Computational Materials Science 150 (2018) 1–8

Contents lists available at ScienceDirect

Computational Materials Science

journal homepage: www.elsevier.com/locate/commatsci

Embedded atom method potentials for Ce-Ni binary alloy T

a a b,⁎ a,⁎
Yawei Lei , Xiaorui Sun , Rulong Zhou , Bo Zhang
a
Institute of Amorphous Matter Science & School of Materials Science and Engineering, Hefei University of Technology, Hefei, Anhui 230009, PR China
b
School of Materials Science and Engineering, Hefei University of Technology, Hefei, Anhui 230009, PR China

A R T I C LE I N FO A B S T R A C T

Keywords: A new embedded-atom method (EAM) potential of Ce-Ni has been constructed based on the experimental and
Embedded-atom-method (EAM) potential first-principles data with a variety of physical properties. The potential accurately reproduces the equilibrium
CeNi alloy lattice constant, cohesive energy, vacancy and interstitial formation energy, elastic constants, phonon dispersion
Metallic glass and radial distribution functions of liquid amorphous structures of the alloys. With this potential, we in-
vestigated the glass forming ability and dynamic behavior of a deep-eutectic liquid Ce80Ni20. The simulated
results of Ce80Ni20 metallic glass shows good agreement with related experimental data eg. the glass transition
temperature and self-diffusion constant of Ni.

1. Introduction
Cerium and Ce-based compounds have drawn a lot of attention
because of their fascinating properties. Due to the easier transition
between the localized and itinerant states of the 4f electron, Ce exhibits
a remarkable isostructural (fcc to fcc) phase transition from the ferro-
magnetic γ-Ce phase to the paramagnetic α-Ce phase under compres-
sion [1].
Cerium-based compounds also have unique characters due to the 4f
electron. For example, pressure-induced superconductivity has been
reported in CeIn3, CePd2Si2 and CeCu2Ga2 [2]; CeAl3 and CeCu6 are
heavy-fermion Kondo lattice compounds without any long-range mag-
netic order [3,4], and the cubic CeB6 and CeAl2 are Kondo lattice
compounds that order magnetically at low temperature [5,6]. CeNi
appears a pressure-induced first-order volume-collapse phase transi-
tion, because of the fluctuation of the charge states of the Ce atoms
between Ce3+ and Ce4+. So CeNi is a typical model to study the me-
chanisms of phase transition caused by the f-electron instability and
strong electron correlations [7].
Ce-based bulk metallic glasses (BMGs) also possess fascinating
properties such as small elastic moduli, exceptionally low glass-transi-
tion temperature close to room temperature and excellent glass forming
ability (GFA), etc. [8–10]. Hence, Ce-based BMGs were developed to
study glass transition, plastic molding, slow dynamics [11] and glass-
forming mechanism [12] in the glassy alloys, especially for Ce80Ni20
which is a deep eutectic system. However, although Ce80Ni20 com-
pounds are expected to possess good glass-forming ability, Meyer et al.
have shown that the synthesis of glassy Ce80Ni20 alloy by splat
quenching and melt spinning is unsuccessful [13]. To know the reason
of this discrepancy between the fact and the expectation, it is needed to
study the microstructural changes during the quenching and melt
process.
Limited by the computer capacity, ab initio molecular dynamics si-
mulation is incapable of simulating the kinetic processes involving lager
model size and long simulation time. However classical molecular dy-
namics is possible to simulate structures with millions of atoms in
reasonable time with classical effective potential. Therefore obtaining
an effective potential for classical molecular dynamics is essential. In
this paper we use force matching method proposed by Ercolessi and
Adams [14] to develop an analytical embedded-atom method potential
for Ce-Ni alloy. The potential is fitted to the calculated forces, energies
and stresses of various structures from experiment data and ab initio MD
simulations [15,16]. The open source program potfit [17] has developed
as a flexible implementation of the force matching method.
To test the accuracy of our EAM potential, we compared the results
calculated with the fitted EAM potential with those obtained from the
density-functional theory (DFT) [18] calculations and experimental
data. High consistence between the results of EAM potential and those
from DFT calculations or experiments enabled us to use this potential to
study the glass formation process and dynamic behavior of Ce80Ni20
melt.

