Journal of Computational Physics 337 (2017) 132–153

Contents lists available at ScienceDirect

Journal of Computational Physics

www.elsevier.com/locate/jcp

A front tracking method for direct numerical simulation

of evaporation process in a multiphase system
Muhammad Irfan, Metin Muradoglu ∗
Department of Mechanical Engineering, Koc University, Rumelifeneri Yolu, Sariyer 34450 Istanbul, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: A front-tracking method is developed for the direct numerical simulation of evaporation
Received 19 August 2016 process in a liquid–gas multiphase system. One-field formulation is used to solve the
Received in revised form 27 December 2016 flow, energy and species equations in the framework of the front tracking method, with
Accepted 13 February 2017
suitable jump conditions at the interface. Both phases are assumed to be incompressible;
Available online 20 February 2017
however, the divergence-free velocity field condition is modified to account for the
Keywords: phase-change/mass-transfer at the interface. Both temperature and species gradient driven
Evaporation evaporation/phase-change processes are simulated. For the species gradient driven phase
Phase change change process, the Clausius–Clapeyron equilibrium relation is used to find the vapor
Front-tracking method mass fraction and subsequently the evaporation mass flux at the interface. A number of
Multi-phase flows benchmark cases are first studied to validate the implementation. The numerical results
The Clausius–Clapeyron relation are found to be in excellent agreement with the analytical solutions for all the studied
One-field formulation
cases. The methods are then applied to study the evaporation of a static as well as a single
and two droplets systems falling in the gravitational field. The methods are demonstrated
to be grid convergent and the mass is globally conserved during the phase change process
for both the static and moving droplet cases.
© 2017 Elsevier Inc. All rights reserved.

1. Introduction

Interfacial flows are frequent occurrences in nature, industrial processes and biological systems. Air/gas bubbles rising
in a water bed under buoyancy, a free-falling rain droplet in air, core annular flows for oil/gas transportation in petroleum
industries, breathing system in living organisms are some of the examples involving multiphase flows; where interfaces
move, deform and even topological changes occur during the course of time. In addition to experiments, the numerical
simulations have become an indispensable tool for the medical and the industrial sector to help improve their products and
system designs for maximum efficiency. Over decades, researchers have developed various numerical techniques to simulate
multiphase flows. It is quite a challenging task to simulate multiphase flows in the sense that sharp property gradients exist
across the interfaces, which evolve and undergo substantial deformations including topological changes.
Harlow and Welch [1] proposed one of the first methods to simulate free surface flows: the marker and cell method.
Hirt and Nichols [2] came up with a memory efficient region-following scheme, called the volume of fluid (VOF) method,
with single value of fluid volume fraction in each mesh cell. Interface can also be viewed as a level set function; the
technique has been successfully implemented by various authors to simulate multiphase flows [3,4]. In order to keep the
level set function continuous and well resolved, the level set function is maintained as a signed distance function for all the

* Corresponding author.
E-mail address: mmuradoglu@ku.edu.tr (M. Muradoglu).

http://dx.doi.org/10.1016/j.jcp.2017.02.036
0021-9991/© 2017 Elsevier Inc. All rights reserved.
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 133

time, without reconstructing the interface [4]. The notion of diffuse interface has also been used to model the interfacial
flows [5]. Jacqmin [6] performed critical test cases to show the ability of the phase field method to simulate two phase
flows. All the above mentioned methods fall under the category of interface capturing method. The second category may
be called as interface tracking method, in which the interface is represented by a separate Lagrangian grid, whereas the
governing equations are solved on the background Eulerian mesh. In some earlier implementations, the moving interface is
tracked by modifying the background Eulerian grid near the interface such that the fixed grid lines follow the interface [7,8].
Tryggvason et al. [9,10] devised a front tracking method, which does not require any re-meshing of the background grid to
track the interface. Rather, the interface or front is explicitly tracked by interpolating the velocity field from the Eulerian
grid onto the interface marker points.
The phase change phenomenon adds another dimension to the interfacial flows and thus makes modeling even more
challenging. Condensation, solidification, dissolution, boiling and vaporization/evaporation are different phase change pro-
cesses that are frequently encountered in nature and industrial applications. Design of heat exchangers and boilers for
efficient heat transfer, achieving uniform material properties through casting process, efficient burning of fuel droplets in
internal combustion engines and the dissolution of drugs in human body are some of the application areas, where a better
understanding of the physical processes and the design parameters is expected to result in increased system efficiencies
and better health standards. Direct numerical simulation is a promising technique to simulate and analyze the designs for
the system performance under a wide range of operating parameters. Also, with the advent of micro and nano-scale ap-
plications, the direct numerical simulation has proven to be an efficient tool in design improvement of the systems, where
experimentation has serious limitations [11].
Researchers have applied different multiphase flow modeling techniques to simulate different phase change phenomena.
Welch and Wilson [12] studied horizontal film boiling problem using a VOF method. Detailed numerical simulation of 3D
evaporating and strongly deformed droplet was performed by Schlottke and Weigand [13] using the VOF method. Film
boiling case has been studied using a level set method by Son and Dhir [14] and Gibou et al. [15]. The level set method, in
combination with the ghost fluid method [16], has been used to simulate evaporation of a moving and deforming droplet
by Tanguy et al. [17]. Another promising tool to model and simulate multiphase flow is the lattice Boltzmann method. Safari
et al. incorporated the temperature [18] and species gradient [19] based phase change models into the lattice Boltzmann
method developed by Lee [20].
The original front-tracking method developed for isothermal multifluid flows by Tryggvason and coworkers [9,10] has
been extended by various researchers to include the mass transfer and the phase change phenomena. For example, Juric
and Tryggvason [21] simulated the film boiling. They used an iterative procedure to set the correct temperature boundary
condition at the interface. The same procedure has also been used to track the flame front of a premixed flame by Qian et
al. [22]. Esmaeeli and Tryggvason [23,24] eliminated the iterative algorithm by setting the interface temperature as the sat-
uration temperature at the system pressure. Koynov et al. [25] performed simulations of a single bubble and bubble swarms
rising due to buoyancy including mass transfer and chemical reactions at different operating conditions. Aboulhasanzadeh
et al. [26] have recently developed a multiscale approach to compute the mass transfer from buoyant bubbles using a
boundary-layer approximation next to the bubble. This approach greatly reduced the overall grid resolution requirement.
The front tracking method with phase change model has mostly been applied to film boiling [21,24] and dendritic solidifi-
cation [27–29]; the phase change being driven solely by the temperature gradient. But little has been done for moving and
deforming liquid droplet vaporization/evaporation. In particular, to the best of our knowledge, the species gradient driven
phase-change process has not been modeled in the front-tracking framework.
In this paper, a front tracking method is developed for the liquid droplet evaporation driven by the temperature or species
gradient. A one-field formulation is used to solve the governing equations in the framework of the finite-difference/front-
tracking method on a fixed, uniform Cartesian grid. Temperature gradient driven phase change model is discussed first.
The implementation is validated using two benchmark cases: The Stefan and the sucking interface problems. The numerical
results of the interface location, temperature profile and the velocity field show excellent agreement with the analytical
results. The cases of 2D static and moving droplet evaporation are then simulated and results are presented. The main nov-
elty of the present work is the species gradient based phase change model. This model can handle a more general situation
where the gradient in the species concentration drives phase change, even if the temperature is the same in both the phases,
e.g., water droplet evaporating in the air at the atmospheric conditions. The Clausius–Clapeyron relation is incorporated to
compute the species mass fraction as well as the evaporation mass flux at the interface. Two strategies are compared for
implementing the species mass fraction boundary condition at the interface. The one that adds the evaporation mass flux as
a source term to the species equation following the strategy used in treating soluble surfactant by Muradoglu and Tryggva-
son [30,31] is easy to implement, is numerically efficient and yields better results as compared to the one that imposes the
species mass fraction at the interface directly as the boundary condition. First, a simplified test case is simulated for which
analytical solution is available for the evaporation mass flux. Numerical results agree very well with the analytical solution
for various values of interface temperature boundary condition. For a 2D static evaporating droplet case, the model predicts
the correct values of equilibrium wet bulb temperature for a water droplet evaporating in the air under various conditions
of dry bulb temperature and relative humidity. These results ensure the correct coupling of the Clausius–Clapeyron relation
with the solution of the flow, energy and species conservation equations. The model is then applied to simulate the evapora-
tion of droplets that move and undergo significant deformation in a gravitational field. The method has been demonstrated
to be grid convergent and the global mass conservation is satisfied for all the above studied cases.
134 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

In the next section, the mathematical formulation of the multiphase flow with phase change is presented. The numerical
method is described in Section 3 for both evaporation models. In Section 4, a detailed discussion is made about the results
of validation cases and other test runs for temperature and species gradient based evaporation. Finally, conclusions are
drawn in Section 5.

