You are on page 1of 20

Journal of Computational Physics 260 (2014) 143162

Contents lists available at ScienceDirect

Journal of Computational Physics


www.elsevier.com/locate/jcp

High-resolution Roes scheme and characteristic boundary


conditions for solving complex wave reection phenomena in
a tree-like arterial structure
Bo-Wen Lin a,b , Pong-Jeu Lu a,b,,1
a
Department of Aeronautics and Astronautics, National Cheng Kung University, Tainan 701, Taiwan
b
Heart Science and Medical Devices Research Center, National Cheng Kung University, Tainan 701, Taiwan

a r t i c l e i n f o a b s t r a c t

Article history: Our understanding of information carried by waves propagating in cardiovascular systems
Received 22 May 2013 has rapidly advanced in recent years due to usage of modeling and simulations. The
Received in revised form 9 October 2013 present research aims at developing a high-resolution numerical scheme that can be
Accepted 12 December 2013
applied to solve mutually interactive one-dimensional (1-D) ows that prevail in tree-
Available online 19 December 2013
like arterial networks. A scheme using nite-volume, Roe-splitting ux-difference method
Keywords: in conjunction with a time-accurate RungeKutta marching method was developed. The
Wave reection coecient theory of characteristics was adopted for the treatment of boundary conditions, resulting
Characteristic boundary condition in a unied formulism using a newly dened reection coecient as a measure to quantify
Approximate Riemann solver the wave reection occurring at the interior and terminal nodes of the tree-like vascular
Finite-volume method structure. The derived boundary condition formulas are convenient for dealing with various
Vascular bifurcation types of boundary specication requirements that may arise in vascular hemodynamics.
Hemodynamics
There are two computational models, an isolated tubular conduit and a tree-like structure
model, that were employed for the present scheme validation and wave simulation
demonstration. The objective was to assess the accuracy of the present scheme in capturing
long-duration, complex, and tiny wave reections and re-reections. The superiority of
the present scheme on capturing discontinuous jumps was rst demonstrated in the
isolated conduit simulation by comparing the results against exact solutions as well as
those obtained by LaxWendroff method. To evaluate and validate the time-accuracy of
the present scheme in treating multiple wave reections and re-reections in tree-like
structures, a linear wave-tracking algorithm was employed. Compared to the LaxWendroff
nite-difference method, the present ux-difference Roe-splitting scheme was shown to
be more robust and accurate, which offers high-resolution with minimal dissipation and
dispersion errors in simulating long-duration, mutually interacting wave reections and
re-reections that commonly occur in the tree-like arterial network.
2014 Elsevier Inc. All rights reserved.

1. Introduction

The numerical methods developed for solving one-dimensional (1-D) tree-like arterial network models hold the key
to unraveling complex wave phenomena associated with human circulatory physiology. Computational 1-D ow models

*
Corresponding author at: Department of Aeronautics and Astronautics, National Cheng Kung University, Tainan 701, Taiwan. Tel.: +886 6 2380300x223;
fax: +886 6 2380308.
E-mail addresses: pjlu@mail.ncku.edu.tw, pjlu888@yahoo.com (P.-J. Lu).
1
Professor of Aeronautics and Astronautics and Director of Heart Science and Medical Devices Research Center, National Cheng Kung University,
1 University Rd., Tainan 701, Taiwan.

0021-9991/$ see front matter 2014 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jcp.2013.12.033
144 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

have been applied to study global hemodynamic characteristics including: hypertension [1], vascular access surgery [2,
3], cerebral perfusion [4,5], coronary circulation [6], cardiovascular interaction [7]; in addition to working as boundary
conditions for three-dimensional analysis of focal pathologies such as stenoses [8,9] and bypass graft juncture ows [10,11].
Human circulation is pulsatile in nature. Pulse waves sent from an upstream contracting heart will propagate, by passing
through numerous vascular bifurcations, into a complex tree-like vascular network where incident waves and reection and
re-reection waves merge and interact. Presently, the characteristics of human hemodynamics in different portions of the
vasculature (arteries, arterioles, capillaries, venules, veins, etc.) are yet completely understood.
Pulse waves propagating along vascular segments have been modeled using 1-D ow conservation laws of mass and
momentum augmented by a structural state equation representing vessel wall mechanics. Numerical methods such as nite-
difference [12] and nite-element schemes [11,13] have been developed to solve this partial differential equation system.
Experimental [14,15] and clinical validation works [16] have shown that 1-D ow model is sucient to capture the pulse
wave propagation phenomenon in large vessels. However, whether or not these schemes are accurate enough to resolve
complex wave interactions existing in a large-scale network structure has not been investigated in detail. In principle, the
design of numerical dissipation and dispersion arising from the discretization of the related differential equations and the
construction of numerical algorithms is essential for successful establishment of a viable computational means for treating
wave mechanics [17,18]. Moreover, the precision and delity of the simulated wave phenomena also depend on the numeri-
cal treatment of the boundary conditions [19,20]. Experiences gained from the past decades in scheme development indicate
that both scheme and boundary condition treatments should comply with wave mechanics, specically the characteristics
theory [21,22] of the associated hyperbolic system. Schemes that fail to satisfy the characteristics theory often introduce
spurious numerical artifacts that obscure true physics. Techniques for specifying boundary conditions in hyperbolic systems
have been extensively studied [22,23] on the basis of characteristic analysis of waves crossing inow/outow boundaries.
This has contributed to substantial improvements in scheme stability, convergence acceleration, and solution accuracy.
A special characteristic boundary condition treatment was devised in the present work. By introducing a novel deni-
tion of reection coecient, the inow/outow conditions were generalized into a unied formulism. This new form of
boundary condition representation is particularly useful in the study of vascular hemodynamics. Nonreective, closed-end,
open-end, and actual physiological conditions, which are partially reective in nature, can all be accounted for by an appro-
priate reection coecient specication. In addition to the convenience provided by this generalization of end conditions
for vascular hemodynamics, the use of reection coecient also has mathematical connections with the theory of charac-
teristics. For the present 1-D ow problem, only one boundary condition at the inow/outow end needs to be specied.
The present proposition using reection coecient ts ideally with this well-posedness requirement of boundary condition
specication for a hyperbolic partial differential equation system.
In practical clinical applications, the modeling of arterial tree structure has to be terminated at a certain nite number
of bifurcation generations. Lumped models [24,25] or structured-tree models [12,26,27] have been used to represent the
overall effect caused by a truncated downstream vascular bed. These existing terminal modeling methods involve a plurality
of parameters (resistances, compliances, inertances, network geometries, etc.) which are unknown a priori and need to
be identied using in-vivo measurements preferred to be non-invasively obtainable. The complexity and diculty of this
identication task is strongly dependent on the terminal modeling and the number of parameters included. For example,
suppose that 3 parameters are required for one terminal end modeling (3-element Windkessel), then a vessel network
composed of 20 terminals will have 60 resistance and compliance parameters that would need to be identied. However,
the non-invasive, continuous monitoring that can be performed at radial, carotid, or femoral arteries is usually very limited.
Generally, less than ten parameters can be monitored for the pressure and ow velocity measurements. This lack of a
sucient number of observations deteriorates rapidly as the tree structure becomes large, resulting in a non-identiability
problem posed for the practical parameter identication task of the modeled vascular tree. No matter how complex these
downstream bed models can be, they should be equivalently translated into one terminal specication that is allowed
by the theory of characteristics for 1-D ow. The present boundary specication method reduces these terminal modeling
parameters into only one reection coecient at one terminal end, which is the utmost a boundary modeling can attain. It is
worth mentioning that the present reection coecient specication method serves not only as a means to reduce system
parameter number but also as a quantication that carries medical implications. For instance, a stronger reection often
corresponds to obstructed microcirculation or stenosed downstream vessels described by a larger reection coecient [28];
and an open-end-like partial reection coecient quanties the effectiveness of the aortic valve function during systolic
ejection.
Another issue associated with the construction of a tree-like 1-D ow solver is the problem of determining the interface
variables for the connected vessel segments at the interior bifurcations. The present characteristic formulism and reection
coecient generalization can also be applied to this interior node connection treatment. The derivations of these constituent
boundary and interface formulas required for a successful establishment of a 1-D tree-like vascular ow solver have been
unied based on the characteristics theory, which is explained in the Boundary Condition Treatment section.
To date, systematic investigations of cardiovascular hemodynamics have been pursued using nite-difference and spec-
tral nite-element methods [11,12,29,30]. Finite-volume method developed along the line of approximate Riemann solvers
[3133], which has been justied as the better approach for treating various wave phenomena (shock and discontinuity
capturing, tiny acoustic waves, complex wave interactions, etc.), has not been developed for the study of pulsatile blood
ows in human vasculature bed. Among various approximate Riemann solvers [3133], Roe-splitting [31] of ux-difference
B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 145

