On stochastic differential inclusions with unbounded right sides

The paper deals with one-dimensional homogeneous stochastic differential inclusions without drift with a Borel measurable right side. Using a new method of explicit solutions, the necessary and sufficient conditions for the existence of weak solutions of the inclusions with locally unbounded right s...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Datum:2006
1. Verfasser: Lepeyev, A.N.
Format: Artikel
Sprache:English
Veröffentlicht: Інститут математики НАН України 2006
Online Zugang:https://nasplib.isofts.kiev.ua/handle/123456789/4445
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Назва журналу:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Zitieren:On stochastic differential inclusions with unbounded right sides / A.N. Lepeyev // Theory of Stochastic Processes. — 2006. — Т. 12 (28), № 1-2. — С.94–105. — Бібліогр.: 23 назв.— англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
id nasplib_isofts_kiev_ua-123456789-4445
record_format dspace
spelling Lepeyev, A.N.
2009-11-10T14:51:45Z
2009-11-10T14:51:45Z
2006
On stochastic differential inclusions with unbounded right sides / A.N. Lepeyev // Theory of Stochastic Processes. — 2006. — Т. 12 (28), № 1-2. — С.94–105. — Бібліогр.: 23 назв.— англ.
0321-3900
https://nasplib.isofts.kiev.ua/handle/123456789/4445
519.21
The paper deals with one-dimensional homogeneous stochastic differential inclusions without drift with a Borel measurable right side. Using a new method of explicit solutions, the necessary and sufficient conditions for the existence of weak solutions of the inclusions with locally unbounded right sides are given.
en
Інститут математики НАН України
On stochastic differential inclusions with unbounded right sides
Article
published earlier
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
collection DSpace DC
title On stochastic differential inclusions with unbounded right sides
spellingShingle On stochastic differential inclusions with unbounded right sides
Lepeyev, A.N.
title_short On stochastic differential inclusions with unbounded right sides
title_full On stochastic differential inclusions with unbounded right sides
title_fullStr On stochastic differential inclusions with unbounded right sides
title_full_unstemmed On stochastic differential inclusions with unbounded right sides
title_sort on stochastic differential inclusions with unbounded right sides
author Lepeyev, A.N.
author_facet Lepeyev, A.N.
publishDate 2006
language English
publisher Інститут математики НАН України
format Article
description The paper deals with one-dimensional homogeneous stochastic differential inclusions without drift with a Borel measurable right side. Using a new method of explicit solutions, the necessary and sufficient conditions for the existence of weak solutions of the inclusions with locally unbounded right sides are given.
issn 0321-3900
url https://nasplib.isofts.kiev.ua/handle/123456789/4445
citation_txt On stochastic differential inclusions with unbounded right sides / A.N. Lepeyev // Theory of Stochastic Processes. — 2006. — Т. 12 (28), № 1-2. — С.94–105. — Бібліогр.: 23 назв.— англ.
work_keys_str_mv AT lepeyevan onstochasticdifferentialinclusionswithunboundedrightsides
first_indexed 2025-11-24T05:32:34Z
last_indexed 2025-11-24T05:32:34Z
_version_ 1850441536975667200
fulltext Theory of Stochastic Processes Vol. 12 (28), no. 1–2, 2006, pp. 94–105 UDC 519.21 ANDREI N. LEPEYEV ON STOCHASTIC DIFFERENTIAL INCLUSIONS WITH UNBOUNDED RIGHT SIDES The paper deals with one-dimensional homogeneous stochastic differential inclusions without drift with a Borel measurable right side. Using a new method of explicit solutions, the necessary and sufficient conditions for the existence of weak solutions of the inclusions with locally unbounded right sides are given. Introduction The following driftless homogeneous stochastic differential inclusion (SDI) is investi- gated: dXt ∈ B(Xt) dWt, t ≥ 0, (1) where B : R → comp(R) is a multi-valued Borel measurable mapping, comp(R) - the set of all non-empty compact subsets of R with the Hausdorff metric which is defined ∀A, B ∈ comp(R) by ρ(A, B) = max(β(A, B), β(B, A)), where β(A, B) = supx∈A(infy∈B |x − y|) is the excess of A over B, and W is a one-dimensional Wiener process. The theory of differential inclusions originated, basically, from two problems. The first one is the generalization of the equations without solutions to inclusions, which have solutions, and treatment of the solutions of the