EN FR
EN FR


Section: New Results

Local Hölder regularity of Set-Indexed processes

Participants : Erick Herbin, Alexandre Richard.

In the set-indexed framework of Ivanoff and Merzbach ( [62] ), stochastic processes can be indexed not only by 𝐑 but by a collection 𝒜 of subsets of a measure and metric space (𝒯,d,m), with some assumptions on 𝒜. In we introduce and study some assumptions (A1) and (A2) on the metric indexing collection (𝒜,d𝒜) in order to obtain a Kolmogorov criterion for continuous modifications of SI stochastic processes. Under this assumption, the collection is totally bounded and a set-indexed process with good incremental moments will have a modification whose sample paths are almost surely Hölder continuous, for the distance d𝒜. Once this condition is established, we investigate the definition of Hölder coefficients for SI processes. We shall denote α˜X(t) and αX(t) for the local and pointwise Hölder exponents of X at t, and α˜X(t) and αX(t) for their deterministic counterpart in case X is Gaussian.

In [18] , a set-indexed extension for fractional Brownian motion has been defined and studied. A mean-zero Gaussian process 𝐁H=𝐁UH,U𝒜 is called a set-indexed fractional Brownian motion (SIfBm for short) on (𝒯,𝒜,m) if

U,V𝒜,𝐄𝐁UH𝐁VH=12m(U)2H+m(V)2H-m(UV)2H,(6)

where H(0,1/2] is the index of self-similarity of the process.

In [12] , α˜X and α˜X have been determined for the particular case of an SIfBm indexed by the collection {[0,t];t𝐑+N}{}, called the multiparameter fractional Brownian motion. If X denotes the 𝐑+N-indexed process defined by Xt=𝐁[0,t]H for all t𝐑+N, it is proved that for all t0𝐑+N, α˜X(t0)=H and with probability one, for all t0𝐑+N, α˜X(t0)=H. A theorem of allows one to extend these results to SIfBm indexed by a more general class than the sole collection of rectangles of 𝐑+N.

Theorem 0.1 Let 𝐁H be a set-indexed fractional Brownian motion on 𝒯,𝒜,m, H(0,1/2]. Assume that the subclasses (𝒜n)n𝐍 satisfy Assumption (A1).

Then, the local and pointwise Hölder exponents of 𝐁H at any U0𝒜, defined with respect to the distance dm or any equivalent distance, satisfy

𝐏U0𝒜,α˜𝐁H(U0)=H=1

and if Assumption (A2) holds,

𝐏U0𝒜,α𝐁H(U0)=H=1.

Consequently, since the collection 𝒜 of rectangles of 𝐑+N with m the Lebesgue measure satisfies (A1) and (A2), we obtained a new result on a classical multiparameter process: the multiparameter fractional Brownian motion 𝐁H satisfy, for T𝐑+N:

𝐏t[0,T],α𝐁H([0,t])=α˜𝐁H([0,t])=H=1.