Thermodynamics and Materials Modeling
Coarse Grain Simulations
DPD Simulations
(Dissipative particle dynamics)
Machine Learning
Data Mining Techniques
1
Partial differential equations
Coarse Grain Simulations
DPD Simulations
DPD Dissipative particle dynamics
KMC Kinetic
Machine Monte Carlo
Learning
Smoothed DPD for fluid surfaces
DataSmoothed
SPH Mining Techniques
particle hydrodynamics
2
Coarse Grain Simulations
DPD Simulations
Machine Learning
Data Mining Techniques
3
Intermolecular/inter-particulate potentials
4
Molecular Solid Intramolecular interactions
Harmonic Harmonic
Potential Potential
2-body 3-body
4-body
torsion Coulomb
Long-
range
Van der Waals
Repulsion and
Pauli
dispersion Intermolecular interactions
5
Ionic Solids
Buckingham Potential (Rather than 6-12 potential)
6
Energy Minimization/Lattice Statics
Static limit at 0K no vibrations
Yields the low temperature 0K phase
Zi are length of bonds, separation distances, angles, positions
7
Elevated Temperatures and Lattice Dynamics -SUV
H A
-pGT
Monte Carlo
Molecular Dynamics
Lattice Dynamics
At p = 0
Quasi-harmonic approximation: at T we can write:
Good up to 2Tm/3 where anharmonic effects
become important
Simple Harmonic Oscillator
Zero point Phonon frequencies
energy at wave vector q
From Chapter 8.2
Frequency can be obtained from the potential
function as a function of atomic positions 8
Lattice dynamics example
Tm = 1536 K
2/3 Tm = 1024 K
N. L. Allan, G. D. Barrera, J. A. Purton, C. E. Sims and M. B.
Taylor, Phys. Chem., Chem. Phys. 2000, 2, 1099
9
-SUV
H A
-pGT
Allows for experimental verification
10
Solid solutions and non-stoichiometric oxides
1) Fix unit cell size (super cell, 4 atoms, 8 atoms, 64 atom etc.)
for a given composition find all random arrangements “k”
2) With lattice vectors and positions as variable at T minimize Gk
3) For all “K” arrangements find the average properties
4) Vary cell size and find convergence with larger cell sizes
N. L. Allan, G. D. Barrera, J. A. Purton, C. E. Sims and M. B.
Taylor, Phys. Chem., Chem. Phys. 2000, 2, 1099
E. Bakken, N. L. Allan, T. H. K. Barron, C. E. Mohn, I. T. Todorov
11
and S. Stølen, Phys. Chem., Chem. Phys. 2003, 5, 2237.
Monte Carlo Method
Periodic Boundary Conditions
Fix T, V, N
“Z” is a state of the system
1) Move atoms at random
2) Calculate “Q”
3) Take the average
This doesn’t work because P(Z) depends on U(Z) and T
Low energy states have more weight
Metropolis Algorithm
Bias the probability with
12
Monte Carlo Method
Periodic Boundary Conditions
Fix T, V, N
“Z” is a state of the system
1) Calculate f(Z) by molecular mechanics with potentials
2) Accept a configuration “Z” if it has a low energy relative to kT with some randomness
3) Calculate the average
1) Start with a random configuration calculate f(Z)
2) Move one atom or molecule or group of molecules
3) Calculate f(Z’) if lower than f(Z) accept
4) If higher than f(Z) calculate exp(-Df/kT) and a random number from 0 to 1
5) If lower than random number accept
6) Repeat
13
Monte Carlo Method
Ising Model Simulation
1) Start with a random configuration
calculate f(Z)
2) Move one atom or molecule or group of
molecules
3) Calculate f(Z’) if lower than f(Z) accept
4) If higher than f(Z) calculate exp(-Df/kT)
and a random number from 0 to 1
5) If lower than random number accept
6) Repeat
14
Molecular Dynamics
1) Generate initial condition with particles identified by position and velocity
2) Calculate the force on each particle using potentials
3) Forces (accelerations) remain constant for a time step, position and velocity change
4) Repeat 3) until temperature is constant
5) After steady state record velocities and positions so that <r2> = 6Dt is found
Time calculation is on the order of nanoseconds.
Neither Monte Carlo nor Molecular Dynamics can calculate the free energy since they ignore
large energy regions of phase space
They can calculate differences in free energy for phase diagram construction
15
Thermodynamic Perturbation Method
Which is more stable, B) MgO or A) a mixture of MnO and MgO?
