BFGS Algorithm with VariableTransit Analysis

[amatamor]

I am analyzing a frame using the BFGS algorithm and variable transit analysis. Regardless of the increment size I specify or the maximum number of iterations opensees always reports failure to converge after 10 iterations, and always reports the results in 0.01 second increments. This behavior different from other algorithms like KryloNewton.

Is that a limitation of the code or is there an option I am missing?

[skamalzare]

If your model fails with other algorithms too, it is most probably because you have not defined the model correctly. There might be a problem in your boundary conditions. Can you converge in the gravity analysis?

Regards,
Soheil

[amatamor]

The model runs fine, both for Gravity and earthquake loading. With a variable transient analysis I think that the algorithm cuts back after not reaching convergence and tries again until it converges. No problem there.

The problem is that the recorder stores the results at a fixed time increment, and the cut back always occurs after 10 iterations so I am concerned that there is a bug in the source code. Other algorithms do not behave this way.

If I solve the same problem with a KrylovNewton algorithm for a large intensity measure I get a different answer. If I try modified Newton my model might take several weeks for a single run. The BFGS algorithm takes 8 min to run.

[fmk]

I am unclear as to what you mean by cutback? Do you have the -time option in the recorder command and can you show me what you mean by cutback in the recorded time data.

the VariableTransientAnalysis only cuts or increases the time step from one analysis to another based on the number of iterations in the last time step. It the step fails, the user in the script must try something else such as cutting the time step.

[amatamor]

I did specify the-time option. My concern is that even though a warning is shown, when using the BFGS algorithm it seems that execution continues without any reduction in step size.

Output from KrylovNewton algorithm when OS prints a warning indicating that the number of iterations has been exceeded (first column is time)

7.82 -0.0115634 -0.0154045 -0.0203649 -0.022737 -0.015406 -0.00656337 -0.00324033 -0.0134686
7.83 -0.0114386 -0.0157158 -0.0210057 -0.0233491 -0.0158204 -0.00658272 -0.00317227 -0.0136988
7.84 -0.0113041 -0.0160403 -0.0216177 -0.0238583 -0.0162025 -0.00664246 -0.00309806 -0.0139115
7.85 -0.0111689 -0.0163659 -0.0221931 -0.0242774 -0.0165445 -0.00674272 -0.00302427 -0.0141076
7.85198 -0.0111427 -0.0164296 -0.0223021 -0.0243506 -0.016607 -0.00676718 -0.00301029 -0.0141444
7.85288 -0.0111309 -0.0164584 -0.022351 -0.0243829 -0.0166348 -0.00677877 -0.00300408 -0.014161
7.85339 -0.0111242 -0.0164748 -0.0223788 -0.0244011 -0.0166506 -0.00678551 -0.00300057 -0.0141704
7.85382 -0.0111187 -0.0164884 -0.0224019 -0.0244162 -0.0166636 -0.0067912 -0.00299768 -0.0141782

Output from BFGS algorithm when OS prints a warning indicating that the number of iterations has been exceeded (first column is time)

7.82 -0.0125199 -0.0154833 -0.0196288 -0.0226785 -0.0159354 -0.00683636 -0.00362136 -0.0137268
7.83 -0.0122963 -0.0158282 -0.0205271 -0.0235771 -0.0164471 -0.00689915 -0.00357844 -0.0140352
7.84 -0.0120511 -0.0161908 -0.0214276 -0.0243483 -0.0169296 -0.00701316 -0.00350292 -0.0143235
7.85 -0.0117959 -0.0165612 -0.0223105 -0.0250103 -0.0173764 -0.0071756 -0.00340392 -0.0145926
7.86 -0.0115425 -0.0169273 -0.0231562 -0.0255839 -0.0177824 -0.00737712 -0.00329564 -0.0148434
7.87 -0.0113032 -0.0172755 -0.0239471 -0.0260895 -0.0181455 -0.00760253 -0.00319592 -0.0150771
7.88 -0.0110909 -0.0175926 -0.0246682 -0.0265456 -0.0184651 -0.00783553 -0.00312213 -0.015295
7.89 -0.0109159 -0.0178684 -0.0253091 -0.0269675 -0.0187398 -0.00806284 -0.0030877 -0.0154984

[fmk]

can you send me the script. fmckenna ATTT berkeley DOTTTTTTTT edu