equations in the sense of the solutions of the corresponding inclusions. This theory was systematized in [13] by A.F. Filippov, who introduced the main properties, which the solutions of the inclusions should satisfy to. The main property is the preservation of the solution set for the equations having solutions in the classic context. The second problem arose in the models described by the differential equations with multi-valued right sides (cf. [3]). The solutions of the equations appeared to be well-defined in terms of the corresponding inclusions. The investigation of multi-valued random processes (cf. [14]) motivated the creation of a technique for stochastic differential equations with multi-valued right sides, that led to the use of the ideas developed in the theory of differential inclusions in stochastic models (cf. [6]). Stochastic differential inclusions as a separate theory were introduced by P. Kree in [19]. N.U. Ahmed, E. Cepa, G. Da Prato, H. Frankovska, M. Kisielewicz, A.A. Levakov, M. Michta, J. Motyl, R. Petterson who considered the stochastic differential inclusions with different right sides and gave sufficient conditions for the existence of weak solutions of SDIs (cf. [1,5,7,17,18,20,23]). The known results contain the existence conditions for weak solutions of stochastic differential inclusions, whose right sides are required to be bounded at least locally. This paper deals with the existence conditions for weak solutions of SDIs with lo- cally unbounded right sides. The main result is the necessary and sufficient existence conditions for weak solutions of SDIs for every initial distribution. 2000 AMS Mathematics Subject Classification. Primary 34A60, 60G44, 60H10, 60J65, 60H99. Key words and phrases. Stochastic differential equations, stochastic differential inclusions, measur- able coefficients. 94 ON STOCHASTIC DIFFERENTIAL INCLUSIONS 95 Preliminaries Solutions of stochastic differential equations with unbounded coefficients. The main idea of the paper is to express the solutions of stochastic differential inclu- sions in terms of the theory of stochastic differential equations (SDEs) and to use the developed technique of the theory. That is why, at first we will recall several important facts from the SDEs theory. Let (Ω,F ,P) be a complete probability space with filtration F = (Ft)t≥0. As usually, assume, that the filtration F satisfies the natural conditions, e.c. it is right-continuous and F0 contains all P - zero subsets of F . For a stochastic process (Xt)t≥0 defined on (Ω,F ,P), we will write (X, F), if X is F-adapted. SDI (1) is, actually, the generalization of SDE of the following type: dXt = b(Xt) dWt, t ≥ 0, (2) where b : R → R - Borel measurable diffusion coefficient, W - one-dimensional Wiener process. It is known that if the process X is a stochastic integral driven by a Wiener process (for example, it is a solution of Eq. (2), then it has continuous modification (if X is a solution of Eq. (2), then its continuous modification is also a solution of (2)). Taking that into consideration, we will deal with the trajectory space C(R+, R) equipped with the Borel σ-algebra of cylindric sets B(C(R+, R)). A stochastic process (X, F), defined on the probability space (Ω,F ,P) with a filtration F = (Ft)t≥0 and trajectories in C(R+, R) is called a weak solution of SDE (2) with initial distribution X0, if there exists a Wiener process (W, F) with W 0 = 0 such that P- a.s. for all t ≥ 0 Xt = X0 + ∫ t 0 b(Xs) dW s. It should be emphasized that the initial distribution X0 in the solution definition can be an arbitrary random value in the space, where the solution exists. The existence con- ditions of the paper will guarantee the existence of solutions for every initial distribution, that means: for every probabilistic measure P̄ on (R,B(R)), there exists a solution X with initial distribution X0 such that P({X0 ∈ C}) = P̄ (C), ∀C ∈ B(R). The idea of the proof of our main result is based on the Engelbert—Schmidt theorem (cf. [12], Theorem 4.17) which states that the weak solutions of Eq. (2) exist for every initial distribution if and only if the following holds: Mb ⊆ Nb, (3) where the sets in (3) are defined for every measurable function v : R → R by Mv = { x ∈ R ∣∣∣∣∣ ∫ U(x) v−2(y) dy = ∞, ∀U(x) - open neighborhood of the point x } (4) Nv = {x ∈ R|v(x) = 0} (5) The solutions of SDE (2) are always non-exploding (cf. [12], Proposition 4.11). Random time change. Now we will focus attention on the method of random time change. Assume that a filtrated