is elegant and simple in formulas as well as strictly compliant to wave mechanics. In the present work, we rst derived
Roes scheme for the 1-D vascular equations. The scheme accuracy was then validated against linear theory or exact solu-
tions obtained using ultra-ne grids. To show the merits of the present Roes scheme, we compared our results with those
computed by LaxWendroff scheme with emphasis placed on the evaluation of dissipation and dispersion errors that may
occur in long-duration numerical simulations of waves propagating in generic tree-like networks.

2. Mathematical model

2.1. 1-D formulation

The blood ow in a tubular duct is modeled by taking the cross-sectional average of the ow variables represented by
NavierStokes equations in three-dimension. The resulting 1-D equations in the longitudinal direction can be expressed as
[12,26]

A q
+ = 0,
t x
 2
q q A p 2 qR
+ + = , (1)
t x A x A
where A is the cross-sectional area of the tube with radius R, and other variables and parameters are ow rate q, blood
density , kinematic viscosity (assumed constant), boundary layer thickness (assumed constant), and static pressure p.
The independent variables x and t represent, respectively, the spatial and time coordinates of the postulated 1-D equations.
To close up the above ow equations with three state variables ( A, q and p), a third equation is necessary, namely the
state equation of the tube structure which is presently modeled as a two-dimensional thin-walled elastic structure described
by tube law [26]:
  
4 Eh A 0 (x)
p (x, t ) p 0 = 1 , (2)
3 r0 (x) A (x, t )

where p 0 is the ambient or tissue pressure, A 0 is the cross-sectional area with radius r0 at zero transmural pressure (i.e., at
p = p 0 ), and E and h are the Youngs modulus and the thickness of the vascular wall, respectively. The term Eh/r0 reects
the stiffness of the elastic blood vessel and has been tted empirically using in-vivo data as a function of the stress-free
radius r0 [26]:

Eh
= k1 exp(k2 r0 ) + k3 , (3)
r0
in which the constants k1 , k2 , and k3 take the values as: k1 = 2 107 g/(s2 cm), k2 = 22.53 cm1 , and k3 = 8.65
105 g/(s2 cm).

2.2. Characteristic form

In order to derive the characteristic form for the 1-D blood ow system, it is helpful to rewrite Eq. (1) into a conservative
form:
Q F
+ = S, (4)
t x
where
       
A q 0 0
Q= , F= , S= = , (5)
q q2
A
+B 2AqR + B dr0
r0 dx S
and
B A p B dr0
= + . (6)
x x r0 dx
The bold-faced variables Q , F and S denote the vectors of state variables, convective uxes, and source terms, respectively.
By dening the ux Jacobin matrix A = F/ Q , an equivalent form of Eq. (4) can be obtained:

Q Q
+A = S, (7)
t x
where
146 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

Fig. 1. The quadratic reconstruction of variables dened on the left (L) and right (R) sides of an interface.
 
0 1
A= 2 . (8)
qA 2 + c 2 2q
A
The eigenvalues of A are
q q
1 = + c, 2 = c, (9)
A A
and the wave speed c is dened by

A p
c= . (10)
A
Physiologically, the wave speed c is much faster than the ow velocity U (= q/ A ), therefore we have real and distinct
eigenvalues i , indicating that the system of Eq. (4) is strictly hyperbolic and subcritical.

3. Numerical methods

3.1. Flux-difference splitting

The semi-discrete form of the partial differential system Eq. (4) is expressed in a general form as

dQi Fni+1/2 Fni1/2


+ = Sni , (11)
dt x
where the superscript n is used to indicate the time step and the subscript i refers to the space indices corresponding to
cell centers. For each i-th cell, the left and right side interfaces are indexed as i 1/2 and i + 1/2, respectively. Different
methods, such as the LaxWendroff and Godunov schemes developed by Olufsen et al. [12] and Brook et al. [34], have been
created for computing the numerical ux at the interfaces.
In the present work, Roes approximate Riemann solver [21] was adopted to construct the numerical ux at the interface.
Following the notation convention used in the original Roes scheme, we dene the left and right interface values using
superscripts L and R placed at the upper right corner of the variable noted, as sketched in Fig. 1. The numerical ux of the
Roes scheme [21,31] thus takes the form
1  1
Fi +1/2 = FiL+1/2 + FiR+1/2 k |k |ek , (12)
2 2
where ek are the right eigenvectors of A, k are the wave amplitudes, and k are the eigenvalues of A or the characteristic
group speeds pertaining to the k-th wave mode (k = 1, 2).
Roe [31] ingeniously dened a constant Jacobian matrix
A that is equivalent to a local linearization of a Riemann prob-
lem at the cell face. Roes approximate Riemann solver was developed on the basis of characteristic decomposition of the
ux-difference while satisfying the following Roe conditions [31]:

(i) F =
AQ;
(ii) As Q L = Q R = Q ,

A(Q L , Q R ) = A(Q);

where () = () L () R .
B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 147

For the present 1-D blood ow problem, the Roes states were derived as (see Appendix A)

1 R 

A= A + 2 AR AL + AL ,
4   (13)


1 R 
q= q + qL + qR A L / A R + qL A R / A L ,
4
where
A and
q are the Roes averages for evaluating the eigenvalues and eigenvectors associated with the Roe matrix

A and
for constructing the numerical ux in Eq. (12).

3.2. Spatial discretization

The state variables dened on the two sides of a cell interface specify the order of spatial accuracy of the scheme. Using
the MUSCL (i.e., the Monotonic Upstream-centered Scheme for Conservation Laws) approach of van Leer [35], a third-order
upwind-based scheme can be devised:

1 1+
QiL+1/2 = Qi + (Qi Qi 1 ) + (Qi +1 Qi ), (14)
4 4
1 1+
QiR1/2 = Qi (Qi +1 Qi ) (Qi Qi 1 ), (15)
4 4
where, on the right side of the equations, the rst term represents a rst-order approximation and the remaining parts are
high-order corrections. For = 1/3, the MUSCL interpolation corresponds to a discretization of third-order spatial accuracy
on uniform grids. Hereinafter, we use = 1/3 in our simulations.

