A Lagrangian Meshless Finite Element Method

applied to Fluid-Structure Interaction Problems

S.R. Idelsohn
(1,2)
, E. Oate
(2)
, F. Del Pin
(1)


. International Center for Computational Methods in Engineering (CIMEC)
Universidad Nacional del Litoral and CONICET, Santa Fe, Argentina
email: sergio@ceride.gov.ar

International Center for Numerical Methods in Engineering (CIMNE)
Universidad Politcnica de Catalua, Barcelona, Spain
email: onate@cimne.upc.es

Key words: Fluid-Structure interaction, Particle methods, Lagrange formulations,
Incompressible Fluid Flows, Meshless Methods, Finite Element Method.

Abstract
A method is presented for the solution of the incompressible fluid flow equations using a
lagrangian formulation. The interpolation functions are those used in the Meshless Finite
Element Method (MFEM) and the time integration is introduced in a semi-implicit way by
a fractional step method. Classical stabilization terms used in the momentum equations are
unnecessary due to the lack of convective terms in the lagrangian formulation.
Furthermore, the lagrangian formulation simplifies the connections with fixed or moving
solid structures, thus providing a very easy way to solve fluid-structure interaction
problems.


1 Introduction

Over the last twenty years, computer simulation of incompressible fluid flow has been
based on the Eulerian formulation of the fluid mechanics equations. However, it is still
difficult to analyze problems in which the shape of the interface changes continuously or in
fluid-structure interactions with free-surfaces where complicated contact problems are
involved.
More recently, Particle Methods in which each fluid particle is followed in a lagrangian
manner have been used [1-4]. The first ideas on this approach were proposed by Monaghan
[1] for the treatment of astrophysical hydrodynamic problems with the so called Smooth
Particle Hydrodynamics Method (SPH). This method was later generalized to fluid
mechanic problems [2-4]. Kernel approximations are used in the SPH method to interpolate
the unknowns.
On the other hand, a family of methods called Meshless Methods have been developed both
for structural [5,6] and fluid mechanics problems [8-10]. All these methods use the idea of
a polynomial interpolant that fits a number of points minimizing the distance between the
interpolated function and the value of the unknown point. These ideas were proposed first
by Nayroles et al. [7], they were later used in structural mechanics by Belytschko et al. [5]
and in fluid mechanics problems by Oate et al. [8-10]. In a previous paper [11] the authors
presented the numerical solution for the fluid mechanics equations using a lagrangian
formulation and a meshless method called the Finite Point Method. Lately, the meshless
ideas were generalized to take into account the finite element type approximations in order
to obtain the same computing time in mesh generation as in the evaluation of the meshless
connectivities [12,13]. This method was called the Meshless Finite Element Method
(MFEM) and uses the Extended Delaunay Tessellation [14] to build the mesh in a
computing time which is linear with the number of nodal points.
In this paper new ideas and results for the solution of a particle method in the field of Fluid-
Structure Interaction (FSI) using the Meshless Finite Element Method are presented. A
more general formulation is used in which all the classical advantages of the FEM for the
evaluation of the unknown functions and derivatives are preserved.
Different strategies have been proposed to solve FSI problems. The selection of the most
effective approach depends largely on the nature of the problem to be analyzed [15].
Depending on the degree of coupling between the equations for the fluid and the structure,
two cases can be distinguished. The first one occurs when there is a strong coupling
between the fluid flow and the elastic deformation of the structure [15-17]. The second case
occurs when there is a weak interaction between the fluid and the rigid deformation of the
structure. In the latter, the solid must undergo large rigid displacements interacting with the
fluid. This is the case for instance of sea-keeping in ship hydrodynamics, rotating turbines,
mills, and other engines with a moving solid inside a fluid. Both cases of FSI are more
easily studied with a lagrangian formulation of the fluid equations, which can be seen as a
solid with a small shear coefficient or vice versa.
The lagrangian fluid flow equations for the Navier-Stoke problem will be revised in the
next section, the Meshless Finite Element Method (MFEM) will be summarized in the
Appendix and both techniques will be used to solve some FSI problems for rigid solids.

2 Governing equations
The mass and momentum conservation equations can be written in a lagrangian
formulations as:

Mass conservation:

0 =

+
i
x
i
u
Dt
D

(1)

Momentum conservation:

i
f
ij
j
x
p
i
x Dt
i
Du
+

= (2)
where is the density
i
u are the Cartesian components of the velocity field, p the
pressure,
ij
the deviator stress tensor,
i
f the source term (normally the gravity) and
Dt
D

represents the total or material time derivative of a function .
For Newtonian fluids the stress tensor
ij
may be expressed as a function of the velocity
field through the viscosity u by
)
3
2
(
ij
l
l
i
j
j
i
ij
x
u
x
u
x
u
u

=
(3)
For near incompressible flows (
l
k
i
i
x
u
x
u

<<

) the term :
0
3
2

i
i
x
u u
(4)

and it may be neglected in eq.(3). Then:

) (
i
j
j
i
ij
x
u
x
u

u (5)

In the same way, the term
ij
j
x

in the momentum equations may be simplified for slow

incompressible flows as:

) ( ) ( ) (
) ( ) ( )) ( (
j
i
j j
j
i j
i
j
i
j
j j
i
j i
j
j
i
j
ij
j
x
u
x x
u
x x
u
x
x
u
x x
u
x x
u
x
u
x x

u u u
u u u
(6)
Then, the momentum equations can be finally written as:

i
j
i
j i
i ij
j i
i
f
x
u
x
p
x
f
x
p
x Dt
Du
u +

= ) ( (7)

Boundary conditions

On the boundaries, the standard boundary conditions for the Navier-Stokes equations are:

ni i j ij
p = on

(8)
n i i
u u = on
n
(9)
t i i
u u = on
t
(10)
where
i
and
i
are the components of the normal and tangent vector to the boundary.