probability space is given. Any increasing right-continuous family A = (At) of P-a.s. finite F-stopping times is called F-time change. Having defined the right-inverse process Tt = inf{s ≥ 0 : As > t}, one can give the main property of a random time 96 ANDREI N. LEPEYEV change ATt = t P-a.s. ∀t ≥ 0. Additionally, if the process A is strictly increasing, then the process T is continuous and TAt = t P-a.s. ∀t ≥ 0. Now we will give some generalization of the random time change that was introduced by H.J. Engelbert and W. Schmidt (cf. [10]). Let A = (At) be a right-continuous in- creasing process with values in [0, +∞] and A∞ = limt→∞ At. Having defined the inverse increasing process Tt = inf{s ≥ 0 : As > t}, we can pathwisely construct the Lebesgue integral ∫ t 0 ZsdAs = ∫ t∧T∞ 0 ZsdAs, ∀t ≥ 0, where Z is an arbitrary measurable process. Lemma 1 ([10], Lemma 1.6). For every non-negative measurable process Z, the fol- lowing holds: ∫ t 0 ZsdTs = ∫ Tt 0 ZAsds, ∀t ≥ 0. Different kinds of random time changes proved to be helpful in the solution of various problems of stochastic analysis. Let us consider a time change which will be used in this article. Let us be given with an arbitrary Borel measurable function v : R → R and a Wiener process (W̃ , F̃) on the probability space (Ω,F ,P) with arbitrary initial distribution. We can define the increasing process T v t = ∫ t+ 0 v−2(W̃s)ds, ∀t ≥ 0 (6) and its right-inverse process Av t = inf{s ≥ 0 : T v s > t}. (7) Additionally, let U(Mv) be the first entry time of W̃ in Mv, U(Mv) = inf{s ≥ 0 : W̃s ∈ Mv}. (8) The following properties will be used in the proof of the main results of the paper. Lemma 2 ([11],Lemma 1,[9],Theorem 3). Under the definitions above, P-a.s.∫ t 0 v−2(W̃s)ds < +∞, ∀t < U(Mv); ∫ U(Mv)+ 0 v−2(W̃s)ds = +∞; A∞ = U(Mv); and T v ∞ = +∞. Right sides of stochastic differential inclusions. In this section, we will closely investigate the right sides of SDI (1). The following notations of measure will be used in the whole paper: l is the Lebesgue measure on R, and l+ is a Lebesgue measure on R+ = [0,∞). Let X and Y be arbitrary sets. The function v : X → Y is called the selection of a multi-valued mapping B of the set X into some set of subsets from Y , if v(x) ∈ B(x), ∀x ∈ X (cf. [4]). The definition can be extended. Let (X, ν) be a space with measure ν, and let Y be an arbitrary set. Then the function u : X → Y is called ν-a.e. selection of the mapping B of the set X into some set of subsets from Y , if u(x) ∈ B(x), for ν-almost all x ∈ X (cf. [22]). Selections of the multi-valued mapping B(x) can be interpreted in two ways 1) Explicit selections of the mapping B : R → comp(R) such functions v : R → R that v(x) ∈ B(x). 2) Composition selections B(X) : R+ × Ω → comp(R) (l+ × P-a.e. composition selections) - such functions u : R+ × Ω → R that u(t, ω) ∈ B(X(t, ω)). The measurability of the considered multi-valued mappings is meant in the sense of the measurability of functions with values in the space (comp(R), ρ) with the Hausdorff ON STOCHASTIC DIFFERENTIAL INCLUSIONS 97 metric. The corresponding σ-algebra of Borel subsets B(comp(R)) is generated by open sets of the form [A]ε = {B ∈ comp(R)|ρ(A, B) < ε}, ∀A ∈ comp(R), ε > 0. Let us denote op(R) and cl(R) - all open and closed sets from R generated by the Euclidian metric. Proposition 1 ([4], Theorem 3.2.). The multi-valued mapping B : R → comp(R) is Borel measurable if and only if B−(U) = {x ∈ R|B(x) ∩ U = ∅} ∈ B(R) holds either for all U ∈ op(R) or for all U ∈ cl(R). Let us define the internal characteristic selection of the mapping B bint(x) = { β(0, B(x)), β(0, B(x)) ∈ B(x); −β(0, B(x)), β(0, B(x)) /∈ B(x); and external characteristic selection of the mapping B bext(x) = { β(B(x), 0), β(B(x), 0) ∈ B(x); −β(B(x), 0), β(B(x), 0) /∈ B(x). Proposition 2. The internal and external characteristic selections of a Borel measurable multi-valued mapping are Borel measurable functions. Proof. Let U(x, r) ⊂ R denote an open ball around x ∈ R with radius r. For every r > 0, {x ∈ R|β(0, B(x)) < r} = {x ∈ R|B(x) ∩ U(0, r) = ∅} = B−(U(0, r)) holds, and {x ∈ R|β(B(x), 0) < r} = R \ {x ∈ R|B(x) ∩ (R \ U(0, r)) = ∅} = R \ (B−(R \ U(0, r)). The functions β(0, B(x)) and β(B(x), 0) are Borel measurable from Proposition 1 and the fact that the mapping B is Borel measurable. Hence, the statement of Proposition 2 holds due to the fact that the internal and external characteristic selections are the compositions of Borel measurable functions. Remark 1. a) The functions bint and bext are