Car-Parrinello Molecular Dynamics Simulations (CPMD): Basics Ursula Rothlisberger EPFL Lausanne, Switzerland Literature Car-Parrinello: • R. Car and M. Parrinello A unified approach for molecular dynamics and density functional Phys.Rev.Lett. 55, 2471 (1985) • D. Marx and J. Hutter Modern Methods and Algorithms of Quantum • P. Carloni, U. Rothlisberger and M.Parrinello The role and perspective of ab initio molecular dynamics in the study of biological systems Acc. Chem.Res. 35, 455 (2002) • U Rothlisberger J. Grotendorst (Ed.), NIC Forschungszentrum Jülich (2000) 15 years of Car-Parrinello simulations in Physics, Chemistry, p.301 and Biology • D. Sebastiani and U. Rothlisberger Advances in density functional based modelling techniques: Recent extensions of the Car-Parrinello approach in P. Carloni, F. Alber ‘Medicinal Quantum Chemistry’, Wiley-VCH, Weinheim (2003) Computational Chemistry: Reviews of Current Trends, J. Leszczynski (Ed.), World Scientific, Vol. 6, (2001) p.33 When Quantum Chemistry Starts to Move... Traditional QC Methods Classical MD Simulations Car-Parrinello MD • improved optimization • finite T effects • thermodynamic & dynamic properties • solids & liquids • parameter-free MD • ab initio force field • no transferability problem • chemical reactions When Newton meets Schrödinger... Sir Isaac Newton (1642 - 1727) Erwin Schrödinger (1887 - 1961) F ma Hˆ Newt-dinger F ma Hˆ The ideal combination for Ab Initio Molecular Dynamics Atoms, Molecules and Chemical Bonds Atoms + e- N protons & neutrons N electrons Chemical Bonds Chemical Reaction Basic Principles of Quantum Mechanics Wavefunctions and Probability Distributions Classical Mechanics: The position and velocity of the particle are precisely defined at any instant in time. Quantum Mechanics: The particle is better described via its wave character, with a wave function (r,t). The square of wave function is a measure for the probability P(r) to find the particle in an infinitesimal volume element dV around r. 2 P ( r ) ( r )dV 2 ( r )dV 1 V The total probability to find the particle anywhere in space integrates to 1. Classical Mechanics positions and momenta have sharp defined values r,v Continous energy spectrum Epot Quantum Mechanics uncertainty xp h relation r , t * r , t * r , t energies are quantized n=3 n=2 n=1 n=0 q 0 Newton`s Equations hw q 0 Schroedinger Equation n, E, m, h0 Classical Mechanics: Particle Motion r F ma ro,vo r(t),v(t) r ( t ) ro v o t 12 at2 v ( t ) v o at Position r and velocity vof a particle can be calculated exactly at any time t. 1 E kin mv 2 2 Continuous energy Goal: Computational method that provides us with a microscopic picture of the structural and dynamic properties of complex systems Solution 1: Time-dependent Schrödinger Eq. for a system of N nuclei and n electrons i ( R1, R2 , R3...RN , r1, r2 , r3 ,...rn , t ) ( R1, R2 , R3...RN , r1, r2 , r3 ,...rn , t ) t not possible! Approximations: 1) Born-Oppenheimer Approximation (1927): mel <<< mp electronic and nuclear motion are separable Exceptions: Jahn-Teller instabilities, strong electron-phonon coupling, molecules in high intensity laser fields nonadiabatic dynamics Product Ansatz for total wavefunction: (R1, R2 , R3...RN , r1, r2 , r3,...rn ) nu (R1, R2 , R3...RN )el (r1, r2 , r3,...rn ) Electronic Schrödinger Eq.: el el (r1, r2 , r3,...rn , R) Eel (r1, r2 , r3,...rn , R) Electronic Hamiltonoperator: ZI 1 2 el 1 / 2 i i I ,i RI ri i j rij Solve electronic Schrödinger Eq. for each set of nuclear coordinates R (R1, R2 , R3,...RN ) E (R ) potential energy surface (PES) Nuclear SchrödingerEq. Hnunu (R1, R2 , R3....RN ) Etotnu (R1, R2 , R3....RN ) Nuclear Hamiltonoperator: ZI ZJ 1 2 nu I E ( R) I 2M I I , J RIJ Nuclear Quantum Dynamics (review: Makri, Ann. Rev. Phys. 50, 167 (1999) E (R ) Empirical parameterization → force field based MD Calculate E ( R) H → Car-Parrinello Dynamics Classical Nuclear Dynamics 2) Most atoms are heavy enough so that their motion can be described with classical mechanics • ratio of the deBroglie wavelength proton: el m p p mel 1/ 2 h of an electron and a 2mE 40 classical approximation is better: m, n, E, T Works surprisingly well in many cases! what cannot be described: • zero point energy effects • (proton) tunneling quantum corrections to classical results (Wigner&Kirkwood) classical MD extended to quantum effects on equilibrium properties and to some extend also to quantum dynamics path integral MD and centroid dynamics First-Principles Molecular Dynamics How do we do that? 1) straight-forward: • solve electronic structure problem for a set of ionic coordinates • evaluate forces • move atoms Born-Oppenheimer Dynamics Car - Parrinello Molecular Dynamics (1985) Lagrangian Formulation of Classical Dynamics L T (qI ) V (qI ) 1 2 L M I RI V (RI ) I 2 Euler-Lagrange Equation: d δL δL dt δqi * δqi * E M I RI RI Car - Parrinello Molecular Dynamics (1985) Extended Lagrangian Formulation 2 L I 1 / 2 M I RI i i i E i , RI ex ij ij r r dr Roberto Car i j ij Michele Parrinello Equations of Motion E M I RI RI i Hi j ij j Can be integrated simultaneously (e.g. with Verlet, Velocity-Verlet algorithm etc..) t 2 RI ( t t ) 2 RI ( t ) RI ( t t ) FI ( t ) O(t 4 ) 2M I Verlet algorithm dt ~0.1-0.2 fs Does this fictitious classical dynamics described via the extended Lagrangian have anything to do with the real physical dynamics??? • if MI ' s K e 0 total energy of the system becomes the real physical total energy K e K I Epot K I Epot can be checked via energy conservation After initial wfct optimization, system is propagated adiabatically and moves within finite thickness Ke over the potential energy surface What’s the price for it ? • systems sizes: few hundred to few thousands of atoms (CP2K) • Time Steps: ~0.1 fs • Simulation Periods: few tens of ps The Quantum Problem Stationary Solutions: Time-independent Schrödinger Eq. ˆ E Variable Separation: Electronic Schrödinger Eq.: ˆ el el (r1 , r2 , r3 ,...rn , R) Eel (r1 , r2 , r3 ,...rn , R) ZI 1 2 ˆ 1 / 2 i Electronic Hamiltonoperator: el i I , i RI ri i j rij Product Ansatz for the wavefunction: el (r1 , r2 , r3 ,...rn ) (r1 )(r2 )(r3 )...