⁎
Corresponding authors.
E-mail address: rlzhou@hfut.edu.cn (R. Zhou).

https://doi.org/10.1016/j.commatsci.2018.03.060
Received 3 January 2018; Received in revised form 22 March 2018; Accepted 24 March 2018
0927-0256/ © 2018 Elsevier B.V. All rights reserved.
Y. Lei et al. Computational Materials Science 150 (2018) 1–8

2. Methodology defined through a least-squares function formed from the differences of

the EAM and DFT (and experiment) values:
2.1. EAM potential
Z = ZF + ZC (7)
The total energy of an atomic system within Embedded-atom
ZF is related to the difference between the EAM and ab initio forces
method [19,20] is represented as the following formula:
calculated for Nα atoms listed in the fitting database:
1
Etot =
2
∑ ϕ (rij ) + ∑ F (ni ) Nα
i,j i (1) (FiαEAM −FiαDFT )2
ZF = ∑ ∑ Wi
i = 1 α = x ,y,z
(FiαDFT )2 + εi (8)
and ni = ∑ ρ (rij )
j (2) Fiα is the αth component of the force on atom i , Wi is the weight asso-
where ϕ (rij ) is a pair-wise potential function depending only on the ciated with each force, and εi is a small value that helping us to avoid
distance (rij ) of atom i and atom j . F (ni ) is the embedding function that extremely small denominators.
represents the energy required to place an atom into the electron cloud ZC gives the fitting errors for energy and stress values:
of density ni , which is the sum of the contributions ρ (rij ) from all the
NC
neighboring atoms. (AiEAM −AiDFT )2
ZC = ∑ Wi
For the pair-wise potential function, the Morse potential was i=1
(AiDFT )2 + εi (9)
adopted, which can be written as:
where NC is the number of energies or stress tensor components for
r −rc
ϕ (r ) = ψ ⎛
⎝ h ⎠
(
⎞ De { 1−e−α (r − re) )2−1} (3)
fitting, and Ai is the value of energy or stress.

where the three parameters De , α and re describe the depth of the po-
tential minimum, the width of the potential minimum and the equili-
brium distance, respectively.
And, the function ψ is a cutoff function, which is defined as
4
⎧ x 4, x < 0
ψ (x ) = 1+x
⎨ 0, x⩾0
⎩ (4)
The parameters r and h in formula (3) describe the cutoff radius and
the smoothing of the potential, respectively. The cutoff function guar-
antees that the potential functions, as well as their derivatives up to the
second order, approach zero smoothly at the cutoff distance rc .
As for the transfer function ρ (r ) , we used an oscillating transfer
function [21],
r −rc 1 + a1cos(κr ) + a2sin(κr )
ρ (r ) = ψ ⎛ ⎞ phases in the database will have little influence on the accuracy of this
⎝ h ⎠ rβ (5)
where the parameters a1 and a2 determine the amplitude of the oscil-
lations, and κ is the wave vector and β controls the decay.
For the embedding function F (n), we chose the Johnson function
[22] which is given by
F (n) = F0 [1−qlnn] nq + F1 n
(6)
F0 , F1 and q are constants that we have to fit. n = ρi / ρ0 . ρ0 is the equi-
librium density. In this paper we use ρ0 = 1.
For Ce-Ni binary system, the number of free parameters of our po-
tential model is 23 with 7 potential functions (pair-wise functions for
Ce-Ce, Ce-Ni and Ni-Ni, transfer functions for Ce and Ni, and embed-
ding functions for Ce and Ni), and every pair and transfer function has
one additional parameter h for the cutoff function ψ . The cutoff radius rc
is kept fixed at 7 Å.
2.2. Computational method
We used the Vienna Ab-initio Simulation Package (VASP) [23,24] to
perform force and energy calculations using projector augmented wave
(PAW) [25] potentials within the generalized gradient approximation
(GGA) [26] with high precision. All the simulation supercells contain
96–128 atoms. A plane-wave cutoff energy of 400 eV was used along
with 2 × 2 × 2 Monkhorst–Pack k-point meshes[15]. Classical MD cal-
culations were performed using the Large-scale Atomic/Molecular
Massively Parallel Simulator (LAMMPS) code [27]. All force-matching
was performed using the POTFIT package, which has previously been
used to optimize tabulated pair and EAM potentials. The fitting error
3. Results and discussion
3.1. Error
We select 116 atomic configurations including 76 low-energy
crystalline structures of pure Ni and various Ce-Ni compounds and 40
highly activated configurations (e.g. 14 structures with point defects,
24 structures with liquid phases, 2 structures with metallic glasses
quenched at different cooling rates). The pure Ce crystalline structures
are not included in our fitting database because there are two iso-
structure phases which are hard to be correctly described simulta-
neously by using density functional method. So, fitting including pure
Ce phase can lead to large errors. However, this potential is mainly
designed for simulating Ce-Ni binary systems, so the absence of pure Ce
potential for Ce-Ni binary systems.
The root mean square (RMS) errors for forces, energies, and stresses
are listed in Table 1. As seen, the RMS errors of the forces, energies and
stresses are all smaller than the values reported in reference [28], which
indicates that our fitting reaches a good accuracy. A graphical re-
presentation of the difference of energies, stresses and forces calculated
by EAM potential and ab initio are shown in Fig. 1. As seen clearly, the
maximum error between the EAM energy and ab initio energy is less
than 25 meV/atom, and the average error is only 5.1 meV/atom. The
average errors of stress and force are 90 bar and 168 meV/Å per con-
figuration, respectively. All these results indicate that the results of the
fitted EAM potential are in good consistence with the ab initio calcu-
lations. The profiles of the functions of our EAM potential are shown in
Fig. 2.
We point out that although the RMS errors and average errors are
very small, it is not enough to judge the quality and transferability of
the fitted potential. Further comparison of various properties calculated
using the fitted potential and the ab initio method is needed.
Table 1
Root mean square errors after the optimization for
forces (in meV/Å), energies (in meV/atom), and stresses
(in kPa).
RMS errors for
Forces (meV/Å) 196.48
Energies (meV/atom) 6.99
Stresses (kPa) 39.00

