Modelling and Simulation in

Materials Science and Engineering

PAPER You may also like

- Non-power-law I–V characteristics in Ca-
Development of a grain boundary pinning model doped polycrystalline Y1xCaxBa2Cu3O7
Ke-Xi Xu, Jing-He Qiu and Li-yi Shi
that considers particle size distribution using the - Flux pinning landscape up to 25 T in
SmBa2Cu3Oy films with BaHfO3 nanorods
phase field method fabricated by low-temperature growth
technique
Yuji Tsuchiya, Shun Miura, Satoshi Awaji
To cite this article: Michael R Tonks et al 2015 Modelling Simul. Mater. Sci. Eng. 23 045009 et al.

- Parameter-space topology of models for

cell polarity
Philipp Khuc Trong, Ernesto M Nicola,
Nathan W Goehring et al.
View the article online for updates and enhancements.

This content was downloaded from IP address 128.189.74.29 on 12/02/2024 at 22:48

 Modelling and Simulation in Materials Science and Engineering

Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 (16pp) doi:10.1088/0965-0393/23/4/045009

Development of a grain boundary pinning

model that considers particle size
distribution using the phase field method
Michael R Tonks1, Yongfeng Zhang1, Aaron Butterfield1,2 and
Xian-Ming Bai1
1
Fuel Modeling and Simulation, Idaho National Laboratory, PO Box 1625, Idaho
Falls, ID 83415, US
2
New address: Electrical and Computer Engineering, Ohio State University, 2015
Neil Ave., Columbus, OH 43210, US

E-mail: [email protected]

Received 23 January 2015, revised 4 March 2015

Accepted for publication 25 March 2015
Published 23 April 2015

Abstract
In this work, we expand a grain boundary (GB) pinning model that considers
a range of different spatial distributions of particles to also account for a
distribution of particle sizes. We begin by developing a phase field model that
describes GB and pore interactions and verify it by comparing to molecular
dynamics simulations. We then develop an analytical pinning model that
considers the impact of the particle size distribution, in terms of the mean and
standard deviation of the particle radius. The analytical model is verified by
comparing to simulation results of our phase field model and those of a simple
Monte Carlo model. A significant finding from the model is that the mean
value of the resistive pressure decreases with increasing standard deviation
of the particle radius.

Keywords: zener pinning, phase field method, grain boundary migration,

grain growth

(Some figures may appear in colour only in the online journal)

1. Introduction

Grain boundaries (GBs) significantly impact material performance, as they act as defect
sources and sinks, inhibit dislocation motion and act as segregation sites for precipitate atoms
and impurities. In addition, GBs migrate to reduce the free energy of the material system,
resulting in increasing grain size over time. While the mobility of a GB is an intrinsic property,

0965-0393/15/045009+16$33.00 © 2015 IOP Publishing Ltd Printed in the UK 1

Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

the actual GB migration is severely influenced by impurities and defects [1, 2]. For example,
particles and pores resist GB motion, slowing and even halting grain growth [3, 4].
The earliest treatment of GB pinning was by Zener [5]. Zener determined the resistive pres-
sure for a given volume fraction of spherical, incoherent, immobile and rigid particles. Zener
was able to derive a simple expression for the pinning pressure by making several simplifying
assumptions; he assumed that the GBs were flat, the particles were randomly distributed, only
particles within one radius influence the GBs and that all particles exert the maximum pinning
force throughout the interaction. With these assumptions he derived the resistive pressure to be
3 γGB f Vn
Prmax =
(1) ,
2 r
where r is the particle radius, γGB is the GB energy, f V is the volume fraction of particles and
the exponent n  =  1. The assumption that GBs are flat is not consistent with actual GB behav-
ior, as GBs form a catenoid-like shape around a pinning object. In addition, as particles pin
GBs, a larger fraction are located on the GBs than indicated by the assumption of flat GBs
intersecting randomly distributed particles. Zener’s model has been modified by later work to
remove these assumptions and improve the model predictions [6–10] by employing various
values of the exponent n. For example, n  =  0.93 has been shown to account for the catenoid
shape around the particles [8] and n  =  1/3 considers particles which primarily fall on GBs
[10].
An alternative approach was taken by [11]. They analyzed these various models from
a stereological perspective and found that they are all significantly hindered by geometric
assumptions about the particle and grain shapes and size distributions and by assumptions
with regards to the spatial distributions of the particles. To address issues with regards to the
spatial distribution of the particles, they investigated the resistive pressure as a function of a
stereological parameter R, which quantifies the amount of GB-particle contact beyond ran-
dom. Using R, they found that any spatial distribution of particles could be represented with
an exponent of n  =  1 according to
γ f
Pr = 2R GB V .
(2)
λ¯p

where λ̄ p is the stereological particle size. Note that λ¯p = 4/3r for monosized spherical par-
ticles [11]. Though this approach accounts for any positioning of particles, it does not address
assumptions with respect to the particle size distributions.
In this work, we seek to provide added insight into the resistance against GB migration due
to particles and pores (note that for the rest of this work we say ‘particle’ to represent pores
and particles, unless explicitly stated) with various size distributions using a combination of
analytical theory and mesoscale simulation. We first summarize a phase field model of pore
and GB interaction and verify it against atomistic simulation results. We also add a term to the
model to apply a constant artificial force to drive GB motion. We then use a simple bicrystal
system with a flat GB to study GB/particle interaction under a constant artificial driving force.
We derive an equation for the resistive pressure based on the fractional coverage of the GBs
by particles, i.e. the percentage of GB area currently covered by particles and compare it with
the phase field results. We then modify the model to account for the particle size distribution
and verify it through comparison to phase field results and results from a simple Monte Carlo
model. Finally, we develop an expression for the resistive pressure for a given particle size
distribution that is a function of the volume fraction, where the spatial distribution of particles
is defined by the stereological parameter R as in equation (2).

