MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let me define $$ J^f_{n}(t) = \, \int_0^t \int_0^{t_1} \ldots \int_0^{t_{n-1}} f(t, t_1, \ldots, t_n) \; dB_{t_n} ...dB_{t_1} $$ where $f:[0,1]^{n+1} \to \mathbb{R}$ is a nice deterministic function. Note that the upper limit of the outer integral appears in the integrand. I would like an estimate on the tail of $\sup_{t \in (0,1]} |J^f_n(t)|$ i.e. $$ \mathbb{P} \left( \sup_{t \in (0,1]} |J^f_n(t)| \geq K\right) \leq \; ? $$ Note that we can write $J^f_n(t)$ in the following form $$ J^f_n(t) = \int_0^t J^f_{n-1}(t,t_1) \; dB_{t_1} $$ which is precisely the form of a Gaussian Volterra process (see e.g. ) except here we have a stochastic kernel.

  • Has anyone encountered a process like this before? Does it have a name?

In the case of a constant integrand, such an estimate is fairly easy to find. For the integrand $f \equiv 1$, I define $$ I_{n}(t):= J^f_n(t) = \int_0^t \int_0^{t_1} \ldots \int_0^{t_{n-1}} 1 \; dB_{t_n} ...dB_{t_1} $$ then we can use the martingale inequality (found e.g. in proposition 4.24 here ) $$ \mathbb{P} \left( \sup_{t \in (0,T]} |M_t| \geq K_1, \langle M_. \rangle _T \leq K_2\ \right) \leq 2 \exp \left( - \frac{K_1^2}{2K_2} \right) $$ recursively applied to $I_n(t)$, then $I_{n-1}(t)$ etc. to get a bound of the type $$ \mathbb{P} \left( \sup_{t \in (0,1]} |I_n(t)| \geq K\right) \leq C_n \exp \left( - K^{d_n} \right) $$ with $C_n$ and $d_n$ positive constants.

  • The problem for $J^f_n(t)$ is that, in general, it is not a martingale or a Guassian process

With this in mind, what tools can we use to analyse $J^f_{n}(t)$

If the function $f(t, t_1, \ldots, t_n)$ splits into $f_0(t)f_1(t_1)\ldots f_n(t_n)$ then we can pass each function outside one of the integrals and we are left with a martingale inside so this makes things considerably easier. This motivates the following question

  • Is there a theory for approximating general $f(t, t_1, \ldots, t_n)$ by products of functions $f_0(t) f_1(t_1) \ldots f_n(t_n)$ ?
share|cite|improve this question
Your $I_n$ is basically a Hermite function, by Ito's classical results on the "multiple Wiener integral". See, e.g. section 5.4.1 of – Steve Huntsman Apr 9 '13 at 20:40
Hi, thanks for the reference. It seems iterated/multiple Ito integrals with constant integrand are quite well understood. It is really when the integrand depends on the upper limit that things become harder. – user31090 Apr 10 '13 at 13:25

The estimate you are interested in has already been studied. The tail bound

$ \mathbb{P} \left( \sup_{t \in [0,1]} | J_n^f (t)| \ge K \right) \le C_1 \exp( -C_2 K^{2/n}) $

was proved by C. Borell. The main tools are infinite dimensional isoperimetric inequalities. It was later proved by M. Ledoux that we even have

$ \lim_{K \to \infty} \frac{1}{K^{2/n}}\mathbb{P} \left( \sup_{t \in [0,1]} | J_n^f (t)| \ge K \right) =-I_n(f) $

where $I_n(f)$ is known explicitly.

Michel Ledoux, A note on large deviations for Wiener chaos

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.