Outflow boundary conditions for three-dimensional

finite element modeling of blood flow and pressure in arteries
Irene E. Vignon-Clementel
a
, C. Alberto Figueroa
a
,
Kenneth E. Jansen
c
, Charles A. Taylor
a,b,
*
a
Department of Mechanical Engineering, Stanford University, E350 Clark Center, 318 Campus Drive,
Stanford 94305-5431, CA 94305, USA
b
Department of Bioengineering and Surgery, Stanford University, E350 Clark Center, 318 Campus Drive,
Stanford 94305-5431, CA 94305, USA
c
Scientific Computation Research Center and the Department of Mechanical, Aeronautical and Nuclear Engineering,
Rensselaer Polytechnic Institute, Troy, NY 12180, USA
Received 23 November 2004; received in revised form 8 February 2005; accepted 13 April 2005
Abstract
Flow and pressure waves emanate from the heart and travel through the major arteries where they are damped, dis-
persed and reflected due to changes in vessel caliber, tissue properties and branch points. As a consequence, solutions to
the governing equations of blood flow in the large arteries are highly dependent on the outflow boundary conditions
imposed to represent the vascular bed downstream of the modeled domain. The most common outflow boundary con-
ditions for three-dimensional simulations of blood flow are prescribed constant pressure or traction and prescribed
velocity profiles. However, in many simulations, the flow distribution and pressure field in the modeled domain are
unknown and cannot be prescribed at the outflow boundaries. An alternative approach is to couple the solution at
the outflow boundaries of the modeled domain with lumped parameter or one-dimensional models of the downstream
domain. We previously described a new approach to prescribe outflow boundary conditions for simulations of blood
flow based on the Dirichlet-to-Neumann and variational multiscale methods. This approach, termed the coupled mul-
tidomain method, was successfully applied to solve the non-linear one-dimensional equations of blood flow with a vari-
ety of models of the downstream domain. This paper describes the extension of this method to three-dimensional finite
element modeling of blood flow and pressure in the major arteries. Outflow boundary conditions are derived for any
downstream domain where an explicit relationship of pressure as a function of flow rate or velocities can be obtained at
the coupling interface. We developed this method in the context of a stabilized, semi-discrete finite element method.
0045-7825/$ - see front matter Ó 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.cma.2005.04.014
*
Corresponding author. Address: Department of Mechanical Engineering, Stanford University, E350 Clark Center, 318 Campus
Drive, Stanford, CA 94305, USA. Tel.: +1 650 725 6128; fax: +1 650 725 9082.
E-mail address: taylorca@stanford.edu (C.A. Taylor).
Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
www.elsevier.com/locate/cma
Flow rate and pressure distributions are shown for different boundary conditions to illustrate the dramatic influence of
alternative boundary conditions on these quantities.
Ó 2005 Elsevier B.V. All rights reserved.
Keywords: Blood flow; Pressure; Boundary conditions; Coupled multidomain method; Finite elements
1. Introduction
In normal and disease states in children and adults, quantification of three-dimensional flow phenomena
and pressure fields is important to understand the response of the cardiovascular system to biomechanical
forces [1]. In the development of the circulatory system in normal subjects and congenital heart disease pa-
tients, flow modulates diameter and pressure controls wall thickness [2,3]. In older subjects, the initiation
and progression of atherosclerotic and aneurismal disease is directly affected by the complex three-dimen-
sional fluid mechanical environment of the major arteries [4,5]. Moreover, the ability to adequately simulate
flow and pressure is needed to model the performance of devices such as heart valves, LVADs (left ventric-
ular assist devices), filters, stents and stent-grafts. Non-invasive three-dimensional flow and pressure data
can also provide important information to determine the significance of an obstruction or predict the out-
come of a procedure [6,7].
In recent years, remarkable progress has been made in simulating blood flow in realistic anatomical
models constructed from three-dimensional medical imaging data. Arguably, accurate anatomic models
are of primary importance in simulating blood flow. However, as we demonstrate in this paper, realistic
boundary conditions are equally important in computing velocity and pressure fields. Yet, this subject
has received far less attention than image-based model construction for three-dimensional simulations.
In contrast, significant progress has been made in devising outflow boundary conditions for solving the
one-dimensional equations of blood flow in elastic vessels. For example, Stergiopulos solved the non-linear
one-dimensional equations of blood flow in a comprehensive model of the arterial system using a lumped
parameter model of the vasculature downstream of each branch in his numerical model [8]. Several groups
have developed and analyzed the coupling of one-dimensional equations with lumped models [9–12]. In
contrast to methods coupling the one-dimensional equations of blood flow to lumped parameter models,
Olufsen developed a distributed downstream model based on calculating the input impedance of an asym-
metric binary fractal tree using WomersleyÕs linear wave theory [13,14] and an algorithm for computing the
impedance of a vascular network first proposed by Taylor [15]. OlufsenÕs approach was to generate a fractal
tree for each outlet starting from a vessel that matched the diameter of the outlet and diminished in size
with each successive generation of vessels until a fixed terminal vessel size was attained. With this method
an impedance for each outlet of the upstream numerical model was computed naturally from linear wave
theory and branching laws. OlufsenÕs distributed model of the downstream vasculature enabled the repre-
sentation of more realistic flow and pressure waveforms than those obtained with lumped parameter mod-
els [16]. Steele and Taylor used a modified version of OlufsenÕs impedance boundary condition to model
blood flow at rest and during simulated exercise conditions [17]. In this case, vascular networks were as-
signed to the outlets of a model of the abdominal aorta, modified to represent the resting flow distribution
of 11 different subjects and then dilated to simulate the effects of lower limb exercise. Vignon and Taylor
developed a multidomain approach to couple one-dimensional equations to different lumped and one-
dimensional boundary conditions [18]. While these one-dimensional methods can be used to compute flow
rate and mean pressure, by design, they cannot be used to simulate complex three-dimensional flow phe-
nomena and pressure losses. Three-dimensional numerical methods have been used to compute velocity
fields and quantify shear forces acting on the surface of blood vessels. However, since most three-dimen-
sional models of blood flow use zero or constant pressure, zero traction, or prescribed velocity profiles
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3777
as outlet boundary conditions, blood pressure is not computed accurately and notably absent from reports
of hemodynamic investigations [19–24].
For simulations of blood flow in large arteries, the outlet boundary conditions represent the downstream
vasculature including smaller arteries, arterioles, capillaries, venules and veins returning blood to the heart.
Clearly, the vast extent and complexity of the circulation precludes a three-dimensional representation of
the entire circuit, yet ignoring the effect of the downstream circulation results in grossly inaccurate predic-
tions of velocity and pressure fields for many problems where the distribution of flow between the major
arteries is unknown. If zero or equal pressures or tractions are used for different outlets, the flow split will
be dictated solely by the resistance to flow in the branches of the domain of interest, neglecting the
dominant effect of the resistance of the downstream vascular beds. An alternative approach is to utilize
three-dimensional models for the major arteries where high-fidelity information is needed, and reduced-
order models to represent the remainder of the system. While closed-loop models are optimal, a simpler
approach is to directly represent the vasculature of the small arteries and arterioles using zero-dimensional
or one-dimensional models. These models can be terminated at the level of the capillary vessels where an
assumption of constant pressure is reasonable. Several groups [6,25–33] have successfully coupled three-
dimensional models to either resistances or more sophisticated zero-dimensional models (lumped models),
but this coupling has been performed iteratively, and generally applied to geometries with few outlets and
low resistances (as seen in the pulmonary vasculature). A further limitation of these methods to couple
three-dimensional and zero-dimensional models is the fact that there is no direct relationship between
the anatomy of the downstream vascular bed and the lumped parameters resulting in difficulties in speci-
fying these parameters and relating them to subsequent physiologic or pathophysiologic changes in the
downstream vasculature. In addition, for many simulations based on three-dimensional imaging data,
anatomic information is available for vessels downstream of the primary region of interest. The incorpora-
tion of such data would improve the accuracy of the models of the downstream vasculature.
Methods to couple three-dimensional and one-dimensional models were first described by Formaggia
et al. [34,35]. While great progress was made in these landmark papers, the coupling was performed for sim-
ple geometries and iteratively. Based on our experience, implicit coupling significantly improves conver-
gence, especially for models with multiple outlets. In addition, these papers did not include coupling
between three-dimensional domains and complex vascular networks as have been incorporated in one-
dimensional numerical solutions of blood flow in arteries. Since the vascular bed from the major arteries
to the capillaries can include tens of millions of blood vessels, non-linear one-dimensional models would
be intractable. However, using linear wave propagation theory the input impedance of the downstream vas-
cular bed can be computed for large complex vascular trees. A method to prescribe the impedance (calcu-
lated using linear wave theory) of these downstream vascular beds at the outlets of three-dimensional
models would enable the specification of realistic boundary conditions for three-dimensional simulations
of blood flow and pressure.
While inadequate outflow boundary conditions and rigid-wall models are the main impediments to the
realistic prediction of pressure in three-dimensional blood flow simulations, in this paper we focus on the
first issue and do not address the issue of wall deformability in the three-dimensional domain. We describe a
new method to prescribe outflow boundary conditions in the context of the finite element method as this
method is particularly well suited for handling complex geometries and boundary conditions inherent in
modeling blood flow [24]. The approach we describe is based on the Dirichlet-to-Neumann (DtN) [36]
and the variational multiscale [37] methods and is an extension of the 1D coupled multidomain approach
we successfully applied with a variety of models of the downstream domain [18]. For one-dimensional prob-
lems, we demonstrated that a DtN map can be calculated for the impedance of complex vascular trees and
that this approach incorporates naturally occurring wave reflections from a downstream bed. Wave prop-
agation in transient and periodic states was simulated and the importance of selecting appropriate bound-
ary conditions was demonstrated for one-dimensional simulations of blood flow. We also noted that the
3778 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
best boundary condition for cardiovascular applications is not the one that exhibits no wave reflection,
since wave reflections naturally arising from downstream beds (from bifurcations, tapering, and variations
in wall properties) should propagate back upstream into the numerical domain. We concluded that, at pres-
ent, impedance-based boundary conditions are the best approach for incorporating natural sites of wave
reflection in the downstream vasculature.
In this paper we present a coupled multidomain approach for three-dimensional finite element simula-
tions of blood flow and pressure. Given the computational expense of three-dimensional numerical meth-
ods and the resolution limits imposed by current imaging technologies, we constrain the three-dimensional
domain to the major arteries, and model the downstream domains with simpler models (Fig. 1). The outlet
boundary conditions are implemented implicitly resulting in good stability and convergence properties at
physiologic pressures. The organization of the paper will begin with the description of our coupled multi-
domain method in three-dimensions and its specialization to resistance and impedance boundary condi-
tions. We then demonstrate this new method on a straight, cylindrical blood vessel, a bifurcation model
with a stenosis on one side, and a subject-specific model of the human abdominal aorta.
2. Methods
2.1. Governing equations (strong form)
The method described can be applied to conservative as well as advective formulations of the incom-
pressible Navier–Stokes equations, and was successfully implemented in both cases. We proceed by defining
the spatial domain as X and its boundary as C. The three-dimensional equations for the flow of an incom-
pressible Newtonian fluid consist of the three momentum balance equations and the continuity equation
(written here in advective form) subject to suitable initial and boundary conditions
Fig. 1. Illustration of the ‘‘Coupled Multidomain Method’’, where information from the downstream (analytical) domain is
incorporated in the upstream (computational) domain through a boundary map. This approach accommodates a variety of
downstream models ranging from pure resistive to complex electrical analogs and impedance models.
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3779
q~v
;t
þq~v Á r~v ¼ Àrp þdivðs
$
Þ þ
~
f
divð~vÞ ¼ 0
s
$
¼ 2lD
$
with D
$
¼
1
2
ðr~v þr~v
T
Þ.
ð1Þ
The primary variables are the fluid velocity ~v ¼ ðv
x
; v
y
; v
z
Þ and the pressure p. The density of the fluid is gi-
ven by q (assumed constant), the external force by
~
f , and the viscosity by l (assumed constant).
In regards to boundary conditions, the boundary C of the spatial domain can be split into a Dirichlet
partition C
g
and a Neumann partition C
h
such that (C = oX = C
g
[ C
h
; C
g
\ C
h
= ;). Considering this,
the velocity at the inlet(s) of the domain is typically specified as
~vð~x; tÞ ¼~v
in
ð~x; tÞ ~x 2 C
in
& C
g
. ð2Þ
A no-slip boundary condition is imposed on the walls (surfaces that are neither inlets nor outlets). We dis-
cuss outflow boundary conditions subsequently. The initial conditions for this problem are given by
~vð~x; 0Þ ¼~v
0
ð~xÞ ~x 2 X. ð3Þ
2.2. Weak form
We define the trial solution and weighting function spaces for the semi-discrete formulation as
S¼ f~vj~vðÁ; tÞ 2 H
1
ðXÞ
n
sd
; t 2 ½0; TŠ;~vðÁ; tÞ ¼~g on C
g
g;
W ¼ f~wj~wðÁ; tÞ 2 H
1
ðXÞ
n
sd
; t 2 ½0; TŠ; ~wðÁ; tÞ ¼
~
0 on C
g
g;
P¼ fpjpðÁ; tÞ 2 H
1
ðXÞ; t 2 ½0; TŠg;
ð4Þ
where H
1
represents the usual Sobolev space of functions with square-integrable values and first derivatives
in X; n
sd
represents the number of spatial dimensions and ~g represents the prescribed Dirichlet boundary
condition.
The weak form reads: find ~v 2 S and p 2 P such that for every ~w 2 W and q 2 P
B
G
ð~w; q;~v; pÞ ¼ 0;
B
G
ð~w; q;~v; pÞ ¼
_
X
~w Á ðq~v
;t
þq~v Á r~v À
~
f Þ þr~w : ðÀp I
$
þs
$
Þ
_ _
d~x
À
_
X
rq Á ~vd~x À
_
C
h
~w Á ðÀp I
$
þs
$
Þ Á ~nds þ
_
C
q~v Á ~nds.
ð5Þ
2.3. Coupled multidomain method
We adopt the following approach to derive appropriate outflow boundary conditions [18,37]. First, we
divide the spatial domain X into an upstream ‘‘numerical’’ domain
´
X, and a downstream ‘‘analytic’’ do-
main X
0
such that
´
X \ X
0
¼ ; and
´
X [ X
0
¼ X. The boundary that separates these domains is defined
as C
B
(Fig. 2). We define a disjoint decomposition of the variables, for example for the unknown solution
vector, Uð~x; tÞ ¼ ½~v; pŠ
T
, ð~x; tÞ 2 X ½0; TŠ,
U ¼
´
U þU
0
with
´
Uj
X
0 ¼ 0 and U
0
j
^
X
¼ 0 ð6Þ
3780 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
so that
U ¼
´
Uj~x 2
´
X; U
0
j~x 2 X
0
and
´
U ¼ U
0
~x 2 C
B
_ _
. ð7Þ
Note also that
~n
0
¼ À
^
~n. ð8Þ
We use a similar decomposition for the weighting functions, and insert these expressions into the previous
variational form:
_
X
ð
^
~w þ~w
0
Þ Á qð
^
~v
;t
þ~v
0
;t
Þ þqð
^
~v þ~v
0
Þ Á rð
^
~v þ~v
0
Þ À
~
f
_ _
þðr
^
~w þr~w
0
Þ : Àð^p þp
0
Þ I
$
þð^s þs
0
Þ
_ _
d~x
À
_
C
h
ð
^
~w þ~w
0
Þ Á Àð^p þp
0
Þ I
$
þð^s
$
þ s
$
0
Þ
_ _
Á ~nds À
_
X
ðr^q þrq
0
Þ Á ð
^
~v þ~v
0
Þ d~x
þ
_
C
ð^q þq
0
Þð
^
~v þ~v
0
Þ Á ~nds ¼ 0. ð9Þ
The disjoint nature of this expression is used to derive a new variational form for the numerical domain.
Specifically, since
´
U vanishes on X
0
and U
0
vanishes on
´
X (and similarly for the weighting functions),
we find:
Fig. 2. The spatial domain is divided between the upstream numerical domain
´
X (main model) and the downstream analytic domain
´
X ¼ [
i
X
0
i
(here the trees), separated by the coupling boundary C
B
¼ [
i
C
Bi
(here in green).
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3781
_
^
X
^
~w Á ðq
^
~v
;t
þq
^
~v Á r
^
~v À
~
f Þ þr
^
~w : ðÀ^p I
$
þ^s
$
Þ d~x À
_
^
C
h
^
~w Á ðÀ^p I
$
þ^s
$
Þ Á
^
~nds
þ
_
X
0
~w
0
Á ðq~v
0
;t
þq~v
0
Á r~v
0
À
~
f Þ þr~w
0
: ðÀp
0
I
$
þs
$
0
Þ d~x À
_
C
0
h
~w
0
Á ðÀp
0
I
$
þs
$
0
Þ Á ~n
0
ds À
_
^
X
r^q Á
^
~vd~x
þ
_
^
C
^q
^
~v Á
^
~nds À
_
X
0
rq
0
Á ~v
0
d~x þ
_
C
0
q
0
~v
0
Á
~
n
0
ds ¼ 0. ð10Þ
Now since (1) holds in X
0
, we obtain the original variational form specialized to the numerical domain
´
X
with the addition of boundary terms accounting for the interface to the analytic domain, X
0
_
^
X
^
~w Á ðq
^
~v
;t
þq
^
~v Á r
^
~v À
~
f Þ þr
^
~w : ðÀ^p I
$
þ^s
$
Þ d~x À
_
^
C
h
^
~w Á ðÀ^p I
$
þ^s
$
Þ Á
^
~nds
þ
_
C
B
~w
0
Á ðÀp
0
I
$
þs
$
0
Þ Á ~n
0
ds À
_
^
X
r^q Á
^
~vd~x þ
_
^
C
^q
^
~v Á
^
~nds À
_
C
B
q
0
~v
0
Á ~n
0
ds ¼ 0. ð11Þ
Here we introduce an approximation to be able to quantify the terms above that come from the down-
stream domain:
ðÀp
0
I
$
þs
0
$
Þj
C
B
% ½M
$
m
ð~v
0
; p
0
Þ þH
$
m
Š
C
B
;
~v
0
j
C
B
% ½
~
M
c
ð~v
0
; p
0
Þ þ
~
H
c
Š
C
B
.
ð12Þ
The operators M ¼ ½M
$
m
;
~
M
c
Š
C
B
and H ¼ ½H
$
m
;
~
H
c
Š
C
B
are defined on the X
0
domain based on the chosen mod-
el to represent the downstream domain (different examples are provided subsequently).
Then using Eqs. (7) and (8), we enforce the continuity of the momentum and mass fluxes across C
B
_
C
B
^
~w Á ðM
$
m
ð
^
~v; ^pÞ þH
$
m
Þ Á
^
~nds ¼ À
_
C
B
~w
0
Á ðM
$
m
ð~v
0
; p
0
Þ þH
$
m
Þ Á ~n
0
ds;
_
C
B
^qð
~
M
c
ð
^
~v; ^pÞ þ
~
H
c
Þ Á
^
~nds ¼ À
_
C
B
q
0
ð
~
M
c
ð~v
0
; p
0
Þ þ
~
H
c
Þ Á ~n
0
ds.
ð13Þ
We thus approximate the boundary terms on C
B
in the weak form (11) by
_
C
B
~w
0
Á ðÀp
0
I
$
þs
$
0
Þ Á ~n
0
ds % À
_
C
B
^
~w Á ðM
$
m
ð
^
~v; ^pÞ þH
$
m
Þ Á
^
~nds;
_
C
B
q
0
~v
0
Á ~n
0
ds % À
_
C
B
^qð
~
M
c
ð
^
~v; ^pÞ þ
~
H
c
Þ Á
^
~nds.
ð14Þ
The resulting weak form for the multidomain method is then
_
^
X
^
~w Á ðq
^
~v
;t
þq
^
~v Á r
^
~v À
~
f Þ þr
^
~w : ðÀ^p I
$
þ^s
$
Þ d~x À
_
^
C
h
^
~w Á ðÀ^p I
$
þ^s
$
Þ Á
^
~nds;
À
_
C
B
^
~w Á ðM
$
m
ð
^
~v; ^pÞ þH
$
m
Þ Á
^
~nds À
_
^
X
r^q Á
^
~vd~x þ
_
^
C
^q
^
~v Á
^
~nds þ
_
C
B
^qð
~
M
c
ð
^
~v; ^pÞ þ
~
H
c
Þ Á
^
~nds ¼ 0.
ð15Þ
We see that the solution in the numerical domain depends on operators defined by the mathematical model
chosen to represent the physics of the downstream domain (see the terms in boxes) but not the solution
variable, ~v
0
and p
0
in the downstream domain.
3782 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
2.4. Derivation of the operators depending on the model chosen to represent the physics
in the downstream domain
In selecting a mathematical model of the downstream domain X
0
, we have the goal of obtaining a rep-
resentation formula whereby the solution on the boundary C
B
can be obtained as a function of the problem
on the interior of the analytic domain and the outlet(s) of the analytic domain. Fundamentally, this repre-
sentation formula will depend on the approximations we introduce in treating the downstream domain. We
consider two possibilities, namely zero-dimensional models and one-dimensional wave propagation theory.
In those cases, we will have a relationship between mean pressure and flow rate at C
B
, but this method can
incorporate more general relationships between the primary variables or functions of them—namely, full
traction relationships instead of just the normal component.
2.4.1. Resistance boundary condition
We first present the results for the simplest zero-dimensional model for the downstream domain, the
resistance. The general concept of resistance is to define a constant relationship between mean pressure
and flow rate P = QR [38], consistent with Poiseuille flow where the flow is fully developed (gradients in
the axial direction are zero). We make the assumption that pressure in the downstream domain is constant
over the cross-sectional area of the inlet boundary C
B
so that
p
0
À~n
0
Á s
$
0
Á ~n
0
¼ p
0
À0 ¼
_
C
B
p
0
ds
_
C
B
ds
¼ ÀR
_
C
B
~v
0
Á ~n
0
ds. ð16Þ
Therefore, the operators M and H are defined as
½M
$
m
ð~v
0
; p
0
Þ þH
$
m
Š
C
B
¼ R
_
C
B
~v
0
Á ~n
0
ds I
$
þs
$
0
À~n
0
Á s
$
0
Á ~n
0
I
$
_ _¸
¸
¸
¸
C
B
;
½
~
M
c
ð~v
0
; p
0
Þ þ
~
H
c
Š
C
B
¼~v
0
j
C
B
.
ð17Þ
Note that the sign of the resistance part of the operators changes if the face is an inlet or an outlet. Here, C
B
is an inlet for X
0
and an outlet for
´
X. In this simple resistance case, the first box representing a term on the
coupling interface C
B
in (15) is modified as follows:
_
C
B
^
~w Á ðM
$
m
ð
^
~v; ^pÞ þH
$
m
Þ Á
^
~nds ¼ À
_
C
B
^
~w Á
^
~n R
_
C
B
^
~v Á
^
~nds þ
^
~n Á ^s
$
Á
^
~n
_ _
ds þ
_
C
B
^
~w Á ^s
$
Á
^
~nds. ð18Þ
2.4.2. Impedance boundary condition
An example of a memory map is the impedance model, which is analogous to the concept of impedance
in electricity. In this paper the downstream domain is approximated using WomersleyÕs linear wave theory
[13,14] in an asymmetric fractal tree to derive a one-dimensional impedance Z(t) assuming periodicity in
time. Note that in this theory ~n Á s
$
Á~n is negligible. We can then derive for an inlet face:
p
0
À~n
0
Á s
$
0
Á ~n
0
% p
0
À0 ¼
_
C
B
p
0
ds
_
C
B
ds
¼ À
1
T
_
t
tÀT
Zðt Àt
1
Þq
0
ðt
1
Þ dt
1
¼ À
1
T
_
t
tÀT
Zðt Àt
1
Þ
_
C
B
~v
0
ðt
1
Þ Á ~n
0
ds dt
1
. ð19Þ
The flow rate at time t depends on the history of the pressure over one period. The representation formula
for the operators is obtained as
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3783
½M
$
m
ð~v
0
; p
0
Þ þH
$
m
Š
C
B
¼
1
T
_
t
tÀT
Zðt Àt
1
Þ
_
C
B
~v
0
ðt
1
Þ Á ~n
0
ds dt
1
I
$
þs
$
0
À~n
0
Á s
$
0
Á ~n
0
I
$
_ _¸
¸
¸
¸
C
B
;
½
~
M
c
ð~v
0
; p
0
Þ þ
~
H
c
Š
C
B
¼~v
0
j
C
B
.
ð20Þ
As before, note that the sign of the impedance part of the operators changes if the face is an inlet or an
outlet. Here, C
B
is an inlet for X
0
and an outlet for
^
X. For this memory case, the first box on the coupling
interface C
B
in (15) is modified as follows:
_
C
B
^
~w Á ðM
$
m
ð
^
~v; ^pÞ þH
$
m
Þ Á
^
~nds ¼ À
_
C
B
^
~w Á
^
~n
1
T
_
t
tÀT
Zðt Àt
1
Þ
_
C
B
^
~vðt
1
Þ Á
^
~nds dt
1
þ
^
~n Á ^s
$
Á
^
~n
_ _
ds
þ
_
C
B
^
~w Á ^s
$
Á
^
~nds. ð21Þ
Remark 1. Pressure and velocities on the boundary become unknown solution variables. Thus this method
enforces an implicitly coupled boundary condition.
Remark 2. The method described above can also be used when the impedance is derived from a (zero-
dimensional) lumped parameter model.
Remark 3. The components of the natural boundary integrals over C
h
in (5) that are not explicitly
described above in (18) and (21) are computed directly from the current solution
_
C
B
^
~w Á ^s
$
Á
^
~n À
^
~w Á
^
~nð
^
~n Á ^s
$
Á
^
~nÞ ds. ð22Þ
These so-called ‘‘consistent boundary conditions’’ are not in fact boundary conditions since they are di-
rectly computed from the solution variables. For a description of this as well as many alternative outflow
boundary conditions see Gresho and Sani [39]. In particular, the magnitude of the tangential tractions
which arise in Eqs. (18) and (21) are very small relative to other terms and could be neglected with minimal
effect on accuracy and stability.
Remark 4. For simplicity, we omit the ^ superscript from the field variables and the corresponding spaces
in the subsequent sections.
2.5. Finite element discretization
We employ a stabilized semi-discrete finite element method, based on ideas developed in Brooks and
Hughes [40], Franca and Frey [41], Taylor et al. [24]. The discrete trial solution and weighting function
spaces for the semi-discrete formulation are given by
S
k
h
¼ f~vj~vðÁ; tÞ 2 H
1
ðXÞ
n
sd
; t 2 ½0; TŠ;~vj
~x2X
e
2 P
k
ðX
e
Þ
n
sd
;~vðÁ; tÞ ¼
^
~g on C
g
g;
W
k
h
¼ f~wj~wðÁ; tÞ 2 H
1
ðXÞ
n
sd
; t 2 ½0; TŠ; ~wj
x2X
e
2 P
k
ðX
e
Þ
n
sd
; ~wðÁ; tÞ ¼
~
0 on C
g
g;
P
k
h
¼ fpjpðÁ; tÞ 2 H
1
ðXÞ; t 2 ½0; TŠ; pj
x2X
e
2 P
k
ð

