Let $B_t$ be the classic Brownian motion. I understand that, if $s>1/2$, almost surely $B_t$ is nowhere $s$-Hölder continuous i.e. almost surely for no point $x$ it happens that $B_t\in C^s(x)$.

Moreover I read a claim that said the same about any translation by a continuous function: given $s>1/2$ and $f$ continuous then almost surely $f+B_t$ is nowhere $s$-Hölder continuous. Is this true and if so why?

  • $\begingroup$ If $s=1/2$, then almost surely there will be points at which the Brownian motion is $s$-Hölder continuous -- see e.g. Remark 1.21 in Brownian Motion by Peter Mörters and Yuval Peres. I have edited your post accordingly. $\endgroup$ Feb 12, 2023 at 3:07

2 Answers 2


The result holds for any bounded function $f$, in the following sense: for any real $s>1/2$, \begin{equation} P^*(A)=0, \end{equation} where \begin{equation} A:=\Big\{\exists t_0\in[0,1]\ \limsup_{t\to t_0}\frac{|W_f(t)-W_f(t_0)|}{|t-t_0|^s}<\infty\Big\}, \end{equation} $P^*$ is the outer probability, $\limsup_{t\to t_0}:=\limsup_{t\to t_0,t\in[0,1]}$, \begin{equation} W_f:=W+f, \end{equation} and $W$ is a standard Wiener process.

The proof is obtained by a straightforward adaptation of the proof of the Paley--Wiener--Zygmund theorem on the almost sure nowhere differentiability of the Brownian motion.

Indeed, since $W$ and $f$ are bounded, \begin{equation} A\subseteq B=\bigcup_{M=1}^\infty B_M, \end{equation} where \begin{equation} B:=\Big\{\exists t_0\in[0,1]\ \sup_{t\in[0,1]}\frac{|W_f(t)-W_f(t_0)|}{|t-t_0|^s}<\infty\Big\}, \end{equation} \begin{equation} B_M:=\Big\{\exists t_0\in[0,1]\ \sup_{t\in[0,1]}\frac{|W_f(t)-W_f(t_0)|}{|t-t_0|^s}\le M\Big\}. \end{equation} Next, \begin{equation} B_M:=B_{M,1}\cup B_{M,2}, \end{equation} where \begin{equation} B_{M,1}:=\Big\{\exists t_0\in[0,1/2]\ \sup_{t\in[0,1]}\frac{|W_f(t)-W_f(t_0)|}{|t-t_0|^s}\le M\Big\}, \end{equation} \begin{equation} B_{M,2}:=\Big\{\exists t_0\in[1/2,1]\ \sup_{t\in[0,1]}\frac{|W_f(t)-W_f(t_0)|}{|t-t_0|^s}\le M\Big\}. \end{equation}

Let $r$ be any integer such that \begin{equation} r>1/(s-1/2). \end{equation} Let then $n$ be any integer $\ge n_r$, where $n_r$ is the smaller integer $q$ such that $2^{q-1}>r$.

Assuming the event $B_{M,1}$ occurs, let $K$ be a random integer in the set $\{1,\dots,2^{n-1}\}$ such that $t_0\in[\frac{K-1}{2^n},\frac K{2^n}]$. Then for $j=1,\dots,r$ \begin{equation} \begin{aligned} &\Big|W_f\Big(\frac{K+j}{2^n}\Big)-W_f\Big(\frac{K+j-1}{2^n}\Big)\Big| \\ &\le\Big|W_f\Big(\frac{K+j}{2^n}\Big)-W_f(t_0)\Big| +\Big|W_f\Big(\frac{K+j-1}{2^n}\Big)-W_f(t_0)\Big| \\ &\le M\Big|\frac{K+j}{2^n}-t_0\Big|^s +M\Big|\frac{K+j-1}{2^n}-t_0\Big|^s\le\frac{2Mj^s}{2^{sn}} \le\frac{2Mr^s}{2^{sn}} \end{aligned} \end{equation} So, \begin{equation} B_{M,1}\subseteq\bigcap_{n\ge n_r}\bigcup_{k=1}^{2^{n-1}}C_{n,k}, \end{equation} where \begin{equation} C_{n,k}:=\bigcap_{j=1}^r \Big\{\Big|W_f\Big(\frac{k+j}{2^n}\Big)-W_f\Big(\frac{k+j-1}{2^n}\Big)\Big| \le\frac{2Mr^s}{2^{sn}}\Big\}. \end{equation} By the independence of the increments of the Wiener process and because the pdf of the normally distributed random variable $W_f\big(\frac{k+j}{2^n}\big)-W_f\big(\frac{k+j-1}{2^n}\big)$ is $\le2^{n/2-1}$, we have \begin{equation} \begin{aligned} P(C_{n,k})&=\prod_{j=1}^r P\Big(\Big|W_f\Big(\frac{k+j}{2^n}\Big)-W_f\Big(\frac{k+j-1}{2^n}\Big)\Big| \le\frac{2Mr^s}{2^{sn}}\Big) \\ & \le\Big(2^{n/2}\frac{2Mr^s}{2^{sn}}\Big)^r=\frac C{2^{(1+a)n}}, \end{aligned} \end{equation} where $C:=(2Mr^s)^r$ and $a:=r(s-1/2)-1>0$. So, \begin{equation} P\Big(\bigcup_{k=1}^{2^{n-1}}C_{n,k}\Big)\le2^{n-1}\frac C{2^{(1+a)n}}\le\frac C{2^{an}} \end{equation} and \begin{equation} P^*(B_{M,1})\le\lim_{n\to\infty}\frac C{2^{an}}=0. \end{equation} So, $P^*(B_{M,1})=0$. Similarly, $P^*(B_{M,2})=0$. So, $P^*(B_M)=0$, for all $M$.

Thus, \begin{equation} P^*(A)\le P^*(B)\le\sum_{M=1}^\infty P^*(B_M)=0, \end{equation} as claimed. $\quad\Box$


If $f$ is absolutely continuous with $\int_0^1 f'(t)^2\,dt<\infty$, then, by Girsanov's formula (see e.g. Theorem 5.1. in [1]), the process $(B_t+f(t))_{t\in[0,1]}$ is a standard Brownian motion with respect to the measure $Q$ whose density $\frac{dQ}{dP}$ with respect to the distribution, say $P$, of the original standard Brownian motion $(B_t)_{t\in[0,1]}$ is given by the formula $$\frac{dQ}{dP}(x)=\exp\left(-\int_0^1 f'(t)dx_t - \frac{1}{2}\int_0^1 f'(t)^2\,dt\right).$$ It also follows that $P$ is absolutely continuous with respect to $Q$.

So, in this case, for any real $s>1/2$, the process $(B_t+f(t))_{t\in[0,1]}$ is $Q$-almost surely nowhere $s$-Hölder continuous, and hence $P$-almost surely nowhere $s$-Hölder continuous, as desired.

The case of general bounded functions $f$ is now considered in the other answer.

[1] Ioannis Karatzas and Steven E. Shreve, Brownian Motion and Stochastic Calculus, Second Edition, 1991.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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