DL_POLY advanced training course

baconossifiedMechanics

Oct 29, 2013 (3 years and 9 months ago)

1,047 views

03/08/2005

NCHC DL_POLY Training Course

1

DL_POLY advanced training course
and summaries of the “2005 Methods in
Molecular Simulation Summer School”

Jen
-
Chang Chen

E
-
mail:
jcc67@caece.net

03/08/2005
1
1

03/08/2005

NCHC DL_POLY Training Course

2

Acknowledgements


Thanks to CCP5 and Marie Curie Actions
for financial support of the summer school.


Thanks to Dr. Bill Smith and Dr. Rong
-
Shan Qin for informative discussions.


I also appreciate Dr. David Willock and
Prof. John Harding for their kind help.

03/08/2005

NCHC DL_POLY Training Course

3

Outlines


DL_POLY: A quick review


Practical workshop 1
-

DL_POLY


Inter
-
atomic potentials


Long ranged forces


Optimization


Hyperdynamics


Non
-
equilibrium MD


Meso
-
scale methods


Practical workshop 2
-

DL_MESO

03/08/2005

NCHC DL_POLY Training Course

4

Suggested Reading I


Understanding Molecular Simulation
, D. Frenkel and
B. Smit, Academic Press 2002.


The Art of Molecular Dynamics Simulation
, D.C.
Rapaport, Cambridge 2004.


Computer Simulation of Liquids
, M.P. Allen and D.J.
Tildesley, Oxford 1989.


Molecular Dynamics Simulation: Elementary
Methods
, J.M. Haile, Wiley 1997.


A Guide to Monte Carlo Simulations in Statistical
Physics
, D.P. Landau and K. Binder, Cambridge 2000.


Molecular Modelling and Simulation
, Tamar Schlick.
Springer Verlag 2002.

03/08/2005

NCHC DL_POLY Training Course

5


Suggested Reading II


Classical mechanics
, H. Goldstein, C. Poole and J. Safko, Addison
Wesley 2002.


Theory of Simple Liquids
, J.
-
P. Hansen and I.R. McDonald,
Academic Press 1986.


Statistical Mechanics of Nonequilibrium Liquids
, D. Evans and
G. Morriss, Academic Press 1990.



Statistical mechanics: a survival guide
, Mike Glazer and Justin
Wark, Oxford 2001.


Statistical Mechanics an Introduction
, David H. Trevena,
Harwood Publishing Ltd 2003.


An introduction to Statistical Thermodynamics
, R. Gasser and
W. Richards, World Scientific 1995.


Statistical Mechanics
, D. A. McQuarrie, Harper Collins 1976.


An introduction to statistical thermodynamics
, Terrell L. Hill,
Dover 1986.


Statistical mechanics : a concise introduction for chemists
, B.
Widom, Cambridge, New York 2002.

03/08/2005

NCHC DL_POLY Training Course

6

A survival guide of DL_POLY I


Please read the user manual and review the DL_POLY
website at first.


http://www.cse.clrc.ac.uk/msi/software/DL_POLY/


Second, read “A DL_POLY tutorial” and “FAQ.”


Finally, join the DL_POLY forum.


http://www.cse.clrc.ac.uk/disco/forums.shtml


A Good program library for MD, MC, LS and LD is in the
following URL.


http://www.ccp5.ac.uk/librar.shtml


If you like to receive the information regarding the
activities of CCP5, check the following link.


http://www.cse.clrc.ac.uk/database/CCP5
-
mailinglist
-
form.jsp


03/08/2005

NCHC DL_POLY Training Course

7

A survival guide of DL_POLY II


Input files

The DL_POLY_2 User Manual, Version
2.14, P.107, 2003



CONFIG
: Fixed format (Mandatory)



CONTROL
: Free format (Mandatory)



FIELD
: Fixed format (Mandatory)



TABLE
: Fixed format (An alternative way


to specify short
-
ranged potentials)


For a restart job
:



REVCON
: Rename this file to the



CONFIG file (Formatted)



REVIVE
: Rename or copy it to the



REVOLD file (Unformatted)



REVOLD
: Unformatted file



Utilize the DL_POLY Java GUI.


