[MITgcm-support] Weird numerical noise leads to the model blowing up

[Fancer Lancer]

If anyone ever will face the same problem, it seems to me, that the reason
of it is the same as it was claimed in the topic discussed here a few
months ago: "Linear internal waves evolution under a free surface and a
rigid lid."
http://mitgcm.org/pipermail/mitgcm-support/2014-January/008764.html

That time Jean-Michel suggested to use the option cg2dUseMinResSol=1 in
order to force the model to utilize the minimum residual solution of 2d
problem. It helped for the most of my numerical runs. But it seems like the
model anyway didn't converged at some point of simulation (see the output):
 cg2d: Sum(rhs),rhsMax =   4.59031695927215E-03  9.54828652355926E-08
 cg3d: Sum(rhs),rhsMax =   4.85750298973343E-01  1.43154617420132E-14
 cg2d: Sum(rhs),rhsMax =   4.69887235641187E-03  9.54041157541186E-08
* cg3d: Sum(rhs),rhsMax =   5.76048634182131E-02  1.23467293119958E-13*
 cg2d: Sum(rhs),rhsMax =   4.56515285889347E-03  9.55727017601895E-08
 cg3d: Sum(rhs),rhsMax =   2.03099361693707E-01  3.40824611669796E-14

That led to the model blowing up afterwards.
As far as I understand the root cause of the problem lays in the problem
setup, in particular I specified the jumping density profile as it was
between two fluids with different densities, in addition I made extremely
high resolution on the transition layer between the fluids.
To sum up I suppose, that the only way to fix the sudden model blowing up
would be to change the initial condition setup.

If someone has any other suggestions welcome to the discussion.

Sincerely,
-Sergey



On Wed, Apr 9, 2014 at 6:12 PM, Fancer Lancer <fancer.lancer at gmail.com>wrote:

> Hi Michael,
> No, the model worked in the non-hydrostatic mode.
> In addition there is the rigid lid on the surface.
>
> Sincerely,
> -Sergey
>
>
> On Wed, Apr 9, 2014 at 5:58 PM, michael schaferkotter <
> schaferk at bellsouth.net> wrote:
>
>> were your experiments done in hydrostatic mode?
>>
>>
>> On Apr 9, 2014, at 7:13 AM, Fancer Lancer wrote:
>>
>> > Good day MITgcmers,
>> >
>> > I use the linear potential theory to specify a small-amplitude internal
>> wave in 2D case. I performed a lot of model runs for waves with different
>> wavelengths for the various pycnocline depths. To cut it short most of them
>> work pretty fine: initial wave propagates as it supposed to be predicted.
>> But there are some experiments, which suddenly blow up with weird noise
>> around the pycnocline (see the attached gif animated figure).
>> >
>> > Does someone know what is the possible reason of that strange behavior?
>> >
>> > It should be noted a few facts about the model setup:
>> > 1) Since the linear theory is used to setup the velocity field the
>> transition layer (pycnocline) between two water layers is supposed to be as
>> thin as it possible. So ideally there shouldn't be the transition layer at
>> all, but should be just the density jump. All the experiments work pretty
>> fine even with the abrupt density jump (no pycnlocline), but some of them
>> suddenly blow up (see the gif-figure).
>> > 2) The linear theory implies that fluid interface perturbations should
>> be infinitely small that's why I use the variable vertical resolution,
>> which is huge around the fluid interface. But it should be noted, that I
>> specify the corresponding time-step to satisfy the CFL condition and the
>> most of the model runs work pretty well.
>> >
>> > Any suggestion concerning the problem would be appreciated.
>> > Many thanks in advance.
>> > Sincerely,
>> > -Sergey
>> >
>> > <Test3.gif>_______________________________________________
>> > 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
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20140417/be999d1c/attachment.htm>