Conversation with Prof. Namas Chandra October 5 2000

Works with ARL researchers in Materials Lab (a few hundred people)
Does both Finite element "macroscopic" and molecular dynamic "microscopic" simulations

Different Programs are used for different problems -- Dyna could be used to study
impact of "stone" on a vehicle; "a different" program for the penetration of an
armor-piercing projectile.

One needs to build in an environment (Portal or Problem Solving Environment PSE) that allows
user to
a) Use single geometry specification for multiple simulations
b) Choose either automatically or with some good user interface, the "correct" simulation
   package
c) Optimize for particular goals. This involving running jobs with multiple parameter sets
   (possibly in parallel). Also one needs to find ways of automatically recognizing "success"
    or "quality" of a given simulation. This is one aspect of datamining. This "quality" is then
    used as objective function for optimization.

This PSE would allow simulation to be run on multiple back-ends either by searching for free-est machine
or by user choice.
Note tend to use non-commercial codes i.e. "free" version of Dyna not commercial one
Raju Namburu (CSM CTA lead) has parallel version of Dyna (Check this)

We discussed parallelization of the molecular dynamics code. There are two cases
1) Statics. Smaller computing job. Particles move relatively small distances (one interatomic
   spacing) during run. Goal is to find equilibrium
2) Dynamics. Time consuming and much larger movement of atoms during a run.
   Notes on Algorithm for this case:
   		Maybe 7,000 lines of Fortran
   		Cut-off forces (3 atomic distances)
   		10,000 particles
   		We estimated that this could parallelize effectively but not be able to  use more than
   		10-20 processors as otherwise surface/volume effects become too large
\