ISSN 2201-7372
Volume 2, Number 2, 2014, 230-363
Β© Copyright 2014 the authors. 230
Strong Large Deviations Principles of Non-Freidlin-Wentzell Type -Optimal Control Problem with Imperfect Information
-Jumps Phenomena in Financial Markets
J. Foukzon
Israel Institute of Technologies, Department of Mathematics, Haifa, Israel
Abstract. The paper presents, a new large deviations principles (SLDP) of non-Freidlin-Wentzell type, corresponding to the solutions Colombeau-Itoβs SDE. Using SLDP we present a new approach to construct the Bellman function π£(π‘, π) and optimal control π(π‘, π) directly by way of using strong large deviations principle for the solutions Colombeau-Itoβs SDE. As important application such SLDP, the generic imperfect dynamic models of air-to-surface missiles are given in addition to the related simple guidance law. A four, examples have been illustrated proposed approach and corresponding numerical simulations have been illustrated and analyzed. Using SLDP approach, Jumps phenomena, in financial markets, also is considered. Jumps phenomena, in financial markets is explained from the first principles, without any reference to Poisson jump process. In contrast with a phenomenological approach we explain such jumps phenomena from the first principles, without any reference to Poisson jump process.
Keywords: Optimal control, Bellman equation, Colombeau-Itoβs SDE,Large deviations principles, Algebra of Colombeau generalized functions, Poisson jump process, Jumps phenomena, in financial markets.
1. Introduction
What new scalable mathematics is needed to replace the traditional Partial Differential Equations (PDE) approach to differential games?
Letβ = (πΊ, π΄. π)be a probability space. Any stochastic process onβπ is aΞ£-measurable
mappingπ: πΊ Γ [0, π] β βπ. Many stochastic optimal control problems essentially come
down to constructing a functionπ’(π‘, π₯) that has the properties: (1) π’(π‘, π₯) = infπΌοΏ½πΜ οΏ½οΏ½ππ ,π·π₯ (π)οΏ½πβ[0,π‘]; {πΌ(π )}πβ[0,π‘]οΏ½οΏ½ and
(2) π’(π‘, π₯) = infπΌοΏ½πΜ οΏ½οΏ½ππ ,π·π₯ (π)οΏ½πβ[0,π‘]; {πΌ(π )}πβ[0,π‘]οΏ½ + π’ οΏ½π‘, ππ‘,π·π₯ (π)οΏ½οΏ½, where πΌ(π‘) β π β π π.
Here πΜ = πΈπΊοΏ½β« οΏ½ποΏ½π0π‘ π ,π·π₯ (π), π οΏ½οΏ½ ππ οΏ½ is the termination payoff: functional,πΌ(π‘)is a control
and ππ‘,π·π₯ (π) is some Markov process governed by some stochastic Itoβs equation driven
by a Brownian motion of the form
ππ‘,π·π₯ (π) = π₯ + β« π οΏ½π0π‘ π ,π·π₯ (π), πΌ(π )οΏ½ππ + βπ·π(π‘, π). (3)
Hereπ(π‘, π) is the Brownian motion. Traditionally the function π’(π‘, π₯) has been computed by way of solving the associated Bellman equation, for which various numerical techniques mostly variations of the finite difference scheme have been developed. Another approach, which takes advantage of the recent developments in computing technology and allows one to construct the functionπ’(π₯, π‘) by way of backward induction governed by Bellmanβs principle such that described in [1]. In paper [1] Equation (3) is approximated by an equation with affine coefficients which admits an explicit solution in terms of integrals of the exponential Brownian motion. Using Colombeau approach proposed in paper [2], [3],[4] we have replaced Equation (3) by Colombeau-Itoβs equation [4-6]: οΏ½ππ‘,π·,ππ₯,π β²(π, π)οΏ½ πβ² = π₯ + οΏ½οΏ½ ππβ²οΏ½ππ‘,π·,πβ² π₯,π (π, π), πΌ(π )οΏ½ π‘ 0 ππ οΏ½πβ² +βπ· οΏ½β« π€0π‘ πβ²(π , π)ππ οΏ½ πβ²+ βποΏ½π(π‘, π)οΏ½πβ². Here π, πβ²β (0,1],π β πΊ
1, π β πΊ2, πΊ1β© πΊ2 = β , where π€(π‘, π) is the white noise on βπ
i.e., π€(π‘, π) = π ππ‘β π(π‘, π) almost surelyin π·β²and π€
noiseon βπi.e., π€
πβ²(π‘, π) = β©π€(π‘, π), ππβ²(π β π‘)βͺ, and ππβ² is a model delta net [2], [4].
Fortunately in contrast with Equation (3) one can solve Equation (4) without any approximation using strong large deviations principleof Non-Freidlin-Wentzelltype [5],[6],[7].
Statement of the novelty and uniqueness of the proposed idea: A new approach, which is proposed in this paper allows one to construct the Bellman functionπ£(π‘, π₯) and optimal control πΌ(π‘, π₯) directly, i.e., without any reference to the Bellman equation, by way of using strong large deviations principle for the solutionsColombeau-Itoβs SDE (CISDE).
2. Proposed Approach
Letβπ = (Ξ©π, πΊπ, ππ), π = 1,2 be a probability spaces such that: Ξ©1β© Ξ©2 = β . Let us
consider m-persons Colombeau-Ito differential gameπΆπΌπ·πΊπ;π(π, π, π, πΊπ(βπ), β1, β2),
withthe termination payoff functional for the i-th player is: οΏ½πΜ ππβ²,ποΏ½ πβ²=ππΊ1ππΊ2οΏ½οΏ½β« π0π πβ²,ποΏ½π₯π‘,π·,ππ₯,π β²(π, π), πΆ(π‘), π‘, ποΏ½ππ‘οΏ½ πβ²οΏ½ + +ππΊ1ππΊ2οΏ½οΏ½β οΏ½π₯π,π·,ππ₯,π β²;π(π, π) β π¦ποΏ½ 2 π π=1 οΏ½ πβ²οΏ½ (1)
and with stochastic nonlinear dynamics: οΏ½πΜπ‘,π·,ππ₯0,πβ²(π, π)οΏ½ πβ² = οΏ½ππβ²οΏ½ππ‘,π·,πβ² π₯,π (π, π), βπ·π€ πβ²(π‘, π), πΆ(π‘), π‘, ποΏ½οΏ½ πβ² + βποΏ½π(π‘, π)οΏ½πβ² (2) π, πβ²β (0,1],π β πΊ 1, π β πΊ2. Here βπ‘ β [0, π]: οΏ½ππβ²(π‘)οΏ½πβ² β βοΏ½π; π0,π·,ππ₯0,πβ²(π, π) = π0 β βπ, βπΊ β (0,1]:π = [(ππβ²)πβ²], π = [(ππβ²)πβ²]; π(π,β,β,β,β), π(π,β,β,β) β πΊπ(βπ), πΆ(π‘) = {πΌ1(π‘), β¦ , πΌπ(π‘)}; πΌπ(π‘) β ππ β βππ, π = 1, β¦ , π,
And mβpersons Colombeau-Ito differential game
πΆπΌπ·πΊπ;π(π, π, π, πΊπ(βπ), π·(π‘), π(π‘), β1, β2) with imperfect measurements and with
imperfect information about the system [5], [6]. The corresponding stochastic nonlinear dynamics is: οΏ½πΜπ‘,π·,ππ₯0,πβ²(π, π)οΏ½ πβ² = οΏ½ππβ²οΏ½ππ‘,π·,πβ² π₯0,π (π, π), βπ·π€ πβ²(π‘, π), π(π‘), πΌ οΏ½π‘, π₯π‘,π·,ππ₯0,πβ² + π·(π‘)οΏ½ , π‘, ποΏ½οΏ½ πβ²+
+βποΏ½π€(π‘, π)οΏ½πβ²; π, πβ²β (0,1], π β Ξ©1, π β Ξ©2 and the playoff for the i-th player is: οΏ½πΜ ππβ²,ποΏ½ πβ²=ππΊ1ππΊ2οΏ½οΏ½β« ππβ²,ποΏ½ππ‘,π·,πβ² π₯0,π (π, π), πΌοΏ½π‘, π·(π‘)οΏ½, π‘, ποΏ½ π 0 ππ‘οΏ½πβ²οΏ½ + +ππΊ1ππΊ2οΏ½οΏ½β οΏ½π₯π,π·,ππ₯0,πβ²;π(π, π) β π¦ποΏ½ 2 π π=1 οΏ½ πβ²οΏ½. (3) Hereπ·(π‘) =(π½1(π‘), β¦ , π½π(π‘)), π(π‘) = οΏ½π1(π‘), β¦ , ππ(π‘)οΏ½and βπ‘ β [0, π]: οΏ½π₯πβ²(π‘)οΏ½πβ² β βοΏ½π; π₯ 0,π·,ππ₯0,πβ²(π, π) = π₯0, βπΊ β (0,1]: π = [(ππβ²)πβ²], π = [(ππβ²)πβ²]; π(π₯,β,β,β,β,β), π(π₯,β,β,β) β πΊπ(βπ)or π(π₯,β,β,β,β), π(π₯,β,β) β πΊπ,ππ (πΈ), πΌ(π‘) = {πΌ1(π‘), β¦ , πΌπ(π‘)}; πΌπ(π‘) β π β βππ, π = 1, β¦ , π, π·(π‘) = {π½1(π‘), β¦ , π½π(π‘)}, π(π‘) = {π1(π‘), β¦ , ππ(π‘)}.
Here β is a field of the real numbers, πΊ(βπ) is the algebra of Colombeau generalized
functions [8],[9],[12] , πΊπ(βπ) = πΊ(βπ) Γ β¦ Γ πΊ(βπ), πΊ
π,π(πΈ) =β±βͺπ,ππ,π(πΈ)(πΈ)is the Colombeau
type algebra[13],[14], E is an appropriate algebra of functions, which is a locally convex vector space over field β΅, πΊπ,ππ (πΈ) = πΊπ,π(πΈ) Γ β¦ Γ πΊπ,π(πΈ), β οΏ½ is the ring of Colombeau
generalized numbers [11], βοΏ½π = βοΏ½ Γ β¦ Γ βοΏ½,π‘ β πΌ
π(π‘)is the control chosen by the i-th
player, within a set of admissible control values ππ.
