a molecular dynamics study of self-diffusion in stoichiometric … · 2018. 5. 15. · a molecular...
TRANSCRIPT
arX
iv:1
802.
0295
6v2
[co
nd-m
at.m
trl-
sci]
13
May
201
8
A Molecular Dynamics Study of Self-Diffusion in
Stoichiometric B2-NiAl crystals
Marcin Mazdziarz, Jerzy Rojek, and Szymon Nosewicz
Institute of Fundamental Technological Research Polish Academy of Sciences,
Pawinskiego 5B, 02-106 Warsaw, Poland
E-mail: [email protected]
Abstract. Self-diffusion parameters in stoichiometric B2-NiAl solid state crystals
were estimated by molecular statics/dynamics simulations with the study of required
simulation time to stabilise diffusivity results. An extrapolation procedure to improve
the diffusion simulation results was proposed. Calculations of volume diffusivity for
the B2 type NiAl in the 1224 K to 1699 K temperature range were performed using
the embedded atom model potential. The results obtained here are in much better
agreement with the experimental results than the theoretical estimates obtained with
other methods.
PACS numbers: 02.70.Ns, 66.30.h, 66.30.Fq
Keywords: NiAl, nickel-aluminium, diffusivity, molecular dynamics, molecular statics,
embedded-atom method, sintering
1. Introduction
For numerical material modelling, identification of model parameters for the material
is one of the key tasks which commonly causes many difficulties. A straightforward
approach to parameter determination relies on direct measurement of the wanted
parameters. In some cases, however it is difficult or very expensive to design experiments
that allow direct measurements of the required parameters. Alternatively, parameters of
the model can be identified by numerical simulations at a lower scale. This methodology
used in multiscale modelling is an appealing option, notably when experimental
measurements are not feasible or experimental data are not accessed [1].
This work presents the determination of the diffusion parameters in stoichiometric
B2-NiAl crystals using the molecular dynamic simulations. The nickel-aluminium (NiAl)
type intermetallics are modern materials with low density and advantageous mechanical
properties. These are often used as matrix materials in composites manufactured using
sintering technologies [2]. There is a growing demand for modelling support of design
and optimisation of sintering processes. Sintering is a diffusion driven process, and
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 2
the knowledge of diffusivity of investigated materials is usually necessary to define
parameters of sintering models [3].
Diffusivity can be estimated using both atomic and atomistic modelling. These
approaches can be used to model sintering directly, however, this approach has some
limitations. Direct simulations of particle sintering are limited only to nanoscale,
whereas in general, average particle sizes in NiAl powder can be three orders of
magnitude bigger, see e.g.[4, 5]. Therefore, atomistic models have been aimed to
simulate diffusion in order to evaluate the diffusion properties of NiAl material which
can be used in sintering models [3]. Generally, the diffusivity of NiAl material was
derived from the so called, static and dynamic simulations. Using molecular statics
(MS) approach and modified analytic embedded-atom method (MAEAM) potential
the activation, formation and migration energies of Ni self-diffusion in intermetallic
compound NiAl have been calculated for five diffusion mechanisms in [6]. However,
diffusivity was not analysed during this study. The results show that Ni diffusion
is predominated by the triple-defect diffusion mechanism since it needs essentially
the lowest migration and activation energy among the five diffusion mechanisms
considered. Point defect energetics and the activation energy in NiAl were determined
by molecular statics in [7, 8, 9, 10]. The effective formation energies for a homogeneous
thermodynamically stable ordered compound B2-NixAl1−x were determined by a
combination of the ab initio electron theory with a generalized grand canonical statistical
approach in [11, 12, 13]. By means of molecular statics simulations and the embedded
atom model (EAM) potentials for the assumed six-jump cycle (6JC) mechanism in
the temperature spectrum from 800 to 1500K [14] have found the pre-exponential
factor, D0=1.3 x 10−5(m2s−1) and the activation energy, Q=3.12(eV/atom) (EM=2.44
and EF=0.68). By means of first-principles density functional theory (DFT) calculations
[15] five postulated diffusion mechanisms have been analysed for Ni in NiAl in the
temperature range from 1200 to 1500K and derived the activation energies Q=2.99-
4.15 (eV/atom) and the pre-exponential factor D0=0.46-1.49 x 10−5(m2s−1). Even for a
simpler monatomic fcc Fe system, the DFT calculated self-diffusionD0 is underestimated
by two orders of magnitude, see [16]. A universal tendency can be seen, that the
activation energy Q determined from molecular statics and first-principles are generally
in good agreement with experimental data but pre-exponential factor D0 is significantly
undervalued, see also discussion in Sec.2. It is worth noting that such calculations do
not straightforwardly provide the thermal properties of a material . The harmonic
approximation accuracy even taking into account, the influence of temperature on
other properties of a material, is far from having reached a satisfactory level [17].
Anharmonic effects can be very significant but there are very few analytic calculations
taking into account higher-order terms using perturbation theory, whereas numerical
molecular dynamics calculations can account for anharmonicity to all orders. Some
effects associated with thermal expansion at constant pressure can be described by the
quasiharmonic approximation but anharmonic effects, which explicitly depend on the
magnitude of the atomic vibrational displacements, are present even at fixed volume;
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 3
see [18]. It seems, therefore, that molecular dynamics is a reliable method that allow
one to take into account all anharmonic effects [19]. The experimentally specified
Debye temperature of nominally stoichiometric NiAl varies from 470K to 560K [20]. In
connection to our research, the relevance of the Debye temperature can be twofold. First,
it can provide an estimation of the temperature above which all phonon states become
occupied. This would mean that below Debye temperature the classical considerations
are not likely to apply. Second, the Debye temperature is widely adopted as an
estimation of the temperature under which the underestimation of anharmonicity in
the QHA, QHAD is not very important, even for systems with reduced symmetry (e.g.
with defects). The stochastic Monte Carlo method is another static approach to study
diffusion. The Monte Carlo simulations represent a broad spectrum of computational
algorithms that rely on repeated random sampling to simulate the time evolution of
some processes which appear in nature [21]. It is a fairly straightforward and helpful
technique that can output a sequence of configurations and the times at which transitions
happen between these configurations [22]. Atomistic diffusion in faced-centered cubic
NiAl binary alloys, Ni containing 0.05, 0.1, 0.15, 0.2 and 0.25 atomic fraction of Al, was
examined by Kinetic Monte Carlo (KMC) method by [23]. Fundamental data obtained
from ab initio computer calculations were used as an input to these simulations. The
derived activation energies QNi were in good agreement with experimental data but
pre-exponential factor D0,Ni is unfortunately again highly undervalued.
