warm start
Page 1 of 2

Author:  anderson [ Fri Sep 03, 2010 11:59 am ]
Post subject:  warm start

I noticed that warm_start is fixed to 0. Does the web run do it automatically, or is the nuclei output wrong?

Author:  strong [ Fri Sep 03, 2010 12:08 pm ]
Post subject:  Re: warm start

it would require a previous run with output_nuclei_full=1,
which is then read in by the next run. not sure whether webrun can handle that
question for Andrey....

Author:  anderson [ Fri Sep 03, 2010 12:13 pm ]
Post subject:  Re: warm start

Okay. The gamma output should be the same in any case, no?

Author:  strong [ Fri Sep 03, 2010 12:20 pm ]
Post subject:  Re: warm start

after a warm start the propagation continues, so not necessarily unless it has already converged

Author:  avladim [ Fri Sep 10, 2010 9:16 pm ]
Post subject:  Re: warm start

I am not sure I understand what you mean by nuclei output being wrong. Warm_start is not used in WebRun. Any run that calculates gamma-rays must calculate cosmic ray propagation first (at least protons and electrons) -- that's how it is set up in WebRun now.

Author:  strong [ Fri Sep 10, 2010 10:44 pm ]
Post subject:  Re: warm start

as long as warm start is not allowed, this is fine.
but warm start is an interesting option, eg. when making long 3D runs for stochastic sources.
If you want that then the best thing would be to get the code.

Author:  anderson [ Thu Dec 09, 2010 5:31 pm ]
Post subject:  Re: warm start

Any work on this yet? I really would like to get accurate nuclei files out.

Author:  avladim [ Thu Dec 09, 2010 6:00 pm ]
Post subject:  Re: warm start

Sorry, Brandon, I did not realize that you were expecting some modifications to the service after the conversation above. I believe that everything that can be done using warm_start=1, can also be done with warm_start=0, so changing it was not on my to do list.

If you give me some details on what you want to accomplish, I am sure I will be able to help you figure out how to do it in WebRun.


Author:  anderson [ Mon Dec 13, 2010 12:35 pm ]
Post subject:  Re: warm start

Interesting. My impression was that without a warm start, some of the nuclei products would not be propagated, and therefore the nuclei output (at earth radius) would be inaccurate. And as Andy said, unless the gamma has converged, it will also be inaccurate. Do you mean to say that warm_start=0 on the server actually performs all the propagation correctly and that warm_start=1 is simply a time saver? That would certainly make sense... I am just used to a version where warm/not warm means the difference between accurate/inaccurate results.

Author:  avladim [ Mon Dec 13, 2010 12:42 pm ]
Post subject:  Re: warm start

You may be confusing warm_start with network_iterations. You need network_iterations=2 to get B/C accurately. But warm_start is not required to get accurate nuclear reactions network and propagation. Here is what the explanatory supplement ( ) says:

warm_start   = 0 read in nuclei file and continue run
Flag to indicate the run is to be started with results output from previous run with output gcr full =1. Useful for long runs which have to be broken up due to CPU time limitations. NB the galdef file should be the same as for the first run, apart from this parameter; the only possible differences should be the values of end timestep and timestep repeat2, and the generation of gamma-rays etc. The grid and selected nuclei, electrons etc. must be the same.

Author:  anderson [ Tue Dec 14, 2010 2:11 pm ]
Post subject:  Re: warm start

Oh wow. They actually accomplish the same thing, I had just overlooked network iterations and was doing things the hard way. Thanks very much.

Author:  avladim [ Tue Dec 14, 2010 2:34 pm ]
Post subject:  Re: warm start

You are welcome. Sorry I did not understand you at first, I am not familiar with the older Galprop version that you are used to.

Author:  strong [ Wed Dec 15, 2010 12:32 am ]
Post subject:  Re: warm start

it would anyway be a good idea to enable the warm start.
it could well be required for small constant timesteps to get accurate results.
the rapid-solution scheme is not necessarily always reliable.

Author:  avladim [ Wed Dec 15, 2010 1:46 am ]
Post subject:  Re: warm start

Andy, why can't one just specify a small time step and run a single calculation?
Implementing warm start in webrun will take some work, because fetching and staging additional files will be necessary, and specifying the run to be used for input will take some interface surgery. I would like to avoid it if it is not necessary.

Author:  strong [ Wed Dec 15, 2010 2:40 am ]
Post subject:  Re: warm start

indeed one should do that, especially now the machines are fast.
warm start would really only be required for following time-dependent runs, and can happily be left to those installing locally.
But we should really encourage the small-timestep runs. I have added something on that in the latest version of the Explanatory Supplement.
It looks like the timesteps are still fixed in Webrun even for advanced mode, could you enable changes ?

I registered this as a bug under

NB the small timesteps in are controlled by timestep_repeat, with e.g. start_timestep=end_timestep = 1000 (yr) and timestep_repeat = 100000. Put timestep_factor = 0.9 to terminate the run appropriately.
One can also use timestep_repeat2 (with timestep_repeat=0) but this is only available in 3D .

Page 1 of 2 All times are UTC - 8 hours [ DST ]
Powered by phpBB® Forum Software © phpBB Group