Hereπ‘ βΌ οΏ½οΏ½π₯π‘,π·,ππ₯0,πβ²;1(π, π)οΏ½
πβ², β¦ , οΏ½π₯π‘,π·,πβ²;π
π₯0,π (π, π)οΏ½
πβ²οΏ½isthe
trajectoryof the Equation (2). Optimal control problem for thei-th player is:
οΏ½πΜΏππβ²,ποΏ½
πβ² = οΏ½minπΌπ(π‘)βπποΏ½maxπΌπ(π‘)βπππΜ ππβ²,πβ ποΏ½οΏ½
πβ² . (4)
We remind now some classical definitions. Let us consider now Itoβs SDE:
πππ‘= π(ππ, π‘)ππ‘ + βππ=1ππ(ππ, π‘)πππ(π‘, π), (5)
π0 = π0(π), π β βπ.
Theorem1.[15]-[16]. Let the vectorsπ(π, π‘), π(π, π‘)be continuous functions of(π, π‘)such that for some constantsπ·and πΆthe following conditionshold:
βπ(π, π‘) β π(π, π‘)β + β |πππ=1 π(π, π‘) β ππ(π, π‘)| β€ π·βπ β πβ, (6)
βπ(π, π‘)β + β |πππ=1 π(π, π‘)| β€ πΆ(1 + βπβ). (7)
Then: (1) For every random variableπ(π)independent of theprocessesππ(π‘, π), π =
1,2, β¦ , πthere exists a solution ππ‘of the Itoβs SDE(5)which is an almost surely
continuous Markov process and
(2)Two solutionsππ‘,1andππ,2(π)is unique up to equivalence: ποΏ½ππ‘,1(π) = ππ,2(π)οΏ½ = 1,for
all π‘ β [0, β) = πΌβ.
Remark1.[15],[17].It well known, that the boundedness assumption on π(π, π‘)and π(π, π‘)can be weakened, but somekind of restriction on the π(π, π‘)and π(π, π‘)is
necessary in order to guarantee the existence of a global solution i.e., a solution defined for allπ‘ β [0, β).If we remove this condition of boundedness, then a solution of Itoβs SDE (5) does exist locally but, in general, blows up (or explodes) in finite time.
Definition1.LetβΜπ = βπβ{π}be the one-point compactification of βπandπΎΜπ =
οΏ½π|οΏ½0, β) β π‘ β¦ π(π‘) β βΜπis continuous and such that
ifπ(π) = π, thenπ(π‘β²) = π for all π‘β²β₯ π‘}.Let βοΏ½βΜποΏ½be the π-field generated by Borel
cylinder sets. Forπ β πΎΜπwe set
π(π) = inf{π‘|π(π‘) = π₯} (8) and call the explosion time of the trajectory π(π‘), π‘ β [0, β).
Definition2.[15].By a solutionππ‘(π)of the equation(5)we mean a
οΏ½πΎΜπ, βοΏ½βΜποΏ½οΏ½- valued random variable defined on a probability
spaceβ = (Ξ©, πΊ, π)with areference family(πΊπ‘)π‘β₯0such that:
(i) there exists an n-dimensional(πΊπ‘)-Brownian motion
πΎ(π‘, π) = οΏ½π1(π‘, π), β¦ , ππ(π‘, π)οΏ½ with πΎ(0, π) = 0,
(iii) if π(π) = ποΏ½ππ(π)οΏ½ is the explosion time of ππ(π)then for almost all π,
ππ‘(π) β π0(π) =
= β« π(πππ π(π), π‘)ππ + βππ=1β« π0π‘ π(ππ(π), π‘)πππ(π , π),(9)
for all π‘ β οΏ½0, π(π)οΏ½.
Theorem2.[15],[16].(1)Given βΜπβcontinuousπ(π, π‘)and π(π, π‘)consider the equation
(5).Then for any probability π onοΏ½πΎΜπ, βοΏ½βΜποΏ½οΏ½with compact support, there exists a
solution
of (5) such thatthe law of π0(π) coincides with π.
(2) Supposeπ(π, π‘)andπ(π, π‘)are locally Lipschitz continuous, i.e., for every π > 0 there exists a constantπ·π> 0 such that
βππ(π, π‘)βππ(π, π‘)β + β οΏ½πππ=1 π,π(π, π‘) β ππ,π(π, π‘)οΏ½ β€ π·πβπ β πβ (10)
for everyπ, π β π©π,π©π΅ = {π|βπβ β€ π}.Then for any probability π onοΏ½πΎΜπ, βοΏ½βΜποΏ½οΏ½with
compact support, there exists a solution
of (5) such thatthe law of π0(π) coincides with π.
Theorem3.[16].Letππ‘,π(π‘), π = 1,2, β¦ be the solutions
of the Itoβs SDEβs
πππ‘,π = πποΏ½ππ‘,π, π‘οΏ½ππ‘ + βππ=1ππ,ποΏ½ππ‘,π, π‘οΏ½πππ(π‘, π), (11)
π0,π = π(π)π₯ β βπ.
Assume that: (i) let the vectors ππ(π, π‘), ππ(π, π‘)becontinuous functions of (π, π‘)such
that for some constantsπ·and πΆ the following conditions hold
βππ(π, π‘) β ππ(π, π‘)β + β οΏ½πππ=1 π,π(π, π‘) β ππ,π(π, π‘)οΏ½ β€ π·βπ β πβ, (12)
(ii)π[π2(π)] < β, (14)
(iii)βπ > 0:
limnββsupβxββ€NοΏ½βππ(π, π‘) β π0(π, π‘)β + β οΏ½πππ=1 π,π(π, π‘) β ππ,0(π, π‘)οΏ½οΏ½ = 0. (15)
Then
limnββsup0β€tβ€TποΏ½ππ‘,π(π) β ππ‘,0(π)οΏ½2 = 0. (16)
Corollary 1.Let ππ‘,π(π‘), π = 1,2, β¦ be the solutionsof the Itoβs SDEβs
πππ‘,π = πποΏ½ππ‘,π, π‘οΏ½ππ‘ + βππ=1ππ,ππππ(π‘, π)π0,π = π(π)π₯ β βπ.(17)
Assume that: (i) Let the vectors ππ(π, π‘),becontinuous functions of (π, π‘) and ππ =
ππππ π‘ such that for some constants π·and πΆ the following conditions hold βππ(π, π‘) β ππ(π, π‘)β β€ π·βπ β πβ (18) βππ(π, π‘)β + β οΏ½πππ=1 π,ποΏ½ β€ πΆ(1 + βπβ), (19) (ii) π[π2(π)] < β, (20) (iii)βπ > 0: limnββsupβxββ€NοΏ½βππ(π, π‘) β π0(π, π‘)β + β οΏ½πππ=1 π,ποΏ½οΏ½ = 0. (21) Then limnββsup0β€tβ€TποΏ½ππ‘,π(π) β ππ‘,0(π)οΏ½2 = 0.(22)
Here ππ‘,0(π) is the solution of the ODE:
Remark 2.Note that Theorem 3 in fact asserts that under conditions (12)-(15)any solutionππ‘(π) of the Itoβs SDE (5) is continuously depend on functions π(π, π‘)andπ(π, π‘).
Note that the assumptions of the Lipschitz continuously (12) and boundedness (13) on π(π, π‘) and π(π, π‘)in the Theorem 3 cannot be weakened.
Theorem. Assume that: (1) Let ππ‘,π(π‘), π = 1,2, β¦ be the solutions of the Itoβs SDEβs
ππ‘,π = πποΏ½ππ‘,π, π‘οΏ½ππ‘ + πποΏ½ππ‘,π, π‘οΏ½ππΎ(π‘, π),
π0,π = π(π), π₯ β βπ.
and letποΏ½π‘,π(π‘), π = 1,2, β¦ be the solutions of the Itoβs SDEβs
ποΏ½π‘,π = ποΏ½ποΏ½ποΏ½π‘,π, π‘οΏ½ππ‘ + ποΏ½ποΏ½ποΏ½π‘,π, π‘οΏ½ππΎ(π‘, π), ποΏ½0,π = π(π), π₯ β βπ. Here πποΏ½ππ‘,π, π‘οΏ½ππΎ(π‘, π) = οΏ½ ππ,ποΏ½ππ‘,π, π‘οΏ½πππ(π‘, π) π π=1 , ποΏ½ποΏ½ποΏ½π‘,π, π‘οΏ½ππΎ(π‘, π) = οΏ½ ποΏ½π,ποΏ½ππ‘,π, π‘οΏ½πππ(π‘, π) π π=1 . (2) The inequalities βππ(π, π‘)β + βππ(π, π‘)β β€ πΎπ(1 + βπβ), βππ(π, π‘) β ππ(π, π‘)β + βππ(π, π‘) β ππ(π, π‘)β β€ πΎπβπ β πβ, οΏ½ποΏ½π(π, π‘)οΏ½ + βποΏ½π(π, π‘)β β€ πΎπ(1 + βπβ),
οΏ½ποΏ½π(π, π‘) β ποΏ½π(π, π‘)οΏ½ + βποΏ½π(π, π‘) β ποΏ½π(π, π‘)β β€ πΎπβπ β πβ,
οΏ½ππ(π, π‘) β ποΏ½π(π, π‘)οΏ½ β€ πΏ1,πβπβ,
βππ(π, π‘) β ππ(π, π‘)β β€ πΏ2,πβπβ
where 0 β€ π‘ β€ π, is satisfied. Then the inequality
sup0β€π‘β€ππ οΏ½οΏ½ππ‘,πβ ποΏ½π‘,ποΏ½2οΏ½ β€ ππΏποΏ½πΏ1,π2 + πΏ2,π2 οΏ½πΈ οΏ½οΏ½ οΏ½ποΏ½π‘,ποΏ½2ππ‘ π
0 οΏ½
is satisfied.
Proof. See Appendix.
Remark 3.[17].If conditions(6)-(7)are valid only in every cylinder ππ Γ πΌβ, with πΆ =
πΆ(π ), π· = π·(π ), one can construct a sequenceof functions ππ(π, π‘)andππ(π, π‘)such that
forβπβ < π
ππ(π, π‘) = π(π, π‘), ππ(π, π‘) = π(π, π‘), (24)
and therefore for eachππ(π, π‘), ππ(π, π‘)satisfy conditions(6)-(7)everywhere in βπ. By
Theorem 1,there exists a sequence of Markov processes ππ‘,π(π)corresponding to the
functionsππ(π, π‘)and ππ(π, π‘).
