Comput. Methods Appl. Mech. Engrg.

191 (2001) 561582

www.elsevier.com/locate/cma

On the coupling of 3D and 1D NavierStokes equations

for ow problems in compliant vessels q
L. Formaggia a, J.F. Gerbeau b, F. Nobile a, A. Quarteroni

a,c,*


D
epartement de Math
ematiques, Ecole
Polytechnique F
ed
erale de Lausanne, CH-1015 Lausanne, Switzerland
b
INRIA Projet M3N Rocquencourt, B.P.105, F-78153 Le Chesnay Cedex, France
Dipartimento di Matematica ``F. Brioschi'', Politecnico di Milano, Piazza Leonardo da Vinci 32, I-20133 Milano, Italy
a

Received 3 January 2000; received in revised form 27 October 2000

Abstract
For the analysis of ows in compliant vessels, we propose an approach to couple the original 3D equations with a convenient 1D
model. This multi-scale strategy allows for a dramatic reduction of the computational complexity and is suitable for ``absorbing''
outgoing pressure waves. In particular, it is of utmost interest for the description of blood motion in the arterial system. 2001
Elsevier Science B.V. All rights reserved.
Keywords: Fluidstructure interaction; Computational uid dynamics; Absorbing boundary conditions; Multi-scale models;
Haemodynamics

1. Introduction
The study of a transient incompressible ow inside a compliant vessel, that is a vessel whose walls deform under the action of the uid with possibly large displacements, is of considerable interest, for instance
in the simulation of blood ow in large arteries [17].
The mathematical problem consists of the 2D or 3D NavierStokes equations describing the uid
motion, coupled with a suitable model that describes the displacement of the vessel wall (hereafter called
``the structure''). We will refer to this coupled system as the 2D/3D uidstructure model.
This problem is interesting from the mathematical and numerical modelling viewpoints, as it exhibits
some peculiar features like the appearance of travelling pressure waves along the vessel. Indeed, even if the
ow is governed by parabolic equations such as the incompressible NavierStokes, the behaviour of the
coupled uidstructure system is in many aspects more akin to that of a hyperbolic problem. As a consequence, an additional complexity arises in the treatment of the ``inow'' and ``outow'' boundaries, where
one would like to have a correct representation of the travelling waves, without spurious reections. The
case of absorbing boundary conditions for compressible ows and more general hyperbolic equations has
been widely studied, see for instance [3,6,7]. However, in our case, where the propagation phenomena are

Supported by the Swiss N.S.F. (Project 21-54139.98) and by MURST Con. 1998, ``Advanced Numerical Methods for Scientic
Computing''.
*
Corresponding author. Tel.: 21-693-5546; fax: 41-21-693-4303.
E-mail addresses: luca.formaggia@ep.ch (L. Formaggia), Jean-Frederic.Gerbeau@inria.fr (J.F. Gerbeau), fabio.nobile@ep.ch
(F. Nobile), alo.quarteroni@ep.ch (A. Quarteroni).
0045-7825/01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 0 4 5 - 7 8 2 5 ( 0 1 ) 0 0 3 0 2 - 4

562

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

due to the uidstructure interaction and not to the uid compressibility, little may be found in the literature.
In this work, this issue has been tackled by coupling the 2D/3D uidstructure problem with a reduced
one-dimensional model, which acts as an ``absorbing'' device for the waves exiting the computational
domain. This reduced model is obtained integrating the uid equations over each section normal to the axis
of the vessel and describes the coupled system in terms of transversally averaged ow rate and pressure.
This work is also motivated by the prospect of implementing multi-scale models for the human cardiovascular system. It is clear that a realistic detailed numerical simulation of the ow in a segment of an
artery (like, for instance, the carotid bifurcation or a trait of a coronary) may not be fully accomplished
without accounting for the interactions with the remaining part of the cardiovascular system. One possibility to this direction is oered by the coupling of the detailed 2D/3D simulation with a simplied 1D or
even lumped parameter modelling of the rest of the global system. The work detailed in this paper represents a necessary step towards this goal.
In Section 2 we will derive the formulation for the 3D uidstructure model. A simplied one-dimensional formulation for the structure is also presented and used for the stability analysis of the coupled
problem. The main result of the section is an energy inequality which, under a condition on the positivity of
the kinetic energy ux at the outow, provides a bound of the global energy of the coupled problem in
terms of initial and boundary data.
Section 3 is devoted to the derivation and analysis of the 1D model of the ow inside a compliant vessel.
Under suitable approximations, we derive a hyperbolic system of equations with source term. A proper set
of boundary conditions is derived by means of standard characteristics analysis. Then the stability of the
system is analysed for the sub-critical case, which is the most relevant for our target applications. Again a
bound on the energy of the system is found in terms of the initial and boundary data, stating the stability of
the 1D model.
Next, we present several strategies for coupling the two models and possible algorithms based on subdomain iterations. Details on the adopted numerical discretisation are given in Section 5. For the 2D/3D
uidstructure models we have employed nite element techniques and we have used an Arbitrary Lagrangian Eulerian formulation to account for the boundary movement in the uid region. The 1D hyperbolic system is solved by a LaxWendro scheme.
Finally, numerical results are provided both for the 3D1D and the 2D1D models' coupling. We
compare the solution behaviour when a homogeneous Neumann boundary condition on the velocity is
prescribed at the outlet with the solution obtained by the coupling. The latter shows a greatly reduced level
of spurious reections. For the 2D1D case a quantitative analysis on the error is given by comparing the
results with that obtained on a vessel of double length.
The main applicative eld addressed here is blood ow in arteries. Yet, the model we have considered
may be extended to the study of vibrations or water hammer eects in pipe network ows. These classes of
problems have been extensively investigated, see for instance the review paper [14,23] for a nite element
space-time formulation of the ow in a cantilevered pipe. The methodology proposed in this work may be
of interest for these situations as well.

