computer simulations of dislocations - purdue universityibilion/...computer simulations of...
TRANSCRIPT
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 1
References
Computer Simulations of Dislocation, V. V. Bulatov and W Cai
Theory of Dislocations, Hirth and Lothe
Introduction to Dislocations, D Hull and D J Bacon
Introduction to Solid State Physics, Kittel
Computer Simulations of Dislocations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 2
1. Introduction to crystal dislocations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 3
1. Introduction to Crystal Dislocations
Overview:
The basic elements and common terminology used to describe perfect crystal
structures.
Dislocation as a defect in the crystal lattice and its properties.
Forces on dislocations and atomistic mechanisms for dislocation motion.
It define a great many properties of crystalline materials: crystals’ ability to yield
and flow under stress, creep and fatigue, ductility and brittleness,
indentation hardness and friction, crystal grows, etc.
In material science, a dislocation is a crystallographic defect, or irregularity,
within a crystal structure.
Transmission Electron Micrograph of Dislocations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 4
1.1 Perfect Crystal Structures
crystal structure=lattice + basis A basis may consist of one or more atoms.
(a) A two-dimensional crystal consisting of two types of atoms (white and gray). (b) The Bravais lattice
is specified by two repeat vectors a and b. (c) The basis contains three atoms.
primitive cell: The smallest parallelepiped with a lattice point at each of its eight
corners (3D).
unit cell: To better reflect the symmetries, certain types of Bravais lattices are
specified by non-primitive lattice vectors a, b and c. The parallelepiped formed
by these vectors is called the unit cell, for example, FCC, BCC.
The Unit cell is larger or equal to the Primitive cell
To avoid confusion, the adopted convention for constructing unit cell is to
associate each crystal structure with the Bravais lattice of highest possible
symmetry and with the basis containing the smallest number of atoms.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 5
(a) The unit cell of a simple cubic Bravais lattice. (b) The unit cell of a body-centered-cubic Bravais
lattice. (c) The unit cell of a face-centered-cubic Bravais lattice.
the lattice points
positions in the SC
lattice
the lattice points positions
of a BCC lattice
the lattice points positions
of an FCC lattice
The lattice and the basis are not uniquely defined.
1.1 Perfect Crystal Structures
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 6
Miller indices
A vector l that connects one point to another in a Bravais lattice can be written
as a linear combination of repeat vectors, i.e.
where i, j and k are integers. The Miller indices notation for this vector is [i j k].
To specify a crystallographic plane, the Miller indices of the direction normal to
the plane are used. The Miller indices of a lattice plane can also be defined to
be a set of integers with no common factors, inversely proportional to the
intercepts of the crystal plane along the crystal axes, but written between
round brackets, i.e. (i j k).
A family of directions (vectors) is written
between angular brackets <i j k>, while a family
of planes is written in curly brackets {i j k}.
321
1:
1:
1::
xxxkji
Negative component is specified by placing a bar over the corresponding index,
i.e.
1.1 Perfect Crystal Structures
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 7
1.2 The concept of crystal dislocations
They are usually thought of as extra lattice planes (edge) inserted in the
crystal that do not extend through all of the crystal, but end in the dislocation
line.
A dislocation is a defect of crystal lattice topology and can be defined by
specifying which atoms are dislocated or mis-connected with respect to the
perfect, defect free structure of the host crystal.
(a) A perfect simple-cubic crystal. (b) Displacement of two half-crystals along cut plane A by lattice
vector b results in two surface steps but does not alter the atomic structure inside the crystal. (c) The
same “cut-and-slip” procedure limited to a part of cut plane A introduces an edge dislocation ⊥.
Cut and slip
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 8
Edge dislocation: a defect where an extra half-plane of atoms is introduced mid way
through the crystal, distorting nearby planes of atoms.
Screw dislocation: atoms round the dislocation line are arranged in a spiral.
In pure screw dislocations, the Burgers vector is parallel to the line direction.
In an edge dislocation, the Burgers vector is perpendicular to the line direction.
Two properties: line direction (sense), the direction running along the bottom of
the extra half plane; the Burgers vector, describes the magnitude and direction
of distortion to the lattice.
(a) An edge dislocation created by inserting a half-plane of atoms B. (b) A screw dislocation created
by a “cut-and-slip” procedure in which the burger vector is parallel to the dislocation line. (c) Mixed
dislocation, a curved dislocation line with an edge orientation at one end (on the left) and a screw
orientation at the other end (on the right).
1.2 The concept of crystal dislocations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 9
In real crystals, dislocations form in many different ways: shearing along crystal
planes, condensation of interstitials (extra atoms in the lattice) or vacancies (empty
atomic sites).
A more precise way to identify the presence of a dislocation is Burgers circuit
test, which consists of a sequence of jumps from atoms to their neighbors. The
Burgers circuit should form a complete loop when it is drawn in a perfect crystal,
and may not end at the starting atom when in a defective crystal.
line sense The start and end points of the
circuits are Si and Ei ,
respectively. Circuit 1 does not
enclose dislocation ⊥ whereas
circuits 2 and 3 do. The sense
vector ξ is defined to point out of
the paper so that all three circuits
flow in the counterclockwise
direction following the right-hand-
rule.
