CALTECH ASCI TECHNICAL REPORT 002

caltechASCI/2000.002

Viscosities of Liquid Metal Alloys from Nonequilibrium Molecular Dynamics Y. Qi, T. Cagin, Y. Kimura, W. Goddard III

8/5/98 Phys. Rev. E

Viscosities of Liquid Metal Alloys from Nonequilibrium Molecular Dynamics Yue Qi, Tahir Cag n,* Yoshitaka Kimura, and William A. Goddard III*
Materials and Process Simulation Center, Beckman Institute (139-74) Division of Chemistry and Chemical Engineering California Institute of Technology, Pasadena, California 91125

Abstract

We have developed a nonequilibrium molecular dynamics (NEMD) approach to predict viscosity by including external shear rates directly into the Hamiltonian equations of motion. Using the quantum Sutton-Chen (Q-SC) many-body potentials for Au and Cu, we applied NEMD to predict the viscosity as a function of shear rates for AuxCu1;x alloys with x ranging from 0 to 100%. This was done for temperatures of 1500K to 2000K. The predicted viscosities are in reasonable agreement with experiment. In particular, we nd that xing the density and changing the temperature leads to very little change in the shear viscosity. Thus, the temperature dependence of viscosity is due mainly to the change in density with temperature. 83.20.Jp, 83.20.Di, 83.50.Ax, 83.80.Pc, 61.25.Mv, 61.66.Dk

*To whom correspondence should be addressed.

19-1

Reliable data on the viscosity of liquid metals is important for designing and optimizing of metallurgical processes, such as re ning, casting, welding, and solidi cation. Unfortunately, experimental data on viscosity is inadequate for the needs of today's technology. Equations have been proposed to estimate the viscosity of liquid metals based on experimental and theoretical relations (e.g. recent work by Hirai1 and Seetharaman2), but these equations may not provide adequate accuracy. Our objective is to provide a bridge between experiment and theory by providing a sound foundation for molecularly based theories of rheology. The nonequilibrium molecular dynamics (NEMD) technique enables us to determine the transport coe cients from the response of the system to nite applied elds.3 4 In NEMD, the viscosity is calculated in the same way that an experimentalist would measure it. The shear viscosity coe cient is obtained directly by evaluating the momentum ux in a system subject to a known applied shear rate _ . NEMD can evaluate the pressure and viscometric function in addition to the shear viscosity, enabling us to study the uid microstructure in the nonequilibrium steady state. NEMD methods have been used to study the properties of such systems as molten NaCl,5 n-alkanes,6 7 water,8 liquid crystals,9 etc. In this paper we applied NEMD to calculate the shear viscosity of Cu-Au alloys subjected to a planar Couette shear ow. We used the quantum Sutton-Chen (Q-SC) many-body force eld (FF)10;12 in which parameters were optimized for Cu and Au by tting to such experimental properties as density, cohesive energy, moduli, and phonon frequencies.12 The parameters for the alloys are obtained using combination rules. These parameters lead to accurate values for surface energies, vacancies energies, and stacking faults. Consequently, we expect them to be accurate for describing shear ow. Section 2 describes the methodology, while Section 3 discusses the results.

1.0 Introduction

2.0 Calculational Methods 2.1 Interaction Potential and Parameters

The Q-SC many-body FF describes the total potential energy of the metal as 2 3 X X X 1 Utot = Ui = 4 Dij 2 V (rij ) ; ci Dii 1=2 5 (1) i
i i i6=j

19-2

where rij is the distance between atoms i and j , V (rij ) is a pairwise potential

V (rij ) =

ij rij

nij

(2)

(accounting for the short-range repulsion) between the cores of atoms i and j , and i is a local density associated with atom i and de ned as X X ij m = (rij ) = (3) i rij j 6=i j 6=i
ij

This term describes the attractive forces of metallic bonding. The sums in Eq. (1) are for rij < 2 ij (a cuto of twice of the cubic lattice parameter of the material). In Eq. (1), D sets the overall energy scale, while c is a dimensionless parameter scaling the attractive metallic bonding term relative to the Pauli repulsive term. is a length parameter leading to a dimensionless form for V and . To extend the Q-SC model to alloys, we used the following combination rules to represent the interaction between a Au-Cu pair p Dij = Di Dj (4) (5) mij = (mi + mj ) 2 nij = (ni + nj ) (6) 2 ( i + j) (7) ij = 2 The parameters used in this simulation are shown in Table 1. These Q-SC parameters were obtained by tting the density, cohesive energy, elastic constant, and phonon dispersion curves. They lead to accurate values for surface energies, vacancy energy, and stacking fault energy. We refer to this modi ed set of SC parameters as the Q-SC FF.

