Converging differences

[benbou]

Hi,
I was wondering if anyone encountered this problem too:
Why would an analysis converge quite fast on a computer, but would never converge on another one. I'm of course talking about an indentical model of a quite complex 3D RC building. The computers which don't converge are older (P4 32-bit).
Could it be caused by some kind of processor precision??

Tks

Benoit

[Romain]

Hi Benoit,

The question that you made is quite interesting.
In fact, depending on you processor precision, you can find some difficulties in the convergence or even small differences on the results.
This fact is more evident when you define properties with very high values. For instance , elements with very high stiffness, etc.

Cheers,
Romain

[seismosoft]

Hi Benoit,

Romain is indeed correct in his comments, in the sense that in models sensitive to numerical precision (due to very large stiffness values in some elements, or very small analysis steps, or very tight convergence criteria, etc), one may indeed obtain different results in computers that have different processors, or different Operative Systems, etc. In such cases, you can decrease those large stiffness values (if they are not strictly needed), or increase analysis step, etc.

One other possible explanation for the behaviour you have observed may be that the SeismoStruct Settings used in one of the computers are different from those in the other (check them carefully) .

Seismostruct Support

[benbou]

Thanks you for the answer.
I'm glad to know that I am not the only one with this kind of behavior.
Is it possible that the very large stiffness elements that may cause problems are the rigid diaphragm or another type of constraints??

Thanks
Benoit Boulanger

[seismosoft]

Yes, it is possible, especially if 'Penalty Functions' are employed to model such constraints.

Seismosoft Support