2
Y. Lei et al. Computational Materials Science 150 (2018) 1–8

one can get the values of the equilibrium cohesive energy E0 , equili-
brium volume V0 , bulk modulus B0 and pressure derivative of the bulk
modulus B0′.
Fig. 3 shows the EOS of the six structures as a function of volume. It
can be seen that the coincidence of the energy curve is excellent, which
means the general agreement between EAM and ab initio calculations.
For each structure, in Table 2, we list the equilibrium cohesive energy,
equilibrium volume and bulk modulus obtained by fitting the three-
order Bich-Murnaghan function. The differences of the equilibrium
volume, cohesive energy and bulk modulus between the EAM and ab
initio calculations are all very small. For alloys, the maximum difference
of the equilibrium volume, cohesive energy and bulk modulus among
all these crystalline phases are −1.27 Å3/cell (−5.9%) for Ce1Ni1, −6
meV/atom (0.1%) for Ce1Ni5 and −16.89 GPa (−14.7%) for Ce1Ni2,
respectively. Such small differences of these values from EAM and ab
initio calculations indicate again that the fitted EAM is of high accuracy.
Also, the volumes are in good agreement with the experimental values.
Fig. 1. Comparison of ab initio and EAM calculations of the average error per So our EAM potential can describe the properties of perfect crystalline
configuration. phase of Ce-Ni intermetallic compound well. The values of pure Ce
phase are also listed in Table 2. Because the pure Ce phases are not
included in our fitting database, the errors of the cohesive energy and
3.2. Equations of state
bulk modulus of pure Ce are a little large. The equilibrium lattice
constant we calculated is 4.625 Å which is closed to the small-volume
We first compare the equations of state (EOS) of several crystalline
Ce while not the large-volume Ce as the potential proposed by Sheng
phases including element crystal Ni and intermetallic compounds of
et al. [30].
Ce1Ni1 (Space group is Cmcm), Ce1Ni2 (Fd3-m), Ce1Ni3 (P63/mmc),
Ce1Ni5 (P6/mmm), Ce2Ni3 (P2-m). The change of the volumes
within ± 8.0% of the equilibrium volume are considered in the calcu- 3.3. Phonon dispersion
lation of the EOS and for each crystal structure at least 9 different vo-
lumes are concerned. Using a least squares fit of the set (V ,E ) to the By using the EAM potential combined with the Phonopy package
Bich-Murnaghan formula [29]: [37], the phonon dispersion curves of Ce1Ni1 are computed, which is as
shown in Fig. 4. The primitive cell of Ce1Ni1 alloy contains eight atoms,
meaning that the dispersion should have 24 lines. As clearly shown in
2 3 2 2 2
9V0 B0 ⎧ ⎡ V0 3 ⎤ ⎡ ⎛ V0 ⎞ 3 ⎤ ⎡ V0 3 ⎤ ⎫ Fig. 4, the calculated phonon dispersion curves do not contain any
E (V ) = E0 + ⎛ ⎞ −1⎥ B0′ + ⎢ V −1⎥ ⎢6−4 ⎛ V ⎞ ⎥ ⎬ imaginary frequency indicating its stability. As seen, the phonon dis-
16 ⎨ ⎢ ⎝ V ⎠ ⎝ ⎠ ⎝ ⎠ ⎦
⎩⎣ ⎦ ⎣ ⎦ ⎣ ⎭ persion curves calculated using our EAM potential are in good agree-
(10) ment with those obtained from ab initio calculations.

