Structural optimization and segregation behavior of quaternary alloy nanoparticles based on simulated annealing algorithm
Lu Xin-Ze1, Shao Gui-Fang2, †, , Xu Liang-You2, Liu Tun-Dong2, Wen Yu-Hua1
Department of Physics, Xiamen University, Xiamen 361005, China
Department of Automation, Xiamen University, Xiamen 361005, China

 

† Corresponding author. E-mail: gfshao@xmu.edu.cn

Project supported by the National Natural Science Foundation of China (Grant Nos. 51271156, 11474234, and 61403318) and the Natural Science Foundation of Fujian Province of China (Grant Nos. 2013J01255 and 2013J06002).

Abstract
Abstract

Alloy nanoparticles exhibit higher catalytic activity than monometallic nanoparticles, and their stable structures are of importance to their applications. We employ the simulated annealing algorithm to systematically explore the stable structure and segregation behavior of tetrahexahedral Pt–Pd–Cu–Au quaternary alloy nanoparticles. Three alloy nanoparticles consisting of 443 atoms, 1417 atoms, and 3285 atoms are considered and compared. The preferred positions of atoms in the nanoparticles are analyzed. The simulation results reveal that Cu and Au atoms tend to occupy the surface, Pt atoms preferentially occupy the middle layers, and Pd atoms tend to segregate to the inner layers. Furthermore, Au atoms present stronger surface segregation than Cu ones. This study provides a fundamental understanding on the structural features and segregation phenomena of multi-metallic nanoparticles.

1. Introduction

Metallic nanoparticles (NPs) are extensively used in chemical industry, petroleum industry, new energy, and environmental protection (such as air purification and sewage treatment).[1,2] Among these NPs, noble metal NPs have attracted much attention due to their excellent catalytic activity and good stability. For example, the available experimental and theoretical studies reveal that platinum (Pt) NPs are excellent and versatile catalysts in various important reactions. In particular, bi- or multi-metallic NPs can make full use of the synergy effects of crystal structures and electronic structures of these metals when Pt alloys with other kinds of metals. Accordingly, the alloy NPs may exhibit higher catalytic activity and selectivity compared with the monometallic ones.[3]

In recent years, most research has focused on bimetallic NPs, such as Pt–Au, Pt–Pd, and Pt–Cu.[48] Nowadays, trimetallic NPs also receive more and more attention.[914] However, the quaternary alloy NPs have rarely been studied. The available research only focuses on the experimental preparation of the NPs rather than their structures and stability.[1517] In the study of the structural stability of NPs, the fundamental method is to calculate the total energy of the system and thereby to predict its stable structures. However, the potential energy surface of NPs is a three-dimensional curved surface. Therefore, the essence of the NP structure optimization is to search the global minimum point (corresponding to the lowest-energy structure) on the three-dimensional energy surface. In addition, the potential energy surface in the three-dimensional space is rugged, and the locally optimal number of potential energy functions grows rapidly as the NP size grows. Thus, the structural optimization of NPs actually belongs to a non-deterministic polynomial problem. The most effective way to solve this problem is to develop a more efficient and more precise global optimization algorithm.

Currently, several theoretical methods, such as molecular dynamic methods, Monte Carlo methods, and density functional theory (DFT) calculations, have been used to explore the stable structures of alloy NPs. The traditional Monte Carlo methods easily drop into a local optimum in the structural optimization of NPs,[1822] therefore it is difficult to find the global optimum. To effectively predict the structural properties of NPs, more intelligent evolutionary algorithms such as the genetic algorithm and the particle swarm optimization (PSO) method[2326] have been introduced. For example, the genetic algorithms have been used to study the stable structures of Pd–Au bimetallic NPs.[2325] Chen proposed a spherical cut-cross genetic evolution to predict the clusters’ structure.[26] In our previous study, the Monte Carlo methods were introduced to predict the stable structure of Pt–Pd bimetallic NPs by changing the exchange sequence.[27] However, there are some uncertainties, such as the operator selection and parameter setting, existing in these random intelligent optimization methods. It is crucial to search for high efficiency algorithms for the structural optimization of NPs.

