[larsschwarzer]

Hi everyone,
I have been trying to write a UDF to manipulate rate coefficients for the thermal (Zeldovich) NO-mechanism. A first test version simply uses the rate coefficients given in the Fluent Theory Guide - to make sure the UDF does what it is supposed to do. After some effort I managed to compile the UDF (big thanks to this Forum!). When I set up my case and start the calculation, Fluent goes busy for several minutes/hours, then returns with an error:

Divergence detected in AMG-Solver: pollutant-0

As I understand, this has something to do with the input parameters of the UDF, i.e. the syntax is correct, but some of the computed values are causing stability issues. I am using the same settings as I would using the built-in Fluent NOx calculation (which of course work fine).

As the error message does not really point to the problem, here are some questions:
1) Has anyone ever encountered the same problem?
2) Does the DEFINE_NOX_RATE macro require parallelization? There is no information on that in the manual.
3) I found an old post here,
http://www.cfd-online.com/Forums/fluent/44944-nox-udf.html
Some of the commands seem to be replaced in the current version of Fluent but the code can be easily adapted. I've been specifically wondering about the if-clause:

#define SMALL_N 1.e-6
...
rf = rr = 0.0;
s2 = s3 = 0.0;
s1 = kf2*MOLECON(NOx, O2);

if (s1 < SMALL_N) {
   rf = 2.*o_eq*kf1*MOLECON(NOx, N2);
   rr = 0.0;
}
else {
   s2 = 1.+ kf3*oh_eq/s1;
   s3 = s1 + kr1*MOLECON(NOx, NO)/s2;
   rf = 2.*o_eq*kf1*MOLECON(NOx, N2)*s1/s3;
   rr = 2.*o_eq*kr1*MOLECON(NOx, NO)*kr2*MOLECON(NOx, NO)/s3;
}

NOx->fwdrate += rf;
NOx->revrate += -rr;

-- does this in any way relate to stability of the solver? The formally correct rates are in the 'else' part. It looks to me as if Fluent is having trouble when dividing a small number by another small value, and that this is being addressed here. I have tried adding this to my code, but the output does not change.

Any help or thoughts on the matter would be appreciated.

Lars

p.s.:
Using Ansys Fluent 15.0 with 3d, double precision solver on parallel Windows 64-bit machine. I can make the UDF-Code available if that would help.

[larsschwarzer]

For those who happen to stumble on this post - I could meanwhile solve some of my problems:

The NOx-Macro does not require special considerations for parallelization, it works 'as is'.

I tried manipulating the value of "SMALL_N" and set it to prohibitively high values, which cancelled most NO production but led to more stability. (The error disappeared).
What was more effevtive was reducing the under-relaxation factor (finding a suitable value is rather alchemy than science to me...) for 'pollutant no' and changing the solution method from second to first order. The UDF runs now, but it's extremely slow.

Does anybody know if the built-in model works with pre-calculated values for NO production from a look-up table?

Enjoy your weekend,
Lars