The most natural dynamic technique to calculate diffusivity is to use molecular
dynamic . An overview of recent molecular dynamics simulations in equiatomic Ni-Al
systems can be found in [24]. In [25], diffusion and interdiffusion in binary metallic
melts, Al-Ni and Zr-Ni, was analysed by molecular dynamics computer simulations
and the mode coupling theory of the glass transition. But the authors claimed
that their ”results not to be quantitative predictions for real Al-Ni or Zi-Ni melts,
but rather for model systems that allow us to understand the relevant mass transport
mechanisms in such melts qualitatively and semiquantitatively”. Molecular dynamics
study of self-diffusion in liquid Ni50Al50 alloy was carried out in [26, 27, 28] and MD
simulation of 2D diffusion in (110) B2-NiAl film in [29]. The first molecular dynamics
simulation of diffusion mechanisms in ordered stoichiometric Ni3Al with the Finnis–
Sinclair interatomic potential was performed by [30], but the author noted that ”results
are at least qualitatively, and in some respect quantitatively, in agreement”. In a
comparative study of embedded-atom methods applied to the reactivity in the NiAl
system [31], the self-diffusion of the liquid Ni-Al mixture was also studied, unfortunately
without giving details of methodology. The consistency of their simulation results with
experiment is rather poor. The authors even stated that ”self-diffusion in a solid state is
much slower and diffusion coefficients are much smaller by several orders of magnitude
than in the liquid state. As a result, MD timescales are generally too short for an
extensive study of these phenomena”, what we refute in this work.
The initial estimation of the volume, surface, and grain-boundary diffusivity for
the B2 type NiAl as the pivotal mechanism of sintering was performed in our previous
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 4
work [32]. That paper presented the methodology , the assumption for the model and
initial results for three temperatures - from 1573 to 1673K. The estimated values were
supposed to be applied to discrete element modelling of the sintering process of NiAl
powder, however such small range of temperatures can be insufficient and can lead to
inconsistency of material data in the context of the connection between atomistic and
microscopic scales. In response to such challenges, the current work concentrates on
improved and extended investigations of volume diffusion determination in NiAl material
in the much wider temperature range - from 1224 to 1699K. Determined materials
parameters will be characterized to better accuracy and agreement with experimental
results. Furthermore, the present paper discusses the inspection of the necessary
simulation time required to ensure the stabilization of diffusivity results. The studied
problem seems to be the one of the major issue of proper determination of diffusivity
results and have a considerable impact on obtained final results. Therefore, it has been
proposed the new procedure to improve the quality of diffusion simulation results, what
is important especially in the analysis of diffusion results in lower temperatures.
2. The mechanism of atomistic diffusion in NiAl
Diffusion is material transport induced by the motion of atoms. A schematic illustration
of volume self-diffusion in the monatomic structure of an atom, from its initial position
into a vacant lattice site, is depicted in Fig.1. The migration energy, EM , has to be
applied to the atom in order to overcome inter-atomic bonds and to move to the new
position.
Figure 1. Schematic representation of the volume self-diffusion process.
Similarly to other solids, self-diffusion is the primary diffusion process in
intermetallics . In binary compounds two tracer self-diffusion coefficients – one for
A atoms and one for B atoms, – are relevant [33]. In the NiAl binary system, we can
identify Ni self-diffusion and Al self-diffusion, where the first one is highly dominative
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 5
in the stoichiometric composition [34, 35, 36]. Interdiffusion, or chemical diffusion,
is different from the tracer diffusion, because it is present in a chemical composition
gradient and not in a homogeneous solid [33].
For the B2 type NiAl, see Fig.3a), various hop mechanisms based on experimental
and theoretical considerations have been suggested [37]. In addition to the dominant
nearest-neighbour (NN) hops in monoatomic structures, next-nearest neighbour (NNN)
hops are also feasible in B2-NiAl due to its open crystal structure. An intriguing property
of B2-NiAl is that NN Al hops from the Al sublattice to the Ni sublattice cannot emerge,
as the end state is predicted to be not mechanically stable. Other suggested mechanisms
include two simultaneous pair-atom (SPA) hop mechanisms, the six-jump cycle (6JC),
two anti-structure bridge (ASB) mechanisms and the triple-defect (TD) sequence.
It is well-known that thermal defects occur at finite temperatures apart from
the constitutional defects. Modelling concepts are usually based on the idea of a
noninteracting point defects gas as suggested by Wagner and Schottky. Thermal defects
in an ordered binary alloy of a fixed composition must occur in a balanced manner in
order to retain the alloy stoichiometry. Hence, since the alloy composition is fixed, they
can occur merely in the composition-conserving combination and any point defects
cannot be a thermal defect alone. There are four basic composition-conserving defects
composed of two types of point defects: Exchange antisite defect (X), Divacancy defect
(D), Triple Ni or simply triple defect (TN) and Triple Al defect (TA), see [38]. It is
worth mentioning that the formation energies of single point defects depend on the
choice of the reference states and they are not insignificant, as opposed to the formation
energies of composition-conserving defects, which do not depend on a specific choice of
the reference state and can be directly compared. Thereby, the first case requires a grand
canonical ensemble, the second a canonical ensemble. In the B2-type intermetallics the
thermal disorder is typical of a triple defect type and the triple defect (TN) formed
by one antisite Ni atom and two Ni vacancies is presumed to be the dominant thermal
defect in NiAl, see [38, 35].
