Does there exists a constant $C>0$ such that

$$ \int_{-1}^1 \lvert x\rvert\lvert\partial_x u\rvert^2 \,dx \geq C\, \lVert u\rVert^2_{H^{1/2}((-1,1))},$$ for all $u\in C^{\infty}_0((-1,1))$?

  • $\begingroup$ Just to confirm, we have $ \| u\|^2_{H^{\frac{1}{2}}((-1,1))} := \int_{-1}^1 \frac{1}{\sqrt{1 + (2 \pi x)^2}} |\hat u(x)|^2 \, dx$? Where $\hat u (x) :=\int_{-1}^1 u(t) e^{-i x t} \, dt$. $\endgroup$
    – Nate River
    Nov 18, 2023 at 17:40
  • 1
    $\begingroup$ The integral on the frequency side must be on the whole $\mathbb R$. Additionally, the integrand factor should be inverted... so no your interpretation of the norm is incorrect $\endgroup$
    – Ali
    Nov 18, 2023 at 21:58
  • $\begingroup$ Oh sorry, made some typos. Thanks for clarifying! $\endgroup$
    – Nate River
    Nov 19, 2023 at 3:42
  • $\begingroup$ I deleted an embarrassingly nonsense answer, thanks for @an_ordinary_mathematician for spotting the mistake. $\endgroup$ Nov 19, 2023 at 16:19

1 Answer 1


It is not true. Start with a function $u$ which vanishes for $x<0$ and is equal to $1$ for $0 \leq x \leq \frac 12$ and then smooth from $x \geq \frac 12$. The Fourier coefficients behave like $1/n$ so that $ u \not \in H^{\frac 12}$. However it can be approximated in $L^2$ with Lipschitz functions $u_\epsilon$ having $\int x |u_\epsilon'|^2 \leq C$ joining $(0,0)$ with $(\epsilon, 1)$ by a straight line. The $H^{\frac 12}$ norms of $u_\epsilon$ blow up since $u \not \in H^{\frac 12}$.

  • $\begingroup$ However the $L^2$ norm is controlled by $\int |x| |u'|^2$. Just write (for $x>0$) $-u(x)=\int_x^1 t^{\frac 12}u'( t ) t^{-\frac 12}\, dt$ so that, by Holder, $|u(x) \leq A |\log x|^{\frac 12}$ with $A^2=\int |x||u'|^2$ $\endgroup$ Nov 21, 2023 at 18:26
  • $\begingroup$ @GeorgioMetafune It is funny that on the other hand the statement is true for even functions (simple complex interpolation). $\endgroup$
    – fedja
    Nov 27, 2023 at 19:49
  • $\begingroup$ @fedja Why is this true? I do not see it. $\endgroup$ Nov 27, 2023 at 21:57
  • $\begingroup$ Consider the linear operator $T_z:L^2\to L^2$ that maps a function $g$ to $|y|^{1-z}$ times the Fourier transform of the even extension of $G(x)=\int_x^1 g(t)t^{-z}\,dt$ from $(0,1)$ to $(-1,1)$ when $\Re z\in[0,1]$. When $\Re z=0$ and $\Re z=1$, we get a bounded mapping (in the latter case I used that $g\mapsto G$ is bounded in $L^2$, by, say, the Schur test with $\varphi(x)=\psi(x)=1/\sqrt x$). So it is bounded for $z=1/2$ as well and that is just what I said. :-) $\endgroup$
    – fedja
    Nov 27, 2023 at 22:58
  • $\begingroup$ @fedja Thank you very much, it is clear now. $\endgroup$ Nov 28, 2023 at 14:28

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.