X
e
Þg.
ð23Þ
Considering this, the weak form becomes: find ~m 2 S
k
h
and p 2 P
k
h
such that for every ~w 2 W
k
h
and
q 2 P
k
h
3784 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
B
G
ð~w; q;~v; pÞ ¼ 0;
B
G
ð~w; q;~v; pÞ ¼
_
X
f~w Á ðq~v
;t
þq~v Á r~v À
~
f Þ þr~w : ðÀp I
$
þs
$
Þgd~x À
_
X
rq Á ~vd~x
À
_
C
h
~wðÀp I
$
þs
$
Þ~nds þ
_
CÀC
B
q~v Á ~nds
À
_
C
B
~wðM
$
m
ð~v; pÞ þH
$
m
Þ Á ~ndC þ
_
C
B

~
M
c
ð~v; pÞ þ
~
H
c
Þ Á ~ndC.
ð24Þ
In this equation, the superscripts h, k of the discrete approximation of the continuous variables have been
omitted for simplicity. Note that the boundary terms are identified in the box. In order to stabilize the
Galerkin method, the formulation becomes [24,42]:
Find ~v 2 S
k
h
and p 2 P
k
h
such that for every ~w 2 W
k
h
and q 2 P
k
h
Bð~w; q;~v; pÞ ¼ 0;
Bð~w; q;~v; pÞ ¼ B
G
ð~w; q;~v; pÞ þ

n
el
e¼1
_
X
e
fð~v Á rÞ~w Á s
M
~
Lð~v; pÞ þrÁ ~ws
C
rÁ ~vgd~x
þ

n
el
e¼1
_
X
e
f~w Á ðq~v
D
Ár~vÞ þð
~
Lð~v; pÞ Á r~wÞ Á ðs
~
Lð~v; pÞ Á r~vÞgd~x
þ

