Stationary solution converge but time dependent doesn't

I've coupled a COMSOL model for fluid dynamics with a very simple pde that model the transport of humidity in air. When I solve it for the stationary case, the solution converge easily, but when I try to solve the time dependent problem, the solution can't reach convergence. This seems very strange to me, the fluid is laminar and incompressible, the mesh is well designed, the initial values are coherent, all the conditions are smooth. Is it even possible that COMSOL isn't able to find a solution for the time dependent problem, while he found very easily the time independent one?

Thank you

• It would be helpful to know more about what "does not converge" mean. Does the solution explode? Does it become periodic and oscillate? – Wolfgang Bangerth Apr 12 '13 at 16:33
• Comsol says: Nonlinear solver did not converge Time:0 Last time step did not converged I've tried to do the same thing with precompiled modules: now I've tried o couple Navier-Stokes equations with heat transfer in solids. Stationary solution converge easily, but time-dependent can't find a solution. For coupling the two models I did a Identity map and then fixed a dirichlet temperature condition on the boundary of the two models for temperature, as T0=mod1.p1(T2) Even if this method works fine for the stationary solver, I don't know if there is an easiest way to couple two different models – Lorenzo Nespoli Apr 13 '13 at 7:42
• Does your solution fail to converge on the very first timestep? Or does it diverge after a number of timesteps? Also, it would help if you could provide the equations your solving and some details about the discretization scheme you used in comsol. – Paul Apr 14 '13 at 4:56
• The solution failed to converge at the very first time step. The equations are written by COMSOL, so I don't know exactly what equation are, but they're supposed to be correct. That are the standard non isothermal flow and heat transfer in solids.The solver settings are the default ones, that are: fully coupled, solver: PARDISO. I've tried to set the solution of the stationary case as input for first time step, but this didn't work neither. Do you know if there is another method for granting T and q continuity in two sections using different models, apart from identity mapping? – Lorenzo Nespoli Apr 14 '13 at 7:18