UPPER BOUNDS FOR THE DENSITY OF SOLUTIONS OF STOCHASTIC DIFFERENTIAL EQUATIONS DRIVEN BY

Description

UPPER BOUNDS FOR THE DENSITY OF SOLUTIONS OF STOCHASTIC DIFFERENTIAL EQUATIONS DRIVEN BY FRACTIONAL BROWNIAN MOTIONS FABRICE BAUDOIN, CHENG OUYANG, AND SAMY TINDEL Abstract. In this paper we study upper bounds for the density of solution of stochastic differential equations driven by a fractional Brownian motion with Hurst parameter H > 1/3. We show that under some geometric conditions, in the regular case H > 1/2, the density of the solution satisfy the log-Sobolev inequality, the Gaussian concentration inequality and admits an upper Gaussian bound. In the rough case H > 1/3 and under the same geometric conditions, we show that the density of the solution is smooth and admits an upper sub-Gaussian bound. Contents 1. Introduction 1 2. Stochastic calculus for fractional Brownian motion 4 2.1. Malliavin calculus tools 4 2.2. Differential equations driven by fBm 5 3. Estimates for solutions of SDEs driven by fBm: the smooth case 7 3.1. Log-Sobolev inequality 11 3.2. Concentration inequality 12 3.3. Gaussian upper bound 13 4. Extension to the irregular case 15 4.1. Increments 15 4.2. Computations in C? 17 4.3. Weakly controlled processes 18 4.4. Rough differential equations 20 4.5. Estimates for the Malliavin derivative 23 4.6. Density upper bound 24 References 26 1. Introduction Let B = (B1, . . . , Bd) be a d dimensional fractional Brownian motion (fBm in the sequel) defined on a complete probability space (?,F ,P), with Hurst parameter H ? First author supported in part

• been analyzed


• sub-gaussian bound


• has been


• differential equations


• smooth vector


• valued stochastic


• gaussian bounds


• hölder continuity


• equations driven

Sujets

• Has Been
• Differential equation
• Smooth vector
• Hölder condition

Informations

• Publié par: pefav
• Nombre de lectures: 21
• Langue: English

Extrait

UPPER BOUNDS FOR THE DENSITY OF SOLUTIONS OF STOCHASTIC DIFFERENTIAL EQUATIONS DRIVEN BY FRACTIONAL BROWNIAN MOTIONS

FABRICE BAUDOIN, CHENG OUYANG, AND SAMY TINDEL

Abstract.In this paper we study upper bounds for the density of solution of stochastic differential equations driven by a fractional Brownian motion with Hurst parameter H >1/3. We show that under some geometric conditions, in the regular caseH >1/2, the density of the solution satisfy the log-Sobolev inequality, the Gaussian concentration inequality and admits an upper Gaussian bound. In the rough caseH >1/3and under the same geometric conditions, we show that the density of the solution is smooth and admits an upper sub-Gaussian bound.

Contents

1. Introduction 2. Stochastic calculus for fractional Brownian motion 2.1. Malliavin calculus tools 2.2. Differential equations driven by fBm 3. Estimates for solutions of SDEs driven by fBm: the smooth case 3.1. Log-Sobolev inequality 3.2. Concentration inequality 3.3. Gaussian upper bound 4. Extension to the irregular case 4.1. Increments 4.2. Computations inC ∗ 4.3. Weakly controlled processes 4.4. Rough differential equations 4.5. Estimates for the Malliavin derivative 4.6. Density upper bound References

1 4 4 5 7 11 12 13 15 15 17 18 20 23 24 26

1.dortitcuonnI LetB= (B 1 , . . . , B d )be addimensional fractional Brownian motion (fBm in the sequel) defined on a complete probability space(Ω,F,P), with Hurst parameterH∈

First author supported in part by NSF Grant DMS 0907326. Third author partially supported by the (French) ANR grant ECRU. 1

2

F. BAUDOIN, C. OUYANG, AND S. TINDEL

(0,1) that it means that. RecallBis a centered Gaussian process indexed byR + , whose coordinates are independent and satisfy (1)E  h B t j −B s j  2 i =|t−s| 2H ,fors, t∈R + . In particular, by considering the family{B H ;H∈(0,1)}, one obtains some Gaussian processes with any prescribed Hölder regularity, while fulfilling some intuitive scaling properties. This converts fBm into the most natural generalization of Brownian motion to this day. We are concerned here with the following class of equations driven byB: t Z 0 t V 0 (X x s )ds+ i= Xd 1 Z 0 X s x )dB s i (2)X t x =x+V i (, wherexis a generic initial condition and{V i ; 0≤i≤d}is a collection of smooth vector fields ofR d  to the fact that fBm is a natural generalization of Brownian motion,. Owing this kind of model is often used by practitioners in different contexts, among which we would like to highlight recent sophisticated models in Biophysics [17, 25, 26]. As far as mathematical results are concerned, equation (2) is now a fairly well under- stood object: existence and uniqueness results are obtained forH > 2 1 thanks to Young integral type tools [28, 23], while rough paths methods [11, 18] are required for 4 1 < H < 2 1 . Numerical schemes can be implemented for this kind of systems [10, 11], and a notion of ergodicity is also available [14, 15]. Finally, the law ofX tx has been analyzed by means of semi-group type methods [1, 20] and its density has also been investigated in [2, 6, 16, 24]. In spite of these advances, concentrations results and Gaussian bounds for the solution to (2) are scarce: we are only aware of the large deviation results [19] in this line of investigation. The current article is thus an attempt to make a step in this direction, by analyzing a special but nontrivial situation. Indeed, we consider here equation (2) driven by a fBm with Hurst parameterH∈ ( 3 1 ,1)we suppose that our vector fields, and V 0 , . . . , V d fulfill either of the following non- degeneracy and antisymmetric hypothesis: Hypothesis 1.1.The vector fieldsV 0 , . . . , V d areC ∞ -bounded, andV 1 , . . . , V d satisfy (i)For everyx∈R d , the vectorsV 1 (x),∙ ∙ ∙, V d (x)form a basis ofR d . (ii)There exist smooth and bounded functionsω k ij such that: d (3)[V i , V j ] = X ω ij k V k ,andω ij k =−ω ki j . k=1

The second assumption (ii) is of geometric nature and actually means that the Levi- Civita connection associated with the Riemannian structure given by the vector fieldsV i ’s is r X Y 2 = 1 [X, Y]. In a Lie group structure, this is equivalent to the fact that the Lie algebra is of compact type, or in other words that the adjoint representation is unitary. Such geometric assump- tion already appeared in the work [3] where it was used to prove a small-time asymptotics of the density.