n
el
e¼1
_
X
e
rq Á
s
M
q
~
Lð~v; pÞ d~x;
ð25Þ
~
Lð~v; pÞ represents the residual vector of the momentum equation:
~
Lð~v; pÞ ¼ q~v
;t
þq~v Á r~v þrp ÀrÁ s
$
À
~
f . ð26Þ
And ~v
D
is a conservation-restoring advective velocity whose expression is given by
~v
D
¼ À
s
M
q
~
Lð~v; pÞ. ð27Þ
The stabilization parameters are defined as follows:
s
M
¼
1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð2c
1
=DtÞ
2
þ~v Á g
$
~v þc
2
m
2
ðg
$
: g
$
þx
2
Þ
_ ;
s
C
¼
q
8s
M
trðg
$
Þ
C
s
C
c
1
and s ¼
s
M
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
~
Lð~v; pÞ Á g
$
Á
~
Lð~v; pÞ
_ ;
ð28Þ
where c
1
and c
2
are constants defined from the one-dimensional scalar model problem of the advection–dif-
fusion equation. x is the frequency of rotation of the reference frame, C
s
C
is a scale factor for s
C
, and g
$
is
the covariant metric tensor
g
$
¼ ð
~
n
;~x
Þ
T
~
n
;~x
. ð29Þ
These non-linear equation (25) are then solved using methods described in [42,43]. The ‘‘coupled multi-
domain method’’ was implemented for different boundary conditions in both the advective and the conserva-
tive forms, explicitly (in the residual only) and implicitly (in both the tangent matrix and the residual). We
observed good numerical convergence and stability properties. In addition, we verified that the relationships
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3785
we imposed as boundary conditions were numerically satisfied and checked for mass conservation between
the inlet and outlets. Subsequent to the verification studies, we solved a series of problems to demonstrate
that realistic pressure fields could be computed and to compare and contrast impedance, resistance and con-
stant pressure boundary conditions. This is described fully in the next section.
3. Results
In the examples below, the profile of the inlet velocity was chosen to be parabolic for subsequent com-
parisons with one-dimensional analysis simulations [18]. For a prescribed input flow, we do not expect that
the pressure fields will be significantly different between time-varying parabolic and Womersley inlet bound-
ary conditions. Furthermore, while the focus of this paper is on the outlet boundary conditions, the method
described could be applied in a similar fashion for inlet boundary conditions to include, for example, a
lumped model of the heart.
We first consider pulsatile flow in a straight vessel with a prescribed flow rate at the inlet mapped into a
parabolic velocity profile and different boundary conditions at the outlet, to specifically observe the influ-
ence of the outlet boundary condition on the solution in the numerical domain. The inlet flow wave is a
representative carotid flow rate [38,44,45]. The nominal radius of 0.3 cm and the vessel length of 12.6 cm
(length to diameter ratio of 20) were chosen to correspond approximately to that of the human common
carotid artery. The solution was computed on a 45,849 element and 9878 node mesh with a time step of
0.002 s, for a total of five cardiac cycles. The pressure pulse in the carotid artery has been reported to be
approximately 70–80/110–120 mmHg for a healthy person [38,46]. The resistance value is the ratio of mean
pressure to mean flow. The impedance is generated following OlufsenÕs approach described in the introduc-
tion: a fractal tree is generated starting from a vessel that matched the diameter of the outlet and diminished
in size with each successive generation of vessels until a fixed terminal vessel size was attained. The length-
to-radius ratio of the tree is chosen so that the zero-frequency of the impedance matches the previous resis-
tance boundary condition. The mean outlet pressure of the previous simulations is chosen to impose the
constant pressure outlet boundary condition. Fig. 3 depicts the results obtained with constant pressure,
resistance and impedance outflow boundary conditions. Flow is displayed for reference, since the incom-
pressibility constraint forces the flow to be instantaneously the same at the outlet as that prescribed at
the inlet. The constant pressure outlet boundary condition results in a pressure pulse that does not have
the correct amplitude or phase (pressure lags flow in reality). Note that this is the most common outlet
boundary condition used for three-dimensional analyses of blood flow. The resistance boundary condition
results in an unrealistically large pressure pulse, and forces pressure and flow to be in phase at the outlet.
However, the resistance boundary condition is an improvement over the constant pressure outlet boundary
condition when studying wave propagation, and does not require any knowledge of the pressure or flow
wave forms before the simulation. Finally, Fig. 3 depicts the results for the impedance boundary condition.
In this case, the pressure wave is propagated down the length of the vessel with little damping or dispersion.
In this case, pressure lags flow and the pressure amplitude and wave form are similar to curves given by
Nichols and OÕRourke [38]. Therefore, in regards to pressure curves, we conclude that the impedance
boundary condition is the most realistic boundary condition examined.
To further illustrate the critical influence of boundary conditions, we performed numerical simulations in
an idealized model of an abdominal aorta bifurcation to the iliac arteries with a 75% area reduction stenosis
on one side, using realistic anatomic dimensions (see Fig. 4a) and inflow wave form for an abdominal aortic
bifurcation. The finite element mesh consists of 94,280 elements and 19,579 nodes. A pulsatile, parabolic
profile was imposed at the inlet, and constant pressure (P = 90 mmHg) or equal impedance boundary con-
ditions (with a resistance of 24,000 dynÁsÁcm
À5
) at the two outlets. The simulation was run for 10 cardiac
cycles, with a CFL number of 1 in the stenosis and thus a time step size of 0.0008 s, with three non-linear
3786 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
Fig. 3. Comparison of flow and pressure wave forms in a carotid artery with a periodic inlet flow and constant pressure, resistance and
impedance outlet boundary conditions. For the pressure boundary condition, inlet pressure varies little from the prescribed constant
outlet pressure and the peak pressure precedes the peak flow. The resistance boundary condition gives rise to an unrealistically large
pressure amplitude. In addition, pressure and flow are in phase at the outlet. For the impedance boundary condition, the range of
pressure is from approximately 85–115 mmHg and the pressure lags flow along the length of the vessel. 1 mmHg = 133.3 Pa,
1 cc/s = 10
À6
m
3
/s.
Fig. 4. (a) Geometric model and (b) flow distribution between normal iliac artery and an iliac artery with a stenosis that reduces the
cross-sectional area by 75% for constant pressure and impedance boundary conditions. For a constant outlet pressure, the flow split is
dictated solely by the resistance to flow due to the geometry of the computational domain (approximately 70% to the normal side, 30%
to the stenosed side). For an impedance boundary condition with a physiologic level of resistance, the flow split (50%/50%) is
principally determined by the downstream demands (represented by equal outlet impedances). These results are consistent with clinical
observations for resting flow conditions and iliac artery stenoses with less than an 85% area reduction.
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3787
iterations per time step. Fig. 4b depicts the dramatic differences in flow split depending on outlet boundary
conditions. With a constant outlet pressure, the flow split is dictated solely by the resistance to flow due to
the geometry of the computational domain (70% flow to the normal side, 30% to the stenosed side). With an
impedance boundary condition, the flow split (50%/50%) is primarily determined by the downstream con-
ditions (represented by equal outlet impedances), which is representative of most clinical cases (i.e. for ste-
noses with less than an 85% area reduction). Note that because the impedance boundary condition
incorporates the history of flow and pressure over a cardiac cycle, it takes longer for the solution to become
periodic (about 6 cycles) than that observed with a constant outlet pressure (within 2 cycles).
The time varying flow and pressure curves at the inlet and outlets for the constant pressure and imped-
ance boundary conditions are shown in Fig. 5. With the constant pressure boundary condition, blood flow
Fig. 5. Inlet and outlet flow and pressure curves during the last two cardiac cycles of the simulation for constant pressure (90 mmHg)
and impedance outlet boundary conditions. Note that for the constant pressure boundary condition, the flow to the normal iliac artery
exceeds that through the stenosed iliac artery for all time points in the cardiac cycle and the inlet pressure varies from approximately
88–96 mmHg. For the impedance outlet boundary condition, the peak flow is greater in the normal iliac artery than the stenosed iliac
artery, yet during late systole and early diastole, the flow is greater in the stenosed iliac artery than the normal iliac artery. Note that for
the impedance outlet boundary condition, the peak pressure for the normal iliac artery is close to that of the inlet pressure, whereas the
peak pressure for the stenosed iliac artery is significantly less than that in the normal iliac artery. Also note that the inlet pressure varies
between approximately 88 and 106 mmHg. 1 mmHg = 133.3 Pa, 1 cc/s = 10
À6
m
3
/s.
3788 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
from the aorta to the normal iliac artery is greater than that to the stenosed iliac artery throughout the car-
diac cycle, following the 70/30 mean flow split noted above. The inlet pressure amplitude for the pressure
boundary condition is not representative of physiologic conditions: it is solely determined by the instanta-
neous pressure gradient from the inlet to the prescribed constant outlets that is needed to drive the flow to
both outlets, and therefore is only a slight variation from the mean value. For the impedance boundary
condition we observe that while the mean flow split is nearly 50/50, the normal iliac artery gets more flow
at peak systole (when the effect of the 75% area stenosis is greatest) and less flow during diastole. We also
note that for a brief period in diastole, the flow in the stenosed iliac artery is actually greater than the inlet
flow, indicating flow is actually drawn from the normal iliac artery—coinciding with the brief period of re-
verse flow in the normal iliac artery. This latter observation is quite interesting and perhaps due to the fact
that during the deceleration phase in late systole, blood is more easily decelerated in the normal iliac artery
than would be expected for the stenosed iliac artery. Again, note that this phenomenon is not observed
with constant outlet pressure boundary conditions. For the impedance boundary conditions, the inlet, nor-
mal iliac and right iliac pressure wave forms exhibit physiologic amplitudes and phase. Note that the peak
systolic pressure at the outlet of the stenosed iliac artery is considerably less than that observed in the nor-
mal iliac artery and that all the pressure waveforms exhibit little difference in diastole when the flow rate is
low.
Fig. 6 provides a dramatic illustration of the differences in the velocity fields between the constant pres-
sure and impedance outlet boundary conditions. Velocity magnitude at peak systole and early diastole is
shown along the symmetry plane of the model. The velocity fields for the two different outlet boundary con-
ditions exhibit substantial differences. We note, in particular, that the peak velocity through the stenosis is
much higher for the simulation performed with the impedance outlet boundary condition as compared to
that performed with the constant pressure outlet boundary condition. This is consistent with the significant
differences in flow through the stenosed iliac artery for the two boundary condition cases examined. Fig. 7
depicts the instantaneous pressure fields for the constant pressure and impedance outlet boundary condi-
tions at peak systole and early diastole. Note the fact that the scales are different for the two boundary con-
ditions and the two time points in the cardiac cycle examined.
Finally, we consider pulsatile flow in a model of the abdominal aorta of a normal subject with geome-
try and flow rates generated from magnetic resonance imaging data [47]. The measured volumetric flow rate
at the level of the diaphragm was specified at the inlet boundary and impedances are specified at all the
outlets. The exit impedances were chosen so that the mean flow rate in each of the branches corresponded
to the measured flow distribution when such data was available and literature data as needed [48,49]. Fig. 8
shows the 3D model. A total of 800,151 elements and 167,850 nodes were utilized and the solution was
computed using a time step of 0.001 s for a total of five cardiac cycles. Fig. 8 depicts the volume flow rate
and pressure at the inlet and representative outlets of the subject-specific abdominal aorta model when
impedance boundary conditions are applied. Blood flows into the different arteries according to the pre-
scribed impedances representing the demands of the downstream trees. Previous simulations of pulsatile
flow in the human abdominal aorta [50] were performed by imposing outflow velocity profiles using
WomersleyÕs theory of pulsatile flow in rigid vessels to match the prescribed volumetric flow rates. For
the branch vessels feeding the viscera (celiac, superior mesenteric, inferior mesenteric and renal arteries),
these flow rates were scaled to some fraction of the inlet flow waveform and reverse flow in the iliac arter-
ies during diastole arose because more flow was pulled from the other branch vessels than was input to
the abdominal aorta. Note that for the present simulations the flow split between the branch vessels
occurs naturally. Finally, to illustrate the results that can be obtained from coupling three-dimensional sim-
ulations of blood flow to impedance boundary conditions, the mean wall shear stress for the subject-specific
model is presented in Fig. 9. Clearly, mean wall shear stress will be highly affected by the outlet boundary
conditions.
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3789
Fig. 6. Velocity magnitude along the symmetry plane of the model at peak systole and early-diastole for the constant pressure and
