JOURNAL OF COMPUTATIONAL PHYSICS

ARTICLE NO.

136, 214226 (1997)

CP975776

Modeling Low Reynolds Number Incompressible Flows Using SPH

Joseph P. Morris, Patrick J. Fox, and Yi Zhu
School of Civil Engineering, Purdue University, West Lafayette, Indiana 47907
E-mail: [email protected]
Received December 16, 1996; revised June 12, 1997

The method of smoothed particle hydrodynamics (SPH) is extended to model incompressible flows of low Reynolds number.
For such flows, modification of the standard SPH formalism is required to minimize errors associated with the use of a quasi-incompressible equation of state. Treatment of viscosity, state equation,
kernel interpolation, and boundary conditions are described. Simulations using the method show close agreement with series solutions for Couette and Poiseuille flows. Furthermore, comparison
with finite element solutions for flow past a regular lattice of cylinders shows close agreement for the velocity and pressure fields.
The SPH results exhibit small pressure fluctuations near curved
boundaries. Further improvements to the boundary conditions may
be possible which will reduce these errors. A similar method to that
used here may permit the simulation of other flows at low Reynolds
numbers using SPH. Further development will be needed for cases
involving free surfaces or substantially different equations of
state. Q 1997 Academic Press

1. INTRODUCTION

The study of incompressible fluid flows in which viscous

forces are either comparable with or dominate inertial
forces has applications to many physical problems. Industrial, biological, and environmental processes often involve
flows of low Reynolds number (Re). Many applications
to the fields of environmental, mechanical, chemical, and
petroleum engineering likewise involve slow viscous incompressible flows through filters, substrates, porous materials, and other potentially deformable structures.
Smoothed particle hydrodynamics (SPH), while originally developed for astrophysical applications [1, 2], has
been applied successfully to a wide range of problems.
SPH is a fully Lagrangian technique in which the numerical
solution is achieved without a grid. An advantage of SPH
is the relative ease with which new physics may be incorporated into the formulation. It is also straightforward to
allow boundaries to move or deform and to model the
interaction of several fluid phases bounded by a free surface. SPH has certain advantages over other fluid dynamical methods, which may, for example, encounter difficulty
with deformable boundaries, multiphase fluids, free surfaces, and the extension to three dimensions. While SPH
is a versatile method, errors can sometimes be larger than
214
0021-9991/97 $25.00
Copyright 1997 by Academic Press
All rights of reproduction in any form reserved.

those obtained using grid-based methods tailored for specific problems. Moreover, SPH can be more computationally expensive than alternative techniques for a given application.
For typical astrophysical applications, SPH is used to
model compressible fluids at high Reynolds number (usually Re $ 103 [4]). Although SPH has been used to model
free surface flows of inviscid incompressible fluids [3], and
some research has been performed using SPH for compressible gases with Reynolds numbers down to five [5],
low Reynolds number (Re # 1) incompressible flows have
not been modeled using SPH. The objective of this paper
is to present modifications of the standard SPH formalism
needed to simulate such flows. Treatment of viscosity,
equation of state, kernel interpolation, and boundary conditions are described. Simulations using the method show
close agreement with series solutions for Couette and
Poiseuille flows and with finite element solutions for flow
past a regular lattice of cylinders.
2. THE METHOD

2.1. Background
Using SPH the fluid is represented by particles, typically
of fixed mass, which follow the fluid motion, advect contact
discontinuities, preserve Galilean invariance, and reduce
computational diffusion of various fluid properties including momentum. The equations governing the evolution of
the fluid become expressions for interparticle forces and
fluxes when written in SPH form. Using the standard approach to SPH [6, 7], the particles (which may also be
regarded as interpolation points) move with the local fluid
velocity. Each particle carries mass m, velocity v, and other
fluid quantities specific to a given problem. The equations
governing the evolution of fluid quantities are expressed
as summation interpolants using a kernel function W with
smoothing length h. For example, the density at particle
a, ra , may be evaluated using

ra 5

Om W
b

ab ,

(1)

MODELING LOW RE INCOMPRESSIBLE FLOWS USING SPH

where Wab denotes

Wab 5 W (rab , h)

(2)

215

usually only important for faster flows involving shocks.