3.3. Flux limiter

The above third-order scheme provides high-order spatial accuracy over a smooth region; however, it introduces spurious
numerical oscillations when discontinuities or shocks are present. In some extreme conditions, although rare for circulatory
hemodynamics, discontinuous ows, such as hydraulic jump or elastic jump [36,37], may occur. Hence, in order to accurately
and robustly capture the physical phenomena, an additional non-linear correction factor called limiter [38,39] needs to be
introduced to preserve the monotonicity by maintaining the total variation diminishing (TVD) property of the scheme.
A minmod TVD limiter [39] is used herein to achieve the monotonicity of Eqs. (14) and (15):

1   1+  
QiL+1/2 = Qi + ri+1/2 (Qi Qi 1 ) + ri +1/2 (Qi +1 Qi ), (16)
4 4
1   1+  + 
QiR1/2 = Qi ri+1/2 (Qi +1 Qi ) ri 1/2 (Qi Qi 1 ), (17)
4 4
Q Q Q Q
where r i+1/2 = Qi+1Q i and r i+1/2 = Qi iQ1 are the ratios of slopes, dened forward (+) or backward () as seen from the
i i 1 i +1 i
two sides of the interface. Here, the slope limiter is chosen as

(r ) = minmod( r , 1), (18)

where the minmod function is controlled by the predetermined parameter :



x if |x|  | y | and xy > 0,
minmod(x, y ) = y if |x| > | y | and xy > 0, (19)

0 if xy  0.

3.4. Time marching

The semi-discrete equation (11) can be written into a system of coupled ordinary differential equations that are time
marched:

dQi
= RHS(Qi ), (20)
dt
where the right-hand-side terms are grouped into an RHS operator consisting of the numerical ux and source terms. Here,
we use 3-stage RungeKutta time-stepping [40] to achieve a second-order accuracy with respect to time when marching
from time level n to n + 1:
148 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

 
Q(1) = Qn + t RHS Qn ,
1     
Q(2) = Qn + t RHS Qn + RHS Q(1) ,
4
n +1
 1     
Q = Q + t RHS Qn + RHS Q(1) + 4 RHS Q(2) .
n
(21)
6

4. Boundary condition treatments

4.1. Time-dependent numerical boundary equations

The characteristic forms of Eqs. (1) and (4) are obtainable according to the characteristics theory. Following the work of
Thompson [23], our characteristic formalism can be expressed as

W1 W1 S
+ 1 = 0, (22)
t x 2c
W2 W2 S
+ 2 + = 0, (23)
t x 2c
where

W = L Q, (24)

with
 
1 2 1
L= , (25)
2c 1 1
which are called the characteristic or Riemann differences. At the boundary cells, the spatial derivatives Li need to be ap-
proximated using upwind differencing in order to comply with the direction of wave propagation. In addition, one particular
physical boundary condition pertaining to certain employed boundary type must be specied for one of these two Li :
 
W1 1 A q
L1 1 = 2 + , (26)
x 2c x x
 
W2 2 A q
L2 2 = 1 , (27)
x 2c x x
where the lower index in Li represents the i-th wave mode propagating with characteristic velocity i . By using the trans-
formation between characteristic and physical (primitive) variables, the physical variables A, q and p can be found in terms
of Li :

A
+ L 1 + L 2 = 0, (28)
t
q
+ 1 L 1 + 2 L 2 S = 0, (29)
t
p c2
+ (L1 + L2 ) = 0. (30)
t A
At the inow or outow boundaries, these time-dependent equations govern the advancement of the physical variables with
respect to time. In the above formulation, the domain of inuences, from which the spatial difference approximations Li
are designed, should be explicitly considered so as to satisfy the direction of wave propagation.

4.2. Boundary conditions

Let the computational domain dened for a 1-D ow be bounded within an interval 0  x  L (Fig. 2(a)). The waves with
i < 0 at x = 0 and i > 0 at x = L are outgoing waves, where one-sided difference using interior stencil points is adopted
for evaluating the corresponding Li . However, the waves with i > 0 at x = 0, and i < 0 at x = L are incoming waves,
where the proper values for computing Li depend on how the physical boundary conditions are specied.
In most physiological conditions, the blood ow velocity is smaller in magnitude than the wave velocity. This leads to
the conclusion that only one boundary condition needs to be imposed at the inow/outow boundary. Physical boundary
conditions commonly arising in circulatory physiology include open-/closed-end and nonreecting boundaries as well as a
newly proposed partial reective boundary condition, which is dened and described below.
B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 149

Fig. 2. (a) Computational domain and characteristic boundary condition specication of an isolated vascular model; (b) the arterial bifurcation model; (c) the
characteristic bifurcation interface treatment of the decomposed parent and daughter vessel segments.

4.2.1. Nonreecting boundary condition


Nonreecting boundary condition has been used in computations of domains with innite extent [41] and for treating
uxes across interfaces, which are supposed to be transparent, between adjacent grid blocks [42]. Nonreecting condition
was rst correctively developed for nonlinear equations by Hedstrom [43]. The mathematic form for this nonreecting
condition reads

W 
= 0. (31)
t boundary
Thus, by incorporating Eq. (31) into Eqs. (22) and (23), the nonreecting boundary condition becomes

L1 |x=0 = S /2c Lnon


1 ,

L2 |x= L = S /2c Lnon


2 . (32)
Note that in Eq. (32), the superscript non denotes nonreecting.

4.2.2. Closed-end boundary condition


For a tube with both ends closed, the incoming pressure is doubled and the incoming ow is zero as the wave encounters
the boundary [44]. Thus, q = 0 is specied at the closed-end boundary at all times. With the use of Eq. (29) and q = 0 always
we have

L1 |x=0 = (2 L2 + S )/1 Lc1 ,


L2 |x= L = (1 L1 + S )/2 Lc2 . (33)
In Eq. (33), the superscript c denotes closed-end.

4.2.3. Open-end boundary condition


For an open-end tube, the pressure equals the given ambient pressure, which is often constant with respect to time.
From Eq. (30), this constant ambient pressure requirement is equivalent to always having p / t = 0 at the end, leading to
L1 + L2 = 0 and

L1 |x=0 = L2 Lo1 ,
L2 |x= L = L1 Lo2 . (34)
In Eq. (34), the superscript o denotes open-end.

4.2.4. Partially reecting and general formalism of boundary conditions


Under physiological conditions, pulse wave reection occurs at sites of bifurcated junctures or places having impedance
mismatch, which are characterized as neither fully reecting nor purely nonreecting conditions [45,46]. Large arteries
gradually bifurcate into smaller arteries from generation to generation, resulting in hundreds of millions of arterioles and
capillaries being embedded in the tissue. In practice, it is not possible to exhaust this tree-like vasculature with millions
of connected 1-D ow branches. Hence, if only concerned about the hemodynamics in larger arteries, the 1-D ow model
must be truncated after a few generations of bifurcations. In the past, terminal conditions were modeled to represent the
overall effect of the truncated downstream vascular bed. Nowadays, Windkessel or lumped-parameter models (0-D) are
frequently adopted [47], which usually require specication of several unknown resistance and compliance parameters. No
matter what type of Windkessel model (3-element, 4-element, etc.) is used, the terminal boundary modeled is partially
reective. Physiologically, any ad-hoc terminal modeling should be associated with some particular partial reection phe-
nomenon characterized by the modeling. According to the theory of characteristics, only one single boundary condition can
150 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

be specied at each end of a natural 1-D vascular ow segment. The combination of mathematical requirements and phys-
iological observations inspires the present concept of implementing a reection coecient into our characteristic boundary
condition treatment in order to properly portray the partial wave reection phenomenon.
A non-linear reection coecient R f is presently dened in the time-domain by

L Lnon
Lci Linon , for 0  R f  1,
R f ,i = i i
i = 1, 2, (35)