2. Mathematical formulation

Consider a liquid–gas multiphase system; both of which are assumed to be incompressible. Fluid flow in each phase is
governed by the Navier–Stokes equations. We can write a single set of governing equations applicable to the whole domain
as long as the jumps in the property fields are properly handled across the interface and surface tension effects are taken
into account appropriately. Then the momentum conservation equations can be written for the entire computational domain
as

∂ρu
+ ∇ · (ρ uu) = −∇ p + ρ g + ∇ · μ(∇ u + ∇ uT ) + σ κ nδ(x − x )d A , (1)
∂t
A

where u and g are the velocity and the gravitational acceleration vectors, respectively, p is the pressure, t is time and ρ and
μ are the discontinuous density and viscosity fields, respectively. The last term on the right hand side represents the body
force due to the surface tension, where σ is the surface tension coefficient, κ is twice the mean curvature, and n is a unit
vector normal to the interface. The surface tension acts only on the interface as indicated by the three-dimensional delta
function δ whose arguments x and x are the point at which the equation is being evaluated and a point at the interface,
respectively.
For a multiphase flow without phase change, the continuity equation satisfies the incompressibility condition throughout
the domain, i.e., ∇ · u = 0. However, for the phase change problem, the divergence-free velocity field condition is modified
at the interface to account for the phase-change/mass-transfer, so the continuity equation becomes
 
1 1 1
∇ ·u= − δ(x − x )q˙ d A  . (2)
hlg ρg ρl
A

The delta function makes the above equation non-zero at the interface and zero elsewhere. In Eq. (2), hlg is the latent heat
of vaporization and q˙ represents the heat flux per unit time at the interface. Subscripts  , l and g represent the interface,
the liquid and gas phases of a multiphase system, respectively. The mass and momentum jump conditions at the interface
are:
ρl (ul − u ) · n = ρ g (u g − u ) · n = m˙ , (3)

m˙ (u g − ul ) = (τ g − τ l ) · n − ( p g − pl )I · n + σ κ n, (4)
where m˙ is the mass flux per unit time across the interface, τ is the stress tensor and I is the identity tensor. Note that the
Marangoni effects are not considered in this study but can easily be incorporated into the present numerical method [32].
The energy equation is solved in the whole domain and is given by
 
∂T ∇ · k∇ T 1 T sat
+ u · ∇T = − 1 − (c p , g − c p ,l ) δ(x − x )q˙ d A  , (5)
∂t ρc p ρc p hlg
A

where T is the temperature, c p is the specific heat at constant pressure and k is the thermal conductivity. Subscript sat
denotes the saturation value of the variable. The last term in the above equation incorporates the thermal effects of phase
change into the energy equation where the coefficient (1 − (c p , g − c p ,l ) T sat /hlg ) is a constant which modifies the latent heat
hlg due to unequal specific heats of the liquid and gas phases. The convection–diffusion equation for the species evolution
in space and time reads as:
∂ Yα
+ u · ∇ Yα = ∇ · Dα∇ Yα α = 1, 2, ..., ns , (6)
∂t
where Y α and D α represent the mass fraction and mass diffusion coefficient of species component α , respectively. In
the present study, we consider only one species, the vapor, but the method can be extended to include many species
straightforwardly. We solve the species equation only for the vapor phase in the gas domain, outside the liquid droplet.
The energy and species jump conditions must be satisfied to ensure the energy and mass conservation across the inter-
face. These are:
 
∂ T  ∂ T 
m˙ hlg = q˙ = k g − kl , (7)
∂n g ∂ n l

∂ Y 
m˙ Y l − m˙ Y g + ρ g D α
 
= 0. (8)
∂ n 
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 135

Fig. 1. (a) The schematic illustration of a Lagrangian grid on an Eulerian background mesh. (b) The staggered grid used in the numerical solution of the
governing equations.

For a mono-component liquid droplet, Y l = 1 and gradients of the species mass fraction are zero. Thus Eq. (8) takes the
form:

∂ Y vap  g
ρg D α ∂ n 
m˙ = . (9)
1 − Y vap


 , is calculated using the Clausius–Clapeyron relation, i.e.,

The vapor mass fraction at the interface, Y vap
  
hlg m vap 1 1
vap = p atm exp −
p − , (10)
R T TB

p
vap m vap

Y vap = , (11)
( patm − p vap )m g + p vap m vap
B
where p  
vap is the saturated vapor pressure corresponding to the interface temperature T , T is the liquid boiling temper-
ature at the ambient pressure conditions, i.e., at p = patm , R is the perfect gas constant, and m vap and m g are the molar
masses of the vapor and gas, respectively.
We also assume that the material properties remain constant following a fluid particle, i.e.,
Dρ Dμ Dk Dc p D Dα
= 0; = 0; = 0; = 0; = 0, (12)
Dt Dt Dt Dt Dt
D
where Dt
= ∂
∂t + u · ∇ is the material derivative. The relevant non-dimensional parameters for this study can be expressed
as
ρl μl μg μc p
γ= ; ζ= ; Sc = ; Pr = ;
ρg μg ρg D k
ρ g u sls c p , g ( T ∞ − T sat )
Re = ; St = , (13)
μg hlg
where γ and ζ represent the density and the viscosity ratios, respectively. Sc, P r, Re and St are the Schmidt number,
Prandtl number, Reynolds number and Stefan number, respectively. u s and l s are appropriately selected velocity and length
scales, respectively, and t s = l s /u s be the time scale.

3. Numerical solution procedure

The governing equations for the flow (Eqs. (1)–(2)), energy (Eq. (5)) and species mass fraction (Eq. (6)) fields are solved in
a coupled form on a fixed, uniform, staggered MAC grid using a finite-difference/front-tracking method [9,10,21,24,33]. The
spatial derivatives in the momentum equations are discretized using a second-order central difference scheme, whereas, the
time integration is performed using a first-order projection method [34]. The solution of the energy and the species equa-
tions is advanced in time using a first order explicit Euler method. All spatial derivatives in the energy and species equations
are approximated using second-order central differences except for the convective terms where a 5th order WENO-Z [35]
scheme is used. The pressure, temperature, species mass fractions and all material properties are stored at the cell centers
on an Eulerian grid (Fig. 1(b)).
The interface or front, separating different phases, is made up of connected marker points and is tracked explicitly [9,10,
33]. Each marker point moves with the local flow velocity, in addition to the velocity due to phase change, Eqs. (16)–(17).
136 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

The piece of the interface between two adjacent marker points is called a front element. The schematic representation of
the Lagrangian grid on the background fixed mesh is shown in Fig. 1(a). The Indicator function I (x, t ) tracks the liquid and
the gas phases both in space and time and is defined as:

1 in droplet phase,
I (x, t ) = (14)
0 in bulk phase.
The indicator function I (x, t ) is computed at each time step using the standard procedure described by Tryggvason et
al. [10], which involves the solution of a separable Poisson equation. Then the material property fields are updated at each
time step as a function of I (x, t )
ρ = ρl I (x, t ) + ρ g (1 − I (x, t )); μ = μl I (x, t ) + μ g (1 − I (x, t ));
k = kl I (x, t ) + k g (1 − I (x, t )); ρ c p = ρl c p,l I (x, t ) + ρ g c p, g (1 − I (x, t ));
D α = D α , g (1 − I (x, t )). (15)
Information needs to be communicated between the fixed grid and the moving interface during the solution process. For
example, the surface tension as well as the heat and mass fluxes are first calculated at the interface and then smoothed
onto the fixed Eulerian grid while solving momentum, energy and species equations, respectively. Similarly, the velocity
field is only available at the fixed Eulerian grid nodes and needs to be interpolated onto the marker points for moving the
interface. A complete description of the smoothing and interpolation procedure can be found in the review paper and the
recent book by Tryggvason et al. [10,33]. We have introduced some modifications in the standard procedure for handling
certain quantities which are discussed in the relevant sections.
Also, to keep the front resolution within the prescribed limits, front restructuring is necessary. The element addition
and deletion is performed using a third-order Legendre polynomial fit to preserve interface curvature during the restruc-
turing [10]. Interface location is updated at each time step by moving the individual marker points; the velocity of each
marker point comprises of the local flow velocity and the velocity of vaporization, i.e.,
dx
= un n , (16)
dt
where
 
1 q˙ 1 1
un = (ul + u g ) · n − + . (17)
2 2hlg ρl ρg
Further details of the front-tracking method are available in a paper by Unverdi and Tryggvason [9] and a review by Tryg-
gvason et al. [10].