Fig. 2. The functions of the EAM potential for the Ce-Ni system. (a) Pair potential functions. (b) Charge density functions. (c) Embedding functions.

3
Y. Lei et al. Computational Materials Science 150 (2018) 1–8

Fig. 3. Energy varying with the change of the volume of Ce-Ni compounds and fcc Ni crystalline structure.

Table 2 For cubic system Ce1Ni2 the errors are a little large, about 40 GPa of
Comparison of ground state properties of Ce-Ni compounds and fcc Ce and Ni. C11, C22 and C33. We know that there are usually big errors between the
Volume (Å3/cell) Cohesive energy (eV/ Bulk modulus (GPa)
calculated elastic constants and the measured results. Even for the first
atom) principles calculations, the errors may be as large as 20 GPa. So, the
accuracy of our EAM potential for the elastic constant is acceptable.
Exp. or ab EAM ab initio EAM ab initio EAM
initio
3.5. Point defect
Ce 26.39 24.73 −5.932 −5.829 40.51 16.03
25.58exp [31] Usually there are large amount of defects in the metallic com-
Ni 10.96 10.91 −5.570 −5.571 192.24 195.31 pounds, so the fitted EAM potential should also be able to describe the
10.95exp [32]
formation of defects well. To this end, we calculated the formation
Ce1Ni1 20.49 20.27 −6.053 −6.049 77.28 65.77
21.54exp [33] energy of Ce and Ni vacancies and interstitials in various Ce-Ni crys-
Ce1Ni2 15.53 15.69 −6.071 −6.069 115.39 98.50 talline phases using the fitted potential and compared the results with
15.60exp [34] those from DFT calculations, according to formula (11) and (12):
Ce1Ni3 14.67 14.64 −6.057 −6.053 132.05 120.92
14.67exp [35] EVac
f = Edefect−Eperfect + μCe (or μNi ) (11)
Ce1Ni5 13.85 13.74 −5.982 −5.976 147.68 148.16
13.72exp [36] EfInt = Edefect−Eperfect −μCe (or μNi ) (12)
Ce2Ni3 18.24 18.34 −6.067 −6.067 93.99 82.50
where Edefect is the total energy of the defective structure with a Ce (or
Ni) vacancy or interstitial; Eperfect is the total energy of the perfect
3.4. Elastic constants crystalline structure with the same sized supercell; μCe and μN i are the
chemical potentials of Ce and Ni obtained from Ce and Ni crystal, re-
We further calculated the elastic constants of several crystalline spectively. All the structures are fully relaxed by using VASP and
phases using the fitted potential and compared them with the DFT re- LAMMPS package respectively to obtain the total energies.
sults. The six diagonal independent elastic constants C11 to C66 of The formation energies of the defective structures are listed in
Ce1Ni1, Ce1Ni2, Ce1Ni3 and Ce1Ni5 phases which are listed in Table 3, Table 4. As seen, although these structures are all highly activated
where DFT values are in brackets. For Ce1Ni3 and Ce1Ni5, which are structures, both ab initio and EAM potential calculated values anasto-
both hexagonal crystal systems, the EAM potential results have good mosed well. For the vacancies, except for the Ce vacancy in Ce1Ni1
agreement with that of ab initio, where the maximal error is about 20 whose formation energy is particularly large, the errors between the ab
GPa. Coincidence degree of orthorhombic crystal Ce1Ni1 is also good. initio and EAM potential results are small. The calculated formation

Fig. 4. Comparison of phonon dispersion curves of Ce1Ni1 alloy calculated by using the present EAM potential with those obtained from ab initio calculation.

4
Y. Lei et al. Computational Materials Science 150 (2018) 1–8

Table 3
Calculated elastic constant Cii (i = 1, 2, 3, 4, 5, 6) of all the Ce-Ni compounds concerned. The values in the parentheses are DFT results.
Ce1Ni1 Error Ce1Ni2 Error Ce1Ni3 Error Ce1Ni5 Error

C11 82(94) −13% 139(180) −23% 186(169) 10% 203(221) −8%

C22 76(118) −36% 139(180) −23% 186(169) 10% 203(221) −8%
C33 83(102) −19% 144(184) −22% 181(184) −2% 276(269) −3%
C44 25(18) 39% 22(40) −45% 32(43) −25% 50(69) −27%
C55 16(30) −47% 22(40) −45% 32(43) −25% 55(69) −20%
C66 10(8) 25% 27(45) −40% 41(50) −18% 49(56) −12%