2
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

2. Pore/GB interaction phase field model

In this study, we used phase field simulations to assist in the development of an expression for
the resistive pressure that accounts for the particle size distribution. To ensure the accuracy of
the model, we verified that the interactions between GBs and particles predicted by the phase
field model were accurate by comparing to MD simulations of Helium (He) bubbles interact-
ing with GBs in Molybdenum (Mo). Thus, we required a model that predicted these interac-
tions, but also allowed us to prevent transport of gas atoms through the bulk or along GBs so
that bubbles do not significantly change size during the simulations.
GB interaction with particles and pores have been extensively modeled with mesoscale
methods. The earliest attempts used Monte Carlo Potts models [9, 12, 13] and later work
applied the phase field method [14–19]. The primary benefit of the phase field model is the
capability to quantitatively model the particle or pore interaction with GBs [14, 20–23].
We employed a simple phase field model of pore and GB interaction based on that pre-
sented in [22]. However, we added the directional model of bulk, surface and GB diffusion
from [24]. In the model, grains are represented by order parameters ηi equal to one within a
corresponding grain and equal to zero in the other grains and within pores, assuming isotropic
GB properties. The pores are represented with a conserved vacancy concentration c that is
equal to one within a pore and some small concentration in the bulk. The variables evolve to
minimize the total free energy in the system, representing microstructure evolution. The free
energy of the system is defined in terms of the model variables as [22]

F=
V
∫ ( κ
2 i
κ
2 )
μf (c, ηi ) + c ∇c 2 + ∑ i |∇ηi|2 dV , (3)

where κc is the interfacial parameter for c, κi is the interfacial parameter for the ith order
parameter and μ is the bulk energy parameter. The bulk energy term is defined as
 ⎛ η 4 η 2 ⎞ ⎛ c4 c 2 ⎞
f (c, ηi ) = ∑ ⎜ i − i ⎟ + ⎜ − ⎟ + a GB ∑ ∑ ηi2ηj2 + as ∑ c 2ηi2, (4)
i ⎝
4 2⎠ ⎝4 2⎠ i j>i i

where aGB is set to 1.5 to ensure a symmetric diffuse interface [25] and as  =  aGB γs/γGB with
surface energy γs and GB energy γGB.
The evolution of the concentration c is defined by a Cahn–Hilliard equation according to
∂c δF
= ∇ ⋅ M∇
(5) ,
∂t δc(r, t )
where r is the spatial position, t is the time and M is the concentration mobility rank two ten-
sor that accounts for directional diffusion along surfaces and GBs. The evolution of the order
parameters ηi is defined by an Allen–Cahn equation, as
∂ηi δF
(6)= −L ,
∂t δηi(r, t )
where L is the scalar order parameter mobility and for this model is assumed to be equal for
all GBs.
The model has a number of parameters that must be determined as functions of measur-
able quantities in order to make a material specific model. These quantities are the GB energy
γGB, surface energy γs, effective vacancy or gas diffusivity Deff and the GB mobility MGB. For
the order parameter values, we use the model presented in [25], which gives the relationships

3
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

3
κi = γGBlGB
(7)
4

4 MGB
L=
(8)
3 lGB

γ
μ = 6 GB .
(9)
lGB

where lGB is the interfacial width that defines the width of the diffuse interfaces in the phase
Q
field model and MGB = M0 e− k bT is the GB mobility with the mobility prefactor M0 and the
activation energy Q. The magnitude of the concentration mobility tensor M can be determined
by the fact that the concentration evolution in the bulk should follow Fick’s law of diffusion
for c  ≪  1, such that
∂ 2F
M = D/ 2 = D/(μ(2as − 1)).
(10)
∂c
The concentration c impacts the free energy in the same manner as ηi, therefore
3
κ c = γslGB.
(11)
4
The gas diffusivity matrix defines the total diffusivity due to bulk, surface and GB diffusion
following the model from [24]. The diffusivity matrix is equal to
ˆb +D
D = Deff (D
(12) ˆs+D
ˆ GB),

where the effective diffusivity

Em
(13) −
Deff = D0 e k bT
with the diffusivity prefactor D0, migration energy Em, Boltzmann constant kb and temperature
ˆ b), surface diffusivity
T. The individual normalized diffusivity tensors for bulk diffusivity (D
ˆ ˆ
(Ds) and GB diffusivity (DGB) are determined as

ˆ b = wb I
D
(14)

ˆ s = wsc 2(1 − c 2 )Ts

D
(15)

