Let $w\colon [0,T]\times\mathbb{T}^d \to \mathbb{R}^n$ be such that $$ \|w\|_{L^\infty(BMO)} := \sup_{t\in[0,T]}\|w(t,\cdot)\|_{BMO} \leq C $$ and $\int_{\mathbb{T}^d} w(t,x)\mathrm{d}x = 0 $ for all $t$.

The corollary from the John-Nirenberg inequality states that $$ \int_{\mathbb{T}^d} e^{p|w(t,x)|} \mathrm{d}x \leq C $$ for $p\leq \frac{c_2}{\|w\|_{L^\infty(BMO)}}$.

Moreover, let $x(t,y)\in C^1([0,T]\times\mathbb{T}^d)$ be such that $x(t,\cdot)$ is a diffemorphism for any $t$ and the Jacobian $J(t,y) = \det D_y x(t,y)$ satisfies $$ 0 < \frac{1}{c} \leq J(t,y) \leq c $$

My problem: I'm trying to show that for a fixed $p>1$ we can choose sufficiently small $t$ such that

\begin{equation}\label{int} \int_{\mathbb{T}^d} \exp\left(p\int_0^t |w(s,x(s,y))|\mathrm{d}s\right)\mathrm{d}y \tag{*} \end{equation} is finite.

So far I tried multiple things that didn't work out, but I believe they may be helpful:

  1. If we had in the exponent just $\int_0^t |w(s,x)|\mathrm{d}s$, then as $$ \left\|\int_0^t |w(s,\cdot)|\mathrm{d}s\right\|_{BMO} \leq \int_0^t \|w(s,\cdot)\|_{BMO} \mathrm{d}s \leq t\|w\|_{L^\infty(BMO)} \leq Ct, $$ we would have the integrability for all $p \leq \frac{c_2}{Ct}$, so choosing sufficiently small $t$ we can take $p$ as big as we want. However, the function $w(t,x(t,\cdot))$ may no longer be in $BMO$, so we cannot do it this way.

  2. The obvious way to deal with (*) is to use convexity of $\exp$ and get rid of the integral in the exponent. Then

$$\int_{\mathbb{T}^d} \exp\left(p\int_0^t |w(s,x(s,y))|\mathrm{d}s\right)\mathrm{d}y \leq \int_{\mathbb{T}^d}\int_0^t e^{p|w(s,x(s,y))|} \mathrm{d}s\mathrm{d} y $$ and then using Fubini and the bounds on $J(t,y)$, we get $$ \int_{\mathbb{T}^d}\int_0^t e^{p|w(s,x(s,y))|} \mathrm{d}s\mathrm{d} y \leq c\int_0^t \int_{\mathbb{T}^d} e^{p|w(s,x(s,y))|} J(s,y) \mathrm{d}y\mathrm{d}s = c\int_0^t \int_{\mathbb{T}^d} e^{p|w(s,x)|} \mathrm{d}x \mathrm{d}s. $$ Now we can apply the John-Nirenberg inequality, but we wouldn't get arbitrary large $p$. I was wondering if there are some functional inequalities, which would allow me to put the dependence of small $t$ again in the exponent, but so far I didn't find any.

  1. My third idea was also incorrect, but maybe it can be fixed somehow. It uses the fact that for nonnegative integrable functions there exist $\xi\in[0,t]$ such that $\int_0^t f(s)\mathrm{d}s \leq tf(\xi)$. In my case $w\in L^1([0,T]\times\mathbb{T}^d)$ and I would have for almost all $y$ $$ \int_0^t |w(s,x(s,y))|\mathrm{d}s \leq t|w(\xi,x(\xi,y))|. $$ Then $$\int_{\mathbb{T}^d} \exp\left(p\int_0^t|w(s,x(s,y))|\mathrm{d}s\right) \mathrm{d} y \leq \int_{\mathbb{T}^d} \exp\left(pt|w(\xi,x(\xi,y))|\right) \mathrm{d} y $$ and I could perform the change of variables in the same way as in 1. and eventually from John-Nirenberg obtain the integrability for $p \leq \frac{c_2}{Ct}$.

However, what I didn't take into account is that my $\xi$ depends on $y$, so instead I have $$ \int_o^t |w(s,x(s,y))|\mathrm{d}s \leq t |w(\xi(y), x(\xi(y),y))| $$ and I don't know if I can do something with it (after the change of variables I would still have the dependence of $x(t,y)$ inside $w$).

I would be really happy from the slightiest hints, as I got out of ideas. I'm also not 100% sure if this is even true, although it looks like it should work...


1 Answer 1


I don't know much about BMO things, but I do know the following version of Jensen's inequality: $$\phi\left(\int f(s) d\mu(s)\right) \leq \int \phi(f(s)) d\mu(s),$$ provided $\mu$ is a probability measure. From that point of view, I'm not sure part 2 of your argument is correct. You should instead put $d\mu(s) = ds/t$ to arrive at: $$ \int_{\mathbb{T}^d} \exp\left(p \int_0^t |w(s,x(s,y)|ds \right) dy \leq \int_{\mathbb{T}^d} {1 \over t}\int_0^t e^{pt|w(s,x(s,y)|}ds dy. $$

According to your corollary, this is finite provided $pt$ is sufficiently small, so I think you're gucci.

  • $\begingroup$ Thank you very much, I forgot about the length of the interval in Jensen's inequality $\endgroup$
    – M_S
    Commented Jan 8, 2021 at 21:53

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.