Absolutely Continuous Representing Measures of Moment Sequences with an Emphasis on the Regularity of the Densities
Philipp J. di Dio · 2024
Read Absolutely Continuous Representing Measures of Moment Sequences with an Emphasis on the Regularity of the Densities by Philipp J. di Dio (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2411-03145.txt.
Description
Abstract In this work we investigate and characterize linear functionals L : ℝ [ x 1 , … , x n ] → ℝ L:\mathds{R}[x_{1},\dots,x_{n}]\to\mathds{R} with absolutely continuous representing measures μ \mu , i.e., d μ ( x ) = g ( x ) d x \mathrm{d}\mu(x)=g(x)\,\mathrm{d}x for some density g g .
Catalog record
| CREATOR | Philipp J. di Dio |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2411-03145
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Absolutely Continuous Representing Measures of Moment Sequences with an Emphasis on the Regularity of the Densities Philipp J. di Dio Abstract In this work we investigate and characterize linear functionals L : ℝ [ x 1 , … , x n ] → ℝ L:\mathds{R}[x_{1},\dots,x_{n}]\to\mathds{R} with absolutely continuous representing measures μ \mu , i.e., d μ ( x ) = g ( x ) d x \mathrm{d}\mu(x)=g(x)\,\mathrm{d}x for some density g g . We focus on the regularity of g g and how it is determined and influenced by the moments s α = L ( x α ) s_{\alpha}=L(x^{\alpha}) . Journal: arXiv 2020 MSC Primary 44A60; Secondary 30E05, 26C05. 1 Introduction Let \mu be a measure on \mathds{R}^{n} with n\in\mathds{N} . For \alpha=(\alpha_{1},\dots,\alpha_{n})\in\mathds{N}_{0}^{n} we call the number s_{\alpha}:=\int_{\mathds{R}^{n}}x^{\alpha}\penalty\ \mathrm{d}\mu(x) with x^{\alpha}:=x_{1}^{\alpha_{1}}\cdots x_{n}^{\alpha_{n}} the \alpha -moment of \mu if it exists. The K -moment problem is: Given finitely or infinitely many real numbers s_{\alpha} , \alpha\in I\subseteq\mathds{N}_{0}^{n} , and a closed set K\subseteq\mathds{R}^{n} . Does there exist a measure \mu on K such that s_{\alpha}=\int_{K}x^{\alpha}\penalty\ \mathrm{d}\mu(x) holds for all \alpha\in I ? For classical results on the moment problem see e.g. [ 1 , 2 , 3 , 20 , 11 , 27 , 23 , 26 , 24 , 33 , 42 , 44 , 43 , 45 , 46 , 47 , 59 , 54 , 55 , 58 , 56 , 61 , 60 , 63 , 68 , 79 , 85 , 82 , 87 ] to name only a few references to demonstrate the rich and long history as well as the variety of applied techniques to this problem. In applications moments and moment problems appear frequently, e.g. in financial mathematics [ 5 , 86 ] , in (algebraic) statistics [ 74 , 88 , 69 , 7 , 23 ] , in optimization [ 62 ] , in shape reconstruction [ 8 , 70 , 71 , 67 , 72 , 36 , 12 , 83 , 35 , 37 , 38 ] , and in pattern recognition [ 52 , 22 , 21 , 83 , 6 ] . Especially for shape reconstruction and pattern recognition the concept of derivatives of moments and moment functionals was developed [ 24 ] , see also [ 66 , 49 , 14 , 73 ] . In these applications moments of absolutely continuous measures are the main focus, i.e., measures \mu of the form \mathrm{d}\mu(x)=g(x)\,\mathrm{d}x for some density function g . In [ 4 ] L^{1} -densities were treated. In the current work we extend this work and look especially at the regularity of the density g . We are here also facing the problem of signed representing measures, see e.g. [ 19 , 75 , 16 , 15 , 81 , 51 , 10 , 57 , 32 , 50 , 80 , 25 ] . The paper is structured as follows. In Section 2 we treat the setup of [ 4 ] in full generality, i.e., truncated moment functionals and sequences ( I is finite resp. finite dimensional vector spaces \mathcal{V} of measurable functions) on general measurable spaces (\mathcal{X},\mathfrak{A}) . In Theorem 2.5 we give the main theorem which is based on/appeared in [ 23 ] . Its proof is significantly easier than the more special version in [ 4 ] since we use here the concept of Dirac approximating families ( Definition 2.1 ) and Richter’s Theorem 2.4 [ 76 ] . In Example 2.3 we give an example of an inner moment functional which can not be represented by an absolutely continuous representing measure. For the rest of Section 2 we discuss and show how from Theorem 2.5 the results in [ 4 ] follow and how Theorem 2.5 also gives absolutely continuous representing measures for boundary moment functionals and on general measurable spaces \mathcal{X} . In Section 3 we then treat linear functionals L:\mathds{R}[x_{1},\dots,x_{n}]\to\mathds{R} such that the representing measures have a density g with \mathrm{supp}\,g\subseteq[0,1]^{n} . For that we employ the technique of derivatives of moments [ 24 ] and the Signed Hausdorff’s Theorem 3.9 [ 45 ] . We characterize linear functionals L having an absolutely continuous (signed) representing measure on [0,1]^{n} and additionally also characterize them with given regularity (differentiability). In Section 4 we then treat linear functionals L:\mathds{R}[x_{1},\dots,x_{n}]\to\mathds{R} without any support restrictions. Because of the flexibility of signed representing measures on all \mathds{R}^{n} we have to restrict the treatment to sequences s=(s_{\alpha})_{\alpha\in\mathds{N}_{0}^{n}} such that f(z):=\sum_{\alpha\in\mathds{N}_{0}^{n}}\frac{i^{|\alpha|}\cdot s_{\alpha}}{\alpha!}\cdot z^{\alpha} is entire. The function f turns out to be the characteristic function \hat{\mu} of the representing measure \mu and (inverse) Fourier transformation, which is well-known and used e.g. in signal processing, provides us with the density g and hence characterizations of (signed) absolutely continuous representing measures. In Section 5 we summarize the main results. 2 Absolutely Continuous Representing Measures of Truncated Moment Functionals 2.1 Main Result To formulate and prove the main results in this section about truncated moment functionals we need the following concept. Definition 2.1 ( [ 23 , p. 3022, general setting (c)] ) . Let (\mathcal{X},\mathfrak{A}) be a measurable space and let \mathcal{V} be a vector space of measurable functions f:\mathcal{X}\to\mathds{R} . A family \mathcal{F}=\{\delta_{\sigma,x}\,|\,x\in\mathcal{X},\ \sigma\in\Lambda\} , \Lambda a set, of measures \delta_{\sigma,x} on \mathcal{X} such that for any x\in\mathcal{X} there exists a sequence (\sigma_{x,i})_{i\in\mathds{N}_{0}}\subseteq\Lambda with \lim_{i\to\infty}\int_{\mathcal{X}}f(y)\penalty\ \mathrm{d}\delta_{\sigma_{x,i},x}(y)=f(x) (1) for all f\in\mathcal{V} is called a Dirac-approximating family for \mathcal{V} . Our guiding example of a Dirac-approximating family is the following which is also absolutely continuous with respect to the Lebesgue measure \lambda . Example 2.2 . Let \Lambda=(0,1) , let \lambda be the Lebesgue measure on \mathds{R} , and let g_{\sigma,x}(y):=\frac{1}{\sqrt{2\pi\sigma^{2}}}\cdot e^{-\frac{(x-y)^{2}}{2\sigma^{2}}} be the Gaussian distribution centered at x\in\mathds{R} with variance \sigma\in\Lambda . Then we have \lim_{\sigma\searrow 0}\int_{\mathds{R}}f(y)\cdot g_{\sigma,x}(y)\penalty\ \mathrm{d}\lambda(y)=f(x) for all f\in C_{b}(\mathds{R},\mathds{R}) bounded continuous functions. Therefore, \{\delta_{\sigma,x}\,|\,\sigma\in(0,1),\ x\in\mathds{R}\} with \delta_{\sigma,x} defined by \mathrm{d}\delta_{\sigma,x}(y):=g_{\sigma,x}(y)\,\mathrm{d}\lambda(y) is a Dirac-approximating family for C_{b}(\mathds{R},\mathds{R}) . \circ We now show that condition ( 1 ) for Dirac-approximating families is crucial in the existence in absolutely continuous representing measures. Example 2.3 . Let \mathcal{X}=[0,2] , let \mathcal{V}=\mathrm{lin}\,\{f_{0},f_{1},f_{2}\} with f_{0}(x)=1 , f_{1}(x)=x , and f_{2}(x):=\begin{cases}x^{2}&\text{for}\ x\in[0,2]\setminus\{1\}\ \text{and}\\ -1&\text{for}\ x=1.\end{cases} Then \lim_{\sigma\searrow 0}\int_{0}^{2}f_{2}(y)\cdot g_{\sigma,1}(y)\penalty\ \mathrm{d}\lambda(y)=1\neq-1=f_{2}(1). In fact, \mathcal{V} has no Dirac-approximation family which is absolutely continuous with respect to the Lebesgue measure \lambda on [0,2] . Hence, the linear functional L:\mathcal{V}\to\mathds{R} represented by \lambda+\frac{8}{3}\delta_{1} is in the interior of the moment cone but can not be represented by an absolutely continuous measure with respect to \lambda . To see this assume \rho(x)\,\mathrm{d}\lambda(x) with \rho\in C([0,2],\mathds{R}) represents L . Then \int_{0}^{2}f_{2}(x)\cdot\rho(x)\penalty\ \mathrm{d}\lambda(x)=L(f_{2})=0 implies \mathrm{supp}\,\rho\subseteq\{0\} which implies L(1)=0 . This is a contradiction since L(1)>0 . \circ We have seen that if ( 1 ) does not hold with a family of absolutely continuous measures then we can construct a moment functional in the interior of the moment cone which has no absolutely continuous representing measure with respect to the Lebesgue measure. We now prove that if there is at least one Dirac-approximating family of absolutely continuous measures with respect to the Lebesgue measure then any moment functional in the interior of the moment cone has an absolutely continuous