Assumption1.Suppose now that the distribution ofπ0(π)hascompact support
inβπ.Then as, well known, that the first exit random timesπ
π(π) of the
processes ππ‘,,π(π)from the set βπβ < π are identical for π β₯ π[15] [18],[19]. Let this
common value beππ(π). Itis also clear that the processes themselves coincide up to
timeππ(π), i.e.
Orin the equivalent form
ποΏ½sup0β€π‘β€ππ(π)οΏ½ππ‘,π(π) β ππ,π(π)οΏ½ > 0οΏ½ = 0, π > π. (27)
Definition1. (i) Let πβ(π)denote the (finite or infinite) limit of the monotone
increasing sequence ππ(π)as π β β. We call the random variable πβ(π)the first exit
time from every bounded domain, or briefly the explosion time. (ii)We now define a new stochastic process ππ‘(π)by setting[17]:
ππ‘(π) = ππ,π(π)for π‘ < ππ(π). (28)
It well known, that this is always a Markov process for π‘ < ππ(π)[18],[19].
We also can to define a new stochastic process ππ‘(π)by setting
ππ‘(π) = π β limnββππ,π(π) (29)
If finite or infinite limit in RHS of Eq.(29) exist. (iii) In general case we set
οΏ½ππ‘,πΊβ²(π)οΏ½
πΊβ² = οΏ½ππ‘,π(π)οΏ½π, π = 1
πΊβ². (30)
We note that the Colombeau-Itoβs equation οΏ½ππ‘,πΊβ²(π)οΏ½ πΊβ² β οΏ½π0,π(π)οΏ½πΊβ² = οΏ½β« ποΏ½πππ π,πΊβ²(π), π οΏ½ππ οΏ½πΊβ² + β οΏ½β« π0π‘ ποΏ½ππ,πΊβ²(π), π‘οΏ½πππ(π , π)οΏ½ πΊβ² π π=1 (31)
is satisfied for all π‘ β οΏ½0, πβ(π)οΏ½.
ππ ,π{π
β(π) = β} = 1. (32)
Assumption2. We assume now
that:(1)βπ β (0,1]:οΏ½ππΊβ²(π, π‘, π)οΏ½πΊβ² β οΏ½π1,πΊβ²(π, π‘, π), β¦ , ππ,πΊβ²(π, π‘, π)οΏ½πΊβ² β πΊπ(βπ) οΏ½or πΊπ,ππ (πΈ)οΏ½,π =
(π1, β¦ , ππ), π β (0,1]:πfor all π‘ β [0, β)and
(2) βπ β (0,1]πthere exist infinite Colombeau constants οΏ½πΆ
ππβ²οΏ½πβ²andοΏ½π·ππβ²οΏ½πβ²such thatβπ β
(0,1]:
(i)(βππΊβ²(π, π‘, π)βπ)πΊβ² β€ οΏ½οΏ½πΆππβ²οΏ½πΊβ²οΏ½ (1 + βπβπ), πβ²β (0,1],(33)
(ii)οΏ½οΏ½ππ,πΊβ²(π, π‘, π) β ππ,πΊβ²(π, π‘, π)οΏ½οΏ½
πΊπβ² β€ οΏ½οΏ½π·ππβ²οΏ½πΊπβ²οΏ½ βπ β πβ(34)
for all π‘ β [0, β)and for all xβ βπand for allyβ βπ.
Definition 2. [4] 1.Let β = (πΊ, π΄. π)be a probability space. Let βπ be the space of nets οΏ½ππ(π)οΏ½πof measurable functions on πΊ.
Let βπ π be the space of nets (ππ)π β βπ , π β (0,1],with the
property that for almost all π β πΊ there exist constants π, πΆ > 0 andπ0 β (0,1] such that |(πβ°)β°| β€ πΆπβπ, π β€ π0.
2.Let ππ is the space of nets(πβ°)β° β βπ , π β (0,1],with the property that foralmost
all π β πΊand all π β β+there exist constantsπΆ > 0and π0 β (0,1] such that |(πβ°)β°| β€
πΆππ, π β€ π
0.Thedifferential algebra πΊπ of Colombeau generalized random variables is
thefactor algebra πΊπ = βπ /ππ . Let us consider now a familyοΏ½ππ‘,π,ππ₯0 ,πβ²οΏ½
πβ² of the solutions Colombeau-Itoβs SDE:
οΏ½πππ‘,,π,ππ₯0,πβ²(π)οΏ½ πβ² = οΏ½ππΊβ²οΏ½π₯π‘,,π,πβ² π₯0 ,π(π), π‘, ποΏ½οΏ½ πβ²+ βποΏ½ππΎ(π‘, π)οΏ½πβ², (35) οΏ½π0,π,,πππ,πΊ β²οΏ½ πβ² = οΏ½ππΊβ² π0(π)οΏ½ πΊβ², β πΊπ , οΏ½π οΏ½π0,π,πβ² π₯0,π οΏ½οΏ½ πβ² = π0 β π οΏ½ π, (36) π‘ β [0, π], π, πβ²β (0,1].
Here (i)πΎ(π‘, π) = οΏ½π1(π‘, π), β¦ , ππ(π‘, π)οΏ½ is n-dimensional Brownian motion,(ii)βπ‘ β
polynomial on variable π = (π₯1, β¦ , π₯π),i.e.
π0,π(π, π‘, π) = βπΌ,|πΌ|β€ππ0,ππΌ (π‘, π)π₯πΌ, (37)
πΌ = (π1, β¦ , ππ), |πΌ| = βππ=1ππ , 0 β€ ππ β€ π, or
(iii)) βπ‘ β [0, π]:οΏ½ππβ²(π₯, π‘, π)οΏ½
πβ² β πΊπ,ππ (πΈ), π0(π, π‘, π) β‘ ππβ²=0(π, π‘, π): βπ β βπisβ-analytic
function on variable π = (π₯1, β¦ , π₯π),i.e.
π0,π(π, π‘, π) = ββπ=1βπΌ,|πΌ|β€ππ0,ππΌ (π‘, π)π₯πΌ, (38) πΌ = (π1, β¦ , ππ), |πΌ| = βππ=1ππ , 0 β€ ππ β€ πand (iv) limβπββββπ0(π, π‘, π)β βπβ = ββ , (v) ππ,πβ²(π(π‘), π‘, π) = ππ,0(ππΊβ²(π‘), π‘, π).(39) Here ππΊβ²(π‘) = οΏ½π₯1,πβ²(π‘), β¦ , π₯π,πβ²(π‘)οΏ½and π₯π,πβ²(π‘) = οΏ½ π₯π,πβ² (π‘) = π₯π(π‘) 1+(πΊβ²)ππππππ(π‘), π β₯ 1 or π₯π,πβ²(π‘) = π₯π(π‘)πππ[π₯π(π‘)]. (40) π = 1, . . , π. Here πππ[π§] β πΆβ(β), π π’πποΏ½πππ[π§]οΏ½ β [βπ(ππ), π(ππ)]
β© βͺ β¨ βͺ β§ πππ[π§] = 1 β π§ β [βπ1(ππ), π1(ππ)] β [βπ(ππ), π(ππ)], πππ[π§] = 0 β π§ β β\[βπ(ππ), π(ππ)], 0 β€ πππ[π§] β€ 1 β π§ β [βπ(ππ), π(ππ)]\[βπ1(ππ), π1(ππ)].
Remark 5.By Theorem 1 for every Colombeau generalized random variableοΏ½ππΊπβ²0(π)οΏ½
πΊβ², β πΊπ such that
οΏ½π οΏ½π0,ππ₯0,πβ²οΏ½οΏ½
πβ² = π0 β π οΏ½π,and independent of the processes π1(π‘, π), β¦ , ππ(π‘, π) there exist
Colombeau generalized stochastic process οΏ½ππ‘,π,πππ,πΊβ²(π)οΏ½ πΊβ², π β²β (0,1], such thatοΏ½π 0,π,πΊβ² ππ,πΊ (π)οΏ½ πΊβ² = οΏ½ππΊβ² π0(π)οΏ½ πΊβ²,and οΏ½ππ‘,,π,πππ,πΊβ²(π)οΏ½
πΊβ²is the solution of the Colombeau-Itoβs SDE (35)-(36),which is an almost
surely continuous Colombeau generalized stochastic process and is unique up to equivalence
οΏ½π οΏ½οΏ½ππ,,π,πΊππ,πΊβ²,1(π) β ππ,,π,πΊππ,πΊβ²,2(π)οΏ½ > 0οΏ½οΏ½
πΊβ² = 0, for all π‘ β [0, β).
Remark 6.One can to construct a sequence of Colombeau generalized functions οΏ½ππβ²,π(π₯, π‘, π)οΏ½πβ² such that for βπ₯β < π:
ππβ²,π(π₯, π‘, π) = ππΊβ²(π₯, π‘, π), πβ²β (0,1], π β (0,1]π,
and therefore for each ππβ²,π(π₯, π‘, π), satisfy conditions (18)-(19) everywhere in βπ. By
Theorem 1, there exists a sequence of Colombeau generalized stochastic processes οΏ½ππ‘,,ππ₯0,πβ²,π(π)οΏ½
πβ²corresponding to Colombeau generalized functionsοΏ½ππβ²,π(π₯, π‘, π)οΏ½πβ². Suppose
now that for each πβ²β (0,1], π β (0,1]π the distribution of π 0,πβ²
π₯0 (π) has compact support
in βπ. Then there exit times of the processes π
π‘,,π,ππ₯0,πβ²,π(π), πβ², π β (0,1], from the set
βπ₯β < π are identical for π β₯ π. Let this common value beππβ²,π(π, π). It is also clear
that the processesοΏ½ππ‘,,π,ππ₯0,πβ²,π(π)οΏ½
πβ²and οΏ½ππ‘,,π,πβ²,π π₯0,π (π)οΏ½
πβ²themselves coincide up to
timeοΏ½ππβ²,π(π, π)οΏ½
οΏ½π οΏ½sup0β€π‘β€ππβ²,π(π,π)οΏ½ππ₯π‘,,,π,π0,π β²,π(π) β ππ‘,,,π,ππ₯0,π β²,π(π)οΏ½ > 0 οΏ½οΏ½
πΊβ² = 0, (41)
for all π β₯ π.