ˆ GB = wGB ∑ ∑ ηiηj TGB,

D
(16)
i=1 j≠i

where wb, ws and wGB are weighting parameters and Ts and TGB are surface and GB projection
tensors, respectively. The surface projection tensor is a function of the concentration accord-
ing to
∇c ∇c
Ts = I −
(17) ⊗ .
∇c ∇c

4
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

The GB projection tensor is a function of the order parameters as

∇ηi − ∇ηj ∇ηi − ∇ηj
TGB = I −
(18) ⊗ .
|∇ηi − ∇ηj | |∇ηi − ∇ηj |
The weights wb, ws and wGB define the relative importance of the bulk diffusion, surface diffu-
sion and GB diffusion, respectively. In order to restrict transport of gas through the bulk and
prevent gas bubble growth/shrinkage, we used the parameter values wb  =  wGB  =  1  ×  10−4
and ws  =  1.0.
To simplify our investigation of the resistive pressure, we apply a constant artificial driving
force to the GBs. The artificial driving force is added to the model by adding an additional
term to equation (6) according to

∂ηi ⎛ δF ⎞
= −L ⎜
(19) + Afd ⋅ ∇ηi⎟.
∂t ⎝ δηi ⎠
To derive an expression for A, we use a similar approach to that taken in [26]. The velocity of a
GB moving due to this driving force is v  =  MGB f d. The velocity of a flat GB (i.e. ∂F/∂ηi  =  0)
that is perpendicular to the x-axis in the phase field model using equation (19) is
∂η ∂η
v = i / i = −LAfd .
(20)
∂t ∂x
If we equate these two expressions for v, substitute in equation (8) and solve for A, we obtain
3
A = − lGB.
(21)
4
The magnitude of the driving force used in this work is f d  =  21.1 MPa in the x-direction. This
driving force is comparable in magnitude to the curvature driving force for a grain size of 0.2
μm.
The phase field equations are solved using the finite element method using the MARMOT
mesoscale modeling code as outlined in [27]. The concentration is solved by splitting the
fourth order Cahn–Hilliard equation into two second-order equations in the manner summa-
rized in [28]. All of the equations are solved simultaneously using implicit time integration.
We employed mesh adaptivity to reduce the computational expense.
To model GB and pore interaction in the Mo–He system, we required values for the GB
mobility and energy in Mo. We also needed the diffusivity prefactor D0 and the migration
energy Em to calculate the diffusivity, as well as the Mo surface energy γs. We obtained these
values (shown in table 1) from molecular dynamics (MD) simulations. In the MD simulations
we used the interatomic potential developed in Ackland et al [29] for pure Mo and that in
Zhang et al [30] for He–Mo. The GB energy and mobility were obtained using the circular-
embedded grain model [31] with 〈100〉 45° tilt GBs. For the surface energy, the result is cal-
culated using a spherical void with a 12 nm diameter.
In order to verify that the phase field model accurately captures the interaction between
pores and GBs, we compared the phase field results to MD simulation results representing
identical Mo systems containing He bubbles. The reduction in volume of a circular grain
with an initial radius of 20 nm was predicted by both simulation approaches, with ten bubbles
randomly distributed along the GB. The 3D domain was 64.4 nm  ×  64.4 nm  ×  1.93 nm with
periodic boundary conditions and the center axis of the circular grain was parallel with the
z-axis. The ten He bubbles had a radius of 0.3 nm (ten He atoms were used to represent the

5
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

Table 1. Properties of the Mo–He system used in the phase field simulations calculated
using MD.
Property Value
D0 3.325  ×  10−10 m2 s−1
Em 0.999 eV
γGB 1.920 J m−2
σs 2.160 J m−2
M0 3.986  ×  10−6 m4 Js−1
Q 1.031 eV

bubbles in the MD simulations). The simulations were carried out at a temperature of 2700
K. Three simulations were run with different random initial positions of the ten He bubbles
using both approaches. The phase field simulations employed a GB width of lGB  =  0.2 nm.
Examples of the phase field and MD domains are shown in figure 1(a).
Both approaches predicted an initially slow reduction in the grain volume (see figure 1(b)),
while all ten bubbles are in contact with the GB. However, the rate increases as the GB releases
from more and more of the ten bubbles until the final rate of reduction is defined by the intrin-
sic GB mobility, as no bubbles are in contact with the GB. The comparison between these
two distinct methods is reasonably good, though the phase field simulations seem to predict
somewhat slower decrease in the grain size at the early stages of the volume reduction. Thus,
we confidently employed phase field models to inform the development of an analytical model
of the resistive pressure that accounts for a particle size distribution.

