[MITgcm-devel] changes to very sensitive evp code

[Martin Losch]

OK, thanks, I'll go ahead with that then,

Martin

On Jan 4, 2013, at 2:05 PM, Jean-Michel Campin <jmc at ocean.mit.edu> wrote:

> Hi Martin,
> 
> In my opinion, you can make these changes and update the output
> (with some doc to explain what did change when you check-in the 
> new output).
> And I can update the adjoint output once the forward one is checked-in.
> 
> Cheers,
> Jean-Michel
> 
> On Fri, Jan 04, 2013 at 10:36:02AM +0100, Martin Losch wrote:
>> Hi Jean-Michel and others,
>> 
>> for completeness, I'd like to implement the modified EVP* of Lemieux et al (2012). In order to be able to use the present (then old) EVP implementation, I have to introduce a few factors, etc. As you know, EVP is very sensitive to everything, so changing
>> SEAICEmassU/SEAICE_deltaTevp
>> to 
>> recip_deltaT = 1. _d 0 / SEAICE_deltaTevp
>> SEAICEmassU*recip_deltaT
>> already breaks the forward testreport for lab_sea.hb87 (only 2 digits of agreement left for the cg2d solver). Anything else that I do, does the same, eg. replacing
>> denom1 = 1./(1.+0.5*dt/tau)
>> with 
>> alph = 2*tau/dt
>> denom1 = alph/(alph+1)
>> has the same effect (in fact the difference between the two expressions is not zero but 1.110223e-16 on my linux computer)
>> 
>> Should I go ahead anyway and update the reference output? Can I ask you to update the adjoint, once I am done? EVP is not tested anywhere else.
>> 
>> Martin
>> 
>> 
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel