Diffuse interface models in uid mechanics

Didier Jamet CEA-Grenoble

Contents
1 Basics of two-phase ow modeling 1.1 Local instantaneous bulk equations . . . . . . . . . . . . 1.1.1 General form . . . . . . . . . . . . . . . . . . . . 1.1.2 Incompressible ow . . . . . . . . . . . . . . . . 1.2 Local instantaneous interfacial equations . . . . . . . . 1.2.1 General form . . . . . . . . . . . . . . . . . . . . 1.2.2 Particular cases . . . . . . . . . . . . . . . . . . . 1.3 Limitations of the discontinuous modeling of interfaces 1.3.1 Physical limitations . . . . . . . . . . . . . . . . . 1.3.2 Numerical limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 2 2 4 4 4 6 7 7 8 8 9 9 9 10 12 13 14 15 17 18 19 19 20 21 21 22 23 24 24 26 28 28 29 29 29 30 33

2 Liquid-vapor ows with phase-change: the van der Waals model of capillarity 2.1 Thermodynamic model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 A mean-eld approximation . . . . . . . . . . . . . . . . . . . . . . . 2.1.2 General equilibrium conditions . . . . . . . . . . . . . . . . . . . . . . 2.1.3 Internal structure of the interface . . . . . . . . . . . . . . . . . . . . . 2.1.4 Surface excess energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.5 Equilibrium of a pherical inclusion . . . . . . . . . . . . . . . . . . . . 2.1.6 Expression for the volumetric free energy . . . . . . . . . . . . . . . . 2.2 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Korteweg stress tensor and surface tension force . . . . . . . . . . . . 2.2.2 Equation of evolution of the temperature . . . . . . . . . . . . . . . . 2.3 Different forms of the momentum balance equation . . . . . . . . . . . . . . 2.4 Boundary conditions and contact angle . . . . . . . . . . . . . . . . . . . . . 3 Two-phase ows of non-miscible uids: the Cahn-Hilliard model 3.1 Thermodynamic model . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 A mean-eld approximation . . . . . . . . . . . . . . . 3.1.2 Equilibrium conditions . . . . . . . . . . . . . . . . . . 3.2 The Cahn-Hilliard equation . . . . . . . . . . . . . . . . . . . . 3.3 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Boussinesq approximation . . . . . . . . . . . . . . . . 3.3.2 Finite density contrast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4 Diffuse interface models and numerical simulation of mesoscopic problems 4.1 Numerical vs physical interface thickness . . . . . . . . . . . . . . . . . . . 4.2 Necessary modication of the diffuse interface models . . . . . . . . . . . 4.3 Liquid-vapor ows with phase-change . . . . . . . . . . . . . . . . . . . . . 4.3.1 Modication of the parameters . . . . . . . . . . . . . . . . . . . . . 4.3.2 Consequences of this modication . . . . . . . . . . . . . . . . . . . 4.4 Two-phase ows of non-miscible uids . . . . . . . . . . . . . . . . . . . .

1 Basics of two-phase ow modeling

In uid mechanics, as in any other discipline in physics, the models developed are relevant only at a given scale. Two-phase ows are very important in many industrial applications, including nuclear engineering. In industrial applications, the scales of interest are rather large, typically of the order of several meters. Thus, very often, their modeling is based on so-called averaged models in which all the physical quantities considered (mass, velocity, energy. . . ) are averaged quantities, either over an averaging volume or over an averaging time interval. At this scale of modeling, the interfaces that separate the bulk phases are not modeled individually. However, all the transfers between the bulk phases occur at the interfaces and these transfers therefore have to be modeled through so-called closure laws. The determination of these closure laws is an important part of the modeling art of researchers and engineers. One of the ingredients necessary in this art is the knowledge of the transfers that occur at the interfaces. The issue is therefore to model these transfers and therefore to model the interfaces. In most common industrial applications, the nest scale considered is that of the bubbles or drops present in the ow. The typical size of the bubbles or drops present depends on the parameters of the uids (e.g. surface tension) and of the ow (e.g. velocity). Nevertheless, a typical size can be from tens of microns to millimeters. This range of length scales is therefore the smallest to be considered in modeling interfacial transfers between the bulk phases. The question is then to determine how the interfaces must be modeled at this scale. At the molecular scale, an interface is a transition zone whose typical size is a few tens of Angstrms1 . Therefore, the typical size of a bubble or droplet is at least 106 times larger that the typical thickness of an interface. This clear scale separation allows to model an interface as a mathematical surface of discontinuity. This modeling hypothesis is the most common and also the most relevant to study interfaces at the scale considered. In the following, we present the classical model used to describe a two-phase ow at the scale of a few inclusions, i.e. bubbles or droplets. This model is made of local instantaneous bulk equations that are presented in section 1.1. The equations of the bulk phases are coupled by the local instantaneous interfacial equations that are presented in section 1.2.

1.1 Local instantaneous bulk equations

1.1.1 General form In uid mechanics, the modeling is based on the assumption of continuum mechanics. A bulk phase is thus characterized by continuous elds of physical variables. For a pure phase, these physical variables can be the density , the velocity v, the pressure P , the temperature T , the specic internal energy u or the specic entropy s. The motion of the uid is described by a set of coupled partial differential equations. They are obtained by applying the following basic physical principles: mass balance, momentum balance, energy balance and entropy balance. In their most general form, these equations read + t ( v) = 0

( v) + ( v v) = T t ( e) + ( e v) = q + (v T ) t ( s) + ( s v) = q s + s t
1 This size gets much larger when the two-phase system approaches a critical point and it gets even innite at a critical point. At the vicinity of a critical point, the scale separation between the typical size of the bulk phases and the interface thickness is no longer valid and the ne structure of the interfacial transition zone must be accounted for.

where T is the stress tensor, q is the heat ux, q s is the entropy ux, s is the entropy source and e is the specic total energy dened by e=u + where v 2 /2 is the specic kinetic energy. These balance equations have to be closed. First, the thermodynamic behavior of the uid has to be specied through the expression for u(s, ). The differential P du = T ds + 2 d thus denes the temperature T and the pressure P . In particular, it denes the equation of state P (, T ) and the specic heat capacity at constant volume Cv = T (T /s) v2 2

The general balance equations have to be closed also by specifying the expression for , q, q s and s . If the uid is Newtonian and follows the Fouriers law (which is actually very common), one has: T = P I + 2 = v + t v ( v) I 3 T q qs = T 2 k ( T) s = + ( v : v) T2 T where is the dissipative stress tensor, I is the identity tensor, is the viscosity of the uid, k is the thermal conductivity. The balance equations are thus closed. It worth noting that, once they are closed, the entropy balance equation and the energy balance equations are equivalent and one of them is therefore no longer necessary. Moreover, using the denition of the specic heat capacity at constant pressure Cp = T (T /s)P q = k

the closed entropy balance equation can easily be written as follows: Cp where denotes the particle derivative. It can be shown that where T = dT = dt (k T) T s P dP +( v : dt v)

d = +v dt t s P

()

1 3

is the coefcient of thermal expansion. The system of balance equations can therefore be written as follows + t ( v) + t dT = Cp dt 1.1.2 Incompressible ow By denition, an incompressible ow is such that the velocity is solenoidal: v =0 ( v) = 0 v (1) (2) (3)

( v v) = P + (k T ) + T T

dP + : dt

The mass balance equation shows that this condition is satised in particular if the density of the uid is constant. For an incompressible ow, the interpretation of the pressure P is a priori an issue. To illustrate this point, let us consider an isothermal ow. In this case, only the mass and momentum balance equations can be accounted for. The main unknowns of the problem are the velocity v and the pressure P . Since the density and the temperature are constant, the equation of state would imply that P (, T ) = cste In this case, no ow would be possible because no pressure gradient would exist in the system. Actually, it can be shown that, for incompressible ows, the pressure is interpreted as a Lagrange multiplier that accounts for the solenoidal constraint. The system of balance equations thus reads v =0

v + 0 (v v) = P + t dT = (k T ) + : v Cp dt

1.2 Local instantaneous interfacial equations

1.2.1 General form It has been shown in the introduction that the equations of motion of the bulk phases are coupled through interfacial transfers. Mathematically, these transfers correspond to boundary conditions at the interface that couple the two bulk phase balance equations derived in the previous section. The issue is then to determine these boundary conditions. This determination is based on the same basic principles as those used to determine the bulk equations of motion. Thus, these boundary conditions actually correspond to interfacial balance equations. It can be shown that these interfacial balance equations are the following: i (v i v i ) n = i (v i v i ) n = m 1 1 2 2
i i n(P2 P1 ) + n ( i i ) = m (v i v i ) + 2 2 1 1