Li Lnon
i
, for 1  R f  0,
Lnon
i
Loi

where R f ,i are the reection coecients respectively specied at each side of the vessel (i.e., at x = 0, i = 1 and at x = L,
i = 2). In Eq. (35), Lnon
i
, Lci and Loi represent nonreecting closed-end and open-end boundary conditions as dened previ-
ously in Eqs. (32)(34). With the introduction of Eq. (35), we can construct a unied characteristic formalism of boundary
conditions:
  non 
L1 |x=0 = Lnon c
1 R f ,1 L1 L1
  for 0  R f  1, (36)
L2 |x= L = Lnon non c
2 R f ,2 L2 L2
  non 
L1 |x=0 = Lnon o
1 + R f ,1 L1 L1
  for 1  R f  0, (37)
L2 |x= L = Lnon non o
2 + R f ,2 L2 L2

in which the reection coecients R f ,i are bounded between 1 and 1. The three nominal types of boundary conditions,
open-end, closed-end, and nonreecting, correspond to the specication of reection coecients of 1, 1, and 0, respec-
tively. However, other partially reective waves occurring in circulatory physiology take on intermediate values between 1
and 1. Note that positive R f represents reected waves of the same sign (e.g. compression incident and reected waves)
while negative R f corresponds to reection of opposite wave types (e.g. compression incident but rarefaction reected
waves).

4.2.5. Physical meaning of the reection coecient R f


To show the usefulness of the reection coecient dened above, we examined specic vascular ow conditions that
have frequently been seen in discussions of pulse waves in hemodynamics. Assuming an inviscid ow in a vessel of constant
diameter (dr0 /dx = 0), the source term of Eq. (4) vanishes (S = 0). Since, in general U  c, the characteristic speed can be
approximated as 1 c and 2 c signifying that the Riemann variables are invariant along the characteristic lines that
propagate upstream and downstream respectively with speeds of c. This situation renders the nonreecting and closed-end
boundary conditions, dened respectively by Eqs. (32) and (33), into

L1 |x=0 = 0 Lnon
1 ,

L2 |x= L = 0 Lnon
2 , (38)

L1 |x=0 = L2 Lc1 ,
L2 |x= L = L1 Lc2 . (39)
Substituting boundary conditions (34), (38), and (39) into (36) and (37) yields

L1 |x=0 = R f ,1 L2 ,
L2 |x= L = R f ,2 L1 . (40)
According to wave intensity analysis [4850], the incremental changes in pressure and velocity of wavelets can be de-
composed into contributions associated with the forward (+) and backward () propagating wave components:
1
dP = (d P c dU ), (41)
2
1
dU = (dU d P / c ). (42)
2
With the aid of Eqs. (28)(30) and the fact that S = 0, 1 c, and 2 c, Eqs. (41) and (42) can be deduced to show that
the ratios of the incremental changes associated with the incident and reected waves are

d P /d P + = L2 /L1 , (43)
dU /dU + = L2 /L1 . (44)
With substitution of Eq. (40) into Eq. (43), the incremental reected pressure component caused by incident wave generated
from within the interior can be written as
B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 151

(d P + /d P )x=0 = R f ,1 ,
(d P /d P + )x= L = R f ,2 . (45)

Similarly, the ratio of the incremental velocity components of a reected and an incident wave at a boundary takes the form

(dU + /dU )x=0 = R f ,1 ,


(dU /dU + )x= L = R f ,2 . (46)

Thus dened, the reection coecient R f reects the ratios of the pressure and velocity components occurring at the
boundary of interest.

4.2.6. Comparison of the present and the classical reection coecient denitions
In the linear theory, reection coecient is dened by the reected pressure to the incident pressure at the terminal
ends, as given by Eq. (45). When considering wave motion in a 1-D tube connected with a terminal resistance Windkessel
model [44], the reection coecient R f can be conveniently formulated as

R Z0
Rf = , (47)
R + Z0

in which, R is the terminal resistance and Z 0 the characteristic impedance, c 0 / A 0 . With this R f denition, closed-end,
open-end, and nonreecting end correspond respectively to R f = 1, 1 and 0. For partially reective ends, however, R f
takes on values between [1, 1]. Note that this denition is made under the assumption that the ow is inviscid (no
viscous dissipation) and the wave speed is much larger than the ow velocity, the fact that S = 0, 1 c, and 2 c.
Our new denition for R f , Eq. (35), can be viewed as a generalization of Eq. (47). There are no restrictions on the zero
source term and on the forward and backward wave speeds relative to the ow convection speed. The similarity between
Eqs. (35) and (47) is that they both are conditioned on R f = 1 for closed-end, 1 for open-end, and 0 for nonreecting end,
respectively. The dissimilarity, however, lies in the denitions of the numerator and denominator of R f , the new denition
uses the differential operator whereas the classic uses the algebraic number.

4.3. Bifurcation condition treatment

4.3.1. Interface conditions


Interface conditions for bifurcated vessels have been numerically treated using different mathematical or physical argu-
ments [7,30]. At a bifurcation juncture (see Fig. 2(b)), the present interface treatment requires that mass conservation and
pressure continuity are satised:

q 1 + q 2 + q 3 = 0, (48)

p1 = p2 = p3 , (49)

where q j , p j are the blood ow rates and pressures representative of the parent ( j = 1) and two daughter ( j = 2, 3) vessels
jointed at the juncture. Inward ow rate of the vessel is dened positive, whereas outward ow rate is negative.
Taking time derivatives of primitive variables p and q in Eqs. (48) and (49) and expressing them by their characteristic
counterparts, Eqs. (29) and (30), leads to

11 L11 + 12 L12 S 1 = 21 L21 + 22 L22 S 2 + 31 L31 + 32 L32 S 3 , (50)

(c 1 )2   (c 2 )2  2  (c 3 )2  3 
L11 + L12 = L1 + L22 = L1 + L32 , (51)
A1 A2 A3
j j
where the indices i and j appearing on the lower and upper corners of variables A j , c j , S j , i , and Li denote the i-th
wave component (i = 1, 2) and j-th vessel branch ( j = 1, 2, 3), respectively (see Fig. 2(c)).
Eqs. (50) and (51) represent the bifurcation interface conditions governed by three equations in terms of six variables
j
(Li , i = 1, 2 and j = 1, 2, 3). According to the theory of characteristics, only incoming wave-related L12 , L21 , and L31 variables
need to be specied, whereas L11 , L22 , and L32 can be approximated by upwind differences using interior points. Thus, at
each jointed bifurcation node, the problem is well-posed with a balanced number of unknowns and equations. Furthermore,
by adopting matrix notation, Eqs. (50) and (51) can be rewritten as
152 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

DX = Y, (52)
where

12 21 31 L12
(c 1 )2 2
D=
A1
(cA22) 0 ,
X = L21 ,
(c 1 )2 2
L31
A1
0 (cA33)

11 L11 22 L22 32 L32 S 1 + S 2 + S 3
2
(c 2 )2
Y=
(cA11) L11 + A2
L22 .
(53)
(c 1 )2 (c 3 )2 3
A1
L11 + A3
L2
c1 c2 c3 ( A 1 c2 c3 + A 2 c1 c3 + A 3 c1 c2 )
The determinant det(D) = A1 A2 A3
> 0, therefore the square matrix is nonsingular, an inversion D1 exists,
and

X = D1 Y. (54)

4.3.2. Physical interpretation of reection coecient at bifurcation


Consider an incident wave that originates from the domain 1 and enters the bifurcated domains 2 and 3 , as shown
in Fig. 2(c). Assume that there are no other incident waves propagating from within domains 2 and 3 that enter 1 , i.e.,

