[MITgcm-support] Sensitivities, OpenAD, tiles (Martin Losch)

Wed Sep 9 03:12:33 EDT 2015

Hi Heriberto,

I see the problem, but I am afraid I cannot help. I didn’t realise that TAF code and OpenAD code use different I/O routines.

 Let’s hope for someone with OpenAD experience to reply.

M.

> On 08 Sep 2015, at 18:03, Heriberto Vazquez <heriberto1mx at gmail.com> wrote:
> 
> Hello Martin
> 
> Thanks for your answer.
> 
> As you say it looks like a benign problem because the outputs looks just as I expected them.
> 
> I did check addummy_in_stepping but unfortunately that subroutine is only called by TAF, in fact is a subroutine only defined for TAF. I can do some modifications in it and I did, but it turns out that adexch is a subroutine only for TAF as well. In pkg/openad, there is a subroutine designed for printing out the adjoint which don't work. I do some modifications and now it works but there is not exchanges in that subroutine. After check out addummy_in_stepping, I tried to implement the same exchanges for OpenAD and using directly exch1_rl (by the way that subroutine is defined in another file in pkg/openad/ called externaldummies) after few changes I can make a progress and it work (the compiling and I can run the adjoint as well) but the output were still with the problem in the edges, in fact now the problem were in adjetan as well (I didn't have problem with adjetan before I set the exchanges in openad_dumpAdjoint).
> 
> I look at places where exchanges were done for TAF and I found that only addummy_in_stepping uses exchanges, so for that reason I don't know where to look at for OpenAD case.
> 
> You mentioned that you faced a similar problem with TAF, did you solve it only with exchanges in addummy_in_stepping? Why in my case without exchanges in my openad_dumpAdjoint I got a good eta but not the other variables? Is there another place where I can take a look at to see if I can do something?
> 
> Thanks in advance for your reply
> 
> Cheers,
> 
> Heriberto
> 
> On Fri, Sep 4, 2015 at 9:00 AM, <mitgcm-support-request at mitgcm.org> wrote:
> Send MITgcm-support mailing list submissions to
>         mitgcm-support at mitgcm.org
> 
> To subscribe or unsubscribe via the World Wide Web, visit
>         http://mitgcm.org/mailman/listinfo/mitgcm-support
> or, via email, send a message with subject or body 'help' to
>         mitgcm-support-request at mitgcm.org
> 
> You can reach the person managing the list at
>         mitgcm-support-owner at mitgcm.org
> 
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of MITgcm-support digest..."
> 
> 
> Today's Topics:
> 
>    1. Sensitivities, OpenAD, tiles (Heriberto Vazquez)
>    2. Re: Sensitivities, OpenAD, tiles (Martin Losch)
> 
> 
> ----------------------------------------------------------------------
> 
> Message: 1
> Date: Thu, 3 Sep 2015 17:48:08 -0700
> From: Heriberto Vazquez <heriberto1mx at gmail.com>
> To: mitgcm-support at mitgcm.org
> Subject: [MITgcm-support] Sensitivities, OpenAD, tiles
> Message-ID:
>         <CABO7yNS7d8Zf865a=PBV47g6=e=qxe8u-u39-XzMe+F5e7F1EA at mail.gmail.com>
> Content-Type: text/plain; charset="utf-8"
> 
> Hello everyone
> 
> I am doing some research in the Gulf of Mexico, so far I have an
> implementation of MITgcm and its adjoint built with OpenAD. Everything
> looks just fine except because in the sensitivity plots I can see the
> tiles' edges with relative high values in sensitivities and in the corner
> of the tiles with the highest values.
> 
> Some peculiarities are the fact that the sensitivity in adjetan works just
> fine without problem with the edges but in sensitivity of adjuvel and
> adjvvel I can see the lines; for adjuvel are more visible the vertical
> lines (north-south) and for adjvvel are more visible the horizontal lines
> (east-west), but for salinity and temperature both lines (horizontal and
> vertical) are visible and in the corners of the tiles appear the biggest
> sensitivities.
> 
> I tried changing advection schemes and the size in the overlapping but the
> lines are still there.
> 
> It looks like is an exchange problem, but I don't know where to look at.
> 
> Do you have any recommendations or suggestions?
> 
> Thanks in advance for your help
> 
> Cheers,
> 
> Heriberto
> Posdoctoral Researcher
> Scripps Institution of Oceanography
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20150903/febfe73c/attachment-0001.htm>
> 
> ------------------------------
> 
> Message: 2
> Date: Fri, 4 Sep 2015 09:13:54 +0200
> From: Martin Losch <Martin.Losch at awi.de>
> To: MITgcm Support <mitgcm-support at mitgcm.org>
> Subject: Re: [MITgcm-support] Sensitivities, OpenAD, tiles
> Message-ID: <3EE7FB3B-BAF7-405E-9261-5D02B469FA19 at awi.de>
> Content-Type: text/plain; charset="utf-8"
> 
> Hi Heriberto,
> 
> I haven?t worked with OpenAD yet, but if your model simulations does not crash, then I assume this this is a benign problem associated only with the output. I had a similar problem with adjoint seaice variables (and TAF code), that weren?t properly exchanged before writing to disk. Please have a look at pkg/autodiff/addummy_in_stepping.F which is the routine that organizes the output of the adjoint variables. You?ll see that there are plenty of ?call adxexch?, but unfortunately only if ALLOW_OPENAD is not defined. I guess you can probably try and fix the code for the OpenAD case youself (and contribute your fixes).
> 
> Martin
> 
> > On 04 Sep 2015, at 02:48, Heriberto Vazquez <heriberto1mx at gmail.com> wrote:
> >
> > Hello everyone
> >
> > I am doing some research in the Gulf of Mexico, so far I have an implementation of MITgcm and its adjoint built with OpenAD. Everything looks just fine except because in the sensitivity plots I can see the tiles' edges with relative high values in sensitivities and in the corner of the tiles with the highest values.
> >
> > Some peculiarities are the fact that the sensitivity in adjetan works just fine without problem with the edges but in sensitivity of adjuvel and adjvvel I can see the lines; for adjuvel are more visible the vertical lines (north-south) and for adjvvel are more visible the horizontal lines (east-west), but for salinity and temperature both lines (horizontal and vertical) are visible and in the corners of the tiles appear the biggest sensitivities.
> >
> > I tried changing advection schemes and the size in the overlapping but the lines are still there.
> >
> > It looks like is an exchange problem, but I don't know where to look at.
> >
> > Do you have any recommendations or suggestions?
> >
> > Thanks in advance for your help
> >
> > Cheers,
> >
> > Heriberto
> > Posdoctoral Researcher
> > Scripps Institution of Oceanography
> > _______________________________________________
> > MITgcm-support mailing list
> > MITgcm-support at mitgcm.org
> > http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> 
> 
> ------------------------------
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> End of MITgcm-support Digest, Vol 147, Issue 2
> **********************************************
> 
> 
> 
> -- 
> Saludos
> 
> Heriberto
> 
> __
> No podemos resolver problemas usando el mismo tipo de pensamiento que usamos cuando los creamos...
> Einstein
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support