It is well-known that the Brownian motion (Wiener process) is almost sure locally $\alpha$-Holder for any $\alpha<1/2$. That is, with probability 1 $$ \sup_{t,s\in[0,1]}\frac{|W_t-W_s|}{|t-s|^{\alpha}}<\infty. $$

On the other hand, Brownian motion is clearly not globally $\alpha$-Holder. Indeed, it follows from the law of iterated logarithm, that $$ \sup_{t\ge0}\frac{|W_t|}{|t|^{\alpha}}=\infty. $$

One can play with the first inequality and with the help of Kolmogorov theorem and Borell-Cantelli obtain something that looks like ``global-local'' weighted Holder $$ \sup_{\substack{t,s\ge0\\ |t-s|\in[0,1]}}\frac{|W_t-W_s|}{(t^\delta\vee s^\delta\vee1)|t-s|^{\alpha}}<\infty. $$ for any arbitrarily small $\delta>0$.

My question is whether a global weighted Holder of the form $$ \sup_{t,s\ge0}\frac{|W_t-W_s|}{(t^\gamma\vee s^\gamma\vee 1)|t-s|^{\alpha}}<\infty $$ holds? What is the optimal weights exponent $\gamma$ (can one say any better than $\gamma=1$)? Does there exist a global version of Kolmogorov continuity theorem that one can directly apply?

  • 2
    $\begingroup$ If $s=1$ and $t$ is large, the quantity in your last line is comparable to $|W_t|/t^{\alpha}$, which as you know has infinite supremum. So it can't hold in the form that you've stated. $\endgroup$ – Nate Eldredge Feb 8 '16 at 19:57
  • $\begingroup$ @NateEldredge Sorry I had a typo. One should read $t^\gamma\vee s^\gamma\vee1$. Then the quantity in my last line is comparable to $|W_t|/t^{\alpha+\gamma}$ which has a finite supremum. $\endgroup$ – Oleg Feb 9 '16 at 10:27

Let $\alpha,\gamma >0$. As Nate notes, for the inequality $$ \sup_{t,s \ge 0}\frac{|W_t-W_s|}{(t^\gamma\vee s^\gamma\vee 1)|t-s|^{\alpha}}<\infty $$ to hold almost surely, it is necessary that $\gamma+ \alpha>1/2$. Of course $\alpha<1/2$ is also needed.

Claim: These conditions are also sufficient.

As observed by the original poster, the claim holds if we restrict attention to $|t-s| \le 1$. This is where the condition that $\alpha<1/2$ is needed. One proof for this case is quite similar to the argument below, considering separately $t \in [2^{n-1},2^n]$.

So it suffices to consider $1<|t-s|$. For integers $0 \le k \le n$ and a constant $R>1$, denote by $A_n(k,R)$ the event that $$ \frac{|W_t-W_s|}{(t^\gamma\vee s^\gamma\vee 1)|t-s|^{\alpha}}\ge R $$ holds for some $s<t$ in $[2^{n-1},2^n]$ that satisfy $2^{k-1} \le t-s \le 2^k$.

Since $\max_{t \le T} W_t$ has the same law as $|W_t|$ (see e.g., Theorem 2.21 in [1]), the standard bound for Gaussian tails (see, e.g., Lemma 12.9 in [1]) implies that $${\bf P}[A_n(k,R)] \le 2^n \cdot\exp(-cR^2 \cdot 2^{2n\gamma+2k\alpha-k}) \le 2^n\cdot\exp(-cR^2 \cdot 2^{2n(\gamma+\alpha-1/2)}) $$ where $c=c(\alpha,\gamma)$ is a constant. Summing this over all pairs of integers $0 \le k \le n$ and then taking $R \to \infty$ proves the claim.

[1] Peter Morters and Yuval Peres, Brownian Motion, Cambridge University Press (2010), available at http://research.microsoft.com/en-us/um/people/peres/brbook.pdf

  • $\begingroup$ Yuval, thanks for your answer. Could you please clarify how does the event $A_n(k,R)$ depend on $R$? Did you mean something like $A_n(k,R):=\{ |W_t-W_s|/|t-s|^{\alpha}(t^\gamma\vee s^\gamma \vee 1)\le R\}$? $\endgroup$ – Oleg Feb 14 '16 at 13:09
  • $\begingroup$ I also did not find Lemma 12.1 in [1]. Did you mean Lemma 12.9? $\endgroup$ – Oleg Feb 14 '16 at 13:31
  • $\begingroup$ Hi Oleg, I corrected the definition of $A_n(k,R)$ and the citation. $\endgroup$ – Yuval Peres Mar 6 '16 at 8:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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