Negative rates during inversion

This forum is dedicated to discuss all problems and suggestions related to the inversion software
Post Reply
ascarrasco
Posts: 2
Joined: Mon Jan 11, 2021 9:14 am

Negative rates during inversion

Post by ascarrasco »

Hi Marc, hi all,

I am trying to do some inversions of ellipticity curve (only) with dinver, using the latest version of the NA algorithm (with the new parameters Ns0, Ns, Nb, Nr, GiveUp, Nw). In this case, I am usually using Ns0=Ns=100000, Nb=0. Nr=50, GiveUp=90 and Nw=2 (still not entirely sure if this is the best set of inv params) for a model with 3 layers over a half-space. Anyway, my question is:

Sometimes, I get a negative Rate and the amount of valid models is less and less. I'm curious about this and I would like to learn what does it mean? Is this a good or a bad issue? How do I know when the process will stop? Should I stop the process when I see I'm permanently getting negative rates (and even "negative" valid models, then)? (because I left one inversion running, it got into negative values and never stopped) Should I change the set of inv params, maybe?

Thanks in advance for your help,
All the best and have you all a good week,
Sebastian
admin
Site Admin
Posts: 822
Joined: Mon Aug 13, 2007 11:48 am
Location: ISTerre
Contact:

Re: Negative rates during inversion

Post by admin »

Hi Sebastian,

Ns0=100000 is a huge number. With a minimum of 1 model the NA can start. Usually I start with 50 random models. From this initial population only the Nr (50 in your case) best models will be kept for the NA stage.

With only 3 layers, you should have 7 parameters (plus or minus depending on the options). If you get negative rates I suspect that a number of cells are declared as "atomic" (they are so small that it is not possible to generate new models in them, a consequence of discrete parameters, paper under preparation in Journal of Seismology for details). If so, the number of active cells may decrease. If you provide a copy of the log (at least a few lines), maybe I can confirm. It happens if you generate a lot of models, which is your case. With 3 layers, 10000 to 2000 models are usually enough.

I note that you have also parameter 'b' in your release. I guess that you are using the branch 3.4 which contains new developments that are not completely finalized. For safety and production, I would recommend branch 3.3 with the latest release 3.3.5.
ascarrasco
Posts: 2
Joined: Mon Jan 11, 2021 9:14 am

Re: Negative rates during inversion

Post by ascarrasco »

Thanks Marc for your answer, I'm attaching the last lines of the log messages.

I think I get the point, I started a run yesterday and it has not stopped yet, so I will stop it, reduce Ns0 and start a new run. Anyway, this is not always happening. I have some runs with a large Ns0 and it runs correctly (probably depending on the parameterization then?)

Also, I will install the suggested version.

Once again, thanks for the answer.
Sebastian
Attachments
negative_ratios_log.txt
(2.68 KiB) Downloaded 762 times
admin
Site Admin
Posts: 822
Joined: Mon Aug 13, 2007 11:48 am
Location: ISTerre
Contact:

Re: Negative rates during inversion

Post by admin »

I fixed a bug on branch 3.4 today that made dinver running indefinitely without producing new models. That was a rather stupid mistake and the correction is now available on the git repository.

You have a lot of models flagged as quarantined (or atomic for the latest release). These cells are not producing models any more. It might be the cause of your strange rates.

You may stay on branch 3.4 but upgrade to the latest modifications.
Post Reply