really selections of the mapping B due to ∀x ∈ R : bint(x) ∈ B(x), bext(x) ∈ B(x). b) For every explicit selection v of the mapping B, the following holds: |bint(x)| ≤ |v(x)| ≤ |bext(x)|, ∀x ∈ R. Additionally, let us define the optimal characteristic selection of the mapping B: bopt(x) = { 0, 0 ∈ B(x); bext(x), 0 /∈ B(x); Remark 2. a) The function bopt is a Borel measurable selection of the mapping B, because ∀x ∈ R : bopt(x) ∈ B(x) and it is a composition of Borel measurable functions. b) |bint(x)| ≤ |bopt(x)| ≤ |bext(x)|, ∀x ∈ R. It is worthwhile to provide a few facts on the integrability of multi-valued mappings. The Lebesgue integral of multi-valued mappings was described in detail by R.J. Auman in [3]. Let B be a multi-valued mapping of R into comp(R) and IB be the set of all Lebesgue integrable selections of B. Then the integral of B over interval [0, t] ⊂ R, t > 0 is the set of integrals of all integrable selections∫ t 0 B(s)ds = { ∫ t 0 u(s)ds : u ∈ IB}. 98 ANDREI N. LEPEYEV It is evident that, for the complete description of the integral, one can deal only with Borel measurable selections. A stochastic integral driven by the Wiener process of multi- valued mappings can be defined in the same way. Namely, let a multi-valued process B : R+ × Ω → comp(R) be defined on the probability space (Ω,F ,P) with filtration F, and let the process be measurable and F-adapted. Then we can define the set IB of all (B(R+)×F)-measurable, F-adapted processes u such that u(t, w) is l+×P-a.e. selection of the mapping B(t, w), and E( ∫ +∞ 0 |u(s)|2ds) < ∞. The stochastic integral driven by a Wiener process over the interval [0, t] ⊂ R, t > 0 of the multi-valued mapping B can be defined by ∫ t 0 B(s)dWs = { ∫ t 0 u(s)dWs : u ∈ IB}. Taking into consideration the fact that the right side of the stochastic differential inclusion (1) is the composition B(X) : R+ × Ω → R and that the selections can be interpreted in two ways, we have a right, for fixed measurable F-adapted process X , to consider IB - the set of all explicit Borel measurable selections v of the mapping B : R → comp(R) such that E( ∫ +∞ 0 |v(Xs)|2ds) < ∞ and introduce the set { ∫ t 0 v(Xs)dWs : v ∈ IB}. Using the excess β between sets from compR, we can define four sets which will be used in the statements of the theorems given below: MB = { x ∈ R ∣∣∣∣∣ ∫ U(x) β(0, B(y))−2 dy = ∞, ∀U(x) - open neighborhood of point x } , MB = { x ∈ R ∣∣∣∣∣ ∫ U(x) β(B(y), 0)−2 dy = ∞, ∀U(x) - open neighborhood of point x } , NB = {x ∈ R|{0} ∈ B(x)}, NB = {x ∈ R|B(x) = {0}}. Remark 3. a) The sets NB andNB are Borel ones, that is easily seen from the equalities {x ∈ R|{0} ∈ B(x)} = {x ∈ R|bint = 0}, {x ∈ R|B(x) = {0}} = {x ∈ R|bext = 0} and Proposition 2. b) The following relations are valid: MB ⊆MB, NB ⊆NB. c) For the sets Mv and Nv defined in (4) and (5) corresponding to the selections bint, bext, and bopt, we can deduce Mbint =MB, Mbext =MB, Nbint =NB, Nbext =NB, Nbopt = Nbint , Mbext =MB ∪ (MB ∩NB) ⊆MB, MB \NB = Mbopt \ Nbopt , which, particularly, implies that the setsMB andMB are closed. d) For every explicit selection v of the mapping B, NB ⊆ Nv ⊆NB, MB ⊆ Mv ⊆MB, (Mv \ Nv) ⊇ (MB \NB). ON STOCHASTIC DIFFERENTIAL INCLUSIONS 99 Weak solutions of stochastic differential inclusions Definitions and properties of weak solutions. Let us recall (cf. [1]) that the stochastic process (X, F) defined on the probability space (Ω,F ,P) with filtration F = (Ft)t≥0 and trajectories in C(R+, R) is called a weak solution of SDI (1) with initial distribution X0, if there exist a Wiener process (W, F) with W 0 = 0 and a measurable F-adapted l+ × P-a.e. selection u : R+ × Ω → R of the composition B(X) : R+ × Ω → comp(R) such that the following holds P- a.s. for all t ≥ 0: Xt = X0 + ∫ t 0 u(s) dW s. (9) The following definition introduces a subset of the set of SDI weak solutions. Definition 1. A stochastic process (X, F) defined on the probability space (Ω,F ,P) with filtration F = (Ft)t≥0 and trajectories in C(R+, R) is called an explicit weak solution of SDI (1) with initial distribution X0, if there exist a Wiener process (W, F) with W 0 = 0 and a Borel measurable explicit selection v : R → R of the mapping B : R → comp(R) such that the following holds P- a.s. for all t ≥ 0: Xt = X0 + ∫ t 0 v(Xs) dW s. (10) Proposition 3. An explicit weak solution of SDI (1) is a weak solution of SDI (1). Proof. Let the stochastic process (X, F) with trajectories in C(R+, R) on the probability space (Ω,F ,P) with filtration F = (Ft)t≥0 and the Wiener process (W, F) be