3.1. Flow solver

The flow equations are solved on an Eulerian grid using the projection method developed by Chorin [34]. It is a
predictor–corrector type method in which we first predict the temporary velocity field by ignoring the pressure effects,
and in the second step, the predicted velocity field is corrected to satisfy the continuity equation, Eq. (2). The momentum
equations can be written in the form:
ρ n+1 un+1 − ρ n un
= An − ∇ p , (18)
t
where A represents the advection, the diffusion, the gravitational and the surface tension force terms and the superscript n
indicates the current time level. The projection method splits the above equation as
ρ n+1 u∗ − ρ n un
= An , (19)
t
ρ n+1 un+1 − ρ n+1 u∗
= −∇ p , (20)
t
where u∗ is the unprojected velocity field, calculated using Eq. (19), by ignoring the pressure effects. Next, we take diver-
gence of Eq. (20) to obtain a Poisson equation for the pressure, i.e.,
1 ∇ · u∗ − ∇ · un+1
∇· n +1
∇p = . (21)
ρ t
For ∇ · un+1 , we use Eq. (2) as
⎡ ⎤n+1
  
1 1 1
∇ · un+1 = − ⎣ δ(x − x )q˙ d A  ⎦ , (22)
hlg ρg ρl
A
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 137

Fig. 2. (a) The method used to approximate the temperature gradients at the interface. (b) Computation of the interface length sk corresponding to the
kth marker point.

where q̇ is computed at (n + 1) time level, i.e., using the front position at (n + 1). We substitute Eq. (22) into Eq. (21) and
solve the resulting Poisson equation for pressure iteratively using a Red–Black Gauss–Seidel method with a Successive Over
Relaxation (SOR). Once u∗ and p are known, the velocity field at the next time level, n + 1, is found using Eq. (20) as:
t
un+1 = u∗ − ∇ p. (23)
ρ n +1
The above algorithm is first order accurate in time. However, it can easily be extended for the second-order accuracy using
a predictor corrector scheme as described by Tryggvason et al. [10,24].

3.2. Temperature gradient based evaporation model

In this model, it is assumed that the interface temperature T  is the same as the saturation temperature T sat , since
pressure fluctuations in the system are small as compared to the absolute pressure. Thus the energy jump condition, Eq. (7),
is used to calculate the heat flux per unit time at the interface, and is rewritten here for the kth marker point as
 
∂ T k ∂ T k
q̇k = kg − kl , (24)
∂n g ∂ n l
where k represents the kth marker point of the interface. A first-order one-sided finite difference discretization of Eq. (24)
yields [24,36]
1
q̇k = [k g ( T g − T sat ) − kl ( T sat − T l )], (25)
ηh
where T g and T l are the temperatures approximated at points (x+ , y + ) and (x− , y − ), in the gas and the liquid domains,
respectively, using a bi-linear interpolation, as shown in Fig. 2(a). These points are at a distance ηh, normal from the kth
marker point (x1 , y 1 ). In Eq. (25), h is the uniform grid spacing and η scales the length of the probe and can be selected
between 1–2 without any significant effect on the results [23,24,28]. Once q̇k is found, the last term of the energy equation,
Eq. (5), is computed first and then smoothed onto the neighbouring fixed grid nodes in a conservative manner. Following
Tryggvason et al. [10,33], for smoothing an interface quantity, say φ , onto fixed grid node (i , j ) in two-dimensions, we
must have
 
φ (s) ds = φi , j (x) d A , (26)
s A
which is approximated as
 sk
φi , j = φk w ki, j , (27)
h2
k

where sk is the length of the piece of the interface between the centers of the front elements sharing the kth marker point
and is calculated as shown by the thick lines in Fig. 2(b); and w ki, j is the weight of the fixed grid node (i , j ) corresponding to
the kth marker point and is calculated using the Peskin’s cosine function [37]. The weights must also satisfy the consistency
condition
138 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

Fig. 3. Schematic illustration of an interface cell showing the nomenclature.


w ki, j = 1. (28)
i, j

Next, for the species field, we need ṁk , which can be computed as,

q̇k
ṁk = . (29)
hlg

3.3. Species gradient based evaporation model

In this model, species concentration gradient at the interface is the only driving force for the phase change. The species
mass fraction at the interface and consequently the evaporative mass flux is computed using the Clausius–Clapeyron relation.
For the kth marker point of the interface, the evaporative mass flux per unit time (ṁk ) is computed using Eq. (9) as

∂ Y g
ρg D α ∂ n k
ṁk = k
, (30)
1 − Y vap

k
where Y vap is obtained from the Clausius–Clapeyron relation, i.e., Eqs. (10)–(11). The species gradient in the gas phase,
normal to the interface, is calculated following the same procedure as discussed for the temperature gradient (see Fig. 2a).
Corresponding to each marker point, we find a point (x+ , y + ) in the gas phase, at a distance ηh normal from the marker
point. The species mass fraction, Y + , is approximated at (x+ , y + ) using a bi-linear interpolation. The species gradient is
then calculated using a first-order one-sided finite difference approximation as

∂ Y  g 1
− Y +) .
k
= (Y vap (31)
∂ n k ηh
 may be applied
To solve the governing equation for the species mass fraction, the vapor mass fraction at the interface Y vap
directly as the boundary condition [38,39] or the mass flux per unit time ṁk can be distributed in a conservative manner
onto a thin layer just outside the liquid droplet [30,31] and then added as a source term to the species equation. Both the
strategies are briefly described below.

3.3.1. Interface mass fraction as the Dirichlet boundary condition

k
In this case, the interface mass fraction, Y vap , is applied as the Dirichlet boundary condition. The first step is to identify
the grid cells which contain interface in their immediate vicinity. An algorithm is devised to find the interface cells based
on the indicator function I , which defines interface as I = 0.5. An array is defined to store interface cells coordinates and is
updated at each time step.
The discretization of convective and diffusive terms of the species equation, Eq. (6), is modified in the interface cells to
k
incorporate Y vap as the boundary condition. Fig. 3 shows one of the several irregular stencil configurations that may arise
depending on the interface location. The convective term is approximated using a first-order upwind scheme [39]. A second
order central difference discretization of the diffusion term for the interface cell (i , j ) in Fig. 3 can be written as
 
∂2Y ∂2Y
( D α ∇ · ∇ Y )i , j = D α (i , j ) 2
+ (32)
∂x ∂ y2 i, j
   

2Y vap 
2Y vap
2Y i −1, j 2Y i , j 2Y i , j −1 2Y i , j
= D α (i , j ) − + + − + , (33)
h(h + h x ) h(h x ) h x (h + h x ) h(h + h y ) h(h y ) h y (h + h y )
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 139

Fig. 4. Treatment of the evaporation mass flux per unit time ṁk as a source term: schematic plot showing (a) the standard symmetric distribution stencil
and (b) its modified version. Subfigure (c) is the contour plot of vapor mass fraction showing the distribution of ṁk onto the fixed Eulerian grid using
modified strategy for a static evaporating droplet case; the convective and diffusive terms of species equation are switched off to better explain the scenario.
(For interpretation of the colors in this figure, the reader is referred to the web version of this article.)

where h is the uniform grid size; h x and h y are the distances between cell center (i , j ) and the interface in x and y
directions, respectively. In the rest of the domain, discretization of the species equation is fairly straight forward.

3.3.2. Evaporation mass flux as a source term

k
Instead of applying Y vap as the Dirichlet boundary condition, the evaporation mass flux per unit time, ṁk , found using
Eq. (30), is distributed as a species source, Ṡ α , onto the adjacent fixed grid nodes, just outside the interface, in a conservative
manner. The conventional symmetric and the modified one-sided distribution stencils are schematically depicted in Figs. 4(a)
and 4(b), respectively. This approach is similar and closely related to the adsorption layer concept developed by Muradoglu
and Tryggvason [30,31] for the treatment of mass exchange between the interface and bulk fluid in the simulation of soluble
surfactant in multiphase flows. The species equation, Eq. (6), is thus modified to account for the evaporative mass transfer
as a source term
∂ Yα S˙α
+ u · ∇ Yα = ∇ · Dα∇ Yα + α = 1, 2, ..., ns . (34)
∂t ρ
The source term Ṡ αi, j at grid node (i , j ) is approximated as [10,33]
 sk
Ṡ αi, j = ṁk w ki, j , (35)
h2
k

where ṁk is evaporation mass flux per unit time computed at the kth marker point, sk is the length of the interface
corresponding to the kth marker point (see Fig. 2b), h is the uniform grid spacing and w ki, j is the weight of grid point
(i , j ). The weight should satisfy the consistency condition in order to conserve the total source strength in going from the
interface to the grid, i.e.,
140 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153


w ki, j = 1. (36)
i j

The weight for the grid node (i , j ), for smoothing a quantity from the kth marker point, can be written as

w̃ ki, j
w ki, j =   . (37)
k
i j w̃ i , j

The non-normalized weight is obtained as a product of one-dimensional distribution functions, i.e.,

w̃ ki, j = d(xk − ih)d( y k − jh), (38)

where (xk , y k ) is the coordinate of the kth marker point and the distribution function d is a slightly modified version of
the Peskin’s cosine function [30,31,37] defined as
  
1
1 + cos( πλx ) if |x| < λ and I < 0.5,
d(x) = 2λ (39)
0 otherwise,
where λ is the width of the layer onto which ṁk is distributed as a mass source, and is selected as λ = 2h in the present
study. We also checked for λ = 3h but no considerable effect on the output parameters is observed. Fig. 4(c) shows the
sample contour plot of the species mass source smoothed onto the fixed grid following the above strategy, i.e., I < 0.5.

