# INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN FLUIDS

Int. J. Numer. Meth. Fluids 2008; 57:31–45
Published online 13 September 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/ﬂd.1610
A hybrid vortex method for the simulation of three-dimensional
ﬂows
Wei Li
∗, †
and Marco Vezza
Department of Aerospace Engineering, University of Glasgow, Glasgow G12 8QQ, U.K.
SUMMARY
This paper presents an integral vorticity method for solving three-dimensional Navier–Stokes equations.
A ﬁnite volume scheme is implemented to solve the vorticity transport equation, which is discretized on
a structured hexahedral mesh. A vortex sheet algorithm is used to enforce the no-slip boundary condition
through a vorticity ﬂux at the boundary. The Biot–Savart integral is evaluated to compute the velocity
ﬁeld, in conjunction with a fast algorithm based on multipole expansion. This method is applied to the
simulation of uniform ﬂow past a sphere. Copyright q 2007 John Wiley & Sons, Ltd.
Received 5 March 2007; Revised 16 July 2007; Accepted 19 July 2007
KEY WORDS: vortex method; ﬁnite volume method; three-dimensional ﬂow; fast summation; sphere
ﬂow
1. INTRODUCTION
Computational ﬂuid dynamic (CFD) methods have been developed for several decades, providing
an efﬁcient tool for the analysis of many fundamental and practical ﬂuid dynamics problems.
Various numerical techniques have been employed to solve the Navier–Stokes equations that
govern viscous ﬂuid ﬂow, by using the Euler description, in which the truncated domain of the
entire ﬂow region is overlaid by a grid system. Despite the huge success on the simulation of
complex ﬂuid ﬂow, standard CFD methods that rely on solving equations of primitive variables,
the velocity and pressure, are susceptible to excessive numerical dissipation of vorticity. If not
controlled, this dissipation ultimately leads to an enhanced spreading of the vorticity support and
a loss of circulation of the vortex structures.
Lagrangian vorticity methods, on the other hand, follow an alternative approach, in which the
vorticity–velocity form of Navier–Stokes equations only needed to be solved within the limited

Correspondence to: Wei Li, Department of Aerospace Engineering, University of Glasgow, Glasgow G12 8QQ,
U.K.

E-mail: lwei@aero.gla.ac.uk
Copyright q 2007 John Wiley & Sons, Ltd.
32 W. LI AND M. VEZZA
vorticity-containing regions, so that they appear to be efﬁcient and self-adaptive [1–4]. Since
the nonlinear convection term is included in the vorticity time derivative, Lagrangian vorticity
methods exhibit little or no numerical dissipation. However, despite these advantages, Lagrangian
vorticity methods experience some challenging problems that hinder their application in general
ﬂuid dynamics. One of these problems is the computational cost in evaluating the velocity of
vortices via the Biot–Savart law. In this process, the integration used to compute velocity with N
vortices is O(N
2
), and is therefore quite slow and usually prohibitive for large N. This situation has
been improved by introducing several acceleration methods, such as the fast summation algorithm,
which is based on the expansion of the kernel function in the Biot–Savart formula [5, 6]. A
hierarchical structure is commonly used in these acceleration methods such that the particle–
particle interaction is converted to the particle–group or group–group interaction, with the overall
computational complexity reduced from O(N
2
) to O(N ln N) or O(N).
Another problem associated with Lagrangian vorticity methods is the simulation of viscous
diffusion. Instead of the random walk approach that has a shortcoming of slow-convergence rate,
a variety of deterministic methods have been developed over the last two decades. In one of these
methods, the concept of diffusive velocity is introduced by deﬁning an equivalent convection veloc-
ity for the diffusion process. The application of this method to the ﬂow around a circular cylinder
and aerofoil [7] shows that smoother results can be obtained if the vortex particles strictly overlap.
Another deterministic method is the particle strength exchange scheme [8], in which the circula-
tions associated with the vortex particles irregularly distributed in the ﬂow are redistributed either
by the use of an integral representation for the Laplacian operator or a more complicated formula
given in [9]. However, after a period of evolution, the vortex particle distribution may become
uneven, thus destroying the overlapping condition, which necessitates a complicated remeshing
process [8, 10].
The current paper introduces a new hybrid approach to resolve some of the remaining difﬁcul-
ties associated with the Lagrangian vorticity methods. Instead of employing overlapping vortex
particles/blobs, a hexahedral grid system is introduced to solve the three-dimensional incompress-
ible Navier–Stokes equations using the ﬁnite volume method. A modiﬁed Biot–Savart formula is
used to calculate the velocity within the parts of the grid system where non-zero vorticity exists.
Owing to the nature of the ﬁnite volume method, this hybrid approach can guarantee the conser-
vation of the vorticity and mass within the entire ﬂow domain. The use of hexahedral elements
requires a proper numerical integration procedure for calculating the velocity. This has been ac-
complished by using closed-form formulas proposed by Suh [11] with essential completion from
this study.
2. PROBLEM STATEMENT
The Navier–Stokes equations governing incompressible ﬂow can be written in primitive variable
(velocity–pressure) form as
*
*t
u + u · ∇u=−
1
j
∇p + v∇
2
u (1)
∇ · u=0 (2)
where u(x, t ) is the velocity ﬁeld and v is the kinematic viscosity.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 33
Taking the curl of Equation (1) yields the unsteady vorticity transport equation
*
*t
x + u · ∇x − x · ∇u=v∇
2
x (3)
for the vorticity ﬁeld x(x, t ) =∇ ×u(x, t ). In viscous ﬂow, a non-slip boundary condition must
be satisﬁed on solid surfaces, including a Dirichlet condition on the normal component of x [10]
(as the tangential derivatives of the velocity vanish at the wall),
c
n
=x · n=0 (4)
and a Neumann condition on the two tangential components of x (cancellation of the tangential
components of the slip velocity at the wall).
The force F on an immersed solid body with surface S can be calculated by the sum of the
pressure and viscous shear forces as
F=−

S
( pn + jn×x) ds (5)
where n is the outward normal of surface S.
There is also a classical technique used to evaluate the force by computing the time derivative
of the linear impulse (external ﬂow around one body):
F
j
=−
dI
dt
(6)
where j is the density and I is deﬁned by
I =
1
2

