CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   Convergence (

Tudi Xiao June 28, 1999 14:03

Hello Everybody,

I am simulating the combustion process in a small furnace with FLUENT 5, the fuel is natural gas, the oxidizer is oxygen (not air), I found beyond 1000 itheration, the computation effort is merely for getting convergence of energy, although it is eventually convergent. My questions are (1). Is this the nature of FLUENT? (2). When the residual is at 10-4E or 10-5E level, reduce underrelaxation parameter to 10-1E or 10-2E surely can accelerate the convergence, but is this reliable?

Like some friends here said, I am using CFD codes aort of "Blindly", thank you people for guiding me out of this.


John C. Chien June 28, 1999 14:44

Re: Convergence
(1). It is really a good question. (2). To obtain a converged solution is itself a black box. So far there is no line-by-line instructions available as how to get a converged solution to a problem. (3). The commercial code is another black box. Some people think that it will save them time by using a black box. (4). For experienced CFD engineers with adequate CFD training and background, it is time saving to use a black box because they don't have to re-invent the wheel ( even if the wheels are not ideal). (5). For those who don't have adequate background and training, using a black box to find a solution in another black box is just like walking on the clouds, you are not sure that you can stay on it all the time. (6). Commercial codes are products of team work. It is hard to know exactly how it actually works. (7). So, the best approach is to use the results as a separate reference, for either design work or analysis. (8). My suggestion is that commercial codes are not good training tools for CFD engineers. By the time one is using commercial code, he should have the background to know what is inside the code. Otherwise, a black box plus a black box is "a black box squared". (9). A well-trained blind person can still move around freely on his own. But,since a black box is always a black box, only the decision to use the code can be a blind one. (10). It is useless to understand a black box.

Dmitriy Makarov June 28, 1999 22:17

Re: Convergence

If this advice will be helpful:

1. May be 1000 iter. is not really big number for natural convection problem.

2. As far as it is natural convection problem, it is better to be solved in the transient mode (if this one exists in your CFD code).

3. In transient solution relaxation usually can be rised.

Best wishes,


Nuray Kayakol June 29, 1999 03:12

Re: Convergence

here some experience from another commercial code.

convergence of energy is obtained after convergence of mass and momentum. Increasing underrelaxation parameters can accelerate convergence at the expense of accuracy. But acceptable solution (below 1% energy imbalance) can be obtained by reducing underrelaxation parameters at the expense of slow convergence rate. Therefore, some scenarios can be tried. Using high underrelaxation parameters at the beginning (say 0.6 or 0.7) and then switching to small one (0.2-0.3) didn't work in one of my case. One can just try the opposite.

One more point. Grid size affects convergence rate. At finer grids the solution may have oscillatory behavior.

Tudi Xiao July 2, 1999 16:33

Re: Convergence
Thank you John, Dmitriy and Nuray for your kind help. This forum is so helpful and also enjoyable, reading discussions and even arguments refreshes my mind, and I feel that all people's personality are really a gift from God. Thanks all people who attend this forum.

All times are GMT -4. The time now is 03:29.