L22 = 0,
L32 = 0. (55)

This expresses that some portion of the incident wave is transmitted (indicated by and while another portion is L21 L31 )
reected (indicated by L12 ) due to the impedance mismatching at the bifurcation. To simplify the algebraic representation,
we assume that the source term of Eq. (4) is negligible (S = 0) and the characteristic speeds can be approximated by 1 c
and 2 c. This approximation gives a simplied form of Eq. (50):
     
c 1 L11 L12 = c 2 L21 L22 + c 3 L31 L32 . (56)

Combining Eqs. (51), (55), and (56) with the replacement of L21 , L31 in terms of L11 , L12 yields

A1   A2  1  A3  1 
L11 L12 = L1 + L12 = L1 + L12 . (57)
c1 c2 c3
Through the introduction of the denition of reection coecient, Eq. (40), and some algebraic manipulation, the reec-
tion coecient for the parent vessel takes the form
A1 A2 A3
L12 c1
c2
c3
R 1f = = . (58)
L11 A1
c1
+ A2
c2
+ A3
c3

Following the same procedure, we can also derive expressions for the reection coecients of daughter vessels 2 and 3:
A2 A1 A3 A3 A1 A2
L21 c2
c1
c3 L31 c3
c1
c2
R 2f = = , R 3f = = , (59)
L22 A1
c1
+ A2
c2
+ A3
c3
L32 A1
c1
+ A2
c2
+ A3
c3

j
where R f denotes the reection coecient derived at the bifurcation for an incident wave traveling in the j-th vessel.
A linearized formulation in the same equation form, but with constant A j and c j , had previously been obtained by Sherwin
et al. [30].

5. Numerical examples

5.1. Example 1: test case with discontinuous ow

In this example, we demonstrate the superiority of the present scheme on capturing discontinuous jumps. The simulated
isolated vascular model is sketched in Fig. 2(a). The ow is assumed inviscid and the cross section of the tube is constant,
thus eliminating physical viscous damping and the amplication mechanism attributed to lumen radius change. This 1-D
computational domain has a length of L = 50 cm with a lumen radius of 0.8 cm and a vessel stiffness of Eh/r0 = 3.75
105 g/(s2 cm) is chosen. Simulation was conducted using nonreecting boundary conditions specied at both inow/outow
B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 153

Fig. 3. Example 1: The computed ow rate waveforms at a point located in the middle of a simulated isolated blood vessel (x = 0.5L); (a) a square ow
prole specied at the inlet of the blood vessel; and results computed respectively by (b): LaxWendroff scheme (LW), (c): the present scheme with TVD
limiter, and (d): the present scheme without TVD limiter. (Solid lines are exact solutions and dashed lines with circles are numerical solutions.)

boundaries (i.e., R f ,1 = R f ,2 = 0). Therefore, the computational domain can literally be taken as an innitely long tube,
which rules out boundary effects and renders the employed scheme as the sole mechanism responsible for solution accuracy.
A square ow prole is prescribed at the inlet of the blood vessel (x = 0), given by

1, 0.05  t  0.1 s
q(x, t ) = at x = 0. (60)
0, otherwise
Since there exists no analytic solution for this problem, we constructed an exact solution using an ultra-ne grid of 20,000
points. Simulation results were then examined against this exact solution. In order to shed some light on the advantage of
the present ux-difference Roe-splitting method, we also compared our solution with solutions obtained using the popular
second-order LaxWendroff (LW) nite-difference method.
In this simulation, we also examined the effect of TVD limiter on discontinuity capturing. Figs. 3(b)(d) illustrate the
simulated ow waveforms obtained using LW and the present TVD/non-TVD MUSCL schemes for a point located in the
middle of the blood vessel (x = 0.5L). Note that the LW scheme introduces spurious numerical oscillations near disconti-
nuities. In contrast, the present TVD MUSCL scheme is able to capture the discontinuities in an essentially non-oscillatory
fashion.

5.2. Example 2: reecting boundary condition

Usage of the present boundary condition dened in terms of reection coecient, as given by Eq. (40), is to be exem-
plied. The purpose is to show that various types of inow/outow reection scenarios can be properly simulated with
relation to actual physical situations.
154 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

Fig. 4. Example 2: The pressure (solid lines) and ow rate (dashed lines) waveforms calculated at the central point of an isolated blood vessel (x = 0.5L);
(a) Flow rate prescribed at the inlet of the vessel (x = 0); (b) results of a closed-end boundary condition (R f ,1 = R f ,2 = 1); (c) results of a solid-end-like
partially reecting boundary condition (R f ,1 = R f ,2 = 0.5); (d) results of an open-end-like partially reecting boundary condition (R f ,1 = R f ,2 = 0.5).

At the vessel inlet (x = 0), a ow rate pulse with a temporal half-sine waveform is imposed. The pulse period is chosen
to be 0.06 s and the prole is given by
 (t 0.02)
sin( ), 0.02  t  0.08 s
q(x, t ) = 0.06 at x = 0. (61)
0, otherwise
The same vascular geometry previously used in Example 1 was adopted; however, in the present case, reective boundary
conditions were employed. Three sets of reection coecients were used and for each scenario, the same R f values were
assigned at both inow/outow ends. The choices of R f as 1, 0.5, and 0.5 respectively represent a closed-end, a partially
reecting end with the same sign (closed-end-like), and a partially reecting end with opposite sign (open-end-like).
The pressure (solid lines) and ow rate (dashed lines) waveforms calculated at a point located at the middle of the blood
vessel (x = 0.5L) are illustrated in Figs. 4(b)(d). For a closed-end conduit (i.e., R f ,1 = R f ,2 = 1), the pulse wave traverses
in the vessel conduit and reects and re-reects upon its arrival at the ends. The rst peak (T = 0.1 s) corresponds to the
incident wave initially passing across the middle point of the conduit. The second peak (T = 0.2 s) is a backward traveling
reection wave known as the reected rst peak from the downstream boundary (x = L). The third peak (T = 0.3 s) is
a forward traveling re-reection wave bounced from the upstream boundary (x = 0). Repeated reections of waves in the
closed conduit are apparent from the persistent, periodic peaks shown in Fig. 4(b). In addition, at each reection, the ratio
of the incident to reected wave amplitudes is unity as specied, with same signs for pressure reection and opposite signs
for ow reection. This signies reversed ow direction occurring when the wave impinges the solid boundaries.
Results corresponding to partially reecting boundary conditions were obtained using reection coecients of R f ,1 =
R f ,2 = 0.5 and R f ,1 = R f ,2 = 0.5. The mechanisms underlying partial wave reection phenomenon have previously been
discussed [51]. A positive partial reection coecient (R f > 0) represents a boundary more closely related to closed-end
nature, while a negative R f value pertains to a boundary more similar to open-end nature. The reected wave amplitudes
B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 155

Fig. 5. Example 3: (a) The long-time exposure of the ow rates recorded at the central point of the simulated isolated blood vessel ow (x = 0.5L) with
both ends closed. Results are obtained using the present scheme on a grid of 100 points. The gradual decay of wave amplitude indicates the numerical
dissipation contained in the scheme; (b) the wave amplitudes simulated by the present scheme have lesser decay rates as compared to those of the
LaxWendroff (LW) scheme, as shown on the grids of 100, 200, and 400 points after the m-th reection.

tend to diminish gradually due to wave energy depletion caused by outgoing wave components upon each impingement at
the ends. Whether the boundaries are open-end-like or closed-end-like can be differentiated by the sign change signatures
associated with the reected ow rate and pressure waveforms, as illustrated in Figs. 4(c) and 4(d).
The mechanisms of wave amplitude reduction and sign change can alternatively be explained using the concept of
reection coecient. There is no loss of energy under non-dissipative reection described by Eqs. (45) and (46). Hence, the
nal values of pressure and ow velocity after the m-th reection can be calculated by


m
 
Pm = P n1 R nf ,
n =1


m
 
Um = U n1 R nf , (62)
n =1

where the superscript n denotes a reections order or number in sequence, P 0 and U 0 represent the initial pressure and ve-
locity values (i.e., before reection occurs), and m is the total number of reection sites that have been visited. For instance,
P 3 = P 0 P 1 P 2 ( R 0f )3 and U 3 = U 0 U 1 U 2 ( R 0f )3 are the nal pressure and ow velocity states for the 3rd reection, assuming
that the reection coecient is constant (R nf = R 0f ). For a positive reection coecient, which represents a closed-end-like
boundary, there is a sign change on the reected ow velocity; on the other hand, for a wave characterized by a negative
reection coecient, indicating an open-end-like boundary, a sign change on the reected pressure occurs.