3.4. Overall solution procedure

The overall solution procedure is briefly outlined below:

i. Heat and mass fluxes per unit time, q̇n and ṁn , are computed using temperature and species fields at time level n,
using Eq. (7) and Eq. (9), respectively.
ii. q̇n is distributed onto the fixed grid using the Peskin’s distribution function [37].
iii. The procedure described in section 3.3.1 or 3.3.2 is used to handle the species mass fraction boundary condition at the
interface.
iv. Interface is advected for the next time level, n + 1, by simply integrating Eq. (16) as xn+1 = xn + tun n , where un is
computed using Eq. (17).
v. Indicator function at new interface position, I n+1 , is calculated based on the new interface location, xn+1 . (ρ c p )n+1
field is also updated based on the new indicator function, i.e., I n+1 .
vi. The energy (Eq. (5)) and species (Eq. (6) or (34)) equations are solved for the new temperature T n+1 and species Y n+1
fields, respectively.
vii. q̇n+1 is calculated and distributed onto the fixed grid following the steps i–ii.
viii. Next, we solve the flow equations for the new velocity field, un+1 , as discussed in Section 3.1. We need the surface ten-
sion term while solving Navier–Stokes equation. We compute the surface tension for each front element at the interface
location, xn , and distribute it onto the neighboring fixed grid nodes using the Peskin’s distribution function [37].
ix. The material property fields are updated for the time level n + 1 using Eq. (15).
x. Restructure the Lagrangian interface grid at each time step to keep the front element size within the prespecified limits.

4. Results and discussion

4.1. Temperature gradient based evaporation model

This model simulates the evaporative multiphase systems where the temperature gradient at the interface drives the
phase change, e.g., the fuel droplet evaporation and burning in the internal combustion engines.

4.1.1. Validation test – 1: the Stefan problem

The Stefan problem is a well-known test case to validate the phase change models [12,14,15,18,23,24,40]. Fig. 5 shows
the schematic of the Stefan problem used in this study. A vertical interface separates the liquid and the vapor phases. Both
the phases are assumed to be incompressible and are initially at rest at the saturation temperature condition, T sat . The
temperature of the left wall, T w , adjacent to the vapor phase, is increased above the saturation temperature. The heat flows
from the wall towards the interface. At the interface, the liquid vaporizes due to the temperature gradient. The interface
moves towards right due to the free flow boundary conditions applied at the right boundary. As a result, a velocity field
is also developed in the liquid phase. However, the liquid temperature stays fixed at the saturation value. The vapor is
assumed to remain stationary, therefore, diffusion is responsible for heat transfer from the wall to the interface. Hence, the
energy equation needs to be solved just in the vapor phase and can be written as

∂T ∂2T
= αg 2 0 ≤ x ≤ x (t ), (40)
∂t ∂x
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 141

Fig. 5. Schematic of the Stefan problem.

where T is the temperature, α g is the thermal diffusivity of the vapor phase and x (t ) is the interface location at time t.
Equation (40) is solved subject to the boundary conditions
T ( x = 0, t ) = T w , T (x = x (t ), t ) = T sat . (41)
Heat flux per unit time at the interface, q̇ , is computed using the energy jump condition

∂ T 
q̇ = k g , (42)
∂n g
where k g is the thermal conductivity of the vapor phase. The analytical solution for the interface location at any time t can
be expressed as

x (t ) = 2β αg t , (43)
where β is the solution of the transcendental equation
c p , g ( T w − T sat )
β exp(β 2 ) erf(β) = √ . (44)
hlg π
The analytical value of temperature at any point x in the vapor domain and at any time instant t is given by the expression
   
T sat − T w x
T g (x, t ) = T w + erf  . (45)
erf(β) 2 αg t
The velocity in the liquid phase can be found analytically using the following relation
 
ρg
ul = 1 − un , (46)
ρl

where un = β α g /t. For a flat interface with uniform velocity field, the momentum jump condition in the normal direction,
Eq. (4), simplifies to the following pressure jump relation
 
ρg 2
p g − pl = −ρ g 1− un . (47)
ρl
The fluid properties used in this numerical test case are: ρ g = 0.5, ρl = 2.5, μ g = 0.007, μl = 0.098, k g = 0.0035, kl =
0.0015, c p , g = c p ,l = 1.0 and hlg = 100. Saturation temperature, T sat , is set as 10 and wall temperature T w = 12 to achieve
the Stefan number to be St = 0.02.
We performed numerical simulations on a 1 × 1 domain. Results are presented for four different density ratios, i.e.,
γ = ρl /ρ g = 5, 10, 20 and 40. Vapor phase density, ρ g , is varied to achieve these values; the thermal diffusivity α g is
also changed as a result. For all the cases, the initial interface location is set as x = 0.1, which corresponds to different
initial times t o (γ ) depending on the density ratio (γ ). The initial temperature field is specified in the vapor phase using the
analytical solution, Eq. (45), at the initial time t o (γ ). First, a grid convergence study is performed for density ratio γ = 20.
The results for interface location and temperature profile show that a 64 × 64 grid resolution gives a very good match with
the analytical results as shown in Fig. 6.
142 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

Fig. 6. Grid convergence results for the Stefan problem for density ratio γ = 20. (Left) the evolution of the interface location, (right) the temperature profile
at t = 208.98. The insets show the enlarged views in each plot.

Fig. 7. Comparison of the analytical and numerical results for the 1D Stefan problem. The evolution of (a) the interface location, (b) the liquid phase velocity,
and (c) the pressure difference with time. Grid: 64 × 64.

Next, the numerical results of interface location, liquid phase velocity and pressure jump are compared with the analyt-
ical solutions for different density ratios. A very good agreement is observed for all the cases, as shown in Fig. 7.

4.1.2. Validation test – 2: The sucking interface problem

The sucking interface problem is also a benchmark case used to validate the temperature gradient based phase change
model and has previously been studied by Welch and Wilson [12] and Guedon [40] to validate their phase change models.
In this problem, a vapor layer is attached to the left wall of the domain while the rest is filled with a liquid as schematically
shown in Fig. 8. Both phases are assumed to be incompressible and are separated by a vertical interface. The vapor phase
is at saturation temperature T sat , and stays at rest throughout the simulation. The liquid temperature T ∞ is higher than
the saturation value, therefore, the phase change will occur at the interface. This will cause the interface to move towards
right and a flow will be developed in the liquid phase. The heat of vaporization, absorbed at the interface due to phase
change, comes from the liquid, which results in the formation of thermal boundary layer at the interface. A complete
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 143

Fig. 8. Schematic of the sucking interface problem.

convection–diffusion energy equation needs to be solved in the liquid phase in the following form

∂T ∂2T
+ u · ∇ T = αl 2 x (t ) ≤ x ≤ 1, (48)
∂t ∂x
subject to the boundary condition
T (x ≤ x (t ), t ) = T sat , (49)
where αl is the thermal diffusivity of the liquid phase. Heat flux per unit time at the interface is computed using the energy
jump condition given by Eq. (7). The sucking interface problem is a more stringent test case as compared to the Stefan
problem since we need to solve the convective term of the energy equation coupled with the Navier–Stokes equations in
addition to the diffusion term. Analytical solutions are available in the literature for this test problem [12,40]. At any time t,
the interface location is calculated using the same expression as for the Stefan problem

x (t ) = 2β αg t ,
where β is the solution of the following transcendental equation
⎡  ⎤
√ ρ2 αg
( T ∞ − T sat ) c p , g kl α g exp −β 2 ρg2 α
⎢ ⎥ c p , g ( T w − T sat )
exp(β 2 ) erf(β) ⎢ ⎥=
l l
⎣β − √  √  ⎦ √ . (50)
ρg αg hlg π
hlg k g παl erfc β ρ √α
l l

The analytical solution for the temperature profile in the vapor phase is given as
   