3 The time splitting

The time integration of equations (7) and (8) presents some difficulties when the fluid is
incompressible or nearly incompressible. In this case, explicit time steps cannot be used.
Even when using an implicit time integration scheme, incompressibility introduces some
wiggles in the pressure solution which must be stabilized. To overcome these difficulties, a
fractional step method has been proposed [18] which consists in splitting each time step in
2 steps as follows:

Split of the momentum equations

+
+ +
+

+
=

n
i
j
ij
i
n
i i i
n
i
n
i
n
i i
f
x
p
x t
u u u u
t
u u
Dt
Du
)
1 1
(
* * 1 1
(11)
where
n n
t t t =
+1
is the time step; ) , (
n n
i
n
i
t x u u = ; ) , (
1 1 1 + + +
=
n n
i
n
i
t x u u and
*
i
u are
fictitious variables defined by the split

A) ij
n
j
n
i
i
n
i i
x
t
p
x
t
t f u u

+ =
*

(12)
C) ) (
1 * 1 n n
i
i
n
i
p p
x
t
u u

=
+ +

(13)


where is parameter equal to cero or one defining a first or second order split,
respepctively [18].

Split of the mass conservation equations

i
i i
n
i
n n n n
x
u u u
t t Dt
D

+
=

+
=

+ + +
) (
* * 1 * * 1 1

(14)

where
*
is a fictitious variable defined by the split

i
i
n
i
n
i
i
n
x
u u
t
x
u
t

+ +
) (
* 1 * 1
* *

(15)


(16)





Coupled equations

From eqs.(13) and (16) the coupled mass-momentum equation becomes:

B) ) -
n n
i
n
p p
x t


1
2
2
2
1
(
*
+
+

(17)

Taking into account eq.(15) above expression can be written as:

B) ) -
n n
i i
i
n n
p p
x x
u
t t


1
2
2 *
2
1
(
+
+

(18)

4 Incompressibility conditions

The simplest way to introduce the incompressibility condition is to write:

= = =
+ 0 1 n n
(19)

Then, the first term of eq. (18) disappears. Nevertheless, in a lagrangian formulation it is
better to evaluate this term in order to avoid possible numerical errors at each time step.
The incompressibility condition is introduced by imposing that at time step
1 + n
t the density
must be equal to the initial one, i.e.

= =
+ 0 1 n

(20)

Due to numerical errors the density
n
is not necessarily equal to
0
and it must be
updated at each time step. A different way to evaluate
n
will be explained in Section 7.
Equation (18) is finally written as:

B) ) -
n n
i i
i
n
p p
x x
u
t t


1
2
2 *
2
0
(
+

(21)

Then, the total time step may be described as follows: starting with the known value
n
u and
n
p from the previous time increment, the computation of the new velocities and the
pressure involves the following five steps:
I) Evaluate the
*
u velocity from (12).
II) Evaluate the new density
n
(see Section 7)
III) Evaluate the pressure
1 + n
p solving the laplacian equation (21).
IV) Evaluate the velocity
1 + n
u using (13)
V) Move the particles to the
1 + n
x position.

5 Spatial discretization

The lagrangian split scheme described in the previous section has two important
advantages:
1) Step I is linear and explicit. The use of a lagrangian formulation eliminates the
standard convection terms present in eulerian formulations. The convection terms are
responsible for non-linearity, non symmetry and non self-adjoint operators which require
the introduction of high order stabilization terms to avoid numerical oscillations. All these
problems are not present in this formulation.
2) In all the five steps described in previous section, the only implicit step is the solution
of the laplacian of pressure (step III). This is a scalar, symmetric and positive definite
equation. Then, it is very easy to solve it using an iterative scheme (such as the conjugate
gradient method).
The big disadvantage of the lagrangian formulation is the permanent updating of the node
positions. That is the reason why standard Finite Element Methods are not useful, as the
process of updating conforming non-structured finite element meshes is expensive
The key of the lagrangian formulation is the efficiency in the mesh updating process. In a
previous paper [11] the authors evaluated the use of a meshless method for this purpose. In
[11] a meshless method based in point collocation was used. This introduces some
difficulty in prescribing the boundary conditions.
Other meshless methods as the Element Free Galerkin Method (EFGM) [5] or the Natural
Element Method (NEM) [19] have difficulties to solve arbitrary point distributions in a 3D
domain due to the complicated shape functions used.
In this paper, the Meshless Finite Element Method (MFEM) proposed in reference [12] will
be used . The method is summarized in the Appendix.
The big advantage of the MFEM compared with the FEM is the possibility of generating
meshes in a computing time of order n , being n the total number of nodes[14]. Compared
with EFGM or NEM, the advantages are the simplicity of the shape functions, which are
coincident with the FEM shape functions in most parts of the domain.
Using the MFEM , the unknown functions are approximated using an equal order
interpolation for all variables as (in matrix form)


U
N
N
N
U N u
T
T
T
T
i i

= =

(22)


P N P N p
T T
p
= =
(23)