03/08/2005

NCHC DL_POLY Training Course

8

A survival guide of DL_POLY II (cont.)


Output files

The DL_POLY_2 User Manual, Version
2.14, P.107, 2003



REVCON
: The restart configuration



OUTPUT
: Simulation results



HISTORY
: The history of atom position,


velocity and force



STATIS
: The history of statistical data



RDFDAT
: The RDF data



ZDNDAT
: The Z density profile



REVIVE
: The accumulated statistical



data (unformatted file)

03/08/2005

NCHC DL_POLY Training Course

9

A survival guide of DL_POLY III


DL_POLY Java GUI provides “analysis” module!


You can write a program to do pre
-
processing or
post
-
processing easily.


For example, I wrote few python scripts to produce
the CONFIG file from GULP program, to extract
statistical data and to generate a multi
-
frame XYZ file.


For visualization, VMD is one of good tools.


http://www.ks.uiuc.edu/Research/vmd/


You need a program to parse the HISTORY
file to a VMD input file.


http://www.ks.uiuc.edu/Research/vmd/plugins/molfile/

03/08/2005

NCHC DL_POLY Training Course

10

Example: XYZ file format

2 ## Total number of atom in this frame

Step 1 ## Comment

Ar 0.0 0.0 0.0 ## Symbol X Y Z

Ar 0.5 0.0 0.0 ## Symbol X Y Z

2 ## Total number of atom in the SECOND frame

Step 2

Ar 0.1 0.1 0.1

Ar 0.5 0.5 0.5

03/08/2005

NCHC DL_POLY Training Course

11

Example: his2xyz.py

#!/usr/bin/env python

## Usage:

## I. chmod 755 his2xyz.py

## II. ./his2xyz.py HISTORY HIS.xyz

import sys,string

atomList=['Si','Al','O_','O','H','C','Cl','Na','Zr','Y_','Mg','K','Ag']

inputFile=open(sys.argv[1],'r')

outFile=open(sys.argv[2],'w')

title=inputFile.readline()

line=inputFile.readline()

while(line!=""):


if string.split(line)[0]=='timestep':



timestep='step= '+string.split(line)[1]+'
\
n'



totalAtom=string.split(line)[2]+'
\
n'




outFile.write(totalAtom)



outFile.write(timestep)


if string.split(line)[0][:2] in atomList:



atomName=string.split(line)[0][:2]+'
\
t'



outFile.write(atomName)



xyzline=inputFile.readline()



outFile.write(xyzline)


line=inputFile.readline()

inputFile.close()

outFile.close()

03/08/2005

NCHC DL_POLY Training Course

12

Practical workshop 1
-

DL_POLY


Subject:
Methane in silicalite


Objectives:


Ensure that you are familiar with DL_POLY.


Calculate diffusion coefficient by using Einstein
equation.


Apply VMD or your favorite visualization tool to
see the process.

03/08/2005

NCHC DL_POLY Training Course

13

Practical workshop 1
-

DL_POLY (cont.)


Step 1: Download “nchc.tar.gz” into your cygwin folder.


Step 2: Do “tar zxvf nchc.tar.gz”


Step 3: Copy all the files into “execute” folder.


Step 4: Run DL_POLY Java GUI to see the CONFIG file.


Step 5: Use DL_POLY Java GUI or your favorite editor to
see the FIELD and CONTROL file.


Setp 6: Run “DL_POLY.X”


Step 6: Check all output files.


Step 7: Run DL_POLY Java GUI to obtain the MSD file of
“C_1” and “H_1”.


Step 8: Use your favorite visualization tool to see the
diffusion process.


./his2xyz.py HISTORY his.xyz

03/08/2005

NCHC DL_POLY Training Course

14

Practical workshop 1
-

DL_POLY (cont.)


