Dishing out the dirt on ReaxFF Force field subgroup meeting 29/9/2003 1 Contents - ReaxFF: general principles and potential functions -All-carbon compounds: Training set - Sample simulation: Ethylene+O2 reactive NVE 2 ReaxFF: general principles and potential functions Hierarchy of computational chemical methods MESO MD ReaxFF QC 10-15 Design FEA Time years Atoms Molecular conformations Electrons Bond formation Grids Grains Empirical force fields Empirical methods: - Allow large systems - Rigid connectivity QC methods: - Allow reactions - Expensive, only small systems Simulate bond formation in larger molecular systems ab initio, DFT, HF Ångstrom Kilometres Distance 3 Current status of ReaxFF program and force fields Program: - 18,000 lines of fortran-77 code; currently being integrated into CMDF - MD-engine (NVT/NVE/limited NPT), MM-engine - Force field optimization methods: single parameter search, anneal - Can handle periodic and non-periodic systems - User manual (under development) available online Force fields Published: hydrocarbons, nitramines, Si/SiO/SiH, Al/AlO Advanced: proteins/CH/CN/CO/NO/NN/NH/OH/OO, MoOx, all-carbon, Mg/MgH In development: SiN/SiC, Pt/PtO/PtN/PtC/PtCo/PtCl, Ni/NiAl/NiC, Co/CoC, Cu/CuC, Zr/ZrO, Y/YO, Ba/BaO, Y-BaZrOH, BH/BB/BN/BC, Fe/FeO Method seems universally available; has been tested now for covalent, ceramic, metallic and ionic materials. 4 Program structure Non-reactive force field Reactive force field 3 1: x1 y1 z1 2: x2 y2 z2 3: x3 y3 z3 4: x4 y4 z4 5: x5 y5 z5 6: x6 y6 z6 4 Atom positions 1: 2 3 4 2: 1 5 6 3: 1 4: 1 5: 2 6: 2 1 2 4 6 Connection table 1: x1 y1 z1 2: x2 y2 z2 3: x3 y3 z3 4: x4 y4 z4 5: x5 y5 z5 6: x6 y6 z6 Atom positions Fixed 3 Bond order 5 BOij f (rij ) 2 1 Determine connections 0 MM or MD routine 1 1.5 2 2.5 Interatomic distance (Angstrom) 3 MM or MD routine 5 ReaxFF: General rules - MD-force field; no discontinuities in energy or forces - User should not have to pre-define reactive sites or reaction pathways; potential functions should be able to automatically handle coordination changes associated with reactions - Each element is represented by only 1 atom type in force field; force field should be able to determine equilibrium bond lengths, valence angles etc. from chemical environment 6 ReaxFFSiO: System energy description Esystem Ebond EvdW aals ECoulomb Eval Etors 2-body 3-body 4-body Eover Eunder multibody E pen Econj ReaxFFCH 7 Bond energy 1. Calculation of bond orders from interatomic distances 3 Bond order (uncorrected) Sigma bond Pi bond 2 Double pi bond Bond order pb o , 2 r ij ' BOij exp pbo,1 Sigma bond ro pb o , 4 rij exp pbo,3 Pi bond ro pb o , 6 rij exp pbo,5 ro 1 Double pi bond 0 1 1.5 2 2.5 3 Interatomic distance (Å) 8 Bond energy 2. Bond order correction for 1-3 bond orders Corrected bond orders Uncorrected bond orders H 0.95 H 1 H H H SBOC=4.16 SBOH=1.17 - Unphysical - Puts strain on angle and overcoordination potentials H H H 0.9 Corrected bond order H S BO 0.8 0.94 0.7 0.6 0.5 H 0.4 H SBOC=3.88 SBOH=0.98 0.3 0.2 0.1 0 0 H - Correction removes unrealistic weak bonds but leaves strong bonds intact - Increases 0.2 0.4computational 0.6 0.8expense 1 as bond orders become Uncorrected bond multibody order interactions 9 3.5 3.7 4.1 4.5 5 6 Bond energy 3. Calculate bond energy from corrected bond orders E bond De BOij exp pbe,1 1 BOij p be,2 De BOij De BOij 3 Bond order (uncorrected) 100 Sigma bond 0 Bond order 2 Bond energy (kcal/mol) Pi bond Double pi bond 1 1 2 2.5 3 -100 -200 Sigma energy Pi energy -300 -400 0 1.5 Double pi energy Total bond energy -500 1 1.5 2 2.5 