U R A NS Study of A ir-L ayer D rag Reduction in a H igh-

Reynolds-Number F lat-Plate Turbulent Boundary L ayer

Zhaoyuan Wang1, Jianming Yang2, Frederick Stern3

IIHR-Hydroscience and Engineering, University of Iowa, Iowa City, IA, 52242-1585, U S A

T he air layer formation in a high Reynolds-number flat plate turbulent boundary layer is
simulated using a two-phase sharp interface C artesian grid solver. T he interface is tracked
by a coupled level set and volume-of-fluid method (C LSV O F) and turbulence is modeled by
a Spalart-A llmaras (SA) turbulence model with a wall function (W F) approach. T he air
layer along the entire test plate is successfully achieved and the drag reduction is
approximately 100% , which agrees very well with the experimental findings. W ith reduced
air flow rate, bubble drag reduction (B D R) is also observed; the computational results also
qualitatively match the experiments. T he transitional region from B D R to A L D R is also
observed in the present simulation. However, the critical air flow rate to form the A L D R is
lower in the simulations than in the experiments. Several possible reasons are likely
accounting for the low critical air flow rate in the simulations, such as SA-W F turbulence
model, three-dimensional instability and surface tension effects. T he critical air flow rate
does not change much with grid refinement.

I. Introduction

B UBBLE drag reduction (BDR) is an important technique that injects gas into the liquid turbulent boundary
layer to form bubbles to obtain drag reduction. This technique can substantially reduce skin friction, which has
great potential applications in ship hydrodynamics. During the past several decades, a large amount of research has
been devoted to the BDR.1 However, most of the studies were conducted at relatively low Reynolds numbers and
small scales. Proper scaling of BDR remains unclear.
In the study by Sanders et al.,2 a set of BDR experiments were conducted for a large scale flat plate turbulent
boundary layer at high Reynolds numbers. It has shown that significant levels of BDR could be achieved only near
the air injector, and limited persistence of BDR exists away from the air injector. This short persistence distance of
BDR makes it impractical for applications. It has also shown that a layer of gas was formed and persisted along the
entire plate at lower flow speeds and higher gas injection rates, which could lead to skin-friction reduction of more
than 80%. Elbing et al.3 continued the study of Sanders et al.2 in an effort to understand the mechanisms underlying
the limited persistence of the BDR and the onset conditions for the air layer drag reduction (ALDR). The
experimental results indicated that ALDR could be established once the air was injected beyond a critical rate, and
more than 80% drag reduction could be obtained over the entire plate. Three distinct regions associated with drag
reduction were observed with air injection rate: BDR, transition and ALDR. It was found that the air layer was
sensitive to the inflow conditions. In the recent work,4 a 1/2 inch (12.7 mm) step was used at the inlet, and the air
was injected from the base of the backward-facing step. This greatly enhances the stability of the air layer. The
ALDR is a potential alternative to BDR, however, the knowledge of ALDR mechanism is quite limited and more
comprehensive studies are needed. Related to ALDR, partial cavity drag reduction (PCDR) is another important
technique to reduce skin friction. PCDR needs potentially lower gas flux compared to ALDR, but un-optimized
cavity flow can lead to significant form drag.4 Partial cavities are sensitive to flow speed and perturbations from the
incoming flow.5
In the present study, URANS (unsteady Reynolds-averaged Navier-Stokes) simulations of ALDR on a large
scale flat plate are performed. The objective is to validate prediction capability of the computational code, CFDShip-
Iowa Version 6.16-8 for ALDR, investigate the mechanism of ALDR, and explore potential applications to ship
hydrodynamics. The simulations are carried out using a sharp interface Cartesian grid solver, with the interface

1
Assistant Research Scientist, AIAA Member.
2
Associated Research Scientist, AIAA Member.
3
Professor of Mechanical Engineering.
1
American Institute of Aeronautics and Astronautics
000000
tracked by a coupled level set and volume-of-fluid (CLSVOF) method and turbulence modeled by a Spalart-
Allmaras (SA) turbulence model9 with a wall function (WF) approach. The experimental data reported by Elbing et
al.3 is used to validate the simulation results. It is a challenge for the numerical simulation of such flow since high
Reynolds numbers, air-water interface, and two-phase turbulence are involved.

