amico 367 The random variables (impulsive) α± k (x), x ∈ R+ are given by the distribution functions G±(A) = P{α± k (x) ∈ A}, A ∈ Rd. The SIP is a particular case of the Random Evolution Process (REP) [2]. In our previous work [3] the SIP was considered on the MRP with the merging phase space Ê = {+,−}. The increments of the SIP (1.1) ∆Sn = Sn − Sn−1 = α± n , n ≥ 0, x ∈ R+ may be interpreted as a success α+ n , or as a failure α− n . This is the natural interpretation of the SIP in the risk theory [5]. The asymptotical behaviour of the SIP in the series scheme (average and diffusion approximation [2]) and the scheme of asymptotically small diffusion [4, 6] is considered. The peculiarity of the MRP on the phase space E = {±x; x > 0} is that the stationary distribution of the Markov chain xn, n ≥ 0 is given in the explicit form (see Section 2). So, the algorithms of averaging (Proposition 4.1) and diffusion approximation (Proposition 5.1) may be realized effectively. Hence, the simplified models of the SIP may be used in applications to the risk problems [5, S.6.5]. 2. Superposition of two renewal processes The renewal processes are defined by sum of positive valued random variables, independent in common and identically distributed [1] (see also [2, S. 8.3]): τ±n = n∑ k=1 θ±k , n ≥ 0, τ±0 = 0, (2.1) P±(t) = P{θ±k ≤ t}, t ≥ 0, P±(0) = 0. The renewal processes can be given by the counting processes: ν±(t) := max{n > 0 : τ±n ≤ t}, t ≥ 0. The superposition of two renewal processes (2.1) is defined by the count- ing process ν(t) = ν+(t) + ν−(t), t ≥ 0. (2.2) 368 Stochastic impulsive processes... The superposition of two renewal processes (2.2) can be characterized by the Markov Renewal Process (MRP) [1, ch. 1] xn, θn, n ≥ 0, given on the phase space E = {±x; x > 0}, (2.3) with the sojourn times θ±n (x) = θ±n ∧ x, x ∈ R+ = (0,+∞). The symbols + or − in (2.3) are fixed the renewal moment of one or another renewal processes (2.1). The continuous component x is fixed the remainder time up to the renewal moment other renewal process in (2.2). The embedded Markov chain xn, n ≥ 0, is given by the matrix of the transition probabilities P (x, dy) = [ P+(x− dy) P+(x+ dy) P−(x+ dy) P−(x− dy) ] (2.4) The specific property of the embedded Markov chain with the transi- tion probabilities (2.4) is existence of the stationary distribution with the densities ρ±(x) = ρP∓(x), P∓(x) := 1− P∓(x), ρ = (p+ + p−) −1, p± := ∞∫ 0 P±(x) dx. (2.5) The stationary distribution on merged phase space Ê = {+,−}, is given by ρ± = ρp∓ = λ±/λ, λ± = 1/p±, λ = λ+ + λ−. 3. Storage Impulsive Process The SIP on superposition of two renewal processes (2.2) is defined by the sum of random variables take values in Euclidean space Rd, d ≥ 1 Sn = u+ n∑ k=1 αk(xk), n ≥ 0, S0 = u ∈ Rd. (3.1) The random variables α± k (x), k ≥ 1, x ∈ R+, are given by the distribution functions on (Rd,Rd) G±(A) = P{α± k (x) ∈ A}, A ∈ Rd. V. S. Koroliuk, R. Manca, and G. D’Amico 369 Example 3.1. The risk process (3.1) constructed by the (positive) input random variables α+ k > 0, in the renewal moments of the renewal process ν+(t), t ≥ 0, and by (negative) output random variables −α− k > 0, in the renewal moments of the renewal process ν−(t), t ≥ 0 that is S(t) = u+ ν+(t)∑ k=1 α+ k − ν−(t)∑ k′=1 α− k′ . The SIP (3.1) can be characterized by the generator of the two compo- nents Markov chain Sn, xn, n ≥ 0. (3.2) Lemma 3.1. The two component Markov chain (3.2) is characterized by the generator given on the vector test function φ(u, x) = (φ+(u, x), φ−(u, x)): Lφ(u, x) = PGφ(u, x)− φ(u, x), (3.3) where the operator P is given by the matrix (2.4) and Gφ(u) = [ G+ 0 0 G− ]( φ+(u) φ−(u) ) = (G+φ+(u),G−φ−(u)), G±φ±(u) := ∫ Rd G±(dv)φ±(u+ v) (3.4) Remark 3.1. The generator (3.3) can be represented in scalar form: L±φ(u, x) = ∫ R G±(dv) x∫ 0 P±(dt)φ±(u+ v, x− t) + ∫ R G∓(dv) ∞∫ x P±(dt)φ∓(u+ v, t− x)− φ±(u, x). (3.5) Proof of Lemma 3.1. The conditional expectation Lφ(u, x) = E[φ(Sn+1, xn+1)− φ(u, x)|Sn = u, xn = ±x] is calculated directly: L±φ(u, x) = E[φ(u+ αn+1, xn+1)− φ(u, x)] = G±P±φ±(u, x) +G∓P±φ∓(u, x), 370 Stochastic impulsive processes... where by definition P±φ(x) := x∫ 0 P±(dt)φ(x− t), P±φ(x) := ∞∫ x P±(dt)φ(t− x). Remark 3.2. The generator (3.3) can be transformed as follows: Lφ(u, x) = Qφ(·, x) +P[G− I]φ(u, x), where by definition Q := P− I, is the generator of the embedded Markov chain xn, n ≥ 0, given by the transition probabilities (2.4). The two component Markov chain, given by the generator (3.3), is characterized by the martingale with respect to the standard σ-algebras Fn := σ{(Sk, xk), 0 ≤ k ≤ n} µn+1 = φ(Sn+1, xn+1)− φ(u, x)− n∑ k=1 Lφ(Sk, xk). (3.6) The martingale characterization (3.6) of the SIP (3.1) will be used in asymptotical analysis on increasing time intervals in the series scheme with the small parameter series ε→ 0 (ε > 0). 4. SIP in the average scheme The SIP in the series scheme with the small parameter series ε → 0 (ε > 0), is considered in the following scaling: Sε(t) = u+ ε [t/ε]∑ k=1 αk(xk), t ≥ 0, ε > 0, u ∈ Rd. (4.1) The averaging behavior of the SIP is analyzed by using a martingale characterization (3.6). Lemma 4.1. The normalized SIP (4.1) can be characterized by the mar- tingale µε(t) = φ(Sε(t), xε(t))− φ(Sε(0), xε(0))− ε[t/ε]∫ 0 Lεφ(Sε(h), xε(h)) dh. V. S. Koroliuk, R. Manca, and G. D’Amico 371 The compensating generator is such that Lεφ(u, x) = ε−1Qφ(·, x) +P(x)Gεφ(u, x), (4.2) where P(x) = [ P+(x) P+(x) P−(x) P−(x) ] , P±(x) = 1− P±(x) Gε = [ Gε + 0 0 Gε − ] , Gε ±φ(u) = ε−1 ∫ Rd G±(dv)[φ(u+ εv)− φ(u)]. (4.3) Lemma 4.2. The generator (4.2)–(4.3) admits the following asymptotic representation Lεφ(u, x) = ε−1Qφ(·, x) +P(x)Gφ(u, x) + δεl (x)φ(u, x) (4.4) with the negligible term |δεl (x)φ(u)| → 0, ε→ 0, φ(u) ∈ C2(Rd). The operator is defined as follows: G = [ G+ 0 0 G− ] , G±φ(u) = g±φ ′(u), g± := ∫ Rd vG±(dv). (4.5) Proof of Lemma 4.2 follows from the Taylor expansion applied to the formula (4.3). Proposition 4.1. The finite-dimensional distributions of the SIP (4.1) in the average scheme converges to Sε(t) ⇒ S0(t) = u+ ĝt, t ≥ 0, ε→ 0, (4.6) where the velocity ĝ = ρ+g+ + ρ−g− = (λ+g+ + λ−g−)/λ. Proof. The generator of the normalized SIP (4.4)–(4.5) has the singular perturbation form. The singular perturbed operator Q = P − I is re- ducible invertible [2, ch. 5] with the projector on the null-space given by the stationary distribution (2.5): Πφ(x) = ρ [ ∞∫ 0 P−(x)φ+(x) dx+ ∞∫ 0 P+(x)φ−(x) dx ] . 