1.2 The concept of crystal dislocations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 10
In circuit 2 in the above picture, the vector connecting starting point S2 to end
point E2 of test circuit 2 is the Burgers vector b associated with the enclosed
dislocation. It needs to emphasize that the direction (or sign) of the Burgers
vector is meaningful only when the sense of the dislocation line is either
explicitly defined or implied by context. For example, if we reverse the direction
of line sense vector ξ and make it point into the plane of the paper. Our right-
hand-rule convention then dictates that, to obtain the Burgers vector, circuit 2
should now run clockwise, from atom E2 to atom S2, which obviously reverses
the direction of the resulting Burgers vector.
The algorithm of constructing Burgers circuits (r0 and vi are input)
1. Define the starting position r0 of the circuit to be the position of atom nS. Set
i := 1.
2. Find atom ni whose position ri is nearest to point ri−1 +vi .
3. Compute the difference between the actual and perfect relative positions of
the atom pair connected by the current translation, Δui := ri −(ri−1 +vi).
4. Increment the step counter i := i +1. If i ≤N, go to step 2.
5. Define the index of the end atom, nE := nN.
1.2 The concept of crystal dislocations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 11
The Burgers vector is obtained either as the difference rN −r0 or, equivalently,
as the sum of vectors
The sum in this equation is a discrete analogue of the equation used in
continuum mechanics to define the Burgers vector of a Volterra dislocation,
where C is any contour enclosing the dislocation line and ∂u/∂l is the elastic
displacement gradient along the contour.
The resulting Burgers vector is unaffected by any deformation and/or
translation of the test circuit as long as such deformation and/or translation
does not make the circuit “cut” through a dislocation line. The Burgers
vector is conserved along any given dislocation. It is an intrinsic property of
the dislocation line that can be regarded as the dislocation’s topological
charge.
1.2 The concept of crystal dislocations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 12
Multiple dislocations: a test circuit drawn around two dislocation lines reveals a
Burgers vector equal to the vector sum of the Burgers vectors of two enclosed
dislocation. Two dislocation lines ξ2 and ξ3 merge into a single line at junction
node P. The Burgers vector of the resulting dislocation ξ1 can be obtained from
Burgers circuit q drawn on cross-section C. Because circuit q can be obtained
from circuit p by deformation and translation without cutting through the
dislocation lines, the Burgers vector revealed by both circuits must be the same,
that is
If we flip the directions of ξ2, ξ3, b2,
b3, the conservation of Burgers
vector can be written as
Analogy to the conservation of
current in an electric circuit: Cross
section B
Cross
section C
1.2 The concept of crystal dislocations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 13
1.3 Motion of a Crystal Dislocation
This section illustrates how the driving force for dislocation motion is obtained from
continuum elasticity theory, whereas the dislocation’s response to this force is
governed by discrete atomistic mechanisms.
The driving force for dislocation motion is applied stresses
(a) A perfect crystal with simple
cubic structure and dimension
Lx ×Ly ×Lz. (b) The top surface is
subjected to a traction force Tx
while the bottom surface is fixed.
An edge dislocation nucleates from
the left surface. In (c) and (d) the
dislocation moves to the right. In
(e) the dislocation finally exits the
crystal from the right surface. The
net result is that the upper half of
the crystal is displaced by b with
respect to the lower half.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 14
The work done by the surface traction is
Where fx is the driving force per unit length on the dislocation line
Generalize to 3D force per unit length at an arbitrary point P on a dislocation line,
we can obtain Peach-Koehler formula [1]
where σ is the local stress field and ξ is the local line tangent direction at point
P. The cross product with ξ ensures that the PK force is always perpendicular to
the line itself.
The significance of the PK formula is that the force acting on a dislocation is
fully defined by the local stress σ on the dislocation, regardless of the origin of
this stress.
The source of the local stress includes: surface traction, other dislocation or
other strain-producing defects interactions.
[1] J.P. Hirth and J. Lothe, Theory of Dislocations, Wiley, New York, 1982
1.3 Motion of a Crystal Dislocation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 15
The motion of dislocation motion is normally known as the sliding along well-
defined crystallographic planes. Sliding by dislocation motion only requires
significant atomic rearrangements near the dislocation core (local), as opposed
to over an entire plane (global). The level of stress required to make a dislocation
move is usually orders of magnitudes lower than the critical stress to break all
bonds on a crystallographic plane.
Theoretical strength: the shear stress needed to displace the upper half of the
crystal relative to the other half.
Small displacement:
In real crystals: 5 3
max10 10
1.3 Motion of a Crystal Dislocation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 16
Another important property of dislocation motion is that the area swept out by a
moving dislocation is proportional to the plastic strain it introduces to the crystal.
bi is the component of Burgers vector b in i direction, ni is the component of
vector n normal to the glide plane A, which simultaneously contains the
dislocation line and the Burgers vector. ΔA=vLΔt is the total area swept out by
dislocations, with total length L and average velocity v, during a period Δt
The plastic strain rate can be related to dislocation density ρ by calculating the
time derivative of the plastic strain.
