I wonder if anyone has come across this before; when I have a homogeneous scalar throughout a domain. I have a value specified for the scalar at a wall of the same value as that in the domain. So I should expect no change in the scalar values throughout the domain right. But, there is a change in the scalar values, and while the solver will converge to the true solution, it does go off on a bit of a wild goose chase. I have been told that the error propagates from the rounding error. However, by looking at the scalar values at the monitoring point the values change by several "degrees". If anyone else has had this problem and has a solution or can offer comment I'd be very grateful.

Many thanks in advance

Dougal MCQueen

... it does go off... what?

... several "degrees"? Do you mean percentages.

Have you tried double precision? Or upwind, as long as the values are constant?

Pascale

Sorry I should have said "units" really. The scalar stands in for potential temperature. By it "goes off", I mean that the solver diverges wrt the scalar value after ~10 iterations, and then comes back towards a sensible value, after ~100 iterations. I start with a value of 293, the solver diverges to a worst value of about 270, and then settles back to within 2 units of the 293 value. I have used higher upwind for velocity and ccct for the scalar as the differencing schemes. In the domain the flow should be strictly straight through a box from the inlet to outlet (atmospheric flow) with a rough floor/ground.

I hope this answers these questions adequately.

Mmm, looks like you use CFX-4.

Actually, I have no idea. I would expect that upwind provides best results, higher upwind the worst and CCCT intermediate. Possibly you have problems with unphysical under- and over shoots.

On the other hand, I remember that with CFX-4 it was difficult to restart calculations from an exsisting solution because internally some unwanted resettings were performed. Consult the helpdesk with this.

Pascale

Hi Dougal,

I'm not sure I completely understand the problem, but I'll give you my 2 cents worth anyway.

If the solution is not converging then your answer could be anything. I think the key to keeping this simulation under control is by stopping the poor residuals. Is the simulation steady state or transient? Any ideas why the residuals "go off"?

The high residuals might be due to the scalar being convected from the inlet to the outlet - that means it takes a certain amount of time before the scalar is "flowing" throughout the entire domain.

Glenn