I AM DOING SOME SIMULATION ABOUT STEADY LAMINAR NATURAL CONVECTION IN A SQUARE CAVITY AT VERY LOW PRANDTL NUMBER (LIQUID METALS), BUT I HAVE SEVERAL PROBLEM IN THE CONVERGENCE OF THE CALCULUS, ALSO FOR VERY RELAXED EQUATIONS (0.01) AND ALSO FOR MULTIGRID TECHNIQUES. THIS PROBLEMS ALSO REMAIN FOR THE TRANSIENT. THERE IS ANYONE WHO CAN TELL ME IF HE HAS THE SAME PROBLEMS?

I NOW THINK THAT CFX IS NOT SUITABLE FOR THE CONVECTION OF LIQUID METALS!!

Hi,

Are you using CFX4 or CFX10? Both codes can do natural convection with low Prandtl number.

Why have you heavily relaxed the equations?

Glenn Horrocks

I have used cfx-5 to validate it against experiments of Tian et al. for an air filled cavity. For this case it was possible to get converged results.

Sorry, was the wrong location. Here again: I have used cfx-5 to validate it against experiments of Tian et al. for an air filled cavity. For this case it was possible to get converged results.

Hi,

I am confused as to what you are trying to do. Are you saying you have successfully validated CFX-5 for an air filled cavity but when you use a low Prandtl number fluid it does not converge?

Glenn Horrocks

No. What I was trying to say is that natural convection in general can be calculated with CFX-5. I have compared my simualtions against some experiments for Pr=0.72 and it worked fine. I have no idea how it works for very small Prandtl numbers, but I guess it maybe better using not such a high under-relaxation factor.

Yes: I validated my codes (cfx 10.0) with the benchmark of DeVahl Davis (natural convection of air Pr = 0.71, ra 1000-10 000- 100 000) and cfx works fine. Then I tried to performs some simulations of Pr = 0.0071 (liquid metals) and cfx 10.0 does not converge!!!!

Hi,

Low Pr fluids have slower time constants than high Pr fluids. This means you might need to use larger timesteps.

There should be no reason why you cannot get CFX to converge with low Pr fluids, it is just a matter of finding the right convergence settings.

Glenn Horrocks

BUT I USE THE STEADY STATE, SO THE TIME STEP DOESN'T MATTER, I THINK. I THINK THE REASON IN THE TYPE OF DISCRETIZATION OF THE ADVECTIONT TERMS: I THINK THE SOUITABLE DISCRETISATION METHOD IS THE CENTRAL DIFFERENCES, BUT IN CFX IT IS NOT AVAIBLE IN THE LAMINAR SIMULATIONS (ONLY FOR LES)!

Dear Tommy,

You must keep in mind that time step matters a lot when you are talking about CFX since they do use false time stepping to approach the steady state solution.. Others solver use plain underelaxation to mimic the same idea.. If you cannot resolve the advection time scales with your time step, the non-linear solver (outer loop) will have a hard time converging or not at all..

Besides, natural convection of low prandtl number can have multiple solutions under certain circumstances.. For example, under the Benard-Rayleigh configuration, or Taylor-Couette is very easy to get different "unstable" steady state solutions depending how you approached the solution..

Good luck, Opaque..

Hi,

Opaque is right - CFX uses a pseudo-timestepping method even for steady state simulations so timestep size is important for steady state simulations too. Have a look at the documentation under achieving convergence.

Also I should point out that it is unlikely that central differencing is a good advection scheme to use. It is quite numerically unstable. The high resolution or hybrid scheme is more likely to be suitable. Again read the documentation on advection schemes.

Glenn Horrocks