The simulations presented here, therefore, use (6) to
evolve density.
2.3. Dynamic Pressure

and
rab 5 ra 2 rb ,

(3)

where ra denotes the position of particle a. The kernel

typically takes the form
W (rab , h) 5

S D

urab u
1
f
,
s
h
h

(4)

where s is the number of dimensions and the function f

is discussed in Section 2.7. Other expressions for quantities
at the particles are obtained by summation involving the
kernel or its derivatives. For example, the most common
SPH expression for the pressure gradient term is
2

S D O S
1
=p
r

52

mb

pa pb
1
=W ,
r 2a r 2b a ab

2.2. Evolution of Density

If (1) is used when modeling incompressible free-surface
flows, particle density is smoothed out at the edge of the
fluid and spurious pressure gradients are induced at the
surface. To avoid this problem, Monaghan [3] initially set
the density to a reference value and evolved particle densities according to the following SPH equation for continuity:

Om v

b ab ? =aWab .

pd 5 pt 2 ph ,

(7)

where pt and ph are the total and hydrostatic pressures,

respectively. Since pressure appears in the NavierStokes
equations only as a gradient, the effect of ph is that of a
body force,

(5)

where pa is the pressure at particle a and =a denotes the

gradient with respect to the coordinates of particle a. For a
kernel of the form in (4), this pressure gradient formulation
conserves momentum exactly, since forces acting between
individual particles are antisymmetric.
The following sections consider specific aspects of the
SPH formalism and the approaches necessary to simulate
incompressible flows at low Reynolds number.

dra
5
dt

For low Reynolds number flows, local variations in the

pressure gradient which force the fluid motion can be very
small, compared with the hydrostatic pressure gradient.
This is of special significance to SPH, since pressure is
obtained using an explicit function of density and is only
accurate to about 1% (see the following section). For many
problems it is simpler to use SPH to model the dynamic
pressure, pd , defined as

(6)

Our test cases do not involve free surfaces, and thus (1)
may be used to evolve particle densities. One disadvantage
of this approach is that r must be evaluated by summing
over the particles before other quantities may be interpolated. However, (6) allows r to be evolved concurrently
with particle velocities and other field quantities, thus significantly reducing the computational effort. While (6)
does not conserve mass exactly ((1) does, provided the
total number and mass of particles are constant), this is

1
1
1
2 =pt 5 2 =pd 2 =ph ,
r
r
r

(8)

1
5 2 =pd 1 F,
r
where F is a force per unit mass. Using this approach,
pressure gradient driven flow through a periodic lattice can
be easily simulated (see Section 3.3). Periodic boundary
conditions are applied to all fluid quantities and the flow
is driven by the effective body force. The dynamic pressure
is modeled using the equation of state (see (10) in Section
2.4). For simplicity, p is used in the following sections to
denote the dynamic pressure pd .
2.4. Equation of State
Most grid-based techniques model the flow of water
as incompressible since the speed of sound in water is
usually large compared with bulk fluid motions (i.e., a very
low Mach number). Using SPH, fluid pressure is an explicit
function of local fluid density and particle motions are
driven by local density gradients. Therefore, it is necessary
to use a quasi-incompressible equation of state to model
such flows with SPH. Since the actual equation of state for
water would require a prohibitively small time step for
stability (by the CFL condition [8]), an artificial state equation is used. The chosen sound speed is low enough to be
practical, yet high enough to limit density fluctuations to
within 3%. A similar approach has been used in grid-based
techniques to model steady incompressible flow [911].
Previous applications of SPH to incompressible flows

216

MORRIS, FOX, AND ZHU

[312] have employed a state equation suggested by Batchelor [13],

p 5 p0

HS D J
r
r0

21 ,

(9)

where c 5 7 and a zero subscript denotes reference quantities. The choice of c 5 7 in (9) causes pressure to respond
strongly to variations in density. Thus, perturbations to the
density field remain small, even at high Reynolds numbers.
However, as the density fluctuations increase, small errors
in density correspond to increasingly larger errors in pressure. For lower Reynolds numbers, more accurate pressure
estimates are obtained using SPH if c is taken to be unity
(as it is in [911]), since errors in density and pressure
remain proportional.
In previous work involving incompressible fluids, the
subtraction of 1 in (9) was introduced to remove spurious
boundary effects at a free surface. It is well established
that SPH is unstable when attractive forces act between
particles [1417, 29]. Consequently, for the simulations
presented in Section 3, this subtraction was found to lead to
numerical instabilities in regions of sustained low pressure.
Since the test simulations (and many applications) have
particles filling all space, this work uses
p 5 c 2r,

(10)

where c is the speed of sound.

The sound speed must be chosen carefully to ensure
both an efficient and accurate solution of a given problem.
The value of c must be large enough that the behavior of
the corresponding quasi-incompressible fluid is sufficiently
close to that of the real fluid, yet it should not be so large
as to make the time step prohibitively small. Using a scale
analysis, Monaghan [3] argued that, for density to vary by
at most 1%, the Mach number of the flow should be 0.1
or less. In fact, for typical smoothing lengths used with
SPH, kernel interpolation is only accurate to within approximately 1%. The principal cause of this variation is
small fluctuations in density which inevitably occur as particles move past one another. Thus, pressure gradients obtained using a high sound speed are potentially noisy. Nevertheless, the velocities obtained are accurate if smoothed
either by XSPH [18] or viscosity. For many applications,
a close estimate of bulk fluid motion is sufficient. However,
for other problems, accurate estimates of the pressure field
are needed. We found that computed pressures are in close
agreement with other techniques when c is chosen such
that the density varies by at most 3%.
It is possible to estimate an appropriate value of c by
considering the balance of forces in the NavierStokes
momentum equation,

dv
1
5 2 =p 1 n =2v 1 F,
dt
r

(11)

where n is the kinematic viscosity and F is a body force

per unit mass. Substituting a velocity scale V0 , a length
scale L0 , and a body force per unit mass of magnitude F
into (11), we find that the square of the sound speed should
be comparable with the largest of
c2 p

