IV. Classical Molecular Dynamics: Telluride July 2009 J. Tully

IV.
Classical Molecular Dynamics Telluride

July 2009
J. Tully
Basic Assumptions:
1. Born-Oppenheimer Approximation
2. Classical mechanical nuclear motion
Unavoidable Additional Approximations:

1. Approximate potential energy surface
2. Incomplete sampling
3. (often) Extrapolate to longer timescales
4. Too few atoms
5. Continuum solvent and other shortcuts
V. Adiabatic “on-the-fly” Dynamics Telluride
July 2009
J. Tully
What about the potential energy surface? i.e., the forces?
empirical (sometimes called “classical” - bad terminology)
semi-empirical (Huckel, valence-bond, PPP-Rossky…)
analytic fit to ab initio (HF, DFT, MC-SCF, CCSD,…)
on-the-fly ab initio (HF, DFT,,…)

July 2009
J. Tully
QM/MM Approach:
empirical force field
ab initio
Brownian
dynamics
dielectric
continuum
generalized Langevin
rigid template
July 2009
J. Tully
What about the potential energy surface? i.e., the forces?
empirical (sometimes called “classical” - bad terminology)
semi-empirical (Huckel, valence-bond, PPP-Rossky…)
analytic fit to ab initio (HF, DFT, MC-SCF, CCSD,…)
on-the-fly ab initio (HF, DFT,,…)

Chemical Dynamics Telluride
July 2009
J. Tully
I. Quantum Dynamics
II. Semiclassical Dynamics
aside: tutorial on classical mechanics
III. The Classical Limit via the Bohm Equations
IV. Classical Molecular Dynamics
V. Adiabatic “on-the-fly” Dynamics
VI. Car-Parrinello Dynamics
VII. Infrequent Events
aside: transition state theory and re-crossing
VIII. Beyond Born Oppenheimer
IX. Ehrenfest Dynamics
X. Surface Hopping
XI. Dynamics at Metal Surfaces
XII. Mixed Quantum-Classical Nuclear Dynamics
July 2009
J. Tully
The Hellman – Feynman Theorem:
d d
E j ( R)   j ( R ) | H el ( R) |  j ( R ) = force
dR dR
subject to  j ( R) |  j ( R)  1
and H el ( R ) |  j ( R )  E j ( R) |  j ( R)
d d H el ( R )
E j ( R)   j ( R) | |  j ( R)
dR dR
d d
  j ( R ) | H el ( R ) |  j ( R )   j ( R ) | H el ( R ) |  j ( R)
dR dR
July 2009
J. Tully
The Hellman – Feynman Theorem:

d d H el ( R )
E j ( R)   j ( R) | |  j ( R)
dR dR
d d
  j ( R ) | H el ( R ) |  j ( R )   j ( R ) | H el ( R ) |  j ( R)
dR dR
d d H el ( R )
E j ( R)   j ( R) | |  j ( R)
dR dR
 d d 
 E j ( R)   j ( R) |  j ( R)   j ( R) |  j ( R) 
 dR dR 
d
   j ( R) |  j ( R)   0
dR  
Pulay corrections
July 2009
J. Tully
I. Quantum Dynamics
X. Surface Hopping
VI. Car-Parrinello Dynamics Telluride
July 2009
J. Tully
d dL L
Euler-Lagrange equations: 
dt dq q
Classical Lagrangian: L  T V  

1
2
M  R 2  E j  R 

M R   E j (R ) / R
Car-Parrinello Lagrangian: L  

1
2
M  R 2    j | H el |  j 
fictitious electronic dynamics
  12  n  n |n    nm [  n |  m   nm ]
n nm
Kohn-Sham orbital
Roberto Car, Michele Parrinello

July 2009
J. Tully
Student problem: Lagrange (undetermined) parameter problem
find the minimum of the function f ( x, y )  x 2  y 2

subject to x  y  2 , using the method of Lagrange parameters
define h( x, y )  f ( x, y )   ( x  y  2)
and minimize wrt x, y and 

July 2009
J. Tully
d dL L
Euler-Lagrange equations: 
dt dq q
Classical Lagrangian: L  T V  

1
2
M  R 2  E j  R 

M R   E j (R ) / R
Car-Parrinello Lagrangian: L  

1
2
M  R 2    j | H el |  j 
  12  n  n |n    nm [  n |  m   nm ]

n nm

M R      j | H el |  j  / R
 nn      j | H el |  j  /  n*   nm m
m
Roberto Car, Michele Parrinello
IV. Classical Molecular Dynamics Telluride
July 2009
J. Tully

CAUTION!
“exact” PES + “exact” quantum dynamics exact
“exact” PES + classical dynamics inaccurate ???
empirical PES + classical dynamics maybe accurate ???
is “first-principles” synonymous with “less accurate”?