3. The analytical resistive pressure model

In Zener’s original model, he considered the resistive force of spherical, incoherent, immobile
and rigid particles on migrating GBs. This resistive force is equal to
Fr = 2πrγGB cos β sin β ,
(22)
where r is the particle radius and β the angle illustrated in figure 2. He determined that the
maximum force occurs at β  =  45°, such that
Frmax = πrγGB.
(23)
Similar expressions have been derived using the argument that the particle takes the place of
some GB area, reducing the GB energy [7]. By making the simplifying assumptions men-
tioned previously, Zener determined that the resistive pressure is equal to equation (1). The
GB velocity is a function of the GB mobility M, the driving pressure P and the resistive pres-
sure Pr according to
v = M (P − Pr ),
(24)
where Pr  ⩽  P.
Zener based his resistive pressure on the volume fraction of particles f V, as this is an
observable and measurable quantity to allow for validation of the model and facilitate its
applicability in real systems. However, to quantify the resistive pressure as a function of the
volume fraction, he had to assume random positioning of particles with respect to GBs.
Here, we derive an expression that makes no assumptions about the positioning of the
particles by basing it directly on the GB fractional coverage f c, or the percentage of GB area

6
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

(a)

2500
Grain volume (nm )

2000
3

1500

1000

500 Phase field

MD
0
0 0.5 1 1.5 2 2.5
Time (ns)

(b)

Figure 1. Verification of the phase field model by comparing to MD results. A shrinking

Mo circular grain with ten He bubbles distributed along the grain boundary is modeled
by both approaches, where examples of the phase field (left) and MD (right) simulation
domains are shown in (a). The volume of the shrinking grain versus time predicted by
both methods is shown in (b). Note that points from three different simulations for each
approach with different initial bubbles positions are shown in the plot. The phase field
results show reasonably good agreement with the MD results.

currently covered by particle area. We begin with the resistive force, from equation (22) and
multiply by the the number of particles per unit area in contact with the GB to get the resistive
pressure
NpGB
Pr = 2πrγGB cos β sin β
(25) ,
AGB
where NpGB is the total number of particles in contact with the GB, AGB is the total GB area and
we assume that all particles interact with the GB with the same angle β. If we substitute in the
expression for the fractional coverage fc = NpGBAb /AGB, we obtain
f
Pr = 2πrγGB cos β sin β c
(26)
Ab

2γ f cos β sin β
= GB c
(27)
r

7
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

Figure 2. Illustration of the interaction between a particle and a GB that meet at an

angle β, where r is the particle radius, v is the GB velocity and Fr is the resistive force
caused by the particle.

where Ab  =  πr2 is the cross section of a single particle on the GB. The max resistive pressure
would occur at β  =  45° with a value of
γ f
Prmax = GB c .
(28)
r
The relationship between the fractional coverage and the volume fraction depends on the spa-
tial distribution of the particles and is discussed in section 5 of this work.
As the GB passes the pinning particles, it does not stay flat but rather takes on a catenoid
shape, as shown in figure 2. To account for the catenoid shape, Zener’s model was modified
with the exponent n  =  0.93 [8], thus increasing the resistive pressure. To account for the cat-
enoid shape in our model, we include a weighting term W to give
γ f cos β sin β
Pr = 2W GB c
(29)
r

γGB fc
Prmax = W .
r
The value for W was determined by comparing to phase field simulations of GB migration
with pinned boundaries.
We modeled a Mo bicrystal with dimensions of 0.2 μm  ×  0.4 μm  ×  0.4 μm, with a flat
GB driven by the artificial driving force at T  =  800 K using the values from table 1. Because
the domain size was larger than in the previous simulation, we employed a larger GB width of
lGB  =  3.1 nm to allow for a larger minimum mesh size to reduce the computational expense.
Various numbers of He bubbles were lined up in the path of the migrating GB on a plane
parallel to the GB plane, as shown in figure 3(a). In each simulation, all bubbles had the same
radius and simulations were run using radii of 7.0, 9.0 and 11.0 nm. The 3D simulations were
run on 480 processors for approximately 48 h.
A plot of the grain volume versus time for various numbers of 11 nm radius bubbles is
shown in figure 3(b). Note that the change in volume of the grain is linear until the GB comes
in contact with the bubbles. At this time, the rate increases as the GB is pulled towards the
bubbles. This is consistent with the resistive force equation (equation (22)) for values of β less
then zero. As the GB moves past the bubbles, the velocity of the GB decreases until the GB
reaches an angle of β  =  45°, where the resistive force is maximum. The GB maintains a fairly

8
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

0.03 20

0.025

Grain Volume (µm )

3
15

(MPa)
0.02

0.015 5 bubbles 10

max
10 bubbles
0.01

Pr
20 bubbles 5 r = 7 nm
0.005 30 bubbles r = 9 nm
no bubbles r = 11 nm
0
0 2000 4000 6000 0 0.02 0.04 0.06 0.08
Time ( µs) GB coverage

(a) (b) (c)

Figure 3. Comparison between the analytical model (equation (29)) and phase
field simulation results modeling a flat GB driven by an artificial driving force and
interacting with aligned bubbles with a single constant radius. An example of a phase
field simulation with twenty r  =  9 nm radius bubbles is shown in (a), the grain volume
of the shrinking grain versus time predicted by the phase field simulations for various
numbers of aligned r  =  11 nm bubbles is shown in (b) (where the linear fits used to
calculate the maximum resistive pressure are shown as dash–dotted lines for the 20 and
30 bubble simulations) and the comparison of the maximum resistive pressure predicted
by the analytical model to that obtained from the phase field simulations is shown in
(c). The fit value of W  =  1.1 seems to be independent of bubble radius and number of
bubbles for this system.

