Toward a multilevel QM/MM methodology for performing molecular dynamic simulations of complex reactive processes

Size: px

Start display at page:

Download "Toward a multilevel QM/MM methodology for performing molecular dynamic simulations of complex reactive processes"

Christina Whitehead
5 years ago
Views:

1 Toward a multilevel QM/MM methodology for performing molecular dynamic simulations of complex reactive processes Michael R. Salazar Department of Chemistry Union University, Jackson, TN msalazar@uu.edu (731) Material and Process Simulation Center California Institute of Technology January 22,

2 Outline Context and motivation Overview of complex reactions Challenges/Opportunities of complex chemical reactions Time-dependent group methodology Accelerated MD with Chemistry (AMolDC) outline Illustrative simulations Computational scaling Summary 2

Context/Motivation of Research How can we take first principles (ab initio) methods for the molecular dynamics (MD) simulations of: A + BC BC' + A' AB + C 3 atoms AC + B 6 species A + B + C 5

3 Context/Motivation of Research How can we take first principles (ab initio) methods for the molecular dynamics (MD) simulations of: A + BC BC' + A' AB + C 3 atoms AC + B 6 species A + B + C 5 channels ABC and extend it to: C 2 H 4 + O 2 products or HMX(β) HMX(δ) HMX(δ) HMX(NO 2 ) HMX(NO 2 ) N 2 O + NO + CO + (intermediate products) N 2 O + NO + CO N 2 + CO 2 + (final gas products) 3

4 Overview of Complex Reactions soot pyrolysis C 2 H 4 + O 2 C 2 H 4 + 2O. C 2 H. 3 + OH + O C 2 H 4 time ~10 2 elementary steps O 2 CO + H 2 O CO 2 + H 2 O ns - µs oxidation exponential growth of species with steps: OH, C 2 H 3, C 2 H 3 OO, H 3 CCH 2, ~10 3 unique chemical species Combustion and detonation processes Oxidation and pyrolysis of fuels HMX/RDX decomposition 4

Current Theory (Kinetics) postulated kinetic mechanism fit rate constants or experiments of elementary reactions or ab initio MD or transition state theory Numerical Approaches to Combustion Modeling

5 Current Theory (Kinetics) postulated kinetic mechanism fit rate constants or experiments of elementary reactions or ab initio MD or transition state theory Numerical Approaches to Combustion Modeling REACTIONS!! H2,O2, H,OH and O reactions! H2+OH=H2O+H 2.14E ! [Emdee et al. 1992] H2+O2=OH+OH 1.70E ! [Miller and Melius 1992] H+O2=OH+O 1.91E ! [Emdee et al. 1992] H+O2+M=HO2+M 1.41E ! [Baulch et al. 1994] for N2 H2/1.25/ H2O/6.0/ CO2/1.90/! relative to N2, based on Baulch et al H+M=H2+M 1.00E ! [Miller and Melius 1992] 2H+H2=2H2 9.20E ! [Miller and Melius 1992] 2H+H2O=H2+H2O 6.00E ! [Miller and Melius 1992] 2H+CO2=H2+CO2 5.49E ! [Miller and Melius 1992] H+OH+M=H2O+M 2.21E ! [Baulch et al. 1992] for N2 H2/1.25/ H2O/6.0/ CO2/1.90/! relative to N2, based on Baulch et al H+O+M=OH+M 6.02E ! [Miller and Melius 1992] H2O/5.0/ OH+H2O2=H2O+HO2 7.08E ! [Emdee et al. 1992] OH+OH=O+H2O 1.23E ! [Emdee et al. 1992] O+HO2=OH+O2 1.74E ! [Emdee et al. 1992] O+H2=OH+H 5.13E ! [Emdee et al. 1992] O+O+M=O2+M 5.04E ! [Tsang and Hampson 1986] corrected for N2 H2/1.25/ H2O/6.0/ CO2/1.90/! relative to N2, based on Baulch et al O+OH+M=HO2+M 1.00E ! [Zhang and McKinnon 1995]!!! HO2 peroxyl reactions! HO2+H=H2O+O 3.00E ! [Baulch et al. 1992] HO2+H=H2+O2 6.61E ! [Emdee et al. 1992] HO2+H=OH+OH 1.40E ! [Miller and Melius 1992] HO2+OH=H2O+O2 7.50E ! [Miller and Melius 1992] HO2+HO2=H2O2+O2 2.00E ! [Miller and Melius 1992]!!! H2O2 reactions! H2O2+M=OH+OH+M 1.21E ! [Baulch et al. 1992] for N2 H2/1.25/ H2O/6.0/ CO2/1.90/! relative to N2, based on Baulch et al H2O2+H=HO2+H2 4.79E ! [Emdee et al. 1992] H2O2+H=OH+H2O 1.00E ! [Emdee et al. 1992] H2O2+O=OH+HO2 9.55E ! [Emdee et al. 1992] H2O2+O=O2+H2O 9.55E ! [Emdee et al. 1992]!!! HCO (aldehyde) reactions 5