V
x ×xdV (7)
3. NUMERICAL IMPLEMENTATION
The governing equations as well as the boundary conditions presented in the previous section are
globally coupled. A ﬁnite volume method is used to solve the vorticity transport equations, while
the boundary conditions are implemented by using vortex sheets on the surface of the solid body.
The velocity evaluation is carried out using a combination of two methods, the direct integration
using numerical integral and the indirect integration using fast summation.
3.1. Finite volume scheme
The vorticity transport equation (3) can be written in the integral form:
*
*t

V
W dV +

S
F(W) · ndS =0 (8)
where n is the unit normal vector of the surface S of the control volume V.
W =[c
1
c
2
c
3
]
T
(9)
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
34 W. LI AND M. VEZZA
Figure 1. The typical hexahedral cell in the domain.
F(W) =

F
1
(W)
F
2
(W)
F
3
(W)

T
(10)
F
1
(W) =

−v
*c
1
*x
u
1
c
2
− u
2
c
1
− v
*c
2
*x
u
1
c
3
− u
3
c
1
− v
*c
3
*x

(11)
F
2
(W) =

u
2
c
1
− u
1
c
2
− v
*c
1
*y
−v
*c
2
*y
u
2
c
3
− u
3
c
2
− v
*c
3
*y

(12)
F
3
(W) =

u
3
c
1
− u
1
c
3
− v
*c
1
*z
u
3
c
2
− u
2
c
3
− v
*c
2
*z
−v
*c
3
*z

(13)
The 3D computational domain is divided into hexahedral cells as shown in Figure 1, and a system
of differential equations is obtained by applying Equation (8) to each cell separately.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 35
Let the values of the quantities associated with each cell be denoted by i, j, k (values at the cell
centre). For each cell Equation (8) can be presented as the form
d
dt
(h
i j k
W
i j k
) + Q(W)
i j k
=0 (14)
where h is the volume of the cell and Q is the ﬂux term.
Q(W)
i j k
= Q
i j k
+ F · S|
i +1/2, j,k
− F · S|
i −1/2, j,k
+F · S|
i, j +1/2,k
− F · S|
i, j −1/2,k
+ F · S|
i, j,k+1/2
− F · S|
i, j,k−1/2
(15)
where S|
i +1/2, j,k
is the normal area on i direction of the interface between cell (i, j, k) and cell
(i + 1, j, k).
(c
i
h)
n+1
i, j,k
=(c
i
h)
n
i, j,k
+
t
2
[3Q
i, j,k
(t
n
, c
n
i
, u
n
) − Q
i, j,k
(t
n−1
, c
n−1
i
, u
n−1
)] (16)
The diffusive ﬂux term is discretized using central differential scheme. For the convective terms,
a third-order QUICK (quadratic upwind interpolation for convective kinematics) [12] scheme is
used.
3.2. Vorticity boundary conditions
The vorticity boundary conditions employed in the present paper are based on vortex sheet method
[10, 13]. A piecewise-continuous vortex sheet is used to replace the body surfaces in order to
cancel the slip velocity. The strength of the sheet is obtained from a Fredholm boundary integral
equation of the second kind:
1
2
Dc(x) ×n +
1

S
1
|x − x

|
3
· (x − x

) ×Dc(x

) dx

=u
slip
(17)
where c(x) is the strength of the vortex sheet and u
slip
is the slip velocity on the surface.
To implement this boundary condition, the body surface is discritized using the surface mesh as
vortex panels. The slip velocity at each panel control point is induced by the free stream, surface
panels and all wake vortex elements within the ﬂow. The vortex strength c(x) is computed by
solving a linear system after evaluating the slip velocity on all the panels. Then the vortex ﬂux to
be emitted into the ﬂow can be obtained as
m
*x
*n
=
Dc(x)
t
(18)
3.3. Velocity evaluations
3.3.1. Direct integration method. In the present method, the vorticity ﬁeld is represented by a
structured grid system consisting of a number of hexahedral elements. The evaluation of Biot–
Savart integral is required to calculate the velocity ﬁeld. Therefore, a robust and accurate algorithm
is needed for the integration over a hexahedral element. Suh [11] proposed an elegant method for
the evaluation of Biot–Savart integral in both two and three dimensions. It transforms the volume
integral into line integrals based on Gauss and Stokes integral theorems and can be applied to
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
36 W. LI AND M. VEZZA
elements with arbitrary number of faces (sides). However, after applying this approach to the
current study, it was found that some limitations need to be imposed onto the original expressions
presented in [11], otherwise, in some cases, one or two terms would produce inﬁnite values. The
other change to the original expressions is that the vorticity distribution within the elements is
simpliﬁed to be uniform in this study rather than Suh’s linear distribution.
According to the Biot–Savart law, the induced velocity due to a vorticity distribution over an
element bounded by planar panels can be expressed as
u=
1

∇ ×

V
x
r
dV =
1

V
x×∇

1
r

dV =
1

x×n
j
N
s

j =1

S
j
1
r
dS (19)
where N
s
is the total number of planar panels bounding the element and n
j
is the unit normal of
the j th panel.
Now
1
r
=e
n
· (∇ ×B) (20)
B=
e
n
×r
r + e
n
· r
(21)
where e
n
=±n such that e =e
n
· r0 then

S
j
1
r
dS =±

S
j
n · (∇ ×B) dS =±

C
BdC (22)
where C is the perimeter of the planar panel. For a quadrilateral panel, it can be expressed as

C
BdC =
4

k=1

l
k
B · l
k
dl (23)
where l
k
is the unit directional vector of side l
k
.
For the evaluation of the line integrals, a local coordinate system (¸, p) is used in the panel, as
shown in Figure 2.
Now we have
r =r
k
+ ¸l
k
(24)
and
r =

r
2
k
− 2¸
p
¸ + ¸
2
(25)
because (n×r) · l
k
=r · (l
k
×n) =r
k
· p
k
.
Hence,

l
k
B · l
k
dl =r
k
· p
k

l
k
0

(¸ − ¸
p
)
2
+ p
2
p
+ e
(26)
After some manipulations, the indeﬁnite integral in Equation (26) can be expressed as

