| d0578d90 | 06-Mar-2015 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
ADJOINTS+EVENTS: The cost function and the sensitivity matches for the example with events (ex3adj_events) compared with the original adjoint example (ex3adj)
i) Added TSAdjointEventMonitor that si
ADJOINTS+EVENTS: The cost function and the sensitivity matches for the example with events (ex3adj_events) compared with the original adjoint example (ex3adj)
i) Added TSAdjointEventMonitor that simply checks if an event had occured in the forward solve at the current time. If true, calls the user postevent routine with the forwardsolve flag set to FALSE.
ii) Added stepnum field in event recorder
iii) Fixed the time interval of the disturbance in ex3adj.
iv) PostStepFunction in ex3adj_events (also added to ex3adj) displays value of delta after every time-step.
v) PostEventFunction in ex3adj_events now handles both the forward and the adjoint solve.
vi) Have commented out TSPostStep() from TSAdjointSolve as we don't need it in the adjoint solve (I think).
show more ...
|
| 73967516 | 04-Mar-2015 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
ADJOINTS+EVENTS: Revaluate the previous event function residual after a postevent call.
Some states might get changed by the postevent user callback routine. Hence, the event function
residual is re
ADJOINTS+EVENTS: Revaluate the previous event function residual after a postevent call.
Some states might get changed by the postevent user callback routine. Hence, the event function
residual is re-evaluated immediately after a call to postevent and this residual will be the
"previous step event residual" when TS completes the next step.
show more ...
|
| 7a728e9f | 04-May-2014 |
Shri Abhyankar <abhyshr@mcs.anl.gov> |
TSEvent: Avoid re-detection of the same event when previous function value is exactly zero.
If a previous event results in an event function value of exactly zero, then the same event
was being dete
TSEvent: Avoid re-detection of the same event when previous function value is exactly zero.
If a previous event results in an event function value of exactly zero, then the same event
was being detected in the next interval. This was because the check for event detection was
based only on the change of function sign. Added an extra guard that checks if the previous
function value is zero and subsequently avoids re-detecting of the previous event.
Observed this behavior in time-based events (t - t_event_time = 0).
show more ...
|