BPFD090623

T H E S C I E N T I S T D O E S N O T S T U D Y N AT U R E B E C A U S E I T I S U S E F U L ;
H E S T U D I E S I T B E C A U S E H E D E L I G H T S I N I T, A N D H E D E L I G H T S
I N I T B E C A U S E I T I S B E A U T I F U L . I F N AT U R E W E R E N O T B E A U T I F U L ,
I T W O U L D N O T B E W O R T H K N O W I N G , A N D I F N AT U R E W E R E N O T
WORTH KNOWING, LIFE WOULD NOT BE WORTH LIVING.
H E N R I P O I N C A R É , S C I E N C E A N D M E T H O D, 1 9 0 8
R AY M O N D E . G O L D S T E I N A N D E R I C L A U G A
BIOLOGICAL PHYSICS
A N D F L U I D DY N A M I C S
( A G R A D U AT E C O U R S E I N 2 4 L E C T U R E S )
D E PA R T M E N T O F A P P L I E D M AT H E M AT I C S A N D
THEORETICAL PHYSICS, UNIVERSITY OF CAMBRIDGE
Copyright © 2023 Raymond E. Goldstein and Eric Lauga
published by department of applied mathematics and

theoretical physics, university of cambridge
First printing, June 2023

Contents
Preface 17
1 Overview (Lecture 1) 19
1.1 Length scales and methodologies 19
1.2 Scaling arguments 20
2 Microscopic Physics (Lectures 2-4) 25

2.1 Review of molecular physics 25
2.2 Van der Waals interactions 28
2.2.1 The pairwise intermolecular potential 28
2.2.2 Interaction of extended objects 30
2.3 Screened electrostatic interactions 31
2.3.1 Poisson-Boltzmann and Debye-Hückel theory 31
2.3.2 Electrostatic interactions between surfaces 34
2.4 Geometrical aspects of screened electrostatics 37
2.4.1 Comparative approach 38
2.4.2 Perturbative approach 39
3 Fluctuations & Brownian Motion (Lectures 5-7) 41

3.1 Equipartition theorem 41
3.2 Fluctuating continuous objects 42
3.2.1 The tense string 42
3.2.2 Continuum limit 43
3.3 Brownian Motion and Diffusion 46
3.4 Polymers and Entropic Forces 48
3.4.1 Freely-hinged chain 49
3.4.2 The freely-jointed chain 50
6
3.5 Polymer statistics and random walks 50

3.5.1 General formulation 50
3.5.2 Flory theory and self-avoidance 52
4 Biological Filaments (Lectures 8-12) 55

4.1 Essentials of differential geometry of curves 55
4.2 Overdamped filament motion 56
4.2.1 A variational principle 57
4.2.2 Elasticity 59
4.2.3 Boundary conditions and biharmonic eigenfunctions 62
4.3 Instabilities and elastohydrodynamics 65
4.3.1 Example 1. Euler Buckling 65
4.3.2 Example 2: Wiggling elastica 68
4.3.3 Example 3: Stretch-coil transition under compressional flows 74
4.3.4 Example 4: Motor-induced oscillations 75
5 Surfaces & Membranes (Lectures 13-16) 81

5.1 Differential geometry of surfaces 81
5.1.1 Surfaces of Revolution 82
5.1.2 Catenoids 82
5.2 Membrane Fluctuations 84
5.2.1 Thermal fluctuations of spherical vesicles 84
5.2.2 Helfrich repulsion 86
5.2.3 Flicker phenomena of red blood cell membranes 88
5.3 Shapes and shape transformations 93
5.3.1 Buckling of two-dimensional vesicles under pressure 93
5.3.2 Tether formation 96
6 Cellular Motion (Lectures 17-20) 103

6.1 Cell swimming 103
6.1.1 The general swimming problem 104
6.1.2 Rate independence and scallop theorem 105
7
6.2 Flagella and the physics of low-Re propulsion 106

6.2.1 Taylor’s model a single waving sheet 106
6.2.2 Taylor’s model; two coupled sheets 109
6.2.3 Resistive Force Theory and drag-based thrust 111
6.2.4 Propulsion by a traveling wave of deformation 112
6.2.5 Bacterial locomotion 114
6.3 Models of synchronisation 116
6.3.1 Bead-spring model 117
6.3.2 The Adler equation 119
6.3.3 The stochastic Adler equation 121
6.4 Swimming Cells 122
6.4.1 Stokeslets 122
6.4.2 Stresslets and Rotlets 123
6.5 Microswimmers in External Fields and Flows 126
6.5.1 Bottom-Heavy Swimmers 127
6.5.2 Swimmers in flows 127
6.5.3 Gyrotaxis in Poiseuille flow 131
6.5.4 Jeffery Orbits in a flow with constant shear rate 133
6.6 Surface-Mediated Interactions 134
6.6.1 Singularities Near Surfaces 134
6.6.2 Attraction and reorientation of stresslets by surfaces 137
6.6.3 Surface-mediated attraction of stokeslets 138
7 Kinetics & Pattern Formation (Lectures 21-24) 141

7.1 Michaelis-Menten kinetics 141
7.1.1 Steady-state approximation 143
7.1.2 Matched asymptotic approximation 144
7.1.3 Cooperativity in Reaction Rates 146
7.1.4 Slaving 147
7.2 Excitable media: Neurons and the FHN model 148
7.3 Front Propagation 153
7.4 Reaction-Diffusion Systems 157
7.4.1 Phenomenology 157
7.4.2 Front Propagation II 161
7.4.3 The Turing Instability 165
8
7.5 Chemotaxis and Instabilities 168

7.5.1 Background 168
7.5.2 Phenomenology of Dictyostelium discoideum 168
7.5.3 The Keller-Segel Model 169
8 Bibliography 173
9 Index 183
List of Figures
1.1 Range of length scales and living systems of interest. 19

1.2 The green alga Volvox. 22
1.3 The aquatic plant Chara corallina. 23
2.1 Temperature dependence of the second virial coefficient. 25

2.2 Components of the intermolecular potential. 27
2.3 Two charged harmonic oscillators. 28
2.4 Stacked thylakoid membranes. 30
2.5 Geometry of atom-slab interaction. 31
2.6 Geometry of slab-slab interaction. 31
2.7 Structure of the phospholipid DPPC. 35
2.8 Structure of a lipid membrane in water. 36
2.9 Membrane interaction energy in DLVO theory. 37
2.10 Electron micrograph of red blood cells. 38
2.11 A ripply surface held at fixed potential. 39
3.1 A string under tension. 42

3.2 Scaling function for variance of fluctuating string. 43
3.3 The freely-hinged chain. 49
3.4 Force-extension curves for freely-hinged and freely-jointed chains. 50
3.5 The freely-jointed chain. 50
4.1 Geometry of a curve in the plane. 55

4.2 Geometry of tangent angle, tangent and normal vectors. 55
4.3 Bending of a filament. 59
4.4 Graphical solution of the transcendental equation. 63
4.5 The first five biharmonic eigenfunctions. 63
4.6 Microfluidic setup to study single actin filaments. 64
4.7 Variance of fluctuating actin filaments. 65
4.8 Buckling of microtubules inside a lipid vesicle. 65
4.9 Geometry of Euler buckling. 65
4.10 Pitchfork bifurcation of Euler buckling. 67
4.11 Filament shapes in Euler buckling. 67
4.12 Numerical and analytical results for Euler buckling. 67
4.13 Stokes and elastohydrodynamic problems. 68
4.14 Solution of EHDII for an infinite filament. 70
4.15 Waveforms of spermatozoa in a viscous fluid. 70
4.16 Solution of EHDII for finite filaments. 71
4.17 Scaling functions for EHDII. 73
10
4.18 The first three critical modes for the stretch-coil instability. 75
4.19 Experimental results on shear-induced buckling. 76
4.20 Setup for the follower force problem. 76
4.21 Components of complex eigenvalue for follower-force problem. 78
4.22 The leading eigenfunction of the follower-force problem. 78
4.23 Separable solution of follower-force problem. 79
5.1 A soap film in the form of a catenoid. 83

5.2 Graphical solution of transcendental equation for catenoids. 83
5.3 Shapes of catenoids in the two branches of solutions. 84
5.4 Areas of the two branches of catenoid solutions. 84
5.5 A perturbed cylinder. 84
5.6 A perturbed circle. 86
5.7 A fluctuating elastic membrane confined between two walls. 87
5.8 Electron micrograph of red blood cells. 88
5.9 Geometry of the calculation for flicker phenomenon. 88
5.10 Key experimental observations on the flicker phenomenon. 90
5.11 Scaling function G for the flicker phenomenon. 93
5.12 Shapes of two-dimensional vesicles under pressure. 96
5.13 Membrane tether pulled from a bilayer lipid vesicle. 96
5.14 Trajectories of Eq. (5.94) in the D − α plane. 97
5.15 Catenoids of maximum height for a range of β. 97
5.16 Shapes of poked membranes in the Monge representation. 99
5.17 Numerical solution of full tether problem. 100
6.1 Examples of microswimmers. 103

6.2 A generic swimmer whose shape is actively deformed. 104
6.3 Two different paths in shape space. 105
6.4 Swimming sequences and the scallop theorem. 106
6.5 Ciliary motions in mussels and green algae. 106
6.6 Geometry of Taylor’s waving sheet model. 107
6.7 Two nearby waving sheets in Taylor’s calculation. 109
6.8 Dissipation ratio in Taylor’s calculation. 110
6.9 Origin of drag anisotropy of slender bodies. 111
6.10 Forces due to prokaryotic and eukaryotic flagella. 112
6.11 Normal and tangential forces on a moving filament segment. 112
6.12 A tilted rod segment moving upwards. 112
6.13 Geometry of Lighthill’s calculation of propulsion. 113
6.14 Kinematics of bacterial swimming. 114
6.15 Dynamics of a rotating helix. 114
6.16 Swimming dynamics of a model bacterium. 116
6.17 Two coupled bead-spring oscillators in the NEL model. 117
6.18 Graphical representation of the Adler equation. 119
6.19 Mechanism of synchronisation in the NEL model. 120
6.20 Effective potential in the Adler model. 120
6.21 Flagellar synchronisation in Chlamydomonas. 121
6.22 Streamlines of a stokeslet. 122
6.23 Flow field around Volvox carteri. 123
6.24 Pushers and pullers. 124
11
6.25 A force dipole. 124

6.26 Pusher and puller stresslet flow fields. 125
6.27 Flow field around E. coli. 126
6.28 Flow field around C. reinhardtii. 126
6.29 Geometry of a bottom-heavy swimmer. 127
6.30 Gyrotaxis of a bottom-heavy organism in channel flow. 132
6.31 Dynamics of the orientation angle in Jeffery orbits. 133
6.32 Jeffery orbits in the laboratory frame. 134
6.33 Stokes singularities near a free surface. 134
6.34 Stokeslet flow fields near stress-free and no-slip surfaces. 135
6.35 Stokeslet and stresslet flows near a no-slip surface. 137
6.36 Flow field around the bacterium E. coli near a surface. 137
6.37 Two spherical microswimmers near a no-slip surface. 139
6.38 Effective potential for bound states. 139
6.39 Infalling dynamics of two Volvox colonies. 139
7.1 Solutions to the Michaelis-Menten dynamics. 143

7.2 Michaelis-Menten law. 143
7.3 The Lineweaver-Burk plot. 144
7.4 Matched asymptotic solutions to MM dynamics. 145
7.5 Schematic sigmoidal binding curve of oxygen to hemoglobin. 146
7.6 Schematic of an axon. 148
7.7 Schematic of an action potential. 149
7.8 Nullclines of the FHN model. 151
7.9 Dynamics of the FHN model I. 152
7.10 Dynamics of the FHN model II. 152
7.11 Dynamics of the FHN model III. 152
7.12 Examples of front propagation. 153
7.13 A potential F (u) with two competing, locally stable states. 154
7.14 The function f (u). 154
7.15 Growth rate for the stable state ū = 0. 155
7.16 Growth rate for the state ū = r. 155
7.17 Stationary front shape for r = 1/2. 156
7.18 Generic form of F (u). 156
7.19 Effective potential in the stable-to-stable problem. 156
7.20 Mechanical analogy in the unstable-to-stable case. 156
7.21 Growth rate curves for reaction-diffusion system. 157
7.22 Growth curves for a pattern-forming system. 158
7.23 Stability analysis of FHN model in the fast-inhibitor limit. 160
7.24 Dynamics of two nearby fronts in the FHN model. 164
7.25 Labyrinthine pattern formation in the FHN model. 164
7.26 Behaviour of the determinant of the stability matrix. 166
7.27 Experimental observation of the Turing instability. 167
7.28 Patterns in the Belousov-Zhabotinski reaction. 168
7.29 Life cycle of Dictyostelium discoideum. 169
7.30 Spiral and target patterns in Dictyostelium discoideum. 169
7.31 The streaming instability in Dictyostelium discoideum. 169
7.32 Growth rate versus k in the Keller-Segel model. 172
List of Tables
1.1 Some physical quantities used in these notes. 20
3.1 Scaling exponent for self-avoiding walks in d dimensions. 53
4.1 The persistence lengths of various biological filaments. 59
6.1 Reynolds numbers for various swimming organisms. 104

15
Dedicated to John O. Kessler (1928-2022)
and Howard C. Berg (1934-2021), whose
enthusiasm for biological physics and deep
physical insight were a profound inspiration.

Preface
Physics and Biology have been intertwined since the dawn of mod-
ern science, from Antony van Leeuwenhoek’s use in 1700 of ad-
vanced optics to reveal the hidden world of microscopic life,1 to 1
van Leeuwenhoek, A. IV. Part of a
Bonaventura Corti’s 1774 discovery of the persistent fluid motion letter from Mr. Antony van Leeuwen-
hoek, concerning the worms in sheeps
inside large eukaryotic cells,2 Robert Brown’s 1828 study of random livers, gnats, and animalcula in the ex-
motion at the microscale3 and Theodor Engelmann’s determination crements of frogs. Philos. Trans. R. Soc.,
22:509–18, 1700
in 1882 of the wavelength dependence of photosynthetic activity.4 2
B. Corti. Osservazioni Microscopische
Although at the time it might have been difficult to define the precise sulla Tremella e sulla Circolazione del Flu-
disciplines of each of these scientists—perhaps they were all ‘natural ido in Una Pianta Acquajuola. Appresso
Giuseppe Rocchi, Lucca, Italy, 1774
philosophers’—in hindsight we can see clearly the way in which their 3
Brown, R. XXVII. A brief account of
discoveries impacted both biology and physics. Despite this long his- microscopical observations made in the
tory of discoveries at the boundary between the two fields, and the months of June, July, and August, 1827,
on the particles contained in the pollen
innumerable fundamental contributions to both disciplines over the of plants; and on the general existence
long arc of time, the field of biological physics as a discipline within of active molecules in organic and in-
the research enterprise of physics has only risen to great prominence organic bodies. Philos. Mag., 4:161–73,
1828
since the postwar era, particularly since the mid 1980s. We are now 4
Engelmann, T.W. Ueber Sauerstof-
at the point that most academic physics departments have an iden- fausscheidung van Pfalnzenzellen im
Mikrospektrum. Pflüger, Arch., 27:
tifiable group in biological physics alongside those in high energy, 485–89, 1882
condensed matter, atomic and astrophysics.
As a subject, biological physics is in a much earlier stage of devel-
opment than the traditional ones found in physics or applied mathe-
matics curricula. This makes it an exciting field in which to work, but
also poses challenges for students and faculty, as there is no univer-
sally accepted canon to follow. Instead, each lecturer has had to cob-
ble together concepts and presentations from a number of disparate
sources to arrive at a syllabus. These lecture notes have been de-
veloped very much in that spirit. They derive from several master’s
level (Part III) courses taught by each of us, sometimes in collabora-
tion with Prof. Ulrich Keyser of the Cavendish Laboratory. We have
attempted to strike a balance between coverage, depth, and pedagogy
in order to address —in the 24 lectures of a Cambridge term—topics
in biological physics that cover length scales from the molecular to
the terrestrial. We presume no prior exposure to the subject, but
assume familiarity with the basic ideas of statistical physics, elec-
tromagnetism, nonlinear dynamics, and fluid mechanics. As this is
primarily a course for Part III students in the mathematical Tripos,
the presentation often includes digressions into the applied mathe-
matical aspects of the problems at hand.
We have organised the topics covered by these notes roughly in or-
18
der of increasing length scales, from the basic molecular forces that
govern the cohesion and interaction of molecules and supramolecu-
lar assembles, the elastic properties of filaments and membranes, the
role of thermal fluctuations in their conformation and interactions,
on to cellular motion, chemical kinetics and pattern formation. Our
approach is to give a sampling of the phenomena and their analy-
sis at each of the length scales of interest. Even with the goal of
brevity in mind, it is likely that covering all of these notes in 24 lec-
tures is a challenge. For further readings, we refer the student to
a number of excellent textbooks that have appeared in recent years
covering various aspects of biological physics5 and physical biology,6 5
P. Nelson. Biological Physics: Energy,
as well as more specialized works on elasticity.7 Much motivation for Information, Life. Chiliagon Science,
Philadelphia, PA, student edition, 2020
the physicist’s approach to biology comes from D’Arcy Wentworth 6
R. Phillips, J. Kondev, J. Theriot, H.G.
Thompson’s classic book,8 and we recommend highly the classic text Garcia, and N. Orme. Physical Biology of
on applied mathematics by Lin and Segel.9 Where possible, we have the Cell. Garland Science, Boca Raton,
FL, 2nd edition, 1998
included references to the original, often old literature, in the belief 7
B. Audoly and Y. Pomeau. Elasticity
that reading such scientific works is highly worthwhile.10 and Geometry: From hair curls to the non-
Associated with these notes is a set of Example Sheets. In many linear response of shells. Oxford Univer-
sity Press, Oxford, UK, 2010
places throughout the text we have deliberately left out certain cal- 8
D.W. Thompson. On Growth and Form.
culational details to encourage the reader to fill them in. At the Dover Publications, Mineola, N.Y., the
same time, we have taken pains to explain in a fair amount of detail complete revised edition, 1992
9
C.C. Lin and L.A. Segel. Mathematics
the ins and outs of various key calculations. While we emphasize Applied to Deterministic Problems in the
analytical calculations throughout the text, there are many contexts Natural Sciences. Macmillan Publishing
in which numerical work is an important adjunct to those calcula- Co., Inc., New York, 1974
10
Goldstein, R.E. Coffee stains, cell re-
tions, whether for something as simple as understanding the detailed ceptors, and time crystals: Lessons from
shape of a function, finding the roots of a transcendental equation, the old literature. Phys. Today, 71:32–38,
or to solve a nonlinear PDE to understand its global behaviour. To 2018
that end we have provided a series of simple Matlab programs for

these various tasks, named according to the figure(s) produced by
them. In addition, we have organized the references so that clicking
on the author names links to a collection of pdfs of all of the refer-
ences (apart from books), while clicking on the article title takes one
to the original journal article.
These lecture notes are meant to be a living document, easily
changed in response to comments and corrections suggested by stu-
dents and colleagues who use them, and also able to expand over
time to cover more topics in biological physics. We can already see
the possibility of chapters on population biology, systems biology,
taxes (chemotaxis, phototaxis), a more in-depth treatment of elastic-
ity, and computation.
We are grateful to George T Fortune for a careful proofreading
of this entire set of notes, and to Thomas Powers, xxx, and yyy for
detailed comments on drafts. Any remaining errors or omissions are
of course our own, and we look forward to hearing from users of
these notes of any further corrections or suggested modifications.
R E Goldstein and E Lauga, July 2023

1
Overview (Lecture 1)
1.1 Length scales and methodologies
The field of biological physics covers phenomena spanning an enor-

mous range of length scales. Historically, there was great emphasis
placed on the cellular scale of microns,1 but as the field broadens 1
Berg, H.C. and Purcell, E.M. Physics of
to include problems ranging from neuroscience to development and chemoreception. Biophys. J., 20:193–219,
1977
even ecology, that range now extends to kilometers (Fig. 1.1).
Thermal ﬂuctuations Other forms of noise

10-9 m 10-6 m 10-3 m 1m 103 m
nm µm mm m km
hydrogen atom
proteins cilia hair ﬁsh

coiled DNA algae ecology
bacteria ciliates stretched DNA
lipid vesicles
neuroscience
pa�ern formation
spermatozoa
Figure 1.1: The range of length scales
considered in these notes and a sam-
In this course we will study methods to quantify the physics of
pling of the living systems of interest.
living systems on many of these scales. While many of the underly-
ing properties of the components of cells, tissues, and organisms can
be described by molecular physics and its continuum generalisations,
life is inherently out of equilibrium and characterized by "noise" from
many sources. At the smallest scales—from nanometers up to hun-
dreds of microns—equilibrium thermal fluctuations may be domi-
nant, but even there, and certainly at larger scales, descriptions must
incorporate such features as biochemical fluctuations within single
organisms and organism-to-organism phenotypic variability. Thus,
the mathematical methods utilized here include not only equilibrium
statistical physics, but also deterministic and stochastic ordinary and
partial differential equations of various types. In addition, many of
the important questions in biological physics involve understanding
the time-dependent geometry of living matter and its constituents.
In order to make these notes self-contained we have included where
appropriate some background material on differential geometry and
geometric evolution, and provided references to more complete treat-
ments in the literature.
20 biological physics and fluid dynamics
One of the key lessons of applied mathematics is the importance of

proper rescaling of the governing equations to reveal the relevant (di-
mensionless) control parameters that determine the behaviour. Not
only is this good practice in solving problems, it also promotes a
point of view that focuses on the size of the parameter space and the
scaling laws that underlie a given phenomenon. It serves to make
clear the conclusions of models in interdisciplinary collaborations.2 2
Goldstein, R.E. Are theoretical ‘re-
A corollary to the above is that, in biological physics, numbers sults’ results? eLife, 7:e40018, 2018
matter. That is to say, gaining an intuitive feel for the typical sizes of
things, energy scales, and time scales is very important in building
a quantitative picture for a given problem. But what system of units
should we use? Many physicists and applied mathematicians have
been educated to use the MKS system, but its use can lead to very
awkward numbers; take for example the bending modulus k c of a
cell membrane, whose typical value might be 4 × 10−20 N·m, thus
having the units of energy. We shall see that the typical forces on the
cellular scale are picoNewtons (pN), and with typical sizes of molec-
ular constituents being nanometres (nm), the scale pN·nm is more
appropriate, for then k c ∼ 40 pN· nm. And, most importantly, with
the Boltzmann constant k B = 1.38 × 10−23 N·m/K and the absolute
temperature T = 300K, thermal energy kB T is ∼ 4 × 10−21 N·m or
k B T ≃ 4 pN · nm. (1.1)
Thus, concisely, the bending modulus is k c ∼ 10 k B T.
Table 1.1 collects the first set of physical quantities that will appear
in these notes. Other tables will follow as we develop the subject.
quantity symbol/definition value alternate units

viscosity of water µ 0.01 g/cm·s (1 cP) 0.001 kg/m·s
density of water ρ 1 g/cm3 103 kg/m3
kinematic viscosity of water ν 0.01 cm2 /s 106 µm2 /s
dielectric constant of water ϵ 80
acceleration of gravity g 10 m/s2 107 µm/s2
size of a bacterium major axis (1 − 3) × 10−4 cm 1 − 3 µm
size of unicellular algae radius 5 × 10−4 cm 5 µm
size of multicellular algae radius (100 − 400) × 10−4 cm 100 − 400 µm
charge of the proton e 1.6 × 10−19 C
membrane thickness δ (3 − 5) × 10−9 m 3-5 nm
membrane elastic modulus kc 4 × 10−20 N·m 10 kB T
Table 1.1: Some physical quantities

used in these notes.
1.2 Scaling arguments
We now turn to several examples that illustrate the power of scal- 3

