information in logfile content #542
-
|
Hello there, In my logfile, I noticed that there is a ratio at the end of the projP line: 5/8., as shown below. This ratio goes from 1/8 to 8/8, then starts over all again. Does anyone know what it means? step= 625 t= 3.05544245e+01 dt=5.0e-05 C= 0.50 elapsedStep= 4.24e+00s elapsedStepSum= 2.72936e+03s thanks, |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
In order to reduce the number of iteration, we use the residual from previous timesteps to build the Krylov subspace and project the current solution out of it. This will give you a better initial guess for linear solver provided the solution is smooth in time and the coefficient of the matrix doesn't change. As the solution evolves, such Krylov subspace needs updates in time as well. You can use more vectors by specifying nVector=20, which might reduce your pressure iteration further. nekRS/examples/kershaw/kershaw.par Line 10 in 4f87e0e This will however increase your memory usage so you should also monitor the OCCA device memory printed in the logfile as well. |
Beta Was this translation helpful? Give feedback.
-
|
Great! Thank you! |
Beta Was this translation helpful? Give feedback.
In order to reduce the number of iteration, we use the residual from previous timesteps to build the Krylov subspace and project the current solution out of it. This will give you a better initial guess for linear solver provided the solution is smooth in time and the coefficient of the matrix doesn't change.
As the solution evolves, such Krylov subspace needs updates in time as well.
The current way is to flush the vectors once it's full. Think about it as the restart GMRES that happens every 8 timesteps.
You can use more vectors by specifying nVector=20, which might reduce your pressure iteration further.
nekRS/examples/kershaw/kershaw.par
Line 10 in 4f87e0e