Interdiffusion in liquid AlCu and NiCu alloys

H. Cheng, Y. J. L, and M. Chen

Citation: The Journal of Chemical Physics 131, 044502 (2009); doi: 10.1063/1.3184614
View online: http://dx.doi.org/10.1063/1.3184614
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/131/4?ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
Correlation between dynamic slowing down and local icosahedral ordering in undercooled liquid Al80Ni20 alloy
J. Chem. Phys. 143, 084508 (2015); 10.1063/1.4929481
Relationship between structure and dynamics in liquid Al1x Ni x alloys
J. Chem. Phys. 143, 084504 (2015); 10.1063/1.4928975
Liquid phase separation and rapid dendritic growth of highly undercooled ternary Fe62.5Cu27.5Sn10 alloy
J. Appl. Phys. 117, 054901 (2015); 10.1063/1.4907214
Measurement and calculation of surface tension for undercooled liquid nickel and its alloy
J. Appl. Phys. 106, 033506 (2009); 10.1063/1.3187793
Chemical and icosahedral short-range orders in liquid and undercooled Al 80 Mn 20 and Al 80 Ni 20 alloys: A
first-principles-based approach
J. Chem. Phys. 123, 104508 (2005); 10.1063/1.1979495

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
131.151.108.239 On: Wed, 27 Jan 2016 17:52:17

THE JOURNAL OF CHEMICAL PHYSICS 131, 044502 2009

Interdiffusion in liquid AlCu and NiCu alloys

H. Cheng, Y. J. L, and M. Chena
Department of Engineering Mechanics, Tsinghua University, Beijing 100084, Peoples Republic of China

Received 3 November 2008; accepted 26 June 2009; published online 22 July 2009
The interdiffusion processes in liquid AlCu and NiCu alloys are studied by using molecular
dynamics simulation method. The MaxwellStefan MS diffusivities are calculated over a wide
composition range with both the GreenKubo method and the Darken relation. Comparisons show
that the Darken relation predicts well the MS diffusivity for NiCu alloy, while overestimates the
value for AlCu alloy, especially in the medium concentration region. Based on the calculated MS
diffusivities and the activities of the alloys, the Fickian interdiffusivities are predicted. The results
show strong dependences on the compositions of the alloys. In addition, the Fickian
interdiffusivities of Al60Cu40 and Ni50Cu50 melts as a function of undercooling are estimated, which
is proved to be beneficial in improving the quantitative predictions of dendrite growth velocity in
solidification. 2009 American Institute of Physics. DOI: 10.1063/1.3184614
I. INTRODUCTION

The solute diffusion in liquid alloy is one of the dominant kinetic processes in solidification. It plays a key role in
quantitative prediction of microstructural evolvement. In directional solidification, the stability of the solid/liquid interface is dominatingly determined by the diffusivity and the
growth velocity.1 For dendrite growth, the solute flux in the
liquid ahead of the dendrite tip J can be described by the
well-known Ficks First Law,1
J=D

c
r

1
ln ai
ci = ci .
i =
RT
ln ci

i = 1, ,N 1,

where Dij is the Fickian interdiffusivity. Equation 2 shows

that the diffusion flux Ji is driven solely by the concentration
gradient c j. However, in systems with ternary or more components, the description of transport diffusion cannot neglect
the influence from other species. It has been demonstrated by
the experiments that N2 molecules diffusion against the concentration gradient in an ideal gas mixture.2,3
The MaxwellStefan MS equation provides a different
description of diffusion, especially for multicomponent
mixtures.4,5 On the basis of the kinetic process of ideal gas, a
flux is produced by the balance between the driving force,
chemical potential gradient i and the frictional drag,
Author to whom correspondence should be addressed. Tel.: 86-162773776.
FAX:
86-1-62781610.
Electronic
mail:
mchen@tsinghua.edu.cn.

0021-9606/2009/1314/044502/6/$25.00

Here is the thermodynamic correction factor describing the

nonideal behavior of the mixture. According to the above
equation, the relation between the MS diffusivity and the
Fickian diffusivity is easily phrased as follows:
D12 = 12 .