(4) 2 n R (5) (6)

where the subscripts 1 or 2 characterize the phase and the superscript i characterizes the interface, n is the unit vector normal to the interface and directed from phase 2 to phase 1, v i is the velocity 4

m (ei ei ) = n [q i q i ] + (n T i ) v i (n T i ) v i 2 1 2 1 2 2 1 1

interface n phase 1 phase 2

Figure 1: Sharp interface modeling. of the interface, m is the mass ux across the interface, is the surface tension and R is the mean radius of curvature of the interface. The interfacial energy balance equation is rarely used in this form and is replaced by the following approximation m (hi hi ) 2 1 n [(k T )i (k 2 T )i ] 1

where h is the specic enthalpy (h = u + P/). This approximation is obtained by neglecting the kinetic energies and the work due to the viscous forces [Delhaye et al., 1981, chap. 5]. This approximation is rather intuitive. Indeed, if we can approximate the enthalpies of the phases at the interface by their values at saturation2 , this interfacial balance equation reads mL n [q i q i ] l v

where L = hsat hsat is the latent heat of vaporization. v l The interfacial balance equations are not sufcient to close the system at the interface: other boundary conditions are necessary. If we assume that the interface is at local thermodynamic equilibrium [Delhaye et al., 1981, chap. 5], these conditions are the following: vi t = vi t 1 2
i i T1 = T 2 i i g1 g 2 =

(7) (8) (9)

m2 2

where t is any unit vector tangential to the interface and g is the specic Gibbs free energy dened by P g=u+ T s The rst condition means that the tangential component of the velocities at the interface are equal. The second condition means that the temperatures of the phases at the interface are equal. The last condition means that the specic Gibbs free enthalpies of the phases at the interface are
2 We remind that, by denition, the conditions at saturation correspond to the conditions of equilibrium of a liquidvapor system with a planar interface (for which capillarity has no effect). We also remind that, for a pure substance (water for instance), at given pressure P , a liquid-vapor equilibrium is possible only for a particular temperature, called the saturation temperature T sat . Thus, T sat is a function of the pressure P : T sat (P ). This relation can of course be inverted, which gives the pressure of equilibrium for a given temperature. The two-phase equilibrium curve T sat (P ) in a (T, P ) plane is called the saturation curve. This curve is described by the Clapeyron relation:

1 (i )2 (i )2 2 1

1 1 ( i n) n i ( i n) n i 2 1 1 2

dP sat L = sat sat dT T (vv vl ) where L is the latent heat of vaporization.

slightly different. To better understand these conditions, let us neglect the viscous term and the kinetic energy term in m2 appearing in the last condition that thus reads
i i g1 = g 2

(10)

The conditions on the temperature and on the Gibbs free enthalpy actually dene the saturation conditions. Therefore, these conditions specify that the interface is locally at saturation. The term in m2 actually accounts for the kinetic energies of the phases at the interface to dene the total energy at the interface. It is important to realize that these conditions are strongly related to the hypothesis that the interface is at local thermodynamic equilibrium. This means that, whatever the transfers across the interface (mass, momentum, energy), the interface can always keep at equilibrium. In other words, the internal processes occuring within the interface are efcient enough to keep the interface at equilibrium. This is not always true, in particular when strong transfers occur. In this case, the internal processes occuring within the interface are no longer efcient enough to keep the interface at local equilibrium. The above relation then have to account for this disequilibrium, which is done through the prescription of a so-called kinetic relation [Truskinovsky, 1993]. This kinetic relation can then be related to jumps in temperature at the interface for instance. The discussion about the kinetic relation goes beyond the objectives of this manuscript, but the main idea is that, as soon as rather strong transfers occur through the interface, the interface can no longer be assumed to be at local equilibrium and the boundary conditions at the interface are strongly related to the internal processes occuring within the interface. It is thus important to investigate these processes. 1.2.2 Particular cases In the previous section, we have presented the general boundary conditions that must be applied at the interface. In this section, we present particular cases that aim at helping ot understand these interfacial boundary conditions. Let us rst consider a system at equilibrium, in which case the velocities of the phases are null (v k = 0, k {1; 2}) and the pressure and temperature of the bulk phase are uniform ( Pk = 0 and Tk = 0, k {1; 2}). The interfacial mass balance equation (4) implies that the speed of displacement of the interface v i n is null as well as the interfacial mass ow rate (m = 0). The interfacial momentum balance equation thus reduces to
i i P1 P 2 =

2 R

(11)

This is the classical Laplace relation. Since the temperature is uniform in each phase and is continuous at the interface, the temperature is uniform in the entire two-phase system; it is simply denoted T . The condition of local equilibrium (10) reads g1 (T, P1 ) = g2 (T, P2 ) If we make a Taylor expansion of this relation in the vicinity of the saturation state T sat (P2 ), one gets sat sat sat g1 ssat (T T sat ) + v1 (P1 P2 ) g2 ssat (T T sat ) 1 2 where we used dg = s dT + v dP

sat where v = 1/ is the specic volume, and where k = k (T sat (P2 ), P2 ) denotes the quantity k at saturation for the pressure P2 used as the reference.

sat sat Since g1 = g2 (equilibrium condition at saturation), one has:

T Using the relation it comes T

T sat +

ssat 1

sat 2 v1 ssat R 2

(12)

L = T sat (ssat ssat ) 1 2 T sat 1 +

sat v1 2 L R

(13)

This is the classical Gibbs-Thompson relation. This analysis shows that the general boundary conditions derived in the previous section are generalization of classical equilibrium conditions. It is worth noting that these equilibrium are often used locally at the interface. After having studied the general boundary conditions in the particular case of equilibrium, we now turn to another important case corresponding to a two-phase system involving nonmiscible uids. For the sake of simplicity, we assume that the system is isothermal. By denition, non-miscible uids are such that they do not mix with each other. Thus (almost) no particle of one uid is present in the other. This means in particular that there is no mass transfer across the interface: m = 0. The interfacial mass balance equation (4) shows that vi n = v i n = vi n 1 2 This shows that the normal component of the velocities are equal at the interface and equal to the speed of displacement of the interface. Moreover, since the tangential component of the velocities are equal at the interface, the velocities are equal at the interface: vi = vi 1 2 In the interfacial momemtum balance equation (5), v k represents the viscous force applied at the interface. The projection of this equation in the tangential direction implies that (n i ) t = (n i ) t 2 1 This means that the tangential force at the interface is continuous. It also means that the interface is not sheared. The projection of equation (5) in the normal direction implies that (n (T i T i )) n = 2 1 2 R

This means that the normal force to the interface is not continous when the interface is curved. Actually, if we remember that surface tension is a force per unit length tangential to the interface, the above relation actually represents a balance of all the forces applied to the interface.

1.3 Limitations of the discontinuous modeling of interfaces

In this section, we discuss the validity of the previous results. 1.3.1 Physical limitations As indicated in the introduction, the above developments are valid if the interface can be modeled as a surface of discontinuity. This assumption is valid as long as the typical thickness of the interface is much smaller that the typical length scale of the phenomena occuring in the surrounding bulk phases. In many applications, this assumption is valid. However, in some situations, it 7

is no longer the case and some of them are now discussed. At a critical point, the properties of both phases become equal. Above the critical point, a two-phase system cannot exist and the uid exists only as a one-phase uid. On the contrary, below the critical point, under certain conditions, the uid can co-exist under two different phases separated by an interface. As the two-phase system system approaches the critical point from below, the thickness of the interface actually increases and becomes innite at the critical point. Thus, just below the critical point, the typical size of the transition layer that separates the bulk phases becomes of the same order of magnitude as the typical size of the bulk phases. Therefore, the interface cannot be modeled as a surface of discontinuity and its internal structure has to be described . Let us consider two air bubbles in liquid water. It is commonly observed that two bubbles can coalesce, i.e. they merge to give rise to a single bubble. If the interfaces are modeled as a surface of discontinuity, just at the moment where they merge, the model is singular. In a sense, one of the interfaces desappears and this is not possible if the interface is discontinuous. To overcome this singularity, one has to study the detailed interaction of the interfaces during their merging. This can be done only by accounting for the internal structure of the interfaces [Lee et al., 2002a,b]. Another situation, actually similar to the previous one, is the description of the creation of a second phase in an initially single-phase system; this is called nucleation. In this case, one has to describe how, from a single phase, an interface is created. This continuous process can be modeled only if the interface under construction is modeled as a continuous medium [DellIsola et al., 1996]. 1.3.2 Numerical limitations In the previous section, we have shown that, in some cases, the idealization of an interface as a surface of discontinuity is physically irrelevant and the detailed structure has to be described. However, in many applications, these phenomena can be neglected or do not occur. Nevertheless, the coupled partial differential equations that describe the two-phase ow are highly non-linear and numerical simulation is often necessary to solve them. The numerical simulation of two-phase ows is very challenging because it is a moving boundary problem. Several numerical techniques exist to solve this kind of problems and their description is beyond the scope of this presentation. These techniques are often difcult to implement numerically, especially in three space dimensions, and sometimes depend on the know-how of the code developper. This is partially due to the lack of a clear mathematical background for some of the methods. Nevertheless, the boundary conditions that must be applied at the moving interfaces need a particular treatment in the numerical algorithm, which is difcult and often tedious. This is why diffuse interface methods can be numerically attractive. If one can come up with a system of partial differential equations that is valid in the entire two-phase system, including within the continuous transition interfacial zones, the motion of the entire two-phase system would be describe by this single system of equations, which thus eliminates the difcult problem of the particular treatment of the boundary conditions at the interfaces. The programming effort would therefore be highly decreased. Moreover, if these equations are obtained from rst principles, the development of accurate numerical schemes can be based on a better mathematical ground [Jamet et al., 2002].

