Dirichlet bc with embed.h

[Louis Steytler]

Hi Everyone,

I would like to use embedded boundaries, and started experimenting with the "karman.c" example. Keeping all else the same, I removed the cylinder and the tracer, and changed the lines:

u.n[embed] = fabs(y) > 0.25 ? neumann(0.) : dirichlet(0.);
u.t[embed] = fabs(y) > 0.25 ? neumann(0.) : dirichlet(0.);

to

u.n[embed] = neumann(0.);
u.t[embed] = neumann(0.);

which correspond to a channel in -0.5 < y < 0.5 with free-slip wall. The expected flow is a uniform velocity in the channel -0.5 < y < 0.5, and that is what I get (see the attached image "neumann.png"). Next, I changed

u.n[embed] = neumann(0.);
u.t[embed] = neumann(0.);

to

u.n[embed] = dirichlet(0.);
u.t[embed] = dirichlet(0.);

In this case all the walls are no-slip. After some development length beyond the inlet, a plane Poiseuille flow is expected in the channel. The computed result is shown in the attached image "dirichlet.png", and surprisingly is the same as in "neumann.png".

I think I might be missing something, why does it seem the dirichlet condition is not enforced with the embedded boundary?

Any advice would be much appreciated.

Thanks very much,

Louis

[Antoon van Hooft]

Hallo Louis,

There seems to be a bug for marginal cases where the boundary lies on cell faces:

the `boundary_flux()` is not assigned to any cell, this is indeed the same as using `neumann(0)`.

You could shift the channel a bit or change the domain size.

Antoon

[Louis Steytler]

Hi Antoon,

That worked.

Thanks very much,

Louis

--
You received this message because you are subscribed to the Google Groups "basilisk-fr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to basilisk-fr...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/basilisk-fr/fb940506-d182-433f-8a4a-0fef7053f6de%40googlegroups.com.