T T
N N = =
(24)

where
T
N are the MFEM shape functions and P U, , the nodal values of the three
components of the unknown velocity, the pressure and the density respectively.
Using the Galerkin weighted residual method to solve equations (12), (21) and (13) with
boundary conditions (8-10) the following integral equations can be written:

A) 0 ) ( ( } ) {(
*
=

d p N dV
x
p
x
f
t
u u N
i
n
j
n
ij
n
ni i
j
n
ij n
i
i
n
i i
V
i

(25)

B) 0 ) ( ) - (
1 1 1
2
2
*
2
0
=

+
)
`

+ + +
d u u N
t
dV p p
x
u
x t t
N
u
i
n
i i
n
i p
V
n n
i
i
i
n
p

(26)

C) 0 ) ( } ( ) {(
1 1 * 1


+ + +
=

d p p N dV p p
x t
u u N
i
n n
i
n n
i
i
n
i
V
i


(27)

where the boundary conditions have been also split.
Integrating by parts some of the terms , the above equations become:

A)
0 ) (
*
=

d N dV
x
u
x
N
p
x
N dV
t
t f u u N
n
ni i
i
n
i
i
i
V V
n
i
i i
n
i i
V
i

(28)

B) 0 )
) (
( ) (
1
1
1
* 0
2
=

+
+
d u N
t
dV
x
p p
u
t x
N
dV N
t
u
n
n p
V i
n n
i
i
p
V
n
p



(29)

C) 0 ) ( )} ( ) {(
1 1 * 1


+ + +
=

d p p N dV p p
x t
u u N
n n
i
n n
i
i
n
i
V
i


(30)

It must be noted than the essential and natural boundary conditions of equations (29) are:

= on p 0 (31)

u
n
on u =
+
0
1

(32)

Discrete equations

Using the approximations (22), (23) and (24) the discrete equations become:

A)

+ =

d N
t
U dV
x
N
x
N t
P dV
x
N
N
t
dV f N t U dV N N U dV N N
n
n i
n
i
i
T
i
i
i
V
n
i
T
p
i
V
i
V
i
n
i
V
T
i i i
V
T
i i
*



(33)




In compact form:

A)
n n T n
u u
KU
t
P B
t
F t U M U M

+ =
*
(34)

In the same way:
B)

+
V
n n
i
T
p
i
p
V
n
n p
T
i
i
p
V
n T
p
V
T
p p
P P dV
x
N
x
N
d u N
t
U dV N
x
N
t
dV N N dV N N
t
u
) ( ) ( ) (
) (
1
1 1 *
0
2

(35)

In compact form:

B)
1 *
2
0
) (
+
= +

n n
n
p
P S P S U
t
U B
t t
M

)
(36)

and:

C)
) ( ) (
1 1 *
1
n n T
p i
n n
i
T
p
V
i
T
i
V
i
n T
i
V
i
P P d N N P P dV
x
N
N
t
U dV N N
U dV N N

+ +
+
(37)
In compact form (noting that

= on p 0 ):

C) ) (
1 * 1 n n T
u
n
u
P P B
t
U M U M

=
+ +
(38)

Where the matrices are:

=
p
p
p
M
M
M
M
0 0
0 0
0 0
(39)

=
V
T
p
dV NN M (40)

=

V V
T
V
T T
dV N
z
N
dV N
y
N
dV N
x
N
B ) ( ; ) ( ; ) ( (41)
dV
z
N
z
N
y
N
y
N
x
N
x
N
S
T T
V
T
) (

(42)

=

+
d u N U
u
n
n
1
)
(43)

=
S
S
S
K
0 0
0 0
0 0
(44)

=


V
nz
T
V
ny
T
V
nx
T
V
z
T
V
y
T
V
x
T T
dV N dV N dV N
dV f N dV f N dV f N F

; ;
1
; ;
(45)

6. Stabilization of the incompressibility condition

In the Eulerian form of the momentum equations, the discrete form must be stabilized in
order to avoid numerical wiggles in the velocity and pressure results. This is not the case in
the lagrangian formulation where no stabilization parameter must be added in equations
(34) and (38). Nevertheless, the incompressibility condition must be stabilized in equal-
order approximations to avoid possible pressure oscillations.
Then, equation (36) must be stabilized if smooth pressure results are important. It must be
noted than pressure oscillations do not influence significantly in the velocity results.
Nevertheless, in most physical problems, pressure is the main result to be obtained. That is
why stabilization of equation (36) must be performed.
The so-called finite calculus (FIC) formulation [20-22] will be choused here as the
stabilization procedure. This formulation is based in the modification of the governing
differential equations of the problem by accepting that the domain where the balance laws
are established (balance of momentum and balance of mass) has a finite size. The modified
equations in the FIC formulation for incompressible fluids are.

Momentum

0
2
=

k
i k
i
x
r h
r
(46)

Mass conservation

0
2
=

k
k
x
r h
r
(47)
where from eq.(1) and (2) the residuals are defined by:

i
j
ij
i
i
i
f
x x
p
Dt
Du
r

+ =
(48)

i
i
x
u
Dt
D
r

+ =

(49)

with
d
n k i , 1 , = where
d
n are the space dimensions of the problem.
Eqs. (46,47) are completed with the boundary and initial conditions. Note that for
consistency, the Neumann boundary condition on

must also be adequately modified by

adding a residual term. The details can be found in [21].
The underlined terms in Eqs. (46,47) introduce the necessary stabilization in the numerical
solution using whatever discretization method. Examples of the application of the FIC
approach the convection-diffusion problems and incompressible problems in solids and
fluid mechanics are presented in [21-22].