(¸ − ¸
p
)
2
+ p
2
p
+ e
= I
1
+ I
2
(27)
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 37
P
r
k
r
k+1
r
ξ k k+1
p
k
Figure 2. Deﬁnition of the local coordinate system.
where
I
1
= ln

(l
k
− ¸
p
) +r
k+1
r
k
− ¸
p

(28)
I
2
requires some special treatment, looking at two different cases:
(1) If ¸<¸
p
or ¸>¸
p
, for 0<¸<l
k
I
2
=
e

p
2
p
− e
2
[ (29)
where
[ = arcsin

p
2
p
− e
2
[p
2
p
l
k
+ e(l
k
− ¸
p
)r
k
+ e¸
p
r
k+1
]
p
2
p
(e +r
k+1
)(e +r
k
)

(30)
(2) If ¸<¸
p
and 0<¸<¸
p
, for 0<¸
p
<l
k
or ¸>¸
p
and ¸
p
<¸<l
k
, for 0<¸
p
<l
k
I
2
=
e

p
2
p
− e
2
(¬ − [) (31)
3.3.2. Indirect integration method. The direct integration method presented in the last section
requires calculations of 12 logarithmic and angular functions for a single hexahedral element. For
a large number of elements, the time required for direct integration becomes excessive. Therefore,
a fast summation algorithm is used in this paper, based on the Taylor expansion [5, 6].
A hierarchical grid system is ﬁrstly introduced to generate a box tree. The root of the tree is the
single box containing all the vortex particles in the ﬂow. Then each box of the tree is divided into
two offspring boxes with equal size by cutting the longest side of the box. This process continues
until the number of vortex particles in the smallest box, the top of the box tree, is less than a
prescribed value.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
38 W. LI AND M. VEZZA
For points y in a single box t centred at y
t
, their inﬂuence at a ﬁx point x can be expressed as
f (x, y) =
x − y
|x − y|
3
=

|k|z−1
1
k!
D
k
y
f (x, y)
y=y
t
(y − y
t
)
k
(32)
where
k =(k
1
, k
2
, k
3
), |k| =k
1
+ k
2
+ k
3
, k! =k
1
!k
2
!k
3
!
D
k
y
=*
|k|
/*y
k
1
1
*y
k
2
2
*y
k
3
3
, y
k
= y
k
1
1
y
k
2
2
y
k
3
3
The degree of its Taylor polynomial is z − 1.
Then the calculation of the velocity u(x) at point x is given by
u(x) =−
1