Table 4 energies using EAM potential is usually smaller than that from ab initio
Formation energies of single vacancy and interstitial in different Ce-Ni com- calculation. The formation energies of Ce defects are usually higher
pounds. than those of Ni defects indicating that it is hard for the formation of Ce
Vacancy Relaxed Vacancy formation energy (eV) Error (eV) defects. For the interstitials, the errors between the calculated forma-
tion energies from EAM potential and ab initio calculations for some
Ab initio EAM cases are a little large. Overall, our EAM potential can give good de-
scription for these defective structures.
Ce VCe 1.255 1.360 0.105
Ni VNi 1.541 1.313 −0.228

Ce1Ni1 VCe 2.675 1.064 −1.611

3.6. Classical MD
VNi 1.534 1.032 −0.502

Ce1Ni2 VCe 0.501 0.228 −0.273

Not only the ordered crystalline structures but also the disordered
VNi 1.929 1.086 −0.843 structures of liquid and glass states are also incorporated into the po-
tential fitting. We first get a good EAM potential by fitting only the
Ce1Ni3 VCe1 2.207 2.006 −0.201
VCe2 1.634 1.095 −0.539 crystalline structures. Then we use this potential to run MD simulations
VNi1 1.468 1.085 −0.383 on the systems containing 80 Ce atoms and 20 Ni atoms. The original
VNi2 1.504 0.857 −0.647 structure of Ce80Ni20 was generated randomly and then equilibrated at
VNi3 1.517 0.872 −0.645 1500 K for 150 ns. Then we quenched it to 300 K at the rate of 1010 K/s
VNi4 1.645 0.924 −0.721
to obtain the glass structures. We collected one configuration every
Interstitial Relaxed Interstitial formation energy (eV) Error (eV) 200 K (totally 11 configurations) from the MD trajectory and in-
corporated them into the datasets to fit a new EAM potential. The
Ab initio EAM process is repeated until the energy error is less than 1% and force error
is less than 10%.
Ce1Ni1 IntCe 1.455 1.108 −0.347
IntNi 0.445 −0.141 −0.586 Then we used the new EAM potential to run a large-scale MD si-
mulation on the Ce80Ni20 system which contains 8000 Ce atoms and
Ce1Ni2 IntCe 3.320 2.029 −1.291
IntNi 1.569 0.721 −0.848 2000 Ni atoms in the supercell. The system was equilibrated at 2500 K
for about 20 ns. The structures in the trajectory of the last 9 ps were
Ce1Ni3 IntCe 3.064 0.764 −2.300
IntNi 2.493 0.205 −2.288
collected for structural analysis. The partial radial distribution func-
tions (RDFs) of Ce80Ni20 are shown in Fig. 5. For comparison, ab initio
MD simulation on a small cell which containing 80 Ce and 20 Ni atoms
was also performed at 2500 K for about 10 ps. The partial pair dis-
tribution functions of Ce80Ni20 from the ab initio MD simulation are also

Fig. 5. Radial distribution functions calculated for liquid Ce80Ni20 at 2500 K using EAM potential (red dotted lines) and ab initio MD (black solid lines).

