You are on page 1of 11

Acta Materialia 240 (2022) 118314

Contents lists available at ScienceDirect

Acta Materialia
journal homepage: www.elsevier.com/locate/actamat

Full length article

Chemical domain structure and its formation kinetics in CrCoNi


medium-entropy alloy
Jun-Ping Du a,∗, Peijun Yu a, Shuhei Shinzato a, Fan-Shun Meng a, Yuji Sato b, Yangen Li a,
Yiwen Fan a, Shigenobu Ogata a,∗
a
Department of Mechanical Science and Bioengineering, Osaka University, Osaka 560-8531, Japan
b
Department of Mechanical Engineering, The University of Tokyo, Tokyo 113-8656, Japan

a r t i c l e i n f o a b s t r a c t

Article history: The formation of local chemical order in medium-entropy alloys and high-entropy alloys (MEAs/HEAs)
Received 30 May 2022 has been strongly suggested in recent experimental observations. Since chemical order can lead to
Revised 19 August 2022
changes in mechanical and functional properties, tailoring of chemical order is a promising approach
Accepted 28 August 2022
for further improving those properties of MEAs and HEAs. However, details remain unclear regarding
Available online 30 August 2022
the atomic structure of the chemical order and the formation kinetics. Here, employing a large-scale
Keywords: Monte Carlo/molecular dynamics hybrid annealing simulation with a neural network potential, we find a
Atomic ordering chemical-domain structure (CDS) after annealing below 800 K in FCC CrCoNi MEA. In addition, the forma-
Kinetic Monte Carlo tion kinetics, such as the formation time and process and time–temperature–chemical-order diagrams of
Artificial neural networks the CDS, were successfully obtained using a kinetic Monte Carlo simulation with artificial neural network
Atomistic modeling acceleration. The findings provide key information for controlling chemical order via thermal processing.
Atomic structure
© 2022 The Author(s). Published by Elsevier Ltd on behalf of Acta Materialia Inc.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)

1. Introduction on atom probe tomography observations of CrCoNi MEA annealed


at 973 K for 384 h, Inoue et al. [26] recently argued that the struc-
Medium-entropy alloys (MEAs) and high-entropy alloys (HEAs) tural unit of local chemical order is a 10-atomic-layer structure (∼
with multiple principal elements constitute a new category of po- 2 nm) with an alternating arrangement of Cr-rich {100} and Co/Ni-
tential structural materials in engineering design that exhibit su- rich {100} atomic layers. However, based on electron diffraction
perior combinations of strength, ductility, and fracture toughness observations of CrCoNi MEA annealed at 873 K and 1273 K for 1 h,
[1–12]. Compositional adjustments of these MEAs and HEAs have Zhou et al. [27] argued that the structural unit of local chemical
produced other superior properties, such as strong corrosion resis- order is an alternating arrangement of Cr-rich {113} and Co/Ni-
tance [13,14], fatigue resistance [15,16], wear resistance [17], and rich {113} atomic layers with the size of 0.6 nm, similar to the
functional properties [12,18]. structural unit of local chemical order found in VCoNi MEA [28].
The existence of chemical short-range order (CSRO) in the an- The differing conclusions among different experiments regarding
nealed equiatomic CrCoNi MEA system has been suggested by both chemical order may be results of differences in experimental con-
experimental [19,20] and theoretical approaches [21–24]. Zhang ditions, such as sample preparation, heat treatment, and analytical
et al. asserted that hardness enhancement in CrCoNi MEA was due method.
to nanometer-sized CSRO with a superlattice diffraction pattern As a complement to experimental research, atomic simulation
formed after aging at 1273 K for 120 h followed by slow furnace has been used to reproduce chemical order with the atomic-scale
cooling [19]. Meanwhile, Yin et al. argued that the CSRO formed resolution, for instance in annealing simulations by density func-
above 873 K in the alloy was negligible or had a negligible effect tional theory (DFT)-based atom-swap Monte Carlo (MC) methods
on its strength and hardness [25]. Therefore, they speculated that with molecular dynamics (MD) sampling (atom-swap-MC-based
the chemical order found by Zhang et al. [19] was formed during MC/MD hybrid method). Although DFT is reliable in the descrip-
the furnace cooling when the system fell below 873 K [25]. Based tion of complicated interatomic interactions in MEAs and HEAs,
the physical size of the simulation model is limited to, at most,
hundreds of atoms, i.e., ∼ 4 nm3 [21,22,24], due to the high com-

Corresponding authors. putational cost of the method. Therefore, although chemical order
E-mail addresses: jpdu@tsme.me.es.osaka-u.ac.jp (J.-P. Du), ogata@me.es.osaka-
in MEAs and HEAs has been determined using the DFT-based MC
u.ac.jp (S. Ogata).

https://doi.org/10.1016/j.actamat.2022.118314
1359-6454/© 2022 The Author(s). Published by Elsevier Ltd on behalf of Acta Materialia Inc. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/)
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

