中文

多维截断矩问题:由矩的导数进行形状与高斯混合重构

泛函分析 2019-07-17 v2 代数几何 概率论 统计理论 统计理论

摘要

本文引入矩与(矩)泛函的导数理论,以通过高斯混合、多面体的特征函数以及多面体的简单函数来表示矩泛函。除其他测度外,我们研究高斯混合、它们由矩进行的重构,尤其是表示矩泛函所需的高斯个数。我们发现存在矩泛函L:R[x1,,xn]2dRL:\mathbb{R}[x_1,\dots,x_n]_{\leq 2d}\to\mathbb{R},其可由(n+2dn)n(n+dn)+(n2)\binom{n+2d}{n} - n\cdot \binom{n+d}{n} + \binom{n}{2}个高斯之和表示,且不能更少。因此,对任意dNd\in\mathbb{N}ε>0\varepsilon>0,我们可找到nNn\in\mathbb{N}使得LL可由(1ε)(n+2dn)(1-\varepsilon)\cdot\binom{n+2d}{n}个高斯之和表示,且不能更少。上界为(n+2dn)1\binom{n+2d}{n}-1

关键词

引用

@article{arxiv.1907.00790,
  title  = {The multidimensional truncated Moment Problem: Shape and Gaussian Mixture Reconstruction from Derivatives of Moments},
  author = {Philipp J. di Dio},
  journal= {arXiv preprint arXiv:1907.00790},
  year   = {2019}
}

备注

arXiv admin note: substantial text overlap with arXiv:1903.00598. Author note: This is part II of arXiv:1903.00598. arXiv:1903.00598 was extended and splitting it into two parts was necessary. Part I contains the Caratheodory number from Hilbert functions parts (now arXiv:1903.00598v2). Part II contains the (Gaussian) mixtures and shape reconstruction from derivatives of moments parts