V 20 nV0 FL0
,
,
,
d L0d d

(12)

Dr
.
r0

(13)

where

d5

The first term in (12) corresponds to that derived by

Monaghan [3]. The second and third terms ensure that
pressure forces are comparable with viscous and body
forces, respectively. These conditions should only be regarded as a guide to an estimate of an appropriate sound
speed. After a simulation has been run initially at low
resolution and the actual variation in p is known, the value
of c can be changed to produce the desired density variation.
2.5. Boundary Conditions
Initial applications of quasi-incompressible SPH involved high Reynolds number simulations of free surface
flows interacting with free-slip boundaries. Such work employed boundary particles which exerted strong repulsive
forces to prevent SPH particles from penetrating solid surfaces [3, 19, 20]. To realistically model flows at lower Reynolds numbers, no-slip boundary conditions are needed. In
addition, for the free surface flows considered by Monaghan [3], boundary particles do not contribute to the
density of the free SPH particles, thus permitting the fluid
to freely leave a solid boundary with no pressure-driven
restoring force. For our work, boundary particles contribute to the density of fluid particles such that pressure decreases when fluid and boundary particles diverge. It is
possible to implement such a boundary condition using
image particles [21]. These images are created by reflecting
fluid particles across the boundary with opposite velocities.
This procedure works well for straight channels, but introduces density errors for curved surfaces. Takeda et al. [5]
achieved a no-slip boundary condition using special boundary terms which mimic a half-space filled with SPH image
particles. While these approaches have proved useful for
compressible and moderate to high Reynolds number
flows, we found they did not give sufficiently stable results
for our simulations.

217

MODELING LOW RE INCOMPRESSIBLE FLOWS USING SPH

In this work, actual SPH particles are used to represent a

no-slip boundary surface. These particles contribute to the
usual SPH expressions for density and pressure gradients,
and their own densities are also evolved. Evolving the densities of boundary particles was found to better capture peak
pressures than if boundary densities were kept constant.
Obstacles within the flow are created by placing all the particles on a regular lattice throughout the computational domain and designating those particles falling within a solid
object to be boundary particles. The advantage of this approach is that a quiet start is guaranteed since the particle
number density is initially constant throughout the computational domain. However, it can lead to an imperfect representation of a curved boundary at low resolution. An alternative would be to place boundary particles on the surface
of each obstacle. One difficulty with this method is filling
the fluid space with a distribution of free particles corresponding to constant density. In this case, the initial configuration can be relaxed to a quiet state before driving
forces are applied. The former approach was used in this
work for simplicity.
A first-order no-slip condition can be created if boundary
particles are moved with the velocity of the object to which
they are attached. However, this approach can produce significant errors in velocity near boundaries. Antisymmetry
can be approximated by extrapolating the velocity of free
particles through the boundary to the position of the boundary particles. Ideally, a local estimate of the velocity gradients at the surface of the boundary would be used to assign
these artificial velocities to interior points; however, such
estimates would require a second summation over the particles and, hence, a substantial increase in the computational effort.
This work uses a simpler approach which is stable, accurate, and requires little extra computation. The method is
similar to that used by Takeda et al. [5] to obtain a functional
form for the viscous force due to a solid boundary. Figure 1
illustrates the concept for a portion of a curved boundary.
For each fluid particle a, the normal distance da to the boundary is calculated. This normal is used to define a tangent
plane (a line in two dimensions) from which the normal distance dB to each boundary particle B is calculated. The velocity of particle a is extrapolated across the tangent plane,
assuming zero velocity on the plane itself, thus giving each
boundary particle velocity vB 5 2(dB /da)va . In practice, the
discrete arrangement of boundary particles may permit a
fluid particle to closely approach the nominal curve describing the boundary. In such circumstances, the magnitude of
vB must be restricted. Accordingly, the following formula is
used to calculate relative velocities between fluid and
boundary particles,
vab 5 bva ,
where

(14)

FIG. 1. Construction of artificial velocity for boundary particles to

simulate a no-slip boundary condition.

b 5 min bmax , 1 1

dB
.
da

(15)

