NUCLEATION THEORY USING EQUATIONS OF STATE
by
ABDALLA A. OBEIDAT
ATHESIS
Presented to the Faculty of the Graduate School of the
UNIVERSITY OF MISSOURI-ROLLA
in Partial FulfillmentoftheRequirementsfortheDegree
DOCTOR OF PHILOSOPHY
in
PHYSICS
2003
Gerald Wilemski, Advisor Barbara N. Hale
Jerry L. Peacher Paul E. Parris
Daniel Forciniti ii
ABSTRACT
Various equations of state (EOS) have been used with the most general Gibbsian form (P form) of classical nucleation theory (CNT) to see if any improvement − could be realized in predicted rates for vapor-to-liquid nucleation. The standard or
S form of CNT relies on the assumption of an incompressible liquid droplet. With − the use of realistic EOSs, this assumption is no longer needed. The P form results − for water and heavy water were made using the highly accurate IAPWS 95 EOS − and the CREOS.TheP form successfully predicted the temperature (T ) and − supersaturation (S) dependence of the nucleation rate, although the absolute value was in error by roughly a factor of 100. The results for methanol and ethanol using alessaccurateCPHB EOS showed little improvement over the S form results. − Gradient theory (GT ), a form of density functional theory (DFT), was applied to water and alcohols using the CPHB EOS. The water results showed an improved
T dependence, but the S dependence was slightly poorer compared to the S form − of CNT. The methanol and ethanol results were improved by several orders of magnitude in the predicted rates. GT and P form CNT were also found to be in − good agreement with a single high T molecular dynamics rate for TIP4P water.
The P formof binary nucleation theory was studied for a fictitious water-ethanol − system whose properties were generated from DFT and a mean-field EOS for a hard sphere Yukawa fluid. The P form was not successful in removing the unphysical − behavior predicted by binary CNT in its simplest form. The DFT results were greatly superior to all forms of classical theory. iii
ABSTRACT
Various equations of state (EOS) have been used with the most general Gibbsian form (P form) of classical nucleation theory (CNT) to see if any improvement − could be realized in predicted rates for vapor-to-liquid nucleation. The standard or
S form of CNT relies on the assumption of an incompressible liquid droplet. With − the use of realistic EOSs, this assumption is no longer needed. The P form results − for water and heavy water were made using the highly accurate IAPWS 95 EOS − and the CREOS.TheP form successfully predicted the temperature (T ) and − supersaturation (S) dependence of the nucleation rate, although the absolute value was in error by roughly a factor of 100. The results for methanol and ethanol using alessaccurateCPHB EOS showed little improvement over the S form results. − Gradient theory (GT ), a form of density functional theory (DFT), was applied to water and alcohols using the CPHB EOS. The water results showed an improved
T dependence, but the S dependence was slightly poorer compared to the S form − of CNT. The methanol and ethanol results were improved by several orders of magnitude in the predicted rates. GT and P form CNT were also found to be in − good agreement with a single high T molecular dynamics rate for TIP4P water.
The P formof binary nucleation theory was studied for a fictitious water-ethanol − system whose properties were generated from DFT and a mean-field EOS for a hard sphere Yukawa fluid. The P form was not successful in removing the unphysical − behavior predicted by binary CNT in its simplest form. The DFT results were greatly superior to all forms of classical theory. iv
ACKNOWLEDGEMENTS
I would like to express my gratitude and appreciation to my research advisor, Dr.
Gerald Wilemski, for his constant support and patience through the time I spent at
UMR. Without his guidance and motivation this work would have never been done.
I would like also to thank the members of my Ph.D. committee Dr. B. Hale, D. P.
Parris, Dr. J. Peacher, and Dr. D. Forciniti for their help and support. I also would like to thank Dr. J-S. Li for his support and suggestions during my thesis work.
I am very thankful to my parents, without their endless love and support, I would not have been neither in UMR nor in this life.
I would like also to thank my roommates and friends, Eyad, Malik, Ahmad, Abdul,
Vikas, and Sabrina, who made my life much easier while staying in the US.
This dissertation is dedicated to the wonderful lady Enas.
This work was supported by the Engineering Physics Program of the Division of Materials Sciences and Engineering, Basic Energy Sciences, U. S. Department of
Energy. v
TABLE OF CONTENTS
Page
ABSTRACT...... iii
ACKNOWLEDGEMENTS ...... iv
LIST OF ILLUSTRATIONS ...... viii
LISTOFTABLES...... xii
SECTION......
1. INTRODUCTION...... 1
1.1. NUCLEATION PHENOMENOLOGY AND BASIC THEORY 1
1.2.BRIEFOVERVIEWOFBINARYNUCLEATION..... 8
1.3.MOTIVATION...... 10
2. EQUATION OF STATE APPROACH FOR CLASSICAL NUCLEATIONTHEORY...... 12
2.1.THEORY...... 12
2.1.1.WorkofFormation...... 12
2.1.2.Gibbs’sReferenceState...... 15
2.3.1.NumberofMoleculesinCriticalNucleus...... 17
3. EQUATIONS OF STATE FOR UNARY SYSTEMS ...... 19
3.1.WATER...... 19
3.1.1.IAPWS-95...... 19
3.1.2.CrossOverEquationofState(CREOS-01)...... 20
3.1.3. JefferyandAustinEOS(JA—EOS)...... 21
3.1.4.CubicPerturbedHardBody(CPHB)...... 22 vi
3.1.5.Peng-Robinson(PR)...... 23
3.2.HEAVYWATER:CREOS-02...... 24
3.3. METHANOL AND ETHANOL: CPHB ...... 24
4. RESULTS OF EOS APPROACH FOR UNARY SYSTEMS . . 25
4.1.WATER...... 25
4.2.HEAVYWATER...... 29
4.3.DISCUSSIONOFWATERRESULTS...... 32
5. GRADIENTTHEORYOFUNARYNUCLEATION...... 36
5.1.BACKGROUND...... 36
5.2.THEORY...... 37
6. RESULTS OF GRADIENT THEORY FOR UNARY NUCLEATION...... 40
6.1.WATERANDTIP4PWATER...... 40
6.1.1. Planar and Droplet Density ProfilesfromGT..... 40
6.1.2.WaterNucleationRates...... 44
6.1.3.TIP4PWaterNucleation...... 47
6.2.COMPARISONOFTHEWATEREOS...... 48
6.3. RESULTS FOR METHANOL AND ETHANOL ...... 52
7. BINARYNUCLEATIONTHEORY...... 55
7.1CLASSICALNUCLEATIONTHEORY...... 55
7.2DENSITYFUNCTIONALTHEORY(DFT)...... 58
7.3SURFACETENSIONANDREVERSIBLEWORK..... 59 vii
7.4DFTFORHARDSPHERE-YUKAWAFLUID...... 60
8. PROPERTIES OF THE MODEL BINARY HARD-SPHERE YUKAWA(HSY)FLUID...... 61
8.1EQUATIONOFSTATE...... 61
8.2FITTEDPROPERTYVALUES...... 66
9. RESULTSOFTHEHSYBINARYFLUID...... 67
9.1.CRITICALACTIVITIESATCONSTANTW*...... 67
9.2. NUMBER OF MOLECULES IN CRITICAL DROPLET . . 69
10.CONCLUSIONS...... 71
APPENDICES......
