Open lgeretti opened 5 years ago
Found the actual reason for the apparent termination issue: near the final time the set increases rapidly and the evolver tries to improve things by splitting (to no avail), consequently producing new working sets indefinitely. The fact that logging at level 1 fails to show this is indication that we should print both the INITIAL and WORKING set sizes, and not only the initial one (currently called working, by the way, which is quite confusing).
Now I need to understand why the error increases near the final time.
It might be the case that we do not recondition the time model, since the te value is very high. We shall in general join the space and time models, then recondition.
If you take experimental/examples/hybrid/space_rendezvous from the working-arch branch, there is a termination issue occurring if the evolution time is low enough (say 113 seconds instead of the nominal 200 seconds). In order for the problem to appear, the system must have performed the first discrete step though.
Since the set shrinks over time, and since larger sets cause the termination issue to appear also at 200 seconds, it seems to be somehow related to the set size.