New Developments in ADF/ReaxFF
Ole Carstensen
Scientific Computing & Modelling [email protected]
Manchester, 14 September 2015
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Amsterdam Density Functional (ADF) development
• Baerends group VU, Amsterdam (>1973) • Ziegler group, Calgary (>1975)• SCM: Spin-off company (1995)
• Currently 15 people (8 senior PhD’s) + EU fellows• Development, testing, debugging, optimizing, porting,
documentation, support, ..– Implement what users want
– New features from academia
• Many academic collaborators / EU networks
Tom Ziegler (1945-2015)
Evert-Jan Baerends
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
ADF modeling suite authors 2014
E.J. Baerends, T. Ziegler, J. Autschbach, D. Bashford, A. Bérces, J.A. Berger, F.M. Bickelhaupt, C. Bo, P.L. de Boeij, P.M. Boerrigter, S. Borini, R. E. Bulo, L. Cavallo, D.P. Chong, L. Deng, R.M. Dickson, A. C. T. van Duin, D.E. Ellis, M. van Faassen, L. Fan, T.H. Fischer, C. Fonseca Guerra, M. Franchini, A. Ghysels, A. Giammona, S.J.A. van Gisbergen, M. Gorbani Asl, A.W. Götz, J.A. Groeneveld, O.V. Gritsenko, M. Grüning, S. Gusarov, F.E. Harris, T. Heine, P. van den Hoek, C.R. Jacob, H. Jacobsen, L. Jensen, E.S. Kadantsev, J.W. Kaminski, G. van Kessel, R. Klooster, F. Kootstra, A. Kovalenko, M.V. Krykunov, E. van Lenthe, J.N. Louwen, D.A. McCormack, E. S. McGarrity, A. Michalak, M. Mitoraj, S.M. Morton, J. Neugebauer, V.P. Nicu, L. Noodleman, V. P. Osinga, S. Patchkovskii, M. Pavanello, P.H.T. Philipsen, D. Post, C.C. Pye, W. Ravenek, M. de Reus, J.I. Rodríguez, P. Romaniello, P. Ros, R. Rüger, P.R.T. Schipper, H. van Schoot, G. Schreckenbach, J.S. Seldenthuis, M. Seth, D.G. Skachkov, J.G. Snijders, M. Solà, M. Swart, D. Swerhone, G. te Velde, P. Vernooijs, L. Versluis, L. Visscher, O. Visser, F. Wang, T.A. Wesolowski, E.M. van Wezenbeek, G. Wiesenekker, S.K. Wolff, T.K. Woo, A.L. Yakovlev
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Overview of the ADF molecular modeling suite
COSMO-RScontinuum methods
mesoscale methods
ReaxFF:reactive MDMonte Carlo
MOPAC
DFTBQM/MM
Tim
esca
le
ADF:DFT BAND: periodic DFT
ADF
BAND
ADF/ReaxFF
Systemsize
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Dr. Olivier Visser Dr. Erik van Lenthe Dr. Alexei Yakovlev Dr. Pier PhilipsenGUI, general ADF, ZORA, COSMO ADF, ReaxFF, optimization BAND, periodic DFTB
Job opening: MSc. Hans van Schoot MSc. Mirko Franchini Dr. Ole CarstensenScientific Software Developer ReaxFF, MD, GPUs ADF, BAND technical GUI, MD, consultancy
SCM software development staf
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Dr. Stan van Gisbergen Dr. Fedor Goumans Dr. Sergio Lopez Lopez General Management Business Development Scientific Partner ManagerSales, Legal, Signatures Marketing, Technical Sales EU projects, collaborations
Mrs. Frieda Vansina Mrs. Kitty Kleinlein Office Manager Office ManagerLicense files, user interactions Bookkeeping, special projects
SCM staf – office / management / business
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
2. New Acceleration Techniques1. ADF/ReaxFF
3. Upcoming Features
Grand canonical Monte Carlo + ADF/ReaxFF
ChemTraYzer: automated event detection Force bias Monte Carlo + ADF/ReaxFF
General, preparation, execution, analysis
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
2. New Acceleration Techniques1. ADF/ReaxFF
3. Upcoming Features
Grand canonical Monte Carlo + ADF/ReaxFF
ChemTraYzer: automated event detection Force bias Monte Carlo + ADF/ReaxFF
General, preparation, execution, analysis
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
ADF/ReaxFF: General
ReaxFF initially developed by Adri van Duin.
Alexei Hans Olivier
ADF/ReaxFF- optimization & parallelization of the original code.- many parameters included (e.g. transition metals)- GUI support- internal parameter optimization via Monte Carlo-...