2. The 3D model
In this section and in the next one, we illustrate the models we have used to describe the uid and the
structure movement and their interaction: the 3D uidstructure interaction problem and the derivation of
the 1D reduced model, respectively.
Let us consider the district illustrated in Fig. 1, which we will denote by X. The vessel has a compliant
wall denoted by Cw . The boundary of the vessel is completed by the sections S1 and S2 , connecting the
district to the rest of the system. More precisely, S1 is the upstream section; S2 is the downstream section,
through which the uid enters and leaves X, respectively.
We will denote by ut; x (with x 2 X, t > 0) the uid velocity eld, by pt; x the pressure and by q the
(constant) uid density. We describe the uid motion through the classical NavierStokes equations

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

563

Fig. 1. Simple compliant tube.

ot u u
ru
div u 0

1
div Tu; p
q

in X;

where T is the Cauchy stress tensor. Hereafter, we will consider the case of Newtonian uids where we can
use the following constitutive relation:
Tu; p

pI lru;

l being the uid viscosity.

Remark 1. Concerning the blood ow problem, in large arteries blood can be assumed to behave as a
Newtonian uid [13,26] while in small vessels and capillaries, its rheology is more complex. An account is
given in [2,16,26].
Whatever PDE model is considered for the vessel wall, it would provide the position of every point on Cw
at any time t > 0, which is typically described by means of the displacement g with respect to the reference
position C0w (see for instance [1,11]).
Then, matching conditions at Cw may be provided as follows:

u g_
2
on Cw ;
T
n pext n U
where pext is the given external pressure, U is the forcing term acting on the wall and n is the outward unit
vector to Cw .
The rst of (2) guarantees the perfect adherence of the uid to the structure while the second one states
the continuity of the stresses at the interface (according to the action and reaction principle).
Both uid and structure equations must be supplied with initial conditions (resp. on X and C0w ) and
boundary conditions (resp. on S1 ; S2 and oS1 ; oS2 ).
About the structure model, we will address two cases:
1. A shell model, i.e. a 3D structure in which one dimension, the thickness, is much lower than the other
ones. In particular we have considered the so-called geometrically exact shell model developed in [20
22]. The main unknowns are the mid-surface position of the shell and its normal (or more precisely a
director vector) at each point. In this model, deformations are described without approximation on
the geometry even when large displacements are considered which is particularly interesting for
the uidstructure interaction problem arising in the study of blood ow.
2. A simplied structural model, which is derived for a cylindrical conguration (generalised string model;
see [17]). Let
C0w fr; h; z: r R0 ; 0 6 z 6 L; 0 6 h < 2pg
be a cylindrical reference surface of radius R0 ; we neglect the longitudinal and angular displacement
while the radial displacement gr gr t; h; z is given by
qw h

o2 gr
ot2

kGh

o2 gr
Eh gr

oz2 1 m2 R20

o3 g r
f t; h; z:
oz2 ot

564

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

Here h is the wall thickness, R0 is the arterial reference radius at rest; k is the so-called Timoshenko shear
correction factor; G is the shear modulus; E is the Young modulus, m is the Poisson ratio (which for an
incompressible material is equal to 1=2); qw the wall volumetric mass, c is a viscoelastic parameter and
nally f is an external forcing term.
Model (3) is basically derived from the equations of linear elasticity for a cylindrical tube with small
thickness, under the hypotheses of plane stresses and membrane deformations (i.e. negligible elastic
bending terms). The term kGho2 gr =oz2 accounts for shear deformations [19] while the term co3 gr =oz2 ot
introduces a viscoelastic behaviour.
For the purpose of the mathematical analysis, we will play with a 2D model obtained by intersecting the
tube X with a plane h h (see Fig. 2). Correspondingly, we will consider the 2D problem arising form the
combination of the 2D NavierStokes equations for the uid with Eq. (3) to describe the motion of
the upper and lower boundaries (with a frozen value of the angle h).
Remark 2. Although not being completely realistic for blood ow problems, this simplied 2D analysis
maintains all the mathematical aspects peculiar to the original coupled uidstructure problem and will
therefore be adopted to test the numerical coupling algorithms we will present in Section 5.
2.1. Energy inequality for the coupled uidstructure problem
Let us consider the 3D NavierStokes equations coupled with the structure model (3) through the
matching conditions (2). In Eq. (3), the forcing term f t; h; z is given by the radial component Ur of the
normal stress exerted by the uid, recast in the reference conguration C0w , that is
s

2 
2
R
ogr
1 ogr
1
on C0w :
f t; h; z Ur

4
R0
R oh
oz
Here we have noted R R0 gr , while the term under square root accounts for the change in the surface
measure passing from Cw to C0w . Clearly, with such a right-hand side, Eq. (3) becomes non-linear and
dicult to handle with. However, as we will show in Section 5, at the numerical level we will decouple the
uid and structure problems, and the right-hand side will be computed from a previous step of the staggered algorithm.
Remark 3. Since, from (2), we have Ur
may note that
n

N
jNj

with N Rer

ogr
eh
oh

T pext I
n
er , where T is the Cauchy stress tensor eld, we

ogr
ez
oz

Fig. 2. Intersecting the cylinder with the plane h h we obtain a 2D geometry with a 1D compliant wall.

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

and the right-hand side of (3) becomes

s

2 
2
R
ogr
1 ogr
Ur
1

R0
R oh
oz

1
T pext I
N
er
R0


Trr

pext Trh

565

1 ogr
og
Trz r
R oh
oz

R
;
R0

which is now linear in gr provided we approximate R by R0 . Moreover, under the same approximation, if T
is an isotropic tensor eld, i.e. T pI, we have T pext I
N
er =R0 p pext and the right-hand side
reduces to the only pressure difference p pext .
We now analyse in more detail the coupled uidstructure problem with the following initial and
boundary conditions:
For the fluid
u u0

for t 0

u0

on S1 ;

in X;
5

T
n 0 on S2 :
For the structure
gr g0 ;

g_ r er u0

gr 0

for t 0 on Cw ;

for z 0; z L:

For the sake of simplicity, and without loss of generality, we will consider only the case pext 0. Let us
rewrite Eq. (3) in the general form
s

