I would like to know about Saturne's capacity to simulate inviscid flows, i.e. Euler flow. I guess this can be made by setting viscosity to zero. I also realize I could try this myself, but my doubt is if such a method can be accepted as a viable "poor man's Euler solver", both in speed and quality. The idea is to be able to solve an inviscid flow first, which should be faster than for nonzero viscosity, and then use the results as a starting flow field for viscous runs. I'm also open to hear alternative methods, including code tweaking (but I'm no expert here though).
usoPreviewPopup
Simulating zero viscosity flows
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
-
Yvan Fournier
Re: Simulating zero viscosity flows
Hello,
This idea for flow initialization has been mentioned in a developer meeting a few months ago, but no work has been started on that aspect to my knowledge. You could set viscosity to zero and consider you have laminar flow, and with the default centered scheme, numerical viscosity would be quasi-zero, but I am not sure in that case if the resolution algorithm would be robust, compared to algorithms designed for Euler flows (I'll let colleagues more into the core numerics of the code answer that).
Currently, you might have better chances of faster initialization setting a somewhat realistic (and non-zero velocity) initial velocity (though the incompressibility condition may be harder to ensure, so initialization may be faster, but it may also diverge).
Also, are you using a steady algorithm ? This is usually much faster, so even for an unsteady (transient) calculation, reaching the initial state with a steady algorithm and the restarting the code using an unsteady algorithm would probably be a good idea.
Finally, reducing default precision of linear solvers from 10 -8 to 10 -5 usually seems quite OK, and will speed up the caclulation.
Best regards,
Yvan
This idea for flow initialization has been mentioned in a developer meeting a few months ago, but no work has been started on that aspect to my knowledge. You could set viscosity to zero and consider you have laminar flow, and with the default centered scheme, numerical viscosity would be quasi-zero, but I am not sure in that case if the resolution algorithm would be robust, compared to algorithms designed for Euler flows (I'll let colleagues more into the core numerics of the code answer that).
Currently, you might have better chances of faster initialization setting a somewhat realistic (and non-zero velocity) initial velocity (though the incompressibility condition may be harder to ensure, so initialization may be faster, but it may also diverge).
Also, are you using a steady algorithm ? This is usually much faster, so even for an unsteady (transient) calculation, reaching the initial state with a steady algorithm and the restarting the code using an unsteady algorithm would probably be a good idea.
Finally, reducing default precision of linear solvers from 10 -8 to 10 -5 usually seems quite OK, and will speed up the caclulation.
Best regards,
Yvan