Interatomic distance (Å) 3 Interatomic distance (Å) 10 Nonbonded interactions - Nonbonded interactions are calculated between every atom pair, including bonded atoms; this avoids having to switch off interactions due to changes in connectivity - To avoid excessive repulsive/attractive nonbonded interactions at short distances both Coulomb and van der Waals interactions are shielded Energy (kcal/mol) 1000 +0.5 +0.5 ECoulomb C 750 qi q j r 1/ 3 ij 3 1/ 3 ij Unshielded Coulomb Shielded Coulomb potential Shielded Coulomb 500 Unshielded vdWaals Shielded vdWaals 250 vdWaals: Shielded Morse potential 0 0 0.5 1 1.5 2 2.5 3 Interatomic distance (Å) 11 Charge calculation method - ReaxFF uses the EEM-method to calculate geometry-dependent, polarizable point charges - 1 point charge for each atom, no separation between electron and nucleus - Long-range Coulomb interactions are handled using a 7th-order polynomal (Taper function), fitted to reproduce continuous energy derivatives. Taper function converges to Ewald sum much faster than simple spline cutoff. NaCl-crystal (33.84x33.84x33.84 Ѓ) Coulomb energy -160000 -170000 Taper 7 Ewald (12.5 Ѓ inner space cutoff) Spline 3 -180000 -190000 -200000 0 5 10 15 20 Outer cutoff 25 30 35 12 Total two-body interaction - Summation of the nonbonded and the bonded interactions gives the two-body interactions - Bond energies overcome van der Waals-repulsions to form stable bonds Energy (kcal/mol) 750 Bond energy Shielded vdWaals energy 500 Total pair energy 250 0 0 1 2 3 -250 -500 Interatomic distance (Å) 13 Valence angle energy 1. General shape a i General shape: j b Modifies equilibrium angle o according to -bond order in bond a and bond b k Eval f ( BOa ) f ( BOb ) f ijk o BOa , BOb Ensures valence angle energy contribution disappears when bond a or bond b dissociates 14 Valence angle energy 2. Bond order/valence angle energy Eval f ( BOa ) f ( BOb ) f ijk o BOa , BOb f ( BOa ) 1 exp 1 BO 2 a a j i b k Eval Eval,max 0 0 0.5 1 1.5 2 Bond order bond a 15 Valence angle energy 3. -Bond order/equilibrium angle Eval f ( BOa ) f ( BOb ) f ijk o BOa , BOb f ( BOa ) 1 exp 1 BOa2 a j b i k neighbours( j ) o ( BOa , BOb ) 180 o,o 1 exp 3 2 BO jn n 1 Equilibrium angle (degrees) 180 160 140 120 100 0 0.5 1 1.5 2 neighbours( j ) BO jn n 1 16 Torsion angle energy 1. General shape a i General shape: j b c l k 1 1 Etors f ( BOa ) f ( BOb ) f ( BOc ) V2 1 cos 2 ijkl f BOb V3 1 cos 3 ijkl 2 2 Ensures torsion angle energy contribution disappears when bond a, b or c dissociates (similar to valence angle) Controls V2-contribution as a function of the -bond order in bond b 17 Torsion angle energy 2. -bond order influence on V2-term a i j b c k l 1 1 Etors f ( BOa ) f ( BOb ) f ( BOc ) V2 1 cos 2 ijkl f BOb V3 1 cos 3 ijkl 2 2 2 f BOb exp 4 1 BOb V2,max V2eff 0 0 0.25 0.5 BOb 0.75 1 18 Overcoordination energy Avoid unrealistically high amounts of bond orders on atoms nbonds nbonds i 1 i 1 Eover f ( BOij ) i i Valencyi 1 1 exp( i ) neighbours BOij nbonds BOi , j (C)=4 BOi , j (C)=5 i 1 Atom energy BOi , j (C)=3 j 1 3 3.5 4 4.5 nbonds BOi , j i 1 19 Computational expense 1000000 Time/iteration (seconds) 100000 10000 x 1000,000 1000 100 QM (DFT) ReaxFF 10 1 0.1 0.01 0 100 200 300 Nr. of atoms 400 20 All-carbon compounds: training set Strategy for parameterizing reactive force fields - Pick an appropriate QC-method - Determine a set of cluster/crystal cases; perform QC - Fit ReaxFF-parameters to QC-data Complications Non-reactive force field ReaxFF Nonbonded Over coordination Nonbonded Valence/ Torsions Bonds Valence/ Torsions Bonds Interatomic distance (Angstroms) Interatomic distance (Angstroms) 21 Binding energies in all-carbon compounds relative to Graphite Relative binding energy (kcal/atom) 120 100 80 60 Reax QC 40 20 0 Diamond C60-buckyball C20-dodeca Hexacyclic C20 Acyclic C20 Bicyclic C17 Cyclic C17 Cyclic C15 Acyclic C14 Tricyclic C13 Cyclic C13 Acyclic C13 Acyclic C12 Tricyclic C10 Cyclic C10 Acyclic C10 Cyclic C9 Acyclic C9 Cyclic C8 C8 cube C8 3ringII C8 3ring Acyclic C8 Cyclic C7 Acyclic C7 Cyclic C6 2_C3 Acyclic C6 Cyclic C5 Acyclic C5 C4 pyramid Cyclic C4 Acyclic C4 Cyclic C3 Acyclic C3 Acyclic C2 - Even-carbon acyclic compounds are more stable in the triplet state; odd-carbon, mono and polycyclic compounds are singlet states - Small acyclic rings have low symmetry ground states (both QC and ReaxFF) - ReaxFF reproduces the relative energies well for the larger (>C6) compounds; bigger deviations (but right trends) for smaller compounds - Also tested for the entire hydrocarbon training set (van Duin et al. JPC-A, 2001); ReaxFF can 22 describe both hydro- and all-carbon compounds Energy (kcal/mol) Bond formation between two C20-dodecahedrons 100 DFT ReaxFF 50 0 Energy (kcal/mol) 1.5 2 2.5 100 DFT ReaxFF 50 0 1.5 2 2.5 C-C distance (Å) - ReaxFF properly describes the coalescence reactions between C20-dodecahedrons 23 Angle bending in C9 - ReaxFF properly describes angle bending, all the way towards the cyclization limit 24 C6+C5 to C11 reaction - ReaxFF properly predicts the dissociation energy but shows a significantly reduced reaction barrier compared to QC 25 3-ring formation in tricyclic C13 -ReaxFF describes the right overall behaviour but deviates for both the barrier height and the relative stabilities of the tetra- and tricyclic compounds 26 Diamond to graphite conversion Calculated by expanding a 144 diamond supercell in the c-direction and relaxing the a- and c axes QC-data: barrier 0.165 eV/atom (LDA-DFT, Fahy et al., PRB 1986, Vol. 34, 1191) E (eV/atom) 0.2 0.15 graphite diamond 0.1 0.05 0 10 15 20 c-axis (Å) -ReaxFF gives a good description of the diamond-to-graphite reaction path 27 Relative stabilities of graphite, diamond, buckyball and nanotubes a: Compound ERef (kcal/atom) EReaxFF Graphite 0.00a 0.00 Diamond 0.8a 0.52 Graphene 1.3a 1.56 10_10 nanotube 2.8b 2.83 17_0 nanotube 2.84b 2.83 12_8 nanotube 2.78b 2.81 16_2 nanotube 2.82b 2.82 C60-buckyball 11.5a 11.3 Experimental data; b: data generated using graphite force field (Guo et al. Nature 1991) - ReaxFF gives a good description of the relative stabilities of these structures 28 Ongoing all-carbon projects - Nanotube failure, buckyball collision (Claudio) - Si-tip/nanotube interactions (Santiago) - Nanotube growth, buckyball polymerization (Weiqiao) - Buckyball/nanotube nucleation (Kevin) - Buckyball/nanotube oscillator (Haibin) - Diamond surface interactions (Sue Melnik) 29 Sample simulation: Ethylene+O2 reactive NVE -12 Ethylene, 36 O2 - Pre-equilibrated at 4000K. Switched off C-O and H-O bonds during equilibration to avoid reactions - Time-step: 0.025 fs.; cannot go much higher due to high temperature + reactive potential - Should react; main expected products H2O, CO2 and CO 30 QuickTime™ and a GIF decompressor are needed to see this picture. 31 - Fast reaction after initiation - Exothermic; temperature rises to 7000K - Energy is not perfectly conserved at elevated temperatures. - Future work: investigate potential; see if energy conservation can be improved. MD-iteration 32 - ReaxFF gets pretty reasonable product distribution; probably slightly too much CO; may need to check CO+0.5O2 to CO2 reaction energy 33