Pyrolysis of an Illinois coal sample (Combustion & Flame 2012)
Cu-metal particle on a ZnO-support with water vapor (Zn/O: Raymand et al., Surf. Sci. 2010)
Hexane cracking on a Fe/H-ZSM5 catalyst (Fe/O: Aryanpour et al.,
JPC-A 2010)
Various parallel ADF/ReaxFF studies on various systems over the last years
new ReaxFF parameters
Adri van Duin
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
ADF/ReaxFF: The GUIfast and easy preparation...
set up complex, heterogeneous systems with Packmol builder
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
ADF/ReaxFF: The GUIfast and easy preparation...
straightforward definition of different temperature regimes, electric field regimes, pressure constraints, bond constraints, etc.
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
ADF/ReaxFF: The GUIfast and easy preparation, execution...
Manage all your jobs (local and remote) from your local desktop computer.Supports: PBS, SLURM, LSF, SGE...
Quickly need more resources? ADF/ReaxFF@CrunchYard:pay-per-use supercomputing.
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
ADF/ReaxFF: The GUIfast and easy preparation, execution and analysis
● Visualize trajectory properties (Temperature, Pressure, Energies...)● Analyze changing molecular composition during a reactive MD run● ...
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Friendly Expert Support Team!
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
2. New Acceleration Techniques1. ADF/ReaxFF
3. Upcoming Features
Grand canonical Monte Carlo + ADF/ReaxFF
ChemTraYzer: automated event detection Force bias Monte Carlo + ADF/ReaxFF
Preparation, execution and analysis
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
New Acceleration TechniquesGrand Canonical Monte Carlo (GC-MC)
Model a system that is in thermodynamic equilibrium(e.g. sorption in multi-component systems)
Task:
MD (ambient P,T)limited by timescales,reactions are rare events
MD (high P,T)High energy processes: thermodynamically not feasible.
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
New Acceleration TechniquesGrand Canonical Monte Carlo (GC-MC)
Model a system that is in thermodynamic equilibrium(e.g. sorption in multi-component systems)
Task:
MD (ambient P,T)limited by timescales,reactions are rare events
MD (high P,T)High energy processes: thermodynamically not feasible.
Monte Carlo (MC) Techniques:Sampling of configurational energetics,reproduce a thermal Boltzmann distribution
Grand canonical MC: stochastical exchange of atoms (reservoir ↔ system) @ const. μ
Reservoir
SystemThermodynamic equilibrium:
μReservoir
= μSystem
System GC-MC
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
New Acceleration TechniquesApplication of (hybrid) GC-MC with ADF/ReaxFF
System
Senftle et al.,J. Chem. Phys. 139, 044109 (2013); http://dx.doi.org/10.1063/1.4815820
Palladium based oxidation catalysts:catalytic behaviour ↔ extend of oxidation
Bulk-ox.
Surface-ox.
Both kinetic and thermodynamic influences are important!
Modelling
1. single MC move
2. Geometry optimization
4. System energy converged at equilibrium?
3. Accept/Reject?
no yes
● Model catalyst: 3nm Pd935
Cluster● Oxygen atoms are added until convergence
● GC-MC requires: - exp. μ
O2 (T, P0) from published tables
- 0 K bond dissociation energy of O2
from a DFT calculation Both kinetic and thermodynamic
influences are important!
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
New Acceleration TechniquesApplication of (hybrid) GC-MC with ADF/ReaxFF
Results
Senftle et al.,J. Chem. Phys. 139, 044109 (2013); http://dx.doi.org/10.1063/1.4815820
GC-MC/MD runsfor different T,P0
Convergence within 15000-25000 MC-steps(for this system)
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
New Acceleration TechniquesApplication of (hybrid) GC-MC with ADF/ReaxFF
Results
Senftle et al.,J. Chem. Phys. 139, 044109 (2013); http://dx.doi.org/10.1063/1.4815820
Identify phases via ρ(r)
Bulk Oxide (300 K, 1 atm)
Bulk Oxide (800 K, 1 atm)
Surface Oxide (1000 K, 1 atm)
Surface Oxide (1200 K, 1 atm)
Phase diagram for Pd-Oxide Formation
Theo. referenceExp. reference GC-MC/MD runs
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
2. New Acceleration Techniques1. ADF/ReaxFF
3. Upcoming Features
Grand canonical Monte Carlo + ADF/ReaxFF
ChemTraYzer: automated event detection Force bias Monte Carlo + ADF/ReaxFF
Preparation, execution and analysis
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
New Acceleration TechniquesUniform-acceptance force-bias Monte Carlo (fbMC) [1,2]
[1] Timonova et al., Phys. Rev. B 81, 144107 (2010); http://dx.doi.org/10.1103/PhysRevB.81.144107 [2] Mees et al.,Phys. Rev. B 85, 134301; http://dx.doi.org/10.1103/PhysRevB.85.134301
Metropolis MC
? ?