In this paper, we will employ the simulated annealing (SA) algorithm to explore the stable structure of Pt–Pd–Cu–Au alloy NPs. The surface segregation behavior of the NPs is investigated by analyzing the atomic distribution, the radial distribution function, and the distribution of atomic coordination number. Further, we discuss the relationship between the atomic distributions and the compositions of the NPs.

2. Methods
2.1. The Q-SC many-body potentials

In this paper, the interatomic interaction in Pt–Pd–Cu–Au quaternary alloy NPs is described by the quantum corrected Sutton–Chen (Q-SC) many-body potentials. According to the Q-SC force field, the total potential energy of the system is expressed as

where Rij is the distance between atoms i and j, ɛ is the energy scale factor, V(Rij) is a pair interaction function defined by

and ρi is a local electron density accounting for cohesion associated with atom i defined by the following equation:

In Eqs. (1)–(3), a is the length parameter, c is a dimensionless parameter, and n and m are integer parameters that satisfy the constraint n > m.[2830] The model parameters for Pt, Pd, Cu, and Au are listed in Table 1.

Table 1.

Potential parameters for Pt–Pd–Cu–Au NPs.[31]

.

For describing the interatomic interaction between different metals, the geometric mean is used to obtain the energy parameter ɛ, while the arithmetic mean is used for the remaining parameters.[5,27]

2.2. The initial configuration

To conduct the atomistic simulation, tetrahexahedral (THH) NPs,[32] covered by high-index facets, are chosen as the initial atomic configuration. Firstly, a large face-centered cubic (fcc) single crystal is created. Subsequently, a square pyramid is used to cap this cube on each face along the direction of the crystal. Finally, THH NPs bounded by 24 {210} facets are obtained. Furthermore, Pt and other metallic atoms (Pd, Cu, and Au) are randomly distributed in the THH NPs by computer-produced random seeds. Figure 1 shows the illustration of THH NPs.

Fig. 1. Schematic illustration of THH NPs.
2.3. The optimization model

Generally, the most stable structure of NPs corresponds to the lowest-energy structure. Therefore, we can transfer the structural optimization of NPs into searching for their lowest-energy structures.

For Pt–Pd–Cu–Au quaternary alloy NP consisting of N atoms, let w0, w1, w2, and w3 denote the proportions of the four types of atoms (element 1, element 2, element 3, and element 4, respectively) and w0+w1+w2+w3 = 1. Subsequently, arabic numerals are used to code different types of atoms, where 0, 1, 2, and 3 represent element 1, element 2, element 3, and element 4, respectively. Therefore, we can define a combinational optimization model for Pt–Pd–Cu–Au NPs as follows:

The parameters m, n, a, and ɛ are calculated according to the following method. Firstly, let λ be the type of atom i and μ be the type of atom j (λ = 0,1,2,3 and μ = 0,1,2,3). Let an identity matrix be

A symmetric matrix is given by

Then we can describe the interaction between atom i and atom j as which can be alternately calculated as follows:

The ci is determined by the equilibrium lattice parameter.

2.4. The SA algorithm

The earliest idea of SA algorithm was proposed by Metropolis et al. in 1953.[33] In 1983, Kirkpatrick and co-authors successfully introduced the annealing thought to the combinatorial optimization field.[34] Annealing thought is a kind of stochastic optimization algorithm based on the Monte–Carlo iterative solution strategy. The starting point for annealing thought is based on the similarity between the annealing processes of a material in physics and the general combinatorial optimization problem. The SA algorithm starts from a high initial temperature. As the temperature decreases, it may search for the objective function’s global optimum in solution space. So far, the SA algorithm has achieved great success in combinatorial optimization, production scheduling, machine learning, circuit design, neural network, and other fields.

