I got an error stating that the residual error was grater than my relative tolerence. The residual error was 2.2e+3 if I remember correctly.
Can I adjust the relative tolerence in my frequency domain and time dependent study by adjusting the value of the tolerence factor?
What is the difference between a tolerance factor of 1 and 1e-6? I still do not know understand what the value tells me about the tolerence factor.
Whats the effect of increasing relative tolerence on the accuracy of the result?
Thank you Liang.
Lets say my relative tolerance is 1.
And I got an error from the simulation stating that the "residual error is greater than the relative tolerance", does that mean that my solution will never converge? Because my relative tolerance is already 1.
At that I was using an extremely fine mesh and with a small time step in my time dependent study.
If you set the relative tolerance to 1, this is unacceptable. You should use a value less than 1 for your simulation(usually it should be less than 0.3, depends on your problems).
For the convergence rate, you can get more fast convergence with increase of relative tolerance. But this will make you loss some accuracy.
For the mesh and time step size, as I already mentioned, the relative tolerance is related to convergence criteria. So, if you get wrong or inappropriate values of tolerance, any adjustments of mesh and time steps are useless.
Thank you Liang Cui for your replies. I am really interested in this since I am trying to converge my simulation.
Here are my questions:
What is the maximum relative tolerance I can set to get at least a good result? (i.e. when I am increasing the relative tolerance)
And what does it mean if I have a relative error which is greater then my relative tolerance?
It is up to your own simulation problem. Generally, it should be less than 0.3. Maybe you can increase the value, but you should check the results very carefully if you have a high relative tolerance.
If the relative error is greater than relative tolerance, the convergence criterion is not satisfied, namely, you can't get convergence.
I used a relative tolerance of 0.2, that should be good. Do you have any references that I could see where it states that the relative tolerance should be less than 0.3? I would like to write it in my thesis and make a reference on it.
@Chao Sui
I only changed the relative tolerance in my time dependant study but not the frequency domain study. The freq. domain worked fine with its the default tolerancd
Thank you for raising the question and for the answers. I would like to ask a question related to relative tolerance. What does setting this relative tolerance to value of 0.001 mean? Does it mean that at each and every grid point, all the variables that are being solved for have to satisfy this relative tolerance as the solver is carrying out iterations to reach a converged solution?
Hi,
I have a question. I am getting error of 0.008 and so my solution is not converged as default setting is 0.001 as infered from the previous discussion.
I am doing stationary study so I couldn't find the settinf to change relative tolerance. Could anyone pls tell me how I can increase relative tolerance in stationary study? What is the relation between tolerance factor and relative tolerance? Thank you
Relative tolerance setting for stationary study can be found within the stationary solver1. there the default value is 0.001 and you can modify as to a value for your simulation.
- Badges
- Science topic