I I. M athematical Model and Numerical M ethods

A . Navier-Stokes E quations
Incompressible viscous flows of two immiscible fluids, e.g., air and water, are governed by the Navier-Stokes
equations:
wu 1
 u ˜’u ’ ˜   pI  T   g , (1)
wt U
’˜ u 0 , (2)
where t is the time, u is the velocity vector, p is the pressure, I is the unit diagonal tensor, ȡ is the density, g
represents the gravity acceleration, and T is the viscous stress tensor defined as
T 2P S , (3)
with ȝ the dynamic viscosity and S the strain rate
S 
1
2

’u   ’u  .
T
(4)

Since the fluid properties are discontinuous across the interface, which is a function of time and space, density
and viscosity are also functions of time and space and only known with given interface position. Their definitions
will be deferred after the introduction of interface representation using the level set function.

B. T urbulence M odeling
In the URANS approach, the turbulent eddy viscosity is obtained by solving a transport equation for an
auxiliary variable Q as proposed by Spalart and Allmaras9
wQ~ Q~
wt
~
d
^ 1
V
`
 u ˜ ’Q~ cb1 :Q~  c w1 f w ( ~ ) 2  ’ ˜ >(Q Q~)’Q~@  cb 2 ’Q~ .
2
(5)

The left hand side of the equation consists of the unsteady and advection terms of the turbulent eddy viscosity. The
terms of right hand side are the production, destruction and diffusion, respectively. The turbulent eddy viscosity is
obtained as
Q T f v1Q , (6)
where,
F3 Q . (7)
f v1 , F
F  cv13
3
Q
The production term is based on the magnitude of vorticity,
Q
: :  2 2 fv2 , : 2: ˜ : ,
N d (8)
:
1
2
 
’u   ’u  , f v 2 1 
T F
1  F f v1
Mariani and Zilliac10 provided an improvement to the production term by suppressing the turbulence, i.e, excessive
production of eddy viscosity in regions where vorticity magnitude exceeds the strain-rate,
Q
: :  f min(0, S  : )  fv2 . (9)
N 2d 2
They concluded the value of f ~ 3.5-4.0 provides best result for wingtip vortex calculations. In present calculations f
= 4.0 is chosen. The destruction term involves a near-wall damping function which is

2
American Institute of Aeronautics and Astronautics
000000
 1/6
ª 1 c 6 º
fw g « 6 w3 6 » ,
¬ g  c w3 ¼
(10)
g r  cw 2 ( r 6  r ),
Q
r
: N 2d 2

where d is the distance to the nearest wall.

The model constants are
cb1 0.1355, cb 2 0.622,
cv1 7.1, V 2 / 3,
cw1 cb1N 2  (1  cb 2 )V 1 3.239, (11)
cw 2 0.3, cw3 2,
N 0.41, C DES 0.65

In the multi-phase flows the molecular eddy viscosity is smoothed across the interface using a Heaviside
function (refer to Ref. 6) to avoid sharp gradients in molecular viscosity in Eq. (5). In order to capture the effects of
viscous boundary layers within the framework of a Cartesian grid solver, a multi-layer wall-function model capable
of switching smoothly between sub-, buffer-, and log-layers is used, details are given in the studies.
8,11

C . Interface M odeling
Defining the interface Ƚ as the zero level set of a signed distance function I , or the level set function, the
position of the interface can be tracked by solving the level set evolution equation
wI
 u ˜’I 0 . (12)
wt
To keep I as a signed distance function in the course of the evolution, we iterate the reinitialization equation for the
level set function12
wI
 S Io   ’I  1 0 , (13)
wW
where W is the pseudo time and S (I0 ) is a numerically smeared-out sign function
Io
S Io  , (14)
I  h2
2
o

with I 0 the initial values of I and h a small distance, usually the grid cell size, to smear out the sign function.
In the CLSVOF method , the volume-of-fluid (VOF) function, F , is defined as the liquid volume fraction in a
cell with its value in between zero and one in a surface cell and zero and one in air and liquid respectively. The
advection equation of F is
wF
 u ˜’ F 0. (15)
wt
The level set function is corrected based on the reconstructed interface using VOF function for mass conservation.
With the level set function defined, the fluid properties, such as density and viscosity, are given by the
following equations:
U U G  (U L  U G ) F
(16)
P P G  (P L  P G ) F
where the subscripts G and L represent gas and liquid phase, respectively.
In terms of jump conditions, the velocity across the interface Ƚ is continuous, as the fluids are viscous and no
phase change is considered here:

3
American Institute of Aeronautics and Astronautics
000000
 >u@ 0, (17)
and the jump condition for stress is

 
ª n ˜  p I  P ’u   ’u T
¬«  ˜ n º¼» VN , (18)

where [ ] indicates the jump at the interface, i.e., f LI  f GI for a variable f with superscript I denotes interface, n is
the unit vector normal to the interface, ı is the coefficient of surface tension, and ț is the local curvature of the
interface. Notice that with a continuous viscosity and velocity field, the stress jump condition Eq. (18) reduces to
> p@ pLI  pGI VN . (19)

D. Numerical M ethods and H igh Performance Computing

The flow equations are discretized on a staggered Cartesian grid with the convection terms approximated by a
third-order QUICK scheme13 and other terms by the standard second-order central difference scheme. A semi-
implicit time-advancement scheme is adopted to integrate the momentum equations with the second-order Crank-
Nicolson scheme for the diagonal viscous terms and the second-order Adams-Bashforth scheme for the convective
terms and other viscous terms. A four-step fractional-step method is employed for velocity-pressure coupling. The
pressure Poisson equation is solved using a semi-coarsening multigrid Poisson solver from the HYPRE library from
Lawrence Livermore National Laboratory.14 Details of the numerical methods can be found in the study by Yang
and Stern.6
The code is parallelized via a domain decomposition (in three directions) technique using the MPI library. All
inter-processor communications for ghost cell information exchange are in non-blocking mode. In general, optimal
load balance can be achieved except for a small amount of overhead due to interface, which may be unevenly
distributed over processors. Parallel I/O using MPI2 have been implemented such that all processors read from and
write to one single file simultaneously, which is much more effective than one or a few processors receive data from
all processors and write to one or a few files and more convenient than every processor writes its own data files.

I I I. Results

A . Computational Setup
The simulations are carried out on a two dimensional (2D) computational domain with the boundary conditions
specified as shown in Figure 1 for both the wet case (without air injection) and the ALDR case. The air injection
geometry, slot A, used in Test 1 in the experiments by Elbing et al.3 is chosen in the ALDR tests. A non-uniform
grid of 256×1024 is used first with the streamwise grid refined near the air injector and the grid in the normal
direction refined near the wall as shown in Figure 2. This initial coarse grid is designed to resolve the boundary
layer, streamwise resolutions are not enough for the water/air interface changes. A wet case without air injection at a
Reynolds number of Re = 2.1×108 is conducted first in order to validate the SA turbulent model with the wall
function approach. The computational result is shown in Fig. 3 along with the EFD (experimental fluid dynamics)
and analytical solutions. As shown in the figure, the computational result matches the power-law fit very well. As
compared to the experimental data, the skin friction coefficient is under predicted. This might be due to grid design
and deficiency of SA model, further investigations using finer grid and more advanced turbulence models will be
considered in the future work.

No slip wall Air Û No slip wall

Û
inlet Water/air outlet
g g
Water inlet Water outlet Water inlet
Slip wall Slip wall

(a) (b)
F igure 1. Boundary conditions for wet plate and A L D R simulations.

4
American Institute of Aeronautics and Astronautics
000000
 F igure 2. G rid structure of the computational domain for the A L D R.

0 .0 0 3
E xp e rim e n ta l
K a rm a n - S ch o e n h e rr E q u a tio n
- 0 .1 4 9
P o w e r la w fit, C F 0 = 0 .0 2 5 R e x
C F D S H IP - V 6 - S A - W F
0 .0 0 2 5

C F0
0 .0 0 2

0 .0 0 1 5

0 .0 0 1
0 0 .5 1 1 .5 2 ( X 1 0 8 ) 2 .5
R ex

F igure 3. Skin friction coefficient of the wet case as a function of Reynolds number (based on
downstream distance).  

B. A ir L ayer D rag Reduction

In the present study, all the simulations are conducted with a Reynolds number of Re = 7.37×107 which
corresponds to Test 1 in Elbing et al. (2008) with a free stream velocity of 6.7 m/s. The largest air flow rate of 15.3
m3/min is used first; later several low air flow rates are employed to investigate the effect on the formation of BDR
and ALDR. The simulation matrix is given in Table 1.