method, the atomic structures of the nanometer-sized chemical or- fect of non-collinear spin configurations on energy does not qual-
der found in experiments [19,26–28] may not be fully captured itatively alter the chemical ordering tendency [35]. The validation
in the physically small simulation model. More importantly, atom- of the NNP (Version 2) and the performance comparison with the
swap-MC-based MC/MD hybrid simulations can only reproduce NNP (Version 1) [35] are presented in Section 1 in Supplemen-
equilibrium chemical order at each temperature and thus do not tary Information. The NNP files used in simulations by LAMMPS
supply any kinematic information, such as the chemical order for- through the LAMMPS interface implemented in n2p2 can be down-
mation time or formation process. To expand the size of the sim- loaded from Ref. [37].
ulation model, the embedded atom method (EAM) [29] has been
used in place of DFT. However, the accuracy of the EAM is ques-
2.2. Monte Carlo/molecular dynamics hybrid simulations
tionable for studying the details of chemical order in general. For
example, a strong attractive interaction in Ni–Cr pairs in CrCoNi
To reveal the temperature-dependent equilibrium chemical or-
MEA has been found in DFT and experimental studies [20–22] but
der in CrCoNi MEA, we performed atom-swap-MC-based MC/MD
has not been reproduced by the EAM. Recently, machine learning
hybrid simulations using LAMMPS code [41] at constant temper-
interatomic potentials [30] based on a DFT training dataset, such
ature and zero pressure with a 2916-atom (N = 2916) Cr–Co–
as a neural network potential (NNP) [31] or a Gaussian approxima-
Ni equiatomic cubic-shaped FCC supercell model and using the
tion potential [32], have rapidly developed and been successfully
newly trained NNP. Initially, we randomly assigned Cr, Co, and
applied to atomic simulations for various materials [33,34]. In our
Ni atoms to the FCC sites with keeping equiatomic concentration,
prior work, we have trained an NNP (Version 1) on the CrCoNi
and applied the atom-swap-MC-based MC/MD hybrid method. One
MEA system that describes well the bulk and defect properties
MC/MD hybrid simulation step consists of three atom-swap MC tri-
of the MEA including local chemical interactions [35]. Generally,
als between atoms of different elements followed by a 10-fs MD
an NNP has a flexible architecture, so it can be used to describe
sampling under NPT (i.e., constant number of atoms, pressure, and
the complex potential surface of a multicomponent alloy system pm
given a well-prepared DFT-based training dataset. In addition, NNP temperature). Warren–Cowley order parameters [42], αimj = 1 − ij
cj ,
computation is much faster than that of DFT, and thus simulation were used to characterize the degree of ordering of element i and
model of a physical size sufficient to contain the nanometer-sized element j in the mth-neighbor coordinate shell, where pm ij
is the
chemical order can be employed. In the present work, we further probability of finding an atom of element j in the mth-neighbor
trained the NNP (Version 1) for CrCoNi MEA [35] with n2p2 code coordinate shell of an atom of element i, and c j is the atomic con-
[36] to improve its performance, and then constructed an NNP centration of element j. When calculating the αimj value, each atom
(Version 2) having a crucial advantage to the EAM (see Section 1 in was re-positioned on the FCC lattice site to extract “on-lattice”
Supplementary Information). In this training, we included not only chemical order. A positive αimj denotes a repulsive tendency be-
the total potential energy of the system but also the atomic force tween element j and element i, and vice versa.
acting on each atom in the DFT training dataset. Then, based
on the NNP (Version 2), we applied the atom-swap-MC-based
MC/MD hybrid method to obtain the temperature-dependent equi- 2.3. Kinetic Monte Carlo simulations
librium chemical order for a simulated annealing of hundreds of
swaps per atom in a large CrCoNi MEA model: 32 nm3 (2916 As noted in Section 2.2, the atom-swap-MC-based MC/MD hy-
atoms). This method newly revealed a chemical-domain structure brid method can study only the thermal equilibrium state. There-
(CDS) in the MEA, which could be corresponding to the conven- fore, we here performed kMC vacancy diffusion simulation in or-
tional domains in alloys. This architecture was found to consist of der to estimate the chemical order formation time and forma-
many few-nanometer-sized superlattice-chemical-ordered-domains tion process at various temperatures. The basic concept of the
(SLCOD) with various orientations of order in single-phase FCC Cr- kMC simulation is the same as in our prior work [43], in which
CoNi MEA at annealing temperatures below 800 K. Especially, a a time–temperature–CSRO relation was derived. The time incre-
CDS with larger-sized SLCOD was found at annealing temperatures ment tkMC corresponding to the average incubation time for a
below 700 K. The SLCOD has an intermetallic-compounds-type su- ln(1/r )
vacancy jump in the kMC simulation is given by tkMC = 12 ,
perlattice chemical order, which consists of an alternating arrange- i=1 νi
ment of Cr-rich and Co/Ni-rich layers in {100} and {110} orienta- where r is a random number uniformly distributed between 0
tions. On the other hand, local CSROs within mostly the second- and 1 such that ln(1/r ) = 1. νi is the jump frequency of any one
nearest-neighbor range of FCC was found at annealing tempera- of the 12 vacancy jump directions to the nearest-neighbor FCC
tures above 800 K. The obtained results are consistent with re- sites. νi = ν0 exp(−E i /kB T ), where ν0 = 1012 s−1 , is an attempt
cent experimental results [19,20,26–28]. In addition to the chem- frequency and is set as the typical atomic vibration frequency. It
ical order analysis at thermal equilibrium, we performed a ki- should be noted that we approximate the attempt frequency by the
netic Monte Carlo (kMC) vacancy diffusion simulation employing a constant value for simplicity, while it should have a local chemical
newly developed very fast neural network scheme of vacancy jump environment dependency which can be explicitly computed using,
rate prediction. The method successfully estimated the chemical for example, minimum energy pathway analysis [44,45]. kB is the
order formation time and the formation process at various tem- Boltzmann constant, and T is the temperature. E i is the activa-
peratures. With these predictions as a basis, time–temperature– tion energy for the vacancy jump given by the minimum energy
chemical-order diagrams were drawn, providing key information path computations using the nudged elastic band (NEB) method
for controlling chemical order via thermal processing. [46,47] with an interatomic potential. However, here the high com-
putational cost of NNP makes it inefficient to compute E i with
2. Methodology the NEB method. Therefore, we newly developed a fast computa-
tional technique of vacancy jump activation energy estimation us-
2.1. Neural network potential for CrCoNi MEA ing an artificial neural network (ANN), which can avoid the time-
consuming minimum energy path computations at each kMC step
The NNP (Version 2) for CrCoNi MEA was trained via n2p2 code (see Fig. 1) and significantly accelerate the kMC simulation. The
[36] based on a DFT training dataset (see Ref. [37]) using collinear ANN predicts the vacancy jump activation energy E as the out-
spin-polarized DFT calculations on VASP [38–40]. However, the ef- put from two inputs: on-lattice (unrelaxed) atomic configuration

2
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

Fig. 1. Illustration of the relaxed structure energy estimator based on artificial neural network predicting vacancy jump activation energy (E) from two inputs: unrelaxed
configuration before the vacancy jump (RI ) and unrelaxed near-saddle atomic configuration (RS ).