(rn ) Effective 1-particle model The Quantum Problem Set of N coupled 1-particle equations: ˆ hi (ri ) i (ri ) ˆh 1 / 2 2 Z I 1 i i I RI ri i j rij Basis Set Expansion: (ri ) c l l Set of algebraic Eqs. Solved iteratively (self-consistent field) l Plane-waves: i r 1 Vcell Choice of QM method: DFT c im e m iG m r ca. 10’000-100’000 FFT DENSITY FUNCTIONAL THEORY Walter Kohn and John Pople Nobelprize in Chemistry 1998 Literature on DFT: Original Papers: • P.Hohenberg, W.Kohn, Phys.Rev.B 1964, 136, 864-871. • W.Kohn, L.J.Sham, Phys.Rev.A 1965, 140, 1133-1138. Textbooks: • W.Kohn, P.Vashista, in Theory of the Inhomogeneous Electron Gas, N.H.March and S.Lundqvist (Eds), Plenum, New York 1983 • R.G.Parr, W.Yang, Density Functional Theory of Atoms and Molecules, Oxford University Press, New York 1989. R.M.Dreizler, E.K.U.Gross, Density-Functional Theory, Springer, Berlin 1990. • W.Kohn, Rev.Mod.Phys. 1999, 71. Density Functional Theory (DFT) Like Hatree-Fock: effective 1-particle Hamiltonian Let’s define a new central variable: x 1 , x 2 , x 3 ... x N r Electron density * r x1 , x2 , x3 ... x N x1 , x2 , x3 ... x N dx1dx2 ...dx N ' Total electron density integrates to the number of electrons: r dr N Theoretical foundations of DFT based on 2 theorems: Hohenberg and Kohn (1964): (Phys.Rev. 136, 864B) • The ground state energy of a system with N electrons in an external potential Vex is a unique functional of the electron density r E Er Vex determines the exact r vice versa: Vex is determined within an additive constant by gs expectation value of any observable (i.e. the H) is a unique functional of the gs density r •Variational principle: The total energy is minimal for the ground state density 0 r of the system Er min E 0 E 0 r Kohn and Sham (1965): (Phy. Rev. 1140, 1133A) The many-electron problem can be mapped exactly onto: •an auxiliary noninteracting reference system with the same density (i.e. the exact gs density) •where each electrons moves in an effective 1-particlepotential due to all the other electrons E i i i r Vion r r d r 2 i (1) (2) ' 1 r r ' ' d r d r E xc r E ion R I 2 r r (4) (3) (5) (1) Kinetic energy of the non interacting system (2) External potential due to ionic cores (3) Hartree-term ~ classical Coulomb energy (4) exchange-correlation energy functional (5) Core -core interaction 2 r 2 i r i Kohn-Sham eqs: 1 2 2 Vion r VH r Vxc r i i i r Exchange and Correlation Exchange-Correlation Hole Universal exchange-correlation functional, exact form not known! local density approximation hom xc r xc r can be determined exactly: Exchange: (P.A.M. Dirac, Proc. Cambridge Phil. Soc. 26, 376 (1930), E.P. Wigner, Trans. Fraraday Soc. 34, 678 (1987)) 1 hom 3 x x 1 3 r C 33 Cx 4 Correlation: (D.M. Ceperly, B.J. Alder, Phys. Rev. Lett. 45, 566 (1980), G.Ortiz, P. Ballone, Phys. Rev. B 50, 1391 (1994)) exact (numerical) results from Quantum Monte Carlo simulations Parametrized analytic forms that interpolate between different density regimes are available (e.g. J.P. Perdew, A. Zunger, Phys. Rev. B. 23, 5084 (1981)) - in principle very crude approximation! - Exc of a non uniform system locally ~ uniform electron gas results - should ‘work’ only for systems with slowly varying density but: atoms and molecules are inhomogeneous systems! - works remarkably well in practice: Performance of LDA/LSDA in general good structural properties: bond lenghts up to 1-2% bond angles ~ 1-2 degrees torsional angles ~ a few degrees vibrational frequencies ~ 10% ( phonon modes up to few %) cheap and