Bivariate Truncated Moment Sequences with the Column Relation X ​ Y = X m + q ⁡ ( X ) XY=X^{m}+q(X) , with q q of degree m − 1 m-1

Seonguk Yoo, Aljaž Zalar · 2024

Read Bivariate Truncated Moment Sequences with the Column Relation X ​ Y = X m + q ⁡ ( X ) XY=X^{m}+q(X) , with q q of degree m − 1 m-1 by Seonguk Yoo, Aljaž Zalar (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-2412-21020.txt.

Description

Abstract . When the algebraic variety associated with a truncated moment sequence is finite, solving the moment problem follows a well-defined procedure. However, moment problems involving infinite algebraic varieties are more complex and less well-understood. Recent studies suggest that certain bivariate moment sequences can be transformed into equivalent univariate sequences, offering a valuable approach for solving these problems.

Catalog record

CREATORSeonguk Yoo, Aljaž Zalar
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE71 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-21020

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Bivariate Truncated Moment Sequences with the Column Relation X ​ Y = X m + q ⁡ ( X ) XY=X^{m}+q(X) , with q q of degree m − 1 m-1 Thanks: The second-named author was supported by the ARIS (Slovenian Research and Innovation Agency) research core funding No. P1-0228 and grant No. J1-50002. Seonguk Yoo, Aljaž Zalar Abstract . When the algebraic variety associated with a truncated moment sequence is finite, solving the moment problem follows a well-defined procedure. However, moment problems involving infinite algebraic varieties are more complex and less well-understood. Recent studies suggest that certain bivariate moment sequences can be transformed into equivalent univariate sequences, offering a valuable approach for solving these problems. In this paper, we focus on addressing the truncated moment problem (TMP) for specific rational plane curves. For a curve of general degree we derive an equivalent Hankel positive semidefinite completion problem. For cubic curves, we solve this problem explicitly, which resolves the TMP for one of the four types of cubic curves, up to affine linear equivalence. For the quartic case we simplify the completion problem to a feasibility question of a three-variable system of inequalities. 1991 Mathematics Subject Classification Primary 47A57, 44A60; Secondary 15-04, 47A20, 32A60 1. Introduction Given a real 2-dimensional multisequence of degree m , \beta\equiv\beta^{(m)}=\{\beta_{00}, \beta_{10}, \beta_{01},\cdots,\beta_{m,0},\beta_{m-1,1},\cdots,\beta_{1,m-1},\beta_{0,m}\} with \beta_{00}>0 , the truncated moment problem (TMP) entails finding necessary and sufficient conditions for the existence of a positive Borel measure \mu such that \operatorname{supp}\,\mu\subseteq\mathbb{R}^{2} and \displaystyle\beta_{i,j}\equiv\beta_{(i,j)}=\int x^{i}y^{j}\,\,d\mu\,\,\,(0\leq i+j\leq 2n;\ i,j\in\mathbb{Z}_{+}). In this context, we refer to \mu a representing measure (rm) for \beta or the moment matrix M(n) defined below. When the order of a moment sequence is even, such as m=2n for some n\in{\mathbb{N}} , it is possible to define the moment matrix M(n)\equiv M(n)(\beta^{(2n)}) of \beta as follows: {M}(n)\equiv{M}(n)(\beta^{(2n)}):=(\beta_{\textbf{i}+\textbf{j}})_{\textbf{i},\,\textbf{j}\in\mathbb{Z}^{2}_{+}:|\textbf{i}|,\,|\textbf{j}|\leq n}, where \textbf{i}+\textbf{j} stands for the coordinate-wise sum and |\textbf{i}| is the sum of coordinates of i . To guarantee the existence of a representing measure for \beta^{(2n)} , it is essential that the matrix M(n) is positive semidefinite. However, there are additional conditions that must be met. To examine these conditions, let \mathbb{R}[x,y]_{n} denote the set of bivariate polynomials