CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   CFX (https://www.cfd-online.com/Forums/cfx/)
-   -   Fatal overflow problem (https://www.cfd-online.com/Forums/cfx/222576-fatal-overflow-problem.html)

Ordo November 29, 2019 09:00

Fatal overflow problem
 
Greetings everyone

While running one of my simulations, I encountered the following error message:

ERROR #004100018 has occured in subroutine FINMES.
Message:
Fatal overflow in linear solver.

The solver stopped running after this and no result files were generated. As far as I understand, this means the residuals are diverging too badly for the solver to continue. After reading up on the message, I quickly came across the FAQ on cfd-online.com

I've checked all the points there and the only one, I'm not too sure about are the initial conditions. I'm doing a steady state analysis of external flow around a building and initialized the flow field with a logarithmic wind velocity profile. This is the same one, I defined at the inlet of my domain. For my first simulations (same velocity profile, but different angles of attack), the solver managed to converge and the results seemed plausible. Now I was setting up a few variants of the case with different maximum velocities, while the distribution (logarithmic profile) qualitatively stays the same.

So for the case where I chose a lower velocity and for an angle of attack of 15 I got the error. What's confusing me, is that it only shows up for this one angle of attack of 15, while the other angles with this lower velocity converge just fine and also generate plausible results.

My next steps were going to be to initialize the problematic case with the results of my first case with the higher velocity, or try using upwind for advection scheme and then using those results as an initial condition for the final run with high resolution.

Any other thoughts, what might be the problem here?

ghorrocks November 29, 2019 18:58

Yes, the FAQ is the place to start with this error. If your model is converging sometimes and not others after only small changes than it suggests your model has poor numerical stability. To improve the numerical stability consider:
* Improve mesh quality - this is the most important one and always seems to get dismissed by noobies. But improving mesh quality ALWAYS makes everything easier.
* Better initial condition - try using one of your converged results from another AOA for an initial condition.
* Smaller time steps
* Double precision solver

The higher velocity case is likely to be less numerically stable, but give it a go. Using upwinding to get an initial condition can also work, but I generally avoid this as I prefer to fix the underlying numerical instability so it can run with an accurate scheme from the start.


All times are GMT -4. The time now is 05:23.