2 Liquid-vapor ows with phase-change: the van der Waals model of capillarity
In this section, we present the van der Waals model of capillarity. This model is a diffuse interface model dedicated to the description of an interface that separates a liquid and a vapor phase of

a pure uid. Extensions to binary mixtures is possible but will not be presented here [Fouillet et al., 2002]. It is interesting to note that this model is the rst diffuse interface developed by van der Waals [van der Waals, 1894].

2.1 Thermodynamic model

Any diffuse interface model is actually a thermodynamic model. Indeed, the internal structure of an interface is mainly an equilibrium feature. Dynamic effects only perturb this equilibrium structure, which is thus important to characterize. 2.1.1 A mean-eld approximation The main issue is the following: is it possible to describe the internal structure of a liquid-vapor interface at equilibrium by considering a classical thermodynamic description of the uid? By classical, we mean that the energy of a uid particle depends only on local variables such as the density and the temperature T . Van der Waals showed that it is actually impossible [Rowlinson and Widom, 1982]: the interface would be sharp and surface tension would be null. That is why non-local terms have to be considered. In the case of a liquid-vapor interface, van der Waals postulated the following thermodynamic description: F = F 0 (, T ) + ( )2 2 (14)

where F is the volumetric free energy of the uid, F 0 is its classical part and is the capillary coefcient. For the sake of simplicity, we will always consider that is constant. F0 A

Figure 2: Illustration of the graph of the classical volumetric free energy F 0 (). It can be shown that this particular form is justied from a molecular point of view. We will not proove this and the interested reader can refer to [Rocard, 1967] for instance. In particular, it can be shown that the value of depends only on the intermolecular potential. 2.1.2 General equilibrium conditions For the sake of generality, we will consider that the volumetric free energy of the uid is given by the general expression F (, T, ). The differential of F thus reads which denes the entropy S, the Gibbs free enthalpy g as well as . The second law of thermodynamics states that a closed and isolated system at equilibrium is such that its entropy is maximum. Mathematically, this reads
V

dF = S dT + g d + d

(15)

[S + L1 U (S, , ) + L2 ] dV = 0 9

(16)

where represents the variation, V is the uid domain S is the volumetric entropy, U is the volumetric internal energy and where L1 and L2 are two constant Lagrange multipliers accounting for the constraints of conservation respectively of the energy and mass of the system. Since U =F +ST one has
V

[(1 + L1 T ) S + (L1 g + L2 ) + L1 ] dV = 0

Now = = Thus
V

() ) (17)

( ) (

[(1 + L1 T ) S + (L1 (g

) + L2 ) ] dV = 0

where we have used the condition (that is discussed in section 2.4) ( ) dV = n dS = 0

where V is the boundary of the domain and n its unit normal outwardly directed and where we have assumed that n = 0 on V . Since equation (17) must be satised for any variation S and , one has: T = g 1 L1 L2 L1 (18) (19)

Since L1 and L2 are constants, the equilibrium conditions read T = cste g = cste

The rst condition means that the temperature of the system is uniform at equilibrium. The second condition means that the generalized Gibbs free enthalpy g=g (20)

is uniform at equilibrium. This latter condition is a generalization of the classical equilibrium condition stating that the Gibbs free enthalpy is uniform at equilibrium. It must be emphasized that these equilibrium condition are valid in the entire two-phase system, including the bulk liquid and vapor phases as well as the interfacial zones. 2.1.3 Internal structure of the interface In this section, we discuss the consequences of the equilibrium conditions derived in the previous section on the internal structure of a liquid-vapor interface. For that purpose, we restrict the analysis to the case where the expression for F (, T, ) is given by (14). In this case, the equilibrium condition (19) reads F 0 (, T0 ) 10
2

= cste

(21)

where T0 is a constant corresponding to the equilibrium temperature of the system; T0 can be viewed as a parameter and is dropped in the following developments. Mathematically, this equation is a differential equation that the density eld (x) must satisfy at equilibrium. To better understand the consequences of this differential equation, we now consider a planar interface at equilibrium at we denote z the coordinate normal to the interface. We thus seek for the function (z) that denes the prole of the interfacial zone. This function must satisfy d2 (22) g 0 () 2 = cste dz where g 0 = F 0 /. g0 A

Figure 3: Illustration of the graph of g 0 (). Very far from the interfacial zone, bulk liquid and vapor phases exist and therefore d2 /dz 2 = 0; likewise, d/dz = 0. This equation shows that g 0 (v ) = cste = g 0 (l ) = geq (23)

where v and l are the densities of the vapor and liquid phases respectively far from the interface. This equation shows that the specic free Gibbs energy of the phases are equal. This corresponds to the condition of equilibrium of the interface discussed in section 1.2.2. By multiplying equation (22) by d/dz and integrate, one gets F 0 () F 0 (v ) geq ( v ) = If we dene one has W () = F 0 () F 0 (v ) geq ( v ) 2 d dz
2

d dz

(24)

(25)

= W ()

which is clearly a differential equation for (z). This equation shows in particular that W (v ) = W (l ) which is equivalent to F 0 (l ) geq l = F 0 (v ) geq v P 0 = g0 F 0 11
0

(26) (27)

Now, using classical thermodynamic relations, it can be shown that the pressure P is given by (28)

Thus equation (27) reads P 0 (v ) = P 0 (l )

(29)

This relation means that the liquid and vapor phases at equilibrium are equal. We recover the classical condition of equilibrium of a planar interface. It is worth noting that the two conditions (23) and (29) are two equations of the two unknowns v and l that can thus be determined. Moreover, these conditions have a simple graphical interpretation on the graph of the function F 0 (). Indeed, the condition (23) means that the two slopes of the tangent to this graph at v and l are equal. The condition (29) means that the y-intercepts of these tangents are equal. Therefore, these two tangents are the same. Thus, v and l are dened by the bi-tangent to the graph of the function F 0 (). Moreover, this allows to show that W () (dened by (25)) is actually the height between F 0 () and its bi-tangent. For the sake of simplicity (as will be shown hereafter), the function W () is often modeled as follows: W () = A ( v )2 ( l )2 (30) where A is a parameter characteristic of the function F 0 () (cf. gure 2). This particular form allows to simplify the differential equation (24): d = dz 2A ( v ) (l )

This differential admits the following solution (z) = where h= where h represents the interface thickness. 2.1.4 Surface excess energy This analysis also allows to determine the energy concentrated at the interface. This energy, denote F ex is dened as follows:
zi +

l + v l v z + tanh 2 2 2h 1 l v 2A (31)

F where

ex

(F (, ) F 0 (v )) dz +
zi

zi

(F 0 (l ) F (, )) dz geq ex
+

ex =

((z) v ) dz +

zi

(l (z)) dz

where z i is any position. Actually it is straightforward to show that F ex does not depend on z i . Using the relations (25) and (24), it can be shown that
+

F ex =

d dz

dz =
v

W () d

(32)

This energy concentrated at the interface is interpreted as the surface tension. It is worth noting that the above expression is general and is valid for any expression for the function F 0 () and therefore W (). In the particular case where W () is given by (30), one nds that the expression for the surface tension is the following: = (l v )3 2A 6 12 (33)

2.1.5 Equilibrium of a pherical inclusion In section 1.2, we showed that surface tension has an effect on the outer bulk phases through the Laplace relation. Is it recovered by the van der Waals model? The equilibrium condition (21) is always valid and is in particular valid for a spherical inclusion (bubble or droplet) at equilibrium. In this case, we consider a spherical system of coordinates whose origin is the center of the inclusion. The only variations that must accounted for are along the radial direction. The equilibrium condition thus reads g d2 2 d = cste 2 dr r dr (34)

where r is the radial coordinate. This equation shows in particular that

s g(s ) = g(s ) = geq v l

(35)

This relation shows that the bulk phase specic Gibbs free enthalpies are equal. It is important s to note that, at this point, the value of geq is unknown and is a priori different from geq (the equilibrium value of a planar interface). This implies in particular that the densities of the bulk phases (s and s ) are different from those of a planar interface. v l By multiplying equation (34) by d/dr and integrating from r = 0 to r yields
s F 0 () F 0 (s ) geq ( s ) = i i

d dr

+
0

d d

(36)

where the subscript i denotes the interior phase (i.e. vapor for a bubble or liquid for a drop). This expression shows in particular that
s F 0 (s ) F 0 (s ) geq (s s ) = e i e i 0

d d

where the subscript e denotes the exterior phase. Using the denition (28) of the pressure, this relation reads:
0 Pi0 s Pe s = 0

d d

The variations of are signicant only in the vicinity of the radius R of the inclusion; thus
0

d d

2 R

d d

Moreover, if the density prole (determined by the integro-differential equation (36)) is only weakly inuenced by the curvature effects the last integral term can be approximated by taking the density prole of a planar interface, in which case

d d

d 2 R

Therefore This is the Laplace relation.

