When I use single phase in simulating tank filling,it converge well.But when I switch to multiphase,error appears: | ERROR #004100018 has occurred in subroutine FINMES. | | Message:| | Fatal overflow in linear solver.

Can any body tell me why?

The convergence fairy is stubborn in multiphase flows. Could be any of a myriad reasons.

You can try to reduce your timestep. Sometimes, with two fluids, I've worked with 1e-5. When the flow will be more stable, you can use a bigger timestep. However, it's better to obtain an initial value file, for example using first a 'zero equation' turbolence model, which is very robust.