Definition3. (i) Let ππΊβ²(π, π, π), πβ², π β (0,1]πdenote the (finite or infinite) limit of the
monotone increasing sequence ππβ²,π(π, π, π)as π β β. We call the generalized random
variable οΏ½ππΊβ²(π, π, π)οΏ½πβ², πβ²β (0,1]the first exit time of the sample function from every
bounded domain, or briefly thegeneralizedexplosion time.
(ii) We now define Colombeau generalized stochastic process οΏ½ππ,,β,πΊππ,πΊβ²(π)οΏ½
πΊβ²by setting
ππ,,π,πΊππ,πΊβ²(π) = ππ,,π,πΊππ,πΊβ²,π(π)for π‘ = π‘(π) < ππΊβ²,π(π, π, π). (42)
(iii)That this is always a Markov process for π‘ = π‘(π) < οΏ½ππΊβ²,π(π, π, π)οΏ½ πΊβ².
(iv) Colombeau generalized stochastic process οΏ½ππ,,π,πΊππ,πΊβ²(π)οΏ½
πΊβ²defined by
setting (42) on the random generalized interval οΏ½0, οΏ½ππΊβ²,π(π, π, π)οΏ½ πΊβ²οΏ½is regular ,if for any π < β, π β βπ, π β (0,1]π:
(ππ ,π{π
πΊβ²(π, π, π) = β})πΊβ² = 1, πβ²β (0,1](43)
(vi) Colombeau generalized stochastic process οΏ½ππ,π,,πΊππ,πΊβ²(π)οΏ½
πΊβ², defined by setting (42)
is a strongly regular if for any π < β, π β βπ,πβ²β [0,1], π β (0,1]π:
(ππ ,,π{π
πΊβ²(π, π, π) = β})πΊβ² = 1. (44)
Remark7.Wenote that: (iii) does not imply (iv).
Proposition1.Assume that Colombeau generalized stochastic processοΏ½ππ,,πΊππ,πΊβ²(π)οΏ½
πΊβ²
defined by setting (42)is a strongly regular. Then (1)βπ, π β (0,1]π, βπΏ, πΏ > 0:
limπΊβ²β0π οΏ½οΏ½ππ‘,,π,πππ,πΊβ²(π) β ππ‘,π,πππ,πΊβ²=0(π)οΏ½ π οΏ½ = 0. (45.a) limπΊβ²β0π οΏ½οΏ½ππ‘,,π,πππ,πΊβ²(π) β ππ‘,π,πππ,πΊβ²=0(π)οΏ½ > πΏοΏ½ = 0. (45.b) (2)βπΏ, πΏ > 0: limπΊβ²β0,πβ0π οΏ½οΏ½ππ‘,,π,πππ,πΊβ²(π) β ππ‘,π=0,πππ,πΊ β²=0(π)οΏ½ 2 οΏ½ = 0. (45.c) limπΊβ²β0,πβ0π οΏ½οΏ½ππ‘,,π,πππ,πΊβ²(π) β ππ‘,π=0,πππ,πΊ β²=0(π)οΏ½ > πΏοΏ½ = 0. (45.d)
Proof. Immediately follows fromTheoremA1.(I) (see appendix A)anddefinitions1,3. Let us consider now a family οΏ½ππ‘,π,ππ₯0 ,πβ²(π)οΏ½
πβ² of the solutions of the Colombeau SDE:
οΏ½πππ‘,π,,ππ₯0,πβ²(π)οΏ½ πβ²,π = οΏ½ππΊβ²,ποΏ½ππ‘,,π,πβ² π₯0 ,π (π), π‘, ποΏ½οΏ½ πβ²+ βπππΎ(π‘, π), (46) οΏ½π0,ππ₯0,πβ²οΏ½ πβ² = π0 β π οΏ½π, π‘ β [0, π], π, πβ², β (0,1],π β (0,1]. π
Here πΎ(π‘) is n-dimensional Brownian motion,
andβπ β (0,1]π, βπ‘ β [0, π] andfor almost al Ο β Ξ© :οΏ½ππβ²,π(π₯, π‘, π)οΏ½πβ² β πΊπ(βπ), π0,0(β, π‘) β‘
ππβ²=0,π=0(β, π‘, π): βπ β βπis a polynomialvector-function on a variable π = (π₯1, β¦ , π₯π) i.e.,
ππ,0,0(π, π‘) = βπΌ,|πΌ|β€πππ,0,0πΌ (π‘)π₯πΌ, πΌ = (π1, β¦ , ππ), |πΌ| = βππ=1ππ , 0 β€ ππ β€ π, and
ππ,πβ²,π(π(π‘), π‘, π) = ππ,0,0οΏ½ππβ²,π(π‘, π), π‘οΏ½. (47)
π₯π,πβ²,π(π‘, π) = π₯π(π‘) 1+πβ²π₯ π2π(π‘)+πβ²οΏ½ππβ« π0π‘ ππ[π₯π(π)]π₯π2π(π)ππ+βπΏππ(π‘)οΏ½ 2 , (48) π = 1, . . , π. Now we let π’π(π‘) = ππβ« πππ[π₯π(π)]π₯π2π(π)ππ + βπΏππ(π‘) π‘ 0 (49)
and rewrite Eq.(46) of the canonical Colombeau-Ito form: οΏ½πππ‘,,ππ₯0,πβ²,π(π)οΏ½ πβ²,π = οΏ½ππβ²,ποΏ½ππ‘,,πβ²,π π₯0 ,π (π), π π‘,,πβ²,,π πΏ (π), π‘οΏ½οΏ½ πβ² + βπππΎ(π‘, π), ππ‘,,πβ²,π,π πΏ (π) = οΏ½π’1,π‘,,ππΏ β²,π(π), β¦ , π’π,π‘,,ππΏ β²,π(π)οΏ½, (50) οΏ½ππ’π,π‘,,ππΏ β²,π(π)οΏ½ πβ² = πποΏ½ππποΏ½π₯π,π‘,ππ₯0 ,πΏβ²,π(π)οΏ½ οΏ½π₯π,π‘,,ππ₯0 ,πΏβ²,π(π)οΏ½ 2π οΏ½ πβ² + βπΏπππ(π‘), (51) π = 1, β¦ , π, οΏ½π₯0,ππ₯0,πβ² οΏ½ πβ² = π0 β π οΏ½ π, π‘ β [0, π], π, πβ², π, πΏ β (0,1].
Theorem3.Let us consider a pair of the Colombeau-Itoβs SDE: οΏ½πππ,,πΊππ,πΊβ²,π(π)οΏ½ πΊβ² = οΏ½ππΊβ² π οΏ½π π‘,,πΊβ² ππ ,πΊ(π), π‘οΏ½οΏ½ πΊβ² + βποΏ½ππΎ(π‘, π)οΏ½πΊβ², (52) οΏ½π0,πΊππ,πΊβ²,ποΏ½ πΊβ² = π0 β π οΏ½ π, π‘ β [0, π], π, πβ²β (0,1], π = 1,2. (53)
Assume now that:(1) Conditions (33) and (34)is satisfied. (2)For a given π > 0, βπ β βπsuch that βπβ β€ π: π
πΊβ²
1 (π, π‘) = π πΊβ² 2(π, π‘).
Letππ,,πΊππ,πΊβ²,π(π), π = 1,2 be a pair of the solutions of the Colombeau- Itoβs SDE (52)-(53) and
letβ±ππ,π‘β²,π(π), π = 1,2be a setβ±ππβ²,π(π) = οΏ½π‘|sup0β€π β€π‘οΏ½ππ,,πΊππ,πΊβ²,π(π)οΏ½ β€ ποΏ½.We let nowπππβ²,π(π) =
supοΏ½π‘|π‘ β β±π,ππβ²,π(π)οΏ½.
(i) ποΏ½ππ,ππ β²,1(π) = ππ,ππ β²,2(π)οΏ½ = 1 and
(ii) π οΏ½sup0β€π β€π1οΏ½ππ,,πΊππ,πΊβ²,1(π) β ππ,,πΊππ,πΊβ²,2(π)οΏ½ = 0οΏ½ = 1.
Proof. A proof of this statement, complete similarly, to a classical case. For example see[15],chapt.2, subsect.6,theorem2.
Let us rewrite now Eq.(50)-Eq.(51) in the next form (with πππ β‘ 1)
οΏ½ππ‘,,ππ₯0,πβ²,π(π, πΏ)οΏ½ πβ²,π = π0 + οΏ½β« π0π‘ πβ²,ποΏ½ππ₯0 ,ππ,,πβ²,π(π, πΏ), ππ,,ππΏ β²,π(π), ποΏ½ πποΏ½ πβ²+ βππΎ(π‘, π), (54) ππ‘,,ππΏ β²,π(π) = οΏ½π’1,π‘,,ππΏ β²,π(π), β¦ , π’π,π‘,,ππΏ β²,π(π)οΏ½, οΏ½π’π,π‘,,ππΏ β²,π(π)οΏ½ πβ² = πποΏ½β« οΏ½π₯π,π,,πβ²,,π π₯0 ,π (π, πΏ)οΏ½2πππ π‘ 0 οΏ½πβ² + βπΏππ(π‘), (55) π = 1, β¦ , π,
Let πΊπ(π), π β βπbea function: (i) πΊπ(π) = πifβπβ β€ π(ii)πΊπ(π) = 0ifβπβ > π.