1) Simulate B using Monte Carlo Metropolis Method, calculate AB
2) Temporarily substitute the potential functions for A and calculate AA
3) Find the difference, AA – AB to determine stability.
This can work if (AA – AB) < kT
If not, then use a coupling parameter between 0 and 1
16
Thermodynamic Integration
17
Quantum mechanical/ab initio methods
1) Electronic wavefunction is independent of the nuclei since electrons are much
smaller and move much faster: Born-Oppenheimer Approximation
2) Solve the Schrodinger equation
Hamiltonian in atomic units:
ri electron positions; da nuclear positions, Za nuclear charge
Kinetic Energy – e- nuc. attraction + e- e- repulsion + Nuc. Nuc. repulsion
3) Solve approximately since true wave function can’t be found directly. Compare
proposed function results with data. Variational Principle: lowest energy wins.
4) Obey Pauli exclusion principle.
18
Density functional theory
1) Ground state can be obtained through minimization of E(r) of r(r)
2) Parallel non-interacting system (NIS)
3) Write the energy functional as
KE of NIS + e- nuc. int. + Coulomb + exchange correlation energy
4) Minimize E[r] to obtain wave functions then iterate to obtain the ground
state density and energy
Exchange energy is energy of swapping identical electrons
Correlation energy is energy of interaction with all other electrons, how is
the movement of one electron impacted by all other electrons
19
Neither of these are known so you use a Low-Density Approximation (LDA)
Molecular Dynamics In order to do molecular dynamics, you do not write a program.
LAMMPS Software is available to perform the calculations. LAMMPS
20
Free access to graphical processing units (GPU)
Colaboratory
21
HOOMD-blue MD simulations
22
Dissipative particle dynamics
Off-lattice particles moving in space and time
A particle is a molecule or fluid region (not a single atom)
Particles experience dissipative and random forces (details are integrated)
Access to long times and distances
Up to 0.1 micron and 50 microseconds
Non-bonded forces within a set cutoff distance are considered for particle “i” interaction with particles “j”
Conservative Force, Dissipative Force, Random Force
Conservative Force- gives particle an identity
Random and dissipative forces act as a thermostat
Momentum is conserved locally so you can work with small numbers of particles
Random force between two interaction particles must be of opposite sign
One random force calculation for each pair of interacting particles
(This differs from Brownian motion where each particle experiences an independent random force)
Particles connected by Hookean springs if desired
Usually N, V and T are kept constant
23
Parallelization in Dissipative particle dynamics
DPD interactions are short range so different processors can run in parallel for very
large systems (micron and milliseconds (1000 microseconds))
http://gpiutmd.iut.ac.ir/en/gal
LCD self-assembly lery/video-gallery
Smectic Crystal
Rod-Shaped particles
1 microsecond time
24
FEM Finite Element
(Difference or Volume)
Method
SPH Smoothed Particle
Hydrodynamics
DPD Dissipative Particle Dynamics
LBM Lattice Boltzmann Method (fluids)
25
26
Dissipative particle dynamics
Coarse Graining Parameter Nm = number of molecules per particle
Speedup time is 1000 Nm8/3 for Nm = 3 you get 20,000X; for 7, 200,000X
Particles are “soft” not LJ particles Repulsive
Drag
Random
All forces act within a sphere of interaction, rc
Thermostat requirements:
w are “weight functions”
g and s are coefficients
White noise function
aij is the repulsion between particles “i” and ”j” 27
Dissipative Particle Dynamics (DPD)
28
29
30
31
32
Some available simulation packages that can (also)
perform DPD simulations are:
•CULGI: The Chemistry Unified Language Interface, Culgi
B.V., The Netherlands
•DL_MESO: Open-source mesoscale simulation software.
•DPDmacs
•ESPResSo: Extensible Simulation Package for the
Research on Soft Matter Systems - Open-source
•Fluidix: The Fluidix simulation suite available from
OneZero Software.
•GPIUTMD: Graphical processors for Many-Particle
Dynamics
•Gromacs-DPD: A modified version of Gromacs including
DPD.
•HOOMD-blue: Highly Optimized Object-oriented Many-
particle Dynamics—Blue Edition
•LAMMPS
•Materials Studio: Materials Studio - Modeling and
simulation for studying chemicals and materials, Accelrys
Software Inc.
•SYMPLER: A freeware SYMbolic ParticLE simulatoR
from the University of Freiburg.
•SunlightDPD: Open-source (GPL) DPD software. 33
34
35
