A comparator on a photodiode channel fires when the noise voltage
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
A comparator on a photodiode channel fires when the noise voltage Z, a standard Gaussian in units of its rms value, exceeds a threshold of 4. The false-alarm probability P=P(Z>4)=3.167124×10−5 is to be estimated by importance sampling, with f=1 for x>4 and 0 otherwise, p(x)=e−x2/2/2π and the shifted exponential proposal qλ(x)=λe−λ(x−4),x>4, where the rate λ>0 is to be chosen. Every draw from qλ has f=1, so the estimate is N−1∑iw(Xi) with w=p/qλ. Evaluate Gaussian tails P(Z>t) with software.
((a)) Show that qλ/p≥c for all x>4 with c=λ2πe4λ−λ2/2, and conclude that the variance of the estimator is finite for every λ>0.
((b)) Show that the second moment M(λ)=∫4∞p2/qλdx equals M(λ)=2πλeλ2/4−4λP(Z>42−2λ), and find the λ∗ that minimises it, to three significant figures.
((c)) Find the relative error of the estimate from N=104 draws, in per cent, at λ∗, at λ=4, and for the shifted unit Gaussian q(x)=e−(x−4)2/2/2π, for which M=e16P(Z>8) with P(Z>8)=6.22096×10−16.