See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222840523

The extended finite element method for two-phase and free-surface flows: A
systematic study

Article in Journal of Computational Physics · May 2011

DOI: 10.1016/j.jcp.2011.01.033 · Source: DBLP

CITATIONS READS

91 629

2 authors, including:

Henning Sauerland
Hitachi Europe GmbH
20 PUBLICATIONS 424 CITATIONS

SEE PROFILE

All content following this page was uploaded by Henning Sauerland on 11 January 2018.

The user has requested enhancement of the downloaded file.

 The extended finite element method for two-phase and
free-surface flows: A systematic study

Henning Sauerlanda,∗, Thomas-Peter Friesa

a Chair for Computational Analysis of Technical Systems, Center for Computational Engineering Science (CCES),
RWTH Aachen University, Schinkelstr. 2, 52062 Aachen, Germany

Abstract
In immiscible two-phase flows, jumps or kinks are present in the velocity and pressure fields
across the interfaces of the two fluids. The extended finite element method (XFEM) is able to re-
produce such discontinuities within elements. Robust and accurate interface capturing schemes
with no restrictions on the interface topology are thereby enabled. This paper investigates dif-
ferent enrichment schemes and time-integration schemes within the XFEM. Test cases with and
without surface tension on moving or stationary meshes are studied and compared to interface
tracking results when possible. A particularly useful setting is extracted which is recommended
for two-phase flows. An extension of this formulation for the simulation of free-surface flows
and of floating objects is proposed.
Keywords: extended finite element method, XFEM, two-phase flows, free-surface flows,
enrichment

1. Introduction

The numerical simulation of time-dependent physical problems with moving interfaces is a

challenging field of computational mechanics. Two-phase flows incorporate kinks in the field
variables across the moving interfaces due to the density and viscosity differences between the
phases. Under consideration of surface-tension effects, the pressure field is even discontinuous.
If these discontinuities occur inside elements they can not be accurately represented by standard
finite element shape functions [1]. Furthermore, for certain flow configurations, topological
changes of the phases may occur which the numerical method has to account for. Two-phase
and free-surface flows have many industrial applications and have therefore a long tradition in
computational mechanics [2, 3, 4]. Applications can be found in the field of liquid sprays, fuel
tank sloshing, molding, film coating, bubble column reactors, steam generators/turbines, ink-jet
printing, hydraulic design of dams, design of ship hulls etc.
One distinguishes two basic approaches for problems with moving interfaces: interface track-
ing and interface capturing. In interface tracking, the interfaces are meshed appropriately such
that element edges align with the interface. The mesh follows the moving interface during the

∗ Corresponding author
Email addresses: [email protected] (Henning Sauerland), [email protected]
(Thomas-Peter Fries)
Preprint submitted to Journal of Computational Physics January 18, 2011
simulation. Accurate results are obtained in the context of classical finite element approxima-
tions. However, topological restrictions for the interface apply as automatic mesh movement can
not, in general, handle topological changes.
In contrast, interface capturing methods describe the interface implicitly, often by level-set
methods or the volume of fluid (VOF) method. The interface is then interpreted as the isoline
of an auxiliary scalar function in the domain. Free movement of the interface is possible, as no
topological restrictions apply. However, as the interface and, consequently, the discontinuities
are now within elements, classical finite element simulations often perform poorly. A possible
remedy is a mesh refinement close to the interface where the error is large. Another alternative is
the use of the extended finite element method (XFEM) [5, 6, 7]. The classical FE approximation
space is locally enriched by functions that enable the exact approximation of discontinuities
(jumps or kinks) within elements. Combining interface capturing with the XFEM thus has the
potential to provide accurate results for problems with moving interfaces without topological
restrictions on the movement.
The XFEM has been used before in the context of fluid mechanics. Chessa and Belytschko
applied the XFEM to two-phase flows with and without surface-tension effects solving the Navier-
Stokes equations [8, 9]. The absolute value of the level-set function (abs-enrichment) is used to
account for the kink in the velocity field. Gross and Reusken [10] consider 3D two-phase flows
with surface tension using the XFEM. The resulting jump in the pressure field is treated by en-
riching the pressure approximation space using a Heaviside function. Additionally, their tetrahe-
dral meshes are adaptively refined near the interface. Kölke [11] uses the signum function of the
level-set field to enrich both the velocity and the pressure function space. Lagrange multipliers
are used to enforce the C0 -continuity of the enriched quantities if required. The intrinsic XFEM
of Fries and Belytschko [12] is applied to two-phase flows in [13]. In contrast to the standard,
extrinsic XFEM approach, the intrinsic formulation does not introduce additional unknowns.
An enrichment approach similar to the XFEM can be found in [14]. Here, Minev et al. use a
Heaviside function as a pressure enrichment and a bubble function enrichment to account for the
discontinuous normal velocity gradient in two-phase flows with surface-tension. Furthermore,
an integration technique is introduced which does not require the knowledge of intersections of
the interface with the element edges. In [15], Coppola-Owen and Codina introduce a new enrich-
ment function for discontinuous pressure gradients in two-phase flows, which is zero at the cut
element nodes and has a constant gradient on each side of the interface. Thereby, the additional
degrees of freedom can be condensed prior to the assembly. Zlotnik and Dı́ez [16] generalize the
abs-enrichment function by Moës et al. [17] for n-phase flow problems, where multiple interfaces
cross an element. They enrich the velocity and pressure approximation space for the numerical
simulation of Stokes flow problems. In [18], combinations of the abs-enrichment by Moës and
the Heaviside-enrichment are numerically studied for mixed finite elements in the framework of
incompressible materials. There, the physical situation is similar to two-phase Stokes flow.
When using the XFEM for the simulation of two-phase flows, several enrichment schemes
can be employed: Velocity and/or pressure fields may be enriched, enrichments for kinks or
jumps may be used, etc. Furthermore, the treatment of the moving interface in time, i.e. the time
integration, has to be considered with special care in the XFEM. Compared to other studies, this
work for the first time, systematically studies the possible alternatives in the context of inter-
face capturing and XFEM for two-phase flows. Another unique aspect is the extension of the
XFEM flow solver to moving meshes and situations with inflow and outflow boundaries (instead
of closed containers). The resulting XFEM two-phase flow solver is applied for the simulation of
free-surface flows and its potential in the simulation of floating bodies is demonstrated. There-
2
 Figure 1: Computational domain.

fore, the given free-surface problem is reformulated as a two-phase problem where one fluid has
negligible density and viscosity. The motion of a floating body is accounted for by means of an
additional interface tracking approach.
The remainder of the paper is structured as follows: Section 2 introduces the governing
equations. Subsequently, the XFEM and different enrichment schemes are introduced in Section
3, followed by description of the used time integration schemes in Section 4. Section 5 introduces
the surface tension formulation and several numerical test cases are shown in Section 6. A final
conclusion is drawn in Section 7.