in \mathbb{R}[x,y] with degree at most n . We arrange the columns of M(n) according to the monomials \{\textit{1},X,Y,X^{2},XY,Y^{2},\ldots,X^{n},\ldots,Y^{n}\} in degree-lexicographic order. For instance, a sextic moment matrix M(3) would appear as follows: \displaystyle\hskip 48.36967pt\textit{1}\hskip 17.07164ptX\hskip 17.07164ptY\hskip 14.22636ptX^{2}\hskip 7.11317ptXY\hskip 9.95845ptY^{2}\hskip 11.38109ptX^{3}\hskip 5.69054ptX^{2}Y\hskip 5.69054ptXY^{2}\hskip 7.11317ptY^{3} \displaystyle\begin{array}[]{r}\ \textit{1}\\ X\\ Y\\ X^{2}\\ XY\\ Y^{2}\\ X^{3}\\ X^{2}Y\\ XY^{2}\\ Y^{3}\end{array}\!\!\left(\begin{array}[]{c|cc|ccc|cccc}\beta_{00}&\beta_{10}&\beta_{01}&\beta_{20}&\beta_{11}&\beta_{02}&\beta_{30}&\beta_{21}&\beta_{12}&\beta_{03}\\ \hline\cr\beta_{10}&\beta_{20}&\beta_{11}&\beta_{30}&\beta_{21}&\beta_{12}&\beta_{40}&\beta_{31}&\beta_{22}&\beta_{13}\\ \beta_{01}&\beta_{11}&\beta_{02}&\beta_{21}&\beta_{12}&\beta_{03}&\beta_{31}&\beta_{22}&\beta_{13}&\beta_{04}\\ \hline\cr\beta_{20}&\beta_{30}&\beta_{21}&\beta_{40}&\beta_{31}&\beta_{22}&\beta_{50}&\beta_{41}&\beta_{32}&\beta_{23}\\ \beta_{11}&\beta_{21}&\beta_{12}&\beta_{31}&\beta_{22}&\beta_{13}&\beta_{41}&\beta_{32}&\beta_{23}&\beta_{32}\\ \beta_{02}&\beta_{12}&\beta_{03}&\beta_{22}&\beta_{13}&\beta_{04}&\beta_{32}&\beta_{23}&\beta_{32}&\beta_{23}\\ \hline\cr\beta_{30}&\beta_{40}&\beta_{31}&\beta_{50}&\beta_{41}&\beta_{32}&\beta_{60}&\beta_{51}&\beta_{42}&\beta_{33}\\ \beta_{21}&\beta_{31}&\beta_{22}&\beta_{41}&\beta_{32}&\beta_{23}&\beta_{51}&\beta_{42}&\beta_{33}&\beta_{24}\\ \beta_{12}&\beta_{22}&\beta_{13}&\beta_{32}&\beta_{23}&\beta_{13}&\beta_{42}&\beta_{33}&\beta_{24}&\beta_{15}\\ \beta_{03}&\beta_{13}&\beta_{04}&\beta_{23}&\beta_{13}&\beta_{05}&\beta_{33}&\beta_{24}&\beta_{15}&\beta_{06}\\ \end{array}\right). When the matrix M(n) exhibits a column relation, it can be written as p(X,Y)={\mathbf{0}} for some polynomial p(x,y)=\sum_{ij}a_{ij}x^{i}y^{j}\in\mathbb{R}[x,y]_{n} , where each monomial is replaced by the column of M(n) indexed with this monomial and \mathbf{0} stands for a zero vector; this concept is known as functional calculus. For convenience, we sometimes refer to this column dependence relation as that of the moment sequence. Although column relations themselves are not polynomials, they can be interpreted as such and provide vital information about a representing measure for M(n) . Specifically, if \beta has a representing measure, the following condition must be satisfied [ 7 ] : p(X,Y)=\mathbf{0}\implies(pq)(X,Y)=\mathbf{0}\text{ for every }q\in\mathbb{R}[x,y]\text{ with }\deg(pq)\leq n. In this case, {M}(n) is said to be recursively generated (rg). Let \mathcal{Z}(p):=\{(x,y)\in\mathbb{R}^{2}\colon p(x,y)=0\} denote the zero set of the polynomial p . The following result helps determine the support of a representing measure for a truncated moment sequence: Proposition 1.1 . ( [ 6 , Proposition 3.1] ) Suppose \mu is a representing measure for \beta^{(2n)} . For p\in\mathbb{R}[x,y]_{n} , \operatorname{supp}\,\mu\subseteq\mathcal{Z}(p)\iff p(X,Y)=\mathbf{0}. Now, we aim to define a set contained the support of representing measures; the algebraic variety of \beta\equiv\beta^{(2n)} is given by \mathcal{V}_{\beta}\equiv\mathcal{V}\left({M(n)}\right):=\bigcap{}_{p(x,y)=0,\,\deg\,p\,\leq n}\;\mathcal{Z}(p). (2) Using Proposition 1.1 the following holds: if M(n) admits a representing measure \mu , then \operatorname{supp}\;\mu\subseteq\mathcal{V}_{\beta} and \operatorname{rank}\;{M}(n)\leq\operatorname{card}\;\operatorname{supp}\;\mu\leq\operatorname{card}\mathcal{V}_{\beta}. The second inequality in the above is known as the variety condition . Here, we introduce the Riesz functional defined on the space of all monomials. For a