J Mater Sci (2013) 48:4438–4445 DOI 10.1007/s10853-013-7262-4

Structure of liquid Cu–Sb alloys by ab initio molecular dynamics simulations, high temperature X-ray diffraction, and resistivity Fengxiang Guo • Yu Tian • Jingyu Qin • Rongfu Xu • Yong Zhang • Hongliang Zheng • Ting Lv • Xubo Qin • Xuelei Tian • Yucheng Sun

Received: 3 December 2012 / Accepted: 21 February 2013 / Published online: 6 March 2013 Ó Springer Science+Business Media New York 2013

Abstract Structure of Cu–Sb melts has been studied by ab initio molecular dynamics simulations, high-temperature X-ray diffraction and resistivity measurements. Over the whole concentration range, heterogeneous coordination numbers are larger than that of homogeneous atoms. This indicates preferential Cu–Sb coordination in Cu–Sb melts. A drop is observed in maximum position of simulated Sb–Sb partial distribution functions around Cu75Sb25, which reveals the rapid increase of Sb–Sb coordination. Around eutectic melts, main peak splitting is observed in both structure factor and simulated total pair distribution functions, which reveals the co-existence of Cu–Sb heterogeneous and Sb–Sb clusters. Abnormal changes in temperature coefficient of resistivity are observed around pure Sb and in compound-forming range, which are well interpreted as reinforcement of Peierls distortion and Cu3Sb compound clusters, respectively. Structural inhomogeneity that results from atomic size effect also has been discussed by analyzing

Fengxiang Guo and Yu Tian contributed equally to this study. F. Guo  J. Qin  R. Xu  Y. Zhang  H. Zheng  T. Lv  X. Qin  X. Tian (&) Key Laboratory for Liquid–Solid Structural Evolution and Processing of Materials (Ministry of Education), School of Materials Science and Engineering, Shandong University, Jingshi Road 17923, Jinan 250061, People’s Republic of China e-mail: [email protected] Y. Tian School of Physics, Shandong University, Jinan, Shandong 250100, People’s Republic of China Y. Sun Weichai Power Company Limited, Weifang 261205, People’s Republic of China

123

concentration dependence of Warren–Cowley parameters and concentration correlation functions. Abbreviations PDF, g(r) Pair distribution function S(q) Structure factor AIMD Ab initio molecular dynamics simulation SRO Short-range order LLST Liquid–liquid structure transitions CN Coordination numbers TCR Temperature coefficient of resistivity CSRO Chemical short-range order CCF Concentration correlation function PCN Partial coordination number

Introduction Local structure in metallic melts tended to show short-range order (SRO) without long-range periodicity [1]. Recent reports indicated that such inhomogeneity also took an important role in understanding formation of metallic glass [2]. It had been well established that, studying local structure and clusters evolution was not only mandatory to probe into atomic re-arrangements during nucleation and crystallization process, but also essential to control microstructure and thus improve materials macro-performance [3, 4]. Cu–Sb binary liquid alloys received active interests both from scientific and technological perspectives. From a fundamental scientific perspective, liquid Cu–Sb alloys exhibited interesting properties such as anomalies in viscosity and thermodynamics [5, 6], indicating their compositional inhomogeneity. Steeb [7] studied temperature dependence of electrical resistivity in liquid Cu–Sb alloys with an electrode–less method, and concluded the

