CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   A question about discrete adjoint method (http://www.cfd-online.com/Forums/main/11929-question-about-discrete-adjoint-method.html)

Thomas August 3, 2006 12:02

A question about discrete adjoint method
 
For a discrete adjoint method, is the adjoint solver should be similar to the N-S flow solver? May I use the explicit multigrid method for N-S flow solver and implicit method for adjoint solver? Thanks in advance.

unimportant August 3, 2006 16:47

Re: A question about discrete adjoint method
 
Depends on what your goals are. If you are after optimization then yes, it is fine to combine the two provided that everything else is the same, i.e. the same turbulence model, same mesh, etc. Most importantly this is fine if you are looking into steady state problems, just make sure that explicit solver has converged fully, otherwise gradients are inaccurate especially around stationary points of functionals.

If your goals are error analysis, superconvergent computation of various functionals, and similar fun stuff, then no, it is not OK to mix the two formulations. You must have the same error accumulation and propagation in order to correctly account for them. Also, for transient analysis it is important to use exactly the same discretizations going forward and backward since you want to make sure that the system is taking the same path in both directions. If different discretization and time marching techniques are used in two directions then directional derivatives computed at initial time using adjoint solver do not represent correctly the sensitivities with respect to terminal conditions.

Good luck.

Thomas August 4, 2006 22:05

Re: A question about discrete adjoint method
 
Thanks

ztdep August 9, 2006 13:37

Re: A question about discrete adjoint method
 
Hi: What is the "adjoint"


All times are GMT -4. The time now is 11:44.