before the vacancy jump RI and unrelaxed near-saddle atomic con- we performed two simulations for each temperature starting from
figuration RS = 12 (RI + RF ), where RF is the on-lattice atomic con- different random initial structures. In the following discussion, we
figuration after the vacancy jump (see Fig. 1). The same descriptors present averaged data of the two simulations except for atomic
of atomic structure in the ANN as in the NNP (Version 2), such structure visualization. It should be worth noting that we annealed
as symmetry functions, were used. The training dataset was con- also at 500 K and 600 K, and found that at these temperatures the
structed using the NEB method [46,47] and the NNP (Version 2) for αimj values are still gradually developing even after 600N MC tri-
72,490 different vacancy configurations in a 107-atom cubic FCC als while maintaining the essential chemical order trend described
supercell (see Section 2 in Supplementary Information). In the NEB in the following discussion (see Section 3 in Supplementary Infor-
calculation, the initial and saddle atomic configurations were fully mation). Figure 2 shows that the αimj value of the first- and second-
relaxed for computation of the energies in these configurations, neighbor coordinate shells, αi1j and αi2j , are well developed at lower
E relax (RI ) and E relax (RS ), respectively. The activation energy was
temperatures and gradually decrease with increasing temperature.
computed as the difference between the ANN outputs of these en-
The first-neighbor coordinate shell demonstrates a repulsive Cr–Cr
ergies: E = E relax (RS ) − E relax (RI ). The constructed ANN’s activa-
interaction (positive αCrCr
1 ) and attractive Cr–Co and Cr–Ni inter-
tion energy prediction showed an average error of only 0.05 eV, as
actions (negative αCrCo and αCrNi
1 1 ), in agreement with the results
confirmed using a test dataset containing 20,0 0 0 different vacancy
of DFT-based MC simulations [21,22] and experiments using X-ray
configurations that are not included in the training dataset (see
absorption fine structure techniques [20]. In contrast, Cr–Cr, Cr–Co,
Supplementary Fig. S8). The calculated 20,0 0 0 activation energies
and Cr–Ni pair interactions of the second-neighbor coordinate shell
for vacancy migration show a large variation, i.e. the range with
show a completely opposite tendency. Notably, above 800 K the
the three standard deviations of the mean is 0.31 eV ∼ 1.38 eV and
range of chemical order is limited to mostly the second-nearest-
0.34 eV ∼ 1.36 eV for the NNP and the ANN, respectively, which
neighbor, that is, only so-called CSRO is formed at such high tem-
are in good agreement with that for DFT, 0.35 eV ∼ 1.24 eV [48].
peratures. In contrast, at 700 K non-negligible finite αimj s can be
The ANN prediction scheme can accelerate the kMC simulation
speed 500 times in the case of the 2916-atom Cr–Co–Ni equiatomic observed over a longer range more than m = 22 (r = 12 Å) show-
cubic-shaped FCC supercell model with one vacancy (see Supple- ing that not only CSRO but also a long-range chemical order be-
mentary Fig. S9) because 200 ∼ 300 times of energy/force eval- yond CSRO is formed.
uations are necessary for one activation energy evaluation while By sampling local chemical arrangements within the range of
only twice ANN evaluations for initial and saddle atomic configura- r = 12 Å at different FCC atom sites in the annealed models, such
tions are needed in ANN prediction strategy. The advantage of the as the fully annealed models at 700 K and partially annealed mod-
ANN scheme becomes more significant for larger models requiring els at 600 K and 500 K (see Section 3 in Supplementary Informa-
a larger number of energy/force evaluations in one activation en- tion), and by carefully observing these chemical arrangements, we
ergy evaluation. found that two types of superlattice chemical order were forming,
The ANN training dataset for activation energy estimation was differing in orientation depending on the sampling (see Fig. 3). One
constructed from the vacancy jump activation energy estimated at is a Cr/CoNi {100} superlattice (Fig. 4(a)) with an alternating ar-
0 K using the NNP with the NEB method [46,47]. The effect of vi- rangement of {100} Cr-rich and Co/Ni-rich layers, as has been sug-
brational entropy was neglected in the kMC simulation, while it gested in prior experiments [26] and DFT computations [23] (see
was considered in the atom-swap-MC-based MC/MD hybrid sim- Section 4 in Supplementary Information). The other is a Cr/CoNi
ulations in the Section 2.2. The effect of the vibrational entropy {110} superlattice (Fig. 4(b)) with an alternating arrangement of
on the activation energy is roughly a few kB T [49], where kB {110} Cr-rich and Co/Ni-rich layers. The latter is newly found in
is the Boltzmann constant, and T is the temperature. Thus, ex- the present study and shows the same stacking pattern of Cr-rich
pected error in time in log-scale is not significant, such as ∼ layers as the intermetallic compound oI6-Ni2 Cr [51] with Immm
log[exp(akB T /kB T )] < 1 (a ∼ 2 [50]). symmetry. The formed superlattice structures provide the possi-
ble atomic pictures of the ordered structures in the order–disorder
3. Results transition found in the CrCoNi MEA [52]. For each of these two su-
perlattices, we can define a unit chemical structure consisting of
3.1. Temperature-dependent chemical order in CrCoNi MEA Cr-rich and Co/Ni-rich layers (see Fig. 4). Since there are six equiv-
alent {110} planes and three equivalent {100} planes in the FCC
The values of Warren–Cowley order parameters, αimj , up to m lattice, six and three equivalent superlattices with different ori-
= 22 are shown in Fig. 2 for models annealed using the atom- entations can exist, respectively. To study their energetic stabili-
swap-MC-based MC/MD hybrid method at various temperatures: ties, we used DFT to compute the energies of these unit chem-
700 K and 800 ∼ 1200 K after 600N and 100N atom-swap MC ical structures. For the structures with the alternative arrange-
trials, respectively, where N = 2916 atoms. To avoid the artificial ment of Cr-rich and Co/Ni-rich layers in CrCoNi MEA, 108-atom
effects of the initial structure selection on the annealed structure, supercells were used for DFT computations. For each case, we pre-

3
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

Fig. 2. Range of chemical order at various temperatures in CrCoNi medium-entropy alloy. Stacked Warren–Cowley order parameters (αimj ) between element i and element
j in the mth-neighbor coordinate shell from the element i are shown as functions of the coordinate shell radius (r) in structures annealed at various temperatures ranging
from 700 K to 1200 K and in random structure. 600N and 100N atom-swap Monte Carlo trials were performed for 700 K and 800 ∼ 1200 K simulations, respectively, where
N = 2916 atoms.