A.IMPORTANTTHERMODYNAMICRELATIONS...... 73
B.DETAILSOFVARIOUSEQUATIONSOFSTATE...... 77
C. PHYSICAL PROPERTIES OF WATER AND HEAVY WATER 89
D. SURFACE TENSION AND WORK OF FORMATION IN DFT 91
BIBLIOGRAPHY...... 95
VITA...... 100 viii
LIST OF ILLUSTRATIONS
Figures Page
1.1. Schematic pressure — volume phase diagram for a pure substance. The green solid line is a true isotherm whose intersections (e) with the bin- odal dome give the equilibrium pressure and volumes of the coexising vapor-liquid phases. Binodal: solid heavy curve; spinodal: red dashed curve...... 1
1.2. The contributions of the surface and volume terms of the free energy of formation versus the cluster size. The free energy of formation has amaximumatthecriticalsize...... 4
1.3. Experimental data for water from Ref.[18] illustrating the inadequate temperature dependence predicted by the classical Becker-Doering the- ory[4], labeled S-form in the figure...... 6
2.1. Same as Figure 1.2 but the free energy of formation is plotted as a functionoftheradiusofthecluster...... 13
2.2. The concept of the reference liquid state using a pressure-density isotherm for a pure fluid. The full circles represent the equilibrium vapor-liquid states, while the diamonds mark the metastable vapor phase and the referenceliquidphase...... 17
4.1. The work of formation for water droplets using the IAPWS-95 EOS withthethreeformsofCNTatT=240,250,and260K...... 25
4.2. Comparison of the experimental rates of Woelk and Strey (open circles) for water with two versions of CNT based on the IAPWS-95 EOS; P- formandS-form...... 26
4.3. Comparison of the experimental rates of Woelk and Strey (open circles) for water down to T=220 K with two versions of CNT based on the CREOS-01andwiththescaledmodel...... 27
4.4. The number of water molecules in the critical cluster as predicted by the nucleation theorem and the P-form calculations. The dashed-line showsthefullagreementwiththeGibbs-Thomsonequation...... 28 ix
4.5. The experimantal rates of heavy water by Woelk and Strey down to T=220 K with the predictions of the P-form of the CREOS-02. . . . . 30
4.6. The P-form results using CREOS-02 at high S compared with two different sets of supersonic nozzle experiments. The scaled model and the empirical function also shown at T=237.5, 230, 222, 215, and 208.8 K from left to right...... 31
4.7.AsinFigure4.4butforheavywater...... 32
4.8. The temperature-density isobars of water using the IAPWS-95 EOS and the CREOS-01 compared to experimental data of Kell and Whal- ley[75]andPetitetetal.[81]...... 33
4.9. The work of formation of water at T=240, 250, and 260 K predicted bytheIAPWS-95andCREOS-01...... 34
4.10. Isothermal compressibility of liquid water at 10 MPa and 190 MPa calculated from the fit of Kanno and Angell[83]...... 35
6.1. The thickness of flat water interfaces at different T using GT, MD simulations[90],andexperimentaldata[91]...... 40
6.2. Density profiles of water droplets predicted by CPHB at different T, forasupersaturationratioof5...... 42
6.3.SameasFigure6.2butatS=20...... 42
6.4. Density profiles of water droplets at T=350 K for different values of thesupersaturationratiousingtheCPHBEOS...... 43
6.5.SameasFigure6.4butusingtheJA-EOS...... 44 x
6.6. Nucleation rate predictions of the CPHB using the P-form and the GT comparedtoexperimentaldataofWoelkandStrey[18]...... 45
6.7. The ratio of the GT work of formation to that of the P-form of CNT asafunctionofsupersaturationratioat260K...... 46
6.8. The number of water molecules in the critical cluster as predicted by the nucleation theorem and the GT calculations. The dashed line rep- resentsfullagreementwithGibbs-Thomsonequation...... 47
6.9. Nucleation rates for GT and two forms of CNT at T=350 K using different EOSs, as shown in the figure, compared with the MD rate for TIP4P water and the result of the P-form of CNT using CREOS-01. . 48
6.10. The predictions of different EOSs for the equilibrium liquid density of water at different T compared to the experimental data generated usingtheIAPWS-95...... 49
6.11. Density of liquid water using the CPHB EOS (stars) at different P (0.1, 50, 100, 150, 190 MPa) compared to the experimental data calculated usingtheIAPWS-95(circles)...... 49
6.12. The predictions of different EOSs for the equilibrium vapor pressure at different T compared to the experimental data calculated by using theIAPWS-95...... 50
6.13. Same as for Figure 6.12 except for the equilibrium vapor density. . . . 50
6.14. Experimental nucleation rates of methanol compared to the predictions ofGTandtheP-formofCNTwiththeCPHBEOS...... 52
6.15.AsinFigure6.14butforethanol...... 53
6.16. Liquid ethanol density vs. P at different temperatures using the CPHB EOS(opensymbols)andexperimentaldata(solidsymbols)...... 53
6.17. Experimental nucleation rates of ethanol compared to calculated rates using the S-form and the P-form of CNT with the CPHB EOS and the P-form of CNT using fittedexperimentaldensitydata[94]...... 54 xi
8.1. The total and partial equilibrium vapor pressures of the HSY model fluidatT=260Kversusmixturecomposition,x...... 62
8.2.P-xphasediagramofthebinaryHSYmodelsystem...... 63
8.3. Surface tension for the pseudo water-ethanol system and measured valuesforwater-ethanolversusethanolmolefraction,x...... 64