good method for transition metals!: e.g. Cr2, Mo2 in good agreement with experiment ( not bound in HF, UHF!) F2 re within 3% (not bound in HF) atomization, dissociation energies over estimated (mainly due to errors for atoms), typically by 10-20% hydrogen-bonding overestimated van der Waals-complexes: strongly overestimated binding (e.g. noble gas dimers, Mg2, Be2: factor 2-4 Re[Å] De (eV) HF 1.465 -19.4 Cr2 CCSD 1.560 -2.9 CCSD(T) 1.621 0.5 (Scuseria 1992) DFT 1.59 1.5 exp 1.679 1.4 Generalized Gradient Approximation (GGA) E fxc r , r GGA xc r r dr E xc r , r GGA xc correction function chosen to fulfill formal conditions for the properties of the ex-corr hole Determination of parameters: • fully non empirical • fit to exact Ex-Corr energies for atoms • fit to experimental data (empirical) man different forms (B88, P86, LYP, PW91, PBE, B3LYP etc..) Time-independent electronic Schrödinger Equation: H E Density-Functional Theory E E r r * i i r i r E 1/ 2 i* r 2i r r Vext r dr 1/ 2 r r ' r r' drdr' Exc r Practical Implementation • periodic boundary conditions • plane wave basis set up to a given kinetic energy cutoff Ecut φ i r 1 cime Vcell m iG m r use of FFT techniques convenient evaluation of different terms in real space (Eex-corr, Eext) or in reciprocal space (Ekin, Ehartree) • typical real space grid: ~1003, ~10000-100000 pws • most of the time: FFT most time consuming step (NMlogM) • for large systems: orthogonalization ~N2 • well parallelizable (over number of electronic states and first index of real space grid Pseudo Potentials Framework • Chemical properties determined by valence electrons • perform atomic all electron calculation ab initio pseudo pseu r all ( r ) pseu r : rc ps φ 0 2 r rdrc rc ae φ 0 r r > rc smooth fct r < rc 2 dr • invert Schrodinger equation r(a.u.) m Zv Vps (r ) erf (r / rc ) Vl (r )Pl r l0 2 Vl (r ) (a br )e (r / rc)2 H pseu ( r ) e all pseu ( r ) ( 1 / 2 2 V ( r )) pseu ( r ) e all pseu ( r ) V ( r ) V hartree ( r ) V exc ( r ) V pseu ( r ) ABINIT CASTEP www.abinit.org [ i Molecular Simulations Inc. ] CPMD [ i i CP2K ] Fhi98md [ i i i M. Parrinello, MPI Stuttgart, Germany and IBM Zurich Research Laboratory, Switzerland www.cpmd.org Free software Fritz-Haber Institute Berlin, Germany fhim@fhi-berlin.mpg.de ] JEEP François Gygi, Lawrence Livermore National Laboratory, USA NWCHEM Pacific Northwest National Laboratory, USA PAW [ i v P.E. Blöchl, Clausthal University of Technology, Germany ] SIESTA VASP [ v ] P. Ordejon, Institut de Ciencia de Materials de Barcelona, Barcelona, Spain [ v J. Hafner, University of Vienna, Austria CPMD (3.9) (CP2K) www.cpmd.org Features (see also online manual): • • • • • • • plane wave basis, pseudopotentials, pbc and isolated systems LDA, LSD, GGAs (single point hybrid fct calcs possible) geometry optimization MD (NVE, NVT, NPT, Parrinello-Rahman) path integral MD different types of constraints and restraints Property calculations: population analysis, multipole moments, atomic charges, Wannier fcts, Fukui fcts etc.. Runs on essentially all platforms.. Most Recent Features: • QM/MM interface • Response function calculations: NMR Chemical shifts, electronic spectra, vibrational spectra • Time Dependent DFT MD in