Seventh International Conference on CFD in the Minerals and Process Industries

CSIRO, Melbourne, Australia

9-11 December 2009

STABILITY OF OSCILLATORY AND PULSATILE PIPE FLOW

Jonathan R A NEBAUER and Hugh M BLACKBURN
Department of Mechanical and Aerospace Engineering, Monash University, Victoria 3800, AUSTRALIA

ABSTRACT
Linear flow stability analysis is concerned with the long
time-limit behaviour of a fluid-dynamical system, namely
the systems tendency to find a steady state, or make a
transition to a turbulent one. It is known that steady
laminar flow in pipes (HagenPoiseuille flow) is linearly
stable to general perturbations at finite Reynolds numbers,
and that single-harmonic oscillatory flow is stable to
axisymmetric perturbations. We extend the analysis to
demonstrate that the pure oscillatory flow is also stable to
general perturbations. We explain how this implies that
all laminar steady and periodic circular pipe flows of this
type are linearly stable. The least stable modes identified
in this study are axially invariant.

moderately careful experiments. There is still debate

about the precise mechanism that leads to transition.
For single-harmonic oscillatory pipe flow, linear stability
analysis for axisymmetric perturbations by Yang and Yih
(1977) also suggests linear stability to perturbations of this
kind for all Reynolds numbers and periods of oscillation.
As for steady pipe flow however, experiments (e.g.
Eckmann and Grotberg 1991) show that transition to
turbulence can occur, often in the form of bursts during
each oscillation.
In the case of pulsatile pipe flow (when an oscillation is
superimposed on a steady mean flow), experiments, e.g.
by Stettler and Fazle Hussain (1986), also demonstrate the
presence of burst-type transition. However in this case,
there is presently no published study of linear stability, a
deficiency that our present work aims to remedy. As we
explain below, it is however sufficient to study the
problem of linear stability of oscillatory flows, since the
stability of the steady flow component is well established.

NOMENCLATURE
Kn axial pressure gradient at frequency harmonic n
D pipe diameter
J0 Bessel function of order 0.
k (integer) azimuthal wavenumber
Lz axial domain length
n (integer) frequency harmonic
p pressure
r radial position in pipe
Real component
R pipe radius
Re Reynolds number
t
time
T pulse period
u velocity vector
u axial velocity component
u area-average flow speed
Ured reduced velocity
Wo Womersley number
axial wavenumber
fluid density
fluid kinematic viscosity

Potentially the pulsatile pipe-flow transition region (as

parameterised either by a Womersley or Reynolds
number) can be used to define the operational limits of
peristaltic devices in biological applications where low
shear stresses are advantageous. Here, high throughput is
required, although turbulent transition presents a problem
as suspensions, colloids and blood may be damaged.
Transition to turbulence also is of interest in arterial flows
as turbulence can induce high spatio-temporal gradients of
wall shear stress, leading to changes in morphology of
endothelial cells and disturbing normal regulation of
transport of blood-borne chemicals to and from the arterial
walls.
BASE FLOWS AND PROBLEM PARAMETERS
All axisymmetric laminar steady and time-periodic
laminar flows of Newtonian fluids in straight circular
pipes driven by an axially constant pressure gradient can
be decomposed into a sum of BesselFourier solutions.
We examine the linear stability of axisymmetric and
axially uniform oscillatory and pulsatile flow of fluid with
density and kinematic viscosity in a circular pipe of
diameter D. These base flows are obtained in closed form
as analytical BesselFourier solutions first published by
Sexl (1930) and later by Womersley (1955):

INTRODUCTION
Oscillatory (zero-mean) and pulsatile (non-zero mean)
incompressible flows in a straight, rigid, circular tube are
canonical phenomena of classical fluid mechanics. In
addition they serve as models of a variety of flows of
engineering and physiological application, for example
peristaltic pumping and arterial flows.
As is well known, steady laminar flow in a circular tube
(HagenPoiseuille flow) is linearly stable to general
infinitesimal disturbances for all Reynolds numbers yet
studied (e.g. Drazin and Reid, 1981; Schmid and
Henningson 1994) but is observed to become turbulent at
bulk flow Reynolds numbers of order 20003000 in

Copyright 2009 CSIRO Australia

K iT J (i 3 2Wo2 r D)

un (r,t) = n 0
1 exp2in t T, (1)
32

2n J 0 (i Wo)

where Wo = (2/)1/2D/2 is a dimensionless frequency

parameter known as the Womersley number, n is a

frequency harmonic, J0 is a complex Bessel function and

Kn is an associated complex axial pressure gradient
amplitude. In the limit as T grows without bound, this
analytical solution asymptotes to the standard parabolic
HagenPoiseuille solution for steady laminar flow in a
circular pipe.