2. Governing equations
Consider a two-dimensional computational domain Ω ⊂ R2 with boundary Γ = ∂Ω. The
boundary is decomposed into a Dirichlet and Neumann boundary, Γu and Γh respectively, forming
a complementary subset of the boundary Γ, i.e. Γu ∪Γh = ΓandΓu ∩Γh = ∅. The normal vector on
Γ is denoted by n. The domain Ω encloses two immiscible Newtonian fluids in Ω1 (t) and Ω2 (t).
These two phases are separated by a moving interface Γd (t), where n̂ is the normal vector on Γd
as shown in Figure 1.
In this work, we also consider numerical examples where the mesh is moving. Therefore,
the governing equations are given in arbitrary Lagrangian-Eulerian (ALE) form [19]. That is,


the convective velocity is defined as u = u − um , with um = x(tn+1 ) − x(tn ) /∆t being the mesh
velocity and ∆t the constant time step. In these cases, the mesh movement will be treated with a
pseudo-structure approach, as for example in [20]. The fluid velocity u(x, t) and pressure p(x, t)
for each phase j = 1, 2 are governed by the instationary, isothermal, incompressible Navier-
Stokes equations in velocity-pressure formulation:
!
∂u
ρj + u · ∇u − f − ∇ · σ = 0 in Ω j (t) × [0, T ],
∂t (1)
∇ · u = 0 in Ω j (t) × [0, T ],
with ρ j being the density of the respective fluid phase. The stress tensor σ is defined by
1 
σ(u, p) = −pI + 2µ j ε(u) with ε(u) = ∇u + (∇u)T , (2)
2
3
where µ j is the corresponding dynamic viscosity and I the identity tensor. The Dirichlet and
Neumann boundary conditions on the boundary Γ are given by

u = û on Γu × [0, T ], (3)
n · σ = ĥ on Γh × [0, T ], (4)

where û and ĥ are prescribed velocity and stress values. At the interface, typically the following
conditions are prescribed

[u]Γd = 0 on Γd (t) × [0, T ], (5)

[ n̂ · σ]Γd = γκ n̂ on Γd (t) × [0, T ]. (6)

γ is the surface tension coefficient, κ the curvature of Γd , and [ f ]Γd defines the jump of f across
the interface Γd . Furthermore, a divergence free initial velocity field is required:

u(x, 0) = û0 (x) in Ω(t = 0). (7)

2.1. Interface description

One can generally distinguish between two approaches for the description of the interface.
Interface tracking methods rely on an explicitly meshed interface, while interface capturing meth-
ods describe the interface implicitly. While the latter group of methods can easily deal with
topological changes of the interface, interface tracking methods are known to be more accurate.
In the framework of the XFEM, typically the level-set method [21, 22]—an interface captur-
ing approach—is used in order to describe the inner-element interface Γd . In combination with
XFEM, the lack of accuracy can be overcome, as will be seen later. Here, the zero-level of
the scalar level-set function φ represents the interface. In this work, φ has the signed-distance
property [21]:

φ(x) = ± min
∗
kx − x∗ k , ∀x ∈ Ω. (8)
x ∈Γd

That is, each point x in the domain stores the shortest distance to the interface. The sign depends
on which of the two fluids is present at x. If the interface Γd is moving throughout the simulation,
φ(x, t) needs to be updated in each step. Therefore, the level-set transport equation
∂φ
+ u(x, t) · ∇φ = 0 in Ω × [0, T ] (9)
∂t
is solved, where u(x, t) is the convective fluid velocity as defined before and

φ(x, 0) = φ̂0 (x) in Ω(t0 = 0) (10)

is the initial level-set function. Equation (9) is hyperbolic, hence, boundary conditions have to
be prescribed at inflow boundaries Γφ :

φ = φ̂ on Γφ × [0, T ]. (11)

Due to the transport of the level-set function, the signed-distance property (8) is lost. A
reinitialization of φ is thus necessary in order to recover (8). Simple reinitialization approaches
modify the interface position during the reinitialization. Performing such an approach in every
4
 Figure 2: Coupling.

time step means that with smaller time steps the overall influence of the error due to the reinitial-
ization increases, as more reinitialization steps are performed. In order to avoid this we specify
a fixed number of reinitialization steps for a computation, irrespective of the time step size.
It is important to note that (1) and (9) pose a strongly coupled problem. That is, there is
a mutual influence of the level-set function φ on the fluid velocity field and vice versa. If a
combination of interface tracking and interface capturing is used, e.g. for two-fluid-structure
interaction (as shall be seen below), the coupling of the fluid and the interface is furthermore
coupled with the moving structure which influences the fluid meshes through a pseudo-structure
approach [20], see Fig. 2.

3. Spatial discretization and XFEM

Due to the density and viscosity differences between the phases in a two-phase flow problem,
we encounter discontinuities in the velocity and pressure fields along the interface. We classify
two different types of discontinuities: weak and strong discontinuities. A strong discontinuity is
characterized by a jump in the function (e.g. the pressure field of a two phase flow with surface
tension), whereas a weak discontinuity features a kink in the function, that is, a jump in the
derivative (e.g. the velocity field in a two phase flow). The XFEM accounts for these jumps and
kinks by enriching the approximation space [5, 6]. The approximation is
X X
gh (x, t) = Ni (x, t)gi + Mi (x, t)ai . (12)
∗
|i∈I {z i∈I
} | {z }
strd. FE approx. enrichment

Ni (x, t) is the standard FE shape function (here: linear) for node i, I is the set of all nodes in the
domain, Mi (x, t) are the local enrichment functions, gi are the nodal variable values, ai are the
additional XFEM unknowns, and I ∗ is the nodal subset of the enrichment. Figure 3 illustrates
which nodes will be enriched for an exemplary interface in a domain. It is seen that all nodes
are enriched which belong to elements cut by the interface. Elements with only some enriched
nodes are called blending elements (cf. Fig. 3). It is well-known in the XFEM that these partly
enriched elements can lead to problems, see e.g. [23, 24].
Mi (x, t) is further specified as follows [25]:
 
Mi (x, t) = Ni (x, t) · ψ(x, t) − ψ(xi , t) ∀i ∈ I ∗ , (13)

5
 Figure 3: Enriched domain.

with ψ(x, t) being the global enrichment function. It is noted that (13) is the so-called shifted
enrichment which ensures Kronecker-δ property of the overall approximation (12). A global
enrichment function that is typically chosen for strong discontinuities is the sign-enrichment:




 −1 : φ(x, t) < 0,


ψsign (x, t) = sign (φ(x, t)) = 
 0 : φ(x, t) = 0, (14)



 1 : φ(x, t) > 0.

It is to be noted that the piecewise-constant enrichment function (14), together with the shifted
formulation (13), does not lead to any problems in blending elements [7]. For weak discontinu-
ities, the abs-enrichment

ψabs (x, t) = abs (φ(x, t)) (15)

was originally proposed in [25, 26]. However, due to problems in blending elements, this ap-
proach only achieves suboptimal convergence rates. A number of different approaches exists
which circumvent problems with blending elements [24, 27, 28, 23, 29]. Here, we employ the
abs-enrichment by Moës et al. [17] which is zero over the blending elements:

X X
ψabs (x, t) = |φi | Ni (x, t) − φi Ni (x, t) . (16)
i∈I i∈I

Remark In general, enrichment functions Mi (x, t) with small support can occur which leads to
an ill-conditioned system matrix. This has to be considered if iterative solvers are applied or if
roundoff errors are dominant. See Reusken [30] and Béchet et al. [31] for two approaches to deal
with this problem.

3.1. Enrichment schemes

We propose different enrichment schemes for the velocity and pressure fields. They are
compared later on with numerical test cases in Section 6.

6
3.1.1. Enrichment of velocity and pressure space (u-abs, p-abs and u-abs, p-sign)
We recall that for two-phase flow problems without surface tension, weak discontinuities in
the velocity and pressure field exist. With surface tension, the pressure field shows a jump at
the interface. It appears natural to employ the abs-enrichment (16) for the velocity field, and—
depending on whether surface tension effects are considered or not—either the sign-enrichment
(14) or the abs-enrichment (16) for the pressure field. Thereby, all discontinuities are accounted
for appropriately in the context of the XFEM.
X X  
uh (x, t) = Ni (x, t)ui + Ni (x, t) · ψabs (x, t) − ψabs (xi , t) ai , (17)
i∈I i∈I ∗
X X h i
h
p (x, t) = Ni (x, t)pi + Ni (x, t) · ψabs/sign (x, t) − ψabs/sign (xi , t) bi . (18)
i∈I i∈I ∗

However, it turned out that enriching the velocity space does not improve the results signifi-
cantly. Instead, the number of iterations in order to solve the non-linear governing equations of
immiscible two-phase flows, see eq. (1), increased noticeably. In some cases, these convergence
problems were severe and no convergence could be achieved. Many test-cases confirmed that
this potential instability has its source in the enrichment of the velocity space only. In contrast,
enriching the pressure field exclusively improves the results dramatically. Hence, we will neglect
the enrichment of the velocity space in the following.