constant velocity for some time at this point of maximum resistive force, until bubbles begin
to be released from the GB. Eventually, the GB releases from all bubbles and its rate returns
to its value before it came in contact with the bubbles. Though all the curves follow this same
trend, the magnitude of the changes in velocity increase with increasing numbers of bubbles.
To simplify the comparison between our phase field results and the analytical equa-
tion defining the resistive pressure as a function of the fractional coverage, we focus on the
time at which the contact angle β is near 45°, when the resistive force is a maximum. To
determine the GB velocity, we fit a line to the phase field data (shown for the 20 and 30 bubble
simulations in figure 3(b)) during this period and use the slope to calculate the velocity. From
this velocity, we calculate Prmax from equation (24) and the value of the artificial driving force.
We determined the value for W by fitting equation (29) to the values for Prmax calculated
from the simulation results with r  =  7 nm, giving W  =  1.1. Using this value, we compared
the model predictions to the phase field results for r  =  9 nm and r  =  11 nm, as shown in fig-
ure 3(c). The phase field results clearly show a linear relationship between Prmax and f c and an
inverse relationship with the bubble radius. In addition, the predicted values compare closely
with the phase field results, which seems to indicate that the value for W is independent of the
value of f c and r for this system.
As a final test of the analytical model and the W value, we compared the analytical model
predictions from equation (29) to the circular grain results from figure 1. Since the bubbles
do not detach from the GB simultaneously, the number of bubbles in contact with the GB
was recorded from the simulation and used to calculate the value of f c over time. As the GB
migrates through the ten He bubbles, the β contact angle is different for each bubble, varying
from approximately  −45° to 45°. However, we found that an average contact angle of 18°
provided the best comparison with the phase field results, probably because the GB spends the
most time near β  =  45°, pushing the average above 0°. The model predictions compare well

9
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

2500

Grain volume (nm )

2000

3
1500

1000

500 Phase field

Analytical
0
0 0.5 1 1.5 2 2.5
Time (ns)

Figure 4. Comparison between the phase field simulation results for a circular grain
with ten bubble and the analytical model (equation (29)), showing excellent agreement.
Note that f c for the model is estimated from the number of bubbles in contact with
the GB and an average contact angle value β  =  18° (as the transient contact angle is
different for each bubble) was found to produce the best comparison with the phase
field results.

with the simulation results, as shown in figure 4, providing some evidence that W is independ-
ent of driving force.

4. Resistive pressure model with a particle size distribution

Both Zener’s pinning model from equation (1) (for randomly distributed particles) and
our new model from equation (29) (for any spatial distribution) define the resistive pres-
sure for particles with the same radius r. However, in reality the particles will have
a distribution of sizes and the variation in the particle radius could impact the resis-
tive pressure. In this section we modify equation (29) to account for the particle size
distribution.
For a set of particles with radii that vary according to a given probability density function
f (r˜), the resistive pressure becomes a stochastic function, as does the GB fractional coverage.
max
To determine the expected, or average, value of the stochastic resistive pressure E (P˜r ) we
must account for the stochastic nature of the fractional coverage. For a given fractional cover-
age f c and known particle radius distribution with mean r̄ and standard deviation σr, the total
maximum resistive pressure due to particles with varying radii r̃ can be expressed by modify-
ing equation (29) to obtain

GB
Np
fc (r˜i )
max
P˜r = WγGB
(30) , ∑
i=1
r˜i

where NpGB is the number of particles on the GB and the fractional coverage for a single par-
ticle fc (r˜i ) = πr˜i2 /AGB with the grain boundary area AGB. The expected value of the resistive
pressure is
 ⎛ ⎞ ⎛ NpGB ⎞
⎜ ˜ max⎟ WγGBπ ⎜
E ∑ r˜i⎟ =
WγGBπ GB
E ⎜Pr ⎟ =
⎜ ⎟ AGB ⎜ i = 1 ⎟
Np r¯. (31)
⎝ ⎠ ⎝ ⎠
AGB

10
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

The fractional coverage can be calculated from

GB
Np
πr˜ 2
f˜c = ∑ i .
(32)
i=1
AGB

As f c is known, it is equal to the expected value according to

 ⎛ GB
⎞
π ⎜
r˜i2⎟
Np
π
˜ ∑
AGB ⎜ ⎟
fc = E (fc ) = E = NpGB(r¯ 2 + σi2 ), (33)
⎝ ⎠
i=1
AGB

where E (r˜i2 ) = r¯ 2 + σi2, from the definition of the standard deviation. Solving this expression
for NpGB and substituting into equation (31) gives