Distances
i
h in Eqs.(46,47) are characteristic length parameters and their values control
the relevance of the stabilization terms. The computation of the characteristic lengths is a
critical issue in the stabilization process [20].
The new terms in the momentum and mass conservation equations stabilize the numerical
solution in presence of high values of the convective terms and incompressibility zones,
respectively. Obviously, in lagrangian flows, as in incompressible solid mechanics
problems, the relevant stabilization term is that of Eq.(47), as the convective terms are zero
in the momentum equations.
For the practical application of the FIC formulation the stabilization term in the mass
balance equation is expressed as a function of the residual of the momentum equations
using Eq.(46) as:

d
n
i i
i
i
k
k
x
r
x
r h
1
2

(50)

where
i
are intrinsic time parameters given by:

u

8
3
2
i
i
h
=
(51)

The modified incompressibility equation is therefore written for the numerical
computations as:

=
=

d
n
i i
i
i
x
r
r
1
0
(52)
The stabilization terms in the momentum equation (46) are dropped hereonwards for the
numerical solution.
It is convenient to rewrite the residual
i
r in Eq.(48) as

i
i
i
x
p
r +

=
(53)

where
i
are pressure gradient projection terms. These terms are considered as additional
nodal variables. The necessary additional equations to match the increase in the number of
unknowns are obtained by expressing that the residual
i
r as defined by Eq.(48), vanishes, in
the average sense, over each element. This can be expressed in weighted integral form as

= +

V
i
i
i
dV
x
p
w 0 ) (
(54)

where
i
w are appropriate weighting functions.
Discretization of the
i
terms using the same MFEM interpolation functions gives:
=
T
i i
N
(55)

where represents the local value of the three components of the pressure gradient.
Eq.(54) leads to an equation system of the form (for
i i
N w = )

0 = + P B M
T

(56)

Eq.(21) is now modified with the new stabilization term as:

i 1
1
2
2 *
2
0
x
) - (

=
+ i
n
i
i n n
i i
i
n
r
t
p p
x x
u
t t
d

(57)

and Eq.(26) becomes noww
+
)
`

=
+
dV
r
t
p p
x
u
x t t
N
V
i
n
i
i n n
i
i
i
n
p
d
i 1
1
2
2
*
2
0
x
) - (

boundary terms
(58)

Integrating by parts, the equivalent to eq.(29) is:

{ } 0 . . ) (
) (
) (
1
1
1
1
1
*
0
2
= + +

+
+
=
+

t b dV
x
p
t x
p p
u
t x
N
dV N
t
n
i
i
n n
i
i
V i
n n
i
i
p
V
n
p
d

(59)

Introducing the discretization of the different fields, and using a compact notation gives:

1 *
2
0
) (
) (
+
+ = +

n n n
n
p
P S S B P S U
t
U B
t t
M

)

(60)

where the new stabilization matrices

B and

S are defined by:

=

V V
z T
V
y
T x T
t
dV N
z
N
t
dV N
y
N
t
dV N
x
N
B

) ( ; ) ( ; ) (
(61)

dV
t z
N
z
N
t y
N
y
N
t x
N
x
N
S
z
T
y
T
x
V
T
) (

(62)

Note that the effect of the stabilization terms is the addition of a new Laplacian matrix

S and a new term in the r.h.s. of Eq.(60) depending on the pressure gradient projection
variables
i
.
The pressure gradient projection may be evaluated explicitly using eq.(56) by:

D)
1 1 1 + +
=
n T n
P B M
(63)

The three steps A),B),C) described before are now completed with a fourth step D) where
the lumped diagonal form of matrix M may be used.

7. Mass conservation

In a lagrangian formulation a new mesh is generated at each time step, and all the
information is transmitted with the nodes or particles. In that way, a local variation in the
volume associated with the particles is used as the correct volume in the next time step. A
permanent update of the initial volume is necessary to avoid large error accumulation.
Thus, the correct evaluation of the first term of equation (36) becomes important in a
lagrangian formulation and will be discussed below.
The term

) (
0 n
p
M (64)

may be evaluated in two different ways:

I) Evaluation via a density update

From the mass conservation equation, the density at time
n
t may be computed as:

i
n
i n n
x
u
t

=


1

(65)

Making use of the spatial discretization (22) and (24) and the Galerkin residual method
gives:

n
i
T
i
V
p
n T
p
V
p
n T
p
V
p
U dV
x
N
N t dV N N dV N N

=



1

(66)

Integrating by parts the last term:

+ =
u
d u N t U dV N
x
N
t dV N N dV N N
n
n p
n T
i
i
T
p
V
n T
p
V
p
n T
p
V
p

1

(67)

or in compact notation:

U t U B t M M
n n
p
n
p

1
+ =


(68)

In order to take into account that the shape functions N are different at each mesh update
the following notation will be used: the shape functions or the matrices evaluated at the
time
n
t will be noted by
n
p
N and
n
p
M . Then equation (68) becomes:

n n n n n n
p
n n
p
U t U B t M M

1
+ =


(69)

where t
n
represents the time increment al time
n
t
Then:


n n n n n n
p
n n n
U t U B M t + = + =
1 1 1

) (
(70)

where the density variation has been defined by:

n n n n
p
n n
U t U B M t

) (
1
=


(71)

representing the variation at time
n
t .
Successive application of eq.(70) for all time steps gives:

{ }
l l
n
l
l l l
p
l
n
l
l n
U t U B M t

) (
1
1 0
1
0
+ = + =

=

=

(72)


The term ) (
0 n
p
M of the r.h.s. of Eq.(36) can be written as:

=
=
n
l
l
p
n
p
M M
1
0
) (
(73)

This means that at each time step
l
t the vector:

{ }
l l l l l
p
l l
U t U B M t

) (
1
=


(74)

must be evaluated, added to the previous one and stored for the next time step.