P(r)
P(r)
select
Sampling structural Phase Space
according to one well known ensemble distribution function (e.g. NVT)
fbMC
?
P(r, F1)
select
P(r, F2)
P(r, F3)
P(r, F4)
Sampling the Dynamics
● each change driven by “instantaneous” and “local” Boltzmann Distributions
● irrespective from distance to equilibrium
● Limits of P(r,F)T >> F → completely random movementT << F → Particle moves exactly in direction of force
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
New Acceleration TechniquesApplication of fbMC/MD [3]
[3] Neyts et al., J. Am. Chem. Soc. 133, 17225 (2011); http://dx.doi.org/10.1021/ja204023c
System
Unique physical properties:● extraordinary thermal conductivity● either metallic or semiconducting
(depending on chirality)● …
Problem: lack of control over fundamental properties
Computational SWNT growth studies(in 2011)
● lack of long range interactions & polarizable charges (prev. MD studies)
● Time scales too short for relaxation→ lots of defects, no healing
Modelling
4 ps - ReaxFF 104 fbMC steps
Modelling of a surface-bound Ni-catalyst
● Catalyst: Ni40
nanoclusterlower eight atoms fixed → Ni(100) plane
● Carbon atom added every 2 ps during ReaxFF dynamics
● virtual reflective boundary: Carbon atoms can only approach the cluster from above
● T = 1000 K
Single walled carbon nanotubes (SWNT) fbMC/ReaxFF [3]
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
New Acceleration TechniquesApplication of fbMC/MD to SWNT growth [3]
[3] Neyts et al., J. Am. Chem. Soc. 133, 17225 (2011); http://dx.doi.org/10.1021/ja204023c
Results
+ Healing of defects during sliding of network (steps 4-5).+ First simulation showing growth of an armchair SWNT with definite chirality on catalyst.
pentagon formation graphitic patches cap development
carbon network slides over catalyst
part of metal gets freed end of sliding,chirality obtained
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
2. New Acceleration Techniques1. ADF/ReaxFF
3. Upcoming Features
Grand canonical Monte Carlo + ADF/ReaxFF
ChemTraYzer: automated event detection Force bias Monte Carlo + ADF/ReaxFF
Preparation, execution and analysis
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Upcoming FeatureAutomated event detection with ChemTraYzer [4]
[4] Döntgen et al., J. Chem. Theory Comput. 11, 2517 (2015); http://dx.doi.org/10.1021/acs.jctc.5b00201
How to extract complex reaction mechanisms?
For example: burning methane..
Source: en.wikipedia.org/wiki/Combustion
Even 'simple' reactions...
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Upcoming FeatureAutomated event detection with ChemTraYzer [4]
[4] Döntgen et al., J. Chem. Theory Comput. 11, 2517 (2015); http://dx.doi.org/10.1021/acs.jctc.5b00201
How to extract complex reaction mechanisms?
Source: en.wikipedia.org/wiki/Combustion
Even 'simple' reactions can turn out to be complicated.
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Upcoming FeatureAutomated event detection with ChemTraYzer [4]
[4] Döntgen et al., J. Chem. Theory Comput. 11, 2517 (2015); http://dx.doi.org/10.1021/acs.jctc.5b00201
How to extract complex reaction mechanisms?
ChemTraYzer:
● Pythonscripts developed by the Leonhard group (Aachen)● Automatic determination of reaction events
- based on framewise comparison of atom connectivities- chemical composition extracted via graph-theory (BFS)
● (reasonable) rate constants
CH4 → .CH
3 +H
.CH
4 +
.OH→ .CH
3 + H
2O
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
Upcoming FeatureAutomated event detection with ChemTraYzer [4]
[4] Döntgen et al., J. Chem. Theory Comput. 11, 2517 (2015); http://dx.doi.org/10.1021/acs.jctc.5b00201
ChemTraYzer and ADF2016/ReaxFF
Already working: ● ChemTraYzer & ADF/ReaxFF trajectories
Work in progress: ● Full GUI support for both setup & analysis● On-the-fly analysis ● “stop trajectory if reaction xy has been sampled”
What it could look like...
Generate trainingssets forReaxFF reparametrization...
New Developments in ADF/ReaxFF Reactive Force FieldsOle Carstensen Manchester 2015
2. New Acceleration Techniques1. ADF/ReaxFF
3. Upcoming: ChemTraYzer
Automated event detection● complex reaction networks● rate constants● generation of trainingsets
fbMC + ADF/ReaxFF dynamicsModel processes that occuron loooong timescales by MC sampling of the dynamics
General, preparation, execution and analysis
Interested in a collaboration, a free trial licence or want to apply at SCM?
GC-MC + ReaxFF (dynamics)study thermodynamic and kinetic properties of multi-compound systems