an explicit weak solution of SDI (1) with respect to some explicit selection v. Then the same process X is a weak solution on the same probability space with the same filtration and Wiener process. Namely, we can take the selection u = v(X) which is a (B(R+) × F)- measurable F-adapted process due to the facts that the function v is Borel measurable, and X is (B(R+)×F)-measurable and F-adapted. For the selection, the following holds: u(t, ω) = v(X(t, ω)) ∈ B(X(t, ω)), ∀t ≥ 0, ω ∈ Ω. The next statement is evident, but it should be emphasized due to its importance. Proposition 4. Every explicit weak solution of the stochastic differential inclusion (1) with respect to some selection v is a weak solution of the stochastic differential equation (2) with diffusion coefficient v. On the other hand, every weak solution of the stochastic differential equation (2) with some diffusion coefficient v is an explicit weak solution of the stochastic differential inclusion (1) with the right side that possesses the explicit selection v. Proof. The proof is easily accomplished from the definitions of an SDI explicit weak solution and an SDE weak solution. The stochastic process (X, F) is trivial if Xt = X0, ∀t ≥ 0,P-a.s. Otherwise, the process is non-trivial. Proposition 5. The stochastic differential inclusion (1) has trivial weak solutions with initial distribution X0 if and only if 0 ∈ B(X(0, ω)) P-a.s., e.c. bint(X0) = 0 P-a.s.. Proof. The sufficiency is easily accomplished if we take the selection u(t) ≡ 0. To prove the necessity, let us take an arbitrary trivial weak solution X with respect to some selection u with initial distribution X0. Then u(t, ω) = 0 l+ × P-a.e.; hence, the definition of a SDI weak solution implies 0 ∈ B(X(t, ω)) l+ × P-a.e.. But B(X(t, ω)) = B(X(0, ω)), ∀t ≥ 0,P-a.s., therefore, 0 ∈ B(X(0, ω)) P-a.s. or, in other words, bint(X0) = 0 P-a.s.. 100 ANDREI N. LEPEYEV Necessary and sufficient conditions for the existence of weak solutions of stochastic differential inclusions. Theorem 1. The stochastic differential inclusion (1) has weak and explicit weak solu- tions for every initial distribution if and only if MB ⊆NB. (11) Proof. We will start with the necessity. Our objective is to prove that condition (11) implies the existence of weak solutions for every initial distribution with respect to the optimal characteristic selection. For the simplicity of notations, let us denote the selection v ≡ bopt. Point a) of Remark 2 shows that the selection v satisfies the conditions of the definition of an explicit weak solution of SDI (1). On the other hand, since the selection v is single- valued, we can consider a stochastic differential equation of type (2) with the diffusion coefficient b ≡ v. From Proposition 4, all the weak solutions of SDE (2) will be the explicit weak solutions of SDI (1). We will use the method of random time change. Let us take an arbitrary Wiener process (W̃ , F̃) on some probability space (Ω,F ,P) with filtration F̃ = (F̃t)t≥0 and an arbitrary probability measure P̄ on R as the initial distribution of the Wiener process. Using W̃ , we can introduce the processes Av and T v and the set U(Mv), as defined in (6), (7), and (8). The process Av = (Av t )t≥0 is a continuous increasing family of F̃-stopping times. From Lemma 2, the process Av t is finite P-a.s. for all t ≥ 0. We can define a process Xt = W̃Av t , ∀t ≥ 0 and a filtration F = (Ft)t≥0 = (F̃Av t )t≥0 and can conclude that (X, F) is a continuous local martingale with square characteristic 〈X〉 = Av (cf. [16]). Let us give the explicit form of Av. The set (Nv \Mv) has the zero Lebesgue measure (from the definition of the set Mv). Hence, W̃ ∈ Mv ∪ (R\Nv) l+×P-a.e. Using Lemma 2, we can conclude that v(W̃s) = 0 for all s < Av ∞ l+ × P-a.e.. Lemma 2 also implies W̃Av∞ ∈ Mv if {Av ∞ < +∞}. Due to condition (11) and point c) of Remark 3, we have Mv ⊆ Nv, and if {Av ∞ < +∞}, then P-a.s. v2(W̃Av∞) = 0. (12) Using the property and the definition of the process T v, we obtain that P-a.s. Av t = ∫ Av t 0 v2(W̃s)dT v s = ∫ T v Av t 0 v2(Xs)ds, where the last equality follows from Lemma 1. Hence, P-a.s. Av t = ∫ t 0 v2(Xs)ds if Av t < Av∞ and ∫ ∞ 0 v2(Xs)ds if Av t = Av∞. On the other hand, if Av t = Av∞, then Xs = WAv∞ for all s ≥ t, and we can conclude from (12) that v2(Xs) = 0 for all s ≥ t, which implies P-a.s. Av t = ∫ t 0 v2(Xs)ds, ∀t ≥ 0. From the Doob theorem (cf. [15], Theorem II.7.1′), there exists a Wiener process (W̄ , F) (on a possibly extended probability space) such that (X, F) is