excited states • History dependent Metadynamics Mixed Quantum-Classical QM/MM- Car-Parrinello Simulations • Fully Hamiltonian QM/MM Car-Parrinello hybrid code QM-Part: CPMD 3.8 pbc, PWs, pseudo potentials (n-1) CPUs MM-Part: GROMOS96 + P3M, AMBER (SYBIL, UFF) 1 CPU Interface Region Quantum Region Classical Region A. Laio, J. VandeVondele, and U. Rothlisberger, J. Chem. Phys. 116, 6941 (2002); A. Laio, J. VandeVondele, and U. Rothlisberger, J. Phys. Chem. B (ASAP article) review in : M. Colombo et al. CHIMIA 56, 11 (2002) QM/MM Car-Parrinello Simulations monovalent pseudo potential QM/MM Lagrangian QM 2 * 1 1 L dr i r i r M I RI E MM EQM / MM 2 i 2I * EQM i , j dr i r j r i , j i i, j e- MM j l k qp - + EQM: DFT qo * 1 1 1 E KS i , RI dr i r i r drVN r r dr dr ' r r ' E xc r included 2i 2 r r' EMM: Standard biomolecular Force Field bonded non bonded E MM E MM E MM in Vext 1 1 bonded EMM kb (rij b0 ) 2 k ( ijk 0 ) 2 k n 1 cos(n ijkl 0 ) b2 2 n 12 6 op op qq nonbonded EMM l m 4 op rop lm 4 0 rlm op rop QM/MM Car-Parrinello in Combination with Response Properties • Variational Perturbation Theory: A. Putrino, D. Sebastiani, M. Parrinello, 113, 7103 (2000) • IR and Raman Spectra • Fukui Functions R. Vuilleumier, M. Sprik J.Chem.Phys. 115, 3454 (2001) • Chemical Shifts D. Sebastiani, M. Parrinello, J. Phys. Chem. A 105, 1951 (2001) • TDDFT: Spectra and Dynamics J. Hutter J.Chem.Phys. 118, 3928 (2003) QM/MM Car-Parrinello in Combination with Excited State Methods • ROKS HOMO-LUMO single excitations T. Ziegler et al. Theor. Chim. Acta 43, 261 (1977) (sum method) CP-version: I. Frank et al. J. Chem. Phys. 108, 4060 (1998) • LR-TDDFT-MD (Tamm-Dancoff Approximation) J. Hutter J. Chem.Phys. 118, 3928 (2003) L. Bernasconi et al. J. Chem.Phys. 119, 12417 (2003) • P-TDDFT-MD I. Tavernelli (to be published) Landau-Zener Surface Hopping Ehrenfest Dynamics m1 m2 t1,2 E(s) = 2E(m) - E(t) Limitations Due to Short Simulation Time • MD as dynamical tool: Real-time simulation of dynamical processes many processes lie outside time range • MD as sampling tool: only small portion of phase space is sampled relevant parts might be missed, pa exp( Fab ) especially if there exist large pb barriers between different important regions (e.g. different conformers) ensemble average have large statistical errors (e.g. relative free energies!) pA pB Techniques from Classical MD: • Sampling at enhanced temperature • Rescaling of atomic mass(es) • Constraints (Ryckaert, Ciccotti, Berendsen 1977) (Sprik & Ciccotti 1998) • Umbrella Sampling (Torrie&Valleau 1977) • Quasi-Harmonic Analysis (Karplus, Jushick 1981) • Reaction Path Method (Elber & Karplus 1987) • ‘Hypersurface Deformation’ (Scheraga 1988, Wales 1990) • Multiple Time Step MD (Tuckerman, Berne 1991) (Tuckerman, Parrinello 1994) • Subspace Integration Method (Rabitz 1993) • Local Elevation (van Gunsteren 1994) •Conformational Flooding (Grubmuller 1995) •Essential Dynamics (Amadei&Berendsen 1996) • Path Optimization (Olender & Elber 1996) • Multidimensional Adaptive Umbrella Sampling (Bartels, Karplus 1997) • Hyperdynamics (Voter 1997) (Steiner, Genilloud, Wilkins 1998) (Gong & Wilkins 1999) • Transition Path Sampling (Dellago, Bolhuis, Csajka, Chandler 1998) • Adiabatic