2 
2
o2 gr
o2 gr
o3 gr
R
ogr
1 ogr
Ur
1
7
a 2 bgr c 2

q~w 2
R0
R oh
ot
oz
o z ot
oz
on C0w and for t > 0, where all the coecients q~w , a, b, and c are positive quantities. We wish to derive an
energy inequality for the coupled problem. The following lemma holds:
Lemma 4. The coupled problem (1), (7) with matching conditions (2) and initial and boundary conditions (5)
and (6) satisfies the following energy equality:
(
)

2

2




1 d
og
og
2
r
r
qkuk2L2 X q~w
ot 2 0 a oz 2 0 bkgr kL2 C0w
2 dt
L Cw

L Cw

2 2
Z
o gr
1
2
2


lkrukL2 X c
q
juj u
n dc 0:
oz ot L2 C0w 2 S2
Moreover, if
Z
2
juj u
n dc > 0
S2

8t > 0;

we obtain the a priori energy estimate

2

2
ogr

ogr

2



T

qkuT k2L2 X q~w

a
ot
2 0
oz
2 0 bkgr T kL2 C0w
L Cw
L Cw
Z T
Z T 2 2
o gr

2
lkruk2L2 X dt 2
c
oz ot 2 0 dt 6 C;
0
0
L Cw
where C is a constant which depends on the initial conditions u0 and g0 .

10

566

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

Proof. We rst multiply Eq. (7) by ogr =ot and integrate on the cylindrical surface C0w , obtaining
(
)



2 2
ogr 2
ogr 2
o gr
1 d
2





bkgr kL2 C0w c
q~w
a


oz ot 2 0
2 dt
ot L2 C0w
oz L2 C0w
L Cw
s


2 
2
Z
R
ogr
1 ogr
ogr
1
R0 dh dz:

Ur

R0
R oh
oz
ot
C0w


zL
ogr ogr
o2 gr ogr 
a
c

oz ot
oz ot ot 
z0

11

Thanks to the specic boundary conditions, the last term on the left-hand side is identically zero, while the
right-hand side, if computed on the actual conguration Cw , becomes
s

2 
2
Z
Z
Z
ogr
1 ogr
ogr
og
R dh dz
Ur 1

Ur r dc
T
n
u dc;
R oh
oz
ot
ot
C0w
Cw
Cw
where in the last equality we have exploited the coupling conditions (2). Then, the following energy equality
holds for the structure problem:
(
)



2 2
Z
ogr 2
ogr 2
o gr
1 d
2






bkgr kL2 C0w c

T
n
u dc:
12
a
q~w
2 dt
ot L2 C0w
oz L2 C0w
oz ot L2 C0w
Cw
Analogously for the uid equations, by multiplying the rst of (1) by u and integrating over X we can derive
the following energy estimate (see e.g. [17]):


Z
Z
1 d
1
2
2
2
q kukL2 X lkrukL2 X
qjuj n
u dc
T
n
T
n
u dc
2 dt
2
S1 [S2
Cw
and, exploiting the boundary conditions, we have
Z
Z
1 d
1
q kuk2L2 X lkruk2L2 X
qjuj2 n
u dc
T
n
u dc:
2 dt
S2 2
Cw

13

R
By summing equalities (12) and (13) the term Cw T
n
u dc cancels out and we obtain (8).
Finally, integrating (8) in time between 0 and T, under the hypothesis (9), we obtain the desired inequality (10). 
The hypothesis (9) is actually satised if S2 is an outflow section, i.e. u
n > 0 for all x 2 S2 . However, in
vascular problems, this assumption is seldom true because the pulsating nature of blood ow might induce
a ow reversal along portions of an artery during the cardiac beat.
We may observe that the ``viscoelastic term'' in (3) allows one to obtain the appropriate regularity of the
velocity eld u on the boundary (see [17]).
In the derivation of energy inequality (10), we have considered homogeneous boundary conditions both
for the uid and the structure. However, the conditions g 0 at z 0 and z L, which correspond to hold
the wall ends xed, are not realistic in the blood ow context. Since the model (3) for the structure is of
propagative type, rst-order absorbing boundary conditions are a better choice, i.e.
r
ogr
a ogr
0 at z 0;
14
ot
q~w oz
r
ogr
a ogr

0 at z L:
15
ot
q~w oz
In this case, the nal estimate (10) is still valid. Indeed the boundary term which appears in (11) now should
read

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

567


zL
ogr ogr o2 gr ogr 
a

oz ot oz ot ot z0
"
r "
 2 
 2 #
 2 
 2 #
p ogr 
ogr 
c q~w d
ogr 
ogr 

q~w a

:
2 a dt
ot z0
ot zL
ot z0
ot zL
This term, integrated in time, would eventually appear on the left-hand side of inequality (10). We may
note, however, that we obtain at both z 0 and z L the following expression:
r
r

2
2
2
p Z T ogr
c q~w ogr
c q~w ogr
s ds
T
0 :
16
q~w a
2 a ot
2 a ot
ot
0
Therefore, this additional term is positive and it depends only on the initial conditions.
Yet, conditions (14) and (15) are not compatible with the homogeneous Dirichlet boundary conditions
for the uid; indeed, if gjz0 6 0 and u 0 on S1 , the trace of ut; x on the boundary is discontinuous and
thus not compatible with the regularity required on the solution of (1) (see e.g. [18]).
A possible remedy consists in changing the condition u 0 on S1 into
(
u
ez g
on S1
T
n T
n
ez 0
with g 0 on oS1 . An energy inequality for the coupled problem can be derived also in this case with
standard calculations, taking a suitable harmonic extension g~ of the non-homogeneous data g such that
g~jS1 g

and

g~jCw 0:

The calculations are here omitted for the sake of brevity.

