In the theory of Hardy spaces of the unit disc, a fact that is implicitely used quite often is that if $f\in H^p, 1<p<\infty$, then there exists a function $F\in H^p$ such that $|f(z)| \leq |F(z)|, \,\, \forall z \in \mathbb{D}$, $ Re F \geq 0$ and $ \Vert F \Vert_p \leq c_p \Vert f \Vert_p $.

To see why this is the case, given $f\in H^p$ define \begin{equation*} F(z) = \int_{\mathbb{T}} |f(\zeta)| \frac{1+\overline{\zeta}z}{1-\overline{\zeta}z} |d\zeta|. \end{equation*}

This is sometimes called Herglotz transform of $|f|$, but the point is that is a bounded linear operator from $L^p(\mathbb{T})$ into $H^p$, as a corollary of the M. Riesz Theorem. Hence $F$ defined like this has the required properties.

I was wondering if the existence of such an $F$ could be also true in the case $p=1$. Although the construction should be completely different because of the Failure of the M. Riesz Theorem for $p=1$.


In general it is impossible. We can take the square root, map the circle conformally to the half-plane, and arrive at the following problem: given any nonnegative $f\in L^2(\mu)$ with finite logarithmic integral, can we find $g\in H^2(\mu)$ with $\Re g\ge f$ and $|\Im g|\le \Re g$ where $d\mu(x)=\frac{dx}{1+x^2}$? Now, if that is possible for all $f$, it is also possible with the $H^2(\mu)$ norm of $g$ bounded by $C\|f\|_{L^2(\mu)}$.

Let's do the usual blow-up now taking some $f\in L^2(dx)$ and considering $f(nx)$ instead of $f$. Then, when we scale back, we'll get majorants $g_n\in H^2(\mu_n)$ such that $\|g_n\|_{H^2(\mu_n)}\le C\|f\|_{L^2(\mu_n)}\le C\|f\|_{L^2(dx)}$ with $d\mu_n(x)=\frac{dx}{1+(x/n)^2}$. We can pass to a subsequence and assume that $g_n$ converge to some $g$ weakly in $L^2(dx)$ on every subinterval of $\mathbb R$. Then we shall have $g\in H^2(dx)$ and we still have $\Re g\ge f, |\Im g|\le \Re g$(the vanishing of the Fourier transform on the negative semi-axis and the comparisons with non-negative functions can be tested by integrating against appropriately chosen $L^2(dx)$ functions, and the norm can only drop).

But for $H^2(dx)$ functions we have $\int_{\mathbb R}|\Re g|^2dx=\int_{\mathbb R}|\Im g|^2$, so we are forced to have $|\Im g|=\Re g$ almost everywhere on the line. But then $g^2\in H^1(dx)$ and $\Re[g^2]=0$ on $\mathbb R$, which is impossible.

Thus, a sufficiently strong bump at one point will give you a counterexample. To figure out exactly how strong is "sufficiently strong", one needs to make all that weak limit nonsense quantitative, which I leave to someone else :-)

  • $\begingroup$ Thanks a lot, this is very nice ! Doing the calculations I don't see whe $\Re g$ is positive, I get instead $| \Im g| \leq |\Re g | $ and $\sqrt{2} |\Re g| \geq f$. Probably I am missing something. $\endgroup$ – an_ordinary_mathematician Sep 19 '20 at 9:12
  • $\begingroup$ @an_ordinary_mathematician The square root of a function with non-negative real part has non-negative real part, doesn't it? $|\Im g|\le|\Re g|$ is insufficient because it isn't preserved under weak limits. $\sqrt 2$ is there, of course, but it is just absorbed into $C$. $\endgroup$ – fedja Sep 19 '20 at 14:16
  • $\begingroup$ So that, combined with the observation about the Herglotz transform above is a quite roundabout proof that the Hilbert transform is unbounded on $L^1(\mathbb{T})$ ! $\endgroup$ – an_ordinary_mathematician Sep 19 '20 at 14:28

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.