Why is the PCG solver giving an incorrect reaction for a remote force load in ANSYS 10.0?
|
A new preconditioner was introduced at 10.0. It is accessed with the default PCGOPT,2. When large CP/CE (i.e., number of terms greater than 10,000) are present there was logic in this new preconditioner to skip those CE's. This was done because of a slow operation in forming the new preconditioner when the number of CP/CE terms is large. Previously we had only seen these large CP/CE with contact elements and thus the slave node was always the target/pilot node which has no stiffness. When the slave node has no stiffness we can skip that CE without affecting the new preconditioner. However, for this model there is a node attached to a beam element as the slave node which did have stiffness associated with it. Thus, the new preconditioner was bad and that is why the PCG solver had erratic convergence when PCGOPT,2 was used. The new PCG preconditioner has been fixed to detect whether the slave node of these large CP/CE has stiffness or not (Ref: Defect 41262.) If it does have stiffness and it is part of a very large CP/CE then we no longer skip that CP/CE when building the preconditioner. Because this can cause a longer runtime, we now warn the user that the current Level of Difficulty might be inefficient and that switching to PCGOPT,1 may make the solution more efficient. If the slave node of the large CP/CE does not have any stiffness then we skip the CP/CE as before. Your model gives the correct reaction force at 11.0. |
||
|