On the number of minima of a random polynomial (Q2483199): Difference between revisions

A random polynomial \(f\) over the reals with \(n\) variables and degree \(d\geq 2\) is considered. The problem is to compute, on average, the number of its critical points (the number of real roots of the system \(Df(x)= 0\)). It is proved; that the upper bound for the expected number of such critical points is \(\sqrt{2}\cdot(d- 1)^{(n+1)/2}\). Using the large deviation principle for the spectral value of large random matrices the upper bound \(K\cdot\exp(-n{\ln(3)\over 4}+{n+ 1\over 2})\ln(d- 1)\) is also derived, where \(K\) is a positive constant. This result indicates that most random polynomials of fixed degree have only saddle-points, which are neither local minima nor maxima. A closed form expression for the expected number of minima (resp. maxima) in terms of hypergeometric functions is also given.