372 Stochastic impulsive processes... To get the limit operator the algorithm of solving of perturbation problem [2, ch. 5] can be used. The generator (4.4)–(4.5) is considered on the perturbed test function φε(u, x) = φ(u) + εφ1(u, x), φ(u) := (φ(u), φ(u)). Let’s calculate: Lεφε(u, x) = ε−1Qφ(u) + [Qφ1 +P(x)Gφ(u)] + δεl (x)φ(u), (4.7) with the negligible term |δεl (x)φ(u)| → 0, ε→ 0, φ(u) ∈ C2(Rd). The first term in (4.7) equals zero, because φ(u) is a constant for the generator Q. The next term in (4.7) gives us a problem of singular per- turbation Qφ1(u, x) +P(x)Gφ(u) = L0φ(u). (4.8) The limit operator L0 is determined by using solvability condition for the equation (4.8) (see [2, ch. 5]) L0Π = ΠP(x)GΠ. (4.9) Let’s calculate (4.9) taking in mind (4.4)–(4.5): L0 = ρ [ G+ ∞∫ 0 P−(x)P+(x) dx+G− ∞∫ 0 P−(x)P+(x) dx +G− ∞∫ 0 P+(x)P−(x) dx+G+ ∞∫ 0 P+(x)P−(x) dx ] = ρ[Gg+p− +G−p+] = ρ+G+ + ρ−G− = (λ+G+ + λ−G−)/λ = ρ+G+ + ρ−G− = Ĝ. That is the limit generator L0φ(u) = ĝφ′(u), defines the deterministic drift in (4.6) S0(t) = u+ ĝt, t ≥ 0. V. S. Koroliuk, R. Manca, and G. D’Amico 373 5. SIP in the diffusion approximation scheme The SIP in the series scheme with the small parameter series ε → 0 (ε > 0), is considered in the following scaling: Sε(t) = u+ ε [t/ε2]∑ k=1 αk(xk), t ≥ 0, u ∈ Rd, (5.1) under the additional Balance Condition (BC): ĝ = ρ+g+ + ρ−g− = 0. (5.2) Proposition 5.1. The SIP (5.1) under BC (5.2) converges weakly Sε(t) ⇒ wσ(t), ε→ 0. The limit Brownian motion wσ(t), t ≥ 0, is defined by the variance σ2 = σ2b + σ2g , (5.3) σ2b = ρ+B+ + ρ−B−, B± = ∫ Rd v2G±(dv), (5.4) σ2g = 2ΠP(x)GR0P(x)GΠ. (5.5) The potential operator R0 is defined by a solution of the equation QR0 = R0Q = Π− I. Remark 5.1. The component of variance (5.5) can be calculated by us- ing the reducible inverse operator R0 to the generator Q of the embedded Markov chain. That is an open problem (see [3]). Proof of Proposition 5.1. As in previous section 3 the martingale charac- terization of the SIP (5.1) is a starting point in asymptotic analysis. Lemma 5.1. The normalized SIP (5.1) can be characterized by the mar- tingale µε(t) = φ(Sε(t), xε(t))− φ(u, x)− ε2[t/ε2]∫ 0 Lεφ(Sε(h), xε(h)) dh. (5.6) The generator L of the two component Markov chain Sεn = Sε(τ εn), xεn = xε(τ εn), n ≥ 0, 374 Stochastic impulsive processes... is represented by Lεφ(u, x) = [ε−2Q+P(x)Gε]φ(u, x), (5.7) Gε = [ Gε + 0 0 Gε − ] , Gε ±φ(u) = ∫ Rd G±(dv)[φ(u+ εv)− φ(u)]. (5.8) Lemma 5.2. The generator (5.7)–(5.8) admit the asymptotic represen- tation on the smooth enough test function φ(u) ∈ C3(Rd) : Lεφ(u, x) = ε−2Qφ(·, x) + ε−1P(x)Gφ(u, x) +P(x)Bφ(u, x) + δεe(x)φ(u, x), (5.9) with the negligible term |δεe(x)φ(u, ·)| → 0, ε→ 0, φ(u, ·) ∈ C3(Rd). Here B = [ B+ 0 0 B− ] , B±φ(u) := 1 2 B±φ ′′(u). Now a solution of singular perturbation problem [2, ch. 5, Proposi- tion 5.2] is used on the perturbed test function φε(u, x) = φ(u) + εφ1(u, x) + ε2φ2(u, x). (5.10) Lemma 5.3. The generator (5.9) on the perturbed test function (5.10) admit the asymptotic representation Lεφε(u, x) = L0φ(u) + δεe(x)φ(u), with the negligible term |δεe(x)φ(u)| → 0, ε→ 0, φ(u) ∈ C3(Rd). The limit generator L0φ(u) = 1 2 σ2φ′′(u), is the generator of the Brownian motion wσ(t), t ≥ 0, with the variance (5.3)–(5.5). Proof. Let’s calculate Lεφε(u, x) = ε−2Qφ(u) + ε−1[Qφ1(u, x) +P(x)Gφ(u)] + [ Qφ2 +P(x)Gφ1(u, x) + 1 2 B(x)φ(u) ] + δεe(x)φ(u), V. S. Koroliuk, R. Manca, and G. D’Amico 375 with the negligible term |δεe(x)φ(u)| → 0, ε→ 0, φ(u) ∈ C3(Rd). The BC (5.2) can be represented as follows: ΠP(x)GΠ = 0. Hence there exists solution of the equation Qφ1(u, x) +P(x)Gφ(u) = 0, that is φ1(u, x) = R0G(x)φ(u). Now the next equation Qφ2 + [ P(x)GR0P(x)G+ 1 2 B(x) ] φ(u) = L0φ(u), gives the limit generator L0φ(u) = 1 2 σ2φ′′(u), by using solvability condition: L0φ(u) = [ ΠP(x)GR0P(x)GΠ+ 1 2 ΠB(x)Π ] φ(u) = 1 2 σ2φ′′(u). 6. Large deviation problem The SIP in the scheme of asymptotically small diffusion is considered in the following scaling: Sε(t) = u+ ε2 [t/ε3]∑ k=1 αk(xk), t ≥ 0, u ∈ Rd (6.1) under additional BC (5.2). Proposition 6.1. The large deviation problem for SIP (6.1) under the BC (5.2) is realized by the exponential generator of asymptotically small diffusion Hφ(u) = 1 2 σ2[φ′(u)]2. (6.2) The variance is defined in (5.3)–(5.5). Proof. The SIP (6.1) can be characterized by the exponential martingale [4, Part I]. 376 Stochastic impulsive processes... Lemma 6.1. The two component Markov process Sε(t), xε(t) := x[t/ε3], t ≥ 0, is characterized by the exponential martingale exp{φ(Sε(t), xε(t))/ε− φ(Sε(0), xε(0))/ε − ε−1 ε3[t/ε3]∫ 0 Hε(Sε(h), xε(h)) dh} = µε(t) is Ft(S, x)-martingale. The exponential generator Hεφ(u, x) = ε−2 ln[e−φ(u,x)/εLεeφ(u,x)/ε + 1]. (6.3) The compensative generator Lεφ(u, x) = [Q+P(x)Gε]φ(u, x), P(x) = [ P+(x) P+(x) P−(x) P−(x) ] , Gε = [ Gε +(x) 0 0 Gε −(x) ] , (6.4) Gε ±φ(u) = ∫ Rd G±(dv)[φ(u+ ε2v)− φ(u)]. Note that the generators Gε ± admit the asymptotic representation Gε ±φ(u) = ε2g±φ ′(u) + ε2δεgφ(u), (6.5) with the negligible form |δεgφ| → 0, ε→ 0, φ ∈ C2(Rd). Note the exponential generator (6.3)–(6.5) is considered on the per- turbing test function φε(u, x) = φ(u) + ε ln[1 + εφ1(u, x) + φ2(u, x)]. (6.6) Lemma 6.2. The compensating operator (6.4) on the perturbing test function (6.6) admits the asymptotic representation Hε Lφ ε(u, x) := e−φ ε/εLεeφ ε/ε = ε[Qφ1 + P (x)Gφ] + ε2[Qφ2 − φ1Qφ1 +Hb(x)φ] + δεl φ, (6.7) with the negligible term |δεl φ| → 0, ε→ 0, φ ∈ C3(Rd). V. S. Koroliuk, R. Manca, and G. D’Amico 377 Proof. The asymptotic representation (6.7) is the consequence of the following asymptotic relations: e−φ ε/εQeφ ε/ε = e−φ/ε[1 + εφ1 + ε2φ2] −1Q[1 + εφ1 + ε2φ2]e φ/ε = εQφ1 + ε2[Qφ2 − φ1Qφ1] + ε2δεqφ, e−φ ε/εP(x)Geφ ε/ε = e−φ/ε[1+εφ1+ε 2φ2] −1P(x)G[1+εφ1+ε 2φ2]e φ/ε = εP(x)Gφ+ ε2[P(x)Gφ1 + P (x)Hb(x)φ] + δεgφ, with the negligible terms δεqφ and δεgφ. Now the singular perturbation problems are used for the equations Qφ1 +P(x)Gφ = 0, Qφ2 − φ1Qφ1 +Hb(x)φ = Hφ. (6.8) The first equation (6.8) under the BC (5.2) has the solution φ1 = R0P(x)Gφ, Qφ1 = −P(x)Gφ. Hence −φ1Qφ1 = Hg(x)φ(u) := 1 2 σ2g(x)[φ ′(u)]2. (6.9) The second equation (6.8) with (6.9) is transformed to Qφ2 + [Hg(x) +Hb(x)]φ(u) = Hφ(u). (6.10) The right part side in (6.10) is determined by the solvability condition: H = Π[Hg(x) +Hb(x)]Π. (6.11) Corollary 6.1. The exponential generator (6.3) admits the asymptotic representation Hεφε(u, x) = Hφ(u) + δεhφ(u), (6.12) with the negligible term |δεhφ| → 0, ε→ 0, φ ∈ C3(Rd). The proof of Proposition 6.1 is finished as follows. We consider (see (6.7)) Hεφε(u, x) = ε−2 ln[1 +Hε Lφ ε(u, x)] = ε−2 ln[1 + ε2Hφ(u) + ε2δεl φ(u)] = Hφ(u) + δεhφ, with the negligible term δεhφ, φ ∈ C3(Rd). The limit exponential gener- ator H is calculated in (6.11) using the representation (see (5.3)–(5.5)) Hg(x)φ(u) = 1 2 σ2g(x)[φ ′(u)]2, σ2g = 2ΠGR0P(x)G1, Hb(x)φ(u) = 1 2 σ2b (x)[φ ′(u)]2, σ2b = ΠP(x)B1. 