[Mira]

Hi,
I am use buoyantboussinesqpimplefoam solver. I found that when the time step changes, the result of numerical calculations changes too. How to understand at what step in time the correct calculation? How to make so that at change of step on time calculations results did not vary? If some one know how to solve the problem it will be very helpful. Thank you.

[piu58]

The time step and the mesh have to check at every simulation. Lower the time step until the changes in your solution are very small. Then it is time step stable.
In the same way you should use finer and finer meshes until the changes are as small as you need.

[Mira]

Thanks for the quick reply! I used the standard tutorial "hot room" and decreased the step from 0.1 to 0.01, 0.001 and 0.0001. I got different results at time step 0.001 and 0.0001. The results at time step 0.1 and 0.01 was the same, But at the time step 1 other results were obtained. I'll try to reduce the size of the grid. Thank you!

[arjun]

PISO solution is time step dependent.

[Mira]

Originally Posted by arjun

PISO solution is time step dependent.

Do you know why?

[FMDenaro]

Generally explicit/implicit FD/FV depend on time step in the sense that the local truncation error is a function of the time and space steps.

[arjun]

Originally Posted by Mira

Do you know why?

major contributor is non linear term in convection. There are other small issues that affect the results to different degrees (for example one might not update the gradient with every corrector to save time while in implicit methods they are computed every iteration)

[SaufiSangi]

Originally Posted by arjun

major contributor is non linear term in convection. There are other small issues that affect the results to different degrees (for example one might not update the gradient with every corrector to save time while in implicit methods they are computed every iteration)

Hi foamers! I study droplets evaporation in OpenFOAM and I also recently noticed that my results are very time step dependent. I can reach time step convergence for Courant numbers like 2, 1, 0.5. I also get nice results. As soon as I decrease the time step of one/two orders of magnitude the results become really strange and totally different. I agree that the problem is the convection term, because if I perform the same simulations suppressing the velocity field (or with a very slow one) the time step dependency does not appear anymore. In literature I never found the time step to be a serious numerical problem (such as the mesh), as long as the code is stable and the Courant number lower than 1.

Do you guys have a satisfying explanation of this? Is it really the PISO method? Can it be due to the numerical diffusion of the convective term, that sums up "more errors" if more time steps are used? I really don't know what to think!


Hope to hear from you guys,

Sangi