5
Y. Lei et al.
shown in Fig. 5 for comparing.
As seen in Fig. 5, the total radial distribution of equilibrium liquid
Ce80Ni20 simulated using the fitted EAM potential (red dotted lines) is
in excellent agreement with that from the ab initio MD simulations
(black solid line) at 2500 K. The difference between the intensity and
the position of the nearest-neighbor distance in the total RDF of the
classic MD and ab initio MD simulation is scarcely less. For the partial
RDFs, the coincidence of Ce-Ce and Ce-Ni partial RDFs is also very
good. The first peak of the Ce-Ce partial PDFs from ab initio simulation
is a little wider than that from the classic MD simulation, which is due
to the fact that Ce atoms in trivalent and tetravalent states usually co-
exist in the Ce-based materials under low pressures while the EAM
potential cannot reflect these two different states of Ce simultaneously.
For the Ni-Ni partial PDFs, the position of first peak is identical but the
height has big error. The reason may be that there are too few Ni atoms
in the simulating cell in the ab initio MD simulation which increases the
uncertainty of the appearance in different positions of Ni atoms.
3.7. Glass forming ability
Using our innovative potential, the glass formation of Ce80Ni20 alloy
was investigated. To obtain the original structure, we constructed a
supercell containing 8000 Ce atoms and 2000 Ni atoms which are
distributed randomly in an orthogonal box. Then the system was heated
up from 300 K to 1500 K at the speed of 1010 K/s and equilibrated at
1500 K for 20 ns. Then the system was quenched to the room tem-
perature at different cooling rate ranging from 109 K/s to 5 × 1012 K/s.
The NPT ensemble (i.e. constant number, pressure and temperature)
was applied in all the simulations. The potential energies as a function
of temperature are shown in Fig. 6. The atomic structures of the final
configurations at 300 K were analyzed using Voronoi analysis method
to examine the crystallinity of the samples, which can be seen in Fig. 6.
In the Voronoi analysis method, the space is partitioned into polyhedra
around atoms by constructing bisecting planes along the lines joining
central atoms and all their neighbors. Voronoi index 〈n3, n4, n5, n6, …〉
is used to designate and differentiate the type of the polyhedron, where
ni represents the number of i-edged faces of the Voronoi polyhedron.
As well known, the melt will transform into crystalline structure
when it is cooled down at a very slow cooling rate, otherwise it will
transform into amorphous phase at a quick cooling rate. As shown in
Fig. 6, when the cooling rate is below 1010 K/s, an obvious ladder ap-
pears at about 400 K in each potential energy curve, indicating the
crystal formation. With the increasing of cooling rate, an inflection
point replaces the ladder. This inflection point is corresponding to the
glass transition temperature Tg. From Fig. 6, we get Tg of Ce80Ni20
Fig. 6. The potential energy as a function of temperature of Ce80Ni20 under
different cooling rate.
6
Computational Materials Science 150 (2018) 1–8
which is 352 K at the cooling rate of 1011 K/s. Experimentally, Zhang
et al. have developed a family of Ce-based BMGs which show excellent
glass forming ability and extremely low glass transition temperature
(∼60 °C) [9], such as quaternary Ce–Al–Cu–X alloys which has an ex-
ceptionally low Tg (about 341–410 K) [10]. As seen, the Tg of Ce80Ni20
obtained from our large scale MD simulation is within the range of the
Tg of typical Ce-based BMGs, which indicates that our CeNi EAM po-
tential has good transferability for the application in the Ce-Ni amor-
phous phases.
Fig. 7 shows the contents of icosahedral and icosahedral-like clus-
ters at 300 K cooled down at different rates. When the cooling rate is
very slow, the contents of icosahedral and icosahedral-like clusters are
very small. On the other hand, the content of icosahedral and icosa-
hedral-like clusters in the structures obtained from a quick cooling is
two or three times higher than those in the structures obtained from a
slow cooling. However, even at the highest cooling rate of 1011 K/s
concerned, the content of icosahedral clusters 〈0, 0, 12, 0〉 in the final
structure is only 0.2%, and those of the icosahedral-like clusters 〈0, 2,
8, 1〉, 〈0, 2, 8, 2〉 and 〈0, 1, 10, 2〉 are only 0.85%, 0.4% and 1.45%
respectively. Such low concentration of icosahedral-like clusters in the
amorphous Ce80Ni20 phase may be the reason that the Ce80Ni20 MGs is
hard to be prepared by splat quenching and melt spinning.
3.8. Liquid dynamics
Atomic diffusion in the low-temperature melt is critical for the glass
formation of materials. The atomic diffusion coefficient can be obtained
from Einstein expression [38] which is
Nm
1
D = lim
t →∞ 6Nm t
∑ [rj (t )−rj (0)]2
j=1 (13)
N
where ∑ j =m1
[rj (t )−rj(0)]2is the mean-square displacement (MSD) of
the tagged particle. The MSD of well equilibrate solids and liquids at
different temperature (300–1800 K every 100 K) were computed and
the results are shown in Fig. 8. The calculated self-diffusion coefficients
of Ce and Ni atoms in the Ce80Ni20 solids and liquids at different
temperatures were calculated using Eq. (13), which are listed in
Table 5. As seen, DNi in the Ce80Ni20 melt varies from 7.5 × 10−10 m2/s
at 700 K to 4.23 × 10−9 m2/s at 1500 K. The experimental values of DNi
in the Ce80Ni20 measured using quasielastic neutron scattering (QENS)
method [13] are within the range of 8.0 × 10−10 m2/s at 775 K to
7.25 × 10−9 m2/s at 1525 K. Clearly, the self-diffusion coefficients
calculated from our large-scale classic MD simulation using the fitted
EAM potential is in good agreement with the experimental measure-
ments, again indicating that the fitted EAM potential can describe the
Ce-Ni binary alloy system correctly.
4. Conclusion
A binary Ce-Ni Embedded Atom Method (EAM) potential has been
constructed. It is fitted to the ab initio data and validated by the cal-
culation of the general properties of some crystalline phases and dis-
ordered liquid and amorphous phases, such as equation of states, va-
cancy and interstitial formation energy, elastic constants, phonon
dispersion, and radial distribution function of Ce80Ni20 melts.
Using our potential, we studied the glass forming ability of Ce80Ni20
at atomic scale. The predicted glass transition temperature Tg of
Ce80Ni20 is about 352 K which is within the Tg values of typical Ce-
based metallic glasses. Structural analysis indicates that the contents of
the icosahedral cluster 〈0, 0, 12, 0〉 and icosahedral-like clusters 〈0, 2,
8, 2〉, 〈0, 1, 10, 2〉 and 〈0, 2, 8, 1〉 are very low in the Ce80Ni20 glasses
which is the reason that the Ce80Ni20 MGs is hard to be prepared by
splat quenching and melt spinning experimentally. The calculated self-
diffusion coefficients are also in good agreement with the experimental
measurements. All these results indicate that our EAM potential of Ce-
Y. Lei et al. Computational Materials Science 150 (2018) 1–8

