CFD Online Discussion Forums

CFD Online Discussion Forums (
-   FLUENT (
-   -   convergence problems (

jeremy May 29, 2002 05:38

convergence problems
I have problems of convergence. the residuals reach a stable value, but the comtinuity doesn't converge under 10-3 (i.e. a value of roughly 3.10-3).

decreasing the underrelaxation factors doesn't change very much. The grid is OK, I thin.

How to improve this convergence? Can I assume that the model is converged?

thank you in advance.


An Modh Coinniolach May 29, 2002 06:10

Re: convergence problems
How close is your initial guess to the final answer?

Gavin May 29, 2002 06:12

Re: convergence problems
Convergence is difficult to define. Some people set a criteria for residuals of continuity say of 10e-03. Others wait for the residyuals to bottom out, i.e the solution is no longer changing. Firstly, do you need second accuracy? Is your grid tets or quads? If it is tets you will need 2nd order accuracy (re: numerical diffusion). Secondly, what i recommend, is to use a surface monitor at some points where the flow is not dominated by one component of flow (i.e do not choose a point right inside an inlet), then monitor the values untill they converge to a constant value. Note that this is only relevant for a steady state flowfield. What are your flow conditions, what is the problem etc? Have you supplied an initial condition, if so this may be close to the actual solution, and the flow field approximation did not need to change by 3 orders of magnitude in order to reach a converged solution (i.e it was almost there in the first place!).

Remember, that to define a solution as converged, you really show have an idea of what the flow should be like, and if possible monitor the progress of certain variables towards known values, either obtained experimentally or an educated guess. But all this depends on the flowfield. Perhaps you would like to elaborate on this and i may be able to help.

jeremy May 29, 2002 06:38

Re: convergence problems
i am a beginner in fluent. How can I see the initaila guess?

jeremy May 29, 2002 06:52

Re: convergence problems
I use quads, and the flow follows the mesh, so I don't need 2nd order.

my model is a rectangular domain, with on top a very little rectangular narrowing. It is almost a flow between to plates (it should be simple!!) on the left: velocity inlet on the right : outflow bottom: wall upper part : pressure inlet :101325 Pa rectangular narrowing: wall

what do you mean by initial condtion? the initialisation?

An Modh Coinniolach May 29, 2002 10:23

Re: convergence problems
Maybe your residuals will not drop much for this type of problem. If you are initialising the flow with an initial guess that is quite close to the final answer, the residual may not drop through three orders of magnitude.

Like the last contributor said, you should monitor the convergence of some other quantities, say mass flow rate and pressure at inlet and outlet. I am looking at turbines and I monitor mass flow rate, Cl on the blade and pressure at inlet and outlet. Also these are my main output quantities, so I need to know that they are converged. I find that it generally takes about 10^-4 for madd flow rate, etc. to converge for my situation.

Once you can show that these other quanties are converged, you are in a much better position.

And now the perennial question, do you have experimental data?

jeremy May 29, 2002 11:13

Re: convergence problems
no, I have no experimental data...

so i have to monitor mass flow rate and pressure at inlet and outlet. I would then know when it converges when a stable value is reached?

thank you for your answers.

Gavin May 30, 2002 06:41

Re: convergence problems
Jeremy, firstly, if your flow is compressible (Vin <100m/s approx), go to report->fluxes (then select your inlet and outlet) -> calculate. Check to see if this value is less than 1% of your inlet mass flow. If it is it may indicate convergence of you solution. Have you monitored velocity at a point like i suggested? If so, how is it going? You say you do not need second order accuracy, how many cells have you, is flow 3d,are boundary layers important? Is flow turbulent? What model are you using i.e. k-e, etc.

All times are GMT -4. The time now is 23:33.