On the principle of duality in Lorentz spaces (Q1363295)

The authors give an implicit-explicit condition for the verification of several inequalities, including some for the maximal function and for the Hilbert transform in one dimension. The Lorentz spaces are defined by one of several semi-norms (which are equivalent under suitable conditions on the weight, in particular, if the weight is nonincreasing). The two of concern here are respectively \(\Lambda_p(\nu)\) and \(\Gamma_p(\nu)\) defined by \[ |f|_{p,\nu}^*=\left(\int_0^{\infty} (f^*(t))^p\nu(t) dt\right)^{1/p} <\infty, \] and \[ |f|_{p,\nu}=\left(\int_0^{\infty}\left(\frac{1}{t}\int_0^t f^*\right)^p\nu(t) dt\right)^{1/p} <\infty, \] where \(f^*\) is the nonincreasing rearrangement of \(f\). Lorentz had characterized the dual spaces of these spaces by giving an equivalent but complicated norm for the dual to \(\Lambda_p(\nu)\). Sawyer gave an inverse Hölder inequality which led to a more explicit expression for the dual, namely that \[ \Lambda_p^*(\nu)=\Gamma_{p'}\left(\left(\frac{1}{x}\int_0^x\nu\right)^{-p'}\nu(x)\right), 1<p<\infty,\frac{1}{p} +\frac{1}{p'}=1, \] subject to the condition \[ \int_0^{\infty}\nu=\infty. \] The authors use ideas from authors cited in the paper to provide an inverse Hölder inequality expressed in terms of a measure defined with a discretization sequence \(\{\mu_k\}\). It is assumed throughout that the weight \(\nu(s)\geq 0\) satisfies the nondegeneracy condition \[ \int_0^{\infty}\frac{\nu(s)}{(s+1)^p} ds <\infty,\int_0^1 s^{-p}\nu(s)=\int_1^{\infty}\nu=\infty. \] They prove the inverse Hölder inequality \[ \sup_{0\leq f\uparrow}\frac{\int_0^{\infty} fg}{\left(\int_0^{\infty}\left(\frac{1}{x}\int_0^x f\right)^p\nu(x) dx\right)^{1/p}}=\left(\int_0^{\infty}\left(\int_0^xg\right)^{p'}\mathcal V(x) dx\right)^{1/p'}, \] where \(1<p<\infty, g(x)\downarrow\). Here \[ \mathcal V(x)=\sum\delta_{\mu_k}(x) V^{-p' +1}(x), \] \(\delta_{\mu}\) denotes the Dirac delta function at point \(\mu\), and \(V\) is defined by \[ \left(\int_0^{\infty}\frac{\nu(s)ds}{(s+t)^p}\right)^{1/p}\equiv\frac{1}{t}V^{1/p}(t). \] If \(0<p\leq 1, g(x)\geq 0\), one has \[ \sup_{0\leq f\uparrow}\frac{\int_0^{\infty} fg}{\left(\int_0^{\infty}\left(\frac{1}{x}\int_0^x f\right)^p\nu(x) dx\right)^{1/p}}=\sup_{t>0}\left(\int_0^t g\right)V^{-1/p}(t). \] This leads to a duality result \[ \Gamma_p^*(\nu)=\Gamma_{p'}(t^{p'}\mathcal V(t)), 1<p<\infty,\frac{1}{p} + \frac{1}{p'}=1. \] The result is implicit in that the dual measure \(\mathcal V\) depends on the points \(\mu_k\); it is explicit in that there is a formula that tells how to compute the \(\mu_k\). The measure \(\mathcal V\) also enters into the necessary and sufficient conditions for estimates \[ |Mf|_{q,\omega} \leq |f|_{p,\nu}, \] and \[ |Hf|_{q,\omega} \leq |f|_{p,\nu}, \] where \(M\) is the usual \(n\)-dimensional Hardy-Littlewood maximal function and \(H\) is the one dimensional Hilbert transform. The precise statements are too complicated for a brief summary; one must see the paper.