July 2009
J. Tully
I. Quantum Dynamics
X. Surface Hopping
Telluride
VII. Infrequent Events July 2009
J. Tully
Molecular Dynamics Time Step: ~ 10-15 s
Feasible number of integration steps: ~ 106 - 109
Maximum Time for Direct MD Integration: ~ 10-6 s
Moreover, need to sample multiple trajectories, different conditions, etc.
Examples:
High activation energy processes
Slow relaxation times (timescale mismatch, glasses, …)
Concerted movement of many atoms (protein folding)
Infrequent Event
Telluride
J. Tully
Many methods to enhance sampling for equilibrium simulations.

Non-Boltzmann sampling, e.g., umbrella sampling
U(x) Un-bias by weighting each

point by exp[+U(x)/kT]
energy
V(x)+U(x)
But what about dynamics?

V(x)
reaction coordinate
Telluride
J. Tully
Reaction Rate Theory (Classical) A

S
1. Define reactant and product regions R1 B

and “equilibrium reaction rate” S < S
S > S
2. Define reaction variable S
and dividing surface S R2

 ps 
3. Compute one-way flux through dividing plane: †
k   G ( ps )   dps
0 m
forward only
fraction of molecules at S = S with momentum ps velocity at S

Telluride
J. Tully
G ( ps ) 
 ... exp[H (s, p , q , p ) / kT ]  (s  s ) ds dq dp
s q q
s
 ... exp[H (s, p , q , p ) / kT ] ds dp


s q s dq dpq
 
s
define: Q0 exp( E0 / kT )  h  M  ... exp[H (s, ps , q, pq ) / kT ] ds dps dq dpq

   
Planck’s constant: conventional
quantum-classical correspondence
E†
E0 sa
Telluride
J. Tully
G ( ps ) 
s q q
s
 ... exp[H (s, p , q , p ) / kT ] ds dp


s q s dq dpq
 
s

   
Planck’s constant: conventional
quantum-classical correspondence
student problem: simple harmonic oscillator: V ( x)  12 K x 2  12 m  2 x 2
show that at high temperatures, the quantum and classical partition functions
are equal if the classical partition function is defined by
 
1
Qclassical    exp  V ( x ) / kT  exp 
  p 2
/ 2mkT  dx dp
h  
Telluride
J. Tully
G ( ps ) 
s q q
s
 ... exp[H (s, p , q , p ) / kT ] ds dp


s q s dq dpq
 
s

   
and Q† exp( E† / kT )  h  ( M 1)  ... exp[H † (s , q, pq ) / kT ] dq dpq

   
where H † ( s , q, pq )  H ( s , ps , q, pq )  ps2 / 2m
   
E†
E0 sa
Telluride
J. Tully
 ps2  Q†
G ( ps )  exp    exp[( E†  E0 ) / kT ]
 2mkT  hQ0
  ps   ps2  Q†  ( E†  E0 ) 

   m  exp  2mkT  hQ exp  
†
k  dps
0
    0  kT 
kT Q†  ( E†  E0 ) 
†
k  exp    transition state theory
h Q0  kT 
Activated Complex Theory
equilibration?
“Lake Eyring” Henry Eyring
Telluride
J. Tully
Transition State Theory is an upper limit to the true reaction rate
Variational Transition State Theory
A
S
VTST: Choose location of dividing surface
to minimize transition state theory rate R1 B
S < S
S > S
But still not the true rate
R2
Recrossing correction
Telluride
J. Tully
Recrossing correction:
rate k  f k†
1. f = fraction of reactive trajectories, f < 1 A

S
2. TST rate depends on location
of dividing surface S. R1 B
3. True rate does not depend on location

of dividing surface S.
R2
4. k  k†
July 2009
J. Tully
I. Quantum Dynamics
X. Surface Hopping
Telluride
VIII. Beyond Born-Oppenheimer July 2009
J. Tully
Ahmed Zewail
Telluride
J. Tully
  r , R    i  r; R   i  R 
Born-Oppenheimer
Telluride
J. Tully
  r, R      r; R    R 
i
i i
Beyond Born-Oppenheimer
Telluride
J. Tully
  r, R      r; R    R 
i
i i
Substitute into TISE, multiply from left by  j  r; R , integrate over r:

*
2
  M 1  2R  j  R   E j  R   j  R   E  j  R  
2 
2
  D ji  R  i  R    2  d ji  R    R i  R 
2 i i j
student problem:
show that dij(R) is anti-Hermitian and its diagonal elements are zero
where nonadiabatic (derivative) couplings are defined by:
 
d ij  R     M 1  *i  r, R   R  j  r, R   d r

 
Dij  R     M 1  *i  r , R    2R  j  r , R   d r

Telluride
J. Tully
15
Potential Energy Curves
for the
Oxygen Molecule
10
energy (eV)
from
R. P. Saxon and B. Liu,
J. Chem. Phys. 1977
0
1 2 3
internuclear distance (A)
Telluride
J. Tully
Nonadiabatic Transitions at Metal Surfaces: Electron-Hole Pairs
Evac
EF
conduction
band
Metal
Telluride
J. Tully
Vibrational Lifetime of CO on Cu(100)
v = 1 v = 0
Molecular Dynamics: t ~ 10-3 s.

Experiment (A. Harris et al.): t = 2.5 ps.
Telluride
J. Tully
H22 (ionic)
H11 (neutral)
off-diagonal coupling: H12

Telluride
J. Tully
 H11 ( R) H12 ( R) 
The Non-Crossing Rule H  R   
 12
H ( R ) H 22 ( R ) 
H11(R) “diabatic” potential energy curves

E(R)
H22(R)
H12(R)
R H. C. Longuet-Higgins
H11 ( R)  H 22 ( R) 1
E  R    [ H11 ( R)  H 22 ( R)]2  4[ H12 ( R)]2
2 2
Telluride
J. Tully
 H11 ( R) H12 ( R) 
H  R   
 12
H ( R ) H 22 ( R ) 
The Non-Crossing Rule

E(R) 2H12
“diabatic” potential energy curves
adiabatic potential
energy curves
R
H11 ( R)  H 22 ( R) 1
E  R    [ H11 ( R)  H 22 ( R)]2  4[ H12 ( R)]2
2 2
Telluride
J. Tully
The non-crossing rule

for more than 1 degree
of freedom:
“Conical Intersection”
N degrees of freedom:
N-2 dimensional “seam”
H11 ( R1 , R2 )  H 22 ( R1 , R2 ) 1
E   R1 , R2    [ H11 ( R1 , R2 )  H 22 ( R1 , R2 )]2  4[ H12 ( R1 , R2 )]2
2 2
Telluride
J. Tully
? The Massey Criterion:
E t  
E(R) 2H12
E  2 H12
?
t  distance/velocity
R  2 H12 / |  ( H11  H 22 ) / R | / R
 2 H12 / | ( H11  H 22 ) / R |
R | ( H11  H 22 ) / R |
 1 adiabatic
4 H122
H. S. W. Massey
Telluride
J. Tully
?
Landau-Zener Approximation
E(R) Assumptions:
1. H11 and H22 linear
? 2. H12 constant
3. Velocity constant
R
 2 H122 
 exp 
 | ( H  H ) / R | 
Pnonad
 R 11 22 
L. D. Landau C. M. Zener
Telluride
J. Tully
How do we simulate dynamics?
• Classical motion induces electronic transitions

• Quantum state determines classical forces
 Quantum – Classical Feedback: Self-Consistency

Telluride
J. Tully
  r, R      r; R    R 
i
i i
Substitute into TISE, multiply from left by  j  r; R , integrate over r:

*
2
  M 1  2R  j  R   E j  R   j  R   E  j  R  
2 
2
  D ji  R  i  R    2  d ji  R    R i  R 
2 i i j
Wave packet methods: Initial wave function: Gaussian wave packet

1/4
 2 
i ( x, 0)   2  exp(ik0 x) exp[( x  x0 ) 2 / a 2 ]
a 
 j ( x, 0)  0 , ji
Telluride
J. Tully
TWO GENERAL MIXED QUANTUM-CLASSICAL

APPROACHES FOR INCLUDING FEEDBACK
Vb Vb
Va Va
Ehrenfest (SCF) Surface-Hopping

Telluride
J. Tully
 (r , t )
i  H el  ( r , t ) R(t)
t
(r , t )   c (t )  (r; R)
i
i i (adiabatic states)
.
dc j dt   V jj c j  R    j (r ; R) |  Ri (r ; R)  ci
i
 i
Telluride
J. Tully
 (r , t )
i  H el  ( r , t ) R(t)
t
Classical path must respond self-consistently to

quantum transitions: “quantum back-reaction”
Ehrenfest and Surface Hopping differ

only in how classical path is defined
July 2009
J. Tully
I. Quantum Dynamics
X. Surface Hopping
Paul Ehrenfest
Telluride
IX. Ehrenfest Dynamics July 2009
J. Tully

i   r, R, t   H  r, R    r, R, t  (1)
t
Self-consistent Field Approximation (fully quantum):

i t 
  r, R, t     r, t    R, t  exp   Er  t ' d t ' (2)
 
