CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   convergence criterion (

Luca June 16, 2000 09:00

convergence criterion
I am a new phoenics user, can anybody tell me what is the GLOBAL CONVERGENCE CRITERION (default value 10e-4) that is in the NUMERICS menu? I think it is something linked to the residuals but I don't know how it is calculated, does anybody know?

John C. Chien June 16, 2000 20:06

Re: convergence criterion
(1). Don't try to understand the definition. There is nothing you can do there. (2). When I run a commercial code, I always set these numbers (you are likely to have one for each equation being solved) to 1.0E-08 that is 0.00000001 Then you can set it and forget about it. (2). Use the total number of iterations to control the convergence. Check the variables of interest to you at several locations of interest, and plot the convergence history. (3). The flow convergence (steady state) is non-uniform, at some locations it is fast and at some other locations it is very solw. To reach convergence, you must be able to see flat part of the history plots of the variables vs iteration numbers. (4). Let the computer do the number crunching hard work, and make sure that you have the converged solution. (mesh size and quality always has a great impact on the convergence of the solution) Do not accept the solution unless it is the fully converged solution.

Duane June 18, 2000 02:43

Re: convergence criterion
Hi Luca,

global convergence USUALLY refers to a volume integration over the entire domain. So for mass conservation it is the sum of all mass flow rates at the boundaries (and source terms) normalized by the inlet mass flow. So 10e-4 is an overall ballance within 0.01%. This is an indication of how well the nonlinear discrete equations have been solved NOT necessarily of the accuracy of the discrete equations in approximating the solution. 1e-4 is typically for industrial work. It is also important to look at the local residuals how well the individual control volume equations have been solved by the nonlinear iterative process.

Good Luck..................Duane

All times are GMT -4. The time now is 13:39.