J Mater Sci (2013) 48:4438–4445

4439

existence of Cu3Sb compound clusters in the melts. Besides these indirect measurements, direct structure studies on Cu–Sb melts were performed by X-ray diffraction. Based on the assumption of concentration independency, Knoll [8] calculated partial structure functions and believed that atoms deviated from statistical distribution in some Cu–Sb melts. Experimental structure factor S(q) and pair distribution functions (PDFs) of Cu–Sb liquid alloys revealed that, segregation into clusters corresponding to associations of Sb atoms and associations of unlike atoms existed in eutectic Cu37Sb63 alloy [9]. This was interpreted as that solid–phase structure began to form before crystallization. The existence of Cu3Sb clusters in Cu80Sb20 was verified in terms of structure relationship between liquid and solid phase [10]. As important elements of leadfree solders, Cu–Sb components had an important influence on interfacial growth kinetics, and thus determined thickness and grain size of intermetallic compounds layer in soldering process [11, 12]. However, unlike other noble– polyvalent metal systems [13, 14], few data for atomic arrangements in Cu–Sb melts existed, which made the clusters evolution unclear and limited their application. Recently, reversible temperature-induced liquid–liquid structure transitions (LLST) [15] were reported in Sb–10 wt% Cu melts [16], which increased nucleation undercooling and refined grains in as-solidified microstructure. In eutectic melts, it is very interesting that both irreversible and reversible LLST existed for different thermal history [17]. These discoveries revealed the complexity of Cu–Sb melts structure and its influence on solidification process. These further indicated the necessity of melts structure studies and added a new ingredient to material science. In this contribution, liquid structure of Cu–Sb alloys has been studied by combining ab initio molecular dynamics (AIMD) simulation with X-ray diffraction to probe into local atomic structure. As structure sensitive parameters, resistivity has been measured to characterize concentration inhomogeneity in the melts.

Experimental The samples of Cu–Sb alloys used in present study were prepared by melting from ingots of 5 N copper and 4 N antimony in an induction furnace with the help of glass

fluxing technology. Taking consideration of the mass loss during alloys melting, the errors in concentration for all the samples were less than 0.5 at.%. X-ray diffraction experiments were performed using the h–h high temperature X-ray diffractometer [10]. The measured scattering intensity could be converted into the coherent scattering intensity per atom in an electron coh unit Ieu ðqÞ using generalized Krogh–Moe–Norman method [18–20], where q = 4psinh/k, and h was the half scattering angle. Compton scattering was corrected using the values given by Cromer and Mann [21] Z1  2 sinðqrÞ 2 coh dr ð1Þ Ieu ðqÞ ¼ f þ h f i 4pr 2 ½qðrÞ  q0  qr 0

where f was scattering factor, q(r) was radial density function, q0 was average number density. Then, Ashcroft–Langreth structure factor could be   coh obtained [22, 23] through the function SðqÞ ¼ Ieu = f 2 ð qÞ   P 2 with f 2 ðqÞ ¼ cj fj ðqÞ, where, cj was atomic fraction, j

and fj(q) was total atomic scattering factor of j component in the alloy. With an accumulation in counts, multiple scanning reduced the total error of S(q) to less than 2 %. Pair distribution function g(r) (PDF) reflecting the real space information, could be obtained by S(q) through Fourier transformation Z1 1 gðrÞ ¼ 1 þ 2 q½SðqÞ  1 sinðqrÞdq ð2Þ 2p rq0 0

The position of the first maximum of g(r) represented the mean first-neighbor distance r1. The coordination numbers (Nmin) were calculated using the method [24] of minimum value, integrating from the nearest zero on the left to the first minimum on the right of the first peak in radial distribution function RDF. The error was less than 2 % for r1 and 5 % for Nmin with the present method. Electrical resistivity was measured using four-electrode method with an error of about 0.5 lX cm, more details could be referred to Ref. [10]. To analyze the partial coordination relationship, AIMD simulations were performed using the Vienna ab initio simulation package (VASP) by implementing the projector augmented wave method [25, 26]. All the simulations were

Table 1 Number densities and coordination numbers of Cu–Sb melts Sb (at.%) ND (AIMD) CNs (AIMD)

0

10 0.075

11.52

0.066 10.71

CNs (XRD)

20 0.062 10.34 10.34

25

33

40

50

63

70

80

90

100

0.059

0.054

0.048

0.044

0.040

0.038

0.036

0.033

0.032

9.87

9.18

8.45

8.18

7.74

7.46

7.22

7.08

6.97

7.18

7.00

Cu80Sb20 data are from Ref. [10]

123

4440

carried out in the canonical ensemble (NVT) through a Nose´ [27] thermostat for temperature control. Hundred atoms were used to simulate the liquid alloys in a cubic cell with periodic boundary condition. Cu90Sb10 and Cu melts were simulated at 100 K above the liquidus, and the other alloys at 1073 K. The number densities (NDs) of the melts were shown in Table 1, which were adjusted to achieve an average external pressure around zero. First, the alloys were equilibrated for 3 ps. Then, 2000 more configurations were simulated for 6 ps with the time step of 3 fs to produce structural function at 1073 K. More details could be referred to Refs. [28, 29].

