Author |
Message |
Patricio A. Greco (pagreco)
Member Username: pagreco
Post Number: 18 Registered: 08-2003
| Posted on Tuesday, February 17, 2004 - 09:21 pm: | |
Mr.Robert Nelson: The attached script works fine with the version 3 but I tried to run V.4 and causes a program crash.This is a coaxial to waveguide transition , the waveguide have a dummy load at the end. The file info.txt have the debug information provided for the operating system. Will you help me to find the error? Let me tell you that the results I found with this script was widely verified.
|
Robert G. Nelson (rgnelson)
Moderator Username: rgnelson
Post Number: 111 Registered: 06-2003
| Posted on Wednesday, February 18, 2004 - 03:58 pm: | |
Greco: I will investigate this problem and try to determine a solution. However, already I see a problem: You have nested cylinders with a shell thickness 1/10000 the size of the domain. In Version 3 the treatment of the z-dimension was completely separate from the treatment of X,Y. The mesh generator created a mesh in X,Y and then extruded this a single cell high across your z-extent. This creates needle-shaped cells 550 times longer in z than their X,Y dimension. Version 4 has tried to establish an equivalence in the treatment of all coordinate dimensions, and eliminate the disparate treatment of Z. As a result, version 4 wants to create a better balanced mesh in this thin cylindrical shell. This implies an enormous number of mesh cells, because the same density propagates a ways into the surrounding material as well. (There is also some kind of a flaw in the version 4.0.1 treatment of this tiny shell, but that will be rectified soon). There is no provision in version 4 to revert to the mesh model of version 3. The only way to recover this kind of needle-cell mesh would be to scale your z-dimension and material parameters so that FlexPDE 4 could be presented a problem in which the z-dimension is comparable to the shell thickness (within perhaps an order of magnitude). Another question that should be addressed is whether this thin shell actually contributes anything to the solution, or whether the cylinders could be merely merged to eliminate the problem. |
Patricio A. Greco (pagreco)
Member Username: pagreco
Post Number: 19 Registered: 08-2003
| Posted on Thursday, February 19, 2004 - 03:47 pm: | |
Robert: I cannot replace the small cilinders because the objetive of these cilinders is to increase the node density in order to define the current jz correctly. I tried with one cilinder and a feature but the program also fails. The error messages are too much less if I define a GRIDARC angle less than arcos(1-(r/e)) where r is the minor radio of the cilinder and e the trickness of the shell. All locks like the cells limits touch the other face of the shell causing errors in the mesh generation. I hope you understand my english , thank you. |
Robert G. Nelson (rgnelson)
Moderator Username: rgnelson
Post Number: 113 Registered: 06-2003
| Posted on Thursday, February 19, 2004 - 04:38 pm: | |
Greco: We are currently checking through the implementation of 3D mesh density controls, and find there are errors. We should be able to better handle the control of density on your cylinder in the 4.0.2 release next week. |
Patricio A. Greco (pagreco)
Member Username: pagreco
Post Number: 21 Registered: 08-2003
| Posted on Monday, March 08, 2004 - 09:20 am: | |
ERRATA: in my last post replace arcos(1-(r/e)) by arcos(1-(e/r)) |