The temperature dependence of diffusivity typically follows an Arrhenius equation
[33] and typically is written as:
D = D0 exp
(
−Q
kBT
)
= D0 exp
(
−EF + EM
kBT
)
, (1)
or in the logarithmic form
log (D) = log (D0)−
(
Q
kB
)(
1
T
)
, (2)
where D is the diffusivity or diffusion coefficient; D0 is the pre-exponential factor; Q is
the activation energy; T is the temperature; kB is the Boltzmann’s constant; EF is the
formation energy and EM is the migration energy.
The thermal equilibrium concentration of point defects is given by the relation:
ND
N= exp
(
−EF
kBT
)
, (3)
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 6
where ND is the number of defects and N is the number of potential defect sites.
Simple Arrhenius behaviour should not however be deemed to be universal. At T >
1500K an upward deviation from this equation can be noted, among others, in B2-NiAl
[14, 38] and bcc-Fe [39].
The activation energy Q for the stoichiometric composition for triple Ni defect (TN)
in the Arrhenius Eq.1 leads to:
Q =
(
ETNi
3
)
+ EM , (4)
where ETNi/3= EF , see [34].
Presently, there exist no directly measured Al tracer diffusion data in NiAl, Al
diffusion cannot be determined experimentally because of the absence of a suitable
isotope, but Ni diffusion in single-crystal NiAl has been determined [40]. For Ni in B2-
NiAl self-diffusion prefactor D0=2.71-3.45 x 10−5(m2s−1) and activation energy Q=2.97-
3.01(eV/atom). It is visible that the spread of results is quite marginal as opposed to
results of self-diffusion in pure Al and in pure Ni. The diffusion prefactor D0 for Al
varies from three orders of magnitude up to eight orders for Ni in different studies [41].
3. Computational methodology
The molecular simulations in this study were made with the use of the Large-scale
Atomic/Molecular Massively Parallel Simulator (LAMMPS) [42] and visualized in the
Open Visualization Tool (OVITO) [43] and Wolfram Mathematica [44].
3.1. Molecular potentials
Two embedded-atom method (EAM) potentials: EAM2002[45] and EAM2009[46] have
been validated for their possible reproduction of NiAl diffusion in the temperature
spectrum from 1224 to 1699K. Using molecular statics approach [47, 48, 49, 50, 51]
the following parameters of NiAl were probed: lattice constant, cohesive energy, bulk
modulus, elastic constants, surface energy, defect energies, migration energies as well as
melting point temperature, see Tab.1. For MS simulations the sample was assumed to
be cube and composed of 103 elementary B2-NiAl cells, see Fig.3a) (∼ 2000 atoms). The
size of the sample allows to fulfil the assumed by Wagner and Schottky condition of non-
interaction of point defects. Energy minimization with periodic boundary conditions
(PBC) applied to all faces of the sample was carried out with a nonlinear conjugate
gradient algorithm [42]. The assumption was, that convergence was reached when the
relative change in the energy and forces, between two successive iterations was less than
10−13. Following the energy minimization procedure an external pressure was applied
to the simulation box to see the volume change effect.
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 7
3.2. Diffusivity simulations
Periodic boundary conditions (PBC) were applied in all simulations. After molecular
statics relaxation, molecular dynamics equilibration for the time period of 1 ns (0.5x106
MD steps, one step = 2 fs) was done. The data gathering simulation covered a time
interval of minimum 1000 ns (500x106 MD steps). Finding the needed simulation time
warranting the stabilization of the results was the first step in the simulation. At a given
temperature and pressure, NPT (constant number of atoms, pressure and temperature)
Nose-Hoover style barostat was used [42, 47]. The diffusivity was calculated as the
average:
Davg. ≡1
2dlimt→∞
⟨
[r(t0 + t)− r(t0)]2⟩
t, (5)
or the instantaneous one:
Dinst. ≡1
2dlimt→∞
∂⟨
[r(t0 + t)− r(t0)]2⟩
∂t, (6)
where, d is the dimensionality (d=2 for surface and grain-boundary diffusion, d=3
for volume); t is the time and⟨
[r(t0 + t)− r(t0)]2⟩
is the ensemble average MSD.
Thermodynamic informations were calculated and stored at intervals of 2 ps (1000 MD
steps).
3.2.1. Volume diffusivity Thermal equilibrium point defect concentration is a function
of temperature and the formation energy EF , see Eq.3. Defect concentration for triple
Ni and Al defect energy found by the molecular potential applied in the MD studies,
see Tab.1, in the temperature interval from 1224 to 1699K is depicted in Fig.2.
TNi
TAl
1300 1400 1500 1600 1700
10-6
10-5
10-4
10-3
Temperature [K]
Defectconcentration
(per
site)
Triple defect concentration
Figure 2. Triple defect concentration for Ni and Al.
Hence, for the cubic computational sample, Fig.3b), using above Eq.3 we can
calculate that one triple Ni defect (TN) for temperature 1224K is in the 223 basic
cells (21296 atoms), for 1244K 213 basic cells (18522 atoms), for 1265K 203 basic cells
(16000 atoms), for 1288K 193 basic cells (13718 atoms), for 1313K 183 basic cells (11664
atoms), for 1341K 173 basic cells (9826 atoms), for 1372K 163 basic cells (8192 atoms),
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 8
for 1406K 153 basic cells (6750 atoms), for 1445K 143 basic cells (5488 atoms), for
1489K 133 basic cells (4394 atoms), for 1540K 123 basic cells (3456 atoms), for 1599K
113 basic cells (2662 atoms) and for 1699K 103 basic cells (2000 atoms), respectively. It
is worth emphasising that the triple defect concentration for Al is about 100 times lower
than for Ni, see Fig.2, what would mean the use of about 100 times larger computational
samples to calculate Al self-diffusion.