5.3. Example 3: numerical dissipation

For an unsteady wave problem, a regular concern is time-accuracy being hampered in long-time simulation due to
numerical dissipation and dispersion errors. As shown in Fig. 4(b) of Example 2, such pulse waves sent into a conduit with
both ends closed will, in theory, reect and re-reect perpetually. However, observations depict simulated wave amplitude
gradually decaying in the course of computation. These numerical errors are attributed to the computational scheme and
the boundary condition treatment, because physical viscous damping was removed intentionally.
A pulse ow rate having the same waveform dened in Eq. (61) was imposed at the vessel inlet (x = 0), but with a
shorter pulse period of 0.02 s. After the initial pulse completely entered the computational domain, the inow end was
closed. The reection coecients, hence, were R f ,1 = R f ,2 = 1, which simulated the closed-end boundary conditions after
introduction of the pulse wave. In accordance to Eq. (62), the reected waves should be non-decaying in the process of
wave propagation.
The time history of the ow rates recorded at the middle point of the blood vessel (x = 0.5L) is sketched in Fig. 5(a).
There is a trend of decay in the peaks due to the presence of numerical dissipation errors. These errors are related to
the scheme design and grid density as well, as illustrated in Fig. 5(b). It is observed that ner grids have better accuracy,
because numerical dissipation is proportional to grid size. Also, simulations of longer duration, indicated by larger reection
number m, yield dampened wave amplitudes. In summary, the present TVD MUSCL schemes performance is uniformly
superior to LW schemes on all the grids and durations tested.
156 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

Fig. 6. A tree-like arterial structure model originally used by Alastruey et al. [52].

Table 1
The parameters and geometric data of the tree-like vascular network model originally used by Alastruey et al. [52].

Edge Length (cm) Radius (mm) c 0 (cm/s) R f ,1 R f ,2


1 50 5.0 560 1 0.19*
2 75 4.0 560 0.24* 0.11*
3 100 1.0 560 0.95* 0.5
4 125 0.8 560 0.96* 0.5
5 100 3.5 560 0.15* 0.5
*
Reection coecients at bifurcations are evaluated using Eqs. (58) and (59).

Fig. 7. Forward (Panels (a) and (b)) and backward (Panels (c) and (d)) pressure wave components computed at the central point of Edge 1. Panels (b) and
(d) are the local enlargements of Panels (a) and (c), respectively. The symbols represent the numerical results computed by the present method using an
ultra-ne grid of 18,000 points. The solid lines are the exact solutions solved by linear wave tracking algorithm [52].

5.4. Example 4: a tree-like vascular network model

A tree-like network model proposed by Alastruey et al. [52] was used herein as a benchmark for scheme validation and
comparison. The tree-like structure has two internal bifurcation nodes and four external boundary nodes connected by ve
edges (Fig. 6). The relevant reection coecients can either be evaluated using Eqs. (58) and (59) at the bifurcation nodes
B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 157

Fig. 8. Long-time history of pressure waveforms recorded at the central point of Edge 1. Solutions are simulated by the LaxWendroff (LW) scheme. The
symbols represent the numerical solutions obtained using a grid of 900 points and the solid lines are the exact solutions solved by the present scheme
using an ultra-ne grid of 18,000 points. Note the pressure scales in each Panel differ by orders of magnitude.

or be prescribed in the form of Eq. (40) at the terminal nodes. All the geometrical data and reection coecients at the
bifurcation/boundary nodes are listed in Table 1.
A unit pulse ow rate was imposed at Node 1 as the inow condition. For a point located in the middle of Edge 1
(Fig. 6), the simulated pressure waveforms were decomposed into wave components in their characteristic directions via
wave intensity theory [48]. The separated forward and backward-going pressure components are shown in Figs. 7(a)(d).
By comparing the simulated results using an ultra-ne grid of 18,000 points to those published by Alastruey et al. [52]
using linear wave tracking algorithm, excellent agreement in data is achieved, indicating a low amount of dissipation and
dispersion errors associated with the present scheme. This reference solution obtained using ultra-ne grid will be taken as
the basis for the subsequent long-duration scheme evaluation.
Following the same simulation procedure, but with a coarser grid of 900 points, we can show the superiority of the
present scheme in capturing tiny and complex reected and re-reected waves typically occurring in multilayers of vascular
branches. The long-time histories of the pressure waveforms calculated by LW and the present schemes are illustrated
in Fig. 8 and Fig. 9, respectively. Deviations in solutions appear and amplify as the simulation time increases. Note that
the scales of the simulated pressure drop in orders from Panels (a) to (d). It is seen that LW scheme introduces spurious
numerical oscillations near places of high temporal gradients. Moreover, for long-time simulation, as sketched in Figs. 8(d)
and 9(d), LW scheme obviously suffers from much larger dispersion and diffusion errors. Evidence of this is the severe
loss of ability in tracking the amplitude, shape, and phase of the propagating waveforms. In contrast, the present MUSCL
Roes scheme is shown to be more capable of capturing the tiny reections and re-reections wave proles. For long-time
simulation, only lag of wave propagation in time is committed, whereas correct waveforms have been virtually maintained.
The built-in TVD property prevents aliasing phenomena from occurring and enables the present Roe-splitting scheme to
perform more robustly than the LW scheme.
158 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

Fig. 9. Long-time history of pressure waveforms recorded at the central point of Edge 1. Solutions are simulated by the present scheme. The symbols
represent the numerical solutions obtained using a grid of 900 points and the solid lines are the exact solutions solved by the present method using an
ultra-ne grid of 18,000 points. Note the pressure scales in each Panel differ by orders of magnitude.

6. Conclusions

The present work proposes a high-resolution Roe-splitting scheme and a unied characteristic boundary condition treat-
ment for solving 1-D vascular hemodynamic ows. The superiority of the Roe-splitting scheme in achieving high-order
accuracy in terms of having small numerical dissipation and dispersion errors has been demonstrated and validated.
There have been two basic boundary condition treatment problems involved in constructing a huge set of 1-D branches
into a tree-like vascular network. One is the terminal condition specication that represents the truncated downstream
microcirculation and the other is the interface connection of bifurcated branches spread in a network. A unied formulism
based on the theory of characteristics is proposed by introducing a parameter, termed reection coecients, to general-
ize the required numerical boundary condition treatments. This newly dened reection coecient is able to measure the
partial wave reection phenomenon occurring at the terminal boundaries (explicitly specied) or interior bifurcation nodes
(implicitly solved). The present characteristic implementation of numerical boundary and interface conditions not only ra-
tionalizes the way information is carried by waves into or out of the constituent 1-D branches, but also enhances the global
accuracy and robustness of solutions of simulated ow in a tree-like network.
The superiority of the present scheme on capturing discontinuous jumps and producing time-accurate simulations has
been demonstrated by comparing the numerical results against those computed using linear wave-tracking algorithm and
the LaxWendroff method. The results show that the proposed Roes scheme can offer an aliasing free high-resolution for
solving long-duration, complex reection and re-reection waves in a tree-like structure. Physiologic pulse ow features
associated with human circulatory system can be better resolved when assisted by the present method. With accurate and
robust computational procedures becoming available, we believe we will be in a better position to understand the genesis
of diseases in connection with pulse wave propagation and reection in major vessels and organs.

