A means by which a table of values f(x0), f(x1),…, f(xm) at the distinct points x1, x2,…, xm, can be approximated (represented) by a function, say
where ϕ
i(
x),
i = 1,2,…,
n, are chosen, and the coefficients
ci, i = 1,2,…,
n, are to be determined. Typically
m >
n. The objective is to choose a fit that reduces the effect of random errors in the data combined with producing a curve that is smooth (no rapid changes or oscillations) between the data points. This is generally referred to as smoothing. The smoothing is often achieved by using low-degree polynomials (with suitable ϕ
i(
x) and the coefficients
ci are frequently determined by the least-squares criterion (
see approximation theory).
Compare interpolation.