No label defined (Q4532115)

The author exposes an algorithm for constructing piecewise linear functions \(g_+(x)\) and \(g_-(x)\) which approximate a given finite probability density \(p(x)\) with an arbitrary accuracy. Let \(M\) denotes the number of grid intervals in \(x\) which are used for constructing piecewise linear functions. The algorithm proposed possesses the following properties: \[ \langle g_+ \rangle (M)= 1 + e_+(M),\quad \langle g_- \rangle (M) = 1 - e_-(M),\quad \tau[g](M) \approx \text{const}, \] where \(\tau[g](M)\) is the time for modeling a random variable, \(\langle g_+ \rangle (M)\) and \(\langle g_- \rangle (M) \) tend to zero as \(M\to\infty\). The algorithm is investigated in detail for the beta-distribution with nonnegative parameters. Moreover, a modification of the algorithm is presented both for infinite densities and infinite intervals.