1. Comparison to LAMMPS-REAX
PuReMD: Purdue Reactive Molecular Dynamics Program H. Aktulga§, J.Fogarty¶, S. Pandit¶, A. Grama§ § Purdue University, West Lafayette, IN 47907, USA ¶University of South Florida, Tampa, FL 33620, USA
1. ReaxFF Overview
Reax Force Field
Works at the atomistic level
Dynamic bonding in ReaxFF vs Static bonding in other classical methods
At the heart of dynamic bonding lies a classical bond order interaction
Bond strength is determined by atom types and distance
Dynamic bonding makes the simulation of reactive systems possible
Dynamic bonding also introduces many challenges:
Updating bond, valence angle, and dihedral angle lists at each step
Updating the partial charge on each atom at each step  QEq
Shorter time-steps need to be used due to chemical reactions
2. Implementation
System
Geometry
Control
Parameters
Force Field
Trajectory
Snapshots
System Status
Update
Program Log
File
PurdueReax
Initialization
Read input data
Initialize data-structs
Compute Bonds
Corrections are applied
after all uncorrected bonds
are computed
Bonded Interactions
Bonds
Lone pairs
Over/UnderCoord
Valence Angles
Hydrogen Bonds
Torsion Angles
3-body & 4-body intrs are
Built from lower order intrs
QEq
Huge sparse linear system
PGMRES(50) or PCG
ILUT-based preconditioners
give excellent performance
vd Waals & electrostatics
Single pass over the
far nbr-list after
charges are updated
Interpolation with cubic
splines for nice speed-up
Evolve the System
Ftotal = Fnonbonded + Fbonded
Update x & v with velocity Verlet
NVE, NVT and NPT ensembles
Reallocate
Fully dynamic and adaptive memory management:
excellent use of resources
can handle very large systems on a single processor
Neighbor Generation
3D-grid based O(n) neighbor generation
Several optimizations for best performance
Init Forces
Initialize the QEq coef matrix
Compute uncorrected bond orders
Generate H-bond lists
3. Performance
Weak Scaling: Bulk Water – 6540 atoms in 40x40x40A3 box / core
Strong Scaling: Bulk Water – atoms in 80x80x80A3 box
4. Applications
Y. Park, H. Aktulga, A. Grama, A. Strachan
“Strain relaxation in Si/Ge/Si nanoscale bars from MD simulations”
J Appl Phys 106, (2009)
J. Fogarty, H. Aktulga, A. van Duin,
A. Grama, S. Pandit
“A Reactive Simulation of the Silica-Water Interface”
J Chem Phys (2010)
Comparison to LAMMPS-REAX
weak scaling
strong scaling
J. Fogarty, H. Aktulga, A. Grama, S. Pandit
“Oxidative Damage in Lipid Bilayers:
A Reactive Molecular Dynamics Study”
Biophysical Society Meeting (2010)
5. Status
Code (PuReMD) fully validated and released to a number of applications groups -- Goddard (CalTech), van Duin (PSU), Pandit (USF), Buehler (MIT), Thompson (Sandia), Germann (LANL), Kent (ORNL), Andzelm (ARL), Quenneville (Spectral Sc. Inc).
Code fully integrated into LAMMPS (currently being fully validated).
Optimizations for various platforms and improvement of algorithms ongoing.