2.2 Planar Couette Flow and the Shear Viscosity

For computational simplicity, the NEMD calculation on sheared uids is carried out for a system subjected to a planar Couette ow (pure shear), with xed density and temperature. We consider that the applied shear is in the xy plane and changes linearly in the y direction. The system can be exactly modeled by applying the isokinetic SLLOD equations of motion for a molecular uid.3 4 19-3

pxi = Fxi ; _

xi = pxi + _ yi _ m i yi = pyi _ m i zi = pzi _ m

i _ pyi ;

(8) (9) (10)

pxi

(11) (12) (13) (14)

pyi = Fyi ; pyi _ pzi = Fzi ; pzi _

where _ is the shear rate, or the velocity gradient, _ (t) = @ux @y

and ux is the velocity in the x direction. Here is the Gaussian thermostat variable (Lagrange multiplier) obtained by solving the constraint equation P P i; i Fi pP _ (t) i pxi pyi = (15) i pi pi Thus, we assume isokinetic energy. The shear viscosity (_ P T ) is de ned as the ratio of the shear component of the stress tensor xy and the applied shear rate _ , ( ) = ; h xy i _ Here where Wxy is the virial
N 1 X pxipyi + W xy = xy i=1 mi

(16)

(17) (18)

and is the strain tensor. Our calculations lead to a square root dependence of viscosity dependence on _ shear rate, p (T ) = 0 + 1 _ (19) 19-4

Wxy = @@U

xy

(where the coe cients depend on temperature and density). Thus, we nd nonNewtonian behavior. This asymptotic square root relationship for 3-dimensional simple uids studied by NEMD under shear identi ed by Evans et. al.14 .

2.3 Method

We considered temperatures of 1500K, 1750K, and 2000K and alloys Aux ;Cu1;x with x = 0, 0.25, 0.5, 0.75, and 1.0. The NEMD simulation was carried out in a cubic box (500 atoms per unit cell) subject to periodic boundary conditions. The unit cell length was based on extroplating the experimental density at the corresponding temperature.15 For alloys we assume that the density is given by the linear combination rule
T (Aux Cu1;x) = T (Au)

x+

T (Cu)

(1 ; x)

(20)

The densities values used in the calculations are listed in Table 2. To ensure that the system is equilibrated for the liquid state, we started with a random face-centered cubic (fcc) alloy, raised the system to 300K above the melting point, and equilibrated by constant temperature, constant volume (ThN) simulation, for 5ps (5000 steps of 1fs) without shear (normal MD). This provided the starting structure for each NEMD simulation. We considered applied shear rates of _ = 2, 1, 0.5, and 0.25 (1/ps). For each shear rate and temperature, we rst equilibrated (with NEMD) for NEQ steps followed by Nmeas = 9NEQ steps which were used to calculate properties. As the shear rate decreases we observed larger uctuations and slower convergence. Thus, we increased the number of steps. The total number of steps ranged from 20,000 (_ = 2) to 100,000 (_ = 0:25), as shown in Table 3. Figure 1 shows the convergence of the calculated viscosity for Cu at 2000K. We also partition each simulation into four segments to calculate a standard deviations for the viscosity. These results lead to the error bars shown in the gures the largest deviation is less than 5% of the average viscosity.