Ω=LxLyLz is the volume of the crystal. Generalization:
z
x y z
NL
L L L , N is the number of dislocations.
1.3 Motion of a Crystal Dislocation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 17
Conservative versus Non-conservative Motion
According to the motion with respect to glide plane, we can assort the dislocation
motion into 2 categories: glide, motion along the glide plane; and climb, motion
perpendicular to the glide plane.
Dislocation glide is often called conservative motion, meaning that the total
number of atoms is conserved, whereas dislocation climb is non-conservative
(exception is found in prismatic loop [1]). A screw dislocation always glides and
never climbs.
The mobility of non-screw dislocations is usually highly anisotropic. At low
temperatures, climb is usually difficult and glide is dominant. However, at high
temperatures or under conditions of vacancy super-saturation, climb can
become dominant instead.
1.3 Motion of a Crystal Dislocation
[1] Introduction to Dislocations, D Hull and D J Bacon
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 18
Dislocation glide mobility may be influenced by both extrinsic factors,
such as impurities acting as obstacles, and intrinsic factors, such as the
interatomic interactions at the dislocation core.
Two fundamental parameters that characerize intrinsic lattice resistance to
dislocation motion are the Peierls barrier and the Peierls stress.
For a screw dislocation, the Burgers vector is parallel to the line direction, hence
the glide plane is not uniquely defined. No insertion or removal. All ways glide and
never climbs. But can change slip plane by cross-slip.
Mixed dislocation: the edge component of Burgers vector determines how many
atoms need to be inserted or removed during climb.
1.3 Motion of a Crystal Dislocation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 19
Peierls valley: the minimum of this
function mark the preferred dislocation
positions.
Consider a straight dislocation moving in its glide plane. The effect of the crystal
lattice on this motion can be represented by an energy function of the dislocation
position, which has the periodicity of the lattice
Peierls barrier: The energy barrier (per
unit length) that a dislocation must
surmount to move from one Peierls valley
to an adjacent one under zero stress.
Peierls stress: In the presence of a non-zero local stress, the force acting
on a dislocation modifies the periodic energy function. As a result, the
actual energy barrier experienced by the dislocation becomes lower than
the Peierls barrier. The critical stress makes the energy barrier vanish
completely, is called Peierls stress. The minimum stress required to make a
straight dislocation move at zero temperature.
1.3 Motion of a Crystal Dislocation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 20
When local stress is higher than the Peierls stress, dislocation motion can
happen without other assistance, like thermal fluctuations. In this case,
dislocation can evolve easily, Molecular Dynamics is suitable for the simulation.
Usually, as Peierls stress for ordinary dislocations in the FCC metals is low, so
that MD model is used.
When the local stress is lower than the Peierls stress, a dislocation
cannot move at zero temperature, but can move at a finite temperature with the
help of thermal fluctuations. In this case, rather than moving the whole straight
dislocation at once, motion begins by throwing a short dislocation segment into
the next Peierls valley.
In this case, the kink pair nucleation is a rare event. The time step needs to be
in relatively large scale. It is inefficient to use Molecular Dynamics to simulate
the dislocation evolution, because in most time steps, the dislocation evolution
can not be seen. Kinetic Monte Carlo method is adopted. Most dislocation
motion in semiconductors, belongs to this situation.
1.3 Motion of a Crystal Dislocation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 21
2. Fundamental of Atomistic Simulations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 22
Interatomic interactions
Boltzmann distribution
Energy minimization
Introduction to Monte Carlo and Molecular Dynaimics
Fundamentally, materials derive their properties from the interaction between their
constituent atoms. To understand the behavior of dislocations, it is necessary and
sufficient to study the collective behavior of atoms in crystals populated by
dislocation.
2. Fundamentals of atomistic simulations
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 23
When put close together, atoms interact by exerting forces on each other. The
variability of interatomic interactions stems from the quantum mechanical motion
and interaction of electrons. Henceforth, rigorous treatment of the interactions
should be based on a solution of Schrödinger’s equation for interacting electrons,
which is usually referred to as the first principles or ab initio theory. Although
accurate, it is a very inefficient method.
The usual way to construct a model of interatomic interactions is to postulate a
relatively simple, analytical functional form, interatomic potential, for the potential
energy of a set of atoms,
where ri is the position vector of atom i and N is the total number of atoms. The
force on an atom is
The hope is that interatomic potential can capture the most essential physical
aspects of atom–atom interaction. The parameters are usually fitted to
experimental or ab initio simulation data
2.1 Interactomic interactions
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 24
The most obvious physical feature of interatomic interactions is that atoms do not
like to get too close to each other. The other important aspect of the interatomic
interaction is that atoms attract each other at longer distances.
where
Here, ε0 is the depth of the energy well and 21/6σ0
is the distance at which the interaction energy
between two atoms reaches the minimum.