T sat − T w x
T g (x, t ) = T w + erf  . (51)
erf(β) 2 αg t
The temperature profile can be found in the liquid phase analytically using the following relation
⎡ ⎤
  
T∞ − T w x β(ρ g − ρl ) αg
T l (x, t ) = T ∞ − ⎣  √  ⎦ erfc √ + . (52)
ρg αg 2 αl t ρl αl
erfc β ρ √α
l l

The horizontal velocity can be found in the liquid phase, at any time t, using the same expression as we used for the Stefan
problem, Eq. (46). Similarly, the pressure jump can be calculated analytically for this case too, using the same expression as
we used for the Stefan problem, i.e., Eq. (47).
The fluid properties considered for this numerical test case are: ρ g = 0.25, ρl = 2.5, μ g = 0.007, μl = 0.098, k g = 0.0035,
kl = 0.0015, c p , g = c p ,l = 10.0 and hlg = 100. The saturation temperature T sat is set as 10, which is also the wall temperature
T w . The liquid is superheated with T ∞ = 12. The simulations are performed on a 1 × 1 domain. The interface is initially
placed at x = 0.1 which corresponds to the initial time t o = 24.7. The temperature is initialized in the domain using the
analytical solution, Eqs. (51) and (52). Numerical and analytical results of the interface location, temperature profile, liquid
phase velocity and pressure jump are compared as shown in Fig. 9. The comparisons clearly show that the numerical results
approach the analytical results as the grid is refined. Also, the numerical values of pressure jump satisfies the analytical
values. Hence our implementation is grid convergent, tracks the interface very well and successfully captures the pressure
jump and the sharp thermal boundary layer.
144 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

Fig. 9. Comparison of the numerical and analytical results for the sucking interface problem, γ = 10.

Fig. 10. The schematic illustration of a planar static droplet evaporation case.

4.1.3. Evaporation of a two-dimensional static droplet

This case simulates the evaporation of a static droplet in a planar configuration. l x and l y denote the dimensions of the
domain in the x and y coordinate directions with the origin marked at the bottom left corner as shown in Fig. 10. The same
nomenclature is also used regarding the geometric dimensions of the domain in the rest of the study. A droplet of initial
diameter do = 0.25 mm is placed at the center of 1 × 1 mm2 domain (l x = l y ). At walls, the Dirichlet boundary condition is
specified for temperature T w ; and vapors are allowed to leave the domain freely. The physical properties considered here
yield the non-dimensional parameters as γ = 5, ζ = 10, P rl = 1.75, P r g = 0.7 and Sc = 1.0. The temperature inside the
liquid droplet is initialized as T sat = 373 K and approximately stays constant during the simulation, whereas the temperature
in the gaseous domain is higher than T sat . This temperature gradient is varied to achieve various values of Stefan number,
St. The length scale l s and time scale t s are selected as do and do2 /α g , respectively. Due to the temperature gradient at the
interface, phase change occurs and a radially outward Stefan flow is generated in the gas domain. The area of the droplet,
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 145

Fig. 11. (a) Comparison of the analytical and numerical results for the normalized d2 plotted versus time for St = 0.025, 0.05 and 0.1. The inset shows
interface at two instants during evaporation at t ∗ = 0 and 9.14. St = 0.05 and Grid: 192 × 192. (b) The mass conservation error is plotted for various grid
resolutions showing the grid convergence as the grid is refined. The inset shows the order of accuracy of the method, St = 0.025. For both figures Sc = 1.0,
P rl = 1.75, P r g = 0.7, γ = 5 and ζ = 10.

and therefore the squared-diameter (d2 ) of a 2D evaporating droplet reduces with time according to the following analytical
expression

dd2 8k g ln(1 + St )
=− √ , (53)
dt ρl c p, g ln(dins / d2 )
where dins is the diameter of the inscribed circle in the computational domain. We numerically integrated Eq. (53) using
MATLAB ode45 solver and compared the time history of the normalized d2 with our simulation results for various values of
Stefan number as shown in Fig. 11(a). A good agreement is observed till 50% of the droplet mass is evaporated for all the
three cases. Also, it is observed that, doubling the Stefan number reduces the time to reach 50% mass reduction by half; the
trend similar to the findings of Safari et al. [18]. The inset shows droplet interface at two time instants during the course of
evaporation for St = 0.05. The spherical symmetry is perfectly maintained during the evaporation. A grid convergence study
is performed for St = 0.025, the global mass conservation error being the target parameter. The global mass conservation
error is defined as
 
  Ml +  M vap 
εmass =  ,
 (54)
M lo

where  Ml and  M vap denote the changes in the mass of the liquid droplet and in the vapor mass, respectively. These
are computed as  Ml = Mlo − Mlt and  M vap = M vapo − M vapt , where Ml and M vap are the liquid and vapor masses,
respectively. The subscript ‘o’ denotes the initial values and ‘t’ represents the corresponding variables at time instant t.
Fig. 11(b) shows the global mass conservation error plotted against the droplet mass reduction for various grid resolutions.
The error approaches to zero as the grid is refined demonstrating the consistency and the grid convergence of our numerical
method. The inset shows the order of convergence for this test case which is above the first order.

4.1.4. Evaporation of a two-dimensional moving droplet

This is the case of an evaporating droplet moving under the action of the gravity. A droplet of initial diameter
do = 0.25 mm is centered at (xc , y c ) = (0.5, 3.6) mm in a 1 × 4 mm2 domain. The initial droplet temperature is T sat = 373 K
whereas gas phase temperature is initialized as 480 K. Domain boundaries are specified as walls where the Dirichlet bound-
ary conditions are specified for temperature and species mass fraction as T w = 480.0 K and Y w = 0, respectively. The
magnitude of the gravitational acceleration g is varied to obtain various combinations of Eotvos ( Eo) and Morton ( Mo)
numbers, which characterize the shape of drop moving in a surrounding fluid. Eotvos and Morton numbers are defined as
Eo = (ρl − ρ g )d2 g /σ and Mo = μ4g (ρl − ρ g ) g /ρ g2 σ 3 , respectively. The method is first tested for the global mass conservation
and grid convergence for a moving droplet that deforms sufficiently while evaporating as may be the case in the combus-
tion environment. The physical properties are selected to have the non-dimensional parameters as Eo = 10, Mo = 10 × 10−4 ,
St = 0.1, Sc = 1, P rl = 5.37, P r g = 1.0, γ = 5 and ζ = 20. The initial droplet diameter do is selected as the length scale

and do / g is the time scale for this case. Fig. 12 shows the contour plots of temperature and species mass fraction and
the global mass conservation error for different grid resolutions. The results clearly indicate that the mass conservation
error decreases as the grid is refined for this more stringent test case. Also the spatial order of accuracy is almost linear
(1st-order). Fig. 13 shows the contour plots of species mass fraction for three separate cases, from left to right, in order of
increasing droplet deformation. The corresponding Eo and Mo numbers are ( Eo, Mo) = (5, 5 × 10−4 ), (10, 10 × 10−4 ) and
(20, 20 × 10−4 ). The other parameters are fixed at St = 0.1, Sc = 1, P rl = 5.37, P r g = 1.0, γ = 5 and ζ = 20 for all the three
cases. It is observed that, as the droplet departs from a spherical shape, it evaporates faster. This trend is justified because
146 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

Fig. 12. The contour plots of (a) temperature and (b) species mass fraction for an evaporating droplet falling under the action of the gravity at t ∗ = 13.416
and 128 × 512 grid resolution. (c) The global mass conservation error is plotted for various grid resolutions showing the grid convergence. The inset shows
the order of accuracy of the method. Eo = 10, Mo = 10 × 10−4 , St = 0.1, Sc = 1.0, P rl = 5.37, P r g = 1.0, γ = 5 and ζ = 20. (For interpretation of the colors
in this figure, the reader is referred to the web version of this article.)

Fig. 13. Evaporation of droplets falling under gravity with varying degree of deformation. Contour plots of the species mass fraction for three separate cases
(from left to right): Eo = 5, Mo = 5 × 10−4 , t ∗ = 13.914; Eo = 10, Mo = 10 × 10−4 , t ∗ = 13.416; Eo = 20, Mo = 20 × 10−4 , t ∗ = 16.44. Grid: 128 × 512.
(For interpretation of the colors in this figure, the reader is referred to the web version of this article.)

evaporation is a surface phenomenon and occurs due to temperature gradient at the interface; and since more surface area
is exposed to high temperature in case of deformed droplets as compared to the spherical one, therefore, evaporation is
enhanced.