Step 9: Utilize EXCEL to open the MSD0.XY and
MSD1.XY and draw the MSD vs. Time plots. Please note
the slope of the plot.


Einstein equation:



03/08/2005

NCHC DL_POLY Training Course

15

Inter
-
atomic potentials


Potentials are algorithms to generate energy
surfaces.


The potential terms should reflect the physics of
the system and often contain a lot of
approximations.


In the same species, different problems might
apply different potentials.


In molecular simulation, you should have a
strong feeling about potentials.


Wrong potentials == Wrong physics (GIGO)

03/08/2005

NCHC DL_POLY Training Course

16

Basic criteria for selecting potentials


Physical meanings


Computing Efficiency (computing cost)


Think about a MD algorithm.


Accuracy


Transferability


are the potential models cheap, accurate and
common in the same species, systems,
phases or configurations?

03/08/2005

NCHC DL_POLY Training Course

17


Where do you find the potentials?


Literatures


Ionic solids:
http://www.ri.ac.uk/DFRL/research_pages/resources/Potential_database/index.html


G. V. Lewis and C. R. A. Catlow, “Potential Models for Ionic Oxides,” J. Phys.
C: Solid State Phys.,18(1985), 1149
-
1161.


C. R. A. Catlow and A. M. Stoneham, “Inoicity in Solids,” J. Phys. C: Solid
State Phys.,16(1983), 4321
-
4338.


Chap. 8~Chap. 11, Lecture Notes in Physics Vol. 166: Computer Simulation of
Solids, edited by C.R.A. Catlow and W.C.Mackrodt, Springer
-
Verlag,1982.


Metals


Embedded Atom Model


Sutton
-
Chen Model, A. P. Sutton and J. Chen, Phil. Mag. Lett., 61(1990), 139.


Polymer


http://msdlocal.ebi.ac.uk/docs/mmrefs.html


Fitting from ab
-
initio calculations of structure


Fitting from experimental data


Direct calculation of each term in the potential

03/08/2005

NCHC DL_POLY Training Course

18

Before you start fitting


What are you going to fit to?


Experiment:


How good is the data you are fitting to?


Is bulk crystal data good enough to describe your
situation?


What data are you using to validate the results?


Calculation:


Were the calculation done properly?


Does the method used include the effects you are
fitting to?


How comprehensive is the set of configurations you
have?


What functional form are you going to use?

Note: This slide copied from the page 5 of Prof. John Harding’s handout.

03/08/2005

NCHC DL_POLY Training Course

19

When you start fitting


Fitting is an optimization problem


Find local
minima.


Make sure that the fitting process cannot get
into unphysical regions of parameter space.


Global fitting processes do not ensure that the
right physics goes to the right part of the
model.


Remember that the potential cutoff is part of
the model. Please quote it in your work.


Note: This slide copied from the page 15 of Prof. John Harding’s handout with modifications.

03/08/2005

NCHC DL_POLY Training Course

20

How do we fit the potentials?


GULP, General Utility Lattice Program,
provides a fitting function!



http://gulp.curtin.edu.au/


We can determine the potentials
parameters by fitting to data from ab
-
initio
calculation, normally by attempting to
reproduce energy surface.


We can derive empirical potentials by
trying to reproduce experimental data.

03/08/2005

NCHC DL_POLY Training Course

21


Ab
-
initio calculations for fitting data


Calculate the energies of a number of different
configurations relevant to the problem needed.


Fit potential using energies and structures


Calculate a dynamics trajectory.


Fit the forces and also properties to those calculated
bt yhe potential form chosen.


You should know about….


The ab
-
initio calculation doesn’t mean absolutely
accurate.


Note: This slide copied from the page 14 of Prof. John Harding’s handout with modifications.

03/08/2005

NCHC DL_POLY Training Course

22

Long ranged forces


Long ranged forces arise from electrostatics
and hydrodynamics.


For electrostatics:




The direct sum is inefficient.


Ewald summation is excellent.


Reaction field has a good physical basis.


Fast multipole method is quite slow.

