> On Sep 14, 2016, at 12:49 PM, Zhekai Deng <[email protected]>
> wrote:
>
> So, in my previous one, I only have constraints on left (fixed 0.5 inlet
> concentration), top and bottom ( the boundary conditions that I applied), but
> not boundary condition on the right. I have implemented an additional one on
> the right to account for the outlet:
>
> phi.faceGrad.constrain(velocityVector*phi.faceValue,where = mesh.facesRight)
>
> but it seems does not help very much.
I was surprised to see that several "corrections" I applied to your script
didn't seem to change things very much. This problem is strangely insensitive
to things that I think should matter.
In contrast, moving the definition of eq out of the sweep loop made a huge
difference and I don't understand why.
> I am not sure what does "velocity is backwards someplace" mean. From the way
> I defined it, the $u \hat{i} = \tilde{z}$ and $u
> \hat{j} = 0$. Both are independent of solution variable. Would it be possible
> to clarify what does "velocity is backwards" mean?
This was just idle speculation that maybe a velocity or shear points left when
it should point right (up vs. down), etc. and that maybe that's what causes the
rippling.
> After the discussion with my colleague, we think it is better to understand
> this as a vector term that only act on the normal direction ($\hat{j}$) of
> the flow domain. The reason for this is that the shear rate term is gradient
> of velocity field which is a tensor. However, we are not entire sure the
> functional dependence of segregation velocity on shear rate tensor. That's
> why we usually call it as local shear rate (du/dz \hat{j} ), and I should
> have been clear on this issue in the beginning.
Thanks for clarifying.
> In summary, it does seems that the initial sweeps give the right "trend"
> toward the solution. However, I am still not sure about why it is not
> convergent. Any suggestion on how to proceed to the next step ? I have
> attached the newest version of the code.
I don't know offhand. With a Péclet number of 100, your problem is almost
completely hyperbolic, which FiPy (and cell-centered Finite Volume) isn't very
good at. Daniel knows more about this and may have some suggestions.
You might consider adding a TransientTerm for artificial relaxation and trying
the VanLeerConvectionTerm.
- Jon
_______________________________________________
fipy mailing list
[email protected]
http://www.ctcms.nist.gov/fipy
[ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
