My free surface calculation runs fine in serial, but when I run it in parallel it crashes on the first timestep.


This is usually due to a partitioning problem. You should not have a parallel partition boundary parallel and close to the free surface itself in the domain. Also, if the free surface moves it should not be able to touch a partition boundary thats parallel to it. The reason is that this will cause numerical difficulties to do with the treatment of the interface and the nodes around it. With the default MeTiS k-way partitioning you really dont have much idea where the partition boundaries will go, and simply choosing a different number of partitions (3 instead of 4, or 7 instead of 8) might cure it. A better method might be to try a different partitioning method that is more geometrical so you can specify the partition boundaries to be perpendicular to the free surface (eg. RCB or specified direction).
Incidentally, you should also try to avoid placing domain interfaces parallel and close to the free surface for similar reasons.





Show Form
No comments yet. Be the first to add a comment!