and the mean flow velocity scale can again appear through
the reduced velocity. This pairing is a sensible choice for
the oscillatory cases in that the Womersley number
appears in the analytical solution for the base flows. We
note that the oscillatory components of the base flow have
(via eq. 1) radial velocity profiles that are only a function
of Wo, r/D, and t. The reduced velocity is then a
premultiplying kinematic factor that describes how far the
bulk flow oscillates along the pipe, expressed in pipe
diameters, but does not alter the velocity profile. For
oscillatory flows we nonetheless would expect a priori
that their stability could be a function of the two
dimensionless flow parameters, Wo and Ured, as well as
axial and azimuthal wavenumbers.

An important point to be made about the base flows under

consideration is that any temporally periodic
axisymmetric laminar pipe flow (including steady flow,
the infinite-period case) can be expressed as a linear sum
of terms of type (1). Since all the flows have the same
boundary conditions, we can consider their linear stability
on a term-by-term basis, one term for each temporal
Fourier harmonic. Again exploiting linearity, we can deal
with general spatial perturbations at each temporal period
as a linear sum of axial and azimuthal Fourier modes, with
wavenumbers = 2D / Lx and k respectively.
One
implication is that we do not here need to examine the
linear stability of the steady flow, since that has been
comprehensively dealt with in previous works it suffices
to examine the stability of the oscillatory components, and
these can be dealt with one temporal harmonic at a time.

In Figure 1 we show radial profiles of axial velocity at ten

phase-points in the base flow cycle and the bulk flow
speed as a function of time for two Womersley numbers.
With increasing Wo, the velocity profile becomes more
like plug flow but with small overshoots near the pipe
wall. This behaviour was first noted by Richardson in
1929 and studied analytically by Sexl (1930).

The area-average or bulk flow speed is a function of time:

u (t) = (8 /D

D /2
0

u(r,t)rdr .

(a)

(2)

Without loss of generality we can ignore the pressure

gradient Kn as a parameter and adjust the phases and
amplitudes of the solutions (1) such that at each temporal
harmonic n, we have

un (t) = An cos(2nt /T) + Bn sin(2nt /T) .

(3)

The n=0 case corresponds to the standard Hagen

Poiseuille solution u(r) =A02[1-(r/R)2], and as stated
above, is also a solution to (1).

(b)

In the present application it is useful to define the peak

bulk flow speed as
u p = max u (t) .
0< tT

Dimensionless parameters

The flows in question generally have two dimensionless

parameters that describe the pulse period and some
measure of the flow speed. Taking u p as a velocity scale

and diameter D as a length scale, the time scale is D / u p .

This leads to one choice of the two dimensionless
parameters as a Reynolds number and reduced velocity,
respectively

Re =

up D

and

U red =

Figure 1: Shows radial velocity profiles at different baseflow phases note the reversed flow lags in the near-wall
region by up to 90o. The area-average flow speed u (t ) is
shown as a function of time for oscillatory base flows with
(a) Wo=10.23; (b) Wo=35.5.

upT
.
D

In the steady flow case, the only parameter is the

Reynolds number, Re. Alternatively, for the oscillatory
components, the viscosity can be confined to the
Womersley number

Wo =

D 2 ,
2T

Copyright 2009 CSIRO Australia

NUMERICAL METHODS

As noted above, time-periodic base flows enter the

problem through the linearised advection terms in
equation (4) the base flows are precomputed at a
moderate number of phase points or time-slices and then
may be accurately reconstructed during timestepping via
Fourier interpolation in time. Since the base flows in the
present problem are analytically defined, this
reconstruction could be avoided, however it is fast, cheap
and accurate when enough time-slices are used in the
present work we have found 64 time-slices are typically
sufficient for accuracy. A small number of leading
Floquet multipliers and eigenmodes are obtained via
Krylov subspace projection. This is followed by Ritz
reconstruction of modes on the full solution space. The
reader is directed to the references supplied above for
further detail.