pared ten structure samples with different random atomic arrange- in the unit chemical structure of the superlattice, and β (k ) repre-
ments of Co and Ni in the Co/Ni-rich layers of the superlattice sents element type at site k in the model, where site k is located
structure while the Cr-rich layers of the superlattice structures are around the site i. Site k is the corresponding site of site K in the
fully occupied by Cr. We then took the average of the energies of unit chemical structure when the unit chemical structure superim-
the ten structure samples. The DFT computations were also per- poses on the model local structure centered on site i after space
formed using collinear spin-polarized DFT calculations on VASP group operation TIs . δ is Kronecker delta: if the element type at
[38–40]. The exchange correlation between electrons was treated site k is the same as that at site K, then δα (K )β (k ) = 1; otherwise
with the generalized gradient approximation in the Perdew-Burke- δα (K )β (k) = 0. If Di > 0.85, we define site i as belonging to super-
Ernzerhof (PBE) form Perdew et al. [53]. An energy cutoff of 350 eV lattice Is = argmax{Is ,TI } Di , while if Di ≤ 0.85, then site i does not
s
was used for the plane wave basis set. A Monkhorst-Pack mesh belong to either of the superlattices. The space-group operation TIs
(4 × 4 × 4 ) was used for the Brillouin-zone integration. The ge- giving the maximum D value provides the orientation of the su-
ometry optimizations were performed until atomic forces were perlattice chemical order. Performing the analysis of the matching
smaller than 0.01 eV/Å. A total energy convergence criterion of degree D across all atoms in each annealed model at each tem-
10−6 eV was used for the electronic structure optimization. Thus, perature, we determined the type of superlattice by D > 0.85 and
the average enthalpy difference between the two superlattices is its orientation at each atom site, and then we assigned a color to
only 0.027 eV/atom, suggesting that the two types of superlattices each atom depending on the type of superlattice and orientation
can co-exist. (see Fig. 5(a)). At temperatures of 700 K and below, we found a
CDS consisting of many nanometer-sized SLCOD of various super-
3.2. Three-dimensional chemical domain formation in CrCoNi MEA lattice orientations, i.e., a kind of “chemical” polycrystal. The size of
1
each SLCOD was estimated by d = (n) 3 , where n is the number
To more objectively confirm whether the Cr/CoNi {100} and of atoms in a SLCOD and  is the averaged atomic volume in the
{110} superlattices were formed at the lower temperatures, we MEA. Fig. 5(b) and (c) show the temperature dependence for the
applied a chemical structure matching analysis to the annealed maximum size and average size of SLCOD. The maximum size of
models. Thus, to determine which superlattice-chemical-order each SLCOD at 500 ∼ 700 K was larger than 2 nm, with SLCOD still con-
atom in a model more likely belongs to, we defined and used tinuing to grow after 600N atom-swaps in 500 K and 600 K. The
a matching degree D for an atom site i in the model. The size of SLCOD at each equilibrium state decreased with increasing
matching degree D for site i in the model was defined as Di = annealing temperature and eventually became CSRO at tempera-
N
1 
Is
tures above 800 K. Notably, the structures obtained by the two in-
max{Is ,TI }
s N
δα (K )β (k) . Here Is represents the Cr/CoNi {100} or dependent simulations at 700 K were almost fully covered by the
Is K=1
two chemical order superlattices. Since the recent experiment by
{110} superlattice, TIs represents the space group operation of su-
Zhou et al. [27] suggests a Cr/CoNi {113} chemical structure with
perlattice Is , NIs represents the number of atom sites in the unit
an alternating arrangement of {113} Cr-rich and Co/Ni-rich layers
chemical structure (Fig. 4) of superlattice Is : NCr/Ni{100} = 14 or
(see Fig. 6), which is similar to the Cr/CoNi {110} but not the same,
NCr/Ni{110} = 19. α (K ) represents element type at site K (Kth site)

4
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

Fig. 3. Observed long-range superlattice chemical order in the CrCoNi medium-entropy alloy. Local chemical structures exhibiting superlattice chemical order were sampled
from various locations within the CrCoNi medium-entropy alloy models fully annealed at 700 K and partially annealed at 600 K and 500 K.

Fig. 4. (a) Cr/CoNi {100} and (b) Cr/CoNi {110} superlattice atomic arrangements.

5
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

Fig. 5. Nanometer-sized chemical-domain structure (CDS) of superlattice-chemical-ordered domains (SLCOD) in CrCoNi medium-entropy alloy. (a) Representative distribu-
tions and (b) maximum size (dmax ) and (c) mean size (d) of Cr/CoNi {100} and Cr/CoNi {110} SLCOD in structures annealed at various temperatures ranging from 700 K to
1200 K and in random structure. 600N and 100N atom-swap Monte Carlo trials were performed for 700 K and 800 ∼ 1200 K simulations, respectively, where N = 2916
atoms. Above 800 K, the CDS was almost at equilibrium.

we applied the chemical structure matching analysis for Cr/CoNi by 0.081 and 0.054 eV/atom, respectively (see Section 5 in Sup-
{113} to the equilibrium structures at different temperatures. We plementary Information), and thus can appear only at the higher
found that the CSRO structures formed in the equilibrium struc- temperatures, 800 K and above. Moreover, the Cr/CoNi {113} can
tures at 800 K and above have subnanometer-sized Cr/CrNi {113} have a higher energy penalty for its further growth beyond the
chemical domain in addition to the subnanometer-sized Cr/CoNi CSRO size (∼ 1 nm) because the number of the unfavored Cr–Cr
{100} and Cr/CoNi {110}, while the volume fraction of the Cr/CoNi nearest-neighbor pairs density rapidly increases with the increase
{113} domain is smaller than that of the Cr/CoNi {100} and Cr/CoNi of domain size as seen in Supplementary Fig. S14, while Cr/CoNi
{110} domains (see Fig. 6 and Supplementary Fig. S13). The Cr/CoNi {110} and Cr/CoNi {100} do not as seen in Fig. 4. Formation kinet-
{113} structure is closely related to the Cr/CoNi {110} structure, be- ics and size distribution of the Cr/CoNi {113} structure at the 873 K
cause the Cr/CoNi {110} superlattice can equivalently be viewed and above is discussed in the Discussion section with a comparison
as Cr/2CoNi {113}, such as an alternative arrangement of one Cr- with experiments.
rich and two Co/Ni-rich {113} layers as shown in Supplementary The above results of CDS formation at the lower temperatures
Fig. S14; nevertheless the Cr/CoNi {113} structure has a higher av- suggests that by controlling thermal processing, such as quenching
erage enthalpy than Cr/CoNi {110} and Cr/CoNi {100} structures rate and annealing time and temperature, we can control the for-

6
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

Fig. 6. Representative distributions of Cr/CoNi {113} chemical structure at 800 K and above in structures shown in Fig. 5. Chemical structure of Cr/CoNi {113} are shown in
the dashed frame and two equivalent {113} planes are shown as a guide to the eye.