6 Current Theory (Dynamics) Classical MD simulations with ReaxFF for potential RDX decomposition products RDX/Estane decomposition 6

7 Challenges/Opportunities of MD Studies Challenges for ab initio MD studies: 1. System Size ~10 3 unique species NAtoms ~ PESs Reactive and non-reactive PESs Ee large dimensionality x 3. Time Scale i ns µs Opportunities: 1. Mechanism Species Branching ratios 2. Kinetics Dynamics determining kinetics rather than kinetics determining dynamics 3. Energetics = tr Bath gas collisional processes O 2 O 2, CO CO 2, Reactive processes C 2 H 3 + O 2 C 2 H 3 O 2, { P [ 2 T + 2Vn, e + J + K ] 2S PFP} + Vn, n Computational scaling factors: O(N) to O(N 7 ) N evaluations ~ for 100 ps 7

8 MD by Time Dependent Groups O 2 H 2 O V Total group Ni group N group Nm 1 l ( t) = Viα ( t) + Vlα, mβ ( t) α 2 α β i l m l C 2 H 3 CO Group interactions (QM/QM/QM ) Governed by spatial cutoffs Differing levels of ab initio theory over differing groups Group-group interactions (MM) Outside of spatial cutoffs Less important to reactive systems Fast-access PES database Interpolation force fields low level ab initio Salazar, M.R. J. Phys. Chem. A 2005, 109,

9 Accelerated MD with Chemistry (AMolDC) Input initial coordinates, T, and P MakeGroups Sum over groups Direct dynamics perform MD start Make groups in the simulation cell Have all atomic forces been calculated? Is there sufficient data to interpolate? Perform quantum chemical calculation limit reached? end interpolate Insert data into PES database PESDatabase 350 Link-listed subcells Interpolation group N i 1 group Nl group N m VTotal ( t) = Viα ( t) + Vlα, mβ ( t) i α 2 l α m l β Wall Clock Simulation time (ps) 9

10 Calculate Forces perform MD start Make groups in the simulation cell Have all atomic forces been calculated? Is there sufficient data to interpolate? Perform quantum chemical calculation limit reached? end interpolate Insert data into PES database 10

11 Discontinuities of V Total O 2 HF MP2 O 2 C 2 H 3 C 2 H 3 at spatial cutoff MCSCF O O V(C 2 H 3 ) + V(O 2 ) V(C 2 H 3 + O 2 ), but V 0,therefore, T 0 Shuffling of groups H. C C H H V Total Time 11

12 Illustrative Simulations N 2 system Reflective walls Variable simulation cell size T=1000K P~10 3 atm Variable spatial cutoff (ε) N Atoms = 12, P~ atm V dv/dr SpatialCutoff = 8.5 bohr SpatialCutoff = 6.9 bohr V T Total Energy V T Total Energy V,dV/dr (a.u.) ε = 6.9 ε = bohr 8.5 bohr Energy (Hartree) Energy (Hartree) r (bohr) time time (a.u.) (a.u.) 12

13 Illustrative Simulations (cont.) N 2 system Reflective walls Variable simulation cell size T=1000K P~10 3 atm Variable spatial cutoff (ε) N Atoms = 300, P~ atm V dv/dr 6 5 Spatial Cutoff = 6.9 bohr V T Total Energy V,dV/dr (a.u.) ε = 6.9 bohr Energy (Hartree) r (bohr) time (a.u.) 13

14 Canonical (NVT) Simulations 220 Atom simulation P ~ 1000 atm T = 1000 K Time step = 0.25 fs Energy (Hartree) V T E_total time (a.u.) T (Hartree) time (a.u.) Focus on thermalized region 14