3.3. Diffusivity determination from simulation results
In order to answer the question of what value of diffusion D should be read from the
simulation, two procedures were proposed and tested:
I Direct: we take the value of the diffusivity D, Eqs.5,6, from the end of the MD
simulation.
II Extrapolation: we approximate the results of the diffusivity D, Eqs.5,6, versus
simulation time t with a function with a limit, e.g. D(t)=a×exp(b/t), and take its
value for t → ∞.
4. Results
Both considered EAM interatomic potentials , see Tab.1, similarly reproduce the
parameters of B2-NiAl, nevertheless, EAM2002 potential predicts the melting point at
1520K since the experimental value is 1911K. Because of our interest in the temperature
spectrum from 1224 to 1699K for subsequent molecular dynamics computations the
EAM2009 potential, predicting a more reasonable melting temperature of 1780K, was
used.
Figure 3. Atomistic models: a) NiAl basic cell, b) simulation box for volume diffusion.
There is no general rule for how long a simulation has to be run before we reach
the long-time asymptotic behaviour predicted by the Einstein-Smoluchowski relation,
Eqs.5,6. As far as we know the required simulation duration ensuring the stabilization
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 9
Table 1. Material parameters of NiAl from two EAM potentials - Molecular statics
simulations. DFT and experimental data taken from [45, 46]. Migration energy∗
calculated for nearest-neighbour (NN) hop mechanism.
EAM2002 EAM2009 Experiment/DFT
Lattice constant NiAl [A] 2.86 2.832 2.88
Cohesive energy [eV] -4.47 -4.51 -4.50
Bulk modulus B [GPa] 160 159 158
Elastic constant C11 [GPa] 200 191 199
Elastic constant C12 [GPa] 140 143 137
Elastic constant C44 [GPa] 120 121 116
Surface energy [Jm−2] 1.52 2.07 ∼2.2
Exchange energy EX [eV] 2.765 2.075 2.65-3.15
Divacancy energy ED [eV] 2.396 2.569 2.18-3.07
Triple Ni defect energy ETNi [eV] 2.281 2.807 1.58-2.83
Triple Al defect energy ETAl [eV] 5.276 4.406 5.44-6.46
Migration energy∗ ENiM [eV] 2.324 2.49 2.58
Migration energy∗ EAlM [eV] 1.476 1.76 ∼1.6
Melting point [K] 1520 1780 1911
of the slope of the mean-squared displacement (MSD) results was not duly discussed
in the literature. For the molecular dynamics analysis of self-diffusion in bcc Fe [39],
the simulation time was ∼ 41 ns, for hcp and bcc Zr [52] ∼ 20-30 ns and for ordered
stoichiometric Ni3Al in [30] was ∼ 30 ns. For the molecular dynamics simulation of
carbon diffusion in α-iron, the simulation time was ∼50 ns. Similarly, as we can observe
in this work, diffusivity decreases with increasing time and moves towards stabilisation,
see [53].
The conditions where molecular dynamics simulations can be used to calculate
highly converged Arrhenius plots for substitutional alloys with various vacancy
concentrations between 1% and 5% were analysed in [54]. It was found that highly
converged results can be obtained when using an elevated temperature range (i.e.
Tsimulation/Tmelting=0.87-0.98) and an extended simulation time longer than 300 ns. The
higher temperature in the simulations, as well as the higher concentration of defects,
resulted in more convergent results and lower statistical error.
Our observations suggest that the needed simulation time that allows the
stabilization of the MSD slope is substantially longer than that used by mentioned
authors. It can be further observed in Figs.4-7, that the needed simulation time
depends on temperature and as it grows, the time decreases. We also observe, that
instantaneous diffusivity, Eq.6, stabilizes faster than the average diffusivity, Eq.5, and
will be used further as a reference value. It is also observed that in conducted calculations
the simulation time at least 1000 ns is needed. For temperature 1244K the simulation
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 10
time had to be extended to 1800 ns and to 2500 ns for 1224K, respectively.
MSD
0 200000 400000 600000 800000 1×106
0
5
10
15
20
Time [ps]
MSD
[Å^2]
a) Mean Squared Displacement - 1699.864K
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
1.×10-6
2.×10-6
3.×10-6
4.×10-6
5.×10-6
6.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
a) Diffusivity - 1699.864K
MSD
0 200000 400000 600000 800000 1×106
0
1
2
3
4
Time [ps]
MSD
[Å^2]
b) Mean Squared Displacement - 1599.52K
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
1.×10-6
2.×10-6
3.×10-6
4.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
b) Diffusivity - 1599.52K
MSD
0 200000 400000 600000 800000 1×106
0.0
0.5
1.0
1.5
2.0
Time [ps]
MSD
[Å^2]
c) Mean Squared Displacement - 1��������
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
1.×10-6
2.×10-6
3.×10-6
4.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
c) Diffusivity - 1540.295K
MSD
0 200000 400000 600000 800000 1×106
0.0
0.2
0.4
0.6
0.8
1.0
Time [ps]
MSD
[Å^2]
d) Mean Squared Displacement - ��� ���
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
3.×10-6
3.5×10-6
Time [ps]
Diffusivity
[Å^2/ps]
d) Diffusivity - ���������
Figure 4. Mean Squared Displacement and Average and Instantaneous Diffusivity
for different temperature (1699-1489K).
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 11
MSD
0 200000 400000 600000 800000 1×106
0.0
0.1
0.2
0.3
0.4
Time [ps]
MSD
[Å^2]
e) Mean Squared Displacement - ������ !