mation of nano-structured CDS while managing the size of SLCOD, ble S6. Next, we correlated the tkMC under the fixed vacancy con-
1
and thus we can tailor the mechanical and functional properties of centration (cv0 = 2916 ) in kMC simulations to an annealing time (ta )
CrCoNi MEA. under an average vacancy concentration (cv ) in a realistic MEA sys-
tem by cv0 dtkMC = cv dta , according to the assumption that the evo-
3.3. Time–Temperature–Chemical-order diagram lution time is inversely proportional to the average vacancy con-
centration. The obtained annealing time (ta ) is strongly dependent
Concrete kinematic information regarding chemical order for- on the cv as shown in Section 6 in Supplementary Information.
mation is essential for tailoring the CDS via thermal process- For instance, in isotherm annealing at 673 K it may require a very
ing. Such information can also provide a comprehensive under- long annealing time, perhaps months or years, to form an equi-
standing of recent experiments conducted under different con- librium state under an equilibrium vacancy concentration (∼ 10−7
ditions. Therefore, the vacancy jump kMC simulations based on appm) at the annealing temperature. However, if an equilibrium
the ANN’s activation energy prediction scheme were performed to vacancy concentration at 10 0 0 K (∼ 10−3 appm) can be maintained
give the time–temperature–αimj (T–T–αimj ) diagrams and the time– at the annealing temperature of 673 K, the annealing time to the
temperature–maximum-domain-size dmax (T–T–dmax ) diagram by equilibrium state becomes hours, which leads to an acceleration of
applying LAMMPS [41] to a 2916-atom Cr–Co–Ni equiatomic cubic- the chemical order formation by ∼ 104 . The equilibrium vacancy
Eq
shaped FCC supercell model with one vacancy. For the initial concentration (cv (T )) [54] in the CrCoNi MEA was estimated by
chemical structure for the T–T–αimj and the T–T–dmax computa- exp(n − 1 − kEvT )
Eq
tions, an annealed structure (αCrCr
1 =0.15) given by the atom-swap- cv ( T ) = B
, where n is 3 for the ternary sys-
n + exp(n − 1 − kEvT )
MC method at a CrCoNi melting temperature of 1530 K estimated B
tem, and Ev = 1.78 eV is the averaged vacancy formation energy
by the NNP was used. To avoid artificial effects of the initial struc-
computed using the NNP, which is in excellent agreement with
ture selection, we performed three simulations at each tempera-
that computed using DFT, 1.82 eV [55]. It should be noted that
ture, starting from different snapshots of the atom-swap MC sim-
for simplicity we employed the average vacancy formation energy
ulation results after well-annealing at 1530 K. The αimj and the
as an approximation to estimate the vacancy concentration, while
dmax values were then computed by taking the average of each
the vacancy concentration should have a chemical order depen-
over the three simulations to give the averaged T–T–αimj diagrams
dency because the distribution of vacancy sites and vacancy en-
and T–T–dmax diagram. To draw the diagrams, the αimj –tkMC and ergies varies with the chemical order and can somewhat affect the
the dmax –tkMC relationships at various temperatures
 (T ) were fitted vacancy concentration [56,57]. When we anneal the MEA samples
to the function, Y = A(T ) − (A(T ) − B ) exp(− C (T )tkMC ), where Y at lower temperatures immediately after cooling down from a near
is αimj or dmax , tkMC is the time given by the kMC simulations, melting temperature, it is reasonable to assume a higher vacancy
both A(T ) and ln(C (T )) are quadratic or exponential functions concentration at the beginning of the isotherm annealing than the
of T , and B is a constant. For instance, A(T ) = a0 + a1 T + a2 T 2 equilibrium vacancy concentration at the annealing temperature
and A(T ) = a0 exp(−T /a1 ) + a2 for αimj and dmax , respectively, and because it takes longer time at such lower temperatures to release
ln(C (T )) = c0 + c1 T + c2 T 2 for both αimj and dmax , where a0 , a1 , a2 , the excess vacancies gained at the solidification process near melt-
ing temperature to approach the equilibrium concentration.
c0 , c1 , and c2 are fitting parameters as listed in Supplementary Ta-

7
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

Fig. 7. Time evolution of chemical ordering in CrCoNi medium-entropy alloy. (a) Time–temperature–αCrCr
1
and (b) time–temperature–maximum-domain-size (T–T–dmax )
diagrams for chemical domain evolution induced by vacancy diffusion in isotherm annealing immediately after cooling down from 1500 K to each isotherm annealing
temperature in CrCoNi medium-entropy alloy.

Therefore, to take the advantage of the high vacancy concen- extremely low. While the slow furnace cooling from high temper-
tration for promoting the chemical order formation, we adopted ature as adopted by Zhang et al. [19] is an alternative heat treat-
a heat treatment process of isotherm annealing immediately af- ment process that takes advantage of the high vacancy concentra-
ter rapid cooling from the near melting temperature of 1500 K to tion formed at high temperature to promote the chemical order
the annealing temperature, so that the vacancy concentration at formation.
Eq
the beginning of the isotherm annealing is cv (T , t = 0 ) = cv (T =
1500 K ). Then, the evolution of cv (T , t ) was obtained by solving
4. Discussion
the diffusion equation for vacancies in a spherical crystalline grain
with a typical radius of 100 μm (see Section 6 in Supplementary
As discussed above, the Cr/CoNi {100} superlattice structure
Information). The time given by the kMC simulations can be re-
 ta suggested by Inoue et al. [26] and the Cr/CoNi {113} superlattice
cv ( T , t )
lated to the annealing time ta by tkMC = dt. Obtained structure suggested by Zhou et al. [27] were both found in the
0 cv0
equilibrium structures as well at different temperatures. These su-
T–T–αCrCr and T–T–dmax diagrams are shown in Fig. 7(a) and (b),
1
perlattice structures were also found in the present study as shown
respectively. The other Warren–Cowley order parameters, such as
above. Moreover, to directly compare our kMC simulation to recent
T–T–αi1j and T–T–αi2j , can be found in Supplementary Fig. S17. No-
experiments for the Cr/CoNi {113} chemical-domain size distribu-
tably these results confirm that there is a one-to-one correspon-
tion, we computed the Cr/CoNi {113} chemical-domain size distri-
dence between changes in these parameters, and thus the degree
bution by annealing at 873 K and 1273 K using the kMC method as
of chemical order can be representatively described by αCrCr 1 . The
displayed in Fig. 9. The chemical ordering at 873 K and 1273 K can
obtained diagrams are insensitive to the value of the radius of
reach an equilibrium state within 1 h (see Supplementary Fig. S20).
the spherical crystalline grain as shown in Supplementary Fig. S18.
From Fig. 9, we can find that the distribution of chemical-domain
Considering the high-temperature isotherm annealing at 1173 K
size given by the kMC simulation shows a peak at around 5 Å and
and 1273 K [19,25], the diagrams indicate that both αCrCr 1 and
average sizes of 5.3 Å and 5.4 Å at 873 K and 1273 K, receptively.
dmax can quickly reach to an equilibrium state within seconds.
The size distribution detected in our isothermal annealing sim-
In isotherm annealing at low temperatures, such as 673 K and
ulation reasonably agrees with the experiment samples prepared
573 K, a CDS with SLCOD can be formed requiring a longer an-
by the isothermal annealing and the subsequent water-quenching
nealing time, about hours. The representative evolution evident in
[27]. Here, it should be noted that the ordering kinetics around
the kMC simulations of the CDS at 1073 K and 673 K is shown
1273 K is quite rapid (see Fig. 7). Thus, the difference in anneal-
in Fig. 8. From Fig. 8, we can see that the formed CDSs have the
ing time at such high temperatures should not lead to a meaning-
same superlattice-chemical-order as obtained in the atom-swap-
fully distinct degree of chemical order. Moreover, since the degree
MC-based MC/MD hybrid simulations. At 1073 K, the maximum
of chemical order formed at such a high temperature is quite lim-
domain size dmax reaches to equilibrium within one second, while
ited (see Fig. 7), further chemical order can be expected to develop
at 673 K, dmax continues to grow to about 19 Å within 20 s.
in the subsequent quenching process. The final degree of chemi-
The choice of the heat treatment process is crucial for the de-
cal order at room temperature strongly depends on the history of
gree of the formed chemical order, since the evolution of the va-
cooling rate in the quenching process.
cancy concentration depends only on time and temperature and
The effects of chemical order on the dislocation plasticity of
cannot be controlled arbitrarily. If a MEA sample has been placed
MEAs and HEAs manifest mainly in the impedance of disloca-
at room temperature for a long enough time (∼ hours), even if it
tion nucleation and slip, leading to sluggish dislocation dynam-
has an equilibrium vacancy concentration of 400 K at the begin-
ics [9,19,28,29]. Moreover, a CDS with SLCOD may offer an addi-
ning of isotherm annealing, the annealing time for the formation
tional contribution to alloy strengthening, since the 12 110 dislo-
of CDS with SLCOD at 673 K requires years (see Supplementary
cation on the {111} plane needs to cut the {100} or {110} Cr-rich
Fig. S19), because the vacancy concentration at 673 K and below is
layers when the dislocation glides. To study these effects, we per-

8
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

Fig. 8. Representative evolution of chemical-domain structure at 1073 K and 673 K as induced by vacancy diffusion in isotherm annealing immediately after cooling down
from 1500 K to the isotherm annealing temperature of 1073 K and 673 K, respectively.