N1

Dij c j,

j=1

i, j = 1, ,N 1.

The frictional drag, as the right term in the above equation, is proportional to the relative velocity of species i and j,
ui u j and the mole fraction c j. ij denotes the MS diffusivity
and can be regarded as an inverse drag coefficient. For a
nonideal liquid mixture, the left term of Eq. 3 is a function
of the component activity ai,

r=R

where c denotes the mole fraction of solute, D is the Fickian

diffusivity, and R the tip radius. More generally, considering
an N-component system with ci being the mole fraction of
the ith component, the Fickian diffusion processes follow a
relation as below,
Ji =

c jui u j
1

i =
ij
RT
j=1;ji

It is noteworthy that the descriptions of the diffusion

process by the Fick and the MS formulations are mathematically equivalent.6 In the limit of ideal mixtures, the MS and
Fickian diffusivities are identical to each other.
Direct measurement of the MS diffusivity using Eq. 3
is difficult because the diffusivity less depends on the concentration. Based on a phenomenological model in binary
systems, Darken7 and Bardeen8 proposed a correlation to related interdiffusivity to the self-diffusivities of species 1 and
2, D1,self and D2,self,
D12 = c1D2,self + c2D1,self S ,

where S is the Manning factor,9 which measures the contribution of cross correlations to D12. If the Manning factor is
postulated to be close to unity, the interdiffusivity is then
simplified into a linear combination of the self-diffusivities.
As a consequence, we have the Darken relation,

131, 044502-1

2009 American Institute of Physics

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
131.151.108.239 On: Wed, 27 Jan 2016 17:52:17

044502-2

J. Chem. Phys. 131, 044502 2009

Cheng, L, and Chen

12 = c1D2,self + c2D1,self .

It relates the MS interdiffusivity to the self-diffusivities of

the species. The Darken relation has been justified in some
multicomponent liquid metal and organic systems,1014 and is
frequently applied to evaluate the MS diffusivity in both experiments and molecular dynamics MD simulations.15,16 In
experiments, the long-capillary LC method has been employed to measure the self-diffusivities of the solutes,12,17,18
which requires a diffusion couple that eliminates the buoyancy convection, then the self-diffusivity is calculated by
measuring the concentration profile of the isotope incidences, consequently the MS diffusivity. For binary systems,
the Fickian diffusivity is indeed more easily attainable in
experiments because its definition relates directly to concentration gradient. Lee et al.19 measured the Fickian diffusivity
of the liquid AlCu alloy by analyzing the concentration profile ahead of a planar interface in directional solidification,
and observed that the fluid flow can seriously affect the experimental results. The interdiffusion of AlCu melts in the
undercooled regime was also studied experimentally by
Tanaka and Kajihara.20
MD simulations provide a convenient way of studying
interdiffusion in liquid alloys. Horbach et al.12 investigated
the diffusion dynamics of Al80Ni20 melts by a combination of
experimental and MD methods. Their results show that the
Manning factor well remains unity over the whole temperature range, and suggest the validation of the Darken relation
for this alloy. Similar conclusions are also obtained in
Al77Ni20Ce3 and linear alkanes systems.11,21 Although it is
justified experimentally or from MD simulations for some
liquid metal and organic systems, the validation of the
Darken relation to other alloy systems is still an open question. In this paper, we calculate the self-diffusivities and interdiffusivities of liquid AlCu and NiCu alloys in MD
simulations by means of velocity and interdiffusion flux autocorrelation functions without the aid of the Darken relation. We obtain the MS and Fickian interdiffusivities as functions of concentration and temperature, and examine the
validity of the Darken relation in AlCu and NiCu alloy
systems.