|k|z−1
a
k
(x, y
t
)[(x
2
A
k
t
− B
k
t
− x
3
C
k
t
+ D
k
t
), (33)
(x
3
E
k
t
− F
k
t
− x
1
A
k
t
+ G
k
t
), (x
1
C
k
t
− H
k
t
− x
2
E
k
t
+ I
k
t
)] (34)
The sums in the above equations are evaluated for all boxes t and indices k before the velocity
calculation and in parallel with the vortex sorting and grid generation process [14].
4. FLOW PAST A SPHERE
The hybrid vortex method presented in previous sections is utilized for the computation of ﬂow past
a sphere. The ﬂow is started impulsively at various Reynolds numbers (Re) (100, 200, and 300)
with two different ﬂow ﬁeld resolutions (medium and high). The ﬂow variables are normalized
using the diameter of the sphere and the free-stream velocity, which is oriented in the positive
x direction. A single-block structured grid (Figure 3) has been created around the sphere. The
number of hexahedral elements varies from a resolution of 36 000 to 64 000 and 108 000. The
radius of the sphere was taken as R
0
=0.5, and the outer radius of the domain was R
1
=5.0.
An impulsive start of the ﬂow is initialized by computing the potential ﬂow past the sphere
body using the vortex sheet method (Equation (17)). The vorticity associated with the surface slip
is then distributed to the boundary cells through the diffusion process (Equation (18)). The surface
slip velocity is recomputed at the beginning of the next time step to provide a continuing source
of vorticity. The calculation was carried out on a DELL PRECISION 470 workstation with two
Intel Xeon (TM) CPU.
The ﬂow past a sphere behaves differently when Reynolds number (Re) changes. In the range of
Re<210–212 [15, 16], the ﬂow remains steady and axisymmetric although separated. Above that,
if Re<270, the ﬂow remains steady but is no longer axisymmetric. When Re>290 [15], the ﬂow
becomes unsteady but still retains time periodicity and planar symmetry. Two (100 and 200) of the
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 39
Figure 3. The structured grid used for the sphere ﬂow simulation.
three Reynolds numbers simulated in this study are below 210, corresponding to the axisymmetric
ﬂow. The other one (300) was used to test the potential of the hybrid vortex method, but due to
the capability of the workstation, only the early stage of the unsteady ﬂow has been simulated.
4.1. Force coefﬁcient history
Figures 4–6 show the drag coefﬁcients Cd = D/(0.5jU
2
¬R
2
) plotted as a function of time, at
three Reynolds number 100, 200, and 300. The drag coefﬁcient results are compared with the
values from the Schiller–Naumann formula [17] Cd =(24/Re)[1 + 0.15 Re
0.687
] and the results
from Johnson and Patel [15]. The drag coefﬁcients are observed to have a quick decline from the
initial overshoot in the range 0<t <0.5, then to approach a steady-state average value. The curves
of the drag coefﬁcient appear to exhibit more ﬂuctuation when Re increases.
4.2. Flow characteristics
Figures 7–9 show the streamlines in the x–y plane at Reynolds number 100, 200, and 300. For
all these three Reynolds numbers, the ﬂow is seen to separate from the surface of the sphere and
form a separation bubble in the wake. A clear axisymmetric ﬂow pattern can be seen in both
ﬂows at Re =100 and 200, with the only difference being the length of the separation bubble and
the position of the vortex. However, for Re =300, even at the early stage of simulation, the ﬂow
has started to show asymmetrical characteristics and unsteadiness. This feature can be seen more
clearly from a three-dimensional view shown in Figure 10.
In Figure 11, the vorticity |c| contours in the x–y plane at several instances for Re =200 are
presented. Again, the axisymmetric ﬂow characteristics can be observed through the development
of the separated ﬂow.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
40 W. LI AND M. VEZZA
Figure 4. Drag coefﬁcient comparison for the ﬂow past a sphere at Re =100, showing computed results
(solid line), Schiller–Naumann formula (dashed line), and results from Johnson and Petel (dotted line).
Figure 5. Drag coefﬁcient comparison for the ﬂow past a sphere at Re =200, showing computed results
(solid line), Schiller–Naumann formula (dashed line), and results from Johnson and Petel (dotted line).
4.3. Effects of numerical parameters
In Figures 12 and 13, the effect of numerical parameters, in terms of time steps and grid resolution,
on total drag coefﬁcients is presented for Re =300. From Figure 12, the sensitivity of the time step
to the computation of drag coefﬁcient can be observed. The use of smaller time steps, dt =0.005
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 41
Figure 6. Drag coefﬁcient comparison for the ﬂow past a sphere at Re =300, showing computed results
(solid line), Schiller–Naumann formula (dashed line), and results from Johnson and Petel (dotted line).
Figure 7. Streamlines of sphere ﬂow in the x–y plane at Re =100.
Figure 8. Streamlines of sphere ﬂow in the x–y plane at Re =200.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
42 W. LI AND M. VEZZA
Figure 9. Streamlines of sphere ﬂow in the x–y plane at Re =300.
Figure 10. Three-dimensional view of the streamlines of sphere ﬂow in the x–y plane at Re =300. The
velocity contour is also shown.
and 0.0025, produced nearly identical drag values near the immediate time region (t =0+) after
the impulsive start. The use of the moderate time step, dt =0.01, appears to be less accurate,
although it is preferred for the long-term simulation.
To evaluate the effect of grid resolution, three grids, namely Grid1 (total grid number is 36 000)
and Grid2 (total grid number is 64 000) and Grid3 (total grid number is 108 000), are used separately
for the simulation of a ﬂow at Re =300. In Figure 13, it can be seen that the coarser grid, Grid1,
causes larger ﬂuctuation in the drag force. At some points around t =3.5, unexpected disturbances
appeared that may trigger an earlier oscillation in the later stage of simulation. The result from
the use of Grid3 shows no signiﬁcant change compared with the use of Grid2, indicating at these
two resolutions, the simulations were converged.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 43
Figure 11. The vorticity contours at four instances of time at Re =200.
Figure 12. Sensitivity of time step on the drag coefﬁcients at Re =300.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
44 W. LI AND M. VEZZA
Figure 13. The effect of grid resolution on the drag coefﬁcient at Re =300.
5. CONCLUSIONS
A hybrid vortex method is proposed in this paper. This approach extends Lagrangian vorticity-
based methods by using a hexahedral grid system in place of the method of overlapping vortex
particles/blobs. The ﬁnite volume method is used to solve the three-dimensional incompressible
Navier–Stokes equations in conjunction with the use of a modiﬁed Biot–Savart formula for the
calculation of ﬂow velocity. This hybrid approach can guarantee the conservation of the vorticity
and mass within the entire ﬂow domain including the prevention of the bleeding of vorticity over
the surface of an immersed body.
A closed-formformula, initially proposed by Suh [11], has been used with essential modiﬁcations
to accomplish the numerical integration over the surfaces of a hexahedral element for the calculation
of ﬂow velocity. A fast algorithm is also used to accelerate the process of velocity evaluation.
The work presented in this paper is the preliminary research and development of this hybrid
approach. The initial application of the method to the case of a sphere at low Re indicates that
predicted drag and ﬂow patterns are in line with the expectations. Future works will concentrate
on extending the capability of this code through implementations of advanced differential schemes
in conjunction with parallization techniques.
ACKNOWLEDGEMENTS
The authors wish to acknowledge the ﬁnancial supports from EPSRC.
REFERENCES
1. Leonard T. Computing three-dimensional incompressible ﬂows with vortex elements. Annual Review of Fluid
Mechanics 1985; 17:523.
2. Marshall JS, Grant JR, Gossler AA, Huyer SA. Vorticity transport on a Lagrangian tetrahedral Mesh. Journal of
Computational Physics 2000; 161:85–113.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd
A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 45
3. Cottet G-H, Koumoutsakos PD. Vortex Methods: Theory and Practice. Cambridge University Press: Cambridge,
2000.
4. Winckelmans GS. Vortex methods. Encyclopaedia of Computational Mechanics, vol. 3. Wiley: New York, 2004.
5. Van Dommelen L, Rundensteiner EA. Fast, adaptive summation of point forces in the two-dimensional Poisson
equation. Journal of Computational Physics 1989; 83:126–147.
6. Draghicescu CI, Draghicescu M. A fast Algorithm for vortex blob interaction. Journal of Computational Physics
1995; 116:69–78.
7. Clarke NR, Tutty OR. Construction and validation of a discrete vortex method for two-dimensional in compressible
Navier–Stokes equations. Computers and Fluids 1994; 23:751–783.
8. Koumoutsakos P, Leonard A. High-resolution simulations of the ﬂow around an impulsively started cylinder
using vortex methods. Journal of Fluid Mechanics 1995; 296:1–38.
9. Shankar S, Van Dommelen L. A new diffusion procedure for vortex methods. Journal of Computational Physics
1996; 127:88–109.
10. Ploumhans P, Winkelmans GS, Salmon JK, Leonard A, Warren MS. Vortex methods for direct numerical
simulation of three-dimensional bluff body ﬂows: application to the sphere at Re =300, 500 and 1000. Journal
of Computational Physics 2002; 178:427–463.
11. Suh J-C. The evaluation of the Biot–Savart integral. Journal of Engineering Mathematics 2000; 37:375–395.
12. Leonard BP. A stable and accurate convection modelling procedure based on quadratic upstream interpolation.
Computer Methods in Applied Mechanics and Engineering 1979; 19:59–98.
13. Koumoutsakos P, Leonard A, Pepin F. Boundary conditions for viscous vortex methods. Journal of Computational
Physics 1994; 113:52–61.
14. Qian L. Towards numerical simulation of vortex-body interaction using vorticity-based methods. Ph.D. Thesis,
University of Glasgow, Glasgow, 2001.
15. Johnson TA, Patel VC. Flow past a sphere up to a Reynolds number of 300. Journal of Fluid Mechanics 1999;
378:19.
16. Tomboulides G, Orszag SA. Numerical investigation pf transitional and weak turbulent ﬂow past a sphere. Journal
of Fluid Mechanics 2000; 416:45.
17. Schiller L, Naumann A.
¨
Uber die grundlegenden Berechungen bei der Schwerkraftaubereitung. Vereines Deutscher
Ingenieure 1933; 77:318.
Copyright q 2007 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2008; 57:31–45
DOI: 10.1002/ﬂd

