Full-Newton and Quasi-Newton Iterative Schemes

When running problems using Implicit solution sheme in LS-DYNA, the default iterative non-linear solver used is the BFGS method that employs a ‘Quasi-Newton’ method in which the global stiffness matrix is reformed only every ILIMIT steps and in between these a relatively inexpensive update to the stiffness matrix is performed. This default stiffness matrix update scheme works very well and relative fast when there is no significant contact is involved and the non-linearily of the problem is moderate. The default method often fails to converge when the non-linearity of the problem grows or when significant amount of contact is involved. For such problems, a more expensive Full-Newton method is recommended by setting ILIMIT=1 which forces LS-DYNA to reform the stiffness matrix at every iterative step to yield a more accurate stiffness estimation. When full-newton method is employed, it is often necessary to set a large maximum allowable reformations using the MAXREF parameter to allow sufficient number of stiffness reformations before terminating or reduction of the solution timestep (when AUTO timestep control is activated). Recommended value of MAXREF depends on the problem but typical values could be around 200 which should be sufficient for any large non-linear problem to converge on a solution.

Related keywords include *CONTROL_IMPLICIT_SOLUTION, *CONTROL_IMPLICIT_SOLVER

• LiFei says:

My question is that how I can figure out the reasons for the problems which are not convergent.Sometimes ,it is very difficult for me to find the answers.Can you give me any advice?And maybe you can introduce some your tips to me.Thank you!

• Suri Bala says:

LiFei,

There are several techniques to debug ill-conditioned implicit problems.
Please look for my post on this topic soon.

Thanks,
Suri Bala

• Jose says:

Hello

In parallel job some times give a error “LCPACK”.
the error give me in the middle of analysis.
which are the general reasons of this error?

thanks

• Suri Bala says:

Jose

This is probably coming from the Linear Solver module due to the presence of unsupported feature in the keyword file. It could also be due to badly (over) constraint model. You can first perform a EIGEN value problem with and wihtout constraints. If you still have the same problem, you can report it to bugs at lstc dot com.

Suri