II) Evaluation via the initial associated volume

Mass conservation implies:


= =
=
) ( ) 0 (
0
n t t
V
n
V
dV dV
(75)

Using the shape functions at the corresponding time step:

n
V
T n
V
T
n t t
dV N dV N


= =
=
) ( ) 0 (
) ( ) (
0 0

(76)

Defining the volume associated to each particle by:

=
=
) (
) ( ) (
n t
V
T n T n
dV N

(77)

eq.(76) becomes:

n T n T
) ( ) (
0 0
=
(78)

which has the meaning of the total mass conservation. Vector
n
may be considered as the
vector containing the volumes associated to each particle. It may be calculated using (77) or
using the Voronoi diagram of the node distribution.
The concept of local mass conservation may be used next. This means that each particle
(node) conserves his own local mass, i.e.:

n
i
n
i i i
=
0 0

(79)

The term ) (
0 n
p
M may be written as:

) ( ) (
0 0 0 0 0 0
= =
n n n n

(80)

where
0
and
n
represent a diagonal matrix with the volume associated to each particle
at time
0
t t = and
n
t t = respectively.
These matrices may be evaluated using the lumped matrices
0

M and
n
M

or directly using
the associated volume to each particle obtained from a Voronoi diagram.

8. Boundary Surfaces

One of the main problems in mesh generation is the correct definition of the boundary
domain. Sometimes, boundary nodes are explicitly defined as special nodes, which are
different from internal nodes. In other cases, the total set of nodes is the only information
available and the algorithm must recognize the boundary nodes. Such is the case in the
lagrangian formulation in which, at each time step, a new node distribution is obtained and
the boundary-surface must be recognized from the node positions.
The use of the MFEM with the Extended Delaunay partition makes it easier to recognize
boundary nodes.
Considering that the node follows a variable ) (x h distribution, where ) (x h is the minimum
distance between two nodes, the following criterion has been used:
All nodes on an empty sphere with a radius ) (x r bigger than ) (x h , are considered as
boundary nodes.
Thus, is a parameter close to, but greater than one. Note that this criterion is coincident
with the Alpha Shape concept [12].
Once a decision has been made concerning which of the nodes are on the boundaries, the
boundary surface must be defined. It is well known that in 3-D problems the surface fitting
a number of nodes is not unique. For instance, four boundary nodes on the same sphere
may define two different boundary surfaces, a concave one and convex one.
In this work, the boundary surface is defined with all the polyhedral surfaces having all
their nodes on the boundary and belonging to just one polyhedron. See Reference [12].
The correct boundary surface may be important to define the correct normal external to the
surface. Furthermore; in weak forms (Galerkin) a correct evaluation of the volume domain
is also important. Nevertheless, it must be noted that in the criterion proposed above, the
error in the boundary surface definition is of order h . This is the standard error of the
boundary surface definition in a meshless method for a given node distribution.

9. Application to Fluid-Structure Interactions

The fluid described above will interact with structures that are in contact with it. Three
different cases of structures will be analyzed. In all three cases, the elastic strains will be
neglected and only rigid solid motions will be considered.



Fixed structures

The first type of examples presented is structures in which there is a fixed wall, for
instance, the recipient in which the fluid is contained. See Figs. 1 and 2.
This kind of structures will be analyzed by adding fixed particles at the boundaring with
velocity 0 =
i
u . These particles will be included in the computation of equations A) and B)
as standard nodes, but during the equation C) the velocity will be fixed to zero.
The inclusion of fixed boundary particles is very important to avoid contact problems.
These fixed particles automatically force the fluid to remain inside a recipient. The moving
particles cannot go across the wall due to the incompressibility condition and not to any
other restriction of velocity or displacement. This condition solves the contact problems
with complicated curved structures. See for instance example 2.

Moving structures with a known velocity

The second type of fluid-structure interaction is between the fluid and a moving wall of
known velocity as a function of the time. This is the case of moving recipients, moving
mills, or moving ships with prescribed velocity.
In this case, moving particles with known velocity are introduced in the domain boundaries.
Note that the term:

U
t
)

(81)

with

=

+
d u N U
u
n
n
1
)

(82)

must be added in equation C) where
1 + n
n
u are the known velocity on the boundaries. See
for instance Figs. 4 and 5.

Moving structures

Finally, the case of moving rigid structures is considered. For instance, the case of a
floating ship originating to water waves (see keeping). In this case, the solid will be
considered as a domain with a high viscosity parameter, much higher than the fluid domain.
For practical problems a value of u
4
10 is enough to represent a solid without introducing
numerical problems (see Figures 5 and 6).

10 Numerical Test

10.1 Water column collapse

This problem was solved by Koshizu and Oka [4] both experimentally and numerically. It
became a classical example to test the validation of the lagrangian formulation in fluid
flows. The water is initially located on the left supported by a removable board. The
collapse starts at time t = 0, when the removable board is slid-up. Viscosity and surface
tension are neglected.
Figure 1 shows the point positions at different time steps. The dark points represent the
free-surface detected with the alpha-shape algorithm with an alpha parameter 1 . 1 = . The
internal points are gray and the fixed points are black.
The water is running on the bottom wall until, near 0.3 sec, it impinges on the right vertical
wall. Breaking waves appear at 0.6 sec. Around 1 sec. the water reaches the left wall.
Agreement with the experimental results of ref. [4] both in the shape of the free surface as
well as in the time development are excellent.