and is therefore quite slow and usually prohibitive for large N . 10]. LI AND M. 57:31–45 DOI: 10. A hierarchical structure is commonly used in these acceleration methods such that the particle– particle interaction is converted to the particle–group or group–group interaction. Ltd. Numer. However. A modiﬁed Biot–Savart formula is used to calculate the velocity within the parts of the grid system where non-zero vorticity exists. (2) is the kinematic viscosity. 6]. which necessitates a complicated remeshing process [8. Instead of the random walk approach that has a shortcoming of slow-convergence rate. One of these problems is the computational cost in evaluating the velocity of vortices via the Biot–Savart law. the integration used to compute velocity with N vortices is O(N 2 ). In one of these methods. VEZZA vorticity-containing regions. The use of hexahedral elements requires a proper numerical integration procedure for calculating the velocity. in which the circulations associated with the vortex particles irregularly distributed in the ﬂow are redistributed either by the use of an integral representation for the Laplacian operator or a more complicated formula given in [9]. so that they appear to be efﬁcient and self-adaptive [1–4]. Int. Another problem associated with Lagrangian vorticity methods is the simulation of viscous diffusion. after a period of evolution. Fluids 2008. This situation has been improved by introducing several acceleration methods. 2. Owing to the nature of the ﬁnite volume method. Instead of employing overlapping vortex particles/blobs. such as the fast summation algorithm. the vortex particle distribution may become uneven. This has been accomplished by using closed-form formulas proposed by Suh [11] with essential completion from this study. However. despite these advantages. The application of this method to the ﬂow around a circular cylinder and aerofoil [7] shows that smoother results can be obtained if the vortex particles strictly overlap. with the overall computational complexity reduced from O(N 2 ) to O(N ln N ) or O(N ). a hexahedral grid system is introduced to solve the three-dimensional incompressible Navier–Stokes equations using the ﬁnite volume method. the concept of diffusive velocity is introduced by deﬁning an equivalent convection velocity for the diffusion process. Lagrangian vorticity methods exhibit little or no numerical dissipation. which is based on the expansion of the kernel function in the Biot–Savart formula [5. The current paper introduces a new hybrid approach to resolve some of the remaining difﬁculties associated with the Lagrangian vorticity methods. a variety of deterministic methods have been developed over the last two decades.1002/ﬂd . PROBLEM STATEMENT The Navier–Stokes equations governing incompressible ﬂow can be written in primitive variable (velocity–pressure) form as * 1 (1) u + u · ∇u = − ∇ p + ∇ 2 u *t ∇ · u=0 where u(x. Lagrangian vorticity methods experience some challenging problems that hinder their application in general ﬂuid dynamics. Since the nonlinear convection term is included in the vorticity time derivative. In this process.32 W. Another deterministic method is the particle strength exchange scheme [8]. J. thus destroying the overlapping condition. Meth. t) is the velocity ﬁeld and Copyright q 2007 John Wiley & Sons. this hybrid approach can guarantee the conservation of the vorticity and mass within the entire ﬂow domain.

Ltd.1002/ﬂd . 1 2 3] T (9) Int. n =x · n=0 (4) and a Neumann condition on the two tangential components of x (cancellation of the tangential components of the slip velocity at the wall).A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 33 Taking the curl of Equation (1) yields the unsteady vorticity transport equation * x + u · ∇x − x · ∇u = ∇ 2 x *t (3) for the vorticity ﬁeld x(x. The velocity evaluation is carried out using a combination of two methods. t) = ∇ × u(x. the direct integration using numerical integral and the indirect integration using fast summation. including a Dirichlet condition on the normal component of x [10] (as the tangential derivatives of the velocity vanish at the wall). A ﬁnite volume method is used to solve the vorticity transport equations. J. Meth. NUMERICAL IMPLEMENTATION The governing equations as well as the boundary conditions presented in the previous section are globally coupled. There is also a classical technique used to evaluate the force by computing the time derivative of the linear impulse (external ﬂow around one body): F where is the density and I is deﬁned by I= 1 2 x × x dV V =− dI dt (6) (7) 3.1. The force F on an immersed solid body with surface S can be calculated by the sum of the pressure and viscous shear forces as F=− S ( pn + n × x) ds (5) where n is the outward normal of surface S. 57:31–45 DOI: 10. Finite volume scheme The vorticity transport equation (3) can be written in the integral form: * *t W dV + V S F(W ) · n dS = 0 (8) where n is the unit normal vector of the surface S of the control volume V . Fluids 2008. W =[ Copyright q 2007 John Wiley & Sons. t). while the boundary conditions are implemented by using vortex sheets on the surface of the solid body. In viscous ﬂow. 3. Numer. a non-slip boundary condition must be satisﬁed on solid surfaces.

57:31–45 DOI: 10. Numer. LI AND M. J. and a system of differential equations is obtained by applying Equation (8) to each cell separately. Copyright q 2007 John Wiley & Sons. ⎡ ⎥ ⎢ F(W ) = ⎣ F2 (W ) ⎦ F3 (W ) ⎡ ⎢ ⎢ ⎢ ⎢ F1 (W ) = ⎢ u 1 ⎢ ⎢ ⎣ u1 ⎡ ⎢ u2 ⎢ ⎢ ⎢ F2 (W ) = ⎢ ⎢ ⎢ ⎢ ⎣ u2 ⎡ ⎢ u3 ⎢ ⎢ ⎢ F3 (W ) = ⎢ u 3 ⎢ ⎢ ⎣ 1 − u1 2− F1 (W ) ⎤T (10) ⎤ − 2 * 1 *x − u2 − u3 3 ⎥ ⎥ ⎥ * 2⎥ ⎥ − 1 *x ⎥ ⎥ * 3⎦ 1− *x ⎤ * 1 *y ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ * 3⎦ *y (11) − 3 * 2 *y 2 (12) − u3 − ⎤ * 1 *z ⎥ ⎥ ⎥ * 2⎥ ⎥ − u2 3 − 2 *z ⎥ ⎥ ⎦ * 3 − *z 1 − u1 3 − (13) The 3D computational domain is divided into hexahedral cells as shown in Figure 1. Ltd. Meth.1002/ﬂd . The typical hexahedral cell in the domain. Int.34 W. VEZZA Figure 1. Fluids 2008.

