Scheduled Downtime
On Friday 21 April 2023 @ 5pm MT, this website will be down for maintenance and expected to return online the morning of 24 April 2023 at the latest

"restart" run and "continuous" run have different results

Gaoyang

New member
Hello everyone:

I'm using MPAS to run a simulateion from 2011-01-08 to 2011-01-18, and generate some restart files like restart.2011-01-09_00.00.00.nc. Then I restart the simulation at 2011-01-09. But the "restart" run and the "continuous" run give different results. The figure below shows the difference between Z500 from the "restart" run and Z500 from the "continuous" run at 2011-01-10. This is very confusing for me, I think they should give identical results.

1702611587200.png

I attach the namelist.atmoephere files of "restart" run and the "continuous" run for reference. Looking forward to your help. Thank you.
 

Attachments

  • restart-namelist.atmosphere.txt
    1.7 KB · Views: 3
  • ctrl-namelist.atmosphere.txt
    1.7 KB · Views: 4
I have run a test case and I can confirm that the results from continuous run and initial run are not bit-4-bit identical.

We will look into this issue and get back to you once we fix the problem. It may take some time. Thanks for your patience.
 
Further investigation of the issue indicates that the difference between results of continuous run and restart run are very small, which might be attributed to the optimization approach when compiling the code.

I don't think there is anything wrong in the 'restart' capability of MPAS.

To obtain bit-4-bit identical results, please recompile MPAS in debug mode (i.e., DEBUG=true)
 
Further investigation of the issue indicates that the difference between results of continuous run and restart run are very small, which might be attributed to the optimization approach when compiling the code.

I don't think there is anything wrong in the 'restart' capability of MPAS.

To obtain bit-4-bit identical results, please recompile MPAS in debug mode (i.e., DEBUG=true)
Thank you for your reply. It is very helpful to me! I will try to recompile in debug mode, hoping it can give bit-4-bit identical results.
 
Please see the attached plot that shows the difference in "height_500hPa" between initial and restart run.
The values are mostly less than 0.05. This seems more like rounding errors.
I run the test using the MPAS 120-km mesh. Note that I read data on MPAS mesh and didn't do any interpolation.
For your plot, did you interpolate MPAS output to lat-lon grids?
 

Attachments

  • test.png
    test.png
    281.9 KB · Views: 2
Please see the attached plot that shows the difference in "height_500hPa" between initial and restart run.
The values are mostly less than 0.05. This seems more like rounding errors.
I run the test using the MPAS 120-km mesh. Note that I read data on MPAS mesh and didn't do any interpolation.
For your plot, did you interpolate MPAS output to lat-lon grids?
Yes, I interpolate the output to lat-lon grids when I plot the figure.
And I'm using 60-km mesh and SST/sea-ice update file.

I also check the data on MPAS mesh from diag.2011-01-10_00.00.00.nc. The difference in "height_500hPa" between initial and restart run can reach more than 1.0 at around 4% cells.

I upload the output files of initial and restart run and corresponding namelist files to your Nextcloud storage for reference (filename: restart_vs_init.tar). Did I do something wrong when restarting the simulation?
 
I found that the difference could be large after the data are remapped to lat-lon grids. However, if you look at them in the native MPAS mesh, the different is very small.
 
Top