Relatively few materials, among them the noble gases (He, Ne, Ar, etc.) and ionic
crystals (e.g. NaCl), can be described by pair potentials with reasonable accuracy.
For most other materials pair potentials do a poor job, especially in the solid state.
Pair potential
A well-known pair potential model that describes both long-range attraction and
short-range repulsion between atoms is the Lennard-
Jones (LJ) potential,
2.1 Interatomic interactions
Hard sphere model
L-J model
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 25
many-body potential energy
Example: Stillinger–Weber (SW) potential for silicon, containing two-body and
three-body terms. Embedded-atom model (EAM), decomposing potential into a
pairwise interaction accounting for the effect of core electrons and a field term
defining the energy required to embed atom into an environment with electron
density.
where
is the local density of bonding electrons supplied by the atoms neighboring with
atom i. Because the embedding function is non-linear, the EAM-like potentials
include many-body effects that cannot be expressed by a superposition of pair-
wise interactions. As a result, EAM potentials can be made more realistic than
pair potentials.
2.1 Interactomic interactions
http://enpub.fulton.asu.edu/cms/potentials/main/main.htm
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 26
Development of an interatomic potential:
Selection of a functional form that can capture the underlying physics of
interatomic interaction in the material of interest (fit specific material properties).
The key issue in developing a good interatomic potential is its transferability, the
ability to accurately describe or predict a material’s property that is not included
in its fitting database.
In the context of dislocation, the physical parameters that matter most are the
relative energies of the most stable crystal structures, elastic constants, point
defect energies and stacking fault energies. An interatomic potential fitted to
these parameters stands a better chance of describing dislocation behaviors
accurately.
Interatomic interactions are usually short range. To improve numerical efficiency,
it is common to truncate interatomic potentials, i.e. by setting the interaction
energy to zero whenever the distance between two atoms exceeds a cut-off
radius rc. In order to smooth out the undesirable artifacts by the truncation
discontinuity of energy at rc, both energy and its derivative should be zero at cut-
off distance.
2.1 Interactomic interactions
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 27
For short-range potentials, force on a given atom can be calculated given the
positions of its neighbors only within the cut-off radius. Another useful property of
interatomic potential models is that it is usually straightforward to partition the
potential energy into local energy contributions for each atom. For the EAM-like
potentials, the local energy of atom i can be defined as
It is easy to check that the sum of local energies of all atoms is equal to the total
potential energy as it should be.
Computation efficiency and accuracy
DFT: less efficient, accurate;
TB: mediate, mediate;
Empirical potential: fast and capable of
handling large number of atoms, less
accurate .
2.1 Interactomic interactions
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 28
2.2 Equilibrium Distribution
In a N atoms system, the total energy is
H is Hamiltonian. In classical mechanics, the instantaneous state of the system is
a microstate fully specified by the positions and momenta of all atoms {ri, pi}
(phase space).
When the system is in thermal equilibrium temperature T (not too low), the
distribution of the states {ri, pi} obeys Boltzmann law.
Where partition function
The system is said to be canonical ensemble.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 29
Statistical enemble: Consider a very large number of replicas of the system,
each with N atoms and described by the same Hamiltonian. At any instance of
time, each system’s state is represented by a point {ri , pi} in the 6N-dimensional
phase space. An ensemble is the collection of all these replicas or, equivalently, of
points in the 6N-dimensional phase space. Each replica (a member of the
ensemble) may evolve with time, which is reflected by the motion of a
corresponding point in the phase space. The replicas do not interact. When the
replicas are distributed according to Boltzmann’s distribution, the ensemble is
called canonical.
A macroscopic quantity A at thermal equilibrium can be expressed as a statistical
average of microscopic functions over the canonical ensemble
where
For harmonic system, where the potential V({ri}) is a quadratic function of the
atomic positions, the average potential energy is . And also we know the
average kinetic energy is , the total energy is
2.2 Equilibrium Distribution
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 30
2.3 Energy Minimization
The Steepest-descent Method
Start from an arbitrary initial configuration R. At every iteration, the force vector F
is computed and R is displaced by a small step along F. The iterations continue
until |F| becomes smaller than a prescribed tolerance .
At low temperature, the kinetic energy is zero and the potential energy is
minimized, providing a good description of the atomic structure of the system.
The idea of the steepest-descent algorithm is that, as long as |F| is non-zero,
V (R) can be further reduced by displacing R in the direction of F.
Conjugate Gradient Relaxation
CGR goes through a series of search directions. The (local) minimum energy
point along each search direction is reached before CGR proceeds to the next
search direction. The search sequence is constructed in such a way that
subsequent search directions “avoid” (i.e. are conjugate to) all previously
searched directions.
Inefficient, local minimum
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 31
This algorithm works best when the potential energy is a quadratic function, i.e.
