From the Schrodinger Equation to molecular Dynamics

In particle methods, the laws of classical mechanics [48, 371] are used, in particular Newton’s
second law. In this chapter we will pursue the question why it makes sense to apply the laws of
classical mechanics, even though one should use the laws of quantum mechanics. Readers that are
more interested in the algorithmic details or in the implementation of algorithms in molecular
dynamics can skip this chapter.

In quantum mechanics, the Schrodinger equation is taking the place of Newton’s equations. But the
Schrodinger equation is so complex that it can be solved analytically only for a few simple cases.
Also the direct numerical solution on computers is limited to very simple systems and very small
numbers of particles because of the high dimension of the space in which the Schrodinger equation
is posed. Therefore, approximation procedures are used to simplify the problem. These procedures
are based on the fact that the electron mass is much smaller than the mass of the nuclei. The idea is
to splitthe Schrodinger equation, which describes the state of both the electrons and nuclei, with a
separation approach into two coupled equations. The influence of the electrons on the interaction
between the nuclei is then described by an effective potential. This potential results from the
solution of the so-called electronic Schrodinger equation. As a further approximation the nuclei
are moved according to the classical Newton’s equations using either effective potentials
which result from quantum mechanical computations (which include the effects of the
electrons) or empirical potentials that have been fitted to the results of quantum mechanical
computations or to the results of experiments. All in all, this approach is a classical
example for a hierarchy of approximation procedures and an example for the use of
effective quantities. In the following, the derivation of the molecular dynamics method
from the laws of quantum mechanics is presented. For further details see the large body of
available literature, for example [

The Schrodinger Equation

Up to the end of the nineteenth century, classical physics could answer the most important
questions using Newton’s equations of motion. The Lagrange formalism and the Hamilton
formalism both lead to generalized classical equations of motion that are essentially
equivalent. These equations furnish how the change in time of the position of particles
depends on the forces acting on them. If initial positions and initial velocities are given, the
positions of the particles are determined uniquely for all later points in time. Observable
quantities such as angular momentum or kinetic energy can then be represented as
functions of the positions and the impulses of the particles. In the beginning of the
twentieth century the theory of quantum mechanics was developed. There, the dynamics of
the particles is described by a new equation of motion, the Schr¨odinger equation. In
contrast to Newton’s equations its solution no longer provides unique trajectories, meaning
uniquely determined positions and impulses of the particles, but only probabilistic
statements about the positions and impulses of the particles. Furthermore, position and
impulse of a single particle can no longer be measured arbitrarily accurately at the same
time (Heisenberg’s uncertainty principle) and certain observables, as for example the
energies of bound electrons, can only assume certain discrete values. All statements that
can be made about a quantum mechanical system can be derived from the state function (or
wave function) Ψ which is given as the solution of the Schr¨odinger equation. Let us
consider as an example a system consisting of N nuclei and K electrons. The time-
dependent state function of such a system can be written in general as
Ψ = Ψ(R1, . . . ,RN, r1, . . . , rK, t),
where Ri and ri denote positions in three-dimensional space R3 associated to the ith nucleus
and the ith electron, respectively. The variable t denotes the time-dependency of the state
function. The vector space (space of configurations) in which the coordinates of the
particles are given is therefore of dimension 3(N + K). In the following we will abbreviate
(R1, . . . ,RN) and (r1, . . . , rK) with the shorter notation R and r, respectively. According to
the statistical interpretation of the state function, the expression

Ψ∗(R, r, t)Ψ(R, r, t)dV1 · · · dV N+K

describes the probability to find the system under consideration at time t in the volume
element dV1 · . . . · dVN+K of the configuration space centered at the point (R, r). By
integrating over a volume element of the configuration space one determines the probability
to find the system in this domain
We assume in the following that nuclei and electrons are charged particles. The
electrostatic potential (Coulomb potential) of a point charge (with elementary charge +e) is

ex 1
where r is the distance from the position of the charged particle and ϵ 0 is the
4 π ϵ0 r
 1
dielectric constant. is also called Coulomb constant. An electron moving in this
4 π ϵ0