Maxwell, J.C. Remarks on the mathe-
matical classification of physical quan-
ing arguments (sometimes called dimensional analysis, as usually tities. Proc. London Math. Soc., s1-3:224,
attributed to Rayleigh3 ) to reveal the essential physics in a problem, 1869
and to point out several particular features of the cellular scale that
will become our primary focus.
Although quantum phenomena will make only a cursory appear-
ance in these notes (when we discuss molecular interactions in Chap-
ter 2), the hydrogen atom provides our first example of scaling ar-
guments. Letting m be the reduced mass of the two-body system
overview (lecture 1) 21
(proton + electron), the kinetic energy is p2 /2m, where p is the mo-

mentum, while the potential energy (in the Gaussian system of units)
is −e2 /r, where r is the separation between electron and proton. If
we invoke the de Broglie relation p = h/λ between the momentum p
and the associated particle wavelength λ, and the quantization con-
dition λ = 2πr/n for some integer n, then p = nh̄/r, there is a single
length scale r that governs the physics, and the total energy is
n2 h̄2 e2
E (r ) ∼ − . (1.2)
2mr2 r
The function E(r ) diverges to +∞ as r → 0 and approaches 0 from
below as r → ∞, having thus a single minimum. We find r by min-
imising E(r ), yielding the minimiser
n2 h̄2 me4
rn∗ = and En (rn∗ ) = − . (1.3)
me2 2n2 h̄2
Here, rn∗ is n2 r B , where r B is the Bohr radius, and En = Ry/n2 , where

Ry is the Rydberg constant. These are the exact values for the hy-
drogen atom. Note that this approach required none of the intricate
analysis of Laguerre polynomials or any other aspects of the full
solution to the Schrodinger equation,4 and even obtains the correct 4
L.D. Landau and E.M. Lifshitz. Quan-
factor of 2 in the Rydberg constant. tum Mechanics. Non-relativistic theory.
Pergamon Press, Oxford, UK, 2nd edi-
A second approach to Eq. 1.2, more in keeping with the ethos tion, 1965
of Applied Mathematics,5 is to balance the two terms on the r.h.s., 5
C.C. Lin and L.A. Segel. Mathematics
so n2 h̄2 /mr2 ∼ e2 /r (ignoring the factor of 2), which implies the Applied to Deterministic Problems in the
Natural Sciences. Macmillan Publishing
existence of a length rn ∼ n2 h̄2 /me2 . If we define a dimensionless Co., Inc., New York, 1974
radius R and dimensionless energy E as
r E
R= and E = , (1.4)
rn me4 /2n2 h̄2
then we obtain the marvellously simple result
1 2
E= 2
− , (1.5)
R R
which is minimised at R = 1, giving E = −1, as above. The ab-
sence of any explicit physical parameters from (1.5) indicates that a
"natural" system of units has been identified.
Let us now turn to an important biophysical question concerning
the role of inertia in the motion of microscopic organisms such as
bacteria.6 Suppose we have a spherical body of radius a and density 6
Purcell, E.M. Life at low Reynolds
ρ equal to that of the surrounding fluid, whose viscosity is µ. Let x (t) number. Am. J. Phys., 45:1–11, 1977
be its position during one-dimensional motion. Then the equation of

motion is
4 3 d2 x dx
πa ρ 2 = −6πµa , (1.6)
3 dt dt
where we have invoked the Stokes drag force 6πµav on a particle
moving with speed v = dx/dt. We wish to answer the question:
if the particle initially is moving with speed v0 , how long will it
take to come to rest under the action of drag, and how far will it
travel during that time? Without solving this explicitly we can sim-
ply assume there is a characteristic time τ and displacement ℓ (the
"coasting length") and balance terms, setting
4 3 ℓ ℓ
πa ρ 2 ∼ 6πµa (1.7)
3 τ τ
to obtain
2 a2
τ= and ℓ = v0 τ, (1.8)
9 ν
where ν = µ/ρ is the kinematic viscosity, with units of length2 /time.
From the structure of Eq. 1.6, we expect exponential decay of the daughter colonies
motion, with decay time τ.

To get a feel for these quantities, let us consider a bacterium with
a ∼ 2 µm; from Table 1.1 we then have τ ∼ 10−6 s: one microsecond!
If the initial speed is, say, v0 = 10 µm/s, the distance traveled is
ℓ ∼ 10−5 µm = 10−2 nm, far less than the size of the bacterium, and
indeed just a small fraction of the Bohr radius (0.05 nm)! We conclude
that inertia is completely irrelevant to the motion of micron-sized
ECM
objects in water. In this simple picture, inertia would only begin to
matter when the coasting length is comparable to the particle radius. somatic cells 250 μm
Setting these equal leads to the critical radius Figure 1.2: The green alga Volvox. It
consists of ∼ 103 biflagellated somatic
9ν
ac ∼ . (1.9) cells embedded in a transparent ex-
2v0 tracellular matrix (ECM), with a small
number of daughter colonies that grow
From this we see that even when the speed is ∼ 500 µm/s, which is up inside the spheroid.
among the highest speeds known for large multicellular algae such
as Volvox carteri shown in Fig. 1.2,7 the critical radius would still be 7
Goldstein, R.E. Green algae as model
∼ 1 cm, an order of magnitude larger than those organisms. Thus, organisms for biological fluid dynam-
ics. Annu. Rev. Fluid Mech., 47:343–375,
there is a vast range of organisms that swim in the low Reynolds 2015
number world. This subject will be covered in detail in Chapter 6.
We are now in a position to understand the relationship between
thermal energy and diffusion, as first done by Einstein8 and Suther- 8
Einstein, A. Investigations on the the-
land.9 The basic idea is to consider a suspension of microscopic ory of the Brownian movement. Ann. d.
Phys., 17:549–560, 1905
particles at concentration c( x ), with diffusion constant D, acted on 9
Sutherland, W. LXXV. A dynamical
by a force F ( x ) = −dϕ/dx, where ϕ is the potential energy. The flux theory of diffusion for non-electrolytes
J of particles is the sum of a Fick’s law contribution and an advective and the molecular mass of albumin.
Phil. Mag., 9:781–785, 1905
term with some speed u,
dc
J = −D + uc. (1.10)
dx
In the viscous regime, the speed u is determined by the force balance
ζu = −dϕ/dx, where ζ = 6πµa for spherical particles. To determine
D by the principles of equilibrium statistical physics we must indeed
be in equilibrium, with J = 0. Integrating (1.10), we obtain
c ∼ exp(−ϕ/Dζ ). (1.11)
If this is to be consistent with the Boltzmann factor of statistical

physics, then Dζ = k B T or
kB T 0.2 µm2
D= ≃ . (1.12)
6πµa ã s
overview (lecture 1) 23
The second relation holds for water at room temperature, where ã is

the radius in microns. This immediately reveals that a small molecule
on the order of 2Å (ã ∼ 2 × 10−4 ) has D ∼ 103 µm2 /s, whereas a 2 µm
microsphere of the kind used to trace flows has D ∼ 0.1 µm2 /s.
From the diffusion equation
∂c
= D ∇2 c, (1.13)
∂t
we see by dimensional analysis that lengths ℓ and time t are related
by Dt ∼ ℓ2 or daughter colonies
ℓ2
t∼ . (1.14)
D
From the values of D derived above, we see that the time it takes
a small molecule to diffuse across a bacterium (ℓ ∼ 1 µm) is 10−3 s,
while that to cross a colony of Volvox (ℓ ∼ 300 µm) is about 103 s ∼
16 minutes. On the other hand, consider the aquatic plant Chara coral-
ECM
lina (Fig. 1.3). Its cells are some of the largest known to science;
they can reach ℓ ∼ 10 cm = 105 µm, giving a diffusive timescale of somatic cells 250 μm
∼ 107 s. Remembering that there are about 10π megaseconds in a

year (3.15 × 107 ), this time is about 4 months!
Clearly, diffusion is not effective for large eukaryotic cells, and Figure 1.3: The aquatic plant Chara
corallina.
some active transport mechanism is needed. This point was em-
phasized eloquently by J.B.S. Haldane many years ago in a famous
essay.10 One such mechanism involves cargo transport by "motor 10
Haldane, J.B.S. On Being the Right
proteins" moving along filaments in the cell, or indeed by the fluid Size, 1927
entrained by their motion, a phenomenon termed cytoplasmic stream-

ing that was first discovered in plants such as Chara as long ago as
1774.11 At larger scales there are ever more complex vascular sys- 11
B. Corti. Osservazioni Microscopische
tems driving by muscular pumps. sulla Tremella e sulla Circolazione del Flu-
ido in Una Pianta Acquajuola. Appresso
As a final exercise in the use of dimensional analysis we ask about Giuseppe Rocchi, Lucca, Italy, 1774
the typical scale of voltages to be expected inside living systems.
We have in mind a situation like that in neurons, in which there
are differences in ionic concentrations across membranes. Since the
chemical potential for a dilute solution of ions at concentration c is
k B T ln c, the difference in chemical potential between the inside and
outside of a membrane must be k B T ln(cin /cout ). Setting this equal to
the electrostatic energy eV for an elementary charge to move through
the voltage difference across the membrane, we find

k T cin
V ∼ B ln . (1.15) 12
Hodgkin, A.L. and Huxley, A.F. A
e cout quantitative description of membrane
current and its application to conduc-
From Table 1.1 we obtain k B T/e ∼ 25 mV, which is indeed the basic tance and excitation in nerve. J. Physiol.,
scale of voltages across neuronal membranes,12 as we shall discuss 117:500–544, 1952
in Chapter 7.
2
Microscopic Physics (Lectures 2-4)
2.1 Review of molecular physics
At the smallest scales of interest in these lectures we are concerned

with the elementary interactions between molecules, and how they
determine interactions between more extended objects such as mem-
branes. To that end, we review the basics features of van der Walls in-
teractions between atoms, and of screened electrostatics, along with
some basic principles in non-ideal gas theory that serve as the basis
of molecular solutions, including those that can separate into distinct
phases. Although it might seem that a discussion of matters related
to equilibrium phase transitions is somewhat off-topic in a course
on biological physics, it is important to note that a rapidly growing 1
Hyman, A.A., Weber, C.A. and
Jülicher, F. Liquid-liquid phase sepa-
body of recent work has shown that the phenomenon of liquid-liquid ration in biology. Annu. Rev. Cell Dev.
phase separation plays a crucial role inside many cells.1 Biol., 30:39–58, 2014
The ideal gas law,
p
pV = Nk B T or =ρ, (2.1)
kB T
relates the pressure p, volume V, and number of molecules N, where
k B is the Boltzmann constant, T is the absolute temperature and ρ =
N/V is the density. It is, of course, only an approximation. At low
densities real gases are described by a "virial expansion", B2(T)
p
≃ ρ + B2 ( T )ρ2 + B3 ( T )ρ3 + · · · . (2.2)
kB T
where the Bi are termed "virial coefficients". Intuitively, the quadratic TB T
term in ρ involves 2-body interactions, the cubic term captures 3-
body effects, and so on. Experimental measurements (Fig. 2.1) of the
second virial coefficient B2 show that it is negative at low tempera-
tures (indicating attraction between pairs) and becomes positive (re-
Figure 2.1: Temperature dependence of
pulsive) at high temperatures, crossing through zero at the so-called the second virial coefficient of a gas.
Boyle point TB .
One of the great early triumphs of statistical physics was to show
how B2 ( T ) arises from the underlying intermolecular potential. Its
derivation gives us an opportunity to review some basic features of
classical statistical physics that will be needed later. For the pur-
pose of this deviation, it is more convenient to work in the canonical
ensemble, although the more systematic approach uses the grand
canonical ensemble.2 We assume that classical statistical physics is 2

D.A. McQuarrie. Statistical Mechanics.
valid, whereby if the energy of a system of N particles of mass m is Harper & Row, New York, 1976
p2i
E= ∑ 2m + ∑ u(ri , rj ), (2.3)
i i< j
where pi is the momentum of particle i. Here, u is the intermolecular

potential that we assume is only a function of the distance rij =
|ri − r j |, and the restriction on the sum avoids double-counting. The
classical partition function for the N-particle system is then
1 1
Z Z
Z= d3N p d3N r e− βE , (2.4)
h3N N!
where β = 1/k B T and the notation d3N p stands for dp1 dp2 . . ., etc.
Recall that the factor of h−3N is a vestige of the fully quantum ap-
proach, where one must measure volumes in phase space, and serves
to render Z dimensionless (recall that the product of a coordinate
and a momentum has units of angular momentum (energy×time),
as does h). The factor of 1/N! arises from the indistinguishability of
the particles.
Recognising that the configurational integral d3N r · · · is V N for a
R
non-interacting system (u = 0), where V is the volume of the system,

and that the momentum integrals can be done explicitly, indepen-
dent of the presence of u, we can factor out the ideal, non-interacting
partition function Z0 , leaving the interacting part Q, as
N
1 V 1
Z
Z= d3N r ∏ e− βuij , (2.5)
N! Λ3 VN i< j
| {z }|
Z
{z }
0
Q
where uij = u(rij ) and Λ = (2πh̄2 /mk B T )1/2 is the thermal de

Broglie wavelength, and we have written the exponential of a sum
as the product of exponentials.
The next step in developing a perturbation theory is to use the
first important method of mathematical physics (adding zero), and
thus to write
e− βuij = 1 + f ij with f ij = e− βuij − 1, (2.6)
and observe that f ij vanishes for a non-interacting system, and serves

as a small parameter for an expansion in density. The quantity
f ij is known as the Mayer f -function.3 From the exact expression 3
Mayer, J.E. Statistical Mechanics of
Condensing Systems. I. J. Chem. Phys.,
∏i< j 1 + f ij we can expand for small f , noting that there are N ( N −

5:67, 1937
1)/2 ∼ N 2 /2 identical terms linear in f ij , leading to
!
1
Z
Q = d r 1 + ∑ f ij + · · ·
3N
(2.7)
VN i< j
N2
Z
≃ 1+ d3 r f (r ) + · · · , (2.8)
2V
where, having done N − 1 integrals, there is one remaining factor
of V in the denominator. From the partition function Z we obtain
microscopic physics (lectures 2-4) 27
the free energy F = −k B T ln Z, and then from F the pressure p =

−(∂F/∂V )T,N in the form of Eq. 2.2, and thus we identify
u(r)
1
Z
B2 = − d3 r e− βu(r) − 1 . (2.9)
2
This is an exact expression for B2 .
A typical intermolecular potential u(r ) has a strong repulsion at
short distances and a long-range attractive tail falling off as r −6 as ur(r)
r
r → ∞ due to van der Waals interactions (the details of which are
discussed below), as in Fig. 2.2. It was van der Waals who, in his
revolutionary 1873 PhD thesis,4 had the crucial idea that the ther- ua(r)
modynamics of liquids and gases potential could be understood by
partitioning the intermolecular potential into two parts: a purely re- Figure 2.2: Decomposition of molecular
interaction potential (red) into purely
pulsive part that leads to "excluded volume" around each molecule,
attractive and purely repulsive compo-
and a purely attractive part whose effects on the thermodynamics nents (dashed).
could be estimated within perturbation theory. We can compare this
4
van der Waals, J.D. Over de Continuiteit
calculation to our results on the virial coefficient.
van den Gas- en Vloeistoftoestand (On the
In van der Waals’ calculation of the contribution of u a to the en- continuity of the gas and liquid state). PhD
ergy of a gas, it is assumed that the probability distribution of parti- thesis, Leiden University, 1873
cles around a given one (the radial distribution function) is uniform,

rather than having the characteristic oscillations due to particle im-
penetrability. Under this assumption the energy Ea is
1
Z
Ea = Nρ d3 ru a (r ) , (2.10)
2
where again the factor of 1/2 avoids double-counting. We then de-
fine the parameter a as
1
Z
a=− d3 r u a (r ) , (2.11)
2
so Ea = − aNρ and the contribution to the pressure is −(∂Ea /∂V ) T,N =
− aρ2 . Thus, a first correction to the equation of state is
p ≃ ρk B T − aρ2 . (2.12)
Further, van der Waals realized that effect of the hard-core interac-
tions could be accounted for by subtracting from the total volume
V an amount proportional to the number of particles N, with an
effective excluded volume per particle b. Putting these two effects
together one has the revised equation of state
( p + aρ2 )(V − Nb) = Nk B T (2.13)
Expanding p for small ρ we find

a
B2 ( T ) = b − . (2.14)
kB T
This has a form that is qualitatively like that seen in experiment (Fig.
2.1), with saturation at high temperatures to a constant, reflecting en-
tropic effects, and a divergence to −∞ at low temperatures as the at-
tractive part of the potential dominates. We conclude that the essence
of the temperature dependence of B2 ( T ) is captured by the ideas of

excluded volume and mean field attraction.
A final point to make is that this result emerges naturally from the
exact form of B2 in Eq. 2.9 if we assume the interaction potential is
+∞ between r = 0 and some radius σ and sufficiently weak beyond
σ that the βu a (r ) ≪ 1. Then,
2πσ3 a
B2 ( T ) = − , (2.15)
3 kB T
where a is defined in (2.11), with the integration from σ to ∞. This
gives a precise meaning to the excluded volume parameter b; σ is
twice the hard-core radius, so b is four times the particle volume.
An important lesson from above is that interaction terms quadratic
in a density or concentration must be interpreted as free energies,
involving energy and entropy, rather than being purely energetic.
Moreover, we see how the integrated strength of the pairwise inter-
action, as appears in (2.11), arises naturally by mean field arguments.
They may also be understood as the zero-momentum values of the
Fourier transform of the potential, but for that transform to be well-
defined it is necessary to remove the singular repulsive contribution
to the potential, as we did in the decomposition of Fig. 2.2.
2.2 Van der Waals interactions
2.2.1 The pairwise intermolecular potential

Van der Waals’ name is of course also associated with the fluctuating-
dipole interactions between neutral objects. To get insight into the
physics responsible for the long-range attraction between neutral
atoms or molecules, we follow essentially verbatim the very nice
derivation by Holstein.5 The model of interacting atoms employed 5
Holstein, B.R. The van der Waals in-
is two charged harmonic oscillators with positive charges that are teraction. Am. J. Phys., 69:441, 2001
fixed in place at some separation R and whose negative charges can

oscillate back and forth under the action of a spring of constant k,
as in Figure 2.3. The artifice of a spring connecting each proton and
k _ k _ Figure 2.3: Geometry of two interacting

+ + "atoms" separated by a distance R, each
x1 x2 with electron and proton connected by
a spring of constant k, with internal dis-
R placements between each electron and
its nucleus of x1,2 .
√
electron introduces a natural oscillation frequency ω0 = k/m for
each "atom", where m is the electron mass. The Hamiltonian H is a
sum of the electron kinetic energy and spring energy H0 ,
p21 1 p2 1
H0 = + mω02 x12 + 2 + mω02 x22 (2.16)
2m 2 2m 2
and the electrostatic interactions H1 ,

2 1 1 1 1
H1 = e + − − . (2.17)
R R − x1 + x2 R − x1 R + x2
This is exact but cumbersome. The physically interesting limit is that

of large separations, with R ≫ | x1 |, | x2 |, where H1 becomes
2e2 x1 x2
H1 ≈ − , (2.18)
R3
whose R−3 fall-off with distance we recognize as a dipole term. This

cross term can be eliminated through the coordinate transformation
x1 ± x2 x+ + x− x+ − x−
x± = √ ⇒ x1 = √ , x2 = √ . (2.19)
2 2 2
The total Hamiltonian then becomes
p21 1 2 2 p2 1 2 2
H≃ + mω+ x+ + 2 + mω− x− , (2.20)
2m 2 2m 2
where the new frequencies are
2 2e2
ω± = ω02 ± . (2.21)
mR3
Having diagonalised the Hamiltonian, the interaction potential u(r )
is just the change to the ground state energy of the system due to the
Coulomb interactions, namely the shift is zero-point energies,
1 1 1
u (r ) = h̄ω+ + h̄ω− − 2 · h̄ω0
2 2 2
2 2
(e /mω0 ) 2
1
≃ − h̄ω0 +··· , (2.22)
2 R6
where we have expanded the frequencies ω± to leading order. Cru-
cially, this is an attractive interaction. In grouping the factors in (2.22),
we have isolated the characteristic energy h̄ω0 that sets the overall
scale for the interaction. It would be typical of an internal excitation
energy from an s state to a p state, since the interaction is fundamen-
tally due to virtual transitions to states with dipole moments. Hence,
h̄ω0 is typically several electron volts, where 1 eV is 40 kB T.
Recognizing the R−6 dependence, we observe that e2 /mω02 must
be a characteristic volume (remember we are working in Gaussian
units!). To check this note that the simplest relationship between an
induced electric dipole moment d and an applied electric field E is
d = αE, (2.23)
where α is the "polarisability". d has units of charge×length (Q · L),

and the units of E are Q/L2 , and thus α ∼ L3 ; it is a volume. For a
hydrogen atom we thus deduce that α ∼ a30 , as the Bohr radius a0 is
the only characteristic length scale.
That it is proper to interpret e2 /mω02 as the polarisability can be
seen by placing the charged harmonic oscillators in Fig. 2.3 in an
electric field E0 x̂ acting to displace the electrons only (recall that the
positive charges are fixed in place), so
H = H0 + eE0 x1 + eE0 x2 . (2.24)

A little manipulation shows that this can be rewritten as
1 1 eE0
H= mω02 z21 + mω02 z22 + · · · ; z1,2 = x1,2 + , (2.25)
2 2 mω02
where the ellipses represents unimportant constants. We thus have

a new pair of oscillators whose equilibrium positions are linearly
shifted by the field. The induced dipole moment is the electron
charge times that shift, namely e2 E0 /mω02 , and thus α = e2 /mω02 .
Putting together all of the pieces, we see that the interaction be-
tween two fluctuating dipoles can be written in the compact form
1 h̄ω0 α2
u (r ) = − . (2.26)
2 r6
It follows that the typical scale of this energy within a condensed
phase of number density n is obtained by setting r −6 ∼ n2 , giving
u ∼ h̄ω0 (αn)2 . (2.27)
For most dense strongly insulating liquids the product αn ∼ 0.1 −

0.2. This is relevant to the calculation of interacting extended bodies,
as we discuss in the next section. The quantity αn also occupies
6
Herzfeld, K.F. On atomic properties
a very important role in the theory of metal-insulator transitions, which make an element a metal. Phys.
which are predicted to occur when it exceeds an O(1) threshold.6 Rev., 29:701–705, 1927
2.2.2 Interaction of extended objects

We now use the basic results from the previous section to understand
7
Bussi, Y., Shimoni, E., Weiner, A.,
Kapon, R., Charuvi, D., Nevo, R., Efrati,
the van der Waals interaction between extended objects. A prime ex- E., Reich, Z. Fundamental helical geom-
ample is the stacks of thylakoid membranes found in chloroplasts of etry consolidates the plant photosyn-
thetic membrane. Proc. Natl. Acad. Sci.
green algae, as shown in Fig. 2.4.7 The balance of forces that de- USA, 116:22366–22375, 2019
Figure 2.4: Thylakoid membranes. A

10 nm thick slice of lettuce chloroplast
imaged with electron microscopy. The
so-called "grana" (G), stacks of disc-
shaped thylakoid membranes are inter-
spersed with unstacked "stroma" thy-
lakoids (SL), all immersed in water and
enclosed in the chloroplast envelope
(black arrow). Scale bar is 200 nm.
From Ref. 7.
8
Sculley, M.J., Duniec, J.T., Thorne,
termine the spacing observed in these structures has been analysed S.W., Chow, W.S. and Boardman, N.K.
along the lines we discuss below.8 The stacking of chloroplast thylakoids.
To calculate the interaction between slabs we start with the inter- Quantitative analysis of the balance of
forces between thylakoid membranes of
action of a single atom with a laterally infinite slab with atom num- chloroplasts, and the role of divalent
ber density n, as indicated in Fig. 2.5. Let us denote the interaction ions. Arch. Bioch. Biophys., 201:339–346,
1980
between two atoms as
C
u11 (r ) = − 6 . (2.28)
r
Working in cylindrical coordinates the interaction u1S between one

atom and the slab is
Z 0 Z 2π Z ∞ q
u1S (z) = n dz′ dϕ rdru11 (z − z′ )2 + r2 (2.29) z
−δ 0 0 y
x
Z z+δ Z ∞
1
= −2πCn dz′ rdr ′2 (2.30) z
z 0 ( z + r 2 )3

πCn 1 1 δ
= − − , (2.31)
6 z3 ( z + δ )3
where we changed variables from z′ to z′′ = z − z′ in the second Figure 2.5: Geometry of an atom inter-
line (and dropped the prime). The power law r −6 of the interparticle acting with a laterally infinite slab of
thickness δ.
potential has become z−3 by integration over three dimensions.
Now we consider two slabs separated by a distance d (Fig. 2.6).
With the result 2.31, the slab-slab interaction uSS (d) is obtained by
integrating u1S (z) over the thickness of the second slab. If A is the
area of that slab, then there are Andz atoms in differential thickness δ
dz, and the interaction energy per unit area VA (z) ≡ uSS (z)/A is
Z d+δ d
VA (d) = dznu1S (z)dz (2.32)
d
δ
A 1 2 1
= − H − + , (2.33)
12π d2 (d + δ)2 (d + 2δ)2
Figure 2.6: Geometry of two slabs of
where have introduced the Hamaker constant9
thickness δ separated by face-to-face
distance d.
A H = π 2 n2 C, (2.34)
9
Hamaker, H.C. The London-van der
with units of energy. Based on our simple picture of the constant C Waals attraction between spherical par-
ticles. Physica, 4:1058–1072, 1937
in Eq. 2.26, we see that A H ∼ (1/2)π 2 h̄ω0 (αn)2 , so with h̄ω0 ∼ 3 eV
and αn ∼ 0.1 − 0.2 as discussed above, we estimate A H ∼ 5 − 25 kB T.
2.3 Screened electrostatic interactions
2.3.1 Poisson-Boltzmann and Debye-Hückel theory

Many of the interesting objects in biology interact both through van
der Waals forces and (screened) electrostatic interactions. The be-
haviour of systems with these two competing forces was first estab-
lished in the classic work of Verwey and Overbeek.10 As the van der 10
Verwey, E.J.W. and Overbeek, J.Th.G.
Theory of the Stability of Lyophobic Col-
Waals forces decay as an inverse power of distance, and the electro-
loids. Elsevier Publishing Company,
static contribution decays exponentially (as we shall see), the com- Inc., New York, 1948
bined potential is attractive at long distances and repulsive at short
distances, leading to a potential minimum. It is this balance of forces
that sets the spacing of thylakoid membranes (Fig. 2.4), for example.
Two elementary charges in vacuum separated by a distance r have
an electrostatic energy (in cgs units) of
e2
E = eϕ(r ) = , (2.35)
r
where ϕ = e/r is the electrostatic potential. Essentially all of liv-
ing matter is infused with water, whose dielectric constant ϵ ∼ 80
reduces the energy scale to
e2
E= . (2.36)
ϵr
While reduced by nearly two orders of magnitude, this interaction
still falls off as an inverse power of distance. A sense of the scale
of E is obtained by finding the distance at which it is comparable to
thermal energy. This Bjerrum length is
e2
λB = ∼ 0.7 nm. (2.37)
ϵk B T
Even pure water has small amounts of ionic species, since water
molecules are in equilibrium with protons (H+ ) and hydroxyl ions
(OH− ) according to the reaction scheme
H2 O ⇌ H+ + OH− . (2.38)
Recall that the pH of a solution is defined as − log10 [H+ ], where

the notation [· · · ] means concentration in moles/litre. Thus, a neu-
tral pH of 7 corresponds to a proton concentration of 10−7 M/l ∼
6 × 1013 cm−3 . These and any additional ions present will screen the
bare electrostatic interactions considered above; mobile charges op-
posite in sign to a given charge cluster around it in a diffuse cloud,
screening it from other distant charges. The standard formalism to
compute the ionic distribution and resulting interactions is Poisson-
Boltzmann theory, or, in its linearised form, Debye-Hückel theory.11 11
Debye, P. and Hückel, E. The the-
They are based on two principles. The first is the Poisson equation ory of electrolytes. I. Lowering of freez-
ing point and related phenomena (En-
relating the electrostatic potential ϕ to the charge density ρ glish translation). Phys. Zeit., 24:185–
206, 1923
4πρ
∇2 ϕ = − . (2.39)
ϵ
The second is the Boltzmann distribution relating the ionic concen-
trations cs of the species s of charge zs e (with zs the "valence") to the
electrostatic potential,
cs = c0s e− βzs eϕ , (2.40)
where c0s is a background concentration. Combining these, we obtain
the Poisson-Boltzmann equation,
4π
∇2 ϕ = −
ϵ ∑ zs ec0s e−βzs eϕ . (2.41)
s
It is simplest to consider the case of a z : z electrolyte, in which

the dissolved neutral species dissociates into two ions of equal and
opposite charges, such as sodium chloride (1 : 1; NaCl → Na+ +Cl− )
or copper sulfate (2 : 2; CuSO4 → Cu2+ +SO24− ). Then we have
8πzec0
∇2 ϕ = sinh( βzeϕ). (2.42)
ϵ
This is a very nonlinear equation for which analytical solutions can
be found only in simplified geometries. Much of the important
physics of screened electrostatics can be seen in the weak field limit,
when βzeϕ ≪ 1, where we keep only the linearised r.h.s. of 2.42 and
obtain the Debye-Hückel equation

∇2 − κ 2 ϕ = 0, (2.43)
where κ −1 is the Debye-Hückel length

1/2
ϵk B T 10 nm
κ −1 = ∼ p . (2.44)
8πz2 e2 c0 c0 [mM]
Here, we have expressed the concentration in the units of millimolar

(mM), as typical concentrations of ions such as sodium are in the
range 1 − 200 mM. We conclude that the screening length is nano-
metric in typical biophysical contexts.
The mathematical problem presented by the Debye-Hückel equa-
tion 2.43 is the modified Helmholtz equation. Let us first solve this in
one spatial dimension, for the half-space above a surface lying in
the x − y plane (z = 0). The relevant general solution of 2.43 in-
volves the functions exp(±κz), and the requirement that the solution
be bounded as z → ∞ leaves only the decaying exponential. If the
surface potential is fixed at ϕ0 , say, then
ϕ(z) = ϕ0 e−κz . (2.45)
The induced charge density σ0 on the surface can be computed in

the usual way (by considering a Gaussian pillbox that straddles the
surface). It obeys
4πσ0 ϵκ
−n̂ · ∇ϕ = ⇒ σ0 = ϕ0 , (2.46)
z =0 ϵ 4π
where n̂ is the outward normal to the surface. An analogous calcula-

tion yields the potential when the surface charge itself is specified,
4πσ0 −κz
ϕ(z) = e . (2.47)
ϵκ
Within Debye-Hückel theory the surface charge and surface poten-
tial are linearly related, but in the more general Poisson-Boltzmann
setting this relationship is nonlinear.
Observe that the DH equation ∇2 − κ 2 ϕ = 0 is the Euler-Lagrange

equation for the functional

1 1
Z
ϵ
G= d3 r (∇ϕ)2 + κ 2 ϕ2 , (2.48)
4π 2 2
where we employ the general Euler-Lagrange formula
δG ∂ ∂(· · · ) ∂(· · · )
=− + , (2.49)
δϕ ∂x ∂ϕx ∂ϕ
where (· · · ) is the integrand of the functional. The DH equation

holds as the minimisation condition on (2.48) for any constant factor
in front of the integral, but with the choice in (2.48), we recognize
the first term as the standard form of the electrostatic energy density
ϵE2 /8π in a medium of dielectric constant ϵ. The second term is the

weak-field approximation of an entropic contribution.
If we integrate by parts the term in 2.48 involving (∇ϕ)2 , we ob-
tain from Green’s first identity a bulk contribution that vanishes by
the DH equation, leaving only the surface term
1
Z
G= dSσϕ, (2.50)
2 S
where we have used 2.46 to express the normal derivative of the

potential in terms of the charge density σ. This is the appropriate
free energy for a system in which the surface charge is specified, and
it is completely determined by those Neumann boundary conditions
on the surface. There is no free lunch, for of course the potential on
the surface must be found from the solution of the DH everywhere.
For situations with fixed surface potential rather than fixed charge,
the surface free energy must be Legendre transformed, which is
equivalent to accounting for the work done against the battery that
held the potential fixed. This new free energy F is
1
Z Z
F = G − dSσϕ = − dSσϕ. (2.51)
S 2 S
The factors of 1/2 in Eqs. 2.50 and 2.51 arise from the linear relation
between surface potential and surface charge. In the more general
case, the free energy is expressed as a "charging integral" of the form
Z Z ϕ0
′ ′
− dS σ (ϕ )dϕ . (2.52)
S
For our simple planar examples above,
G 2πσ02 F ϵκϕ02
= and =− , (2.53)
A ϵκ A 8π
where A is the total area of the surface.
2.3.2 Electrostatic interactions between surfaces

Now we move on to calculate the electrostatic interaction between
two surfaces, considering both the case of fixed charged density
and the case of fixed potential. Although the case of a specified
charge density is more biophysically relevant than a specified poten-
tial, which one would imagine is appropriate to an electrochemical
context, the comparison between the two serves to illustrate some
general features of the way that quadratic field theories determine
interactions between surfaces. This takes on added relevance for
neutral, dipolar membranes, which are believed to polarize the in- 12
Marcelja, S. and Radić, N. Repulsion
of interfaces due to boundary water.
tervening water molecules and produce a "hydration repulsion" be- Chem. Phys. Lett., 42:129–130, 1976
tween them,12 measured by clever experiments some years ago.13 13
Lis, L.J., McAlister, M., Fuller, N.,
For two surfaces held at the same potential ϕ0 and located at Rand, R.P. and Parsegian, V.A. Inter-
actions between neutral phospholipid
±d/2, the potential is the symmetric combination of the fundamental bilayer membranes. Biophys. J., 37:657,
exponential solutions found previously, 1982
cosh(κz)
ϕ = ϕ0 . (2.54)
cosh(κd/2)
Using this we find the charge density at the plate at z = d/2. Here,
−n̂ · ∇ = d/dz, thus
ϵκϕ0
σ (d/2) = tanh(κd/2) . (2.55)
4π
At the bottom plate −n̂ · ∇ = −d/dz, but with sinh(−d/2λ) =
− sinh(d/2λ) the charge density is the same. As the charge and
potential do not vary with position over these flat surfaces the sur-
face integration will just give a factor of the surface area A. The free
energy per unit area V (d) ≡ F (d)/A is
ϵκϕ02
V (d) = − tanh(κd/2). (2.56)
4π
In the limit d → ∞ we recover twice the single-surface result 2.53.
Subtracting the "self energy" of the surfaces, we obtain the repul-
sive interaction potential VR (d) = V (d) − V (∞),
ϵκϕ02

κd
VR (d) = 1 − tanh . (2.57)
4π 2
At large argument, tanh approaches unity from below, so this is

clearly a repulsion, as expected. In detail, if κd ≫ 1 we note that
tanh(z) ≃ 1 − 2e−2z + · · · , and thus
ϵκϕ02 −κd
VR (d) ≃ e , (2.58)
2π
which therefore has the same exponential decay as the electrostatic
potential itself. A second important point is that VR (d) is bounded
as d → 0; when the two charged surfaces are brought closer together,
their induced charges decrease fast enough to avoid a singularity.
Now we consider the case of two surfaces with fixed charge den-
sity σ0 . A simple calculation shows that the required potential is
4πσ0 cosh(κz)
ϕ( x ) = , (2.59)
κϵ sinh(κd/2)
with corresponding potential energy per unit area
4πσ02

κd
VR (d) = coth −1 . (2.60)
κϵ 2
The function coth approaches unity from above at large argument,

and thus we again have a repulsive interaction. However, this time
phosphatidylcholine
there is a divergence at short distances due to the fixed charge den- head group
sities that are brought ever closer together.

Let us estimate the scale of energies involved in the case of charged
hydrocarbon tails
lipid membranes. Lipid molecules are "amphiphilic", with a struc- C16 (palmitic acid)
ture that is a fusion of parts that are hydrophilic (having an affin- Figure 2.7: The chemical structure
of the phospholipid DPPC (dipalmi-
ity for water) and hydrophobic (repelling water). Figure 2.7 shows toylphosphatidylcholine).
the structure of the lung surfactant dipalmitoylphosphatidylcholine
(DPPC), which has two 16-carbon hydrophilic "tails" and a phos-
phatidylcholine head group. In diagrams of this type, the zig-zag
line represents the repeating group CH2 , terminated by the methyl
group CH3 , with the hydrogen atom labels omitted for clarity. These
particular hydrocarbon chains are said to be "saturated" because ev-
ery carbon atom is 4-fold coordinated, bonded to two carbon atoms
and two hydrogen atoms (the maximum possible), unlike "unsatu-
rated" lipids with one or more C−C double bonds, thereby having
carbon atoms that have only a single bond to a hydrogen atom.
When placed in water, lipids self-organize into structures that iso-
late the oily hydrocarbon tails from contact with water molecules,
leaving the hydrophilic headgroups in contact with water. This "hy-
drophobic effect" plays a central role in cellular biology, dictating the 14
C. Tanford. The hydrophobic effect:
formation of micelles and biological mem-
structure of lipid assemblies, the folding of proteins, and localization branes. Wiley, New York, 1973
of channels in membranes.14 The simplest structure is a "micelle", a
water
spherical or cylindrical structure with the tails on the inside and the head groups
heads facing outwards. hydrocarbon

tails
Bilayer membranes have the structure shown in Fig. 2.8. They
are essentially thin fluid sheets a few nm thick in which typically water
(but not always) the lipid molecules are free to diffuse laterally. A Figure 2.8: Structure of a lipid mem-
brane in water.
typical charged head group has a single elementary charge in an
area of ∼ 1 nm2 . If the Debye-Hückel screening length is ∼ 1 nm,
then 4πσ02 /κϵ ∼ 40 mN/m, which can be compared to the surface
tension of water (80 mN/m). Expressed another way, 40 mN/m is
40 pN·nm/nm2 , or about 10 kB T/nm2 .
Returning to the motivational problem of systems such as stacks
of thylakoid membranes, we now add the van der Waals interaction
2.33 between two membranes to the screened electrostatic repulsion
2.60. A convenient set of nondimensionalisations involves scaling
distances with the membrane thickness δ and the potential energy
with the Hamaker constant. Thus defining
d 12πδ2
x= , K = κδ, V= (VA + VR ), (2.61)
δ AH
we obtain

Kx 1 2 1
V ( x ) = Γ coth −1 − 2 + − , (2.62)
2 x ( x + 1)2 ( x + 2)2
where
24π 2 σ02 δ2
Γ= . (2.63)
ϵκA H
Using our estimates from above, and taking a membrane thick-

ness of 3 − 6 nm, we find Γ ∼ 100 and K > 5. While the large size
of Γ seems to imply a very large mismatch between the electrostatic
and van der Waals scales of energy, the exponential attenuation of
the former allows a competition with the latter at distances beyond a
few screening lengths. The graph shown in Fig. 2.9, reveals an essen-
tial feature of this competition, namely a minimum in V at values of
x of order unity. This would correspond to the equilibrium spacing
of the thylakoid membranes. Moreover, the global minimum of the
energy is at contact (x = 0), where the van der Waals term diverges.
As K increases (corresponding to an increase in salt concentration)
5 Figure 2.9: Interaction energy 2.62 be-

K=5
tween membranes for Γ = 30 and var-
5.5 ious values of K, as shown. Obtained
V using the program Figure_209_DLVO.m.
6
VR
0
6.5
7 VA
-5
0 0.5 1 1.5 2
x
the minimum at larger x disappears, leaving only the minimum at

contact. This implies the well-known "salting-out" effect in colloidal
suspensions, whereby an increase of dissolved salt leads to precipi-
tation of the suspended particles.
2.4 Geometrical aspects of screened electrostatics
In the previous section we considered only the simplest planar ge-

ometries, but many of the interesting structures in biology have non-
trivial curvatures. As the Debye-Hückel screening length is nano-
metric, while the radius of curvature R of structures within the cell
is often much larger, it is natural to exploit that separation of length
scales to develop a systematic, perturbative approach using κR as
a small parameter. Of particular interest historically has been the
contribution of screened electrostatic interactions to the elasticity of
membranes. Intuitively, bending a charged surface brings the sur-
face charges closer together, producing a repulsion that effectively
increases the stiffness of the surface.
In Chapter 5 we develop the necessary differential geometry of
surfaces to deal with detailed problems in membrane physics. Here,
we simply recall that at every point on a surface there are two prin-
cipal radii of curvature, termed R1 and R2 , and from these we can
construct two important geometric quantities; the mean curvature

1 1 1
H= + (2.64)
2 R1 R2
and the Gaussian curvature

1
K= . (2.65) 15
Canham, P.B. The minimum energy
R1 R2
of bending as a possible explanation of
the biconcave shape of the human red
In the early 1970s, Canham15 and Helfrich16 independently intro- blood cell. J. Theor. Biol., 26:61–76, 1970
duced a model for the elasticity of "fluid" membranes of fixed area, 16
Helfrich, W. Elastic properties of
composed of lipids that are free to diffusive within the sheet, that lipid bilayers: theory and possible ex-
periments. Z. Naturforsch. Teil C, 28:693–
have vanishing shear modulus. They proposed that the characteris- 703, 1973
tic biconcave discoid shapes of red blood cells (erythrocytes), shown
in Fig. 2.10, are minimisers of this energy under the constraint of a
fixed enclosed volume of water less than the equivalent sphere (the
sphere having the same surface area as the membrane).
This functional is a surface integral of a quadratic form in the
curvatures,
1 1
Z
E = dS k c ( H − H0 )2 + kc K . (2.66) Figure 2.10: Electron micrograph of red
2 2 blood cells.
Here, k c and k̄ c are the bending modulus and Gaussian curvature

modulus of the membrane, and H0 is the spontaneous curvature that
incorporates the possibility of a preferred curvature in the ground
state. It is often a constant, but can vary spatially if the membrane
composition is inhomogeneous. From the Gauss-Bonnet theorem,
the second integral in 2.66 is a constant that depends only on the
topology of the closed surface of interest.
The problem of quantifying the electrostatic contribution to the
elastic moduli and H0 can be addressed several ways. The first17 is 17
Winterhalter, M. and Helfrich, W. Ef-
a method by which comparison of the energies of planes, cylinders, fect of surface charge on the curvature
elasticity of membranes. J. Phys. Chem.,
and spheres yields those quantities directly. The second is a bound- 92:6865–6867, 1988
ary perturbation method for nearly planar surfaces,18 while the third 18
Goldstein, R.E., Pesci, A.I. and
utilizes a "multiple-scattering" method19 introduced in the context of Romero-Rochín, V. Electric double
layers near modulated surfaces. Phys.
the wave equation.20 We outline the essential features of the first two Rev. A, 41:5504–5515, 1990
approaches, leaving full calculations to the Example Sheets. 19
Duplantier, B., Goldstein, R.E.,
Romero-Rochín, V. and Pesci, A.I.
Geometrical and topological aspects
2.4.1 Comparative approach of electric double layers near curved
surfaces. Phys. Rev. Lett., 65:508–511,
For simplicity we focus on the problem of fixed surface charge den- 1990
sity σ0 . On dimensional grounds we expect the elastic constants
20
Balian, R. and Bloch, C. Distribu-
tion of eigenfrequencies for the wave
(units of energy) to scale as equation in a finite domain: I. Three-
dimensional problem with smooth
4πσ02 boundary surface. Ann. Phys. (N.Y.), 60:
k c , k̄ c ∼ , (2.67) 401–447, 1970
ϵκ 3
and based on the estimate 4πσ02 /ϵκ ∼ 10 kB T/nm2 above Eq. 2.61
and an assumed screening length of 1 nm, these moduli are on the
order of 10 kB T. Likewise, as the screening length is the only length
scale in the problem, we expect H0 ∼ κ −1 . The purpose of these
calculations is to make these estimates precise.
As we have already solved the problem for a single plane, we
next consider the cylinder. We wish to solve the modified Helmholtz
equation 2.43 inside and outside a cylinder of radius R, assuming
axisymmetry. The Debye-Hückel equation becomes
∂2

2 ∂ 2
r 2 + r − (κr ) ϕ = 0. (2.68)
∂r ∂r
As this is homogeneous in powers of r, the solution is a function of

κr. The two solutions are modified Bessel functions K0 (κr ) for the outer
problem (decaying at infinity) and I0 (κr ) for the inner problem (well-
behaved at the origin). We solve the problem in those two domains.
Using the identity I0′ (z) = I1 (z), the inner solution is found to be
4πσ I0 (κr )
ϕ (r ) = . (2.69)
ϵκ I1 (κR)
The (uniform) free energy per unit area of the inner surface is
1 1 4πσ02 I0 (κR)
σϕ = . (2.70)
2 S 2 ϵκ I1 (κR)
As outlined above, the limit of interest is κR ≫ 1. We then appeal to

the asymptotic results for modified Bessel functions [Abramowitz &
Stegun21 (henceforth, "A&S") # 9.7.1], 21
M. Abramowitz and I.A. Stegun.
Handbook of Mathematical Functions with
ez µ − 1 (µ − 1)(µ − 9) Formulas, Graphs, and Mathematical Ta-

Iν (z) = √ 1− + +··· , (2.71) bles. Dover Publications Inc, Mineola,
2πz 8z 2!(8z)2 N.Y., 1965
where µ = 4ν2 . From these results, we see that the exponential

prefactors of the asymptotic forms cancel in the ratio 2.70, leaving a
simple expansion in inverse powers of κR,
I0 (κR) 1 3
∼ 1+ + +··· . (2.72)
I1 (κR) 2κR 8(κR)2
Carrying through an expansion of the energy density for a spherical

surface also yields an expansion in 1/κR, allowing a direct compari-
son to be made with the Helfrich formula. Precise calculations must
account for the finite thickness of the membrane and the possibility
of a charge density difference between the two sides. We leave it to
the student (Example Sheet 1) to complete these calculations.
2.4.2 Perturbative approach

z
Next, we sketch the basic features of a perturbative approach to find- y
x
ing the energetics of electric double layers near a non-flat boundary. ϕ0
Let the surface position be z = h( x ), where h is a single-valued height h(x)
function independent of y, and the surface is held at a fixed potential Figure 2.11: A ripply surface held at
ϕ0 (Fig. 2.11). Thus, we must solve the Debye-Hückel equation for fixed potential.
the potential ϕ( x, z) subject to the boundary condition
ϕ( x, h( x )) = ϕ0 , (2.73)
and ϕ = 0 at z = ∞. The mathematical challenge arises from the fact

that in general we do not know the Green’s function of the modified
Helmholtz operator for a domain bounded by an arbitrary function
h( x ). But, we can perturbatively connect the solutions at finite h( x )
to those at h = 0, where we know the solution. This "boundary
perturbation theory" is a method of broad applicability.
We introduce a dimensionless small parameter ε as a counting
device, and expand the surface boundary condition in a Taylor series,
1
ϕ( x, ϵh( x )) ≃ ϕ( x, 0) + εh( x )ϕz ( x, 0) + ε2 h( x )2 ϕzz ( x, 0) + · · · ,
2
(2.74)
where subscripts denotes partial differentiation. We expect that the
solution itself—in the bulk—also has an expansion in powers of ε,
ϕ( x, z)) ≃ ϕ(0) ( x, z) + εϕ(1) ( x, z) + ε2 ϕ(2) ( x, z) + · · · . (2.75)