a weak solution of SDE (2) with initial distribution P̄ . This process X is an explicit weak solution of SDI (1) from Proposition 4. From Proposition 3, this process is a weak solution of SDI (1). ON STOCHASTIC DIFFERENTIAL INCLUSIONS 101 Let (X, F) be an arbitrary weak solution of inclusion (1) with respect to some selection u, e.c. (9) holds. Then a square variation 〈X〉 = Au has the form Au t = ∫ t 0 u2(s)ds, ∀t ≥ 0 P − a.s. Let us define a process T u t = inf{s ≥ 0 : Au s > t}, that is right-inverse to the process Au, and let us introduce a process (Wu t )t≥0 = (XT u t )t≥0 with filtration F u = (Fu t )t≥0 = (FT u t )t≥0, where X∞ = limt→∞ Xt if {Au∞ < +∞}. We can conclude that (Wu, Fu) is a continuous local martingale such that 〈Wu〉t = t ∧ Au∞ and the initial distribution W v 0 = X0. Hence, (Wu, Fu) is the Wiener process stopped in Au ∞ (cf. the proof of Lemma 2 in [8]). Using Lemma 1, we get∫ t∧Au ∞ 0 b−2 ext(W u s )ds = ∫ Au T u t 0 b−2 ext(W u s )ds = ∫ T u t 0 b−2 ext(Xs)dAu s = ∫ T u t 0 b−2 ext(Xs)u2(s)ds ≤ T u t . (13) The last inequality follows from the fact that since u(t, ω) ∈ B(X(t, ω)) for l+×P-almost all (t, ω) ∈ R+ × Ω, bext(X(t, ω)) ≥ u(t, ω) for l+ × P-almost all (t, ω) ∈ R+ × Ω. Let us take any point x0 /∈ NB and consider a weak solution X of SDI (1) with initial distribution X0 = x0 that is non-trivial from Proposition 5. The non-triviality of the solution implies P({Au∞ > 0}) > 0. Therefore, there exists t > 0 such that P({Au ∞ > t}) > 0. Taking into consideration T u t < +∞ if {Au ∞ > t} and inequality (13), we can conclude P({ ∫ t 0 b−2 ext(W u s )ds < +∞, Au ∞ > t}) > 0. (14) Hence, from Theorem 1 in [9], there exists an open neighborhood G of the point x0 such that the function b−2 ext is integrable over G, and x0 /∈MB. From the proof of Theorem 1, it is seen that condition (11) is valid if weak solutions exist only for constant initial distributions x0 ∈ R. Corollary 1. SDI (1) has weak and explicit weak solutions for every constant initial distribution x0 ∈ R if and only if the right side of the inclusion satisfies condition (11). Corollary 2. If stochastic differential inclusion (1) has weak solutions for every initial distribution, then the inclusion has explicit weak solutions with respect to the selection bopt for every initial distribution. Conducting the proof of the sufficiency for the selections bint and bext, one can conclude Theorem 2. If MB ⊆NB (15) or MB ⊆NB, (16) then the stochastic differential inclusion (1) has weak and explicit weak solutions for every initial distribution. Corollary 3. For every initial distribution, condition (15) guarantees the existence of an explicit weak solution of the stochastic differential inclusion (1) with respect to the internal characteristic selection bint. At the same time, for every initial distribution, condition (16) guarantees the existence of an explicit weak solution of the stochastic differential inclusion (1) with respect to the external characteristic selection bext. 102 ANDREI N. LEPEYEV Lemma 3. If the right side of inclusion (1) is closed (its graph is a closed subset of R 2), then it satisfies condition (15). Proof. We will denote U(x, C) - open ball around x ∈ R with radius C > 0. Since the family of such open balls is the fundamental system of neighborhoods of R, we can use the open balls in the definitions of the sets MB and MB. Let x∗ ∈ MB. We use the rule of contraries. Assume that x∗ /∈NB, e.c. (x∗, 0) does not belong to the graph. The closure of the graph implies ∃U(0, C1), U(x∗, C2) such that U(0, C1) ∩ {y|y ∈ B(x), x ∈ U(x∗, C2)} = ∅ and, for all y ∈ B(x), x ∈ U(x∗, C2), the inequality |y| ≥ C1 holds. Therefore, β(0, B(x)) ≥ C1 or β(0, B(x))−2 ≤ C−2 1 . Hence, ∫ U(x∗,C2) β(0, B(x))−2 dx < ∞. Thus, we have a conflict with the definition of the setMB. Corollary 4. The stochastic differential inclusion (1) has weak and explicit weak solu- tions for every initial distribution, if its right side B is closed. The next theorem clarifies the existence of explicit solutions for every explicit Borel measurable selection of the SDI right side. It is known that the selections always exist for every multi-valued Borel measurable mapping (cf. [4], Theorem III.6). Theorem 3. The stochastic differential inclusion (1) has explicit weak solutions with respect to every Borel measurable selection for every initial distribution if and only if MB ⊆NB. Proof. The proof of the sufficiency can be conducted similarly to that in Theorem 1. One needs to take an arbitrary explicit Borel measurable