T able 1 Simulation M atrix

Volumetric gas
Injection geometry Re water (U ’ ) T urbulence model G rid F low type and status
injection rates (Q *)
0.5184
0.0864 A L D R, completed
0.0432
0.0406
1024 256 0.0398 T ransitional, completed
Slot A 7.37 107 (6.7 m/s) SA-W F 0.0389
0.0345
0.0259 B D R, completed
0.0086
2048 256 0.0864 A L D R, completed
16384 256 0.0864 A L D R, completed
Note: Q*=Qa/(U’· Ainj)

The drag reduction along the plate with an air flow rate of 15.3 m3/min is shown in Fig. 4 where drag reduction
of nearly 100% is achieved over the entire plate, which matches the EFD data very well. As shown in Fig. 5, the air
layer is formed along the whole length of the plate. It should be noted that at the early stage of the computations
water spots are found on the plate and gradually were swept to downstream with the incoming flow. The air flow
rate is then reduced to 2.55 m3/min (the lowest in Test 1 of Ref. 3) where a thinner air layer is achieved with the
drag reduction approximately 100%. As discussed in Ref. 3, the increase of air flow rate does not apparently
increase the drag reduction when air layer is formed. The velocity vector fields with the interface profile overlapped
at two different locations along the plate are plotted in Fig. 6. Near the air injector, velocity changes sharply across
the interface, whereas in the downstream velocity field across the interface is smooth.

5
American Institute of Aeronautics and Astronautics
000000
 120
ALDR

100
EFD_air_15.3
EFD_air_2.55
CFD_air_15.3
80
CFD_air_2.55
CFD_air_1.1985
CFD_air_1.173

%DR
CFD_air_1.1475
60 CFD_air_1.02
CFD_air_0.765
CFD_air_0.255
Transitional
40

20

BDR

0
0 2 4 6 8 10
X-Xinj(m)
F igure 4. % D R versus X-X inj with different air flow rates Q a . U ’ =6.7 m/s.

F igure 5. A ir layer profile along the plate. U ’ =6.7 m/s, Q a =15.3 m3/min.

0.0015

0.0005
0.001

0.0005 0
X

0
-0.0005

-0.0005
-0.001

-0.001
0.106 0.107 0.108 0.3015 0.302 0.3025 0.303 0.3035
a) Z b) Z
F igure 6. V elocity vector fields along the plate. a) Near the air injector; b) In the middle of plate. U ’ =6.7 m/s,
Q a =15.3 m3/min.
In order to investigate the effect of air flow rate on the formation of the ALDR, the air flow rate is then further
reduced to a quite small value of 0.255 m3/min. Then the air flow rate is gradually increased to 0.765, 1.02 and
1.1475 m3/min later. As shown in Fig. 4, bubble drag reduction is obtained when the air flow rate is decreased to
0.255, 0.765, 1.02 m3/min. For each case, high levels of drag reduction are achieved only near the air injector and
then decay rapidly with downstream distance. Figure 4 also indicates that the increase of air flow rate apparently
increases %DR near the air injector and its effect becomes insignificant with downstream distance. This agrees with
the BDR features observed in the experiments. As the air flow rate reaches 1.1475 m3/min, apparent %DR increase
is achieved along the entire plate, which is close to the transitional region obtained in the experiment with an air
flow rate of 2.55 m3/min. It should be noted that the threshold air flow rate necessary to form ALDR is smaller in
the present simulation than in the experiments. This deviation from the experiment is probably due to the
compressed air injection used in the experiment which is difficult to model in the present study. The grid
6
American Institute of Aeronautics and Astronautics
000000
resolutions, surface roughness, in flow conditions and surface tension effect are also possible reasons, which will be
investigated in the future work. The interface profiles along the plate for BDR, transitional, and ALDR are shown in
Fig. 7. The predicted bubble size is almost one order of magnitude larger than the EFD results. %DR versus air flow
rate plot at x = 6 to 7 m is shown in Fig. 8. BDR is observed over the lower-range of air injection rates, where %DR
decreases with downstream distance. The transitional region shows a Rapid %DR increase with air flow rates in
most sections of the plate. For the ALDR, high %DR is obtained with no apparent decay with downstream distance
over the entire plate. The experimental results indicate that the critical air flow rate increases with Reynolds number.
Since the initial coarse grid is designed for the boundary layer and not enough to resolve the interface changes,
the streamwise grid is refined to 2048 and 163824, respectively. The air flow rate is 2.55 m3/min (transitional region
in Test 1of Ref. 3). Although interface instabilities are observed on the refined grids, %DR is still more than 90% as
shown in Fig. 8. Grid refinement in the wall normal direction might be needed to further resolve the interface
instability. Another very possible reason is due to the two-phase interfacial turbulence modeling. In the present
study, the single-phase based RANS equations have been used. The effect of turbulence on the interface and the
interface induced turbulence are not considered. For example, the eddy viscosity is found to be over-predicted near
the interface in RANS (similarly for LES see the work by Liovic and Lakehal15). As a result, the small deformations
of the interface are smeared out. This problem is especially serious when the interface motion owing much to the
turbulence induced disturbances. As demonstrated in a plane water jet test by Shirani et al.16 using a single-phase
based RANS model, the jet does not spread but keeps a uniform cross section throughout the entire jet length. A
similar problem is encountered here, the interface does not break up into bubbles but remains a smooth air layer at
low air flow rates. As a result of Reynolds-averaging process (or filtering for LES) of RANS, some additional terms
should emerge in the momentum and interface advection equations. Hong and Walker17 have developed a set of
Reynolds-averaged equations for the two-phase interfacial flows. However, these equations do not represent a
closed system of equations and no model is introduced. Shirani et al. 16 developed a model to include the interface
effects in the standard RANS models, and a model for the correlation of the mean fluctuations of VOF with velocity.
These models have been applied for the simulation of a plane water jet, very promising results are obtained. This
model will be implemented in the future work. It should be noted that it is the first attempt for the two-phase
interfacial turbulence modeling considering the interactions of the interface and turbulence. Further tests are
required0.002
to determine the coefficients of models and derive more sophisticated
0.002 models.
X