Direct integration method. Fluids 2008.2. j. j. n−1 n−1 . The slip velocity at each panel control point is induced by the free stream. Therefore. To implement this boundary condition. 13]. j.k+1/2 − F · S|i. surface panels and all wake vortex elements within the ﬂow. j. Suh [11] proposed an elegant method for the evaluation of Biot–Savart integral in both two and three dimensions. 3.k−1/2 (15) (14) where S|i+1/2. The evaluation of Biot– Savart integral is required to calculate the velocity ﬁeld. Meth.A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 35 Let the values of the quantities associated with each cell be denoted by i.k − F · S|i. a third-order QUICK (quadratic upwind interpolation for convective kinematics) [12] scheme is used. j−1/2.1. Vorticity boundary conditions The vorticity boundary conditions employed in the present paper are based on vortex sheet method [10. k) and cell (i + 1.k =( n i h)i. j.1002/ﬂd *x Dc(x) = t *n (18) . k (values at the cell centre). Velocity evaluations 3.u ) − Q i. A second-order Adams–Bashforth scheme is used to advance in time: ( n+1 i h)i. The vortex strength c(x) is computed by solving a linear system after evaluating the slip velocity on all the panels. J. the vorticity ﬁeld is represented by a structured grid system consisting of a number of hexahedral elements. k). j. j.k (tn−1 . j. In the present method. 57:31–45 DOI: 10.k (tn . a robust and accurate algorithm is needed for the integration over a hexahedral element. It transforms the volume integral into line integrals based on Gauss and Stokes integral theorems and can be applied to Copyright q 2007 John Wiley & Sons. 2 n n i . For the convective terms. j.k − F · S|i−1/2.k + F · S|i.3.k + t [3Q i.3. Numer. Q(W )i jk = Q i jk + F · S|i+1/2. u )] i (16) The diffusive ﬂux term is discretized using central differential scheme. Int. A piecewise-continuous vortex sheet is used to replace the body surfaces in order to cancel the slip velocity.k + F · S|i. j. j.k is the normal area on i direction of the interface between cell (i. j+1/2. For each cell Equation (8) can be presented as the form d (h i jk Wi jk ) + Q(W )i jk = 0 dt where h is the volume of the cell and Q is the ﬂux term. Ltd. The strength of the sheet is obtained from a Fredholm boundary integral equation of the second kind: 1 1 Dc(x) × n + 2 4 1 · (x − x ) × Dc(x ) dx = u slip |x − x |3 (17) S where c(x) is the strength of the vortex sheet and uslip is the slip velocity on the surface. the body surface is discritized using the surface mesh as vortex panels. Then the vortex ﬂux to be emitted into the ﬂow can be obtained as m 3. j.

a local coordinate system ( . Meth. after applying this approach to the current study. it was found that some limitations need to be imposed onto the original expressions presented in [11]. Numer. the induced velocity due to a vorticity distribution over an element bounded by planar panels can be expressed as u= 1 ∇× 4 1 x dV = r 4 x×∇ V V 1 1 dV = x × n j r 4 Ns j=1 Sj 1 dS r (19) where Ns is the total number of planar panels bounding the element and n j is the unit normal of the jth panel. Hence. otherwise. J. in some cases. p) 2 + 2 p +e = I1 + I2 (27) Int. the indeﬁnite integral in Equation (26) can be expressed as d ( − Copyright q 2007 John Wiley & Sons. According to the Biot–Savart law. For the evaluation of the line integrals. 57:31–45 DOI: 10. it can be expressed as B dC = C 4 k=1 lk B · lk dl (23) where lk is the unit directional vector of side lk .1002/ﬂd . Ltd. Now 1 = en · (∇ × B) (20) r en × r B= (21) r + en · r where en = ± n such that e = en · r 0 then Sj 1 dS = ± r n · (∇ × B) dS = ± Sj C B dC (22) where C is the perimeter of the planar panel. as shown in Figure 2. one or two terms would produce inﬁnite values. For a quadrilateral panel. Now we have r = rk + lk and 2 r = rk − 2 p (24) + 2 (25) because (n × r) · lk = r · (lk × n) = rk · pk . VEZZA elements with arbitrary number of faces (sides). ) is used in the panel. The other change to the original expressions is that the vorticity distribution within the elements is simpliﬁed to be uniform in this study rather than Suh’s linear distribution. LI AND M. However. Fluids 2008.36 W. B · lk dl = rk · pk lk 0 lk d ( − p) 2 + 2 p +e (26) After some manipulations.

For a large number of elements.A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 37 P rk r r k+1 k pk k+1 ξ Figure 2.3. based on the Taylor expansion [5. The direct integration method presented in the last section requires calculations of 12 logarithmic and angular functions for a single hexahedral element. Deﬁnition of the local coordinate system. A hierarchical grid system is ﬁrstly introduced to generate a box tree. Then each box of the tree is divided into two offspring boxes with equal size by cutting the longest side of the box. 57:31–45 DOI: 10. J. Copyright q 2007 John Wiley & Sons. for 0< <lk I2 = where ⎛ = arcsin ⎝ (2) If < and 0< < p. Numer. looking at two different cases: (1) If < p or > p . for 0< I2 = − e2 ( − ) (31) 3. This process continues until the number of vortex particles in the smallest box. Indirect integration method. 6]. Int. Ltd. where I1 = ln (lk − p ) + rk+1 rk − p (28) I2 requires some special treatment. The root of the tree is the single box containing all the vortex particles in the ﬂow. is less than a prescribed value. the time required for direct integration becomes excessive. Fluids 2008. Therefore.2. a fast summation algorithm is used in this paper. Meth. 2 p e 2 p − e2 ⎞ ⎠ p <lk (29) − e2 [ 2l p k 2 (e p + e(lk − p )rk +e p rk+1 ] + rk+1 )(e + rk ) p (30) p for 0< p <lk or > e 2 p and p< <lk . the top of the box tree.1002/ﬂd .