8.4. Variation of the partial molecular volume of p-water with composition. 65
8.5.SameasFigure8.4.butforp-ethanol...... 65
9.1. Critical activities of p-water (1) and p-ethanol (2) needed to produce aconstantworkofformationof40kT...... 67
9.2. The number of molecules of each component of the critical droplet as a function of the p-water activity using version 1 and version 2 of the CNT,aswellastheDFT...... 69
9.3. The number of molecules of each component of the critical droplet as a function of the p-water activity using version 3 of the CNT and the DFT...... 70
9.4. The number of molecules of each component of the critical droplet as a function of the p-water activity using versions 1, 2, and 3 of the CNT. 70
A.1. Schematic depiction of a spherical critical nucleus in a metastable gas phase...... 76 xii
LIST OF TABLES
Tables Page
B.1. The coefficientsvaluesoftheidealgaspart...... 78
B.2. The coefficientsandparametersoftheresidualpart...... 79
B.3. The other coefficientsandparametersoftheresidualpart...... 81
B.4. The coefficientsoftheCREOSequationofstate...... 83
B.5. The coefficientsofCREOS-01andCREOS-02EOSs...... 84
B.6. The coefficientsandparametersoftheJA-EOS...... 85
B.7. The C parameters for water, ethanol, and methanol of the CPHB EOS. 87
B.8. The parameters of the CPHB EOS used for water, ethanol, and methanol...... 88 1. INTRODUCTION
1.1 NUCLEATION PHENOMENOLOGY AND BASIC THEORY This thesis is primarily concerned with the accuracy of theories of vapor-to-liquid nucleation based on equations of state of real fluids. Nucleation refers to the kinetic processes involved in the initiation of first order phase transitions in nonequilibrium systems. Two phase equilibrium states for a pure substance, e.g. vapor and liquid, occur at unique pairs of temperature T and pressure P . For two-phase vapor- liquid equilibrium, the pressure is referred to as the equilibrium vapor pressure Pve.
Now, if the actual pressure of the vapor Pv is larger than the equilibrium vapor pressure, the vapor is said to be supersaturated. This new state of the vapor is either metastable or unstable. These two types of states are distinguished by their location with respect to the mean-fieldspinodal,whichisillustratedinFigure1.1.
c riti c a l p P oi nt
T 6 T 5
M
e e l T =T t 4 c b a S a T t 3 t a e b e P aS t e l e T e 2 Unstable M T 1 0 0 V V V le ve
Figure 1.1. Schematic pressure — volume phase diagram for a pure sub- stance. The green solid line is a true isotherm whose intersections (e) with the binodal dome give the equilibrium pressure and volumes of the coexising vapor-liquid phases. Binodal: solid heavy curve; spinodal: red dashed curve.
This figure shows a P -V phase diagram for a pure fluid with several isotherms 2 based on a van der Waals equation of state (EOS). The heavy dome-shaped curve denotes the binodal line, the locus of two-phase, vapor-liquid equilibrium states, which ends at the critical point. The true isotherms are flat in the two-phase re- gion inside the dome. All mean-field EOSs are similar in that they artificially treat the fluid as a homogeneous phase with a continuously varying density inside the two-phase region. This unphysical oversimplification generates the "van der Waals loops" shown by the isotherms. The spinodal boundary separates mechanically sta- ble regions (metastable states for which (∂P/∂V )T 0) from mechanically unstable ≤ regions (for which (∂P/∂V )T > 0) as determined by the slope of the isotherms of the mean-field EOS. If the supersaturated vapor is in contact with its own liquid, it will condense until the vapor again reaches saturation. If a container of volume V contains only pure vapor, at a suitably large super- saturation value S = Pv/Pve, droplets will start to form at an appreciable rate as a result of collisions among vapor molecules. This process of forming a droplet is known as homogeneous nucleation. If impurities are also present in the container, the supersaturated vapor will first condense on them in a process referred to as het- erogeneous nucleation. Since nucleation plays a key role in many fields ranging from atmospheric applications to materials science, the study of nucleation has a long his- tory and is currently receiving much attention stimulated by the development of new experimental and theoretical techniques to measure and predict nucleation rates. The first comprehensive treatment of the thermodynamics of the nucleation process is due to Gibbs[1]. Gibbs showed that the reversible work required to form a nucleus of the new phase consists of two terms: a bulk (volumetric) term and a surface term. Later, in 1926 Volmer and Weber[2] developed the first nucle- ation rate expression, by arguing that the nucleation rate should be proportional to the frequency of collisions between condensable vapor molecules and small droplets (critical clusters) of the new phase of a size, the critical size, that just permits spon- taneous growth. A more detailed kinetic approach for the process of vapor-to-liquid nucleation was subsequently developed by Farkas[3]. The theory of Volmer, We- ber, and Farkas was extended a few years later by Becker and Döring[4], Frenkel[5], Zeldovich[6], and is now known as classical nucleation theory (CNT). 3
