Could anyone please give some basic idea on how to interpret the following : " It takes 14.8 Cray CPU seconds per iteration, or 7x10^-5 ~ 1.0x10^-4 CPU seconds per point and per iteration, and the results are obtained in the order of 1000 to 2000 iterations. The preliminary solution structure emerges after about 1000 iterations with CFL number of O(0.1) and it takes another 1000 iterations to have the solution fully settled." My questions are: If I have a simple cube as the computational domain with say grids like (x y z)=(90 81 35) and side = 10 inches and say there is a 0.5 inch diameter jet at X=5 on the YZ-plne and I want to solve for pressure and velocity on the last face (YZ plane at X=10). For this problem

How is the time per iteration calculated?

How will I know in how many iterations results will be obtained?

How will I know the solution is fully settled?

Kindly give some basic idea on how I would appreciate any references that would put things in perspective.

Thank you all for your time. Sorry about the length of the post. But please please help.

Betty

Time per iteration is normally determined by issuing a system call at the start and end of the iteration loop. cpu_time(), second(), or some other intrinsic function to return elapsed CPU time.

The number of iterations is meaningless as a measure of solution convergence.

One may judge the solution is converged ("settled" as you put it) by examining residual histories, force and moment histories, or other measures of the change in the solution. Residuals should drop several orders of magnitude, ideally to machine zero.

Thank you, that helped.

What is the criteria for setting the number of iterations? Do I pick a random number and then monitor the solution for convergence? Please correct me if I am wrong.

Thanks again Betty

Yes, you have to pick a number of iterations and monitor the convergence.

For example, say I run 1,000 iterations at a CFL number of 1, and I run the same problem for 1,000 iterations at a CFL number of 0.1 - the former case will be much further evolved (larger time steps), but both ran an equal number of iterations.

That is why I say the number of iterations is meaningless: is it problem-dependent and CFL-dependent.

Thanks pc