Results and discussion Partial pair distribution functions Partial PDFs of Cu–Cu, Cu–Sb, Sb–Sb pairs for different compositions are shown in Fig. 1a–c, respectively. PDFs of pure Cu and Sb are also exhibited for comparison. To analyze the structure evolution in the melts, the position of

J Mater Sci (2013) 48:4438–4445

first maximum in PDFs and partial coordination numbers are shown in Fig. 1d. The first peaks in gCuCu(r) and gCuSb(r) are symmetrical and their positions are almost independent of concentration. While the firsts peak in gSbSb(r) are asymmetrical. And a large gap between the second and asymmetrical first peak is observed in each gSbSb(r) curve. In this gap, there is a small hump in PDF of pure Sb and gSbSb(r) of adjacent Sb-rich melts. It should be noted that, the position of the small hump in gSbSb(r) is about 0.44 nm, as denoted by vertical bar. It is very close to the second-neighbor distance in crystalline Sb with a rhombohedral A7–structure, which originates from a Peierls distortion of a simple cubic lattice due to electronic instability [30]. Then, this hump is believed to be a signature of Sb clusters with Peierls distortion surviving in Sb-rich Cu–Sb melts [31], following the viewpoint of structural features inheritance [32]. As shown in Fig. 3d, over all the concentration, NCuSb and NSbCu are larger than NCuCu and NSbSb, respectively. These indicate the preference of heterogeneous coordination for atoms in the melts. It means that Sb atoms would prefer to heterogeneously coordinate with Cu atoms rather than construct Peierls

Fig. 1 (Color online) Partial PDFs of Cu–Sb alloys in AIMD. a gCuCu(r), b gCuSb(r), c gSbSb(r), d position (solid symbols) of maximum in partial PDFs and partial coordination numbers (open symbols). Cu80Sb20 curves are from Ref. [10]

123

J Mater Sci (2013) 48:4438–4445

4441

in maximum position of total PDF may not be observed until the percentage of Sb–Sb coordination is high enough, since it denotes statistical atomic arrangement in the melts (this will be discussed later). When Sb concentration is larger than 30 at.%, as Sb concentration increases further, Sb–Sb coordination increases. Due to the large radius and high concentration of Sb atoms, large distortion would occur in FCC-like melts structure. Then, atomic arrangements and rSbSb in Sb-rich melts trend to those of liquid Sb. Then, concentration behavior of rSbSb is well interpreted.

Total pair distribution functions

Fig. 2 (Color online) Experimental and simulated total PDFs. The circles and solid lines denote experimental and simulated results, respectively. Cu80Sb20 results are from Ref. [10]

distortion with Sb atoms in these melts. Then, the weakening hump could be well interpreted as that Peierls distortion decreases with decreasing Sb concentration when Sb concentration is higher than 80 at.%. In AIMD results of Cu–Sb melts, concentration behavior of the first–neighbor distance of Sb–Sb pairs (rSbSb) is very interesting. As shown in Fig. 1d, when Sb concentration is no more than 25 at.%, rSbSb is about 25 % larger than the diameter of Sb atoms (0.305 nm). This large deviation implies that Sb atoms rarely locate in the first neighbor shell of Sb atoms. In the same concentration range, the small NSbSb also implies that the percentage of Sb atoms with homogeneous coordination is very low. To understand such phenomena, analysis on local atomic arrangement of Cu-rich melts is necessary. As it is known, crystalline Cu has FCC structure. Then, it is reasonable to expect that liquid Cu around melting point may display some characteristics similar to FCC solid Cu, such as coordination number, atomic geometrical arrangement. Similarly, from viewpoint of liquid–solid structure correlation, the existence of Cu3Sb clusters with a FCC-like structure [33] was verified in the melts around Cu80Sb20 [10, 34]. Taking into consideration that both liquid Cu and Cu80Sb20 melts have FCC-like structure [35], the anomalously large rSbSb in Cu-rich melts is interpreted as that Sb atoms substitute discontinuously for Cu atoms during alloying [10]. When Sb concentration increases to the concentration corresponding to Cu3Sb, the newly added Sb atoms will emerge in the first coordination shell of Sb atoms. This leads to the rapid increase of Sb–Sb homo-coordination and the rapid decrease of rSbSb. Therefore, around Cu–25 at.% Sb, a significant drop in rSbSb is observed. However, the change