4.2. Species gradient based evaporation model

This model depicts a more general situation where phase change occurs owing to the species concentration gradient
across the interface. A number of tests are performed to validate the numerical solution algorithm. The computation of
evaporation mass flux is an important parameter which depends on the correct coupling of the interface temperature with
the species mass fraction at the interface through the Clausius–Clapeyron relation. Also, the implementation of species
boundary condition at the interface is very crucial for the species solver to ensure the global mass conservation.

4.2.1. Validation test – 1: Evaporation mass flux

An efficient numerical phase change model must be able to predict evaporation mass flux accurately, since it is an im-
portant parameter in the phase change phenomenon that determines the rate of evaporation and consequently the interface
location. The problem setup consists of a container filled with water upto a certain level. Air fills the rest of the container.
The air–water interface is marked as y = 0. Based on the interface temperature, vapor mass fraction at the interface, Y vap
 ,
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 147

Fig. 14. A schematic illustration of the validation case for the evaporation mass flux.

Fig. 15. Comparison of the numerical and the analytical non-dimensional evaporation fluxes for various values of mass number B.

is computed using the Clausius–Clapeyron relation. Sufficiently far from the interface, at y = L, the air is dry, i.e., Y vap
L
= 0.
At the interface, water vapor diffuses into the air and a Stefan flow is setup in the gas phase that convects the vapor away
from the interface. Hence, a complete convective–diffusive species equation needs to be solved for the vapor field in the gas
domain. The water is assumed to be motionless. The schematic of this test case is shown in Fig. 14. For this simplified test
case, the interface temperature is assumed to stay constant during the simulation, therefore Y vap is constant too. Also, it

is assumed that the liquid evaporated is replenished exactly and continuously, therefore the location of the interface stays
fixed. Furthermore, the surrounding gas is assumed to be insoluble in water so there is no net transport of surrounding gas
into the container. The physical properties of water and air are used for this test case. For one complete simulation where
interface temperature stays constant, these physical properties are also assumed to be constant, but for different interface
temperature boundary conditions, the properties are varied accordingly.
The mass conservation principle applied over a small control volume for vapor source yields the following differential
equation [41],

dY vap  g
ṁ vap = ṁ vap Y vap − ρ g D α  , (55)
dy 
where ṁ vap is the evaporation flux per unit time at the interface. Equation (55) is solved subject to the following boundary
conditions
 
Y vap  y =0 = Y vap

, Y vap  y = L = Y vap
L
, (56)

which results in the analytical solution given by

ρg D α
ṁ vap = ln(1 + B ), (57)
L
where B is the mass number given as
 −YL
Y vap vap
B= . (58)
1 − Y vap


The evaporation mass flux per unit time is non-dimensionalized by the scaling factor ρ g D α / L. An increase in the interface
 and therefore the mass number B is also increased since Y L = 0. Fig. 15 compares the numer-
temperature increases Y vap vap
ical results of the non-dimensional evaporation mass flux with the analytical solution for various values of mass number B.
148 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

Fig. 16. (a) Comparison of the two strategies for implementing vapor mass fraction boundary condition at the interface. The global mass conservation error
is plotted at various grid resolutions demonstrating the grid convergence. The strategy 3.3.2 shows lesser errors as compared to the strategy 3.3.1. (b) The
global mass error is plotted against the normalized grid sizes for 20% and 40% droplet mass evaporation for strategy 3.3.2 displaying the spatial order of
accuracy greater than 1. DBT = 313 K and RH = 10%.

The computational results approach the analytical solution as the grid is refined, which shows that our model is grid con-
vergent and precisely computes the convective and diffusive mass transfer components. For high values of mass number B,
i.e., high evaporation rates, much finer grid resolutions are required to reach the analytical values which is expected and
self explanatory.

4.2.2. Validation test – 2: Temperature comparison with psychrometric chart values

For this test case, the computational setup consists of a liquid droplet of initial diameter do = 0.25 mm held stationary
at the center of a 1 × 1 mm2 domain. Initially, the temperature (dry bulb temperature) is the same throughout the domain
and the phase change occurs due to the species gradient at the interface resulting in a low temperature at the interface.
This goes lower and lower as evaporation proceeds until a steady state temperature condition is attained at the interface,
called the wet bulb temperature. The liquid droplet also comes into equilibrium with the wet bulb temperature. This wet
bulb temperature is a function of the dry bulb temperature (DBT) and the relative humidity (RH) in the air. The Dirichlet
boundary conditions are specified at the domain boundaries for both the temperature and the vapor mass fraction, i.e., T g
and Y vap , respectively. Y vap can be computed as Y vap = ωh /(1 + ωh ), where ωh is the humidity ratio which is a function of
dry-bulb temperature and relative humidity; and can be read from a psychrometric chart. At the interface, the vapor mass
fraction boundary condition Y vap  may be specified using any of the strategies discussed in Sections 3.3.1 or 3.3.2. Both of

these are first compared for the global mass conservation for the case with DBT =313 K and RH =10%. The physical prop-
erties of air and water are used for all the following cases unless otherwise stated except for the liquid density ρl which is
taken as 10 kg/m3 . kl is modified accordingly to obtain the thermal diffusivity value αl for water. do and do2 / D α are selected
as the length and the time scales, respectively, for all the cases related to the static droplet evaporation. The comparison
shown in Fig. 16(a) suggests that, for the same grid size, the adsorption layer concept of distributing mass flux in the im-
mediate outer vacinity of the interface and then adding as source term (Section 3.3.2) results in better mass conservation as
compared to directly imposing Y vap  as the interface boundary condition for species mass fraction (Section 3.3.1). However,

for both the methods, the trends of global mass conservation error clearly show the grid convergence on grid refinement.
Fig. 16(b) shows the global mass conservation error plotted against the non-dimensional grid size for the strategy 3.3.2. Two
sets of data points are used corresponding to 20% and 40% mass losses during the droplet evaporation. The spatial order of
accuracy is more than one in this case. For all the cases to follow, we stick with the strategy of Muradoglu and Tryggva-
son [30,31] (Section 3.3.2) for implementing vapor mass fraction boundary condition at the interface. Various cases are then
simulated with dry bulb temperatures in the range 283 K–313 K and relative humidities 10–90%, and the resulting wet bulb
temperatures are compared with the psychrometric chart values. A grid convergence study is performed first to select a
suitable grid resolution applicable to the rest of the simulations. Fig. 17 shows the grid convergence results for the wet bulb
temperature and the variation of squared-diameter (d2 ) with time corresponding to DBT = 283 K—313 K and RH = 10%. It
is found that a 128 × 128 grid resolution yields grid convergent results for both the indicators. Moreover, the temperature
results converge faster as compared to the interface location. The results of the normalized d2 are also compared with the
analytical solution. The analytical expression derived for the variation of d2 with time for a 2D evaporating droplet case is
given by

dd2 8ρ g D α ln(1 + B )
=− √ , (59)
dt ρl ln(dins / d2 )
where B is defined by Eq. (58) and dins is the diameter of the inscribed circle in the computational domain. Equation (59) is
solved using MATLAB ode45 solver. The analytical and numerical profiles for the variation of normalized d2 with time show
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 149

Fig. 17. Grid convergence study. (Left) Numerical results of wet bulb temperatures converge to the psychrometric chart values for DBT = 283 K–313 K
and RH = 10%. (Right) The simulation results for the variation of normalized d2 with the non-dimensional time t ∗ compared with the analytical solution.
DBT = 313 K, RH = 10%.

Fig. 18. Contour plots of temperature (top row) and species mass fraction (bottom row) for an evaporating droplet at non-dimensional times (from left to
right) t ∗ = 0.424, 4.24 and 42.4; the wet bulb temperature is reached in the right most plot of the top row. DBT = 313 K, RH = 50% and Grid: 128 × 128.
(For interpretation of the colors in this figure, the reader is referred to the web version of this article.)

Fig. 19. Temperature profiles plotted along the horizontal center line of the domain width at various non-dimensional times t ∗ . The steady state temperature
condition is achieved inside the evaporating droplet. DBT = 313 K, RH = 50%.
150 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

Fig. 20. Numerical results of wet bulb temperatures compared with the psychrometric chart values for various combinations of dry bulb temperatures and
relative humidities. Grid: 128 × 128.