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
3.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
e) Diffusivity - 1445.47K
MSD
0 200000 400000 600000 800000 1×106
0.20
0.25
0.30
0.35
Time [ps]
MSD
[Å^2]
f) Mean Squared Displacement - "#$%&'()*
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
3.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
f) Diffusivity - +,-./0234
MSD
0 200000 400000 600000 800000 1×106
0.20
0.21
0.22
0.23
0.24
0.25
0.26
Time [ps]
MSD
[Å^2]
g) Mean Squared Displacement - 56789:;<=
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
3.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
g) Diffusivity - 1372.287K
MSD
0 200000 400000 600000 800000 1×106
0.19
0.20
0.21
0.22
0.23
Time [ps]
MSD
[Å^2]
h) Mean Squared Displacement - 1341.452K
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
3.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
>) Diffusivity - 1341.452K
Figure 5. Mean Squared Displacement and Average and Instantaneous Diffusivity
for different temperature (1445-1341K).
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 12
MSD
0 200000 400000 600000 800000 1×106
?@ABC
DEFGH
0.180
IJKLM
0.190
NOPQR
Time [ps]
MSD
[Å^2]
i) Mean Squared Displacement - STUVWXYZ[
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
3.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
i) Diffusivity - \]^_`abcd
MSD
0 200000 400000 600000 800000 1×106
efgij
klmno
0.180
Time [ps]
MSD
[Å^2]
p) Mean Squared Displacement - 1288.341K
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
3.×10-6
Time [ps]
Diffusivity
[Å^2/ps]
q) Diffusivity - 1288.341K
MSD
0 200000 400000 600000 800000 1×106
rstuv
wxyz{
0.180
Time [ps]
MSD
[Å^2]
|) Mean Squared Displacement - }~�������
Avg.
Inst.
0 200000 400000 600000 800000 1×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
Time [ps]
Diffusivity
[Å^2/ps]
�) Diffusivity - ���������
Figure 6. Mean Squared Displacement and Average and Instantaneous Diffusivity
for different temperature (1313-1265K).
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 13
MSD
0 500000 1.0×106 1.5×106
0.162
0.164
0.166
0.168
�����
�����
Time [ps]
MSD
[Å^2]
l) Mean Squared Displacement - 1244.0217K
Avg.
Inst.
0 500000 1.0×106 1.5×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
Time [ps]
Diffusivity
[Å^2/ps]
l) Diffusivity - 1244.0217K
MSD
0 500000 1.0×106 1.5×106 2.0×106 2.5×106
0.162
0.164
0.166
0.168
�����
� ¡¢£
¤¥¦§¨
Time [ps]
MSD
[Å^2]
m) Mean Squared Displacement - 1224.443K
Avg.
Inst.
0 500000 1.0×106 1.5×106 2.0×106 2.5×106
0
5.×10-7
1.×10-6
1.5×10-6
2.×10-6
2.5×10-6
Time [ps]
Diffusivity
[Å^2/ps]
m) Diffusivity - 1224.443K
Figure 7. Mean Squared Displacement and Average and Instantaneous Diffusivity
for different temperature (1244-1224K).
The volume diffusivity for different temperatures from our MD computations,
using Direct(I) and Extrapolation(II) procedure of diffusivity determination from MD
simulation results, is summarized in Tab.2 and Fig.8. Our results fit better to the
averaged experimental data, [40], than these from DFT calculations for triple defect
mechanism (TN) [15]. The conformity of our simulation results to experimental results
from 1700K to ∼1400K is excellent. The use of the Extrapolation procedure(II)
improves diffusion simulation results especially for lower temperatures. At lower
temperatures, the agreement is worse, but better still than the results of the DFT
calculations.
Thereafter, using volume diffusivity results for different temperatures from our
molecular dynamics simulations, Tab.2, and the Arrhenius equation, Eq.2, Ni in B2-NiAl
self-diffusion prefactor D0 and activation energy Q for different temperature intervals
was determined, see Tab.3 and Fig.9. The numerical results have been confronted with
experimental values taken from [40]. For the temperature interval 1699-1489K the
calculated D0 and Q perfectly match the experimental results for Direct(I) procedure of
diffusivity determination from MD simulation results. The Extrapolation(II) procedure
extends this interval into the range 1699-1372K. As we can see, the new approach
improves agreement with experimental data, however the inconsistency problem is still
visible in lower temperatures. Moreover, our NiAl diffusion Tsimulation/Tmelting=0.69-
0.95 is much wider than that suggested by the authors in [54], the low temperature
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 14
Table 2. Volume diffusivity for different temperatures from our molecular dynamics
computations (MDI, Direct procedure I and MDII, Extrapolation procedure II), DFT
[15] calculations and from the experiments [40].
Temperature[K] MDI-D[m2/s] MDII-D[m2/s] Exp.-D[m2/s] DFT-D[m2/s]
1224 1.1688E-16 4.5937E-17 1.4659E-17 1.7081E-18
1244 1.6368E-16 7.001E-17 2.2895E-17 2.6801E-18
1265 2.9969E-16 1.2856E-16 3.6542E-17 4.2985E-18
1288 2.9300E-16 1.3336E-16 5.9743E-17 7.0636E-18
1313 3.2353E-16 1.8037E-16 1.0030E-16 1.1923E-17
1341 3.7173E-16 2.1841E-16 1.7347E-16 2.0738E-17
1372 4.2486E-16 2.4826E-16 3.1014E-16 3.7299E-17
1406 6.3452E-16 4.1901E-16 5.7570E-16 6.9678E-17
1445 6.3920E-16 1.0412E-15 1.1153E-15 1.3590E-16
1489 1.8723E-15 2.255E-15 2.2690E-15 2.7852E-16
1540 3.8250E-15 3.0116E-15 4.8868E-15 6.0459E-16
1599 7.1946E-15 7.7964E-15 1.1251E-14 1.4039E-15
1699 3.5237E-14 3.8241E-15 4.0480E-14 5.1172E-15
MD
MD*
Exp.
DFT
1300 1400 1500 1600 1700
0
1.×10-14
2.×10-14
3.×10-14
4.×10-14
Temperature [K]
Diffusivity
[m^2/s]
a) Diffusivity
MD
MD*
Exp.