impedance outlet boundary conditions.
3790 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
Fig. 7. Pressure at peak systole and early diastole for the constant pressure and impedance outlet boundary conditions.
1 mmHg = 133.3 Pa.
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3791
Fig. 8. Pressure and flow wave forms in a subject-specific model of the human abdominal aorta with a measured periodic inlet flow rate
and impedance outlet boundary conditions. 1 mmHg = 133.3 Pa, 1 cc/s = 10
À6
m
3
/s.
3792 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
4. Conclusions
We have successfully developed and implemented a method to prescribe outflow boundary conditions
intended for three-dimensional finite element simulations of blood flow based on the Dirichlet-to-Neumann
and variational multiscale methods. As long as the effect of the downstream domain can be represented by
an explicit function of pressure as a function of flow rate or velocity, the methods described can be used to
couple the upstream three-dimensional numerical domain with the downstream analytic domain. The
numerical results presented demonstrate that physiologic values of pressure can be attained and illustrate
the importance of using appropriate boundary conditions. A long straight cylindrical blood vessel is used to
demonstrate that realistic blood pressures can be calculated using this new method and that the choice of
the outlet boundary conditions has a significant effect on the computed pressure field. The next example, an
otherwise symmetric model of an abdominal aortic bifurcation with an obstruction in one iliac, demon-
strates that the choice of outlet boundary conditions can dramatically affect the flow distribution, velocity
and pressure fields. The final example, a patient specific abdominal aorta model, demonstrates that this ap-
proach can be used for complex multi-branched systems.
Results have been shown for idealized and patient specific 3D geometries, using physiological values,
coupled to more sophisticated downstream models that incorporate memory-effects like the impedance
boundary condition. Lumped parameter models more complex than a simple resistance could be easily
incorporated in a manner similar to that used for the impedance boundary condition.
Fig. 9. Mean wall shear stress over one cardiac cycle for the subject-specific abdominal aorta model depicted in Fig. 8.
1 dyn/cm
2
= 0.1 Pa.
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3793
The stability of the solutions proved to be sensitive to the number of frequencies used to generate the
impedance boundary condition, due to the convolution with the flow rate history. Although these results
are very encouraging, for a more realistic representation of flow and pressure further analysis of the gen-
eration and validation of the impedance functions is needed. The impedance functions were generated using
a binary fractal tree that did not incorporate differences between the branching patterns in different organs.
Alternate models based on physical measurements of branching patterns would aid in the specification of
realistic impedance functions. A further limitation of the impedance boundary condition is that it is derived
based on the assumption of periodicity. An alternate outflow boundary condition that incorporates tran-
sient, non-periodic flows or a coupling to the non-linear one-dimensional equations of blood flow may be
needed for non-periodic three-dimensional flow problems.
It is important to note that the present implementation does not incorporate deformable vessels in the
three-dimensional domain and, as such, does not model wave propagation from the three-dimensional to
one-dimensional domains. However, the analyses performed clearly indicate the need for appropriate
boundary conditions in computing pulse pressures. Realistic values of blood pressure (mean and pulse) will
be essential for obtaining accurate loading and displacement in fluid-solid interaction methods. Finally, the
method described may be extensible to other biological and non-biological flow phenomena where the
downstream model can be described as a function that links the primary variables or their derived quan-
tities at the coupling interface.
Acknowledgements
This material is based upon work supported by the National Science Foundation under Grant No.
0205741 and a predoctoral fellowship from the American Heart Association. The authors gratefully
acknowledge Dr. Mette Olufsen for the use of her methods to compute input impedance of vascular trees
and Dr. Farzin Shakib for his linear algebra package (http://www.acusim.com). Finally, the authors grate-
fully acknowledge the assistance of Dr. Nathan Wilson for assistance with software development, Dr. Mary
Draney for acquiring the subject-specific abdominal aorta MRI data, and Hyun Jin Kim for constructing
the abdominal aorta model.
References
[1] C.A. Taylor, M.T. Draney, Experimental and computational methods in cardiovascular fluid mechanics, Ann. Rev. Fluid Mech.
36 (2004) 197–231.
[2] A. Kamiya, T. Togawa, Adaptive regulation of wall shear stress to flow change in the canine carotid artery, Am. J. Physiol. 239 (1)
(1980) H14–H21.
[3] H. Wolinsky, S. Glagov, Comparison of abdominal and thoracic aortic medial structure in mammals. Deviation of man from the
usual pattern, Circulation Res. 25 (6) (1969) 677–686.
[4] S. Glagov, C.K. Zarins, D.P. Giddens, D.N. Ku, Hemodynamics and atherosclerosis. Insights and perspectives gained from
studies of human arteries, Arch. Pathol. Lab. Med. 112 (10) (1988) 1018–1031.
[5] C.K. Zarins, C.A. Taylor, Hemodynamic Factors in Atherosclerosis. Vascular surgery: A Comprehensive Review, W.S. Moore
and W.B.S. Company, 2002, pp. 105–118.
[6] C.A. Taylor, M.T. Draney, J.P. Ku, D. Parker, B.N. Steele, K. Wang, C.K. Zarins, Predictive medicine: computational techniques
in therapeutic decision-making, Comput. Aided Surgery 4 (5) (1999) 231–247.
[7] National Heart, Lung and Blood Institute report of the task force on research in pediatric cardiovascular disease, US Department
of Health and Human Services, Public Health Service, National Institutes of Health, 2002.
[8] N. Stergiopulos, D.F. Young, T.R. Rogge, Computer simulation of arterial flow with applications to arterial and aortic stenoses,
J. Biomech. 25 (12) (1992) 1477–1488.
[9] L. Formaggia, F. Nobile, A. Quarteroni, A. Veneziani, Multiscale modelling of the circulatory system: A preliminary analysis,
Comput. Visualization Sci. 2 (2/3) (1999) 75–83.
3794 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796
[10] J. Wan, B.N. Steele, S.A. Spicer, S. Strohband, G.R. Feijoo, T.J.R. Hughes, C.A. Taylor, A one-dimensional finite element
method for simulation-based medical planning for cardiovascular disease, Comput. Methods Biomech. Biomed. Engrg. 5 (3)
(2002) 195–206.
[11] W.H. Ruan, M.E. Clark, M.D. Zhao, A. Curcio, A hyperbolic system of equations of blood flow in an arterial network, SIAM J.
Appl. Math. 64 (2) (2003) 637–667.
[12] S.J. Sherwin, L. Formaggia, J. Peiro, V. Franke, Computational modelling of 1D blood flow with variable mechanical properties
and its application to the simulation of wave propagation in the human arterial system, Int. J. Numer. Methods Fluids 43 (6–7)
(2003) 673–700.
[13] J.R. Womersley, Oscillatory motion of a viscous liquid in a thin-walled elastic tube—I: the linear approximation for long waves,
The Philos. Mag. 7 (1955) 199–221.
[14] J.R. Womersley, Oscillatory flow in arteries: the constrained elastic tube as a model of arterial flow and pulse transmission, Phys.
Med. Biol. 2 (1957) 178–187.
[15] M.G. Taylor, Wave transmission through an assembly of randomly branching elastic tubes, Biophys. J. 6 (1966) 697–716.
[16] M.S. Olufsen, Structured tree outflow condition for blood flow in larger systemic arteries, Am. J. Physiol. 276 (1999) H257–H268.
[17] B.N. Steele, C.A. Taylor, Simulation of blood flow in the abdominal aorta at rest and during exercise using a 1-D finite element
method with impedance boundary conditions derived from a fractal tree, in: Proceedings of the 2003 ASME Summer
Bioengineering Meeting, Key Biscayne, FL, 2003.
[18] I. Vignon, C.A. Taylor, Outflow boundary conditions for one-dimensional finite element modeling of blood flow and pressure
waves in arteries, Wave Motion 39 (4) (2004) 361–374.
[19] J.R. Cebral, M. Castro, O. Soto, R. Lohner, P.J. Yim, N. Alperin, Finite element modeling of the circle of Willis from magnetic
resonance data, Proceedings of SPIE—The International Society for Optical Engineering 5031 (2003) 11–21.
[20] M. Oshima, R. Torii, T. Kobayashi, N. Taniguchi, K. Takagi, Finite element simulation of blood flow in the cerebral artery,
Comput. Methods Appl. Mech. Engrg. 191 (6–7) (2001) 661–671.
[21] K. Perktold, G. Rappitsch, Computer simulation of local blood flow and vessel mechanics in a compliant carotid artery
bifurcation model, J. Biomech. 28 (7) (1995) 845–856.
[22] J. Stroud, S. Berger, D. Saloner, Numerical analysis of flow through a severely stenotic carotid artery bifurcation, J. Biomech.
Engrg. 124 (1) (2002) 9–20.
[23] G.R. Stuhne, D.A. Steinman, Finite-element modeling of the hemodynamics of stented aneurysms, J. Biomech. Engrg. 126 (3)
(2004) 382–387.
[24] C.A. Taylor, T.J.R. Hughes, C.K. Zarins, Finite element modeling of blood flow in arteries, Comput. Methods Appl. Mech.
Engrg. 158 (1998) 155–196.
[25] G. Pennati, F. Migliavacca, G. Dubini, R. Pietrabissa, M. de Leval, A mathematical model of circulation in the presence of the
bidirectional cavopulmonary anastomosis in children with a univentricular heart, Med. Eng. Phys. 19 (3) (1997) 223–234.
[26] F. Migliavacca, M.R. de Leval, G. Dubini, R. Pietrabissa, R. Fumero, Computational fluid dynamic simulations of
cavopulmonary connections with an extracardiac lateral conduit, Med. Engrg. Phys. 21 (3) (1999) 187–193.
[27] F. Migliavacca, P.J. Kilner, G. Pennati, G. Dubini, R. Pietrabissa, R. Fumero, M.R. de Leval, Computational fluid dynamic and
magnetic resonance analyses of flow distribution between the lungs after total cavopulmonary connection, IEEE Trans. Biomed.
Engrg. 46 (4) (1999) 393–399.
[28] E.B. Shim, R.D. Kamm, T. Heldt, R.G. Mark, Numerical analysis of blood flow through a stenosed artery using a coupled
multiscale simulation method, Comput. Cardiol. (2000) 219–222.
[29] A. Quarteroni, S. Ragni, A. Veneziani, Coupling between lumped and distributed models for blood flow problems, Comput.
Visualization Sci. 4 (2) (2001) 111–124.
[30] G. Guadagni, E.L. Bove, F. Migliavacca, G. Dubini, Effects of pulmonary afterload on the hemodynamics after the hemi-Fontan
procedure, Med. Engrg. Phys. 23 (5) (2001) 293–298.
[31] K. Lagana, G. Dubini, F. Migliavacca, R. Pietrabissa, G. Pennati, A. Veneziani, A. Quarteroni, Multiscale modelling as a tool to
prescribe realistic boundary conditions for the study of surgical procedures, Biorheology 39 (3–4) (2002) 359–364.
[32] A. Quarteroni, A. Veneziani, Analysis of a geometrical multiscale model based on the coupling of ODEs and PDEs for blood flow
simulations, Multiscale Model. Simul. 1 (2) (2003) 173–195.
[33] K. Lagana, R. Balossino, F. Migliavacca, G. Pennati, E.L. Bove, M.R. de Leval, G. Dubini, Multiscale modeling of the
cardiovascular system: application to the study of pulmonary and coronary perfusions in the univentricular circulation, J.
Biomech. 38 (5) (2005) 1129–1141.
[34] L. Formaggia, J.F. Gerbeau, F. Nobile, A. Quarteroni, On the coupling of 3D and 1D Navier–Stokes equations for flow problems
in compliant vessels, Comput. Methods Appl. Mech. Engrg. 191 (2001) 561–582.
[35] L. Formaggia, J. Gerbeau, F. Nobile, A. Quarteroni, Numerical treatment of defective boundary conditions for the Navier–Stokes
equations, SIAM J. Numer. Anal. 40 (1) (2002) 376–401.
[36] D. Givoli, J.B. Keller, A finite element method for large domains, Comput. Methods Appl. Mech. Engrg. 76 (1) (1989)
41–66.
I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796 3795
[37] T.J.R. Hughes, Multiscale phenomena: GreenÕs functions, the Dirichlet-to-Neumann formulation, subgrid scale models, bubbles
and the origins of stabilized methods, Comput. Methods Appl. Mech. Engrg. 127 (1995) 387–401.
[38] W.W. Nichols, M.F. OÕRourke, McDonaldÕs Blood Flow in Arteries: Theoretical, Experimental and Clinical Principles, Oxford
University Press, New York, 1998.
[39] P.M. Gresho, R.L. Sani, Incompressible Flow and the Finite Element Method, Wiley, 2000.
[40] A.N. Brooks, T.J.R. Hughes, Streamline upwind Petrov–Galerkin formulation for convection dominated flows with particular
emphasis on the incompressible Navier–Stokes equations, Comput. Methods Appl. Mech. Engrg. 32 (1982) 199–259.
[41] L.P. Franca, S.L. Frey, Stabilized finite element methods II. The incompressible Navier–Stokes equations, Comput. Methods
Appl. Mech. Engrg. 99 (2–3) (1992) 209–233.
[42] C.H. Whiting, K.E. Jansen, A stabilized finite element method for the incompressible Navier–Stokes equations using a
hierarchical basis, Int. J. Numer. Methods Fluids 35 (2001) 93–116.
[43] K.E. Jansen, C.H. Whiting, G.M. Hulbert, Generalized-{alpha} method for integrating the filtered Navier–Stokes equations with
a stabilized finite element method, Comput. Methods Appl. Mech. Engrg. 190 (3-4) (2000) 305–319.
[44] J.R. Cebral, R. Lohner, O. Soto, P.J. Yim, On the modelling of carotid blood flow from magnetic resonance images, in:
Proceedings of the 2001 Summer Bioengineering Meeting, 2001.
[45] D.W. Holdsworth, C.J.D. Norley, R. Frayne, D.A. Steinman, B.K. Rutt, Characterization of common carotid artery blood-flow
waveforms in normal subjects, Physiol. Meas. 20 (1999) 219–240.
[46] S.Z. Zhao, X.Y. Xu, A.D. Hughes, S.A. Thom, A.V. Stanton, B. Ariff, Q. Long, Blood flow and vessel mechanics in a
physiologically realistic model of a human carotid arterial bifurcation, J. Biomech. 33 (8) (2000) 975–984.
[47] B.T. Tang, C.P. Cheng, P.S. Tsao and C.A. Taylor, Subject-specific finite element modeling of three-dimensional pulsatile flow in
the human abdominal aorta: comparison of resting and exercise conditions, in: Proceedings of the 2003 Summer Bioengineering
Meeting, Key Biscayne, FL, 2003.
[48] W.F.M. Ganong, Review of Medical Physiology, Appleton & Lange, Englewood Cliffs, 1995.
[49] A.C. Guyton, Physiology of the Human Body, Saunders College Publishing, San Francisco, 1984.
[50] C.A. Taylor, T.J.R. Hughes, C.K. Zarins, Finite element modeling of three-dimensional pulsatile flow in the abdominal aorta:
relevance to atherosclerosis, Ann. Biomed. Engrg. 26 (6) (1998) 1–14.
3796 I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796