Fig. 9. Cr/CoNi {113} unit chemical structure given by the kinetic Monte Carlo (kMC) simulation. (a) and (b) show probability distribution of the chemical-domain size of
the Cr/CoNi {113} unit at 873 K and 1273 K, respectively, given by the present kMC simulation and those given by the inverse Fast Fourier transform image (IFFT) and the
dark-field (DF) images in experiment [27]. The inset tables in (a) and (b) list the averaged domain sizes (d). (c) Representative chemical-domain structure obtained by kMC
at 873 K and 1273 K in those equilibrium states after 1 h annealing (see Supplementary Fig. S20). The atoms in each CSRO Cr/CoNi {113} unit are only shown and colored
according to the orientations of the two {113} planes in the unit chemical structure.

9
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

formed nanoindentation simulation and observed the delay in the Declaration of Competing Interest
first pop-in due to the existence of CDS with chemical domain
boundary [19] (see Section 7 in Supplementary Information). The authors declare that they have no known competing finan-
Meanwhile, It has been questioned whether the effect of chemi- cial interests or personal relationships that could have appeared to
cal order on mechanical properties is negligible. We speculate that influence the work reported in this paper.
the effect could strongly depend on the heat treatment protocol.
The CrCoNi MEA samples prepared by water quenching to room CRediT authorship contribution statement
temperature from annealed temperatures above 873 K do not show
a measurable yield strength variant [25,26]. In contrast, the sam- Jun-Ping Du: Conceptualization, Formal analysis, Methodology,
ples prepared by furnace cooling from annealed temperature above Writing – original draft, Writing – review & editing. Peijun Yu:
873 K to room temperature show both yield strength and hard- Methodology. Shuhei Shinzato: Methodology, Writing – review
ness enhancement [19]. From our T−T−chemical-order diagrams, & editing. Fan-Shun Meng: Methodology. Yuji Sato: Methodol-
we can deduce that the chemical order in the samples preserved ogy. Yangen Li: Methodology. Yiwen Fan: Methodology. Shigenobu
by quick water quenching was not well-developed and thus shows Ogata: Conceptualization, Formal analysis, Writing – original draft,
little effect on the mechanical properties. While in the samples Writing – review & editing.
preserved by furnace cooling more strong chemical order should
be developed during the relatively slow furnace cooling process,
Acknowledgments
especially at temperatures below 873 K, and thus may enhance
the yield strength and the hardness. Note that a very recent ex-
This work was supported by the Element Strategy Initia-
periment [58] found the degree of chemical order in the furnace-
tive for Structural Materials (ESISM) of MEXT, Grant Number JP-
cooled CrCoNi MEA is indeed promoted compared to that in the
MXP01121010 0 0. P.Y. and S.O. acknowledge the support by JSPS
water-quenched one. In addition, a wavy dislocation slip motion
KAKENHI Grant No. JP18H05453. S.O. acknowledges the support by
was found in the water-quenched one, while the planar dislocation
JSPS KAKENHI Grant Nos. JP17H01238 and JP17K18827. The calcula-
slip motion was found in the furnace-cooled one. However, a clear
tions were performed on the large-scale computer systems at the
difference in the yield strengths could not be found [58]. Possibly,
Cybermedia Center, Osaka University, and the Large-scale parallel
the degree of chemical order has not yet been higher enough to
computing server at the Center for Computational Materials Sci-
cause an increase in yield strength. Another possible reason is that
ence, Institute for Materials Research, Tohoku University.
planar slip implies the presence of a diffuse antiphase boundary;
although the dislocation slip through the CSRO structure generates
Supplementary material
resistance to the dislocation due to breaking the favorable solute-
solute bonds [59], the diffuse antiphase boundary reduces the re-
Supplementary material associated with this article can be
sistance to the dislocation in it due to the solute–dislocation inter-
found, in the online version, at doi:10.1016/j.actamat.2022.118314.
action [60,61], and these two effects on the yield strength cancel
each other out.
References
The T−T−chemical-order diagrams in the present study provide
the information to design heat treatment protocols for promoting [1] J.-W. Yeh, S.-K. Chen, S.-J. Lin, J.-Y. Gan, T.-S. Chin, C. T.-T. Shun, S. H. Tsau,
the chemical order in the CrCoNi MEA. For example, we can per- Y. Chang, Nanostructured high-entropy alloys with multiple principal ele-
form rapid cooling from higher temperature (like 1500 K) to 673 K ments: novel alloy design concepts and outcomes, Adv. Eng. Mater. 6 (5)
(2004) 299–303.
with a high cooling rate to maintain the high vacancy concentra-
[2] B. Cantor, I. Chang, P. Knight, A. Vincent, Microstructural development in
tion, followed by an isothermal annealing at 673 K for like a couple equiatomic multicomponent alloys, Mater. Sci. Eng. A 375 (2004) 213–218.
of hours, and then a furnace cooling to room temperature. Such [3] B. Gludovatz, A. Hohenwarter, D. Catoor, E.H. Chang, E.P. George, R.O. Ritchie,
A fracture-resistant high-entropy alloy for cryogenic applications, Science 345
heat treatment protocol may lead to a higher degree of chemical
(6201) (2014) 1153–1158.
order and more apparent effects on the strength. [4] B. Gludovatz, A. Hohenwarter, K.V. Thurston, H. Bei, Z. Wu, E.P. George,
R.O. Ritchie, Exceptional damage-tolerance of a medium-entropy alloy CrCoNi
at cryogenic temperatures, Nat. Commun. 7 (2016) 10602.
5. Conclusion [5] F. Granberg, K. Nordlund, M.W. Ullah, K. Jin, C. Lu, H. Bei, L. Wang,
F. Djurabekova, W. Weber, Y. Zhang, Mechanism of radiation damage reduc-
tion in equiatomic multicomponent single phase alloys, Phys. Rev. Lett. 116
In summary, we used a newly developed neural network po- (13) (2016) 135504.
tential based on DFT to find novel CDS structure at temperatures [6] Z. Zhang, H. Sheng, Z. Wang, B. Gludovatz, Z. Zhang, E.P. George, Q. Yu,
below 800 K in CrCoNi MEA. The CDS consists of two types of su- S.X. Mao, R.O. Ritchie, Dislocation mechanisms and 3D twin architectures gen-
erate exceptional strength-ductility-toughness combination in CrCoNi medi-
perlattices, such as Cr/CoNi {100} and Cr/CoNi {110}, with different um-entropy alloy, Nat. Commun. 8 (2017) 14390.
orientations, while at temperatures 800 K and above CSRO with [7] G. Laplanche, A. Kostka, C. Reinhart, J. Hunfeld, G. Eggeler, E. George, Reasons
different local structures, such as Cr/CoNi {100}, Cr/CoNi {110}, and for the superior mechanical properties of medium-entropy CrCoNi compared
to high-entropy CrMnFeCoNi, Acta Mater. 128 (2017) 292–303.
Cr/CoNi {113}, appeared. Moreover, we used a newly developed [8] S. Yoshida, T. Ikeuchi, T. Bhattacharjee, Y. Bai, A. Shibata, N. Tsuji, Effect of ele-
fast kMC computation technique based on an ANN and success- mental combination on friction stress and Hall-Petch relationship in face-cen-
fully drew time–temperature–chemical-order diagrams that predict tered cubic high/medium entropy alloys, Acta Mater. 171 (2019) 201–215.
[9] Q. Ding, Y. Zhang, X. Chen, X. Fu, D. Chen, S. Chen, L. Gu, F. Wei, H. Bei, Y. Gao,
the annealing-temperature-dependent time necessary for chemical
et al., Tuning element distribution, structure and properties by composition in
order formation. Such non-empirical prediction of CDS and its for- high-entropy alloys, Nature 574 (7777) (2019) 223–227.
mation kinetics opens new doors for the rigorous management of [10] Z. Li, S. Zhao, R.O. Ritchie, M.A. Meyers, Mechanical properties of high-entropy
alloys with emphasis on face-centered cubic alloys, Prog. Mater. Sci. 102 (2019)
chemical order and thus mechanical and functional properties of
296–345.
multicomponent alloys. [11] E.P. George, W. Curtin, C.C. Tasan, High entropy alloys: a focused review of
mechanical properties and deformation mechanisms, Acta Mater. 188 (2020)
435–474.
Data Availability [12] X. Yan, P.K. Liaw, Y. Zhang, Order and disorder in amorphous and high-entropy
materials, Metall. Mater. Trans. A 52 (6) (2021) 2111–2122.
[13] N. Hua, W. Wang, Q. Wang, Y. Ye, S. Lin, L. Zhang, Q. Guo, J. Brechtl, P.K. Liaw,
The data used in this work are available from the authors upon Mechanical, corrosion, and wear properties of biomedical Ti–Zr–Nb–Ta–Mo
request. high entropy alloys, J. Alloys Compd. 861 (2021) 157997.

