Time steps in transient run with heat transfer

Linda

I have done a transient simulation i CFX. The flow is a turbulent flow in an (almost) straight pipe. The heat loads are linearly ramped up from around 100 K to 600 K during one second.

When the boundary layer at the wall is resolved (y+<1)I need to set a time step of about 1.5*10^-5 s. (This is the size of the physical time step from the steady state run) Otherwise the solver gives the following error: "floating point exception", and is then terminated.

If y+ is about 5 instead the time step needs to be about 0.004 s for the solution to converge. This is also a smaller time step than is desired.

It is obvious that there is a connection between the resolution of the grid, and the resolution in time. Can anyone explain why this is? There is not supposed to be any limitations of the maximum size of the time step since the solver in CFX uses an implicit method for the integration of time.

I would be grateful for an insight in this problem.

Thank you /Linda