Fig. 21. Contour plots of the vapor mass fraction for a moving, deforming and evaporating droplet at four time instants t ∗ = 0.89, 5.37, 9.84 and 14.31. Grid:
128 × 512. The non-dimensional parameters are Eo = 10, Mo = 10 × 10−5 , Sc = 1.0, P rl = 54.08, P r g = 0.15, γ = 5 and ζ = 20. (For interpretation of the
colors in this figure, the reader is referred to the web version of this article.)

good agreement as the grid is refined, as shown in Fig. 17. Some sample contour and line plots are then presented for the
case with DBT = 313 K and RH = 50%. Fig. 18 shows the contour plots for the evolution of temperature and species mass
fraction fields. A better illustration of the temperature evolution may be seen in Fig. 19, which shows temperature profiles
along the horizontal center line of the domain at various time instants during the evaporation. Once an equilibrium wet bulb
temperature is attained, all the heat absorbed just results in the phase change. Next, we made three sets of runs by keeping
the relative humidity fixed at 10%, 50% and 90% respectively, and varying the dry bulb temperature in the range 283 K–313 K
for each case. In the final batch of runs, the temperature is held constant at 283 K and 313 K and the relative humidity is
varied from 10% to 90% for the each case. The comparison of the numerical results with the psychrometric chart results are
shown in Fig. 20. Numerical results are in excellent agreement with the psychrometric chart values, however, some deviation
is observed at high dry bulb temperatures and high relative humidities. This may be attributed to our constant properties
assumption, since the thermophysical properties vary significantly at high temperatures and high relative humidities [42].

4.2.3. Evaporating droplet falling under gravity

This section simulates the planar case of an evaporating droplet that moves due to the gravitational acceleration g, and
deforms during the journey. The initial diameter, do , of the droplet is 0.25 mm. The domain size is selected as 1 × 4 mm2 .
The droplet is initially placed with its center at (0.5, 3.6) mm and starts from the rest. Temperature is initialized as 371 K
in the whole domain whereas the saturation temperature is T sat = 373 K. The domain boundaries are set as walls, and
the Dirichlet boundary conditions are specified for the temperature and vapor mass fraction at walls as T g = 371 K and
Y vap = 0, respectively. The vapor mass fraction boundary condition at the interface is implemented following the strategy
of Muradoglu and Tryggvason [30,31]. The other physical properties are selected to have the relevant non-dimensional
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 151

Fig. 22. (Left) The global mass conservation error for a moving deforming evaporating droplet at various grid resolutions. (Right) The global mass error
versus the non-dimensional grid size after 15% and 25% loss in the droplet mass. The relevant nondimensional parameters are Eo = 10, Mo = 10 × 10−5 ,
Sc = 1.0, P rl = 54.08, P r g = 0.15, γ = 5 and ζ = 20.

Fig. 23. Contour plots of the vapor mass fraction for two interacting droplets at various time instants (t ∗ = 2.68, 5.37 and 14.31). The non-dimensional
parameters are Eo = 10, Mo = 10 × 10−5 , Sc = 1.0, P rl = 54.08, P r g = 0.15, γ = 5 and ζ = 20. Grid: 192 × 512. (For interpretation of the colors in this
figure, the reader is referred to the web version of this article.)


parameters as Eo = 10, Mo = 10 × 10−5 , Sc = 1.0, P rl = 54.08, P r g = 0.15, γ = 5 and ζ = 20. do and do / g are the
appropriate length l s and time t s scales, respectively; and the velocity scale is calculated as u s = l s /t s . Fig. 21 shows the
contour plots of the vapor mass fraction at different time instants during the life of droplet, as it falls due to the gravity.
The global mass conservation results are shown in Fig. 22. It is observed that the global mass conservation error reduces as
the grid is refined. The order of accuracy is close to one for this deformed droplet evaporation case which shows the ability
of our method to handle highly deformed interfaces. The method is overall second order accurate in space but the spatial
accuracy reduces to first order for the global mass conservation mainly due to the smoothing of discontinuous fields such
as evaporation mass source at the interface.
Finally, some sample results are presented for the multiple droplets that move, evaporate, deform and interact with each
other. Two droplets are initially placed with their centers 2do apart in a 1.5 × 4 mm2 domain and move under the action of
the gravity. Temperature is initialized as 363 K in the whole domain whereas all other properties are same as for the single
droplet case. Fig. 23 shows the snapshots of their path history as they start from the rest. The global mass conservation
error plots are shown in Fig. 24.

5. Conclusions

A finite-difference/front-tracking method is developed to simulate the liquid–vapor phase change phenomenon. A one-
field formulation is used to solve the governing equations on a fixed uniform Eulerian grid whereas the interface is
152 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153

Fig. 24. The global mass conservation error plotted for two interacting and evaporating droplets at various grid resolutions. Error reduces as the grid is
refined.

represented by a separate Lagrangian grid. Different interpolation and distribution schemes are used to communicate be-
tween these two grids. The Navier–Stokes equations are solved using a projection method, whereas the energy and species
equations are solved in time using an explicit Euler method.
Both the temperature and the species gradient based phase change models are studied. The temperature gradient based
model is validated against the benchmark cases, i.e., the Stefan and the sucking interface problems. Interface location,
temperature profile and the liquid phase velocity are the parameters under investigation. Results are demonstrated to be
grid convergent and are in excellent agreement with the analytical solutions. Two-dimensional static and moving droplet
cases are then simulated in a planar configuration. For the static case, the global mass conservation and grid convergence
studies are performed and very good results are obtained. The spherical symmetry of the droplet is maintained during
the course of evaporation. Moving droplets take different shapes depending on the Eo and Mo numbers. Simulations are
performed for the substantially deforming droplet cases and the numerical method is demonstrated to be grid convergent.
Species gradient based evaporation model employs the Clausius–Clapeyron equilibrium relation to compute species mass
fraction at the interface and subsequently the evaporation mass flux. Species mass fraction boundary condition at the inter-
face is implemented using two different strategies. The one that considers evaporative mass flux as a source term is found
to be numerically efficient, easy to implement, gives better global mass conservation results and can easily be extended to
3D configuration. The numerical results of the non-dimensional evaporation mass flux is compared with the analytical so-
lution for a simplified case and excellent agreement is observed on grid refinement. The global mass conservation and grid
convergence checks are performed for a two-dimensional static droplet case. Results for both the indicators are excellent.
In the steady state condition, the droplet temperature attains equilibrium with the air temperature at the interface, i.e., the
wet bulb temperature, and can be read from a psychrometric chart. For various combinations of the dry bulb temperatures
and relative humidities, the numerical results of wet bulb temperatures compare very well with the psychrometric chart
values. Finally, simulations are performed for the highly deformed droplets falling in a gravitational field. Both single and
two interacting droplets cases are studied. Global mass conservation is ensured for both the cases and mass conservation
error converges to zero on grid refinement. We have observed that, for the novel species gradient based evaporation model,
approximately 40% of the total computational time is utilized by the phase change solver for a static droplet evaporation
case.
The current implementation is general and can easily be extended to incorporate more than one product species as
is typically the case in droplet evaporation followed by a chemical reaction. This has application to a lot of real world
engineering problems including fuel droplet evaporation and combustion in the internal combustion engines. It is also
straightforward to extend the present numerical method to 3D geometries.

Acknowledgements

The first author is supported by The Higher Education Commission of Pakistan. Computations are performed at HPC
facility of the Koc University and TUBITAK-ULAKBIM HPC centre.

References

[1] F.H. Harlow, J.E. Welch, Numerical calculation of time dependent viscous incompressible flow of fluid with free surface, Phys. Fluids 8 (12) (1965)
2182–2189, http://dx.doi.org/10.1063/1.1761178.
[2] C. Hirt, B. Nichols, Volume of fluid (VOF) method for the dynamics of free boundaries, J. Comput. Phys. 39 (1) (1981) 201–225,
http://dx.doi.org/10.1016/0021-9991(81)90145-5.
[3] S. Osher, J.A. Sethian, Fronts propagating with curvature-dependent speed: algorithms based on Hamilton–Jacobi formulations, J. Comput. Phys. 79 (1)
(1988) 12–49, http://dx.doi.org/10.1016/0021-9991(88)90002-2.
[4] M. Sussman, P. Smereka, S. Osher, A level set approach for computing solutions to incompressible two-phase flow, J. Comput. Phys. 114 (1) (1994)
146–159, http://dx.doi.org/10.1006/jcph.1994.1155.
 M. Irfan, M. Muradoglu / Journal of Computational Physics 337 (2017) 132–153 153