10
J.-P. Du, P. Yu, S. Shinzato et al. Acta Materialia 240 (2022) 118314

[14] C. Lee, C. Chang, Y. Chen, J. Yeh, H. Shih, Effect of the aluminium content of [37] The neural network potential for CrCoNi medium-entropy alloy, https://sites.
Alx CrFe1.5 MnNi0.5 high-entropy alloys on the corrosion behaviour in aqueous google.com/view/nnp-crconi/.
environments, Corros. Sci. 50 (7) (2008) 2053–2060. [38] G. Kresse, J. Hafner, Ab initio molecular dynamics for liquid metals, Phys. Rev.
[15] M.A. Hemphill, T. Yuan, G. Wang, J. Yeh, C. Tsai, A. Chuang, P. Liaw, Fatigue B 47 (1) (1993) 558.
behavior of Al0.5 CoCrCuFeNi high entropy alloys, Acta Mater. 60 (16) (2012) [39] G. Kresse, J. Hafner, Ab initio molecular-dynamics simulation of the liq-
5723–5734. uid-metal–amorphous-semiconductor transition in germanium, Phys. Rev. B 49
[16] T.-N. Lam, S.Y. Lee, N.-T. Tsou, H.-S. Chou, B.-H. Lai, Y.-J. Chang, R. Feng, (20) (1994) 14251.
T. Kawasaki, S. Harjo, P.K. Liaw, Enhancement of fatigue resistance by over- [40] G. Kresse, J. Furthmüller, Efficient iterative schemes for ab initio total-energy
load-induced deformation twinning in a CoCrFeMnNi high-entropy alloy, Acta calculations using a plane-wave basis set, Phys. Rev. B 54 (16) (1996) 11169.
Mater. 201 (2020) 412–424. [41] S. Plimpton, Fast parallel algorithms for short-range molecular dynamics, J.
[17] M.-H. Chuang, M.-H. Tsai, W.-R. Wang, S.-J. Lin, J.W. Yeh, Microstructure and Comput. Phys. 117 (1) (1995) 1–19.
wear behavior of Alx Co1.5 CrFeNi1.5 Tiy high-entropy alloys, Acta Mater. 59 (16) [42] J. Cowley, An approximate theory of order in alloys, Phys. Rev. 77 (5) (1950)
(2011) 6308–6317. 669.
[18] X. Yan, Y. Zhang, Functional properties and promising applications of high en- [43] Z. Shen, J.-P. Du, S. Shinzato, Y. Sato, P. Yu, S. Ogata, Kinetic Monte Carlo simu-
tropy alloys, Scr. Mater. 187 (2020) 188–193. lation framework for chemical short-range order formation kinetics in a mul-
[19] R. Zhang, S. Zhao, J. Ding, Y. Chong, T. Jia, C. Ophus, M. Asta, R.O. Ritchie, ti-principal-element alloy, Comput. Mater. Sci 198 (2021) 110670.
A.M. Minor, Short-range order and its impact on the CrCoNi medium-entropy [44] Y.N. Osetsky, L.K. Béland, A.V. Barashev, Y. Zhang, On the existence and origin
alloy, Nature 581 (7808) (2020) 283–287. of sluggish diffusion in chemically disordered concentrated alloys, Curr. Opin.
[20] F. Zhang, S. Zhao, K. Jin, H. Xue, G. Velisa, H. Bei, R. Huang, J. Ko, D. Pagan, Solid State Mater. Sci. 22 (3) (2018) 65–74.
J. Neuefeind, Local structure and short-range order in a NiCoCr solid solution [45] Y. Song, R. Malek, N. Mousseau, Optimal activation and diffusion paths of per-
alloy, Phys. Rev. Lett. 118 (20) (2017) 205501. fect events in amorphous silicon, Phys. Rev. B 62 (23) (20 0 0) 15680.
[21] J. Ding, Q. Yu, M. Asta, R.O. Ritchie, Tunable stacking fault energies by tailoring [46] G. Henkelman, B.P. Uberuaga, H. Jónsson, A climbing image nudged elastic
local chemical order in CrCoNi medium-entropy alloys, Proc. Natl. Acad. Sci. band method for finding saddle points and minimum energy paths, J. Chem.
115 (36) (2018) 8919–8924. Phys. 113 (22) (20 0 0) 9901–9904.
[22] A. Tamm, A. Aabloo, M. Klintenberg, M. Stocks, A. Caro, Atomic-scale properties [47] G. Henkelman, H. Jónsson, Improved tangent estimate in the nudged elastic
of Ni-based FCC ternary, and quaternary alloys, Acta Mater. 99 (2015) 307–312. band method for finding minimum energy paths and saddle points, J. Chem.
[23] Z. Pei, R. Li, M.C. Gao, G.M. Stocks, Statistics of the NiCoCr medium-entropy Phys. 113 (22) (20 0 0) 9978–9985.
alloy: novel aspects of an old puzzle, npj Comput. Mater. 6 (2020) 122. [48] S. Zhao, T. Egami, G.M. Stocks, Y. Zhang, Effect of d electrons on defect prop-
[24] F. Walsh, M. Asta, R.O. Ritchie, Magnetically driven short-range order can erties in equiatomic NiCoCr and NiCoFeCr concentrated solid solution alloys,
explain anomalous measurements in CrCoNi, Proc. Natl. Acad. Sci. 118 (13) Phys. Rev. Mater. 2 (1) (2018) 013602.
(2021). [49] H. Mehrer, Diffusion in Solids: Fundamentals, Methods, Materials, Diffusion–
