scientific article; zbMATH DE number 1236334
Let \(\langle Y\rangle = \int y(x) p(x) dx\) denote the expected value of the function \(Y=y(x)\) of the random variable \(X\) having density \(p(x)\). The authors show theoretically and by example that error bounds for estimating \(\langle Y\rangle\) by the Monte Carlo method are exponential small. That is, for \(\delta>0\), let \(X_i\), \(i=1,\dots,n\) be a random sample from density \(p(\cdot)\), then \[ \begin{matrix}\text{ Pr}({1\over n}\sum_{i=1}^n y(X_i) \leq \langle Y\rangle-\delta\\ \text{Pr}({1\over n}\sum_{i=1}^n y(X_i) \geq \langle Y\rangle+\delta\end{matrix}\Biggr\} \leq e^{-n\alpha_\delta} \] for some value \(\alpha_\delta\) that depends on \(\delta\) and on the variance of the random variable \(Y\). The coefficient \(\alpha_\delta\) asymptotically increases to a limiting value with \(n\). Thus the limiting value, which is independent of \(n\), provides a more conservative bound. The derivation of this result is based on considerations of the distribution \(q(x)\), \[ q(x)=p(x)e^{-\lambda_0-\lambda y(x)} \] which has maximal entropy under the constraint that its mean equal \(\int y(x)q(x) dx\), see \textit{N. Agmon, Y. Alhassid}, and \textit{R. D. Levine} [J. Comput. Phys. 30, 250-258 (1979; Zbl 0406.65034)].
- Distributions with maximum entropy subject to constraints on their L-moments or expected order statistics
- A Characterization of probability densities with expected log likelihood
- scientific article; zbMATH DE number 4094576
- Estimation by Maximum Entropy Subject to Second‐Order Conditions
- Density reconstructions with errors in the data
- An efficient algorithm to compute maximum entropy densities
- Sample dependence in the maximum entropy solution to the generalized moment problem
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4223744)