The result for scaling the proposal density with dimension given in the paper appears to be generally useful and correct. Two comments:
1. There are some potential problems using the autocorrelation function to assess chain efficiency: it is quite possible for the autocorrelation to be small but the separation between effectively independence samples to be large. Jo has confirmed that similar problems can happen with the spectral estimator.
2. Results are restricted to using ND Gaussians as the proposal density. In highN (N>2) this is probably rather unrobust, and it may be better to use a lower dimensional proposal.
I've written up these comments in detail in the cosmomc notes
http://cosmologist.info/notes/cosmomc.ps.gz
Similar comments apply to [arxiv]astroph/0310723[/arxiv]
[astroph/0405462] Fast and reliable MCMC for cosmological parameter estimation
Authors:  Joanna Dunkley, Martin Bucher, Pedro G. Ferreira, Kavilan Moodley, Constantinos Skordis 
Abstract:  Markov Chain Monte Carlo (MCMC) techniques are now widely used for cosmological parameter estimation. Chains are generated to sample the posterior probability distribution obtained following the Bayesian approach. An important issue is how to optimize the efficiency of such sampling and how to diagnose whether a finitelength chain has adequately sampled the underlying posterior probability distribution. We show how the power spectrum of a single such finite chain may be used as a convergence diagnostic by means of a fitting function, and discuss strategies for optimizing the distribution for the proposed steps. The methods developed are applied to current CMB and LSS data interpreted using both a pure adiabatic cosmological model and a mixed adiabatic/isocurvature cosmological model including possible correlations between modes. For the latter application, because of the increased dimensionality and the presence of degeneracies, the need for tuning MCMC methods for maximum efficiency becomes particularly acute. 
[PDF] [PS] [BibTex] [Bookmark] 

 Posts: 1618
 Joined: September 23 2004
 Affiliation: University of Sussex
 Contact:

 Posts: 41
 Joined: November 22 2004
 Affiliation: ITP Heidelberg
 Contact:
[astroph/0405462] Fast and reliable MCMC for cosmological p
The scaling is indeed useful (and easy to implement). Funny enough, the adaptive
stepsize I used in AnlalyzeThis! provided roughly the same acceptance rate.
But as the adaptive stepping in AnalyzeThis! stops after "sensing" the distribution (i.e. some extended burnin), the formula is still useful for the stepproposal after "burnin".
stepsize I used in AnlalyzeThis! provided roughly the same acceptance rate.
But as the adaptive stepping in AnalyzeThis! stops after "sensing" the distribution (i.e. some extended burnin), the formula is still useful for the stepproposal after "burnin".