Author |
Message |
Zhang, Lulu (lulu)
New member Username: lulu
Post Number: 2 Registered: 10-2003
| Posted on Tuesday, April 13, 2004 - 10:33 pm: | |
There are a few errors in flexpde such as: RMS error, Max error. The user guide says that "FlexPDE applies a consistancy check to integrals of the PDE's over the mesh cells. From this it estimates the relative uncertainty in the solution variables and compares this to an accuracy tolerance". Would some detailed information be provided on this aspect? How the RMS error and Max error are estimated? I also found that when the time step is split automatically in order to let the error be less than errorlim, sometimes the timestep could be very small (say less than 1e-15), then a error information box pumps out saying "Error: Floating-Point Divided by Zero". And in my problem, I let the regrid be off because the regridding process caused erratic results when I transfer the output into another file. Is there any way that I can handle this "Floating-Point Divided by Zero" problem? |
Robert G. Nelson (rgnelson)
Moderator Username: rgnelson
Post Number: 135 Registered: 06-2003
| Posted on Wednesday, April 14, 2004 - 08:35 pm: | |
I have written a Technical Note for inclusion in future releases of the Help system which discusses the subject of FlexPDE error estimates. The note is attached to this posting. In regard to the timestep catastrophe, when FlexPDE starts a time-dependent problem, it first tries to find a valid starting timestep by comparing one-step and two-step solutions over a trial timestep, and cutting this trial step size until the one-step and two-step solutions agree within the ERRLIM tolerance. This process assumes that there will be some timestep value at which the evolution of the solution will be smooth. If you have included a discontinuous jump in problem conditions at the start time, there may be no timestep, no matter how small, at which the solution is continuous in time. In this case, you may see a catastrophic cutting of the timestep, until the computed values fall below machine roundoff, and some numeric error ensues. You should always pose your problem with continuous conditions across the start time. |
|