0 Pi0 s Pe s

The graphical determination of the condition of equilibrium of a spherical inclusion is illustrated in gure 4. 13

W 2/R

2/R bubble drop

Figure 4: Graphical determination of the states of a spherical inclusion at equilibrium. 2.1.6 Expression for the volumetric free energy The study of of the equilibrium conditions developed in the previous sections shows the importance of the dependence in of the volumetric free energy F 0 on the determination of the internal structure of the interface and on the value of the surface tension. At a given temperature T 0 , the expression for F 0 () is the following (cf. relations (25) and (28)): F 0 () = W () + geq Peq

where Peq is the saturation pressure at the temperature considered. The equilibrium conditions studied in the previous sections show that the values of Peq and of geq have no inuence the results for an isothermal system. Therefore, very often, one takes F 0 () = W (). Because the equilibrium condition on the temperature is simply T = cste, in the previous sections, the temperature of the system has been considered as a parameter. However, the effects of the temperature are important to account for accurately. Indeed, in many applications, phasechange occurs because the system is heated, which means that the temperature gradients drive the phase transition. Moreover, the latent heat is also an important thermodynamic property that has to be accounted for. The general expression (15) for the differential of F shows that its dependence in T is related to the entropy: s= In particular, the latent heat L is such that L = T (sv sl ) = T 1 v (T ) F T (v (T )) 1 l (T ) F T (l (T )) (37) 1 F T

where v (T ) and l (T ) are the vapor and liquid densities at saturation. Moreover, the heat capacity at constant volume Cv (another important thermo-physical property) is given by s Cv = T (38) T Thus the dependence in T of F must be consistent with the data L(T ) and Cv(T ). Without any approximation, it is difcult to provide an expression for F 0 (, T ) that satises this consistency. To keep the simplicity of the polynomial expression (30) for W (), we can make the following approximation: F 0 (, T ) = W (, T ) + geq (T ) Peq (T ) 14

If the polynomial approximation (30) for W () is reasonnable for the range of temperature considered, one has W (, T ) = A(T ) ( v (T ))2 ( l (T ))2 where A(T ) is related to the isothermal compressibilities of the phases at saturation (cf. (74) and (75)) and also to the surface tension (cf. (33)). In order to simplify the analysis, we consider that the densities at saturation are independent of the temperature: v = cste and l = cste. Moreover, if we assume that the heat capacities at constant volume of the phases at saturation are independent of T , it can be shown that the following expressions for geq (T ) and Peq (T ) hold: geq (T ) = g0 + Peq (T ) = P0 + Cvv /l Cvl /v T ln 1/v 1/l T T0 (T T0 ) T T0 (T T0 )

where the subscript 0 refers to the values at a reference temperature T0 . It must be emphasized that the expression for F 0 (, T ) allows to determine the equation of state P 0 (, T ) as well as the heat capacity Cv() by their expressions (28) and (38).

T T0 Cvv Cvl L0 T ln + 1/v 1/l T0 1/v 1/l

2.2 Equations of motion

In the previous section, we showed that the thermodynamic model of van der Waals in which the energy of the uid is supposed to depend not only on T and but also on allows to (i) describe the internal structure of a liquid-vapor interface and (ii) to recover a surface tension (interpreted as an energy concentrated at the interface). However, we are very often interested in dynamic situations. Since the thermodynamic behavior of the uid is not classical, the form of the corresponding equations of motion are unknown and certainly not trivial. The main issue is to determine how the dependence of the energy in modies the classical mass, momentum and energy balance equations. Do extra terms in these equations must be accounted for? In which equations? What are their form? Different techniques can be found in the litterature to derive these equations. The most appropriate is certainly the use of the Hamiltons principle that allows to nd the non-dissipative equations of motion. The Hamiltons principle allows to clearly make appear the energy exchanges (since the system is conservative in energy for an isothermal ow). However, its application is somewhat technical and goes beyond the purpose of this presentation. Therefore, we present an easier derivation that nevertheless gives the same results. The application of the general principles of balance of mass, momentum and energy and the second law of thermodynamics read d = v (39) dt dv = T (40) dt de = q + (v T ) (41) dt ds = q s + s (42) dt If we dot-product equation (40) by v and then substract the equation obtained to the equation (41), we get du = q+T : v (43) dt 15

which is the equation of evolution of the internal energy. The expression for the internal energy is u= and the pressure P is dened by P = so that the differential of u reads du = T ds + has P d + d (46) F F (45) F +T s (44)

Using equations (39), (42) and (43) to express d/dt, ds/dt and du/dt in the above relation, one d dt (47)

The last term of this equation is transformed as follows d dt

q+T :

v = T (

q s + s ) P

v+

= = = = =

+v t + i v j ,ij t ( ) + (i v j ,j ),i (i v j ),i ,j t t ( ) + ( v ) i,i v j ,j i v j,i ,j t t ( ) + ( v ) (v ) ( ) : t t

where ,i /xi and where the Einstein convention on the repeated indices has been used. Thus d = dt d dt d ( dt ) ( ) : v

Substituting this relation in equation (47) allows to express the entropy source s as follows 1 T d dt 1 T2 d dt 1 [T + (P T

s =

(48) The second law of thermodynamics states that s 0 for any motion. The following expressions for q and T satisfy this condition3 :
3 These relations are not the most general. For instance, the thermal conductivity k is in general a tensor of order two of the form k = k1 I + k2 /( )2 . This relation expresses that the thermal conductivity in the normal and tangential directions to the interface can be different. Likewise, a Newtonian behavior is very restrictive compared to the general expressions found for the dissipative stress tensor in which ve different viscosity coefcients appear. It is worth noting that, despite its simplicity, the method used to derive these expressions, and especially the expression for q, might not be the most rigorous from a fundamental point of view. Indeed, using the Hamiltons principal, it can be shown that the term d/dt is actually not a heat ux per say but is rather a work since it has no contribution to the entropy source and is thus a conservative contribution. This term is known as the intersticial working [Dunn and Serrin, 1965].

qs

q+

q+

T+

) I +

] :

16

q = T = (P +

where k must be positive and where is the dissipative stress tensor that must satisfy : v0

d k T dt ) I

(49) + (50)

A classical Newtonian uid satises this condition. Using these closure relations, the system of balance equations that describes the motion of the uid is the following: d = v (51) dt dv = P + ( ) ( ) + (52) dt de d = + (k T ) + (v T ) (53) dt dt 2.2.1 Korteweg stress tensor and surface tension force In the most classical case where the energy of the uid is expressed by (14), = and The momentum balance equation thus reads dv = P0 + dt
2

P = P 0 (, T )

( )2 2

( )2 2

) +

(54)

The stress tensor ( ) is called the Korteweg stress tensor [Korteweg, 1901]. We show in the following that this stress tensor implies a tension force in the tangential direction to the interface, interpreted as the surface tension force. To simplify the analysis, we consider a planar interface at equilibrium. We denote z and x the coordinates respectively normal and tangential to the interface (because of symmetry, only one tangential direction can be accounted for). Thus, all the variables depend only on z. Let us rst dene P = P 0 2 + ( )2 2 so that, at equilibrium, the momentum balance equation (55) reduces to d dP + dz dz By integration, one simply gets P (z) = P where P in the pressure in the bulk phases. d dz
2

d dz

=0

17

The expression for the stress tensor (at equilibrium) is (cf. equation (50)) T eq = P I d dz
2

ez e z

where ez is the unit vector in the direction normal to the interface. Let us analyze the forces applied in the normal and tangential direction to the interface. In the direction normal to the interface, the force applied is Fz = ez T eq = P + d dz

ez

= P ez This relation shows that the force in the normal direction does not depend on z and is equal to the pressure of the bulk phases. In the direction tangential to the interface, one has Fx = ex T eq = P ex = P d dz
2

ex

The force in the tangential direction varies with z as shown in gure 5. In particular, the force force applied is lower within the interface than outside. Thus, the pessure is lower and a net tension force (opposed to the pressure) is applied in the tangential direction. The overall tangential tension F is 2 + d F = dz dz This expression is the same as the energy concentrated at the interface (cf. equation (32)). It is worth noting that it is the case only at equilibrium. Fx

Interface ez ex

Figure 5: Tangential force to the interface. 2.2.2 Equation of evolution of the temperature The entropy balance equation reads T ds = dt (k 18 T) + : v

Since is constant, s = s0 . Therefore T ds = Cv 0 dT + T Thus Cv 0 s0 s0 d

T

dT = dt

(k

T) T

d + : dt

(55)

This equation is very similar to the classical equation of evolution of the temperature in a single-phase uid. However, the interpretation of the term in d/dt is very different. Indeed, for a single-phase incompressible uid, this term vanishes (by denition). In the case of a liquid-vapor interfacial zone, (d/dt) represents the rate of vaporization denoted c . Indeed, let us consider an interface where vaporization occurs. If we follow a liquid uid particle in its motion, as it crosses the interface to become vapor, its density drastically decreases, therefore d/dt < 0. Now, T (s0 /)T is analogous to the latent heat (cf. (37)). Therefore, this term can be approximated by c L /(l v ). This term is thus a spreading of the latent heat source over the interfacial zone.