, where G is a 3Nx3N symmetric matrix. The conjugate condition means
that any two search direction must satisfy
Global Minimization
Neither the CGR nor the steepest-descent algorithm guarantees finding the global
minimum of an energy function. If the global minimum is the target, an inefficient
approach is to run minimization many times starting from randomly selected initial
configurations. Then, the lowest local minimum found from a series of relaxations
provides an upper bound on the global minimum.
Line search
2.3 Energy Minimization
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 32
A useful method is simulated annealing (SA), which mimics the thermal annealing
of a material. SA simulation starts at a high temperature, making it easier for the
system to escape from the traps of local minima and explore a larger volume in
the configurational space. The SA search then slowly reduces the temperature to
zero. At the end of this procedure, the system converges to a local minimum that
is likely to be less dependent on the initial configuration, and is more likely to be
the global minimum. Tight upper bound on the global minimum can be obtained by
performing SA several times.
SA is usually used together with MC and MD
methods to search for the global minimum of the
potential function.
2.3 Energy Minimization
[1] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi. Optimization by
simulated annealing. Science, 220(4598): 671–680, 1983.
[2] S. Kirkpatrick. Optimization by simulated annealing: quantitative
studies. Journal of Statistical Physics, 34(5/6): 975–986, 1984.
The optimal form of T(t) is system dependent. In practice, is
commonly used.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 33
2.4 Monte Carlo
Traditional numerical integration (i.e. quadrature) in 6N-dimensional phase space
is prohibitively expensive.
Therefore, Monte Carlo (MC) and Molecular Dynamics (MD) come to the rescue.
Simulate the motion of atoms at a finite temperature in such a way that the fraction
of time the system spends in each state satisfies Boltzmann’s distribution.
If the microscopic function depends only on the atomic configuration R, i.e.
A=A(R), the ensemble average can be written in terms of an integral over the
configurational space (only position, no moment),
where
The idea of the Monte Carlo method is to generate a stochastic process in the
configurational space such that the states visited during the simulation satisfy the
equilibrium distribution ρ(R).
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 34
In the case, the ensemble average A can be replaced by an average of A(R) over
the trajectory of the Monte Carlo simulation. Most Monte Carlo methods simulate a Markov process, meaning that the
probability of moving to a specific state in the next step is only a function of the
current state and is independent of the states visited in the preceding steps. A
Markov process is specified by its transition probability matrix, π(R,R’), which is
the probability density of visiting state R’ in the next step if the system is currently
at R. π satisfies the normalization condition,
If the probability distribution of the system at step n is ρ(n)(R), then the probability
distribution at step n + 1 is
Hence, to sample the equilibrium distribution ρ(R), the Markov process should
satisfy
Here, Metropolis algorithm is used by choosing detailed balance condition
2.4 Monte Carlo
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 35
it states that the net rate of transitions from state R to state R’ is exactly the same
as the net rate of the reverse transitions from R’ to R, when the stationary
distribution ρ is reached. The following restriction on the transition probabilities is
reached
In Metropolis algorithm, π is chosen as
Where acceptance ratio
Two sub-steps of a single Monte Carlo step is included. First, a trial move from
the current configuration R to another configuration R’ is selected with probability
α(R,R’). Usually, the trial moves displace one or several atoms in a random
direction, R’ =R+δR. If the potential energy of the trial state R’ is lower than the
energy of the current state R, the move is accepted. However, if the potential
energy of the new state is higher by dV =V (R’)−V (R), the move is accepted with
probability Pacc = exp(−βdV ). Otherwise, the move is rejected and the system
remains in the old state R for yet another step.
2.4 Monte Carlo
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 36
The algorithm computing the average of function A(R), for a N-atom system with
initial positions ri
It is a common practice to adjust Δ empirically so that the fraction of accepted
moves stays close to 50% during the simulation. The requirement of designing a
MC procedure is (1) trial moves proceed unbiased, i. e. α(R, R’)=α(R’, R); (2) the
simulation trajectory can span the entire configurational space.
2.4 Monte Carlo
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 37
2.5 Molecular Dynamics
Unlike the Monte Carlo method, which generates artificial trajectories spanning
the configurational space and complying with Boltzmann’s distribution, molecular
dynamics (MD) attempts to simulate the “true” dynamics of atoms while also
preserving Boltzmann’s statistics. The classical equations of motion for a system
with Hamiltonian H({ri , pi}) are
If only the potential is the function of atom position while the kinetic energy is the
function of momentum, the equation of motion can be written as
Molecular dynamics, at its heart, is simply the numerical integration of this
Newton’s second law.
The Verlet Algorithm
It is based on the following symmetric finite-difference approximation to the
acceleration,
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 38
which leads to
Using this equation repeatedly, atomic positions are computed step by step. The
atomic velocity does not appear explicitly in Verlet algorithm. It is evaluated by
The Velocity Verlet Algorithm
The numerical estimation of the atomic velocities obtained in the Verlet algorithm
is not very accurate. The velocity Verlet algorithm is proposed to address this
problem
The equations of motion of a Hamiltonian system should conserve the system’s
energy.
