\magnification=1200
\baselineskip=20pt
\nopagenumbers
\font\big=cmr12 scaled \magstep2

\centerline{\bf STANFORD UNIVERSITY}
\centerline{\bf DEPARTMENT OF STATISTICS}
\centerline{\big BERKELEY-STANFORD JOINT COLLOQUIUM}
\bigskip

\baselineskip=12pt
\centerline{4:00 p.m., Tuesday, Apr 12, 2005}
\centerline{Room 60, Evans Hall, Berkeley}
%\centerline{(Cookies at 3:45 in 1st Floor Lounge)}

\bigskip
\baselineskip=15pt
\centerline{\sl Art B. Owen }
\centerline{Department of Statistics}
\centerline{Stanford University}
\bigskip
\centerline{\bf Quasi-Monte Carlo for Markov Chain Monte Carlo}
\bigskip
Abstract:

There has been considerable progress recently in developing
improvements to Monte Carlo methods.
Quasi-Monte Carlo (QMC) methods improve MC accuracy by 
using a more balanced sampling. Markov chain Monte Carlo (MCMC)
widens the applicability of MC to problems where we cannot
sample IID from the desired distribution.
The published intersection between these two methods is 
conspicuously small.  
We show that some but not all QMC points give rise to consistent
estimates in Metropolis-Hastings algorithms, including the
Gibbs sampler.  The suitable points are completely uniformly 
distributed (CUD).  In some numerical examples, variance
reduction factors seen range from just over 2 to just
over 240.



This is joint work with  Seth D. Tribble

\bye