Hello all,

I am having convergence problems. There is a persistent warning in the .info file for the models which fail. The warning reads:

*** WARNING #038 *** UNNORMALISED RESIDUAL CANNOT BE REDUCED TO 1.000000E-06 IN EQ: PCOR

ROUND-OFF ERROR RESCUE NOT ATTEMPTED; SOLUTION RETURNED WITHOUT UPDATE

Has anybody encountered this warning? Does anybody understand this warning and how to remedy it?

Also, I am getting the inflow at the outlet boundary warning. I have increased the mesh resolution at the outlet, but this doesnt seem to help with that warning.

Will an atm pressure boundary at the domain outlet fix everything?

thank you for any advice.

David Aquilina

David, there are 2 things to check: first - are your inlet and outlet bc connected? sometimes when refining a mesh on forgets to merge the vertices, so you have an internal wall. flow can't get out of your domain and that is the reason that the presuure sweeps fo incompressible flow didn't converge. the second point is: increas the number of pressure sweeps to 2000 or even higher. when you have very long channels in your domain or in other words- the way the flow has to follow is very long the solver needs lots of pressure sweeps for the first 1 to 10 iterations.

hope this helps,

thomas

I had a similar problem with a geometry that was very long - tube like. The adapco folks recommended that I try the AMG solver.

Thank you both Thomas and Paul for you advice.

I increased the number of pressure sweeps to 2000, as Thomas suggested. The model is still solving but there are no warnings or errors in the info file, so things are looking good and promising.

I will try out the AMG solver to see if there is any difference.

I appreciate the help, David

<font face="Courier new">Thomas,

How long is too long ? Any aspect ratio recommendations ?

Murali</font>

Murali, I wouldn't say the domain is too long. The conjugate gradient solver needs lots of pressure sweeps for long channels - that is my experience. I am not a code developer, so I do not really know why! But as Paul mentioned, in such cases, the use of the AMG-solver (if you have a fortran-90 compiler installed) gives you a much better performance.

concerning the aspcect ratios, i would recommend the values in the manual.

cheers,

Thomas