Discretizing \(L_p\) norms and frame theory (Q2102130)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Discretizing \(L_p\) norms and frame theory |
scientific article |
Statements
Discretizing \(L_p\) norms and frame theory (English)
0 references
28 November 2022
0 references
Given \(1 \leq p < \infty\) and a probability space \(\Omega,\) the authors consider the problem of discretizing the \(L_p\) norm on an \(N\)-dimensional subspace of \(L_p (\Omega).\) Mathematically, this problem can be stated as follows. Let \(X^N \subseteq L_p (\Omega)\) be an \(N\)-dimensional subspace. The \(L_p\) norm of \(f \in X^N\) is given by \[ \| f \|^p_p = \int_{\Omega} |f(t)|^p \ d\mu(t). \] Given \(A < 1 < B,\) the problem is to choose sampling points \(\{t_j\}_{j = 1}^M \subseteq \Omega\) such that for all \(f \in X^N\) \[ A \| f \|^p_p \leq \frac1M \sum_{j = 1}^M |f(t_j)|^p \leq B \|f\|^p_p . \] In the study of the above problem, the authors consider a Nikol'skii-type inequality. An \(N\)-dimensional subspace \(X^N \subseteq L_p(\Omega)\) is said to satisfy a \((p, \infty)\)-Nikol'skii-type inequality for \(1 \leq p < \infty\) if there exists \(\beta > 0\) such that \[ \|x\|_{L_{\infty}} \leq \beta N^{1/p} \| x \|_{L_p} \quad \forall x \in X^N. \tag{1} \] The motivation for considering this is that previously it has been proved that for \(1 \leq p < \infty\) there are certain entropy conditions on \(X^N \subseteq L_p(\Omega)\) which guarantee that the \(L_p\)-norm on \(X^N\) can be discretized using \(M\) of the order of \(N(\log(N))^2.\) Even though these entropy conditions can be technical, they imply that the subspace satisfies a \((2, \infty)\)-Nikol'skii-type inequality which in turn implies a \((p, \infty)\)-Nikol'skii-type inequality for some \(1 \leq p < 2\). For \( 1\leq p < 2,\) the authors construct a class of subspaces \(X^N \subseteq L_p (\Omega)\) which uniformly satisfy (1) but the \(L_p\)-norm on \(X^N\) cannot be discretized using \(M\) on the order of \(N\) sampling points. It is shown that the problem of construction of frames that do stable phase retrieval is connected to the problem of discretizing the \(L_1\)-norm on a subspace of \(L_1(\Omega).\) In particular, the authors prove that to sample a continuous Parseval frame to obtain a frame which does stable phase retrieval, it is necessary to simultaneously discretize both the \(L_1\)-norm and the \(L_2\)-norm on the range of the analysis operator of the continuous frame.
0 references
discretization
0 references
Marcinkiewicz type discretization
0 references
phase retrieval
0 references
frames
0 references
0 references
0 references
0 references
0 references