When using the density-based solver with the implicit solution formulation in steady-state you can accelerate the convergence of your solution on highly-stretched and anisotropic meshes (like the one used when modeling external aerodynamic problems) by selecting the Convergence Acceleration For Stretched Meshes in the Solution Methods task page. For further information and theoretical background on this solution acceleration option, see Convergence Acceleration for Stretched Meshes in the Theory Guide. The Convergence Acceleration For Stretched Meshes option provides an optimum solution convergence of the implicit solution method.
To apply convergence acceleration for stretched meshes, perform the following:
Specify the solver options by selecting Density-Based and Steady in the General task page.
Select Implicit from the Formulation drop-down list and enable Convergence Acceleration For Stretched Meshes in the Solution Methods task page (Figure 28.8: The Solution Methods Task Page for the Density-Based Implicit Formulation).
Extra settings for CASM can be set using the following text command:
solve → set → convergence-acceleration-for-stretched-meshes/
Enter yes in response to the Use convergence acceleration for stretched meshes (CASM)? question. You will also be asked for a cut-off on the CFL value
multiplier. By default this value is set to 100. Typically, you do
not need to adjust this value. But if convergence difficulties are
encountered and reduction of the CFL value alone does not help improve
convergence, then it is advisable to reduce this CFL multiplier cut-off
to a lower value (for example from 100 to 50, 20 or 10).
The use of CASM can typically give a much faster convergence over the standard solution method. In general, when using CASM, you do not need to specify a very large CFL value as you do with the standard solution method. A CFL value between 5 and 10 is typically used for converging most flow problems. When the Convergence Acceleration For Stretched Meshes option is selected, the solver will run when appropriate with a variable local CFL value proportional to the cell aspect ratios. Therefore, when the cell aspect ratio nears unity (typically far from walls), the local cell CFL value will be the same as the value that you supplied. However, as the cell is stretched and the cell aspect ratio increases (near walls), the local cell CFL value will be multiplied by the cell aspect ratio value. This is true until the cell stretching is beyond the multiplier cut-off value specified using the text command. The proportional change in CFL value on highly stretched cells helps accelerate the solution especially on highly packed and stretched meshes like the one used in modeling external flow problems.
When the cell aspect ratio is selected by default, the density
based implicit solver will operate with an explicit relaxation of
0.5 which can be adjusted from the Explicit underrelaxation
value entry in solve → set → expert
The convergence of the solution is mainly controlled by adjusting the CFL value. Therefore, if convergence problems are encountered, lowering the CFL value will help improve the convergence. Additional solution parameters to be adjusted for more conservative solution settings are:
lowering the density-based implicit solver explicit
relaxation (solve → set → expert)
adjusting the CFL multiplier cut-off value to a lower
value (solve → set → convergence-acceleration-for-stretched-meshes/)
This solution convergence method is very aggressive. Therefore it is of paramount importance to start with a good initial guess especially if you start with second order spatial discretization. To get a good starting solution with the guess you provide, you are advised to use the full multi-grid initialization method (see Full Multigrid (FMG) Initialization).
Note: When selecting the Convergence Acceleration For Stretched Meshes option then the Pseudo-Transient solution method will not be available. You can either use Convergence Acceleration For Stretched Meshes or the Pseudo-Transient solution method. These two options cannot be used at the same time. Both methods help in obtaining faster convergence on anisotropic meshes. But one method requires that you enter a CFL value, while the other requires that you enter a pseudo-time step value to march the solution to convergence. It is up to you to select the method with which you are most comfortable.
Convergence Acceleration For Stretched Meshes shows an advantage over the standard solution method, particularly with stretched meshes with low Y+ values (near unity). The use of Convergence Acceleration For Stretched Meshes results in an alteration of the numerical dissipation of the selected flux scheme. This change may slightly impact the monitored loading level if compared with the solution obtained without the use of this option.