2.3 Different forms of the momentum balance equation

In the form (52) of the momemtum balance equation, the Korteweg tensor might not be the most appropriate. Indeed, from a numerical point of view for instance, the discretization of this term is not straightforward. Moreover, we will see that there exists an equivalent from in which only Laplacian and gradient operators appear, which is generally much easier to implement. The following identity holds: ( ) = ( ) ( ) +

Given the expression (45) for P and the differential of F (15), one has P = g +s T Thus, equation (52) reads dv = dt (g ) s T +

This form of the momemtum balance equation makes clearly appears the two conditions of equilibrium (18)-(19). This shows in particular that if any of these thermodynamic equilibrium conditions is not satised, it triggers a uid motion. Moreover, since the thermodynamic equilibrium conditions appear in this equation, it shows its thermodynamic consistency. Moreover, from a numerical point of view, Jamet et al. [2002] showed that, through a detailed analysis of the discretized energy exchanges, this form allows to get rid of the so-called parasitic currents [Brackbill et al., 1992]. These non-physical currents, concentrated in the close vicinity to the interface, are induced by numerical truncation that are very difcult to eliminate without a detailed analysis of the energy exchanges. The thermodynamic consistency of this model gives a framework to develop accurate numerical schemes.

2.4 Boundary conditions and contact angle

So far, in the determination of the equilibrium conditions or of the equations of motion, we have not studied the boundary conditions that must be applied on the density eld. In this section,

19

we show that particular boundary conditions arise from the existence of the capillary term in the energy functional and that these boundary conditions are related to the contact angle4 . Let us consider a liquid-vapor system in contact with a solid wall. Let us consider an energy of interaction between the solid and the uid U s (per unit surface area) and let us assume that this energy depend only on the local density of the uid at the boundary (this assumption can be justied by a mean-eld approximation [Gouin, 1998]). Therefore, the total internal energy of the system is U (S, , ) dV +
V V

U s () dA

where V is the boundary of the uid domain V . Following the developments made in section 2.1.2, the application of the second law of thermodynamics to determine the equilibrium conditions of the system yields
V

[S + L1 U (S, , ) + L2 ] dV +
V

L1 U s () dA = 0

where we remind that L1 is the constant Lagrange multiplier accounting for the constraint of conservation of the total internal energy. This yields (cf. section 2.1.2): [(1 + L1 T ) S + (L1 (g ) + L2 ) ] dV + dU s +n d dS = 0

The last surface integral is a term that did not appear in the study developed in section 2.1.2. Since the above condition must be satised for any variation , the following condition must be satised at the boundary V : dU s n= (56) d To illustrate the physical meaning of this boundary condition, let us consider the following assumptions: the expression for F (, , T ) is given by (14) so that = and U s () is assumed to be linear so that dU s /d = = cste. The boundary condition therefore reads n = (57)

Since and are constant, this consdition imposes the value of the normal derivative of . As illustrated in gure 6, this imposes the value of the contact angle. The interested reader can refer to [Seppecher, 1996, Jacqmin, 2000] for instance where this boundary condition has been studied. It is worth noting that the boundary condition (56) is an equilibrium boundary condition. However, this condition can be extended to out-of-equilibrium conditions to recover a variation of the contact angle with the speed of displacement of the contact line (a variation that is observed experimentally). This form is thermodynamically coherent since it ensures that the entropy of the system increases.

3 Two-phase ows of non-miscible uids: the Cahn-Hilliard model

In the previous section, we presented the van der Waals model of capillarity dedicated to the modeling of an interface that separates the liquid and vapor phase of the same pure substance. Many two-phase systems are made of different substances, for instance air and water or oil and water. In this case, an interface separates two phases that are made of different species. Nonmiscibles phases are phases for which no mass transfer from one phase to the other exist.
4 When a drop of liquid is put in contact with a solid wall, it is observed that the angle between the liquid-gas interface and the solid surface is characteristic of the triplet solid-liquid-gas. For instance, for the same liquid and gas, if we change the nature of the solid, the same volume of liquid either tends to spread on the surface ( < 90 ) or to retract ( > 90 ). This tendency is related to the energy of interaction between the solid and the liquid. If the solid has more afnity with the liquid than with the gas, the system tends to minimize its energy by increasing the area of contact between the liquid and the solid ( < 90 ).

20

Figure 6: Illustration of the contact angle boundary condition.

3.1 Thermodynamic model

From a physical point of view, why do these two species tend to be separated? At the molecular level, the energy of interaction of two molecules of different species is larger than that one two molecules of the same species. Thus, since the system tends to minimize its energy, if the species are separated, the energy of the system is weaker. This simple reasoning shows that the driving force of the transition, which eventually gives rise to the existence of an interface, is the amount of one species into the other. If not too many molecules of one species is present is the other, the energy of the system is not that increased but if more molecules are added, the energy gets to large and the system tends to separate into two different phases: one rich in the rst species and the other rich in the other species. In this case, the relevant thermodynamic variable is the concentration of one species in the mixture. 3.1.1 A mean-eld approximation For the sake of simplicity, we will rst consider that the density of the mixture is constant for any value of the concentration of one species in the mixture; it is denoted 0 . Let c denote the mass fraction (or concentration) of one species in the mixture. By analogy with the van der Waals model, Cahn and Hilliard [Cahn and Hilliard, 1958, 1959a,b] postulated that the free energy of the system F is given by F = F 0 (c) + ( c)2 2 (58)

where F 0 (c) is the classical part of the energy and is the capillary coefcient. F0 A 0 A

Figure 7: Illustration of the graph of the functions F 0 (c) and 0 (c). However, it can be shown that this particular form is justied from a molecular point of view. Indeed, using a mean-eld approximation, it can be shown that the attractive energy of interaction of molecules of different types gives rise to this form for the energy of the mixture and that depends only on the inter-molecular potentials. 21

                                                                            
vapor liquid solid n c c

3.1.2 Equilibrium conditions Since we deal with a mixture, we can assume that thermal effects are negligible. We can thus assume that the temperature of the system is imposed to a constant. Since the system is assumed to be isothermal, the equilibrium is characterized by a minimum of a minimum of its free energy. Mathematically, this reads:
V

(F (c, c) + L 0 c) dV = 0

where L is a constant Lagrange multiplier accounting for the fact that the system is closed and that the total mass of each species is constant, which reads 0 c dV = cste
V

Following the same developments as those presented in section 2.1.2, one nds that the equilibrium condition is the following: F c F c = cste

This condition is very similar to the condition (19) found for the van der Waals model. In the particular case where F (c, c) is given by the expression (58), this equilibrium condition simply reads 0 (c)
2

c = cste

(59)

where

dF 0 (60) dc This equation is the same as that obtained for the van der Waals model and the same conclusions hold. 0 (c) =

In particular, for a planar interface at equilibrium, it is found that the equilibrium conditions correspond to the double-tangent to the graph of the function F 0 (c). This condition denes the mass fraction of the phases at equilibrium of a planar interface, c1 and c2 , as well as the chemical potential of a planar interface eq . It is worth emphasizing that c1 and c2 are not equal to 0 or 1; actually, from a physical point of view c1 and c2 cannot be exactly equal to 0 or 1. Their value actually depend on the physical system considered. It is then convenient to introduce the double-well function W (c) dened as the difference between F (c) and its double-tangent: W (c) = F (c) (F (c1 ) + eq (c c1 )) This function is very often approximated by a polynomial of degree 4: W (c) = A (c c1 )2 (c c2 )2 It is worth noting that this particular form can be justied from a mean-eld approximation close to a critical point. It must be emphasized that this approximation is valid when c1 c2 , which means in particular that it is not valid for c1 0 and c2 1. This is important because, often, Cahn-Hilliard models are used with c1 = 0 and c2 = 1, which actually only corresponds to a renormalization of the true mass fraction. Moreover, the study of the equilibrium of a spherical inclusion implies that same results as in the van der Waals model: (i) the chemical potentials of the bulk phases surrounding the curved interface are equal 0 (ci ) = 0 (ce ) = s e 22

and (ii) there is an equivalent of the Laplace relation P 0 (ci ) P 0 (ci ) = where the pressure P 0 is dened by P 0 (c) = c dF 0 F 0 (c) dc 2 R

In this case, the pressure P 0 does not have the interpretation of a physical pressure. A Taylor expansion of these equilibrium conditions of a spherical inclusion around the equilibrium state of a planar interface allow to show that s e eq + 1 2 ci c e R (61)

It is worth noting that the form of this equilibrium condition is actually similar to the GibbsThompson condition (12). This equilibrium condition shows in particular that the value of the chemical potential is proportional to the curvature of the interface. This explains how an interface tends to get spherical. Indeed, let us consider an closed interface whose shape is initially irregular. If, at each point of the interface, the interfacial zone is at local thermodynamic equilibrium, the above equilibrium condition is satised locally. These means in particular that, along the interface, the chemical potential is not uniform. According to the Cahn-Hilliard equation, this yields a diffusion mass ux and therefore a mass diffusion. This mass diffusion makes the overall system evolve and, since the Cahn-Hilliard equation is thermodynamically coherent, this evolution tends to make the system get closer to an equilibrium state. If we consider a spherical inclusion at equilibrium, the chemical potential along the interface is constant, therefore no mass ux exist and the system keeps at rest.