Numerical simulations have shown that good results are

obtained if bmax is approximately 1.5. If the boundary is in
motion, va in (14) should be replaced by the fluid velocity
relative to the boundary. The artificial velocity vB is used to
calculate viscous forces (see Section 2.6)), but it is not used
to evolve boundary particle positions. Consequently, the actual boundary velocity is used in (6) such that densities remain consistent with (1). For concave surfaces, a similar
method could be used in which the tangent plane is constructed by considering the nearest point on the curve to
each boundary particle (rather than fluid particle). The two
methods give identical results for a plane surface.
There are alternative techniques which take boundary
curvature into account. For example, Monaghan [19, 20] introduced special boundary particle formulations in the context of high Reynolds number free-surface flows which explicitly account for local curvature effects when calculating
the contribution from boundary particles. This approach
can reduce the noise otherwise introduced by a discrete
boundary representation. Similar improvements to the
boundary conditions employed here, incorporating approaches taken by other authors for high Reynolds number
problems, are the focus of further investigation.
2.6. Viscosity
Most implementations of SPH employ an artificial viscosity first introduced to permit the modeling of strong
shocks [6, 7]. This viscosity is incorporated into the momentum equation
dva
52
dt

O m Srp 1 rp 1 P D = W
b

a
2
a

b
2
b

ab

ab ,

(16)

218

MORRIS, FOX, AND ZHU

where

2ace ab 1 be 2ab
, if vab ? rab , 0;
rab
Pab 5
0,

DJ O
a

(19)

rab Wab
urab u ra

(20)

and we can simplify (19) to

DJ O
a

b
2
b

ab

ab

(22)

ab

a b

(18)

mb (ea 1 eb)rab ? =aWab

vab ,
rarb (r 2ab 1 0.01h2)

=aWab 5

1
= ? e= v
r

a
2
a

ab

where Fa is the body force evaluated at particle a.

where e 5 rn is the dynamic viscosity. This hybrid expression combines a standard SPH first derivative with a finite
difference approximation of a first derivative. By taking a
Taylor expansion about particle a, it can be shown that this
expression is appropriate [26]. This formulation conserves
linear momentum exactly, while angular momentum is only
approximately conserved. If the kernel takes the form of
(4) then

HS

(17)

and rab is the average density of particles a and b. The

0.01h2 term is included to keep the denominator nonzero.
Although this formulation has been used to model real
viscosity [4, 22], it produced inaccurate velocity profiles for
our simulations. The advantage of (16) is that it guarantees
conservation of angular momentum, which is important
for applications involving relatively large fluid velocities
or an unbounded fluid edge. Since this work involves low
velocities and SPH particles fill all space, a more realistic
form of viscosity has been adopted. Previous expressions
used to calculate realistic viscous forces have typically involved nested summations over the particles [23, 24] (and,
hence, twice the computational effort), or have directly
employed second derivatives of the kernel [5]. The disadvantage of using second derivatives is that interpolation is
much more susceptible to error at low resolution (especially for low-order spline kernels [25]). Our method employs an SPH estimation of viscous diffusion which is similar to an expression used in [26] to model heat conduction,
1
= ? e= v
r

otherwise,

hvab ? rab
e ab 5 2
,
r ab 1 0.01h2

HS

O m Srp 1 rp D = W
m (e 1 e )v
1 W
1O
S
D1F ,
rr
r r

dva
52
dt

mb (ea 1 eb )vab 1 Wab

.
rarb
rab ra

Thus, the momentum equation is written as

(21)

2.7. The Kernel

There are many possible choices of the function f in (4).
Most SPH simulations employ a cubic spline kernel [27, 28],

f (s) 5

10
7f

1 2 3s2 /2 1 3s3 /4, if 0 # s , 1;

(2 2 s)3 /4,

if 1 # s , 2;

(23)

if s $ 2

0,

(here normalized for two dimensions), since it resembles

a Gaussian while having compact support. However, it has
been shown that SPH can be unstable to transverse modes
when kernels with compact support are used [14, 17, 29].
As higher-order splines more closely approximating a
Gaussian are employed, these instabilities are reduced.
One reason for the poor performance of lower-order
splines is that the stability properties of SPH depend
strongly upon the second derivative of the kernel. The
second derivative of the cubic spline is a piecewise-linear
function, and, accordingly, the stability properties are inferior to those of smoother kernels. For example, when the
quintic spline [27] is used,

f (s) 5

7
478f

(3 2 s)5 2 6(2 2 s)5 1 15(1 2 s)5,

0 # s , 1;

(3 2 s) 2 6(2 2 s) ,

1 # s , 2;

(3 2 s)5,

2 # s , 3;

0,

s $ 3,

(24)

transverse mode instabilities are negligible [14, 29]. For

most applications using a cubic spline, these instabilities are
not important since the resulting perturbations in density
typically peak at about 1%. However, for a quasi-incompressible equation of state, such variations are significant
(see Section 2.3). Thus, for simulations involving very low
Reynolds numbers, it was found that a cubic spline rapidly
produced significant noise in the pressure and velocity
fields, whereas simulations employing the quintic spline
remained stable. Although the quintic spline is computationally more intensive (by approximately a factor of 2),
it was found to be the most reliable for our simulations.

