MICRESS Forum

Hi:
I am doing calculation with elastic coupling, but I don't konw how to choose the following numberical paramters, especially the former two parameters? Are they related with the grid spacing dx or anything esle?
# Other numerical parameters
# ==========================
# convergence criteria for BiCGStab-solver (stress calculation)
# and (average) strain calculation? (real) (real)
1E-6 1E-6
# max. number of iterations for BiCGStab-solver
# and (average) strain iterations? (int) (int)
100 3
# Phase minimum?
1.00E-05
# Interface thickness (in cells)?
5.00
Thank you
swh

The stress and strain field is derived from the mechanical equilibrium condition. This condition defines a set of linear equations for the local displacements and the corresponding matrix is solved by the iterative BiCGSTAB-algorithm. The convergence criterion defines the termination condition for the iterative solver and is a measure of how close the approximated solution comes to the exact solution. Unfortunately, it is not straight forward to recommend a general value which leads to an acceptable solution. One has to compare different simulations in order to get an idea whether the convergence is sufficient enough for the particular case. With decreasing convergence criteria the results comes closer to the exact one, but of course the computation takes longer times.
In order to solve the elastic equilibrium for inhomogeneous materials a kind of "predictor/corrector" algorithm is implemented. The second number is the convergence criterion for this second part (named as "strain iteration"), again one has to play around a bit. Usually a sufficient solution can be achieved with strain iterations in the order of 10.

The maximum numbers just limit the number of iterations. Of course in this case the solver can end up with a "bad" solution for the stress and strain field, one has to check. However, limiting the number of iterations avoid that MICRESS will stuck in case of some (temporal) convergence problems. The screen output gives information whether convergence mismatches appeared and also about the number of iterations.