English

Mean squared error minimization for inverse moment problems

Optimization and Control 2012-09-03 v1

Abstract

We consider the problem of approximating the unknown density uL2(Ω,λ)u\in L^2(\Omega,\lambda) of a measure μ\mu on ΩRn\Omega\subset\R^n, absolutely continuous with respect to some given reference measure λ\lambda, from the only knowledge of finitely many moments of μ\mu. Given dNd\in\N and moments of order dd, we provide a polynomial pdp_d which minimizes the mean square error (up)2dλ\int (u-p)^2d\lambda over all polynomials pp of degree at most dd. If there is no additional requirement, pdp_d is obtained as solution of a linear system. In addition, if pdp_d is expressed in the basis of polynomials that are orthonormal with respect to λ\lambda, its vector of coefficients is just the vector of given moments and no computation is needed. Moreover pdup_d\to u in L2(Ω,λ)L^2(\Omega,\lambda) as dd\to\infty. In general nonnegativity of pdp_d is not guaranteed even though uu is nonnegative. However, with this additional nonnegativity requirement one obtains analogous results but computing pd0p_d\geq0 that minimizes (up)2dλ\int (u-p)^2d\lambda now requires solving an appropriate semidefinite program. We have tested the approach on some applications arising from the reconstruction of geometrical objects and the approximation of solutions of nonlinear differential equations. In all cases our results are significantly better than those obtained with the maximum entropy technique for estimating uu.

Keywords

Cite

@article{arxiv.1208.6398,
  title  = {Mean squared error minimization for inverse moment problems},
  author = {Didier Henrion and Jean-Bernard Bernard Lasserre and Martin Mevissen},
  journal= {arXiv preprint arXiv:1208.6398},
  year   = {2012}
}
R2 v1 2026-06-21T21:57:47.911Z