MALLIAVIN CALCULUS FOR FRACTIONAL DELAY EQUATIONS

Description

Niveau: Supérieur, Doctorat, Bac+8
MALLIAVIN CALCULUS FOR FRACTIONAL DELAY EQUATIONS JORGE A. LEON AND SAMY TINDEL Abstract. In this paper we study the existence of a unique solution to a general class of Young delay differential equations driven by a Holder continuous function with parameter greater that 1/2 via the Young integration setting. Then some estimates of the solution are obtained, which allow to show that the solution of a delay differential equation driven by a fractional Brownian motion (fBm) with Hurst parameter H > 1/2 has a C∞-density. To this purpose, we use Malliavin calculus based on the Frechet differentiability in the directions of the reproducing kernel Hilbert space associated with fBm. 1. Introduction The recent progresses in the analysis of differential equations driven by a fractional Brownian motion, using either the complete formalism of the rough path analysis [3, 10, 18], or the simpler Young integration setting [25, 33], allow to study some of the basic properties of the processes defined as solutions to rough or fractional equations. This global program has already been started as far as moments estimates [13], large deviations [16], or properties of the law [2, 21] are concerned. It is also natural to consider some of the natural generalizations of diffusion processes, arising in physical applications, and see if these equations have a counterpart in the fractional Brownian setting.

• algebraic structure


• then


• stochastic differential


• brownian motion


• a1 ?


• young integration


• differential equations driven


• rn-valued random variable


• standard brownian

Sujets

• Algebraic structure
• Stochastic differential equation
• Brownian motion

Informations

• Publié par: profil-zyak-2012
• Nombre de lectures: 11
• Langue: English

Extrait

MALLIAVIN CALCULUS FOR FRACTIONAL DELAY EQUATIONS

´ JORGE A. LEON AND SAMY TINDEL

Abstract.In this paper we study the existence of a unique solution to a general class of YoungdelaydifferentialequationsdrivenbyaHo¨ldercontinuousfunctionwithparameter greater that 1/ Then2 via the Young integration setting. some estimates of the solution are obtained, which allow to show that the solution of a delay differential equation driven by a fractional Brownian motion (fBm) with Hurst parameterH >1/2 has aC ∞ -density. Tothispurpose,weuseMalliavincalculusbasedontheFre´chetdifferentiabilityinthe directions of the reproducing kernel Hilbert space associated with fBm.

1.Introduction

The recent progresses in the analysis of differential equations driven by a fractional Brownian motion, using either the complete formalism of the rough path analysis [3, 10, 18], or the simpler Young integration setting [25, 33], allow to study some of the basic properties of the processes defined as solutions to rough or fractional equations. This global program has already been started as far as moments estimates [13], large deviations [16], or properties of the law [2, 21] are concerned. It is also natural to consider some of the natural generalizations of diffusion processes, arising in physical applications, and see if these equations have a counterpart in the fractional Brownian setting. Some partial developments in this direction concern pathwise type PDEs, such as heat [7, 11, 12, 30], wave [28] or Navier-Stokes [4] equations, as well as Volterra type systems [5, 6]. As we shall see, the current paper is part of this second kind of project, and we shall deal with stochastic delay equations driven by a fractional Brownian motion with Hurst parameter H >1/2. Indeed, we shall consider in this article an equation of the form: dy t =f(Z t y )dB t +b(Z t y )dt, t∈[0, T],(1) whereBis ad-dimensional fractional Brownian motion with Hurst parameterH >1/2, f:C 1 γ ([−h,0];R n )→R n×d andb:C 1 γ ([−h,0];R n )→R n satisfy some suitable regularity conditions,C 1 γ designates the space ofγ-H¨lder continuous functions of one variable (see o Section 2.1 below) andZ y t : [−h,0]→R n is defined byZ t y (s) =y t+s . In the previ- ous equation, we also assume that an initial conditionξ∈ C 1 γ is given on the interval [−h,0]. Notice that equation (1) is a slight extension of the typical delay equation which

Date: December 7, 2009. 2000Mathematics Subject Classification.60H10, 60H05, 60H07. Key words and phrases.Delay equation, Young integration, fractional Brownian motion, Malliavin calculus. J.Le´onispartiallysupportedbytheCONACyTgrant98998.S.Tindelispartiallysupportedbythe ANR grant ECRU. 1

2

´ JORGE A. LEON AND SAMY TINDEL

is obtained for some functionsfandbof the following form: f:C 1 γ ([−h,0];R n )→R n×d ,withf(Z y t ) =σ Z − 0 h y t+θ ν(dθ)  ,(2) for a regular enough functionσ, and a finite measureνon [−h, special case of0]. This interest will be treated in detail in the sequel. Our considerations also include a function fdefined byf(Z t y ) =σ(Z y t (−u 1 ), . . . ,Z t y (−u k )) for a givenk≥1, 0≤u 1 < . < u . . k ≤h and a smooth enough functionσ:R n×k →R n×d . The kind of delay stochastic differential system described by (1) is widely studied when driven by a standard Brownian motion (see [20] for a nice survey), but the results in the fractional Brownian case are scarce: we are only aware of [8] for the caseH >1/2 andf(Z y ) =σ(Z y (−r)),0≤r≤hand the further investigation [9] which establishes, a continuity result in terms of the delayr . Asfar as the rough case is concerned, an existence and uniqueness result is given in [22] for a Hurst parameterH >1/3, and [31] extends this result toH >1/can be thus seen as a step in the study4. The current article of processes defined as the solution to fractional delay differential systems, and we shall investigate the behavior of the density of theR n -valued random variabley t for a fixed t∈(0, T], whereyis the solution to (1). specifically, we shall prove the following More theorem, which can be seen as the main result of the article:

Theorem 1.1.Assumefandbehdt´rcetavirevianyoesof.LetrdervehadFdeunbo t∈(0, T]be an arbitrary time, andybe the unique solution to (1) inC 1 κ (R n ), for a given1/2< κ < H. Then the law ofy t is absolutely continuous with respect to Lebesgue ∞ measure inR n , and its density is aC-function.

Notice that this kind of result, which has its own interest as a natural step in the study of processes defined by delay systems, is also a useful result when one wants to evaluate the convergence of approximation schemes in the fractional Brownian context. We plan to report on this possibility in a subsequent communication. Let us say a few words about the strategy we shall follow in order to get our Theorem 1.1. First of all, as mentioned before, there are not too many results about delay systems governed by a fractional Brownian motion. In particular, equation (1) has never been considered (to the best of our knowledge) with such a general delay dependence. We shall thus first show how to define and solve this differential system, by means of a slight variation of the Young integration theory (called algebraic integration), introduced in [10] andalsoexplainedin[21].Thissettingallowstosolveequationslike(1)inH¨olderspaces thanks to contraction arguments, in a rather classical way, which will be explained at Section 3.1. In fact, observe that our resolution will be entirely pathwise, and we shall deal with a general equation of the form dy t =f(Z t y )dx t +b(Z t y )dt, t∈[0, T],(3) for a given pathx∈ C 1 γ ([0, T];R d ) withγ >1/2, where the integral with respect toxhas to be understood in the Young sense [32]. Furthermore, in equations like (3), the drift termb(Z y  Thus,usually harmless, but induces some cumbersome notations.) is  for sake of simplicity, we shall rather deal in the sequel with a reduced delay equation of the type: Z t f(Z y s )dx y t =a+ s , t∈[0, T]. 0

MALLIAVIN CALCULUS FOR FRACTIONAL DELAY EQUATIONS

3

Once this last equation is properly defined and solved, the differentiability of the solution y t in a pathwise manner, similarly to thein the Malliavin calculus sense will be obtained case treated in [26]. Finally, the smoothness Theorem 1.1 will be obtained mainly by bounding the moments of the Malliavin derivatives ofy. This will be achieved thanks to a careful analysis and some a priori estimates for equation (1).

2.Algebraic Young integration

The Young integration can be introduced in several ways (convergence of Riemann sums, fractional calculus setting [33]). We have chosen here to follow the algebraic ap- proach introduced in [10] and developed e.g. in [12, 21], since this formalism will help us later in our analysis.

2.1.Increments.us begin with the basic algebraic structures which will allow us toLet define a pathwise integral with respect to irregular functions: first of all, for an arbitrary real numberT >0, a topological vector spaceVand an integerk≥1 we denote byC k (V) (or byC k ([0, T];V)) the set of continuous functionsg: [0, T] k →Vsuch thatg t 1 ∙∙∙t k = 0 whenevert i =t i+1 for somei≤k−1. Such a function will be called a (k−1)-increment, and we will setC ∗ (V) =∪ k≥1 C k (V). An important elementary operator isδ, which is defined as follows onC k (V): k+1 δ:C k (V)→ C k+1 (V),(δg) t 1 ∙∙∙t k+1 = X (−1) k−i g t 1 ∙∙∙t ˆ i ∙∙∙t k+1 ,(4) i=1 ˆ wheret i  Ameans that this particular argument is omitted. fundamental property ofδ, which is easily verified, is thatδδ= 0, whereδδis considered as an operator fromC k (V) toC k+2 (V will denote). WeZC k (V) =C k (V)∩KerδandBC k (V) =C k (V)∩Imδ. Some simple examples of actions ofδ, which will be the ones we will really use through- out the paper, are obtained by lettingg∈ C 1 (V) andh∈ C 2 (V for any). Then, s, u, t∈[0, T], we have

(δg) st =g t −g s ,and (δh) sut =h st −h su −h ut .(5) Furthermore, it is easily checked thatZC k (V) =BC k (V) for anyk≥1. In particular, the following basic property holds:

Lemma 2.1.Letk≥1andh∈ ZC k+1 (V) there exists a (non unique). Thenf∈ C k (V) such thath=δf.

Observe that Lemma 2.1 implies that all the elementsh∈ C 2 (V) such thatδh= 0 can be written ash=δffor some (non unique)f∈ C 1 (V). Thus we get a heuristic interpretation ofδ| C 2 (V) : it measures how much a given 1-increment is far from being an exact increment of a function, i.e., a finite difference.

Remark2.2.first elementary but important link between these algebraic struc-Here is a tures and integration theory: letfandgtwo smooth real valued function on [0be , T]. Define thenI∈ C 2 (V) by I st = Z t s df v Z s v dg w ,fors, t∈[0, T].