PDF is an important physical quantity in physics of fluids, since it is directly measurable to reveal the characteristics of local structures in liquid. What’s more, in principle, various properties of liquid metals and alloys can be estimated from PDF when coupled with an appropriate theory [36]. Experimental total PDFs are obtained by Fourier transformation from S(q) measured by X-ray diffraction. While simulated total PDFs are obtained [37] by weighting the partial PDFs with the X-ray scattering factors, following 2 gtotal ðrÞ ¼ ðc2Cu fCu gCuCu ðrÞ þ 2cCu cSb fCu fSb gCuSb ðrÞ 2 2 þ cSb fSb gSbSb ðrÞÞ=ðcCu fCu þ cSb fSb Þ2

ð3Þ

where ca, cb are concentration of a and b atoms, respectively. fCu, fSb are the X-ray scattering factors of the two species when scattering wave vector equals to zero. As shown in Fig. 2, simulated total PDFs are in good agreement with experimental results in both height and maximum position (r1) of primary peaks, except for that of Cu20Sb80 with a larger maximum. In Sb-rich melts, simulated PDFs exhibit main peaks with obvious shoulders, while experimental ones show smooth main peaks. This deviation may be due to that experimental PDFs denote the statistically averaging of atomic distribution in the melts. In fact, as shown in Table 1, the coordination numbers (CNs) in AIMD and experiments are comparable. Even in Cu20Sb80 melts, as indicated by vertical line in Fig. 2, the humps on the right side of principal peaks are significant in both experimental and simulated PDF, which were interpreted as the signature of the existence of Peierls distortion [36]. These facts verify that the present simulations are reliable and precise enough for atomic structure analysis of Cu–Sb melts. As expressed in Eq. (3), the total PDF is the sum of the partial PDFs weighted by the X-ray scattering factor and concentration of each element. Then, when Sb concentration is no less than 80 at.%, the shape of simulated total PDF is quite close to that of liquid Sb. As shown in Fig. 2, r1 almost keep a constant of 0.300 nm in this region. These indicate the existence of Sb–Sb homogeneous clusters [38]

123

4442

J Mater Sci (2013) 48:4438–4445

Fig. 3 Temperature dependence of resistivity for liquid Cu–Sb alloys. Data for Cu80Sb20 are from Ref. [10]

in Sb-rich melts. While in Cu-rich melts, when Sb concentration increases to 10 at.%, r1 increases from 0.245 nm in liquid Cu to 0.2525 nm for the addition of larger Sb atoms. It should be noticed that, r1 almost keeps a constant of 0.2625 nm in Sb concentration range of 20–40 at.%. The value is quite close to that of Cu–Sb heterogeneous coordination radius 0.265 nm. This could be well interpreted as the existence of Cu–Sb heterogeneous clusters with high structure stability in the melts, since this range is corresponding to Cu–Sb compound forming region in solid phase. When Sb concentration ranges from 50 to 80 at.%, simulated PDFs show a broad main peak with an obvious shoulder. In Cu50Sb50, on the right of maximum of PDF, a significant shoulder is observed around 0.300 nm, which implies some new coordination increases. While in Cu37Sb63 melts, an obvious splitting is observed in primary peak of PDF curve. In Cu20Sb80, the maximum and shoulder locate around 0.300 and 0.265 nm, respectively. In a previous study of glassy Zr50Ni50 and Zr60Ni40 alloy prepared by mechanical alloying [39], such a splitting was attributed to the presence of two structure units with different kinds of SRO in the amorphous phases. In fact, split main peaks positions locate around 0.265 and 0.300 nm, which are corresponding to first-neighbor distance of Cu–Sb and Sb–Sb pairs, respectively. Then, present AIMD