Remark We want to emphasize that the problems with the velocity enrichment are in terms of
robustness. From a theoretical point of view, an enrichment of the velocity field also leads to a
more accurate approximation of the solution provided that a stable solution can be found, which
is, however, often not possible.

Remark The method described in this work is mainly applied to physical problems dominated
by gravitational forces, i.e. small errors in the pressure can lead to large errors in the velocity.
Therefore, an improvement of the pressure approximation is more beneficial than modifying the
velocity approximation.

3.1.2. Abs-enrichment of the pressure space (p-abs)

The first enrichment scheme used in the following only considers the enrichment of the pres-
sure space using the abs-enrichment (16). Only enriching the pressure space, together with
the fully implicit pressure treatment, leads to an important advantage when using time-stepping
schemes. In [32], it is discussed that some integrals in the weak formulation (see below) involve
functions to be evaluated at time level tn and tn+1 . It is shown that it is important in this case to
use a more elaborate quadrature which considers the location of the interface at both time levels,
Γd (tn ) and Γd (tn+1 ). Otherwise, it is not possible to achieve optimal convergence. Due to the fully
implicit treatment of the pressure, we have to deal only with one interface position at each time
step if only the pressure is enriched. The normal element subdivision (see e.g. [5]) together with
standard quadrature rules can then be used in the XFEM.
X
uh (x, t) = Ni (x, t)ui , (19)
i∈I
X X  
h
p (x, t) = Ni (x, t)pi + Ni (x, t) · ψabs (x, t) − ψabs (xi , t) bi . (20)
i∈I i∈I ∗

7
3.1.3. Sign-enrichment of the pressure space (p-sign)
Using the sign-enrichment (14) for the pressure space would be a reasonable choice for two-
phase flows under consideration of surface tension effects. Even though we are experiencing
only a weakly discontinuous pressure field for two phase flows without surface tension, one
may use the sign-enrichment also in this case (considerations from functional analysis allow us
to use discontinuous pressure spaces anyway). Using the sign-enrichment instead of the abs-
enrichment, the pressure is approximated by an even larger approximation space which also
allows the pressure to be strongly discontinuous (instead of only weakly discontinuous). Thus, it
is reasonable to assume that this enrichment leads to even better results—with of without surface
tension—compared to p-abs which is confirmed in the numerical results.
X
uh (x, t) = Ni (x, t)ui , (21)
i∈I
X X h i
h
p (x, t) = Ni (x, t)pi + Ni (x, t) · ψ sign (x, t) − ψ sign (xi , t) bi . (22)
i∈I i∈I ∗

Remark The application of a discontinuous enrichment to a weakly discontinuous field can be

justified by the fact that weak discontinuities are also covered by the discontinuous enrichment.
At this point we also want to refer to a publication by Idelsohn et al. [33]. In numerical studies
they show that in incompressible two-phase flows, a pressure jump occurs not only due to surface
tension effects, but also due to viscosity differences. The theoretical foundation of this interface
condition was shown in [34]. The pressure jump induced by the viscosity jump is assumed to
be small. Nevertheless, this finding can be stated as another reason why the sign-enrichment
performs better compared to the abs-enrichment even in cases without surface tension.

4. Temporal discretization

A detailed study of time integration in the XFEM can be found by Fries and Zilian in [32].
Here, two different time discretization methods are applied to the coupled system ((1) and (9)).
On the one hand, the recommended optimal one-step time-stepping scheme from [32] is used
and, on the other hand, the computationally more expensive discontinuous Galerkin method in
time (space-time elements).

4.1. Time-stepping with the trapezoidal rule

As proposed in [32], the trapezoidal rule is used as a time-stepping scheme. The Navier-
Stokes equations (1) may be written as

∂u
= F(u, t). (23)
∂t
This partial differential equation is then replaced by

un+1 − un 1
= θ · F(un+1 , tn+1 ) + (1 − θ) · F(un , tn ), θ= , (24)
∆t 2

8
with time step size ∆t. The SUPG/PSPG-stabilized weak form, for simplicity written without the
insertion of (24) follows to: Find uh ∈ Shu and ph ∈ Shp such that ∀wh ∈ Vuh , ∀qh ∈ Vhp :
Z ! Z
∂uh
wh · ρ j + uh · ∇uh − f dΩ + ε(wh ) : σ(uh , ph ) dΩ
Ω ∂t Ω
Z nel Z !
h h
X
h h 1 h
+ q ∇ · u dΩ + τs u · ∇w + ∇q
Ω e=1 Ωel
e ρj
" ! # (25)
∂uh h
· ρj + u · ∇uh − f − ∇ · σ(uh , ph ) dΩ
∂t
Z Z
= wh · ĥ dΓ + γκwh · n̂ dΓ,
Γh Γd

with nel being the number of elements and ε and σ defined in eq. (2). The pressure and the
continuity equation are treated fully implicitly and the force term is assumed to be stationary
and constant. Shu , Vuh , Shp , and Vhp are the velocity and pressure function spaces spanned by
the finite element shape functions and the enrichments as discussed in the previous section. The
stabilization parameter τs is chosen according to [35]:
 ! !2 !2 − 21
 2 2 2|uh
| 2 4ν 
τs =  + + 2  (26)
∆t he he
with ν = µ/ρ the kinematic viscosity and he the element length (see e.g. [36] for details). The
density and viscosity in elements which are not cut by the interface are given by the respective
phase occupying those elements. However, in cut elements µ and ρ are averaged using µ1 , µ2 , ρ1 ,
ρ2 , and the relative ratio of the area occupied by each phase in the element.
In the discretized domain, the discrete level-set function values are interpolated in the same
way as the unknowns in the standard FEM,
X
φh (x) = Ni (x)φi , (27)
i∈I

using the standard linear FE shape functions Ni (x). Due to the convective type of the level-set
transport equation (9), the SUPG stabilization is applied here, too. The weak formulation is then
defined as: Find φh ∈ Shφ such that ∀wh ∈ Vφh :
Z !
∂φh
wh · + uh · ∇φh dΩ
Ω ∂t
Xnel Z  h  ∂φh ! (28)
h h h
+ τs u · ∇w · + u · ∇φ dΩ = 0.
e=1 Ωel
e ∂t

The stabilization parameter τs is given by (26) with ν = 0.

Fig. 4 shows the interaction of the fields. The flowchart shows the outer loop over time
(time loop, superscript n) and the inner loop representing the coupling between the level-set
transport equation and Navier-Stokes equations (level-set loop, superscript i); mesh motion is
not considered in Fig. 4. The level-set loop is executed until the condition ||φi+1 − φi || < ε is
satisfied for a specified bound ε. The Navier-Stokes equations are solved in a series of Picard
iterations.
9
Figure 4: Solver flowchart.

