Best practice for selection of substeps in Structural models
Venkata asked about the best practice for setting the number of substeps in Structural models.
The solver is configured, by default, to attempt to obtain convergence of each substep within 26 iterations. If it fails to find convergence by then, it will divide the time step in half, which is called bisection, and try again. There are other triggers that will make it bisect before waiting for 26 iterations to accrue. There are commands that can override the number 26 and force the solver to continue iterating for longer.
If you are too aggressive and request too large an initial time step (too few initial number of substeps), it might happen that the solver bisects before it reaches convergence on the initial substeps. That is just a waste of iterations. If you notice this, then the corrective action if you solve again from the beginning is to increase the number of initial substeps.
If you are too conservative and request too small an initial time step (too many initial number of substeps), you will have done more iterations than the minimum needed to get to the next increment.
However, the solution logic is very good at automatically increasing the size of the time step if it sees convergence is happening quickly. The cost of being too aggressive is usually much higher than the cost of being too conservative, so be conservative in the initial number of substeps (use a larger number).