II. MD SIMULATIONS
A. Diffusivity

In MD simulations, a variety of transport properties can

be calculated by means of the GreenKubo method. The selfdiffusivity is defined by the integral over the velocity autocorrelation function,
Di,self =

1
3

12 =

1
3Nc1c2

Jt J0dt,

with the autocorrelation function of the interdiffusion flux,

Jt = c1 uit c2 u jt. N is the total number of particles.
i2

j1

If the velocity cross correlation uit u j0 , i j can be neglected, the average interdiffusion flux is accordingly simplified as,
12 =

c2
3

u 1t u 10dt +

c1
3

u 2t u 20dt.

10
This corresponds with the Darken relation. Therefore, the
MD simulations provide another way to test the Darken relation by comparing the MS diffusivities from Eqs. 9 and
10. That is, check the negligibility of the velocity cross
correlation.
The Fickian diffusivity is also presented in this work by
using Eq. 5, in which the activity of liquid alloys is given
by

ai = exp

i i T, P
,
RT

11

where i is the chemical potential and i T , P is the standard chemical potential, R the gas constant, and T the absolute temperature. The chemical potential is calculated by the
PANDAT software in our simulations.
B. Simulation details

The embedded atom method is employed to describe the

atomic interaction of AlCu and NiCu alloys. The potential
functions are developed by Cai and Ye.24 Simulations are
performed under the isothermal-isobaric N-P-T ensemble.
The systems consist of 4000 atoms covering the whole concentration range. The atoms are arranged into face-center
cubic fcc lattice in a cubic box initially. The periodic
boundaries are applied in three coordinate directions and the
timstep is set as 1.0 fs. First, the system runs for 200 000
steps at zero pressure and 3000 K to be melt completely, and
then is quenched to the desired temperature. In each sampling, we record the velocities of all atoms for 500 000 steps
after 100 000 steps equilibrating.
The autocorrelation function of the interdiffusion flux is
computed using overlapped measurements. We average 6000
time-steps samples for each interval 0.1 ps to generate
accurate statistics. After about 400 fs, the autocorrelation
function decays to zero as illustrated in Fig. 1. Further increase in the simulation time cannot noticeably improve the
results.
III. RESULTS AND DISCUSSIONS

u it u i0dt,

where u it denotes the velocity vector of species i at time t

and is the ensemble average. In the case of interdiffusion,
the MS diffusivity is defined by,22,23

A. MS diffusivity

On the basis of the GreenKubo relation described in

Eq. 9, we calculate the MS diffusivities of liquid AlCu
alloys at 1500 K with compositions ranging from 2% to 99%
Al, as shown in Fig. 2. With the increase in the Al concen-

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
131.151.108.239 On: Wed, 27 Jan 2016 17:52:17

044502-3

Interdiffusion in liquid alloys

FIG. 1. Decaying of the autocorrelation function of the interdiffusion flux

with time steps.

J. Chem. Phys. 131, 044502 2009

FIG. 3. Self-diffusivity of AlCu alloy vs Al composition. The circles and

triangles are the self-diffusivities of Cu and Al components in liquid AlCu
alloy respectively. Solid lines are guide to eyes.

tration, the MS diffusivity rises quadratically from 6.36

109 to 2.11 108 m2 s1. Up to now, there is no published data in both experiments and molecular simulations
available in literature to test this result. Moreover, the MS
diffusivity of the AlCu alloy is also calculated using the
Darken relation based on their self-diffusivities. Figure 3
shows the dependences of the self-diffusivities of Al and Cu
in liquid AlCu alloys on composition. As the Al concentration increases, both of the self-diffusivities increase in a
similar way to the MS diffusivity. The differences between
them show rather small. Ejima et al.18 measured the tracer
diffusivity of Cu in liquid pure Al by using the LC method
and gave the Arrhenius relation in the temperature range
from 976 to 1260 K as D = D0 expQ / RT with the preexponential factor, D0 = 1.10 107 m2 s1 and activation
energy Q = 23.8 kJ mol1. If we extrapolate this relation to
1500 K, the tracer diffusivity of Cu in liquid Al equals to
1.63 108 m2 s1, which agrees qualitatively with the
present result in the case of 99% Al. With the selfdiffusivities of Al and Cu, the MS diffusivities based on the
Darken relation can be calculated and are also plotted in Fig.
2. The comparisons of the two MS diffusivities indicate that

