Opened 12 years ago

Closed 11 years ago

#110 closed help (fixed)

1km simulation failing to converge

Reported by: gdevine Owned by: lois
Component: UM Model Keywords:
Cc: Platform:
UM Version:

Description

Hi,

I have just run a 1km resolution simulation for a region of Africa. Unfortunately the model produces 'junk'

output from the very first timestep and in the .leave file shows this….

Atm_Step: Timestep 1
==============================================
initial Absolute Norm : 3.63954480017502346
GCR( 2 ) failed to converge in 200 iterations.
Final Absolute Norm : 0.687325985659843416E-02
==============================================

I have had similar problems before at 12km but was able to get around it by (1) halving the timestep and (2) upping the number of iterations in the GCR. I have tried this with the 1km but it doesn't appear to help (timestep currently at 2.5 minutes). Do you have any suggestions as to what may be wrong here? The LBCs from the 4km run, ancillary files, and the generated start dump (.astart) all look okay. Is it possible that some of my ancillary files are not compatible with each other. Currently most of the ancillaries are provided by the start dump, except for orography/ozone/land-sea mask. The run ID is xcrfg.

Any help much appreciated.
Gerry

Change History (5)

comment:1 Changed 12 years ago by gdevine

Hi,

I ran this again, and this time I cut the timestep to ~1minute - seems to help. The results seem more sensible but I do notice some issues with the boundaries. I will get back to you when I look into this a bit more.

Thanks
Gerry

comment:2 Changed 12 years ago by gdevine

Hi again,

I have looked at the output from the most recent run (xcrff) I have done (described above). The timestep I have used was 90 seconds. This run seems to work for the first 6 hours or so (with some issues at the boundaries evident when viewing in XCONV), at which point the 'GCR failing to converge' problem occurs (with 'junk being produced from that point onwards). Do you have any suggestions where to go from here? I am assuming this timestep is short enough? What else could I look out for in the .leave file, that might help me pinpoint the problem? Also, I have been using the same set of mods for this run that I have been using for 12km and 4km runs. Is this correct or are there any specific mods when running at 1km? The run I.D is xcrff.

Apologies for the barrage of questions.
Gerry

comment:3 Changed 12 years ago by lois

  • Owner changed from um_support to lois
  • Status changed from new to assigned

Two things you may want to look at Gerry.

In Chang's 1km over UK sample job xceqf she is using a time step of 30 seconds. So you may want to consider going as low as this. She also include some extra mods for 1km runs which are available under umx on HPCx.

So could you consider your case withe a lower timestep and then looking at what these mods might do for your particular case.

If you are still having problems then we will take a closer look.

Lois

comment:4 Changed 12 years ago by gdevine

Hi Lois,

Thanks for getting back to me so quickly. I will give a go with the shorter timestep and extra mods. I will let you know how I get on.

Gerry

comment:5 Changed 11 years ago by ros

  • Resolution set to fixed
  • Status changed from assigned to closed

No further comments have been made to this query so I'm closing this query. I assume that the suggestion of using a lower timestep fixed this problem.

Note: See TracTickets for help on using tickets.