y )[(x2 Ak − B k − x3 C k + D k ). The vorticity associated with the surface slip is then distributed to the boundary cells through the diffusion process (Equation (18)). The calculation was carried out on a DELL PRECISION 470 workstation with two Intel Xeon (TM) CPU. Ltd. J. A single-block structured grid (Figure 3) has been created around the sphere. When Re>290 [15]. their inﬂuence at a ﬁx point x can be expressed as 1 k D y f (x. which is oriented in the positive x direction. The surface slip velocity is recomputed at the beginning of the next time step to provide a continuing source of vorticity. the ﬂow remains steady but is no longer axisymmetric. LI AND M. (33) (34) |k| −1 (x3 E k − F k − x1 Ak + G k ). 57:31–45 DOI: 10.38 For points y in a single box f (x. k k! = k1 !k2 !k3 ! k k k D k = * /*y1 1 *y2 2 *y3 3 . VEZZA centred at y . 4. The number of hexahedral elements varies from a resolution of 36 000 to 64 000 and 108 000. In the range of Re<210–212 [15. k2 . y) y=y (y − y )k −1 k! (32) x−y = |x − y|3 |k| |k| = k1 + k2 + k3 . |k| W. (x1 C k − H k − x2 E k + I k )] The sums in the above equations are evaluated for all boxes and indices k before the velocity calculation and in parallel with the vortex sorting and grid generation process [14]. The ﬂow variables are normalized using the diameter of the sphere and the free-stream velocity. Meth. Above that. Numer. k3 ). and 300) with two different ﬂow ﬁeld resolutions (medium and high). and the outer radius of the domain was R1 = 5. Fluids 2008. y k k y k = y1 1 y2 2 y3 3 The degree of its Taylor polynomial is − 1. the ﬂow becomes unsteady but still retains time periodicity and planar symmetry. FLOW PAST A SPHERE The hybrid vortex method presented in previous sections is utilized for the computation of ﬂow past a sphere. Int.0. The ﬂow past a sphere behaves differently when Reynolds number (Re) changes.1002/ﬂd . An impulsive start of the ﬂow is initialized by computing the potential ﬂow past the sphere body using the vortex sheet method (Equation (17)). if Re<270. The radius of the sphere was taken as R0 = 0. 16]. Two (100 and 200) of the Copyright q 2007 John Wiley & Sons. the ﬂow remains steady and axisymmetric although separated. Then the calculation of the velocity u(x) at point x is given by u(x) = − 1 4 ak (x. The ﬂow is started impulsively at various Reynolds numbers (Re) (100.5. y) = where k = (k1 . 200.

4.2. This feature can be seen more clearly from a three-dimensional view shown in Figure 10. Ltd. However. Numer.A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 39 Figure 3. 4.1. For all these three Reynolds numbers. 200. three Reynolds numbers simulated in this study are below 210. The structured grid used for the sphere ﬂow simulation. with the only difference being the length of the separation bubble and the position of the vortex. Flow characteristics Figures 7–9 show the streamlines in the x–y plane at Reynolds number 100.15 Re0. Again. J. the ﬂow is seen to separate from the surface of the sphere and form a separation bubble in the wake. The curves of the drag coefﬁcient appear to exhibit more ﬂuctuation when Re increases. corresponding to the axisymmetric ﬂow. The drag coefﬁcients are observed to have a quick decline from the initial overshoot in the range 0<t<0. Force coefﬁcient history Figures 4–6 show the drag coefﬁcients Cd = D/(0. even at the early stage of simulation.5. The drag coefﬁcient results are compared with the values from the Schiller–Naumann formula [17] Cd = (24/Re)[1 + 0.5 U 2 R 2 ) plotted as a function of time. then to approach a steady-state average value. Meth. The other one (300) was used to test the potential of the hybrid vortex method.687 ] and the results from Johnson and Patel [15]. 57:31–45 DOI: 10. the vorticity | | contours in the x–y plane at several instances for Re = 200 are presented. 200. and 300. Copyright q 2007 John Wiley & Sons. and 300. only the early stage of the unsteady ﬂow has been simulated. the ﬂow has started to show asymmetrical characteristics and unsteadiness. at three Reynolds number 100. Fluids 2008. the axisymmetric ﬂow characteristics can be observed through the development of the separated ﬂow. Int. for Re = 300. but due to the capability of the workstation. In Figure 11.1002/ﬂd . A clear axisymmetric ﬂow pattern can be seen in both ﬂows at Re = 100 and 200.

4. Ltd. on total drag coefﬁcients is presented for Re = 300. showing computed results (solid line). Fluids 2008. LI AND M. the sensitivity of the time step to the computation of drag coefﬁcient can be observed. Drag coefﬁcient comparison for the ﬂow past a sphere at Re = 200. the effect of numerical parameters. in terms of time steps and grid resolution. Numer. Int. J.1002/ﬂd . and results from Johnson and Petel (dotted line). 57:31–45 DOI: 10. dt = 0. The use of smaller time steps. Schiller–Naumann formula (dashed line). Effects of numerical parameters In Figures 12 and 13. VEZZA Figure 4.3. Schiller–Naumann formula (dashed line). Meth.005 Copyright q 2007 John Wiley & Sons. From Figure 12. Figure 5. showing computed results (solid line). Drag coefﬁcient comparison for the ﬂow past a sphere at Re = 100.40 W. and results from Johnson and Petel (dotted line).

Schiller–Naumann formula (dashed line). J. Numer.A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 41 Figure 6. showing computed results (solid line). Ltd.1002/ﬂd . Int. Streamlines of sphere ﬂow in the x–y plane at Re = 200. Figure 7. Drag coefﬁcient comparison for the ﬂow past a sphere at Re = 300. Figure 8. Copyright q 2007 John Wiley & Sons. Fluids 2008. 57:31–45 DOI: 10. Meth. Streamlines of sphere ﬂow in the x–y plane at Re = 100. and results from Johnson and Petel (dotted line).