2.5. Coding

The structural optimization of NPs belongs to a discrete problem. Therefore, we use an integer coding method. For an NP with N atoms, let w1, w2, w3, w4 denote the proportions of four types of atoms, respectively, and w1+w2+w3+w4=1. Use numbers 0, 1, 2, and 3 to represent the Pt, Pd, Cu, and Au atoms, respectively. All the coordinates of atoms in the NP are numbered from 1 to N. The coding scheme is defined as follows:

The corresponding numbers of the four types of atoms are p0 = N × w0, p1 = N × w1, p2 = N × w2, and p3 = N × w3. Each sequence of 0, 2, 2, and 3 represents a solution of the problem in the solution space, and it also represents a structure of the Pt–Pd–Cu–Au quaternary NP. To construct an initial feasible solution, various atoms are randomly distributed in the N coordinate space subject to the above constraints.

2.6. Neighborhood searching

Through continuous searching iteration, a neighborhood searching makes further adjustments to the feasible solution so as to approach an optimal solution. In this paper, we use the SA algorithm for neighborhood construction. With a feasible solution with N atoms, we randomly exchange two atoms’ positions to generate new solutions and calculate the corresponding fitness. To avoid useless searches, two exchanged atoms must be different types. As shown below, atom 0 and atom 2 exchange will result in a new explanation, and this exchange is effective. However, when atom 3 exchanges with another atom 3, there is no change to the current solution, so this exchange is invalid.

There are ρ (ρ = ρ0 ρ1 ρ2 ρ3) solutions for each state. As the particle size increases, the neighborhood solutions grow exponentially. However, we change only a pair of atoms each time. During the late stage of the search, the optimization speed is slow. Therefore, for the larger size NP, we make the following improvements.

In the SA search process, according to the search generation and cluster size, we determine a sequence position parameter for each search. Early in the search, to accelerate the optimization progress, each of the variable parameters is assigned a larger progressively decreasing constant K. To obtain the optimal result, iterations are terminated when K reaches 1. In this paper, the setting of K is as follows:

where No is the optimal size of the clusters (i.e., the number of atoms in the clusters), imax is the largest search generation, and i is the current search generation.

2.7. Judgement

In the SA algorithm, we choose the Q-SC potential function as the fitness function. When the algorithm runs, the current solution is updated if we find a lower energy than the energy of the current solution. If the mutated energy is higher than the current solution’s energy, the new solution is accepted based on a probability calculated as follows:

where ΔE is the particle energy difference between the new solution and the current solution; k0 is the temperate change rate (generally between 0–1); and T is the current temperature. During the annealing treatment, as the temperature decreases, the probability of accepting a worse solution gradually becomes less and less. Let L be the iteration of each temperature and Tmin be the operation end temperature. Through repeated iterations, a new solution is generated, the energy difference is calculated. Furthermore, the resulting solution is accepted or discarded, and T is gradually reduced. When T is reduced to the value which satisfies the operator termination criterion, T < Tmin, the annealing operation ends. At the same time, the current solution is accepted as the optimization solution. The SA algorithm’s flow chart is shown in Fig. 2.

Fig. 2. Flowchart of SA.
3. Results and discussion
3.1. Stable structure

In order to verify the efficiency of the improved SA algorithm, the stable structures of THH Pt–Pd–Cu–Au quaternary alloy NPs with 443 atoms and 1417 atoms are studied. Figures 3(a) and 3(b) show the lowest-energy configurations of Pt–Pd–Cu–Au quaternary alloy NPs obtained by the SA algorithm. The Pt–Pd–Cu–Au quaternary atomic ratios of 0.25:0.25:0.25:0.25 are selectively illustrated in this figure to simplify the discussion. Meanwhile, the experimental results illustrate that Au and Cu tend to distribute on the surface, Pt tends to occupy the middle layers, and Pd tends to occupy the inner layers.

This stable structure of Pt–Pd–Cu–Au NPs is caused by the balance of different energies. Table 2 shows the properties of Pt, Pd, Cu, and Au. Owing to the difference of cohesive energy and surface energy, Au and Cu tend to segregate on the surface, Pt tends to distribute in the middle layers, and Pd preferentially distributes in the inner layers. This segregation will produce a lower total energy.

