Odd results with Calculix buckling solution

Fatmac

when changing the number of modes in the analysis settings I get a different value for Mode 1 when i select 1 mode to when I select 3 modes. Using the internal solve produces the result I expected with mode 1 being the same value regardless of the number of modes selected in the analysis settings. Be good to know if there is something I'm doing incorrectly in the calculis settings or if I am misunderstanding the what the "nimber of modes" option does

Victor

You're right that it shouldn't do that. Some ideas:

Are you using the CCX binary that comes with Mecway or another one? The Mecway one has automatic shift point searching and most others don't. It shouldn't make much different here though because buckling factors are slightly greater than 1.

Are you using Pastix with mixed precision? A common source of inaccuracies.

Try reducing the accuracy parameter in the *BUCKLE card. It's a bit awkward since *BUCKLE seems to have to be at the start of the step section so you can put it in custom model definition like:

*STEP
*BUCKLE
1,1e-5

and also omit *STEP and *BUCKLE from being generated.

Does it have anything to do with the C3D8R? I'm a bit leery of that element for some reason.

Could you post the liml file? Even if you solve the problem, I'm still curious what's unique about this model to cause that.

Fatmac

I tried with and with out the C3D8R elements so i could compare with the internal solver and i still get the same affects when I select only 1 mode. I have mixed precsion switched off. I have tried switching solver with the key words but I get the feeling the buckling always uses pastix. Im using version 27 - not tired in older versions-the file with out results is 16MB so may be too big to post here

Fatmac

I have striped down the model i was using. It had contact and 2 beams in it -I simplified it to one beam and a simple load. The difference between mode 1 when 1 mode is asked for and when 3 mode is asked for is now only decimal places so perhaps it was due to over complication in the model. I have attached the striped down file -hope it is not too large

Victor

Thanks @Fatmac. It looks like the accuracy parameter fixes it. I get a reliable 88.88 1st mode with 1e-5 whether requesting 1 mode or 3. This is with the MKL solver and no Pastix but I did see a difference without setting accuracy (89.01 and 88.88).

@Disla has suggested accuracy should be reduced by default and it looks like this is confirmation of that.