CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Siemens (
-   -   convergence error in transient simulations (

Ossi Kaario November 29, 2000 10:29

convergence error in transient simulations

How to make sure in transient simulations that e.g. momentum, turbulence, energy or scalar equations are sufficiently convergenced !? And what to do if it seems that they are not ?


John C. Chien November 29, 2000 12:28

Re: convergence error in transient simulations
(1). Very good question. (2). The answer is : I don't know. (3). I think, the real solution of transient problem changes from time to time, and problem to problem. (4). If you have analytical solution of an oscillating plate, then, it is possible to know whether the simulation is converged for each time step. But ,in general, it is not easy to find out whether the solution is converged for each time step. (5). At the same time, you could change the number of inner loop iterations, or even the mesh arrangement and density, to see how the solution react to these changes. (6). In the explicit methods, reducing the time step is the only way to improve the accuracy.(converge to real transient solution) (7). In implicit methods, I guess, you could also reduce the time step to improve the convergence and accuracy of the solution. (8).On the other hands, if the mesh is fixed, and the flow field is changing rapidly, then I guess, you will always have some kind of convergence problem.(in addition to the accuracy problem.)

Ossi Kaario November 30, 2000 06:38

Re: convergence error in transient simulations

if one has made grid independency as well as time step independency checks, how should one then charge for the convergence for each equation during each time step ?

Pat Neuman November 30, 2000 09:43

Re: convergence error in transient simulations
Convergence is a very difficult parameter to quantify. The two parameters to watch closely in the file are the PISO corrections and the CORRAUNT number.

The PISO corrections are the number of inner loops that John mentioned. They should be between 3 and 5 for a stable solution. If they are big, the time step should be reduced.

The CORRAUNT number is the number of cells a molecule passes through in a time step. Values from 10 to 100 are reasonable.

If these two numbers are reasonable, you should be fairly confident that each iteration is converging.

All times are GMT -4. The time now is 00:42.