The integral is estimated from antithetic pairs with uniform on

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 integral I=∫01f(u) duI=\int_0^1f(u)\,\mathrm{d}u is estimated from antithetic pairs Y=12[f(U)+f(1−U)]Y=\tfrac12\big[f(U)+f(1-U)\big] with UU uniform on (0,1)(0,1), and costs are compared at equal numbers of evaluations of ff. Three integrands are considered: f1(u)=u2f_1(u)=u^2, f2(u)=sin⁡πuf_2(u)=\sin\pi u and f3(u)=(u−12)3f_3(u)=(u-\tfrac12)^3. Let ρa\rho_{\rm a} be the correlation coefficient of f(U)f(U) and f(1−U)f(1-U). ((a)) Before calculating anything, predict for each function whether antithetic pairs lower the variance at equal cost, raise it or leave it unchanged, with a one-line reason. ((b)) For f1f_1, find ρa\rho_{\rm a} and the factor by which the variance at equal cost is reduced. ((c)) For f2f_2, find ρa\rho_{\rm a} and the same factor. ((d)) For f3f_3, find ρa\rho_{\rm a} and the variance of YY, and say why the factor 1/(1+ρa)1/(1+\rho_{\rm a}) cannot be quoted as a finite number here.

Browse the problem library