Floating-point exception - erroneous arithmetic operation when running with Thompson microphysics and DEBUG=true

[Peiyang]

Hello all,

Recently, I was testing the capabilty of MPAS-A model in simulating tropical cyclones. However, when I compiled the model with 'make gfortran CORE=atmosphere PRECISION=single AUTOCLEAN=true DEBUG=true', and ran the model with Thompson microphysics (with the 'convection_permitting' physics suite), I always encountered the "Floating-point exception - erroneous arithmetic operation" error at the first time step.

Here are some details:

• I compiled the code with gcc v9.5.0, mpich 3.3.2, hdf5 1.10.11, netcdf v4.7.4, netcdf-fortran v4.5.4 + p-netcdf v1.10.0, hdf5 v1.10.11, pio v2.5.4.
• The pbs log file points the error to file 'src/core_atmosphere/physics/physics_wrf/module_mp_thompson.F' line 1860. The model would write history and diag files at the initial timestep but fail afterwards. I did some tests and realized the problem has something to do with 'mvd_r(k).gt.100.E-6' in the if statement.
• I tested the (a) the 'mesoscale_reference' suite and (b) the 'convection_permitting' suite overridden with 'WSM6' microphysics, and both worked.
• I tested both 60-3 km and 60-15km grids, and both showed the same behavior.
• Finally, I tested Thompson microphysics (with the 'convection_permitting' physics suite) with DEBUG option been turn off, the model ran smoothly.
• I'm attaching the namelist and log files associated with the failed run for your reference.

This problem is quite confusing to me. I cannot tell if the results given by Thompson microphysics and DEBUG=true are reasonable as of now. Could it just be compiler issues or something more important. Any thoughts on this would be helpful. Thank you!

 

[mgduda]

Thanks very much for the detailed description of the issue! I think you're correct that the floating-point exception is related to the 'mvd_r' variable, and looking through the code, it seems likely that at some levels this variable is being used where its value has never been set. As a quick work-around (note: I haven't tried to validate the correctness of this fix!), I think you could add some code like the following

      do k = kts, kte
         mvd_r(k) = 0.
      end do

around line 1597 of module_mp_thompson.F.

We'll work on getting a fix in place for this in an upcoming bugfix release. Thanks again for pointing this out!

 

[Peiyang]

Thanks very much for the detailed description of the issue! I think you're correct that the floating-point exception is related to the 'mvd_r' variable, and looking through the code, it seems likely that at some levels this variable is being used where its value has never been set. As a quick work-around (note: I haven't tried to validate the correctness of this fix!), I think you could add some code like the following

      do k = kts, kte
         mvd_r(k) = 0.
      end do

around line 1597 of module_mp_thompson.F.

We'll work on getting a fix in place for this in an upcoming bugfix release. Thanks again for pointing this out!

Thank you for the quick work-around. Good to know it is just a initialization issue. And the fix does the trick!

BTW, I also encountered the issue given in this thread. I noticed that static.nc for the 60-3 km grid can be generated when compiling init_atmosphere with DEBUG=false, but not when DEBUG=true. I'm currently using your fix given in the thread, but I would like to let you know about the similar DEBUG thing.