Timestep size and Convergence
Does time-step size affect the convergence? Like, if a del t is beyond some limit, will it diverge?
If so, How to calculate the optimum time-step size for an unsteady flow problem? |
The CFL condition should get you your answer about the stability of your solution and the time step to use
|
Sorry, I am working with ANSYS, where Courant number can't be found before going about the solution. Also, My discretization schemes are Second-order implicit. Hence, Will Courant number affect?
And, Thank you so much for your reply, sir. |
You can actually specify the courant number in fluent with the density based solver
|
Quote:
The time step enters into the accuracy and stability of a numerical scheme. If a scheme is consistent and stable, you have convergence. Explicit schemes are subject to stability constraints that depend both on the convective part (CFL) and the diffusive part of the equations. Implicit schemes are generally unconditionally stable and the time step affects the accuracy. What kind of flow problem are you simulating? |
Yes sir, I am actually working on pressure based solver. 2D Laminar flow over bluff-bodies.
|
2D laminar flow over multiple bluff bodies.
|
what is exactly your problem? The solution is always diverging or only for a threshold in the value of dt?
|
Mine, Flow over multiple square cylinders at low Reynolds number cases.
For del t 0.01, I have run it for more than 20000 timesteps. The solution has not converged. So, I tried to run at del t 0.1, But even then the problems dont converge if sbyD i.e., the distance between cylinders/D is 3 or lesser than that. I have tried to give for that del t 0.1 and some 30000 timesteps. Yet I find the residuals actually diverge after 20000 time steps. And It never look like converging. Hence, My doubt. the del t and convergence has any relation? |
But do you mean convergence towards a steady state? And how you are sure that your flow problem has a physical steady state at that Re number?
A single bluff body has onset of unsteady vortex shedding for Re as low as 40-50 |
Sir, I am working at Re 100.
Yes. Convergence towards a steady state. And I am trying to validate a simulation and experimental work already published. Hence, I think it should give proper vortex shedding at that Re. This is the residuals for del t = 0.1 for Re 100 flow over two cylinders separated by 3D. It actually start diverging lately. That I don't know whether it will converge eventually or not (i.e., vortex will start shedding steadily). https://www.dropbox.com/s/ow6ul3wikg...duals.png?dl=0 |
Diverge for further iterations at this time or you see divergence for longer time? Your flow will develop a fully unsteady solution
|
For a longer time.
It looks like the vortex never sheds also. I have attached the vorticity contours also at the end of 2300 s. https://www.dropbox.com/s/p4bc7krs50...02cyl.jpg?dl=0 |
the flow appears no yet fully developed... it could be due to an excess of numerical viscosity, what about the spatial discretiazion?
|
Spacial Discretization:
Gradient - Least Square Cell based Pressure - Second order Momentum - Second order upwind |
Quote:
Shift your mesh in any direction by 1.0E-10 m and try again. |
Quote:
|
Quote:
I suppose that he wants you to "destroy" the grid symmetry with respect to the body. However, how about the cell Re number? It should be O(1) everywhere to ensure a very low numerical diffusion. What about the inflow profile? |
Quote:
The mesh is actually symmetrical only about X axis. Why does it impact? |
Quote:
The cell Re number seems adequate, what about the inflow velocity? |
All times are GMT -4. The time now is 03:13. |