Bias MD (Marchi, Ballone 1999) • Metadynamics (Laio, Iannuzzi, Parrinello PNAS 99, 12562 (2002), PRL 90, 23802 (2003) Development of Enhanced Sampling Methods Configurational Sampling • multiple time step sampling • classical bias potentials and forces • double thermostatting • parallel tempering Two Dimensional Free Energy Surface with torsional potential bias Sampling of Rare Reactive Events Electronic Bias Potentials • Finite Electronic Temperature • Vibronic Coupling • Charge Restraint T = 500K EA = 30 kcal/mol Peroxynitrous Acid 48ps 1kcal/mol J. Chem. Phys. 113 4863 (2000), J. Chem. Phys. 115 7859-7864 (2001), J. Phys. Chem. B 106, 203-208 (2002), J. Am. Chem. Soc. 124, 8163 (2002) Constraints Lagrangian: L T V f (r {N}) f0 (r {N}) 0 Equations of motion: miri fi gi gi i • freeze out fast motions increase integration time step ( linear speed up) • constrain slowest motion guide system ‘manually’ over barrier (condition: slowest part of reaction coordinate is known, all other degrees of freedom have time to equilibrate along the path) ( free energy differences via thermodynamic integration) 2 dF F12 F( 2 ) F(1) d' 1 d' integral replaced by a discrete set of points (R)= ’ for a simple distance constraint (R)= lRI-RJl: dF ' d' Umbrella Sampling: Bias Potentials (Torrie&Valleau 1977) f (p, q) H (Grubmuller 1995, Voter 1997, Karplus 1997, Wilkins 1998…) ‘Ideal’ Bias: ‘Golden Rules’ f (p, q) H f (p, q)e H(p,q) H e H(p,q) H f (p, q)e'H'(p,q)H(p,q) 'H' e'H'(p,q)H(p,q) 'H' • high overlap with original ensemble • close match PES or free energy surface • low dimensionality • computationally inexpensive Sampling Error in ab initio MD: Methyl Group Rotation in Ethane C2H6 (500K, 7.25 ps) Probability Distribution HCCH EA = 2.8 kcal/mol 0.0135 0.00991 p (0120 ) 0.996 p (120 240 ) p (240 360 ) FMAX 4.5 kcal mol Atomic Bias Potentials Methyl Group Rotation in Ethane C2H6 (500K, 7.25 ps) Before correction After correction Torsional Bias 0.0017au p (0120 ) 0.338 kcal p (120 240 ) 0.331 FMAX 0.02 mol p (240 360 ) 0.331 Vbias 1/ 2V0 (1 cos(3)) Bias Potentials from Classical Force Field 400 300 200 100 0 Peroxynitrous Acid ONOOH -100 -100 0 100 200 300 400 Trajectory in biased space (48 ps) Free Energy Surface J. VandeVondele, U.R. J. Chem. Phys. 113 4863 (2000) CAFES: Canonical, Adiabatic Free Energy Sampling • Partitioning into reactive system / environment adiabatic decoupling E m t m m m R E Slow dynamics of the reactive subsystem R different temperatures TR/TE (2 thermostats) ρ REAL (x R ) ρ CAFES (x R ) TR TE Sampling efficiency at TR can be estimated Ea= 20 kcal/mol, TE=300K, TR=1200K -> 1013 J. VandeVondele, U.R. J. Phys. Chem. B 106, 203 (2002) Nucleophilic substitution with anchimeric assistance • QMMM SPC/CPMD • CAFES 100 / 2000K / 300K • ~22 kcal/mol • shows that the reaction coordinate is not simple Transition State Path Sampling Given: - initial state A - final state B - one path connecting the two generate the ensemble of ‘reactive paths’ calculate transition rates A general path has : L 1 f ({x}) ( x0 ) p( x x 1 ) 0 A reactive path has : L 1 f AB ({x}) hA ( x0 )hB ( xL ) ( x0 ) p( x x 1 ) 0 Dispersion Interactions in DFT MM QM Suggested Remedies • add -C6/r6 -term (with