[MITgcm-support] Smaller deltaTmom speeds up the model?!

[Martin Losch]

I think you can safely use 1e-7 (or 1e-8). 1e-13 is probably required for adjoint simulations, where we assume that the solver is exactly self-adjoint.

Martin

> On 20. Jun 2018, at 15:49, Naughten, Kaitlin A. <kaight at bas.ac.uk> wrote:
> 
> Hi Martin,
> 
> I have cg2dTargetResidual=1e-13. I see the default is 1e-7 so my value is pretty strict isn't it? I will try something like 1e-8 as a middle ground.
> 
> Many thanks,
> Kaitlin
> From: MITgcm-support <mitgcm-support-bounces at mitgcm.org> on behalf of Martin Losch <Martin.Losch at awi.de>
> Sent: 20 June 2018 13:26:35
> To: MITgcm Support
> Subject: Re: [MITgcm-support] Smaller deltaTmom speeds up the model?!
>  
> Hi Kaitlin,
> 
> what is your value for cg2dTargetResidual?
> If you think that you spent too much time in cg2d (or SOLVE_FOR_PRESSURE), you could reduce cg2dTargetResidual to maybe 1.e-7. I have seen even smaller values in some of the expensive high-res configuration, e.g. here <http://wwwcvs.mitgcm.org/viewvc/MITgcm/MITgcm_contrib/llc_hires/llc_4320/input/data?revision=1.14&view=markup>
> but you’ll sacrifice accuracy for speed.
> 
> Martin
> 
> 
> > On 20. Jun 2018, at 14:09, Naughten, Kaitlin A. <kaight at bas.ac.uk> wrote:
> > 
> > Hi Martin,
> > 
> > Thanks, as always, for your prompt and helpful support here. Indeed, the big difference between the two simulations is in the SOLVE_FOR_PRESSURE timer, which I understand is the elliptic solver. The simulation with deltaTmom=60 spends much less time than deltaTmom=600 in this section, by a factor of 6. And yes, deltaTfreeSurf=deltaTmom by default.
> > 
> > All the best,
> > Kaitlin
> > From: MITgcm-support <mitgcm-support-bounces at mitgcm.org> on behalf of Martin Losch <Martin.Losch at awi.de>
> > Sent: 19 June 2018 14:33:14
> > To: MITgcm Support
> > Subject: Re: [MITgcm-support] Smaller deltaTmom speeds up the model?!
> >  
> > Hi Kaitlin,
> > 
> > see answers below
> > 
> > > On 19. Jun 2018, at 15:12, Naughten, Kaitlin A. <kaight at bas.ac.uk> wrote:
> > > 
> > > Hi everyone,
> > > 
> > > I am experimenting with timesteps and have discovered something strange. I have deltaT=600, and for the first few months of simulation I need to have deltaTmom=60 so that some areas with an unstable water column in the initial conditions don't cause the model to blow up. After the first few months, I can remove deltaTmom=60 (so it defaults back to deltaT=600) and the model is stable.
> > > 
> > > But when I do this, the model slows down! To confirm this was the case, I ran two simulations starting from the same timestep (1 year in). The simulation with deltaTmom=60 completed 22464 timesteps in 20 minutes. The simulation with deltaTmom=deltaT=600 only completed 8496 timesteps. 
> > > 
> > > I have two questions about this:
> > > 
> > > 1) Is this behaviour as counterintuitive as it seems? I'm wondering if larger momentum timesteps are somehow causing the elliptic solver to converge more slowly, which more than compensates for the reduced number of timesteps?
> > you can see what time is spent where by inspecting the profiling at the end of STDOUT.0000. If the elliptic solver really uses all the time in your case with deltaTmom=deltaT, then that should be obvious from the profiling numbers.
> > I think, deltaTfreeSurf is also set to deltaTmom (please check in STDOUT.0000), so that your elliptic solver now sees a 10 times shorter timesteps, i.e. the tendencies are 10 times smaller and the route to convergence is shorter, so I wouldn’t be surprised, if that’s really what’s happening.
> > > 
> > > 2) Since a smaller deltaTmom is so much faster, is there anything wrong with having a different value for deltaTmom and deltaT? Some of my colleagues thought that having different timesteps for ocean tracers and momentum might be dangerously unphysical.
> > You’ll damp all of your fast waves and probably suppress the seasonal cycle. But it also modifies your water masses and stratification. See relevant papers about tracer acceleration (e.g.,
> > Danabasoglu, G., 2004: A comparison of global ocean general circulation model solutions obtained with synchronous and accelerated integration methods. Ocean Modelling, 7, 323- 341.
> > Bryan, K. 1984: Accelerating the convergence to equilibrium of ocean-climate models. Journal of Physical Oceanography, 14(4), 666-67
> > ) for a discussion.
> > 
> > Martin
> > 
> > > 
> > > Many thanks,
> > > Kaitlin Naughten
> > > This message (and any attachments) is for the recipient only. NERC is subject to the Freedom of Information Act 2000 and the contents of this email and any reply you make may be disclosed by NERC unless it is exempt from release under the Act. Any material supplied to NERC may be stored in an electronic records management system.
> > > _______________________________________________
> > > MITgcm-support mailing list
> > > MITgcm-support at mitgcm.org
> > > http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support
> > 
> > _______________________________________________
> > MITgcm-support mailing list
> > MITgcm-support at mitgcm.org
> > http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support
> > This message (and any attachments) is for the recipient only. NERC is subject to the Freedom of Information Act 2000 and the contents of this email and any reply you make may be disclosed by NERC unless it is exempt from release under the Act. Any material supplied to NERC may be stored in an electronic records management system.
> > _______________________________________________
> > MITgcm-support mailing list
> > MITgcm-support at mitgcm.org
> > http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support
> This message (and any attachments) is for the recipient only. NERC is subject to the Freedom of Information Act 2000 and the contents of this email and any reply you make may be disclosed by NERC unless it is exempt from release under the Act. Any material supplied to NERC may be stored in an electronic records management system.
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support