Stability analysis

The stability analysis problem is solved in primitive

variables. Starting from the incompressible Navier
Stokes equations

t u = u u p + 2 u,

u = 0,

where p is the kinematic or modified pressure, it is

proposed that u=U+u where U is the base flow whose
stability is examined and u is an infinitesimal
perturbation. Upon substitution and retaining terms linear
in u, the linearized NavierStokes equations are obtained:

t u = u U U u p + 2 u

(4)

Discretization

We note that in the present problem, the base flow is Tperiodic, i.e. U(t+T)=U(t). Because in incompressible
flows the pressure is not an independent variable, and all
terms are linear in u, we can write this evolution equation
in symbolic form

Spatial discretization and time integration is handled using

a cylindrical coordinate spectral element method with
mixed explicit/implicit time stepping, as outlined in
Blackburn and Sherwin (2004). The domain is discretized
into spectral elements in the meridional semi-plane that
runs from the pipe axis to the outer radius in the radial
direction and a finite length of pipe Lz in the axial
direction, as shown for example in Figure 2.

t u = L( u)
where L is a linear operator with T-periodic coefficients
through the influence of the base flow. Correspondingly
the stability analysis of this equation is a linear temporal
Floquet problem (Iooss & Joseph 1990). Writing the state
evolution of u over one period as

u(t + T) = A(T)u(t)

Fourier modal structure is assumed in the azimuthal

direction with integer wavenumbers k, and as a result of
linearization, each azimuthal mode can be dealt with
independently.
In the axial direction we use real
wavenumbers = 2D/Lz. Because of the approach taken
to spatial discretization in the axial direction, the Floquet
eigensolution for any domain length can contain modes
for both = 0 (i.e. modes that are axially invariant) and
=m2D/Lz (where m is an integer). Typically, there are a
number of multipliers for =0 that are larger in magnitude
than the first axially variant mode and we compute
sufficient modes to be assured that we obtain the leading
mode for =2D/Lz as well.

(5)

where A(T) is the system monomodry matrix, we obtain a

Floquet eigenproblem

A(T) uj (t) = j uj (t) ,

where u j (t ) are phase-specific Floquet modes and j are
Floquet multipliers (which in general occur in complexconjugate pairs). Stability of the problem is assessed from
the Floquet multipliers: unstable modes have multipliers
that lie outside the unit circle in the complex plane (i.e. ||
> 1), while stable modes lie inside (i.e. || < 1).
We use a time-stepping based methodology outlined in
Tuckerman and Barkley (2000), given detailed
explanation in Barkley, Blackburn and Sherwin (2008),
and previously used in studies of various oscillatory flows
(e.g. Blackburn 2002; Blackburn et al. 2005; Blackburn
and Sherwin 2007) in order to solve the Floquet
eigenproblem. A key point about the approach is that a
system monomodry matrix A(T) is not explicitly
constructed; rather, a Krylov method is used that is based
on repeated application of the state transition operator (5)
whose action is obtained by integrating the linearised
NavierStokes equations forward in time over interval T.
By varying the Krylov dimension and ensuring sufficient
resolution we are typically able to resolve a moderate
number (e.g. four) of the leading (least stable) Floquet
modes. Here we have in the main concentrated on the
dominant mode. All the multipliers for the leading modes
found here are real, and positive.

Copyright 2009 CSIRO Australia

Figure 2: An example of a spectral element mesh in the

meridional semi-plane for the cylindrical-coordinate
formulation used here. Internal node points are drawn in
selected elements for a tensor-product polynomial shape
function order N=6. The axial domain length corresponds
to an axial wavenumber = 2/0.5 = 12.57.

RESULTS

We note that this collapse is not seen in their data for nonaxially invariant modes, i.e. > 0. This may be observed
in Figure 4, where no collapse of data with Womersley
number is found. Our results, to be discussed below, also
show that for > 0, there are (as expected a priori) again
two dimensionless groups. We can explain this effect
using the following physical reasoning. Axially invariant
modes do not see the influence of the kinematic parameter
Ured, which as we previously noted describes how far the
oscillatory flow moves back and forth along the pipe.
This leaves only Wo as a parameter. However, for > 0,
the modes also have a finite axial dimension and this
allows an interaction so that stability is characterized both
by Wo and Ured.

Comparison to data of Yang & Yih (1977)

We first examine stability to axisymmetric perturbations

(k = 0), as dealt with previously by Yang & Yih (1977).
Yang & Yih found that axially invariant modes ( = 0)
were the least stable, and provided results for a range of
Reynolds numbers, dimensionless frequencies, and axial
wavenumbers. We have re-interpreted their dimensionless
groups as Womersley number and reduced velocity. A
comparison between their results and ours for = 0,
k = 0 is shown in Figure 3. In all cases, the flow is stable
( < 1), but only marginally so at large Wo.

For purposes of comparison, we show in Figure 5

illustrative examples of both axially invariant and axially
variant (but axisymmetric) Floquet mode shapes These are
drawn for one phase point in the base flow cycle. Both
mode shapes are obtained at Wo = 97, Ured = 2.5. We note
that for k = 0, = 0, continuity requires Floquet modes in
which only axial velocity components can be non-zero.

