# Polynomial of best approximation

Jump to: navigation, search

A polynomial furnishing the best approximation of a function in some metric, relative to all polynomials constructed from a given (finite) system of functions. If is a normed linear function space (such as or , ), and if is a system of linearly independent functions in , then for any the (generalized) polynomial of best approximation (*)

defined by the relation exists. The polynomial of best approximation is unique for all if is a space with a strictly convex norm (i.e. if and , then ). This is the case for , . In , which has a norm that is not strictly convex, the polynomial of best approximation for any is unique if is a Chebyshev system on , i.e. if each polynomial has at most zeros on . In particular, one has uniqueness in the case of the (usual) algebraic polynomials in , and also for the trigonometric polynomials in the space of continuous -periodic functions on the real line, with the uniform metric. If the polynomial of best approximation exists and is unique for any , it is a continuous function of .

Necessary and sufficient conditions for a polynomial to be a best approximation in the spaces and are known. For example, one has Chebyshev's theorem: If is a Chebyshev system, then the polynomial (*) is a polynomial of best approximation for a function in the metric of if and only if there exists a system of points , , at which the difference assumes values and, moreover, The polynomial (*) is a polynomial of best approximation for a function , , in the metric of that space, if and only if for , In , the conditions are sufficient for to be a polynomial of best approximation for , and if the measure of the set of all points at which is zero, they are also necessary; see also Markov criterion.

There exist algorithms for the approximate construction of polynomials of best uniform approximation (see e.g. , ).