Fig. 7. The fraction of the icosahedral cluster 〈0, 0, 12, 0〉 and icosahedral-like clusters at 300 K under the cooling rate of 1011 K/s of Ce80Ni20 structure.

Fig. 8. MSD of liquid Ce80Ni20 at different temperatures.

Table 5 2016YFB0300500) the Fundamental Research Funds for the Central

Calculated self-diffusion coefficients of Ni (DNi) and Ce (DCe) in the Ce80Ni20 Universities (No. JZ2016HGPB0671). The first-principles molecular
solids and liquids at different temperatures. The unit of the self-diffusion dynamic simulations are supported by the supercomputing system in
coefficient is 10−9 m2/s. the Supercomputing Center of University of Science and Technology of
T 300 K 400 K 500 K 600 K 700 K 800 K 900 K 1000 K China.
DNi 0.01 0.07 0.19 0.45 0.75 1.07 1.45 1.63

T 1100 K 1200 K 1300 K 1400 K 1500 K 1600 K 1700 K 1800 K References

DNi 2.10 2.51 2.64 3.70 4.23 4.39 4.42 4.63
[1] J. Allen, R.M. Martin, Kondo volume collapse and the γ→ α transition in cerium,
T 300 K 400 K 500 K 600 K 700 K 800 K 900 K 1000 K Phys. Rev. Lett. 49 (15) (1982) 1106.
DCe 0.003 0.01 0.22 0.40 0.64 0.87 1.13 1.40 [2] N. Mathur, et al., Magnetically mediated superconductivity in heavy fermion
compounds, Nature 394 (6688) (1998) 39.
T 1100 K 1200 K 1300 K 1400 K 1500 K 1600 K 1700 K 1800 K
[3] K. Andres, J.E. Graebner, H.R. Ott, 4f-Virtual-bound-state formation in CeAl3 at low
DCe 1.67 2.00 2.09 2.84 3.21 3.44 3.75 3.90 temperatures, Phys. Rev. Lett. 35 (26) (1975) 1779–1782.
[4] Y. Ōnuki, Y. Shimizu, T. Komatsubara, Magnetic property of a new Kondo lattice
intermetallic compound: CeCu6, J. Phys. Soc. Jpn. 53 (4) (1984) 1210–1213.
Ni has high accuracy and good transferability. So, it may have wide [5] K. Winzer, Giant kondo resistivity in (La, Ce) B6, Solid State Commun. 16 (5) (1975)
521–524.
applications in the study of various Ce-Ni binary systems. [6] B. Barbara, et al., On the magnetic ordering of CeAl2, Solid State Commun. 24 (7)
(1977) 481–485.
Acknowledgments [7] A. Mirmelstein, et al., Effect of chemical and external pressure on the structure of
intermetallic compound CeNi, J. Alloys Compd. 444–445 (2007) 281–284.
[8] B. Zhang, et al., Properties ofCe-based bulk metallic glass-forming alloys, Phys. Rev.
This work was supported by the National Natural Science B 70 (22) (2004).
Foundation of China (No. 51571079), China MOST 973 Program (No. [9] B. Zhang, et al., Amorphous metallic plastic, Phys. Rev. Lett. 94 (20) (2005)
205502.
2015CB856800), National Key Research and Development Project (No.