0 0

-0.002 0.002 -0.002

0.46 0.48 0.5 0.46 0.48 0.5

a) Z b) Z
X

-0.002

0.46 0.48 0.5

c) Z

F igure 7. Interface profile along the plate with various air flow rates (magnified view at the middle part),
U ’ =6.7 m/s. a) B D R (Q a =1.02 m3/min), bubble size: 0.48 to 4.5 mm; b) T ransitional (Q a =1.20m3/min); c)
A L D R (Q a =2.55 m3/min).

7
American Institute of Aeronautics and Astronautics
000000
 100
ALDR

80

%DR (at x= 6.0~7.34 m) 60

CFD: Uf=6.7m/s, solid, coarse

CFD: Uf=6.7 m/s, solid, medium
40 EFD: Uf=6.7m/s, solid
EFD: Uf=6.7m/s, porous
Transitional EFD: Uf=11.1m/s,porous
EFD: Uf=13.3m/s, solid
EFD: Uf=13.3 m/s, porous
EFD: Uf=20.0 m/s, solid
EFD: Uf=6.5m/s, new
CFD: Uf=6.7m/s, solid, fine
20
BDR

0
0 0.02 0.04 0.06 0.08 0.1
2 -1
q(m s )
F igure 8. % D R versus air flow rate plot.

I V . Summary and F uture Wor k

The air layer formation in a high Reynolds-number flat plate turbulent boundary layer is simulated using a two-
phase URANS solver. The air layer along the entire test plate is successfully achieved and the drag reduction is
approximately 100%, which agrees with the experimental findings very well. With reduced air flow rate, BDR is
also observed; the computational results also qualitatively match the experiments. The transitional region from BDR
to ALDR is also observed in the present simulation. However, the critical air flow rate to form the ALDR is lower in
the simulations than in the experiments. Several possible reasons are likely accounting for the low critical air flow
rate in the simulations, such as SA-WF turbulence model, three-dimensional instability and surface tension effects.
The critical air flow rate does not change much with grid refinement.
In the future work, the computational results will be further verified and validated. The low air flow rate issue
will be investigated. The RANS model in the framework of two-equation k-epsilon model proposed by Shirani et
al.16 will be implemented and evaluated in order to account for the interactions between the interface and turbulence.
Advanced turbulence models, such as SST Two-Equation model, hybrid RANS/LES, will be considered. Three
dimensional LES simulations at relatively low Reynolds numbers will also be conducted. Three-dimensional
instability and surface tension effect will be investigated. The mechanism of the formation of the ALDR, transitional
region and BDR, flow structure, and interface instability, will be analyzed in the future work.

Acknowledgments
The Office of Naval Research under Grant N00014-08-1-0620, administered by Dr. Patrick Purtell, sponsored
thLV UHVHDUFK XQGHU WKH SURMHFW ³Two-Phase URANS/DES Simulations of Large Scale/RE Air Layer Drag
5HGXFWLRQ´. The simulations were performed using DoD HPC resources.

