Accuracy of OpenSeesMP results

This forum is for issues related to parallel processing
and OpenSees using the new interpreters OpenSeesSP and OpenSeesMP

Moderator: selimgunay

Post Reply
aaataha
Posts: 15
Joined: Sun Oct 13, 2019 1:44 pm

Accuracy of OpenSeesMP results

Post by aaataha » Fri Oct 15, 2021 6:58 pm

Hello,

I am using openseesMP for modeling a large soil medium (partitioned) and noticed a difference in the results compared to sequential analysis. What could be the reason for this deviation of openseemMP results?

fmk
Site Admin
Posts: 5883
Joined: Fri Jun 11, 2004 2:33 pm
Location: UC Berkeley
Contact:

Re: Accuracy of OpenSeesMP results

Post by fmk » Thu Nov 04, 2021 9:42 pm

The programs will not give the same results. Typically the difference is small, a difference in some insignificant digits. Hoewever if matrix is poorly conditioned at stages in a nonlinear analysis the results can be larger at the end. Check on differences after first few steps, if very different your model the model you built with SP os different from MP.

Reasons for differences if same model:
1) different solver, i.e. Mumps in MP, something else in sequential. Here the difference is due to the LU decomposition being different.
2) even with same solver there can be difference. In parallel, the partitioned matrices accumulate the matrix differently (in that the element contributions are added in a different order to the matrix). If exact math were performed there would be no difference in the results, however there is roundoff and that causes differences in the matrix entries before the LU decomposition starts.

Possible differences if different model:
1) transient problem and mass added to node in different partitions twice.
2) ditto with nodal loads

Post Reply