Auto time stepping, also known as time step optimization, aims to reduce the solution time especially for nonlinear and/or transient dynamic problems by adjusting the amount of load increment. If nonlinearities are present, automatic time stepping gives the added advantage of incrementing the loads appropriately and retreating to the previous converged solution (bisection) if convergence is not obtained. The amount of load increment is based on several criteria including the response frequency of the structure and the degree of nonlinearities in the analysis.
The load increment within a step is controlled by the auto time stepping procedure within limits set by you. You have the option to specify the maximum, minimum and initial load increments. The solution will start with the "initial" increment but then the automatic procedure can vary further increments within the range prescribed by the minimum and maximum values.
You can specify these limits on load increment by specifying the initial, minimum, and maximum number of substeps that are allowed. Alternatively, since a step always has a time span (start time and end time), you can also equivalently specify the initial, minimum and maximum time step sizes.
Although it seems like a good idea to activate automatic time stepping for all analyses, there are some cases where it may not be beneficial (and may even be harmful):
Problems that have only localized dynamic behavior (for example, turbine blade and hub assemblies), where the low-frequency energy content of part of the system may dominate the high-frequency areas.
Problems that are constantly excited (for example, seismic loading), where the time step tends to change continually as different frequencies are excited.
Kinematics (rigid-body motion) problems, where the rigid-body contribution to the response frequency term may dominate.