We set now ππΊπβ²,π(π, π, π‘) = ππΊβ²,π(πΊπ(π), πΊπ(π), π‘). LetοΏ½ππ‘,,ππ₯0,πβ²,π(π, πΏ, π)οΏ½ πβ² = οΏ½οΏ½ππ‘,,πβ²,π π₯0,π (π, πΏ, π)οΏ½ πβ², ππ‘,,πβ²,,π πΏ (π, π)οΏ½
be a family of the solution of the Colombeau-Itoβs SDE: οΏ½ππ‘,,ππ₯0,πβ²,π(π, πΏ, π)οΏ½ πβ² = π0+ οΏ½β« ππβ²,π π οΏ½π π,,ππ₯0 ,πβ²,π(π, πΏ, π), ππ,,ππΏ β²,π(π), π, ποΏ½ π π‘ 0 ποΏ½πβ²+ (56) +βππΎ(π‘, π), ππ‘,,ππΏ β²,π(π, π) = οΏ½π’1,π‘,,ππΏ β²,π(π, π), β¦ , π’π,π‘,,ππΏ β²,π(π, π)οΏ½, οΏ½π’π,π‘,,ππΏ β²,π(π, π)οΏ½ πβ² = πποΏ½β« οΏ½πΊποΏ½π₯π,π,,ππ₯0 ,πβ²,,π(π, πΏ, π)οΏ½οΏ½ 2π ππ π‘ 0 οΏ½ πβ² + +βπΏππ(π‘), (57)
π = 1, β¦ , π,
Definition4.(1) Let οΏ½οΏ½ππ‘,,ππ₯0,πβ²,π(π, πΏ, π)οΏ½
πβ²,ποΏ½π=1 β
be a sequence of the solution of the Colombeau-Itoβs SDE(54)- (55). Letβ±π,ππβ²,π(π, πΏ) be a set
β±π,ππβ²,π(π, πΏ) = οΏ½π‘|sup0β€π β€π‘οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ, π)οΏ½ β€ ποΏ½. (58)
(2) We let now
ππ,ππ β²,π(π, πΏ) = supοΏ½π‘|π‘ β β±π,ππβ²,π(π)οΏ½, (59)
ππ,πββ²,π(π, πΏ) = limπββππ,ππ β²,π(π, πΏ).(60)
(3) LetποΏ½π,,πΊππ,πΊβ²,π(π, πΏ)be a net of the stochastic processes defined
by setting
ποΏ½π,,πΊππ,πΊβ²,π(π, πΏ) = ππ,,πΊππ,πΊβ²,π(π, πΏ, π)iff π‘ < ππΊ,πΊπβ²,π(π, πΏ)(61)
(4) LetοΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½
πΊβ²οΏ½be Colombeaugeneralized stochastic process defined by setting
οΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½
πΊβ²οΏ½ = οΏ½οΏ½ποΏ½π‘,,πβ²,π
ππ,πΊ (π, πΏ)οΏ½
πΊβ²οΏ½. (62)
Remark5.We note that according to the Theorem3 βπ(π β₯ π)one obtain π οΏ½sup0β€π‘β€π
πΊ,,πΊβ²,π
π΅ (π)οΏ½πππ,,πΊπ,πΊβ²,π(π, πΏ, π) β ππ,,πΊππ,πΊβ²,π(π, πΏ, π)οΏ½ > 0οΏ½ = 0,
Therefore definitions (61)-(62) is correct. Definition5.LetοΏ½ππ‘,ππ₯0 ,πβ²,ποΏ½
πΊβ²be a family of the solutions Colombeau-Itoβs SDE (56)-(57).
(1) A familyοΏ½ππ‘,ππ₯0 ,πβ²,ποΏ½
πΊβ², π, πβ²β (0,1],π β (0,1]
οΏ½limπββπ οΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½ > ποΏ½οΏ½
πΊβ² = 0.(63)
Or in the next equivalent form οΏ½ποΏ½ππΊ,πΊββ²,π(π, πΏ) = βοΏ½οΏ½ πΊβ² = 1. (64) (2) A familyοΏ½ππ,πΊππβ² ,πΊ,ποΏ½ πΊβ²is a strongly regular if βπβ², πβ² β [0,1], βπ, π β (0,1] π: οΏ½limπββπ οΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½ > ποΏ½οΏ½ πΊβ² = 0.(65)
or in the next equivalent form: βπβ², πβ²β [0,1], βπ, π β (0,1] π:
οΏ½ποΏ½ππΊ,πΊββ²,π(π, πΏ) = βοΏ½οΏ½
πΊβ² = 1.(66)
Definition6. LetοΏ½ππ‘,ππ₯0 ,πβ²,ποΏ½
πΊβ²bea family of the solutions Colombeau-Itoβs SDE (56)-(57). A
family οΏ½ππ‘,ππ₯0 ,πβ²,ποΏ½ πΊβ², π, π β²β οΏ½0,1], π β (0,1] πisanon-regular if βπ‘β²βπ‘ β₯ π‘β²: οΏ½lim πββπ οΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½ > ποΏ½οΏ½ πΊβ² β 0. (67)
Or in the next equivalent form οΏ½ποΏ½ππΊ,πΊββ²,π(π, πΏ) < βοΏ½οΏ½
πΊβ² = 1. (68)
Proposition2. Assume that Colombeau generalized stochastic process οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½
defined by setting (62) is a strongly regular. Then (1) limπΊβ²β0 πβ0,limπΏβ0π οΏ½οΏ½ππ,,πΊβ²,π ππ,πΊ (π, πΏ) β π π ππ,πΊ(π)οΏ½ποΏ½ = 0 (69.a) (2) βπ > 0: limπΊβ²β0 πβ0,limπΏβ0π οΏ½οΏ½ππ‘,,πΊβ²,π ππ,πΊ (π, πΏ) β π π‘ ππ,πΊ(π)οΏ½ > ποΏ½ = 0. (69.b) Here:ππ‘ππ,πΊ(π) = π π‘,,πΊβ²=0,π=0 ππ,πΊ (π, πΏ = 0).
Proof. Immediately follows fromTheorem3 andTheoremA.1 (see appendix A). Proposition3.LetοΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½
πΊβ² = οΏ½οΏ½ππ‘,πβ²,π π₯0 ,ποΏ½
πβ², οΏ½ππ‘,,πβ²,π
πΏ (π)οΏ½
πβ²οΏ½be a family of the solutions
Colombeau-Itoβs SDE (56)-(57)withππ[π§] β‘ 1.A family οΏ½ππ‘,ππ₯0 ,πβ²,ποΏ½
πΊβ², π, πβ²β (0,1],π β (0,1] πis
regular.
Proof. Assume that: process οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½
πΊβ²is a non-regular. Therefore(π π ,,π{π πΊβ²(π, π) < β})πΊβ² = 1 and consequently οΏ½ππ ,,ποΏ½π ππΊβ²(π,π),,πβ²,π ππ,πΊ (π, πΏ) = βοΏ½οΏ½ πΊβ² > 0. (70)
But the other hand from Eq.(56)-Eq. (57) we obtain
οΏ½ππ πΊβ²(π,π),,πβ²,π π₯0,π (π, πΏ)οΏ½ πβ² = π0 + οΏ½β«0ππΊβ²(π,π)ππβ²,ποΏ½ππ,,ππ₯0 ,πβ²,π(π, πΏ), ππ£,,ππΏ β²,π(π), π£, ποΏ½ ππ£οΏ½ πβ² + + οΏ½βππΎ(ππΊβ²(π, π), π)οΏ½ πβ², (71) ππ£,,ππΏ β²,π(π) = οΏ½π’1,π£,,ππΏ β²,π(π), β¦ , π’π,π£,,ππΏ β²,π(π)οΏ½, οΏ½π’π,π πΊβ²(π,π),,πβ²,π πΏ (π)οΏ½ πβ² = π οΏ½β« οΏ½π₯π,π£,,πβ²,,π π₯0 ,π (π, πΏ)οΏ½2πππ£ ππΊβ²(π,π) 0 οΏ½πβ²+ οΏ½βπΏπποΏ½ππΊβ²(π, π)οΏ½οΏ½πβ², π = 1, β¦ , π, (72)
From (70) and Eq. (71)-Eq. (72) we obtain οΏ½π0,,π₯0οΏ½οΏ½ π πβ²,ποΏ½ππ£,,ππ₯0 ,πβ²,π(π, πΏ), ππ£,,ππΏ β²,π(π), π£, ποΏ½ π ππβ²(π,π) 0 π£ = βοΏ½οΏ½πΊβ² = 0, and therefore οΏ½π0,,π₯0οΏ½π ππΊβ²(π,π),,πβ²,π π₯0,π (π, πΏ) = βοΏ½οΏ½ πβ² = οΏ½π0,,π₯0οΏ½π0+ οΏ½ π πβ²,ποΏ½ππ£,,ππ₯0 ,πβ²,π(π, πΏ), ππ£,,ππΏ β²,π(π), π£, ποΏ½ π ππβ²(π,π) 0 π£ + βππΎ(ππΊ β²(π, π), π) = βοΏ½οΏ½ πΊβ² = 0, οΏ½π0,,0οΏ½π’ π,ππΊβ²(π,π),,πβ²,π πΏ (π) = βοΏ½οΏ½ πβ² = οΏ½ππ ,,ποΏ½π οΏ½ οΏ½π₯ π,π,,πΊππ ,πΊβ²,,π(π, πΏ)οΏ½ ππ ππ£ + βπΏπποΏ½ππΊβ²(π, π)οΏ½ ππΊβ²(π,πΊ) 0 = βοΏ½οΏ½πΊβ² = 0. Thus οΏ½ππ ,,ποΏ½π ππΊβ²(π,π),,πβ²,π ππ,πΊ (π, πΏ) = βοΏ½οΏ½ πΊβ² = 0.
But this is the contradiction. This contradiction completed the proof.