219

MODELING LOW RE INCOMPRESSIBLE FLOWS USING SPH

Morris [29] presents a discussion of alternative kernels

which have superior stability properties to standard cubic
splines with no extra computational expense. It is not clear,
however, if these kernels represent a significant improvement over the quintic spline for a given region of support
for this application.

velocities are evolved according to (22). To ensure stability

of the integration scheme, the time step is limited according
to the conditions set forth in Section 2.8. No-slip boundary
conditions are simulated by assigning artificial velocities to
the boundary particles using (14).
3. MODEL TESTING AND VERIFICATION

2.8. Time Integration

A modified Euler technique was used to perform the
time integration. For stability, several time step criteria
must be satisfied, including a CFL condition [8],
Dt # 0.25

h
,
c

(25)

and additional constraints due to the magnitude of particle

accelerations fa [7],
Dt # 0.25 min
a

SD
h
fa

1/2

h2
.
n

The first test case is Couette flow between infinite plates

located at y 5 0 and y 5 L. The system is initially at rest.
At time t 5 0, the upper plate moves at constant velocity
V0 parallel to the x-axis. The series solution for the timedependent behavior of this flow is
vx (y, t) 5

S D S

y 2V
V0
nf
n2f 2
0
y1
(21)n sin
y exp 2n 2 t ,
L
L
L
n51 nf
(28)

(26)

and viscous diffusion,

Dt # 0.125

3.1. Couette Flow

(27)

Equation (27) is based upon the usual condition for an explicit finite difference method simulating diffusion. At sufficiently high resolution (small h) or large viscosity, (27) is
typically the dominant time constraint. The choice of kernel
and the arrangement of particles influences the coefficients
in (25)(27). In particular, different splines can have different effective resolution lengths for the same value of h.
For example, use of a cubic spline (which is narrower than
a quintic for the same smoothing length) may require
slightly smaller coefficients in the above expressions.

where vx is the fluid velocity in the x-direction. The flow

was simulated using SPH for n 5 1026m2s21, L 5 1023m,
r 5 103kgm23, V0 5 1.25 3 1025ms21, and with 50 particles
spanning the channel. This corresponds to a Reynolds
number of 1.25 3 1022, using
Re 5

(29)

Figure 2 shows a comparison between velocity profiles

obtained using (28) and SPH at several times including the
steady state solution (t 5 y). The results are in close
agreement (within 0.5%), confirming the accuracy of the
approach used to evaluate viscous and boundary forces
with SPH. Lower resolution simulations completed with
20 particles spanning the channel were found to agree to
within approximately 2% of the series solution values.

2.9. Summary of the Method

A summary of the method to simulate quasi-incompressible flow at low Reynolds number using SPH can now be
presented. The particles are placed on a hexagonal lattice
filling all space and assigned a constant initial density ( r0 ).
Boundary particles are defined as those initially lying inside
obstacles within the flow field and beyond solid walls. The
equation of state for the fluid is given by (10) and the sound
speed c is chosen such that density fluctuations are at most
3%. In the case of periodic flow driven by an applied pressure
gradient, the dynamic pressure (7) is used and the applied
pressure differential is simulated by an applied body force
(8). For stability at low Reynolds numbers, a quintic spline
kernel is used with a smoothing length of 1.5 times the initial
nearest neighbor distance. Particle densities are evolved using (6) to reduce the computational effort, and the particle

V0L
.
n

3.2. Poiseuille Flow

The second test case is Poiseuille flow between stationary
infinite plates at y 5 0 and y 5 L. The fluid is initially at
rest and is driven by an applied body force F parallel to
the x-axis for t $ 0. The series solution for the transient
behavior is
vx (y, t) 5

y
4FL2
F
y(y 2 L) 1
3
3
2n
n50 n f (2n 1 1)

sin

D S

(30)

fy
(2n 1 1) f n
(2n 1 1) exp 2
t .
L
L2
2

SPH was used to simulate Poiseuille flow for n 5 1026m2s21,

L 5 1023m, r 5 103kgm23, F 5 1024ms22, and 50 particles

220

MORRIS, FOX, AND ZHU

FIG. 2. Comparison of SPH and series solutions for Couette flow (Re 5 1.25 3 1022).

spanning the channel. This corresponds to a peak fluid

velocity V0 5 1.25 3 1025ms21 and a Reynolds number of
1.25 3 1022 using (29). A comparison between velocity
profiles obtained using (30) and SPH appears in Fig. 3.
The results are again in close agreement, with the largest
discrepancy being about 0.7% for the steady state solution.
Lower resolution simulations completed with 20 particles

spanning the channel agreed to within approximately 2%

of the series solution values.
3.3. Flow through a Periodic Lattice of Cylinders
The Couette and Poiseuille simulations tested the interaction between viscous and body forces and the effective-

FIG. 3. Comparison of SPH and series solutions for Poiseuille flow (Re 5 1.25 3 1022).