I.E. Vignon-Clementel et al. / Comput. Methods Appl. Mech. Engrg. 195 (2006) 3776–3796

3777

Flow rate and pressure distributions are shown for different boundary conditions to illustrate the dramatic influence of alternative boundary conditions on these quantities. Ó 2005 Elsevier B.V. All rights reserved.
Keywords: Blood flow; Pressure; Boundary conditions; Coupled multidomain method; Finite elements

1. Introduction In normal and disease states in children and adults, quantification of three-dimensional flow phenomena and pressure fields is important to understand the response of the cardiovascular system to biomechanical forces [1]. In the development of the circulatory system in normal subjects and congenital heart disease patients, flow modulates diameter and pressure controls wall thickness [2,3]. In older subjects, the initiation and progression of atherosclerotic and aneurismal disease is directly affected by the complex three-dimensional fluid mechanical environment of the major arteries [4,5]. Moreover, the ability to adequately simulate flow and pressure is needed to model the performance of devices such as heart valves, LVADs (left ventricular assist devices), filters, stents and stent-grafts. Non-invasive three-dimensional flow and pressure data can also provide important information to determine the significance of an obstruction or predict the outcome of a procedure [6,7]. In recent years, remarkable progress has been made in simulating blood flow in realistic anatomical models constructed from three-dimensional medical imaging data. Arguably, accurate anatomic models are of primary importance in simulating blood flow. However, as we demonstrate in this paper, realistic boundary conditions are equally important in computing velocity and pressure fields. Yet, this subject has received far less attention than image-based model construction for three-dimensional simulations. In contrast, significant progress has been made in devising outflow boundary conditions for solving the one-dimensional equations of blood flow in elastic vessels. For example, Stergiopulos solved the non-linear one-dimensional equations of blood flow in a comprehensive model of the arterial system using a lumped parameter model of the vasculature downstream of each branch in his numerical model [8]. Several groups have developed and analyzed the coupling of one-dimensional equations with lumped models [9–12]. In contrast to methods coupling the one-dimensional equations of blood flow to lumped parameter models, Olufsen developed a distributed downstream model based on calculating the input impedance of an asymmetric binary fractal tree using WomersleyÕs linear wave theory [13,14] and an algorithm for computing the impedance of a vascular network first proposed by Taylor [15]. OlufsenÕs approach was to generate a fractal tree for each outlet starting from a vessel that matched the diameter of the outlet and diminished in size with each successive generation of vessels until a fixed terminal vessel size was attained. With this method an impedance for each outlet of the upstream numerical model was computed naturally from linear wave theory and branching laws. OlufsenÕs distributed model of the downstream vasculature enabled the representation of more realistic flow and pressure waveforms than those obtained with lumped parameter models [16]. Steele and Taylor used a modified version of OlufsenÕs impedance boundary condition to model blood flow at rest and during simulated exercise conditions [17]. In this case, vascular networks were assigned to the outlets of a model of the abdominal aorta, modified to represent the resting flow distribution of 11 different subjects and then dilated to simulate the effects of lower limb exercise. Vignon and Taylor developed a multidomain approach to couple one-dimensional equations to different lumped and onedimensional boundary conditions [18]. While these one-dimensional methods can be used to compute flow rate and mean pressure, by design, they cannot be used to simulate complex three-dimensional flow phenomena and pressure losses. Three-dimensional numerical methods have been used to compute velocity fields and quantify shear forces acting on the surface of blood vessels. However, since most three-dimensional models of blood flow use zero or constant pressure, zero traction, or prescribed velocity profiles

A further limitation of these methods to couple three-dimensional and zero-dimensional models is the fact that there is no direct relationship between the anatomy of the downstream vascular bed and the lumped parameters resulting in difficulties in specifying these parameters and relating them to subsequent physiologic or pathophysiologic changes in the downstream vasculature. / Comput. While inadequate outflow boundary conditions and rigid-wall models are the main impediments to the realistic prediction of pressure in three-dimensional blood flow simulations. Engrg. capillaries. in this paper we focus on the first issue and do not address the issue of wall deformability in the three-dimensional domain. anatomic information is available for vessels downstream of the primary region of interest. The approach we describe is based on the Dirichlet-to-Neumann (DtN) [36] and the variational multiscale [37] methods and is an extension of the 1D coupled multidomain approach we successfully applied with a variety of models of the downstream domain [18].3778 I. We also noted that the . While great progress was made in these landmark papers. a simpler approach is to directly represent the vasculature of the small arteries and arterioles using zero-dimensional or one-dimensional models. but this coupling has been performed iteratively.E. The incorporation of such data would improve the accuracy of the models of the downstream vasculature. for many simulations based on three-dimensional imaging data.25–33] have successfully coupled threedimensional models to either resistances or more sophisticated zero-dimensional models (lumped models). Based on our experience. we demonstrated that a DtN map can be calculated for the impedance of complex vascular trees and that this approach incorporates naturally occurring wave reflections from a downstream bed. venules and veins returning blood to the heart. For one-dimensional problems. neglecting the dominant effect of the resistance of the downstream vascular beds. We describe a new method to prescribe outflow boundary conditions in the context of the finite element method as this method is particularly well suited for handling complex geometries and boundary conditions inherent in modeling blood flow [24]. However. These models can be terminated at the level of the capillary vessels where an assumption of constant pressure is reasonable. using linear wave propagation theory the input impedance of the downstream vascular bed can be computed for large complex vascular trees. For simulations of blood flow in large arteries. Mech. While closed-loop models are optimal. If zero or equal pressures or tractions are used for different outlets. especially for models with multiple outlets. the flow split will be dictated solely by the resistance to flow in the branches of the domain of interest. non-linear one-dimensional models would be intractable. these papers did not include coupling between three-dimensional domains and complex vascular networks as have been incorporated in onedimensional numerical solutions of blood flow in arteries. the outlet boundary conditions represent the downstream vasculature including smaller arteries. blood pressure is not computed accurately and notably absent from reports of hemodynamic investigations [19–24]. Clearly. yet ignoring the effect of the downstream circulation results in grossly inaccurate predictions of velocity and pressure fields for many problems where the distribution of flow between the major arteries is unknown. In addition. implicit coupling significantly improves convergence. the vast extent and complexity of the circulation precludes a three-dimensional representation of the entire circuit. Methods to couple three-dimensional and one-dimensional models were first described by Formaggia et al. and reducedorder models to represent the remainder of the system. [34. Vignon-Clementel et al. Since the vascular bed from the major arteries to the capillaries can include tens of millions of blood vessels. Several groups [6. Wave propagation in transient and periodic states was simulated and the importance of selecting appropriate boundary conditions was demonstrated for one-dimensional simulations of blood flow. the coupling was performed for simple geometries and iteratively. and generally applied to geometries with few outlets and low resistances (as seen in the pulmonary vasculature). 195 (2006) 3776–3796 as outlet boundary conditions. Methods Appl. An alternative approach is to utilize three-dimensional models for the major arteries where high-fidelity information is needed. arterioles. In addition.35]. A method to prescribe the impedance (calculated using linear wave theory) of these downstream vascular beds at the outlets of three-dimensional models would enable the specification of realistic boundary conditions for three-dimensional simulations of blood flow and pressure.

and was successfully implemented in both cases. Given the computational expense of three-dimensional numerical methods and the resolution limits imposed by current imaging technologies. The organization of the paper will begin with the description of our coupled multidomain method in three-dimensions and its specialization to resistance and impedance boundary conditions. a bifurcation model with a stenosis on one side. where information from the downstream (analytical) domain is incorporated in the upstream (computational) domain through a boundary map. and variations in wall properties) should propagate back upstream into the numerical domain. We proceed by defining the spatial domain as X and its boundary as C. We concluded that. at present. and a subject-specific model of the human abdominal aorta. 1). tapering. 1. This approach accommodates a variety of downstream models ranging from pure resistive to complex electrical analogs and impedance models. best boundary condition for cardiovascular applications is not the one that exhibits no wave reflection.1. Engrg. Methods 2. Methods Appl. 195 (2006) 3776–3796 3779 Fig. Vignon-Clementel et al. cylindrical blood vessel.I. The outlet boundary conditions are implemented implicitly resulting in good stability and convergence properties at physiologic pressures. and model the downstream domains with simpler models (Fig. Governing equations (strong form) The method described can be applied to conservative as well as advective formulations of the incompressible Navier–Stokes equations. 2. Illustration of the ‘‘Coupled Multidomain Method’’. In this paper we present a coupled multidomain approach for three-dimensional finite element simulations of blood flow and pressure. / Comput. since wave reflections naturally arising from downstream beds (from bifurcations. we constrain the three-dimensional domain to the major arteries. The three-dimensional equations for the flow of an incompressible Newtonian fluid consist of the three momentum balance equations and the continuity equation (written here in advective form) subject to suitable initial and boundary conditions .E. We then demonstrate this new method on a straight. Mech. impedance-based boundary conditions are the best approach for incorporating natural sites of wave reflection in the downstream vasculature.

Methods Appl. vz Þ and the pressure p. The density of the fluid is giv ven by q (assumed constant).t þ q~ Á r~ ¼ Àrp þ divðs Þ þ ~ v v v f $ divð~ ¼ 0 vÞ $ ð1Þ 1 v v with D ¼ ðr~ þ r~T Þ. First. x vð~ v x. tÞ ¼ ~in ð~ tÞ ~ 2 Cin & Cg . w v. Weak form We define the trial solution and weighting function spaces for the semi-discrete formulation as S ¼ f~ vðÁ. tÞ 2 H ðXÞ. t 2 ½0. g nsd 1 W ¼ f~j~ðÁ. ð~ tÞ 2 X  ½0. Engrg. Considering this.3. The initial conditions for this problem are given by ~ x.3780 I.~ pÞ ¼ 0. vð~ v xÞ x 2. and a downstream ‘‘analytic’’ do0 0 b b main X 0 such that X \ X ¼ . The boundary that separates these domains is defined as CB (Fig. vy . 0Þ ¼ ~0 ð~ ~ 2 X.~ tÞ ¼ ~ on Cg g.~ pÞ ¼ w v. / Comput.2.t þ q~ Á r~ À ~Þ þ r~ : ðÀp I þ sÞ d~ BG ð~. the velocity at the inlet(s) of the domain is typically specified as ~ x. T Š. Mech. $ 2 s ¼ 2l D $ The primary variables are the fluid velocity ~ ¼ ðvx . Vignon-Clementel et al. we b divide the spatial domain X into an upstream ‘‘numerical’’ domain X. vj~ vðÁ. and X [ X ¼ X. x. Cg \ Ch = . q. ww w 0 P ¼ fpjpðÁ. f In regards to boundary conditions. The weak form reads: find ~ 2 S and p 2 P such that for every ~ 2 W and q 2 P v w BG ð~. where H1 represents the usual Sobolev space of functions with square-integrable values and first derivatives in X. tÞ 2 H ðXÞ . We discuss outflow boundary conditions subsequently. the external force by ~. b U ¼ U þ U0 b with U jX0 ¼ 0 and U 0 jX ¼ 0 ^ ð6Þ .E. T Š. T Šg.37]. Uð~ tÞ ¼ ½~ pŠ . ~ðÁ. v. q. x. 195 (2006) 3776–3796 q~. ð2Þ A no-slip boundary condition is imposed on the walls (surfaces that are neither inlets nor outlets). T Š. We define a disjoint decomposition of the variables. À rq Á~d~ À v x n v n w X Ch $ $ C 1 n ð3Þ ð4Þ ð5Þ 2. t 2 ½0. and the viscosity by l (assumed constant). nsd represents the number of spatial dimensions and ~ represents the prescribed Dirichlet boundary g condition. for example for the unknown solution T vector.). 2). the boundary C of the spatial domain can be split into a Dirichlet partition Cg and a Neumann partition Ch such that (C = oX = Cg [ Ch. tÞ ¼ ~ on Cg g. v v f w x w v $ $ X Z Z Z ~ Á ðÀp I þ sÞ Á ~ds þ q~ Á ~ds. Z n o ~ Á ðq~. tÞ 2 H 1 ðXÞ sd . t 2 ½0. Coupled multidomain method We adopt the following approach to derive appropriate outflow boundary conditions [18.

