'nfast' and 'ndtfast' on Barotropic time step

Report or discuss software problems and other woes

Moderators: arango, robertson

Post Reply
Message
Author
yj7054
Posts: 9
Joined: Mon Sep 24, 2018 7:40 pm
Location: CSIRO - Hobart Site

'nfast' and 'ndtfast' on Barotropic time step

#1 Post by yj7054 » Fri Nov 30, 2018 2:53 am

Dear all,

nfast: Number of barotropic timesteps needed to compute time-averaged barotropic variables centered at time level n+1
ndtfast: Number of barotropic timesteps between each baroclinic timestep.

I am confused with the time-stepping in ROMS.

For example, I set dt = 150s, ndtfast = 30 in ocean.in, so I got dtfast = 5s, and I also got nfast = 42 computed by set_weights.F. In main3d.F, there are 42 barotropic time-steps for each dt, rather than 30 steps. SO my question is that are there 12 time steps(M*-M) double counted for each dt? For the next baroclinic step n+2, the barotropic step starts from m=M or from m=M* on the figure attached?

Could anyone give me a further explain? Thank you very much!
Attachments
The split time stepping used in the model.
The split time stepping used in the model.
The split time stepping used in the model.png (15.72 KiB) Viewed 881 times

User avatar
wilkin
Posts: 506
Joined: Mon Apr 28, 2003 5:44 pm
Location: Rutgers University
Contact:

Re: 'nfast' and 'ndtfast' on Barotropic time step

#2 Post by wilkin » Fri Nov 30, 2018 11:32 am

Read this explanation on WikiROMS:
https://www.myroms.org/wiki/Numerical_S ... g_Overview
and the Shchepetkin papers.
John Wilkin: DMCS Rutgers University
71 Dudley Rd, New Brunswick, NJ 08901-8521, USA. ph: 609-630-0559 jwilkin@rutgers.edu

yj7054
Posts: 9
Joined: Mon Sep 24, 2018 7:40 pm
Location: CSIRO - Hobart Site

Re: 'nfast' and 'ndtfast' on Barotropic time step

#3 Post by yj7054 » Sun Dec 02, 2018 5:46 am

wilkin wrote:Read this explanation on WikiROMS:
https://www.myroms.org/wiki/Numerical_S ... g_Overview
and the Shchepetkin papers.
Hi John,

Thank you for your reply.

I read the explanation before but it was complex for me. Now I think I got the main idea roughly:

There are M* barotropic steps for each baroclinic step, and the fast-time averages among these M* steps are transmitted to the baroclinic mode, as well as used as the initial values for barotropic mode during the next baroclinic step.

User avatar
arango
Site Admin
Posts: 1081
Joined: Wed Feb 26, 2003 4:41 pm
Location: IMCS, Rutgers University
Contact:

Re: 'nfast' and 'ndtfast' on Barotropic time step

#4 Post by arango » Sun Dec 02, 2018 4:19 pm

Nope, there are actually M* barotropic timesteps :!: The reason for it is that the cosine-square shape filter needs to provide time-averaged values of ubar, vbar, and zeta centered at the n+1 baroclinic timestep, which gives us second-order temporal accuracy for time-averaged barotropic motions. You may check routine set_weights.F to see how the time weights are computed. The above diagram is not clear enough.

yj7054
Posts: 9
Joined: Mon Sep 24, 2018 7:40 pm
Location: CSIRO - Hobart Site

Re: 'nfast' and 'ndtfast' on Barotropic time step

#5 Post by yj7054 » Sun Dec 02, 2018 10:26 pm

arango wrote:Nope, there are actually M* barotropic timesteps :!: The reason for it is that the cosine-square shape filter needs to provide time-averaged values of ubar, vbar, and zeta centered at the n+1 baroclinic timestep, which gives us second-order temporal accuracy for time-averaged barotropic motions.
Hi Hernan,

Thank you very much!

So the time-averaged values of ubar, vbar, and zeta you mentioned above will be used as initial value for the barotropic modes from m=M to m=M+M* during next baroclinic step n+2, right?

If I am wrong, where does the initial value come from during the next baroclinic step n+2?

User avatar
arango
Site Admin
Posts: 1081
Joined: Wed Feb 26, 2003 4:41 pm
Location: IMCS, Rutgers University
Contact:

Re: 'nfast' and 'ndtfast' on Barotropic time step

#6 Post by arango » Mon Dec 03, 2018 2:14 am

Again, Nope :!: The governing equations are timestep from n (right-hand-side terms) to n+1 (left-hand-side term: time rate of change). The vertically-integrated equations are integrated with smaller timestep to resolve fast dynamics due to gravity wave phenomena. For 3D total momentum coupling, the ubar, vbar, and zeta need to be time-averaged at n+1. So we need to timestep the vertically integrated equations beyond the time associated with n+1 so the time-averaged quantities are centered exactly at n+1. This is achieved with the cosine-squared filter so we don't need to go all the way to n+2 to achive such average. It is cleaver and efficient :!: Maybe you need read the literature of split-explicit timestepping or find someone at your intitution to explain it to you.

Post Reply