Appendix A

The conservative equation form of a 1-D blood ow can be expressed as


B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 159

Q F
+ = S, (A.1)
t x
where
     
A q 0
Q= , F= , S= , (A.2)
q q2
A
+B 2AqR + B dr0
r0 dx

and

4 Eh A0 A
B= . (A.3)
3 r0
The bold-faced variables Q , F, and S denote vectors corresponding to state variables, convective uxes, and source terms,
respectively. Because q = Au, Eq. (A.2) can alternatively be written as
   
A Au
Q= , F= . (A.4)
Au Au 2 + B
Dening the ux Jacobin matrix as A = F/ Q , one can arrive at a non-conservative from of Eq. (A.1) given below:

Q Q
+A = S, (A.5)
t x
where
   
0 1 0 1
A= 2 = , (A.6)
qA 2 + c 2 2q
A
u 2 + c 2 2u

and

2 2 Eh A0
c = . (A.7)
3 r0 A
Roe [31] has constructed a special constant Jacobian matrix

A, evaluated using some intermediate values between Q L and


Q R , to approximate this non-linear Riemann problem while satisfying the following conditions:

(i) F =
AQ;
(ii) As Q L = Q R = Q ,

A(Q L , Q R ) = A(Q);

where () = () L () R .
In order to construct this Roe-averaged Jacobian matrix

A, a parameter vector Z can be chosen following the original


work of Roe [31]:
   
z1 1
Z= = A , (A.8)
z2 u
and Eq. (A.4) can be reexpressed in terms of zi as
   
z12 z1 z2
Q= , F= , (A.9)
z1 z2 z22 + z1
where

4 Eh A0
= . (A.10)
3 r0
By adopting the denition of arithmetic average, indicated by crowning an arbitrary variable a with an over-bar,
 
a = a L + a R /2, (A.11)
it can be shown that the following identities are true for taking difference operations of any associated scalar quantities a
and b:

(a + b) a + b, (A.12)


(ab) ab + ba. (A.13)
Hence, applying difference operator on Q , and F yields
160 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

Q

BZ, (A.14)
where
 

2 z1 0
B= , (A.15)
z2 z1
and

F

CZ, (A.16)
where
 

z2 z1
C= . (A.17)
2 z2

From condition (i), the linearized Jacobian matrix

A can be obtained:

A =

B1
    
z2 z1 1 z1 0
=
2 z2 2 z12 z2 2 z1
 
0 1
= z22 2 z2 . (A.18)
+ 2z1
z12 z1

Substituting Eq. (A.8) into (A.18) we have


 

0 1
A= , (A.19)

u +

c 2 2
2

u
where
z2

u , (A.20)
z1
and

2 Eh A0

c 2
3 r0

A

2 Eh A0
= = . (A.21)
2 z1 3 r0 z1
From Eq. (A.20), we can derive the Roes state for

u as

uL AL + uR AR

u= . (A.22)
AL + AR
Similarly, from Eq. (A.21), the averaged cross-sectional area

A can be derived accordingly as


1  L  R 2

A = (z1 )2 = A + A
4
1  
L
= A +2 AL AR + AR . (A.23)
4
Using the denition of q = Au, we can derive the Roes state for q:

q=


A
u = z1 z2
1  L  R  L  L  
= A + A u A + uR AR
4
 
1 
= qL + qR + qL A R / A L + qR A L / A R . (A.24)
4
The consistency condition (ii) required by the Roes states can also be checked to be correct:

1

A = ( A + 2 A A + A) = A,
4 (A.25)
1  

q = (q + q + q A / A + q A / A ) = q.
4
B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162 161

