[MITgcm-support] Nonlinear free surface and the adjoint

[Christopher L. P. Wolfe]

Disabling the sigma code removed a lot of the recomputions; the ones remaining look pretty minor. Thanks to Gael and Patrick.

On May 29, 2014, at 3:44 PM, gael forget <gforget at MIT.EDU> wrote:

> Hi Christopher,
>> I’m trying to do sensitivity studies using a nonlinear free surface. TAF gives warnings about extensive recomputations that disappear when I turn off the NLFS using #undef NONLIN_FRSURF in CPP_OPTIONS.
> Activating NONLIN_FRSURF increases tape storage greatly but 
> should not introduce extensive recomputations anymore.
>> I’ve heard that ECCOv4 uses the NLFS in both forward and adjoint mode,
> NLFS is indeed used in forward and adjoint for ECCO v4. 
>> so either the recomputations are not that bad or there’s a way to make them go away. Does anyone know which is the case?
> One thing to check is that DISABLE_SIGMA_CODE is defined.
> If that is not the issue, I am not sure where the recomputation you get may come from.
>> If neither, is there a flag I could set to use the NLFS for the forward integration but have it turned off in the adjoint?
> Cheers,
> Gael
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support