Sunteți pe pagina 1din 53

CHAPTER 2

Low-Reynolds-Number Flows
HOWARD A. STONE*a AND CAMILLE DUPRAT*b
a
Department of Mechanical and Aerospace Engineering, Princeton
University, Princeton, NJ 08544, USA; b École polytechnique, Palaiseau,
Paris, France
*
E-mail: hastone@princeton.edu, camille.duprat@ladhyx.polytechnique.fr

2.1 Introduction
In this chapter, we provide a brief description of some of the main results of
low-Reynolds-number hydrodynamics. In particular, we introduce the gen-
eral subject and provide derivations and explanations of some of the fluid
dynamics themes that are used in later chapters. Specifically, we discuss
general theoretical principles, describe various problems involving the mo-
tion of rigid and deformable particles in fluids, and introduce slender-body
theory. One of our main goals in this chapter is to share the physical intu-
ition and underlying mathematics of common low-Reynolds-number flow
situations.
Readers will enjoy the classic article by Purcell with the attractive title
“Life at low Reynolds number.”1 Also, for a stimulating illustration of the
properties of low-Reynolds-number flows, readers are encouraged to watch
the classic film Low Reynolds Number Flows by Taylor; 2 the examples given
there of viscous flow phenomena are even instructive to non-fluid dynam-
icists! Not surprisingly, there are entire books on the subject, 3–5 although
their primary focus is on the motion of rigid particles as it occurs in flows of
suspensions.
To gain further perspective on the wide range of questions that arise in
low-Reynolds-number motions, we illustrate some typical flow problems in

RSC Soft Matter No. 4


Fluid–Structure Interactions in Low-Reynolds-Number Flows
Edited by Camille Duprat and Howard A. Stone
C The Royal Society of Chemistry 2016

Published by the Royal Society of Chemistry, www.rsc.org

25
26 Chapter 2

Figure 2.1. Consider the sedimentation of an object, which is a standard


problem because gravity always causes motion of nonneutrally buoyant par-
ticles in a fluid. For example, when a sphere of radius a sediments at speed
V adjacent to a vertical wall (see Figure 2.1a), the sphere rotates as if it were
rolling down the wall and maintains a constant separation distance from the
wall. In fact, for separation distances much smaller than the sphere radius,
i.e. h  a, the rotation rate is U ≈ V/4a. 10
Many problems involve elastic, deformable objects, where viscous and
pressure stresses from the flow are responsible for the deformation, as will
be illustrated in each of the chapters of this book. For example, when a sus-
pension of flexible fibers is sheared, there are hydrodynamic interactions
among the fibers, and the individual fibers deform and constantly reorient
(Figure 2.1b). Also, in recent years, there have been many examples of mi-
crofluidic flows involving soft objects or boundaries. For example, when an
elastic element partially blocks the flow, the element is deformed and alters
the pressure distribution in the channel (Figure 2.1c). The behavior of flexi-
ble fibers in flows will be discussed in detail in Chapter 4, by Pak and Lauga,
and Chapter 5, by Lindner and Shelley. Finally, there are many examples of
the motion of soft objects such as drops, bubbles, cells, and vesicles. When
these objects are forced to move in a pressure-driven flow, deformation away
from a spherical shape imparts a velocity or lift component transverse to a
boundary (Figure 2.1d). Such problems will be discussed further in Chapter
9, by Vlahovska, and Chapter 10, by Abkarian and Viallat.
There are some typical results for low-Reynolds-number flows that have
wide use. Here are five such results that will be discussed further in this
chapter:

(1) A sphere of radius a that translates at a speed V far from any walls and
in a large bath of fluid experiences a drag force of magnitude

FD Z 6πmaV. (2.1)

This result is known as the Stokes drag formula, which has several
characteristic features:

(i) The magnitude corresponds to the product of the surface area 4πa2
and a typical shear stress mV/a.
(ii) The drag force varies linearly with the particle radius and linearly
with the particle speed.

Readers with a broader knowledge of fluid mechanics can contrast


feature (ii) with the high-Reynolds-number-flow limit, where the drag
force varies approximately with the square of the particle radius and
the square of the speed.
(2) Because of gravity, a sphere of radius a and density rs sediments in a
viscous fluid of density r with a net force of (4π/3)a3 (rs K r) g. Using
Low-Reynolds-Number Flows 27

Figure 2.1 (a) Sedimentation of a sphere near a vertical plane wall. 6 (b) Snapshot of
a suspension of flexible fibers in a shear flow. The various deformations
and orientations of the fibers are evident.7 Reprinted from Tornberg
and Shelley, Simulating the dynamics and interactions of flexible fibers
in Stokes flows, J. Comput. Phys., 2004, 196, 8–40, with permission from
Elsevier. (c) The bending of an elastic fiber in a pressure-driven channel
flow. The scale bars are 100 µm. Note that there is a small gap, typi-
cally about 10 µm, between the fiber and the walls above and below.8
Reprinted by permission from Wexler et al., Bending of elastic fibres in
viscous flows: the influence of confinement, J. Fluid Mech., 2013, 720,
517–544. (d) A shear flow drives the drift of a deformed vesicle (equiv-
alent radius 31 µm) away from a wall as the vesicle moves from left
to right. Figure reprinted with permission from Abkarian et al.,9 Phys.
Rev. Lett., 2002, 88, 068103. Copyright (2002) by the American Physical
Society.
28 Chapter 2

eqn (2.1), the sedimentation speed V is given by


2a2 (rs K r) g
VZ , (2.2)
9m
i.e. V ∝ a2 . Therefore, doubling the particle radius causes an increase
in the sedimentation speed by a factor of 4.
(3) Brownian motion refers to movements of suspended objects due to
thermal bombardment by the molecules of the fluid, with a ther-
mal energy of kB T (kB is Boltzmann’s constant and T is the absolute
temperature). As derived by Einstein in 1905, a sphere of radius a un-
dergoing Brownian motion is characterized by a translation diffusion
coefficient (dimensions length2 /time)
kB T
Dm Z , (2.3)
6πma
which utilizes the Stokes drag formula (2.1). Equation (2.3) is referred
to as the Stokes–Einstein equation. One use of this result in physical
chemistry is to measure the diffusion coefficient of a small particle
and then use eqn (2.3) to identify an effective (or hydrodynamic) radius
for the particle.
(4) A circular disk of radius a translating at a speed V edgewise parallel
to a nearby plane, with separation distance h  a, experiences a drag
force that is approximately given by the product of the surface area
of the disk πa2 and the shear stress mV/h, i.e. Fd ≈ πa2 mV/h. Clearly,
halving the separation distance doubles the drag.
(5) A long slender rod of radius a and length `, with `  a, when translat-
ing at speed V experiences a drag force F that

(i) is approximately proportional to its length (the largest dimension);


(ii) is approximately twice as great when the rod is oriented perpendic-
ular (⊥) to the flow direction than when it is parallel (k) to the flow
direction, i.e.

8πm`V
F⊥ ≈ ≈ 2Fk . (2.4)
ln(2`/a)
The logarithmic factor will be discussed further when slender-body
theory is described in Section 2.8.

In this chapter, we start with a description of the equations of motions in


Section 2.2, and then analyze some common elementary flows in Section 2.3.
We discuss in Section 2.4 one of the most striking features of low-Reynolds-
number flows, which is their kinematic reversibility. We then illustrate some
mathematical features of Stokes flow in Section 2.5. We provide the main
results on the translation and rotation of objects in flows, starting with
spheres in Section 2.7, and then consider elongated objects by presenting
slender-body theory in Section 2.8. Also, we describe the rotational motion
Low-Reynolds-Number Flows 29

of orientable particles in a Stokes flow by introducing Jeffery orbits in Sec-


tion 2.9. In Section 2.10, we briefly discuss the deformation of drops in
extensional and shear flows and indicate how the deformation leads to a drift
of the drop away from a nearby planar boundary. Finally, in Section 2.11 we
make a few remarks about the influence of small inertial effects.

2.2 Equations of Motion


For a single-phase Newtonian fluid of constant density r and constant vis-
cosity m, the incompressible fluid velocity field v and the pressure p are
described by the continuity and Navier–Stokes equations,
 
vv
V · v Z 0 and r C v · Vv Z KVp C m V2 v C rfb , (2.5)
vt
where fb denotes the body force per unit mass. It is worth recognizing that
each term in the Navier–Stokes equation has dimensions of force/volume.
The Navier–Stokes equation describes the change in the linear momentum
of the fluid, and the left-hand side of the equation denotes the inertial (mass
times acceleration) term in Newton’s equations of motion applied to the
fluid. A constitutive equation has been used to linearly relate the state of
mechanical stress on a fluid element to the local velocity gradient, which
distinguishes so-called Newtonian fluids with viscosity m. Since local differ-
ences in stress lead to a net force (per unit volume) on the fluid, the viscous
terms enter through second derivatives of the velocity distribution, i.e. the
m V2 v term in eqn (2.5b).
Boundary and initial conditions are then needed to solve these two cou-
pled partial differential equations. The solutions generally require numerical
methods except for relatively simple geometries and various special cases,
such as the case of low-Reynolds-number motions discussed here and the
case of idealized inviscid flows. We will give some of the simpler analytical
low-Reynolds-number results in this chapter.

2.2.1 The Reynolds Number


Consider the motion of a sphere of radius a at speed V. Over the time it takes
the sphere to translate through its radius, the typical inertial terms in eqn
(2.5b) are O (rV 2 /a). In contrast, the typical magnitude of the viscous term
on the right-hand side of eqn (2.5b) is O (mV/a2 ). We now compare these two
quantities:
inertial terms rV 2 /a rVa
Re Z Z Z , (2.6)
viscous terms mV/a2 m
which defines the Reynolds number, Re .†

† For a historical discussion of the Reynolds number, see Rott. 11


This ratio of variables was iden-
tified in 1883 by Osborne Reynolds (1842–1912), a professor at Manchester University, who
was characterizing the different motions observed in pressure-driven single-phase flow in a
pipe. Some years later Arnold Sommerfeld, a leading German physicist, introduced the name
“Reynolds number” to identify this dimensionless ratio.
30 Chapter 2

An alternative interpretation of the Reynolds number is based on time


scales. In this case, we write
rVa a2 /n time for vorticity to diffuse distance a
Re Z Z Z , (2.7)
m a/V time to translate distance a
where we have introduced the kinematic viscosity n Z m/r (the reader can
verify that n has dimensions length2 /time). The numerator in eqn (2.7) is a
typical vorticity diffusion time. It is useful in fluid mechanics to think about
vorticity, or local rotation of fluid elements, as a flow property that diffuses.
Also, the denominator in eqn (2.7) is the typical time it takes to translate
through a body length. A low-Reynolds-number flow, with Re  1, then corre-
sponds to a motion where the viscous effects or the vorticity diffuses much
faster than a particle or a moving boundary can change its position. The
fluid rapidly, or instantaneously in the ideal limit, establishes a velocity
distribution for a given translation speed and geometry of the boundaries.
The “complexity” of a fluid motion increases as the Reynolds number is
increased. For example, for uniform flow past objects, the flow is initially
laminar, and as the speed is progressively increased, the flow first develops
a downstream wake of increasing length, after which there is a transition
to turbulence, i.e. the velocities in at least some regions of the flow become
time-dependent and fluctuate stochastically. The Reynolds number for the
transition from a laminar to a turbulent flow state is known to depend on
the magnitude of fluctuations in the inlet flow, but, for flow in a circular
pipe in common laboratory conditions it is on the order of 2000 (based on
the mean velocity and the pipe diameter). A similar evolution of the flow
structure from a laminar to a turbulent state occurs in other well-studied ge-
ometries (e.g. Taylor–Couette flow, where one cylinder rotates steadily inside
a second cylinder, and flow over a flat plate).

2.2.2 Stokes Equations


In this chapter, we focus on results characterized by a low Reynolds number,
Re  1, for which eqn (2.5) reduces to the continuity and Stokes equations,
V · v Z 0 and 0 Z KVp C m V2 v C rfb . (2.8)
It is often convenient
 to work in terms of the stress tensor σ Z KpI C
T
m Vv C (V v) for a Newtonian fluid, for which the momentum equation
of low-Reynolds-number hydrodynamics can be written
V · σ C rfb Z 0. (2.9)
In this flow limit, there can still be considerable complexity owing to the
presence of (i) suspended particles, whose distribution affects the flow,
which in turn influences the particle distribution, and (ii) elastic boundaries,
whose shape depends on the flow, which in turn influences the boundary
shape. These topics are discussed in later chapters, each of which highlights
Low-Reynolds-Number Flows 31

some aspects of fluid-structure interactions. Finally, we note that for most of


the discussion in this chapter we neglect the effect of body forces, i.e. fb Z 0.
There are a few other mathematical properties that can be helpful to
note:

(i) Take the divergence of eqn (2.8b) and use the continuity equation.
Then, we learn that the pressure is harmonic:
V2 p Z 0. (2.10)
(ii) Take the curl of eqn (2.8b). Then, we observe that the vorticity vector
ω Z V ∧ v is also harmonic:
V2 ω Z 0. (2.11)
(iii) Consider a two-dimensional incompressible  flow, expressed in Carte-
sian coordinates, i.e. v Z vx (x, y), vy (x, y) . The continuity equation is
vvx /vxCvvy /vy Z 0. We observe that the equation is satisfied automat-
ically if we introduce a stream function j(x, y) such that (the minus
sign can be placed before either component of velocity)
vj vj
vx Z and vy Z K . (2.12)
vy vx
The reader can verify that for a two-dimensional flow the vorticity
vector is perpendicular to the plane of the flow, i.e. ω Z uz ez , where
uz Z vvy /vx K vvx /vy or uz Z KV2 j. Thus, using eqn (2.11), we find
that the stream function is determined by
V4 j Z 0. (2.13)
Again, understanding the structure of the Laplace equation and its
relatives is important. Equation (2.13) indicates that we can solve for
the scalar field j(x, y) and then construct the velocity field from eqn
(2.12) and the pressure field from the Stokes equations rather than
working directly from the continuity and Stokes equations for the two
components of velocity and the pressure.
(iv) A convenient decomposition when solving eqn (2.8) is to denote the
position vector by r and to note the identify (verify it)
V2 (pr) Z 2 Vp C r V2 p Z 2 Vp, (2.14)
2
since V p Z 0 by eqn (2.10). For simplicity, let us assume, without loss
of generality here, that fb Z 0. Then, by superposition, we can write
the solution to the Stokes equations (2.8) as the sum of a particular
solution involving the pressure field and a homogeneous solution in
the form
1
v (r) Z pr C vh (r), (2.15)
2m
where
V2 vh Z 0. (2.16)
Again, the Laplace equation appears prominently.
32 Chapter 2

Much is known about the structure of the solutions of Laplace’s equation.


Because of eqn (2.10), (2.11), (2.13), and (2.16), as well as the structure of the
Stokes equations, the mathematical features related to the Laplace equation
figure prominently when solving low-Reynolds-number flow problems.
We now consider the common case of a constant body force, for exam-
ple the case of gravity, fb Z g, where g is the gravitational vector. Since g is
constant, we can always absorb it into the pressure by an appropriate re-
definition.‡ So, from now on, unless otherwise stated, we will work with the
continuity and Stokes equations (2.8) and (2.9) in the form
V · vZ0 and m V2 v Z Vp or V · σ Z 0. (2.17)

2.3 Elementary Flows


2.3.1 Channel Flows
We make a few brief remarks on elementary laminar flows in channels here.
It is useful to develop physical intuition for the kinds of flows that occur
in common situations. In particular, when there is relative translation, or
sliding, between two planar parallel surfaces the velocity distribution is lin-
ear, which is referred to as simple shear flow (Figure 2.2a). More commonly,
pressure gradients are used to drive flows through channels, for which the
velocity distribution is parabolic (Figure 2.2b).
In many channel flows, the distance along the flow direction ` is very
different from the typical distance `⊥ transverse to the flow direction. In dis-
cussing these flows, we find it useful to make approximations that highlight
this difference in length scales, with `⊥  `. The Stokes equations indicate a
balance between viscous stresses mvc /`⊥ 2 , where vc is a typical speed, and the
pressure gradient, Dp/`, where ` is the distance along the flow over which the
pressure drop Dp occurs. Therefore, from eqn (2.9) we have the estimate
`⊥ 2 Dp
vc ≈ . (2.18)
m `
The corresponding flow rate in the channel is Q ≈ vc ! (cross-sectional area).
Thus, for a circular pipe of radius a Z `⊥ , the pressure-drop-versus-flow-rate
relation is (where we have introduced a constant π/8 that results from a
detailed calculation)
πa4 Dp
QZ , (2.19)
8m `
which is known as the Poiseuille formula. We note that for a given Dp the
dependence of the flow rate Q on the fourth power of the pipe radius, a4 , im-
plies that small changes in radius have a significant impact. For example, for

‡ Forexample, if z is vertical and g Z Kgez , we denote the dynamical pressure by pd Z p C rgz,


in which case the Stokes equations can be written in terms of pd , i.e. m V2 v Z Vpd . It is common
in the literature not to make this distinction between p and pd unless gravitational body forces
actually enter the flow problem.
Low-Reynolds-Number Flows 33

Figure 2.2 Elementary channel flows. (a) Shear flow. (b) Parabolic velocity distri-
bution for pressure-driven flow in a channel or pipe. (c) Pressure-driven
flow along the z direction of a rectangular channel of height h and
width w O h: average velocity across the short (h) direction for flow in
a rectangular channel. For narrower cross sections, i.e. h/w  1, the av-
erage velocity profile becomes flat in the center of the channel and only
changes to zero in the neighborhood of the side walls. After Stone.12 (d)
Flow through a hole. Adapted from Happel and Brenner.3

a given pressure drop, a 10% change in radius leads to an approximately 40%


change in flow rate; in this situation we conclude that even modest shape
changes can have dramatic effects on the flow.
Next, we consider simplifications of the Stokes equations. In the case of
pressure-driven flow along a uniform rectangular channel of height h and
width w O h, the velocity profile along the flow direction is unidirectional,
i.e. v Z v(x, y)ez , where v(x, y) satisfies
dp
m V2 v Z , (2.20)
dz
where dp/dz is a constant. Assuming no-slip boundary conditions, the ve-
locity distribution is approximately parabolic in the short direction and, for
h  w, the velocity distribution is independent of the transverse coordinate y
except near the side walls. Therefore, a plot of the velocity profile, averaged
over the thin direction, is flat in the middle of the channel, as illustrated in
Figure 2.2c. Finally, for the flow rate through a channel with h ! w, we have
Q ∝ (h3 w/m)(Dp/`), which is the analogue of eqn (2.19). For the case of a nar-
row channel, the geometric factor h3 indicates that small changes in height
can have a significant impact on the flow rate for a given pressure drop.
34 Chapter 2

2.3.2 Darcy’s Approximation: Description of Porous Media


In many situations involving complex geometries, such as flow through a
porous medium, it is common to introduce an approximation that captures
the average velocity variations rather than the detailed velocity distribution.
In this spirit, we consider pressure-driven flow in any channel-like configura-
tion. We approximate the viscous resistance using the average fluid velocity
hvi and introduce the permeability k (dimensions length2 ), in which case the
viscous term in the Stokes equations is approximately

mhvi
m V2 v → K . (2.21)
k

The minus sign is needed, since viscous effects will act oppositely to the
mean flow direction. We expect that the permeability k will be character-
ized by the length scales characteristic of the largest viscous effects; in the
language introduced above for channel flow, we have k Z O `⊥ 2 . Then, the


continuity and Stokes equations (2.8) simplify to

mhvi
V · hvi Z 0 and Z KVp C rfb , (2.22)
k

where the latter is commonly known as Darcy’s equation or Darcy’s approx-


imation.§ For flow in a channel of height h  w, k Z h2 /12, while for flow in
a circular pipe of radius a, k Z a2 /8. More generally, for flow in a geometry
with a typical pore scale `⊥ , k Z O `⊥ 2 , with a prefactor O(1/10).


For a constant permeability and a constant body force, taking the di-
vergence of Darcy’s equation (2.22b) yields V2 p Z 0, from which we deter-
mine the pressure field. With the pressure known, the velocity then follows
from Darcy’s approximation. This ordering of steps means that boundary
conditions are placed on the pressure distribution.
Since a constant body force, which commonly exists for gravitationally
driven flows, can be incorporated into an effective pressure, Darcy’s ap-
proximation is commonly written hvi ∝ Vp. Consequently, the pressure is
equivalent to the velocity potential for motions with constant viscosity and
permeability. Moreover, it follows that such motions have zero vorticity,
i.e. V ∧ hvi Z 0. This latter statement is only true for the “average” descrip-
tion, since the flow in the narrow dimension, which has been incorporated
into the definition of the permeability, can be expected to have an approxi-
mately parabolic velocity distribution and so has vorticity. Finally, for those
geometries where the Darcy description is appropriate but where there are
permeability variations, the pressure distribution follows from V · (k Vp) Z 0.

§ Foran interesting discussion about Henry Darcy (1803–1858), a leading 19th-century French
engineer who made significant innovations in fluid mechanics, soil science, and the develop-
ment of clean water systems for cities, see Philip. 13
Low-Reynolds-Number Flows 35

2.3.3 Flow Through a Hole in a Wall: Sampson’s Solution


Another model problem concerns pressure-driven flow through a circu-
lar orifice in a thin plane wall (Figure 2.2d), which was first treated by
Sampson. ¶,14 In this case a pressure drop Dp forces a fluid flux Q through
a circular hole of radius a, with the magnitude of the flux given by
a3 Dp
QZ . (2.23)
3m
The basic form (though not the factor of 3) can be deduced by dimensional
analysis.
For pressure-driven flow through an orifice of radius a and length `, such
as is the case for a pore in a sheet of thickness ` (Figure 2.2d), we can add
the entrance and exit effects associated with the Sampson result (2.23) to the
Poiseuille result (2.19) and so arrive at 16,17
a3 Dp 1
QZ . (2.24)
3m 1 C (8/3π)(`/a)
Perhaps surprisingly, this simple superposition is within about 1% of the
full numerical solution even for a short orifice with `/a Z 1/8. In practice,
the velocity profile, even at the entrance and exit of the pore, is never too far
from the parabolic velocity distribution, at least qualitatively.

2.4 Kinematic Reversibility


The reader may note that the topic of this section was discussed recently
in a similar fashion in lecture notes by one of the authors for a Les Houches
summer school. 18 Also, an excellent discussion of kinematic reversibility can
be found in Lauga and Powers 19 and in Chapter 4 of this book, by Pak and
Lauga.

2.4.1 Observations
There is a feature of Stokes flows that surprises many people when it is first
encountered. Stokes flows in or around boundaries of fixed shape have the
property that the flow looks the same, i.e. that the streamline pattern is
identical, when the driving force is reversed. For example, consider a low-
Reynolds-number flow past a cavity as shown in Figure 2.3a,b. The stream-
lines are symmetric about the vertical midplane. From the images alone, it
is impossible to deduce whether the flow is from left to right or right to left
(the reference for the figure indicates that the flow is from left to right (Van
Dyke, 20 p. 6)). Moreover, similar experiments with flows over obstacles that
have a vertical plane of symmetry (e.g. Figure 2.3c,d) also exhibit streamline
patterns with fore–aft symmetric streamlines consistent with the geometric

¶ The analogous two-dimensional problem of flow through a slit was treated by Hasimoto. 15
36 Chapter 2

Figure 2.3 Experimental images that illustrate kinematic reversibility in low-


Reynolds-number flows by showing fore–aft symmetric streamlines in
flows past various cavities and objects. The flow is in the plane of the
page. (a) and (b) With the cavity height as the length scale, Re Z 0.01.
(a) For width/height = 2, almost the whole of the cavity is filled by one
region of closed streamlines. (b) For width/height = 3, the eddies are
confined to the corners. (c) and (d) Flow past objects attached to a plane;
the flows are fore–aft symmetric. For (c), with the side of the square
as the length scale, Re Z 0.02. For (d), Re Z 0.014. Notice that in each
corner there is a region of recirculating flow, where theory suggests an
infinite sequence of progressively smaller and weaker eddies as the cor-
ner is approached. (e) and (f) Flow through a so-called Tesla valve. The
two images look identical even though the pressure gradient has been
reversed in the second one. At low Reynolds numbers the streamlines
have the same topology independent of the flow direction, and so the
pressure-drop-versus-flow-rate relation is independent of the flow direc-
tion. References: For (a)–(d), all images were obtained from An Album
of Fluid Motion20 and are used by permission of Parabolic Press. For (e)
and (f), we thank Patrick Tabeling for the use of these images.

symmetry. Again, from the images alone it is not possible to know whether
the flow is from left to right or right to left.
Finally, we consider a more complicated channel shape, which is known as
the Tesla valve (after a patent filed by the great inventor Nikola Tesla in about
1920), as shown in Figure 2.3e,f. The flow is driven by a pressure difference
Low-Reynolds-Number Flows 37

between the two ends of the valve. For large Reynolds numbers, say a few
hundred or more, there is a difference in flow rate between the flows in the
two different directions, i.e. the design acts as a diode. However, when the
Reynolds number is approximately unity or smaller, then the streamlines are
the same, independent of the flow direction; see, for example, Figure 2.3e,f.
As a consequence, the corresponding pressure-drop-versus-flow-rate relation
is also independent of the flow direction. For an example of how viscoelastic
effects at low Reynolds numbers can change this kind of reversible dynamics,
the reader is referred to Groisman et al. 21

2.4.2 Mathematical Reasons for Kinematic Reversibility


How do we rationalize the experimental results in Figure 2.3? We begin with
the Stokes equations (2.17) for a Newtonian fluid, which are linear equa-
tions relating the velocity and pressure fields. Thestress is related
 linearly
T
to the velocity and pressure fields, i.e. σ Z KpI C m Vv C (Vv) . Now, con-
sider causing a flow by applying steady forces at the boundaries; for example,
on some surfaces S we apply known stresses τ(r), where r is the position
vector:

n · σ Z τ(r) for r ∈ S. (2.25)

Next, we consider changing the sign of these surface stresses, i.e. τ → Kτ.
Thus, according to (2.25), we change the sign of σ, which for a Newtonian
fluid means that we reverse the signs of v and p, for example v → Kv (we
are free to add a constant to the pressure field). These facts imply that the
flow field is exactly reversed when the boundary forcing is reversed, and
consequently the streamlines are identical in the two cases: only the flow
direction along the streamlines changes. We refer to this reversal of the
flow field as “kinematic reversibility” to respect the thermodynamic fact
that viscous flows always dissipate energy and so are thermodynamically
irreversible.

2.4.3 Examples of Kinematic Reversibility


There are many examples related to this topic, including the possibility of
breaking the constraints of reversibility. Here, we list a few examples.

(1) Perhaps the best-known example of kinematic reversibility in a low-


Reynolds-number flow is the apparent “unmixing” of a viscous liquid,
as beautifully illustrated in the Taylor movie Low Reynolds Number
Flows: 2 the annular space between two concentric cylinders is filled
with a viscous fluid, a small spot or line of colored dye is introduced,
and the inner or outer cylinder is rotated slowly by a few revolutions in
one direction so that the dye appears to mix. Finally, the same cylin-
der is rotated backwards by the same number of revolutions (i.e. the
38 Chapter 2

surface stresses change sign), after which one observes that the dye
has returned to its original position, save for the influence of a small
amount of molecular diffusion. It appears that the reversal of the flow
has “unmixed” the fluid.
(2) When a single sphere is subject to a force parallel to a vertical plane
wall, the sphere translates parallel to the wall and rotates as if rolling
down the wall (Figure 2.1a): the force and the velocity are collinear and
there is no component of velocity perpendicular to the wall. One way
to draw this conclusion is to consider the case of a downwards force
parallel to the wall so that the particle translates downwards. Sup-
pose that you thought that there would also be a motion of the sphere
perpendicular to, and away from, the wall. If you exactly reversed the
driving force parallel to the wall, the velocity of the sphere would re-
verse. Hence, you would then have the sphere translating upwards
parallel to the wall but the transverse component would be towards
the wall, i.e. the velocities change sign when the applied force changes
sign. However, the two situations (upward versus downward forces par-
allel to the wall) are the same hydrodynamic problem, so you cannot
have motion away from the wall in one case and towards the wall in the
other case. Therefore, there is no component of velocity perpendicular
to the wall, and the sphere maintains a constant separation distance
from the wall.
(3) When two identical spheres, i.e. with the same radius and the same
density, sediment in a viscous fluid at a speed such that the Reynolds
number is small, their separation distance does not change; this case
is discussed below as part of Figure 2.8b. As a consequence of hydrody-
namic interactions, the spheres may rotate as they sediment and their
direction of fall may be at an angle to the vertical, but they have no
component of relative velocity.
(4) In a low-Reynolds-number pressure-driven flow of neutrally buoyant
spheres of radius as in a straight tube of radius a  as , the parti-
cles are observed to migrate to a radial position approximately 0.6a
from the centerline, as first documented experimentally by Segré
and Silberberg.22 According to the equations of Stokes flow, the con-
straints of reversibility preclude any transverse migration of a sphere
off its original streamline. It is now recognized that inertial effects, i.e.
the fact that the Reynolds number may be small but it is still finite, can
explain this experimentally observed cross-streamline migration. In
addition, deformation away from a spherical shape can lead to drift in
the direction normal to the wall (e.g. Figure 2.1d), as discussed briefly
in Section 2.10.
In a problem of this type, where transverse migration occurs, it is
important to (i) determine the typical speed V⊥ of migration across
the streamlines and (ii) determine how far (z) along a pipe one needs
to go to make an effective measurement of the phenomenon. Clearly,
if we seek to document transverse displacements comparable to the
Low-Reynolds-Number Flows 39

Figure 2.4 Shape changes of two model hinged swimmers in the plane. (a) The sim-
ple hinge or “scallop,” which has a single joint that opens and closes.
No net translation occurs after one cycle in a Newtonian fluid at low
Reynolds numbers. (b) The two-hinged or “Purcell’s” swimmer, which
alternately rotates its two arms at the ends so as to achieve a cycle in
shape space (I → II → III → IV → I). Following the cycle of movements
illustrated in the figure, the object is capable of net translation in the
horizontal direction. Adapted from Becker et al.23

pipe radius a, which take a time a/V⊥ , then for a mean flow speed hvi
we must have an experimental apparatus longer than z O hvia/V⊥ .
(5) As a preview of the kinds of question that are raised when the move-
ment of self-propelled microorganisms is considered (see Chapter 4,
by Pak and Lauga), we consider the translation due to the actuation
of a hinged object here. Purcell1 noted that a simple single-hinged ob-
ject, which first opens and then closes, is not capable of net translation
after one opening–closing cycle because of the reversibility constraint;
see Figure 2.4a. In other words, for such a simple object, the move-
ments over the first half of the cycle are exactly canceled by those over
the second half of the cycle, which are driven by an exact reversal of
the surface stresses. No net translation is possible.
Purcell made the observation that a two-hinged object undergo-
ing a cyclic set of deformations is capable of translating because it
evades the constraint of kinematic reversibility. The idea of the two-
hinged object is illustrated in Figure 2.4b and is known as the Pur-
cell swimmer. Although one can deduce that it is possible for the
object to translate, a detailed calculation is necessary to determine
the direction and speed of translation.23 Perhaps not surprisingly, re-
searchers interested in small-scale robotics need to understand kine-
matic reversibility and dynamics such as that documented in this
example.
(6) In the low-Reynolds-number mixing of viscous fluids, breaking the
constraints of kinematic reversibility is one of the keys to effective
stirring so as to effectively disperse one fluid throughout the other
fluid.24 In other words, it is necessary to avoid the “unmixing” feature
described in item (1) above. Experts will emphasize that molecular dif-
fusion is eventually responsible for “mixing” at the smallest length
scales.
40 Chapter 2

(7) Finally, we note that when a flexible filament is deformed by viscous


stresses, the object does not revert back to its original shape upon
reversal of the flow. Taylor provided an illustration of this interplay
of elasticity and viscous stresses in the wonderful film Low Reynolds
Number Flows. 2 The basic idea is shown in Figure 2.5: a cylinder is
filled with syrup and the inner cylinder is rotated. Three examples
are given: (a) a square is drawn with colored syrup, and kinematic re-
versibility is shown by rotating the cylinder first clockwise through a
given angle and then counterclockwise through the same angle; (b) a
rigid ring (with a lighter spot on it to indicate orientation) is displaced
and rotates, and will go back to its original position and orientation
upon reversing the flow; (c) a flexible but inextensible fiber (yarn) is de-
formed (the filament buckles) upon reversal of the flow and does not
revert back to its original shape. These images capture both the con-
cept of kinematic reversibility and how the effect can be broken. The
coupling of viscous stresses and elasticity in the dynamics of slender
filaments is discussed more in Chapter 3, by Stone and Duprat.

Perhaps not surprisingly, many physical features can destroy kinematic


reversibility, for example elasticity of the boundaries (as illustrated in Figure
2.5), viscoelasticity of the fluid, surface tension acting at an interface, defor-
mation of suspended particles, and finite inertia (Reynolds number) of the
flow, as already indicated in example (4) above. All of these ideas have been
studied, though many questions remain.

2.5 Mathematical Features


2.5.1 The Lorentz Reciprocal Theorem
This section is specialized but serves to introduce a theoretical idea—
the Lorentz Reciprocal Theorem—that is often helpful in a variety of
low-Reynolds-number flow problems for getting insight and for bypassing
detailed flow calculations. In the interests of clarity, we give the detailed
steps here. The basic idea is similar to Green’s second identity, often intro-
duced in undergraduate advanced calculus courses when the Laplace equa-
tion is discussed. In this case, the integral identity was introduced in 1896
by the Dutch physicist Lorentz (see the English translation, 25 which was
published in the centenary of this landmark publication); for a historical
discussion of this important paper, the reader is referred to Kuiken. 26
Suppose we want to solve the Stokes equations in a given domain but pos-
sibly for two different sets of boundary conditions. For example, we may
wish to contrast the flow around and the drag on a rigid sphere with no-
slip boundary conditions with the corresponding flow and drag when there
is slip on the surface of the sphere. For one problem, we denote the solution
by (v, s), and for the other, (v̂, σ̂). With f and f̂ denoting the body forces, the
Low-Reynolds-Number Flows 41

Figure 2.5 Breaking kinematic reversibility with an elastic filament in a viscous


flow. In the experiment, a cylinder is filled with syrup and the inner
cylinder is rotated. (a) A square is drawn with colored syrup (so there is
no surface tension between the two phases). The cylinder first rotates
clockwise through a given angle and then counterclockwise through
the same angle (the images should be looked at clockwise, starting
at the upper left). The colored syrup returns to its original position
and shape. (b) A rigid ring, which includes a lighter spot to indicate
orientation, is displaced and rotates. The ring returns to its original
position and orientation upon reversing the flow. (c) A flexible but in-
extensible fiber (in this case a piece of yarn) is deformed—the filament
buckles—upon reversal of the flow and so the fiber does not revert back
to its original shape. This last example illustrates the breaking of kine-
matic reversibility. Adapted from the film Low Reynolds Number Flows by
Taylor.2
42 Chapter 2

equations for the two cases are


V · σ C fZ0 and V · σ̂ C f̂ Z 0. (2.26)
Taking the inner product of the first equation with v̂ and that of the second
equation with v and subtracting yields
(V · σ) · v̂ K (V · σ̂) · v C f · v̂ K f̂ · v Z 0. (2.27)
The chain rule of differentiation applied to tensorial quantities allows us to
write k
V · (σ · v̂) Z (V · σ) · v̂ K σ : Vv̂, (2.28)
and similarly for V · (σ̂ · v). For an incompressible flow of a Newtonian fluid,
σ : Vv̂ Z σ̂ : Vv (see the footnote, which the reader is encouraged to work out
as an example.∗∗ ) Thus, using eqn (2.28) and the equivalent result for V ·
(σ̂ · v), upon rearrangement of eqn (2.27) we have
V · (σ · v̂) K V · (σ̂ · v) C f · v̂ K f̂ · v Z 0. (2.29)
Finally, integrating eqn (2.29) over the fluid domain V and using the Diver-
gence Theorem to convert volume (V) integrals to surface (S) integrals leads
to the integral identity
Z Z Z Z
n · σ · v̂ dS C f · v̂ dV Z n · σ̂ · v dS C f̂ · v dV, (2.30)
S V S V

where n is the unit normal directed outward from the fluid domain and
S denotes all bounding surfaces of the domain. In the most common
applications, we take f Z f̂ Z 0, in which case we have
Z Z
n · σ · v̂ dS Z n · σ̂ · v dS. (2.31)
S S

This result was first given by Lorentz in 1896!


For one elementary example of slip flow on a sphere, see section 2 of
Stone. 27 For applications of the idea to various problems involving particle
motions, see Happel and Brenner 3 and Leal.28 We will illustrate the use of
this integral identity below (e.g. in Section 2.5.5).

2.5.2 A Point Source: An Idea Illustrated with the Laplace


Equation
Understanding the fluid response due to a localized force, or point force,
is fundamental to understanding the mathematical features of many Stokes
flows. To begin with it is simpler to consider the Laplace equation, V2 f Z 0,

k The“double dot” notation for two second-order tensors means A : B Z Aij Bj i .


∗∗ Sinceσ is a symmetric second-order tensor and Ê is the symmetric part of the velocity
gradient tensor Vv̂, then σ : Vv̂ Z σ : Ê. Also, for a Newtonian fluid of constant viscosity,
σ Z KpI C 2mE. Since I : Ê Z V · v̂ Z 0, then σ : Ê Z 2mE : Ê. Similarly, σ̂ : Vv Z σ̂ : E Z 2mÊ : E.
Therefore we conclude that s : Vv̂ Z ŝ : Vv. The reader should verify all of these statements.
Low-Reynolds-Number Flows 43

which has many properties in common with the Stokes equations. Let n de-
note the unit normal vector defined on an arbitrary surface Sp and directed
away from the domain V.
The field f(r) created by a specified surface flux j(r) on Sp is the solution
to the boundary value problem
V2 f Z 0 with K n · Vf Z j(r) on Sp , f→0 as r Z |r| → ∞, (2.32)
where the minus sign in the flux boundary conditions follows from the con-
vention that the flux is positive when directed from high to low f. The net
flux Q from the object (surface Sp ) is a constant and is defined as
Z Z
QZK n · Vf dS Z j(r) dS Z constant. (2.33)
Sp Sp

Now, we can draw a conclusion about the characteristics of the field f(r)
far from theR object, i.e. r Z |r| → ∞. Starting with the Laplace equation, we ob-
serve that V V2 f dV Z 0. Denoting the two surfaces that bound the domain V
by Sp , the surface of the object, and S∞ , which refers to a spherical bounding
surface of radius r at some large distance, we then obtain the following using
the Divergence Theorem:
Z Z Z
V2 f dV Z 0 ⇒ n · Vf dS C n · Vf dS Z 0. (2.34)
V Sp S∞

Since the particle has a constant total flux (eqn 2.33), we conclude that
n·Vf dS Z constant. Since dS Z O (r2 ) for a large bounding surface of ra-
R
S∞
dius r (think spherical coordinates), then, based on this last integral, it also
follows that at large distances |n · Vf| Z O (rK2 ). Therefore, upon integration
we conclude that f Z O (rK1 ) as r → ∞. This rK1 decay, which is normally
described as a “slow algebraic decay,” is characteristic of a point source (or
sink) in three dimensions. The property is a feature of the Laplace equation,
as this argument demonstrates, and the conclusion was arrived at without
solving a specific boundary value problem in detail.
Returning to the model problem, far from the object we only see it as a
“point source,” which is characterized by the total flux Q rather than the
detailed surface flux. By definition, the point-source solution fps (r) Z c/r,
where c is a constant, satisfies the Laplace equation with a localized, or delta
function, forcing:
V2 fps C Qd(r) Z 0. (2.35)

RRecall that the delta function is defined formally as d(r) Z 0 for r 6Z 0, with
V
d(r) dV Z 1. By integrating eqn (2.35) over the domain V and using the
Divergence Theorem, we arrive at
Z Z Z
2
V fps dV C Q d(r) dV Z 0 ⇒ n · Vfps dS Z KQ, (2.36)
V V S∞

where we remember that n is outward from the domain. With fps Z c/r,
then n · Vfps Z Kc/r2 and the spherical integration in eqn (2.36) gives a
44 Chapter 2

factor 4πr2 , so that 4πc Z Q. Therefore, we have determined the detailed


point-source solution,
Q
fps Z , (2.37)
4πr
which describes the field far from any object with net flux Q. Notice how
the mathematical properties were deduced using geometrical features of the
Laplace equation rather than by solving a detailed boundary value problem.

2.5.3 Far-Field Decay of the Stokes Equations


With the above background based on a study of the Laplace equation, we
can now ask about the solution to the Stokes equations when a particle sed-
iments in an unbounded fluid domain. In this case there is a net force Fext
on the particle. Away from the particle, we observe fluid motion, and it is
common to speak of this as a flow driven by the (point) force exerted by
the particle. We now follow the same steps as shown in Section 2.5.2 for
the Laplace equation to deduce some basic properties of the velocity and
pressure distributions in this Stokes flow.
We refer to the fluid domain as V and the bounding surfaces as S, which
consists of the boundary of the particle Sp and a boundary at large distances
S∞ . Since inertia is neglected in the Stokes flow limit, then the sum of all
forces on the particle equals zero: Fext C FH Z 0, where the hydrodynamic
H
R
force is F Z K Sp n · σ dS, with n being the unit normal directed away from
the fluid domain. The flow field results from imposing boundary conditions
appropriate to a translating rigid object and solving the Stokes equations, so
one form of the problem statement is

V · σ Z 0, with v Z V on Sp and v → 0 as r → ∞. (2.38)

We integrate V · σ Z 0 over the fluid domain and use the Divergence


Theorem to arrive at
Z Z
n · σ dS Z K n · |{z}
σ |{z}
dS , (2.39)
Sp S∞
| {z } O(rK2 ) O(r2 )
ZFext

where we have indicated typical orders of magnitude, which we now describe.


The integral on the left-hand side of eqn (2.39) is the force Fext the fluid exerts
on the particle, which is a constant vector. So, the right-hand side of eqn
(2.39) must also be a constant: since dS grows as r2 , where r is the radius
of a large sphere representing the surface S∞ , we observe from the integral
that |σ| Z O (rK2 ). With this variation of the stress field, given the form of the
constitutive equation for a Newtonian fluid, we then conclude that
 
far field of a point force: p Z O rK2 and |v| Z O rK1 . (2.40)
Low-Reynolds-Number Flows 45

The slow algebraic decay |v| Z O (rK1 ) of the velocity field is a characteristic
feature of a localized force in an unbounded low-Reynolds-number flow. ††

2.5.4 The Point-Force Solution to the Stokes Equations


In
 pfthepfcasepf of the Stokes equations, the point-force solution, or Stokeslet,
p , v , σ , corresponds physically to the flow field far from the particle,
produced by a sedimenting (or forced and translating) particle in a fluid oth-
erwise at rest. Far from the particle, the detailed shape does not matter. The
continuity and Stokes equations for a point force Fext at the origin are
V · vpf Z 0 and V · σpf C Fext d (r) Z 0. (2.41)
The solution to these equations yields the pressure (a scalar), the velocity
field (a vector), and the stress field (a second-order tensor), each of which
is linear in the forcing Fext . In particular, it can be shown (for example, the
use of Fourier transforms is a convenient solution approach, and a direct
solution is given in Section 2.7.4; see eqn (2.79)) that
Fext · r
p pf (r) Z , (2.42)
4πr3 
1 I rr
vpf (r) Z Fext · J(r), where J(r) Z C 3 , (2.43)
8πm r r
3 rrr
spf (r) Z Fext · K(r), where K(r) Z K . (2.44)
4π r5
In some publications, the symmetric second-rank tensor J(r) is referred to as
the Green’s function for Stokes flow, in others it is referred to as the Oseen
tensor, and in yet others it is referred to as a Stokeslet. Notice that the tensor
K is of third rank and maps the force Fext to the second-rank stress tensor
σpf .
There are two consequences of these point-force formulae that are often
utilized.

(i) Notice how the tensor J representative of the point force decays as rK1
in the far field (r → ∞). Also, the the pressure field p pf and the ten-
sor K representative of the corresponding stress field decay as rK2 . We
deduced both of these properties in the previous section, Section 2.5.3.
(ii) Consider the fluid speed at some distance parallel to the point force
versus the speed at the same distance perpendicular to the point force,
by examining eqn (2.43). At a distance r from a point force, we com-
pare the speed vk , where the location is in the direction of the point
force (i.e. r · Fext Z rF ext ), with the speed v⊥ , where the location is per-
pendicular to the direction of the same point force (i.e. r · Fext Z 0). We
observe that

†† Exercisefor the reader: show that if a torque is applied to an isolated particle, the velocity
decays in the far field as |v| ≈ rK2 , which is still algebraic but is one power of r faster than in
the case of an applied force.
46 Chapter 2
vk
Z 2. (2.45)
v⊥
We will find on several occasions that this factor of 2 that appears
on comparing “parallel versus transverse speeds” is fundamental to
a variety of low-Reynolds-number flows.

2.5.5 An Integral Equation Representation of the Solution to


the Stokes Equations
Suppose that we wish to solve some incompressible Stokes flow problem for
(v, σ) in an unbounded domain V bounded internally by the surface Sp of
some object, i.e. we solve V · σ Z 0 and V · v Z 0. For most shapes, the so-
lution requires numerical means. In this section, we indicate one common
approach to a numerical solution in the form of an integral equation.
Consider the point-force problem, which we denote by V · σpf C Fd (r) Z 0,
where r Z x K y is the position vector between two points, say a field point
x and the location y of the point force. Recall the Reciprocal Theorem, eqn
(2.30), where in this case we use the point-force field in place of {v̂, σ̂}. This
integral identity simplifies to
Z Z Z
n · σ · vpf dSy Z n · σpf · v dSy C F · d(x K y)v dVy , (2.46)
S S V
where y is the integration variable. We take S as representing both the surface
of the particle Sp and a bounding surface at infinity S∞ . However, note that
we can neglect contributions from a surface at “infinity,” since the kernels of
the two surface integrals in eqn (2.46) decay there at least as fast as O (rK3 )
and dS grows as O (r2 ) as r → ∞. Hence, substituting into the point-force so-
lution (2.43) and (2.44) and using the properties of the delta function, we
obtain

1 if x ∈ V  Z Z
1/2 if x ∈ Sp F · v(x) Z F · n · σ · J dSy K F · n · K · v dSy , (2.47)
0 if x 6∈ V
 Sp Sp

where we have allowed for different possibilities for the choice of the position
vector x and have assumed that the surface is smooth, which establishes the
factor of 1/2 for positions on the surface, x ∈ Sp . The notation in eqn (2.47)
makes it clear that y is the integration variable. Since the vector F is otherwise
arbitrary in eqn (2.47), we conclude that

1 if x ∈ V  Z Z
1/2 if x ∈ Sp v(x) Z n · σ · J dSy K n · K · v dSy , (2.48)
0 if x 6∈ V
 Sp Sp

which is an integral equation that involves the velocity and surface stress
distributions on the surface Sp of an object.
The integral equation (2.48) for the velocity is often a convenient starting
point for numerical solutions. For example, if a rigid particle translates, then
Low-Reynolds-Number Flows 47

the velocity on the surface is known and eqn (2.48) is an integral equation of
the first kind for the unknown distribution of surface stresses n · σ. An excel-
lent reference for the many applications of these integral equation methods
to low-Reynolds-number motions is Pozrikidis. 29
For translation of a smooth rigid body at a velocity V, the second inte-
gral on the right-hand side of eqn (2.48) can be integrated by noting that
the stress version of the point-force equation is
V · K C d (x K y) I Z 0 for y ∈ S. (2.49)
Thus, using the Divergence Theorem, for points on the surface Sp sur-
rounding the particle volume Vp , where the velocity v Z V, we have (x ∈ Sp )
Z Z Z
1
n · K · v dSy Z K V · K dV y · V Z d (x K y) dVy I · V Z V. (2.50)
Sp Vp Vp 2
In this way, eqn (2.48) reduces to
Z
VZ n · σ · J dSy (valid for x ∈ Sp , V), (2.51)
Sp

which represents an integral equation of the first kind for the unknown
surface stress distribution n · σ .

2.6 General Features Related to the Motions of


Objects in Viscous Flows
Another class of problems of great importance concerns the flow of particle-
laden fluids. The particles may be rigid spheres, more complex rigid shapes,
or deformable objects. The latter may be liquid droplets in which interfacial
tension resists shape changes; capsules, which are liquid droplets encap-
sulated by an elastic shell; or vesicles or cells, which are bound by lipid
monolayers, bilayers, or multilayers.

2.6.1 Decomposing an External Flow into Simpler Problems


The generic fluid mechanics problem is sketched in Figure 2.6. A parti-
cle, with translational velocity Vp and angular rotation vector U p , is im-
mersed in an external flow v∞ . If r is the position vector measured from
the center of the sphere, then locally, in a region sufficiently close to the
sphere, the undisturbed velocity field can be constructed using a Taylor
series expansion,
v∞ (r) Z v∞ (0) C r · Vv∞ (r) (0) C . . .
1
Z v∞ (0) C ω ∞ (0) ∧ r C E∞ (0) · r C . . . , (2.52)
2
where
1 ∞ T

ω ∞ Z V ∧ v∞ and E∞ Z Vv C (Vv∞ ) . (2.53)
2
48 Chapter 2

Figure 2.6 (a) A particle is immersed in an external flow field v∞ (r). The particle
has a translational velocity Vp and an angular rotation vector U p . In the
neighborhood of the particle the flow field can be written as a Taylor
series, which identifies the local flow velocity v∞ (0), the local vorticity
vector ω ∞ (0), and the local rate-of-strain tensor E∞ (0). (b) Decompo-
sition of a simple shear flow into a rotating flow, which represents the
contribution of vorticity, and a pure straining flow oriented at 45◦ to the
flow direction, which represents the contribution of the rate of strain.

The first term on the right-hand side of eqn (2.52), v∞ (0), represents a
local uniform motion. The linear variations of velocity are captured by the
vorticity vector ω ∞ and rate-of-strain tensor E∞ ; the local vorticity ω ∞ rep-
resents twice the angular velocity of the fluid and is equivalent to the an-
tisymmetric part of the velocity gradient tensor (this topic is standard in
graduate textbooks). Also, the rate-of-strain tensor E∞ indicates the rate of
stretching of fluid elements, as well as the orientations of maximum exten-
sion and compression. This description highlighting only the linear varia-
tions in the local velocity field in the vicinity of a particle is expected to be
a good approximation when the largest particle dimension is smaller than
the typical distance over which the velocity gradient changes. In particular, a
simple shear flow, as sketched in Figure 2.6b, can decomposed into a locally
rotating (vortical) flow and a straining or extensional flow oriented at 45◦ to
the flow direction. In this case, the magnitude of the vortical flow is equal to
the magnitude of the straining flow.
There are three common types of problems that can be discussed for par-
ticles in viscous flows: (i) motion of the particle in an external applied flow,
(ii) the squirming or swimming of an object produced by a velocity distribu-
tion on the particle surface, and (iii) the flow field created by the particle. We
briefly discuss these items in the next three subsections.

2.6.2 Generalized Forces and Velocities: Forces, Torques,


and the Stresslet Tensor
There are many problems that involve the translation and rotation of parti-
cles in a fluid (e.g. Figures 2.6a and 2.7a). In some cases the particles move
Low-Reynolds-Number Flows 49

Figure 2.7 Types of problems involving particles in flow. (a) Motion of particles
due to applied forces and torques as they relate to the external flow
represented via a uniform motion, vorticity, and a rate-of-strain field.
(b) Squirming motions produced by a velocity distribution on the parti-
cle surface. (c) Flow fields: experimental results and model of the flow
field due to the swimming of a biflagellated algal cell Chlamydomonas
reinhardtii. The example shown here shows the time- and azimuthally
averaged flow field measured in an experiment on a swimming cell (left)
and a model based on three point forces (right). Figure reprinted with
permission from Drescher et al.,33 Phys. Rev. Lett., 2010, 105, 168101.
Copyright (2010) by the American Physical Society.

because of an external flow, in others the particles experience a net force or


torque, and in other cases the particles, for example swimming microorgan-
isms, are self-propelled and so experience no net force or torque but achieve
net translation and/or rotation by deforming their body in nonreciprocal
ways; see Chapter 4, by Pak and Lauga.
Because of the linearity of the Stokes equations, we can consider the gen-
eral problem illustrated in Figure 2.6 to be decomposed into three simpler
problems, as sketched in Figure 2.7a. If the velocity field v is considered as
the sum of an external flow v∞ and a disturbance flow vd , i.e. v (r) Z v∞ (r) C
vd (r), then the contributions to the disturbance flow come independently
from
50 Chapter 2
1
Vp K v∞ (0) , U p K ω ∞ (0) , and E∞ (0) . (2.54)
2
Typically we think of the slip velocity Vp K v∞ (0) as arising from an ap-
1
plied external force Fext and the difference in rotation rates U p K ω ∞ (0)
2
as arising from an applied external torque Lext . We know that the force and
torque are given in terms of the stress tensor σ by the surface integrals
Z Z
ext ext
F Z n · σ dS and L Z r ∧ n · σ dS, (2.55)
Sp Sp

where the moment is measured from the center of mass of the particle and
n is directed outward away from the fluid domain, consistent with its use
earlier in this chapter. The torque is related to the antisymmetric part of the
first moment of the surface stress distribution. 5,30
Next, recall that the rate-of-strain tensor E∞ (0) characterizes the rate of
stretching of fluid elements. As a rigid particle, or a soft object that resists
deformation, is incapable of deforming similarly to the local fluid, then the
third subproblem here (due to E∞ ) is related to additional stresses in the
fluid, and so to a disturbance flow, associated with the resistance to rate of
strain of the undisturbed motion. This effect is a change in the flow due to a
“stresslet”; in other words, the particle creates a change in the flow field that
influences the effective stress in the fluid.30 Thus, it is common to measure
the symmetric part of the first moment of the surface stress distribution and
so define the stresslet tensor,
Z
1
SZ (rn · σ C n · σ r) dS. (2.56)
2 Sp

Because the Stokes equations are linear, for a rigid particle there is a linear
mapping between the generalized forces mentioned in the previous para-
graphs and the corresponding velocities. For example, if a particle translates
and rotates in a fluid otherwise at rest, i.e v∞ Z 0, then we expect that a force
is required such that

Fext Z m RFV · Vp C RFU · U p ,



(2.57)

and a similar equation can be written for the torque on the particle. The
coefficients Rij in eqn (2.57) are second-order tensors, as they map one vector
to another vector. For example, for a sphere, the tensor RFV is proportional
to the identity tensor RFU Z 0.
More generally, in the case where there is a far-field motion v∞ (r), the
most general linear mapping of velocities to the generalized forces can be
written as
 ext     V K v∞ (0) 
F RFV RFU RFE p
Lext  Z m RLV RLU RLE  ·  1
U p K ω ∞ (0) , (2.58)

S RSV RSU RSE 2

E (0)
Low-Reynolds-Number Flows 51

where S is the stresslet tensor (a second-order tensor), which characterizes


the resistance of the particle to straining motions. The second-order ten-
sors (RFV , RFU , RFU , RLU ), the third-order tensors (RFE , RLE , RSV , RSU ), and the
fourth-order tensor (RSE ) that appear in eqn (2.58) characterize a rigid body
in a general linear flow. These tensors have special mathematical proper-
ties; for example, many are symmetric tensors, as may be shown using the
Reciprocal Theorem.31 In particular, with recourse to index notation where
convenient,

RFV Z RTFV , RLU Z RTLU , RFU Z RTLV , (2.59a)


(RFE )kij Z (RSV )ij k , (RLE )kij Z (RSU )ij k , (RSE )ij k` Z (RSE )k`ij . (2.59b)

This formalism for relating generalized forces and generalized veloci-


ties is very useful in developing numerical algorithms for the motion of
suspensions. 32

2.6.3 Squirming Motions Produced by a Velocity


Distribution on the Particle Surface
A second type of particle motion problem is common in studies of swimming
microorganisms, where it is referred to as a “squirming” motion. In this case
the particle is force-free (and often torque-free as well) but can still translate
and rotate through the fluid. The particle motion is created by velocities that
are specified on the surface of the particle (Figure 2.7b). The challenge is
to relate the surface velocities to the macroscopically observed translational
and rotational speeds.
For example, in the case of a force-free and torque-free sphere of radius
a with surface Sp , suppose that in the reference frame of the sphere there
are surface velocities vs (with amplitudes sufficiently small that the shape
remains nearly spherical). Then, the spherical squirmer has a translational
speed34
Z
1
Vp Z K vs dS. (2.60)
4πa2 Sp
The topic of squirming is discussed at more length by Pak and Lauga in
Chapter 4.
To derive this last result, we can provide a nice example of the Recipro-
cal Theorem, eqn (2.31), which shows the utility of the integral equation
representation. Specifically, we imagine a force-free sphere that has a pre-
scribed surface velocity distribution, for example vs Z vs (q) eq , where q Z 0 is
the “front” of the sphere and q Z p is the back. From the standpoint of an
observer in the laboratory, the velocity on the surface is v Z Vp C vs , where
Vp is the unknown translation velocity of the sphere.
To determine the translation velocity Vp of this squirmer, we begin with
eqn (2.31). We assume the other problem to be considered is that of the
translation of a rigid sphere with velocity V̂. In this case it is known that
52 Chapter 2

the stress distribution on the surface of the sphere is n · σ̂


σ Z K3mV̂/2a.
Substituting these conditions into eqn (2.31) yields

Z Z
3mV̂ 3mV̂
n · σ dS · V̂ Z K · 4πa2 Vp K · vs dS. (2.61)
S a 2a S

The first integral on the left-hand side equals zero, since the sphere is force-
free. Eliminating the constant V̂ yields eqn (2.60).

2.6.4 Flow Fields Due to Forces


A third type of problem focuses on the flow field created by particles or by
representative forces. The flow field can influence the velocities and ori-
entations of particles in the fluid and can also influence heat and mass
transfer. Thus, there is great interest in being able to calculate the velocity
field due to a set of forces. For example, in Figure 2.7c we show flow fields
from experiments on a swimming biflagellated algal cell Chlamydomonas
reinhardtii. Also shown is a model based on three point forces representing a
biflagellated swimmer, which reproduces the major features of the flow.

2.7 Translation and Rotation of Spheres


2.7.1 Flow Field Around a Translating Sphere in an
Unbounded Fluid
The most basic problem involving the hydrodynamics of suspended particles
concerns the translation and rotation of a sphere of radius a in a bath of
liquid. We denote the translational velocity by V and the rotational velocity by
U. When the Reynolds number is small, the velocity field can be determined
using standard (though laborious) methods. In the frame of reference of the
sphere, the problem statement is

V · vZ0 and m V2 v Z Vpd (2.62)

with boundary conditions

v Z U ∧ r on r Z a and v → KV as r → ∞. (2.63)

Most textbooks describe the axisymmetric solution in spherical coordinates


(r, q), which is best done by making all variables dimensionless; here q is
measured from the front of the sphere.
Because the equations and boundary conditions are linear in v, we can
consider separately the problems of translation and rotation. For example,
Low-Reynolds-Number Flows 53

for translation (with V Z |V|) the solution is


a3
 
3a
vr (r, q) Z KV cos q 1 K C 3 , (2.64a)
|2r {z 2r }
|{z}
uniform
flow disturbance caused
by the sphere
a3
 
3a
vq (r, q) Z V sin q 1 K K 3 , (2.64b)
4r 4r
3mVa
pd (r, q) Z p∞ C cos q. (2.64c)
2r2
Since the “1” in eqn (2.64a) and (2.64b) is the uniform motion far from the
sphere, we can conclude that the velocity disturbance decays at large dis-
tances in proportion to rK1 (the instantaneous streamlines are sketched in
Figure 2.8a). This result has already been deduced in Section 2.5.3 for all
unbounded three-dimensional configurations where a net force acts on the
fluid. Also, the pressure is high in front of the particle (q Z 0) and low behind
the particle (q Z π). The pressure decays as rK2 .
The hydrodynamic force FH exerted by the fluid on the sphere is calculated
by integrating the stress tensor over the surface of the sphere S:
Z
FH Z n∗ · σ dS, (2.65)
S

where n is the unit normal directed from the sphere surface into the fluid
domain. This integral has contributions from both the pressure and the vis-
cous stresses. The result (due to Stokes in 1851) is FH Z K6πmaV and is
known as the Stokes drag formula. Most significantly, the order of magni-
tude of this result is the product of a typical shear stress mV/a and the surface
area of a sphere 4πa2 .

2.7.2 Sedimentation
When a sphere of radius a and density rs O r sediments in a fluid, the net
gravitational force on the sphere is Fext Z (4πa3 /3) (rs K r) g. Thus, balanc-
ing the gravitational force with the viscous drag FH Z K6πmaV, the terminal
sedimentation velocity is
2a2 (rs K r) g
VZ . (2.66)
9m
This formula was already provided at the outset in Section 2.1 and has the
distinctive feature that the sedimentation speed varies as the square of the
particle size.
As for typical orders of magnitude, the reader can verify that a sphere of ra-
dius 1 µm sedimenting in water (m Z 10K3 Pa s) has a fall speed of about V Z
2 µm sK1 . The corresponding Reynolds number is about 10K6 . On the other
hand, a particle of similar density with radius 0.1 mm Z 100 µm sediments
at 1 cm sK1 , for which the corresponding Reynolds number is about unity,
54 Chapter 2

Figure 2.8 (a) Streamlines of the flow around a sphere translating at speed V in a
fluid at rest. (b) Two identical rigid spheres sediment in a fluid with a
velocity modified in magnitude and direction from the case of an iso-
lated sphere. In particular, the separation distance remains constant in
an ideal low-Reynolds-number motion in the absence of wall effects.

and so is near the upper limit where the equations of low-Reynolds-number


hydrodynamics are a rational approximation.
When two particles interact hydrodynamically, their sedimentation speed
and direction are altered (Figure 2.8b). Specifically, two spheres do not sim-
ply sediment aligned with the gravity vector but rather, because of hydro-
dynamic interactions, the sedimentation velocity is directed at an angle be-
tween the vertical and the axis of the particles (Figure 2.8b); also, the spheres
maintain a constant separation distance (see the discussion of reversibility
in Section 2.4.2). In some sense, two identical spheres serve as an approxi-
mation to an extended object that also sediments at an angle to the vertical;
see Section 2.8. For a complete overview of such hydrodynamic interactions,
the reader is referred to Happel and Brenner 3 and Guazzelli and Morris. 5

2.7.3 Some Historical Uses of the Stokes Drag Formula


(i) The simple hydrodynamic drag formula FH Z K6πmaV has played a
role in many basic developments in fundamental physics. For exam-
ple, Millikan measured the sedimentation speed of small oil drops in
an electric field and so determined the field needed to balance the
force of gravity. To interpret the results he needed the Stokes drag
formula to deduce the magnitude of the electrical force, and so the
electrical charge, on a small droplet. For this work, he won the Nobel
Prize in Physics in 1923.
(ii) As a second example, Einstein provided the original argument relating
the experimentally observed Brownian motion of suspended objects to
Low-Reynolds-Number Flows 55

the thermal fluctuations of surrounding liquid molecules. The ther-


mal energy is kB T, where kB is Boltzmann’s constant and T is the
absolute temperature. In one dimension (x), the translation diffusion
coefficient Dt is defined in terms of the mean square displacement hx2 i
according to (in the long-time limit)
hx2 i Z 2Dt t. (2.67)
Einstein defined the resistance coefficient of a sphere of radius a
as z Z F ext /V Z 6πam, where the result of Stokes has been used, and
predicted
kB T kB T
Dt Z Z . (2.68)
z 6πam
The experiments confirming Einstein’s predictions were performed by
Perrin, for which Perrin won the Nobel Prize in Physics in 1926. With
respect to orders of magnitude, it is useful to note that a sphere of
radius 1 µm in water at room temperature has a diffusion coefficient
Dt ≈ 0.2 µm2 sK1 ; consequently, the influence of Brownian motion is
continuously evident in the motion of particles of this size.
(iii) As a third example, we note that rapid centrifugation is an impor-
tant experimental procedure for measuring forces on small particles
whose density is different from that of the surrounding fluid. In the
early 1900s, Svedberg constructed an ultracentrifuge capable of ac-
celerations of 105 g. When a fluid undergoes rigid-body rotation with
an angular speed U, the radially directed external force on a parti-
cle of density rs and radius a at a distance r from the rotation axis
is F ext Z (4πa3 /3) (rs K r) U2 r, which can also be characterized by the
acceleration U2 r. The force resisting the motion is given by the Stokes
drag, 6πmav, where v Z dr/dt is the speed of the particle. This force
balance can be integrated, but it suffices to note that the ratio of
speed to acceleration establishes a time scale, which is tSved Z v/U2 r Z
(1/r)(dr/dt)(6πma/U2 ) here. This type of approach was used by Sved-
berg to determine the sizes of biological molecules (in particular, in
the discovery of the uniform molecular mass of proteins) and the dis-
tributions of their sizes, for which he won the Nobel Prize in Chemistry
in 1926.

So we see that the Stokes drag formula has been fundamental to at least
three Nobel Prizes!

2.7.4 Representation of the Solution with Vectors


We can convert the above solution, eqn (2.64a)–(2.64c), to vector notation.
Alternatively, it is possible to solve eqn (2.62) and (2.63) using methods that
work directly with vectors; here we consider only the problem of translation
and work in the laboratory reference frame in which we observe the trans-
lating sphere. To illustrate the main ideas, we scale all lengths by the sphere
56 Chapter 2

radius a, velocities by the speed of the sphere V, and pressures by mV/a. Thus
we seek to solve
V2 v Z Vp, V · v Z 0, (2.69)
with boundary conditions |v| → 0 as |r| Z r → ∞ and v Z V̂ at r Z 1 (here V̂ is a
unit vector in the direction of motion).
We outline a solution using ideas of vector calculus appropriate to linear
partial differential equations (the reader is encouraged to carry out the de-
tails of the calculation). We seek the vector field v r; V̂ . With respect to a
spherical coordinate system, the radial position in the fluid is r and the an-
gular position (say q) is specified by (r/r) · V̂. Therefore, since the solution for
the vector v must be linear in the forcing V̂, and the only other vector in the
problem statement is r, then the solution must have the vector form
v(r) Z V̂f (r) C V̂ · rrg(r), (2.70)
where f (r) and g(r) are two unknown functions of the scalar variable r. To
determine these functions requires a few steps of algebra (index notation is
helpful in the manipulations of vector calculus). To proceed, it is helpful to
note that Vr Z r/r and Vf (r) Z (r/r)f 0 , where f 0 Z df /dr. We now substitute
eqn (2.70) into the continuity equation, which yields
1 0
f C 4g C rg 0 Z 0. (2.71)
r
Next, it follows from the Stokes equations that V2 p Z 0 (e.g. recall eqn
(2.10)), from which we can conclude, using linearity in V̂, that the pressure
has the form
 
V̂ · r V̂ 3V̂ · rr
p(r) Z a 3 , so Vp Z a K . (2.72)
r r r5
Using eqn (2.70), and after some algebra, we find
   
2 2 0 00 6 0 00
V v Z V̂ f C f C 2g C V̂ · rr g Cg . (2.73)
r r
Then, substituting eqn (2.72) and (2.73) into the Stokes equations yields two
equations:
2 0 a 6 0 3a
f C f 00 C 2g Z 3 and g C g 00 Z K 5 . (2.74)
r r r r
These two ordinary differential equations are straightforward to solve by
finding particular and homogeneous solutions. We find
c b a b
f (r) Z K 3 , g(r) Z 3 C 5 . (2.75)
r 3r 2r r
These expressions involve three constants, c, a, b. Substituting eqn (2.75) into
eqn (2.71) yields c Z a/2, so that we have arrived at
   
a I rr b I 3rr
v(r) Z V̂ · C 3 K V̂ · 3 K 5 . (2.76)
2 r r 3 r r
Low-Reynolds-Number Flows 57

The no-slip boundary condition v Z V̂ at r Z 1 then yields the two equations


1 Z a/2 K b/3 and a Z K2b. Hence, b Z K3/4 and a Z 3/2. Therefore we
find    
3 I rr 1 I 3rr
v(r) Z V̂ · C 3 C V̂ · 3 K 5 , (2.77)
4 r r 4 r r
which is the vectorial equivalent of the disturbance velocity in eqn (2.64a)
and (2.64b).
To conclude this section, we can revert to dimensional variables, and
replace the translation velocity V by the external force acting on the parti-
cle, using V Z (6πma)K1 Fext . Then, the velocity field around the translating
sphere is
a2 ext
   
1 ext I rr I 3rr
v(r) Z F · C 3 C F · 3K 5 . (2.78)
8πm r r 24πm r r

2.7.5 The Limit of a Point Force: A Stokeslet


Suppose that the sphere gets smaller and smaller, i.e. a → 0, but we maintain
the same external force. This limit is referred to as a point force, and the
corresponding velocity field is vpf (r). According to eqn (2.78), in this limit
   
pf 1 ext I rr ext 1 I rr
v (r) Z F · C 3 Z F · J, where JZ C 3 ,
8πm r r 8πm r r
(2.79)
which was simply stated in eqn (2.43). Equation (2.79) is the Stokeslet
velocity field, whose instantaneous streamlines are sketched in Figure 2.8a.

2.7.6 A Rotating Sphere and a Point Torque: The Rotlet


Of course, it is possible to work out the details of the velocity distribution for
the case of a rotating rigid sphere. Let U denote the rotational velocity. Then,
the velocity distribution is
a3 U ∧ r
v(r) Z (2.80)
r3
and the corresponding pressure field is a constant (which can be taken as
zero with no loss of generality).
Note that the velocity distribution in this case decays as rK2 , which is
characteristic of a hydrodynamic “dipole,” and this decay is one power of
r faster than in the example of the hydrodynamic force or monopole (re-
call the footnote in Section 2.5.3). The corresponding hydrodynamic torque
LH Z K Sp r∧(n · σ ) dS (again using n as the unit normal directed away from
R

the fluid domain) that resists the rotation of the sphere is


LH Z K8πma3 U. (2.81)
It is significant that the hydrodynamic response to a torque varies as the
cube of the particle size. Similar ideas apply for nonspherical shapes, where
58 Chapter 2

it is the largest characteristic length that typically controls the torque. Fi-
nally, combining the two previous results, the response of the fluid to a point
torque Lext Z KLH has the velocity distribution vpt (r), where
Lext ∧ r
vpt (r) Z . (2.82)
8πmr3
Equation (2.82) is frequently referred to as the velocity field due to a rotlet.

2.7.7 Resistance to Rate of Deformation: The Stresslet


As discussed above in Section 2.6.2, when a force-free object is unable to
deform at the same rate as a fluid element, forces are exerted on the fluid,
but this can happen in such a way that the overall response is force-free.
As one example, we can think about a droplet immersed in an extensional
flow, for which surface tension resists deformation, effectively pulling back
against the flow with equal and opposite forces, and the drop can remain
nearly spherical, as sketched in Figure 2.9a.
As a more concrete example, imagine placing a spring in a viscous fluid,
stretching the spring by pulling on both ends, and then removing the forces
so that the spring relaxes. At every instant of time, during both extension and
relaxation, there are equal and opposite forces acting on the fluid. Far from
the spring, it is reasonable to expect that the flow field produced is that due
to equal and opposite point forces. As the spring is shrunk to a point, the
corresponding model “singular” flow is referred to as a “stresslet” velocity
field.
We should make it clear at the outset that the term “stresslet” is used in
two distinct ways in the field, which can be confusing. When discussing the
mechanics of suspensions, where the focus is on the effective stress in the
suspension, for example in the case of the effective viscosity, the stresslet

Figure 2.9 Stresslet. (a) A drop in an extensional flow field resists stretching. (b)
Two equal and opposite forces act in a fluid.
Low-Reynolds-Number Flows 59

(tensor) refers to the second-order tensor that characterizes the resistance


of the particle to stretching. For example, our discussion in Section 2.6.2
was consistent with this use of the term “stresslet,” and we introduced the
stresslet tensor S. On the other hand, in the subject of swimming microor-
ganisms (or active particles), or those areas where the focus is on the flow
field associated with suspended particles, the term “stresslet” characterizes
the velocity distribution in the far field around the object (see Chapter 4, by
Pak and Lauga). In this latter use, the stresslet is a vector!
More generally, to describe this kind of force-free flow situation we con-
sider two forces F Z Fe and KF, where e is a unit vector, spaced a distance Ddd
apart, where d is also a unit vector; see Figure 2.9b. We use e as a unit vector
associated with the direction of the force. The configuration is clearly force-
free by construction, and the idealized limit is constructed by letting D → 0;
this limit is often called a Stokes dipole. The symmetric part of the Stokes
dipole is called the stresslet tensor.
Using the principles of tensor analysis, we observe that the flow field will
correspondingly involve two vectors e and d and so can be represented by a
second-order tensor, which is equivalent to the stresslet tensor S introduced
in Section 2.6.2. In the absence of applied torques, this tensor should be
symmetric, and we also will assume that it is traceless, which a more detailed
analysis would show follows from incompressibility of the flow field. So we
can write
 
S0 2
SZ edd C d e K e · d I , (2.83)
2 3

where S0 is a constant. Thus, in this description, the stresslet tensor charac-


terizes the forces (or stresses) associated with active particles.
Next we must determine the corresponding flow field associated with the
forces or stresses acting on the fluid. In some of the literature it is this ve-
locity field, a vector, which is termed the stresslet. There are two approaches:
(i) solve for the flow around a sphere of radius a in an extensional flow, i.e.
v∞ (r) Z r · E∞ at large distances, determine the disturbance flow, and then
take the limit in which the sphere radius shrinks; the stresslet is the domi-
nant part of this flow field in the far field r/a  1. (ii) Determine the force-
free solution to the Stokes equations compatible with a specified (traceless)
second-order tensor. We will be content with (ii).
Recall Section 2.2.2, where we showed that the solution to the Stokes equa-
1
tions could be written in the form v (r) Z pr C vh (r), where V2 p Z 0 and
2
V2 vh Z 0. If we focus on the far field and demand that the flow be linear in
a second-order traceless tensor S, then we can construct scalar and vector
harmonic functions:

1 r
p (r) Z aS : VV and vh (r) Z bS · . (2.84)
r r3
60 Chapter 2

Note that p (r) Z 3a(S : rr/r5 ). Since V · v Z 0, one can show that b Z 0. Thus,
we find that the flow field, which we shall call vstr (r), must have the form
rrr
vstr (r) Z S : , (2.85)
r5
where we have dropped the prefactor since it can be absorbed into S. There
are several features of this result:

(i) The flow field decays in proportion to rK2 as r → ∞. This form is


characteristics of “dipoles.”
(ii) For a rigid sphere in an incompressible extensional flow E∞ , we have,
in dimensionless form, S Z (20πa3 /3)E∞ . 4,5
(iii) For those cases characterized by equal and opposite forces, as in eqn
(2.83) and Figure 2.9b, we can write the flow field as
 
rrr (e · r) (dd · r) r 1 e·d r
vstr (r) Z S : 5 Z S0 K , (2.86)
r r5 3 r3
which is the starting point for the corresponding discussion in
Chapter 4.

2.7.8 Other Shapes


Analytical results are available for a number of other representative shapes,
such as thin circular disks, and prolate and oblate spheroids. Approximate
results for long slender shapes, such as elongated prolate spheroids and long
circular cylinders, are described in Section 2.8. Otherwise, numerical simula-
tions are needed, and one approach to such numerical simulations is to use
the integral equation representation (2.51).
To appreciate that at low Reynolds numbers often the only effect on the
force–velocity relation is a modest change in the prefactor, we give some
results for translation of a thin circular disk of radius a. In particular, for
broadside and edgewise motions, respectively, we have
32
Fext Z 16maV and Fext Z maV. (2.87)
3
Not surprisingly, the force scales with maV, which is required on dimensional
grounds. The prefactor for edgewise motion is 2/3 the value for broadside
motion; evidently, edgewise translation is an “easier” motion for producing
a given speed. Also, the case of broadside translation produces nearly the
same drag (compare 16 with 6π) as a sphere of the same radius, so we have
the takeaway message that shape does not have a big impact on this type of
Stokes flow.
In addition, in-plane and out-of-plane rotation of a thin circular disk, with
angular rotation vector U , require a torque35
32 3
Lext Z ma U. (2.88)
3
Low-Reynolds-Number Flows 61

Perhaps surprisingly, the torque is independent of the rotation direction.


Here, however, we observe a change by about a factor of 2 from the torque
on a sphere of the same radius.

2.8 Slender-Body Theory


There are many examples where long slender objects, such as a needle, trans-
late in a fluid (e.g. Figures 2.1(b) and 2.10). For a cylinder of radius a and
length 2`, we define the aspect ratio as 3 Z a/`. The slender-body approxi-
mation is generally useful whenever 3  1. An excellent summary of various
aspects of slender-body theory has been given by Lauga and Powers. 19
We will describe several examples. Perhaps the most familiar uses of these
ideas in low-Reynolds-number flows are in the topics of swimming microor-
ganisms, which are discussed in Chapter 4, by Pak and Lauga, and the hydro-
dynamics of suspensions of rod-like particles. For example, the deformation
of elastic filaments by viscous flows occurs in a variety of small-scale appli-
cations; see, for example, Chapter 3, by Stone and Duprat, and Chapter 5, by
Lindner and Shelley.

2.8.1 Resistive Force Theory: Brief Summary


To begin with, we state some basic results for the relationship between the
force on and translation speed of a slender object. Then, we provide an
approximate argument for the form of the result.
In the simplest form of the slender-body approximation, known as resis-
tive force theory, we assume that locally the fluid motion is similar to a flow
past a cylinder for which the hydrodynamic force/length f H has different
components normal (⊥) and tangential (k) to the local body axis. Thus, if
the body translates with velocity V Z V⊥ , Vk , then we write the force/length
on the body as (e.g. Batchelor 36 and Cox 37 )
f H Z Kz⊥ V⊥ K zk Vk , (2.89)
where the resistance coefficients are dependent on the aspect ratio:
4πm 2πm
z⊥ Z and zk Z . (2.90)
1 1
ln (23K1 ) C ln (23K1 ) K
2 2
This purely local approximation is a great simplification but does not allow
the force/length at a given location to depend in any way on the velocities at
other positions along the slender object.
In general, the additive constants that appear in the logarithmic
  terms in
1
eqn (2.90) have slightly different values from those shown G , as such
2
values depend on the detailed cross-sectional shape of the slender particle.
We observe for 3  1 that z⊥ ≈ 2zk , as stated earlier (for example, see eqn (2.4)
and (2.45)), i.e. motion perpendicular to the centerline has approximately
twice the resistance of motion parallel to the centerline.
62 Chapter 2

Moreover, notice that the force/length on a slender particle depends only


weakly on the cross-sectional dimension, which appears in the argument
of the logarithm. Consequently, the total force on the particle is approxi-
mately the product of the force/length and the length. Thus, the net force
is proportional to the longest dimension of the slender object and is only
weakly dependent on the cross-sectional dimension. Finally, we note that
there have not been very many critical tests of these results. However, Roden-
born et al.,38 inspired by the topic of swimming microorganisms, reported an
experimental study of slender helical shapes and characterized the regime
of validity of the resistive force theory, which, in fact, appears to be more
limited than might be naively expected.

2.8.2 Resistive Force Theory: Derivation


The idea for the approximations (2.89) and (2.90) can be seen by imagining
that a chain of spheres is distributed side-by-side along the body centerline.
For a slender object, the velocity field then follows from an integral equa-
tion representation of the force/length fext distributed along the centerline
(s); see, for example, eqn (2.51):
Z `
(x K x0 ) (x K x0 )
Z  
ext 1 ext I
v(x) Z f · J dS ≈ f · C 3 ds, (2.91)
S 8πm K` x K x0 |x K x0 |
where we have used the Green’s function for 3D Stokes flow in the kernel
of the integral and x0 is a position vector to a location on the centerline
(position s).
Next, we assume that the force density is uniform and perform the integra-
tion along the centerline. Because of the apparent singularity in the kernel, it
is necessary to exclude a region O(a) to either side of the field point x. These
steps yield (e.g. Batchelor 36 and Cox 37 )
1
ln 23K1 (I C ll ) · fext ,

v(x) Z (2.92)
4πm
where l is a unit vector indicating the orientation of the centerline (here as-
sumed independent of position). We note that the integration of 1/s terms
yields the logarithmic term in eqn (2.92), and a factor of 2 arises from
integrating on either side of the singularity.  
K1 1
Equation (2.92) can be inverted by noting that (I C ll ) Z I K ll .
2
Therefore, the relationship between the force/length and the local velocity is
 
ext 4πm 1
f Z I K ll · v. (2.93)
ln(23K1 ) 2
From this expression we draw the conclusion that motions parallel to the
centerline, v kll, have half the resistance of motions perpendicular to the cen-
terline, v ⊥ll. For 3  1, the results are consistent to leading order in ln (2/3)
with the expressions given in eqn (2.90).
Low-Reynolds-Number Flows 63

2.8.3 Transport and Sedimentation of Slender Objects


When a long slender object sediments or otherwise translates in a viscous
fluid, the translation speed depends on the orientation of the particle relative
to the force; in particular, a slender particle sediments approximately twice
as fast when its axis is aligned with the force as compared with when it is
aligned with its axis perpendicular to the force (Figure 2.10a and b), and the
factor of two is motivated by the discussion in Section 2.5.4 as well as the
previous section. As a consequence, when a slender particle is at an angle
with respect to the vertical, it sediments with a velocity vector directed at
an angle between the vertical and the axis of the particle, i.e. it drifts at a
constant angle q (Figure 2.10c);‡‡ this effect is illustrated in the movie Low
Reynolds Number Flows. 2 This property of drag anisotropy (larger resistance
when moving perpendicular to the flow than parallel to the flow) is common
to all long, thin bodies of revolution translating in a large bath of fluid.

Figure 2.10 A slender fiber translates in a direction (a) perpendicular to its long
axis and (b) aligned with its long axis. (c) A slender particle sediments
with its long axis at an angle a to the direction of gravity, in which case
the direction of translation V is oriented at an angle a K q from the
vertical. (d) Sedimentation of a slender particle adjacent to a vertical
wall with “snapshots” illustrating the dynamics: two cases are shown
that differ in the initial orientation of the fiber. Reference: Russel
et al.39

‡‡ Exercisefor the reader: a needle is oriented at an angle a relative to the vertical. Assume
that the resistance to translation parallel to the axis of the needle is half the resistance to
translation perpendicular to the axis. Show that the sedimentation angle q satisfies
1 tan a
tan q Z tan a, so tan(a K q) Z ,
2 2 C tan2 a
and determine the maximum angle from the vertical that the trajectory can make. Hint: use
force balances for the directions parallel and perpendicular to the needle axis. You should
find that the maximum sedimentation speed corresponds to an orientation angle a ≈ 55◦ and
a trajectory of translation at about 19.5◦ .
64 Chapter 2

Finally, we mention several examples of slender-body motions. As one il-


lustration of the rich dynamics that can result when particles interact hydro-
dynamically with walls, we show in Figure 2.10d the position and orientation
when a slender particle sediments near a wall. 39 The fall speed and particle
orientation change with time, and there are two types of trajectories possible
as the particle first moves towards and then “rebounds” away from the wall.
The reader is referred to Russel et al.39 for a discussion of the dynamics.
Second, consider the case of an object whose symmetry is that of a helix;
this is particularly relevant to the study of the swimming of some microor-
ganisms, and is thus described in detail in Chapter 4. As pointed out by
Taylor in the movie Low Reynolds Number Flows, each segment of the helix
behaves similarly to an obliquely falling fiber. Thus, imagine a corkscrew
shape that sediments vertically in a large bath of viscous fluid: the corkscrew
rotates while it translates. In this case an external torque would be required
to stop the rotation that accompanies translation. Using the mathematical
characterization of eqn (2.58), RFU 6Z 0. In a similar spirit, motion parallel to
the axis of the helix (a “screwing” motion) leads to longitudinal propulsion,
i.e. a helix that is rotating, for example owing to an applied torque, translates
through a viscous fluid. Finally, in a shear flow, helices exhibit a net drift due
to the drag anisotropy. 40

2.9 The Jeffery Orbits of an Elongated Particle


Consider a force-free, rigid, orientable particle, such as a long cylinder or
slender rod, in a Stokes flow. Because the particle is force-free, we expect
that the particle will translate with the local velocity at the particle’s center
of mass. For orientable objects, we are also concerned with the rotation rate
and orientation of the particle; this was first considered by Jeffery in 1923.41
Thus, it is necessary to understand the velocity gradient in the neighborhood
of the particle, which requires knowledge of the rate-of-strain tensor E and
the rotation tensor U (or the vorticity vector ω Z V ∧ v),
1 T
 1 T

EZ Vv C (Vv) and UZ Vv K (Vv) . (2.94)
2 2
Also, we let l (t) denote the unit vector aligned with the instantaneous axis of
the rigid particle. The calculation that we describe in this section, known as
the “Jeffery orbit,” is the evolution equation for l (t).
To begin, recall from the discussions of kinematics in any introductory
fluid mechanics course that a local fluid element, labeled d`` for example,
is stretched and reoriented by the local velocity gradients; see Figure 2.11a.
Thus, with D/Dt denoting the material time derivative following the local
flow, a line element d`` in a fluid flow changes according to
D d`` 1
Z d`` · Vv Z ω ∧ d`` C d`` · E. (2.95)
Dt 2
The first term on the right-hand side of eqn (2.95) represents rotation of the
line element at the local fluid rotation rate (which is half the local vorticity),
Low-Reynolds-Number Flows 65

Figure 2.11 (a) Kinematics of a line element that translates with the local fluid ve-
locity, and rotates and stretches with the local velocity gradient. (b)
Orientation of a slender particle in a simple shear flow.

while the second term on the right-hand side, which involves the rate-of-
strain tensor E, is responsible for stretching (or compression) of the fluid
line element.
An orientable force-free rigid particle responds similarly to a line element,
but a rigid particle is not capable of stretching. Furthermore, the applied
strain rate may rotate the particle at a different rate from a fluid element. We
now introduce an evolution equation for l(t) that resembles eqn (2.95) but
accounts for these two differences between rigid and fluid particles. Thus,
we use eqn (2.95) for the rate of change of orientation l of the rigid particle
(so we now think of l rather than d``), but we eliminate the change of length
by writing the rate-of-strain term as
l · E K l · E · ll . (2.96)
The reader can check by taking the inner product of eqn (2.96) with l that
this manner of writing the rate-of-strain contribution introduces no change
of length.§§
Finally, we introduce a constant prefactor b to account for the difference
in the rotation rate caused by the stretching component of the flow and,
motivated by the form of eqn (2.95), we have an evolution equation for l,

dll 1
Z ω ∧ l C b (ll · E K l · E · ll ) (2.97)
dt 2

§§ To verify this statement, note that


(ll · E K l · E · ll) · l Zll · E · l K l · E · ll2 Z 0,
since l 2 Z 1.
66 Chapter 2

(note that we have switched to the ordinary derivative since we focus only on
changes in orientation with time).¶¶ The constant b is related to the aspect
ratio re (length/radius) of an axisymmetric spheroid, as first determined by
Jeffery:41
re 2 K 1
bZ 2 , where 0 ≤ b ≤ 1. (2.98)
re C 1
For other shapes, a common approximation is to replace re by an effective
aspect ratio for the particle.42 The main result of this section is eqn (2.97),
which is an evolution equation for the orientation of a particle l (t) given a
flow, as represented via the vorticity ω and the rate-of-strain tensor E.