selection v of the mapping B and use the fact that Mv ⊆MB ⊆NB ⊆ Nv, which implies W̃ ∈ Mv ∪ (R \ Nv) l+ × P-a.e. and if {Av∞ < +∞}, then P-a.s. v2(W̃Av∞) = 0. The proof of the sufficiency can be conducted by the steps of the necessity proof of Theorem 1. Having taken an arbitrary point x0 /∈NB, one can investigate the behavior of explicit weak solutions with respect to the selection v(x) = { bext(x), x = x0; bint(x), x = x0. Using the arguments of the proof of Theorem 1, one can conclude that a solution with initial distribution x0 with respect to the selection v is non-trivial. Therefore, P({ ∫ t 0 b−2 ext(W v s )ds < +∞, Av ∞ > t}) > 0. Hence, Lemma 1 in [9] implies the existence of an open neighborhood G of x0 such that the function b−2 ext is integrable over G, and x0 /∈MB. Remark 4. If the right side of SDI (1) is single-valued, then the existence conditions from Theorems 1, 2, and 3 are coincide and equal to condition (3) (the Engelbert—Schmidt theorem). The next definition will use the set P(R ∪ {±∞}) of all non-empty subsets from R ∪ {±∞}. Definition 2. The multi-valued mapping B : R → P(R ∪ {±∞}) is locally integrable in the wide sense (locally integrable in the narrow sense), if its every Borel measurable selection is locally integrable (there exists its Borel measurable locally integrable selec- tion). ON STOCHASTIC DIFFERENTIAL INCLUSIONS 103 The upcoming statements will be dealing with the local integrability of the multi- valued mapping B−2 which can be defined for every x ∈ R by B−2(x) := {z = 1 y2 ∈ R+ ∪ {+∞}, ∀y ∈ B(x)}. Remark 5. The local integrability in the narrow sense of a multi-valued mapping is equivalent to the local integrability of its internal characteristic selection. At the same time, the local integrability in the wide sense of a multi-valued mapping is equivalent to the local integrability of its external characteristic selection. Theorem 4. The stochastic differential inclusion (1) has weak and explicit weak non- trivial solutions for every initial distribution if and only if the mapping B−2 is locally integrable over R in the narrow sense. Proof. The proof mostly repeats the steps of the proof of Theorem 1, that is why we will stop on the principal points about the non-triviality of solutions. The construction of a solution can be conducted for the selection v(x) ≡ bext(x). By Remark 5, the function v−2 is locally integrable, that implies Mv =MB = ∅ and, hence, Av∞ = U(Mv) = ∞. Therefore, the constructed solution cannot be trivial, otherwise would be: v(Xt) = 0, ∀t ≥ 0,P-a.s. and Av t = 0, ∀t ≥ 0,P-a.s., that is, we have a conflict with Av∞ = ∞. The proof of the necessity is easily accomplished due to the existence of non-trivial solutions for every initial distribution, which implies the fulfilment of (13) and, hence, (14). Combining the proofs of Theorems 3 and 4, we can formulate the following: Theorem 5. The stochastic differential inclusion (1) has explicit weak non-trivial solu- tions with respect to every Borel measurable selection for every initial distribution if and only if the mapping B−2 is locally integrable over R in the wide sense. Examples. 1. The theorems of this paper allow us to investigate SDIs of type (1) with locally unbounded right sides. For instance, let the right side of SDI (1) be the countable union of straight strophoids B(x) = ±x √ 2a + x 2a − x , a ∈ Z, x ∈ R. It can be easily verified that the right side satisfies the conditions of Theorem 1, hence, the inclusion has explicit weak solutions (from Proposition 3, they are weak solutions as well) for every initial distributions. Furthermore, the right side satisfies the conditions of Theorem 3. Hence, the inclusion has weak and explicit weak solutions with respect to every Borel measurable selection of B for every initial distributions. On the other hand, by Theorem 4, there exists a probabilistic measure P̄ such that there is no non-trivial weak solution of the inclusion with P̄ as its initial distribution. 2. SDI (1) with the right side B(x) = { [|x| 14 , |x| 15 ], |x| ≤ 1; [|x| 14 , |x| 13 ], |x| > 1; possesses non-trivial explicit weak solutions (by Proposition 3, they are weak solutions as well) for every initial distribution from Theorem 4 due to the local integrability of the mapping B−2 in the narrow sense. Its external characteristic selection bext(x) = { |x| 15 , |x| ≤ 1; |x| 13 , |x| > 1; 104 ANDREI N. LEPEYEV is locally integrable in power (-2). Moreover, the inclusion possesses non-trivial explicit weak solutions with respect to every Borel measurable selection of B for every initial distributions from Theorem 5, because the internal characteristic selection bint(x) = |x| 14 is locally integrable in power (-2). 