the Darken relation works well in low concentration region

10% Al, whereas are observed to be remarkably larger
than those by the GreenKubo method for mediumconcentration systems, and this departure reaches its maximum at 70% Al. Further increases the concentration, the difference between the two diffusivities shrinks continuously.
The results suggest that the Darken relation becomes less
accurate for predicting the MS diffusivity in the medium
concentration range, where the validity of the Darken relation is less proved. A possible reason of this discrepance is
that the cluster structure may display better ordering in this
liquid alloy of medium concentration, which may strengthen
the velocity cross correlation.
We repeat the same simulations in NiCu alloy. Figure 4
shows the concentration dependence of the MS diffusivities
by the GreenKubo method and the Darken relation. Both of
the diffusivities slowly decrease with the increase in Ni concentration. Although the difference between them shows
somewhat large in the medium concentration range 40%
70%, the Darken relation acceptably approximates the MS

FIG. 2. MS diffusivity of AlCu alloy as a function of concentration. The

solid circles are the simulated results by the GreenKubo method and the
solid triangles are the values from the Darken relation. Solid lines are guide
to eyes.

FIG. 4. MS diffusivity of NiCu alloy as a function of concentration. The

open circles are the results obtained by the GreenKubo method and the
open triangles are the values from the Darken relation. Solid lines are guide
to eyes.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
131.151.108.239 On: Wed, 27 Jan 2016 17:52:17

044502-4

Cheng, L, and Chen

J. Chem. Phys. 131, 044502 2009

FIG. 5. Thermodynamic correction factor vs concentration. a AlCu alloy and b NiCu alloy.

diffusivity, implying that the velocity cross correlation contributes less to the interdiffusion behavior than that in the
AlCu alloy. The equilibrium phase diagram of NiCu alloy
shows an infinite mutual dissolution between Ni and Cu atoms. That may partially account for the weak velocity cross
correlation.
B. Fickian diffusivity

If the thermodynamic correction factor is determined,

the Fickian diffusivity can be calculated by Eq. 5 based on
the simulated MS diffusivity. Before that, the activities of Al,
Cu in AlCu melts and Ni, Cu in NiCu melts are calculated
at 1500 K by the PANDAT software. The thermodynamic
correction factors are subsequently obtained over the whole
concentration range, as shown in Figs. 5a and 5b. The
thermodynamic correction factors for Al and Cu in AlCu
alloy show a good consistence, which is also the case for Ni
and Cu in NiCu alloy. Moreover, they all strongly depend
on the alloy composition, and especially in the medium concentration region where the values show positively or negatively far from unity, exhibiting the characteristics of highly
nonideal mixtures. Associated with these simulation results,

Figs. 6a and 6b show the Fickian diffusivities of the

AlCu and the NiCu melts at 1500 K. For AlCu alloy, the
Fickian diffusivity is characterized by a nonmonotonic function of composition and the diffusion is enhanced up to the
maximum at 50% Al with the increase in Al composition.
Due to the negative deviation from the ideal mixture, the
Fickian diffusivity of the NiCu alloy reaches the minimum
at 50% Ni before rising with the concentration increase.
Also, the absence of published experimental data restricts
further validating of the present predictions.
The temperature dependences of the Fickian diffusivity
of Al60Cu40 and Ni50Cu50 melts are also studied. Figure 7a
illustrates the Fickian diffusivity of Al60Cu40 in temperatures
range from 970 to 1500 K. With the decrease in temperature,
DAlCu is reduced from 2.38 108 to 1.14 108 m2 s1. The
Arrhenius relation describing the temperature dependence is
given by D0 = 9.16 108 m2 s1 and Q = 17.0 kJ mol1. Although the liquid AlCu system is intensively used to study
the diffusion dynamics of melts, most measurements concentrated on low Cu-composition systems 10 at. % Cu, and
the reported values of the Fickian diffusivity vary from 2.4
109 to 5.5 109 m2 s1.19,25,26 The results show that the
Fick-diffusion behavior strongly depends on the alloy com-