10.2 Fixed ship under external waves

This example is a very schematic representation of a ship when it is hit by an external wave
(Fig. 2). The ship cannot move and initially the free surface is horizontal with a rectangle
on the right wall to produce a big wave. Fixed nodes represent the ship as well as the wall
recipient.
The example was created in order to test the suitability of the method to solve contact
problems with curved walls correctly. It is interesting to see the crash of the waves under
the ship prow and the rebound of the wave on 3.15 seconds. It is also interesting to see the
different contact walls with the internal and external ship surfaces and the moving free
surface at the bottom of the ship.

10.3 Moving ship with known velocity

In this case (Fig. 3), the same ship of the previous example is now moving at a fixed
velocity. All the nodes representing the ship have an imposed velocity. The free surface,
which was initially horizontal, takes a correct position at the bottom of the ship, and again,
the correct contact problem is realistically solved in the curved prow.

10.4 Rotating water mill

A schematic representation of a water mill is presented in Fig. 4.The blades of the mill have
an imposed rotating velocity, while the water is initially in a stationary and flat position.
Fluid structure interactions with free-surfaces and fragmentation are well reproduced in this
example.

10.5 Solid falling into a recipient with water.

In this example the fluid is interacting with a solid that is totally free, without any imposed
velocity. Figure 5 represents a free cube falling down into a recipient full of water. The
solid cube was modeled by introducing a high viscosity parameter in the element in the
following way: all the polyhedral elements formed by nodes contained in the solid have a
high viscosity value. The other elements are inviscid.
The example represents correctly the contact problem when the cube hits the water and also
the different speed during the falling process.

10.6 Solid floating on a free surface

The last example of Figure 6 represents a very interesting problem of fluid structure
interaction when there is a weak interaction between the fluid and a large rigid deformation
of the structure. In this case, there is also a free-surface problem, representing a schematic
case of see-keeping in ship hydrodynamics.
The example shows an initially stationary recipient with a floating piece of wood in which
a wave is produced on the left side. The wave intercepts the wood piece producing a
breaking wave and moving the floating wood.
All the previous examples are only schematic representations of real problems. Only the
first example has an experimental reference. The rest are presented here in order toe
evaluate the suitability of the method to solve problems other methods have difficulties to
solve.

11 Conclusions

Lagrangian formulation and the Meshless Finite Element Method are an excellent
combination to solve fluid mechanic problems, especially fluid-structure interactions with
moving free-surface and contact problems.
Breaking waves, collapse problems, and contact problems can be solved easily without any
additional constraint.
Furthermore, the Meshless Finite Element Method presented, as opposed to other methods,
has the advantages of a good meshless method concerning the easy introduction of the
nodes connectivity in a bounded time of order n . The method proposed also shares some
advantages with the FEM such as: a) the simplicity of the shape functions, b)
0
C continuity
between elements, c) an easy introduction of the boundary conditions, and d) symmetric
matrices.
The Finite Calculus (FIC) formulation can be successfully used in a lagrangian formulation
in order to eliminate spurious pressure oscillations.
Both the lagrangian formulation and the MFEM are the key ingredients to solve fluid-
structure interaction problems including with free-surface, breaking waves and collapse
situations.












Figure 1 Water column collapse at different time steps.










Figure 1.(Continuation):Water column collapse at different time steps.








Figure 2. Fixed ship under external waves














Figure 3. Moving ship with known velocity.




















Figure 4. Rotating water mill.












Figure 5. Solid cube falling into a recipient with water.


















Figure 6. Solid floating on a free surface.




Appendix

All the shape functions
i
N described in this paper are based on the Meshless Finite
Element Method (MFEM). A full description of the MFEM may be found in Ref. [12].
Nevertheless and for the sake of completeness a summary is presented in this Appendix.
The MFEM combines a particular finite element subdivision in polyhedral shape called
the Extended Delanay Tessellation and ad-hoc shape functions for this kind of polyhedra.

The Extended Delaunay Tessellation (EDT)

Let a set of distinct nodes be: N = {n
1,
n
2
, n
3
,,n
n
} in R
3
.

a) The Vorono diagram of the set N is a partition of R
3
into regions V
i
(closed and
convex, or unbounded), where each region V
i
is associated with a node n
i
, such
that any point in V
i
is closer to n
i
(nearest neighbor) than to any other node n
i
. See
Fig.7 for a 2-D representation. There is a single Vorono diagram for each set N.
b) A Vorono sphere within the set N is any sphere, defined by 4 or more nodes, that
contains no other node inside. Such spheres are also known as empty
circumspheres.
c) A Delaunay tessellation within the set N is a partition of the convex hull of all
the nodes into regions
i
such that = U
i ,
where

each
i
is the tetrahedron
defined by 4 nodes of the same Vorono sphere. Delaunay tessellations of a set N
are not unique, but each tessellation is the dual of the single Vorono diagram of
the set.