15 Canonical (NVT) Simulations V E_total Energy (Hartree) T(Hartree) time (a.u.) time (a.u.) Temp (K) time (a.u.) Exceedingly discontinuous potential and total energy; however, smooth and continuous kinetic energy, smooth and continuous temperature, and, therefore, canonical (NVT) simulations. 15

16 MakeGroups perform MD start Make groups in the simulation cell Have all atomic forces been calculated? Is there sufficient data to interpolate? Perform quantum chemical calculation limit reached? end interpolate Insert data into PES database 16

17 Computational Scaling of MakeGroups Module Cost of MakeGroups for 3000 atoms = 440 sec. for 1000 MD steps MakeGroups(NSubCells,NGroups, Groups,Members) CPU time (seconds) O(N) O(N 2 ) Order of magnitude reduction of cost no MakeGroups 1 subcell 27 subcells 64 subcells 125 subcells 1000 subcells System Size (NAtoms) 17

18 PESDatabase perform MD start Make groups in the simulation cell Have all atomic forces been calculated? Is there sufficient data to interpolate? Perform quantum chemical calculation limit reached? end interpolate Insert data into PES database 18

19 PESDatabase Module perform MD start Make groups in the simulation cell Have all atomic forces been calculated? Is there sufficient data to interpolate? Perform quantum chemical calculation limit reached? end interpolate Insert data into PES database PESDatabase for Complex Systems Loop over subcells { Loop over Groups in subcells { r q P; Group = C H ) } } ( 2 3O A B C D E Groups Interpolation Module Fast Sorting Routine r r q, E, E, r),( q, E, E, ),... ( 1 r 2 q r, E, E Grid 19

20 perform MD start Make groups in the simulation cell Have all atomic forces been calculated? Is there sufficient data to interpolate? Perform quantum chemical calculation limit reached? end interpolate Insert data into PES database Interpolation 20

21 Interpolation Methodology Challenges diverse interactions o collisional processes o reactive processes diverse chemical species o free radicals o excited electronic states o closed shell species diverse PES topologies o flat asymptotic region o barriers, TS regions diverse grid spacings Solution: Completely general adaptive interpolation methodology 21

22 Fig. 2, M. Salazar Local, Optimized Interpolants Local vs. global interpolation 1 2 [ d ] 2 ip r N E( p) = ci + i= 1 speed vs. accuracy 2 d11 + r 2 d N 1 + r 2.. d1n + r c 1 E1.. = d + NN r cn EN Optimization of local interpolants y, bohr x, bohr M.R. Salazar, Chem. Phys. Lett., 359, 460 (2002). log (rms error (a.u.)) dv_{3b}/ds_{1} dv_{3b}/ds_{2} dv_{3b}/ds_{3} Delta F. Colavecchia, W.J. Stevens, J.P. Burke, M.R. Salazar, G.A. Parker, and R.T Pack, J Chem. Phys., 118, 5484 (2003). Error (a.u.) Density (#/bohr^3) Highly repulsive Al + + H 2 22

23 ReaxFF with AMolDC Need a smooth FF that can facilitate reactive events Link ReaxFF with PESDatabase and interpolation Investigations: 1. Accuracy of interpolants as function of: Dimensionality Grid density 2. Computational timings of: PESDatabase Interpolate perform MD start Make groups in the simulation cell Have all atomic forces been calculated? Is there sufficient data to interpolate? Perform quantum chemical calculation ReaxFF limit reached? end interpolate Insert data into PES database 23

24 Summary 1. Reactive PESs Varying levels of theory that change over the groups (QM/QM/QM /MM) Reactive higher levels group Ni group Nl group Nm 1 Collisional lower levels VTotal ( t) = Viα ( t) + Vlα, mβ ( t) 2 2. System Size Assembly of total potential by time-dependent groups MakeGroups Linked-listed subcell division PESDatabase i α l α m l β 3. Time Scale PESDatabase Interpolate Multidimensional, local, optimized interpolants Classical MD Interpolated Car-Parrinello system size in N 2 molecules 24

Toward a multilevel QM/MM methodology for performing molecular dynamic simulations of complex reactive processes

Toward a multilevel QM/MM methodology for performing molecular dynamic simulations of complex reactive processes Michael R. Salazar Department of Chemistry Union University, Jackson, TN 38305 msalazar@uu.edu