ROMS_AGRIF periodic domains and mpi -- user error or bug?

[madelinem]

Hi,

I am using ROMS_AGRIF version 3.0 in an EW_PERIODIC configuration and with MPI.

I find that when I have EW_PERIODIC and NP_XI > 1, at the first time step the w-velocities at the eastern boundary (the full vertical slice) are set equal to negative values of the initial temperature field. In my configuration, I am using a linear EOS with alpha = -1 and beta = 0, so this may be related to density. When I instead tile in the y-direction, NP_ETA >1, this problem does not arise.

I then checked to see what would happen if I set NS_PERIODIC (with undef EW_PERIODIC). Similarly, when NP_ETA > 1, the northern boundary w-velocities are set to the negative temperature, but this time tiling in the x-direction (NP_XI>1) has no effect.

I also striped the temperature field in my initial conditions (set each temperature equal to its i-index value) just to check that I understood which negative temperatures were going into the vertical velocity field. In the EW_PERIODIC case, it is picking up temperatures from i=2, no matter how many processors I use. In the NS_PERIODIC case, I get temperatures corresponding to the i-index.

Does anyone have any ideas where this problem could be coming from? It seems it must be related to the tile indexing somehow. I can provide more info and files if that would help. Any suggestions would be very much appreciated!

Madeline

[patrickm]

See answer and bug fix report in ROMS_AGRIF forum:
https://gforge.inria.fr/forum/?group_id=735

Patrick

[madelinem]

Thanks so much, Patrick!