10
 Figure 5: Space-time slab.

4.2. Discontinuous Galerkin method in time

The second time discretization used in this work is the discontinuous Galerkin method in
time [37] resulting in the use of space-time elements. Here, the time interval is divided into
subintervals In = (tn , tn+1 ) where tn and tn+1 belong to an ordered set of time levels 0 = t0 < t1 <
· · · < tN−1 < tN = T . The space-time slab Qn is then defined as the domain enclosed by the
computational domains at time levels tn and tn+1 , say Ω(tn ) and Ω(tn+1 ) and Pn , where Pn is the
surface described by the boundary Γ as t traverses In (cf. Fig. 5). In case of a fixed mesh Ω(tn )
and Ω(tn+1 ) are equal.
The stabilized space-time formulation
  for the Navier-Stokes equations (1) can then be
 written
h −
   
as follows: Given u , find u ∈ Su and p ∈ S p such that ∀w ∈ Vu , ∀q ∈ Vhp :
h h h h h h h
n n n n n

Z ! Z
∂uh
wh · ρ j + uh · ∇uh − f dQ + ε(wh ) : σ(uh , ph ) dQ
Qn ∂t Qn
Z nel Z !
h h
X ∂wh h h 1 h
+ q ∇ · u dQ + τs + u · ∇w + ∇q
Qn e=1 Qn
e ∂t ρj
" ! #
∂uh h h h h (29)
· ρj + u · ∇u − f − ∇ · σ(u , p ) dQ
∂t
Z  +  +  − 
+ wh · uh − uh dΩ
n n n
Ω(tn )
Z Z
= wh · ĥ dP + γκwh · n̂ dP
(Pn )h (Pn )d

using the following notations:

 ±
uh = lim u(tn ± ε),
n ε→0
Z Z Z Z Z Z
. . . dQ = . . . dΩ dt, . . . dP = . . . dΓ dt.
Qn tn Ωh (tn ) Pn tn Γh (tn )

Formulation
 + (29) is subsequently
   applied  to all space-time
 slabs Q1 , Q2 , . . . , QN−1 starting
h h h h h
with u 0 = û0 . Su n , Vu n , S p n , and V p n are suitable velocity and pressure function
spaces for each time slab composed by the XFEM approximation as discussed in Section 3.1.
The stabilization parameter τs is given by (26).
11
 (a) Element with real interface. (b) Subdivision with real inter- (c) Subdivision with linear inter-
face. face.

Figure 6: Reference element decomposition for integration purposes in hexahedral space-time elements.

Thestabilized space-time
  formulation of the
 level-set transport equation (9) follows as: Given
h −
 
h h h h h
u and φ find φ ∈ Sφ such that ∀w ∈ Vφ :
n n n
Z !
∂φhh h h
w · + u · ∇φ dQ
Qn ∂t
nel Z ! !
X ∂wh ∂φh
+ τs + uh · ∇wh · + uh · ∇φh dQ (30)
e=1 Qn
e ∂t ∂t
Z  +  +  − 
+ wh · φh − φh dΩ = 0
n n n
Ω(tn )

The stabilization parameter τs is given by (26) with ν = 0.

More details on special considerations when using time-stepping schemes or the space-time
method with the XFEM can be found in [32].
It should be noted that the quadrature of the weak forms (25), and (29) requires special
consideration when using the XFEM. The used approximation spaces are enriched with discon-
tinuous functions. Therefore, some of the integrands in the weak formulations are discontinuous
as well. However, standard Gauss quadrature requires smoothness of the integrands. In order
to obtain an accurate numerical quadrature, the cut elements are subdivided for integration pur-
poses such that they align with the interface. Standard Gauss rules can then be applied in these
subelements resulting from the decomposition. For details we refer to e.g. [5, 6].
We only go into details for the discontinuous Galerkin method in time, leading to the use
of 3D hexahedral space-time elements. In order to achieve subcells for integration purposes,
the hexahedra are decomposed into tetrahedra, cf. (Fig. 6(b)). Then, using the level-set values
at the time levels tn and tn+1 of the space-time slab, a planar interface in each tetrahedron is
reconstructed assuming a linear interpolation, see Fig. 6(c). Consequently, these tetrahedra are
then decomposed into tetrahedra or pentahedra aligned with the interface (Fig. 7). Note that the
planar interfaces in the reference element are generally curved in the real element. Standard
Gauss rules can finally be applied in the subcells of these decomposed 3D space-time elements.
In this sense, the moving interfaces in the space-time domain are consistently considered for.

12
 (a) Triangular interface. (b) Quadrilateral interface.

Figure 7: Possible interface cutting planes in tetrahedral subcells.

5. Surface tension

In this work, we consider surface tension effects in some test cases. Fluid particles which
are placed on the interface experience an inbound force (cohesion force). This is because of the
unsymmetric arrangement of the neighboring particles at the interface, leading to unbalanced
intermolecular forces. The surface tension term in (25) and (29) depends on the curvature κ of
the interface Γd . In [13] it is shown, that the explicit computation of the curvature by taking
advantage of the signed-distance property of the level-set function (κ = ∆φ with ||∇φ|| = 1) is
delicate. Therefore, we reformulate the surface tension term by means of the Laplace-Beltrami
operator [38] and thereby avoid an explict computation of κ:
Z Z
h
γκw · n̂ dΓ = − γ∇id · ∇wh dΓ,
Γd Γd (31)
with ∇ f = ∇ f − (∇ f · n̂) n̂.

id is an identity mapping on the interface. For details on the Laplace-Beltrami operator we refer
the interested reader to e.g. [13].

6. Numerical examples

6.1. Sloshing tank

The first test case is taken from [13]; similar test cases are studied in [3, 11]. We consider a
rectangular tank with height 1.5 m and width 1.0 m. The two phases are initially separated by a
sinusoidal interface (cf. Fig. 8)

Γd = {(x, y) : y = 1.01 + 0.1 · sin((x − 0.5) · π), 0 ≤ x ≤ 1}. (32)

The physical parameters are ρ1 = 1000 kg/m3 , ρ2 = 1 kg/m3 , µ1 = 1 kg/m/s, µ2 = 0.01 kg/m/s
and the gravitational force fy = −g = −1.0 m/s2 is applied. No surface tension effects are
considered, that is, a weak discontinuity occurs across the interface. Slip boundary conditions
are prescribed along the walls of the tank and p = 0 N/m2 is set along the upper boundary.
Fig. 8 shows instances of the resulting interface movement using the XFEM with p-sign
enrichment and a domain discretized with 80×120 elements. The simulation spans 20 s by 25600
time steps, leading to a fine time step size of ∆t = 0.00078125. In the following comparisons,
13
 Figure 8: Tank sloshing: computational domain, interface and pressure solution.

either the interface height at the left boundary is plotted over the time or the interface position in
the left half of the tank from x = 0 to x = 0.5 at t = 20 s is shown. If not mentioned otherwise
the time-stepping scheme (24) is used.

6.1.1. Comparison of interface tracking and interface capturing

First, we take a look on how interface tracking compares with interface capturing on a fixed
mesh, discretized with 40 × 60 quadrilateral elements. For interface tracking, a classical FEM
is used (no enrichments are needed as the kinks in the fields align with the element edges). The
movement of the inner nodes is realized by a pseudo structure approach (see e.g. [20]). Here,
the mesh is treated as an ideal elastic material which deforms due to the interface motion. Fur-
thermore, the space-time formulation is used in this case. For interface capturing, we also show
results obtained with the classical FEM. The kinks can then obviously not be considered appro-
priately. XFEM results are obtained by employing either the p-abs or p-sign approach. Figure
9 depicts the temporal convergence plot of the interface height at the left boundary for all the
methods using different time step sizes: ∆t = 0.2, 0.1, 0.05, 0.025, 0.0125, 0.00625, 0.003125 s.
The interface tracking results are as expected very accurate, already for large time steps
(Fig. 9(a)). Even though the interface capturing FEM can not account for the kinks across the
interface appropriately in this case, the sloshing frequency and amplitude can be predicted quite
well for small time steps, see Fig. 9(b). The fairly fine mesh may diminish the drawback of the
FEM not being able to account for the discontinuities in this case. Nevertheless, the following
studies will show the big advantage in using the XFEM.
Figures 9(c) and 9(d) show a first trend that the p-sign enrichment for the XFEM leads to
better results than p-abs, although we are only experiencing a weak discontinuity for this test
case. Remarkably good results are also obtained for rather large time steps. The advantage of the
sign- over the abs-enrichment in this case may be explained with the larger approximation space
which is provided by the sign enrichment (cf. Section 3.1).