Definition7.CISDE(35)-(36) is β οΏ½-dissipative if there exist
Lyapunov candidate functionοΏ½ππβ²(π, π‘)οΏ½πβ²: β οΏ½πΓ [0, π] β β οΏ½and positive infinite
Colombeau constants πΆοΏ½ = [(πΆπβ²)πβ²] β β οΏ½+,
πΜ = [(ππβ²)πβ²] β β οΏ½+,such that:
(1)βπβ²β (0,1] βΆπ
β,πβ² = limπ ββοΏ½inf βπ±β>π ππβ²(π, π‘)οΏ½ = β, and
οΏ½οΏ½πΜπβ²(ππβ², π‘; ππβ²)οΏ½ πβ²οΏ½ β€ πΆΜ οΏ½οΏ½ππβ²(ππβ², π‘)οΏ½πβ²οΏ½ (73) is satisfied.Here οΏ½οΏ½πΜπβ²(ππβ², π‘; ππβ²)οΏ½ πβ²οΏ½ β‘ οΏ½οΏ½πππβ²(πππ‘πβ², π‘)οΏ½ πβ²οΏ½ + οΏ½ οΏ½οΏ½ πππβ²(ππβ², π‘) ππ₯π,πβ² ππ,πβ²(ππβ², π‘)οΏ½ πβ² οΏ½ π π=1 . (74)
Or in the next equivalent form:
CISDE (35)-(36) is β οΏ½-dissipative if there existLyapunovcandidate
functionοΏ½ππβ²(π, π‘)οΏ½πβ²: β οΏ½πΓ [0, π] β β οΏ½and positive infinite Colombeau constants πΆοΏ½ =
[(πΆπβ²)πβ²] β β οΏ½+,
πΜ = [(ππβ²)πβ²] β β οΏ½+, such that:
(1)βπβ²β (0,1] βΆπ
β,πβ² = limπ ββοΏ½inf βπ±β>π ππβ²(π, π‘)οΏ½ = β, and
(2β²)βπβ²β (0,1]βπ πβ²[(ππβ² β βπ) β§ (βππβ²β β₯ ππβ²)]the inequality οΏ½πΜπβ²(ππβ², π‘; ππβ²)οΏ½ πβ² β€ ((πΆπβ²)πβ²)οΏ½ππβ²(ππβ², π‘)οΏ½πβ² (75) is satisfied. Here οΏ½πΜπβ²(π₯πβ², π‘; ππβ²)οΏ½πβ² β‘ οΏ½πππβ²ππ‘οΏ½ππβ²,π‘οΏ½οΏ½ πβ² + οΏ½β πππβ²οΏ½ππβ²,π‘οΏ½ ππ₯πβ² ππ,πβ²(ππβ², π‘) π π=1 οΏ½ πβ². (76)
Definition 8. CISDE (35)-(36) isa stronglyβ οΏ½-dissipative if
Lyapunov candidate functionοΏ½ππβ²(π, π‘)οΏ½
πβ²: β οΏ½πΓ [0, π] β β οΏ½,
πβ² β [0,1]andpositive finite Colombeau constants
πΆοΏ½ = [(πΆπβ²)πβ²] β β οΏ½+, πΜ = [(ππβ²)πβ²] β β οΏ½+,such that:
(1)βπβ²β (0,1] βΆπ
β,πβ² = limπββοΏ½inf βxβ>πππβ²(π, π‘)οΏ½ = β,and(2)β[(ππβ²)πβ²]([(βππβ²β)πβ²] β₯ πΜ) the
inequality
οΏ½οΏ½πΜπβ²(ππβ², π‘; ππβ²)οΏ½
is satisfied. Here οΏ½οΏ½πΜπβ²(ππβ², π‘; ππβ²)οΏ½πβ²οΏ½ β‘ οΏ½οΏ½πππβ²(πππ‘πβ², π‘)οΏ½ πβ² οΏ½ + οΏ½ οΏ½οΏ½πππππ₯β²(ππβ², π‘) π,πβ² ππ,πβ²(ππβ², π‘)οΏ½ πβ² οΏ½ π π=1 . (78) Proposition4.LetοΏ½ππ,,πΊππ,πΊβ²,π(π)οΏ½
πΊβ²) be generalized stochastic process satisfying
Colombeau-Itoβs SDE(35)-(36) on the time interval[π , π]and οΏ½ππΊβ²,π(π, π)οΏ½
πΊβ²- is a
generalizedrandomvariableequaltothetimeatwhichthesamplefunctionofthegeneralized process οΏ½ππ,,πΊππ,πΊβ²,π(π)οΏ½
πΊβ²) first leaves the bounded neighborhood π, and
letοΏ½ππΊβ²,π(π, π‘, π)οΏ½πΊβ² = οΏ½minοΏ½ππΊβ²,π(π, π), π‘οΏ½οΏ½πΊβ².Supposemoreover
thatβπβ²β (0,1] βΆπ οΏ½π π,,πΊβ²,π ππ,πΊ (π) β ποΏ½ = 1.Then οΏ½π οΏ½ππβ²οΏ½ππππΊβ²,ππ,πΊ(π,π‘,π),,πΊβ²,π(π), ππΊβ²,π(π, π‘, π)οΏ½ β ππβ²οΏ½ππ,,πΊππ,πΊβ²,π(π), π οΏ½οΏ½οΏ½ πβ² = = οΏ½π οΏ½οΏ½ πΜπβ²οΏ½ππ’,,πΊππ,πΊβ²,π(π), π’οΏ½ ππ’ ππΊβ²,π(π,π‘,π) π οΏ½οΏ½πβ².
Proof. Similarly as the proof of the corresponding classical result, see [17]Lemma 3.2. Theorem4. (1) Assume that: (i)for CISDE (35)-(36) the inequalities(33)and(34) is satisfied and (ii)CISDE (35)-(36) isβ οΏ½-dissipative.
Then (1) Colombeau generalized stochastic process οΏ½ππ,,πΊππ,πΊβ²,π(π)οΏ½ πΊβ², π
β²β (0,1] , π β
(0,1] πdefined by setting (42) is regular, and (2) the inequality οΏ½π οΏ½ππβ²οΏ½ππ,,πΊππ,πΊβ²,π(π), π‘οΏ½οΏ½οΏ½
πβ² β€ οΏ½π οΏ½ππβ²οΏ½πππ,,πΊβ²,π ππ,πΊ (π), π‘
0οΏ½οΏ½οΏ½πβ²exp[(πΆπβ²)πβ²(π‘ β π‘0)](79)
is satisfied.
Proof.(1) From (76) it follows that the Colombeau generalized functionοΏ½ππβ²(ππβ², π‘)οΏ½ πβ² =
οΏ½ππβ²(ππβ², π‘)οΏ½
Proposition4,forοΏ½ππΊβ²,π(π, π‘, π)οΏ½ πΊβ² = οΏ½minοΏ½ππΊβ²,π(π, π), π‘οΏ½οΏ½πΊβ²wehave οΏ½π οΏ½ππβ²οΏ½ππππΊβ²,,π,ππ,πΊ (π,π‘,π),,πΊβ²(π), ππΊβ²,π(π, π‘, π)οΏ½ expοΏ½β(πΆπβ²)πβ²οΏ½ππΊβ²,π(π, π‘, π) β π‘0οΏ½οΏ½οΏ½οΏ½ πβ² β οΏ½π οΏ½ππβ²οΏ½ππ π,,πΊβ²,π ππ,πΊ (π), π‘ 0οΏ½οΏ½οΏ½πβ² = οΏ½π οΏ½β«πππΊβ²,π(π,π‘,π)πΜπβ²οΏ½ππ’,,πΊππ,πΊβ²,π(π), π’οΏ½ ππ’ π οΏ½οΏ½πβ² β€ 0. (80)
This, together with the inequalitiesοΏ½ππΊβ²,π(π, π‘, π)οΏ½
πΊβ² β€ π‘, οΏ½ππβ²(ππβ², π‘)οΏ½ πβ² β₯ 0, implies οΏ½π οΏ½ππβ²οΏ½ππ πΊβ²,π,π(π,π‘),,πΊβ² ππ,πΊ (π), π πΊβ²,π(π, π‘, π)οΏ½οΏ½οΏ½ πβ² β€ οΏ½π οΏ½ππβ²οΏ½ποΏ½π π,π,πΊβ² ππ,πΊ (π), π‘ 0οΏ½οΏ½οΏ½πβ²exp[(πΆπβ²)πβ²(π‘ β π‘0)] (81)
From (81) one derive the estimate
οΏ½ποΏ½ππΊβ²,π(π, π) < π‘οΏ½οΏ½πβ² β€ β€ exp[(πΆπβ²)πβ²(π‘ β π‘0)] οΏ½π οΏ½ππβ²οΏ½ποΏ½ππ,,πΊβ² ππ,πΊ (π), π‘ 0οΏ½οΏ½οΏ½πβ² οΏ½infβπ±ββ₯n,u>π‘0ππβ²(π, π’)οΏ½ πΊβ²
Letting π β β and making use of theDefinition7we now get(64). (2)Assume that CISDE (35)-(36) is a strongly
β
οΏ½-dissipative. Then(1) Colombeau generalized stochastic process = οΏ½οΏ½ππ,,πΊππ,πΊβ²(π)οΏ½ πΊβ²οΏ½ π
β²β
[0,1], π β [0,1],πdefined by setting (42) is a strongly regular and (2) the inequality
οΏ½π οΏ½ππβ²οΏ½ππ,,πΊππ,πΊβ²(π), π‘οΏ½οΏ½οΏ½ πβ² β€ β€ οΏ½π οΏ½ππβ²οΏ½ππ π,,πΊβ² ππ,πΊ (π), π‘ 0οΏ½οΏ½οΏ½πβ²exp[(πΆπβ²)πβ²(π‘ β π‘0)](82) is satisfied.
Theorem5.Weset nowπππ[π] β‘ 1, π = 1, β¦ , π. For any solution
οΏ½ππ‘,,ππ₯0,πβ²,π(π, πΏ)οΏ½
πβ² = οΏ½π₯1,π‘,,πβ²,π π₯0,π , β¦ , π₯
π,π‘,,ππ₯0,πβ²,ποΏ½πβ²
ofa stronglyβοΏ½βdissipative CISDE(46)-(48) and anyβ-valued parametersπ1, β¦ , ππ, there
exist finite Colombeau constant πΆοΏ½β²= οΏ½οΏ½πΆ
πβ²β²οΏ½πβ²οΏ½ > 0, such that βπ[π = (π1, β¦ , ππ)], the
inequality lim πβ0 πβ²β0 πβ0 οΏ½πβ²ποΏ½β0 limπΏβ0πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ) β ποΏ½ π οΏ½ β€ πΆΜβ²βπΌ(π‘, π)βπ (83)
is satisfied. Or in the next equivalent form: for a sufficiently smallπ β 0and for a sufficiently small π β 0, πβ²β 0such that
πβ² π β 0,the inequality οΏ½οΏ½limπΏβ0πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ) β ποΏ½ π οΏ½οΏ½ πβ²οΏ½ β€ πΆΜ β²βπΌ(π‘, π)β2 (84) is satisfied.
