Summability of the coefficients of a multilinear form (Q2119385)

The author lets \(Z_p=\ell_p\) if \(1\le p<\infty\) and \(Z_p=c_0\) if \(p=\infty\), considers a positive integer \(m\) and a subset \(\Lambda\) of \(\mathbb{N}^m\) and investigates how the smallest value of \(s\) for which the series \(\sum_{(j_1,\dots,j_m)\in\Lambda}|T(e_{j_1},\dots,e_{j_m})|^{s}\) is summable for every \(m\)-linear mapping \(T: Z_{p_1}\times Z_{p_2}\times \dots\times Z_{p_m}\to \mathbb{C}\) is related to \(p_1,\dots,p_m\) and \(\Lambda\). Given \(\mathbf{p}=(p_1,\dots,p_m)\) and \(\lambda\in\mathbb{R}\), \(\left|\frac{1}{ \lambda\mathbf{p}}\right|:=\sum_{i=1}^m\left|\frac{1}{\lambda p_i}\right|\). For a subset \(\Lambda\) of \(\mathbb{N}^m\) and a positive integer \(n\), the author lets \[ \psi_\Lambda(n)=\max\{\mathrm{card}((A_1\times\dots\times A_m)\cap \Lambda): A_i\subset \mathbb{N}, \ \mathrm{card}(A_i)\le n\} \] and defines the combinatorial dimension of \(\Lambda\), \(\dim(\Lambda)\), by \[ \dim(\Lambda):=\limsup_{n\to\infty}\frac{\log \psi_\Lambda(n)}{\log n}. \] When there is \(C>0\) such that \(\psi_\Lambda(n)\le Cn^{\dim(\Lambda)}\) for every \(n\), then \(\dim(\Lambda)\) is said to be exact. The Hardy-Littlewood exponent of \(\Lambda\) of order \(\mathbf{ p}\), \(HL(\Lambda, \mathbf{p})\), is defined as the infimum of the set of all \(s\ge 1\) for which there is \(C>0\) such that for all \(m\)-linear mappings \(T\colon Z_{p_1}\times Z_{p_2}\times\dots\times Z_{p_m}\to \mathbb{C}\), \[ \left(\sum_{(j_1,\dots,j_m)\in\Lambda}|T(e_{j_1},\dots,e_{j_m})|^{s}\right) ^{1/s}\le C\|T\|. \] Given an infinite set \(\Lambda\subset \mathbb{N}^m\) and \(\mathbf{p}=(p_1,\dots, p_m)\) with \(\left|\frac{1}{\mathbf{p}}\right|<1\), the author shows that \[ HL(\Lambda,\mathbf{p})^{-1}\ge \frac{\dim(\Lambda)+1}{2\dim(\Lambda)}-\left| \frac{1}{ \dim(\Lambda)\mathbf{p}}\right|\qquad \hbox{ when }\qquad \left|\frac{1}{\mathbf{p}}\right|<\frac{1}{2} \] and \[ HL(\Lambda,\mathbf{p})^{-1}\ge 1-\left|\frac{1}{\mathbf{p}}\right|\qquad \hbox{ when }\qquad\left|\frac{1}{\mathbf{p}}\right|>\frac{1}{2}. \] In addition, when \(\left|\frac{1}{\mathbf{p}}\right|<\frac{1}{2}\) and \(\dim(\Lambda)\) is exact, the series \[ \left(\sum_{(j_1,\dots,j_m)\in \Lambda}|T(e_{j_1},\dots,e_{j_m})|^{s}\right)^{1/s} \] converges for \(s=\left(\frac{\dim(\Lambda)+1}{2\dim(\Lambda)}-\left|\frac {1}{\dim(\Lambda)\mathbf{p}}\right|\right)^{-1}\). The author also shows that for \(m\ge 2\), \(d\in [1,m]\) and if one of the following holds \begin{itemize} \item[(i)] \(m\) is even and \(d\ge 3/2\); \item[(ii)] \(m\) is odd and \(d\ge 3/2+\frac{1}{2\lfloor m/2\rfloor}\); \item[(iii)] \(m/d\) is an integer, \end{itemize} then there is a subset \(\Lambda\) of \(\mathbb{N}^m\) with \(\dim(\Lambda)=d\) so that for all \(p\in [2,\infty]\) with \(m/p\le 1/2\) \[ HL(\Lambda,\mathbf{p})^{-1}=\frac{d+1}{d}-\left|\frac{1}{2\mathbf{p}}\right|. \] In the final section of the paper, for \(m\ge 2\) and \(d\in[1,m]\) the author introduces \(\Gamma_{\mathrm{mult}}(m,d)\) as the set of all \(p\ge 1\) such that for all \(\epsilon>0\) there is \(C>0\) such that for all \(n\) in \(\mathbb{N}\) there is \(\Lambda\) in \(\{1,\dots,n\}^m\) with \(\mathrm{card}(\Lambda)\ge C^{-1}n^d\) and \(T=\sum_{(j_1,\dots,j_m\in\Lambda}\epsilon(j_1,\dots,j_m)x_{j_1}\dots x_{j_m}\) with \(|\epsilon(j_1,\dots,j_m)|=1\) such that \(\|T\|_{\mathcal L(^m \ell_p)}\le Cn^\epsilon\) and defines \(\gamma_{\mathrm{mult}}(m,d)\) to be the supremum of \(p\) in \(\Gamma_{\mathrm{mult}}(m,d)\). It is shown that \[ \gamma_{\mathrm{mult}}(m,d)\le \min\left(m-\lceil d\rceil+1,\frac{2m}{ d+1}\right). \]