References
1
Merkle, C., and Deutsch, S., "Microbubble Drag Reduction in Liquid Turbulent Boundary Layers," Appl. Mech. Rev., Vol.
45, No. 3, 1992, pp. 103-127.
2
Sanders, W.C., Winkel, E.S., Dowling, D.R., Perlin, M., and Ceccio, S.L., "Bubble Friction Drag Reduction in a High-
Reynolds-Number Flat-Plate Turbulent Boundary Layer," J. F luid Mech., Vol 552, 2006, pp. 353-380.
3
Elbing, B.R., Winkel, E.S., Lay, K.A., Ceccio, S.L., Dowling, D.R., and Perlin, M., "Bubble-Induced Skin-Friction Drag
Reduction and the Abrupt Transition to Air-Layer Drag Reduction," J. F luid Mech., Vol. 612, 2008, pp. 201-236.
4
Ceccio, S.L, "Evaluation of Air Cavity Lubrication at High Reynolds Numbers," Air Layer Drag Reduction Progra m Review
Meeting, Stanford University, Stanford, California, September 15, 2009.
5
Amromin, E., and Mizine, I., "Partial Cavitation as Drag Reduction Technique and Problem of Active Flow Control,"
Marine Technology, Vol. 40, 2003, pp. 181-188.
8
American Institute of Aeronautics and Astronautics
000000
 6
Yang, J., and Stern, F., "Sharp Interface Immersed-Boundary/Level-Set Method for Wave±Body Interactions," Journal of
Computational Physics, Vol. 228, 2009, pp. 6590-6616.
7
Wang, Z., Yang, J., Koo, B., and Stern, F., "A Coupled Level Set and Volume-of-Fluid Method for Sharp Interface
Simulation of Plunging Breaking Waves," International Journal of Multiphase F low, Vol. 35, 2009, pp. 227-246.
8
Bhushan, S., Carrica, P.M., Yang, J., DQG6WHUQ)³6FDODELOLW\6WXG\DQGLarge Grid Computations for Surface Combatant
Using CFDship-,RZD´ International Journal of High Performance Computing, 2010 (under revision).
9
Spalart, P. R., and Allmaras, S. R., "A One-Equation Turbulence Model for Aerodynamic Flows," AIAA-Paper 92-0439, No.
1, 1992, pp. 5-21.
10
Mariani-D, J., Zilliac, G. G., Chow, J. S., and Bradshaw, P., "Numerical/Experimental Study of a Wingtip Vortex in the
Near Field," AIAA Journal , Vol. 33, No. 9, 1995, pp. 1561-1568.
11
Shih, T. H., Povinelli, L.A., and Liu, N. S., "Application of Generalized Wall Function for Complex Turbulent Flows,"
Journal of Turbulence, Vol. 4, 2003, pp. 1-16.
12
Sussman, M., Smereka, P., and Osher, S., "A level set approach for computing solutions to incompressible two-phase flow,"
Journal of Computational Physics, Vol. 114, 1994, pp. 146-159.
13
Leonard, B.P., ³$VWDEOHDQGDFFXUDWHmodeling SURFHGXUHEDVHGRQTXDGUDWLFLQWHUSRODWLRQ´ Comput. Methods Appl. Mech.
Engrg., Vol. 19, 1979, pp. 58-98.
14
Falgout, R.D., Jones, J.E., and Yang, U.M., "The Design and Implementation of HYPRE, a Library of Parallel High
Performance Preconditioners," Numerical Solution of Partial Differential Equations on Parallel Computers , A.M. Bruaset and
A. Tveito, eds., Springer-Verlag, Vol. 51, 2006, pp. 267-294.
15
Liovic, P., and Lakehal, D., "Multi-Physics Treatment in the Vicinity of Arbitrarily Deformable Gas±Liquid Interfaces,"
Journal of Computational Physics, Vol. 222, 2007, pp. 504-535.
16
Shirani E, Jafari A, and Ashgriz N, "Turbulence Models for Flows with Free Surfaces and Interfaces," AIAA Journal, Vol.
44, No. 7, 2006, pp. 1454-1462.
17
Hong, W.L., and D.T. Walker, D.T., "Reynolds-Averaged Equations for Free-Surface Flows with Application to High-
Froude-Number Jet Spreading," J. F luid Mech., Vol. 417, 2000, pp. 183-209.

9
American Institute of Aeronautics and Astronautics
000000