Methods Appl. / Comput. since U vanishes on X 0 and U 0 vanishes on X (and similarly for the weighting functions).I. C The disjoint nature of this expression is used to derive a new variational form for the numerical domain. separated by the coupling boundary CB ¼ [i CBi (here in green). 195 (2006) 3776–3796 3781 b Fig.E. b b Specifically.t Þ þ qð~ þ~0 Þ Á rð~ þ~0 Þ À ~ þ ðr~ þ r~0 Þ : Àð^ þ p0 Þ I þð^ þ s0 Þ d~ w w v v v v v f w w p s x $ X Z Z   ^ ^ À ð~ þ ~0 Þ Á Àð^ þ p0 Þ I þð^ þ s0 Þ Á ~ds À ðr^ þ rq0 Þ Á ð~ þ~0 Þ d~ w w p s n q v v x $ $ $ Ch X Z ^ q v v n ð9Þ þ ð^ þ q0 Þð~ þ~0 Þ Á ~ds ¼ 0. Engrg. Vignon-Clementel et al. 2.t þ~0. ð8Þ We use a similar decomposition for the weighting functions. we find: . so that n b x b U ¼ U j~ 2 X. The spatial domain is divided between the upstream numerical domain X (main model) and the downstream analytic domain b X ¼ [i X0i (here the trees). and insert these expressions into the previous variational form: Z     ^ v ^ ^ ^ ^ ð~ þ ~0 Þ Á qð~. Mech. x ð7Þ Note also that ^ ~0 ¼ À~ n n. U 0 j~ 2 X0 x and o b U ¼ U 0 ~ 2 CB .

H c ŠCB are defined on the X 0 domain based on the chosen mod$ $ el to represent the downstream domain (different examples are provided subsequently). v $ $ $ $ ~ jCB v 0 ~ v ~ % ½M c ð~0 . Mech. p0 Þ þ H m ŠCB . p0 Þ þ H c ŠCB . Methods Appl. we obtain the original variational form specialized to the numerical domain X 0 with the addition of boundary terms accounting for the interface to the analytic domain. Then using Eqs. q ~ v. p CB The resulting weak form for the multidomain method is then Z Z ^ ^ v f ^ ^ ^ ^ ^ ~ Á ðq~. p n w Z CB 0 0 0 $ $ q ~ Á ~ ds % À v n Z CB CB $ $ ð14Þ ^ ^ ~ n ^ðM c ð~ ^Þ þ H c Þ Á ~ds. p0 Þ þ H c Þ Á ~0 ds. p Z CB CB $ $ ð13Þ ~ v ~ q0 ðM c ð~0 .t þ q~ Á r~ À ~Þ þ r~ : ðÀ^ I þ^Þ d~ À ^ ^ ^ ~ ~ Á ðÀ^ I þ^Þ Á ~ds w v w p v w p s x s n ^ X $ $ þ Z CB ~0 Á ðÀp0 I þs0 Þ Á ~0 ds À n w $ $ Z ^ X ^ r^ Á ~d~ þ q v x Z ^ C ^ Ch $ $ ^ ^ ^~ Á ~ds À qv n Z CB q0~0 Á ~0 ds ¼ 0. ð12Þ ~ ~ The operators M ¼ ½M m . M c ŠCB and H ¼ ½H m .E. p0 Þ þ H m Þ Á ~0 ds. ~0 and p 0 in the downstream domain.t þ q~0 Á r~0 À ~Þ þ r~0 : ðÀp0 I þs 0 Þ d~ À ~0 Á ðÀp0 I þs 0 Þ Á ~0 ds À r^ Á ~d~ v v v f w x n q v x w w þ $ $ $ $ ^ X0 C0 X h Z Z Z ^ ^ rq0 Á~0 d~ þ v x q0~0 Á n0 ds ¼ 0. Engrg. v . n v n w Z CB $ $ ^ ^ ~ n ^ðM c ð~ ^Þ þ H c Þ Á ~ds ¼ À q ~ v. v n ð11Þ Here we introduce an approximation to be able to quantify the terms above that come from the downstream domain: ðÀp0 I þ s0 ÞjCB % ½M m ð~0 . w v w p v w p s x s n ^ X $ $ À Z CB ^ ^ ^ ~ Á ðM m ð~ ^Þ þ H m Þ Á ~ds À w v. q ~ v.t þ q~ Á r~ À ~Þ þ r~ : ðÀ^ I þ^Þ d~ À ^ ^ ^ ~ ~ Á ðÀ^ I þ^Þ Á ~ds w v w p v w p s x s n $ $ $ $ ^ ^ Ch X Z Z Z ^ ~0 Á ðq~0. p n $ $ Z ^ X ^ Ch $ $ ^ r^ Á ~d~ þ q v x Z ^ ^ ^~ Á ~ds þ qv n ^ C Z CB ^ ^ ~ n ^ðM c ð~ ^Þ þ H c Þ Á ~ds ¼ 0. w v. X Z Z ^ ^ v f ^ ^ Á ðq~. we enforce the continuity of the momentum and mass fluxes across CB Z Z ^p ^ Á ðM m ð~ ^Þ þ H m Þ Á ~ds ¼ À ^ ~ ~0 Á ðM m ð~0 . p ð15Þ We see that the solution in the numerical domain depends on operators defined by the mathematical model chosen to represent the physics of the downstream domain (see the terms in boxes) but not the solution variable. (7) and (8).3782 I. 195 (2006) 3776–3796 Z Z ^ ^ v f ^ ^ Á ðq~. ð10Þ qv n v ~ þ ^~ Á ~ds À ^ C X0 C0 b Now since (1) holds in X 0 . n CB We thus approximate the boundary terms on CB in the weak form (11) by Z Z ^ ^ ^ ~ Á ðM m ð~ ^Þ þ H m Þ Á ~ds. Vignon-Clementel et al.t þ q~ Á r~ À ~Þ þ r~ : ðÀ^ I þ^Þ d~ À ~ Á ðÀ^ I þ^Þ Á ~ds. / Comput. ~0 Á ðÀp0 I þs0 Þ Á ~0 ds % À w n v.

4.E. we have the goal of obtaining a representation formula whereby the solution on the boundary CB can be obtained as a function of the problem on the interior of the analytic domain and the outlet(s) of the analytic domain. full traction relationships instead of just the normal component. w w w s n v n v. the resistance. we will have a relationship between mean pressure and flow rate at CB. Derivation of the operators depending on the model chosen to represent the physics in the downstream domain In selecting a mathematical model of the downstream domain X 0 . The general concept of resistance is to define a constant relationship between mean pressure and flow rate P = QR [38].I. namely zero-dimensional models and one-dimensional wave propagation theory. We consider two possibilities.4. n ð18Þ CB CB $ $ CB CB $ CB $ Therefore. the operators M and H are defined as  Z . Vignon-Clementel et al. v n CB ð16Þ ð17Þ CB ~ c ð~0 . p0 Þ þ H c Š ¼ ~0 jC . 195 (2006) 3776–3796 3783 2. this representation formula will depend on the approximations we introduce in treating the downstream domain. Methods Appl. 2. We make the assumption that pressure in the downstream domain is constant over the cross-sectional area of the inlet boundary CB so that Z p0 ds ds CB n n p À~ Á s Á~ ¼ p À 0 ¼ Z $ 0 0 0 0 0 CB ¼ ÀR Z ~0 Á ~0 ds. CB b is an inlet for X 0 and an outlet for X. / Comput. but this method can incorporate more general relationships between the primary variables or functions of them—namely. consistent with Poiseuille flow where the flow is fully developed (gradients in the axial direction are zero). ~ ½M v v B CB Note that the sign of the resistance part of the operators changes if the face is an inlet or an outlet. Fundamentally. the first box representing a term on the coupling interface CB in (15) is modified as follows:  Z  Z Z Z ^ Á ~ds þ ~ Á ^ Á ~ ds þ ^p ^ Á ðM m ð~ ^Þ þ H m Þ Á ~ds ¼ À ^ Á~ R ^ ^ ^ ^ s n ^ ^ ^ ~ n ~ ~ n ~ Á ^ Á ~ds. Here. In this simple resistance case. Mech.1. Engrg. In those cases. Resistance boundary condition We first present the results for the simplest zero-dimensional model for the downstream domain.

.

0 0 0 0 0 0 0 0 ~ Á ~ ds I þs À ~ Á s Á ~ I .

p Þ þ H m ŠCB ¼ R v n . v n n ½M m ð~ . .

Note that in this theory ~ Á s Á~ is negligible. In this paper the downstream domain is approximated using WomersleyÕs linear wave theory [13. ¼À Zðt À t1 Þ n ð19Þ v T tÀT CB The flow rate at time t depends on the history of the pressure over one period. which is analogous to the concept of impedance in electricity. Impedance boundary condition An example of a memory map is the impedance model.14] in an asymmetric fractal tree to derive a one-dimensional impedance Z(t) assuming periodicity in time. $ $ $ $ $ $ 2.4.2. We can then derive for an inlet face: n n Z $ p0 ds Z 1 t CB 0 0 0 0 0 p À~ Á s Á~ % p À 0 ¼ Z n n Zðt À t1 Þq0 ðt1 Þ dt1 ¼À $ T tÀT ds CB Z Z 1 t ~0 ðt1 Þ Á ~0 ds dt1 . The representation formula for the operators is obtained as .

Engrg.3784 0 0 I. Methods Appl.E. p Þ þ H m ŠCB v $ $ 0 0  Z t . Mech. Vignon-Clementel et al. 195 (2006) 3776–3796 ½M m ð~ . / Comput.

Z .

1 0 0 0 0 0 0 ~ ðt1 Þ Á ~ ds dt1 I þs À ~ Á s Á ~ I .

. ¼ Zðt À t1 Þ n n n v $ $ $ $ .

the first box on the coupling interface CB in (15) is modified as follows:  Z t  Z Z Z ^ ^ ^ ^ ^ 1 ^ ^ ^ ^ ~ 1 Þ Á ~ds dt1 þ ~ Á ^ Á ~ ds ~ Á ðM m ð~ ^Þ þ H m Þ Á ~ds ¼ À ~ Á~ w w n vðt v. t 2 ½0. the magnitude of the tangential tractions which arise in Eqs.5. (18) and (21) are very small relative to other terms and could be neglected with minimal effect on accuracy and stability. we omit the ^ superscript from the field variables and the corresponding spaces in the subsequent sections. w s n w nðn s nÞ ð22Þ CB $ $ These so-called ‘‘consistent boundary conditions’’ are not in fact boundary conditions since they are directly computed from the solution variables. Remark 2. In particular. þ w s n ð21Þ CB $ Remark 1. Remark 4. T tÀT CB CB 0 ð20Þ ~ v ~ ½M c ð~ . tÞ ¼ ~ on Cg g. Thus this method enforces an implicitly coupled boundary condition. ~jx2Xe 2 P k ðXe Þ sd . tÞ 2 H 1 ðXÞ sd . The components of the natural boundary integrals over Ch in (5) that are not explicitly described above in (18) and (21) are computed directly from the current solution Z ^ ^ ^ ^^ ^ ~ Á ^ Á ~ À ~ Á ~ ~ Á ^ Á ~ ds. The discrete trial solution and weighting function spaces for the semi-discrete formulation are given by ^ Sk ¼ f~ vðÁ. t 2 ½0. t 2 ½0. p n Zðt À t1 Þ n n s n $ $ $ T tÀT CB CB CB Z ^ ^ ~ Á ^ Á ~ds.~ ~ e 2 P k ðXe Þ sd . pjx2Xe 2 P k ðXe Þg. Finite element discretization We employ a stabilized semi-discrete finite element method. The method described above can also be used when the impedance is derived from a (zerodimensional) lumped parameter model. tÞ 2 H 1 ðXÞ. tÞ 2 H 1 ðXÞ sd . Franca and Frey [41]. h w Considering this. T Š.~ tÞ ¼ ~ on Cg g. v As before. [24]. ww w w 0 h ð23Þ  Pk ¼ fpjpðÁ. For this memory case. Pressure and velocities on the boundary become unknown solution variables. g h n n n n W k ¼ f~j~ðÁ. For simplicity. vj~ vjx2X vðÁ. Remark 3. T Š. 2. T Š. For a description of this as well as many alternative outflow boundary conditions see Gresho and Sani [39]. p Þ þ H c ŠCB ¼ ~ jCB . Taylor et al. CB is an inlet for X 0 and an outlet for X. note that the sign of the impedance part of the operators changes if the face is an inlet or an ^ outlet. ~ðÁ. the weak form becomes: find ~ 2 Sk and p 2 Pk such that for every ~ 2 Wk and m h h h q 2 Pk h . based on ideas developed in Brooks and Hughes [40]. Here.

Vignon-Clementel et al. Lð~ pÞ represents the residual vector of the momentum equation: ~ v. Methods Appl.I. In this equation. vÞg x rq Á sM ~ Lð~ pÞ d~ v. ~ n qðM c ð~ pÞ þ H c Þ Á ~dC .t þ q~ Á r~ þ rp À r Á s À~. x is the frequency of rotation of the reference frame. q Z Xe Xe ~ v.42]: Find ~ 2 Sk and p 2 Pk such that for every ~ 2 W k and q 2 Pk v w h h h h Bð~. and g is $ the covariant metric tensor nx nx g ¼ ð~. Bð~. we verified that the relationships .43].~. $ ð29Þ These non-linear equation (25) are then solved using methods described in [42. w v. q ð27Þ The stabilization parameters are defined as follows: 1 sM ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi . Note that the boundary terms are identified in the box. C sC is a scale factor for sC. We observed good numerical convergence and stability properties. 195 (2006) 3776–3796 3785 BG ð~. w v. the formulation becomes [24.~ pÞ ¼ f~ Á ðq~. Z Z ~Þ þ r~ : ðÀp I þ s Þg d~ À rq Á~d~ BG ð~. $ D ð26Þ And ~ is a conservation-restoring advective velocity whose expression is given by v D ~¼ À v sM ~ Lð~ pÞ. Mech. The ‘‘coupled multidomain method’’ was implemented for different boundary conditions in both the advective and the conservative forms. 2 v v ð2c1 =DtÞ þ~ Á g ~ þ c2 m2 ðg : g þx2 Þ $ $ $ sC ¼ q C sC 8sM trðgÞ c1 $ sM and  ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi . n w CB $ $ Z CB ~ v. / Comput.~ pÞ ¼ BG ð~. x.~ pÞ ¼ 0. explicitly (in the residual only) and implicitly (in both the tangent matrix and the residual). q. Lð~ pÞ Á g ÁLð~ pÞ $ ð28Þ where c1 and c2 are constants defined from the one-dimensional scalar model problem of the advection–diffusion equation. f~ Á ðq~ Ár~ þ ðLð~ pÞ Á r~Þ Á ðLð~ pÞ Á r~ d~ w v vÞ w s ~ v. q.t þ q~ Á r~ À f w v. w v v v w x v x $ $ X X Z Z ~ðÀp I þ s Þ~ds þ À n q~ Á ~ds v n w À Z Ch $ $ CÀCB ð24Þ ~ðM m ð~ pÞ þ H m Þ Á ~dC þ v.E. In addition.~ pÞ þ w v. v. þ þ nel XZ e¼1 nel X e¼1 nel XZ e¼1 D Xe ~ v. Engrg. fð~ Á rÞ~ Á sM Lð~ pÞ þ r Á ~sC r Á~ d~ v w vg x w ð25Þ ~ v. ~ v. q.~ÞT~.~ pÞ ¼ 0. k of the discrete approximation of the continuous variables have been omitted for simplicity. q. w v. v v f v Lð~ pÞ ¼ q~. s ~ v. In order to stabilize the Galerkin method. the superscripts h. q.

