Time discretisation scheme for steady state case
Hi, All!
Tell me, please, what is the time discretisation scheme use for steady state problems in CFX: First Order Backward Euler or Second Order Backward Euler? Thanks :) Andrey |
Re: Time discretisation scheme for steady state ca
What do you think?
|
Re: Time discretisation scheme for steady state ca
It depends on the application, but generally the second order scheme is used.
|
Re: Time discretisation scheme for steady state ca
Mmm... I don't know, were in GUI I can check or change it. How it depends on the application? My usual application is subsonic and transonic flow over airfoils, wings, airplanes, etc.
Thanks… Andrey |
Re: Time discretisation scheme for steady state ca
Hi,
Umm - in steady state flow the transient terms are zero so there is no discretisation. That is why you got a sarcastic reply from Cyclone. CFX uses a psuedo-transient approach to converge to a steady state simulation and that would use a simplified type of first order discretisation. But you should not take any notice of any time related stuff in a steady state flow as the underlying equations do not include all the transient terms. You need to do a transient simulation for that. Glenn Horrocks |
Re: Time discretisation scheme for steady state ca
Hi!
Sorry for my English, guys. Sarcasm is not good. I understand difference in transient and pseudotransient problems. I interested in steady state (pseudotransient) problem. Would you say, that Eqn. 16-20 (p. 282 in CFX-Solver Theory Guide) are not important for steady applications, but Eqn. 7-9 (280) are right for them? So CFX uses for steady problems First order scheme in time (pseudotime :). Is this right? But Mehul, why do you say: «generally the second order scheme is used»? Thank you for your time Andrey |
Re: Time discretisation scheme for steady state ca
CFX uses the first order backwards Euler scheme to implement implicit relaxation. Since time accuracy does not matter:
- no inner iterations are performed, so the flow is not forced to balance within a timestep, only at steady state convergence. - the physical timescale can be set per-equation, so sometimes this is called 'false' timestepping because different equations have different false time values. - the equations are assembled and solved once per iteration/time step. - some under-relaxations for explicit contributions (such as 2nd order corrections for advection) are increased, relative to the true transient values, to help with stability. Make sense? |
Re: Time discretisation scheme for steady state ca
Yes. Thanks a lot :)
Andrey |
All times are GMT -4. The time now is 01:29. |