[5] D.M. Anderson, G.B. McFadden, A.A. Wheeler, Diffuse-interface methods in fluid mechanics, Annu. Rev. Fluid Mech. 30 (1) (1998) 139–165,
http://dx.doi.org/10.1146/annurev.fluid.30.1.139.
[6] D. Jacqmin, Calculation of two-phase Navier Stokes flows using phase-field modeling, J. Comput. Phys. 155 (1) (1999) 96–127, http://dx.doi.org/10.1006/
jcph.1999.6332.
[7] G. Ryskin, L.G. Leal, Numerical solution of free-boundary problems in fluid mechanics. Part 1. The finite-difference technique, J. Fluid Mech. 148 (1984)
1–17, http://dx.doi.org/10.1017/S0022112084002214.
[8] J. Glimm, Nonlinear and stochastic phenomena: the grand challenge for partial differential equations, SIAM Rev. 33 (4) (1991) 626–643,
http://dx.doi.org/10.1137/1033137.
[9] S.O. Unverdi, G. Tryggvason, A front-tracking method for viscous, incompressible, multi-fluid flows, J. Comput. Phys. 100 (1) (1992) 25–37,
http://dx.doi.org/10.1016/0021-9991(92)90307-K.
[10] G. Tryggvason, B. Bunner, A. Esmaeeli, D. Juric, N. Al-Rawahi, W. Tauber, J. Han, S. Nas, Y.-J. Jan, A front-tracking method for the computations of
multiphase flow, J. Comput. Phys. 169 (2) (2001) 708–759, http://dx.doi.org/10.1006/jcph.2001.6726.
[11] M. Woerner, Numerical modeling of multiphase flows in microfluidics and micro process engineering: a review of methods and applications, Microfluid.
Nanofluid. 12 (6) (2012) 841–886, http://dx.doi.org/10.1007/s10404-012-0940-8.
[12] S.W. Welch, J. Wilson, A volume of fluid based method for fluid flows with phase change, J. Comput. Phys. 160 (2) (2000) 662–682, http://dx.doi.org/
10.1006/jcph.2000.6481.
[13] J. Schlottke, B. Weigand, Direct numerical simulation of evaporating droplets, J. Comput. Phys. 227 (10) (2008) 5215–5237, http://dx.doi.org/10.1016/
j.jcp.2008.01.042.
[14] G. Son, V. Dhir, Numerical simulation of film boiling near critical pressures with a level set method, J. Heat Transf. 120 (1) (1998) 183–192,
http://dx.doi.org/10.1115/1.2830042.
[15] F. Gibou, L. Chen, D. Nguyen, S. Banerjee, A level set based sharp interface method for the multiphase incompressible Navier Stokes equations with
phase change, J. Comput. Phys. 222 (2) (2007) 536–555, http://dx.doi.org/10.1016/j.jcp.2006.07.035.
[16] R.P. Fedkiw, T. Aslam, B. Merriman, S. Osher, A non-oscillatory Eulerian approach to interfaces in multimaterial flows (the ghost fluid method), J. Com-
put. Phys. 152 (2) (1999) 457–492, http://dx.doi.org/10.1006/jcph.1999.6236.
[17] S. Tanguy, T. Menard, A. Berlemont, A level set method for vaporizing two-phase flows, J. Comput. Phys. 221 (2) (2007) 837–853, http://dx.doi.org/
10.1016/j.jcp.2006.07.003.
[18] H. Safari, M.H. Rahimian, M. Krafczyk, Extended lattice Boltzmann method for numerical simulation of thermal phase change in two-phase fluid flow,
Phys. Rev. E 88 (2013) 013304, http://dx.doi.org/10.1103/PhysRevE.88.013304.
[19] H. Safari, M.H. Rahimian, M. Krafczyk, Consistent simulation of droplet evaporation based on the phase-field multiphase lattice Boltzmann method,
Phys. Rev. E 90 (2014) 033305, http://dx.doi.org/10.1103/PhysRevE.90.033305.
[20] T. Lee, Effects of incompressibility on the elimination of parasitic currents in the lattice Boltzmann equation method for binary fluids, Comput. Math.
Appl. 58 (5) (2009) 987–994, http://dx.doi.org/10.1016/j.camwa.2009.02.017.
[21] D. Juric, G. Tryggvason, Computations of boiling flows, Int. J. Multiph. Flow 24 (3) (1998) 387–410, http://dx.doi.org/10.1016/S0301-9322(97)00050-5.
[22] J. Qian, G. Tryggvason, C. Law, A front tracking method for the motion of premixed flames, J. Comput. Phys. 144 (1) (1998) 52–69, http://dx.doi.org/
10.1006/jcph.1998.5991.
[23] A. Esmaeeli, G. Tryggvason, Computations of explosive boiling in microgravity, J. Sci. Comput. 19 (1) (2003) 163–182, http://dx.doi.org/10.1023/
A:1025347823928.
[24] A. Esmaeeli, G. Tryggvason, Computations of film boiling. Part I: numerical method, Int. J. Heat Mass Transf. 47 (25) (2004) 5451–5461,
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2004.07.027.
[25] A. Koynov, J.G. Khinast, G. Tryggvason, Mass transfer and chemical reactions in bubble swarms with dynamic interfaces, AIChE J. 51 (10) (2005)
2786–2800, http://dx.doi.org/10.1002/aic.10529.
[26] B. Aboulhasanzadeh, S. Thomas, M. Taeibi-Rahni, G. Tryggvason, Multiscale computations of mass transfer from buoyant bubbles, Chem. Eng. Sci. 75
(2012) 456–467, http://dx.doi.org/10.1016/j.ces.2012.04.005.
[27] D. Juric, G. Tryggvason, A front-tracking method for dendritic solidification, J. Comput. Phys. 123 (1) (1996) 127–148, http://dx.doi.org/10.1006/
jcph.1996.0011.
[28] N. Al-Rawahi, G. Tryggvason, Numerical simulation of dendritic solidification with convection: two-dimensional geometry, J. Comput. Phys. 180 (2)
(2002) 471–496, http://dx.doi.org/10.1006/jcph.2002.7092.
[29] N. Al-Rawahi, G. Tryggvason, Numerical simulation of dendritic solidification with convection: three-dimensional flow, J. Comput. Phys. 194 (2) (2004)
677–696, http://dx.doi.org/10.1016/j.jcp.2003.09.020.
[30] M. Muradoglu, G. Tryggvason, A front-tracking method for computation of interfacial flows with soluble surfactants, J. Comput. Phys. 227 (4) (2008)
2238–2262, http://dx.doi.org/10.1016/j.jcp.2007.10.003.
[31] M. Muradoglu, G. Tryggvason, Simulations of soluble surfactants in 3d multiphase flow, J. Comput. Phys. 274 (2014) 737–757, http://dx.doi.org/10.1016/
j.jcp.2014.06.024.
[32] S. Nas, M. Muradoglu, G. Tryggvason, Pattern formation of drops in thermocapillary migration, Int. J. Heat Mass Transf. 49 (13–14) (2006) 2265–2276,
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2005.12.009.
[33] G. Tryggvason, R. Scardovelli, S. Zaleski, Direct Numerical Simulations of Gas–Liquid Multiphase Flows, Cambridge University Press, 2011.
[34] A.J. Chorin, Numerical solution of the Navier–Stokes equations, Math. Comput. 22 (1968) 745–762.
[35] R. Borges, M. Carmona, B. Costa, W.S. Don, An improved weighted essentially non-oscillatory scheme for hyperbolic conservation laws, J. Comput. Phys.
227 (6) (2008) 3191–3211, http://dx.doi.org/10.1016/j.jcp.2007.11.038.
[36] H.S. Udaykumar, W. Shyy, M.M. Rao, ELAFINT: a mixed Eulerian Lagrangian method for fluid flows with complex and moving boundaries, Int. J. Numer.
Methods Fluids 22 (8) (1996) 691–712, http://dx.doi.org/10.1002/(SICI)1097-0363(19960430)22:8<691::AID-FLD371>3.0.CO;2-U.
[37] C.S. Peskin, Numerical analysis of blood flow in the heart, J. Comput. Phys. 25 (3) (1977) 220–252, http://dx.doi.org/10.1016/0021-9991(77)90100-0.
[38] F. Gibou, R.P. Fedkiw, L.-T. Cheng, M. Kang, A second-order-accurate symmetric discretization of the Poisson equation on irregular domains, J. Comput.
Phys. 176 (1) (2002) 205–227, http://dx.doi.org/10.1006/jcph.2001.6977.
[39] Y. Sato, B. Ničeno, A sharp-interface phase change model for a mass-conservative interface tracking method, J. Comput. Phys. 249 (2013) 127–161,
http://dx.doi.org/10.1016/j.jcp.2013.04.035.
[40] G.R. Guedon, Two-Phase Heat and Mass Transfer Modeling: Flexible Numerical Methods for Energy Engineering Analyses, Ph.D. thesis, Politecnico Di
Milano, Italy, 2013, http://hdl.handle.net/10589/82788.
[41] W.M. Kays, M.E. Crawford, Convective Heat and Mass Transfer, third ed., McGraw–Hill, New York, USA, 1993.
[42] P. Tsilingiris, Thermophysical and transport properties of humid air at temperature range between 0 and 100 ◦ C, Energy Convers. Manag. 49 (5) (2008)
1098–1110, http://dx.doi.org/10.1016/j.enconman.2007.09.015.