given moment sequence \beta\equiv\beta^{(m)} , the linear functional \Lambda_{\beta} is given by \Lambda_{\beta}\left(\sum_{0\leq i+j\leq m}a_{i,j}x^{i}y^{j}\right)=\sum_{0\leq i+j\leq m}a_{i,j}\beta_{i,j}, where all a_{i,j} are real numbers and i,j\in\mathbb{Z}_{+} . We say that \Lambda_{\beta} is K -positive for a closed set K\in\mathbb{R}^{2} if \Lambda_{\beta}(p)\geq 0\ \ \text{for all }p\in\mathbb{R}[x,y]_{m}\text{ such that }p|_{K}\geq 0. If, in addition, the conditions p|_{K}\geq 0 and p|_{K}\not\equiv 0 imply \Lambda_{\beta}(p)>0 , then \Lambda_{\beta} is said to be strictly K -positive . When K=\mathbb{R}^{2} , we simply refer to \Lambda_{\beta} as positive rather than K -positive. The K -positivity of \Lambda_{\beta} is a necessary condition for \beta to have a K -representing measure , i.e., a rm supported on K . Conversely, M. Riesz’s classical theorem shows that K -positivity is also sufficient to guarantee the existence of K -representing measures for infinite moment sequences. This result was later extended to \mathbb{R}^{n} by E. K. Haviland. Similar results are available for the truncated moment problem, see the reference [ 9 ] . One of the most significant results in truncated moment theory is the Flat Extension Theorem. This theorem states that if M(n) has a rank-preserving positive extension M(n+1) , then \beta^{(2n)} possesses a \operatorname{rank}M(n) -atomic representing measure [ 6 ] . The extension M(n+1) is referred to as a flat extension . A notable special case arises when \operatorname{rank}M(n)=\operatorname{rank}M(n-1) . In this situation, M(n) is referred to as flat , and \beta^{(2n)} has a unique \operatorname{rank}M(n) -atomic representing measure. We now provide a brief overview of how to find a flat extension of M(n) . Notice that each rectangular block of M(n) with the same degree moments forms a Hankel matrix. To construct an extension M(n+1) , consider the following form: M(n+1)=\begin{pmatrix}M(n)&B\\ B^{*}&C\end{pmatrix}, where B and C are Hankel matrices with some new moments of degree 2n-1 and 2n , respectively. To ensure that a prospective moment matrix M(n+1) is positive semidefinite, we need the following classical result: Theorem 1.2 . ( [ 1 , 21 ] ) Let A , B , and C be matrices of complex numbers, with A and C being square matrices. Then \begin{pmatrix}A&B\\ B^{*}&C\end{pmatrix}\geq 0\iff\begin{cases}A\geq 0,\\ B=AW\text{ for some }W\\ C\geq W^{*}AW.\end{cases}{\iff\begin{cases}A\geq 0,\\ B=AW\text{ for some }W,\\ C\geq B^{*}A^{\dagger}B,\end{cases}} where A^{\dagger} stands for the Moore-Penrose inverse of A . Moreover, \operatorname{rank}\begin{pmatrix}A&B\\ B^{*}&C\end{pmatrix}=\operatorname{rank}A\iff C=B^{\ast}A^{\dagger}B. Although finding the positive extension may appear straightforward, verifying that the C -block is indeed a Hankel matrix is not trivial. Recently, many intriguing interactions between moment theory and algebraic geometry have been uncovered. Solving truncated moment problems can be interpreted as finding the roots of a system of multivariate polynomial equations. By Richter’s result [ 17 ] (see also [ 20 , Theorem 1.24] ), up to recently more often credited to Bayer and Teichmann [ 2 ] , if a moment sequence \beta^{(2n)} has one or more representing measures, then at least one of these measures must be finitely atomic. Consequently, if a real sequence \beta^{(2n)} is associated with a finitely atomic representing measure \mu , it can be expressed as \mu=\sum_{\ell=1}^{r}\rho_{\ell}\delta_{(x_{\ell},y_{\ell})}, where r\leq\dim\mathbb{R}[x,y]_{2n} . Our task is to determine positive numbers \rho_{1},\ldots,\rho_{r} (referred to as densities ) and points (x_{1},y_{1}),\ldots,(x_{r},y_{r}) (referred to as atoms ) of the

Read the complete text