−e x 1
potential has the potential energy V (r) = . Neglecting spin and relativistic
4 π ϵ0 r
interactions and assuming that no external forces act on the system, the Hamilton operator
associated to the system of nuclei and electrons is given as the sum over the operators for
the kinetic energy and the Coulomb potentials,

2 K
e2 K 1 e2 K N Zj
H ( R, r ) : 
2me
  rk 
4
 
4
 
k 1 o k  j rk  rj o k 1 j 1 rk  r j

e2 N
Z k Zi 2 N 1
4 0

k j Rk  R j
 
2 k 1 M k
 Rk

Here, Mj and Zj denote the mass and the atomic number of the jth nucleus, me is the mass

h
of an electron and ℏ= with h being Planck’s constant. ||rk – rj|| are the distances between
2π
electrons, ||rk – Rj|| are distances between electrons and nuclei and ||Rk −Rj|| are distances
between nuclei. The operators ΔRk and Δrk stand here for the Laplace operator with respect
to the nuclear coordinates Rk and with respect to the electronic coordinates r k In the
following we will denote the separate parts of (2.2) in abbreviated form (written in the same
order) with
H = Te + Vee + VeK + VKK + TK.
The meanings of the individual parts are the following: Te and TK are the operators of the
kinetic energy of the electrons and of the nuclei, respectively. V ee, VKK and VeK refer to the
operators of the potential energy of the interactions (thus the Coulomb energy) between
only the electrons, between only the nuclei, and between the electrons and the nuclei,
respectively. The state function Ψ is now given as the solution of the Schrodinger equation
∂ ψ [ R , r ,t ]
ⅈℏ =Hψ [ R ,r , t ]
∂t
where i denotes the imaginary unit. The expression ΔRkΨ(R, r, t), which occurs in HΨ,
stands there for ΔYΨ(R1, . . . ,Rk−1,Y,Rk+1, . . . ,RN, r, t)|Rk that is, the application of the
Laplace operator to Ψ seen as a function of Y (the kth vector of coordinates) and the
evaluation of the resulting function at the point Y = Rk. The operators Δrk and later ∇Rk
and others are to be understood in an analogous way.
In the following we consider the case that the Hamilton operator H is not explicitly time-
dependent, as we already assumed in (2.2). Then, the separation approach
Ψ(R, r, t) = ψ(R, r) · f(t)
of Ψ with a function ψ = ψ(R, r) that does not depend on time and a function f = f(t) that
depends on time when substituted into (2.4) gives rise to
df (t )
i  ( R, r )  f (t )H  ( R, r )
dt
since H does not act on f(t).3 A formal division of both sides by the termψ(R, r) · f(t) _= 0
yields
1 df (t ) 1
i  H  ( R, r )
f (t ) dt  ( R, r )
The left hand side contains only the time coordinate t, the right hand side only the
coordinates in space. Therefore, both sides have to be equal to a common constant E and
(2.7) can be separated.We obtain the two equations
1 df (t )
i E
f (t ) dt
And
H  ( R, r )  E ( R, r )
The differential equation (2.8) describes the evolution over time of the wave function. Its
general solution reads
f (t )  ce iEt / 
Equation (2.9) is an eigenvalue problem for the Hamilton operator H with the energy
eigenvalue E. This equation is called time-independent (or stationary) Schr¨odinger
equation. To every energy eigenvalue En there is one (or, in the case of degenerated states,
several) associated energy eigenfunctions ψn. Also, for every energy eigenvalue En, (2.10)
yields a time-dependent term fn. The solution of the time-dependent Schrodinger equation
(2.4) is then given as a linear combination of the energy eigenfunctions ψn and the
associated time-dependent terms fn of the form

 ( R, r , t )   ce  iEnt /  n ( R, r )
n

cn   * ( R, r ) ( R, r , 0)dRdr
with the weights n
Similar to the time-dependent
Schr¨odinger equation, (2.9) is so complex that analytical solutions can only be given for a
few very simple systems. The development of approximation procedures is therefore a
fundamental area of research in quantum mechanics. There exists an entire hierarchy of
approximations that exploit the different physical properties of nuclei and electrons [417,
626, 627]. We will consider these approximations in the following in
more detail.
A derivation of classical Molecular Dynamics

