different numbers of cores affect the convergence

This forum is for issues related to parallel processing
and OpenSees using the new interpreters OpenSeesSP and OpenSeesMP
Post Reply
alirezacivil
Posts: 23
Joined: Mon Feb 16, 2015 8:39 pm

different numbers of cores affect the convergence

Post by alirezacivil » Sat May 27, 2017 5:53 pm

I am using OpenseesSP to push a huge 3D model to a certain point.

When I change the number of cores or I run it in different computer, it affects the point which the model is diverging.

I wanted to know what is happening and how I can improve the convergence?


It should be noted that results are identical just the point which is not converging changes.

fmk
Site Admin
Posts: 5860
Joined: Fri Jun 11, 2004 2:33 pm
Location: UC Berkeley
Contact:

Re: different numbers of cores affect the convergence

Post by fmk » Fri Jun 02, 2017 11:49 am

the issue is numerical .. at the point of convergence iissue you probably have a matrix that is ill-conditioned .. small changes effect the result of Ax=b computation .. when on different computers and when run in parallel the b and A matrices are not created in exactly the same order (if exact precsion they would be the same, but due to roundoff there will be differences) .. these differences are effecting the reponse in a big way due to conditioing of the matrix.

an option might be to use initial stiffness iterations.

Post Reply