6.1.2. Influence of the enrichment

Next, we figure out what we get from the enrichment by comparing XFEM with p-sign to
standard FEM interface capturing. Here, also results on moving meshes are compared, where an
artificial mesh motion is prescribed on the node in the very domain center. The whole mesh is
then deformed according to an ideal elastic material. This scenario is important e.g. for fluid-
structure interaction problems as shall be seen later. Uniformly refined meshes with 10 × 15,
20 × 30, 40 × 60 and 80 × 120 elements and a time step ∆t = 0.0125 s are used. Taking a look
14
(a) FEM, interface tracking, space-time (b) FEM, interface capturing, time-stepping

(c) XFEM, p-abs, time-stepping (d) XFEM, p-sign, time-stepping

Figure 9: Temporal convergence: interface tracking, interface capturing.

15
 (a) FEM, fixed mesh (b) XFEM, p-sign, fixed mesh

(c) FEM, moving mesh (d) XFEM, p-sign, moving mesh

Figure 10: Spatial convergence: standard FEM, XFEM p-sign.

16
 (a) FEM, fixed mesh (b) XFEM, p-sign, fixed mesh

(c) FEM, moving mesh (d) XFEM, p-sign, moving mesh

Figure 11: Temporal convergence: standard FEM, XFEM p-sign.

at the spatial convergence of the interface position in the left half of the domain, Fig. 10, one
can see that the XFEM performs significantly better than the standard FEM in terms of accuracy.
Especially on the coarse meshes, the advantage of the enrichment in the XFEM can clearly be
seen. Fig. 10(b) shows almost the same interface position with XFEM for the different spatial
resolutions. In case of an additionally moving mesh (Fig. 10(c), 10(d)) the differences between
standard FEM and XFEM are even more eminent. While the XFEM results are comparable to
those on the fixed meshes, oscillations in the order of the amplitude of the sloshing are observed
in the FEM results.
We obtain similar results for the temporal convergence study, see Fig. 11. Here, a 40 × 60
element mesh is used and the same time step sizes as in Section 6.1.1. We conclude that the
enrichment has a dramatic impact on the quality of the results.

Remark In Fig. 10(c) and 11(c) the interface with FEM is very disturbed and lies outside of the
plotted range for the coarsest mesh and the largest time step respectively.

17
 (a) XFEM, p-abs, fixed mesh (b) XFEM p-sign, fixed mesh

(c) XFEM, p-abs, moving mesh (d) XFEM p-sign, moving mesh

Figure 12: Spatial convergence: XFEM p-abs, XFEM p-sign.

6.1.3. Comparison of p-abs and p-sign enrichment

We already mentioned that although only a weak discontinuity exists in this test case (which
can be represented with the p-abs enrichment), the p-sign enrichment performs better. In order to
substantiate this assumption, let us compare the spatial convergence of XFEM with the p-abs and
p-sign enrichment. The same meshes and time step as in Section 6.1.2 are used. Fig. 12 confirms
that the p-sign enrichment is the enrichment of choice, especially for XFEM on moving meshes.
Comparing Fig. 12(a) and 12(c) there are large deviations between the results with and without
mesh motion on the two coarsest meshes for the p-abs enrichment. Also in the case of a fixed
mesh, the results using p-sign are slightly better compared to p-abs. In conclusion, the p-sign
enrichment will be used in the following test cases.

6.1.4. Comparison of space-time and time-stepping

At last, we want to compare the two time integration schemes, space-time and time-stepping
for the tank sloshing test case. An 40 × 60 element mesh is used and the same time step sizes as
in Section 6.1.1.
The results in Fig. 13 confirm the findings of [32]. The overall error level using the space-
18
 (a) XFEM, p-sign, space-time, fixed mesh (b) XFEM, p-sign, time-stepping, fixed mesh

(c) XFEM, p-sign, space-time, moving mesh (d) XFEM, p-sign, time-stepping, moving mesh

Figure 13: Temporal convergence: space-time, time-stepping.

time method on fixed and moving meshes is lower compared to the results obtained with the
time-stepping scheme. However, one can hardly perceive differences in the results using the
different time step sizes, except for the case of the largest ∆t. For sufficiently small time steps,
very good results are achieved with the time-stepping scheme.
Taking into account the computational efficiency, space-time loses against time-stepping as it
requires approximately four times the computational time. Considering the computational costs
and the burden to deal with 4D space-time elements for problems in three spatial dimensions—
particularly regarding the necessity of 4D element subdivision—leads us to the conclusion to
prefer the time-stepping scheme.

6.2. Collapsing water column

We are further considering the second test case from [13], see [39, 40, 41] for similar settings.
The initial state is characterized by a water column of size 0.146 × 0.292 m at the left boundary
of the 0.584 × 0.438 m domain (cf. Fig. 14). Differing from the previous papers, the corner of the
water column is rounded in order to ease the computation of this test case with interface tracking
for comparison. The fluid properties are ρ1 = 1000 kg/m3 , ρ2 = 1 kg/m3 , µ1 = 10−3 kg/m/s,
19
 Figure 14: Collapsing water column: computational domain, interface and pressure solution.

µ2 = 10−5 kg/m/s. No surface tension is considered and a volume force fy = −g = −9.81 m/s2
is applied. Slip boundary conditions are assumed along the walls and p = 0 N/m2 is set along
the upper boundary. The computational domain is discretized with 127 × 95 elements and the
simulation spans 0.25 s using 6400 time steps.
In Fig. 14, the interface position and the pressure solution is shown at different time instances.
The results are in very good agreement with the solutions given in [13].
Figure 15 shows a comparison of our results with experimental data by Martin and Moyce
[39] and interface tracking results by Walhorn [40]. The dimensionless water column p width
δ = x⋆ /a and height δ = y⋆ /b are plotted over the dimensionless times τδ = t 2g/a and
p
τβ = t 2g/b. x⋆ (t) and y⋆ (t) are the intersections of the interface with the bottom and left
boundary. a = x⋆ (t = 0) and b = y⋆ (t = 0) are the initial water column width and height. The
slope of the curve in Fig. 15(a) (dimensionless width of the water column over time) is predicted
very well and almost coincides with the results by Walhorn. However, the water column in
the experiment expands slower in the beginning. This could be explained with the time which
is required to remove the gate which initially separates the water column from the remaining
experimental domain. Thereby, the collapse of the column is slightly delayed. The evolution of
the water column height agrees very well between the simulations and the experiment.
We carried out the same studies for this test case as for the tank sloshing case. The findings
coincide with those described in the last section. For the sake of completeness, Figure 16 shows
some results. The interface position is plotted after 0.25 s using 800 time steps and different
fixed meshes. It is obvious that standard FEM interface capturing can not accurately predict the
position of the water front tip, in particular see Fig. 16(b). On the contrary, the XFEM results
using the sign-enrichment (cf. Fig. 16(c)) are in good agreement with the accurate FEM interface
tracking results (cf. Fig. 16(a)). It is seen that the differences between standard FEM and XFEM
are not as large as in the tank sloshing case. One should note that the FEM interface tracking
approach did work only on the two finest meshes. The method failed for the two coarser meshes

20
 (a) Water column width. (b) Water column height.

Figure 15: Collapsing water column: comparison with experimental data by Martin and Moyce [39].

due to mesh tangling.