The basic kinetic mechanism assumed by these authors was that small clusters grow and decay by the absorption or emission of single molecules. In this theory the clusters are treated as spherical droplets. As in Gibbs’s treatise, the work of formation of a critical droplet consists of volumetric and surface contributions, but in the absence of knowledge of the microscopic cluster properties, bulk thermodynamic properties are used to evaluate the two contributions. Gibbs’s result for W is
W = Aγ Vl(Pl Pv) ,(1) − − 2 4π 3 W =4πr γ r (Pl Pv) , (2) − 3 − and it strictly applies to just the droplet of critical size. The surface term Aγ represents the free energy needed to create a surface. This term always opposes droplet formation. The volume term -Vl(Pl Pv) represents the stabilizing free − energy obtained by forming a fragment of new phase. The negative sign before the volume term ensures that new phase formation ultimately lowers the free energy of the system. In developing the kinetics of nucleation, it is necessary to know the free energy of formation of droplets of noncritical size. The simplest approximation is to assume that Gibbs’s result for W applies to all sizes and to rewrite it in terms of n, the number of molecules in the droplet. In terms of the spherical liquid droplet model, the surface area and volume are straightforward to rewrite since r n1/3 ∝ for this model. It is customary to assume that the droplet is an incompressible liquid and to replace the pressure difference by the difference in chemical potential between the old and new phases at the same pressure Pv, as explained in more detail later. It is generally a very good approximation to treat the vapor phase as ideal, so that the chemical potential difference can then be expressed in terms of the supersaturation ratio S. One other approximation is needed: the surface tension of a planar interface is used to evaluate the surface term because the surface tension of microscopic droplets is unknown. With these simplifications the free energy of formation of a cluster of n molecules is
∆G γ A 2/3 = ∞ n ln S = θn n ln S ,(3) kT kT − − 4 where γ is the surface tension of a planar gas-liquid interface, A is the surface ∞ area of the cluster, which is estimated from the liquid number density ρl,andθ = 1/3 2/3 (36π) γ ρ−l /kT. The dependence of ∆G on n is illustrated in Figure 1.2. ∞
surface term Free Energy of Formation Free Energy 0 n 0 n*
volume term
Figure 1.2. The contributions of the surface and volume terms of the free energy of formation versus the cluster size. The free energy of formation has a maximum at the critical size .
Asseeninthefigure, ∆G has a maximum at the value n = n∗,knownasthe critical size. If the cluster size n is less than n∗, the surface term is dominant. As a result, the cluster has a higher tendency to shrink, i.e., to reduce its free energy, than to grow, i.e., to increase its free energy. On the other hand if n>n∗,the volume term is dominant, and the cluster has a higher tendency to grow than to shrink. From the extremum condition, [d∆G/dn] =0, one obtains the simple n∗ relation for the critical size n∗,
2θ (n )1/3 = , (4) ∗ 3lnS which is equivalent to the Kelvin equation for the critical radius r∗ 2γ r∗ = ∞ .(5) ρlkT ln S 5
The barrier height is equal to the work of formation of the critical droplet, ∆G∗ =
W ∗. Volmer and Weber[2] in 1926 argued that the nucleation rate depends expo- nentially on the work of formation of the droplet. The nucleation rate expression, which derives from the work of Becker-Döring[4], Frenkel[5], Zeldovich[6], is often referred to as Becker-Döring theory. The expression is given by Abraham[7], for example, as
W ∗ JCL = J0 exp , (6) − kT µ ¶ with the pre-exponential factor
2 2γ Pv J = ∞ v , (7) 0 πm l kT r µ ¶ where m is the mass of a condensible vapor molecule, vl(= 1/ρl) is the molecular volume, Pv is the vapor pressure, and the barrier height of nucleation is
16πγ3 W ∗ = ∞ 2 . (8) 3(kTρl ln S) As seen from the nucleation expression, all the inputs to the theory are exper- imental quantities which makes the theory easy and popular to use. For many years, it was impossible to measure nucleation rates accurately. Instead, what was usually determined experimentally was the critical supersaturation at which nucle- ation was observable at a significant rate, whose value was typically not known to better than one or two orders of magnitude. (One can see from Eqs.(6) and (8) that J depends sensitively on S, but that S is rather insensitive to changes in J.) These critical supersaturation measurements were generally in good agreement with the predictions of CNT for many substances. Over the past twenty years, the de- velopment of new experimental techniques with improved precision has allowed the accurate measurement of nucleation rates for many substances[8—16]. Comparison of these results with the predictions of CNT has shown that the theory is usually in error, giving rates that are too low at low temperatures and too high at high temperatures[10, 17, 18], as illustrated in Figure 1.3. 6
1010 260 K 250 K H O 2 Expt S-form
9 10 240 K ) -1 230 K s -3 108
J (cm 107 T=220 K
106 5 10152025 S
Figure 1.3. Experimental data for water from Ref.[18] illustrating the inadequate temperature dependence predicted by the classical Becker- Doering theory[4], labeled S-form in the figure.