max r¯ 2 γ f
E (P˜r ) = W GB c
r¯ + σr
2 2
r¯

1 γ f
= W GB c .
()
2
(34) σ
1 + r̄r r¯

This expression is equivalent to equation (29) when the standard deviation is zero and it indi-
cates that the value of the resistive pressure decreases as the square of the ratio of the standard
deviation to the mean of the particle radius increases. Thus, equation (29) defines the upper
bound of the pinning pressure and is a good approximation for small ratios of σr to r̄ .
As with the deterministic model, we verified this model by comparing to phase field simu-
lation results. These simulations were similar to those from the previous section, but the bub-
ble radii were randomly sampled from a normal distribution with a given mean and standard
deviation. We conducted a series of simulations with a mean bubble radius r̄ = 11 nm and
standard deviation σr  =  2 nm (see figure 5(a)), which predicted resistive pressures in good
agreement with the analytical model (figure 5(b)). However, accurately verifying the statistics
from the phase field simulation results would require hundreds of simulations, which would
be computationally prohibitive. Therefore, we developed a Monte Carlo model for further
verification. In the model, random radii r̃i are generated from a normal distribution for NpGB
bubbles in contact with the GB. The stochastic resistive pressure is generated according to
equation (30) (with W  =  1.1) and the total stochastic fractional coverage is calculated from
equation (32). For each number of bubbles for a given bubble radius mean and standard devia-
tion, the stochastic model was applied for 10 000 different sets of random bubble radii.
max
The expected values of the resistive pressure E (P˜r ) predicted by equation (34) compared
well with the values predicted by the Monte Carlo model and by the phase field simulations
with random bubble radii, as shown in figure 5(b). A significant finding from this work is that
for a given fractional coverage, the mean value of the resistive pressure is a function of the
standard deviation, as shown in equation (34) and demonstrated in figure 5(c). The results also
demonstrate that the impact of a given particle radius standard deviation decreases as the mean
radius increases (see figure 5(c)).

5. Model with a particle size distribution in terms of the volume fraction

Equation (34) defines the resistive force due to spherical particles with radii that vary accord-
ing to a normal distribution as a function of the fractional coverage f c. However, f c is difficult
11
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

20
12 σr = 0 σ =0
r
15 10 σr = 2 σr = 2
15

(MPa)
(MPa)
(MPa)
8 σ =4 σr = 4
r
10 10
6

max
Pmax
Pmax

Pr
r
r
Monte Carlo 4
5 5
Analytical 2 r̄ = 7 nm r̄ = 11 nm
Phase field
0 0
0.02 0.04 0.06 0.08 0 0.01 0.02 0.03 0 0.02 0.04 0.06 0.08
GB coverage GB coverage GB coverage

(a) (b) (c) (d)

Figure 5. Verification of our resistive pressure model that is a function of the GB

fractional coverage and considers particle size distributions (equation (34)). An
example of a phase field simulation with bubble radii following a normal distribution
with a mean radius of r̄ = 11 nm and standard deviation of σr  =  2 nm is shown in (a), a
comparison of the analytical model, Monte Carlo results and the phase field simulations
for r̄ = 11 nm and σr  =  2 nm is shown in (b) and the analytical model shows excellent
agreement with the phase field simulations and the Monte Carlo model. The variation of
the resistive pressure with standard deviation for two mean radii is shown in (c), where
the line shows the analytical model and the symbols show the mean values from the
Monte Carlo simulations.

to measure and changes with time. In general, the volume fraction f V is a much more valu-
able means of quantifying the particle density. To modify equation (34) to be a function of f V
requires a relationship between the GB fractional coverage f c and the volume fraction f V for
various spatial distributions of particles. The stereological parameter R from [4] can represent
any spatial distribution, therefore we developed a relationship between f c and f V using R for
particles with a given size distribution.
The stereological parameter R quantifies the amount of GB-particle contact beyond ran-
dom, i.e.
f GB
R= V ,
(35)
fV
where f VGB is the volume fraction of particles in a volume for which average particles could
be in contact with the GB, i.e. VGB = 2AGB r¯, where AGB is the GB area. When the particles are
randomly distributed, f VGB = fV and R  =  1. From this definition of R, we can derive an expres-
sion for the GB coverage f c that is a function of the known volume fraction f V,
3RfV r¯(r¯ 2 + σr2 )
fc =
(36) ,
2 r¯ 3 + 3r¯σr2
where the derivation is shown in the appendix. If we substitute this expression into equa-
tion (34), we obtain