Energy Conservation
2.5 Molecular Dynamics
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 39
The Verlet and the velocity Verlet algorithms belong to a class of the symplectic
integrators[1] that preserve the energy particularly well. Selection of a proper time step t is an important issue in MD simulations. The total
energy of the system
may “drift away” or even diverge when the selected time increment is too large.
For MD simulations of solids, Δt around 0.01 of the inverse Debye frequency νD
is usually a safe choice. Δt is commonly of the order of 10-15.
2.5 Molecular Dynamics
[1] D. Frenkel and B. Smit. Understanding Molecular Simulation: From Algorithms to Applications.
Academic Press, San Diego, CA, 2002
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 40
3. Case Study of Static Simulation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 41
set up initial positions of atoms based on solutions of the continuum elasticity
theory.
Boundary conditions.
MD simulation of dislocations.
3. Case study of static simulation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 42
An atomistic structure is specified by the positions xi of all atoms. A good way to
describe the new structure is by specifying the displacement vector
for every atom i. We can use the continuum solution u(xi) to approximate the
displacement of atom x far away from the dislocation center by assuming the
crystal is a continuum linear elastic solid.
3.1 Setting up an Initial Configuration
( , ) , , 0, 02
z x yu x y b u u except x y
The reasons for the continuum linear elasticity theory to break down in the
dislocation core:
(1) In a discrete crystal structure, relative displacements between neighboring
atoms in the dislocation core can be very large and vary rapidly from one atom
to its neighbors.
(2) The interactomic interactions in the core are highly non-linear.
The equilibrium atomic positions should be determined by minimizing the
interactomic potentials energy of the entire system.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 43
[112]
[110]
BCC Tantalum
Cut and displaced by . Then relax. ( , )2
zu x y b
Free to move fixed
Periodic in z
Software: MD++
3.1 Setting up an Initial Configuration
Finnis-Sinclair potential for BCC tantalum.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 44
# Create single screw dislocation in BCC Ta
# under fixed boundary condition
setnolog
setoverwrite dirname = runs/ta-screw
#Read in potential file (Finnis-Sinclair potential for Ta)
potfile = ~/Codes/MD++/potentials/ta_pot readpot
#Create Perfect Lattice Configuration
latticestructure = body-centered-cubic latticeconst = 3.3058 #(A)
makecnspec = [ 1 1 -2 10 1 -1 0 16 0.5 0.5 0.5 5 ]
makecn finalcnfile = perf.cn writecn
#Create a single screw dislocation b = [111]
# remove all atoms beyond radius 32
# z y x0 y0 Rc rem
input = [ 3 2 0.0334 -0.0208 32 0 ] makecylinder
# 3 : specifies the cylinder axis is along z axis
# 2 : specifies the "local" y axis is the same as our y axis
# (x0,y0) : specifies the center of the cylinder in "local"
# coordinate system in scaled coordinates
# Rc : speficies the cut-off radius
# rem : 0: atoms outside Rc will be removed, 1: atoms to be fixed
# introduce the dislocation
input = [ 1 # activate makedislocation
3.3058 # lattice constant (in Angstrom)
0 0 0.8660 # Burgers vector (in unit of a)
0 0 1 # line direction vector
0 1 0 # normal vector of cut plane
2.65 -1.28 0 # a point on dislocation (in Angstrom)
0.339 # Poisson's ratio (for nonscrew dislocation)
0 # 0 commit displacement, 1 store in memory
]
makedislocation eval
# fix all atoms beyond radius 23.6
# z y x0 y0 Rc fix
input = [ 3 2 0.0334 -0.0208 23.6 1 ] makecylinder
#Plot Configuration
atomradius = 1.0 bondradius = 0.3 bondlength = 0
atomcolor = cyan highlightcolor = purple backgroundcolor = gray
bondcolor = red fixatomcolor = yellow
color00 = "orange" color01 = "red" color02 = "green"
color03 = "magenta" color04 = "cyan" color05 = "purple"
color06 = "gray80" color07 = "white" color08 = "blue"
plot_color_windows = [ 2
-10 -8.0 6 #color06 = gray80
-8.0 -6.0 8 #color08 = blue
1 #1: draw fixed atoms
]
#plot_limits = [ 1 -10 10 -0.05 10 -10 10 ]
plot_atom_info = 1 plotfreq = 10
rotateangles = [ 0 0 0 1.5 ]
win_width = 600 win_height = 600
#win_width = 140 win_height = 140
openwin alloccolors rotate saverot refreshnnlist eval plot
#sleep quit
#Conjugate-Gradient relaxation
conj_ftol = 1e-10 conj_itmax = 1000 conj_fevalmax = 10000
conj_fixbox = 1
relax
eval finalcnfile = relaxed.cn writeall = 1 writecn
sleep quit
3.1 Setting up an Initial Configuration
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 45
3.2 Boundary Conditions
Free surface: no constraint on the motion of any atom on boundaries. Completely
ignores the effects of atoms outside the simulation volume and introduces
unnecessary surfaces.