LI AND M. In Figure 13. unexpected disturbances appeared that may trigger an earlier oscillation in the later stage of simulation. Fluids 2008. appears to be less accurate. Copyright q 2007 John Wiley & Sons. The use of the moderate time step. Int. Streamlines of sphere ﬂow in the x–y plane at Re = 300. J. Meth. namely Grid1 (total grid number is 36 000) and Grid2 (total grid number is 64 000) and Grid3 (total grid number is 108 000). causes larger ﬂuctuation in the drag force.42 W. Ltd.1002/ﬂd . produced nearly identical drag values near the immediate time region (t = 0+) after the impulsive start. dt = 0. The result from the use of Grid3 shows no signiﬁcant change compared with the use of Grid2. Figure 10. are used separately for the simulation of a ﬂow at Re = 300. VEZZA Figure 9. Grid1.5. it can be seen that the coarser grid. although it is preferred for the long-term simulation. and 0. indicating at these two resolutions. the simulations were converged.01. The velocity contour is also shown. Numer.0025. At some points around t = 3. Three-dimensional view of the streamlines of sphere ﬂow in the x–y plane at Re = 300. three grids. To evaluate the effect of grid resolution. 57:31–45 DOI: 10.

57:31–45 DOI: 10. Numer. Copyright q 2007 John Wiley & Sons. Meth.A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 43 Figure 11. Int. Fluids 2008. Ltd. J. The vorticity contours at four instances of time at Re = 200. Figure 12.1002/ﬂd . Sensitivity of time step on the drag coefﬁcients at Re = 300.

Future works will concentrate on extending the capability of this code through implementations of advanced differential schemes in conjunction with parallization techniques. VEZZA Figure 13. ACKNOWLEDGEMENTS The authors wish to acknowledge the ﬁnancial supports from EPSRC. Grant JR. The ﬁnite volume method is used to solve the three-dimensional incompressible Navier–Stokes equations in conjunction with the use of a modiﬁed Biot–Savart formula for the calculation of ﬂow velocity. Numer. LI AND M. initially proposed by Suh [11]. This approach extends Lagrangian vorticitybased methods by using a hexahedral grid system in place of the method of overlapping vortex particles/blobs.44 W. 5. A fast algorithm is also used to accelerate the process of velocity evaluation. 161:85–113. Leonard T. Copyright q 2007 John Wiley & Sons. J. Meth. Computing three-dimensional incompressible ﬂows with vortex elements. Ltd. 17:523. This hybrid approach can guarantee the conservation of the vorticity and mass within the entire ﬂow domain including the prevention of the bleeding of vorticity over the surface of an immersed body. has been used with essential modiﬁcations to accomplish the numerical integration over the surfaces of a hexahedral element for the calculation of ﬂow velocity. The initial application of the method to the case of a sphere at low Re indicates that predicted drag and ﬂow patterns are in line with the expectations. Gossler AA. A closed-form formula. 2. Annual Review of Fluid Mechanics 1985. Vorticity transport on a Lagrangian tetrahedral Mesh. 57:31–45 DOI: 10. CONCLUSIONS A hybrid vortex method is proposed in this paper. Fluids 2008. REFERENCES 1. Journal of Computational Physics 2000. The effect of grid resolution on the drag coefﬁcient at Re = 300. Marshall JS.1002/ﬂd . The work presented in this paper is the preliminary research and development of this hybrid approach. Huyer SA. Int.

Ltd. Salmon JK. 3. Fast. A fast Algorithm for vortex blob interaction. Uber die grundlegenden Berechungen bei der Schwerkraftaubereitung. 8. 15. Leonard A. Journal of Computational Physics 1995. Rundensteiner EA. Vereines Deutscher Ingenieure 1933. Patel VC. Orszag SA. 127:88–109. Journal of Computational Physics 1994. 378:19. Thesis. Tutty OR. Cambridge University Press: Cambridge. Journal of Computational Physics 1996. 113:52–61. Koumoutsakos P. A new diffusion procedure for vortex methods. 77:318. Boundary conditions for viscous vortex methods. 14. Winckelmans GS. A stable and accurate convection modelling procedure based on quadratic upstream interpolation. Numer.D. Construction and validation of a discrete vortex method for two-dimensional in compressible Navier–Stokes equations. Ploumhans P. Warren MS. ¨ 17. Glasgow. Koumoutsakos P. 16. Journal of Engineering Mathematics 2000. adaptive summation of point forces in the two-dimensional Poisson equation. Suh J-C. Computer Methods in Applied Mechanics and Engineering 1979. University of Glasgow. 13. 500 and 1000. 19:59–98. Draghicescu CI. Pepin F. 5. 6. 7. Vortex methods for direct numerical simulation of three-dimensional bluff body ﬂows: application to the sphere at Re = 300.1002/ﬂd . High-resolution simulations of the ﬂow around an impulsively started cylinder using vortex methods. 178:427–463. 83:126–147.A HYBRID VORTEX METHOD FOR THE SIMULATION OF THREE-DIMENSIONAL FLOWS 45 3. J. Wiley: New York. Tomboulides G. 11. Van Dommelen L. Ph. Vortex Methods: Theory and Practice. Journal of Computational Physics 2002. Draghicescu M. Towards numerical simulation of vortex-body interaction using vorticity-based methods. 23:751–783. Computers and Fluids 1994. Int. Clarke NR. 2004. Vortex methods. Johnson TA. Naumann A. 37:375–395. Journal of Fluid Mechanics 1995. Numerical investigation pf transitional and weak turbulent ﬂow past a sphere. Cottet G-H. 416:45. 9. 2001. Encyclopaedia of Computational Mechanics. 4. Journal of Computational Physics 1989. The evaluation of the Biot–Savart integral. 116:69–78. Schiller L. Flow past a sphere up to a Reynolds number of 300. 10. Leonard BP. Leonard A. Journal of Fluid Mechanics 2000. Winkelmans GS. vol. Meth. 12. Shankar S. Journal of Fluid Mechanics 1999. 57:31–45 DOI: 10. Van Dommelen L. 2000. Koumoutsakos PD. 296:1–38. Fluids 2008. Copyright q 2007 John Wiley & Sons. Leonard A. Qian L.