03/08/2005

NCHC DL_POLY Training Course

23

Ewald sum I


U=Reciprocal space term +


Real space term


Self
-
interaction term

03/08/2005

NCHC DL_POLY Training Course

24

Ewald sum II

+

-

+

-

+

-

+

-

-

+

-

-

+

=

+

C

C+G

-
G

03/08/2005

NCHC DL_POLY Training Course

25

Ewald sum III


The screened charge (C+G) terms
are summed directly and represent
short ranged

interactions in real
space.


Poisson’e equation is used to solve
the purely gaussian terms and this
provides a
short ranged

solution in
reciprocal space.

03/08/2005

NCHC DL_POLY Training Course

26

The adaptations of Ewald sum


Ewald
-
Kornfeld method deals with
dipolar molecules


Smoothed particle mesh Ewald is a
particle mesh method using FFT to
handle reciprocal space.


Hautman
-
Klein method for surfaces
and interfaces.

03/08/2005

NCHC DL_POLY Training Course

27

Smoothed Particle Mesh Ewald I


SPME is the conversion of the
reciprocal space component of the
Ewald sum into a FFT form.

03/08/2005

NCHC DL_POLY Training Course

28

Smoothed Particle Mesh Ewald II


SPME is generally faster than conventional
Ewald sum in most applications. Algorithm
scales as O(NlogN)


In some case, the 3D array can be large.


For efficiency, always use a proper FFT
package.

03/08/2005

NCHC DL_POLY Training Course

29

Optimization


Optimization is the methodology to find
the minimum energy and/or to search the
saddle points.


A saddle point is a minimum in all conjugate
directions except one where it is a maximum.


Minimization methods:


Golden section method, direction set method,
simplex method, conjugate gradient methods
(steepest descent method), Newton
-
Raphson
(variable metric) method, Levenberg
-
Marquandt
method, genetic algorithms and simulated annealing
method.

03/08/2005

NCHC DL_POLY Training Course

30

How to select a minimization method


The selection criteria depend on the
following issue.


Derivative data


Memory storage


Noisy

03/08/2005

NCHC DL_POLY Training Course

31

Minimization method

Methods

Conditions

Golden section method
Direction set method

Simplex method

1.
You only care about function value.

2.
Function is very noisy.

3.
You have storage problems.

Conjugate gradient methods

Steepest descent method

1.
You

d like to obtain function values and 1
st

derivatives.

2.
Function is reasonably well behaved.

Newton
-
Raphson method
Levenberg
-
Marquandt method

1.
You want to obtain function values, 1
st

and 2
nd

derivatives.

2.
You don

t care about noise and memory storage.

Genetic algorithms

1.
Your system has many local minima.

2.
You

d like to evaluate the global one and obtain
only the function values.

Genetic algorithms

Simulated annealing method

1.
Your system has many local minima.

2.
You

d like to evaluate the global one and obtain
the derivatives.

03/08/2005

NCHC DL_POLY Training Course

32

Methods for finding the saddle point


This part is very important for hyper
-
dynamics.


Nudged elastic band method


The NEB method is to find the location of the two
basins separated by the saddle points.


Dimer methods and Mode
-
following methods


If we only know the starting basin, we can use dimer
method or mode
-
following method to find all the
saddle points that lead out of the basin.

03/08/2005

NCHC DL_POLY Training Course

33

Hyperdynamics


References:


Dr. Arthur F. Voter
http://www.t12.lanl.gov/home/afv/


A.F. Voter, F. Montalenti and T.C. Germann,
"Extending the Time Scale in Atomistic Simulation of
Materials," Annu. Rev. Mater. Res., 32 (2002), 321
-
346.


Hyperdynamics is a method to deal with the
simulation of processes that consist of rare events.
The major concern is to know the long timescale (ms
-

μ
s) behaviors rather than short timescale (ps).

03/08/2005

NCHC DL_POLY Training Course

34

Rare
-
event and Transition state theory


Rare event or infrequent event system


The system spends the most of time exploring a
single energy basin. At some point in time, transitions
to neighboring basins are very rapid.