195 (2006) 3776–3796 we imposed as boundary conditions were numerically satisfied and checked for mass conservation between the inlet and outlets. Furthermore. we do not expect that the pressure fields will be significantly different between time-varying parabolic and Womersley inlet boundary conditions.0008 s.45]. The resistance value is the ratio of mean pressure to mean flow. using realistic anatomic dimensions (see Fig. 3.6 cm (length to diameter ratio of 20) were chosen to correspond approximately to that of the human common carotid artery. Subsequent to the verification studies. The impedance is generated following OlufsenÕs approach described in the introduction: a fractal tree is generated starting from a vessel that matched the diameter of the outlet and diminished in size with each successive generation of vessels until a fixed terminal vessel size was attained. 4a) and inflow wave form for an abdominal aortic bifurcation. and forces pressure and flow to be in phase at the outlet. with three non-linear . Flow is displayed for reference. For a prescribed input flow. to specifically observe the influence of the outlet boundary condition on the solution in the numerical domain. and constant pressure (P = 90 mmHg) or equal impedance boundary conditions (with a resistance of 24. The lengthto-radius ratio of the tree is chosen so that the zero-frequency of the impedance matches the previous resistance boundary condition. for example. for a total of five cardiac cycles. Results In the examples below.44.3 cm and the vessel length of 12.E. a lumped model of the heart. A pulsatile. we conclude that the impedance boundary condition is the most realistic boundary condition examined. the resistance boundary condition is an improvement over the constant pressure outlet boundary condition when studying wave propagation.579 nodes. / Comput.3786 I. while the focus of this paper is on the outlet boundary conditions. 3 depicts the results for the impedance boundary condition. Fig. To further illustrate the critical influence of boundary conditions. Finally. and does not require any knowledge of the pressure or flow wave forms before the simulation. we performed numerical simulations in an idealized model of an abdominal aorta bifurcation to the iliac arteries with a 75% area reduction stenosis on one side. the method described could be applied in a similar fashion for inlet boundary conditions to include. The constant pressure outlet boundary condition results in a pressure pulse that does not have the correct amplitude or phase (pressure lags flow in reality). resistance and constant pressure boundary conditions.002 s. This is described fully in the next section. The mean outlet pressure of the previous simulations is chosen to impose the constant pressure outlet boundary condition. the profile of the inlet velocity was chosen to be parabolic for subsequent comparisons with one-dimensional analysis simulations [18]. Methods Appl. However. The resistance boundary condition results in an unrealistically large pressure pulse. in regards to pressure curves. since the incompressibility constraint forces the flow to be instantaneously the same at the outlet as that prescribed at the inlet. The finite element mesh consists of 94. with a CFL number of 1 in the stenosis and thus a time step size of 0. We first consider pulsatile flow in a straight vessel with a prescribed flow rate at the inlet mapped into a parabolic velocity profile and different boundary conditions at the outlet. The solution was computed on a 45. Vignon-Clementel et al. Mech.280 elements and 19. The pressure pulse in the carotid artery has been reported to be approximately 70–80/110–120 mmHg for a healthy person [38. pressure lags flow and the pressure amplitude and wave form are similar to curves given by Nichols and OÕRourke [38]. parabolic profile was imposed at the inlet. 3 depicts the results obtained with constant pressure.000 dynÁsÁcmÀ5) at the two outlets. we solved a series of problems to demonstrate that realistic pressure fields could be computed and to compare and contrast impedance. In this case. Engrg. Therefore. The inlet flow wave is a representative carotid flow rate [38. resistance and impedance outflow boundary conditions. The nominal radius of 0. the pressure wave is propagated down the length of the vessel with little damping or dispersion.849 element and 9878 node mesh with a time step of 0. Note that this is the most common outlet boundary condition used for three-dimensional analyses of blood flow.46]. The simulation was run for 10 cardiac cycles. In this case. Fig.

3 Pa. Methods Appl. inlet pressure varies little from the prescribed constant outlet pressure and the peak pressure precedes the peak flow. Vignon-Clementel et al. The resistance boundary condition gives rise to an unrealistically large pressure amplitude. . 4. 195 (2006) 3776–3796 3787 Fig. For the impedance boundary condition. Mech. For an impedance boundary condition with a physiologic level of resistance. 3. These results are consistent with clinical observations for resting flow conditions and iliac artery stenoses with less than an 85% area reduction. pressure and flow are in phase at the outlet. Engrg. Comparison of flow and pressure wave forms in a carotid artery with a periodic inlet flow and constant pressure.I. the flow split is dictated solely by the resistance to flow due to the geometry of the computational domain (approximately 70% to the normal side. Fig. / Comput. 1 cc/s = 10À6 m3/s. the range of pressure is from approximately 85–115 mmHg and the pressure lags flow along the length of the vessel. For a constant outlet pressure.E. (a) Geometric model and (b) flow distribution between normal iliac artery and an iliac artery with a stenosis that reduces the cross-sectional area by 75% for constant pressure and impedance boundary conditions. 30% to the stenosed side). For the pressure boundary condition. resistance and impedance outlet boundary conditions. the flow split (50%/50%) is principally determined by the downstream demands (represented by equal outlet impedances). 1 mmHg = 133. In addition.

the flow to the normal iliac artery exceeds that through the stenosed iliac artery for all time points in the cardiac cycle and the inlet pressure varies from approximately 88–96 mmHg. Methods Appl. 30% to the stenosed side). With the constant pressure boundary condition. the flow split is dictated solely by the resistance to flow due to the geometry of the computational domain (70% flow to the normal side. Also note that the inlet pressure varies between approximately 88 and 106 mmHg. With an impedance boundary condition. Note that for the constant pressure boundary condition. whereas the peak pressure for the stenosed iliac artery is significantly less than that in the normal iliac artery. Mech.3788 I. the peak pressure for the normal iliac artery is close to that of the inlet pressure. the flow split (50%/50%) is primarily determined by the downstream conditions (represented by equal outlet impedances). Vignon-Clementel et al. Engrg. With a constant outlet pressure. The time varying flow and pressure curves at the inlet and outlets for the constant pressure and impedance boundary conditions are shown in Fig. 5. / Comput.3 Pa.E. Inlet and outlet flow and pressure curves during the last two cardiac cycles of the simulation for constant pressure (90 mmHg) and impedance outlet boundary conditions. 1 mmHg = 133.e. yet during late systole and early diastole. for stenoses with less than an 85% area reduction). Note that for the impedance outlet boundary condition. the peak flow is greater in the normal iliac artery than the stenosed iliac artery. blood flow Fig. which is representative of most clinical cases (i. 1 cc/s = 10À6 m3/s. 4b depicts the dramatic differences in flow split depending on outlet boundary conditions. the flow is greater in the stenosed iliac artery than the normal iliac artery. it takes longer for the solution to become periodic (about 6 cycles) than that observed with a constant outlet pressure (within 2 cycles). Fig. 5. 195 (2006) 3776–3796 iterations per time step. Note that because the impedance boundary condition incorporates the history of flow and pressure over a cardiac cycle. For the impedance outlet boundary condition. .

Again. 9. these flow rates were scaled to some fraction of the inlet flow waveform and reverse flow in the iliac arteries during diastole arose because more flow was pulled from the other branch vessels than was input to the abdominal aorta. 6 provides a dramatic illustration of the differences in the velocity fields between the constant pressure and impedance outlet boundary conditions. / Comput. Finally. Vignon-Clementel et al. Note the fact that the scales are different for the two boundary conditions and the two time points in the cardiac cycle examined. This is consistent with the significant differences in flow through the stenosed iliac artery for the two boundary condition cases examined. The measured volumetric flow rate at the level of the diaphragm was specified at the inlet boundary and impedances are specified at all the outlets. indicating flow is actually drawn from the normal iliac artery—coinciding with the brief period of reverse flow in the normal iliac artery.E. note that this phenomenon is not observed with constant outlet pressure boundary conditions. mean wall shear stress will be highly affected by the outlet boundary conditions. Blood flows into the different arteries according to the prescribed impedances representing the demands of the downstream trees.001 s for a total of five cardiac cycles. Fig. We note. the flow in the stenosed iliac artery is actually greater than the inlet flow. the normal iliac artery gets more flow at peak systole (when the effect of the 75% area stenosis is greatest) and less flow during diastole. Previous simulations of pulsatile flow in the human abdominal aorta [50] were performed by imposing outflow velocity profiles using WomersleyÕs theory of pulsatile flow in rigid vessels to match the prescribed volumetric flow rates. The exit impedances were chosen so that the mean flow rate in each of the branches corresponded to the measured flow distribution when such data was available and literature data as needed [48. Methods Appl. 7 depicts the instantaneous pressure fields for the constant pressure and impedance outlet boundary conditions at peak systole and early diastole. that the peak velocity through the stenosis is much higher for the simulation performed with the impedance outlet boundary condition as compared to that performed with the constant pressure outlet boundary condition. Clearly. We also note that for a brief period in diastole.850 nodes were utilized and the solution was computed using a time step of 0. the mean wall shear stress for the subject-specific model is presented in Fig. inferior mesenteric and renal arteries). Velocity magnitude at peak systole and early diastole is shown along the symmetry plane of the model. 8 depicts the volume flow rate and pressure at the inlet and representative outlets of the subject-specific abdominal aorta model when impedance boundary conditions are applied. Fig. Fig. and therefore is only a slight variation from the mean value. Note that for the present simulations the flow split between the branch vessels occurs naturally. Fig.151 elements and 167. A total of 800. to illustrate the results that can be obtained from coupling three-dimensional simulations of blood flow to impedance boundary conditions. Note that the peak systolic pressure at the outlet of the stenosed iliac artery is considerably less than that observed in the normal iliac artery and that all the pressure waveforms exhibit little difference in diastole when the flow rate is low. Finally. in particular. we consider pulsatile flow in a model of the abdominal aorta of a normal subject with geometry and flow rates generated from magnetic resonance imaging data [47]. normal iliac and right iliac pressure wave forms exhibit physiologic amplitudes and phase. For the impedance boundary conditions. This latter observation is quite interesting and perhaps due to the fact that during the deceleration phase in late systole. 195 (2006) 3776–3796 3789 from the aorta to the normal iliac artery is greater than that to the stenosed iliac artery throughout the cardiac cycle. following the 70/30 mean flow split noted above. the inlet. . Engrg.I. Mech. superior mesenteric. The velocity fields for the two different outlet boundary conditions exhibit substantial differences.49]. For the branch vessels feeding the viscera (celiac. The inlet pressure amplitude for the pressure boundary condition is not representative of physiologic conditions: it is solely determined by the instantaneous pressure gradient from the inlet to the prescribed constant outlets that is needed to drive the flow to both outlets. blood is more easily decelerated in the normal iliac artery than would be expected for the stenosed iliac artery. For the impedance boundary condition we observe that while the mean flow split is nearly 50/50. 8 shows the 3D model.

Engrg. Mech. 6. Vignon-Clementel et al.E. Velocity magnitude along the symmetry plane of the model at peak systole and early-diastole for the constant pressure and impedance outlet boundary conditions.3790 I. / Comput. Methods Appl. . 195 (2006) 3776–3796 Fig.

Mech.3 Pa.I. . Pressure at peak systole and early diastole for the constant pressure and impedance outlet boundary conditions. / Comput.E. 1 mmHg = 133. Engrg. Vignon-Clementel et al. Methods Appl. 195 (2006) 3776–3796 3791 Fig. 7.

1 mmHg = 133. Vignon-Clementel et al. Engrg. 1 cc/s = 10À6 m3/s. 8. / Comput.3792 I. . Mech. Pressure and flow wave forms in a subject-specific model of the human abdominal aorta with a measured periodic inlet flow rate and impedance outlet boundary conditions.3 Pa. Methods Appl.E. 195 (2006) 3776–3796 Fig.

The next example. Mech.E.1 Pa.I. . demonstrates that this approach can be used for complex multi-branched systems. Methods Appl. As long as the effect of the downstream domain can be represented by an explicit function of pressure as a function of flow rate or velocity. the methods described can be used to couple the upstream three-dimensional numerical domain with the downstream analytic domain. 4. 9. a patient specific abdominal aorta model. Results have been shown for idealized and patient specific 3D geometries. an otherwise symmetric model of an abdominal aortic bifurcation with an obstruction in one iliac. 8. using physiological values. A long straight cylindrical blood vessel is used to demonstrate that realistic blood pressures can be calculated using this new method and that the choice of the outlet boundary conditions has a significant effect on the computed pressure field. Vignon-Clementel et al. The final example. Lumped parameter models more complex than a simple resistance could be easily incorporated in a manner similar to that used for the impedance boundary condition. The numerical results presented demonstrate that physiologic values of pressure can be attained and illustrate the importance of using appropriate boundary conditions. Conclusions We have successfully developed and implemented a method to prescribe outflow boundary conditions intended for three-dimensional finite element simulations of blood flow based on the Dirichlet-to-Neumann and variational multiscale methods. 195 (2006) 3776–3796 3793 Fig. velocity and pressure fields. Engrg. Mean wall shear stress over one cardiac cycle for the subject-specific abdominal aorta model depicted in Fig. 1 dyn/cm2 = 0. demonstrates that the choice of outlet boundary conditions can dramatically affect the flow distribution. coupled to more sophisticated downstream models that incorporate memory-effects like the impedance boundary condition. / Comput.