Due to limits of CNT, there has been much efforttoimprovetheclassicalmodel, but most of the newer models take the form of correction factors to CNT. [19, 20]. In addition, the improvements of these models are often substance specific, which limits their applicability. One of the most successful and most general treatments of the temperature dependence of nucleation rates is the so-called scaled model of Hale[21, 22] introduced in 1986. The scaled model is based on CNT, and it yields a universal dependence of nucleation rate on Tc/T 1.Thetwoparametersof − this model are the nearly universal constant Ω, which is interpreted as the excess 26 3 1 surface entropy per molecule, and the constant rate prefactor J0( 10 cm− s− ). ≈ The value of Ω for nonpolar substances is around 2.2, whereas for polar materials it is about 1.5. For later use, and as an example, Ω is 1.476 for heavy water and 1.470 for water. The model works well for many substances for which the CNT fails. In the scaled model, the nucleation rate is given by the expression, 7
3 16π 3 Tc 2 J = J0 exp Ω 1 /(ln S) . (9) − 3 T − Ã µ ¶ ! The most fundamental approach to improving CNT is through the development of microscopic theories. The goal of any microscopic theory is to avoid the overly simplistic use of bulk thermodynamic properties in evaluating the free energy of clus- ter formation. There are several different types of microscopic approaches, which will only be mentioned here. Molecular dynamics (MD) computer simulations have been used to explore properties of small molecular clusters, e.g. by Gubbins and coworkers[23, 24] and Tarek and Klein[25], and to directly simulate nucleation, as in the work of Rao and Berne[26], Yasuoka and Matsumoto[28], and ten Wolde and Frenkel[29]. Monte Carlo (MC) computer simulations have also been used ex- tensively to calculate free energies of cluster formation, e.g. by Lee et al.[27, 30], Hale et al.[31, 32], and Oh and Zeng[33], and to examine the subcritical cluster size distribution directly[34]. Hybrid approaches like those of Weakliem and Reiss[35], Schaff et al.[36] that combine MC or MD simulation results with analytical theory have also been developed. A brief review by Reiss[37] discusses other approaches by many other authors not mentioned here. Another important approach known as the density functional theory (DFT)[38— 40] will be discussed later in detail. To briefly summarize here, in DFT the free energy of the nonuniform system, F [ρ(r)], is written as a functional of the local density ρ(r) at each position r in the fluid. The presence of the nucleus renders the fluid inhomogeneous. The inhomogeneity is characterized by the density that varies continuously from its value at the center of the nucleus to its value in the metastable mother phase far from the nucleus. The properties of the critical nucleus are determined by finding the density profile that minimizes the nonuniform fluid’s free energy. Cahn and Hilliard[40] were the first who developed a type of DFT for nucleation theory. They proposed the Helmholtz potential to be
c 2 F [ρ (r)] = dr f0 [ρ (r)] + [ ρ (r)] ,(10) 2 ∇ Z ³ ´ 8
where f0 is the Helmholtz free energy density of the homogenous fluid of density ρ, ρ is the gradient of the density, and c is the so-called influence parameter related ∇ to the intermolecular potential. Because the Helmholtz potential above depends on the gradient of the density, minimizing Eq.(10) one obtains a differential Euler-Lagrange equation. This theory is called gradient theory (GT ), or square gradient theory, because of the form of the free energy functional. The first GT was actually devised many years earlier by van der Waals[41] to describe the structure of planar interfaces. To apply the GT ,one needs a well-behaved EOS. It should have the form of a cubic equation, similar in spirit to the van der Waals EOS, that describes the system as a single homogeneous phase whose density varies continuously throughout the two-phase region. A more general form of DFT was developed and applied to nucleation theory by Oxtoby and coworkers[42, 43]. It is a molecular theory that explicitly uses an intermolecular potential. The theory uses a hard sphere fluid as a reference state and treats the attractive intermolecular potential as a perturbation. The theory is developed in terms of the grand potential Ω, which is written in the perturbation theory as the following functional of the density,
Ω [ρ (r)] = dr (fh [ρ (r)] µρ (r)) + drdr0φ ( r r0 )ρ(r)ρ(r0) ,(11) − att | − | Z ZZ Here, fh is the Helmholtz free energy density of the hard sphere fluid, φatt is the long-range attractive part of the potential, and µ is the chemical potential. The simpler GT canbederivedfromthemoregeneralDFT by expanding the density in a Taylor series and retaining only the leading nonzero terms. Minimization of Eq.(11) generally leads to an integral Euler-Lagrange equation, which must be solved for the density profile of the nonuniform system.
1.2 BRIEF OVERVIEW OF BINARY NUCLEATION Many of the above considerations apply as well to homogeneous nucleation of binary systems, commonly referred to as binary nucleation, but there is a major difference as well. In binary nucleation, the initial metastable phase and the final phase are two component systems. Thus, the kinetics of nucleation involves the 9 formation of clusters of the new phase that generally contain both components. To apply CNT to binary systems, the most important quantity needed to predict nucleation rates is the composition of a critical nucleus. If the surface tension is known as a function of the composition and if ∆P , the difference in pressure inside and outside the droplet, is known, then the critical radius can be calculated using the Laplace formula, ∆P =2γ/r∗. Assuming the droplet is incompressible, the
Gibbs-Thomson equations can be derived: ∆µ = 2γvi/r. The differences in the i − chemical potential between liquid and vapor phases are represented by ∆µi,whilevi is the molecular volume of component i in the liquid. From the two Gibbs-Thomson equations, one can determine the composition and the critical radius of the droplet. In 1950, Reiss[44] proposed a theory based on kinetic and thermodynamic argu- ments showing that the binary nucleation rate is determined by the passage over a saddle point in the two-dimensional droplet size space. Later, Doyle[45] used this theory to study the sulfuric acid-water system, but the Gibbs-Thomson equations he found contained a term involving the compositional derivative of the surface ten- sion. Because these terms were small for the sulfuric acid-water system, they had essentially no effect on the calculated critical cluster compositions. When Doyle’s equations were subsequently applied to strongly surface active systems, such as ethanol-water or acetone-water, these terms became very important for water-rich cluster compositions. As a result, the theoretically calculated vapor compositions needed to produce experimentally observed nucleation rates were many orders of magnitude lower in the concentration of the surface active component than the ex- perimental concentrations. Renninger, Hiller, and Bone[46] argued that Doyle’s treatment of the Gibbs-Thomson equations was inconsistent. Wilemski[47] pro- posed a revised classical theory in which the Gibbs surface adsorption equation was used to cancel the derivative of the surface tension, thus permitting the conventional Gibbs-Thomson equations to be recovered. It is interesting to note that the con- ventional Gibbs-Thomson equations had been used in the original, early work on binary nucleation by Flood[48] and by Döring and Neumann[49], but had then been forgotten. 10