Especially in this case, also enriching the velocity field turned out to be very unfavorable
and led to unstable results. These findings encourage our choice to only enrich the pressure
approximation space using the sign enrichment (14), thus producing a robust method.

6.3. Rising Bubble

So far only test cases with weak discontinuities have been presented. The first test case from
[42] shows the ability of the proposed method to deal with strong discontinuities. The rectangular
domain is of size 1.0 × 2.0 m with an initially circular bubble with diameter d = 0.5 m, see
Fig. 17. The properties of the fluids are ρ1 = 100 kg/m3 , ρ2 = 1000 kg/m3 , µ1 = 1 kg/m/s,
µ2 = 10 kg/m/s, fy = −g = −0.98 m/s2 and surface tension coefficient γ = 24.5 kg/s2 . The
characteristic dimensionless Reynolds and Eötvös numbers follow to:
p
ρ2 gdd
Re = = 35, (33)
µ2
gρ2 d2
Eo = = 10 (34)
γ
The spatial resolution is 80 × 160 elements and ∆t = 0.002 s. No slip boundary conditions
are assumed at the top and bottom boundary, slip boundary conditions are used along the vertical
walls and p = 0 N/m2 is specified at the upper boundary. With the given parameters, the bubble
should belong to the ellipsoidal regime, see [43]. That means, the surface tension effects domi-
nate the bubble shape and no break up should occur. Fig. 17 shows the rising bubble over time.
In Fig. 18(a) the bubble shape at t = 3.0 s is compared with the results from [42], showing very
good agreement.
For a more accurate comparison Fig. 18(a) depicts the rise velocity of the bubble
R
Ω
v dx
vrise = R 1 (35)
Ω
1 dx
1

plotted over time. This quantity is also in excellent agreement with the simulation results by
Hysing et al. [42].
21
 (a) FEM, interface tracking

(b) FEM, interface capturing (c) XFEM, p-sign

Figure 16: Spatial convergence: Water column interface.

Figure 17: Rising bubble: computational domain, interface and pressure solution.

22
 (a) Bubble shape. (b) Rise velocity.

Figure 18: Rising bubble: comparison of the results with simulation data from [42].

6.4. Free-surface flows

Free-surface flows can be found in many fields such as hydraulic design of dams and weirs,
analysis of waterways or the flow around ship hulls. In all these cases it is important to be able to
accurately predict the unknown movement of the free fluid surface. Due to the large density and
viscosity differences (e.g. between water and air), generally the inertia of the gas can be ignored
and the fluid movement is almost independent with respect to the gas. Interface tracking methods
often utilize this fact by only considering the fluid phase with an explicitly meshed, moving free-
surface, see e.g. [44, 45]. Then, a kinematic condition has to ensure that the normal fluid velocity
matches the normal velocity of the mesh at the free-surface.
In this work, we treat free-surface flows as two-phase flow problems, where one fluid has
negligible density and viscosity compared to the other phase. The original free-surface flow
is then to be seen as a special case of two-phase flows and the numerical strategy from above
applies. It is noted that the resulting large ratios of density and viscosity pose a physically ill-
conditioned problem. It is recalled that in the context of the XFEM, already the enrichment of
elements may lead to a numerically-based ill-conditioning, see Section 3. The situation can be
improved by the approaches summarized in [7]. We find, however, that robust solutions with
direct solvers do not need any special treatment.

Remark Similar to the handling of holes in [46] one could think of nullifying the degrees of
freedom in the air phase instead of considering a two-phase flow problem. This approach would
have wide consequences on the enrichment approach. Hence, that approach is not further con-
sidered in this work.

In contrast to the preceding test cases, the computational domain is not completely enclosed
by a wall, but an inflow and outflow boundary exists, which is crossed by the interface. In order
to maintain a constant fluid height at the inflow, Dirichlet boundary conditions for the level-set
transport equation have to be prescribed there. Furthermore, surface tension effects are neglected
and the proposed p-sign enrichment is used for all cases. Both free-surface configurations con-
sidered herein observe the overflow of an obstruction. Similar cases have been investigated
experimentally for example by Forbes [47] and Chanson [48].

23
 Figure 19: Flow over a circular bump.

Figure 20: Circular bump: Detail of the mesh.

6.4.1. Flow over a circular bump

For the first test case, the domain Ω is of length 35.0 m and height 1.8 m. The interface
between Ω1 and Ω2 has an initial height of y0 = 1.0 m (cf. Figure 19).
The density of the fluids in Ω1 and Ω2 is ρ1 = 10.0 kg/m3 and ρ2 = 0.01 kg/m3 , respectively,
and viscosity is µ1 = 0.08 kg/m/s and µ2 = 0.008 kg/m/s. The gravitational forces are fy =
−g = −1.0 m/s2 . The values are chosen such that no breaking waves occur in order to enable
a comparison with interface tracking. No-slip conditions are applied at the lower boundary and
slip conditions at the upper boundary. At the inflow and outflow boundary the velocity profile is
prescribed by
 1
0.65 · (y/y0 ) 4 m/s, y < y0


u= , v = 0 m/s (36)

0.65 m/s, y ≥ y0

and at the upper left corner the pressure p = 0 N/m2 . At the obstacle, the Reynolds number in
Ω1 is
ρ·u·d
Re = ≈ 40 (37)
η
with d the obstacle height, u ≈ 0.7 m/s over the bump and the Froude number is
u
Fr = √ ≈ 0.8 (38)
g·L
with L ≈ 0.75 m the minimal water depth. A detail of the structured quadrilateral mesh with
4648 nodes and 4455 elements is shown in Figure 20. The time-stepping scheme described in
Section 4.1 is used with a time step ∆t = 0.02 s.
Figure 21 shows the results after 1000 time steps (t = 20 s). Behind the obstacle, waves
emerge at the surface and after approximately 800 time steps, a quasi-steady free-surface forma-
tion is obtained as it is shown in Figure 21. Furthermore, the results of the XFEM are compared
24
Figure 21: Circular bump: Pressure field and interface position in the interval x ∈ [−2, 10] - XFEM (solid), DSD
(dashed).

Figure 22: Flow over a rectangular bump.

with those computed with a stabilized space-time deforming spatial domain (SST-DSD) method
(interface tracking) and solving the elevation equation for the free-surface position as used in
[49].
It can be seen that the results of the XFEM with the p-sign enrichment and interface tracking
method are in good agreement.

6.4.2. Flow over a rectangular bump

The second free-surface test case considers a similar problem, but now the overflown obstacle
is of rectangular shape. The computational domain Ω is 39.0 m long and 1.3 m high. The initial
free-surface elevation is y0 = 0.8 m (cf. Figure 22).
The following physical parameters are given: ρ1 = 1.0 kg/m3 , ρ2 = 0.001 kg/m3 , µ1 =
0.01 kg/m/s and µ2 = 0.001 kg/m/s. The gravitational forces are fy = −g = −9.81 m/s2 . No-slip
conditions are applied at the lower boundary and slip conditions at the upper boundary. At the
inflow and outflow the following velocity profile is prescribed
 1
1.2 · (y/y0 ) 4 m/s, y < y0


u= , v = 0 m/s (39)
1.2 m/s,
 y ≥ y0

and at the upper left corner the pressure p = 0 N/m2 . The dimensionless numbers are Re ≈ 64
and Fr ≈ 0.7. A structured quadrilateral mesh with 5923 nodes and 5672 elements is used
(cf. Figure 23). Here, again the trapezoidal time-stepping scheme with a time step ∆t = 0.02 s is
applied.
Figure 24 again shows the comparison between the mesh-independent quasi-steady surface
formation with the XFEM and the SST-DSD interface tracking results after t = 20 s. It can be
seen that both methods lead to almost identical free-surface shapes.
25
 Figure 23: Rectangular bump: Detail of the mesh.