The governing equation does not depend on ε; if we substitute

2.75 into 2.43, at every order we will have
(∂ xx + ∂zz − κ 2 )ϕ(n) ( x, z) = 0. (2.76)
Merging the expansion of the boundary condition with the expan-

sion of the solution we arrive at a sequence of boundary conditions
for each order of solution. At leading order, we recover the boundary
condition at a flat surface,
O(ε0 ) : ϕ(0) ( x, 0) = ϕ0 . (2.77)
Clearly, the solution for all ( x, z) is
ϕ(0) ( x, z) = ϕ0 e−κz . (2.78)
At first order we find the boundary condition

(0)
O ( ε1 ) : ϕ(1) ( x, 0) = −h( x )ϕz ( x, 0) = κh( x )ϕ0 . (2.79)
Thus, the problem with a boundary condition on a ripply surface has

become one with a flat boundary but an inhomogeneous potential.
At quadratic order we have a similar kind of result,
1 (0) (1)
O ( ε2 ) : ϕ(2) ( x, 0) = − h2 ( x )ϕzz ( x, 0) − h( x )ϕz ( x, 0). (2.80)
2
A convenient way to solve these boundary-value problems is to
work in Fourier space. Let us define the Fourier transforms of the
potential and h with respect to x (keeping the z-dependence of ϕ),
Z Z
ϕ̂(m) (k, z) = dxeikx ϕ(m) ( x, z) and ĥ(q) = dxeiqx h( x ). (2.81)
The modified Helmholtz equation (∂ xx + ∂zz − κ 2 )ϕ = 0 becomes

∂zz − κq2 ϕ̂(n) (q, z) = 0 , where κq2 = κ 2 + q2 , (2.82)
and is solved by
ϕ̂(n) (q, z) = ϕ̂(n) (q, 0)e−κq z . (2.83)
To solve the sequence of boundary-value problems, we need the
Fourier transform of ϕ at z = 0; this can be determined directly from
the order-by-order boundary conditions. For example, at order ϵ1 we
have from (2.79)
ϕ̂(1) (q, 0) = κϕ0 ĥ(q). (2.84)
To complete the calculation, it is necessary to expand such quantities
as the surface normal vector
−h x êx + êz
n̂ = − p (2.85)
1 + h2x
in order to compute the surface charge as

ϵ
σ( x) = − n̂ · ∇ϕ( x, ϵh( x )) . (2.86)
4π
Details are left for the Example Sheet.
3
Fluctuations & Brownian Motion (Lectures 5-7)
3.1 Equipartition theorem
In Chapter 2 we used the standard formulation of classical statistical

physics to derive the second virial coefficient. Now we wish to use
that same starting point to discuss one of the most important results
in equilibrium statistical physics; the theorem of equipartition.
We consider a system whose potential energy U (q N ) depends on
a set of generalized coordinates q N . As in the derivation of (2.5), the
partition function can be factorised into an ideal part Z0 that only
depends on temperature and the "configurational" partition function
Z
N
Z= dq N e− βU (q ) . (3.1)
We have written this as an integral over continuous variables q; in

the discrete case Z is a sum of Boltzmann factors.
Recall from elementary statistical physics that the probability p( E)
that a system has energy E is simply
e− βU
p( E) = , (3.2)
Z
and the expectation value of a quantity A (denoted by ⟨ A⟩) is
1
Z
⟨ A⟩ = dq N A(q N )e− βU . (3.3)
Z
When U is quadratic in the q N , as in the N = 1 case
1 2
U= kq , (3.4)
2
then R∞ 1 2 − βkq2 /2
−∞ dq 2 kq e

1 2 ∂ ln Z
kq = R∞ − βkq2 /2
=− . (3.5)
2
−∞ dqe
∂β
The final form on the r.h.s. of (3.5) tells us that ln Z is a generating
function. Now we change coordinates in Z to obtain
s
Z ∞ Z ∞
− βkq2 /2 2 2
Z= dqe = dxe− x , (3.6)
−∞ βk −∞
and thus
1
ln Z = − ln β + (terms independent of β). (3.7)
2
Hence, −∂ ln Z/∂β = 1/2β = k B T/2 and the average energy per

degree of freedom is
1 2 1
kq = k B T. (3.8)
2 2
This is the equipartition theorem; when the energy is a quadratic
form in the coordinate, the average energy of that degree of freedom
is k B T/2. It is clear from this derivation that if the energy 3.4 were
replace by a sum over a discrete set of modes, each with a quadratic
contribution, since the partition function would factorise into a prod-
uct of terms, each mode separately would satisfy the theorem.
3.2 Fluctuating continuous objects

h(x)
3.2.1 The tense string
L
Now we apply the equipartition theorem to fluctuating continuous
objects such as strings, biological filaments and membranes. The
m m
simplest example is a string hanging from two pulleys, with masses
Figure 3.1: A string under tension.
attached at its two ends so that gravity produces a tension γ in the
string. If the string is at a finite temperature T, then it will exhibit
thermal fluctuations which we characterise by a height function h( x )
as shown in Fig. 3.1. Relative to a straight string, the excess energy E
of the string is the product of the tension γ and the excess arclength,
Z L q
E=γ 2
dx 1 + h x − L , (3.9)
0
where h x = ∂h/∂x. Expanding the square root under the assumption

of a weakly-sloping curve (|h x | ≪ 1), we have
Z L
γ
E≃ dxh2x . (3.10)
2 0
While 3.10 is quadratic in the displacement function h( x ), we can

not apply the equipartition theorem directly because there are no
identifiable independent degrees of freedom. This is where a Fourier
(or other appropriate) representation is useful. Suppose we have
Dirichlet boundary conditions h(0) = h( L) = 0. Then we can write
∞ nπx
h( x ) = ∑ an sin L
. (3.11)
n =1
To be clear, with the dual views of ensemble averages and time aver-
ages in statistical mechanics, this representation holds for any partic-
ular member of the ensemble of systems or at any particular moment
in time for a single string as it fluctuates; we view the amplitudes an
as the new fluctuating quantities and, in the usual way in statisti-
cal physics, we draw conclusions about the statistical properties of h
from those of the set { an }. By the orthogonality of the modes,
∞ nπ 2
γL
E=
4 ∑ L
a2n . (3.12)
n =1
Note the explicit appearance of the system size L.

fluctuations & brownian motion (lectures 5-7) 43
The form of 3.12 is the desired sum of independent quadratic con-

tributions, so if the string is in thermal equilibrium at temperature
T, then the probability distribution of each mode is proportional to
exp(−γ(nπ )2 a2n /4Lk B T ), from which we deduce that ⟨ an ⟩ = 0 and
(by equipartition) the mean squared mode amplitudes are
2k B TL
⟨ a2n ⟩ = . (3.13)
γπ 2 n2
We trace the n2 in the denominator of 3.13 to the role of tension in

the energy, which involves two powers of the derivative of h.
It is of interest to calculate the average envelope of the fluctuating
string, the variance of the displacement ⟨h2 ( x )⟩,
∞ ∞ mπx nπx
⟨h2 ( x )⟩ = ∑ ∑⟨ am an ⟩ sin sin
m n L L
∞
k B TL 2 sin2 (nπx/L)
=
γ π2 ∑ n2
n =1
kB T
= LF (ξ ), (3.14)
γ
where the final form defines the scaling function F shown in Fig. 3.2,
which depends on x and L only through the ratio ξ = x/L. The
linear scaling of the variance with system size L is reminiscent of
the scaling ℓ2 ∼ t for diffusion discussed in Chapter 1, and for the
analogous quantity in random walks, as we show later.
0.3 Figure 3.2: Scaling function for variance

of fluctuating string. Obtained using
the program Figure_302_stringvar.m.
0.2
F (9)
0.1
0
0 0.5 1
9 = x=L
3.2.2 Continuum limit
In many cases we will consider a system large enough that it is rea-

sonable to consider a continuum of modes. as an example, we return
to the tense string. Adopting a complex notation
h( x ) = ∑ e−iqx ĥ(q), (3.15)

q
and to be clear on the algebraic steps, we first write each factor of h x

in the energy in terms of its Fourier coefficients as
Z L
γ ′
E= dx ∑ ∑(−iq)(−iq′ )e−i(q+q ) x ĥ(q)ĥ(q′ ). (3.16)
2 0 q q′
The integration over x is performed using the identity

Z L
dxe−ipx = Lδp,0 , (3.17)
0
where δp,0 is the Kronecker delta. For a real function h( x ), we have

ĥ(−q) = ĥ∗ (q), and thus the energy takes on the simple form
γL
E=
2 ∑ q2 |ĥ(q)|2 . (3.18)
q
Equation 3.18 is the generalization of 3.12, so equipartition yields

D E k T 1
|ĥ(q)|2 = B 2 . (3.19)
γL q
In addition to the thermal average of any quantity, we may also

be interested in the system average, denoted by an overbar, where
Z L
1
(· · · ) = dx (· · · ). (3.20)
L 0
The thermal average of the system average has a very compact form
in the case of the height function,
⟨h2 ⟩ = ∑⟨|ĥ(q)|2 ⟩. (3.21)

q
These results are expressed as discrete Fourier sums. Moving to

the continuum limit, we replace sums by integrals using the relation
1 dq
Z
L ∑→ 2π
. (3.22)
q
Combining (3.19) and (3.21), we find
kB T dq
Z
⟨ h2 ⟩ = . (3.23)
2πγ q2
In passing to the continuum limit, we may find it necessary to

introduce cutoffs to these kinds of integrals to obtain finite results. A
small-scale cutoff on a length a represents the finite molecular size,
below which the continuum theory may not make sense, while a
large-scale cutoff represents, for example, the finite system size L.
The corresponding cutoff wavevectors are qmax = π/a and qmin =
π/L. Accounting for such cutoffs, the variance 3.23 becomes

2
kB T 1 1
⟨h ⟩ = −
2πγ qmin qmax
kB T
= ( L − a) . (3.24)
2π 2 γ
In this case, the molecular cutoff scale a may be set to zero without
introducing any divergence in the average of interest, but, as men-
tioned earlier, the variance grows linearly with the large-scale cutoff
L and there is no sensible L → ∞ limit.
The average energy of the filament is proportional to the average
mean variance of the filament slope. A simple calculation shows
kB T
Z
γL 2
⟨ E⟩ = ⟨h ⟩ = L dq
2 x 4π

kB T L
= −1 . (3.25)
4 a
This result shows that the energy density ⟨ E⟩ diverges as the small-
1
Planck, M. Ueber das Gesetz der
scale cutoff is taken to zero. This "ultraviolet" divergence is exactly Energieverteilung im Normalspektrum.
what the quantum theory of blackbody radiation1 managed to tame Ann. Phys., 309:553–563, 1901
by introducing a discreteness to the modes at small scales.
If we generalize this calculation to d spatial dimensions and thus
a d − 1 dimensional surface, we find
d d −1 q k B T kB T q d −2
Z Z
⟨ h2 ⟩ ∼ ∼ dq (3.26a)
(2π )d−1 γq2 γ q2

 k B T 1 q d −3 − q d −3 , d ̸ = 3
γ 3− d min max
∼ (3.26b)
∼ k B T ln ( L/a) , d = 3.
γ
The behaviour at d = 3 is interesting and subtle. The interface

"width" clearly diverges, both as the molecular cutoff a → 0 and in
the limit of an infinite system, but it is a very weak divergence. Tak-
ing L to be 1 cm and a to be 1 nm, we have L/a = 107 , but ln( L/a)
is only ∼ 16. If we take the surface tension γ = 72 mN/m of an
air-water interface, and note that 1 mN/m is 1 pN·nm/nm2 , then the
width of such a macroscopic air-water interface is ⟨h2 ⟩1/2 ∼ 1 nm,
i.e. it is molecular in size, despite the "formal" divergence.
For pedagogical purposes,2 it is instructive to consider how the 2
J.S. Rowlinson and B. Widom. Molecu-
above is modified when there is a symmetry-breaking field in the lar Theory of Capillarity. Oxford Univer-
sity Press, Oxford, 1982
energy that penalises large excursions in the height function h. If
we view the surface bounded by the height function h( x, y) as the
liquid-vapor interface across which there is a density difference ∆ρ,
and recall that the energy density for a column of fluid of height h is
∆ρgh2 /2, then the energy functional is
Z Z
γ h i
E= dxdy (∇h)2 + 2ℓ−
c
2 2
h . (3.27)
2
p
where ℓc = 2∆ρg/γ is the capillary length (for water, ℓc ∼ 5 mm).
Now we expand the surface deformation as
h (x) = ∑ e−iq·x ĥ(q), (3.28)

q
and obtain
γA
E=
2 ∑(q2 + 2ℓ−c 2 )|ĥ(q)|2 , (3.29)
q
where A = L2 is the surface area. Again by equipartition we find

kB T 1
⟨|ĥ(q)|2 ⟩ = − 2
. (3.30)
γA 2ℓc + q2
This shows that the capillary length provides a cutoff on what would
otherwise be divergent fluctuation amplitudes as q → 0.
Introducing both small-scale and large-scale cutoffs 2π/a and 2π/L,
and assuming a circular geometry in Fourier space, the average vari-
ance of the displacement field is
" #
kB T qdq kB T 1 + 2 (π ℓc /a)2
Z
2
⟨h ⟩ = = ln . (3.31)
2πγ 2ℓ − 2
c +q
2 4πγ 1 + 2 (π ℓc /L)2
The thermodynamic limit L → ∞ is now possible at finite g, with
" 2 #
k B T π ℓ c
⟨ h2 ⟩ ∼ ln 1 + 2 . (3.32)
4πγ a
3.3 Brownian Motion and Diffusion

3
Brown, R. XXVII. A brief account of
Now we start to consider the dynamical properties of objects subject microscopical observations made in the
to thermal fluctuations. The simplest example are microspheres that months of June, July, and August, 1827,
on the particles contained in the pollen
execute Brownian motion (named after the botanist Robert Brown, of plants; and on the general existence
whose 1828 paper3 and a follow-up in 18294 are two of the great- of active molecules in organic and in-
est of all papers in the old scientific literature). We introduce and organic bodies. Philos. Mag., 4:161–73,
1828
study the simplest examples of stochastic ODEs to describe this pro- 4
Brown, R. XXIV. Additional remarks
cess. These "Langevin equations" have random forces that represent on active molecules. Philos. Mag., 6:
161–66, 1829
the effects of collisions with water molecules, and have an amplitude
that is determined by appealing to equilibrium statistical physics ar-
guments in suitable limits.
A convenient starting point for the description of Brownian mo-
tion is the dynamics of a microsphere that is acted on by an optical 5
Ashkin, A. Acceleration and trapping
of particles by radiation pressure. Phys.
trap and subject to thermal fluctuations. First invented in the 1970s Rev. Lett., 24:156–159, 1970
at Bell Laboratories,5 and extended later,6 optical trapping (or "laser 6
Ashkin, A. Optical trapping and ma-
tweezers" 7 ) localises a particle near the waist of a tightly-focused nipulation of neutral particles using
lasers. Proc. Natl. Acad. Sci. USA, 94:
laser beam through gradients in the electric field. This is a conse- 4853–4860, 1997
quence of the general result that dielectric objects experience a force 7
Neuman, K.C. and Block, S.M. Opti-
moving them towards regions of higher electric field. In this case, the cal trapping. Rev. Sci. Instrum., 75:2787–
2809, 2004
microspheres that are typically used have higher dielectric constants
than water at optical frequencies, but lower values than water at zero
frequency. In addition to gradient forces, radiation pressure acts to
push the particle along the optical path, so only for sufficiently high
gradients is the particle trapped. This requires microscope objectives
with high Numerical Aperture (NA) (those with a very short focal
length relative to their diameter).
Close enough to the trap centre, the effective potential experienced
by the particle is quadratic in displacement from the centre. If we
restrict consideration to motion in one lateral dimension the equation
of motion of a trapped microsphere is the Langevin equation
ζ ẋ = −kx + ξ (t), (3.33)

where ζ = 6πµa is the Stokes drag coefficient and ξ (t) is the ran-
dom force. For µm sized spheres and moderate laser power, k ∼ 10
fN/nm, so that when the particle is displacement is a fraction of the
particle size (1 µm = 103 nm), the trapping force will be in the pN
range. From (3.33) we see that there is a relaxation time
ζ
τ= . (3.34)
k
If a ∼ 1 µm and k is as above, τ ∼ 2 ms.
It is important to understand the meaning of the random force
ξ (t). First, note that the typical velocity of a water molecule is
√
found by equipartition to be 3k B T/m ∼ 600 m/s, and hence it takes
only about 3 × 10−13 s for a water molecule to traverse its own size
(0.2 nm), which in turn implies that the collision rate of molecules
with a test particle is perhaps 1020 s−1 . These time scales are vastly
shorter than anything we measure when tracking the microsphere’s
motion, and it is thus natural to exploit this separation of time scales
in the analysis. A second point regards what it means to "solve" a
stochastic ODE such as (5.41). As we shall see below, for any par-
ticular realisation of the random noise ξ (t) the Langevin equation is
simply a linear ODE with a time-dependent force. Yet, in the typical
experiment we perform averages over many stochastic trajectories
of the particle to arrive at the average behaviour. This process cor-
responds to averaging over the realisations of ξ (t). Denoting such
averages with angular brackets, we assume ⟨ξ ⟩ = 0 by symmetry.
To see these issues in action, first simplify (3.33) to
ẋ + τ −1 x = η (t), (3.35)
where η = ξ/ζ. Assuming the initial condition x (0) = x0 , use of an

appropriate integrating factor yields the solution
Z t
′
x (t) = x0 e−t/τ + dt′ e−(t−t )/τ η (t′ ). (3.36)
0
If we average this result over realisations of the noise, then the deter-
ministic term is unchanged, while the integral term vanishes by the
assumed zero value of the mean of the noise, leaving
⟨ x (t)⟩ = x0 e−t/τ . (3.37)
Thus, on average the particle’s trajectory is the deterministic one,

relaxing exponentially to the trap centre.
After the mean position of the particle loses memory of its ini-
tial condition, it still fluctuates in the harmonic trapping potential,
exhibiting a nonzero variance that follows from (3.36) as
Z t Z t
′ ′′ )/τ
⟨( x (t) − x0 e−t/τ )2 ⟩ = dt′ dt′′ e−(2t−t −t ⟨η (t′ )η (t′′ )⟩. (3.38)
0 0
Now we make use of the separation of time scales in the Langevin

approach, and assume that that the correlation ⟨η (t′ )η (t′′ )⟩ inside the
integral is a sharply-peaked function of the difference |t′ − t′′ |, decay-
ing much faster than any relevant timescale of the particle. Calling
this function ϕ(t′ − t′′ ), making the change of variables s = t′ + t′′

and q = t′ − t′′ (with Jacobian 1/2), we obtain the r.h.s. of (3.38) as
Z 2t Z ∞
1 −2t/τ
e dses/τ dqϕ(q), (3.39)
2 0
| −∞ {z }
Γ
where the final integral is just a number we label Γ. The variance of

interest is then
Γτ
⟨( x (t) − x0 e−t/τ )2 ⟩ = (1 − e−2t/τ ). (3.40)
2
In the long time limit (t/τ → ∞), the result simplifies to
Γτ
⟨ x ( t )2 ⟩ = . (3.41)
2
Here we appeal to equipartition, since the linear force law (3.33)
arises from an energy kx2 /2, which in turn implies ⟨ x2 ⟩ = k B T, so
2k B T
Γ= . (3.42)
ζ
A further test of the result is found by examining the result (3.40)

in the short-time limit t/τ ≪ 1. If we assume for convenience that
x0 = 0, then
2k T
⟨ x2 (t)⟩ ≃ B t + · · · . (3.43)
ζ
This is the expected result for a one-dimensional random walk, with
diffusion constant D, where ⟨ x2 ⟩ = 2Dt. We therefore deduce
kB T
D= (3.44)
ζ
and the Stokes-Einstein relation is recovered. An illuminating calcu-

lation in the Example Sheet involves the Langevin description of a
free Brownian particle with inertia.
Looking back at the calculation above it is clear that the separation
of scales between the noise and the dynamical variable was key to
implementing the connection with equipartition. Since the integral
Γ of the noise correlation is the only quantity that was needed, it is
possible to assume that the noise is delta-function correlated in time.
This leads to the standard assumptions on the Langevin noise in the
original form (3.33) or the scaled dynamics (3.35),
⟨ξ ⟩ = 0, ⟨ξ (t)ξ (t′ )⟩ = 2Dδ(t − t′ ), (3.45a)

⟨η ⟩ = 0, ⟨η (t)η (t′ )⟩ = 2k B Tδ(t − t′ ). (3.45b)
3.4 Polymers and Entropic Forces
In this section we will develop basic concepts of polymer physics,

with emphasis on the concept of an entropic spring. Polymers are
structures composed of repeating subunits held together with cohe-

sive forces that may be of various types. In the case of DNA there
are true covalent bonds between the nucleotides, while in the case
of microtubules, non-covalent bonds hold together the individual
protein subunits and the whole structure can grow and shrink over
time through reversible polymerisation. Polymers can be linear se-
quences, as in the case of proteins, or branched or even crosslinked
to form networks. We consider a few of the simplest models of linear
polymers.
3.4.1 Freely-hinged chain

The simplest model of a polymer has N identical links of length b, b
each of which can be oriented in one of two directions. To include
the effects of an external force, we imagine the polymer hangs from
a support in a gravitational field and a mass m is attached to the last
segment (Fig. 3.3). Note that there is no resistance to bending in this g
model, and indeed there are no internal forces at all.
With a reference energy at zero extension, the potential energy of
the freely-hinged chain is E = −mgz, where the extension z is
m
N Figure 3.3: A freely-hinged chain, hang-
z= ∑ bsn , (3.46) ing under gravity from a support.
n =1
and sn = ±1 denotes the orientation of the nth link. The configura-

tional partition function is
!
N
Z= ∑ ··· ∑ exp βFb ∑ sn , (3.47)
s1 =±1 s N =±1 n =1
where F = mg is the gravitational force on each link. By the inde-

pendence of the sn , this can can be simplified to
N
Z= ∏ ∑ e f sn = [2 cosh( f )] N , (3.48)
n=1 sn =±1
where f = βFb. From this we calculate the mean extension as a

function of the force, using Z as a generating function,
∂ ln Zn
⟨z⟩ = ⟨b ∑ sn ⟩ = b = L tanh( f ), (3.49)
n ∂f
where L = Nb is the fully-extended length. Figure 3.4 shows the
normalized extension ⟨z⟩/L versus f , revealing a linear regime at
small f and saturation for f ≫ 1. Expanding (3.49) for small f yields
Nb2
⟨z⟩ ≃ F+··· , (3.50)
kB T
the linear relationship of a Hookean spring with spring constant
kB T
k= . (3.51)
Nb2
The appearance of k B T reminds us that this is an entropic spring; the
resistance to extension arises because of the loss of entropy in the
extended state. Indeed, in the fully-extended state there is only one
configuration possible (all sn = +1), and hence zero entropy.
1 Figure 3.4: Force-extension curves for

the freely-hinged chain (green) and the
freely-jointed chain (blue). Obtained
with Figure_304_polymers.m.
hzi=L
0.5
0
0 1 2 3 4 5
f
3.4.2 The freely-jointed chain

It is natural to ask whether the qualitative features of the free-hinged
chain are found in more realistic models of polymers. To answer b
this, we consider the freely-jointed chain in which each link may rotate θ
around the previous one on a cone of angle 2θ (Fig. 3.5). Now the
g
extension is
N
z= ∑ b cos θn , (3.52)
n =1
with partition function m
Figure 3.5: The freely-jointed chain.
N Z π
Z = ∏ 2π
0
dθn sin θn e f cos θn
,
n =1
N
4π sinh( f )
= . (3.53)
f
The average again follows by differentiation, and has the simple form
1
⟨z⟩ = LL( f ) where L( f ) = coth( f ) − . (3.54)
f
Here, L is the so-called Langevin function well-known in the the-
ory of magnetism. Figure 3.4 compares this functional relationship
with that of the freely-hinged chain, showing qualitative consistency.
There is again a linear relation at small forces, with spring constant
3k B T
k= . (3.55)
Nb2
The larger entropic spring constant reflects the additional degrees of
freedom of the chain.
3.5 Polymer statistics and random walks
3.5.1 General formulation

8
M. Doi and S.F. Edwards. The Theory
Next we study a more general approach to polymer statistics, follow- of Polymer Dynamics. Clarendon Press,
ing the discussion in Doi and Edwards.8 Consider an arbitrary free Oxford, UK, 1986
polymer of segment length b, where rn (n = 1, . . . , N) is the position

of the start of each segment and
rn +1 = rn + ζ n , (3.56)
with ζ n a random displacement of fixed length (|ζ n | = b). The "end-

to-end" displacement R of the polymer,
N
R ≡ r N − r0 = ∑ ζn (3.57)
n =1
is a quantity of great interest. If, as usual, we denote an average over

realisations of random variables (in this case, the {ζ n }), then
N
⟨R⟩ = ∑ ⟨ζ n ⟩ = 0 (3.58)
n =1
by symmetry. The average second moment can be calculated in the

usual way as
N N
⟨ R2 ⟩ = ∑ ∑ ⟨ζ m · ζ n ⟩ = Nb2 , (3.59)
m =1 n =1
where we have made us of the relation ⟨ζ m · ζ n ⟩ = δmn b2 that holds

for the independently distributed random variables {ζ m }. The simi-
larity with the Langevin formalism is apparent.
Let us formulate the problem more generally, without the spe-
cific requirement that the segments be of a fixed length. We start by
defining the probability that a polymer with N segments will have
segment positions at {rk } as
1
P({rk }) = G ({rk }) with G = e− βU ({rk }) . (3.60)
Z
The energy is typically a sum of near-neighbor interactions and a
contribution from an external potential,
N
U ({rk }) = ∑ Uj (rj−1 , rj ) + W ({rk }). (3.61)
j =1
When W = 0, this is just a "random flight" model. Now define the

probability distribution p j (R j ) of segment lengths
p(R j ) = exp[− βUj (R j )] where R j = r j − r j −1 , (3.62)

R
and assume it is normalized ( dRp(R) = 1).
The fixed end-to-end-vector partition function G is defined as the
integral over all degrees of freedom of the internal segments such
that the end position is R, assuming for convenience that the start of
the first segment is at the origin,
Z
G(R; N ) = dR N G ({Rk })δ(r N − R),
!
Z N N
= dR N ∏ p(R j )δ ∑ Rk − R , (3.63)
j =1 k =1
where the notation is that of (2.4), namely dR N = dR1 dR2 · · · dR N .

Now we use an integral representation of a delta function,
d3 k ik·P
Z
δ (P) = e , (3.64)
(2π )3
to express G for a system of identical segments as
d3 k −ik·R
Z
G= e K (k; N ), (3.65)
(2π )3
where the so-called "characteristic function" is
Z N
K (k; N ) = dR p(R)eik·R . (3.66)
As an example, consider a fixed-length segment, for which

1
p (R) = δ (|R| − b) , (3.67)
4πb2
and hence N
sin(kb)
K (k; N ) = . (3.68)
kb
In the limit of large N, it is the small-k behaviour that is important,
and if we expand K we find
N
k 2 b2

K (k; N ) ≈ 1− +... ∼ exp(− Nk2 b2 /6). (3.69)
6
For more general distributions, we will obtain the form in 3.69, but
with b having the interpretation of an appropriate second moment
of p(R). Inverse Fourier transforming in (3.65), we obtain
3/2
−3R2

3
G(R; N ) = exp . (3.70)
2πb2 N 2Nb2
This Gaussian distribution is the central feature of ideal random
walks, and is not restricted to a fixed-length distribution p. It is
essentially a consequence of the central limit theorem that requires
only a finite second moment of the underlying distribution.
3.5.2 Flory theory and self-avoidance

We may interpret the argument of the exponential in (3.70) as an
effective (entropic) free energy,
3R2
Fentropic ( R) = −k B T ln G( R; N ) = k B T . (3.71)
2Nb2
This has the form of a (Hookean) entropic spring, with a minimum at
R = 0 reflecting the fact that stretching the chain lowers its entropy.
In a root-mean-square sense, the typical end-to-end distance obeys
the scaling form
R ∼ Nν, (3.72)
with ν = 1/2 as expected for an ideal random walk. Let us now
consider how excluded-volume interactions change this free energy,
adopting the scaling point of view first introduced by Flory9 and 9

P.J. Flory. Principles of Polymer Chem-
extended by many others.10 The effects of self-avoidance are intro- istry. Cornell University Press, Ithaca,
NY, 1953
duced by means of a short-ranged (indeed, δ function) interaction 10
P.G. de Gennes. Scaling Concepts in
between segments along the chain of the form Polymer Physics. Cornell University
Press, Ithaca, NY, 1979
Z N Z N
1
vk B T dn dm δ(Rn − Rm ), (3.73)
2 0 0
where for convenience we have included k B T in the prefactor.

Using a mean-field argument like that used to calculate the in-
ternal energy of a gas in the approach of van der Waals (Eq. 2.10),
we estimate the contribution of these interactions to the free energy
using the local segment concentration N/R3 ,
N vN 2
vk B T · N · ∼ k B T . (3.74)
R3 R3
The total free energy in this Flory theory is thus
3R2 vN 2

F ( R) = k B T + . (3.75)
2N ℓ2 R3
We see a competition between entropy, which favours the smallest R,

and excluded-volume effects that tend to swell the chain. Differenti-
ating to find the optimum, R∗ , we obtain
R∗ ∼ N 3/5 . (3.76)
As the exponent ν = 3/5 > 1/2, excluded-volume interactions have

swollen the chain from its ideal size (3.72).
It is instructive to generalise this to arbitrary spatial dimension d.
The only change needed to (3.75) is the form of the local concentra-
tion in the excluded volume interaction, which becomes N/Rd . The
new balance of terms is
R2 N2
∼ d, (3.77)
N R
which leads immediately to the predicted exponent
3
ν= . (3.78)
d+2
These are compared to the known values in Table 3.1. The case of
d = 1 is trivial, as such a self-avoiding chain must be fully stretched
out, while for d > 4 the probability of a chain contacting itself is so
low as to leave the Gaussian scaling unchanged. Apart from the case
d > 4 we see that Flory theory is remarkably accurate.
d νFlory νexact comments

1 1 1 exact result
2 3/4 3/4 exact solution
3 3/5 0.589 . . . numerical
4 1/2 1/2 upper critical dimension
>4 3/(d + 2) 1/2 excluded volume irrelevant
Table 3.1: Scaling exponent for self-
avoiding walks in d dimensions.
4
Biological Filaments (Lectures 8-12)
Living systems are replete with filaments, from the polymers such
as actin and microtubules that comprise the cytoskeleton to cilia and
flagella that power the motility of microorganisms. At larger scales
we find slender, undulating organisms such as worms and snakes. In
this chapter we discuss the basic differential geometry of filaments
in two and three dimensions, their elasticity and principles by which
their motion can be understood. We also discuss some key insta- 1
Powers, T.R. Dynamics of filaments
bilities that are relevant in biological phenomena. Aspects of the and membranes in a viscous fluid. Rev.
analyses presented here find application in the theory of pattern for- Mod. Phys., 82:1607–1631, 2010
mation, discussed in Chapter 7. There are many excellent reviews on 2
S. Lipschutz. Differential Geometry.
the subject,1 and concise compendia on differential geometry.2 Schaum’s Outline Series. McGraw-Hill
Education, New York, 1969
4.1 Essentials of differential geometry of curves
We start the basic differential geometry of curves in the plane. Con-

sider a curve r(α) in a plane, parameterized by α ∈ [0, 1] (Fig. 4.1).
For generality, the range of α is taken to be fixed; it can be any mono-
tonically increasing label along the curve, such as the arclength s nor- r(α)
malized by curve length L. Later we will specialise to the arclength
representation where appropriate. Let rα ≡ ∂r/∂α be the unnormal-
ized tangent vector, and introduce the metric 0 α 1
Figure 4.1: A curve in the plane. The
g = rα · rα . (4.1) vector r(α) is a map between the unit
interval and the blue curve.
Then the differential of arclength is
√
ds = |dr| = gdα, (4.2)
t
and the unit tangent to the curve is
θ n
rα
t̂ = √ . (4.3)
g
The normal vector n̂ at each point is orthogonal to the tangent,

and its components are given by Figure 4.2: The unit tangent and normal
vectors to a curve, and the tangent an-
n̂i = ϵij t̂ j , (4.4) gle θ defined with respect to the x axis.
where ϵij is the Levi-Civita symbol (ϵ12 = +1, ϵ21 = −1, ϵ11 = ϵ22 =
0). The pair (t̂, n̂) is a local coordinate system (Fig. 4.2) that rotates
as it moves along the curve according to the Frenet-Serret equations

! ! !
∂ t̂ 0 −κ t̂
= , (4.5)
∂s n̂ κ 0 n̂
where the curvature κ can be positive or negative. Adopting the

convention in Fig. 4.2, where the tangent angle θ is with respect to
the x axis, we have n̂ = cos θex + sin θey and t̂ = − sin θex + cos θey ,
from which we find
∂θ
κ= , (4.6)
∂s
where, in general,
∂ 1 ∂
= √ . (4.7)
∂s g ∂α
In the "Monge representation", where we have a single-valued
function h( x ) as a function of the external coordinate x, the posi-
p p
tion vector is r( x ) = ( x, h), the tangent is (1/ 1 + h2x , h x / 1 + h2x ),
and θ = tan−1 h x , and we have the familiar result
h xx
κ= . (4.8)
(1 + h2x )3/2
With the formulation above, we can calculate two elementary geo-
metric features of curves, their length L and the area A they enclose,
if closed. While we will often deal with open curves (representing,
say, elastic filaments), there are times when closed curves are of in-
terest, whether an actual closed loop of DNA or an interface between
two states in a reaction-diffusion system, or a simplified model of a
cell membrane in two dimensions. The length is trivially
Z 1
√
L= dα g, (4.9)
0
while the area is Z 1

1
A= dαr × rα , (4.10)
2 0
where in two dimensions the cross product is the scalar (a × b) =
ϵij ai b j . The description of space curves is quite similar to the planar
case, only now there is an orthonormal triad of the tangent t̂, normal
n̂ and binormal b̂ = t̂ × n̂ that evolves with arclength as
    
t̂ 0 κ 0 t̂
∂   
n̂ = −κ 0 τ  n̂ , (4.11)
 
∂s
b̂ 0 −τ 0 b̂
where now the curvature κ is strictly ≥ 0 and τ is the torsion.
4.2 Overdamped filament motion
Many of the problems of interest in the biological physics of fila-

ments take place at the cellular scale. If we formulate a Reynolds
number for these problems,
UL
Re = , (4.12)
ν
biological filaments (lectures 8-12) 57
where U is a typical fluid speed, L is a typical length scale for the

motion, and the kinematic viscosity of water ν ∼ 106 µm2 /s (Table
1.1), then with U ∼ 1 − 100µm/s and L ∼ 1 − 100µm, we have Re ∼
10−6 − 10−2 , which is well in the Stokesian regime where inertia can
be neglected. It follows that the equations of motion for filaments
should be first order in temporal derivatives, reflecting a balance of
viscous and elastic forces.
4.2.1 A variational principle

It is pedagogically useful to formulate the dynamics through a varia-
tional principle, as this provides a way to connect forces to an under-
lying energy functional, and to assure that the dynamics has an ap-
propriate reparameterisation invariance. With this in mind, we first
review the manner in which dissipative contributions can be intro-
duced into the usual Lagrangian formulation of classical dynamics.3 3
H. Goldstein. Classical Mechanics.
Consider a system specified by a single coordinate q and its time Addison-Wesley, Boston, MA, 2nd edi-
tion, 1980
derivative q̇. In the usual manner, the Lagrangian is
L(q, q̇) = T − V , (4.13)
where T and V are the kinetic and potential energies. From the
R
action S = dtL, the minimum action principle δS = 0, yields the
Newtonian equation of motion
d ∂L ∂L
− = 0. (4.14)
dt ∂q̇ ∂q
With T = mq̇2 /2 the Euler-Lagrange equation becomes
∂V
mq̈ = − . (4.15)
∂q
In the limit of vanishing inertia (Re = 0), simply dropping the kinetic
energy term leaves us with no dynamics, because dissipative forces
are not captured in either T or V . Instead, we must introduce a gener-
alised force associated with dissipation. This quantity is related to the
rate at which viscous forces do work. For example, if, as in the case
of a sphere moving through a viscous fluid, the viscous force is ζ q̇,
where ζ is a drag coefficient, the rate of dissipation (force×velocity)
is ζ q̇2 . Thus, we introduce the Rayleigh dissipation function
1 2
R= ζ q̇ , (4.16)
2
proportional to the dissipation rate, and obtain the new variational
principle in which the derivative of R with respect to velocity is a
generalised force,
d ∂L ∂L ∂R
− =− . (4.17)
dt ∂q̇ ∂q ∂q̇
In the overdamped limit, we then have the Aristotelian law
∂V
ζ q̇ = − . (4.18)
∂q
Now we ask how to generalise these ideas to moving curves. We

start by formulating the dissipation rate. The simplest generalization
of the Stokes drag on a sphere to a moving curve involves a local,
isotropic drag coefficient ζ, so that
1 √
Z
R= ζ dα g r2t , (4.19)
2
√
and the generalized viscous force is −δR/δrt = −ζ g rt , and thus
(renaming V as E , the energy functional of the curve or filament), the
equation of motion for isotropic drag is
1 δE
ζrt = − √ , (4.20)
g δr
where the r.h.s. is a functional derivative. The main result of hav-

ing formulated the dissipation rate in a reparameterisation invariant
manner is the metric factor in 4.20.
For a long, slender object of length L and radius a in three dimen-
sions, the calculation of the drag is a complicated nonlocal problem,
but often the dominant behaviour is well-described by Resistive Force
Theory4 (RFT) in which there are local drag coefficients ζ ⊥ and ζ ∥ for 4
J. Gray and G.J. Hancock. The propul-
motion perpendicular and parallel to the filament, and ζ r for rota- sion of sea-urchin spermatozoa. J. Exp.
Biol., 32:802, 1955
tional motion. These can be understood most simply by appealing
to the known drag coefficients for prolate ellipsoids.5 If the ellipsoid 5
Oberbeck, A. Ueber stationäre Flüs-
has length L and radius a, then if L/a ≫ 1 the drag coefficients are sigkeitsbewegungen mit Berücksichti-
gung der inneren Reibung. J. Reine.
2πµ 4πµ Angew. Math., 81:62–80, 1876
ζ∥ = , ζ⊥ = (4.21)
ln( L/a) − 1/2 ln( L/a) + 1/2
For many applications it is sufficiently accurate to assume ζ ⊥ = 2ζ ∥ .
These can also be derived in a systematic manner through an asymp-
totic expansion in the aspect ratio of the filament.6 The rate of energy 6
Keller, J.B. and Rubinow, S.I. Slender-
dissipation is a simple generalisation of 4.19, body theory for slow viscous flow. J.
Fluid Mech., 75:705–714, 1976
√
2
1
Z 2
2
R= dα g ζ ∥ t̂ · rt + ζ ⊥ (n̂ · rt ) + b̂ · rt , (4.22)
2
and the general filament equation of motion becomes
1 δE
ζ ∥ t̂t̂ + ζ ⊥ I − t̂t̂ · rt = − √

, (4.23)
g δr
where I = t̂t̂ + n̂n̂ + b̂b̂ is the unit dyad. If we use the asymptotic
result ζ ⊥ = 2ζ ∥ , then the projection operator on the l.h.s. of 4.23 is
proportional to I − t̂t̂/2. The inverse of this operator is
−1
1
I − t̂t̂ = I + t̂t̂, (4.24)
2
and therefore we can write the equation of motion 4.23 as
1 1 δE
rt = − I + t̂t̂ · √ . (4.25)
ζ⊥ g δr
The above analysis defines the dynamics of the filament centreline,

but does not address any twisting deformations that can occur. For
long, slender filaments the balance of forces for a filament rotating

at angular frequency ω about its long axis has the simple form
m = ζ r ω, (4.26)
where m is the moment and ζ r = 4πµa2 is the rotational drag coeffi-

cient. We shall discuss this at greater length below.
4.2.2 Elasticity
In this section we use the general formulation developed above to
study filaments governed by the elastic energy7 7
L.D. Landau and E.M. Lifshitz. Theory
of Elasticity. Butterworth-Heinemann,
Z L
1 Oxford, UK, 3rd edition, 1986
E= A ds κ 2 , (4.27)
2 0
where A is the bending modulus. This energy penalizes bending

in any direction away from a straight filament, which is the ground
state. As in our discussion of membrane elasticity, it is useful to ex-
press A in terms of thermal energy. Since it has units of energy·length,
we can write A = k B TL p , where L p is the persistence length. The
persistence length has a simple physical meaning; if a filament has
length L p and is bent on the scale of its own length, then the energetic
cost is thermal energy, since E ∼ A · L p · 1/L2p ∼ A/L p ∼ k B T.
filament Lp radius length L/L p

DNA 50 nm 1 nm nm-m 10−2 − 107
actin 10 − 15 µm 3.5 nm 1 − 103 µm 0.1 − 102
microtubules 5 mm 12 nm 1 − 100 µm 10−4 − 10−2
Table 4.1: The persistence lengths of
Filaments governed by the elastic energy 4.27 are termed "semi- various biological filaments.
flexible", and are found throughout the cell. Table 4.1 summarizes
the three most important examples. We shall not delve into the de- extension
tailed derivation of 4.27 here, but note that resistance to bending
fundamentally arises from the resistance to stretching. As shown in
Fig. 4.3, the outer edge of a bent filament is under extension and the compression
inner edge is compressed. The resistance to stretching or compres-
Figure 4.3: A bent filament viewed in
sion is quantified by the Young’s modulus Y as the proportionality longitudinal cross section. Red dashed
between the force per unit area applied to an object and its fractional line indicates neutral plane.
change in length,
force ∆L
=Y . (4.28)
area L
Y clearly has units of energy/volume, and by comparison to the
modulus we deduce that A ∼ Y · length4 . As the only length scale of
relevance for a filament is its radius a, we deduce that
A ∼ Ya4 . (4.29)
If we take as an example the case of microtubules, we deduce Y ∼

109 Pa, which is similar to that of plastics.
In order to derive the filament dynamics we need the functional
derivative of the energy 4.27. This functional is of the general form
Z
E= dα f (r, rα , rαα , . . .). (4.30)
√
The quantities rα and rαα show up in the metric factor g and the
elastic term, which can be written in a covariant manner as
2
1 rα
κ 2 = √ ∂α √ . (4.31)
g g
The functional derivative (or “Fréchet" derivative) of 4.30 is
δE ∂f ∂ ∂f ∂2 ∂ f
= − + 2 +··· , (4.32)
δr ∂r ∂α ∂rα ∂α ∂rαα
where each successive term arises from one additional integration by
parts in the usual manner.
For the remainder of our discussion we will confine ourselves to
curves in the plane. A straightforward (if tedious) calculation of this
to 4.27 for such filaments yields the result
1 δE

1 3
−√ = A κss + κ n̂. (4.33)
g δr 2
Note that this force is purely normal to the filament. As the curva-
ture involves two derivatives of position, κss involves four; this is a
hallmark of elasticity that is distinct from the effects of tension (as
in a violin string), which produces a restoring force proportional to
the curvature itself. One might imagine that with an energy that
is (curvature)2 the fourth derivative would be the only term in the
force, but this is not the case. The nonlinear term κ 3 appears even
though this theory is based fundamentally on linear elasticity the-
ory. It is a geometric nonlinearity that arises from the fact that the
coordinate r and arclength s are not independent. And it is intuitive
physically, if we think of a closed circular loop, where the curvature
is constant (κss = 0) and yet there is uniform outward force which
would only be relaxed at vanishing curvature (infinite radius).
A small technical aside is needed here. It is tempting (and fre-
quently done in the literature) to adopt a simplified picture in which
the energy (4.27) as a functional of the quantity r(s), ignoring the fact
that s and r are not independent. In this calculation, we simply write
1
Z
E [r] = A ds r2ss , (4.34)
2
and deduce the force

∂ 1 2
−∂ss Ar = − Arssss . (4.35)
∂rss 2 ss
A simple calculation using the Frenet-Serret equations shows that

− Arssss = A κss − κ 3 n̂ + 3Aκκs t̂, (4.36)
which differs from (4.33) in the coefficient of the cubic term in κ

and the presence of a tangential component. But those extra terms
comprise a total derivative,
1 δE

3 2
−√ = − Arssss − ∂s Aκ t̂ . (4.37)
g δr 2
We shall now see that the difference between the two calculations
represents a tension-like contribution.
If we were to use E directly as our energy functional, the dynamics
will not necessarily preserve the local arclength along the filament.
To enforce local inextensibility we introduce a Lagrange multiplier
function Λ(α) that represents the internal tension,8 and define 8
Goldstein, R.E. and Langer, S.A. Non-
linear dynamics of stiff polymers. Phys.
Z 1
√ Rev. Lett., 75:1094–1097, 1995
F =E− dα gΛ(α). (4.38)
0
The functional derivative of F is

1 δF 1 δE
− Λt̂ s

−√ = −√
g δr g δr

1
= A κss + κ 2 n̂ + Λκ n̂ − Λs t̂. (4.39)
2
The new terms involving Λ can be interpreted as a Young-Laplace

force in the normal direction (tension × curvature) and a tangential
"Marangoni" force (gradient of tension).
In order to find Λ(s), we appeal to some general results for the
dynamics of curves in the plane. The equation of motion 4.25 for
filaments is of the form
rt = U n̂ + W t̂, (4.40)
where U = U (r, κ, . . .) and W = W (r, κ, . . .) are the normal and tan-

gential velocities. If the filament is locally inextensible, then the metric
√
must be time-independent. Computing the time derivative of ln g
to highlight the reparameterisation invariance, and written out to
emphasise that α is time-independent (rαt = rtα ), we have
√
∂ ln g rα rαt
= √ · √ = t̂ · ∂s rt = Ws + κU, (4.41)
∂t g g
and therefore the incompressibility constraint is Ws = −κU. From

our filament equation we deduce that

1 1 2Λs
U= A κss + κ 3 + Λκ , W = − , (4.42)
ζ⊥ 2 ζ⊥
where the factor of 2 in W arises from the drag coefficient ratio.

From 4.41 and 4.42 we find that Λ obeys an inhomogeneous elliptic
second-order ODE at each instant of time,

1 2 A 1 3
∂ss − κ Λ = κ κss + κ . (4.43)
2 2 2
We now see that the difference between the “proper" (4.33) and sim-
plified (4.36) forces is a redefinition of the tension.
The general dynamics 4.40 is an intrinsic equation of motion, and it
is of interest to the time evolution of geometric quantities such as the
tangent angle θ and the curvature κ. As these are scalar quantities
they are often easier to study (both analytically and numerically),
particularly when a filament is free in space and its absolute position
(and possibly its orientation) are unimportant. The tangent vector

evolves as ∂t̂/∂t = (Us − κW )n̂ and therefore with t̂ and n̂ defined
above 4.6, we have ∂t̂/∂t = −θt n̂, and thus
θt = −Us + θs W. (4.44)
A bit more work shows that the curvature obeys the PDE

κt = − ∂ss + κ 2 U + κs W . (4.45)
In both (4.44) and (4.45) we see that W enters as an advective term.
4.2.3 Boundary conditions and biharmonic eigenfunctions

A final general point to make in discussing filaments is the bound-
ary conditions at their ends. While in some problems there are obvi-
ous forces or torques applied at the filament ends, the case of "free"
ends with vanishing force and torque are extremely common and de-
serve attention. It is easiest to think about these problems when the
filament is only weakly displaced from straight, where the Monge
representation holds and we can approximate the elastic energy as
Z L
A
E≃ dx h2xx . (4.46)
2 0
To find the boundary conditions we compute the change δE when

h → h + δh, paying close attention to the surface terms that arise by
repeated integration by parts

RL L RL
δE = A 0 dxh xx δh xx = A h xx δh x − 0 dxh3x δh x
0

L L RL
= A h xx δh x −h xxx δh + 0 dxh4x δh . (4.47)
0 0
In the usual way, the functional derivative of E is just the coefficient

of δh inside the final integral,
δE
= Ah4x . (4.48)
δh
If the surface terms are to vanish for arbitrary δh and δh x , then
h xx (0) = h xx ( L) = 0, and h xxx (0) = h xxx ( L) = 0. (4.49)
These are the boundary conditions of a free filament. The quantity 9

Wiggins, C.H., Riveline, D., Ott, A.
and Goldstein, R.E. Trapping and wig-
Ah xx is a torque and Ah xxx is a force; the free ends are torque-free gling: elastohydrodynamics of driven
and force-free. When these (or related9 ) conditions hold at both ends, microfilaments. Biophys. J., 74:1043–
the operator ∂4x is self-adjoint operator and its eigenvalues are real. 1060, 1998
Its biharmonic eigenfunctions W satisfy an equation of the form
AW4x = k4 W, (4.50)
just as sin qx and cos qx are eigenfunctions satisfying f xx = −q2 f .

For the "free-free" case (both ends free), a simple superposition of
sin kx and cos kx will not satisfy boundary conditions that require
successive derivatives to vanish. Instead, the functions are
W ( x ) = A sin kx + B cos kx + D sinh kx + E cosh kx. (4.51)

Computing the derivatives Wxx (0) and Wxxx we deduce that E = B

and D = A. Then the problem reduces to the 2 × 2 system
! ! !
sinh kL − sin kL cosh kL − cos kL A 0
= . (4.52)
cosh kL − cos kL sinh kL + sin kL B 0
Setting the determinant to zero we obtain the condition

1
cos kL = . (4.53)
cosh kL
1 Figure 4.4: Solution of the transcen-

dental relation 4.53. Blue curve is
cos kL, green is 1/ cosh kL, and red
dots represent the numerically ob-
cos kL, 1= cosh kL
0.5 tained crossing points. Obtained with

Figure_404_and_405_biharmonic.m.
-0.5
-1
0 : 2: 3:
kL
The graphical solution to this transcendental relation is shown in

Fig. 4.4. There is an infinite discrete set of solutions kL analogous to
the trigonometric numbers nπ in ordinary Fourier series with Dirich-
let boundary conditions. Beyond the first solution at k = 0, the cross-
ing points are ever closer to the values (2n + 1)π/2 for n ≥ 1. Even
though these values are like those of a pure Fourier series, the eigen-
functions are qualitatively rather different, as seen in Fig. 4.5.
2 Figure 4.5: The first five biharmonic

eigenfunctions Wn ( x ) for free-free
boundary conditions. Obtained with
1 Figure_404_and_405_biharmonic.m.
Wn (x)
-1
-2
0 0.5 1
x=L
Having established a complete set of orthonormal eigenfunctions

of the operator ∂4x , we can now illustrate their utility in the prob-
lem of thermal fluctuations of semiflexible polymers. If the filament
shape h( x ) is expressed as a sum of biharmonic eigenfunctions,
h( x ) = ∑ a n W ( n ) ( x ), (4.54)
n
we can, using repeated integration by parts as in (4.47), rewrite the

bending energy as a sum over those modes,
Z L L
1 1
Z
A dxh2xx = A dxhh4x
2 0 2 0
Z L ∞ ∞
1
= A dx ∑ ∑ am an λn W (m) W (n)
2 0 m =1 n =1
1 ∞
2 n∑
= A λn a2n , (4.55)
=1
where in the last relation λn is the nth eigenvalue, and we have used
the orthonormality the Ws to simplify the result. Once again we
have found an energy expressed as a sum of independent quadratic
contributions, and thus from equipartition we obtain
kB T k T 16L4
⟨ a2n ⟩ = ≃ B , (4.56)
Aλn A (2n + 1)4 π 4
where in the last relation we have used the approximate form of the
roots of the transcendental equation (4.53). The variance is
kB T W ( n ) ( x )2
⟨h2 ( x )⟩ =
A ∑ λn
. (4.57)
n
Figure 4.6: Microfluidic setup to study

An experiment to study the dynamics of semiflexible filaments in single actin filaments. From Ref. 10.
imposed flow fields was done some time ago, using a microfluidic
setup depicted in Fig. 4.6.10 Single actin filaments can be held at 10
Kantsler, V. and Goldstein, R.E. Fluc-
the crossing point of two perpendicular channels meeting at a four- tuations, dynamics, and the stretch-coil
transition of single actin filaments in ex-
fold junction by means of the pressure difference ∆p between the tensional flows. Phys. Rev. Lett., 108:
inlets and outlets. This also allows the flow to be changed from com- 038103, 2012
pressional to extensional for a filament oriented along either of the
orthogonal directions. The vector field show illustrates the hyper-
bolic flow profile that can be achieved. At right is an image of a
fluorescently labelled actin filament and the tracing of its centreline.
The variance of the filament deviation h( x ) from its straight con-
figuration, normalised by its value at the endpoint, is shown in Fig.
4.7 for these experiments with actin, compared with the contribution
Figure 4.7: Normalised variance of fluc-

tuating actin filaments, from Ref. 10..
from the fundamental mode n = 1 in (4.57). The somewhat unusual

shape "W" shape arises from the nodes in the first mode, themselves
an indirect consequence of the free-filament boundary conditions.
4.3 Instabilities and elastohydrodynamics
Here we work through four pedagogically important problems in

elastohydrodynamics that use the formalism described above.
4.3.1 Example 1. Euler Buckling

Our first example is not itself dynamical, but serves to illustrate some
useful ways of analyzing bifurcations in elastic systems. It is the
classic problem of Euler buckling, in which an elastic filament subject
to thrusting forces at its ends deforms at a critical force.
Figure 4.8: Buckling of microtubules

growing inside a synthetic lipid vesicle.
Scale bar is 5 µm . From Ref. 11.
11
Fygenson, D.K., Marko, J.F. and
Libchaber, A. Mechanics of micro-
tubule-based membrane extension.
Inside of cells, microtubules often interact with cell membranes Phys. Rev. Lett., 79:4497–4500, 1997
and deform them. A series of experiments in the 1990s11 provided
some of the nicest illustrations of the forces at work in these con-
texts. Figure 4.8 shows an experimental realization of this instability
involving microtubules grown by inducing polymerisation of tubu-
lin monomers in solution inside of lipid vesicles. As they grew and
the deformed membrane exerts larger and larger thrusting forces on y
their ends, the microtubules buckle.
Our analysis of Euler buckling will be done using the tangent- θ s
angle representation of the curve to illustrate an alternate approach h(x)
to the Monge representation. Using the geometry of Fig. 4.9, the F x
coordinates x (s) and y(s) of a point at arclength s are related to Figure 4.9: Geometry of Euler buckling.
the tangent angle through dx/ds = cos θ and dy/ds = sin θ. If we
take the left end of the filament to be at the origin (x (0) = 0), then
the difference between the fully extended length L and the projected
distance between the endpoints is x ( L) and x (0) is
Z L Z L
dx
L − x ( L) = ds 1 − = ds (1 − cos θ (s)) . (4.58)
0 ds 0
The energy functional for the filament includes the bending energy
and the work done by the force F in displacing the filament ends,
Z L
1 2
E[θ ] = ds Aθs − F (1 − cos θ ) . (4.59)
0 2
The equilibrium condition δE/δθ = 0 is
Aθss + F sin θ = 0, (4.60)
which is the equation of motion of a pendulum in which arclength s

stands for time. We rescale the arclength to α = s/L to obtain
FL2
θαα + f sin θ = 0, with f = . (4.61)
A
Let us consider clamped boundary conditions, θ (0) = θ (1) = 0.
From everyday experience we expect the buckled shape beyond some
threshold force to have a single maximum in the middle of the fila-
ment, with θ = 0 there as well. This suggests a trial function
θ (s) ≃ a sin (2πα) . (4.62)
Expecting that the amplitude of deflection will be small near the

threshold, we linearise (4.61) to obtain
θαα ≃ − f θ. (4.63)
The function (4.62) will only solve this equation provided f = f c ,

where the critical force is
A
f c = 4π 2 , or Fc = 4π 2 . (4.64)
L2
To put this in perspective, consider microtubules, for which L p =
5 mm ∼ 5 × 103 µm. Using k B T = 4 × 10− 3 pN·µm, we have A ∼
20 pN·µm2 . Thus, microtubules whose length is ∼ 20 µm will buckle
under a force of 1 − 2 pN. We return to this estimate later when dis-
cussing the effects of molecular motors on cytoskeletal filaments.
The linearised problem (4.63) gives us no information on the am-
plitude a of the mode that becomes unstable at f c . Near the bifur-
cation, we expect that nonlinearities will stabilize the buckled shape
at some finite amplitude. We can continue with the the single-mode
approximation (4.62) and use a variational approach. First, let us
assume we are close to the threshold force and write
f = f c (1 + ϵ ) , (4.65)
where ϵ is a bifurcation control parameter, and second, expand in

the energy functional (4.59) the force contribution, using cos θ = 1 −
θ 2 /2 + θ 4 /24 + · · · . If we define the dimensionless E = LE/A and

integrate over α we obtain the energy to leading order in ϵ and a,

fc 1 1
E≃ − ϵa2 + a4 + . . . , (4.66) a
2 2 32
where we have ignored an unimportant constant additive term.
This is a Landau theory for the bifurcation, an expansion in powers
of the “order parameter" a, where the coefficient of the quadratic
term can change sign as we increase the thrusting force. When ϵ < 0 0 ε
(below the bifurcation), E has a single minimum as a function of a,
namely a = 0, while above there are two mirror-image minima at
√
a± = ± 8ϵ, (4.67)
Figure 4.10: Pitchfork bifurcation of Eu-
for ϵ ≥ 0. The square root dependence on the control parameter is ler buckling amplitude versus dimen-
the hallmark of a pitchfork bifurcation, as in Fig. 4.10. sionless control parameter ϵ.
2 0.5
(a) (b) Figure 4.11: Euler buckling. (a)
Numerically obtained solutions θ (s)
1
for ϵ = 0.001, 0.01, 0.05, 0.1, 0.2, 0.3, 0.7,
with clamped boundary condi-
3(s)
y=L
0
tions. (b) The corresponding
filament shapes. Obtained with
-1
Figure_411_and_412_Eulerbuckle.m.
-2 0
0 0.5 1 0 0.5 1
s=L x=L
While the general problem (4.61) can be solved analytically, it is

a two-point boundary value problem that can be solved numerically
in Matlab with a simple code, with results shown in Fig. 4.11.
One important lesson from this type of analysis is the accuracy of
the single-mode approximation for the shape (4.62). It is already ap-
parent from Fig. 4.11(a) that θ is qualitatively like the mode sin(2πs/L);
Fig. 4.12 compares the numerically obtained maximum amplitude of
θ (s) with the single-mode result (4.62) obtained near the bifurcation.
The agreement is excellent even for values of ϵ as large as ∼ 0.3.
Figure 4.12: Maxima of θ (s) versus ϵ

from numerics (colored symbols, corre-
2
sponding to curves in Fig. 4.11), com-
pared to analytical approximation near
bifurcation (dashed). Obtained with
3max
Figure_411_and_412_biharmonic.m.
1
0
0 0.5 1
0
After buckling, the deformed rod behaves like a Hookean spring,

whose properties we find by comparing the compressed length to
the applied force. The right end of the filament moves a distance
La2
Z L
1
∆x ≃ ds θ 2 ≃ +··· (4.68)
0 2 4
√
As a = 8ϵ near the bifurcation, we obtain
Fc
F − Fc = ∆x, (4.69)
2L
and thus beyond the buckling the effective spring constant is Fc /2L.
4.3.2 Example 2: Wiggling elastica

Now we move on to the first of three examples of "elastohydrody-
namics". This is the problem of the shape of a filament whose end is
oscillated periodically. The first example of such a calculation is due
to Machin.12 At the time of his calculation, it was not known whether 12
Machin, K.E. Wave propagation along
the observed waveforms of sperm flagella were a consequence of me- flagella. J. Exp. Biol., 35:796–806, 1958
chanical actuation occurring at one end (the head) or whether there

were active forces distributed along the length of the flagella. His
result, derived below, showed clearly that the latter was the case,
even before the actual discovery of the microscopic mechanism. This
13
Wiggins, C.H. and Goldstein, R.E.
calculation served as the basis for much future work, including the Flexive and propulsive dynamics of
self-propulsion13 that arises from such actuated filaments. elastica at low Reynolds number. Phys.
For this problem, it is useful to make a comparison to the classi- Rev. Lett., 80:3879–3882, 1998
cal examples in viscous fluid dynamics known as Stokes Problems I

and II. Outlined by Stokes in his famous paper on viscosity,14 these 14
Stokes, G.G. On the effect of the in-
involve viscous fluids driven by the motion of a boundary parallel to ternal friction of fluids on the motion
of pendulums. Trans. Camb. Philos. Soc.,
itself, in an impulsive or oscillatory manner, as shown in Fig. 4.13. IX, part II:8–106, 1851
Figure 4.13: Stokes and elastohydro-

dynamic problems. Stokes problems
SI and SII (left) involve impulsively-
moved or oscillated walls bounding a
viscous fluid. EHD problems I and II
a filament is allowed to relax from an
initially bent state (or subject to an im-
pulsive transverse flow, or its end is os-
cillated up and down periodically.
For Stokes Problems I and II (SI & SII), we start from the Navier-
Stokes equations for an incompressible fluid with density ρ, viscosity
µ, pressure p and velocity field u,
ρ (ut + u · ∇u) = −∇ p + µ∇2 u, (4.70)
and recognize for wall-driven flow that the nonlinear term vanishes
by symmetry, and we can ignore the pressure gradient. If the wall
moves along êy , then the velocity will be of the form u = u( x, t)êy . It
follows that the function u obeys the diffusion equation
ut = νuyy , (4.71)
where again ν = µ/ρ is the kinematic viscosity.

In SI, we set the wall in motion at t = 0 with speed U. By the

linearity of 4.71, it is clear that U is the only velocity scale U. The
√
diffusion equation itself has no intrinsic length other than νt, which
follows from the scalings we discussed in Chapter 1 (1.14). It follows
that the velocity should have the form of a similarity solution
x
u = U f (χ) , χ= √ , (4.72)
νt
where the function f obeys the ODE
1
f χχ = − χ f χ . (4.73)
2
We solve this by setting g = f χ to obtain g = A exp(−χ2 /4), and fix
A by the boundary conditions f (0) = 1, f (∞) = 0, yielding

u x
= 1 − erf √ , (4.74)
U 2 νt
where erf is the error function. This solution exhibits a diffusively-
moving transition region that separates the fluid moving with the
wall from that which has yet to be set in motion.
For SII, we consider the situation after any transients have decayed
away. If the wall is oscillated at frequency ω then from the diffusion
√
equation 4.71 the only characteristic length is ∼ 2ν/ω (where we
the factor of 2 is for later convenience); this is the viscous penetration
length. We therefore expect a similarity solution of the form
n o x
u = URe eiωt F (χ) , χ = √ . (4.75)
2ν/ω
The function F then obeys the ODE Fχχ = 2iF, which whose solution
is F ∼ eλχ , with λ± = ±(1 + i ). Only the root with negative real part
is relevant, given the boundary conditions at infinity, yielding
u = Ue−χ cos (χ − ωt) ., (4.76)
an spatially-damped traveling wave moving outward from the wall.

Now we turn to elastohydrodynamic problems, where a filament
is held at one and and either subject to a transverse flow or oscil-
lated. In either case, we work in the small-amplitude limit for which
points on the filament are at ( x, h( x, t), where the height function h
is assumed to have negligible gradients |h x | ≪ 1, so (nearly all) met-
ric factors in the dynamics can be ignored. In either case, the PDE
governing small-amplitude deviations of the filament is
ζ ⊥ ht = − Ah xxxx . (4.77)
We focus on the case of EHDII and, as with SII, consider the situa-
tion post-transients in which the left-hand side is forced as h(0, t) =
h0 cos(ωt), with zero torque (h xx (0, t) = 0), and the distant end is
free, with h xx ( L) = h xxx ( L) = 0. By analogy with SII, dimensional
analysis shows there is an elastohydrodynamic length
1/4
A
ℓ(ω ) = , (4.78)
ζ⊥ω
and we expect a similarity solution of the form

n o x
h = h0 Re eiωt F (χ) , χ = . (4.79)
ℓ(ω )
Then the scaling function obeys an ODE similar to that in SII (4.73),
Fχχχχ = −iF. (4.80)
Seeking exponential solutions F ∼ eλχ gives four roots to the equa-

tion λ4 = −i. These are
λ1 = C8 − iS8 , λ2 = S8 + iC8 ,
λ3 = −C8 − iS8 , λ4 = −S8 − iC8 , (4.81)
where C8 = cos(π/8) = 0.924 . . . and S8 = sin(π/8) = 0.383 . . ..

For the semi-infinite filament (L → ∞), only the two roots λ3,4 with
negative real part are relevant, and we find
h∞ (χ) 1 n −C8 χ o
= e cos (ωt + S8 χ) + e−S8 χ cos (ωt − C8 χ) . (4.82)
h0 2
Here, in contrast to SII, we see traveling waves in both directions, but
the right-ward wave dominates since S8 < C8 . Figure 4.14 shows the
infinite-length solution at various points through a cycle.
1 Figure 4.14: The L → ∞ solution 4.82

to EHDII plotted at integer multiples
of ωt = π/4 through a complete cycle.
Obtained with Figure_414_EHD.m.
h1 =h0
-1
0 2 4 6
@
There are three points to emphasise. First, when L → ∞ there are 15

Gadêlha, H. and Gaffney, E.A. Flag-
no remaining parameters of the problem when expressed as h/h0 ellar ultrastructure suppresses buckling
versus χ; the curves in Fig. 4.14 are "universal" in this limit. This instabilities and enables mammalian
sperm navigation in high-viscosity me-
means that the rapid spatial decay of the amplitude is also universal; dia. J. Roy. Soc. Interface, 16:20180668,
the amplitude is negligible beyond one wavelength of the pattern. 2019
This fact is in sharp contrast to the typical observed waveform of
spermatazoa (Fig. 4.15),15 in which oscillations extend nearly to the
distal end of the filament without attenuation. This fact led Machin
to conclude that sperm flagella can not be actuated only at the prox-
imal end and must instead be driven by energy input throughout
their length. This is indeed correct, as proven by later discoveries of
"motor proteins" that cross-connect the microtubule filaments within
flagella and by consuming ATP translate filaments relative to one Figure 4.15: Waveforms of spermatozoa
another, inducing bending moments throughout the flagellum. in a viscous fluid. From Ref. 15.
A second point is the value of the EH penetration length itself.

Writing A = k B TL p and assuming a large aspect ratio so that the
logarithmic denominator in the drag coefficient is ∼ 5, we find
!1/4
L̃ p
ℓ(ω ) ∼ 0.7 µm, (4.83)
f
where f is measured in Hz and L̃ p is the persistence length in mi-

crons. So, filament oscillations on the experimentally achievable and
biophysically relevant frequency scale of 1 − 100 Hz can bend fila-
ments, with L p in the range of microns to millimeters, on lengths of
microns. This can be done by optical trapping methods.
This calculation can be extended to finite filaments. The problem
is still the similarity solution (4.80) with boundary conditions F (0) =
1, Fχχ (0) = Fχχ (L) = Fχχχ (L) = 0, where L = L/ℓ(ω ) is the scaled
filament length. This linear boundary value problem can be solved
by standard schemes (e.g. bvp4c.m in Matlab) which are used to
compute the filament shapes shown in Fig. 4.16.
1 Figure 4.16: Filament shapes for var-

ious dimensionless lengths L. As
0 in Fig. 4.14, the waveforms are
1
-1 shown at 8 equally spaced inter-
vals over a cycle. Obtained with
1 Figure_416_and_417_Machin.m.
4
0
h(x)=h0
-1
1
2
0
-1
1
L=1
0
-1
0 1 2 3 4 5 6 7
@
When L ≤ 1 the filament is nearly straight throughout the oscil-

lation cycle, pivoting about the point χ = 2/3. By L = 2 a small
amount of undulation is observed, and the bending waves are well
developed by L = 4, eventually producing the traveling-wave forms
of the semi-infinite solution. This leads naturally to a discussion of
the propulsive force arising from these motions. We can appeal to
an irreversibility argument popularised by Purcell,16 now known as Purcell, E.M. Life at low Reynolds
16
the "scallop theorem". Since the Stokes equation 0 = −∇ p + µ∇2 u number. Am. J. Phys., 45:1–11, 1977
has no explicit time derivative, if some particular function u repre-

sents the solution to a problem as time t advances forward, it is an
equally valid solution when t → −t (colloquially, "when the movie
is played backwards"). If we took the solution of EHDII for vanish-
ing L, where it reduces to a pivoting rod, and imagine that motion
moved the rod, say, to the left, then reversing time must produce
movement to the right at an equal and opposite speed. But the rod’s
motion is indistinguishable in the two cases, so the only allowable
speed is zero. This is referred to as the “scallop theorem" as a means
of highlighting that an object with a single degree of freedom exe-
cuting cyclic motion will exhibit no net propulsion. As the rod starts
to flex during the oscillations, time reversal invariance is broken, so
the two movies are not equivalent, and net propulsion can occur.
EHDII provides a particularly simple way to understand this ef-
fect, by means of the time-averaged x-component of the force asso-
ciated with the filament oscillations. Here, we project the left hand
side of the general force law (4.23) on to the x axis, yielding for our
problem of a curve in the plane17 17
Camalet, S. and Jülicher, F. and Prost,
J. Self-organized beating and swim-
F = −êx · ζ ∥ t̂t̂ + ζ ⊥ n̂n̂ · rt ≃ ζ ⊥ − ζ ∥ h x ht + · · · . (4.84) ming of internally driven filaments.
Phys. Rev. Lett., 82:1590–1593, 1999
where we have used the leading order tangent and normal vectors
t̂ ≃ êx + h x êy , n̂ ≃ −h x êx + êy , (4.85)
and written rt ≃ ht êy . Integrating this over the filament yields the
total propulsive force generated. Using (4.77), we have
Z L Z L
A
dxh x ht = − dxh x h4x . (4.86)
0 ζ⊥ 0
This would seem to be a formidable integral, but a minor miracle

occurs in that the quantity h x h4x is a total derivative,

d 1
h x h4x = h x h3x − h2xx , (4.87)
dx 2
so the total propulsive force Fp can be expressed in terms of the fila-
ment shape at its two ends. If the distant end is free, then h xx ( L) =
h3x ( L) = 0, leaving only a contribution from the oscillated end,
A
Fp = ζ ⊥ − ζ ∥ h x (0, t)h xxx (0, t), (4.88)
ζ⊥
where we have made explicit the time dependence of the boundary
terms. Scaling lengths with ℓ(ω ) and assuming the form (4.79), Eq.
(4.88) can be written as

Fp = h20 ω ζ ⊥ − ζ ∥ Re{eiωt Fχ (0)}Re{eiωt Fχχχ (0)}, (4.89)
The prefactor indeed has units of a force (viscosity×length×velocity).

Significantly, net propulsion only arises when the parallel and per-
pendicular drag coefficients differ, as they do for long slender fila-
ments (ζ ⊥ = 2ζ ∥ ). The quadratic dependence on the height function
h will give rise to a nonzero mean value of the force averaged over a
cycle. Denoting such an average by ⟨ Fp ⟩ we find
1
⟨ Fp ⟩ = ζ ⊥ − ζ ∥ h20 ωY (L), (4.90)
2
where L = L/ℓ(ω ) and
Y (L) = FχR (0) Fχχχ

R
(0) + FχI (0) Fχχχ
I
(0), (4.91)
Figure 4.17: Scaling functions Y (L)

0.5 (blue), V (L) (black, ×5 for clar-
V
ity), and Z (L) (red) for the propul-
sive force, swimming speed, and
Y (L); V (L); Z(L)
0.4 Y efficiency in EHDII. Obtained with
Figure_416_and_417_Machin.m.
0.3
Z
0.2
0.1
0
0 5 10
L
with F (χ) = F R (χ) + iF I (χ). The function Y (L) is shown as the blue
curve in Fig. 4.17. The propulsive force indeed tends to zero as the L
decreases and the shapes are time-reversal-invariant, consistent with
the scallop theorem. As L increases, Y displays a maximum followed
by attenuated oscillations as it asymptotes to a finite value as L → ∞.
This saturation indicates that beyond L ∼ 8 no more of the filament
is set in motion and no further propulsive force is gained.
In the regime where the propulsive force has saturated any ad-
ditional length added to the filament serves only to increase the
drag for motion along the x axis, and one therefore expects that
the propulsion speed of the filament would peak at an intermediate
value of L. As we have assumed in EHDII that the filament is nearly
straight (|h x | ≪ 1, it is consistent to assume that the average lateral
propulsive speed ⟨v x ⟩ of the undulating filament can be deduced by
the simple balance of force ζ ∥ L⟨v x ⟩ = ⟨ Fp ⟩, or
⟨ Fp ⟩ h2 ω
⟨v x ⟩ = = 0 V (L), (4.92)
ζ∥ L 2ℓ(ω )
where V (L) = Y (L)/L. As seen in Fig. 4.16, the peak in ⟨ Fp ⟩ implies
a peak in speed, and ⟨v x ⟩ vanishes as L → ∞ as anticipated.
The average power P∥ associated with the lateral motion is P∥ =
v x ⟨ Fp ⟩ ∼ ⟨ Fp ⟩2 . It can be compared to the power
Z L
P⊥ = ζ ⊥ ⟨ dxh2t ⟩ (4.93)
0
generated by transverse motions. Together, these define the "effi-
ciency" ε = P∥ /P⊥ of the propulsive mechanism, which has the form
2
h0
ε∼ Z (L), (4.94)
ℓ(ω )
where the scaling function is
Y 2 (L)
Z (L) = RL . (4.95)
L 0 dχ| F (χ)|2
Figure 4.17 shows that the efficiency also displays the peaked be-
haviour of the other scaling functions, suggesting an optimum length.
4.3.3 Example 3: Stretch-coil transition under compressional flows

Our second example of elastohydrodynamics is based on the experi-
mental geometry in Fig. 4.6, in the case in which a filament initially
aligned along one channel’s long axis is subject to a compressional 18
Young, Y.-N. and Shelley, M.J. Stretch-
coil transition and transport of fibers
flow inwards from its two ends. The analysis of this "stretch-coil in cellular flows. Phys. Rev. Lett., 99:
transition" was first given by Young and Shelley,18 confirmed at the 058303, 2007
microscale by later experiments we discuss below. When a filament
is subject to a viscous flow of velocity u parallel to its long axis, it
will experience a tangential viscous force per unit length of ζ ∥ u. In
the geometry of Fig. 4.6, the in-plane flow is the of a hyperbolic stag-
nation point, with u = (γ̇x, −γ̇y), where γ̇ is the imposed shear rate,
and γ̇ < 0 for a compressional flow and > 0 for extensional flow of a
filament oriented along x. In equilibrium the force density −ζ ∥ γ̇x
must balance gradients in the tension σ ( x ) within the filament,
σx = −ζ ∥ γ̇x, (4.96)
which, subject to the vanishing of σ at the filament ends (here taken

at ± L/2), integrates to yield
1 L2

σ( x) = − x2 ζ ∥ γ̇, (4.97)
2 4
a relation noted much earlier by Batchelor.19 In the presence of this 19
Batchelor, G.K. Slender-body theory
nonuniform tension, the filament energy functional is for particles of arbitrary cross-section in
Stokes flow. J. Fluid Mech., 44:419–440,
Z L/2 1970
1
E= dx Ah2xx + σ( x )h2x . (4.98)
2 − L/2
Just as with Euler buckling, where a concentrated force is applied

to one of the filament, here in the presence of a distributed force
we expect a buckling transition for sufficiently large (negative) shear
rate. The linear stability problem of interest derives from our basic
formulation (4.25), but now augmented with an external flow field u
for relative motions of the filament,
1
rt − u = n̂n̂ + 2t̂t̂ · − Arssss + σt̂ s , (4.99)
ζ⊥
where u is to be evaluated at the filament. In the tensorial term on
the r.h.s., we have specialized to a filament deforming in the plane
and assumed the asymptotic limit for the drag coefficient ratio. If
we now use the standard Monge relations (4.85) for t̂ and n̂ of a
weakly-deformed filament, and perform the rescalings
πx h
χ= , τ = γ̇t, H= , (4.100)
L L
then a very careful derivation of the linearised dynamics yields
2
π 2
4|Σ| ( Hτ + sgn(γ̇) H ) = − H4χ + Σ − χ Hχχ − 4χHχ ,
4
(4.101)
where the control parameter of the problem is
µγ̇L4
Σ= √ , (4.102)
π 3 ln( L/a e)
and the logarithm in the denominator arises from the RFT drag coef-
ficient. Note that the tension term on the r.h.s. of (4.101) is not a total
derivative; the linearised equation of motion is not variational. This
arises from the anisotropic drag and the fact that the background
flow that enters the drag force is the source of the tension.
The determination of the threshold for the stretch-coil instability
is a matter of a numerical solution of the the eigenvalue problem that
comes from a separable solution H (χ, τ ) = exp(ωτ ) F (χ), namely
2
π 2
F4χ − Σ − χ Fχχ − 4χFχ = 4Σ (ω − 1) F. (4.103)
4
Our general expectation is that there will be a discrete set of modes
analogous to the biharmonic eigenfunctions discussed earlier in this
chapter, each of which will be stable at low γ̇ and attain a posi-
tive growth rate ω for sufficiently large shear rate. In the program
Figure_418_stretchcoil.m we find the different critical values Σ∗n
at which ω = 0 by appropriate choice of the initial guess for the
iterative solution.
1 Figure 4.18: The first three

critical modes for the stretch-
2 3 coil instability. Obtained with
Figure_418_stretchcoil.m.
F (@)
n=1
-1
!:=2 0 :=2
@
Such analysis yields the critical value Σ∗ = −0.3932 for the first
unstable mode, Σ2∗ = −1.9876 for the second mode, Σ3∗ = −4.955 for
the third, etc. Those higher order shapes (Fig. 4.18) are visible in the
montage of filament shapes shown in Fig. 4.19. The amplitude of
the buckled filaments, taken as 1 − L/L, where L is the end-to-end
displacement, displays a thermal-fluctuation-rounded pitchfork bi- 20
H. Manikantan and D. Saintillan.
furcation (Fig. 4.19). A theoretical treatment of the combined effects Buckling transition of a semiflexible fil-
ament in extensional flow. Phys. Rev. E,
of thermal fluctuations and nonlinearities on the stretch-coil transi- 92:041002, 2015
tion has recently been developed.20
4.3.4 Example 4: Motor-induced oscillations

Our final example of elastohydrodynamics concerns the effect of
molecular motors translating along biological filaments on the shapes
of those filaments. Motivation for this comes from observations of
the dynamics of microtubules during development of the oocytes
(egg cells) of the fruit fly Drosophila melanogaster. At intermediate
stages of this multi-day process the oocyte is on the order of 100 µm
Figure 4.19: (a) Montage of buckling at

various values of the shear rate. Scale
across, with multitudes of microtubules reaching toward the center, bars are 3 µm and times are dimension-
less values τ since first frame of each
anchored at the periphery. Molecular motors such as kinesin translo-
row. Control parameter values are (a)
cate along those filaments, carrying nanometric cargo that plays a Σ = 0.55, (b) Σ = 1.9, and (c) Σ = 47.
role in establishing the future body plan of the organism. As that (b) Order parameter versus dimension-
less shear rate. Dashed lines indicate
cargo moves, it entrains fluid, producing complex swirling patterns instability thresholds for modes shown.
of fluid motion that can also back-react on the filaments, bending Gray band is noise floor. From Ref. 20.
them. Most of the motor-filament pairs in the cytoskeleton are "po-
lar", in the sense that the motors walk in a unique direction along the
filament. In the case of these anchored microtubules, it is typically
from the anchored end toward the free end. As the motor walks in
that direction, there is a corresponding force on the filament directed
back toward the anchoring point. This is a compressive force akin to
that of Euler buckling. And since motors can exert forces on the pi-
coNewton scale and the microtubules can be tens of microns long,
our estimates of the Euler buckling force (4.64) show the filaments 21
Herrmann, G. and Bungay, R.W. On
can indeed be deformed by a single motor. But there is a crucial the stability of elastic systems sub-
jected to nonconservative forces. J. Appl.
difference between Euler buckling and the motor-filament situation; Mech., 31:435–440, 1964
in the latter, the force is always tangential to the filament. This is an 22
De Canio, G., Lauga, E. and Gold-
example of a "follower force", and it is known to have fundamentally stein, R.E. Spontaneous oscillations of
elastic filaments induced by molecular
distinct properties from Euler buckling.21 Chief among them is a motors. J. R. Soc. Interface, 14:20170491,
lack of a variational structure to the dynamics, even in Stokes flow.22 2017
As we shall see below, this can be traced to the existence of boundary
conditions that produce a non-self-adjoint operator.
The simplest setup of the problem is shown in Fig. 4.20. A fila-
ment of length L is clamped at the origin along the x axis, and subject
to a point force of magnitude Γ tangential to its distal end. We again
use the standard form of elastohydrodynamics (4.25) in which the
elastic force density is written as
− Arssss − (Λrs )s , (4.104)
where Λ is the internal tension. The boundary conditions at the

origin are simply r(0, t) = 0 and rs (0, t) = 0, while at the distal Figure 4.20: Setup for the follower force
problem.
end the torque-free condition requires rss ( L, t) = 0. The condition of
force balance at the distal end can be written as
− Arsss − Λ( L, t)rs ( L, t) = −Γrs ( L, t), (4.105)

which is satisfied by Λ( L, t) = Γ and rsss ( L, t) = 0. The tension itself

must satisfy the usual differential equation (4.43).
We focus on the stability problem for the straight filament, for
which the linearised dynamics takes a simple form. The solution to
the tension equation is simply the constant Λ = Γ. We rescale via
x At h ΓL2
χ= , τ= , H= , σ= , (4.106)
L ζ ⊥ L4 L A
where the temporal rescaling reflects the characteristic bending re-

laxation time of the filament and σ > 0 for a compressive force. The
governing equation is thus
Hτ = − Hχχχχ − σHχχ , (4.107)
with H (0, τ ) = Hχ (0, τ ) = Hχχ (1, τ ) = Hχχχ (1, τ ) = 0. Note that

the tension term by itself would correspond to a "backwards" diffu-
sion problem that would be ill-posed without the stabilising fourth
derivative from bending elasticity. Were it not for the nontrivial
boundary conditions for this problem, we could simply consider sep-
arable solutions of the form for plane waves, H ∼ exp(ωτ ) exp(iQχ),
and deduce a growth rate ω for a mode of wavevector Q of the form
ω = σQ2 − Q4 , which displays a band of unstable modes (those with
√
ω > 0 in the range 0 ≤ Q ≤ σ, with a real growth rate ω. But in re-
ality we can not ignore the boundary terms, and therein lies the key
to the unexpected features of this problem. To see this, we observe
that in the linearised approximation, where the tension is a constant,
the (rescaled) energy of the system is
Z 1
1
2

E= dχ Hχχ − σHχ2 (4.108)
2 0
The time derivative of E itself follows directly from the equation of

motion (4.107) and several integrations by parts,
1 1 1
Eτ = Hχχ Hχτ − Hχχχ Hτ − σHχ Hτ
0 0 0
Z 1
+ dχ ( Hχχχχ + σHχχ) Hτ . (4.109)
0
The first two surface terms vanish identically at both ends by virtue
of the boundary conditions, as does the third term Hχ Hτ at χ = 0,
but Hχ (1, τ ) Hτ (1, τ ) is not forced to vanish by any boundary condi-
tions. Using (4.107) to simplify the integral in (4.109) we obtain
Z 1
Eτ = − dχ ( Hχχχχ + σHχχ)2 − σHχ (1, τ ) Hτ (1, τ ). (4.110)
0
While the integral is clearly nonpositive, the sign of the surface term
is indefinite, so we can not conclude that Eτ < 0 and indeed the
surface term represents the injection of energy into the filament that
sustains oscillations. Stated another way, the surface term breaks the
self-adjointness of the Sturm-Liouville operator ∂χχχχ + σ∂χχ , and
therefore its eigenvalues need not be real.
The eigenvalue problem of interest is obtained by assuming a sep-

arable solution to (4.107) H (χ, τ ) = exp(ωτ ) G (χ), leading to
Gχχχχ + σGχχ = −ωG, (4.111)
with boundary conditions G (0) = Gχ (0) = Gχχ (1) = Gχχχ (1) =

0. While this can be worked out semi-analytically, its numerical
solution provides the student with an opportunity to gain experi-
ence with numerical methods for such problems. In the program
Figure_421_and_423_followerforce.m we implement a finite differ-
ence representation of the operator, with careful attention to the one-
sided stencils needed to implement the boundary conditions. Figure
4.21 shows the real and imaginary parts of the largest eigenvalue
obtained from the matrix representation of the operator.
200 Figure 4.21: Components of com-

plex eigenvalue for follower-force
Im(!) problem as a function of dimen-
100 sionless motor force σ. Red circle
represents onset of supercritical
Hopf bifurcation. Obtained with
!
0 Figure_421_and_423_followerforce.m.
Re(!)
-100
0 10 20 30 40 50
<
The eigenvalue ω is purely real and negative up to σ ≃ 20, as

it is for σ = 0, but beyond that value the imaginary part grows.
The real part becomes positive at σ∗ ≃ 37.69 with a finite imaginary
part; the classic signature of a Hopf bifurcationHopf bifurcation. It
is interesting to see how the character of the leading eigenfunction
behaves as the motor strength σ is increased. Figure 4.22 shows the
normalized real part of the eigenfunctions G (χ) for a range of values
3 Figure 4.22: Normalised real part

of the leading eigenfunction of the
follower-force problem. Curves are
2 for a range of σ from 0 to 45, color-
coded from red to blue. Obtained with
Figure_421_and_423_followerforce.m.
G(@)
0 0.5 1
@
of σ. For small σ the mode is a simple monotonic function, while at

large σ, beyond the bifurcation, it develops a wavelike form.
As an example, we construct the full waving solution as Re[exp(ωτ ) G (χ)],
using the complex leading eigenfunction G found numerically for
σ = 45, and plot it during one cycle of motion. As shown in Fig. 4.23
the dynamics have the character of flagellar motion.
2 Figure 4.23: Separable solution of

follower-force problem at σ = 45 plot-
ted during one cycle. Obtained with
1
Figure_421_and_423_followerforce.m.
H(@)
-1
-2
0 0.5 1
@
5
Surfaces & Membranes (Lectures 13-16)
In these four lectures we consider the properties of biological mem-

branes. We begin with a review of the basic differential geometry
of surfaces and some variational aspects in order to understand the
pedagogical example of minimal surfaces. Examples of fluctuations
of closed objects are studied to understand how global constraints
such as fixed enclosed volume affect the spectra. Applications to
lipid vesicles are given, along with the physics of intermembrane
repulsion due to constrained fluctuations. In the later parts of the
chapter we consider the mathematical description of the equilibrium
shapes of vesicles and several examples of shape transformations.
5.1 Differential geometry of surfaces
We begin with a brief summary of differential geometry. Consider a

surface traced out by the vector x(uα ) through internal coordinates
uα , α = 1, 2. The surface has a metric tensor with components
gαβ = xα · xβ , (5.1)
where here subscripts denote differentiation, so xα ≡ ∂x/∂uα . The xα

are the unnormalized tangent vectors to the surface. They define the
unit normal n̂ as, say,
x1 × x2 x ×x
n̂ = = 1√ 2 , (5.2)
|x1 × x2 | g
or its opposite if we switch 1 ↔ 2. Here, g = det( g) is the determi-

nant of the metric tensor,
g = ϵαβ g1α g2β , (5.3)
where ϵ is the Levi-Civita symbol and the summation convention

holds. The inverse metric tensor, with coefficients gαβ is the usual
!
−1 1 g22 − g12
g = . (5.4)
g − g21 g11
The coefficients hαβ of the second fundamental form are defined

by the normal and tangent vectors as
hαβ = −n̂α · xβ . (5.5)

If we denote by Tr the conventional trace of a matrix, then the mean

and Gaussian curvature, H and K, of the surface are
1 α
H = Tr h (5.6a)
2 β
K = det hα β , (5.6b)
where hα β = gαµ hµβ .
5.1.1 Surfaces of Revolution

Let us apply these formal results to a situation that arises in many
contexts, an axisymmetric surface. Assuming it is obtained as a sur-
face of revolution by rotating around the z axis, it is described by
x = ζ (z) [cos θ ê1 + sin θ ê2 ] + zê3 . (5.7)
A simple calculation shows that the normal vector is
cos θ ê1 + sin θ ê2 − rz ê3

n̂ = p , (5.8)
1 + rz2
while the metric tensor has the simple form

!
ζ2 0
g= . (5.9)
0 1 + ζ z2
Then, the coefficients of the second fundamental form are

p !
− 1/ζ 1 + ζ 2 0
z
hα β = , (5.10)
0 ζ zz /(1 + ζ z2 )3/2
and thus the mean and Gaussian curvatures are

!
1 ζ zz 1
H= − p , (5.11a)
2 (1 + ζ z2 )3/2 ζ 1 + ζ z2
ζ zz
K=− 2
. (5.11b)
ζ (1 + ζ z2 )
5.1.2 Catenoids
Now we ask how these geometric quantities arise in variational cal-
culations involving surfaces. The simplest context is that of surfaces
endowed with an energy per unit area (surface tension), for which
the area functional A[ζ ] is of interest,
Z d q
A[ζ ] = dz ζ 1 + ζ z2 . (5.12)
−d
where for convenience we have dropped the 2π from the θ integra-

tion. A simple calculation yields the functional derivative of A,
δA
= −2ζ H, (5.13)
δζ
where H is the mean curvature (5.11a).

surfaces & membranes (lectures 13-16) 83
If the surface of interest is endowed with surface tension γ, its

surface energy is proportional to γA, and the quantity −γδA/δζ
is proportional to the Young-Laplace pressure difference across the
surface. In the special case H = 0, the pressure difference vanishes
and the surface is in equilibrium. This minimal surface satisfies
ζ zz 1
2
= . (5.14)
1 + ζz ζ
A simple calculation shows that the shape
ζ (z) = a cosh z/a (5.15)
is a solution to (5.14) for a range of the parameter a. This the equation

of a catenoid, an experimental example of which is shown in Fig. 5.1,
where the surface of revolution spans two wire loops.
The solutions (5.15) must solve the boundary condition that they
meet the wires, ζ (±d) = R, where R is the wire radius. If we define Figure 5.1: A soap film in the form of
the dimensionless variables a catenoid, made visible with dissolved
fluorescein illuminated with cyan light.
a d
α= , D= , (5.16)
R R
then the boundary condition can be written as
α cosh( D/α) − 1 = 0. (5.17)
Viewing the radius of the hoops as a fixed quantity and the sepa-
ration 2d as adjustable, we see in Fig. 5.2(a) that this transcenden-
tal equation has two solutions for small D that merge into one at
D = Dc = 0.6627, and there are no solutions for larger D. The two
branches of solution, labelled α± , are shown in Fig. 5.2(b) to merge
together in what is known as a saddle-node bifurcation.
1
(a) (b)
,+
0.5
, cosh(D=,) ! 1
0.75
0:8
Dc ,c
0 0.5
,
0:6
0:5
-0.5 0:3 0.25 ,! Dc

D = 0:1
0
0 0.5 1 0 0.2 0.4 0.6
, D
Figure 5.2: Graphical solution of
the catenoid equation (a) and solu-
The two branches of solutions correspond to catenoids of very
tion branches versus nondimensional
different shapes, as shown in Fig. 5.3. At any given separation, the ring separation (b). Obtained with
upper branch (α+ ) has a lower area then that of the lower branch Figure_502_and_504_catenoids.m.
(α− ), and is the true least-area surface. The areas of the two branches
of solution can be calculated analytically in terms of the solutions
α± . Expressed in dimensionless form as A± = A± /R2 , they are

α± 2D 2D
A± = sinh + , (5.18)
2 α± α±
Figure 5.3: Catenoidal solutions at

separations D = 0.1, 0.3, 0.5, 0.662 for
the two branches of solutions in Fig.
5.2(b). The coloured circles repre-
sent the imaginary supporting loops,
seen in cross section. Obtained with
Figure_503_catenoidsolutions.m.
and are shown in Fig. 5.4. The upper and lower branches of solution
meet at the critical area Ac = 1.199 . . .. A more involved calculation
is necessary to show the stability of the two branches, which are
indicated in the figure. We shall return to these catenoidal solutions
in our study of membrane tethers in Sec. 5.3.2.
(c) Figure 5.4: Areas of the two

