 |
|
|
|
|
|
|
|
| |
|
Data
|
|
|
|
|
|
|
|
|
|
|
| |
|
Crystal
|
|
|
|
|
| |
|
|
Interior |
|
Dynamic Atoms |
|
|
| |
|
|
Edges |
|
Dynamic Atoms |
|
|
| |
|
|
Ghosts |
|
Atoms |
|
|
| |
|
|
Equilibrium State |
|
Atoms |
|
|
| |
|
|
|
|
|
|
|
| |
|
|
Potential, KE, Temp |
|
Total over all nodes |
|
|
| |
|
|
Ground Potential, Cut_temp |
|
Total over all nodes |
|
|
| |
|
|
Control |
|
Global control info |
|
|
| |
|
|
Physics |
|
Global physics info |
|
|
| |
|
|
|
|
|
|
|
| |
|
Atoms
|
|
|
|
|
| |
|
|
N, Bufsize |
|
|
|
|
| |
|
|
Position |
|
|
|
|
| |
|
|
Neighbor List |
|
|
|
|
| |
|
|
Species |
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
Dynamic Atoms
|
|
List of atoms that take part in dynamics |
|
|
| |
|
|
n, bufsize |
|
|
|
|
| |
|
|
Position, Prev Pos |
|
|
|
|
| |
|
|
Neighbor List |
|
|
|
|
| |
|
|
Momentum |
|
|
|
|
| |
|
|
Species |
|
|
|
|
| |
|
|
Force, Potential, KE |
|
|
|
|
| |
|
|
Angle? |
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
Box
|
|
List of particles in cubic region of space |
|
|
| |
|
|
|
|
|
|
|
| |
|
Direction
|
|
One of the cardinal directions, relative to the node
decomposition. |
|
|
| |
|
|
|
|
ZMin, YMin, XMin, XMax, YMax, ZMax |
|
|
| |
|
|
dir_sign |
|
-1 for ZMin, YMin, XMin; 1 for ZMax, YMax, XMax. |
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
|
Other
|
|
|
|
|
|
|
|
|
|
|
| |
|
Boxit
|
|
|
|
|
| |
|
|
|
|
Each
Particle belongs to a box.
|
|
|
| |
|
|
|
|
Each
box has size of (maximum cutoff / 2). That means each atom
can possibly only interact with atoms in the 8 surrounding
boxes.
|
|
|
| |
|
|
|
|
Want
outermost boxes to encompass borders...
hmm.
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
Neighborlist
generation is most efficient when boxes are Cutoff/2+eps. I
think stay with this, make box code robust.
|
|
|
| |
|
|
|
|
Cleanest
and easiest to think about when boxes exactly border ROI.
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
If the cutoff-radius
Rc is equal to the grid-cell side length D, the volume of the sphere of non-zero interactions around a given atom is
4/3 pi r^3, whereas the interactions are computed inside a volume
(3D)^3 for local interaction subgrids. The ratio of these volumes, which is an estimate of the algorithmic efficiency of the force computation, is approximately 0.155. If the code uses
5x5x5 local subgrids and a grid size of Rc/2, the algorithmic efficiency is a much better 0.268.
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
Update
neighborlists using skin
|
|
|
| |
|
|
|
|
By
monitoring grid changes? With D=Rc/2, skin of Rc/2?
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
Shared
boxes are particles outside ROI but within 1 box of it. Exchange
synchronizes information between nodes for these boxes
|
|
|
| |
|
|
|
|
Excess
boxes are particles more than one box outside of ROI. Exchange
transfers atoms in these boxes to the neighboring node.
Ignore case where atom should move beyond neighbor (that is,
more than one node at a time).
|
|
|
| |
|
Neighbors
|
|
Constructed
from boxes
|
|
|
| |
|
|
|
|
First
order neighborlist --
|
|
|
| |
|
|
|
|
2nd
order neighborlist
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|