Figure 3. Floquet multiplier data derived from Yang &

Yih (1977) figure 1, for the dominant axially invariant
axisymmetric modes. Compilation of comparable results
obtained from present computations shown as a solid line
( = 0, k = 0).

(a) k = 0, = 0.

(b) k = 0, = 16.

Figure 5 Comparison of axisymmetric (k = 0) Floquet

mode shapes; (a) axially invariant, = 0; (b) with axial
wavenumber = 16. Visualized as isosurfaces of
positive/negative axial velocity component.

Our data compare well with those from Yang & Yih, but
also consideration of this figure brings out a significant
point that has previously been unremarked. Through not
adopting Womersley number as a dimensionless group
with which to represent their data, Yang & Yih apparently
did not notice that for axially invariant modes, it becomes
the only active parameter and collapses three sets of their
data around a single curve (the slight discrepancies seen
are attributable noise in our digitization of their figures).
We have confirmed this by running analyses at a wide
range of the two dimensionless parameters, and find that
all our leading multiplier data for = 0, k = 0, fall on a
single curve when plotted against Womersley number.

Stability of axially invariant modes ( = 0)

Yang & Yih (1977) dealt only with the stability of

axisymmetric modes, k = 0. Since there is no known
equivalent to Squires theorem (Squire 1933) in
cylindrical coordinates, there is no reason to expect a
priori that axisymmetric modes are the least stable. We
commence examination of results for general modes by
studying the effect of azimuthal wavenumber k on axially
invariant modes, = 0; as explained above. The only
remaining parameter in these cases is the Womersley
number.
Figure 6 shows that axisymmetric modes are in fact the
least stable, since Floquet multipliers reduce
monotonically in magnitude with increasing azimuthal
wavenumber at all values of Wo. This shows that all
axially invariant modes are stable.

Figure 4.
Yang & Yih (1977) dominant Floquet
multiplier data for axisymmetric modes (k = 0) with = 4.
Note that here (as opposed to Figure 3) there is no
collapse with Womersley number.

Copyright 2009 CSIRO Australia

Stability of axially variant modes ( > 0)

Next we turn to examine the question of whether modes

with axial structure are more, or less, stable than those
without. Recall that in these cases, we expect both
Womersley number and reduced velocity as control
parameters. We note that Yang and Yihs results suggest
that for axisymmetric modes, axially invariant cases ( =
0) are the least stable for a fixed value of reduced velocity.
As shown in Figure 9, our results also support this
conclusion.
At fixed Womersley number, Floquet
multipliers decrease monotonically as axial wavenumber
is increased (axial wavelength is decreased).
A
comparison of typical mode shapes for axisymmetric
axially invariant and axially variant modes was shown in
Figure 5.

Figure 6 Dominant Floquet multipliers for axially

invariant modes, = 0, both axisymmetric (k = 0) and
non-axisymmetric (k > 0) as functions of Womersley
number.
Figure 7 illustrates the distinction between axisymmetric
and non-axisymmetric Floquet modes, but where both are
axially invariant, = 0. Again we note that since these
modes are axially invariant, the only non-zero velocity
component must be in the axial direction.

Figure 9 Dominant Floquet multipliers for axisymmetric

modes (k = 0) at Ured = 2.5, showing the effect of axial
wavenumber.
(a) k = 0, = 0.

(b) k = 1, = 0.
In order to demonstrate that there are in fact two control
parameters when mode shapes are axially variant, we
show in Figure 10 the leading Floquet multipliers for
axisymmetric modes (k = 0) at a fixed axial wavenumber
= 16. For axially invariant modes, we re-iterate that all
results would collapse to a single function of Womersley
number. Here, two distinct curves are obtained, one for
each reduced velocity.

Figure 7 Comparison of (a) axisymmetric (k = 0) and (b)

non-axisymmetric (k = 1) Floquet mode shapes, both
axially invariant, = 0. Visualized as isosurfaces of
positive/negative axial velocity component. Computed at
Wo = 97.
To demonstrate that the dominant axially invariant,
axisymmetric mode is significantly less stable than the
subdominant modes, we show in Figure 8 the Floquet
multipliers of the three leading modes for = 0, k = 0.

Figure 10 Dominant Floquet multipliers for axially

variant but axisymmetric modes, with two values of
reduced velocity. These values are computed at fixed =
16.

Figure 8 Floquet multipliers for the leading three (a, b,

c) axially invariant and axisymmetric modes ( = 0, k =
0). Curve a is the same as the curve labelled k = 0 in
Figure 6.

