The number is to be estimated by Monte Carlo with uniform on and

PHYSICS · Multi-part · Fermi University Physics · Mathematical Methods for Physics · Probability, statistics and stochastic methods · Probability, statistics and stochastic methods

Error, Chance and the Machine, Ch Monte Carlo integration and variance reduction

Problem

The number ln⁡2=∫01du/(1+u)\ln2=\int_0^1\mathrm{d}u/(1+u) is to be estimated by Monte Carlo with UU uniform on (0,1)(0,1) and f(u)=1/(1+u)f(u)=1/(1+u). The control variate is g(u)=ug(u)=u, whose mean μg=12\mu_g=\tfrac12 is known, so that each draw contributes f(U)−c (g(U)−μg)f(U)-c\,\big(g(U)-\mu_g\big) to the average. ((a)) Show that var(f)=12−(ln⁡2)2\mathrm{var}(f)=\tfrac12-(\ln2)^2 and cov(f,g)=1−32ln⁡2\mathrm{cov}(f,g)=1-\tfrac32\ln2, evaluate both, and find the optimal coefficient c∗c^*. ((b)) Find the correlation coefficient ρfg\rho_{fg}, and the factor by which the variance per draw at c∗c^* is smaller than the plain variance var(f)\mathrm{var}(f). ((c)) A student takes c=1c=1, as in the example f=eUf=e^{U}, g=1+Ug=1+U of the text. Find the variance per draw at c=1c=1 and compare it with plain sampling. ((d)) Explain in two or three sentences why a negative c∗c^* causes no difficulty.

Browse the problem library