Let me define the Fourier transform of a function $u$, say in the Schwartz space $\mathscr S(\mathbb R)$ as $ \hat u(\xi)=\int_{\mathbb R} e^{-2iπ x\cdot \xi} u(x) dx. $ The Hilbert transform $\mathscr H$ is the Fourier multiplier $-i \operatorname{sign} \xi$ and can be expressed equivalently as the convolution with $ \operatorname{pv}\frac1{π x}. $ We have $$ (\mathscr H u)(x)=\lim_{\epsilon\rightarrow 0_+}\int_{\vert x-y\vert\ge \epsilon} \frac{u(y)}{π(x-y)} dy. $$ That operator can be realized as a unitary skew-adjoint operator on $L^2(\mathbb R)$: this is easy when we look at the Fourier multiplier realization $-i\operatorname{sign}(D)$ given above. A related operator is the so-called Hardy $\mathscr H_a$ operator, whose kernel is $$ k(x,y)=\frac{H(x)H(y)}{π(x+y)}, \quad H=\mathbf 1_{\mathbb R_+}. $$ It is easy to prove that $\mathscr H_a$ is a selfadjoint operator on $L^2(\mathbb R)$ with operator-norm less than 1 and slightly more difficult to prove that its operator-norm is actually equal to 1. On the other hand, one may look at the discrete Hardy operator $\mathscr H_{ad}$, say on $\ell^2(\mathbb N^*)$ by the kernel $$ J(k,l)=\frac{1}{π(k+l)}. $$ To prove that $\mathscr H_{ad}$ is bounded selfadjoint with operator norm less than 1 is not difficult, using the boundedness of the discrete Hilbert transform. It is also possible to prove that the norm of $\mathscr H_{ad}$ is actually 1.

Finally my question: What are the explicit links between the discrete results and the "continuous" results mentioned above? In particular, these not-so-trivial results on operator-norms equal to 1 for Hardy operators are more classical in the discrete case. Is it possible to actually deduce their continuous version from the discrete ones? References are welcome.


Your Answer

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

Browse other questions tagged or ask your own question.