Copyright 2009 CSIRO Australia

CONCLUSION
The present study extends the work of Yang & Yih
(1977), who examined the linear stability of axisymmetric
disturbances to oscillatory pipe flows.
Our new
contribution here is the extension to non-axisymmetric
disturbances, however these are generally more stable
than equivalent axisymmetric cases. In agreement with
Yang & Yih we have found that the least stable
disturbances are provided by axially invariant modes.

DRAZIN, P.G. and REID, W.H., (1981), Hydrodynamic

Stability, Cambridge University Press.
ECKMANN, D.M. and GROTBERG, J.B. (1991),
Experiments on transition to turbulence in oscillatory
pipe flow, J Fluid Mech., 222, 329350.
IOOSS, G. and JOSEPH, D.D. (1990), Elementary
Stability and Bifurcation Theory, 2nd edn. Springer.
SCHMID, P.J and HENNINGSON, D.S. (1994),
Optimal energy density growth in HagenPoiseuille
flow, J Fluid Mech., 277, 197255.
SEXL, T. (1930), ber den von E. G. Richardson
entdeckten annulareffekt, Z. Phys, 61, 179-221.
SHERWIN, S.J. and BLACKBURN, H.M. (2005),
Three-dimensional instabilities and transition of steady
and pulsatile flows in a stenotic tube, J Fluid Mech., 533,
297327.
SQUIRE, H.B. (1933), On the stability of threedimensional disturbances of viscous fluid between parallel
walls, Proc Roy Soc Lond A, 142, 621628.
STETTLER, J.C. and FAZLE HUSSAIN, A.K.M.,
(1986), On transition of the pulsatile pipe flow, J Fluid
Mech., 170, 169197.
TUCKERMAN, L.S. and BARKLEY, D. (2000),
Bifurcation analysis for timesteppers, in Numerical
Methods for Bifurcation Problems and Large-Scale
dynamical systems, eds Doedel & Tuckerman, Springer,
453466.
WOMERSLEY, J.R. (1955), Method for the
calculation of velocity, flow and viscous drag in arteries
when the pressure gradient is known, J Physiol., 127,
533563.
YANG, W.H. and YIH, C.-S., (1977), Stability of timeperiodic flows in a circular pipe, J Fluid Mech., 170,
497505.

A significant new result, not recognised in past work, is

that the stability of these axially invariant modes is
dependent only upon a single control parameter, the
Womersley number, which is also the only parameter
needed to describe the radial velocity profiles of the base
flows. When modes with axial variation are considered, a
second control parameter comes into effect. We have
used the reduced velocity as this second control parameter
this can be interpreted as a measure of the number of
pipe diameters that the oscillatory bulk flow travels in one
oscillation period. Modes without axial variation (thus,
without an axial length scale) do not respond to changes in
reduced velocity.
Base flows with larger reduced
velocities are more stable than those with smaller reduced
velocities.
Our study suggests that all single-harmonic oscillatory
pipe flows are linearly stable, but approach instability
asymptotically with increasing Womersley number. This
is analogous to the behaviour of steady pipe flows, which
are observed to be linearly stable at all finite Reynolds
numbers, but less stable at high Reynolds number.
Finally, we note (apparently for the first time) that since
all periodic laminar pulsatile flows can be described as the
superposition of steady flow and oscillations at different
temporal harmonics, it suffices to study separately the
linear stability of each of the temporal harmonics. Since
steady pipe flow is well-known to be linearly stable at all
finite Reynolds numbers, and our results show that singleharmonic oscillatory pipe flows are stable to both
axisymmetric and non-axisymmetric disturbances at all
finite values of the two control parameters, we have the
general result that all time-periodic laminar pulsatile flows
in circular pipes are linearly stable.
REFERENCES
BARKLEY, D., BLACKBURN, H.M. and SHERWIN,
S.J. (2008), Direct optimal growth for timesteppers, Int
J Num. Meth. Fluids, 57, 14351458.
BLACKBURN, H.M. (2002), Three-dimensional
instability and state selection in an oscillatory
axisymmetric swirling flow, Phys. Fluids, 14, 39833996.
BLACKBURN, H.M., MARQUES, F. and LOPEZ, J.M.
(2005), Symmetry breaking of two-dimensional timeperiodic wakes, J Fluid Mech., 522, 395411.
BLACKBURN, H.M. and SHERWIN, S.J. (2004),
Formulation of a Galerkin spectral elementFourier
method for three-dimensional incompressible flows in
cylindrical geometries, J Comput. Phys. 197, 759778.
BLACKBURN, H.M. and SHERWIN, S.J., (2007),
Instability modes and transition of pulsatile stenotic flow:
pulse-period dependence, J Fluid Mech., 573, 5788.

Copyright 2009 CSIRO Australia