FIG. 6. Concentration dependence of the Fickian diffusivities of AlCu and NiCu alloys. a AlCu alloy and b NiCu alloy.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
131.151.108.239 On: Wed, 27 Jan 2016 17:52:17

044502-5

J. Chem. Phys. 131, 044502 2009

Interdiffusion in liquid alloys

FIG. 7. Fickian diffusivities of Al60Cu40 and Ni50Cu50 alloys as a function of undercooling. Solid lines are the Arrhenius fitting to simulated results. a
Al60Cu40 and b Ni50Cu50.

position. The main reason is that the AlCu melts seriously

deviate from the ideal mixture in the medium concentration
region. Numerically, the thermodynamic correction factor far
from unity dominates the behavior of the Fickian diffusivity
defined by Eq. 5.
For Ni50Cu50 alloy, the simulated temperature covers a
range from 1500 to 2200 K. Figure 7b illustrates the Fickian diffusivity of Ni50Cu50 melts as a function of the reciprocal of temperature. The value of the diffusivity rises from
1.62 109 to 5.63 109 m2 s1 when the temperature increases from 1500 to 2200 K, which fits the Arrhenius relation with D0 = 8.03 108 m2 s1 and Q = 48.9 kJ mol1. Although the NiCu alloy is a typical binary alloy for studying
the dendrite growth, there is still a lack of experimental data
for the diffusivities required by the dendrite growth models,
for example, the LiptonKurzTrivedi LKT model.27 In
some cases, it is only estimated within order of magnitude.
Willnecker et al.28 measured the dendrite growth velocity of
Ni-30%Cu alloy, and the value of D = 6.0 109 m2 s1 is
used in their theoretical analyses, which is also adopted in
predicting the solidification velocity of Ni-50%Cu alloy by

FIG. 8. Dendrite growth velocity of Ni50Cu50 alloy as a function of

undercooling.

Song et al.29 However, for the systems with low Cu composition, Algoso et al. argued that the above diffusivity level is
less efficient for describing the growth velocity in experiments. They optimized some parameters including the diffusivity to pursuit a better agreement between the experiment
and the theory, such as D = 5.0 108 m2 s1 for Ni-5%Cu
and D = 4.4 108 m2 s1 for Ni-10%Cu.30 In order to better
figure out to what degree the diffusivity affects the dendrite
growth, we recalculate the dendrite growth velocity of Ni50%Cu alloy by LKT model, where the calculation parameters given by Song et al. are employed except the diffusivity
which is replaced by the value in present work. The results
are plotted in Fig. 8. The comparison shows that the dendrite
growth velocity using the simulated diffusivity begins to exceed Songs value when the undercooling is larger than 123
K.29 In average, the difference between them remains around
10%. A better estimation of the diffusivity will be helpful to
the quantitative analyses of the dendrite growth kinetics.
IV. CONCLUSIONS

The MS diffusivities of liquid AlCu and NiCu alloys

are predicted by using MD simulations with the GreenKubo
method and the Darken relation. The results show that the
Darken relation provides an acceptable approximation to the
MS diffusivity for NiCu system over the whole composition
range, whereas it overestimates the MS diffusivity for AlCu
alloys in the medium concentration region. The validity of
the Darken relation relies on the alloy type and composition.
The Fickian diffusivities of AlCu and NiCu alloys are
obtained on the bases of the MS diffusivities, and display
strong composition dependences. The behavior may attribute
to the nonidealness of the mixtures of the two alloys, which
is scaled by the thermodynamic correction factor as a function of activity. As an example, the Fickian diffusivities of
Al60Cu40 and Ni50Cu50 alloys as functions of undercoolings
are calculated and employed to predict the dendrite growth
velocity. A better estimation of the diffusivity is proved to be
beneficial in improving the theoretical prediction of the dendrite growth kinetics.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
131.151.108.239 On: Wed, 27 Jan 2016 17:52:17