Fixed boundary: fix atoms on the periphery of the simulation volume in equilibrium
positions that they would occupy in an infinite solid.
Flexible boundary: allows the atoms in the boundary layer to adjust their positions
in response to the motion of inner atoms.
Periodic boundary: embeds the simulation volume into an infinite, periodic array of
replicas or images. It completely eliminates surface effects and maintains
translational invariance of the simulation volume. No point in space is treated any
more specially than others.
Primary cell
Image cell
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 46
Minimum image convention: the relative displacement vector between atoms i and
j is taken to be the shortest of all vectors that connect atom i to all periodic replicas
of atom j. The potential cut-off distance is sufficiently small so that no more than
one replica of atom j falls within the cut-off radius of atom i.
Scaled coordinates:
Consists of repeat vectors of the simulation cell
Whenever there is an atom at position s=(sx, sy, sz), there are also atoms at
positions s=(sx+n1, sy+n2, sz+n3).
The distance between two atoms with scaled coordinates is
3.2 Boundary Conditions
x
y
(x,y) c2
c1
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 47
The separation vector in the real space is
PBC cannot eliminate the artifacts due to inevitably small number of atoms.
It can also introduce its own artifacts, i.e. in the existence of defect. The remedy
will be discussed later.
3.2 Boundary Conditions
Other types of boundary conditions can be constructed “inside” PBC.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 48
3.3 Data Analysis and Visualization
It is necessary to perform data filtering: Save memory, extract important information.
Identify crystal defects:
1. Locate higher energy spot.
Low signal-to-noise ratio.
Suppress unwanted noise by partial steepest descent.
The snapshots of MD simulation are used to initiate
steepest descent paths towards underlying local
minima. Limit the steepest descent relaxation to small
number of iterations to preserve system’s
configuration.
2. Centro-symmetry deviation (CSD) for center symmetric crystals (FCC, BCC, but
not HCP, diamond cubic)
Np=4 Np=6
CSD parameter is 0 in perfect crystal.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 49
Partial dislocations
Stacking fault
Remove
plane and relax
(110)
3.3 Data Analysis and Visualization
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 50
Example: Frank-Read source
The problem of dislocation multiplication is important in the theory of crystal
deformations. Frank-Read source is such a multiplication mechanism of
dislocation.
Schematic representation of the operation of a Frank-Read source. A straight dislocation segment is bowed out by the driving shear stress with two pinning points. After a loop forms, a new dislocation segment is born.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 51
Frank-Read source, Shockley partials and stacking faults
Perfect crystal Dislocation loop
Remove a plane of
atoms on [-1 1 0]
x[-1 1 0] z[1 1 -2]
y[1 1 1]
relax
Apply shear
stress σxy
Stacking faults
Plotted according to
centro-symmetry
deviation parameters
Example: Frank-Read source
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 52
Frank-
Read
source
Shockley partials
The reason for the
dissociation of the perfect
dislocation is that the
motion of the atom along
the path a to c involves a
larger dilatation normal to
the slip plane, and hence
a larger misfit energy
than does motion along
the path a to b to c.
Example: Frank-Read source
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 53
4 Case Study of Dynamic Simulation
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 54
Build a simulation cell consisting of two rectangular slabs with dimensions:
Upper: Lx1=30/2[1 1 1], Ly1=10[-1 0 1], Lz1=8[1 -2 1]
Lower: Lx2=29/2[1 1 1], Ly2=10[-1 0 1], Lz2=8[1 -2 1]
The lower slab has one atomic plane fewer in x direction than the upper one.
Reshape both slabs to have the same length Lx=29.5[1 1 1]/2 and relax.
y [-1 0 1]
x [1 1 1]
z [1 -2 1]
Periodic boundary
condition is applied
in x and z direction
4.1 Setting up an Initial Configuration
Create a planar misfit interface between two crystals. Subsequent energy
minimization would automatically lead to dislocation formation.
Vacuum gap,
create free surface
Finnis-Sinclair potential M.W. Finnis and J.E. Sinclair, Philosophical
Magazine A, 50(1): 45-55, 1984
Plot in terms of CSD
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 55
4.2 Initializing Atomic Velocities
Thermalization: bring the system to a state of thermal equilibrium at a finite
temperature.
Equilibration run: A preliminary MD simulation starting from an arbitrary initial
configuration can bring the system to a state of thermal equilibrium after some
period of time.
Control temperature by scaling atomic velocities
Instantaneous temperature:
Set initial temperature and atom velocities:
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 56
0 200 400 600 800 1000
-200
0
200
400
600
800
1000
1200
1400
1600
1800
2000
En
erg
ies (
eV
)
t (fs)
kinetic
potential
total
0 200 400 600 800 1000
100
200
300
400
500
600
Te
mp
era
ture
(K
)
t (fs)
Initial temperature T*=600K
At the beginning of the equilibrium run, the atomic positions are at a local
potential energy minimum. During equilibration, the average potential energy
increases while the kinetic energy decreases by the same amount so that the
total energy is conserved.