max r¯ 3 3 γ f
E (P˜r ) = WR GB V
r¯ + 3r¯σr 2
3 2
r¯

1 3 γ f
= WR GB V .
()
2
(37)
1+3
σr 2 r¯
r̄

12
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

The expression is again a function of the square of the ratio of the standard deviation to the
mean of the particle radius, as in equation (34). When the standard deviation of the particle
radius is equal to zero, this expression becomes
3 γ f
Prmax = WR GB V ,
(38)
2 r
which is equivalent to equation (2) for W  =  1, from [11]. For particles with the standard devia-
tion of the particle radius equal to zero and a random spatial distribution (R  =  1), the expres-
sion becomes identical to Zener’s equation (1) for W  =  1.
We verified this model by again comparing to Monte Carlo simulation results. We
randomly selected the particle radii as done in the model from the previous section,
however we now also randomly selected a uniformly distributed center location for
each particle. For a given GB location, only particles for which their center location
is within their radius of the GB location contribute to the maximum resistive pressure
calculated with equation (34) (with W  =  1 .1), though all particles are included in the
max
calculation of the volume fraction. A number of GB locations were employed and P̃r
was calculated for 10 000 sets of randomly selected particle radii and locations. The
analytical model compared well with the expected value of the Monte Carlo simulation,
as shown in figure 6.
Equation (37) predicts the resistive pressure for particles with a spatial distribution
defined by the stereological parameter R and with a given size distribution with a mean
particle radius of r̄ and a radius standard deviation of σr. As the square of the ratio of the
standard deviation of the radius to its mean increases, the resistive pressure decreases, as
shown in figure 6. Thus, a given standard deviation has a smaller impact on a system with
a larger mean particle radius. In addition, equation (2) [11] provides an upper bound for
the resistive pressure. From this result, it can be concluded that previous work that has not
considered the impact of the particle size distribution has tended to over predict the resis-
tive pressure, though not significantly if the ratio of the standard deviation to the mean
particle radius was small. While we have verified equation (37) in this work, comparisons
with experimental grain growth data with known particle size distributions are required to
validate the model.
It can facilitate experimental validation to capture the particle pinning effect through an
effective mobility Meff, rather than the resistive pressure as done in equation (24). Thus, we
end this work by defining an effective mobility for particles with a spatial distribution defined
by R and with a given particle size distribution. To accomplish this, equation (24) is equated to
the GB velocity of an effective system without a resistive pressure and we solve for Meff, i.e.
Meff P = M (P − Pr )
(39) ⎛ P⎞
Meff = M ⎜1 − r ⎟.
⎝ P⎠
If we assume that the driving force is due to curvature, such that P  =  2γGB/d, where d is the
grain size, we can obtain a final expression for the effective mobility by substituting equa-
tion (37) into equation (39) to obtain
⎛ ⎞
⎜ 1 3 WRfV d ⎟
Meff = M ⎜1 −
(40) ⎟
⎜ 1 + 3 r¯r 4 r¯ ⎟⎠
⎝
σ 2
()
13
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

8 20
σr = 0 σr = 0
σr = 1 σr = 1
6 15

(MPa)

(MPa)
σ =2 σ =2
r r

4 10
Pmax

Pmax
r

r
2 5
r̄ = 7 n r̄ = 11 nm
5 10 15 0.02 0.04 0.06
Volume fraction −3
Volume fraction
x 10

Figure 6. Verification of the analytical model that is a function of the volume fraction
and accounts for the impact of a particle size distribution (equation (37)) using Monte
Carlo simulations for particles with random spatial distribution, where the dashed lines
show the analytical model and the symbols show the the mean values from the Monte
Carlo simulations. The analytical model compares well with the Monte Carlo results for
an average radii of r̄ = 7 nm (left) and r̄ = 11 nm (right).

6. Conclusions

Particles and pores resist GB motion, applying an opposing pressure. Here we add to
the significant work that has been done in the past looking at the impact of different
spatial distributions of particles with respect to the GBs by developing an expression
that accounts for a particle size distribution. We first developed a phase field model that
accounts for pore and GB interaction and verified it by comparing to MD simulation
results. We then developed an analytical model for the resistive pressure that is a function
of the GB fractional coverage rather than the particle volume fraction. We used results
from phase simulations of a flat GB driven with an artificial driving force interacting with
pores to determine a parameter value to account for the GB catenoid shape that forms
around particles. We then modified the model to account for a particle size distribution
and verified it by comparing to phase field and Monte Carlo simulation results. Finally, we
developed an expression for the resistive pressure for a given particle size distribution that
is a function of the volume fraction, where the spatial distribution of particles is defined
by the stereological parameter R as in [11]. A significant finding from the model is that the
mean resistive pressure decreases as the standard deviation of the bubble radius increases.
Note that the model needs to be validated by comparing to experimental grain growth data
with known particle size distributions.

Acknowledgments

This work was funded by the Department of Energy Nuclear Energy Advanced Modeling and
Simulation program. This manuscript has been authored by Battelle Energy Alliance, LLC
under Contract No. DE-AC07-05ID14517 with the US Department of Energy. The United
States Government retains and the publisher, by accepting the article for publication, acknowl-
edges that the United States Government retains a nonexclusive, paid-up, irrevocable, world-
wide license to publish or reproduce the published form of this manuscript, or allow others to
do so, for United States Government purposes.

14
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

Appendix A.

Here, we derive an expression for the fractional coverage f c in terms of the stereological
parameter R and the volume fraction f V for particles with a distribution of particles sizes. For
this derivation, we assume that all particles in the material have a radius that follows the same
distribution with mean r̄ and standard deviation σr. We define R as
f GB
R= V ,
(A.1)
fV
where f VGB is the volume fraction of particles in a volume for which average particles could be
in contact with the GB, i.e. VGB = 2AGB r¯. This volume fraction can be calculated according to
NpGBV¯b
f VGB =
(A.2) ,
VGB
where NpGB is the number of particles in contact with the GB and V̄b is the mean value of the
bubble volume. The total volume fraction can be expressed as
NpV¯b
fV =
(A.3) ,
Vm
where Np is the total number of particles and Vm is the total volume of the material. Substituting
equations (A.2) and (A.3) into equation (A.1) gives
NpGB Vm
R=
(A.4) .
Np 2AGB r¯