044502-6

ACKNOWLEDGMENTS

This work is financially supported by the National Natural Science Foundation of China Grant Nos. 50395101 and
50701027 and the National 973 Program Grant No.
2009CB219805. The computations are carried out at the
Tsinghua National Laboratory for Information Science and
Technology, China.
W. Kurz and D. J. Fisher, Fundamentals of Solidification, 3rd ed. Trans
Tech, Netherlands, 1989.
J. B. Duncan and H. L. Toor, AIChE J. 8, 38 1962.
3
R. Krishna and J. A. Wesselingh, Chem. Eng. Sci. 52, 861 1997.
4
R. Krishna and J. M. van Baten, Ind. Eng. Chem. Res. 44, 6939 2005.
5
R. Taylor and R. Krishna, Multicomponent Mass Transfer Wiley, NewYork, 1993.
6
D. S. Sholl, Acc. Chem. Res. 39, 403 2006.
7
L. S. Darken, Trans. Am. Inst. Min. Metall. Eng. 175, 184 1948.
8
J. Bardeen, Phys. Rev. 76, 1403 1949.
9
J. R. Manning, Phys. Rev. 124, 470 1961.
10
A. I. Skoulidas and D. S. Sholl, J. Phys. Chem. B 105, 3151 2001.
11
R. Krishna and J. M. van Baten, Chem. Eng. Technol. 29, 761 2006.
12
J. Horbach, S. K. Das, A. Griesche, M. P. Macht, G. Frohberg, and A.
Meyer, Phys. Rev. B 75, 174304 2007.
13
I. V. Belova, G. E. Murch, R. Filipek, and M. Danielewski, Acta Mater.
1

J. Chem. Phys. 131, 044502 2009

Cheng, L, and Chen

53, 4613 2005.

Y. Yamazaki, Y. Iijima, and M. Okada, Acta Mater. 52, 1247 2004.
15
S. Rehfeldt and J. Stichlmair, Fluid Phase Equilib. 256, 99 2007.
16
R. Krishna and J. M. van Baten, Ind. Eng. Chem. Res. 45, 2084 2006.
17
M. Klassen and J. R. Cahoon, Metall. Mater. Trans. A 31, 1343 2000.
18
T. Ejima, T. Yamamura, N. Uchida, Y. Matsuzaki, and M. Nikaido, J. Jpn.
Inst. Met. 44, 316 1980.
19
J. H. Lee, S. Liu, H. Miyahara, and R. Trivedi, Metall. Mater. Trans. B
35B, 909 2004.
20
Y. Tanaka and M. Kajihara, Mater. Trans. 47, 1 2006.
21
A. Griesche, M. P. Macht, and G. Frohberg, J. Non-Cryst. Solids 353,
3305 2007.
22
G. A. Fernndez, J. Vrabec, and H. Hasse, Int. J. Thermophys. 25, 175
2004.
23
M. J. J. van de Ven-Lucassen, T. J. H. Vlugt, A. J. J. van der Zendan, and
P. J. A. M. Kerkhof, Mol. Phys. 94, 495 1998.
24
J. Cai and Y. Y. Ye, Phys. Rev. B 54, 8398 1996.
25
H. M. Tensi and C. Mackrodt, Zeit. fur Metall. 81, 367 1990.
26
B. N. Bhat, J. Cryst. Growth 28, 68 1975.
27
J. Lipton, W. Kurz, and R. Trivedi, Acta Metall. 35, 957 1987.
28
R. Willnecker, D. M. Herlach, and B. Feuerbacher, Phys. Rev. Lett. 62,
2707 1989.
29
G. S. Song, M. H. Lee, W. T. Kim, D. H. Kim, Z. Z. Zhang, X. Lin, G.
C. Yang, and Y. H. Zhou, Mater. Trans., JIM 41, 1569 2000.
30
P. R. Algoso, W. H. Hofmeister, and R. J. Bayuzick, Acta Mater. 51,
4307 2003.
14

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
131.151.108.239 On: Wed, 27 Jan 2016 17:52:17