Substituting (2) into (1), multiplying on the left by   R,t  and integrating
over R gives the SCF equation for the electronic wave function  r,t :
   r, t  2
i
t
  
2me i
 i  r , t   VrR  r , R   r , t 
2  (3)
where VrR  r , R     R, t VrR  r, R   R, t  d R

 * (4)
Telluride
J. Tully
Substituting (2) into (1), multiplying on the left by  r,t  and integrating
over r gives the equivalent SCF equation for the nuclear wave function   R,t :
   R, t  2
i    M 1  2R  R , t 
t 2  (5)
  *  r , t  H el  r , R    r , t  d r  R , t 
Thus, the potential energy function governing the nuclei becomes
  r, t  H el  r, R   r, t  d r instead of the adiabatic energy Ej(R).

 *
The Ehrenfest method consists of taking classical limit of Eq. 5,

i.e., running classical trajectories subject to this potential function
Telluride
J. Tully
 (r , t )
i  H el  ( r , t ) R(t)
t
..
M R(t )    R  (t ) | H el | (t ) 
Vb
Va
Telluride
J. Tully
..
M R(t )    R  (t ) | H el | (t ) 
Problem:
single configuration
 average path
July 2009
J. Tully
I. Quantum Dynamics
X. Surface Hopping
Telluride
X. Surface Hopping July 2009
J. Tully
 (r , t )
i  H el  ( r , t ) R(t)
t
Vb
Va
Telluride
J. Tully
Multi-Configuration Wave Function:
 (r, R, t )  
j
j (r, R )  j (R, t )
Substitute into Schrodinger Eq and take classical limit:
 Surface Hopping
However, a rigorous classical limit has not been achieved !

Telluride
J. Tully
One Approach: Multi-Configuration Bohm Equations:
 (r, R, t )  
j
j (r, R )  j (R, t )
 j (R, t )  A j ( R , t ) exp  i S j ( R , t ) 
 1  2
 2
R Aj
Sj   ( R S j )  E j (R ) 
2
2M 2M Aj
small h 

Sj   21M ( R S j )2  E j (R )
 motion on potential energy surface j

Telluride
J. Tully
  1
An   R An R  An 2R Sn 
2M

m
Am   n R  m  R exp  i ( Sm  S n ) 
Surface Hopping:
Evaluate all quantities along a single path
Sum over many stochastic paths
Telluride
J. Tully
Multi-Configuration Theory: Surface Hopping

 (r , t )
i  H el  ( r , t )
t R(t)
..
1] M R(t )    R E k , i.e., motion on single p.e.s.
2] Stochastic “hops” between states so that probability = |ck|2
3] Apply instantaneous “Pechukas Force” to conserve energy
4] “Fewest Switches”: achieve [2] with fewest possible hops:

Telluride
J. Tully
|c2|2 = 0.8 |c2|2 = 0.7
2
3
tk tk+1
Stochastic Fewest Switches algorithm (2-state):

 | c2 (k ) |2  | c2 (k  1) |2
 , | c2 (k ) |2  | c2 (k  1) |2
P21   | c2 (k ) |2
0 , | c2 (k ) |2  | c2 (k  1) |2

Telluride
J. Tully
Telluride
J. Tully
0.8
0.7
Probability of upper state
0.6
0.5 Surface
Hopping Exact QM
0.4
0.3
0.2
0.1
0 5 10 15 20 25 30
momentum (a.u.)
Telluride
J. Tully
Telluride
J. Tully
1.0
Probability of upper state
0.8
Surface Exact QM
0.6 Hopping
0.4
0.2
0.0
-4 -3 -2 -1 0 1
loge kinetic energy (a.u.)
Telluride
J. Tully
SHORTCOMINGS OF SURFACE HOPPING
1] Trajectories are independent

Trajectories should talk to each other
quantum wave packet surface hopping
Fundamental approximation, but required to make practical

Telluride
J. Tully
2] Forbidden Hops (or frustrated hops)
Hopping algorithm calls for a hop but

there is insufficient kinetic energy
Etotal
?
 probability on state k = |ck|2

Telluride
J. Tully
3] Detailed Balance?
What are the populations of the quantum

states at equilibrium?
Telluride
J. Tully
Detailed Balance: N1 P12 = N2 P21  Equilibrium
• Long Timescales
• Multiple Transitions
h
• Relaxation Processes
• Infrequent events
e.g., nonradiative transition
vs. reaction on excited state
Telluride
J. Tully
“In theories in which the reservoir is treated classically and

its effects on the system described in terms of random
functions instead of noncommuting operators, it follows that
Wmn = Wnm. This is a serious shortcoming of all semiclassical
theories of relaxation.” K. Blum, Density Matrix Theory
and Applications, 2nd Ed., (Plenum, NY, 1996).
However
This is not true for either Ehrenfest or Surface Hopping