123

results indicate the coexistence of Cu–Sb associations and Sb-homogeneous clusters around eutectic alloys. It should be noticed that, the shoulder on the right of primary peak in Cu50Sb50 melts strengthens and becomes the principal peak in Cu20Sb80, while the primary peak decreases in height and becomes a shoulder on the left of principal peak. This reveals the transition of dominant clusters from heterogeneous clusters to Sb–Sb homogeneous ones. Resistivity As a parameter sensitive to structure transition [40, 41], electrical resistivity of liquid Cu–Sb alloys has been measured by four-electrode methods. As shown in Fig. 3a, temperature coefficient of resistivity (TCR) is negative when Sb concentration ranges from 20 to 33 at.%. It is consistent with the fact that compound clusters exist in these melts [10]. When Sb concentration is no less than 80 at.%, there is an obvious inflection in each resistivity curve. This is interpreted as the signature of reinforcement of Peierls distortion during cooling process [42]. It is a reasonable explanation since Peierls distortion bounds free electrons and weakens resistivity change with temperature. As stated in ‘‘Partial pair distribution functions’’ section, AIMD results indicate that Peierls distortion survives in the melts at 1073 K. It is not in contradiction with Wang’s

J Mater Sci (2013) 48:4438–4445

4443 Table 2 Position (q1) and height (h1) of main peaks in structure factor Sb (at.%)

0

20

63

80

100

˚ -1) q1 (A

3.00

2.95

2.88

2.40

2.15

h1

2.48

2.50

2.00

1.54

1.39

˚ -1) qs (A

2.40

2.92

hs

1.33

1.21

Subscript ‘‘s’’ denotes split peaks. Cu80Sb20 data are from Ref. [10]

Fig. 4 (Color online) S(q) of different concentrations by X-ray diffraction (thick line) and AIMD (thin line). Experimental temperatures are 1003, 985, and 1048 K, respectively. Cu80Sb20 data are from Ref. [10]

reports [43], in which Peierls distortion was verified to survive still at 1023 K by X-ray absorption spectroscopy. While present inflection temperature in resistivity in liquid Sb is about 1040 K. This deviation may be attributed to different experimental methods. It should be noticed that, the inflection temperature decreases with increasing Cu concentration. This is interpreted as that the addition of Cu atoms decreases the thermal stability of Peierls distortion for preferential heterogeneous coordination, as denoted in ‘‘Partial pair distribution functions’’ section. As shown in Fig. 3b, the concentration inhomogeneity in the melts is also denoted by the obvious maximum in isothermal resistivity. As temperature increases, the maximum decreases and shifts to high-Sb side. This is interpreted as that compounds clusters are destroyed and the melts trend to be more homogeneous at high temperature. In Sb-rich melts, the addition of Cu atoms suppresses the re-formation of Peierls distortion and thus increases free electrons density. Then, resistivity decreases with increasing Cu concentration when Sb concentration is higher than 70 at.%. Structure factor of Cu–Sb melts S(q) curves of Cu–Sb melts measured by high temperature X-ray diffraction (thick lines), calculated by AIMD (thin lines) are plotted in Fig. 4, respectively. As it is shown, calculated S(q) curves are in good agreement with experimental ones, although there is little deviation in position and height of primary peaks. The small fluctuations in front of primary peaks in calculated S(q) curves may be attributed to low atomic numbers in AIMD simulations. The shapes of both experimental and simulated S(q) of Cu80Sb20 melts are quite similar to that of liquid Cu, except

