[MITgcm-devel] problem with 59c

[Martin.Losch at awi.de]

Hi Jinlun,

we are talking about the LSR solver on the C-grid. As far as I can see, the EVP solver has not been used for the high 
resolution cubed sphere yet. In one of her paper Elizabeth (J.Comp.Phys.,2001) claims that the EVP solver does not 
need zmin = 4.e8 as opposed to other methods, and then she cites you, as personal communication, that zmin=0 works 
too for other solvers, (yours?). So far I have used zmin = 0 for all my evp runs.

Dimitris has actually isolated the problem in the cs510 and it is zmin=0, with zmin=4e8 (the old value, that I removed 
recently from the code) it works.

So we are puzzled, oh well.

Martin

Martin Losch
Alfred Wegener Institute 
Postfach 120161, 27515 Bremerhaven, Germany; 
Tel./Fax: ++49(0471)4831-1872/1797



----- Original Message -----
From: Jinlun Zhang <zhang at apl.washington.edu>
Date: Tuesday, July 3, 2007 5:46 pm
Subject: Re: [MITgcm-devel] problem with 59c

> I am not sure what sea ice model is used here. If it is EVP, I 
> don't 
> really know. But if it is LSR, zmin has to be set to zero all the 
> time. 
> And if things blow up, zmin=0 is unlikely the problem. Something else.
> Jinlun
> 
> Martin Losch wrote:
> 
> > I propose the following: introduce a runtime parameter "zetaMin" 
> that  
> > defaults to zero (because I think it's a good idea to have zero  
> > limiting/regularizing by default) that you set to 4e8 in your 
> runs.  
> > zmin>0 is not needed by the EVP solver (according to Elizabeth)
> >
> > If you want to experiment with zmin = 0, why don't you try  
> > seaiceadvscheme = 33 with diff1 = 0, but I don't see why this 
> should  
> > affect the stability, also you try to increase the accuracy of 
> the  
> > solver (although you probably don't want that because of cpu 
> time,  
> > right?)
> >
> > in general, setting zmin = 0 should have an effect on the 
> velocities  
> > and only after that on HEFF. (btw, can't find your pdf).
> >
> > Jinlun, you once recommened setting zmin = 0 (and I agree with 
> your  
> > recommendation). What's your opinion on this problem?
> >
> > Martin
> >
> > On 2 Jul 2007, at 23:53, Dimitris Menemenlis wrote:
> >
> >> Martin, more on the 59c problem:
> >>
> >> The "SEAICEuseDynamics=.false." problem is not "directly" 
> related  to 
> >> the 59c crash.  That is, "SEAICEuseDynamics=.false." crashes in  
> >> Arctic Ocean even with pre-59c code.
> >>
> >> What causes trouble in 59c is the change from ZMIN=4e8 to ZMIN=0.
> >>
> >> Figure http://ecco2.jpl.nasa.gov/data1/cube/cube66/crash.pdf
> >> shows sea ice thickness around Antarctica right before crash.  
> If  
> >> you blow up the pdf file, you will see that there are several  
> >> regions with grid-scale noise in HEFF.  Also shown are time 
> series  
> >> of HEFF at and near region of maximum thickness showing that  
> >> thickness reaches 2500 m before model crashes.
> >>
> >> So what to do?  Go back to ZMIN=4e8?  Change sea ice advection  
> >> scheme or diffusivity?  (I am using default advection scheme and 
> 
> >> diffusivity.) Artificially impose a limit on HEFF?  Something else?
> >>
> >> D.
> >
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>