Decoupling inequalities with exponential constants (Q6046117)

The authors study decoupling inequalities for random homogeneous polynomials. For a given Banach space \(X\), the authors study polynomials \(P:\mathbb{C}^n\to X\) of the form \[ P(z)=\sum_{\alpha\in\Lambda\subseteq\mathbb{N}_0^n}x_\alpha z_1^{\alpha_1}\cdots z_n^{\alpha_n}\,, \] where we write \(z=(z_1,\ldots,z_n)\) and \(\alpha=(\alpha_1,\ldots,\alpha_n)\), and where \(x_\alpha\in X\) for each \(\alpha\in\Lambda\). Such a polynomial is called \(m\)-homogeneous if \(|\alpha|=m\) for each \(\alpha\in\Lambda\) with \(x_\alpha\not=0\), and is called tetrahedral if each variable appears with exponent at most one. The decoupling inequalities studied here make use of the associated symmetric \(m\)-linear operator \(M:\mathbb{C}^{n\times m}\to X\) defined such that \(M(z,\ldots,z)=P(z)\) for each \(z\in\mathbb{C}^n\). Under the condition that the underlying Banach space \(X\) has finite cotype, the authors establish estimates for \((\mathbb{E}\lVert P(\xi)\rVert^p)^{1/p}\) for a \(m\)-homogeneous tetrahedral polynomial \(P\) in terms of \(M(\xi^{(1)},\ldots,\xi^{(m)})\), where \(\xi,\xi^{(1)},\ldots,\xi^{(m)}\) are independent and identically distributed, thus reducing the complexity of the dependence in the resulting expression. Under stronger geometric conditions on the space \(X\), the authors also compare random polynomials with independent random sums, reducing the complexity of the dependence yet further. Finally, the authors also give analogous inequalities comparing \(P(\xi)\) with \(M(\xi^\prime,\xi,\ldots,\xi)\), in which only one instance of \(\xi\) is replaced by an independent copy \(\xi^\prime\).