lecture march25
TRANSCRIPT
-
7/24/2019 Lecture March25
1/14
Module-3
Ab Initio MolecularDynamics
March 25
-
7/24/2019 Lecture March25
2/14
Liouvilles Theorem
x
pxdXt
dXt+
Phase space volume occupied by a collection of members ofthe ensemble will be preserved during the time evolution
-
7/24/2019 Lecture March25
3/14
Time Reversibility
x
px +t
t
Hamiltons equation ofmotion is invariant undertime reversal :
t t
PI= MIRI
dt MI
RI
d(t) =PI
dRI
d(t)
= H
PI
, d(PI)
d(t)
= H
RI
original
Hamiltonseqns. of motion
RI =
H
PI
PI = H
RI
External forces makesystem out of equilibrium,
then the resultingdynamics is irreversible
-
7/24/2019 Lecture March25
4/14
Position Verlet Algorithm Velocity Verlet Algorithm
Leap-frog algorithm
Reversible multiple time step algorithms
time reversible
numerical stability (Energy cons.) symplectic: phase space volumeconservation
Properties of integrators:
Homework:Read about the first three integrators (above) from the bookUnderstanding Molecular Simulation, Frenkel & Smith
-
7/24/2019 Lecture March25
5/14
Derivation of velocity Verlet
RI(t+t) =RI(t) + RI(t)t+1
2t
2RI(t)
RI(t) =RI(t+t) RI(t+t)t+1
2t
2RI(t+t)
RI(t+t) = RI(t) + t
2MI[FI(t) + FI(t+t)]
Let us now make a reverse step from t+t t
Substituting the first eqn. in the second, we get,
-
7/24/2019 Lecture March25
6/14
History
Alder & Wainwright(JCP, 1957)
five-term virial expression.
Some further investigation for both 32 molecule and
larger systems will be made on the present calculators,
but a satisfactory determination of the detailed behavior
in the apparent transition region will require higher
speed equipment. The possibility
that
a similar phe-
nomenon for hard spheres in two dimensions may have
been missed in the original Monte Carlo calculations
l
will also be investigated.
Work performed under the auspices of the U.
S.
Atomic
Energy Commission.
1
M. N. Rosenbluth and A.
W.
Rosenbluth,
J.
Chern. Phys. 22,
881
1954).
2
Metropolis, Rosenbluth, Rosenbluth, Teller, and Teller,
J.
Chern. Phys. 21, 1087 1953).
3 B. J. Alder and T. Wainwright, J. Chern. Phys. 27,1208 1957).
4 W. W.
Wood and F. R. Parker,
J.
Chern. Phys. 27,
720
1957).
This paper discusses the Monte Carlo method in some detail, as
well as giving computational results for Lennard-Jones molecules.
6
Kirkwood, Maun, and Alder, J. Chern. Phys. 18, 1040 1950).
Phase Transition for a Hard
Sphere System
B.
J. ALDER AND
T.
E. WAINWRIGHT
University of California Radiation LalJoratory
Livermore
California
Received August 12, 1957
A
CALCULATION of molecular dynamic motion
has been designed principally to study the re-
laxations accompanying various nonequilibrium phe-
nomena. The method consists of solving exactly to the
number of significant figures carried) the simultaneous
classical equations of motion of several hundred par-
ticles
by
means of fast electronic computors. Some of the
details as they relate to hard spheres and to particles
having square well potentials of attraction have been
described.
l
2
The method has been used also to calculate
equilibrium properties, particularly the equation of
state of hard spheres where differences with previous
Monte Carl0
3
results appeared.
detected.
t
became apparent
that
some long runs were
necessary at intermediate densities, accordingly the
IBM-704 was utilized where, for 32 particles, an hour is
required for 7000 collisions. Larger systems
of
108, 256,
and 500 particles can also conveniently be handled; in
an
hour 2000, 1000, and 500 collisions, respectively, can
be calculated. The results for 256 and 500 particles are
not now presented due to inadequate statistics.
The equation of state shown in Fig. 1 of the ac-
companying paper6 for
32
and
108
particles
is
for the
intermediate region
of
density, where disagreement was
found with the previous Monte Carlo results. The
volume, v
is
given relative to the volume of close
packing, Vo Plotted also are the more extended Monte
Carlo results; the agreement between these three sys-
tems is within the present accuracy of the pressure
determination. This agreement provides an interesting
confirmation
of
the postulates of statistical mechanics
for this system.
Figure 1 of the accompanying paper shows two
separate and overlapping branches.
In
the overlapping
region the system can,
at
a given density, exist in two
states with considerably different pressures. As the
calculation proceeds the pressure is seen to jump sud-
denly from one level to the other. A study of the posi-
tions of the particles reveals that as long as the system
stays on the lower branch
of
the curve the particles are
all confined to the narrow region in space determined
by
their neighbors, while on the upper branch of the curve
the particles have acquired enough freedom to exchange
with the surrounding particles. Since the spheres are
originally in ordered positions, the system starts out on
the lower branch; the first jump to the upper branch can
require very many collisions. The trend, as expected,
is
that at
higher densities more collisions are necessary for
the first transition, however, there are large deviations.
At vlvo 1.60, 5000 collisions were required;
at
1.55,
25000; while at 1.54 only 400; at 1.535, 7000;
at
1.53,
75 000; and at 1.525,
95
000. Runs in excess of 200 000
collisions at vivo of 1.55 and 1.53 have not shown any
P
HVSI CA I
REVIE '
VOLUME
f36,
NLJM8RR
2A l9
OCTOB
ER 1964
Correlations in
the
Motion
of
Atoms
in
Liquid
Argon*
A.