Here thevector-function πΌ(π‘, π) = (π1(t,π),β¦,ππ(t,π))is the solutionof the differential
master equation:
πΌΜ(π‘, π)= π[ππ(π, π‘)]πΌ(π‘, π) + ππ(π, π‘), πΌ(0, π) = ππβ π, (85)
Hereπ = π[ππ(π, π‘)] is a Jacobian i.e.,Jis π Γ π-matrix:
π[ππ(π, π‘)]= JοΏ½πππ,π(π, π‘)/ππ₯ποΏ½π=π. (86)
Proof. We let now
ππ‘,,ππ₯0,πβ²,πβ π = ππ,,πΊππ,πΊβ²,π. (86)
οΏ½πππ‘,,ππ₯0,πβ²,π(π, πΏ)οΏ½ πβ² = οΏ½ππβ²,ποΏ½ππ‘,,πβ²,π π₯0 ,π (π, πΏ) + π, π π‘,,πβ²,π,π πΏ (π), π‘, ποΏ½οΏ½ πβ² + +βπππΎ(π‘, π), ππ‘,,πβ²,π,π πΏ (π) = οΏ½π’1,π‘,,ππΏ β²,π,π(π), β¦ , π’π,π‘,,ππΏ β²,π,π(π)οΏ½, (87) οΏ½ππ’π,π‘,,ππΏ β²,π,π(π)οΏ½ πβ² = π οΏ½οΏ½π₯π,π‘,,πβ²,,π,π π₯0 ,π (π, πΏ)οΏ½2ποΏ½ πβ² + βπΏπππ(π‘), π = 1, β¦ , π, οΏ½π₯0,ππ₯0,πβ²,ποΏ½ πβ² = π0 β π οΏ½π, π‘ β [0, π], π, πβ², π, πΏ β (0,1].
Thus we need to estimate the quantity lim πβ0 πβ²β0 πβ0 οΏ½πβ²ποΏ½β0 limπΏβ0πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½ π οΏ½.
Application of the Theorem B.4 (see Appendix B) to Eq.(87) gives the inequality (83) directly.
Theorem 6.(Strong large deviations principle) [5],[7]. Assume that CISDE (35)-(36) is a stronglyβ οΏ½-dissipative. Then:
(1) For any solution
οΏ½ππ‘,,ππ₯0,πβ²,π(π)οΏ½
πβ² = οΏ½π₯1,π‘,,πβ²,π π₯0,π , β¦ , π₯
π,π‘,,ππ₯0,πβ²,ποΏ½ πβ²
of a stronglyβοΏ½βdissipative CISDE(35)-(40) and anyβ-valued parametersπ 1, β¦ , ππ, there
exist finite Colombeau constant πΆοΏ½β²= οΏ½οΏ½πΆ
πβ²β²οΏ½πβ²οΏ½ > 0, such that βπ[π = (π1, β¦ , ππ)]the
inequality lim πβ0 πβ²β0 πβ0 οΏ½πβ²ποΏ½β0 πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π) β ποΏ½ π οΏ½ β€ πΆΜβ²βπΌ(π‘, π)βπ(8 8)
is satisfied. Or in the next equivalent form: for a sufficiently small π β 0 and for a sufficiently small π β 0, πβ²β 0 such that
πβ²/π β 0,the inequality οΏ½οΏ½πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π) β ποΏ½ π οΏ½οΏ½ πβ²οΏ½ β€ πΆΜ β²βπΌ(π‘, π)β2.
is satisfied.
(2) For any solution
οΏ½ππ‘,,ππ₯0,πβ²,π(π)οΏ½
πβ² = οΏ½π₯1,π‘,,πβ²,π π₯0,π , β¦ , π₯
π,π‘,,ππ₯0,πβ²,ποΏ½πβ²
of a stronglyβοΏ½βdissipative CISDE(35)-(40) andanyβ-valued parametersπ 1, β¦ , ππ, there
exist finite Colombeauconstant πΆοΏ½β²= οΏ½οΏ½πΆ
πβ²β²οΏ½πβ²οΏ½ > 0, such that βπ[π = (π1, β¦ , ππ)]the
inequality
limπβ0πποΏ½οΏ½ππ‘,,πΊππ,πΊβ²=0,π=0(π) β ποΏ½ π
οΏ½ β€ πΆΜβ²βπΌ(π‘, π)βπ (89)
issatisfied.Here the vector-function πΌ(π‘, π) = (π1(t,π),β¦,ππ(t,π))is the solutionof the
differential master equation:
πΌΜ(π‘, π)= π[ππ(π, π‘)]πΌ(π‘, π) + ππ(π, π‘), πΌ(0, π) = ππβ π, (90)
whereπ = π[ππ(π, π‘)]is aJacobian i.e.,Jis π Γ π-matrix:
π[ππ(π, π‘)]= JοΏ½πππ,π(π, π‘)/ππ₯ποΏ½π=π. Proof1.Fromtheequality πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π) β ποΏ½ π οΏ½ = πποΏ½οΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π) β πππ,πΊπ,,πΊβ²,π(π, πΏ)οΏ½ + + οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ) β ποΏ½οΏ½ π οΏ½, by using the triangle inequality, one obtain
οΏ½πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π) β ποΏ½ π οΏ½ β€ οΏ½πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π) β ππ,,πΊππ,πΊβ²,π(π, πΏ)οΏ½ π οΏ½ + +οΏ½πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π, πΏ) β ποΏ½ π
οΏ½.Therefore statement (1) immediately follows fromTheoremA1 (see appendix A), Proposition 2andTheorem5.
πποΏ½οΏ½ππ,,πΊππ,πΊβ²=0,π=0(π) β ποΏ½ π
οΏ½ = πποΏ½οΏ½οΏ½ππ,,πΊππ,πΊβ²=0,π=0(π) β ππ,,πΊππ,πΊβ²,π(π)οΏ½ + οΏ½ππ,,πΊππ,πΊβ²,π(π) β ποΏ½οΏ½ π
οΏ½, byusingthetriangle inequality, one obtain
οΏ½πποΏ½οΏ½ππ,,πΊππ,πΊβ²=0,π=0(π) β ποΏ½ π οΏ½ β€ οΏ½πποΏ½οΏ½πππ,,πΊπ,πΊβ²=0,π=0(π) β ππ,,πΊππ,πΊβ²,π(π)οΏ½ π οΏ½ +οΏ½πποΏ½οΏ½ππ,,πΊππ,πΊβ²,π(π) β ποΏ½ π
οΏ½.Therefore statement (2) immediately follows fromTheoremA1 (see appendix A), Proposition 1andstatement (1).
Remark.5. We note that in general case the inequality οΏ½οΏ½πΏπβ²(π‘)οΏ½
πβ²οΏ½ β‘ οΏ½οΏ½limΞ΅β0πΞ©οΏ½οΏ½ππ,,πΊππ,πΊβ²(π) β ππ,,πΊππ,πΊ=0β² οΏ½ π
οΏ½οΏ½
πβ²οΏ½ β 0
is satisfied, see Example 1. Example 1.Figures 1-2. π₯Μπ‘π₯0 ,π = βπ β οΏ½π₯
π‘π₯0 ,ποΏ½3β π β οΏ½π₯π‘π₯0 ,ποΏ½2β π β π₯π‘π₯0 ,πβ π β π‘π β (91)
βπ β π‘πβ sin(Ξ© β π‘π) + βππ€(π‘) , π₯
0π₯0 ,π = π₯0.
From Eq.(91) and general differential master equation (90) one obtain the next linear differential master equation:
π’Μ(π‘) = β(3ππ2+ 2ππ + π)π’(π‘) β (π β π3+ π β π2+ π β π) β (92)
βπ β π‘πβ π β π‘πβ sin(Ξ© β π‘π) , π’(0) = π₯ 0.β π.
From the differential master equation (92) one obtain the transcendental master equation:
Figure 1.The solution of the Equation (8) in a comparison with a corresponding solution π(π) of the ODE (10).
Figure 2. πΏ(r)versusR.
οΏ½π₯0 β π(π‘)οΏ½expοΏ½βοΏ½3π β π2(π‘) + 2π β π(π‘)οΏ½ β π‘οΏ½ β
β β« [π β ππ‘ π+ π β ππβ sin(Ξ© β ππ) + π β π3(π‘) + π β π2(π‘)]
0 Γ (93)
Γ expοΏ½βοΏ½3π β π2(π‘) + 2π β π(π‘)οΏ½ β (π‘ β π)οΏ½ππ = 0.
Example 1.Numerical simulation: Figures 1 and 2.
π = 1, π = 5, π = 1, π = π = β2, π = π = π = 2, πΊ = 5, π₯0. = 0, π = 5, π = π/0.001. πΏ(π) = limπβ0π οΏ½οΏ½π₯π‘π₯0 ,πβπ₯π‘π₯0 ,π=0οΏ½ 2 οΏ½. (94) π₯Μπ‘0 = βπ(π₯π‘0)3β π(π₯π‘0)2β ππ₯π‘0 β π β π‘πβ π β π‘πβ sin(Ξ© β π‘π).(95)
Let β = (Ξ©, πΊ, π)be a probability space. Let us consider now
mβpersons Colombeau-Itoβs stochastic differential gameCIDGπ;π(π, π, π, πΊπ(βπ), β)with
nonlinear dynamics: οΏ½πΜπ‘,ππ₯0,πβ²(π)οΏ½ πβ² = οΏ½ππβ²οΏ½ππ‘,πβ² π₯0,π(π), πΆ(π‘), π‘οΏ½οΏ½ πβ²+ βποΏ½π(π‘, π)οΏ½πβ²(96) Hereπ, πβ²β (0,1], π βͺ 1; βπ‘ β [0, π]: οΏ½π πβ²(π‘)οΏ½ πβ² β π οΏ½π; π0,ππ₯0,πβ²(π) = π0 β βπ, ππβ² = οΏ½ππΊβ²,π, β¦ , ππΊβ²,ποΏ½, π = [(ππβ²)πβ²], π = [(ππβ²)πβ²]; π(π₯,β,β), π(π₯,β,β) β πΊπ(βπ), πΆ(π‘) = {πΌ 1(π‘), β¦ , πΌπ(π‘)}; πΌπ(π‘) β ππ β βππ, π = 1, β¦ , π.
Hereπ‘ βΌ πΌπ(π‘), is the control chosen by the i-th player, within a set of admissible
i-th player is οΏ½πΜ ππβ²,ποΏ½ πβ²=π οΏ½οΏ½β« π0π π2β²,ποΏ½π₯π‘,ππ₯0,πβ² (π), πΆ(π‘), π‘οΏ½ππ‘οΏ½ πβ²οΏ½ + (97) +π οΏ½οΏ½οΏ½ οΏ½π₯π,ππ₯,πβ²;π(π) β π¦ποΏ½ 2 π π=1 οΏ½πβ²οΏ½.
