Sparse moments of univariate step functions and allele frequency spectra (Q2135091)

The authors study the univariate moment problem of piecewise-constant density functions on the interval \([0, 1]\). Also, they study its consequences for an inference problem in population genetics as application of their results. Let \(M(A)\) be the convex cone of all moments \((m_a)_{a\in A}\) of the form \(m_a=\int x^a d\mu\) where \(A\) be a finite collection of non-negative integers and \(\mu\) is a nonnegative Borel measure on the unit interval \([0,1]\). In addition, suppose \(M(A)_k\) is the Euclidean closure of the set of moments \((m_a)_{a\in A}\) where \(m_a = \int_0^1 x^a f (x) \,dx\) and \(f\) is a nonnegative step function with at most \(k\) discontinuities. Then, \[ M_k(A) = M(A) \text{ if and only if } k \geq |A|-1. \] In other words, the authors show that, up to closure, any collection of \(n\) moments is achieved by a step function with at most \(n -1\) breakpoints(, and moreover, this bound is tight). They use this to show that any point in the \(n\)-th coalescence manifold in population genetics can be attained by a piecewise constant population history with at most \(n-2\) changes. Also, they prove that if we restrict to the moments of monotone density functions (instead of arbitrary step functions), only half as many break points are needed to obtain \(M_k(A)=M(A)\). Moreover, they show that both the moment cones and the coalescence manifold are projected spectrahedra. In addition, they explained the connections to the problem of finding a nearest point on them as a semidefinite program. In last section, they discuss some open questions concerning the Moment cones and Coalescence Manifolds, for example, biologically plausible choices for step functions, etc.