DFT
1300140015001600
10-18
10-17
10-16
10-15
10-14
1/Temperature[K]
Diffusivity
[m^2/s]
b) Diffusivity
Figure 8. Volume diffusivity for different temperatures: a) normal scale, b) Arrhenius
plot.
results are very noisy and probably have little physical meaning for diffusion, simply
the diffusion lengths are too short.
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 15
Table 3. Ni in B2-NiAl self-diffusion prefactor D0 and activation energy Q
for different temperature intervals from our molecular dynamics simulations (MDI,
Direct procedure I and MDII, Extrapolation procedure II). Experimental D0=2.71-
3.45 x 10−5(m2s−1) and Q=2.97-3.01(eV/atom), see [40].
MDI MDII
Temp. interval[K] D0[m2/s] Q[eV/atom] D0[m
2/s] Q[eV/atom]
1699-1599 3.51317E-3 3.70975 3.90457E-3 3.71324
1699-1540 9.33994E-5 3.18766 1.82291E-3 3.6036
1699-1489 2.76108E-5 3.01506 3.40504E-5 3.03987
1699-1445 9.66945E-5 3.19004 1.44477E-5 2.92018
1699-1406 1.42095E-5 2.92579 2.60354E-5 3.00134
1699-1372 3.61679E-6 2.73949 2.57373E-5 2.99977
1699-1341 7.74201E-7 2.53194 9.39717E-6 2.86412
1699-1313 1.86944E-7 2.34261 2.97039E-6 2.71067
1699-1288 5.10362E-8 2.17135 1.17749E-6 2.58861
1699-1265 1.38062E-8 2.00049 3.94811E-7 2.4458
1699-1244 8.46639E-9 1.93715 2.5614E-7 2.38976
1699-1224 6.42842E-9 1.90179 2.05872E-7 2.36171
1.00E-09
1.00E-08
1.00E-07
1.00E-06
1.00E-05
1.00E-04
1.00E-03
1.00E-02
1.00E-01
1.00E+00
D0 [
m2/s]
a)
MDI
MDII
Exp.
0
1
2
3
4
Q
b)
MDI
MDII
Exp.
Figure 9. Self-diffusion prefactor D0 a) and activation energy Q b) for different
temperature intervals.
5. Conclusions
The volume diffusivity in the 1224K to 1699K temperature range, has been estimated
in the studies reported in this paper by direct molecular statics/dynamics simulations
applying the embedded atom model potential.
We can conclude that:
• The diffusivity in solid state B2-NiAl intermetallic can be successfully quantified
from direct MS/MD simulations.
• The instantaneous diffusivity stabilizes faster than the average one.
• The simulation times needed to achieve stabilised asymptotic diffusivity must be
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 16
in the order of microseconds.
• The application of the extrapolation procedure improves diffusion simulation
results.
Some findings in this paper, especially relating to the simulated by MD diffusivity
of stoichiometric solid state B2-NiAl intermetallic in the 1224K to 1699K temperature
range and the study of required simulation time ensuring the stabilization of diffusivity
results, are the first to be reported and are hopeful that will be confirmed by further
research studies. The estimated values will be used in the discrete element modelling of
a sintering process of NiAl powder [3]. The methodology developed in this study will
be included in future multiscale sintering modelling.
ACKNOWLEDGMENTS
This work was supported by the National Science Centre (NCN – Poland) Research
Project: DEC-2013/11/B/ST8/03287. Additional assistance was granted through the
computing cluster GRAFEN at Biocentrum Ochota and the Interdisciplinary Centre for
Mathematical and Computational Modelling of Warsaw University (ICM UW).
References
[1] A. Cocks, Constitutive modelling of powder compaction and sintering, Prog Mater Sci 46 (2001)
201–229. doi:10.1016/S0079-6425(00)00017-7.
[2] S. Nosewicz, J. Rojek, S. Mackiewicz, M. Chmielewski, K. Pietrzak, B. Romelczyk, The influence
of hot pressing conditions on mechanical properties of nickel aluminide/alumina composite, J.
Compos. Mater. 48 (29) (2014) 3577–3589. doi:10.1177/0021998313511652.
[3] S. Nosewicz, J. Rojek, M. Chmielewski, K. Pietrzak, Discrete element modeling and experimental
investigation of hot pressing of intermetallic NiAl powder, Adv. Powder Technol. 28 (7) (2017)
1745 – 1759. doi:10.1016/j.apt.2017.04.012.
[4] B. J. Henz, T. Hawa, M. Zachariah, Molecular dynamics simulation of the kinetic sintering of Ni
and Al nanoparticles, Mol. Simul. 35 (2009) 804–811. doi:10.1080/08927020902818021.
[5] J. Rojek, S. Nosewicz, M. Mazdziarz, P. Kowalczyk, K. Wawrzyk, D. Lumelskyj, Modeling of a
Sintering Process at Various Scales, Procedia Eng. 177 (2017) 263–270, {XXI} Polish-Slovak
Scientific Conference Machine Modeling and Simulations {MMS} 2016.September 6-8, 2016,
Hucisko, Poland. doi:10.1016/j.proeng.2017.02.210.
[6] G.-X. Chen, J.-M. Zhang, K.-W. Xu, Self-diffusion of Ni in B2 type intermetallic compound NiAl
, J. Alloys Compd. 430 (2007) 102 – 106. doi:10.1016/j.jallcom.2006.04.052.
[7] Y. Mishin, D. Farkas, Atomistic simulation of point defects and diffusion in B2 NiAl, Philos. Mag.
A 75 (1997) 169–185. doi:10.1080/01418619708210289.
[8] Y. Mishin, D. Farkas, Atomistic simulation of point defects and diffusion in B2 NiAl, Philos. Mag.
A 75 (1997) 187–199. doi:10.1080/01418619708210290.
[9] Y. Mishin, D. Farkas, Atomistic simulation of point defects and diffusion in B2 NiAl , Scr. Mater.