[25] B. Yin, S. Yoshida, N. Tsuji, W. Curtin, Yield strength and misfit volumes of Controlled Processes, Vol. 155, Springer Science & Business Media, 2007.
NiCoCr and implications for short-range-order, Nat. Commun. 11 (2020) 2507. [50] M. Mantina, Y. Wang, R. Arroyave, S. Shang, L. Chen, Z. Liu, A first-principles
[26] K. Inoue, S. Yoshida, N. Tsuji, Direct observation of local chemical ordering in a approach to transition states of diffusion, J. Phys. 24 (30) (2012) 305402.
few nanometer range in CoCrNi medium-entropy alloy by atom probe tomog- [51] W. Steurer, J. Dshemuchadse, Intermetallics: Structures, Properties, and Statis-
raphy and its impact on mechanical properties, Phys. Rev. Mater. 5 (8) (2021) tics, Vol. 26, Oxford University Press, 2016.
085007. [52] K. Jin, S. Mu, K. An, W.D. Porter, G.D. Samolyuk, G.M. Stocks, H. Bei, Thermo-
[27] L. Zhou, Q. Wang, J. Wang, X. Chen, P. Jiang, H. Zhou, F. Yuan, X. Wu, Z. Cheng, physical properties of Ni-containing single-phase concentrated solid solution
E. Ma, Atomic-scale evidence of chemical short-range order in CrCoNi medi- alloys, Mater. Des. 117 (2017) 185–192.
um-entropy alloy, Acta Mater. 224 (2022) 117490. [53] J.P. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made
[28] X. Chen, Q. Wang, Z. Cheng, M. Zhu, H. Zhou, P. Jiang, L. Zhou, Q. Xue, F. Yuan, simple, Phys. Rev. Lett. 77 (18) (1996) 3865.
J. Zhu, et al., Direct observation of chemical short-range order in a medium-en- [54] Z. Wang, C. Liu, P. Dou, Thermodynamics of vacancies and clusters in high-en-
tropy alloy, Nature 592 (7856) (2021) 712–716. tropy alloys, Phys. Rev. Mater. 1 (4) (2017) 043601.
[29] Q.-J. Li, H. Sheng, E. Ma, Strengthening in multi-principal element alloys with [55] H. Guan, S. Huang, J. Ding, F. Tian, Q. Xu, J. Zhao, Chemical environment and
local-chemical-order roughened dislocation pathways, Nat. Commun. 10 (2019) magnetic moment effects on point defect formations in CoCrNi-based concen-
3563. trated solid-solution alloys, Acta Mater. 187 (2020) 122–134.
[30] V.L. Deringer, M.A. Caro, G. Csányi, Machine learning interatomic potentials as [56] A.A. Belak, A.V.d. Ven, Effect of disorder on the dilute equilibrium vacancy con-
emerging tools for materials science, Adv. Mater. 31 (46) (2019) 1902765. centrations of multicomponent crystalline solids, Phys. Rev. B 91 (22) (2015)
[31] J. Behler, M. Parrinello, Generalized neural-network representation of high-di- 224109.
mensional potential-energy surfaces, Phys. Rev. Lett. 98 (14) (2007) 146401. [57] X. Zhang, S.V. Divinski, B. Grabowski, Ab initio prediction of vacancy energetics
[32] A.P. Bartók, M.C. Payne, R. Kondor, G. Csányi, Gaussian approximation poten- in HCP Al-Hf-Sc-Ti-Zr high entropy alloys and the subsystems, Acta Mater. 227
tials: the accuracy of quantum mechanics, without the electrons, Phys. Rev. (2022) 117677.
Lett. 104 (13) (2010) 136403. [58] H. Hsiao, R. Feng, H. Ni, K. An, J. Poplawsky, P. Liaw, J. Zuo, Chemical short-
[33] J. Behler, G. Csányi, Machine learning potentials for extended systems: a per- range ordering in a CrCoNi medium-entropy alloy, (2022). arXiv preprint arXiv:
spective, Eur. Phys. J. B 94 (7) (2021) 1–11. 2206.02004.
[34] J. Behler, Four generations of high-dimensional neural network potentials, [59] E. Antillon, C. Woodward, S. Rao, B. Akdim, T. Parthasarathy, Chemical short
Chem. Rev. 121 (2021) 10037–10072. range order strengthening in a model FCC high entropy alloy, Acta Mater. 190
[35] P. Yu, J.-P. Du, S. Shinzato, F.-S. Meng, S. Ogata, Theory of history-dependent (2020) 29–42.
multi-layer generalized stacking fault energy–a modeling of the micro-sub- [60] S. Patu, R. Arsenault, Strengthening due to non-random solid solutions, Mater.
structure evolution kinetics in chemically ordered medium-entropy alloys, Acta Sci. Eng. A 194 (2) (1995) 121–128.
Mater. 224 (2022) 117504. [61] A. Abu-Odeh, D.L. Olmsted, M. Asta, Screw dislocation mobility in a face–
[36] A. Singraber, T. Morawietz, J. Behler, C. Dellago, Parallel multistream training centered cubic solid solution with short-range order, Scr. Mater. 210 (2022)
of high-dimensional neural network potentials, J. Chem. Theory Comput. 15 (5) 114465.
(2019) 3075–3092.

11

You might also like