3. The 1D model
In the case where the boundary C0w of the reference conguration X0 is a cylinder of radius R0 , a simplied 1D model can be obtained integrating, at each time t > 0, the NavierStokes equations (1) over each
section St; z normal to the axis z of the cylinder. In the sequel, At; z denotes the area of St; z and
uz t; z the axial velocity, while Qt; z and pt; z are the ow rate (or ux) and the mean pressure in every
section, given by
Z
Z
1
 z
uz t; z dr; pt;
pt; z dr:
Qt; z
At; z St;z
St;z
Finally u Q=A denotes the mean axial velocity.
The 1D model which is obtained reads (see [5])
8
oA oQ
>
>

0;
>
<
ot
oz
 2
>
oQ o
Q
A op
Q
>
>

a
KR 0:

:
ot oz
q oz
A
A

17

The constant KR is a resistance parameter which accounts for uid viscosity, while a, sometimes indicated
Coriolis coefcient, is dened as
Z
 Z
 2
2
aA
uz dr
uz dr
S

568

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

and accounts for the fact that the ux of momentum computed with the averaged quantities (i.e. Q2 =A) is in
general dierent from the actual one. Even if, in general, a is a function of z and t, we have taken it constant
and equal to one, which corresponds to assume a flat velocity prole.
 A and Q). For its closure, a third
System (17) is a system of two equations in three unknowns (p,
equation is provided by a suitable wall model relating the radial displacement (and henceforth the area A)

to the mean pressure p.
Here we consider a simple algebraic relation between pressure and area of the type
p pext wA with

dw
>0
dA

and

wA0 0;

18

where pext is the constant external pressure and A0 pR20 is the reference area. Then, system (17) turns out
to be hyperbolic and possesses two distinct eigenvaluesc
where c21 A

k1;2 u  c1 ;

A owA
:
q oA

19

The corresponding eigenfunctions are the characteristic variables, here given by

Z
W1;2 u 

A
A0

c1 s
ds:
s

20

They are not constant along the characteristic curves dzi =dt ki , i 1; 2 because of the presence of a
source term.
An example of the algebraic relation of type (18) can be derived considering the static counterpart of
Eq. (3), neglecting the spatial derivative and considering the pressure jump p pext as forcing term. In such
a case, we have
p

pext

Eh gr
:
1 m2 R20
2

Since A pR0 gr , we obtain

p

pext

p
b A

p;

where

p
b 1 pm2hEA0 ;
p
p  b A0 :

21

Henceforth, in view of (18), we have that

p
wA b A

p :

The characteristic variables now are

s
2p
p pext p
W1;2 u  2
q

22

p
p :

23

In our simulations we have adopted this algebraic model. A presentation of other models that make use of
dierential laws linking p to A and its time derivatives is reported in [5].
Remark 5. In the blood ow context, other algebraic relationships between p and A, possibly featuring a
better accordance with experiments, can be found in [9,12,24].
When considering a 2D geometry like the one shown in Fig. 2, Eqs. (17) are still valid provided we now
take A 2R0 gr . The algebraic relation (22) becomes

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

569

8
<b

hE
;
21 m2 R20
wA bA p ; where
: 
p bA0
p p
and the characteristic variables are W1;2 u  2 1=q p pext p


p
p .

3.1. Energy conservation for the 1D model

Here we derive an a priori estimate for the solution of system (17) in the interval I a; b under the
hypotheses that 8t > 0; k1 > 0; k2 < 0 (sub-critical ow regime) and the area A maintains positive.
Remark 6. The assumption k1 > 0 and k2 < 0 is sound in haemodynamics since c1 is much greater than j
uj
in physiologic conditions.
We will consider the following initial and boundary conditions:
A0; z A0 ;

initial conditions :

boundary conditions :

Q0; z Q0 ;

W1 g1

at z a;

W2 g2

at z b:

a 6 z 6 b;

24
25

We dene the energy of the 1D model, for each t > 0, as

1
Et q
2

At; z
u2 t; z dz

Z
a

WAt; z dz;

26

where
Z
WA

A0

wf df:

27

Owing to (18) we may observe that

WA0 W0 A0 0

W00 A > 0 8A > 0:

and

Therefore WA is always positive and Et is a positive function for all Q and A > 0 at each t > 0.
The following lemma holds:
Lemma 7. In the case a 1, system (17), supplied with an algebraic pressurearea relationship like in (18),
satisfies the following conservation property:
Z
ET qKR

Z
a

u dz dt

T
0


Q p

b
1 2 
u  dt E0:
pext q
2
a

28

Obviously, E0 depends only on the initial data A0 and Q0 .

Proof. Let us multiply the second equation of (17) by u and integrate over I. We will analyse separately the
four terms.
First term
Z
I1

oA
u
1
u dz
ot
2

b
a

o
u2
dz
A
ot

b
a

oA 2
1 d
u dz
ot
2 dt

Z
a

1
A
u dz
2
2

b
a

u2

oA
dz:
ot

29

570

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

Second term
Z b
oA
u2
I2 a
u dz
oz
#
" aZ
Z b
b
oA
u 2
o
u
u dz
dz
a
A
u2
oz
oz
a
a
#
" Z
Z
Z
1 b oA
u 2
1 b oA 3
3 b 2 o
u
a
dz
A
u
u dz
u dz
2 a
oz
2 a oz
2 a
oz
#
" Z
Z
1 b oQ 2
1 b oA
u3
a
dz :
u dz
2 a oz
2 a
oz

30

Now, using the rst equation in (17) we obtain

" Z
b #
b

a
oA 2
3 
I2
u  :
u dz A
2
ot
a
a

31

Third term. Using (18) and the fact that pext is independent of z we have
" Z
Z
Z b
b
A op
1 b o
1
oQ
p pext 
wA dz 
p
u dz
A 
u dz
I3
q a oz
q
oz
a q oz
a
Again, using the rst of (17) we have
"Z
" Z
b #
b
b

1
oA
1
d
wA dz 
p pext Q
WA dz 
p
I3
q a ot
q dt a
a

b #

pext Q :

b #

pext Q :

33

Fourth term
Z b
Z b
Q
KR u dz KR
u2 dz:
I4
A
a
a

34

In the case a 1, by summing the four terms we obtain the following equality:

b
Z b
Z b
Z b
1 d
d
1 2 
q
u  0:
A
u2 dz
WA dz qKR
u2 dz Q p pext q
2 dt a
dt a
2
a
a
Integrating Eq. (35) in time between 0 and T we obtain the desired result.

