reactive force field & molecular dynamics simulations...
TRANSCRIPT
Reactive Force Field &�Molecular Dynamics Simulations �
(Theory & Applications) Ying Li
Collaboratory for Advanced Computing & Simulations Department of Chemical Engineering & Materials Science
Department of Computer Science
Email: [email protected]
Feb 19th, 2013
Outline
• Theory: Simulation method: Parallel reactive molecular dynamics (MD)
RMD ReaxFF
• Applications: 1. Combustion of aluminum nanostructures: Aluminum
(Al) nanoparticle aggregates & Al nanorods 2. Shock-induced detonation and associated reaction
pathway on high explosives: RDX
Research Background
• Objective: Understanding the combustion mechanism of energetic materials by atomistic simulations
• Challenge: Reactive atomistic simulations • Large scale (multimillion atoms) • Long time (nanosecond)
Reactive Force Fields (RMD)
• Ceramic Al2O3 (Vashishta et al., 2008)
• Al EAM Potential (Voter and Chen, 1987)
• Hybrid Potential for Al-O system
Reactive Force Fields (RMD) • Ceramic Al2O3 (Vashishta et al., 2008)
• Metallic Al (Voter and Chen, 1987)
Validation of Al2O3 Potential Cohesive energy, elastic constants and melting
Liquid structure
P. Vashishta et al., J. Appl. Phys. 103, 083504 (2008)
Validation of EAM Potential
Cohesive energy, lattice constant, bulk modulus and elastic constants
A. F. Voter et al., Mat. Res. Symp. Porc. 82, 175 (1987)
Validation of RMD for AlxOy Comparison of bond length and bond angle in AlxOy fragments
W. Wang, PhD Thesis, USC (2008) P. Politzer et al., J. Phys. Chem. A, 105, 7473-7480 (2001)
QM
Classification of ReaxFF Potential
Bonded
Non-bonded
E = Elp + Eover + Eunder + Ebond + Eval + Epen + Ecoa
+ Etors + Econj + Ehbond + EvdWaals + Ecoulomb
• Coordination, 2-body, 3-body, 4-body, hydrogen bonding, and nonbonding energies
• Reactive bond order potential energy: Ebond({rij},{BOij}) → Bond breakage & formation
Bond Energy
Developed by Goddard, van Duin, et al. (Caltech)
Bond Energy (ReaxFF: Term 4)
Variable Charge Problem (ReaxFF: Term12)
• Charge-equilibration (QEq) → Charge transfer Determine atomic charges {qi | i = 1, ..., N} every MD step to minimize EES(rN,qN) with charge-neutrality constraint: Σi qi = 0
O2 dissociation on Al(111)
ReaxFF: Periodic Table Published ReaxFF Force Fields: - Hydrocarbons (van Duin, Dasgupta, Lorant & Goddard, JPC-A ‘01, 105, 9396; van Duin & Sinninghe
Damste, Org. Geochem. ‘03, 34, 515) - Si/SiO2 (van Duin, Strachan, Stewman, Zhang, Xu & Goddard, JPC-A ‘03, 107, 3803) - Nitramines/RDX/TATP (Strachan, van Duin, Chakraborty, Dasgupta & Goddard, PRL 2003, 91, 09301;
Strachan, van Duin, Kober & Goddard, JCP ‘05, 122, 054502; Han, Strachan, van Duin & Goddard, in preparation; van Duin, Dubnikova, Zeiri, Kosloff & Goddard, submitted to JACS)
- Al/Al2O3 (Zhang, Cagin, van Duin, Goddard, Qi & Hector, PRB ‘04, 69, 045423) - Ni/Cu/Co/C (Nielson, van Duin, Oxgaard, Deng & Goddard, JPC-A ‘05, 109, 493) - Pt/PtH (Jacob, van Duin, Niemer & Goddard, submitted to JPC-A; Chen, Lusk, Kee, van Duin &
Goddard, submitted to JCP) - Mg/MgH (Cheung, Deng, van Duin & Goddard, JPC-A ‘05, 109, 851)
: not currently described by ReaxFF
Validation of Reax-FF
• 1,600 structures & 40 reactions for H, C, N & O by density functional theory (DFT)
• Good agreement with DFT for RDX decomposition pathways
A. Strachen et al., Phys. Rev. Lett. 91, 098301 (2003)
Parallelization • Entire system is divided
into sub-system and each sub-system is assigned to one processor
• Each processor calculates force on its resident atoms
• Neighboring atoms information is copied to calculate the force on the surface
Linked List Cell Normal Traversal: • Search space: Full system • Computation: O(N2)
Linked List Cell • Length of each cell is at least equal to rcut • Search space: Only neighboring cells • Computation: O(N)
Scalability of Parallel ReaxFF MD
Parallel efficiency is 0.957 on 786,432 BG/Q cores for ReaxFF on 8,455,716,864 atoms RDX
N
P
Weak scaling N = 10752 P
Outline
• Theory: Simulation method: Parallel reactive molecular dynamics (MD)
RMD ReaxFF
• Applications: 1. Combustion of aluminum nanostructures: Aluminum
(Al) nanoparticle aggregates & Al nanorods 2. Shock-induced detonation and associated reaction
pathway on high explosives: RDX
Aluminum Nanostructures Experimental Background
Vapor deposited Al-nanorod
C. Li, et al., Chem. Mater. 19, 5812-5814 (2007)
Close packed Al-NP aggregate
Y. Gan et al., Combust. Flame, 158, 354–368 (2011)
Aluminum nanoparticle (Al-NP) core – metallic Al shell – 2~4 nm amorphous Al2O3
Y. A. Kotov et al., Nanotech. in Russia, 4 (2009)
Aluminum combustion reaction
microscale vs. nanoscale
Y. Sun et al., Defense Science J., 4, 56 (2006)
Oxidation of Al-NP Aggregates
• Questions: 1. After oxidation, Al-NP agglomerate or fragment? 2. What’s the size effect on the oxidation of Al-NP aggregates? 3. What are the reaction pathways (or intermediate products)?
In air
With CuO
Y. Ohkura, et al., Combust. Flame, 158, 2544-2548 (2011)
Small fragments
Large agglomerations
Al Nanorods Oxidation
S. K. Cheah et al., Nano Letters 10, 9, 3230-3233 (2009)
• Al nanorods synthesis • Improved oxidation due to larger available surface area
• Questions: 1. What’s the size effect on the oxidation of Al-NRs? 2. What’s the oxidation mechanism for Al-NRs?
RDX
Single Molecule (CH2-N-NO2)3
1,3,5-Trinitroperhydro-1,3,5-triazine
1 unit cell a = 13.182, b = 11.574, c = 10.709Å
a = b = g = 90˚ Z=8 molecules (168 atoms) per unit cell
Space Group 61 Pbca
RDX: cyclotrimethylenetrinitramine “Research Department Explosive”
Applications: civil mining and military defense
Motivation: Shock-induced Detonation
• PBXN 109 shock sensitivity test • Ingredients: 64% RDX, 20% aluminum, etc.
Scientific questions: 1. What is the reaction time? 2. What is the reaction pathway (or intermediate
products)?
Reaction Time
Simulation reveals two-stage reactions in RDX detonation Large C-&O-rich clusters explain the slow release of CO
Fast reaction Slow reaction
Quantum Mechanical Validation QM (VASP)
QM confirms the stability of the large clusters at 1,300 K
ReaxFF MD