for the peak in front of primary peaks. Pre-peak in experimental S(q) is generally in relation with medium-range order (MRO), which was verified to be Cu3Sb clusters with chemical short-range order (CSRO) in Cu80Sb20 melts [10]. The existence of MRO indicates concentration inhomogenity in Cu-rich melts. While peaks in front of primary peak in simulated S(q) are not related to structure information. A pronounced change in the shape of experimental and simulated S(q) curves occur around the eutectic alloys. Compared with that of Cu80Sb20, experimental principal peak shifts slightly to the low q side, with obviously decreasing height and broadened shape, as shown in Table 2. And a pronounced first peak emerges on the low q side of principal peak, which is very close to the position of primary peak in liquid Sb. While in simulated S(q) curve, primary peak splitting is also observed. However, the first peak is a little higher than the second one, which is different from that of experimental results. This difference may be attributed to the different research methods. Nevertheless, taking peak positions into consideration, the first peaks are believed to be associated with Sb-homogeneous associations in the melts. And it is reasonable to conclude that the second peaks denote Cu–Sb heterogeneous clusters. Thus, main peak splitting suggests coexistence of clusters with geometric arrangement similar to that of liquid Sb and Cu. This is in good agreement with results of PDFs by AIMD in ‘‘Total pair distribution functions’’ section. It is also consistent with previous work of Kaban [14], in which this was interpreted as segregation into clusters corresponding to associations of Sb-atoms and associations of unlike atoms in eutectic alloy. As shown in Fig. 4, primary peak splitting is also observed in both experimental and simulated S(q) curves of liquid Cu20Sb80. In simulated curve, the first peak is higher and the second one is lower than that of experimental results, respectively. And the positions of the two peaks almost keep the same as those in eutectic melts, respectively. The peak corresponding to heterogeneous clusters exhibits obvious decreasing in height, which indicates the decreasing atomic percentage of Cu–Sb heterogeneous clusters. While the first peak in relation with Sb–Sb homogeneous clusters rapidly increases in the height and

123

4444

J Mater Sci (2013) 48:4438–4445

Fig. 6 Warren–Cowley parameters and negative maxima in gCC(r) Fig. 5 Concentration correlation function gCC(r) in Cu–Sb alloys

becomes principal peak. It indicates Sb clusters play a dominant role in the geometric structure of Cu20Sb80 melts. Nevertheless, the present results reveal the coexistence of Sb–homogeneous clusters and Cu–Sb associations in a large concentration range in Sb-rich melts. Consistent with the change of simulated total PDF in ‘‘Total pair distribution functions’’ section, the change of maximum position in S(q) reveals local clusters evolution in Sb-rich melts. Chemical short-range order Atomic arrangements in liquid alloys usually are associated with ordering structure, such as chemical SRO and topological SRO. To describe the deviation from random distribution of atomic structure, the concentration correlation function (CCF) is defined as gCC ðr Þ ¼ cCu cSb ½gCuCu ðr ÞþgSbSb ðr Þ  2gCuSb ðr Þ

ð4Þ

When there is a preferential arrangement for homogeneous or heterogeneous atoms at a given distance, corresponding positive or negative peaks will appear in gCC(r), respectively. As shown in Fig. 5, there is an obvious small positive peak around 0.235 nm and a large negative peak around 0.265 nm on each gCC(r) curve, which indicates that heterogeneous coordination is preferential in the melts. Considering their positions in gCuCu(r), they are attributed to be Cu–Cu