Vorono Circle
Delaunay Triangulation
Vorono Diagram

Figure 7: Vorono diagram, Vorono circle and Delaunay triangulation
for a 4 nodes distribution in 2D.
The computing time required for evaluation of all these 3 entities is of order n

, with
1.333. Using a very simple bin organization, the computation time may be reduced to a
simple order n.
As stated above, the Delaunay tessellation of a set of nodes is non-unique. For the same
node distribution, different triangulations (actually tetrahedrations, as it refers to 3-D) are
possible. Therefore, an interpolation based on the Delaunay tessellation is sensitive to
geometric perturbations of the position of the nodes. On the other hand, its dual, the
Vorono diagram, is unique. Thus, it makes more sense to define meshless shape functions
based on the unique Vorono diagram than on Delaunay tessellations. Furthermore, in 3-D
problems the Delaunay tessellation may generate several tetrahedra of zero or almost zero
volume, which introduces large inaccuracies into the shape function derivatives. The time
to obtain a good mesh via a Delaunay tessellation becomes then an unbounded iterative
operation.
These drawbacks appear in the so-called degenerated case, which is the case where more
than 4 nodes (or more than 3 nodes in a 2-D problem) are on the same empty sphere. For
instance, when 5 nodes are on the same sphere, 5 tetrahedra may be defined satisfying the
Delaunay criterion, but some of them may have zero or almost zero volumes, called slivers,
as seen in Fig. 8:

Figure 8: Five nodes on the same sphere and possible zero or almost zero volume
tetrahedron (sliver) on the right.
In order to overcome above drawbacks, a generalization of the Delaunay tessellation will be
defined.
Definition: The Extended Delaunay tessellation within the set N is the unique partition of
the convex hull of all the nodes into regions
i
such that = U
i ,
where

each
i
is the
polyhedron defined by all the nodes laying on the same Vorono sphere.
The main difference between the traditional Delaunay tessellation and the Extended
Delaunay tessellation is that, in the latter, all the nodes belonging to the same Vorono
sphere define a unique polyhedron. With this definition, the domain is divided into
tetrahedra and other polyhedra, which are unique for a set of node distributions. Fig. 9 for
instance, is a 2-D polygon partition with a triangle, a quadrangle and a pentagon. Fig. 10 is
a classical 8-nodes polyhedron with all the nodes on the same sphere.

Figure 9: Two-dimensional partition in polygons.
The triangle, the quadrangle and the pentagon are each inscribed on a circle
For non-uniform node distributions, considering infinite precision, only 4 nodes are
necessary to define a sphere. Other nodes close to the sphere may define other spheres very
close to the previous one. In order to avoid this situation, which may hide polyhedra with
more than 4 nodes, a parameter will be introduced. In such a way, the polyhedra are
defined by all the nodes of the same sphere and nearby spheres with a distance between
center points smaller than .

Figure 10: Eight-node polyhedron. All nodes are on the same sphere.
The parameter avoids generating zero volume or near zero volume tetrahedra. When is
large, the number of polyhedra with more than 4 nodes increases, and the number of
tetrahedra with near zero volume decreases, and vice versa.
The Extended Delaunay tessellation leads to a domain partition which: (a) is unique for a set of
node distributions; (b) is formed by polyhedra with no zero volume, and (c) is obtained in a
bounded time of order n. Then, it satisfies the conditions for a meshless method.

The Meshless Finite Element shape functions

Once the domain partition in polyhedra is defined, shape functions must be introduced to
solve a discrete problem. In fluid and solid mechanics problems typically, C
0

continuitions
shape functions are chosen. If possible, shape functions must be locally supported in order
to obtain band matrices. They must also satisfy two criteria in order to have a reasonable
convergence order, namely partition of unity and linear completeness.
In order to define the shape functions inside each polyhedron the non-Sibsonian
interpolation is used [23].
Let P = {n
1
, n
2
, , n
m
} be the set of nodes belonging to a polyhedron. The shape function
N
i
(x) corresponding to the node n
i
at an internal point x is defined by building first the
Vorono cell corresponding to x in the tessellation of the set P U {x} and then by
computing:

=
=
m
j j
j
i
i
i
h
s
h
s
N
1
) (
) (
) (
) (
) (
x
x
x
x
x (83)
where s
i
(x) is the surface of the Vorono cell face corresponding to node the node n
i
and
h
i
(x) is the distance between point x and the node n
i
(Figure 11).

x
n
i
s
i
h
i

Figure 11: Four nodes and arbitrary internal point x Vorono diagram. Shape function
parameters
Non-Sibsonian interpolations have the following properties [19].

1) 0 N
i
(x) 1 (1)
2)
i
N
i
(x) = 1 (2)
3) N
i
(n
j
) =
ij
(3)
4) x =
i
N
i
(x) n
i
(4)
Furthermore, the particular definition of the non-Sibsonian shape function for the
limited set of nodes on the same Vorono sphere, adds the following properties:
5) On a polyhedron surface, the shape functions depend only on the nodes of this surface
6) On triangular surfaces (or in all the polygon boundaries in 2-D), the shape functions are
linear.
7) If the polyhedron is a tetrahedron (or a triangle in 2-D) the shape functions are the
linear finite element shape functions.
8) Due to property 5, the shape functions have C
0
continuity between two neighboring
polyhedra. See Figure 12.
9) As a matter of fact, because all the element nodes are on the same sphere, the evaluation
of the shape functions and its derivatives becomes very simple.


Figure 12 .C
0
continuity of the shape function on a 2-D node connection.
The method Meshless Finite Element Method (MFEM) defined here is both a meshless
method and a Finite Element Method. The algorithm steps for the MFEM are:

1) For a set of nodes, compute all the empty spheres with 4 nodes.
2) Generate all the polyhedral elements using the nodes belonging to each sphere and
the nodes of all the coincident and nearby spheres.
3) Calculate the shape functions and their derivatives, using the non-Sibsonian
interpolation, at all the Gauss points necessary to evaluate the integrals of the weak
form.