378 Stochastic impulsive processes... 7. Conclusion This paper contains the three simplified approximation schemes for the SIP given by a sum (1.1) of random variables defined on the Markov renewal process: average (Section 4), diffusion approximation (Section 5) and the scheme of asymptotically small diffusion (Section 6). The considered simplification schemes may be effectively used in ap- plications, partially in financial mathematics [7]. The initial SIP (1.1) or (3.1), defined by two distribution functions G±(A) of jumps and two dis- tribution functions P±(t), given the renewal processes, may be simplified in the average scheme by the deterministic drift process (4.6), given by the four constants ρ± and g± which are the first moments of the renewal times (ρ±) and the average jumps (g±). The fluctuations of the SIP on increasing time intervals in diffusion approximation scheme (Section 4) is described by the limit Brownian motion (Proposition 5.1) given by the variance (5.3)–(5.5) defined by the first two moments of jumps (g± and B±). The scheme of asymptotically small diffusion (Section 6) represents the exponential generator of large deviations used in the analysis of asymptotically small probabilities (see [4]). The initial definition of the SIP given in Section 2 may be easily interpreted as the logistic problem. The positive jumps α+ k (x), k ≥ 0 are interpreted as a profits and the negative jumps α− k (x), k ≥ 0 are the losses. The SIP on increasing time intervals may be approximated by the well-known emery drift (Section 3) and in addition by the Brownian mo- tion process of fluctuation. References [1] V. S. Korolyuk, V. V. Korolyuk, Stochastic Models of Systems, Kluwer Academic Publisher, 1999. [2] V. S. Koroliuk, and N. Limnios, Stochastic Systems in Merging Phase Space, Singapore: VSP, 2005. [3] V. S. Koroliuk, R. Manca, G. D’Amico, Storage impulsive processes in merging phase space // Journal of Mathematical Sciences, 196 (2013), No. 5, 644–651. [4] J. Feng and T. G. Kurtz, Large Deviation for Stochastic Processes, AMS, 131, RI, 2006. [5] W. Feller, Introduction to Probability Theory and its Applications, Vol. 2, NY: Willey, 1966. [6] V. S. Koroliuk, Random evolutions with locally independent increments on in- creasing time intervals // Journal of Mathematical Sciences, 179 (2011), No. 2, 273–289. [7] A. N. Shyryaev, Essentials of Stochastic Finance: Facts, Models, Theory, World Scientific, 1999. V. S. Koroliuk, R. Manca, and G. D’Amico 379 Contact information Vladimir S. Koroliuk Institute of Mathematics of the NASU, 3, Tereshchenkivska Str., Kyiv, 01601, Ukraine E-Mail: vskorol@yahoo.com Raimondo Manca Faculty of Economics, Sapienza University of Rome Via del Castro Laurenziano, 9 00161 - Rome, Italy E-Mail: raimondo.manca@uniroma1.it Guglielmo D’Amico Department of Pharmacy, University “G. d’Annunzio” of Chieti-Pescara, Via dei Vestini - 66013 Chieti, Italy E-Mail: g.damico@unich.it