Negative Eigenvalues

[anantha]

Greetings Dr. Jorgen and others:

I am modelling a visco-hyperelastic device in Abaqus. I am doing a nonlinear large displacement analysis under displacement controlled loading. During the analysis, which is incrementing itself at a time step of 0.005 seconds, to a total of 2 seconds, I get these warnings.

System has "n" negative eigenvalues. Explanations are suggested after the first occurence of this message. And the analysis continues on without much trouble. Sometimes however, the analysis aborts with the error "too many increments attemped for the time step"

I wanted to know if these warnings had any great effect on the final results and also if the results could be taken as almost correct, if not accurate ?

I was also wondering as to why this error came only under displacement controlled loading. For example, I applied a displacement of 1 inch * sin Pi*t at a node and carried the analysis, and the reactive force on the node was 0.5 kips. During this analysis, sometimes I get the above said warning messages. However, If i applied a load of 1 kips * sin pi*t on the same node and ran the analysis, the analysis never gave me any warnings even though the displacements at those nodes were something like 2 or 2.5 inches. Can anyone please tell me the reason for the same ?

Also, is it possible to model a tension-only solid element in abaqus using only CAE ? I want to model a tension-brace in a frame, which has to be a solid element , and should have zero compression stiffness.

I am sorry if this has been a very very long post. I hope to have stated my problems clearly.

Thank you very much
Anantha

[Jorgen]

Hello Anantha,

It is not uncommon to get warnings about negative eigenvalues when running FE simulations, and if your simulation reaches the final time that you are interested in, then your solution is perfectly accurate.

Sometimes the negative eigenvalue warnings lead to more global instability and premature termination of the simulation. In this case your simulation is likely accurate up till the point where you start to get convergence problems.

There are no hard rules about the influence of displacement control vs. load control on the appearance of negative eigenvalues. The problem is more related to geometric distortions and the constitutive response.

Your question about tension-only solid elements is interesting. I am not aware of general and simple solution to that. What you can do is to write a UMAT what only has stiffness in tensile loadings. Depending on what material model you are interested in, this might be simple to do.

Best of luck,
Jorgen