In the following we will derive, starting from the time-dependent Schrodinger equation
(2.4), the equations of classical molecular dynamics by a series of approximations. We
follow [417] and [626, 627].
The TDSCF approach and Ehrenfest’s Molecular Dynamics
First, we decompose the Hamilton operator (2.3) as follows: We set
H H e  TK
with the electronic Hamilton operator
H e : Te  Ve

Ve : Vee  Vek  VKK

is just the operator for the potential energy of the entire system. The wave function Ψ(R, r,
t) depends on the coordinates of the electrons and of the nuclei as well as on time. First, we
separate the wave function into a simple product form
 i t 
 ( R, r , t )   ( R, r , t ) : x( R, t ) (r , t ) exp   Ee (t )dt 
  ' '

  t0 
of the contribution of the nuclei and electrons to the full wave function Ψ. It is assumed that
the nuclear wave function χ(R, t) and the electronic wave function φ(r, t) are normalized

 x ( R, t ) x( R, t )dR  1
*
for any _ point in time t, that means that both and

 (r , t ) ( r , t )dR  1
*
hold. The phase factor
E e is chosen in the form

E e (t )    * (r , t ) x* ( R, t )H e (r , t ) x( R, t )dRdr

which is convenient for the following derivation of a coupled system of equations.

Now, we insert (2.14) into the time-dependent Schrodinger equation (2.4) with Hamilton
operator H, multiply from the left with φ(r, t) and χ* (R, t) and integrate over R and r.
Finally, we require conservation of energy, that
is,
d *
dt 
 H dRdr  0

and obtain thereby the coupled system of equations


i
t
 
2
k 2me
rk    x ( R, t )V ( R, r ) x( R, t )dR  
*
e

i
x
t
 
2
k 2mk
Rk x     (r, t )H
*
e 
( R, r ) ( r , t ) dr x

These equations constitute the foundation for the TDSCF approach (time-dependent self-
consistent field) introduced by Dirac in 1930, see [181, 186]. Both unknowns again obey a
Schr¨odinger equation, but now with a time-dependent effective operator for the potential
energy which arises as an appropriate average of the other unknown. These averages can
also be interpreted as quantum mechanical expectation values with respect to the operators
Ve and He and give a mean field description of the coupled dynamics. As a next step the
nuclear wave function χ is to be approximated by classical point particles. For this, we first
write the wave function χ as
i 
x( R, t ) A( R, t ) exp  S ( R, t ) 
 
with an amplitude A > 0 and a phase factor S, both real [187, 427, 536]. Substitution into
the equation for the nuclei in the TDSCF system (2.17) and separating real and imaginary
parts leads to the coupled system of equations
S N 1 N
1 Rk A
 (RK S ) 2   H e dr   2 
t k 2M K k 2M K A

A N 1 N
1
 (Rk A)(Rk S )   A( Rk S )  0
t k Mk k 2M k

S5
T
    
Rk   , , ,
   Rk    Rk    Rk  
The abbreviation (RK S )
2
Here  1 2 3 
denotes the

scalar product of RK S with itself and (Rk A)(Rk S ) denotes the scalar product of

the vectors (Rk A) and (Rk S ) This system corresponds exactly to the second equation in
5
the TDSCF system (2.17) in the new variables A and S The only term that directly
depends on  is the right hand side of equation (2.19). In the limit  → 0 equation (2.19)
gives
S N 1
  Rk S     *H e dr  0
2

t k 2M k

Setting
 S  
Rk R1 S ,...,  RN S  , this is isomorphic to the Hamilton-Jacobi for this is
isomorphic to the Hamilton-Jacobi form
S
H  R,  R S   0
t
of the equations of motion of classical mechanics with the classical Hamilton function
H ( R, P )  T ( P )  V ( R )

With P  ( P1 ,..., PN ) , where one puts

 Pk (t )  RK S  R  t  , t 
Here, R corresponds to generalized coordinates and P to their conjugated moments.
Pk   Rk V  R 
Newton’s equations of motion ˙ associated to equation (2.22) are then
dPk
 Rk   *H e dr
dt

k
 (t )  R  *H  dr
MkR k e

:  Rk Ve Ehr  R (t ) 

