Chronology | Current Month | Current Thread | Current Date |
[Year List] [Month List (current year)] | [Date Index] [Thread Index] | [Thread Prev] [Thread Next] | [Date Prev] [Date Next] |
where r ranges over the interior of the disk, i.e.
0 <= r <= R . Note the singularly divergent charge
density on the outer edge of the disk.
On Fri, 9 Feb 2001, John S. Denker wrote
At 08:12 AM 2/9/01 -0500, Ludwik Kowalski wrote:
1) Is it appropriate to perform simple averaging for usual
cylindrical cells (dr=const)?
No.
and went on to give the essential theory. To put it into practice
here's what I did (and, I assume, what others are doing as well.)
The key difference is the first derivative term in the cylindrical
Laplacian which involves r itself. This means that we need a
radial coordinate which can be simply generated as an index along
the side of the "universe."
27 ^ (+ r direction)
26 |
25 c
24 d a b -> (+z direction
23 e
22
21 <-(radial index)
The derivatives at the position of cell a are then numerically
implemented as follows:
Laplacian (V) = (c+e-2a) + (c-e)/(2*r_a) + (b+d-2a)
which corresponds respectively to the terms in John D's
... the Laplacian is ...
(d/dr)^2 + (1/r) (d/dr) + (d/dz)^2.
where r_a is the radial index for cell a. Setting the Laplacian =
0 and solving for a we get
a = (b+c+d+e)/4 + (c-e)/(2*r_a)
The only difference from the usual Cartesian implementation is the
last term which becomes progressively less importan (as one should
expect) as r gets bigger.
Once the calculation is finished, finding the Laplacian for those
cells whose value was specified as a part of the boundary
condition gives the charge density (times some appropriate
constant) in those cells.