RE: advise on error shown

[ryancwl95]

Hi,

I am attempting to analyze a monopile embedded in sand, which is modeled using the SanisandMS material model. The pile is subjected to a load of approximately 2.5 meters at the pile head. The analysis successfully runs until a movement of 1.45 meters is achieved at the pile head. However, after this point, an error is displayed.

Intel oneMKL ERROR: Parameter 5 was incorrect on entry to DGELS.
WARNING KrylovAccelerator::accelerate() -
error code -5 returned by LAPACK dgels
WARNING AcceleratedNewton::solveCurrentStep() -the Accelerator failed in accelerate()
StaticAnalysis::analyze() - the Algorithm failed at step: 6832 with domain at load factor 34165

Could you explain what this error means and how I can prevent it?

My current analysis settings are:
numberer RCM
system Mumps
test NormDispIncr 1.0e-003 30 1
algorithm KrylovNewton
constraints Auto
integrator LoadControl 5
analysis Static

[STKO Team]

It seems the Krylov accelerator found some wrong floating-point values (probably coming from a division by zero or similar).
Maybe you have some convergence issue that leads to that.
Try using a simple newton (without the krylov accelerator and see if you still have those NANs)

[ryancwl95]

Hi,

By simple Newton, you mean the most basic one? Without modification or line search.

algorithm Newton <-initial> <-initialThenCurrent>

Would there be any suggested input for <-initial> <-initialThenCurrent> that can help with this convergence issue?

Sincerely,
Ryan

[STKO Team]

try with just algorithm Newton

[ryancwl95]

Hi,

I tried the simple Newton approach, but non-convergence seems to happen earlier than when I use the KrylovNewton algorithm.

The error i received is:

WARNING: CTestNormDispIncr::test() - failed to converge
after: 1000 iterations current Norm: nan (max: 0.001, Norm deltaR: nan)
NewtonRaphson::solveCurrentStep() -the ConvergenceTest object failed in test()
StaticAnalysis::analyze() - the Algorithm failed at step: 0 with domain at load factor 6898.75
OpenSees > analyze failed, returned: -3 error flag
Analysis terminated at step 1383
LOAD/CONSOLIDATION ANALYSIS COMPLETED..

Do you think that I should be reducing the tolerance in test NormDispIncr, or perhaps change to a different test?

[STKO Team]

Ok so it means it's not strictly related to the Krilov accelerator.
Try to look at the output of your model and see if some parts is blowing up. Those NANs typically come from numerical instabilities

[ryancwl95]

Hi,

Thank you for your reply.

Is there a particular result type or result types that I can view in the STKO post-processor that would help me identify what is "blowing up" and causing this numerical instability? Or maybe a specific "result type" output particular to the sanisandMS material?

[STKO Team]

Typically displacement (or accelerations in transient), try to see pieces of your model are displacing in strange ways