Figure 24: Rectangular bump: Pressure field and interface position in the interval x ∈ [−2, 10] - XFEM (solid), DSD
(dashed).

One main advantage of the XFEM for free-surface flow problems is the ability to deal with
topological changes without remeshing, in contrast to interface tracking techniques. In order to
illustrate this ability, the flow over the rectangular bump is carried out under slightly different
conditions. The initial free-surface height is set lower to y0 = 0.7 m and the inflow velocity is
increased:
 1
1.6 · (y/y0 ) 4 m/s, y < y0


u= , v = 0 m/s. (40)
1.6 m/s,
 y ≥ y0

This results in larger dimensionless numbers: Re ≈ 85 and Fr ≈ 0.8, that is still sub-critical flow
conditions. The other parameters, boundary conditions, and the mesh remain unchanged. Figure
25 depicts the evolution of the free-surface and the pressure field over time. For this configuration
the surface is rapidly varying and no quasi-steady state is obtained during the simulation time of
20 s. The free-surface folds, enclosing the lighter phase in the denser one and at the same time
the breaking waves move back and forth. Using the XFEM with the proposed sign-enrichment
for the pressure, we obtain a flexible method which can also cope with very unsteady free-surface
flow problems. These results can not be reproduced adequately by an interface tracking scheme
such as the SST-DSD.

6.5. Circular object falling onto a free surface

In Section 6.1.2, we also validated the XFEM with p-sign enrichment on moving meshes.
The following test case is an outlook of what kind of problems can be dealt with the proposed
method. We consider a solid circular object falling onto a free surface. In Figure 26, the setting
is depicted. Rather than considering one fluid with a free surface, the previously investigated
and validated approach of employing a two-phase flow where one fluid has negligible density
and viscosity is realized again. For the structure, a simple rigid body model is assumed. No-slip
boundary conditions are prescribed on the object and slip conditions on the surrounding walls.
A similar case can be found for example in [50].
This case from the field of fluid structure interactions combines the interface tracking and
interface capturing approaches. It is not possible to compute the immersing of the structure in
26
Figure 25: Evolution of the pressure field for the rectangular bump test case with breaking waves.

27
 Figure 26: Falling object: computational domain, interface and pressure solution.

the fluid completely with a standard interface tracking method due to an incompatibility of the
no-slip condition on the object and the free-surface [51]. In order to overcome this problem, the
falling object is modeled with a pseudo-structure ALE approach and the surrounding two-phase
problem is modeled with the XFEM. That is, interface tracking is used in order to track the fluid-
structure interface and interface capturing in order to consider for the fluid-fluid interface. The
simulation code couples the fluid solver, level-set transport and the structural computation of the
mesh deformation as depicted in Fig. 2.
The following physical parameters are used: ρ1 = 100 kg/m3 , ρ2 = 1 kg/m3 , µ1 = 0.1 kg/m/s,
µ2 = 0.001 kg/m/s, fy = −g = −1.0 m/s2 , no surface tension is considered and the solid body
has a total mass of 2 kg. The underlying mesh consists of 7600 quadrilateral elements and a time
span of 12 s with ∆t = 0.01 s is calculated.
Figure 26 shows some frames from the simulation results. The interaction between the solid
object and the fluid surface is observed until an equilibrium state is reached.

7. Conclusion

In this work, an XFEM for two-phase and free-surface flows is systematically designed with
respect to different criteria: Enrichment schemes, time integration and fixed/moving meshes.
During our numerical studies we found that it is not advisable to enrich the velocity approxi-
mation space as it does not improve the results significantly, but may lead to severe convergence
problems. Furthermore, the required number of iterations for the solution of the governing equa-
tions may increase considerably. On the other hand, the enrichment of the pressure field is
28
essential. We recommend the sign-enrichment of the pressure field even if no surface tension
effects are considered.
In agreement with earlier investigations [32], the semi-implicit time-stepping scheme is cho-
sen over the, slightly more accurate, but computationally more expensive, space-time approach.
This choice is also encouraged by the burden of dealing with 4D space-time elements in the
XFEM when three spatial dimensions are considered.
Motivated by the application of the proposed method to floating body simulations, the nu-
merical studies were also carried out on artificially moving meshes. Especially in this case, the
advantage of the proposed sign-enrichment scheme over the abs-enrichment of the pressure field
could be shown.
On the whole, the numerical results show the great flexibility of the proposed method to
accurately compute two-phase and free-surface flow problems in different configurations, in-
cluding surface tension effects, topological changes, fixed and moving meshes. The method has
recently been extended to three spatial dimensions. Then, parallelization and load balancing in
the context of the XFEM are challenging aspects. The use of iterative solvers in 3D is generally
advisable, too. Then strategies have to be developed in order to circumvent the problem of the
ill-conditioning of the system matrix (cf. Section 3). Results are to be presented in a forthcoming
publication.

8. Acknowledgement

The authors gratefully acknowledge the computing resources provided by the AICES grad-
uate school and RWTH Aachen University Center for Computing and Communication. The
authors also wish to acknowledge the support of the German Science Foundation in the frame
of the Emmy-Noether-research group “Numerical methods for discontinuities in continuum me-
chanics”.
[1] S. Ganesan, G. Matthies, L. Tobiska, On spurious velocities in incompressible flow problems with interfaces,
Comp. Methods Appl. Mech. Engrg. 196 (2007) 1193 – 1202.
[2] M. Sussman, P. Smereka, S. Osher, A level set approach for computing solutions to incompressible two-phase flow,
J. Comput. Phys. 114 (1994) 146 – 159.
[3] T. Tezduyar, M. Behr, J. Liou, A new strategy for finite element computations involving moving boundaries and
interfaces - the deforming-spatial-domain/space-time procedure: II. Computation of free-surface flows, two-liquid
flows, and flows with drifting cylinders, Comp. Methods Appl. Mech. Engrg. 94 (1992) 353 – 371.
[4] C. Hirt, B. Nichols, Volume of fluid (VOF) method for the dynamics of free boundaries, J. Comput. Phys. 39 (1981)
201 – 225.
[5] N. Moës, J. Dolbow, T. Belytschko, A finite element method for crack growth without remeshing, Internat. J.
Numer. Methods Engrg. 46 (1999) 131 – 150.
[6] T. Belytschko, T. Black, Elastic crack growth in finite elements with minimal remeshing, Internat. J. Numer. Meth-
ods Engrg. 45 (1999) 601 – 620.
[7] T. Fries, T. Belytschko, The extended/generalized finite element method: An overview of the method and its
applications, Int. J. Numer. Methods Fluids (2010) DOI: 10.1002/nme.2914.
[8] J. Chessa, T. Belytschko, An extended finite element method for two-phase fluids, ASME J. Appl. Mech. 70 (2003)
10 – 17.
[9] J. Chessa, T. Belytschko, An enriched finite element method and level sets for axisymmetric two-phase flow with
surface tension, Internat. J. Numer. Methods Engrg. 58 (2003) 2041 – 2064.
[10] S. Groß, A. Reusken, An extended pressure finite element space for two-phase incompressible flows with surface
tension, J. Comput. Phys. 224 (2007) 40 – 58.
[11] A. Kölke, Modellierung und Diskretisierung bewegter Diskontinuitäten in randgekoppelten Mehrfeldsystemen,
Dissertation, Technische Universität Braunschweig (2005).
[12] T. Fries, T. Belytschko, The intrinsic XFEM: A method for arbitrary discontinuities without additional unknowns,
Internat. J. Numer. Methods Engrg. 68 (2006) 1358 – 1385.