To make the above equation be a function of f c and f V requires expressions for NpGB and Np.
Solving equation (33) for NpGB gives
fc AGB
NpGB =
(A.5) .
π (r¯ 2 + σr2 )
To obtain an expression for Np, we calculate the stochastic volume fraction from
Np
4π
f˜V =
(A.6) ∑ r˜i3.
3Vm i = 1

The volume fraction f V is known and is equal to the expected value of the stochastic volume
fraction, i.e.
 ⎛ Np ⎞
4π ⎜
E ⎜ ∑ r˜i3⎟⎟ =
4π
fV = E (f˜V ) = NpE (r˜ 3). (A.7)
3Vm ⎝ i = 1 ⎠ 3Vm

To determine the expected value of r̃ 3, we assume r̃ is a normally distributed variable with

r˜ = r¯ + σrz˜, where σr is the standard deviation of r̃ and z̃ is a standard normal variable (z¯ = 0
and σz  =  1). Thus the expected value of r3 can be obtained from
E (r 3) = E (r¯ 3 + 3r¯ 2σrz + 3σr2z 2r¯ + σr3z 3)
(A.8)
= r¯ 3 + 3σr2r¯,

15
Modelling Simul. Mater. Sci. Eng. 23 (2015) 045009 M R Tonks et al

where the expected value of z2 is one and of z3 is zero. Substituting this expression into equa-
tion (A.7) and solving for Np gives
3 fV Vm
Np =
(A.9)
4 π (r¯ 3 + 3r¯σr2 )
We then substitute equations (A.5) and (A.9) into equation (A.4) to give

2 fc (r¯ 3 + 3r¯σr2 )
R=
(A.10) .
3 fV r¯(r¯ 2 + σr2 )
By solving this equation for f c we obtain
3RfV r¯(r¯ 2 + σr2 )
fc =
(A.11) .
2 r¯ 3 + 3r¯σr2

References

[1] Cahn J W 1962 Acta Metall. 10 789–98

[2] Michels A, Krill C, Ehrhardt H, Birringer R and Wu D 1999 Acta Mater. 47 2143–52
[3] Doherty R, Srolovitz D, Rollett A and Anderson M 1987 Scr. Metall. 21 675–9
[4] Patterson B R and Liu Y 1990 J. Am. Ceram. Soc. 73 3703–5
[5] Zener C quoted by Smith C 1948 Trans. Met. Soc. AIME 175 15–51
[6] Gladman T 1966 Proc. R. Soc. Lond. A 294 298–309
[7] Ashby M, Harper J and Lewis J 1969 Trans. Met. Soc. AIME 245 413–20
[8] Hellman P and Hillert M 1975 Scand. J. Met. 4 211
[9] Srolovitz D, Anderson M, Sahni P and Grest G 1984 Acta Metall. 32 793–802
[10] Hillert M 1988 Acta Metall. 36 3177–81
[11] Liu Y and Patterson B 1996 Acta Mater. 44 4327–35
[12] Anderson M, Grest G, Doherty R, Li K and Srolovitz D 1989 Scr. Metall. 23 753–8
[13] Miodownik M, Holm E A and Hassold G N 2000 Scr. Mater. 42 1173–7
[14] Fan D, Chen L-Q and Chen S-P P 1998 J. Am. Ceram. Soc. 81 526–32
[15] Moelans N, Blanpain B and Wollants P 2005 Acta Mater. 53 1771–81
[16] Moelans N, Blanpain B and Wollants P 2006 Acta Mater. 54 1175–84
[17] Moelans N, Blanpain B and Wollants P 2007 Acta Mater. 55 2173–82
[18] Millett P, Wolf D, Desai T, Rokkam S and El-Azab A 2008 J. Appl. Phys. 104 033512
[19] Wang N, Wen Y and Chen L-Q 2014 Phil. Mag. Lett. 94 794–802
[20] Millett P C, El-Azab A, Rokkam S, Tonks M and Wolf D 2011 Comput. Mater. Sci. 50 949–59
[21] Millett P C, El-Azab A and Wolf D 2011 Comput. Mater. Sci. 50 960–70
[22] Millett P C and Tonks M 2011 J. Nucl. Mater. 412 281–6
[23] Chang K and Chen L-Q 2012 Modelling Simul. Mater. Sci. Eng. 20 055004
[24] Ahmed K, Yablinsky C, Schulte A, Allen T and El-Azab A 2013 Modelling Simul. Mater. Sci. Eng.
21 065005
[25] Moelans N, Blanpain B and Wollants P 2008 Phys. Rev. B 78 024113
[26] Tonks M R, Zhang Y, Bai X and Millett P C 2014 Mater. Res. Lett. 2 23–8
[27] Tonks M, Gaston D, Millett P, Andrs D and Talbot P 2012 Comput. Mater. Sci. 51 20–9
[28] Zhang L, Tonks M R, Gaston D, Peterson J W, Andrs D, Millett P C and Biner B S 2012 J. Comput.
Phys. 236 74–80
[29] Ackland G J and Thetford R 1987 Phil. Mag. A 56 15–30
[30] Zhang Y, Millett P C and Tonks M 2011 Comput. Mater. Sci. 50 3224–9
[31] Tonks M R, Zhang Y, Biner S, Millett P C and Bai X 2013 Acta Mater. 61 1373–82

16