The predictions of either version of CNT for ideal binary mixtures are fairly rea- sonable, but for mixtures with a component that strongly segregates on the droplet surface, e.g. alcohol-water or acetone-water mixtures, problems arise. Doyle’s ver- sion of CNT predicts unrealistic results, as just noted, while the revised binary CNT gives rise to unphysical behavior that violates the nucleation theorem[50] for binary systems. In an important step to resolving these difficulties, Laaksonen[51, 52] pro- posed a so-called explicit cluster model to study water-alcohol systems. The model makes realistic predictions for the vapor concentrations while predicting physical behavior for the nucleus composition, in accord with the nucleation theorem. Compared to unary nucleation, less work on microscopic theories of binary nucle- ation has been performed. Zeng and Oxtoby[43] extended the DFT to treat binary nucleation for Lennard-Jones mixtures. Talanquer and Oxtoby have used the GT to study highly nonideal binary systems with parametrized hard-sphere—van der Waals EOS[53]. Napari and Laaksonen have recently performed DFT calculations for a site-site interaction model that simulates systems with a highly surface active component.[54] Hale and Kathmann have performed Monte Carlo simulations to calculate the free energies of formation of sulfuric acid-water clusters[55]. 1.3 MOTIVATION The principal goal of this thesis is to test a form of classical nucleation the- ory closest in spirit to the original pioneering work of Gibbs. The usual forms of CNT are well-known to provide a poor quantitative description of the temperature dependence of measured nucleation rates, although the predicted dependence on supersaturation is generally quite satisfactory. To explore this, Gibbs’s original for- mula was used to calculate nucleation rates for several different substances: water and heavy water, methanol, ethanol. Significant improvement in the predicted tem- perature dependence of the nucleation rate was realized only for water and heavy water. This appears to be due to the extraordinary isothermal compressibility of these two substances at the low temperatures where nucleation rates are generally measured. The other materials studied are much less compressible at low tem- peratures, and the customary approximation of an incompressible fluid, universally used in the classical theory, is valid for these substances. The implementation of 11
Gibbs’s original formula requires the use of an accurate equation of state for the fluid properties. In the case of water, two different EOS were used, but each accurately treated the anomalously high compressibility of fluid water. With various equations of state available, it was possible to test a nonclassical theory of nucleation known as gradient theory. A second goal of this thesis is to de- termine whether or not the predicted temperature dependence of the nucleation rate would be improved by this simplest form of density functional theory. Reasonably good results were found for water using a so-called CPHB EOS, but the gradient theory results for , methanol, and ethanol were only slightly improved compared to the predictions of classical theory. Finally, the application of Gibbs’s original formula to binary nucleation was ex- plored. The goal of this aspect of the work is to see whether certain unphysical aspects of classical binary nucleation theory could be alleviated by using a more exact formulation of the theory. A key difficulty in carrying out this phase of the research was that for the most interesting binary systems, such as the ethanol-water system, there are no accurate EOSs in the temperature range of interest. To sur- mount this difficulty, a model system was devised with properties resembling those of the ethanol-water system. The EOS for the model system consists of a binary hard sphere fluid contribution plus an attractive term of the van der Waals form. The bulk surface tension was computed as a function of mixture composition using density functional theory for a planar interface. To facilitate the DFT calculations, attractive potentials of the Yukawa form were employed. The results showed that Gibbs’s original formula, with the bulk surface tension, also suffered from the same unphysical behavior as simpler forms of the classical binary theory. 12
2. EQUATION OF STATE APPROACH FOR CLASSICAL NUCLEATION THEORY
2.1 THEORY In this chapter, three different versions of classical nucleation theory (CNT) are explored to study nucleation rates of water and heavy water. For two of these versions,anovelapproachbasedondifferent equations of state is used to calculate the work of formation of a critical droplet, W ∗, which is then used to evaluate the nucleation rate. The theoretical predictions are compared with the experimental rates of water and heavy water[18]. The theoretical results are also compared with the predictions of the scaled model of Hale[21]. The number of molecules in a critical cluster are compared with the experimental data using the nucleation theorem[50]. 2.1.1 Work of Formation.ConsideravolumeV containing N molecules of vapor at a chemical potential µv and pressure Pv. The Helmholtz free energy of this vapor is
Fi = Nµ PvV. (12) v −
Upon forming a droplet with n molecules, if we ignore the very small changes in µv and Pv,thefinal Helmholtz free energy of the system is
Ff =(N n)µ + nµ (V Vl)Pv VlPl + Aγ , (13) − v l − − − where µl is the chemical potential of a molecule at the internal pressure Pl of the droplet, Vl isthevolumeofthedroplet,A is its surface area, and γ is the surface tension. The difference in the free energy between the initial and final systems is
∆F = Ff Fi = n(µ µ ) (Pl Pv)Vl + Aγ . (14) − l − v − − (It should be noted that Eqs.(13) and (14) are not quite rigorous since they fail to include the surface excess number of molecules[50]. As shown in Appendix A, the final results below are, nevertheless, correct.)
This free energy difference has a maximum at a specificradius,r∗,whenµl = µv, 13
Max 4π 3 2 ∆F = (Pl Pv) r∗ +4πr∗ γ ,(15) − − 3 which represents the free energy barrier required to be overcome to form a spher- ical critical droplet of radius r∗. Using results of Appendix A, Eq.(14) can be approximated as the following sum of a surface and a volume term,
4πr3 ∆µ ∆F = +4πr2γ, (16) − 3 vl where ∆µ is the difference in chemical potential between the initial metastable phase and the final stable phase and vl is the molecular volume of the stable phase. Figure 2.1 schematically shows the dependence of this free energy as a function of the droplet radius.
surface term
MAX ∆F Free Energy of Formation Free Energy 0 r 0 r*
volume term
Figure 2.1. Same as Figure 1.2 but the free energy of formation is plotted as a function of the radius of the cluster.