2.9.1 Particle Rotation in the Plane of a Simple Shear Flow


We next describe the time-dependent orientation dynamics for specific flow
situations. The orientation of a particle can be described by the two angles
(f(t), q(t)) identified in Figure 2.11b.
As one model, we first consider a simple shear flow in a plane, so that
v Z ġyex is the undisturbed background flow; see Figure 2.11b. We assume
the particle begins oriented in this plane (i.e. q Z π/2) and so remains in
this plane, and hence we consider only two dimensions in this section. The
corresponding vorticity vector and rate-of-strain tensor are
 
ġ 0 1
ω Z Kġez and EZ . (2.99)
2 1 0
We describe the orientation of the particle by the angle f(t) measured
from the flow direction, so that l Z (cos f, sin f); the typical time scale for
tumbling is obviously ġK1 . Substituting into eqn (2.97), we find l · E Z
(ġ/2) (sin f, cos f) and l · E · l Z (ġ/2) sin f cos f. Substituting these results
into (2.97) and considering the x component yields
d ġ bġ
sin f K 2 sin f cos2 f ,

cos f Z K sin f C (2.100)
dt 2 2
which can be simplified to
df ġ  ġ
Z 1 K b 1 K 2 cos2 f Z (1 C b cos (2f)) . (2.101)
dt 2 2
This ordinary differential equation can be integrated analytically (with
f(0) Z 0) to obtain
 
1Kb 1 p
√ tan f(t) Z tan 2
tġ 1 K b . (2.102)
1 K b2 2
In terms of the aspect ratio re , this result can be written slightly more
simply as  
ġt
tan f(t) Z re tan . (2.103)
re C re K1

¶¶ Using eqn (2.97), verify that l · (dl/dt) Z 0.


Low-Reynolds-Number Flows 67

Figure 2.12 Jeffery orbits for a particle in the plane of a two-dimensional shear
flow. (a) Particle orientation f(t) in the plane of a simple shear flow for
re Z 5, 7; larger values of re correspond to longer-period orbits. (b) Plot
showing the change in the orientation of a fiber at equal time incre-
ments as the fiber rotates in the flow. The fiber begins with orientation
f Z 0 (aligned with the y axis), rotates rapidly towards the flow direc-
tion (f Z π/2), and then moves increasing rapidly away from the flow
direction.

The solution is a periodic function of time, where the typical time scale
is set by ġK1 . As an example, the solution for the angular orientation f(t),
Figure 2.12a, indicates that for f O 0 near the y axis (where f Z 0), an elon-
gated particle (e.g. b ≈ 0.9 or re ≈ 5) rapidly reorients and then approaches
alignment with the direction of flow more slowly. A small perturbation may
push the particle past the flow axis, in which case the particle first slowly
changes orientation and then rapidly flips again to approach alignment with
the flow direction. Not surprisingly, as is made clear by eqn (2.103), shorter,
less elongated particles flip with shorter periods and in a more smooth
fashion, while longer particles have long periods and rather more abrupt
rotational changes.

2.9.2 Three-Dimensional Particle Rotation in a Simple Shear


Flow
In this case it is convenient to use the same notation, but now to
let q(t) denote the angle the particle axis makes with the z axis; the
calculation in the previous section refers to q Z π/2 (see Figure 2.11).
Then, for the three-dimensional orientation of the particle we write l Z
(cos f sin q, sin f sin q, cos q). Thevorticity vector
 and rate-of-strain tensor are
0 1 0
as above, for example E Z (ġ/2)  1 0 0 . We can then calculate
0 0 0
and l · E · l Z 2ġ sin2 q sin f cos f.

l · E Z ġ sin q sin fex C sin q cos fey
(2.104)
68 Chapter 2

Figure 2.13 Jeffery orbits in a shear flow. Particle orientations in three dimensions
for different values of the orbit constant C.

Again, it is clear that the typical time scale for particle tumbling is ġK1 .
We then find an equation for f(t) that is identical to eqn (2.101) and an
equation for q(t):
dq ġb
Z sin (2q) sin (2f) . (2.105)
dt 2
Since f(t) is known (eqn (2.103)), it is straightforward to integrate to find
q(t). Indeed, as first found by Jeffery,
K1/2
tan q Z Cre re 2 cos2 f C sin2 f , (2.106)

where the orbit constant C indicates the initial off-axis orientation of the
particle; 0 ! C ! ∞, with C Z 0 corresponding to the axis of vorticity, perpen-
dicular to the shear plane, and C Z ∞ corresponding to the plane of the shear
flow. The particle continuously traverses a fixed orbit {f(t), q(t)}, which is the
Jeffery orbit (Figure 2.13). In the Stokes flow limit for an isolated particle, the
orientation is a periodic function of time and no steady orientations exist.
A particle only deviates from this trajectory if external forces are applied or
other hydrodynamic effects are present. Note that the trajectory for the case
C Z ∞ has been included in Figure 2.13.
Low-Reynolds-Number Flows 69

Figure 2.14 Drop deformation in shear flows. (a) Schematic illustration of the de-
formation of a drop from a spherical to an ellipsoidal shape. The typ-
ical dimensions are indicated, including the length L and the breadth
B. (b) Experimental images of a spherical drop (image 1) and the drop
deforming in a shear flow (images 2–4) at a capillary number Ca Z 0.4;
the two fluids have equal viscosities.43 For smaller deformations, the
drop aligns close to the extensional axes at 45◦ to the direction of the
shear flow, but the drop orients closer to the flow direction as the de-
formation increases. Figure reprinted with permission from Sibillo
et al.,43 Phys. Rev. Lett., 2006, 97, 054502. Copyright (2006) by the
American Physical Society.

These ideas have been applied in several directions, including the themes
of suspension rheology for fibers, comparisons with detailed numerical sim-
ulations to better understand the limitations of the theory, wall effects,
inertial effects, and fiber flexibility.

2.10 Drop Deformation and Drift in Linear Flows


When a neutrally buoyant, nearly spherical liquid drop is immersed in a
shear flow of a second immiscible fluid, the drop tends to translate at the lo-
cal liquid speed. The interfacial tension g at the fluid–fluid interface tends to
resist deformation, while viscous stresses from the flow tend to cause defor-
mation, principally along the extensional components of the flow. For small
deformations, the drop is approximately ellipsoidal. Here we describe briefly
some of the main features of this effect, such as the magnitude of the de-
formation and the rate at which nonspherical shapes drift across the flow
streamlines.

2.10.1 Drop Deformation


When a rigid particle translates at the local fluid speed it tends to rotate at
half the local vorticity—this is a local rigid motion—but it resists the stretch-
ing of fluid elements that is generated by the extensional part of the flow.
The resistance to stretching generates a disturbance in the local velocity
distribution, which is known as a stresslet.
70 Chapter 2

In the most common case, we have a shear flow (Figure 2.14a), which can
be decomposed into a local rotation and a local extensional/compressional
motion oriented at 45◦ degrees to the flow direction; the two contributions
are of equal magnitude (Figure 2.6b). It is the extensional part of the flow,
E∞ , that tends to deform a drop.
We denote the local shear rate by ġ, the continuous-phase viscosity by
m, and the drop viscosity by lµ. The viscous stresses have magnitude mġ.
Drop deformation causes a small change in the shape of the drop (Figure
2.14b). If the radius of the undeformed drop is a, then we can denote a
typical magnitude of the deformation by da, where d is a dimensionless quan-
tity. We then expect surface tension to produce a typical force/area of 2gk,
where k is the mean curvature, which is approximately aK1 (1 C d) in mag-
nitude. Thus, shape changes cause a change in the capillary stress that is
approximately O (gd/a). Balancing this stress against the viscous stresses
that drive deformation indicates a shape change d that is approximately
given by
gd mġa
mġ ≈ ⇒ d≈ Z Ca , (2.107)
a g
where Ca is known as the capillary number. This argument illustrates that
doubling the shear rate or the drop radius should double the magnitude of
the deformation. We also expect that the viscosity ratio between the two flu-
ids will influence the magnitude of the deformation; typically we find that it
is the larger of the two viscosities that dominates the deformation.
Studies of drop deformation in the manner discussed here were initi-
ated by Taylor in 1934,44 and this phenomenon has been studied extensively
over the years; see, for example, the reviews by Rallison45 and Stone.46 In
particular, Taylor presented theory and experiments that showed, in a man-
ner consistent with the above estimate, that for small changes away from
a spherical shape, the drop deformation D, defined in terms of the long
(L) and short (B) axes (Figure 2.14a), for example D Z (L K B)/(L C B), is
approximately
19l C 16
DZ Ca . (2.108)
16 (l C 1)
For small deformations, the drop is approximately aligned with the exten-
sional flow axis, which is at 45◦ to the shear flow direction; as the drop
deformation increases as the shear rate (capillary number) increases, the
drop becomes more aligned with the flow direction (Figure 2.6b).

2.10.2 Drift Transverse to Streamlines: A Stresslet Near a Wall


When a neutrally buoyant rigid sphere of radius a translates in a low-
Reynolds-number shear flow near a rigid planar wall, the sphere moves par-
allel to the wall. The kinematic reversibility of the flow (see e.g. Section 2.4.2),
coupled with the symmetry of the geometry, precludes any translation of the
sphere normal to the plane. The same conclusions hold for a liquid drop that
retains a spherical shape.
Low-Reynolds-Number Flows 71

Figure 2.15 Drift of a drop normal to a shear flow adjacent to a wall.

However, if the droplet deforms, then at small capillary numbers, the flow
changes the shape by an amount O (Ca ), as described in the previous section.
The ellipsoidal shape breaks the symmetry of the flow configuration, and
so motion normal to the plane can occur in a manner consistent with the
Stokes equations (Figure 2.15); in fact, the drop moves away from the plane.47
The next question we address is estimating the drift speed in the direction
normal to the plane.
Here we give an approximate argument for the speed of drift when the
droplet is at a distance h from the wall, with h  a. Since the droplet is
force-free, we learned in Section 2.7.7 that the typical far-field flow is that due
to a stresslet, which decays with distance as rK2 . In the presence of the wall,
the detailed velocity can be constructed using an appropriate image system,
which has a velocity with an order of magnitude ġa(a/h)2 . As we noted above,
a rigid sphere would maintain a constant separation distance from the wall
but would translate and rotate. The symmetry is broken by the drop defor-
mation, which we expect to be O (Ca ); see eqn (2.108). Thus, we anticipate a
drift speed normal to the plane of ġa(a/h)2 Ca . Consequently, the separation
distance h(t) changes approximately according to

dh  a 2
≈ c1 ġa Ca , (2.109)
dt h
where c1 is a constant. Drops further from the wall cross streamlines more
slowly.
72 Chapter 2

Upon integration, eqn (2.109) yields h3 ∝ t. This result for the drift speed
was measured experimentally by Smart and Leighton,47 and those authors
also reported faster drift speeds at higher capillary numbers, in a man-
ner qualitatively consistent with the capillary number dependence in eqn
(2.109). The same kinds of ideas apply to other deformable objects such as
vesicles and cells, and so the topic receives more discussion in Chapter 9, by
Vlahovska, and Chapter 10, by Abkarian and Viallat.

2.11 Inertial Effects and Limitations of the Stokes


Flow Description: The Oseen Approximation
2.11.1 General Remarks for Cases Where Forces Act on
Particles
The equations of Stokes flow should be expected to approximate the Navier–
Stokes equation when the Reynolds number is sufficiently small, i.e., for-
mally, Re  1. In practice, the results are often a good approximation even
for Re ≈ 1. For example, the Stokes drag formula for a rigid sphere is within
about 10% of the measured drag when Re Z 1.
In the case of steady flow, it might be expected that the neglected inertia
term in the Navier–Stokes equation, Re v · Vv (e.g. eqn (2.5) in dimensionless
terms, or see eqn (2.110) below), would be easy to add back, at least in a
perturbative sense. Unfortunately, this approach is not always so straightfor-
ward, as we now describe. Consequently, there are cases where the Stokes
equations do not represent a uniformly valid approximation to the Navier–
Stokes equation: the inability to solve the two-dimensional problem of uni-
form flow past a cylinder—Stokes’s paradox—is the classical example (see
e.g. Leal48 ). Here we just summarize a few ideas, but the subject is much
richer and more complex than the brief remarks below indicate.
To understand some ways in which inertia can influence the dynamics, we
consider steady flow past a particle. The Navier–Stokes equation (2.5), with
a characteristic pressure representative of a Stokes flow and neglecting the
body force, can be written in dimensionless form as
Re v · Vv Z KVp C V2 v. (2.110)
This equation is to be solved with v → KV̂ as r Z |r| → ∞ and with v Z 0 on
the particle surface Sp ; given that the discussion is in dimensionless terms,
here V̂ is a unit vector in the direction of translation of the particle.
Provided the Reynolds number is sufficiently small, we expect that the in-
ertia term can be neglected, in which case we have all of the results described
in previous sections. If this is true, then we can seek a regular perturbation
solution for the velocity distribution in the form
v (r, Re ) Z v0 (r) C Re v1 (r) C O Re 2 .