The nuclei move now according to the laws of classical mechanics in an effective potential
given by the electrons. This so-called Ehrenfest potential V Ehreis a function of the nuclear
coordinates R at time t. It results from an averaging over the degrees of freedom of the
electrons, weighted by He, where the nuclear coordinates are kept constant at their current
positions R(t). There is still the wave function χ of the nuclei in the equation for the
electrons in the system for the TDSCF approach (2.16). Consistency requires it to be
replaced by the position of the nuclei. Thus, if one replaces the probability density of the
nuclei |χ(R, t)|2 by the product of delta functions.
Πkδ(Rk − Rk(t)) in the limit _ → 0 in (2.16), then one obtains for example for the position
operator Rk with

 x  R, t  R x  R, t  dR 
 0
*
k  R (t ) k

the classical position Rk(t) as limit of the quantum mechanical expectations. Here, the delta
functions are centered in the instantaneous positions R(t) of the nuclei given by (2.25). For
(2.16), this classical limit process leads to a time-dependent wave equation for the electrons
R (t ) (r , t ) 2
i    r  R (t )  r , t   Ve  R  t  , r   R (t )(r , t )  H e  R(t ), r   R (t )(r , t )
t 2me k
that move in a self-consistent way with the nuclei, if the classical nuclei are propagated by
(2.25). Note that now He and therefore the wave function φ of the electrons depend
parametrically via Ve on the positions R(t) of the nuclei. The nuclei are thus treated as
classical particles, whereas the electrons are still treated using quantum mechanics. In
honor of Ehrenfest, who first posed the question how Newton’s classical dynamics could be
derived from Schrodinger’s equation, one often calls approaches that are based on the
equations
  (t )  R V Ehr ( R (t ))
MkR k k e

 R ( t )(r , t )
i H e  R(t ), r   R (t )(r , t )
t
Ehrenfest molecular dynamics. Alternatively, one finds such approaches in the literature
under the name QCMD (quantum-classical molecular dynamics model) [104, 204, 447].
Note again that the wave function φR(t) of the electrons is here not equal to the wave
function φ in (2.16), since an approximation was introduced by the limit process for the
positions of the nuclei. The wave function of the electrons depends implicitly on R via the
coupling in the system, which we expressed by the parametric notation φR(t). In the
following we will omit this parametrization for the sake of simplicity and will denote, if
clear from the context, the electronic wave function just by φ.
Expansión in the adiabatic basis
The TDSCF approach leads to a mean field theory. One should keep in mind that
transitions between different electronic states are still possible in this setting. This can be
seen as follows: We expand the electronic wave function φ from (2.31) for fixed t in an
appropriate basis {φj} of the electronic states

 R  r , t    c j  t  j  R (t ), r 
j 0

with complex coefficients

 c (t )
j
and
 j
cj  t 1
2

. The
 c  t  j
2

describe explicitly
how the occupancy of the different states j evolves over time. A possible orthonormal basis,
called adiabatic basis, results from the solution of the time-independent electronic
Schrodinger equation
H e  R, r   j  R , r   E j  R   j  R , r 
where R denotes the nuclear coordinates from equation (2.25) at the chosen time t. The
values {Ej} are here the energy eigenvalues of the electronic Hamilton operator He(R, r),
and the {φj} are the associated energy eigenfunctions.
For (2.30) and (2.31) one obtains with the expansion (2.32) the equations of motion in the
adiabatic basis (2.33) as [447, 626, 627]
  t    c  t  R E  c  t  c (t )  E  El  d k jl
2 *
Mk R k j k j
 j l j
j j ,l
ic j  t   c j  t  E j  i cl  t  R k (t )d k jl
k ,l

with the coupling terms given as

d k jl    j * Rk l

d k jj  0

Here, we used the properties

   R, r  
j
*
Rk H el ( R, R )dr   El  R   E j  R     j *  R, r Rkl  R, r  dr
N

  j  R, r l  R, r  dr   R k  t    j  R, r  Rkl  R, r  dr
* *

k 1

of the adiabatic basis and used furthermore that φ and R in V Her e (R(t)) can be treated as
independent variables. This implies that the time-dependent wave function can be
represented by a linear combination of adiabatic states and that its evolution in time is
described by the Schrodinger equation (2.31). Here, |cj(t)|2 is the probability density that
the system is in state φj at time point t.
Restriction to the Ground State