Discrete approximations to continuous density functions that are \(L_ 1\) optimal (Q1064774): Difference between revisions

Let f(x), \(a\leq x\leq b\), be a continuous density function. The corresponding distribution function F(x) is to be approximated by a piecewise constant function \(G(x)=\sum_{(x<x_ i)}g(x_ i)\). The 2n parameters \(x_ i\) and \(D_ i=G(x_ i)\) for \(i=1,2,...,n\) are to be determined, such that the \(L_ 1\)-distance \(\int | F(x)-(x)| dx\) is minimal. From the characterization of a best approximation as a critical point, it follows that \(D_ i=F(x_ i+x_{i+1})/2\) and \(F(x_ i)=(D_{i-1}+D_ i)/2\) holds for \(i=1,2,...,n\). The author shows that all parameters are given, if \(x_ 1\) is known. Therefore, the approximation problem is reduced to a one-dimensional optimization problem.