Definition 9.CIDGπ;π(π, π, π, πΊπ(βπ), β) (96)-(97) is a strongly βοΏ½βdissipative if
CISDE(96) is a stronglyβοΏ½βdissipative. Theorem.7. Suppose that:
(1)CIDGπ;π(π, π, π, πΊπ(βπ), β)(96)-(97) isa stronglyβ οΏ½βdissipative,
(2)π0(β, πΆ, π‘) β‘ ππΊβ²=0(β, πΆ, π‘): βπ β βπis a polynomial
on a variableπ = (π₯1, β¦ , π₯π)and a linear function on a
variableπΆ(π‘) = {πΌ1(π‘), β¦ , πΌπ(π‘)}i.e.,ππ,π(π, πΆ, π‘) =
βπ,|π|β€πππ,ππ (π‘)ππ+ βπ=ππ ππ.π(π‘)πΌπ(π‘), π = (ππ, β¦ , ππ), |π| = β πππ=π π , 0 β€ ππ β€ π,
(3)π0(β, πΆ, π‘) β‘ ππΊβ²=0(β, πΆ, π‘): βπ β βπis a polynomial
on a variableπ = (π₯1, β¦ , π₯π)and a linear function on a
variableπΆ(π‘) = {πΌ1(π‘), β¦ , πΌπ(π‘)}i.e.,ππ,π(π, πΆ, π‘) =
βπ,|π|β€πππ,ππ (π‘)ππ+ βπ=ππ ππ,π(π‘)πΌπ(π‘), π = (ππ, β¦ , ππ), |π| = β πππ=π π , 0 β€ ππ β€ π.
Then For any solution
οΏ½ππ‘,,ππ₯0,πβ²; πΆοΏ½(π‘)οΏ½ = οΏ½οΏ½π₯1,π‘,,ππ₯0,πβ², β¦ , π₯π,π‘,,ππ₯0,πβ²οΏ½ ; {πΌ1(π‘), β¦ , πΌπ(π‘)}οΏ½(98)
of theCIDGπ;π(π, π, π, πΊπ(βπ), β) (96) -(97) and any β-valued
parameterπ οΏ½π(1), π(2)οΏ½ = οΏ½οΏ½π 1 (1), β¦ , π π (1)οΏ½, οΏ½π 1 (2), β¦ , π π
(2)οΏ½οΏ½there exist finite Colombeau
constant πΆΜβ²= οΏ½οΏ½πΆ
πβ²β²οΏ½πβ²οΏ½ > 0, such thatβποΏ½π = οΏ½π(π), π(π)οΏ½οΏ½ the inequalities
(1)lim πβ0 πβ²β0 οΏ½πβ²ποΏ½β0 πποΏ½οΏ½ππ,,πΊππ,πΊβ²(π) β π(π)οΏ½ π οΏ½ β€ πΆΜβ²οΏ½πΌοΏ½π‘, π(π)οΏ½οΏ½π, (2) lim πβ0 πβ²β0 οΏ½πβ²ποΏ½β0 πποΏ½οΏ½πππ,,πΊπ,πΊβ²(π) β π(π)οΏ½ π οΏ½ β€ πΆΜβ²οΏ½π½οΏ½π, π(π)οΏ½οΏ½π,(99.a)
(3) lim πβ0 πβ²β0 οΏ½πβ²ποΏ½β0 πποΏ½οΏ½πππ,,πΊπ,πΊβ²(π) β ποΏ½ π οΏ½ β€ πΆΜβ²βπ½(π, π)βπ, (4)lim πβ0 πβ²β0 οΏ½πβ²ποΏ½β0 οΏ½πΜ ππβ²,ποΏ½ β€ |ππ(π, π, 0)| + βπΌ(π, π)β2,π = 1, β¦ , π, (5)limπβ0πποΏ½οΏ½ππ,,πΊππ,πΊβ²=0(π) β π(π)οΏ½ π οΏ½ β€ πΆΜβ²οΏ½πΌοΏ½π‘, π(π)οΏ½οΏ½π, (6)π₯π’π¦πΊβππποΏ½οΏ½πππ,,πΊπ,πΊβ²=0(π) β π(π)οΏ½ π οΏ½ β€ πͺοΏ½β²οΏ½π½οΏ½π, π(π)οΏ½οΏ½π, (7) limπβ0πποΏ½οΏ½πππ,,πΊπ,πΊβ²=0(π) β ποΏ½ π οΏ½ β€ πΆΜβ²βπ½(π, π)βπ, (8)limπβ0οΏ½πΜ ππβ²=0,ποΏ½ β€ |ππ(π, π, 0)| + βπΌ(π, π)β2,π = 1, β¦ , π
issatisfied.Or in the next equivalent form: for a sufficiently small π β 0, πβ² β 0such that πβ² π β 0,the inequalities (1)οΏ½οΏ½lim infπβ0πποΏ½οΏ½ππ,,πΊππ,πΊβ²(π) β π(π)οΏ½ π οΏ½οΏ½ πβ²οΏ½ β€ πΆΜ β²οΏ½πΌοΏ½π‘, π(π)οΏ½οΏ½2, (2)οΏ½οΏ½lim infπβ0πποΏ½οΏ½πππ,,πΊπ,πΊβ²(π) β π(π)οΏ½ π οΏ½οΏ½ πβ²οΏ½ β€ πΆΜ β²βπ½(π, π)β2, (99.b) (3)οΏ½οΏ½lim infπβ0ππ΄οΏ½οΏ½ππ»,πΊπ,πΊβ²(π) β ποΏ½ π οΏ½οΏ½ πβ²οΏ½ β€ πΆΜ β²βπΌ(π, π)β2. (4) οΏ½πΜ ππβ²,ποΏ½ πβ² β€ |ππ(π, π, 0)| + βπΌ(π, π)β 2,π = 1, β¦ , π issatisfied.Here πππ,,πΊπ,πΊβ²(π) = β« π0π πΊβ²οΏ½πππ»,πΊπ,πΊβ²(π), πΆ(π‘), π‘οΏ½ ππ‘. Here a functionπΎ(π‘, π) = {πΌ(π‘, π), π½(π‘, π)}π = οΏ½οΏ½π1(π‘, π), β¦ , ππ(π‘, π)οΏ½; οΏ½π1(π‘, π), β¦ , ππ(π‘, π)οΏ½οΏ½πis the solution
πΎΜ(π‘, π)= ποΏ½ποΏ½0(π, π‘)οΏ½πΎ(π‘, π) + ποΏ½0(π, π‘) + β©π (π‘), πΆοΏ½π(π‘)βͺ,(100)
πΌ(0, π) = ππβ π, π½(0, π) = 0.
And with the playoff of thei-th player is:
πΜπ’ = |ππ(π, π, π)| + βπΌ(π, π)β2.(101)
Here
ποΏ½0(π, π‘) = {π(π, 0, π‘); π(π, 0, π‘)}t(102)
and
π = ποΏ½ποΏ½π(π, π‘)οΏ½(103)
isJacobian i.e., J is (π + π) Γ (π + π)-matrix: ποΏ½ποΏ½π(π, π‘)οΏ½= JοΏ½πποΏ½π,π(π, π‘)/ππ₯ποΏ½π=π.(104)
Proof. Let us rewrite Eqs.(96)-(97) of the next equivalent form (i) οΏ½πΜπ‘,ππ₯0,πβ²(π)οΏ½
πβ² = οΏ½ππβ²οΏ½ππ‘,πβ²
π₯0,π(π), πΆ(π‘), π‘οΏ½οΏ½
πβ²+ +βποΏ½π€(π‘, π)οΏ½πβ² (105)
(ii) (ii) πΜπ‘,,πΊπ0,πΊβ²(π) = πππΊβ²οΏ½ππ»,πΊπ0,πΊβ²(π), πΆ(π‘), π‘οΏ½ + +βπΊοΏ½π€(π‘, π)οΏ½πΊβ².(106)
Then the playoff of the i-th player is οΏ½πΜ ππβ²,ποΏ½ πβ²=π οΏ½οΏ½π§π,,πΊβ²,π π0,πΊ (π)οΏ½ πβ²οΏ½ + π οΏ½οΏ½β οΏ½π₯π,πβ²;π π₯,π (π) β π¦ ποΏ½ 2 π π=1 οΏ½ πβ²οΏ½. Here π§π‘,,πΊπ0,πΊβ²,π(π) = β« π0π‘ πΊβ²,ποΏ½πππ»,πΊ0,πΊβ²(π), πΆ(π‘), π‘οΏ½ ππ‘, π0 = 0.
The inequalities (99) immediately follow from Eq.(105), Eq.(106),Theorem 6and definitions.
with a small white noise.
(1)π₯Μ1 = π₯2, π₯Μ2 = βππ₯23+ πΌ1(π‘) + πΌ2(π‘) + βππ€(π‘, π); π > 0, (107)
π‘ β [0, π], π₯1 (0) = π₯10, π₯2 (0) = π₯20 ; π βͺ 1;
(2) πΌ1(π‘) β [βπ1, π1], πΌ2(π‘) β [βπ2, π2];
(3)π½π = π₯12(π), π = 1,2.
Optimal control problem for the first player is: min
πΌ1(π‘) β [βπ1, π1] οΏ½
max πΌ2(π‘) β [βπ2, π2]
[π₯12(π)]οΏ½(108)
and optimal control problem for the second player is: max
πΌ2(π‘) β [βπ2, π2] οΏ½
min πΌ1(π‘) β [βπ1, π1]
[π₯12(π)]οΏ½. (109)
Using Equation (100) one obtained the corresponding linear master game: (1)π’Μ1 = π’2, π’Μ2 = β3ππ22π’2β ππ23 + πΌοΏ½1(π‘) + πΌοΏ½2(π‘), (110)
π’1 (0) = π₯10β π1, π₯2 (0) = π₯20β π2 ;
(2)πΌοΏ½1(π‘) β [βπ1, π1], πΌοΏ½2(π‘) β [βπ2, π2];
(3)π±π= π’12(π), π = 1,2.
Optimal control problem for the first player is: min
πΌοΏ½1(π‘) β [βπ1, π1] οΏ½
max πΌοΏ½2(π‘) β [βπ2, π2]
[π’12(π)]οΏ½, (111)