3.2 The Cahn-Hilliard equation

In the previous section, we derived the equilibrium condition for a Cahn-Hilliard uid. The issue is to determine how the system behaves out of equilibrium. To start this analysis, we will rst simplify the system by assuming that the velocity of the mixture is null. In this case, the evolution of the concentration evolves through a mass diffusion equation that reads c = t j

where j is the diffusion mass ux, whose expression must be determined. The evolution of the free energy of the mixture is given by F = t q f

where q is the heat ux and f is the energy dissipation. The second law of thermodynamics imposes that f 0. Using the same developments as those presented in section 2.2 for the van der Waals model, one nds that f = j where = ( ) q+ F c c j ( t )

23

F c The following expression for j ensures that f 0 = j = ( )

where is called the mobility. It is worth noting that 0 and is any function of the thermostatic parameter of the system (i.e. c and c)5 . In the particular case where F (c, c) is given by (58), the mass diffusion equation reads c = 0 (c) 2 c (62) t This equation is called the Cahn-Hilliard equation. It is worth noting that it is a generalized mass diffusion equation that is valid in the entire two-phase system. The Cahn-Hilliard equation is a generalized mass diffusion equation. In particular, it shows that the local diffusion mass ux is proportional to the gradient of the generalized chemical potential (0 (c) 2 c).

3.3 Equations of motion

In the previous section, we derived the Cahn-Hilliard equation. This equation is the only equation that has to be accounted for if the velocity of the mixture is null. In this section, we derive the equations of motion of the mixture when this velocity is not null. 3.3.1 Boussinesq approximation In this section, we assume that the density of the mixture does not depend on the composition of the mixture: (c) = 0 . This the density is constant, the general mass balance equation of the mixture + ( v) = 0 t degenerates to the solenoidal condition v =0 In this case, we know that, in the momemtum balance equation, the pressure is the Lagrange multiplier associated to this solenoidal constraint. Using a similar approach as that developed in section 2.2 for the van der Waals model, it can be shown that the momemtum balance equation reads v + 0 v v = P ( c c) + 0 t where can be expressed as the classical viscous stress tensor. The Cahn-Hilliard equation (62) has to be modied to account for the convection term as follows: c +v c= 0 (c) 2 c t

5 In

the general case, the mobility is a tensor of the form = 1 I + 2 c c/( c)2 )

and j = (

24

It can be shown that this system of equations is thermodynamically consistent in the sense that the total energy of the system is a decreasing function of time: d dt F 0 (c) +
V

0 v 2 dV = ( c)2 + 2 2

0 (c)

+ :

v dV < 0

To account for buoyancy effects the gravity term g (where g is the acceleration of gravity) has to be added in the momentum balance equation. If = 0 in this force, this force has no effect on the ow: it only modies the pressure (the pressure P can be replaced by the pressure (P 0 g z), which does not modify the structure of the equations). For the gravity to have an effect on the ow and in particular to account for a buoyancy effect, variations of the density must be accounted for. However, these variations are a priori not compatible with the incompressibility approximation. That is why the Boussinesq approximation is generally used: the variation of the density is neglected except in the gravity term where it is linearized. The momentum balance equation therefore reads 0 where = We summarize the system of equations: v =0 c +v t v + 0 v t c= [ ]
2

v + 0 v t

v= P

( c 1 0 d dc

c) +

+ 0 (1 + (c c0 )) g

(63) (64) (65) + 0 (1 + (c c0 )) g c1 + c 2 2 (66) (67)

= 0 (c) v= P ( c

c) +

0 (c) = 4 A (c c1 ) (c c2 ) c

It is worth noting that, like in the van der Waals model (cf. section 2.3), the stress form of the momentum balance equation can be transformed into an equivalent potential form in which the generalized chemical potential appears [Jacqmin, 1999]: 0 v + 0 v t v = P + c+ P =P + F This form of the momentum balance equation shows the inuence of the curvature on the momemtum balance equation. Indeed, we showed in the previous section that, at equilibrium of a spherical inclusion, the value of the chemical potential is proportional to the local interface curvature (cf. (61)). Therefore, the term c represents a spreading (over the interface thickness) of a force proportional to the interface curvature and oriented in the direction normal to the interface approximated by c. This interpretation is very close to the Continuous Surface Force commonly used in sharp interface numerical methods [Brackbill et al., 1992]. This system of equations is particularly attractive numerically. Indeed, we one has a numerical code dedicated to the simulation of incompressible ows, it can be very easily generalized to two-phase capillary ows. Indeed, one only needs to implement a source term in the momentum 25 + 0 (1 + (c c0 )) g (68)

balance equation and a convection-diffusion-like equation for c (the Cahn-Hilliard equation). An example is given in gure 8. The system simulated is the impact of a heavier droplet on a solid grid. This gure shows in particular that topological changes are automatically accounted for. Using standard second order schemes for the discretization in space, the interface is captured by about 4 mesh cells. However, Jacqmin [1999] shows that it is possible to reduce it to about 2 by using a more complex scheme.

Figure 8: Numerical simulation of the impact of a droplet on a grid using the Cahn-Hilliard model with a Boussinesq approximation. Because of its simplicity, simulating three dimensional systems is particularly easy, even on parallel computers. An illustration of the complex three dimensional systems that can be simulated is shown in gure 9.

Figure 9: Numerical simulation of a complex Rayleigh-Taylor instability in a I-shape reservoir using a Cahn-Hilliard diffuse interface model. 3.3.2 Finite density contrast The case where the bulk phases have very different densities (for instance air and water at room temperature a 1 kg/m3 and w 1000 kg/m3 ) is rather different. From a physical point of 26

view, in this case, the velocity is not solenoidal [Galdi et al., 1991]. Therefore, a priori, the pressure can no longer be introduced as a Lagrange multiplier accounting for this constraint. However, from a practical point of view, the solenoidal constraint on the velocity eld is attractive: it allows to use standard numerical methods such as the projection method. The natural choice is thus to use the following equations v =0 (69) dc = dt [ ] + (c) g (70) (71)

dv = P + c+ dt where (c) is generally a linear function (c) (c) = 1 +

It must be emphasized that this system does not satisfy the mass balance equation (1). Indeed, from equations (69) and (70), it is straightforward to show that + t ( v) = d dc [ ] = 0

c c1 (2 1 ) c2 c 1

(72)

However, if d/dc = cste, this equation can be written in a conservative form, which shows that (by integration over the entire volume), the total mass of the system is conserved (an important property that many sharp interface numerical methods do not satisfy). Despite its simplicity, it can be shown that this system is ill-posed, in the sense that no monotonically decreasing energy is associated to this system of equations; it thus violates the second law of thermodynamics. This may be the price to pay to conserve the easiness of the numerical implementation. . . However, it is possible to develop a diffuse interface model with a nite density contrast that is thermodynamically coherent. One of the main issue is to dene incompressibility ow when the velocity eld is not solenoidal. Lowengrub and Truskinovsky [1998] showed that, in this case, it is relevant to use the thermodynamic incompressibility: the density is independent of the pressure. The main thermodynamic variables are the mass fraction c and the thermodynamic pressure P and the relevant thermodynamic potential is not the free energy F but the specic Gibbs free enthalpy g(c, P, c) whose expression is: g(c, P, c) = f 0 (c) + P ( c)2 + 2 (c)

where the linearity in P is due to the incompressibility assumption. The coefcient is equivalent to a capillary coefcient but its dimension is not that of ([] = []/[]). Lowengrub and Truskinovsky [1998] show that the corresponding system of balance equations is the following: + t where dc = dt ( v) = 0 ( ) c) +

dv = P dt =

( c

P d 1 df 0 2 dc dc 27

( c)

This system is thermodynamically coherent and has been used to study pinch-off and reconnection of interfaces for instance [Lee et al., 2002a,b]. However, this sytem is much more coupled than the incompressible model (or the thermodynamically incoherent model (69)-(71)). Indeed, the density (c) appears in many terms and in particular in expression for the Korteweg stress tensor and in the Laplacian part of . Moreover (and certainly most importantly), the pressure P appears in the expression for , which induces a complex coupling between the Cahn-Hilliard and momentum balance equation.