Transition state theory


It assumes a saddle hyper
-
surface, once it has
crossed it, cannot return.

03/08/2005

NCHC DL_POLY Training Course

35

Hyperdynamics methods


Temperature Accelerated Dynamics


Uses a high
-
temperature simulation to construct a faithful
representation of a low temperature trajectory by extrapolating
the processes of high
-
temperature back to low temperature and
assuming that the escape probabilities from a basin are govern
by first order kinetics.


Parallel replica


Replicate a basin into many processors. Then run all systems
until one shows a transition. Stop all processors and replicate
new state to all processors and continue.


Bias potential hyperdynamics


Add a bias potential. Provided this goes to zero at the saddle
-
plane, the dynamics of the biased system has the same relative
escape probabilities as the real system, but runs faster.

03/08/2005

NCHC DL_POLY Training Course

36

Non
-
equilibrium MD


References:


D. J. Evans and G. P. Morriss, “Non
-
Newtonian Molecular dynamics,” Computer
Phys. Rep., 1(1984), 297
-
343.


D. J. Evans and G. P. Morriss, Statistical
Mechanics of Nonequilbrium Liquids,
Academic Press, 1990.


http://www.phys.unsw.edu.au/~gary/book.html

03/08/2005

NCHC DL_POLY Training Course

37

NEMD


Enable an understanding of irreversible processes.


Calculation of transport properties.


Field dependent transport properties


Two main approaches


Direct NEMD


Replicate the non
-
equilibrium process exactly as it occurs in nature.


It’s not useful for obtaining bulk transport coefficients because fluid
is not homogeneous, and density and temperature gradients are
often present making interpretation difficult.


Synthetic Field NEMD


Invent a fictitious field that drives the system away from equilibrium.


Linear response theory can then be used to obtain the transport
coefficient.

03/08/2005

NCHC DL_POLY Training Course

38

Meso
-
scale methods I


References:


D. M. Heyes, J.B. Baxter, U. T
ü
zün and R. S. Qin,
“ Discrete
-
element method simulation: from micro to
macro scales,” Philosophical Transactions of the
Royal Society London A

, 362(2004),1853
-
1865.


S. Succi, The Lattice Boltzmann Equation, Oxford
Science Publications, 2001.


P. Espanol,P. Warren, “Statistical
-
mechanics of
dissipative particle dynamics,” Europhys. Lett., Vol.
30(1995), 191
-
196.


J. Monaghan, “Smoothed Particle Hydrodynamics,”
Annu. Rev. Astron. Astrophys., 30 (1992), 543
-
574.

03/08/2005

NCHC DL_POLY Training Course

39

Meso
-
scale

Length: 10nm ~ 10
μ
m and Time: 10ns ~ 1ms

S. O. Nielsen et al., “Coarse grain models and the computer simulation of soft materials,”
J. Phys. Con. Matter,16 (2004), R481
-
R512.


03/08/2005

NCHC DL_POLY Training Course

40

Meso
-
scale methods II


Hydrodynamics and thermodynamics are the
most important roles.


There are three meso
-
scale methods


Lattice Boltzmann Equation


Dissipative Particle Dynamics


Smoothed Particle Hydrodynamics

03/08/2005

NCHC DL_POLY Training Course

41

Lattice Boltzmann Equation


LBE is a course grained lattice gas automation.


Fluid properties are mapped onto a discrete lattice.


Physical state at each lattice are described by a set of
particle distribution functions.


Macroscopic fluid variables are defined via moments of
the distribution function.


System evolution toward its equilibrium though the
relaxation of the distribution function to its equilibrium
form.


CFD people see LBE as a numerical solver for the
Navier
-
Stokes equations.

03/08/2005

NCHC DL_POLY Training Course

42

Dissipative Particle Dynamics


N particles move in continuous space and discrete time
-
steps. => Newton’s equation of motion! => MD


The particles are acted upon by three two
-
particle forces.
Each of these forces conserves momentum and angular
momentum and acts as pairwise.