D. Arch. Hemodynamics and atherosclerosis. Wang.P. Steele. does not model wave propagation from the three-dimensional to one-dimensional domains. Zarins. Hemodynamic Factors in Atherosclerosis. Parker. However. C. Acknowledgements This material is based upon work supported by the National Science Foundation under Grant No. Lab. [7] National Heart. the analyses performed clearly indicate the need for appropriate boundary conditions in computing pulse pressures.com). Ku. Taylor. Realistic values of blood pressure (mean and pulse) will be essential for obtaining accurate loading and displacement in fluid-solid interaction methods. Mary Draney for acquiring the subject-specific abdominal aorta MRI data. Insights and perspectives gained from studies of human arteries. [2] A.S. Formaggia. Draney. Glagov. Methods Appl. B. J. Public Health Service. Comput. D. Experimental and computational methods in cardiovascular fluid mechanics. 0205741 and a predoctoral fellowship from the American Heart Association. the authors gratefully acknowledge the assistance of Dr. A. J. C. 112 (10) (1988) 1018–1031. D. Multiscale modelling of the circulatory system: A preliminary analysis. M. Am. Farzin Shakib for his linear algebra package (http://www.A. Zarins. Kamiya. 2002. for a more realistic representation of flow and pressure further analysis of the generation and validation of the impedance functions is needed. Lung and Blood Institute report of the task force on research in pediatric cardiovascular disease. Engrg. Mech. 36 (2004) 197–231. Finally. 195 (2006) 3776–3796 The stability of the solutions proved to be sensitive to the number of frequencies used to generate the impedance boundary condition. T.R. Finally. as such. Veneziani. F. Giddens. 2 (2/3) (1999) 75–83. J. [5] C. Aided Surgery 4 (5) (1999) 231–247. Alternate models based on physical measurements of branching patterns would aid in the specification of realistic impedance functions. Biomech. Taylor.K. C. 25 (6) (1969) 677–686. non-periodic flows or a coupling to the non-linear one-dimensional equations of blood flow may be needed for non-periodic three-dimensional flow problems. Ku.N. the method described may be extensible to other biological and non-biological flow phenomena where the downstream model can be described as a function that links the primary variables or their derived quantities at the coupling interface. [8] N. It is important to note that the present implementation does not incorporate deformable vessels in the three-dimensional domain and.T. Rogge. National Institutes of Health. Comput.3794 I. [6] C. / Comput. Ann.B. Glagov.acusim. Company. Although these results are very encouraging. Stergiopulos. Circulation Res.T. [9] L.A.P. Adaptive regulation of wall shear stress to flow change in the canine carotid artery. Physiol.K. Togawa. References [1] C. Vascular surgery: A Comprehensive Review. Zarins. W. 105–118. An alternate outflow boundary condition that incorporates transient. A. Wolinsky. 239 (1) (1980) H14–H21. Deviation of man from the usual pattern. Visualization Sci. Fluid Mech. Pathol. [3] H. pp. Draney. Comparison of abdominal and thoracic aortic medial structure in mammals. Nobile.S. Quarteroni. 25 (12) (1992) 1477–1488. Rev.N. Mette Olufsen for the use of her methods to compute input impedance of vascular trees and Dr. Med. Dr. The impedance functions were generated using a binary fractal tree that did not incorporate differences between the branching patterns in different organs. T. and Hyun Jin Kim for constructing the abdominal aorta model.E. M. Nathan Wilson for assistance with software development. [4] S. K. A further limitation of the impedance boundary condition is that it is derived based on the assumption of periodicity. . US Department of Health and Human Services. Young. S. The authors gratefully acknowledge Dr. D.A.K. due to the convolution with the flow rate history. 2002. Vignon-Clementel et al. Taylor. Computer simulation of arterial flow with applications to arterial and aortic stenoses.F. Moore and W. Predictive medicine: computational techniques in therapeutic decision-making.

A. Engrg. A mathematical model of circulation in the presence of the bidirectional cavopulmonary anastomosis in children with a univentricular heart. L. Stuhne. Berger. Quarteroni. Peiro. Biomech. 191 (2001) 561–582.J. Shim. . Outflow boundary conditions for one-dimensional finite element modeling of blood flow and pressure waves in arteries. Mech. Finite element modeling of blood flow in arteries. Numerical analysis of blood flow through a stenosed artery using a coupled multiscale simulation method. Saloner. Proceedings of SPIE—The International Society for Optical Engineering 5031 (2003) 11–21. R. Gerbeau. Stroud. R. 191 (6–7) (2001) 661–671. Med. Am. [35] L. [34] L.A. Fumero.D. Bove. Engrg. Comput. Comput. Phys. Quarteroni. Engrg. Computational modelling of 1D blood flow with variable mechanical properties and its application to the simulation of wave propagation in the human arterial system. [16] M. Simulation of blood flow in the abdominal aorta at rest and during exercise using a 1-D finite element method with impedance boundary conditions derived from a fractal tree.R. 28 (7) (1995) 845–856.R. Engrg. Vignon. D. P. Womersley.R. Dubini. Numerical treatment of defective boundary conditions for the Navier–Stokes equations. [13] J. Mark. Migliavacca. Appl. G. [28] E. G.R. Pennati. Methods Appl. M. Pietrabissa. E. Bove. Castro. T. Taylor. F. Migliavacca. G. Oscillatory motion of a viscous liquid in a thin-walled elastic tube—I: the linear approximation for long waves. C. Dubini. Yim. G. F. [36] D. Womersley. Gerbeau. F. Mech.J. Steele. [27] F. Biomech. M. Numerical analysis of flow through a severely stenotic carotid artery bifurcation. Numer. [18] I. Veneziani. T. Quarteroni. Methods Appl. 5 (3) (2002) 195–206. [33] K. Engrg. Taylor. R. Wan. [25] G. (2000) 219–222. Med. Migliavacca. [20] M. Multiscale modeling of the cardiovascular system: application to the study of pulmonary and coronary perfusions in the univentricular circulation.R. Structured tree outflow condition for blood flow in larger systemic arteries. G. G. Pietrabissa. Biomed. Fumero. Mech.F. [15] M. Mag.A. SIAM J.I. Formaggia.A. 124 (1) (2002) 9–20. Spicer. R. R. Comput. 64 (2) (2003) 637–667. 195 (2006) 3776–3796 3795 [10] J.R. S. [17] B.B. Ruan. Kamm. SIAM J. Biomech. R. de Leval. Int. Biorheology 39 (3–4) (2002) 359–364. 76 (1) (1989) 41–66. Engrg. [29] A.N. Kilner.S. Biophys. Migliavacca.J. Med. Methods Appl.R. Ragni. Finite element modeling of the circle of Willis from magnetic resonance data. [19] J. Wave Motion 39 (4) (2004) 361–374. G. Keller. R. R. Dubini. R. A hyperbolic system of equations of blood flow in an arterial network. Multiscale Model. Lohner. [12] S. 4 (2) (2001) 111–124. Steinman. Taniguchi. [24] C. S. [21] K. Taylor. Soto. S.H.A. R. Analysis of a geometrical multiscale model based on the coupling of ODEs and PDEs for blood flow simulations. J. A one-dimensional finite element method for simulation-based medical planning for cardiovascular disease. Pennati. Nobile. Taylor. Formaggia.R. 7 (1955) 199–221. Methods Appl. J. M. J. K. T. Alperin. Comput. Franke. Strohband. Sherwin. C. F. Eng. Phys. N.N. Multiscale modelling as a tool to prescribe realistic boundary conditions for the study of surgical procedures. FL. Computational fluid dynamic and magnetic resonance analyses of flow distribution between the lungs after total cavopulmonary connection. Zarins. Anal. Steele. Biomed. G. Kobayashi. J.E. E. Math. / Comput. S. 19 (3) (1997) 223–234. J. Oscillatory flow in arteries: the constrained elastic tube as a model of arterial flow and pulse transmission. V. C. 21 (3) (1999) 187–193. Dubini. J. Biomech. Methods Biomech. 23 (5) (2001) 293–298. M. Quarteroni. Phys. Quarteroni. Effects of pulmonary afterload on the hemodynamics after the hemi-Fontan procedure. Hughes.L. F. N. Numer. Givoli. Perktold. Dubini. A. Rappitsch. Engrg. Mech. Methods Fluids 43 (6–7) (2003) 673–700. G. P. Comput. [22] J. Comput. Engrg. Balossino. IEEE Trans. A. Biol.J. Cardiol. Migliavacca. Hughes. Computational fluid dynamic simulations of cavopulmonary connections with an extracardiac lateral conduit. C. 126 (3) (2004) 382–387. A. M. G. Coupling between lumped and distributed models for blood flow problems.B. Veneziani. A finite element method for large domains.A. Heldt.K.J. 46 (4) (1999) 393–399. A. 2003.R. Pennati. Pietrabissa. Curcio. J. J. 158 (1998) 155–196. de Leval. Nobile. de Leval. Dubini. [14] J. Feijoo. Pietrabissa. Computer simulation of local blood flow and vessel mechanics in a compliant carotid artery bifurcation model. Comput. F. A. 40 (1) (2002) 376–401. Physiol. Lagana. Engrg. J. A. 1 (2) (2003) 173–195. [23] G. 2 (1957) 178–187.L. Lagana. Visualization Sci. J.D. Finite element simulation of blood flow in the cerebral artery. M. Simul. Torii. Migliavacca. M. Taylor. Engrg. Veneziani. Mech. Clark. Key Biscayne. 6 (1966) 697–716. Vignon-Clementel et al. Methods Appl.G. T. [31] K. Wave transmission through an assembly of randomly branching elastic tubes. Cebral.R. R. [32] A.G. Finite-element modeling of the hemodynamics of stented aneurysms. The Philos. B. D. Guadagni. 276 (1999) H257–H268. [26] F. Engrg. Phys. [11] W. Oshima. On the coupling of 3D and 1D Navier–Stokes equations for flow problems in compliant vessels. in: Proceedings of the 2003 ASME Summer Bioengineering Meeting. Formaggia. Takagi. J. [30] G. Pennati. Med.E. de Leval. Olufsen. G. O. A. Zhao. 38 (5) (2005) 1129–1141.

Key Biscayne. Whiting. Mech.J. Whiting.J.E.H. Vignon-Clementel et al. Generalized-{alpha} method for integrating the filtered Navier–Stokes equations with a stabilized finite element method.3796 I. Norley. 99 (2–3) (1992) 209–233. Jansen. Engrg. Guyton. 1984. Taylor. in: Proceedings of the 2003 Summer Bioengineering Meeting.Z. A stabilized finite element method for the incompressible Navier–Stokes equations using a hierarchical basis. 32 (1982) 199–259. Int.K. C.M.R. in: Proceedings of the 2001 Summer Bioengineering Meeting. Appleton & Lange. Sani. Engrg. Yim. Ann. Stanton. G. Rutt. Franca. OÕRourke.C. San Francisco. Tsao and C.L. Frey.W.L.J.M. Ganong.N. [43] K. Taylor. Gresho. Experimental and Clinical Principles. the Dirichlet-to-Neumann formulation. Methods Appl. The incompressible Navier–Stokes equations. Frayne. Nichols.W. 26 (6) (1998) 1–14. FL. Mech. Blood flow and vessel mechanics in a physiologically realistic model of a human carotid arterial bifurcation. . J. [49] A.J. Oxford University Press. Xu. Biomed.E. Engrg. Engrg. Long. Mech. K.V. [44] J. [41] L. Physiology of the Human Body. Engrg. 1995. B. Methods Appl. Hughes. C. Englewood Cliffs. X.A. New York. P. T.D. Soto.E. subgrid scale models. [46] S. / Comput. Methods Fluids 35 (2001) 93–116.S. Zhao. 195 (2006) 3776–3796 [37] T. Zarins. Steinman. Tang. R.A. 127 (1995) 387–401. B. McDonaldÕs Blood Flow in Arteries: Theoretical.J. S. D.A. Cheng.M. Meas.T.Y. [38] W. Hughes. Comput. 33 (8) (2000) 975–984. Brooks. 190 (3-4) (2000) 305–319.K. Hughes. Hughes. [40] A. [42] C. [50] C. Saunders College Publishing. Subject-specific finite element modeling of three-dimensional pulsatile flow in the human abdominal aorta: comparison of resting and exercise conditions.F.A. Streamline upwind Petrov–Galerkin formulation for convection dominated flows with particular emphasis on the incompressible Navier–Stokes equations.P. bubbles and the origins of stabilized methods. Comput. Characterization of common carotid artery blood-flow waveforms in normal subjects.R. Multiscale phenomena: GreenÕs functions. J. Biomech. 2003. Physiol. [48] W. Holdsworth. Comput. Thom. [45] D. Mech. Lohner. T. Comput. A. Incompressible Flow and the Finite Element Method. 20 (1999) 219–240. Wiley. 2000. Stabilized finite element methods II. C.R. Jansen. P.R. Methods Appl. Q. R. Hulbert. S. M. Finite element modeling of three-dimensional pulsatile flow in the abdominal aorta: relevance to atherosclerosis. Ariff. 1998. [39] P. Numer.H. R. Cebral. C. Review of Medical Physiology. [47] B. A. O. Methods Appl.F. Mech.P. Engrg. On the modelling of carotid blood flow from magnetic resonance images. 2001.D. Methods Appl.

Sign up to vote on this title
UsefulNot useful