Telluride
J. Tully
Excited State Population vs. Inverse Temperature

0
kΤ - exp(-ΔΕ/kΤ)
ΔΕ 1-exp(-ΔΕ/kΤ)
-2
Ehrenfest
loge(population)
-4
Surface Hopping
-6
Boltzmann
-8
2-State System
E = 34.6 kJ/mole
-10
0 0.001 0.002
1/T (K)
Telluride
J. Tully
Many Quantum States

[classical bath] --- C --- C --- C --- C --- C --- C --- C --- C --- H
Random force and friction classical quantum

(5000 oK) (Morse) (Morse)
Ehrenfest (SCF) Surface Hopping
V=0
V=1
V=2
V=3
Priya Parandekar

IV. Classical Molecular Dynamics: Telluride July 2009 J. Tully

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IV. Classical Molecular Dynamics: Telluride July 2009 J. Tully

Uploaded by

Copyright:

Available Formats

IV.

Classical Molecular Dynamics Telluride

Unavoidable Additional Approximations:

What about the potential energy surface? i.e., the forces?

empirical (sometimes called “classical” - bad terminology)

semi-empirical (Huckel, valence-bond, PPP-Rossky…)

analytic fit to ab initio (HF, DFT, MC-SCF, CCSD,…)

on-the-fly ab initio (HF, DFT,,…)

What about the potential energy surface? i.e., the forces?

empirical (sometimes called “classical” - bad terminology)

semi-empirical (Huckel, valence-bond, PPP-Rossky…)

analytic fit to ab initio (HF, DFT, MC-SCF, CCSD,…)

on-the-fly ab initio (HF, DFT,,…)

The Hellman – Feynman Theorem:

The Hellman – Feynman Theorem:

Roberto Car, Michele Parrinello

Student problem: Lagrange (undetermined) parameter problem

find the minimum of the function f ( x, y )  x 2  y 2

and minimize wrt x, y and 

  12  n  n |n    nm [  n |  m   nm ]

V. Adiabatic “on-the-fly” Dynamics

“exact” PES + “exact” quantum dynamics exact

“exact” PES + classical dynamics inaccurate ???

empirical PES + classical dynamics maybe accurate ???

is “first-principles” synonymous with “less accurate”?

Molecular Dynamics Time Step: ~ 10-15 s

Feasible number of integration steps: ~ 106 - 109

Maximum Time for Direct MD Integration: ~ 10-6 s

Moreover, need to sample multiple trajectories, different conditions, etc.

Many methods to enhance sampling for equilibrium simulations.

U(x) Un-bias by weighting each

But what about dynamics?

Reaction Rate Theory (Classical) A

1. Define reactant and product regions R1 B

fraction of molecules at S = S with momentum ps velocity at S

 ... exp[H (s, p , q , p ) / kT ] ds dp

 ... exp[H (s, p , q , p ) / kT ] ds dp

student problem: simple harmonic oscillator: V ( x)  12 K x 2  12 m  2 x 2

 ... exp[H (s, p , q , p ) / kT ] ds dp

and Q† exp( E† / kT )  h  ( M 1)  ... exp[H † (s , q, pq ) / kT ] dq dpq

  ps   ps2  Q†  ( E†  E0 ) 

Activated Complex Theory

Transition State Theory is an upper limit to the true reaction rate

Variational Transition State Theory

1. f = fraction of reactive trajectories, f < 1 A

3. True rate does not depend on location

Substitute into TISE, multiply from left by  j  r; R , integrate over r:

Nonadiabatic Transitions at Metal Surfaces: Electron-Hole Pairs

Vibrational Lifetime of CO on Cu(100)

Molecular Dynamics: t ~ 10-3 s.

off-diagonal coupling: H12

H11(R) “diabatic” potential energy curves

The Non-Crossing Rule

The non-crossing rule

? The Massey Criterion:

How do we simulate dynamics?

• Classical motion induces electronic transitions

 Quantum – Classical Feedback: Self-Consistency

Substitute into TISE, multiply from left by  j  r; R , integrate over r:

Wave packet methods: Initial wave function: Gaussian wave packet

TWO GENERAL MIXED QUANTUM-CLASSICAL

Ehrenfest (SCF) Surface-Hopping