Houjun:
> Indeed my question is not that clear. The upper panel is actually a
> 'potential' (phi, solving a Poisson's equation), the lower panel is the
> x-component of ExB (E=-\nabla phi, B is out of paper), a derivative quantity.
> This ExB would be used to transport plasma density, (and this plasma density
> field is then feedback to Poisson's equation through rhs and coef.).
>
> The relevant literatures assume different kinds of BCs, e.g., Neumann for both
> potential and plasma density in y (one mentioned extrapolation too) and
> periodic in x. Those simulations usually show a good symmetry in x, but mine
> shows asymmetry quickly near the top boundary. Although we'll be mainly
> interested in what'll happen in the interior, this boundary is a nuisance.
> Don't know how you can enforce \partial rho (density) / \partial y = 0 at the
> top boundary directly in the transport equation, except modifying the flow
> field near the boundary?
I think you already fixed the problem, but as a general rule: The questions
you ask, namely which boundary condition to choose, is not a mathematical one.
It is one of modeling the system you want to simulate. What boundary condition
is appropriate depends on what system you try to describe, not whether the
results of a simulation appear right or wrong, or convenient or not.
Best
W.