Finnis-Sinclair potential approximated by a harmonic function.
4.2 Initializing Atomic Velocities
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 57
4.3 Stress Control
Apply stress to make the dislocation move.
1. Applying forces on surface atoms.
2. Applying stress in simulations with PBC in all
three directions -- Parrinello-Rahman method:
applies stress to the periodic supercell by
dynamically adjusting its shape so that the
virtual stress fluctuates around a specified
value.
In PR approach, the atomistic model acquires
9 additional degrees of freedom, which are
the components of matrix h. For a system with
N atoms, the total number of degrees of
freedom in PR approach is 3N+9.
[1] J. A. Zimmerman, E. B. Webb, J. J. Hoyt, R. E. Jones, P. A. Klein, and D. J. Bammann. Calculation of stress in atomistic
simulation. Modelling and Simulation in Materials Science and Engineering, 12(4): S319–332, 2004.
[2] A. M. Pendas. Stress, virial, and pressure in the theory of atoms in molecules. Journal of Chemical Physics, 117(3): 965–
979, 2002.
[3] J. Gao and J. H. Weiner. Excluded-volume effects in rubber elasticity. 1. Virial stress formulation. Macromolecules, 20(10):
2520–2525, 1987.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 58
The equations of motion
where
is the stress one wishes to maintain during the simulation,
is its hydrostatic pressure component, is the matrix describing the shape of the
periodic cell at the beginning of the simulation– this matrix defines the reference
frame in which the orientation of the applied stress is expressed. W is a “mass”
parameter that determines how fast the box changes its shape in response to the
imbalance between the desired and the actual stress.
Feedback, adjust h when the virial stress
becomes different from the desired value.
4.3 Stress Control
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 59
4.4 Temperature control
Maintain the constant temperature during dislocation motion–
1)scale the atomic velocities at periodic intervals during the simulation.
2)Nose-Hoover thermostat, which mimics heat exchange between a simulation
volume and its surroundings.
3)Berendsen.
Ensembles:
Micro-canonical or NVE ensemble: fixed or periodic boundary conditions with
fixed repeat vectors. Constant E.
Canonical or NVT ensemble: a system exchanging heat with an external
thermostat will maintain a constant temperature T.
Isothermal isobaric or NPT ensemble: the system interacts with an external
barostat, its total volume will adjust so as to maintain a constant pressure.
Grand-canonical or μVT ensemble: mass exchange with a large external reservoir
will cause the total number of atoms in the system to fluctuate so that its chemical
potential remains close to that of the massostat.
The Nose-Hoover thermostat is to reproduce the canonical ensemble. The total
system is assumed to consist of two interacting sub-systems, the atomistic
system plus the additional fictitious degrees of freedom. The total system is
closed but the atomic sub-system is open.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 60
Introducing a dynamical variables s that serves as a scaling factor of time. It
distorts the atomic trajectories, but maintains the temperature of the atomistic sub-
system close to a preselected value.
feedback
The Nose-Hoover equations of motion are derived from the standard variational
principle of classical mechanics. The trajectories conserve the following total
energy of the extended system.
Atomistic sub-system Heat bath
4.4 Temperature control
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 61
The original Verlet algorithm has to be modified to take Nose-Hoover thermalstat.
where
Solving these two equations for the atomistic positions at time
The equation for the fictitious parameter is
where
in this case study.
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 62
4.5 Extracting Dislocation Velocity
A simple definition of the dislocation position would be the center of mass of
the core atoms. Here, the core atoms are defined as atoms whose CSD
parameter is within certain domain.
Drag coefficient
Transient period,
measures
dislocation’s
effective mass
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 63
After the system achieves thermal equilibrium, shear stresses σxy = 300Mpa are
applied on both free surfaces. The total force is 9.287 eV/A. As the top surface
contains more atoms than the bottom surface (720 versus 696), the force per atom
fx is difference on the two boundaries: on the top surface fx=0.01290 eV/A; on the
bottom surface, fx=-0.01334 eV/A.
Example
CCOORRNNEELLLL U N I V E R S I T Y
MAE 715 – Atomistic Modeling of Materials
N. Zabaras (04/02/2012) 64
From the linear relationship between dislocation core position and time, a
constant velocity under specific stress is predicted. Finding the slope, for the case
at temperature 300K and shear stress 300 Mpa, the velocity is approximately
1176 m/s.
50 100 150 200 250 300
200
400
600
800
1000
1200
Data: Data24_B
Model: Line
Equation: y = A + B*x
Weighting:
y No weighting
Chi^2/DoF = 2614.81905
R^2 = 0.9829
A 125.13333 ±47.60437
B 3.70686 ±0.24447
stress (Mpa)
ve
locity (
m/s
)
The dislocation velocity increases nearly, but not exactly, linearly with shear
stress. Fitting this curve, we can obtain a viscous drag coefficient B=σb/v. In this
case, B = 7.17e-5 pa.s.
CDS=[1.5, 10]A