Applying the Laplace equation, which governs the pressure drop across a curved interface, specifically Pl Pv =2γ/r∗(see Appendix A), we then obtain −
16π γ3 W ∗ = 2 ,(17) 3 (Pl Pv) − 14
where W ∗ is the minimum free energy, i.e., the reversible work, required to form a critical droplet of radius r∗. To apply the above formula, which is known as Gibbs formula, one has to know the exact surface tension at that radius and the pressure inside the droplet. Lack- ing knowledge of the exact surface tension, the first approximation is to use the experimental surface tension of a flat interface, i.e., set γ = γ to obtain ∞
16π γ3 W ∗ = ∞ 2 . (18) 3 (Pl Pv) − We call this equation the P form. − Usually the pressure inside the droplet is found approximately by making the assumption that the droplet is incompressible. In this case, we can replace ∆P =
Pl Pv with (µ µ (Pv))/vl, which follows from the thermodynamic identity − v − l
Pl
∆µ = µ µ (Pv)=µ µ (Pv)= vldP, (19) v − l l − l ZPv when the molecular volume vl is assumed to be constant and the condition of un- stable equilibrium between the critical droplet and the metastable vapor is used. Note that this definition of ∆µ is identical to Kashchiev’s[56]. Equation (18) then becomes
3 2 16π γ vl W = ∞ . (20) ∗ 3 (∆µ)2 We call this equation the µ form. This form is most useful if the chemical potential − difference can be found from an equation of state. However, ∆µ is more commonly found using simpler, but approximate thermodynamic relations. If we assume the supersaturated and saturated vapors are ideal gases and that the droplet is a tiny piece of incompressible bulk liquid, then it is easily shown (e.g. in Appendix A) that
∆µ = kT ln S vl(Pv Pve) , (21) − − where k is the Boltzmann constant, T is the absolute temperature, and S is the 15 supersaturation value defined as the ratio of the actual vapor pressure to the equi- librium vapor pressure Pve, i.e., S = Pv/Pve. It is customary to neglect the Pv term, which is almost always extremely small. Equation (20) then reduces to the most familiar form used in CNT,
3 2 16π γ vl W = ∞ . (22) ∗ 3 (kT ln S)2 For simplicity we call this equation the S form. − Applying the first two forms of W ∗ requires knowledge of the actual pressure and chemical potential inside the droplet. Usually this information is unavailable, and for this reason experimentalists compare their results with rates predicted using the S form because the supersaturation ratio is readily determined from the − experimental data.
A less approximate way to evaluate the P form of W ∗ involves calculating the − internal pressure Pl using the equation
Pl
kT ln S = vldP , (23)
PZve which follows from Eq.(19) and the conditions for stable and unstable equilibrium,
µl(Pve)=µv(Pve) ,(24)
µl = µv(Pv) , (25) along with the ideal gas approximation,
µ (Pv) µ (Pve)=kT ln S. (26) v − v The integral on the right-hand-side of Eq.(23) can be evaluated quite accurately if the liquid density or, equivalently, the molecular volume is known as a function of pressure. If the pressure dependence of the density is not available from direct measurements, it may be calculated using the measured liquid isothermal compress- ibility, preferably as a function of pressure. 2.1.2 Gibbs’s Reference State. A more comprehensive approach for cal- culating the pressure and chemical potential differences needed in the P and − 16
µ forms of W ∗ involves using a complete equation of state (EOS).Acom- − plete EOS consists of a functional representation, either analytical or tabular, of the Helmholtz free energy F of the substance as a function of density and tempera- ture. From the Helmholtz free energy, the pressure and the chemical potential are readily calculated from standard thermodynamic identities. Thus, F contains all the information needed to calculate the work formation of a droplet using the first two forms.
The calculation of the internal pressure Pl from an EOS follows Gibbs’s[1] original reasoning[56—58]. Upon forming a droplet within a homogeneous fluidwithuniform chemical potential and temperature, the droplet may be so small that its internal state may not be homogeneous even at the center of the drop. The meaning of the internal pressure and density of the droplet is then obscured, and these values are difficult to determine. To overcome this difficulty, Gibbs introduced the concept of the reference state as the thermodynamic state of a bulk phase whose internal pressure Pref and density ρref are determined by the same conditions that exist for the new phase and the mother phase, i.e., by assuming that the temperature andthechemicalpotentialarethesameeverywhereinthenonuniformsystem. In mathematical terms, the pressure inside the droplet is calculated such that the chemical potentials are equal in both the metastable vapor and reference liquid phases
µv(ρv)=µl(ρref ) ,(27) where ρv is the density of the supersaturated vapor and ρref is the density of the reference liquid state. As a practical matter, one always calculates differences in chemical potential, and because Eq.(27) involves phase densities that generally differ by many orders of magnitude it is convenient to rewrite this equation as an equality of chemical potential differences measured from the common equilibrium state, for which
P(ρve)=P(ρle) ,(28)
µv(ρve)=µl(ρle) ,(29) where ρve and ρle are the equilibrium vapor and liquid densities, respectively. After 17 subtracting the equilibrium value of µ from both sides of Eq.(27), we obtain
µ (ρ ) µ (ρ )=µ (ρ ) µ (ρ ) . (30) v v − v ve l ref − l le
The chemical potentials are calculated from µ =(∂f/∂ρ)T ,wheref is the appro- priate Helmholtz free energy density for the EOS.Onceρref has been found by solving Eq.(30), the reference pressure Pref is straightforward to calculate from the EOS. Figure 2.2 shows the concept of the reference state.
(ρ ,P ) ref ref P =pressure of metastable region v P =pressure at which µ(ρ ) = µ (ρ ) ref l ref v v
P (ρ ,P ) v v P eq
ρ
Figure 2.2. The concept of the reference liquid state using a pressure- density isotherm for a pure fluid. The full circles represent the equilib- rium vapor-liquid states, while the diamonds mark the metastable vapor phase and the reference liquid phase.
