Author |
Message |
Wijb Sommer (wijb)
Member Username: wijb
Post Number: 4 Registered: 12-2007
| Posted on Friday, January 18, 2008 - 04:12 am: | |
Is it a bad idea to constrain the maximum timestep in a time dependent flow/ dispersion problem, by requesting a monitor plot every dt times? I observe that if the timestep is determined automatically, the timestep grows, but at a certain moment it drops down to E-15 or so. Sometimes it builds up again, else it drops below the halt value. If a monitor is requested say every 1 (second; total runtime is about 170000 seconds) the timestep is more steady and doesn't go to a very small value. |
Robert G. Nelson (rgnelson)
Moderator Username: rgnelson
Post Number: 1047 Registered: 06-2003
| Posted on Friday, January 18, 2008 - 11:32 am: | |
It is sometimes effective to use the Monitor control to limit timestep, but it is better to discover why the automatic control is not working correctly. If you set the ERRLIM control too large, it can allow oscillations to set in that suddenly cause a sharp drop in timestep until the solver recovers. For this reason, it is sometimes the case that small ERRLIM runs faster overall than large ERRLIM. Try reducing the ERRLIM. If there are times at which sources change abruptly, you can put a monitor or plot at the switch time to be sure that FlexPDE does not step over the switch with a large timestep. If there are parameters that are strong functions of the variables, so that the solution appears smooth but parameters suddenly jump, you may be able to control it with small ERRLIM, or you may need to RESOLVE the offending parameter. In some well-behaved problems, you can use SELECT FIXDT to lock the timestep at the value given in the TIME statement. Don't do this if there are sudden changes at widely spaced times, or you will have to run the entire problem at the minimum step requirement.
|
|