Three inter
-
particle forces


Conservative Force


This force conserves energy and is derived from a potential energy
similar to that in MD and is a soft
-
repulsive potentials.


Dissipative Force


This force is to slow the particles and always removes energy, as
measured from their centre
-
of
-
mass frame.


Random Force


This force acts between all pairs of particles and is uncorrelated
between different pairs. It adds energy to the system on average.
Together with the dissipative force, this acts in some sense as a
thermostat for the system.

03/08/2005

NCHC DL_POLY Training Course

43


Smoothed Particle Hydrodynamics


SPH is a particle method and it doesn’t need
grids/meshes to calculate spatial derivatives.


Physical system can be represented by a small
set of properties defined at a finite set of discrete
points.


The interpolation procedure allows any function
at one of these points to be expressed in term of
its values at neighboring points through an
interpolation kernel.


Advantage


Spatial gradients are calculated analytically and we don’t
need to utilize finite differences method and grids.

03/08/2005

NCHC DL_POLY Training Course

44

Applications


Lattice Boltzmann


Complex boundary system


Multi
-
component phase separation


Thermodynamic problem


Dissipative Particle Dynamics


Colloid processing


Polymer system: Interactions between polymer and fiber


Biological system: Interactions between membranes and
lipid acid.


Smooth Particle Hydrodynamics


The large displacement and high speed issue


For example, explosion simulation

03/08/2005

NCHC DL_POLY Training Course

45

Introduction to DL_MESO program


DL_MESO is a general purpose mesoscopic
simulation package developed at Daresbury
Laboratory by Dr. Rongshan Qin.


The program contains three methods: Lattice
-
Boltzmann Equation,
Dissipative Particle Dynamics
and Smoothed Particle Hydrodynamics
.


Current version is DL_MESO 1.2.


DPD will include from Version 2.0 and SPH will
release from Version 3.0.


It is free for academic research.


The code is written by C++ with MPI.

03/08/2005

NCHC DL_POLY Training Course

46

How do we apply the DL_MESO


http://www.ccp5.ac.uk/dlmeso/code.html


For a academic researcher, please
download the license. Sign it and then
fax/mail it to Dr. Qin. Dr. Qin will send you
the code.


Commercial organizations interested in
acquiring the code should contact Dr. Qin.

03/08/2005

NCHC DL_POLY Training Course

47

Practical workshop 2
-

DL_MESO


Thanks to Dr. Rong
-
Shan Qin for providing us all sort
of course materials.


The binary code will expire on August 07.


In this workshop, we have six exercises.


LBE


2D pressure


2D shear


3D phase separation


3D shear


Flow over obstacle


DPD


Phase separation

03/08/2005

NCHC DL_POLY Training Course

48

Practical workshop 2
-

DL_MESO (cont.)


Step 1: Do “tar zxvf nchcmeso.tar.gz”


Step 2: Run “java xdlmeso” The GUI can only generate
the LBE input files at this version.


Step 3: To edit a file. Run “java delmeso.”


Step 4: We need two file to run LBE: lbin.sys and
lbin.spa. “Define System” generates “lbin.sys”, and “Set
Space” creates “lbin.spa”


Step 5: Go to the exercise folder: 2D_Pressure;
2D_Shear; 3DPhaseSep; 3DShear; FlowOverObstacle;
and SIMPLE_DPD.


Step 6: Use an editor to see the input files. All the input
files are well
-
prepared by Dr. Qin.


Step 7: To run LBE. Go into the exercise folder, and type
“../slbe.exe”

03/08/2005

NCHC DL_POLY Training Course

49

Practical workshop 2
-

DL_MESO (cont.)


Step 8: Run “java lbe2dplot” to plot the
calculation results.


Step 9: In order to run DPD, go into the
“SIMPLE_DPD” folder and type
“../sdpd.exe”.


Step 10: Use VMD to open the “outpos.dat”
file, and you can observe the dynamic
process.