The MFEM is a truly meshless method because the shape functions depend only on the
node positions. Furthermore, steps 1 and 2 of the node connectivity process are bounded
with n
1.33
, avoiding the mesh "cosmetics" often needed in mesh generators.
The number of Gauss points necessary to compute the element integrals depends, to a great
extent, on the polyhedral shape of each element. Note that for an irregular node
distribution, there remains a significant amount of tetrahedra ( in the examples, more than
85% of the elements remains tetrahedral) with linear shape functions, for which only one
Gauss point is enough. For the remaining polyhedra, the integrals are performed dividing
them into tetrahedra and then using a single Gauss point in each tetrahedron. This
subdivision is only performed for the evaluation of the integrals and cannot be considered
as a tetrahedral mesh because it is not conforming. The use of one Gauss point on each
tetrahedron guarantee that the computing time in the evaluation of the matrices requires the
same effort than the FEM.

References

[1] R.A. Gingold and J.J. Monaghan, Smoothed particle hydrodynamics, theory and
application to non-spherical stars, Mon. Nat. Roy. Astr. Soc., 181, (1997), 375389.
[2] J. Bonet and S. Kulasegaram, Convection and stabilization of smooth particle
hydrodynamics methods with applications in metal forming simulation, International
Journal for Numerical Methods inEngineering, (1999).
[3] G.A. Dilts, Moving least squares particle hydrodynamics. i. consistency and stability,
International Journal for Numerical Methods in Engineering, 44, (1999), 11151155.
[4] S. Koshizuka and Y. Oka Moving particle semi-implicit method for fragmentation of
incompressible fluid, Nuclear Engineering Science, 123, (1996), 421434.
[5] T. Belytschko, Y. Liu, and L. Gu, Element free galerkin methods, International Journal
for Numerical Methods in Engineering, 37, (1994), 229256
[6] S. De and K.J. Bathe, The method of finite spheres, Computational Mechanics, 25
(2000), 329-345.
[7] B. Nayroles, G. Touzot and P. Villon, Generalizing the fem: Diffuse approximation and
diffuse elements, Computational Mechanics, 10, (1992), 307318.
[8] E. Oate, S.R. Idelsohn, O.C. Zienkievicz and Taylor R.L, A finite point method in
computational mechanics. Applications to convective transport and fluid flow,
International Journal for Numerical Methods in Engineering, 39(22), (1996a), 3839
3886.
[9] E. Oate, S.R. Idelsohn, O.C. Zienkievicz, Taylor R.L. and C. Sacco, A stabilized finite
point method for analysis of fluid mechanics problems, Computer Methods in Applied
Mechanics and Engineering, 39, (1996a), 315346.
[10] R.L. Taylor, O.C. Zienkiewicz, E. Oate, and S.R. Idelsohn, Moving least square
approximations for solution of differential equations, Internal Report 74, CIMNE,
Barcelona, Spain, (1996).
[11] S.R. Idelsohn, M.A. Storti and E. Oate, Lagrangian formulations to solve free surface
incompressible inviscid fluid flows, Comput. Methods Appl. Mech. Engrg. 191,(2001)
583-593
[12] S.R. Idelsohn, E. Oate, N. Calvo and F. Del Pin, The meshless finite element method,
Submitted to Int. J. for Numerical Methods in Engineering, (2001).
[13] H. Edelsbrunner and E.P. Mucke, Three-dimensional alpha-shape. ACM Transactions
on Graphics, 3, (1994), 4372.
[14] N. Calvo, S.R. Idelsohn and E. Oate, The extended Delaunay tesselation, Submitted
to Computer Method in Applied Mechanics and Engineering, (2002).20
[15] K.J.Bathe, H.Zhang and S.Ji, Finite element analysis of fluid flows coupled with
structural interactions, Comp. and Structures 72 (1999) 1-16
[16] H.Zhang and K.J.Bathe, Direct and iterative computing of fluid flows fully coupled
with structures., Comput. Fluid and Solid Mechanics, (2001), 1440-1443.
[17] S.Rugonyi and K.J.Bathe, , CMES, Vol 2, No2, (2001) 195-212.
[18] R.Codina, Pressure Stability in fractional step finite element methods for
incompressible flows, Journal of Comput. Phisycs, 170,(2001) 112-140.
[19] Sukumar N, Moran B, Semenov AYu and Belikov VV. Natural neighbour Galerkin
Methods. International Journal for Numerical Methods in Engineering. 2001; 50:1-27.
[20] E. Oate Derivation of stabilized equations for advective-diffusive transport and fluid
flow problems. Comput. Methods Appl. Mech. Engrg.,1998, 151:1-2, 233--267.
[21] E. Oate. A stabilized finite element method for incompressible viscous flows using a
finite increment calculus formulation. Comput. Methods Appl. Mech. Engrg.,2000,
182:1--2, 355--370.
[22] E. Oate, Possibilities of finite calculus in computational mechanics. (2002) Submitted
to Int. J. Num. Meth. Engng.
[23] Belikov V and Semenov, A. Non-Sibsonian interpolation on arbitrary system of points
in Euclidean space and adaptive generating isolines algorithm. Numerical Grid
Generation in Computational Field Simulation, Proc. of the 6th Intl. Conf. Greenwich
Univ. July 1998.