39 (1998) 625 – 630. doi:10.1016/S1359-6462(98)00206-1.
[10] Y. Mishin, A. Y. Lozovoi, A. Alavi, Evaluation of diffusion mechanisms in NiAl
by embedded-atom and first-principles calculations, Phys. Rev. B 67 (2003) 014201.
doi:10.1103/PhysRevB.67.014201.
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 17
[11] B. Meyer, M. Fahnle, Atomic defects in the ordered compound B2-NiAl: A combination
of ab initio electron theory and statistical mechanics, Phys. Rev. B 59 (1999) 6072–6082.
doi:10.1103/PhysRevB.59.6072.
[12] C. Jiang, L.-Q. Chen, Z.-K. Liu, First-principles study of constitutional point defects in
B2 NiAl using special quasirandom structures , Acta Mater. 53 (2005) 2643 – 2652.
doi:10.1016/j.actamat.2005.02.026.
[13] K. Marino, E. Carter, The effect of platinum on defect formation energies in β-NiAl , Acta Mater.
56 (2008) 3502 – 3510. doi:10.1016/j.actamat.2008.03.029.
[14] S. Divinski, C. Herzig, On the six-jump cycle mechanism of self-diffusion in NiAl, Intermetallics 8
(2000) 1357–1368. doi:10.1016/S0966-9795(00)00062-5.
[15] K. A. Marino, E. A. Carter, First-principles characterization of Ni diffusion kinetics in β-NiAl,
Phys. Rev. B 78 (2008) 184105. doi:10.1103/PhysRevB.78.184105.
[16] H. Wang, X. Gao, H. Ren, S. Chen, Z. Yao, Diffusion coefficients of rare earth elements in fcc
Fe: A first-principles study, J. Phys. Chem. Solids 112 (Supplement C) (2018) 153 – 157.
doi:10.1016/j.jpcs.2017.09.025.
[17] K. Lejaeghere, J. Jaeken, V. Van Speybroeck, S. Cottenier, Ab initio based thermal
property predictions at a low cost: An error analysis, Phys. Rev. B 89 (2014) 014304.
doi:10.1103/PhysRevB.89.014304.
[18] M. Forsblom, N. Sandberg, G. Grimvall, Anharmonic effects in the heat capacity of Al, Phys. Rev.
B 69 (2004) 165106. doi:10.1103/PhysRevB.69.165106.
[19] B. Grabowski, L. Ismer, T. Hickel, J. Neugebauer, Ab initio up to the melting
point: Anharmonicity and vacancies in aluminum, Phys. Rev. B 79 (2009) 134106.
doi:10.1103/PhysRevB.79.134106.
[20] A. Y. Lozovoi, Y. Mishin, Point defects in NiAl: The effect of lattice vibrations, Phys. Rev. B 68
(2003) 184113. doi:10.1103/PhysRevB.68.184113.
[21] S. Yip, Handbook of Materials Modeling, Handbook of Materials Modeling, Springer Netherlands,
2007. doi:10.1007/978-1-4020-3286-8.
[22] C. Weinberger, G. Tucker, Multiscale Materials Modeling for Nanomechanics, Springer Series in
Materials Science, Springer International Publishing, 2016. doi:10.1007/978-3-319-33480-6.
[23] D. R. Alfonso, D. N. Tafen, Simulation of Atomic Diffusion in the Fcc NiAl Sys-
tem: A Kinetic Monte Carlo Study, J. Phys. Chem. C 119 (2015) 11809–11817.
doi:10.1021/acs.jpcc.5b00733.
[24] A. V. Evteev, E. V. Levchenko, I. V. Belova, G. E. Murch, Molecular dynamics simulation of
surface segregation, diffusion and reaction phenomena in equiatomic Ni-Al systems, Phys. Met.
Metall. 113 (13) (2012) 1202–1243. doi:10.1134/S0031918X12130017.
[25] P. Kuhn, J. Horbach, F. Kargl, A. Meyer, T. Voigtmann, Diffusion and interdiffusion in binary
metallic melts, Phys. Rev. B 90 (2014) 024309. doi:10.1103/PhysRevB.90.024309.
[26] A. Kerrache, J. Horbach, K. Binder, Molecular-dynamics computer simulation of crys-
tal growth and melting in Al50Ni50, EPL (Europhys. Lett.) 81 (2008) 58001.
doi:10.1209/0295-5075/81/58001.
[27] E. V. Levchenko, A. V. Evteev, I. V. Belova, G. E. Murch, Molecular dynamics study of density,
surface energy and self-diffusion in a liquid Ni50Al50 alloy, Comput. Mater. Sci 50 (2010) 331 –
337. doi:10.1016/j.commatsci.2010.08.022.
[28] E. V. Levchenko, A. V. Evteev, T. Ahmed, A. Kromik, R. Kozubski, I. V. Belova, Z.-K. Liu, G. E.
Murch, Influence of the interatomic potential on thermotransport in binary liquid alloys: case
study on NiAl, Philos. Mag. 96 (2016) 3054–3074. doi:10.1080/14786435.2016.1223893.
[29] A. V. Evteev, E. V. Levchenko, I. V. Belova, G. E. Murch, Molecular dynamics
simulation of diffusion in a (110) B2-NiAl film , Intermetallics 19 (2011) 848 – 854.
doi:10.1016/j.intermet.2011.01.010.
[30] J. Duan, Atomistic simulations of diffusion mechanisms in stoichiometric Ni3Al, J. Phys.: Condens.
Matter 18 (2006) 1381. doi:10.1088/0953-8984/18/4/022.
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 18
[31] V. Turlo, F. Baras, O. Politano, Comparative study of embedded-atom methods applied to
the reactivity in the NiAl system, Modell. Simul. Mater. Sci. Eng. 25 (6) (2017) 064002.
doi:10.1088/1361-651X/aa6cfa.