7
Y. Lei et al.
[10] B. Zhang, et al., “Soft” bulk metallic glasses based on cerium, Appl. Phys. Lett. 85
(1) (2004) 61–63.
[11] Y. Zhao, B. Zhang, Evaluating the correlation between liquid fragility and glass-
forming ability in the extremely strong Ce-based bulk metallic glasses, J. Appl.
Phys. 122 (11) (2017) 115107.
[12] Y. Zhao, et al., Anomalous packing state in Ce-Ga-Cu bulk metallic glasses,
Intermetallics 84 (2017) 25–29.
[13] S.M. Chathoth, M.M. Koza, A. Meyer, Complex atomic dynamics in a deep-eutectic
binary metallic melt, Mater. Chem. Phys. 136 (2–3) (2012) 296–299.
[14] F. Ercolessi, J.B. Adams, Interatomic potentials from first-principles calculations:
the force-matching method, EPL (Europhysics Letters) 26 (8) (1994) 583.
[15] H.J. Monkhorst, J.D. Pack, Special points for Brillouin-zone integrations, Phys. Rev.
B 13 (12) (1976) 5188.
[16] P. Brommer, F. Gähler, Potfit: effective potentials from ab initio data, Modell.
Simul. Mater. Sci. Eng. 15 (3) (2007) 295.
[17] P. Brommer, et al., Classical interaction potentials for diverse materials from ab
initio data: a review of potfit, Modell. Simul. Mater. Sci. Eng. 23 (7) (2015) 074002.
[18] W. Kohn, L.J. Sham, Self-consistent equations including exchange and correlation
effects, Phys. Rev. 140 (4A) (1965) A1133.
[19] M.S. Daw, M.I. Baskes, Semiempirical, quantum mechanical calculation of hy-
drogen embrittlement in metals, Phys. Rev. Lett. 50 (17) (1983) 1285–1288.
[20] M.S. Daw, M.I. Baskes, Embedded-atom method: derivation and application to
impurities, surfaces, and other defects in metals, Phys. Rev. B 29 (12) (1984) 6443.
[21] S. Chantasiriwan, F. Milstein, Higher-order elasticity of cubic metals in the em-
bedded-atom method, Phys. Rev. B 53 (21) (1996) 14080.
[22] R.A. Johnson, Alloy models with the embedded-atom method, Phys. Rev. B 39 (17)
(1989) 12554–12559.
[23] G. Kresse, J. Hafner, Ab initiomolecular dynamics for liquid metals, Phys. Rev. B 47
(1) (1993) 558–561.
[24] G. Kresse, J. Furthmüller, Efficient iterative schemes for ab initio total-energy cal-
culations using a plane-wave basis set, Phys. Rev. B 54 (16) (1996) 11169.
Computational Materials Science 150 (2018) 1–8
[25] G. Kresse, D. Joubert, From ultrasoft pseudopotentials to the projector augmented-
wave method, Phys. Rev. B 59 (3) (1999) 1758.
[26] Y. Wang, J.P. Perdew, Correlation hole of the spin-polarized electron gas, with exact
small-wave-vector and high-density scaling, Phys. Rev. B 44 (24) (1991)
13298–13307.
[27] S. Plimpton, Fast parallel algorithms for short-range molecular dynamics, J.
Comput. Phys. 117 (1) (1995) 1–19.
[28] D. Schopf, et al., Embedded atom method potentials for Al-Pd-Mn phases, Phys.
Rev. B 85 (5) (2012).
[29] J.-P. Poirier, Introduction to the Physics of the Earth's Interior, Cambridge
University Press, 2000.
[30] H.W. Sheng, et al., Highly optimized embedded-atom-method potentials for four-
teen fcc metals, Phys. Rev. B 83 (13) (2011).
[31] W. Zachariasen, F. Ellinger, The crystal structures of cerium metal at high pressure,
Acta Crystall. Sect. A: Crystal Phys., Diffract., Theor. General Crystall. 33 (1) (1977)
155–160.
[32] J. Häglund, et al., Theory of bonding in transition-metal carbides and nitrides, Phys.
Rev. B 48 (16) (1993) 11685.
[33] J.-L. Bobet, et al., Hydrogenation of CeNi: hydride formation, structure and mag-
netic properties, Intermetallics 14 (2) (2006) 208–212.
[34] V. Paul-Boncour, et al., Structural characterization of RNi2 (R≡ La, Ce) inter-
metallic compounds and their hydrides, J. Less Common Metals 131 (1–2) (1987)
201–208.
[35] D.T. Cromer, C.E. Olsen, The crystal structure of PuNi3 and CeNi3, Acta Crystall. 12
(9) (1959) 689–694.
[36] J. Tang, et al., Strongly enhanced paramagnetism in CeNi5—xGax and LaNi3Ga2, J.
Alloys Compd. 207 (1994) 241–244.
[37] D. Alfè, PHON: A program to calculate phonons using the small displacement
method, Comput. Phys. Commun. 180 (12) (2009) 2622–2633.
[38] M.P. Allen, D.J. Tildesley, Computer Simulation of Liquids, Oxford University Press,
2017.