(*0.247 nm) and Cu–Sb (0.263–0.265 nm) interaction, respectively. Consistent with resistivity results, these indicate that Cu–Sb and Sb–Sb clusters with CSRO exist in liquid Cu–Sb systems. As Sb concentration increases, firstneighbor distance of Sb–Sb pairs decreases and Sb–Sb interaction increases, which would produce increasing positive peaks. As it is shown in Fig. 5, the negative peaks become narrow with increasing Sb concentration. It should be mentioned that, some atomic interaction can not produce the corresponding peaks in gCC(r) curves since the distances of these bonding pairs are too close to distinguish their own effects. Taking the distance of Sb–Sb pairs and the wide negative peak in gCC(r) into consideration, it is reasonable to conclude that the positive peaks denoting Sb–Sb interaction is covered by the effect of Cu–Sb interaction in gCC(r). In order to get insight into the SROs in liquid Cu–Sb alloys, the partial coordination number (PCN) is obtained by integrating partial PDF within the first minimum, as shown in Table 3. The Warren–Cowley parameter is a parameter characterizing the preference of atomic distribution. It can be derived by PCNs as a = 1–Nab/ [cb(Nab ? Naa)], where cb is the atomic fraction of b atoms in the melts. When there is a preference of separation or aggregation for two species at given concentration, corresponding positive or negative values appear, respectively. As shown in Fig. 6, the melts show significant negative values of Warren–Cowley parameter over all the concentration, reflecting the aggregation tendency between Cu and Sb atoms. This is in good agreement with negative

Table 3 Partial coordination numbers of Cu–Sb melts Sb (at.%)

10

20

25

33

40

50

63

70

80

90

NCuCu

9.38

7.92

7.09

5.99

4.58

3.51

2.44

1.73

1.29

0.57

NCuSb

1.03

2.84

3.31

3.95

4.41

5.01

5.7

6.06

6.27

6.83

NSbCu NSbSb

14.64 0.15

11.34 1.45

9.93 2.16

8.02 3.05

6.61 3.23

5.01 4.02

3.35 4.94

2.60 5.25

1.57 5.86

0.76 6.24

123

J Mater Sci (2013) 48:4438–4445

peaks in gCC(r) indicating preference of heterogeneous coordination. The Warren–Cowley parameter exhibits a negative maximum around Cu80Sb20, indicating the strongest chemical interaction between Cu and Sb atoms. While CCFs exhibits negative maximum around Cu50Sb50, implies largest deviation from random distribution for chemical interaction and atomic size difference. This deviation indicates atomic size effect takes an important role in local atomic arrangement and clusters evolution.

Summary Structure inhomogeneity in Cu–Sb melts has been probed by AIMD simulation, high temperature XRD and resistivity measurement. By AIMD simulation, local atomic coordination and clusters evolution are revealed by the abnormal change of Sb–Sb first-neighbor distance. Main peak splitting of S(q) and simulated total PDFs indicate the coexistence of Cu–Sb heterogeneous clusters and Sb–Sb homogeneous clusters in Sb concentration range of 50–80 at.%. These imply solid-phase structures begin to form before crystallization. Resistivity results are well interpreted as the existence of Cu3Sb compounds clusters and Peierls distortion in Cu- and Sb-rich melts, respectively. Concentration inhomogeneity is discussed in terms of gCC(r) denoting the deviation of atomic arrangements from free distribution and the Warren–Cowley parameters characterizing interaction between heterogeneous atoms. CSRO effects between Cu and Sb are revealed by negative Warren–Cowley parameters and negative peaks corresponding to Cu–Sb interaction in CCFs. The positions of negative maximums in these two parameters deviate, which may be attributed to atomic size effect in Cu–Sb melts. Acknowledgements This study was supported by the National Natural Science Foundation of China (Grant Nos. 50971083, 50971082).

References 1. Schick M, Brillo J, Egry I, Hallstedt B (2012) J Mater Sci 47:8145. doi:10.1007/s10853-012-6710-x 2. Tian H, Liu H, Zhang C, Zhao J, Dong C, Wen B (2012) J Mater Sci 47:7628. doi:10.1007/s10853-012-6676-8 3. Kaban I, Kohler M, Ratke L, Hoyer W, Mattern N, Eckert J, Greer AL (2011) Acta Mater 59:6880 4. Kaban I, Curiotto S, Chatain D, Hoyer W (2011) Acta Mater 58:3406

