1
Molecular Dynamics of Lennard Jones System using Global Arrays
2
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
4
There are 3 classes of parallelization: Atom, Force and Spatial Decomposition.
5
The current implementation is based on force decomposition and uses global
11
The entire force matrix (N x N) is divided into multiple blocks (m x m) for
12
dynamic load balancing, where m is the block size and N is the total number of
13
atoms/particles. The force between two atoms/particles can be approximated by
14
Lennard Jones potential energy function.
16
Block Topology (of Force Matrix)
17
--------------------------------
19
Say for example if there are 4 block and 100 atoms, the size of the force
20
matrix is 100x100 and each block size is 50x50. (Dynamic Load Balancing) Each
21
process (say 2 process example) gets one block each, and once it is done with
22
a block, it gets the next block in the queue (atomic task array).
24
Using Newton's laws of equation and Velocity-Verlet algorithm, the velocities
25
and coordinates are updated for the next time step.
27
Refer to the following for more information on MD algorithm:
29
http://www.physics.buffalo.edu/phy516/lectures.html
31
An output file output.dat is created in Molden format, when the variable
32
WRITE_TO_FILE is set. For optimum performance, BLOCK SIZE should be set
33
appropriately depending upon the problem size (number of atoms/particles).
37
- http://www.physics.buffalo.edu/phy516/lectures.html
38
- http://www.cs.sandia.gov/~sjplimp/main.html
43
run "make examples" or "make global/examples/lennard-jones/lennard.x".