The calculated viscosities for the various CuxAu1;x alloys (x = 0%, 25%, 50%, 75%, and 100%) are shown in Figure 2 for various shear rates and temperatures. This clearly shows the nonNewtonian behavior of liquid metals and alloys. 19-5

3.0 Results and Discussion 3.1 Comparison with Experiment

Fitting Eq. (19) to the calculations leads to the viscosities at zero shear rate, 0 in Table 4. We nd good agreement with experiment18 for Cu (within 3.6%) and fair agreement for Au (within 14.6%). This may be because the FF parameters for Cu lead to a better description of the stacking faults and other properties than for Au or it may be because the densities are inaccurate. The 0 and 1 obtained for the alloys CuxAu1;x are shown in Figure 3. The viscosity isotherm follows the trend of the liquidus and the density, as predicted by the Hirai estimation formula.1 2 The relation between ln( ) and 1=T is linear, as shown in Figure 4. The viscosity in this region can be expressed as

3.2 Temperature Dependence

G = A exp RT

(21)

The values for A and G are tabulated in Table 5 compared with the measured values.17;19 The energy of activation, G, is a function of the composition of the melt. To understand the origin of the temperature dependence, we considered the speci c alloy Cu1Au3 and considered the dependence of viscosity on temperature, density, and structure. Figure 5a shows that starting with two di erent random fcc structures leads to an 0 within 1.4%. This suggests that the precision in the calculated viscosity (using the same structure and density) is 1.5%. We considered simulations using a xed density ( xed at the value appropriate for 1750K), but di erent temperatures (1500-2000K). Figure 5b shows that the largest di erence in 0 is about 1.3%. This indicates that 0 depends strongly on density (i.e. pressure) the dependence on temperature is mostly through the change of density with temperature. Thus, using a xed temperature and di erent densities (corresponding to 1500-2000K) gives essentially the same 0 as allowing the temperature and density to to readjust together. This shows that the dominant factor in determining the viscosity of liquid metals is the density! Thus, the variation of viscosity with temperature is caused by the variation in density with temperature. This implies that the dependence of viscosity of liquid metals on pressure can easily obtained from the temperature dependence at low pressure. Thus, to calculate the viscosity using NEMD we must be sure to have the correct density at that temperature. Figure 3b shows that the nonNewtonian coe cient, 1, (viscosity as a function of shear rate) is almost independent of T and structure. 19-6

4.0 Summary

We calculated the viscosity of liquid Au-Cu alloys from NEMD simulations using the Q-SC many-body FF. The agreement with experimental viscosities of Cu at di erent temperature suggest that the NEMD and Q-SC can provide useful predictions. We nd that the temperature of the viscosity of liquid metal is determined by the density (and composition). We also nd that the viscosity isotherm of the alloy follows the trend of the liquidus and the density curve. This research was supported partially by a grant (ARO-DAAH 95-1-0233) from the Army Research O ce and by grants from the NSF (ASC 92-17368 and CHE 9512279). In addition, support for the Materials Simulation Center (MSC) facilities came from DOE-ASCI, ARO-DURIP (DAAG55-97-1-0140), Chevron Petroleum Technology Co., Asahi Chemical, Owens-Corning, Exxon, Asahi Glass, Nippon Steel, Avery Dennison, BP Chemical, and the Beckman Institute. Some calculations were carried out at the Illinois National Center for Supercomputing Applications (NCSA), funded by the National Science Foundation (NSF).

5.0 Acknowledgements

19-7

References
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19.

