A sharp estimate on the norm of the continuous square function (Q4529682)

The author considers the problem of estimating the norms of certain square functions in \(L^2(w) =\left\{f:\int | f| ^2 w<\infty\right\}\) where \(w\) is an \(A_2\) weight on \(\mathbb R\). Both the continuous square function \[ S(f)(x)=\left(\int_{| x-y| <t} | f \ast\psi_t | ^2 \, dy \,\frac{dt}{t^3}\right)^{1/2} \] as well as the wavelet square function \[ S_{\psi}f(x)=\left(\sum_{x\in I\in \mathcal{D} } | \langle f ,\phi_I\rangle | ^2 \,\frac{1}{| I| }\right)^{1/2} \] are considered. Here \(\mathcal{D}\) denotes the collection of dyadic intervals and \(\psi_I=2^{j/2}\psi(2^jx-k)\) when \(I=I(j,k)=[k/2^j,(k+1)/2^j)\), while \(\psi_t(x)=\psi(x/t)/\sqrt{t}\). It is shown that, under suitable conditions on \(\phi\), one has \(\| S(f)\| _{L^2(w)}\leq C_1 \| w\| _{A_2}\| f\| _{L^2(w)}\) while \(\| S_{\psi}(f)\| _{L^2(w)}\leq C_2 \| w\| _{A_2}\| f\| _{L^2(w)}\). The condition imposed on \(\psi\) in both cases is that \(\psi\) is a \(C^1\) function supported in \([0,1]\) such that the functions \(\psi_I\) form an orthonormal (wavelet) basis for \(L^2(\mathbb{R})\), that is, with respect to Lebesgue measure. It is a minor technical point that no such \(\psi\) actually exists: one can always generate a wavelet basis with a finite collection of such functions and redefine the square functions accordingly. Orthonormality plays a convenient role in establishing the bound for the wavelet square function \(S_\psi\), as do regularity and vanishing mean (which follows from orthogonality plus regularity). The bound for \(S(f)\) is established by means of renormalization -- which does not affect \(\| w\| _{A_2}\) -- and a limiting argument. The results represent the best possible dependence of the square function norms as functions of \(\| w\| _{A_2}\).