MODELING LOW RE INCOMPRESSIBLE FLOWS USING SPH

221

FIG. 5. Paths for comparison of SPH and FEM solutions.

FIG. 4. Single cylinder within a periodic lattice.

ness of the no-slip boundary condition in the SPH model.

However, these flows are one dimensional and do not
produce variations in dynamic pressure. A more challenging test of the method involves flow through a square lattice
of cylinders. This particular configuration has been studied
extensively [3032] as a simple model of flow through
fibrous porous media. To solve this problem with SPH, a
single cylinder and its associated volume within the lattice
is considered (Fig. 4). Flow is driven by a pressure gradient
(modeled using an effective body force F ), and periodic
boundary conditions are applied to model an infinite periodic array.
3.3.1. Low Reynolds Number
Periodic flow past a cylinder was simulated using SPH
for L 5 0.1 m, n 5 1026m2s21, a 5 2 3 1022m, F 5 1.5 3
1027ms22, and c 5 5.77 3 1024ms21. Replacing L with a in
(29) and taking the velocity scale to be V0 5 5 3 1025ms21
gives Re 5 1. The SPH simulation was run using approximately 3000 particles placed on a hexagonal lattice with a
nearest neighbor distance of 0.002 m. The cylinder was
modeled by considering all particles within its perimeter
to be of the type described in Section 2.5. The particles
started from rest and steady state was reached after approximately 1500 steps. To investigate long-term behavior,
the simulation was continued for another 6000 steps such
that particle arrangements became disordered. The problem was also modeled using a finite element method (FEM)
program for steady incompressible viscous flow. Velocity
and pressure distributions from the two solutions were
compared by plotting values within one nearest neighbor
distance of the four paths described in Fig. 5. The results
were also compared using contour plots. As the FEM employs a mesh to obtain a solution, it is relatively easy to
obtain contour plots. The corresponding plots for SPH

were obtained by interpolating the particle quantities to a

50 by 50 array of grid points using the quintic kernel.
Smoothing lengths of 1 and 3 grid spacings were used for
the velocity and pressure, respectively. A greater amount
of smoothing was needed to remove small fluctuations
from the pressure field. Contours generated using this
method are inaccurate in the immediate vicinity of the cylinder.
Figure 6 shows a comparison of velocity profiles obtained
using SPH and FEM for paths 1 and 2 defined in Fig. 5.
The results obtained using SPH are in close agreement
with those from the FEM throughout the flow domain.
Corresponding contour plots of velocity magnitude are
shown in Fig. 7. Good agreement is obtained for the bulk
of the flow, although the contour smoothing method inaccurately represents SPH velocities near the cylinder.

FIG. 6. Comparison of SPH and FEM velocity profiles along paths

1 and 2 for Re 5 1.

222

MORRIS, FOX, AND ZHU

FIG. 7. Contour plots of velocity magnitude using (a) FEM and (b) SPH for Re 5 1 (contour lines are labeled in units of 1024m/s).

Figure 8 shows the dynamic pressure along paths 3

and 4 (Fig. 5). The arc of path 3 was taken 0.002 m
(one SPH nearest neighbor distance) beyond the cylinder
boundary since the SPH boundary particles may not
necessarily lie on the boundary surface itself. The SPH
dynamic pressure profile shows small local fluctuations
in the vicinity of the cylinder. Elsewhere, the SPH and

FEM solutions are in close agreement. The peaks in the

pressure obtained using SPH on the boundary itself fell
short of the FEM results by approximately 8%. The FEM
better captures pressure extrema since grid-stretching
increases resolution in the vicinity of the cylinder. Corresponding contour plots of pressure given in Fig. 9 show
that good agreement is again obtained for the bulk of
the flow. This simulation was repeated with twice the
particle resolution (approximately 11,000 particles) and
peak pressures were reproduced to within 5%. Pressure
values in the immediate vicinity of the boundary, however, still exhibited small fluctuations.
3.3.2. Very Low Reynolds Number

FIG. 8. Comparison of SPH and FEM pressure profiles along paths

3 and 4 for Re 5 1.

Flow through a periodic lattice of cylinders was also

solved for L 5 0.1 m, n 5 1024m2s21, a 5 2 3 1022m, F 5
5 3 1025ms22, and c 5 1 3 1022ms21. Taking V0 5 1.5 3
1024ms21, this gives Re 5 0.03. Once again, the simulation
was initialized with zero velocity and the steady state was
reached after approximately 1500 steps. However, the initial lattice was relatively unchanged at this point. To demonstrate the long-term behavior of the method, the simulation was run for another 300,000 steps. The final particle
configuration, shown in Fig. 10, was typical of those observed in these simulations once the initial lattice had broken up. A comparison of velocities along paths 1 and 2
appears in Fig. 11 and velocity contour plots are shown
in Fig. 12. The results obtained using SPH are in close
agreement with those obtained by the FEM. The flux

MODELING LOW RE INCOMPRESSIBLE FLOWS USING SPH

223

FIG. 9. Contour plots of pressure using (a) FEM and (b) SPH for Re 5 1 (contour lines are labeled in units of 1026Pa).

through the lattice is 1.70 3 1024m/s, which agrees to within

3% with the analytical solution of Drummond and Tahir
[31] for Stokes flow. A comparison of pressure fields presented in Figs. 13 and 14 also shows close agreement for
the bulk of the flow, with similar fluctuations as observed
for Re 5 1. Although pressure extrema on the boundary

were not fully captured by SPH, the discrepancies were

somewhat smaller (about 5%). The simulation was repeated for fewer time steps with twice the resolution (approximately 11,000 particles) and peak pressures were obtained with less than 4% error. Once again, however, small
pressure fluctuations were observed near the boundary.

FIG. 10. Final particle positions corresponding to the results presented for Re 5 0.03. Fluid and boundary particles are shown in black
and gray, respectively.

FIG. 11. Comparison of SPH and FEM velocity profiles along paths
1 and 2 for Re 5 0.03.

224

MORRIS, FOX, AND ZHU

FIG. 12. Contour plots of velocity magnitude using (a) FEM and (b) SPH for Re 5 0.03 (contour lines are labeled in units of 1024m/s).

4. CONCLUSIONS AND FUTURE WORK

Extensions of the method of smoothed particle hydrodynamics (SPH) have been presented which allow the simulation of incompressible flows for low Reynolds numbers.
Results suggest that the proposed equation of state, viscosity formulation, boundary conditions, and interpolation

FIG. 13. Comparison of SPH and FEM pressure profiles along paths
3 and 4 for Re 5 0.03.

kernel result in a method which is stable and gives accurate

estimates of velocity for problems of the type considered.
Peak pressures on solid boundaries are smoothed out at
lower resolutions, but are achieved to within 5% for simulations involving over 10,000 particles. However, pressures
in the vicinity of a circular boundary exhibited local fluctuations of several percent. Further improvements to the
boundary conditions may be possible which will reduce
these errors (see Section 2.5). Additional tests involving
convex and concave boundary surfaces are planned.
Straightforward extensions of the method are possible
which will increase its utility. For example, it is possible
to simulate the dispersion of a solute through the lattice
of cylinders described in Section 3.3. A quantity which
does not influence the flow (e.g., the concentration of a
dilute solute) may be evolved independently from one cell
to the next while still assuming the velocity and density
fields are periodic. As most of the computational expense
is associated with locating and summing over nearest
neighbors, the increase in work is moderate. A method
similar to that used here may permit the simulation of
other flows at low Reynolds numbers using SPH. However,
further development will be needed for cases involving
free surfaces or substantially different equations of state.
The extension to three-dimensional problems is, in theory,
straightforward. Obtaining steady-state results is relatively
inexpensive, however, extended simulations in threedimensions involving many crossing times at very low
Reynolds numbers will probably necessitate the use of

MODELING LOW RE INCOMPRESSIBLE FLOWS USING SPH

225

FIG. 14. Contour plots of pressure using (a) FEM and (b) SPH for Re 5 0.03 (contour lines are labeled in units of 1023Pa).

supercomputers. A modification of the method to include

surface tension is currently being undertaken to simulate
the flow of multiphase fluids in porous media.
ACKNOWLEDGMENT/DISCLAIMER
This work was sponsored by the Air Force Office of Scientific Research,
USAF, under Grant F4962096-10020. The views and conclusions contained herein are those of the authors and should not be interpreted
as necessarily representing the official policies or endorsements, either
expressed or implied, of the Air Force Office of Scientific Research or
the U.S. Government.

REFERENCES
1. L. B. Lucy, A numerical approach to the testing of the fission hypothesis, Astron. J. 83, 1013 (1977).

9.
10.
11.

12.