RAHMAN
Argonee
National Laboratory, Argo@me,
I/Blois
(Received 6
May
1964)
A
system
of
864
particles
interacting with a
I.ennard-Jones
potential
and
obeying
classical
equations
of
motion
has been studied on
a
digital
computer (CDC 3600)
to simulate molecular
dynamics
in
l i qu id a rgon
at 94 .
4'K
and a
density
of
1.374
g
cm
'.
The
pair-correlation f un ct io n a n d
t he c on st an t o f
self-di6usion are
found to
agree
well with
experiment;
the latter is
15
lower than
the
experimental
value. Th e
spectrum
of
the
velocity
au tocor rela tion func t ion shows a broad
maximum
in
the
frequency
region &o
=
0.25
(h&T/h).
The
shape
of
the
Van Hove
function
G,
(r,
t)
attains a
maximum
departure
from a
Gaussian
at about
s=3
0.
X1 0
sec
and
becomes a Gaussian
again
at about 10
sec.
The
Van
Hove function
Gg(r,
t)
h as b ee n
com-
pared
with
the
convolution
approximation
of
Vineyard,
showing
that
this
approximation
gives
a too
rapid
decay
of
Gd,
(r,
t)
with time. A
delayed-convolution
approximation
has
been
suggested
which
gives
a
better 6t
with
Gz(r,
I);
this delayed
convolution
makes
Gz(r, t}
decay
as
t4
at
short
times
and
as
I
at
long
times.
I.
INTRODUCTION
'N
recent
years
considerable use has been made
of
~ ~
large digi tal computers to
study
various
aspects
of
molecular dynamics
in solids, liquids,
and
gases.
The
following
is
a
description
of a
compu te r e xper iment
on
liquid
argon
(using
the CDC
3600)
to
study
the
space
and
time
dependence of
two-body cor re la ti on s
which
system,
namely,
94 .
4'K
and
1.374
g
cm
'.
A less
exhaustive
study,
at
130'IZ
and
1.
16
g
cm
',
is
men-
tioned
briefly
at the end.
II.
METHOD
OF COMPUTATION
The
calculations
reported
here
were
based on
the
following
ingredients.
A. Rahman (Phys. Rev. 1964)
first MD first realistic MD
-
7/24/2019 Lecture March25
7/14
e
o
g
e
n
s
e
v
a
.
1
0
I
0
5
IO
A
I
I
FIG.
Theoretical
(solid line)
, and experim
ental(dotte
d
line),x-ray
scatteringi
ntensitiesfo
rliquidwa
ter.Thelatte
rare
takenfrom
Narten,Ref
.29,andref
erto25C.
gooo)
,
onc2
,
nn2
o
n
s
.
2
g
-
7/24/2019 Lecture March25
8/14
-
7/24/2019 Lecture March25
9/14
MD Algorithm
a) Initialize coordinates and velocities
b) Velocity Verlet-1: R(t) !R(t+"t)
c) Compute force (acceleration) for R(t+"t)
d) Velocity Verlet-2: V(t)!V(t+"t)
e) Go to step (b) or stop
RI(t+ t) = RI(t) +t
2
hRI(t) + RI(t+ t)
iRI(t+t) = RI(t) + RI(t)t+
1
2RI(t)t
2
& compute force
-
7/24/2019 Lecture March25
10/14
Time Step = 1
=c
= 4000 cm1
= 8
.33 10
15s 10 fs
t /10 = 1 f s
0 10 20 30 40 50Time (fs)
0.5
0.6
0.7
0.8
0.9
1
r[H-H](Ang.)
t
fastest motion
determines thetime step
vibrational freq. of H2:
period of oscillation of H2:
Time step should at least (equally) divide a periodfor better capture of the features of dynamics
-
7/24/2019 Lecture March25
11/14
Initial State
Initial coordinates:X-ray structure,chemical intuition, arbitrary/random
structures
Initial velocities: from Boltzmanndistribution of velocities for a given
initial temperature T0=> this can be
generated from Box-Muller sampling(see next page)
-
7/24/2019 Lecture March25
12/14
uniform
rand. number
between 0 & 1
Box-Muller Sampling
f(v) =
M
2kBT
1/2
exp
Mv2
2kBT
X=p2 ln 0
2cos(21)
Y =p2 ln 0
2sin(21)
=pkBT/M
uniform rand.
numbers
between 0 & 1
This is a gaussian with
standard deviation
Maxwell distribution of velocities:
velocity
f(
v)
0X Y
X and Y are two random velocities obtained bypicking randomly from the distribution f(v)
all quantities areknown for a givensystem; M is themass of an atom
-
7/24/2019 Lecture March25
13/14
If you have two atoms, we require 6 values
of velocities (each for a degree of freedom) For this, we generate 6 uniform random
numbers from computers:
Substituting these random numbers in thetwo Box-Muller equations above [a pair ata time], we obtain, 6 values for velocities
(with a probability according to Boltzman
distribution):
Assign these values of X and Y to eachdegree of freedom (in any order that you
like)
1, , 6
X1, Y1, X2, Y2,X3, Y3
-
7/24/2019 Lecture March25
14/14
Temperature
Equipartition principle
Instantaneous
temperature T(t) = 1
3NkB
N
XI
MIR2
I
limt
1
t
tX
NX
I=1
1
2MI
R2I()
T= limt
1
t
tX
T()Macroscopic
temperature
1
3N*
N
XI=1
1
2
MIR2
I+ = 12
kBT
1
3N