damping function) (LeSar 1984 ,Sprik 1996, Scoles 2001, Parrinello 2003, Wang, York2004… • specially designed (local) functionals (PW91, PBE, mPBE, X3LYP, …) • density partitioning schemes (Wesolowski 2003…) • nonlocal correlation functionals for special cases (Langreth,Lundvist 2000, 2003…) • perturbation calculation of dispersion forces (Kohn 1998, Szalewicz 2003…) Optimized Effective Atom Centered Potentials VˆtotKS Vˆext Vˆhartree Vˆxc Vˆ NL Expansion in linear combination of atom-centered (nonlocal) potentials Vˆ NL r , r' V eff r RI , r' RI I Vˆ eff r , r' V loc r ( r r' ) V nl r , r' V Analytic pseudopotentials by Goedecker et al. loc r r2 Z ion r erf e xp 2 r 2rloc rloc 2 2 4 6 r r r C C C3 C4 1 2 rloc rloc rloc Vl nl l 3 m l j , h1 r , r' Ylm rˆ plh r hlhj plj r'Ylm* rˆ' plh r r l 2( h1) e xp( r 2 ) / 2rl2 ) Optimization Penalty Functional P nr , i d 3rw ( r )F n(r) dP F dn( r ) d 3rw ( r ) d j n( r ) d j V Hˆ (j 1) Linear density response calculated via first order perturbation theory with perturbation Hamiltonian For Vˆ nl : 2 w F R ref P disp Rref Eint Rref Eint Rref 2 N ions ref I I I eff i j 1 = -0.00352 2 = 3.280 BLYP OECP MP2 Is this potential transferable??? BLYP BLYP OECP OECP MP2 MP2 BLYP z = 3.3A OECP E=32 meV/atom exp z = 3.35A E=35 meV/atom Reference system: Ar2 1 additional f-channel: 1 = -0.00206 2 = 2.902 BLYP BLYP OECP OECP MP2 Klopper et al. J.Chem.Phys. 101, 9747 (1994) MP2 What about the intramolecular geometry?? Bond lengths in benzene: << 0.01 A What about the electronic properties?? Dipole moment: benzene-Ar Quadrupole moment: benzene Polarizability: argon xx- yy benzene zz benzene-Ar BLYP OECP MP2 0.047 0.035 0.037 -5.35 -5.50 -6.46 12.30 39.18 55.0 12.31 38.45 58.1 11.15 35.07 59.2 Formaldimine. Excited state dynamics after excitation S0→S1 The region of conical intersection, CI, is reached only in case of non-thermostatted trajectories. relaxation to product geometry α Φ Φ start on S1 back to reactant geometry Increasing kinetic energy α ω ω Φ minimum on S1 α Landau-Zener SH energy Classical treatment for the derivation of an analytical formula for the transfer rate which is valid for any value of the coupling matrix element spanning the range between adiabatic and nonadiabatic ET. Um (q) Um (q*) Fm (q*) q , m D, A UA UD Fm (q*) q* q 0 q q U m (q ) q q q* q v *t H DA Tvib U D (q*) ( H0 (t ) Vcoupl ) H0 (t ) FD v *t D D FA v *t A A V VDA D A VAD A D The asymptotic value for the survival probability of the electron for remaining at the donor PD D U , D 2 The donor survival probability is 2 PD e 2 1 VDA v * FD FA Units: atomic units used throughout Transition Rate Constants Reactive Flux Correlation Function (t ) hA (0)h B k( t ) k A Be t / rxn hA (0) 1rxn (k A B kB A ) • Can be calculated with trajectories starting at the TS • Is difficult if a RC/TS cannot be defined. Rate constants in the TPE dx ( x )h ( x )h ( x ) C (t ) dx ( x )h ( x ) 0 0 0 A 0 B 0 A 0 t dC (t ) k A B dt hB (t ) AB C (t ) C (t ) hB (t ) AB Contrary to direct MD, the computational efficiency does not depend on the height of the barrier. • C(t) = the fraction of trajectories of length t, starting in A, that arrives in B • Can be calculated with a reversible work calculation.