(2.111)
However, in the far field r  1, where the velocity approaches a uniform
flow, we know that the disturbance velocity decays as rK1 . Thus, we compare
Low-Reynolds-Number Flows 73

the inertia and viscous terms in eqn (2.110) and observe that
O (Re v · Vv) O (Re rK2 )
Z Z Re r for r  1. (2.112)
O (V2 v) O (rK3 )
This ratio is no longer small when r O Re K1  1. We are left with the distress-
ing conclusion that although the Reynolds number is small, inertial effects
can still be significant sufficiently far from the particle; this result stems from
the algebraic decay of the velocity field and the fact that the viscous term
decays faster than the inertial term.kk
Fortunately, in 3D flow problems involving applied forces, the force–
velocity relations that we obtained in earlier sections still hold to leading
order and inertial effects can be shown to produce a lower-order correction,
although the details are generally challenging to work out (e.g. Van Dyke49 ).
In particular, the velocity field is represented by eqn (2.111) close to the par-
ticle (in an “inner” region, r ! Re K1 ), but in an “outer” region a different
representation of the velocity field is needed. In such cases, the Stokes equa-
tions are not a uniformly valid approximation to the Navier–Stokes equation.
It is common to refer to the distance O (Re K1 ) as the Oseen distance.

2.11.2 Inertial Corrections to the Force on a Sphere


Working out the details of any particular flow problem is challenging and re-
quires the use of regular, and often also singular, perturbation methods. One
feature that we can highlight concerns the change in the force on a translat-
ing particle (in 3D) when small inertial effects are taken into account. Most
commonly, this question is discussed by direct calculation of the changes
in the velocity and pressure fields at O (Re ) and then the correction to the
force is calculated (e.g. Van Dyke49 ). One alternative way to obtain the force
more directly, which can work in at least some cases, is to use the Reciprocal
Theorem, eqn (2.30) (see e.g. Leal28 ).
To pose a specific question, we consider a steadily translating sphere as
observed in the laboratory reference frame. Then, in place of the problem
posed above as part of eqn (2.110), we have v Z V̂ on Sp as the boundary con-
dition, with v → 0 as r Z |r| → ∞, and the Navier–Stokes equation is written in
the form
Re v K V̂ · Vv Z KVp C V2 v.

(2.113)

kk It
is important to recognize that the failure of the Stokes flow approximation is typically only
a feature of flow problems involving forces. For example, if we consider the flow due to a
rotating sphere in a fluid at rest in the far field (see eqn (2.80)), then the flow now decays in
the far field in proportion to rK2 . Making the same argument as in eqn (2.112), we have

O (Re v · Vv) O (Re rK5 )


2
Z Z Re rK1 for r  1.
O (V v) O (rK4 )

This ratio is always small, so that for a rotating particle in a fluid otherwise at rest the Stokes
equations are a uniformly valid approximation to the Navier–Stokes equation.
74 Chapter 2

The term KV̂ · Vv is introduced to account for the local acceleration term in
the Navier–Stokes equation, and the left-hand side of eqn (2.113) acts as the
negative of an effective body force in the equation.
To apply the Reciprocal Theorem, we consider as an auxiliary problem
(v0 , σ 0 ) the corresponding known Stokes flow with v0 Z V0 on Sp , and with
v0 → 0 as r Z |r| → ∞. From eqn (2.113), we identify the “body force”
 term for
use in the Reciprocal Theorem, eqn (2.30), as f Z KRe v K V̂ · Vv, and so
we find that the force on the particle F (Re ), compared with the Stokes drag
F0 , is Z
  
F · V0 K F0 · V̂ Z Re v K V̂ · Vv · v0 dV. (2.114)
V
As written, this equation is an exact representation of the force on the parti-
cle at any Reynolds number. The formula can now be developed by using a
perturbation expansion for the velocity field. In this way, it turns out that
the integral can be shown to be O(1), so that we have the result that the
correction to the force is O (Re ). For example, for a translating sphere it
can be shown that for small but finite Reynolds numbers the hydrodynamic
force–velocity relation, in dimensional form, is
 
3
F (Re ) Z K6πmaV 1 C Re C O Re 2 ln Re .

(2.115)
8
The correction term O (Re 2 ln Re ) given here is small compared with the O (Re )
correction to the force, since
Re 2 ln Re
lim Z Re ln Re → 0. (2.116)
Re →0 Re
On the other hand, it is the appearance of logarithmic factors that is one of
the main complications in carrying out such analyses.

2.11.3 The Oseen Approximation


The above ideas emphasize the role of the convective term in the far field,
even when the Reynolds number is small. Consider the form of the steady
translation problem outlined in eqn (2.110) for flow past a sphere. In the
case of unbounded steady 3D flows, an approximation to the Navier–Stokes
equation that is valid at leading order everywhere, including the far field, is
KRe V̂ · Vv Z KVp C V2 v, (2.117)
which is known as the Oseen approximation. Discussion of the details of this
kind of problem would take us too far afield, however, so we refer the reader
to the specialized literature.48

2.12 Concluding Remarks


We have only touched on the many examples and application areas involv-
ing low-Reynolds-number flows. There are many other interesting and useful
fluid mechanics topics that we have not discussed, such as lubrication the-
ory, thin-film dynamics, coating flows, and rheology, simply because they
Low-Reynolds-Number Flows 75

were of less direct relevance to the topics discussed in later chapters of


this book. The references will lead the reader to the broader fluid dynamics
literature beyond what we have been able to touch on here.
This chapter has hinted at a variety of examples, many discussed in the
chapters that follow, of low-Reynolds-number flows in biological problems.
To mention just a few, we can list physiological flows, the movement and
swimming of bacteria, and the development of biofilms, as well as the mo-
tion of red blood cells and the rheology of blood and other cellular sus-
pensions. Of course, to really have an impact in these other fields, it is
important to understand the biological and medical questions so that one
can address the meaningful part of the flow and/or transport problem. Low-
Reynolds-number flows similarly impact on a wide range of problems in
colloid science, materials science, and engineering, especially those involv-
ing all manner of types of mixing of viscous materials, thin-film flows, and
flows in confined spaces; rheological questions where mixtures are used;
flows of colloidal suspensions, etc.; and the associated questions involving
heat and mass transfer. Although geophysics questions are generally on large
length scales, those that involve highly viscous fluids, such as flows in the
Earth’s mantle, are generally analyzed with the Stokes equations since the
corresponding Reynolds number is small; of course, flows in porous rocks
and other porous materials are generally studied using Darcy’s law. Here
we should also mention the subject of poroelasticity, which is a topic that
naturally links slow viscous flows and elastic deformations.
It should be clear from this list that one of the extra benefits that derives
from the study of low-Reynolds-number flows is the wide range of poten-
tial colleagues one can find in many other departments at a university or in
industry.

References
1. E. M. Purcell, Life at low Reynolds number, Am. J. Phys., 1977, 45, 3.
2. G. I. Taylor, Low Reynolds Number Flows, National Committee for Fluid
Mechanics Films, 1967.
3. J. Happel and H. Brenner, Low Reynolds Number Hydrodynamics, Marti-
nus Nijhoff Publishers, The Hague, 1983.
4. S. Kim and S. J. Karrila, Microhydrodynamics, Butterworth-Heinemann,
Boston, 2005.
5. E. Guazzelli and J. F. Morris, A Physical Introduction to Suspension Dynam-
ics, Cambridge University Press, Cambridge, 1st edn, 2012.
6. L. E. Becker, G. H. McKinley and H. A. Stone, Sedimentation of a
sphere near a plane wall: weak non-Newtonian and inertial effects, J.
Non-Newtonian Fluid Mech., 1996, 63, 201.
7. A.-K. Tornberg and M. J. Shelley, Simulating the dynamics and interac-
tions of flexible fibers in Stokes flows, J. Comput. Phys., 2004, 196, 8.
8. J. S. Wexler, P. H. Trinh, H. Berthet, N. Quennouz, O. Du Roure, H. E.
Huppert, A. Lindner and H. A. Stone, Bending of elastic fibres in viscous
flows: the influence of confinement, J. Fluid Mech., 2013, 720, 517.
76 Chapter 2

9. M. Abkarian, C. Lartigue and A. Viallat, Tanktreading and unbinding of


deformable vesicles in shear flow: determination of the lift force, Phys.
Rev. Lett., 2002, 88, 8103.
10. A. J. Goldman, R. G. Cox and H. Brenner, Slow viscous motion of a sphere
parallel to a plane wall – I. Motion through a quiescent fluid, Chem. Eng.
Sci., 1967, 22, 637.
11. N. Rott, Note on the history of the Reynolds number, Annu. Rev. Fluid
Mech., 1990, 22, 1.
12. H. A. Stone, CMOS Biotechnology, Springer Series on Integrated Circuits
and Systems, 2007, vol. 22, pp. 5–30.
13. J. R. Philip, Desperately seeking Darcy in Dijon, Soil Sci. Soc. Am. J., 1995,
59, 319.
14. R. A. Sampson, On Stokes’s current function, Philos. Trans. R. Soc., A,
1891, 182, 449.
15. H. Hasimoto, On the flow of a viscous fluid past a thin screen at small
Reynolds numbers, J. Phys. Soc. Jpn., 1958, 13, 633.
16. H. L. Weisberg, End correction for slow viscous flow through long tubes,
Phys. Fluids, 1962, 5, 1033.
17. Z. Dagan, S. Weinbaum and R. Pfeffer, An infinite-series solution for the
creeping motion through an orifice of finite length, J. Fluid Mech., 1982,
115, 505.
18. H. A. Stone, 2012 Les Houches Summer School on Soft Interfaces, Oxford
University Press, 2014.
19. E. Lauga and T. R. Powers, The hydrodynamics of swimming microor-
ganisms, Rep. Prog. Phys., 2009, 72, 096601.
20. M. V. Dyke, An Album of Fluid Motion, The Parabolic Press, Stanford, CA,
1982.
21. A. Groisman, M. Enzelberger and S. R. Quake, Microfluidic memory and
control devices, Science, 2003, 300, 955.
22. G. Segré and A. Silberberg, Radial particle displacements in Poiseuille
flow of suspensions, Nature, 1961, 189, 209.
23. L. E. Becker, S. A. Koehler and H. A. Stone, On self-propulsion of micro-
machines at low Reynolds number: Purcell’s three-link swimmer, J. Fluid
Mech., 2003, 490, 15.
24. J. M. Ottino, The Kinematics of Mixing, Cambridge University Press,
Cambridge, 1st edn, 1989.
25. H. A. Lorentz, A general theorem on the motion of a fluid with friction
and a few results derived from it, J. Eng. Math., 1996, 30, 19.
26. H. K. Kuiken and H. A. Lorentz, Sketches of his work on slow viscous and
some other areas in fluid mechanics and the background against which
it arose, J. Eng. Math., 1996, 30, 1.
27. H. A. Stone, Interfaces: in fluid mechanics and across disciplines, J. Fluid
Mech., 2010, 645, 1.
28. L. G. Leal, Particle motions in a viscous fluid, Annu. Rev. Fluid Mech.,
1980, 12, 435.
Low-Reynolds-Number Flows 77

29. C. Pozrikidis, Boundary Integral and Singularity Methods for Linearized


Viscous Flow, Cambridge University Press, Cambridge, 1st edn, 1992.
30. G. K. Batchelor, The stress system in a suspension of force-free particles,
J. Fluid Mech., 1970, 41, 545.
31. E. J. Hinch, Note on the symmetries of certain material tensors for a
particle in Stokes flow, J. Fluid Mech., 1972, 54, 423.
32. J. F. Brady and G. Bossis, Stokesian dynamics, Annu. Rev. Fluid Mech.,
1988, 20, 111.
33. K. Drescher, R. E. Goldstein, N. Michel, M. Polin and I. Tuval, Direct mea-
surement of the flow field around swimming microorganisms, Phys. Rev.
Lett., 2010, 105, 168101.
34. H. A. Stone and A. D. T. Samuel, Propulsion of microorganisms by
surface distortions, Phys. Rev. Lett., 1996, 77, 4102.
35. J. P. Tanzosh and H. A. Stone, A general approach for analyzing the ar-
bitrary motion of a circular disk in a Stokes flow, Chem. Eng. Commun.,
1996, 148–150, 333.
36. G. K. Batchelor, Slender-body theory for particles of arbitrary cross-
section in Stokes flow, J. Fluid Mech., 1970, 44, 419.
37. R. G. Cox, The motion of long slender bodies in a viscous fluids, Part 1.
General theory, J. Fluid Mech., 1970, 44, 791.
38. B. Rodenborn, C.-H. Chen, H. L. Swinney, B. Liu and H. P. Zhang, Propul-
sion of microorganisms by a helical flagellum, Proc. Natl. Acad. Sci. U. S.
A., 2013, 110, E338.
39. W. B. Russel, E. J. Hinch, L. G. Leal and G. Tieffenbruck, Rods falling
near a vertical wall, J. Fluid Mech., 1977, 83, 273.
40. H. Fu, T. Powers and R. Stocker, Separation of microscale chiral objects
by shear flow, Phys. Rev. Lett., 2009, 102, 158103.
41. G. B. Jeffery, The motion of ellipsoidal particles immersed in a viscous
fluid, Proc. R. Soc. A, 1923, 102, 161.
42. F. P. Bretherton, The motion of rigid particles in a shear flow at low
Reynolds number, J. Fluid Mech., 1962, 14, 284.
43. V. Sibillo, G. Pasquariello, M. Simeone, V. Cristini and S. Guido, Drop de-
formation in microconfined shear flow, Phys. Rev. Lett., 2006, 97, 054502.
44. G. I. Taylor, The formation of emulsions in definable fields of flow, Proc.
R. Soc. A, 1934, 146, 501.
45. J. M. Rallison, The deformation of small viscous drops and bubbles in
shear flows, Annu. Rev. Fluid Mech., 1984, 16, 45.
46. H. A. Stone, Dynamics of drop deformation and gup in viscous fluids,
Annu. Rev. Fluid Mech., 1994, 26, 65.
47. J. R. Smart and D. T. Leighton Jr., Measurement of the drift of a droplet
due to the presence of a plane, Phys. Fluids A, 1991, 3, 21.
48. L. G. Leal, Advanced Transport Phenomena, Cambridge University Press,
Cambridge, 2010.
49. M. Van Dyke, Perturbation Methods in Fluid Mechanics, The Parabolic
Press, Stanford, CA, 1975.

S-ar putea să vă placă și