Diverging simulation (Nan)

Always verify after a simulation:
• boundary conditions at inlet(s), outlet(s) and on solid walls
• whether backflow exists at inlet(s)/outlet(s)
• velocity vectors on relevant cut planes in the domain
• Y+: in line with turbulence model ?
• µt/µ: maximum value and within a cut
• Cartesian plots perpendicular to walls: boundary layers well captured?
• physics of flow solution close to stagnation points, separation areas and in wakes

In case of divergence the cause could be:
• Poor mesh quality: the mesh should not have negative, concave or twisted cells. Make sure the mesh quality is adequate (Show mesh quality statistic).
• Incorrect boundary condition.
• Unrealistic initial condition?
• Inappropriate turbulence model (without wall functions Y+ <5 and at least 6 layers at y+ < 10).
• Reduce the CFL number.
Look also at quantities like pressure and temperature in the domain: minimum or maximum values may be quite unphysical. For example you might have temperatures locally as low as 100 K while the physics of the application don't explain such values.
If the simulation diverged and gives NaN, it may not be possible anymore to look at it in the results analysis. In such case I run the simulation again but with a number of iterations such that the simulation will stop before the NaN.

If the case is not diverging but not well converging either:
• Unsteady flow features? The residual of a steady simulation will stop reducing after some time and show (periodic) oscillations. – It is very important to define probes at relevant locations (e.g. wake) to confirm the oscillations.
• Too high expansion ratios in the boundary layer can lead to force oscillations, reduced stability and inaccuracy. – Check that the expansion ratio in the boundary layer is below 1.2.

Always verify after a simulation:
• boundary conditions at inlet(s), outlet(s) and on solid walls
• whether backflow exists at inlet(s)/outlet(s)
• velocity vectors on relevant cut planes in the domain
• Y+: in line with turbulence model ?
• µt/µ: maximum value and within a cut
• Cartesian plots perpendicular to walls: boundary layers well captured?
• physics of flow solution close to stagnation points, separation areas and in wakes

In case of divergence the cause could be:
• Poor mesh quality: the mesh should not have negative, concave or twisted cells. Make sure the mesh quality is adequate (Show mesh quality statistic).
• Incorrect boundary condition.
• Unrealistic initial condition?
• Inappropriate turbulence model (without wall functions Y+ <5 and at least 6 layers at y+ < 10).
• Reduce the CFL number.
Look also at quantities like pressure and temperature in the domain: minimum or maximum values may be quite unphysical. For example you might have temperatures locally as low as 100 K while the physics of the application don't explain such values.
If the simulation diverged and gives NaN, it may not be possible anymore to look at it in the results analysis. In such case I run the simulation again but with a number of iterations such that the simulation will stop before the NaN.

If the case is not diverging but not well converging either:
• Unsteady flow features? The residual of a steady simulation will stop reducing after some time and show (periodic) oscillations. – It is very important to define probes at relevant locations (e.g. wake) to confirm the oscillations.
• Too high expansion ratios in the boundary layer can lead to force oscillations, reduced stability and inaccuracy. – Check that the expansion ratio in the boundary layer is below 1.2.