M. Hirai, Iron and Steel Inst. of Japan Intl. 33, 251 (1993). S. Seetharaman and D. Sichen, Iron and Steel Inst. of Japan Intl. 37, 109 (1997). D. M. Heyes, Comp. Phys. R. 8, 71 (1988). D. J. Evans and G. P. Morriss, \Statistical Mechanics of Nonequilibrium Liquid," Academic Press, London, (1990). T. Koishi, Y. Arai, Y. Shirakawa, and S. Tamaki, J. Phys. Soc. Japan 66, 3188 (1997). J. Mundy, S. Balasubramanian, K. Bagchi, J. I. Siepmann, and M. L. Klein, Faraday Discussions 104, 17 (1996). W. Allen and R. L. Rowley, J. Chem. Phys. 106, 10273 (1997). S. Balasubramanian, C. J. Mundy, and M. L. Klein, J. Chem. Phys. 105, 11190 (1996). S. Sarman, P. T. Cummings, and D. J. Evans, Intl. J. Thermophysics 15, 1125 (1994). A. P. Sutton and J. Chen, Phil. Mag. Lett. 61, 139 (1990). H. Ra i-Tabar and A. P. Sutton, Phil. Mag. Lett. 63, 217 (1991). Y. Kimura, T. Cagin, and W. A. Goddard III, \The Quantum Sutton-Chen ManyBody Potential for Properties of FCC metals," Phys. Rev., to be submitted. J. Evans, Phys. Lett. A 74, 229 (1970). J. Evans and H. J. M. Hanley, Phys. Rev. A 20, 1648 (1979) D. J. Evans, H. J. M. Hanley, S. Hess, Phys. Today 37, 26 (1984). Handbook of Chemistry & Physics 51st edition, Chemical Rubber Co. (1971). Handbook of Chemistry & Physics 74th edition, CRC Press (1991). Smithells Metals Reference Book, 6th editon, Butterworths Co. (1983). Handbook of Physico-Chemical Properties at High Temperatures, The 140th Committee of Japan Society for Promotion of Science, Iron and Steel Inst. of Japan Intl., Tokyo, (1988). E. Gebhardt and G. Worwag, Z. Metallkunde 42 358, (1951)

19-8

Table 1. Quantum Sutton-Chen (Q-SC) force eld parameters.12

D (meV) Cu Au 5.7921 7.8052 c 84.843 53.581 m 5 8 n 10 11 (A) 3.6030 4.0651

Table 2. The densities (g/cm3) used in the simulation were extrapolated from reference 15 using the experimental values at lower temperatures and Eq. (20). T (K) Cu Cu3Au1 Cu1Au1 Cu1Au3 Au 1500 7.953 10.240 12.518 14.812 17.085 1750 7.716 9.960 12.202 14.444 16.687 2000 7.481 9.689 11.900 14.107 16.316

Table 3. Simulation steps for various shear rates.

Shear Rate (1/ps) 2 2000 20000 1 4000 40000 0.5 8000 80000 0.25 16000 160000

NEQ a Ntotal = NEQ + Nmeasb

a Steps to equilibrate. b Steps for measurement of properties.

19-9

Table 4. Viscosity at zero shear rate,

Cu T (K) Sim. Exper. 1500 3.68 3.6 1750 2.82 2.72 2000 2.24 2.22

(mPas). The experimental data at these temperatures were calculated from fomulae given in reference 18.
0

Cu3Au1 Cu1Au1 Cu1Au3 Au Sim. Sim. Sim. Sim. Exper. 2.64 2.88 3.97 5.86 4.5 2.29 2.47 3.21 4.56 3.7 1.87 2.04 2.70 3.67 3.2

Table 5. Temperature dependence of viscosity calculated for CuxAu1;x alloys (see

Eq. 21). Cu 24.15 0.5298 Cu3Au1 16.56 0.7071 Cu1Au1 16.05 0.7711 Cu1Au3 18.78 0.8782 Au 23.69 0.949

G (kcal/mol) A (mPas)

19-10

Figure Captions Figure 1. Dependence of calculation viscosity on length of NEMD simulation. For Figure 2.

Figure 3. Figure 4. Figure 5.

shear rates of 2/ps and 0.5/ps on Cu at 2000K. (500 atoms per periodic cell). The calculated viscosities for various shear rates. (a) Cu. (b) Cu3Au1 (x = 25%). (c) Cu1Au1 (x = 50%). (d) Cu1Au3 (x = 75%). (e) Au. The calculated values open symbols (at 6= 0)] were based on three independent runs. The lled symbols (at = 0) are calculated from experiment18 using the A and G tted to the experimental data. The lines are based on Eq. (19). (a) The viscosity at zero shear rate ( 0 ) for Cu-Au alloys at various temperatures. (b) The nonNewtonian coe cient ( 1) in Eq. (19) for Cu-Au alloys at various temperatures. Viscosity of liquid Cu-Au alloys as a function of temperature. Viscosity of Au3Cu1. (a) Correct density for each simulation temperature, but starting from two di erent random structures. (b) The density is kept xed at the value of 1750K, but the calculation temperature is varied. (c) The calculation temperature is xed, but the density is varied among the values appropriate for 1500K, 1750K, and 2000K.

19-11