Table 2.

Materials’ properties calculated with the Q-SC potentials.[31]

.

Furthermore, we calculate the pair correlation function of Pt–Pd–Cu–Au quaternary alloy NPs, N(r). The pair correlation function N(r) is defined as the counted atom number at distance r from the center of mass. For two atom sizes (443 atoms and 1417 atoms), the atomic distribution tendencies are almost the same. Figures 3(c) and 3(d) illustrate that Au and Cu preferentially segregate on the surface, Pt tends to occupy the middle layers, while Pd tends to occupy the inner layers.

Fig. 3. Atomic configurations of Pt–Pd–Cu–Au quaternary alloy NPs with (a) 443 atoms and (b) 1417 atoms. For each pair of columns, the first and the second columns represent the whole and the cross-section atomic arrangements of the NPs, respectively. Pt atoms are in orange, Pd atoms in red, Cu atoms in gray, and Au atoms in yellow. Panels (c) and (d) show the corresponding pair correlation functions.
3.2. Core-shell distribution

The interior distribution of the Pt–Pd–Cu–Au quaternary alloy NPs with atomic ratio of 0.25:0.25:0.25:0.25 is also explored. As shown in Fig. 4, the alloy NPs are divided into several layers (a layer is also called a shell) according to the distance from the core to the atom. Evidently, there are five, seven, and nine layers in the NPs with 443 atoms, 1417 atoms, and 3285 atoms, respectively. These layers (shells) are numbered according to their distances to the center. Figure 4 shows that Au and Cu atoms tend to distribute at the outermost layers, Pt atoms tend to occupy the middle layers, and Pd atoms preferentially locate in the inner layers.

Fig. 4. Calculated atoms’ ratios in each shell: (a) 443 atoms, (b) 1417 atoms, (c) 3285 atoms.
3.3. Surface segregation analysis

Noble metal NPs have broad applications in chemical and petroleum industries due to their excellent catalytic performances. Generally, chemical reactions take place on the surface, and the distribution of the atoms on the surface of multi-alloy NPs is always related to the chemical properties of the component elements. Therefore, it is vital to study Pt–Pd–Cu–Au quaternary alloy NPs’ surface atomic distributions.[35]

According to Figs. 3 and 4, we have learned that Cu and Au atoms preferentially segregate to the surface. To further analyze the segregation phenomena, two cases are investigated: (i) common segregation, and (ii) competitive segregation. At the same time, we selected two NPs of 443 atoms and 1417 atoms for comparison.

Figure 5 shows Pt–Pd–Cu–Au quaternary alloy NPs’ surface segregation curves when both Cu and Au compositions change from 0 to 50%. The segregation degree is measured by the surface atomic fraction, which is defined as C = NA/NS (A is Cu or Au), where NA is the number of surface A atoms and NS is the number of surface atoms. In Figs. 5(a) and 5(b), the left black dotted lines represent the max segregation, while the others denote no segregation. In Fig. 5, we can see that for small amounts of Cu and Au, the Cu and Au atoms tend to segregate to the surface, and reach the max segregation. We define this segregation as the common segregation. When the amounts of Cu and Au grow to the surface-to-volume ratio, they maintain the max segregation. With the compositions further growing, the surface Cu fraction begins to drop. We call this behavior the competitive segregation. This phenomenon is related to the lower surface energy of Au. By comparing Fig. 5(a) with Fig. 5(b), we find that the segregation degree will increase with the atomic size increasing. Herein, the surface segregation degree of Au is stronger than that of Cu with the same amounts of Cu and Au atoms. This effect is also associated with the lower surface energy of Au.

