[esp-r] CFD problem - reprise

[Andrew Cowie]

Hi all,

Once again i find myself having trouble with the CFD in ESP-r.  As always, basic info first:
OS: 64-bit CentOS 5.8
ESP-r: "Release branch" Release 11.11, revision 8518, compiled with GNU compilers

The problem: When running CFD sims (type 4 conflation, handshaking type does not appear to affect the problem), some of the time the CFD calcs will seemingly stop at some random point.  All of the CFD results (from the monitoring file) from this point on are NaN.  This causes the sim that it is currently on to prematurely "converge", and all further sims "converge" in 20 iterations (coincidentally, for those who remember my previous CFD problem, this is exactly what would happen when i ran type 2 thermal conflation, except it would do it all the way through every time).  I have found a number of clues to the problem, but i can't seem to pin it down:

 *
In the CFD monitoring file, the NaN's always start at the W value, i.e. the entry will always be:

XXXX   YYYY ZZZZ NaN  NaN  NaN  NaN  NaN

           NaN  NaN  NaN  NaN  NaN  NaN  NaN

where XXXX is the iteration where the error occured, YYYY and ZZZZ are correct as far as i can tell.

 *   In the ACC actions file, in all sims after the error has occured (it will always occur in the actual simulation, not the preliminary CFD), when the ACC algorithms are invoked all FORorNAT numbers come out to be 0.  However the handshaking calcs appear to be fine.
 *   I have ran it in gdb with the debugging markers, and it did not come up with any errors, so no obvious numerical issues.
 *   The errors will occur consistently in the same place for the same sim.

Thats it, i havn't found any other patterns to the errors.  As far as i can tell though it is a tad more likely to happen the warmer the climate in the sim.  Also, it only seems to happen for models with windows.



Any ideas?  As i'm sure you can tell, im not the worlds best debugger so any input at all would probably be helpful.



Warmest regards,

Andy Cowie