Middle 

East Technical University – Department of Metallurgical & Materials Engineering

Monte Carlo Methods at the Mesoscale
Prepared and Presented by :

Assoc.Prof.Dr. Caner Şimşir

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 1/14

 Modeling Grain Growth
 The driving force for grain growth is the minimization of the
free energy of the system.
 Each boundary has a positive energy, so a model is needed
in which the boundary is eliminated or reduced.
 Ising model with zero-magnetic field is such kind of a model:

J N
E     si s j
2 i 1 j|Z
 At low temperatures, Ising model will evolve to eliminate the
boundary to lower the energy of the system.
 Ising Model has only two possible “orientations” thus too few DOF to model grain growth.
 Potts model is a variant of Ising model, in which Q possible values for “spins” are allowed.

1 N N
E    Esi , s j 1   si , s j    Ei  si 
2 i 1 j|Z i 1
 Fi(Si) is an energy associated with specific orientations and can be used to model
recrystallization and abnormal grain growth. For isotropic g-b energy it is reduced to:
N
E0
H
2
  1  
i 1 j| Z
si , s j


2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 2/14
 The Monte Carlo Potts Model

 A trial is picking a site at random and changing its spin to Q-1 of other possible spins. The
energy change is calculated, and Metropolis procedure is employed.
 Potts model can be used at T>0 but is quite often used for T=0. In system consisting of discrete
events, one can associate time with a MC trial. MC time step ΔtMC=1 MCS/N.
 Suppose T=0 and with isotropic surface energy :
 A site is randomly chosen, say encircled 9.
 The energy of the site is 3 E0.
 Any change except 91 will cayse 6E0.
 If 91, the new energy will be 3E0 and move is accepted.
 Site is updated and the procedure is repeated.

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 3/14

 The Monte Carlo Potts Model
 Potts model is inherently a lattice-based
model. A poor choice of lattice to incorrect
and even nonsensical results.
 Consider a square lattice with the nearest-
neighbors:
 The energy of the grey square is 1 E0.
 If greywhite energy is 3 E0.
 ΔE=2E0, will be rejected at T=0.
 A straight boundary inhibits grain
growth !
 A triangular lattice with first nearest
neighbors is more reasonable.

 The general effects of lattice topology on growth in 2D can be understood by comparing the
energy anisotropy of the underlying lattice, where anisotropy is defined as ratio of the
maximum to minimum g-b energy in the Wulff plot.
 Sq(1) has the largest anisotropy, which would yield anisotropic grain growth.
 Similar lattice effects have also been seen in 3D :
 sc(1) and sc(1,2) shows inhibited growth. sc(1,2,3) leads to normal growth.
 fcc(1), fcc(1,2) and hcp(1) all show inhibited grain growth.
 Finite temperatures can roughen the boundaries to reduce lattice dependencies.

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 4/14

 The Monte Carlo Potts Model
 Although it is easy to implement, Potts model
with standard MC sampling and Metropolis
algorithm becomes inefficient as the
calculation proceeds.
 # of “active” sites / # inactive sites is
proportional with <Sg>/<Ag>.
 Assuming circular grains :

 
1/2
Rg  Ag /  S g  2 Rg

Sg 1 Sg 1
 2
Ag  R  t 
Ag Rg g
Ag t1/2

 On a boundary with only one other spin

adjacent, the only acceptable change has a
probability of 1/(Q-1). Thus, the net
probability of a successful spin (P*) is :

1
P*  1/2
t  Q  1

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 5/14

 The N‐Fold Way
 In the N-fold way, every trial is accepted yet still yields the same information.
 Activity of a site is defined as
ai  
*
P E
si  si

 s
i  s *
i 
 
 At T=0K, ai = 0 for a site inside the grain. ai = 1 for a site at the boundary. ai = 2 for a site at the
junction.
 The total activity is derived from the probabilities of all possible flips in the system :
N
A   ai
i 1

 The N-fold way proceeds in the same manner as the Kinetic Monte Carlo calculations :
(a) Pick a site with probability ai
(b) Flip si to one of the si of energetically possible values
(c) Re-calculate activities
(d) Advance time
(e) Repeat

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 6/14

 The N‐Fold Way
 In any kinetic MCM, we must advance time at each step. For Potts model, we also need to
connect it to the standard MC time τ.
 Since A is the total # of flips (@T=0K), the average probability of a potentially successful flip
per site on the simulation lattice is <a>=A/N.
 In the MC Potts model, # of trials per MC step is N/MCS. Potts model does not have a specific
time scale, but we can associate some time τ with a MC step, in which # of flips per time is N/τ.
 The probability of a successful flip in an MCS is N/τ x <a>. However, only 1/(Q-1) of those will
be successful. So the average # of successful flips per MCS is
N a A

 Q  1  Q  1
 The average time associated with a successful flip is the inverse of the rate :

t 
 Q  1
A
 We want an expression that is stochastic and the average is (Q-1)τ/A. Consider a random
number R (0,1), averaging over many of such random numbers <-ln(R)>=1. Thus,
 Q  1
t NFW  ln 
A
 Each step in N-fold way corresponds to many time steps in equivalent MC simulation in which
nothing changes.
2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 7/14
 The N‐Fold Way
 N-fold way yields the same answers as
conventional MC, but in fraction of computer
time.
 For simulations starting from random initial
conditions, at short times, no grains have
started to develop CPM is more efficient.
 However, as the grains grows CPM is
getting more and more inefficient, while
NFW becomes more and more efficient.
 Typically, one start with CPM for a few
hundreds of MCS and switch to NFW for
efficiency.
 NFW is equally applicable to T>0 and
anisotropic interactions.

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 8/14

 Example Applications : Zener Pinning

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 9/14

 Example  Applications: Recrystallization

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 10/14

 Example  Applications: Abnormal Grain Growth

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 11/14

 Example  Applications: Recrystallization

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 12/14

 Applications in Materials Research
 Biphasic Materials
 Sintering
 Magnetocaloric Effect
 Ferroelectric Materials
 Biological Materials

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 13/14

 Middle East Technical University – Department of Metallurgical & Materials Engineering

Thanks for listening.
Prepared and Presented by :

Assoc.Prof.Dr. Caner Şimşir

2019 Spring METE 350 – Multi‐Scale Modeling and Simulation in Materials Science & Engineering 14/14