dear everyone,

GalPaK3D has been around for over 5 years, but it keeps on evolving and you have been added to this mailing as you are/have been a user of this tool, primarily so that we can keep you informed of updates and new features.

# Regarding this mailing list:

(1) you can unsubscribe to this mailing list at anytime. Follow the link provided in the subscription email or follow the link at the very end of this message//

(2) If you move institute, don't forget to change your email address.

# if you are shy to send questions to me or to the mailing list, there is new forum. The aim is to have a public point to post issues and/or clarifications on the proper usage of the software.

# The documentation is not complete, but we are trying to bring it up to date. Feel free to send requests!

# GalPaK3D v1.20 is out since October and I invite you to check the CHANGELOG to learn about the new features. We try to link the documentation directly from the CHANGELOG file.

# This email galpak3d@python.org  is meant to communicate new infos regarding new releases to its users and thus is not meant for developers and intensive discussions on how to run/debug the code or implement new features.

# If you which to request new features, the best would be to start a new thread on the new forum!.

# ps: dont forget to refresh the http://galpak3d.univ-lyon1.fr/ webpage if you dont see these changes//

Here are a few tips in passing:

# There are several new kinematics models (Freeman, Spano, NFW etc.) and there is a new DiskModel class since v1.9.0.

# One can read the model/instrument parameters from a config file (saved by the default) / Still experimental, please cross-check and send reports //

# There is a compute_stats() method saving BICs etc in stats.dat to help you choose the best model // Also still experimental! //

# The 'mass' kinematic profile is no longer available currently since the Freeman disk was introduced. However, we are considering to put it back in the next release for comparison purposes/

# Use the 'tanh' rotation curve [default] to have more robust estimates for Vmax. The arctan is often degenerate with the turnover radius.

Regarding the MCMC sampler:

# One can use the new autorun API to tune automatically the random_scale parameter. Note, this is not fool proof..

# It is also possible to tune the random_scale manually for each parameter by passing a vector to random_scale in run_mcmc. It is multiplicative to the default values.

# GalPak3D is now compatible with the popular Emcee. Here are a note on this: emcee with its many walkers is very good at probing the posterior properly. But emcee is awfully slow to converge because it uses a Normal/Gaussian proposal distribution. In GalPaK3D we have used a Cauchy/Lorentzian proposal distribution which converges a lot faster thanks to its broader wings.

So, currently, one can use the autorun API to tune automatically the randomscale parameter, and find a solution close to the global minimum. Then use this solution as initial parameter for emcee and voilà. 

Perhaps, we'll manage to add pymultinest as MCMC solver, but feel free to volunteer on this and/or try other MCMC solvers. I am very interested also in pymc3, with the NUTS solver, but couldn't get it to work. This one has both a fast convergence and good posterior sampling.

# If you d like to participate in adding new MCMC solvers, we can add you on the git-lab locally@my institute. Hopefully, the git repo should become public soon.

All the best,

Nicolas