Once W ∗ has been evaluated, the nucleation rate can be calculated using Eq.(6). Comparisons of the calculated rates with experimental values will be made in later sections for various substances. 2.1.3 Number of Molecules in the Critical Nucleus. In addition to the nucleation rate, another physical quantity of interest is the size of the critical nu- cleus, which is experimentally determinable from measured nucleation rates using 18 the nucleation theorem in the approximate form[50, 59],
∂ ln J n∗ . (31) ≈ ∂ ln S
The experimentally determined values of n∗ can be compared with the theoretical values based on the different forms of W ∗ using the rigorous form of the nucleation theorem[56] ∂W∗ = ∆n∗/ (1 ρ /ρ ) . (32) ∂∆µ − − v l For the formation of liquid droplets in a dilute vapor, Eq.(32) reduces to
∂W∗ = n∗ . (33) ∂∆µ −
The critical number n∗ can also be computed from more classical considerations. 3 Since the volume of a spherical critical nucleus is V ∗ =4πr∗ /3, one can calculate the number of molecules in the nucleus from the relation n∗vl = Vl. Applying the
Gibbs-Thomson or Kelvin equation, Eq.(5), for r∗,onefinds
2 3 32πvl γ n = ∞ , (34) ∗ 3(kT ln S)3 which is equivalent to Eq.(4). To implement the approach outlined above, there is clearly a need for a satis- factory EOS. There are many possible candidates in the literature. Not all of these are suitable for use in the EOS approach because they are not sufficiently accurate. Curiously, these less accurate EOSs are actually the only ones suitable for the gradient theory calculations presented later. For completeness all of the EOS used in this thesis are presented in the next chapter. 19
3. EQUATIONS OF STATE FOR UNARY SYSTEMS
3.1 WATER Five EOSsforwaterwereusedinthedifferentphasesofthisthesiswork. 3.1.1 IAPWS 95.ThisEOS was published by the International Associa- − tion for the Properties of Water and Steam [60, 61]. It is an analytical equation based on a multiparameter fit of all the experimental data available at temperatures above 234 K. It is very accurate and therefore highly suitable for use in the EOS approach, but only for T 234 K. This limitation strictly applies to the low T— ≥ low P vapor-liquid equilibrium states. Liquid densities at high P and low T are in good agreement with the few experimental data available. The low T —lowP vapor behavior also is reasonable. This EOS has one other significant drawback. It fails to provide a continuous representation of single phase fluid states in the metastable and unstable regions of the phase diagram, and is, therefore, unsuitable for use in gradient theory calculations. In the IAPWS 95 EOS,thespecific Helmholtz free energy f is represented in − dimensionless form as φ = f/RT ,andφ is separated into an ideal part, φ0 and a residual part φr,i.e,
f = φ = φ0(δ, τ)+φr(δ,τ) ,(35) RT 3 where δ = ρ/ρc, τ = Tc/T with Tc =647.096 K, ρc =322kg/m ,andR = 0.46151805 kJ/(kg K).Thesubscriptc designates a value at the critical point. We also have the following definitions
8 0 0 0 0 0 0 γi τ φ =ln(δ)+n1 + n2τ + n3 ln(τ)+ ni ln(1 e− ), (36) i=4 − X
7 51 54 c r di ti di ti δ i di ti 2 2 φ = niδ τ + niδ τ e− + niδ τ exp αi(δ εi) βi(τ γi) i=1 i=8 i=52 − − − − X56 X X ¡ ¢ bi + ni∆ δΨ, (37) i=55 X 20 with
2 2 ai ∆ = θ + Bi[(δ 1) ] , (38) −
1 2 2β θ =(1 τ)+Ai[(δ 1) ] i , (39) − −
2 2 Ψ =exp Ci(δ 1) Di(τ 1) . (40) − − − − All the values of coefficients and¡ parameters of φ0 and φr arelistedinAppendixB.¢ 3.1.2. Cross Over Equation of State (CREOS 01).MostEOSs − are attempts to improve the van der Waals EOS to give better representations of the properties of real systems, but these equations generally fail to reproduce the singular behavior observed at the critical point. This failure stimulated a search for a new type of EOS that could describe classical mean-field behavior far away from the critical region and smoothly cross over to the singular behavior near the critical point. New equations with this capability have been developed by Kiselev and Ely for water, which they termed CREOS 01[62]. Since the − concept behind the crossover EOS is to get the right behavior near a critical point, to make this equation work at low temperatures, the scenario of a second critical point at low temperature[67] was exploited by Kiselev and Ely[62]. Even though the CREOS 01 equation is a cubic equation, it describes only the liquid states of − the system. Since it does not provide any representation of the vapor states, it is unsuitable for use in GT calculations for vapor-to-liquid nucleation. In the CREOS equation, the Helmholtz free energy of the system is cast in terms of Landau theory[63] as
ρ A(T,ρ)=∆A(τ,∆η)+ µo(T)+Ao(T) , (41) ρc where A is the dimensionless Helmholtz free energy, A = ρA/ρcRTc,andwhere
τ = T/Tc 1, ∆η = ∆ρ = ρ/ρ 1,andµ (T),andAo(T ) are analytical functions − c − o of T . 21
5 2 α α 2 ∆i ∆i ∆A(r, θ)=kr − R (q) aψ0(θ )+ cir R (q)ψi(θ) , (42) Ã i=1 ! X f τ = r(1 b2θ2) , (43) −
β β+ 1 ∆ρ = kr R− 2 (q)+d1τ. (44) All the coefficients and parameters of CREOS 01 are given in Appendix B. − 3.1.3 Jeffery and Austin EOS (JA EOS).Jeffery and Austin[64] have − developed an analytical equation of state to describe water. It has several interesting properties, but also an important drawback. Similar to the CREOS equation, it predicts a low temperature critical point associated with two metastable phases of supercooled water. It also provides a continuous description of single-phase states in the two phase region, similar to the van der Waals and other cubic EOSs. However, since it does not accurately predict the low temperature vapor-liquid binodal line, it is suitable for quantitative use in gradient theory calculations only for a small range of higher temperatures. The JA EOS consists of three parts. The first part, developed by Song and − Mason[65, 66], is a generalized van der Waals EOS of the specificform,
PSM a 1 =1+ α b∗ ρ + αρ 1 . (45) ρRT − − RT 1 λbρ − ³ ´ 2 2· − ¸ 6 2 Here, a is the van der Waals constant, (a =27R Tc /64Pc =0.5542 Pa m mol− ), λ is a constant equal to 0.3241,b∗ and α are related to the Boyle volume, vb, through
α =2.145vb,b∗ =1.0823vb,andb is a function of temperature given by
b(T) T T =0.2exp b3( + b4) b1 exp b5 + b2 ,(46) v − T − T b µ b ¶ µ b ¶ where Tb is the Boyle temperature. The second part of the JA EOS incorporates the effects of hydrogen bonding. − This effect was first treated approximately by Poole et al.[67]. Jeffery and Austin modified the results of Poole et al. to get this part of the Helmholtz free energy as