Fig. 5. Surface atomic fractions of Cu and Au with the Cu (Au) composition changing from 0 to 50% at particle sizes of (a) 443 atoms and (b) 1417 atoms.
3.4. Coordinated sites analysis

Furthermore, we investigate the preferred sites of different metal atoms. We select Pt–Pd–Cu–Au quaternary alloy NPs of 1417 atoms and analyze the preferred sites of different metal atoms (Pt, Pd, Cu and Au) by the distribution of atoms at different positions. Firstly, we change the atomic ratio of one metal; secondly, the atomic ratios of the other three metals change with the first one, and these three metals have the same atomic ratio. Figure 6 shows the concentrations of Pt and Pd atoms at different positions (different-coordinated sites), and figure 7 shows the concentrations of Cu and Au atoms at different positions (different-coordinated sites).

Fig. 6. Pt, Pd, Cu, and Au concentrations at different coordination sites as a function of composition of (a)–(d) Pt or (e)–(h) Pd.
Fig. 7. Pt, Pd, Cu, and Au concentrations at different coordination sites as a function of composition of (a)–(d) Cu or (e)–(h) Au.

As shown in Figs. 6(a)6(d), the atomic ratio of Pt increases and the atomic ratios of the other metals decrease with the increasing atomic number of Pt. It is found that the number of Pd atoms decreases most rapidly in the high coordination sites, the number of Au atoms decreases most rapidly in the middle coordination sites, whereas the number of Cu atoms decreases most rapidly in the low coordination sites.

As shown in Figs. 6(e)6(h), the atomic ratio of Pd increases and the ratios of the other metals decrease with the increasing atomic number of Pd. It is found that, the number of Pt atoms decreases most rapidly in the high coordination sites, the number of Au atoms decreases most rapidly in the middle coordination sites, whereas the number of Cu atoms decreases most rapidly in the low coordination sites.

As shown in Figs. 7(a)7(d), the atomic ratio of Cu increases and the atomic ratios of the other metals decrease with the increasing atomic number of Cu. It is found that, the number of Pt atoms decreases most rapidly in the high coordination sites, the number of Au atoms decreases most rapidly in the middle coordination sites, whereas the number of Pd atoms decreases most rapidly in the low coordination sites.

As shown in Figs. 7(e)7(h), the atomic ratio of Au increases and the ratios of the other metals decrease with the increasing atomic number of Au. It is found that, the number of Pt atoms decreases most rapidly in the high coordination sites, the number of Pd atoms decreases most rapidly in the middle coordination sites, whereas the number of Cu atoms decreases most rapidly in the low coordination sites.

By comparing Fig. 6 with Fig. 7, we can find some interesting phenomena. The number of Pt atoms with the coordination numbers of 9 and 10 decreases slower than that of Pt atoms with the coordination numbers of 11 and 12. By comparison of the results in Figs. 3(c) and 3(d), we can find that the Pd atoms preferentially occupy high coordination sites (i.e., 11-fold or 12-fold coordination sites). The Pt atoms prefer to occupy the low coordination sites (9-fold or 10-fold coordination sites), the Au atoms tend to locate at middle coordination positions (6-fold or 8-fold coordination sites), and the Cu atoms tend to locate at the lowest coordination sites (3-fold or 5-fold coordination sites). The analysis of coordination number indicates that Au and Cu prefer to occupy the surface, Pt prefers to occupy the middle layers, while Pd tends to distribute in the inner layers.

4. Conclusion

We investigated the structural features and the segregation behavior of THH Pt–Pd–Cu–Au quaternary alloy NPs by using the SA algorithm. Three NPs consisting of 443 atoms, 1417 atoms, and 3285 atoms were selected to display the stable structures of the NPs. The simulated results show that Au and Cu atoms preferentially segregate to the surface, Pt atoms tend to occupy the middle layers, and Pd atoms segregate to the inner layers. With the growing particle size, the distribution of Cu atoms changes slightly. The distribution of atomic coordination number reveals that Au and Cu atoms always decrease most slowly in low coordination sites, Pd atoms decrease most slowly in high coordination sites, and Pt atoms decrease most slowly in middle coordination sites. These results are of importance not only for the experimental preparation of noble metal alloy NPs but also for further exploring the multifunctional properties of alloy NPs. Our study also confirms that the developed SA algorithm is suitable for predicting the stable structure of multimetallic NPs.

