The Generalized Moment Problem with Complexity Constraint

Abstract

In this paper, we present a synthesis of our differentiable approach to the generalized moment problem, an approach which begins with a reformulation in terms of differential forms and which ultimately ends up with a canonically derived, strictly convex optimization problem. Engineering applications typically demand a solution that is the ratio of functions in certain finite dimensional vector space of functions, usually the same vector space that is prescribed in the generalized moment problem. Solutions of this type are hinted at in the classical text by Krein and Nudelman and stated in the vast generalization of interpolation problems by Sarason. In this paper, formulated as generalized moment problems with complexity constraint, we give a complete parameterization of such solutions, in harmony with the above mentioned results and the engineering applications. While our previously announced results required some differentiability hypotheses, this paper uses a weak form involving integrability and measurability hypotheses that are more in the spirit of the classical treatment of the generalized moment problem. Because of this generality, we can extend the existence and well-posedness of solutions to this problem to nonnegative, rather than positive, initial data in the complexity constraint. This has nontrivial implications in the engineering applications of this theory. We also extend this more general result to the case where the numerator can be an arbitrary positive absolutely integrable function that determines a unique denominator in this finite-dimensional vector space. Finally, we conclude with four examples illustrating our results. Mathematics Subject Classification (2000). Primary 30E05; Secondary 49K40.

The Circulant Rational Covariance Extension Problem: The Complete Solution

“…In the circulant rational covariance extension problem we consider Hermitian circulant matrices (7) which can be represented in form (8) where is the nonsingular cyclic shift matrix…”

“…In [21] and [22], it was shown that there exists a for each assignment of , and in [1] it was finally proved that this assignment is unique and smooth, yielding a complete parameterization suitable for tuning. Consequently, the rational covariance extension problem reduces to a trigonometric moment problem, where, for each , the remaining problem is to determine a unique such that (5) In [3] and [4], a convex optimization procedure to determine these was introduced, a result that has then been generalized in several directions [5]- [7], [9], [10], [17], [19], [23], [24].…”

Interior point solutions of variational problems and global inverse function theorems

“…Under mild technical conditions, it can be shown that F is proper [22,23]. Since J c is strictly convex, it has at most one critical point that must then be a minimizer.…”

“…Moreover the minimizer is an interior point, and the moment problem (11) is well-posed. A very complete analysis of this moment problem from a variational viewpoint is given in [23]. We return below with more general examples of wellconnected pairs of problems, where f need not generate an exact 1-form.…”

Semidefinite programming for gradient and Hessian computation in maximum entropy estimation

“…For more details on this approach the interested reader is referred to e.g. Byrnes and Lindquist [5], Borwein and Lewis [2], [3], [4], Georgeou [7], Mead and Papanicolaou [11], and Tagliani [12], [13]. As early as in [11], it was recognized that such entropy methods may outperform classical Padé-like approximations.…”