Title :
On an open problem related to the strict local minima of multilinear objective functions
Author :
Liang, Xue-Bin ; Wu, Li-De
Author_Institution :
Dept. of Comput. Sci., Fudan Univ., Shanghai, China
fDate :
11/1/1997 12:00:00 AM
Abstract :
This paper gives a combinatorial proof of a “yes” answer to an open question presented by Vidyasagar (ibid. vol.40, 1995), stated as follows: “Given a multilinear polynomial E(x): [0, 1] n→ℛ, is it true that Eb(x)=E(x)-bt x has a strict local minimum over the discrete set {0, 1}n for almost all b of sufficiently small norm?” The given combinatorial proof is completed directly by providing a sufficient condition for a conjecture on the strict local minima of multilinear polynomials, also postulated in Vidyasagar, to hold. In addition, a simple counter-example is presented to demonstrate that the conjecture may be not true if the provided sufficient condition is not satisfied
Keywords :
combinatorial mathematics; neural nets; optimisation; polynomials; analogue neural networks; discrete optimisation; multilinear objective functions; multilinear polynomial; objective function; strict local minima; sufficient condition; Automatic control; Computer science; Entropy; Neural networks; Polynomials; State-space methods; Sufficient conditions;
Journal_Title :
Automatic Control, IEEE Transactions on