32

35

In order to draw an energy inequality from (28), we need to investigate the sign of the dierent terms.
The rst two of them are clearly positive. Concerning the third one let us analyse the homogeneous case (i.e.
the case where g1 g2 0).
We will rewrite the boundary term in (28) as a function of A, wA and c1 (which, in its turn, depends on
A and wA, see (19)).
If g1 g2 0 in (25), then
Z A
Z A
c1 f
c1 f
df 0 ) ut; a
df;
at z a; W1 u
f
f
A0
A0
Z A
Z A
c1 f
c1 f
df 0 ) ut; b
df:
at z b W2 u
f
f
A0
A0
Then, using (18)

b
1 2 
Q p pext q
u  F At; a F At; b;
2
a

36

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

where
Z
F A A

"
Z A
2 #
c1 f
1
c1 f
df wA q
df
:
f
2
f
A0

A
A0

571

37

We recall that we are assuming k1 > 0 and k2 < 0 (sub-critical ow regime). Then, from (19), we have
j
uj < c1 which implies at both z a and z b
Z A


c1 f 

df
38

 < c1 A:
f
A0
We are now in the position to conclude with the following result:
Lemma 8. If the function wA is chosen in such a way that F A > 0, for all A > 0 which satisfy (38), then the
energy inequality
Z T Z b
ET qKR
39
u2 dz dt 6 E0
0

holds for system (17).

The pressurearea relationship given in (22) satises the hypotheses of Lemma 8. In fact, in this case
p
WA b A

p
A0 ;

and
Z

A0

c1 f
df
f

A0

c21

s
b
f
2q

3=4

b p
A
2q

s
b  1=4
A
df 4
2q

1=4

A0

Condition (38) becomes

 s
 s


b  1=4
b 1=4
4 1=4
4 1=4

1=4 
A0  <
A
A and then A0 < A1=4 < A0 :
4


2q
2q
5
3

40

On the other hand we have

s
 
 q
2 
b  1=4
b  1=4
1=4
1=2
1=4
A
A
F A A4
A0
b A1=2 A0
16
A0
2q
2 2q
s




2 
b  1=4
1=4
1=4
1=4
1=4
A A
4b
A0
A1=4 A0
A1=4 A0
4 A1=4 A0
2q
s
2 

b  1=4
1=4
1=4
A A
5A1=4 3A0 :
4b
A0
2q
Then, condition F A > 0 gives
3 1=4
A1=4 > A0 ;
5
whose satisfaction is a consequence of (40). Therefore, in that case the 1D model is stable.
More generally, under the same relation (22), we can prove an energy estimate also in the case of nonhomogeneous boundary conditions. We have the following result:

572

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

Lemma 9. If the pressurearea relationship is given by (22), and the boundary data satisfy
r
p
4
2q

g1 t >

and

r
p
g2 t < 4
2q

8t > 0;

41

then there exists a positive function G Gg1 ; g2 ; p such that

Z
ET qKR

u2 dz dt 6 E0

T
0

Gg1 t; g2 t; p dt:

42

Proof. We will consider only the case where g1 6 0 and g2 0, since the most general case may be derived in
a similar fashion. We recall that relationship (22) satises the hypotheses on F A of Lemma 8. Then from
(35) we obtain the following inequality:

Z b
Z b
d
WA dz qKR
u2 dz 6 Q p
dt a
a
a



1
3 
uj  :
6 Aj
ujjwAj qAj
2
za

q d
2 dt

A
u2 dz


1 2 
u 
pext q
2
za
43

We have then at z a (from (23) and (25))

s
2p
u 2
p pext p
q

p
p g1

thus, from (18)

u g1

s
2p
wA p
2
q

p
p :

44

p p
On the other hand, k1 u 1= 2q p pext p > 0, and then
u >

1 p
p wA p :
2q

45

Combining (44) and (45) we have

1 p
p wA p < g1
2q

s
2p
wA p
2
q

p
p

and then
p
wA p < f1 g1 ; p ;

46

where
f1 g1 ; p

p
2q
4 p
g1
p:
3
3

The assumptions (41) are necessary conditions for the eigenvalues of system (17) being of opposite
sign.

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

573

Furthermore, we deduce from (22) that wA > p . The following inequalities thus hold at z a:


jwAj 6 max p ; jf12 g1 ; p p j ;
s
s



2p p
2 p

p  6 jg1 j 2
wA p
p f1 g1 ; p ;
j
uj  g1 2


q
q
A

1
1
2
wA p 6 2 f14 g1 ; p
b2
b

and the right-hand side in (43) may

thus be expressed by a positive function of the boundary data g1 and the
p
reference pressure at rest p b A0 , which we have indicated by Gg1 ; p . We obtain then the desired
stability inequality. 
3.2. Entropy function for the 1D model
If we take again the dierential problem (17) and we carry out derivations similar to those illustrated in
(29)(35), considering now the integrals over an arbitrary interval z; z dz  a; b, we may derive a relation analogous to (35). Then, passing to the limit as dz ! 0, the following pointwise dierential relation
may be deduced
 



o 1
o
1 2
2
qA
u WA
Q wA q
u
47
qKR u2
ot 2
oz
2
which can be interpreted as an entropy balance equation for the hyperbolic system. Indeed, the function
1
sA; Q qA
u2 WA
2

48

in an entropy for system (17), with an associated ux



1 2
u :
Fs A; Q Q wA q
2

49

The term qKR u2 may be recognised as a dissipative term.

4. Coupling the 3D model with the 1D model
We consider now two domains X3D and X1D as in Fig. 3 and we solve, in the rst, the 3D uidstructure
model while in the second we consider the simplied 1D model.
On the right-hand side of Ca the 1D model supplies the quantities: Aa , Qa , pa wA and
ua Q=A. We dene, then, the same quantities also on the left-hand side of Ca as
Z
1
u
n dr;
Aa jCa j; ua
jCa j Ca
Z
1

pa
p dr; Qa jCa j
ua :
jCa j Ca

Fig. 3. Coupling a 3D model with a 1D model.

574

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

Moreover, we assume that at Ca the wall displacement is a function of the section area,
gjCa gAa ;

50

g beingp
a given function. For instance, we can assume that Ca is a circle in which case
gjCa Aa =p R0 er .
It is a priori reasonable to look for the continuity of the following quantities at the interface Ca :
A area : Aa Aa ;
B mean pressure : pa pa ;
C flux :

Qa Qa ;

D entering characteristic :

s
2p
 pext p
pa
ua 2
q

p
p W1 a :

Since in the 1D model viscous terms have been (partially) neglected, the variable p can be either interpreted
as a mean pressure or as a mean normal stress. The condition [B] may then be replaced with the continuity
of the averaged normal stress, i.e. r T
n
n. Analogously, the characteristic variable on the left-hand side
can be calculated using the averaged normal stress in place of the mean pressure. We have then two
conditions alternative to B and D, respectively. Namely,
B1 averaged normal stress :
D1 char: entering variable :

ra pa ;
s
2p
ra pext p
ua 2
q

p
p W1 a :

In view of the splitting procedure that will be described in Section 4.1 to solve the coupled 3D1D model,
we are allowed to enforce, at the interface point a, only those conditions that will generate well-posed
individual sub-problems in X3D and X1D .
To this aim, we advocate four dierent sets of coupling conditions:
Interaction Model 1: conditions A, B, D. We note that B and D imply the continuity of u. With the further continuity of A we obtain that of Q. Thus also C is satised.
Interaction Model 2: conditions A, C, D. We note that A and C imply the continuity of u. If we further
 Thus also B is satised.
add D we have the continuity of p.
Interaction Model 3: conditions A, B1, D1. We note that B1 and D1 imply the continuity of u and with
the continuity of A we obtain that of Q. Thus also C is satised.
Interaction Model 4: conditions A, C, D1. We note that A and C imply the continuity of u. If we further
add D we have the continuity of r. Thus also B1 is satised.
4.1. Sub-domain iterations between 1D and 3D models
Each Interaction Model presented above can be split into two sub-problems:
A 1D problem with condition D (or D1) as boundary condition at the interface and an absorbing condition on the right end (i.e. zero entering characteristic variable). With this choice, the hyperbolic system
(17) is well posed, as we have shown in Lemma 9.
A 3D uidstructure problem with condition A as boundary condition for the structure equation and
condition B (B1) or C, depending on which interaction model we consider, as boundary condition for
the NavierStokes equations. On the inow section S1 we might impose ordinary Dirichlet or natural
conditions; for instance we could assign the velocity prole for the uid and the displacement of the wall.
With these choices the structure problem turns out to be well posed, while conditions B or C as such are
not sucient to close the uid equations since they only provide average and not pointwise values on Ca .

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

575

A possible way to impose such defective boundary conditions to the uid equations, known as ``do
nothing'' approach, is proposed in [10] and analysed in the context of blood ow in [17]. We have adopted
this approach to impose the condition [B1] in the simulations presented in Section 6.
An alternative approach based on Lagrange multipliers is investigated in [4].
We can then iterate between the two sub-domains to get the global solution at each time step. For instance, referring to the Interaction Model 3 (which is the one we have implemented) the iterative algorithm
reads as follows: given the solution of the coupled problem (let us say un , pn , gn for the 3D model and Qn , An
for the 1D one) and setting u0n1 un , p0n1 pn , and g0n1 gn
1. solve the 1D model (17) with condition D1 at the interface, evaluating W1 a as a function of
u0n1 ; p0n1 ; g0n1 ; we then obtain a solution Qn1
and A1n1 ;
1
2. solve the 3D problem with boundary conditions A and B1 at the interface, evaluating Aa and ra as
functions of A1n1 and Q1n1 . We obtain a solution u1n1 ; p1n1 ; gn1
1 .
We iterate until the coupling conditions are satised within a xed tolerance. We can eventually add a
relaxation step on the variable W1 a .
5. Finite element discretisation of the 3D and 1D models
In this section we will present the schemes used for the numerical solution of the coupled problem together with some numerical results on the interaction between the 1D reduced model and the full uid
structure coupled model for both the 2D and the 3D cases.
The numerical solution of the 1D model has been obtained by a standard LaxWendro scheme [8].
Writing the system (17) in conservative form:
oU o
FU BU;
ot oz
where


A
U
;
Q

51

2
FU 4 Q
a
A
2

A0

c21 y dy

the LaxWendro scheme may be written as

"
n

oF
Dt2 o
n1
n
B
U U Dt
oz
2 oz

"

5;

BU

oF
oU

2

oU
oz

0
KR

#
Q ;
A

!

oF
oB
B
oU
oU

#n
oF oU
B
;
oU oz

52

where Un denotes the numerical solution at time step t tn .

We have used standard linear nite elements for the space discretisation. Even if the temporal discretisation is explicit, we need to invert a tridiagonal mass matrix. A characteristic boundary treatment has
been adopted and an extrapolation of the outgoing characteristic variable has been used to solve the
compatibility condition.
For the 2D simulations of the uidstructure interaction problem (1)(3), we have considered a conforming nite element discretisation. In particular, for the structure equation (3) we have used P1 nite
elements in space and a Newmark scheme in time, that is


 
1
n1
n
n
2
n1
g g Dtg_ Dt bN f
bN f n ;

2
53


g_ n1 g_ n Dt cN f n1 1 cN f n
gn ; g_ n being an approximation of gr tn ; g_ r tn and
fn

kGh o2 gn
qw h oz2

E
qw 1

m2 R20

gn c

o2 g_ n ^ n
U ;
oz2

576

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

^ n is the forcing term at time tn . The parameters bN and cN of the scheme have been taken equal to
where U
bN 1=4 and cN 1=2. We have shown in [15] that, with this choice, the Newmark scheme applied to
Eq. (3) is unconditionally stable (at least in the case of homogeneous Dirichlet boundary conditions).
For the uid we have adopted an ``Arbitrary Lagrangian Eulerian'' description and we have considered a
Finite Element approximation with P1 isoP2 elements for the velocity eld and P1 elements for the pressure.
The non-linear term is treated in a semi-implicit way and the equations are solved through the Yosida
projection scheme described in [25].
We have considered the following implicit algorithm to couple the two equations.
Given the solution un , pn , gn , g_ n , at time tn :
1. We extrapolate the displacement g and the velocity g_ of the structure
g0n1 gn Dtg_ n ;

g_ 0n1 g_ n :

2. We update the mesh by solving a Laplace problem which provides the displacement sxn of each point
xn 2 Xn as follows:

Ds 0
in X;
s gn1 gn on Cw ;
and setting xn1 xn sxn .
3. We assign the velocity of the uid on the interface Cw
u0n1 g_ 0n1 er :
^ n1 of Eq. (3).
4. We solve the uid equations and we evaluate the forcing term U
0
n1
5. We solve the structural equation obtaining a better estimate g~1 and ~g_ n1
of the displacement and the
1
velocity of the arterial wall, which allows to set
gn1
x~
gn1 1
1
g_ n1 x~
g_ n1 1
1

xg0n1 ;
xg_ 0n1 ;

where x is the relaxation parameter.

n1
n1
6. We go back to 2 and we iterate until the convergence of gkn1 , g_ n1
k , uk , pk .
The algorithm presented converges if x is taken suitably small. We have observed that the highest value
of x under which we have convergence depends on the wall mass qw and the length L of the tube but seems
independent of the elastic constants of the structural model and of Dt. Table 1 shows the critical value of x
with respect to qw and L.
For the 3D simulations, we have used the code SPECTRUM(TM). As for the 2D model, the moving
uid domain is handled with an ALE formulation. At each time step, the code solves iteratively the coupled
problem by treating simultaneously the uid and the structure equations in the rst step (``stagger'') (enforcing the adherence of the uid to the structure with a Lagrange multiplier technique), and in the second
step it accounts for the deformation of the grid.
Concerning the coupling with the reduced 1D model, we have considered Model 3 described in Section 4.
We have made just one sub-iteration between 1D and 3D (2D) model at each time step.
Table 1
Values of x which guarantee convergence of the algorithm
qw
qw
qw
qw
qw
qw

50
10
5
1
0:5
0:1

L2

L6

L 10

x61
x61
x61
x61
x 6 0:9
x 6 0:4

x61
x61
x61
x 6 0:4
x 6 0:2
x 6 0:05

x61
x 6 0:9
x 6 0:6
x 6 0:1
x 6 0:09
x 6 0:01

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

577

We have noticed that a ``brute force'' imposition of the continuity of the area A causes numerical instability, probably because of the discrepancy in the structural laws used in the two models. The matter is
still a subject of investigation. For the computations we are going to illustrate that we have relaxed the
constraint on the continuity of A by adopting, in the 2D case, the absorbing boundary conditions (14), (15)
and, in the 3D case, natural boundary conditions on the transversal displacement.

Fig. 4. Pressure pulse entering at the inow and homogeneous Neumann conditions at the outow; 3D simulation with SPECTRUM(TM). The displacement of the structure has been magnied by a factor 10.

578

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

6. Numerical results and model assessment

We will present two numerical tests.
Test 1: Pressure pulse in a compliant vessel. In the 2D case, we have considered a rectangular domain of
height 1 cm and length L 6 cm. The uid is initially at rest and an overpressure of 15 mm Hg
2
(2  104 dyn=cm ) has been imposed at the inlet for 0.005 s. The viscosity of the uid is equal to 0.035 P, its
3
2
density is 1 g=cm , the Young modulus of the structure is equal to 0:75  106 dyn=cm , its Poisson coef3
cient is 0.5, its density is 1:1 g=cm and its thickness is 0.1 cm.
In the 3D case, our computation has been made on a cylindrical domain of radius R0 0:5 cm and
length L 5 cm, with the following physical parameters: uid viscosity: 0.03 P, uid density: 1 g=cm3 , the
Young modulus of the structure: 3  106 dyn=cm2 , the Poisson coecient: 0.3 and structure density:
3
2
1:2 g=cm . Again, an overpressure of 10 mm Hg (1:3332  104 dyn=cm ) is imposed at the inlet for 0.005 s.
We have rst simulated the uidstructure interaction model without any coupling with the 1D reduced
model. In this case, natural boundary conditions for the uid (i.e. null normal stress) have been imposed on
the outlet. Figs. 4 and 5 show the uid pressure and the domain deformation in the 3D and the 2D case,
respectively. For the sake of clarity, the displacements shown in Fig. 4 are magnied by a factor 10.

Fig. 5. Pressure pulse entering at the inow and homogeneous Neumann conditions at the outow; 2D simulation. Solutions every 5
ms. Note how the pressure is reected at the outow boundary, giving rise to a backward travelling wave with negative amplitude.

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

579

A pressure wave is generated (together with a deformation wave at the same velocity), which is reected
at the outow section. Such a reection is clearly a numerical side-eect due to the boundary conditions
imposed at the outow.
In Figs. 6 and 7 we have considered the coupling with the 1D reduced model at the outow section. Here
the reections are greatly reduced because of the pressure wave is quite well absorbed by the 1D model.
Test 2: Pressure step in a compliant vessel error analysis. We have considered a rectangular domain
X 0; 12  0; 1, divided into two sub-domains X1 0; 6  0; 1 and X2 6; 12  0; 1. We have rst

Fig. 6. Coupling 3D SPECTRUM(TM) simulation with the 1D reduced model.

580

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

Fig. 7. Coupling 2D simulation with the 1D reduced model; solution every 5 ms. Note how the pressure wave exits the domain with
almost no spurious reections.

Fig. 8. Test 2: we compare the solution obtained solving the uidstructure interaction problem in X with the one obtained solving the
same FSI problem in X1 , coupled with the 1D model at the outlet.

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

581

Fig. 9. Relative error in L2 -norm both for pressure and velocity eld, between the solution of the coupled 2D/1D model and the
solution of the uidstructure interaction problem on a vessel of double length.

simulated the 2D uidstructure interaction problem in X, when a pressure step of 5000 dyn=cm2 is imposed at the inlet, 8t > 0, on a uid initially at rest. We have then considered the uidstructure problem
only in X1 , with the same conditions at the inlet and the coupling with the 1D reduced model at the outlet
(see Fig. 8).
Let u; p and u1 ; p1 be the velocities and pressures computed in the rst and second cases, respectively.
Fig. 9 shows the relative error in the L2 -norm between the two solutions restricted to X1 , i.e.
ep

kp

p1 kL2 X1

kpkL2 X1

eu

ku

u1 kL2 X1

kukL2 X1

The simulation has been stopped before the pressure wave reaches the outow of X. Both errors are less
then 2.5% at all times and the maximum corresponds to the instant at which the wave reaches the outow
section of X1 .
7. Conclusions
In this work we have shown that the coupling between a 3D uidstructure model and a 1D reduced
model is an eective way to greatly reduce spurious reections of the pressure waves on the articial
boundaries. We have proposed several strategies for coupling the two models. We have considered solution
algorithms based on sub-domain iterations and we have derived an energy estimate for the sub-problems
thus obtained.
Acknowledgements
The authors wish to thank Dr. Alessandro Veneziani for many fruitful discussions.
References
[1] P.G. Ciarlet, Introduction to Linear Shell Theory, Gauthier-Villars, Paris, 1998.
[2] G.R. Cokelet, The rheology and tube ow of blood, in: R. Skalak, S. Chen (Eds.), Handbook of Bioengineering, McGraw-Hill,
New York, 1987.
[3] B. Engquist, A. Majda, Numerical radiation boundary conditions for unsteady transonic ow, J. Comput. Phys. 40 (1) (1981)
91103.
[4] L. Formaggia, J.F. Gerbeau, F. Nobile, A. Quarteroni, Numerical treatment of defective boundary conditions for the Navier
Stokes equation, INRIA Research Report No. 4093, January 2001.

582

L. Formaggia et al. / Comput. Methods Appl. Mech. Engrg. 191 (2001) 561582

[5] L. Formaggia, F. Nobile, A. Quarteroni, A. Veneziani, Multiscale modelling of the circolatory system: a preliminary analysis,
Comp. Vis. Science 2 (1999) 7583.
[6] M.B. Giles, Non-reecting boundary conditions for Euler equation calculations, AIAA J. 28 (12) (1990) 20502058.
[7] D. Givoli, Non-reecting boundary conditions, J. Comput. Phys. 94 (1991) 129.
[8] E. Godlewski, P.-A. Raviart, Numerical Approximation of Hyperbolic Systems of Conservation Laws, Applied Mathematical
Sciences, vol. 118, Springer, New York, 1996.
[9] K. Hayashi, K. Handa, S. Nagasawa, A. Okumura, Stiness and elastic behaviour of human intracranial and extracranial arteries,
J. Biomech. 13 (1980) 175184.
[10] J.G. Heywood, R. Rannacher, S. Turek, Articial boundaries and ux and pressure conditions for the incompressible Navier
Stokes equations, Int. J. Numer. Methods Fluids 22 (1996) 325352.
[11] T.J. Hughes, The Finite Element Method, Linear Static and Dynamic Finite Element Analysis, Prentice-Hall, Englewood Clis,
NJ, 1987.
[12] G.L. Langewouters, K.H. Wesseling, W.J.A. Goedhard, The elastic properties of 45 human thoracic and 20 abdominal aortas in
vitro and the parameters of a new model, J. Biomech. 17 (1984) 425435.
[13] D.A. McDonald, Blood Flow in Arteries, third ed. edited by W.W. Nichols, M.F. O'Rourke, Edward Arnold, London, 1990.
[14] S. Mittal, T.E. Tezduyar, Parallel nite element simulation of 3D incompressible ows: uidstructure interactions, Int. J. Numer.
Methods Fluids 21 (1995) 933953.
[15] F. Nobile, Fluidstructure interaction problems in hemodynamics, Master's Thesis, Technical University of Milan, 1998 (in
Italian).
[16] K. Perktold, M. Resch, H. Florian, Pulsatile non-Newtonian ow characteristics in a three-dimensional human carotid bifurcation
model, ASME J. Biomech. Engrg. 113 (1991) 463475.
[17] A. Quarteroni, M. Tuveri, A. Veneziani, Computational vascular uid dynamics: problems, models and methods, Comp. Vis.
Science 2 (2000) 163197.
[18] A. Quarteroni, A. Valli, Numerical Approximation of Partial Dierential Equations, Springer Series in Computational
Mathematics, vol. 23, Springer, Berlin, 1994.
[19] H. Reismann, Elastic Plates: Theory and Application, Wiley, New York, 1988.
[20] J.C. Simo, D.D. Fox, On a stress resultant geometrically exact shell model, Part I: formulation and optimal parametrization,
Comput. Methods Appl. Mech. Engrg. 72 (1989) 267304.
[21] J.C. Simo, D.D. Fox, M.S. Rifai, On a stress resultant geometrically exact shell model, Part II: the linear theory; computational
aspects, Comput. Methods Appl. Mech. Engrg. 73 (1989) 5392.
[22] J.C. Simo, D.D. Fox, M.S. Rifai, On a stress resultant geometrically exact shell model, Part III: computational aspects of the
nonlinear theory, Comput. Methods Appl. Mech. Engrg. 79 (1989) 2170.
[23] A.S. Tijsseling, Fluidstructure interaction in liquid lled pipe systems: a review, J. Fluids Struct. 10 (1996) 109146.
[24] A. Tozeren, Elastic properties of arteries and their inuence on the cardiovascular system, ASME J. Biomech. Engrg. 106 (1984)
182185.
[25] A. Veneziani, Mathematical and numerical modelling of blood ow problems, Ph.D. Thesis, University of Milan, 1998.
[26] X.Y. Xu, M.W. Collins, C.J.H. Jones, Flow studies in canine aortas, ASME J. Biomech. Engrg. 114 (11) (1992) 504511.