29
[13] T. Fries, The intrinsic XFEM for two-fluid flows, Int. J. Numer. Methods Fluids 60 (2008) 437 – 471.
[14] P. Minev, T. Chen, K. Nandakumar, A finite element technique for multifluid incompressible flow using eulerian
grids, J. Comput. Phys. 187 (2003) 255 – 273.
[15] A. H. Coppola-Owen, R. Codina, Improving eulerian two-phase flow finite element approximation with discontin-
uous gradient pressure shape functions, Int. J. Numer. Methods Fluids 49 (2005) 1287 – 1304.
[16] S. Zlotnik, P. Dı́ez, Hierarchical X-FEM for n-phase flow (n¿2), Comp. Methods Appl. Mech. Engrg. 198 (2009)
2329 – 2338.
[17] N. Moës, M. Cloirec, P. Cartraud, J. Remacle, A computational approach to handle complex microstructure ge-
ometries, Comp. Methods Appl. Mech. Engrg. 192 (2003) 3163 – 3177.
[18] G. Legrain, N. Moës, A. Huerta, Stability of incompressible formulations enriched with X-FEM, Comp. Methods
Appl. Mech. Engrg. 197 (2008) 1835 – 1849.
[19] T. Hughes, W. Liu, T. Zimmermann, Lagrangian-Eulerian finite element formulation for incompressible viscous
flows, Comp. Methods Appl. Mech. Engrg. 29 (1981) 329 – 349.
[20] T. Tezduyar, M. Behr, J. Liou, A new strategy for finite element computations involving moving boundaries and
interfaces - the deforming-spatial-domain/space-time procedure: I. The concept and the preliminary numerical
tests, Comp. Methods Appl. Mech. Engrg. 94 (1992) 339 – 351.
[21] S. Osher, R. Fedkiw, Level Set Methods and Dynamic Implicit Surfaces, Springer Verlag, Berlin, 2003.
[22] J. Sethian, Level Set Methods and Fast Marching Methods, 2nd Edition, Cambridge University Press, Cambridge,
1999.
[23] J. Chessa, H. Wang, T. Belytschko, On the construction of blending elements for local partition of unity enriched
finite elements, Internat. J. Numer. Methods Engrg. 57 (2003) 1015 – 1038.
[24] T. Fries, A corrected XFEM approximation without problems in blending elements, Internat. J. Numer. Methods
Engrg. 75 (2008) 503 – 532.
[25] T. Belytschko, N. Moës, S. Usui, C. Parimi, Arbitrary discontinuities in finite elements, Internat. J. Numer. Methods
Engrg. 50 (2001) 993 – 1013.
[26] N. Sukumar, D. Chopp, N. Moës, T. Belytschko, Modeling holes and inclusions by level sets in the extended
finite-element method, Comp. Methods Appl. Mech. Engrg. 190 (2001) 6183 – 6200.
[27] P. Laborde, J. Pommier, Y. Renard, M. Salaün, High-order extended finite element method for cracked domains,
Internat. J. Numer. Methods Engrg. 64 (2005) 354 – 381.
[28] A. Hansbo, P. Hansbo, An unfitted finite element method, based on Nitsche’s method, for elliptic interface prob-
lems, Comp. Methods Appl. Mech. Engrg. 191 (2002) 5537 – 5552.
[29] R. Gracie, H. Wang, T. Belytschko, Blending in the extended finite element method by discontinuous Galerkin and
assumed strain methods, Internat. J. Numer. Methods Engrg. 74 (2008) 1645 – 1669.
[30] A. Reusken, Analysis of an extended pressure finite element space for two-phase incompressible flows, Comput.
Vis. Sci. 11 (2008) 293 – 305, 10.1007/s00791-008-0099-8.
[31] E. Béchet, H. Minnebo, N. Moës, B. Burgardt, Improved implementation and robustness study of the X-FEM for
stress analysis around cracks, Internat. J. Numer. Methods Engrg. 64 (8) (2005) 1033 – 1056.
[32] T. Fries, A. Zilian, On time integration in the XFEM, Internat. J. Numer. Methods Engrg. 79 (2009) 69 – 93.
[33] S. Idelsohn, M. Mier-Torrecilla, N. Nigro, E. Oñate, On the analysis of heterogeneous fluids with jumps in the
viscosity using a discontinuous pressure field, Comput. Mech. 46 (2010) 115 – 124.
[34] M. Kang, R. P. Fedkiw, X.-D. Liu, A boundary condition capturing method for multiphase incompressible flow, J.
Sci. Comput. 15 (2000) 323 – 360.
[35] F. Shakib, T. Hughes, Z. Johan, A new finite element formulation for computational fluid dynamics: X. The com-
pressible Euler and Navier-Stokes equations, Comp. Methods Appl. Mech. Engrg. 89 (1991) 141 – 219.
[36] S. Mittal, On the performance of high aspect ratio elements for incompressible flows, Comp. Methods Appl. Mech.
Engrg. 188 (2000) 269 – 287.
[37] J. Donea, A. Huerta, Finite Element Methods for Flow Problems, John Wiley & Sons, Chichester, 2003.
[38] S. Hysing, A new implicit surface tension implementation for interfacial flows, Int. J. Numer. Methods Fluids 51
(2005) 659 – 672.
[39] J. Martin, W. Moyce, An experimental study of the collapse of liquid columns on a rigid horizontal plane, Philo-
sophical Transactions of the Royal Society of London 244 (1952) 312 – 324.
[40] E. Walhorn, Ein simultanes Berechnungsverfahren für Fluid-Struktur-Wechselwirkungen mit finiten Raum-Zeit-
Elementen, Dissertation, Technische Universität Braunschweig (2002).
[41] M. Cruchaga, D. Celentano, T. Tezduyar, Collapse of a liquid column: Numerical simulation and experimental
validation, Comput. Mech. 39 (2007) 453 – 476.
[42] S. Hysing, S. Turek, D. Kuzmin, N. Parolini, E. Burman, S. Ganesan, L. Tobiska, Quantitative benchmark compu-
tations of two-dimensional bubble dynamics, Int. J. Numer. Meth. Fluids 60 (11) (2009) 1259 – 1288.
[43] R. Clift, J. Grace, M. Weber, Bubbles, drops and particles, Academic Press, New York, NY, 1978.
[44] M. Behr, F. Abraham, Free-surface flow simulations in the presence of inclined walls, Comp. Methods Appl. Mech.
30
 Engrg. 191 (2002) 5467 – 5483.
[45] R. J. Labeur, G. N. Wells, Interface stabilised finite element method for moving domains and free surface flows,
Comp. Methods Appl. Mech. Engrg. 198 (2009) 615 – 630.
[46] C. Daux, N. Moës, J. Dolbow, N. Sukumar, T. Belytschko, Arbitrary branched and intersecting cracks with the
extended finite element method, Internat. J. Numer. Methods Engrg. 48 (2000) 1741 – 1760.
[47] L. K. Forbes, Critical free-surface flow over a semi-circular obstruction, J. Engrg. Math. 22 (1) (1988) 3 – 13.
[48] H. Chanson, Free-surface flows with near-critical flow conditions, Canadian J. Civil Engrg. 23 (6) (1996) 1272 –
1284.
[49] I. Güler, M. Behr, T. Tezduyar, Parallel finite element computation of free-surface flows, Comput. Mech. 23 (1999)
117 – 123.
[50] J. Yang, F. Stern, Sharp interface immersed-boundary/level-set method for wave-body interactions, J. Comput.
Phys. 228 (17) (2009) 6590 – 6616.
[51] R. W. Yeung, P. Ananthakrishnan, Viscosity and surface-tension effects on wave generation by translating bodies,
J. Engrg. Math. 32 (1997) 257 – 280.

31

View publication stats