[kaush]

Hi

I am trying to use a combination of JST and ROE scheme for the RANS solution of an aerodynamic problem. I notice that the ROE scheme performs poorly in terms of the convergence in the drag counts for higher angle of attacks compared to the JST scheme. I am now wondering if to run the ROE scheme to a certain degree of convergence and then start the JST scheme for final convergence. Will the spatial order of accuracy of the ROE scheme be preserved? Do i get a more accurate solution than just running the JST scheme for the beginning?

Regards

Kaush

[FMDenaro]

No, the final accuracy will depend on the scheme you use until to convergence. However, this is in my opinion a false problem for RANS where the magnitude of the model overcomes that of the local truncation error.

[kaush]

I did not quite understand the last statement. Is it true even If I get a convergence of upto 5 decades in the residue and use a different scheme to smoothen? Actually the problem can be viewed as converging to the attractor using different time operators (for example the ROE scheme or JST scheme) so the answer seems surprising for me

[sbaffini]

Also, as soon as you change your scheme, your residuals will suddenly jump up. This may be useful only if you have problems in starting a case and propagating the main flow information with a given scheme. But this might as well indicate a problem in the scheme.

[FMDenaro]

Originally Posted by kaush

I did not quite understand the last statement. Is it true even If I get a convergence of upto 5 decades in the residue and use a different scheme to smoothen? Actually the problem can be viewed as converging to the attractor using different time operators (for example the ROE scheme or JST scheme) so the answer seems surprising for me

The magnitude of the modelling does not depend on convergence... in RANS, even if you would be able to vanish the LTE the solution will strongly depend on the model

[kaush]

Thanks I always thought of this as a way of residual smoothing