[32] M. Mazdziarz, J. Rojek, S. Nosewicz, ESTIMATION OF MICROMECHANICAL NiAl
SINTERING MODEL PARAMETERS FROM THE MOLECULAR SIMULATIONS, Int. J.
Multiscale Comput. Eng. 15 (4) (2017) 343–358. doi:10.1615/IntJMultCompEng.2017020289.
[33] H. Mehrer, Diffusion in Solids: Fundamentals, Methods, Materials, Diffusion-Controlled
Processes, Springer Series in Solid-State Sciences, Springer Berlin Heidelberg, 2007.
doi:10.1007/978-3-540-71488-0.
[34] D. Gupta, Diffusion Processes in Advanced Technological Materials, Springer Berlin Heidelberg,
2010. doi:10.1007/978-3-540-27470-4.
[35] C. Herzig, S. Divinski, Essentials in diffusion behavior of nickel- and titanium-aluminides,
Intermetallics 12 (2004) 993 – 1003. doi:10.1016/j.intermet.2004.03.005.
[36] V. P. Ramunni, Diffusion behavior in Nickel-Aluminum and Aluminum-Uranium diluted alloys ,
Comput. Mater. Sci 93 (2014) 112 – 124. doi:10.1016/j.commatsci.2014.06.039.
[37] Q. Xu, A. Van der Ven, Atomic transport in ordered compounds mediated by local disorder: Dif-
fusion in B2−NixAl1−x, Phys. Rev. B 81 (2010) 064303. doi:10.1103/PhysRevB.81.064303.
[38] P. A. Korzhavyi, A. V. Ruban, A. Y. Lozovoi, Y. K. Vekilov, I. A. Abrikosov, B. Johansson,
Constitutional and thermal point defects in B2 NiAl, Phys. Rev. B 61 (2000) 6003–6018.
doi:10.1103/PhysRevB.61.6003.
[39] M. I. Mendelev, Y. Mishin, Molecular dynamics study of self-diffusion in bcc Fe, Phys. Rev. B 80
(2009) 144111. doi:10.1103/PhysRevB.80.144111.
[40] S. Frank, S. Divinski, U. Sodervall, C. Herzig, Ni tracer diffusion in the B2-compound
NiAl: influence of temperature and composition, Acta Mater. 49 (2001) 1399 – 1411.
doi:10.1016/S1359-6454(01)00037-4.
[41] C. Campbell, A. Rukhin, Evaluation of self-diffusion data using weighted means statistics, Acta
Mater. 59 (2011) 5194 – 5201. doi:10.1016/j.actamat.2011.04.055.
[42] S. Plimpton, Fast Parallel Algorithms for Short-Range Molecular Dynamics, J. Comput. Phys.
117 (1995) 1 – 19. doi:10.1006/jcph.1995.1039.
[43] A. Stukowski, Visualization and analysis of atomistic simulation data with OVITO - the Open
Visualization Tool (http://ovito.org/), Modelling Simul. Mater. Sci. Eng. 18 (2010) 015012.
doi:10.1088/0965-0393/18/1/015012.
[44] W. R. Inc., Mathematica, Version 11.0, champaign, IL, 2017.
[45] Y. Mishin, M. J. Mehl, D. A. Papaconstantopoulos, Embedded-atom potential for B2 − NiAl,
Phys. Rev. B 65 (2002) 224114. doi:10.1103/PhysRevB.65.224114.
[46] G. P. Pun, Y. Mishin, Development of an interatomic potential for the Ni-Al system, Philos. Mag.
89 (2009) 3245–3267. doi:10.1080/14786430903258184.
[47] E. B. Tadmor, R. E. Miller, Modeling Materials: Continuum, Atomistic and Multiscale Techniques,
Cambridge University Press, 2011.
[48] M. Mazdziarz, T. D. Young, P. D luzewski, T. Wejrzanowski, K. J. Kurzydlowski, Computer
modelling of nanoindentation in the limits of a coupled molecular–statics and elastic scheme, J.
Comput. Theor. Nanosci. 7 (2010) 1172. doi:10.1166/jctn.2010.1469.
[49] M. Mazdziarz, T. D. Young, G. Jurczak, A study of the affect of prerelaxation
on the nanoindentation process of crystalline copper, Arch. Mech. 63 (2011) 533.
doi:http://www.ippt.gov.pl/~tyoung/2011.myj.pdf.
[50] J. Cholewinski, M. Mazdziarz, G. Jurczak, P. D luzewski, DISLOCATION CORE RECON-
STRUCTION BASED ON FINITE DEFORMATION APPROACH AND ITS APPLI-
CATION TO 4H-SiC CRYSTAL, Int. J. Multiscale Comput. Eng. 12 (2014) 411–421.
doi:10.1615/IntJMultCompEng.2014010679.
[51] M. Mazdziarz, M. Gajewski, Estimation of Isotropic Hyperelasticity Constitutive Models to Ap-
proximate the Atomistic Simulation Data for Aluminium and Tungsten Monocrystals, Computer
A Molecular Dynamics Study of Self-Diffusion in Stoichiometric B2-NiAl crystals 19
Modeling in Engineering & Sciences 105 (2) (2015) 123–150. doi:10.3970/cmes.2015.105.123.
[52] M. I. Mendelev, B. S. Bokstein, Molecular dynamics study of self-diffusion in Zr, Philos. Mag. 90
(2010) 637–654. doi:10.1080/14786430903219020.
[53] K. Tapasa, A. Barashev, D. Bacon, Y. Osetsky, Computer simulation of carbon dif-
fusion and vacancy-carbon interaction in α-iron, Acta Mater. 55 (1) (2007) 1 – 11.
doi:10.1016/j.actamat.2006.05.029.
[54] X. Zhou, R. Jones, J. Gruber, Molecular dynamics simulations of substitutional diffusion, Comput.
Mater. Sci 128 (2017) 331 – 336. doi:10.1016/j.commatsci.2016.11.047.