4 Diffuse interface models and numerical simulation of mesoscopic problems

4.1 Numerical vs physical interface thickness
In section 1.3, we showed that, in uid mechanics, diffuse interface models are attractive for two main reasons: (i) they allow to study peculiar physical phenomena where sharp interface models fail and (ii) they can be easily implemented numerically, which virtually eliminates the difcult and tedious numerical treatment of the moving boundaries. Whatever the reason, when we have to solve the equations of motion numerically, these equations are discretized in time and space. In particular, if we consider the discretization in space, the interfaces have to be captured by the mesh. To illustrate this point, let us take a particular example of a regular mesh where the size of a mesh cell is x in all directions. Since the continuous partial differential equations are discretized on this mesh, all the spacial variations must be captured, so that the numerical truncation errors do not make the system degenerate. In particular, the interface structure must be captured. This means that x must be smaller than the interface thickness h. Typically, for standard discretization schemes, one has x h/4 so that the bi-Laplacian of the Cahn-Hilliard equation is well approximated. Even for higher order schemes, x is of the order of h. The diffuse interface models presented in the previous sections (either for a liquid-vapor system of for non-miscible phases) have been derived based on physical arguments: the main thermodynamic variable are physical variables (mass density or mass concentration c), the energy functional is physical and the equations of motion are based on physical rst principles of conservation. Therefore, the internal structure of the interface is physical and in particular the interface thickness is also physical. This approach is relevant for any study where the interfaces must be modeled as continuous transition zones: coalescence and rupture of interfaces, nucleation, etc. In these cases, the typical interface thickness is of the order of 109 m and x 109 m. Therefore, assuming that the maximum size of the mesh is of the order 106 mesh cells, the typical size of the 3 3D domain studied is about 107 m3 for a regular mesh. This is extremely small, even though it is relevant for this kind of physical processes occuring at the scale of an interface. However, when diffuse interface models are used for numerical convenience, if the physical model is used as it is, one must restrict the studies to domains whose typical size is 107 m. Now, in many applications, the typical size of the inclusions (bubbles or droplets) of the two-phase system is much larger than this, by at least one order of magnitude. At this point, different strategies are possible: (i) diffuse interface models are abandonned as well as their potential ease of use, (ii) adaptative mesh renement techniques may be developed to capture the interfacial zones or (iii) diffuse interface models are adapted to the numerical simulation of mesoscopic problems. In the latter case, the adaptation of the model must be such that the value of the interface thickness is no longer dictated by physical arguments but rather by numerical arguments. The third solution will be studied in the subsequent sections. Nevertheless, the second solution deserves to be discussed. Indeed, we believe that it is a promising strategy because many physical phenoma occur close to the interface and a ne discretization is therefore necessary to capture them. However, using mesh renement to capture the internal structure of an interface might not be the most relevant solution. Indeed, if one is interested in problems whose typical size is that of a bubble or droplet (mesocopic scale), the sharp interface approximation is the most 28

relevant. This means that all the physical processes of interest occur at the scale of the inclusion (bubble or droplet) and not of the interface structure. Thus, there is a clear and justied scale separation. Now, using a mesh renement technique for these problems means that a complex numerical technique is used to capture phenomena of no physical interest that have been introduced to simplify the numerical implementation. Thus, seeking for a way to use diffuse interface models with an articial interface thickness appears as the most relevant solution.

4.2 Necessary modication of the diffuse interface models

In the previous section we showed that, for many problems where the relevant scale is the radius of an inclusion, the scale separation with the interface thickness is justied. In this case, it is almost impossible to use physical diffuse interface models on a regular mesh: much too many mesh points would be necessary only to capture the interfacial zones. Therefore, the typical size of the interfaces has to be adapted so that a reasonnable mesh can be used. In this case, the mesh is more or less given, and therefore x is given. Thus, the interface thickness h must be adapted so that the interface structure can be captured by the mesh. The interface thickness h should therefore be a free parameter whose value can be chosen arbitrarily. In the subsequent sections, we study if and how this is possible. We begin with the van der Waals model and then we study the Cahn-Hilliard model.

4.3 Liquid-vapor ows with phase-change

4.3.1 Modication of the parameters In section 2.1.2, we showed that, with the van der Waals model, the interface thickness is given by 1 (73) h= l v 2 A

where is the capillary coefcient and A is a coefcient that characterizes the function W () and therefore the free energy of the uid F () (see equation (25)). The goal is that h can be chosen arbitrarily. Now, the interface thickness is a consequence of the diffuse interface model; it is not a primary parameter of the model but rather a secondary parameter. The primary parameters of the model are l , v , and A. Therefore, these are the parameters on which one might have a degree of freedom to x the value of h arbitrarily. Among the primary parameters, is the only non-classical parameter: all the others are involved in the properties of the bulk phases. In particular, the parameter A is characteristic not only of the thermodynamic behavior of the uid within the interface but also within the bulk phases: the function F 0 () (in which the parameter A appears) is valid for any value of and in particular for the values of reached within the bulk phases. In particular, it can be shown that the isothermal compressibility of the bulk phases at saturation are given by P P = 2 A v (l v )2 = 2 A l (l v )2

(74) (75)

Thus, the only parameter clearly associated to the interface is and it appears as the most obvious parameter that can be modied to increase the interface thickness. Equation (73) shows that should be increased to increase h. Now, we have also shown that the expression for the surface tension is the following: = (l v )3 2A 6 29 (76)

P0 modied Peq

Figure 10: Modication of the equation of state P 0 () induced by the increase of the interface thickness. This expression shows that if is increased while all the other parameters of the models are kept constant, the value of is increased by the same factor of h. Surface tension is an important physical parameter. Indeed, through the Laplace and Gibbs-Thompson relations (cf. equations (11) and (13)), this interfacial property has an inuence on bulk properties such as the pressure, the temperature and thus all the other bulk properties. Therefore, modifying implies that the overall ow at the mesoscopic scale (of interest) is modied as well. This cannot be acceptable. To overcome this issue, at least one of the other parameters of the model has to be modied as well. The parameter A is chosen to be modied. Equations (31) and (33) show that, in order to increase h and to keep constant, must be increased and A must be decreased proportionally. It can be shown that, for given values of (physical) and h (numerical), and A are given by = 3 h 2 (l v )2 12 (l v )4 h

A=

This analysis shows that the overall thermodynamic behavior of the uid must be modied and not only its non-classical part in ( )2 . This means in particular that the classical part of the equation of state F 0 () must be modied. The main consequences of this modication are analyzed in the following section. 4.3.2 Consequences of this modication In the previous section, we showed that to make the interface thickness a free parameter of the system, it is necessary to modify the thermodynamic behavior of the uid. In particular, it is necessary to modify the classical part of the free energy F 0 (). This is an unexpected consequence of the procedure. Indeed, the initial goal was to modify only the interface thickness. By doing this, we implicitly accepted to violate the detailed physics of the interface layer. But we wanted to modify only interface structure without modifying the outer bulk properties. The above analysis shows that it is not possible. Now, the issue is to study whether the modication of the free energy F 0 () has important consequences on the behavior of the two-phase system at the mesoscopic scale. This study is rather difcult. Very often, this is done through an asymptotic analysis of the system. The goal is to determine to which sharp interface model the diffuse interface model is equivalent. In the particular case where a diffuse interface model is used mainly for numerical convenience, the sharp interface model is known (cf. section 1.2) and we want the diffuse interface model to mimic this given sharp interface model as well as possible. In particular, we aim at recovering the same boundary conditions made of interfacial mass, momentum, energy balance equations and, at rst approximation, local thermodynamic equilibrium of the interface. The issue is thus to determine whether the modication of the parameter A modies these boundary conditions. 30

The modication of A does not modify any property at saturation. In particular, the saturation pressure and the densities at saturation are not modied. However, it modies the variations of the pressure with the density around the states at saturation v and l . In particular, it modies the compressibility of the bulk phases. Compressibility has an effect on the motion of the uid only if the typical velocity of the uid is of the order of magnitude of the speed of sound (that depends on the compressibility of the uid). In many applications, the bulk velocities are very small compared to the speed of sound and, even though the latter is decreased by decreasing A, this remains valid. However, thermal expansion effects are important to account for because they are at the origin of natural convection. The expression for the thermal expansion coefcient is T = 1 (P/)T (P/T )

This expression shows that, if (P/)T is decreased, (P/T ) must be decreased as well to conserve the value of T . Now, the decrease of (P/T ) is equivalent to the decrease of (dP sat /dT ). Because of the Clapeyron relation L dP sat = dT T (vv vl )

in order the keep the value of L constant, it is necessary to modify the value of T . Actually, a thorough analysis of this issue [Fouillet, 2003] shows that only the value of the temperature of reference has to be modied. Since only temperature differences are important, the increase of the reference temperature is not very important. Moreover, it has been shown that this modication mainly inuences the time scale at which the interfacial zone gets back to the saturation temperature. This time is physically extremely small and an increase of the time generally barely has any inuence. This short analysis shows that the issue of the numerical increase of the interface thickness is not trivial. It also shows that, even though the initial goal was to modify only a property of the internal structure of the interface, it lead to modify important properties of the bulk phases as well; it actually lead to modify all the variation of the equation of state P (, T ). This is mainly because the graph of this function (or equivalently the function F 0 (, T )) inuences bulk properties as well as the internal structure of the interface. However, using these modications, the van der Waals model can be used to simulate complex problems such as nucleate boiling on a heated surface [Jamet and Fouillet, 2005] as shown in gure 11. This model can be extended to dilute binary mixtures as shown in gure 12. The inuence of the addition of a small amount of an extra component has been shown to have an important inuence on the wall heat exchange coefcient (cf. gure 13), which is commonly observed experimentally. Nevertheless, it is very difcult to use this model to get quantitative results. One of the main issues is gravity. Indeed, it has been shown that all the pressure scales had to be modied: (P/) and (P/T ). These modications are acceptable as long as no external pressure scale is imposed. But gravity imposes an external scale of pressure variation. To illustrate the issue, let us consider a vapor bubble on a heated wall. Because of gravity, the pressure at the top of the bubble is lower than the pressure at the bottom of the bubble. The local thermodynamic equilibrium of the interface is determined by the Clapeyron relation which determines the equilibre pressure as a function of the local temperature. So that the interface remains at equilibrium, the temperature must vary in the direction of the gravity eld: lower at the top of the bubble than at its bottom: g g T = (dP sat /dT ) (P/T ) 31

Figure 11: Numerical simulation of nucleate boiling using the van der Waals diffuse interface model [Jamet and Fouillet, 2005]. The color eld represents the temperature and the interface corresponds to iso-contours of the density eld.