13.
14.

zengleichungen der mathematischen physik, Math. Ann. 100, 32

(1928).
A. J. Chorin, A numerical method for solving incompressible flow
problems, J. Comput. Phys. 2, 12 (1967).
E. Turkel, Preconditioned methods for solving the incompressible and
low-speed compressible equations, J. Comput. Phys. 72, 277 (1987).
P. Tamamidis, G. Zhang, and D. N. Assanis, Comparison of pressurebased and artificial compressibility methods for solving three-dimensional steady incompressible viscous flows, J. Comput. Phys. 124,
1 (1996).
J. J., Monaghan, Simulating gravity currents with SPH lock gates,
Applied Mathematics Reports and Preprints 95/5, Monash University, 1995. (unpublished)
G. K. Batchelor, An Introduction to Fluid Dynamics. (Cambridge
Univ. Press, Cambridge, UK, 1967).
J. P. Morris, A study of the stability properties of SPH, Applied
Mathematics Reports and Preprints 94/22, Monash University, Melbourne, Australia, 1994. (unpublished)

2. R. A. Gingold and J. J. Monaghan, Smoothed particle hydrodynamics:

Theory and application to non-spherical stars, Mon. Not. R. Astron.
Soc. 181, 375 (1977).

15. D. S. Balsara, Von-Neumann stability analysis of Smoothed Particle

Hydrodynamics: Suggestions for optimal algorithms, J. Comput. Phys.
121, 357 (1995).

3. J. J Monaghan, Simulating free surface flows with SPH, J. Comput.

Phys. 110, 399, (1994).

16. J. W. Swegle, D. L. Hicks and S. W. Attaway, Smoothed Particle

Hydrodynamics stability analysis, J. Comput. Phys. 116, 123 (1995).

4. P. Artymowicz and S. H. Lubow, Dynamics of binary-disk interaction.

1. Resonances and disk gap sizes, Astrophys. J. 421, 651 (1994).

17. J. P. Morris, Stability Properties of SPH, Publ. Astron. Soc. Aust. 13,
97 (1996).

5. H. Takeda, S. M. Miyama and M. Sekiya, Numerical simulation of

viscous flow by Smoothed Particle Hydrodynamics, Prog. Theor.
Phys. 92(5), 939 (1994).

18. J. J. Monaghan, On the problem of penetration in particle methods,

J. Comput. Phys. 82, 1 (1989).

6. W. Benz, Smooth particle hydrodynamics: A review, in NATO Workshop, Les Arcs, France, 1989.

19. J. J. Monaghan, Improved modelling of boundaries, Applied Mathematics Reports and Preprints 95/30, Monash University, Melbourne,
Australia, 1995. (unpublished)

7. J. J. Monaghan, Smoothed Particle Hydrodynamics, Annu. Rev.

Astron. Astrophys. 30, 543 (1992).
ber die partiellen differen8. R. Courant, K. Friedrichs, and H. Lewy, U

20. J. J. Monaghan, Simulating gravity currents with SPH: III boundary

forces, Applied Mathematics Reports and Preprints 95/11, Monash
University, Melbourne, Australia, 1995. (unpublished)

226

MORRIS, FOX, AND ZHU

21. L. D. Libersky, A. G. Petschek, T. C. Carney, J. R. Hipp, and F. A.

Allahdadi, High strain Lagrangian hydrodynamics, J. Comput. Phys.
109, 67 (1993).
22. S. T. Maddison, J. R. Murray, and J. J. Monaghan, SPH simulations
of accretion disks and narrow rings, Publ. Astron. Soc. Aust. 13,
66 (1996).
23. O. Flebbe, S. Munzel, H. Herold, H. Riffert, and H. Ruder, Smoothed
Particle Hydrodynamics: Physical viscosity and the simulation of accretion disks, Astrophys. J. 431, 754 (1994).
24. S. J. Watkins, A. S. Bhattal, N. Francis, J. A. Turner, and A. P.
Whitworth, A new prescription for viscosity in Smoothed Particle
Hydrodynamics, Astron. Astrophys. Suppl. Ser. 119, 177 (1996).
25. L. Brookshaw, A method of calculating radiative heat diffusion in
particle simulations, Proc. Astron. Soc. Aust. 6, 207, (1985).
26. J. J. Monaghan, Heat conduction with discontinuous conductivity,

Applied Mathematics Reports and Preprints 95/18, Monash University, Melbourne, Australia, 1995. (unpublished)
27. Schoenberg, I. J., Contributions to the problem of approximation of
equidistant data by analytic functions, Q. Appl. Math. 4, 45 (1946).
28. Monaghan, J. J. and Lattanzio, J. C., A refined particle method for
astrophysical problems, Astron. Astrophys. 149, 135 (1985).
29. J. P. Morris, Analysis of Smoothed Particle Hydrodynamics with Applications, Ph.D. thesis, Monash University, Melbourne, Australia, 1996.
30. N. Epstein and J. H. Masliyah, Creeping flow through clusters of
spheroids and elliptical cylinders, Chem. Enj. J. 3, 169 (1972).
31. J. E. Drummond and M. I. Tahir, Laminar viscous flow through
regular arrays of parallel solid cylinders, Int. J. Multiphase Flow 10,
515 (1984).
32. G. W. Jackson and D. F. James, The permeability of fibrous porous
media, Can. J. Chem. Eng. 64, 364 (1986).