4445 5. Ning S, Bian X, Ren Z (2010) Phys B 405:3633 6. Gierlotka W, Jendrzejczyk-Handzlik D (2009) J Alloys Compd 484:172 7. Steeb S, Maier U, Godel D (1969) Phys Chem Liq 1:221 8. Knoll W, Steeb S (1973) Phys Chem Liq 4:27 9. Halm Th, Neumann H, Hoyer W (1994) Z Naturforsch 49A:530 10. Guo F, Zheng H, Qin J, Qin X, Lv T, Jia Y, Xu R, Tian X (2012) J Non-Cryst Solids 358:3327 11. Zhao N, Pan X, Ma H, Wang L (2008) Acta Metall Sin 4:467 12. Chen BL, Li GY (2004) Thin Solid Films 462–463:395 13. Kaban I, Hoyer W, Il’inskii A, Slukhovskii O, Slyusarenko S (2003) J Non-Cryst Solids 331:254 14. Kaban I, Hoyer W, Halm Th (2001) J Non-Cryst Solids 288:96 15. Greaves GN, Wilding MC, Fearn S, Langstaff D, Kargl F, Cox S, Van QV, Maje´rus O, Benmore CJ, Weber R, Martin CM, Hennet L (2008) Science 322:566 16. Zu FQ, Mao LN, Hu CM, Chen J, Huang ZY, Liu YC, Liu MQ (2010) Mater Sci Tech 26:1353 17. Huang Z, Zu F, Li X (2012) Cryst Res Technol 47:953 18. Krogh-Moe J (1956) Acta Crystallogr 9:951 19. Norman N (1957) Acta Crystallogr 10:370 20. Waseda Y (1980) The structure of non-crystalline materials. McGraw–Gill, New York 21. Cromer DT, Mann JB (1967) J Chem Phys 47:1892 22. Enderby JE (1980) Philos Trans R Soc London B 290:553 23. Barnes AC, Hamilton MA, Buchanan P, Saboungi ML (1999) J Non-Cryst Solids 250–252:393 24. Zhao X, Bian X, Bai Y, Li X (2012) J Appl Phys 111:103514 25. Blo¨chl PE (1994) Phys Rev B 50:17953 26. Kresse G, Joubert D (1999) Phys Rev B 59:1758 27. Nose´ S (1984) J Chem Phys 81:511 28. Qin J, Gu T, Yang L, Bian X (2007) Appl Phys Lett 90:201909 29. Liu D, Pan SP, Qin JY, Gu TK (2011) J Appl Phys 109:093519 30. Otjacques C, Raty JY, Coulet MV, Johnson M, Schober H, Bichara C, Gaspard JP (2009) Phys Rev Lett 103:245901 31. Zhang C, Wei Y, Zhu C (2005) Chem Phys Lett 408:348 32. Wang WH (2012) J Appl Phys 111:123519 33. Kim TH, Lee GW, Gangopadhyay AK, WHyers R, Rogers JR, Goldman AI, Kelton KF (2007) J Phys Condens Matter 19:455212 34. Knoll W, Steeb S (1973) Phys Chem Liq 4:39 35. Tian XL, Zhan CW, Hou JX, Chen XC, Sun JJ (2010) J Mater Sci Technol 26:69 36. Wu YN, Zhao G, Liu CS, Zhu ZG (2006) J Phys: Condens Matter 18:4471 37. Yang L, Qin J, Pan S, Bian X (2011) J Non-Cryst Solids 357:3207 38. Guo F, Tian Y, Zheng H, Li T, Liu X, Xu X, Qin J, Qin X, Tian X (2013) Mater Lett 95:128 39. Petzoldt P, Scholz B, Kunze HD (1988) Mater Sci Eng 97:25 40. Yi J, Bai HY, Zhao DQ, Pan MX, Wang WH (2011) Appl Phys Lett 98:241917 41. Korkmaz SD, Korkmaz S¸ (2010) Comp Mater Sci 48:466 42. Wang Q, Chen XM, Lu KQ (2001) J Phys: Condens Matter 13:8445 43. Wang Q, Li CX, Wu ZH, Wang LW, Niu XJ, Yan WS, Xie YN, Wei SQ, Lu KQ (2008) J Chem Phys 128:224501 44. Waseda Y, Ohtani M (1974) Phys Stat Sol B 62:535 45. Waseda Y, Suzuki K (1971) Phys Stat Sol B 47:581

123