Figure 12: Numerical simulation of nucleate boiling using an extension of the van der Waals diffuse interface model to binary mixtures [Jamet and Fouillet, 2005].The color eld represents the mass fraction of a dilute substance and the interface corresponds to iso-contours of the density eld. Since (P/T ) is decreased because of the increase of the interface thickness, T is increased. If the thermal conditions of the system more or less impose the temperature gradient (if the heat ux is imposed at the heated wall for instance), the disequilibrium of the bubble interface is modied and the dynamics of the bubble as well.

32

12 10 8 DT (K) 6 4 2 0 0 0.5 1 1.5 t (s)

mixture pure fluid h (J/m^2/s/K)

160000 140000 120000 100000 80000 60000 40000 20000 0

mixture pure fluid

2.5

0.5

1.5 t (s)

2.5

Figure 13: Comparison of the mean temperature wall of of the instantaneous mean heat ux with and without the presence of a dilute substance [Jamet and Fouillet, 2005].

4.4 Two-phase ows of non-miscible uids

In the case of non-miscible phases, the issue is the same: how to make the interface a free parameter without modifying the ow at the mesoscopic scale? The solution is the same as in the liquid-vapor case: increase and decrease A proportionally. However, in this case, the consequences are different. The variations of the classical chemical potential 0 (c) must be modied. We showed in the previous section that this modication might not be critical as long as no external scale of variation of the chemical potential is imposed. In common two-phase ow applications, it is rare that any external scale of variation of the chemical potential is imposed and the situation is thus easier that in the liquid-vapor case. However, there does exist a difculty. Indeed, the mobility is related to the time scale at which a system goes back to equilibrium. Since the interface thickness is modied, if is not modied, the time scale at which the interfacial zone goes back to equilibrium is modied. Therefore, must be increased to ensure that the interface keeps close to equilibrium. Finding the optimal value for is not straightforward. Now, in section 3.1.2 we showed that the chemical potential of equilibrium of a spherical inclusion depends on the radius of curvature of the system. Therefore, if two spherical inclusion of different radii at put close to each other, because of this difference of chemical potentials, a diffusion mass ux develops in between the inclusions, making the smaller inclusion shrink and the bigger expand (cf. gure 14). This diffusion mass ux is proportional to the difference of curvature of the inclusions and to the mobility , the latter being modied. Thus the time scale at which the process occurs is decreased. This problem is generally overcome by making vary so that its value in the bulk phases vanishes, thus eliminating this parasitic coarsening. However, it can be shown that, because of numerical truncation errors (and also other fundamental reasons that cannot be developed here) this parasitic coarsening cannot be totally eliminated.

Figure 14: Mass diffusion between inclusions of different sizes. The color eld represents the generalized chemical potential and the interfaces are represents by iso-contours of the mass fraction. Even though the issue of the numerical increase of the interface thickness might be less critical 33

in the Cahn-Hilliard model than in the van der Waals model, it is nevertheless a real difculty to ensure that these modications do not involve modications on the mesoscopic characteristics of the ow. This analysis shows that the adaptation of physical diffuse interface models to mesoscopic problems is rather difcult and tricky. We showed in particular that it is important to know the sharp interface model that the diffuse interface model must mimic. It is indeed a reference that helps to develop the equivalent diffuse interface model. Moreover, we showed that the difculty comes from the fact that we use only physical variables (mass density or mass fraction c) and that these physical variables are important to characterize (i) the interface structure (that is aimed at being modied) and (ii) bulk phases properties (pressure, chemical potential, etc). We showed that modifying one feature without modifying the other is difcult and tricky. The system lacks degrees of freedom. This degree of freedom can come from the introduction of another parameter whose main goal would be to characterize only the interface structure and not the bulk properties. The phase-eld variable often used in other applications of diffuse interface models can be interpreted as such a variable. It has recently been shown [Jamet and Ruyer, 2004] that such a phase-eld modeling is possible for liquid-vapor ows. The introduction of the phase-eld indeed allows to easily decouple the interface properties from the bulk properties.

References
J. U. Brackbill, D. B. Kothe, and C. Zemach. A continuum method for modeling surface tension. J. Comp. Phys., 100:335354, 1992. J. W. Cahn and J. E. Hilliard. Free energy of a nonuniform system. I. interfacial free energy. J. Chem. Physics, 28(2):258267, 1958. J. W. Cahn and J. E. Hilliard. Free energy of a nonuniform system. II. thermodynamic basis. J. Chem. Physics, 30(5):11211124, 1959a. J. W. Cahn and J. E. Hilliard. Free energy of a nonuniform system. III. nucleation in a twocomponent incompressible uid. J. Chem. Physics, 31(3):688699, 1959b. J.-M. Delhaye, M. Giot, and M. L. Riethmuller. Thermohydraulics of two-phase systems for industrial design and nuclear engineering. Hemisphere Publishing Corporation, 1981. F. DellIsola, H. Gouin, and G. Rotoli. Nucleation of spherical shell-like interfaces by second gradient theory: Numerical simulations. Eur. J. Mech. B/Fluids, 15(4):545568, 1996. J. E. Dunn and J. Serrin. On the thermodynamics of intersticial working. Arch. Rational Mech. Anal., 88:88133, 1965. C. Fouillet. Gnralisation des mlanges binaires de la mthode du second gradient et application la simulation numrique directe de lbullition nucle. Thse de doctorat, Universit Paris VI, 2003. C. Fouillet, D. Jamet, and D. Lhuillier. A continuous interface model for the direct numerical simulation of phase-change in two-component liquid-vapor ows. In 2002 Joint ASME/European Fluid Engineering Division Summer Conference, Montreal, Canada, July 14-18, 2002. G. P. Galdi, D. D. Joseph, L. Preziosi, and S. Rionero. Mathematical problems for miscible and compressible uids with korteweg stresses. European Journal of Mechanics B Fluids, 10(3):253 267, 1991. H. Gouin. Energy of interaction between solid surfaces and liquids. J. Phys. Chem. B, 102:1212 1218, 1998. doi: 10.1021/jp9723426.

34

D. Jacqmin. Calculation of two-phase Navier-Stokes ows using phase-eld modeling. J. Comp. Phys., 155:132, 1999. D. Jacqmin. Contact-line dynamics of a diffuse uid interface. J. Fluid Mech., 402:5788, 2000. D. Jamet, J. U. Brackbill, and D. Torres. On the theory and computation of surface tension: The elimination of parasitic currents through energy conservation in the second gradient method. J. Comp. Phys., 182:262276, 2002. doi: 10.1006/jcph.2002.7165. D. Jamet and C. Fouillet. Direct numerical simulations of nucleate boiling ows of binary mixtures. In 11th International Topical Meeting on Nuclear Thermal-Hydraulics (NURETH-11), Popes Palace Conference Center, Avignon, France, October 2-6, 2005. Paper # 364. D. Jamet and P. Ruyer. A quasi-incompressible model dedicated to the direct numerical simulation of liquid-vapor ows with phase-change. In 5th International Conference on Multiphase Flows, ICMF04, Yokohama, Japan, May 30-June 4, 2004. D. J. Korteweg. Sur la forme que prennent les quations du mouvement des uides si lon tient compte des forces capillaires causes par des variations de densit considrables mais continues et sur la thorie de la capillarit dans lhypothse dune variation continue de la densit. Arch. Nerl. Sci. Exactes Nat., 6:124, 1901. H. G. Lee, J. S. Lowengrub, and J. Goodman. Modeling pinchoff and reconnection in a hele-shaw cell I: The models and their calibration. Phys. Fluids, 14(2):492513, 2002a. H. G. Lee, J. S. Lowengrub, and J. Goodman. Modeling pinchoff and reconnection in a hele-shaw cell II: Analysis and simulation in the nonlinear regime. Phys. Fluids, 14(2):514545, 2002b. J. Lowengrub and L. Truskinovsky. Quasi-incompressible cahn-hilliard uids and topological transitions. Proc. R. Soc. Lond. A, 454:26172654, 1998. Y. Rocard. Thermodynamique. Masson, Paris, 1967. J. S. Rowlinson and B. Widom. Molecular theory of capillarity. Oxford University Press, New York, 1982. P. Seppecher. Moving contact line in the Cahn-Hilliard theory. Int. J. Engng. Sci., 34(9):977992, 1996. L. Truskinovsky. Kinks versus shocks. In R. Fosdick J. E. Dunn and M. Slemrod, editors, Shock Induced Transitions and Phase Structure in General Media, IMA Series in Math. and its Appl., pages 185229. Springer Verlag, 1993. van der Waals. Thermodynamische Theorie der Kapillaritt unter Voraussetzung stetiger Dichteanderung. Z. Phys. Chem., 13:657725, 1894. English translation in J. Statist. Phys., 20, 197.

35