3. Consider SDE of type (2) with the diffusion coefficient b(x) = ⎧⎪⎨ ⎪⎩ Arcth(x), x ∈ (−∞,−1) ∪ (1, +∞); Arth(x), x ∈ (−1, 0) ∪ (0, +1); 1, x ∈ {−1, 0, +1}; where Arth, Arcth - area hyperbolic tangent and cotangent, correspondingly. This equa- tion does not have a weak solution for the initial distribution x0 = 0 from Theorem 1 in [2]. Using one of the standard procedures (cf. [12,21]), we can generalize the equation. For instance, let us construct the right part B of the corresponding inclusion as a minimal close hull of the diffusion coefficient b excluding values {±∞}. As a result, we obtain SDI with the right side B(x) = ⎧⎪⎪⎨ ⎪⎪⎩ Arcth(x), x ∈ (−∞,−1) ∪ (1, +∞); Arth(x), x ∈ (−1, 0) ∪ (0, +1); {0, 1}, x = 0; 1, x ∈ {−1, 1}; Weak solutions of the initial SDE will be treated in the sense of weak solutions of the constructed SDI. The weak solutions exist for every initial distribution, since, for this right side, MB =NB = {0}. Bibliography 1. N.U. Ahmed, Nonlinear stochastic differential inclusions on Banach space, Stochastic Anal. Appl. 12 (1994), 1–10. 2. S. Assing, T. Zenf, On stochastic differential equations without drift, Stoch. and Stoch. Reports. 36 (1991), 21–39. 3. R.J. Auman, Integrals of set-valued functions, Journal of Math. Anal. and Appl. 12 (1965), 1–12. 4. M.C. Castaing, M. Valadier, Convex analysis and measurable multifunctions, Lecture Notes in Math., vol. 580, Springer, Berlin, 1977. 5. E. Cepa, Equations differentielles stochastiques multivoques, Siminaire Probabilities 29 (1995), 86–107. 6. E.D. Conway, Stochastic equations with discontinuous drift, Trans. Amer. Math. Soc. 157 (1971), 235–245. 7. G. Da Prato, H. Frankovska, A stochastic Filippov theorem, Stochastic Anal.Appl. 12 (1994), 409–426. 8. H.J. Engelbert, J. Hess, Stochastic Integrals of Continuous Local Martingales, I, II, Math. Nachr. 97,100 (1980,1981), 325–343,249–269. 9. H.J. Engelbert, W. Schmidt, On behaivior of certain functionals of the Wiener process and applications to stochastic differential equations, Lecture Notes in Cont. and Inf. Sciences, Springer., Berlin, 1981, pp. 47–55. 10. H.J. Engelbert, W. Schmidt, On Solutions of One-Dimensional Stochastic Differential Equa- tions Without Drift, Z.Wahrscheinlichkeitstheorie verw.Geb. (1985), 287–314. 11. H.J. Engelbert, W. Schmidt, On one-dimensional stochastic differential equations with gener- alized drift, Lect. Notes Control Inf. Sci. Springer-Verlag, Berlin. 69 (1985), 143–155. 12. H.J. Engelbert, W. Schmidt, Strong Markov continuous local martingales and solutions of one- dimensional stochastic differential equations, I, II, III, Math. Nachr. 143,144,151 (1989, 1989, 1991), 167–184,241–281,149–197. 13. A.F. Filippov, Differential equations with discontinuous right sides, Nauka, Moscow, 1985. 14. F. Hiai, H. Umegaki, Integrals, conditional expectations and martingales of multivalued func- tions, J. Multivar. Anal. 7 (1977), 149–182. ON STOCHASTIC DIFFERENTIAL INCLUSIONS 105 15. N. Ikeda, S. Watanabe, Stochastic Differential Equations and Diffusion Rrocesses., North- Holland, Amsterdam, 1981. 16. N. Kazamaki, Changes of time, Stochastic Integrals, and Weak Martingales, Z.Wahrscheinlich- keits-theorie verw. 22 (1972), 25–32. 17. M. Kisielewicz, Stochastic differential inclusions, Discuss. Math., Differ. Incl. 17 (1997), no. 1– 2, 51–65. 18. M. Kisielewicz, M. Michta, J. Motyl, Set valued approach to stochastic control. I. Existence and regularity properties. II. Viability and semimartingale issues, Dynam. Systems Appl. 12 (2003), no. 3-4, 405–431,433–466. 19. P. Kree, Diffusion equation for multivalued stochastic differential equation, J. Func. Anal. 49 (1982), 73–90. 20. A.A. Levakov, Stochastic differential inclusions, J. Differ. Eq. 2(33) (1997), 212–221. 21. A.N. Lepeyev, On one-dimensional homogeneous stochastic differential equations and inclu- sions without drift, Vesti NAN Belarusi. Ser. phis.-math. navuk. (2005), no. 2, 42–48. 22. A.N. Lepeyev, Existence theorems for weak solutions of one-dimensional homogeneous stochas- tic differential equations and inclusions with unbounded coefficients, Doklady NAN Belarusi 49 (2005), no. 3, 47–52. 23. R. Petterson, Existence theorem and Wong-Zakai approximations for multivalued stochastic differential equation, Probability and Mathematical Statistics. 17 (1997), no. 1, 29–45. E-mail : ALepeev@iba.by