GALPROP
https://galprop.stanford.edu/forum/

paper on Bayesian analysis with galprop
https://galprop.stanford.edu/forum/viewtopic.php?f=3&t=110
Page 1 of 1

Author:  strong [ Fri Nov 05, 2010 1:58 am ]
Post subject:  paper on Bayesian analysis with galprop

a new paper on using galprop in a multinested Bayesian code is at
http://arxiv.org/abs/1011.0037

the supplementary data (galdef files, parameter chains, data files) can be found at see http://www.g-vo.org/pub/GALPROP

Author:  tomassetti [ Wed Nov 17, 2010 5:02 am ]
Post subject:  Re: paper on Bayesian analysis with galprop

Results on the propagation parameters are given in terms of Best-Fit (i.e. from Chi^2 minimization) and Posterior-Mean (i.e. expectation values of the posterior pdf's). In some cases they differ a lot. So...stupid question: which one really "matters"? What's the "ultimate" set of CR propagation parameter that can be determined (with Galprop) at best of our knowledge on the basis of existing data? BestFit (as I'd guess) or PosteriorMean ? Why have you quoted both...do I miss some relevant physical differences/complementarity in the two?

Author:  strong [ Wed Nov 17, 2010 5:07 am ]
Post subject:  Re: paper on Bayesian analysis with galprop

(NB they are not chi^2 but probabilities)
The posterior mean is definitely the one to use since it takes account of the full posterior, whereas the maximum posterior could be a local maximum or a narrow peak with little probability content.
Why we quote both: to illustrate the difference.
There is no 'ultimate' parameter set since we only considered a particular class of models, and did not include e.g. convection and other effects, or some other types of data which will give more constraints e.g. pbar (but we intend to in future).

Author:  tomassetti [ Wed Nov 17, 2010 5:22 am ]
Post subject:  Re: paper on Bayesian analysis with galprop

So.. the fact that the best-fit parameters are within the uncertainty band of the posterior pdfs is to be taken as a remarkable/untrivial result. Is that correct?

Author:  strong [ Wed Nov 17, 2010 5:27 am ]
Post subject:  Re: paper on Bayesian analysis with galprop

not at all remarkable, this happens in well-behaved distributions, which we have here.

Author:  tomassetti [ Wed Nov 17, 2010 6:30 am ]
Post subject:  Re: paper on Bayesian analysis with galprop

Ok... thank you very much.

Author:  tomassetti [ Sun Nov 21, 2010 11:08 am ]
Post subject:  Re: paper on Bayesian analysis with galprop

Another point. In that paper, the fragmentation chain starts from Max_Z = 14 (silicon). So I deduce that the quantities under study (B/C, 10Be/9Be, pbar/p) are not significantly affected by the presence of Z>14 nuclei... correct? If yes... can you quantify "not significantly" with some number? 1%?

Author:  strong [ Sun Nov 21, 2010 11:13 am ]
Post subject:  Re: paper on Bayesian analysis with galprop

It is completely negligible, certainly below 1%

Author:  avladim [ Mon Nov 22, 2010 1:09 am ]
Post subject:  Re: paper on Bayesian analysis with galprop

I recently tested it by making several runs in WebRun with different max_Z. I was using the "Conventional Model" example. I also stopped at max_Z=14, because the spectra in QuickPlots for B/C and 10Be/9Bebecame indistinguishable from higher max_Z at this point. Certainly well within the uncertainty in the experimental data.

Page 1 of 1 All times are UTC - 8 hours [ DST ]
Powered by phpBB® Forum Software © phpBB Group
https://www.phpbb.com/