Journal of Operator Theory
Volume 69, Issue 1, Winter 2013 pp. 257-277.
On the closure of positive flat moment matricesAuthors: Lawrence Fialkow (1) and Jiawang Nie (2)
Author institution: (1) Department of Computer Science, State University of New York, New Paltz, New York 12561, U.S.A.
(2) Department of Mathematics University of California San Diego, 9500 Gilman Drive, La Jolla, CA 92093, U.S.A.
Summary: Let y≡y(2d)={yi}i∈Zn+,|i|⩽ denote a real n-dimensional multisequence of degree 2d, y_{0} > 0. Let L_{y}:\mathbb{R}_{2d}[x_{1},\ldots,x_{n}] \mapsto \mathbb{R} denote the Riesz functional, defined by L_{y}( \sum_{|i|\leqslant 2d} a_{i}x^{i}) = \sum a_{i}y_{i}, and let M_{d}(y) denote the corresponding moment matrix. Positivity of L_{y} plays a significant role in the Truncated Moment Problem and in the Polynomial Optimization Problem, but concrete conditions for positivity are unknown in general. M_{d}(y) is {\it{flat}} if \mathrm{rank}M_{d}(y) = \mathrm{rank}M_{d-1}(y); it is known that if M_{d}(y) is positive semidefinite and flat, then y has a representing measure (and L_{y} is positive). Let \mathcal{F}_{d} := \{y\equiv y^{(2d)}: M_{d}(y)\succeq 0 \mbox{ is flat}\}. If y\in \overline{\mathcal{F}}_{d} (the closure), then y does not necessarily have a representing measure, but L_{y} is positive, so M_{d}(y)\succeq 0 and, moreover, \mathrm{rank}M_{d}(y)\leqslant \mathrm{dim}\mathbb{R}_{d-1}[x_{1},\ldots,x_{n}]. We prove, conversely, that these positivity and rank conditions for M_{d}(y) are sufficient for membership in \overline{\mathcal{F}}_{d} in two basic cases: when n=1, d\geqslant 1, and when n=d=2.
DOI: http://dx.doi.org/10.7900/jot.2010may11.1890
Keywords: truncated moment sequence, Riesz functional, K-positivity, flat moment matrix, representing measure, polynomial optimization
Contents Full-Text PDF