Always check that the problem has been set up correctly. This is the most common cause of difficulties running CFD software. Also, failure to read the manual is the most common cause of error in problem definition!
Good general guidelines for Best Practice in CFD are given in the Ercoftac Report [2 ], and this is well worth reading.
Once the problem definition has been checked and, if necessary, corrected, it is desirable to run the job through any data check facilities that may be offered by the CFD software being used. Also a check on the solution after one or more iterations may reveal further set-up errors or confirm that the problem has been formulated correctly.
Running the solver initially with a coarse mesh may also help to identify setup issues which could inhibit convergence.
Has the job been classified correctly according to the physics involved? For example has the problem been classed as incompressible, when it is really compressible. An illustration of this would be a problem with an inlet Mach number of less than 0.3, but which gets much larger than this in the interior of the flow. As a further example, defining a laminar problem when turbulent conditions exist is also likely to lead to convergence problems.
Another example where the setup of the physics of a problem is important is with transient behaviour. Many Fluid Dynamics processes are inherently unsteady, both with random turbulence, and large-scale instabilities, such as the vortex shedding behind a circular cylinder. Well-known examples of flows with large scale instabilities include:
Experience has shown that for this type of flow, as the ability to resolve the spatial and time variations improves, then the results are also more likely to show large scale time variations. A non-converged solution may be an indicator of an unstable process with a transient solution. However, this conclusion should be the last conclusion, having eliminated all other possibilities.
Indications of transient behaviour may be periodic fluctuations in the residuals, and regular variations in the solution variables. Things to consider include:
Have consistent units been used throughout the model and correct physical parameters specified, using smooth and self-consistent physical properties?
If curve fitting of experimental data has been used to define material properties, are sensible and consistent material properties predicted at all stages of the solution? Solution variables that are used in the calculation of material properties may take on very different values at the early stages of a solution compared to their final values. This could result in material properties being determined from extrapolated data.