References

[1] P. Reymond, N. Westerhof, N. Stergiopulos, Systolic hypertension mechanisms: Effect of global and local proximal aorta stiffening on pulse pressure,
Ann. Biomed. Eng. 40 (2012) 742749.
[2] W. Huberts, K. Van Canneyt, P. Segers, S. Eloot, J.H.M. Tordoir, P. Verdonck, F.N. van de Vosse, E.M.H. Bosboom, Experimental validation of a pulse wave
propagation model for predicting hemodynamics after vascular access surgery, J. Biomech. 45 (2012) 16841691.
[3] J. Alastruey, K.H. Parker, J. Peiro, S.J. Sherwin, Can the modied Allens test always detect sucient collateral ow in the hand? A computational study,
Comput. Methods Biomech. Biomed. Eng. 9 (2006) 353361.
[4] K. Devault, P.A. Gremaud, V. Novak, M.S. Olufsen, G. Vernieres, P. Zhao, Blood ow in the circle of Willis: modeling and calibration, Multiscale Model.
Simul. 7 (2008) 888909.
[5] J. Alastruey, K.H. Parker, J. Peiro, S.M. Byrd, S.J. Sherwin, Modelling the circle of Willis to assess the effects of anatomical variations and occlusions on
cerebral ows, J. Biomech. 40 (2007) 17941805.
[6] J. Lee, N.P. Smith, The multi-scale modelling of coronary blood ow, Ann. Biomed. Eng. 40 (2012) 23992413.
[7] L. Formaggia, D. Lamponi, M. Tuveri, A. Veneziani, Numerical modeling of 1D arterial networks coupled with a lumped parameters description of the
heart, Comput. Methods Biomech. Biomed. Eng. 9 (2006) 273288.
[8] N. Stergiopulos, D.F. Young, T.R. Rogge, Computer simulation of arterial ow with applications to arterial and aortic stenoses, J. Biomech. 25 (1992)
14771488.
[9] F. Cassot, M. Zagzoule, J.P. Marc-Vergnes, Hemodynamic role of the circle of Willis in stenoses of internal carotid arteries. An analytical solution of a
linear model, J. Biomech. 33 (2000) 395405.
[10] E. Marchandise, M. Willemet, V. Lacroix, A numerical hemodynamic tool for predictive vascular surgery, Med. Eng. Phys. 31 (2009) 131144.
[11] B.N. Steele, J. Wan, J.P. Ku, T.J.R. Hughes, C.A. Taylor, In vivo validation of a one-dimensional nite-element method for predicting blood ow in
cardiovascular bypass grafts, IEEE Trans. Biomed. Eng. 50 (2003) 649656.
[12] M.S. Olufsen, C.S. Peskin, W.Y. Kim, E.M. Pedersen, A. Nadim, J. Larsen, Numerical simulation and experimental validation of blood ow in arteries with
structured-tree outow conditions, Ann. Biomed. Eng. 28 (2000) 12811299.
[13] S.J. Sherwin, L. Formaggia, J. Peiro, V. Franke, Computational modelling of 1D blood ow with variable mechanical properties and its application to the
simulation of wave propagation in the human arterial system, Int. J. Numer. Methods Fluids 43 (2003) 673700.
[14] J. Alastruey, A.W. Khir, K.S. Matthys, P. Segers, S.J. Sherwin, P.R. Verdonck, K.H. Parker, J. Peiro, Pulse wave propagation in a model human arterial
network: Assessment of 1-D visco-elastic simulations against in vitro measurements, J. Biomech. 44 (2011) 22502258.
[15] D. Bessems, C.G. Giannopapa, M.C.M. Rutten, F.N. van de Vosse, Experimental validation of a time-domain-based wave propagation model of blood ow
in viscoelastic vessels, J. Biomech. 41 (2008) 284291.
[16] P. Reymond, F. Merenda, F. Perren, D. Ruefenacht, N. Stergiopulos, Validation of a one-dimensional model of the systemic arterial tree, Am. J. Physiol.,
Heart Circ. Physiol. 297 (2009) H208H222.
[17] J. Berland, C. Bogey, O. Marsden, C. Bailly, High-order, low dispersive and low dissipative explicit schemes for multiple-scale and boundary problems,
J. Comput. Phys. 224 (2007) 637662.
[18] C. Hirsh, Numerical Computation of Internal and External Flow, vol. 1, Wiley, New York, 1990, pp. 301318.
[19] T. Colonius, Modeling articial boundary conditions for compressible ow, Annu. Rev. Fluid Mech. 36 (2004) 315345.
[20] C. Hirsh, Numerical Computation of Internal and External Flow, vol. 2, Wiley, New York, 1990, pp. 344401.
[21] P.L. Roe, Characteristic-based schemes for the Euler equations, Annu. Rev. Fluid Mech. 18 (1986) 337365.
[22] T.J. Poinsot, S.K. Lele, Boundary-conditions for direct simulations of compressible viscous ows, J. Comput. Phys. 101 (1992) 104129.
[23] K.W. Thompson, Time-dependent boundary conditions for hyperbolic systems, II, J. Comput. Phys. 89 (1990) 439461.
[24] N. Xiao, J.D. Humphrey, C.A. Figueroa, Multi-scale computational model of three-dimensional hemodynamics within a deformable full-body arterial
network, J. Comput. Phys. 244 (2013) 2240.
[25] W. Kroon, W. Huberts, M. Bosboom, F. van de Vosse, A numerical method of reduced complexity for simulating vascular hemodynamics using coupled
0D lumped and 1D wave propagation models, Comput. Math. Methods Med. (2012), http://dx.doi.org/10.1155/2012/156094.
[26] M.S. Olufsen, Structured tree outow condition for blood ow in larger systemic arteries, Am. J. Physiol., Heart Circ. Physiol. 276 (1999) H257H268.
[27] W. Cousins, P.A. Gremaud, Boundary conditions for hemodynamics: The structured tree revisited, J. Comput. Phys. 231 (2012) 60866096.
[28] J. Liu, L.-J. Yuan, Z.-M. Zhang, Y.-Y. Duan, J.-H. Xue, Y.-L. Yang, Q. Guo, T.-S. Cao, Effects of acute cold exposure on carotid and femoral wave intensity
indexes: evidence for reection coecient as a measure of distal vascular resistance, J. Appl. Physiol. 110 (2011) 738745.
[29] M. Esmaily Moghadam, I.E. Vignon-Clementel, R. Figliola, A.L. Marsden, A modular numerical method for implicit 0D/3D coupling in cardiovascular
nite element simulations, J. Comput. Phys. 244 (2013) 6379.
[30] S.J. Sherwin, V. Franke, J. Peiro, K. Parker, One-dimensional modelling of a vascular network in spacetime variables, J. Eng. Math. 47 (2003) 217250.
[31] P.L. Roe, Approximate Riemann solvers, parameter vectors, and difference schemes, J. Comput. Phys. 43 (1981) 357372.
[32] B. Einfeldt, On Godunov-type methods for gas-dynamics, SIAM J. Numer. Anal. 25 (1988) 294318.
[33] S. Osher, F. Solomon, Upwind difference schemes for hyperbolic systems of conservation laws, Math. Comput. 38 (1982) 339.
[34] B.S. Brook, S.A.E.G. Falle, T.J. Pedley, Numerical solutions for unsteady gravity-driven ows in collapsible tubes: evolution and roll-wave instability of a
steady state, J. Fluid Mech. 396 (1999) 223256.
[35] B. van Leer, Towards the ultimate conservative difference scheme. V. A second-order sequel to Godunovs method, J. Comput. Phys. 32 (1979) 101136.
[36] J.B. Grotberg, O.E. Jensen, Biouid mechanics in exible tubes, Annu. Rev. Fluid Mech. 36 (2004) 121147.
[37] T.J. Pedley, B.S. Brook, R.S. Seymour, Blood pressure and ow rate in the giraffe jugular vein, Philos. Trans. R. Soc. Lond. B 351 (1996) 855866.
[38] A. Harten, High resolution schemes for hyperbolic conservation laws, J. Comput. Phys. 49 (1983) 357393.
[39] P.K. Sweby, High resolution schemes using ux limiters for hyperbolic conservation laws, SIAM J. Numer. Anal. 21 (1984) 9951011.
[40] C.W. Shu, S. Osher, Ecient implementation of essentially non-oscillatory shock-capturing schemes, J. Comput. Phys. 77 (1988) 439471.
[41] A. Barry, J. Bielak, R.C. Maccamy, On absorbing boundary-conditions for wave-propagation, J. Comput. Phys. 79 (1988) 449468.
[42] M.B. Giles, Nonreecting boundary-conditions for Euler equation calculations, AIAA J. 28 (1990) 20502058.
[43] G.W. Hedstrom, Nonreecting boundary conditions for nonlinear hyperbolic systems, J. Comput. Phys. 30 (1979) 222237.
[44] J. Alastruey, K.H. Parker, J. Peiro, S.J. Sherwin, Lumped parameter outow models for 1-D blood ow simulations: Effect on pulse waves and parameter
estimation, Commun. Comput. Phys. 4 (2008) 317336.
[45] S.E. Greenwald, A.C. Carter, C.L. Berry, Effect of age on the in vitro reection coecient of the aortoiliac bifurcation in humans, Circulation 82 (1990)
114123.
[46] R.D. Latham, N. Westerhof, P. Sipkema, B.J. Rubal, P. Reuderink, J.P. Murgo, Regional wave travel and reections along the human aorta: a study with
six simultaneous micromanometric pressures, Circulation 72 (1985) 12571269.
[47] N. Stergiopulos, B.E. Westerhof, N. Westerhof, Total arterial inertance as the fourth element of the windkessel model, Am. J. Physiol., Heart Circ. Physiol.
276 (1999) H81H88.
[48] K.H. Parker, C.J.H. Jones, Forward and backward running waves in the arteries: Analysis using the method of characteristics, J. Biomech. Eng. 112 (1990)
322326.
162 B.-W. Lin, P.-J. Lu / Journal of Computational Physics 260 (2014) 143162

[49] A.W. Khir, A. OBrien, J.S.R. Gibbs, K.H. Parker, Determination of wave speed and wave separation in the arteries, J. Biomech. 34 (2001) 11451155.
[50] K.H. Parker, An introduction to wave intensity analysis, Med. Biol. Eng. Comput. 47 (2009) 175188.
[51] D.L. Newman, S.E. Greenwald, T.B. Moodie, Reection from elastic discontinuities, Med. Biol. Eng. Comput. 21 (1983) 697701.
[52] J. Alastruey, K.H. Parker, J. Peiro, S.J. Sherwin, Analysing the pattern of pulse waves in arterial networks: a time-domain study, J. Eng. Math. 64 (2009)
331351.

You might also like