Reference
1Murray R W 2008 Chem. Rev. 108 2688
2Zhou Z YTian NLi J TBroadwell ISun SG 2011 Chem. Soc. Rev. 40 4167
3Wu W KZhang L NRen H RZhang KLi HHe Y Z 2015 Phys. Chem. Chem. Phys. 17 13380
4Deng LHu W YDeng H QXiao S F 2010 J. Phys. Chem. 114 11026
5Huang RWen Y HZhu Z ZSun S G 2012 J. Phys. Chem. 116 8664
6Yun KCho Y HCha P RLee JNam H SOh J SChoi J HLee S C 2012 Acta Mater. 60 4908
7Shao G FZheng W XTu Na NaLiu T DWen Y H 2015 Acta Phys. Sin. 64 013602 (in Chinese)
8Gao X XJia Y HLi G PCho S JKim H 2011 Chin. Phys. Lett. 28 033601
9Chen M SCai YYan ZGath K KAxnanda SWayne G D 2007 Surf. Sci. 601 5326
10Pal UGarcia-Serrano JCasarrubias-Segura GKoshizaki NSasaki TTerahuchi S 2004 Sol. Energ. Mat Sol. 81 339
11Sekhon J SVerma S S 2011 Plasmonics 6 311
12Wang LYamauchi Y 2010 J. Am. Chem. Soc. 132 13636
13Guo S JZhang SSun X LSun S H 2011 J. Am. Chem. Soc. 133 15354
14Zhu L SZhao S J 2014 Chin. Phys. 23 063601
15Lumley R NMorton A JPolmear I J 2002 Acta Mater. 50 3597
16Cao PZhang M LHan WYan Y DChen L J 2013 T. Nonferr. Metal. Soc. 23 861
17Mazumder VChi M FMore K LSun S H 2010 Angew. Chem. Int. Edit. 49 9368
18Iglesias OLabarta A 2001 Phys. Rev. 63 184416
19Qin L JZhang Y HHuang S PTian H PWang P 2010 Phys. Rev. 82 075413
20Yuge K 2010 J. Phys.: Condens. Matter 22 245401
21Cho S H 2005 Phys. Med. Biol. 50 N163
22Guo J YXu C XHu A MOakes K DSheng F Y 2012 J. Phys. Chem. Solids 73 1350
23Radillo-Diaz ACoronado YPéreza L AGarzón I L 2009 Eur. Phys. J. 52 127
24Paz-Borbo’n L OGuptab AJohnston R L 2008 J. Mater. Chem. 18 4154
25Bruma AIsmail RPaz-Borbón L OArslan HBarcaro GFortunelli ALia Z YJohnston R L 2013 Nanoscale 5 646
26Yu J MChen Z HNi YLi Z X 2012 Hum. Reprod. 27 25
27Liu T DZheng J WShao GFan T EWen Y H 2015 Chin. Phys. 24 033601
28Wen Y HZhang YZheng J CZhu Z ZSun S G 2009 J. Phys. Chem. 113 20611
29Ikeda HQi YCagin TSamwer KJohnson W LGoddard W A 1999 Phys. Rev. Lett. 82 2900
30Sankaranarayanan S K R SBhethanabotla V RJoseph B 2005 Phys. Rev. 71 195415
31Kimura YQi YCagin TGoddard III W1998Technical ReportVol. 3PasadenaCaltech ASCI1291–29
32Tian NZhou Z YSun S GDing YWang Z L 2007 Science 316 732
33Metropolis NRosenbluth A WRosenbluth M NTeller A H 1953 J. Chem. Phys. 21 1087
34Kirkpatrick SGelatt C DJr.Vecchi M P 1983 Science 220 671
35Li Z WKong X SLiu WLiu C SFang Q F 2014 Chin. Phys. 23 106107