1.25 e r- branches of solutions of the catenoid
Ac sup cal equation versus dimensionless
unstable c ri t
i
ring spacing D. Obtained with
1 Figure_502_and_504_catenoids.m.
l e
0.75 ab
A
st
0.5
l D$ Dc
ca
r i ti
0.25 bc
su
0
0 0.2 0.4 0.6 0.8
D
5.2 Membrane Fluctuations
Next we consider some important problems in the statistical physics

of fluctuating membranes: equilibrium thermal fluctuations of bi-
layer lipid vesicles, fluctuation-induced repulsion between nearby
membranes, and the dynamics of red blood cell fluctuations.
5.2.1 Thermal fluctuations of spherical vesicles L
Before delving into details of this calculation, we note an important R(z)

feature of fluctuations associated with closed objects whose interior z
volume is a constant. We start with the simple problem of a weakly
Figure 5.5: A perturbed cylinder.
perturbed cylindrical body of length L, as in Fig. 5.5. The volume V
and surface area S of the body are
Z L Z L q
V=π dzR2 (z), S = 2π dzR 1 + R2z . (5.19)
0 0
Let R0 be the cylinder radius prior to any perturbations. As we will

be considering the use of equipartition to describe thermal fluctua-
tions, it is necessary to work to second order in deviations from the
circle. Under such perturbations, the mean radius ρ0 will need to

differ from R0 in order to satisfy volume conservation at second or-
der. To see this, assume the body is perturbed by a sum of Fourier
modes along its length, but remains axisymmetric,
R = R̄ + ∑ an sin qn z, (5.20)
n
where R̄ is to be determined and qn = 2πn/L. To quadratic order in

the amplitudes an , the volume is
!
1
V ≃ πL R̄ + ∑ an
2 2
(5.21)
2 n
Thus, if V is be unchanged from its unperturbed value πLR20 to this

order in an , we must have
1
R̄ ≃ R0 −
4R0 ∑ a2n . (5.22)
n
If we now compute the surface area to the same order and impose
volume conservation, we find that the surface energy γS is
π γL h i
γS = 2πLR0 γ +
2 R0 ∑ (qn R0 )2 − 1 a2n + · · · . (5.23)
n
The constraint of fixed volume has led to the crucial term −1 inside
the square brackets instead of simply a factor of q2 associated with
tension without conservation of volume. Its significance can be seen
if we apply equipartition to this energy, from which we conclude that
k B TR0
⟨ a2n ⟩ = h i. (5.24)
πLγ (qn R0 )2 − 1
Clearly, there is a divergence when qn R0 = 1 and the result is not

physical for qn R0 < 1. These indicate an instability at qn R0 = 1,
which is, for n = 1, exactly Plateau’s criterion that a cylinder longer
than its circumference is unstable under the action of surface tension.
The calculation above assumed axisymmetry. Suppose now that
we break circular symmetry, using the simpler context of a perturbed
circular domain. If its unperturbed radius is R0 , and the perturbed
boundary is given as a function of the polar angle θ by
!
r( θ ) = R̄ + ∑ an sin nθ êr (5.25)
n
then the constraint of fixed enclosed area yields the same result as in
(5.22), and the line energy γL is
πγ
2R0 ∑
2
γL = 2πR0 γ + n − 1 a2n , (5.26)
n
and hence
k B TR0
⟨ a2n ⟩ = . (5.27)
πγ (n2 − 1)
Now the reason for the divergence at n = 1 is not an instability, but

rather that an perturbation of the form sin θ is an infinitesimal trans-
lation of the circle. This is not true for large a1 , where the deformed
shape is like a cardioid, as shown in Fig. 5.6. For infinitesimal a1 ,
such a perturbation costs no energy (in this setup), and would natu-
rally have a divergent amplitude if endowed with energy k B T.
We can apply these results to understand the spectrum of thermal
Figure 5.6: A perturbed circle for
fluctuations of a lipid bilayer vesicle, a problem studied experimen- a1 = 0.1, 0.2, . . . , 0.8. Obtained with
tally some time ago1 and analyzed in detail by Helfrich,2 , Milner and Figure_506_perturbedcircle.m.
Safran,3 and Zhong-can and Helfrich.4 Consider a spherical vesicle 1

Schneider, M.B., Jenkins, J.T. and
of radius R perturbed by small-amplitude fluctuations, so that points Webb, W.W. Thermal fluctuations of
on the surface can be labelled as large quasi-spherical bimolecular phos-
pholipid vesicles. J. Phys., 45:1457–1472,
1984
r(θ, ϕ) = ( R + η (θ, ϕ)) êr . (5.28) 2
Helfrich, W. Size distributions of vesi-
cles: the role of the effective rigidity
where θ, ϕ are the usual spherical coordinates. The necessary calcu- of membranes. J. Physique, 47:321–329,
lation involves expressing the change in the elastic energy 1986
3
Milner, S.T. and Safran, S.A. Dynam-
1
Z ical fluctuations of droplet microemul-
E= kc dS(2H )2 , (5.29) sions and vesicles. Phys. Rev. A, 36:
2 4371–4379, 1987
up to quadratic order in η. Note that we have, for simplicity, ne-
4
Zhong-can, O.-Y. and Helfrich, W.
Bending energy of vesicle membranes:
glected any contribution from spontaneous curvature and from the General expressions for the first, sec-
Gaussian curvature term in (2.66) due to its topological character. ond, and third variation of the shape
energy and applications to spheres and
The calculation itself is a somewhat tedious exercise in expanding cylinders. Phys. Rev. A, 39:5280–5288,
the energy functional for weak perturbations around the sphere, ul- 1989
timately expressing the relative displacement η/R of the radius as
an expansion in spherical harmonics Ylm (θ, ϕ), leading to the result
1
E ≃ 8πk c + k c ∑ l (l − 1) (l + 1) (l + 2) | alm |2 . (5.30)
2 l,m
The l-dependent coefficient is vanishes identically for both l = 0 and

l = 1, even without adjustment of the mean radius. This is a special
result associated with the scale invariance of the elastic energy (5.29)
in the absence of the spontaneous curvature term. When H0 ̸= 0, the
mean radius of the vesicle must be adjusted to conserve volume at
second order, as discussed above. The coefficients scale with mode
number as l 4 , as expected from elasticity.
5
Helfrich, W. Steric interactions of fluid
5.2.2 Helfrich repulsion membranes in multilayer systems. Z.
Naturforsch. Teil A, 33:305–315, 1978
An important consequence of thermal fluctuations is the phenomenon 6
Helfrich, W. and Servuss, R.-M. Undu-
of "fluctuation-induced interactions" between elastic objects. These lations, steric interaction and cohesion
of fluid membranes. Nuovo Cim., 3D:
were first proposed by Helfrich5 in the context of multilayer systems, 137–151, 1984
and rederived later using simplified arguments6 that we repeat here.
7
Marko, J.F. and Siggia, E.D. Fluctua-
More recently, similar considerations have been applied to the struc- tions and supercoiling of DNA. Science,
ture of DNA in tightly-packed chromosomes.7 265:506–508, 1994
We consider a two-dimensional elastic membrane initially free in
space, but later confined between two impenetrable walls spaced a
distance 2d apart, as in Fig. 5.7. Starting from the usual elastic energy
E = (k c /2) dS(2H )2 , adding for generality a tension contribution
R
R
σ dS, and assuming small gradients, we have
2
1
Z
E≃ d2 x k c ∇2 h + σ (∇h)2 . (5.31)
2
2d
If we decompose the shape in the usual manner as
h (r) = ∑ e−iq·r ĥ(q), (5.32) h(x)

q
then the energy is

Figure 5.7: A fluctuating elastic mem-
1 brane confined between two walls.
E = A ∑ k c q4 + σq2 |ĥ(q)|2 , (5.33)
2 q
and by equipartition
kB T
⟨|ĥ(q)|2 ⟩ = . (5.34)
A (k c q4 + σq2 )
It follows that for the unconstrained membrane the variance of dis-
placements is
k T qdq
Z
⟨ h2 ⟩ = B (5.35)
2π k c q4 + σq2
√
If σ = 0 (a free membrane), using the cutoff qmin = π/ A, we find
kB T
⟨ h2 ⟩ = A. (5.36)
4π 3 k c
The fact that the variance grows with the area A of the membrane
led Helfrich and Servuss to argue that there will be a scale A∗ at
which the variance will be of order d2 , at which point the membrane
strongly interacts with the walls. The precise value of the prefactor
c in the statement ⟨h2 ⟩ = cd2 varies with the assumptions of the
calculation, so we leave it unspecified at the moment. Then,

kc
A∗ = 4π 3 c d2 . (5.37)
kB T
The next step involves the idea that each patch of membrane of
area A∗ acts as a quasi-independent "particle" confined in the gap
of size 2d, bouncing around in thermal equilibrium like a particle
in a confined one-dimensional ideal gas. The pressure exerted on
the wall can be deduced from the average force the wall experiences
from collisions of the constituent "particles" with some assume mass
m. That force is is just the ratio of the momentum transfer ∆p =
m∆v = 2mv to the average time interval ∆t between collisions with
the wall. Since the average velocity vanishes by symmetry, we use
mv ∼ m⟨v2 ⟩1/2 , and set ∆t = 4d/⟨v2 ⟩1/2 and thus F = k B T/2d.
Dividing by the area A∗ we obtain the pressure P = F/A∗ , or
( k B T )2 1
P∼ . (5.38)
8π 3 ck c d3
The energy per unit area is the integral of P with respect to d,
1 kB T kB T
Vrep (d) = . (5.39)
16π 3 c k c d2
Written in this manner we see naturally that the repulsive interac-

tion can be compared to the van der Waals attraction (2.33) between
membranes, whose leading order behaviour for nearby membranes
has the same power-law dependence, leading to the sum

1 kB T 1 AH 1
Vrep + VA ≃ k B T − . (5.40)
16π 3 c k c 12π k B T d2
In this expression we have grouped terms to make the relevant di-

mensionless ratios apparent. Based on this comparison, and exclud-
ing any other types of intermembrane interactions, one would con-
clude that if van der Waals interactions dominate (large A H then the
membranes adhere to each other, while if the fluctuation-induced re-
pulsion dominates (small k c ) they are repelled to infinity. The prob-
8
Lipowsky, R. and Leibler S. Unbinding
lem becomes more complicated in the presence of screened electro- transitions of interacting membranes.
statics and if one includes the role of thermal fluctuations.8 Phys. Rev. Lett., 56:2541–2544, 1986
5.2.3 Flicker phenomena of red blood cell membranes

Next we consider a classic problem in the dynamics of cell mem-
branes, the "flicker phenomenon" of red blood cells. The name de-
rives from the persistent shimmering observed at the junction of
the rim and the centre of erythrocytes, particularly when viewed
in phase contrast microscopy. While historical explanations invoked
specific biochemical reactions within the cytoplasm of the cells, it is 9
Brochard, F. and Lennon, J.F. Fre-
now understood that the phenomenon arises from thermal fluctua- quency spectrum of the flicker phe-
nomenon in erythrocytes. J. Physique,
tions of the red blood cell (RBC) membrane. The first comprehensive 36:1035–1047, 1975
treatment of this phenomenon was by Brochard and Lennon,9 whose
analysis we follow here with significant simplifications.
Recall the typical shapes of RBCs, as shown in Chapter 2 (Fig. 5.8).
It is apparent that the central region of an RBC consists of roughly
parallel sections of the membrane bounding a thin disc of cytoplasm.
The cytoplasm itself contains a complex soup of molecular species,
and has a different index of refraction from the surrounding water.
When light passes through the RBC, the phase shift experienced will
depend on the product of the local thickness and the index of re-
fraction difference from the surrounding water, and it is this phase
difference that is measured in phase contrast microscopy. It follows Figure 5.8: Electron micrograph of red
blood cells.
that dynamical properties of those fluctuations are of interest, and
we must find the equation of motion for membrane deformations.
Since the experimental method is sensitive to variations in the
cytoplasm thickness, the relevant mode of deformation is the anti-
symmetric "peristaltic" mode shown schematically in Fig. 5.9, so if d
h(x)
is the equilibrium thickness, the thickness variation δd = 2h. The d/2
key experimental observations involve measurements of the spatio-
temporal correlation function of the thickness fluctuations. -d/2
Before presenting their results, it is useful to make a slight digres- h(x)
sion to discuss some important features of correlation functions. We Figure 5.9: Geometry of the calculation
for flicker phenomenon.
frame the discussion in the context of the Langevin equation (5.41)
governing the one-dimensional motion of a particle in an optical trap
(modelled as a harmonic potential),
ζ ẋ = −kx + ξ (t), (5.41)
where ζ is the drag coefficient, k is the trap stiffness, and ξ (t) is the
random noise. We define the Fourier transform pairs
Z
x̃ (ω ) = dte−iωt x (t), (5.42a)
dω iωt
Z
x (t) = e x̃ (ω ). (5.42b)
2π
Recall that the choice of where the factor of 2π is located in these
relations is a matter of convention, or it can be split symmetrically
√
between the two as a factor of 1/ 2π. The only constraint is that the
transform of the inverse transform yields the function, or
Z ∞
dωeiωt = 2πδ(t), (5.43)
−∞
where δ is the Dirac delta. A Fourier transform of (5.41) yields
(iωζ + k) x̃ (ω ) = ξ̃ (ω ). (5.44)
As before, we can assume that the noise is delta function correlated,

and thus the power spectrum P(ω ) = ⟨| x̃ (ω )|2 ⟩ is
2k B Tζ 2D
P(ω ) = = 2 . (5.45)
2 2
ω ζ +k 2 ω + τ −2
where τ = ζ/k is the relaxation time, and we used ⟨|ξ̃ (ω )|2 ⟩ =

2k B Tζ. This result can be used three ways. First, there is the general
result connecting the integral of P(ω ) to equilibrium fluctuations,
Z ∞
dω
P ( ω ) = ⟨ x 2 ⟩, (5.46)
−∞ 2π
where the r.h.s. is the expectation in equilibrium. We find
Z ∞
dω 2k B Tζ k T
= B , (5.47)
−∞ 2π ω 2 ζ 2 + k2 k
as we expect from the equipartition result ⟨ 21 kx2 ⟩ = 12 k B T.

A second connection involves the temporal correlation function of
x (t), which follows from the analysis in Sec. 3.3 as
k B T −t/τ
⟨ x (t) x (0)⟩ = e . (5.48)
k
The celebrated Wiener-Khinchine theorem, derived first by Wiener10 10
Wiener, N. Generalized harmonic
and extended by Khinchine,11 states that the power spectrum and analysis. Acta Math., 55:117–258, 1930
11
Khintchine, A. Korrelationstheorie
the autocorrelation function are Fourier transform pairs. With the der stationären stochastischen Prozesse.
optical trap example, this is verified by direct calculation: Math. Ann., 109:604–615, 1934
Z ∞
dω 2D k T
eiωt = B e−t/τ . (5.49)
−∞ 2π ω 2 + τ −2 k
Third, we can define a "susceptibility" Υ of the system as the con-
stant of proportionality between the response of the system to an
external force and the amplitude of that force. From (5.44) we have
Υ = (1/ζ )/(iω + τ −1 ) which leads to the result that the power spec-
trum is proportional to the imaginary part of the response function,
2k B T
P(ω ) = − ℑ {Υ} . (5.50)
ω
This is the generalised Fluctuation-Dissipation Theorem.
With these definitions, we may now summarize the findings of
Brochard and Lennon. Using a spectrum analyser they determined
from the microscopic observations the quantities shown in Fig. 5.10.
Figure 5.10: Key experimental observa-

(a) (b) tions on the flicker phenomenon in ery-
throcytes. (a) Frequency spectrum of
fluctuations at a given point, (b) spa-
tial correlation function (control curve
at zero separation verifies that cell is
unperturbed), and (c) length scale of
(c) correlations. From Ref. 9.
First, they found that the fluctuations in thickness at a given posi-

tion vary with frequency ω according to
⟨δd2 (ω )⟩ ∼ ω −4/3 , (5.51)
as in Fig 5.10(a). Second, at a given frequency, the spatial correlation

is an oscillatory, decaying function of separation [Fig 5.10(b)], and
third, the characteristic length scale ℓ(ω ) of oscillations in the spatial
correlation function decays with frequency approximately as
ℓ(ω ) ∼ ω −1/6 . (5.52)
In summary, the autocorrelation function has the scaling form

−4/3 R
C ( R, ω ) ∼ ω G . (5.53)
ℓ(ω )
A complete treatment of the coupled problem of fluid flow in a
region bounded by two fluctuating membranes is an algebraically
complex problem related to that involved in Taylor’s waving sheet
model of microorganism motility discussed in Chapter 6, but as the
mean intermembrane spacing d is significantly smaller than the lat-
eral extent of the membranes, we can adopt the simplified geometry
of a slab with small thickness variations and use lubrication theory
to determine the fluid flows. Wile a typical lipid membrane must
be thought of as a fluid, the RBC membrane has a complex network
of embedded polymers comprising the so-called "spectrin network"
that gives it more the characteristics of a soft solid, and thus it is
appropriate to consider no-slip boundary conditions on the fluid.
Let us first review the essential results of lubrication theory for

the viscous fluid flow between parallel no-slip surfaces at z = ±d/2.
We assume that the dominant component of the fluid velocity is that
perpendicular to the surface normal, and write
u( x, y, z) ≃ u(x⊥ , z)ê⊥ , (5.54)
where x⊥ = ( x, y) is the in-plane coordinate and ê⊥ is an in-plane

unit vector. In this regime, the viscous term ∇2 u in the Stokes equa-
tion is dominated by z-derivatives and thus
∂ 2 u (x⊥ , z ) ∇ p
= ⊥ , (5.55)
∂z2 µ
where ∇⊥ is the in-plane component of the gradient. Combining this
relationship with the no-slip conditions on u at z = ±d/2 it follows
that u can be written in the separable form
" #
d 2 ∇ ⊥ p (x⊥ )
u (x⊥ , z ) = − − z2 . (5.56)
2 2µ
R d/2
From this we obtain the in-plane fluid flux J(x⊥ ) = −d/2 dzu,
d3
J=− ∇ p, (5.57)
12µ ⊥
and the depth-averaged velocity ū = J/d, which obeys Darcy’s Law,
J d2
ū = =− ∇ p. (5.58)
d 12µ ⊥
Equation 5.56 captures the essence of the dynamics relevant to
the flicker phenomenon, which requires an equation of motion for
deformations of the membrane shape away from the flat equilibrium.
The relevant PDE for the shape evolution can be derived in three
steps. First, note that conservation of fluid volume implies
∂
(d + 2h) = −∇⊥ · J. (5.59)
∂t
Second, from the lubrication theory above we have (5.57) relating the
flux to gradients in the pressure. Finally, the pressure p at the inter-
face is related to the displacement field h( x ) via the usual functional
derivative. In the long-wavelength approximation, this yields
p = k c ∇4⊥ h. (5.60)
Assembling all the pieces, and dropping the subscript ⊥ for conve-
nience, the equation of motion for h is
k c d3 6
ht = ∇ h. (5.61)
24µ
This unusual form, with 6 derivatives, thus arises from a combina-
tion of a conservation law, lubrication flow, and elasticity. By dimen-
sional analysis we deduce immediately from (5.65) that for perturba-
tions at frequency ω there is a length scale
1/6
k c d3

ℓ(ω ) = , (5.62)
24µω
which provides an explanation of the ω −1/6 behaviour shown in Fig.

5.10, provided we can identify this length with the scale of the de-
cay of the correlation function. If we ignore any lateral boundary
conditions and Fourier decompose the displacement field via
h = h0 e−iq·x−iωt , (5.63)
where x is an in-plane coordinate and q is an in-plane wavevector,

we obtain
−iω = −(k c d3 /24µ)q6 , (5.64)
indicating exponential decay of the modes for all q. They are stable.
To find the response function Υ, we introduce an external pressure
field pext that forces the system, and obtain the dynamics
k c d3 6 d3 2
ht − ∇ h= ∇ pext . (5.65)
24µ 24µ
By analogy with the optical trapping result, we obtain
d3 q2 1
Υ(q, ω ) = . (5.66)
24µ iω + kc d3 q6
24µ
From the generalised Fluctuation-Dissipation theorem (5.50) we ob-

tain the power spectrum
k B Td3 q2 /12µ
⟨|δd(q, ω )|2 ⟩ = 3 2 . (5.67)
ω 2 + k24µ
cd
q12
By a straightforward calculation, the correlation function of inter-

est is just an inverse transform of the power spectrum,
d2 q
Z
C ( R, ω ) = ⟨|δd(q, ω )2 ⟩eiq·R . (5.68)
(2π )2
Written out explicitly, we obtain

Z ∞
k B Td3 eiqR cos θ q3 dq
Z 2π
C ( R, ω ) = dθ . (5.69)
(2π )3 6µ
3 2
0 0 2 kc d 12
ω + 24µ q
By a change of variables to k = ℓ(ω )q, this can be written as

Z ∞ ik( R/ℓ) cos θ 3
k B Td3
Z 2π
e k dk
C ( R, ω ) = dθ , (5.70)
(2π )3 6µω 2 ℓ4 0 0 1 + k12
The θ integral can be done using the relation

Z 2π
eikξ cos θ = J0 (kξ ), (5.71)
0
where J0 is the zeroth order Bessel function. To normalise the corre-

lation function, we examine the remaining integral at R/ℓ = 0,
Z ∞ Z ∞
dk k3 1 du π
= = √ , (5.72)
0 1 + k12 4 0 1 + u3 6 3
yielding a result that can be written in the compact form

√
3 k B T ℓ2 ( ω )

R
C ( R, ω ) = G , (5.73)
8π 4 k c ω ℓ(ω )
where the scaling function G (χ) is

√ Z
6 3 ∞ J0 (χq)
G (χ) = dq q3 , (5.74)
π 0 1 + q12
and the prefactor sets G (0) = 1. Figure 5.11 shows the scaling func-
tion G (ξ ) obtained by numerical integration of (5.74). This result con-
1 Figure 5.11: Scaling function G of

the spatial correlation function of
membrane fluctuations as a function
of distance scaled by elastohydrody-
namic length ℓ(ω ). Obtained with
0.5
G(@)
Figure_511_flicker.m.
0 2: 4:
@ = R=`(!)
firms that the correlation function depends only on the ratio R/ℓ(ω ),
as found experimentally, and that therefore ℓ(ω ) is the length scale
governing the decay of correlations. The prefactor ℓ2 (ω )/ω displays
the ω −4/3 dependence seen in experiment, and the scaling function
G (χ) shows the exact experimental form shown in Fig. 5.10(b).
5.3 Shapes and shape transformations
In this last section of the chapter we study two problems involving

the shapes of elastic membranes. These are chosen to illustrate some
key points regarding the physics of bending and the competition
between elasticity and tension.
5.3.1 Buckling of two-dimensional vesicles under pressure

The general problem of finding the equilibrium shapes of fluid mem-
branes governed by the Helfrich elastic energy (2.66) is a challenging
one by virtue of the strong nonlinearities in the governing equations
and the typical difficulties associated with the differential geome-
try of surfaces. To gain insight into some of the basic physics in-
volved, we study the problem of "two-dimensional vesicles", which 12
Tadjbakhsh, I. and Odeh, F. Equilib-
are treated as closed elastic filaments governed by the bending en- rium States of Elastic Rings. J. Math.
Anal. Appl., 18:59–74, 1967
ergy (4.27), subject to a pressure difference between the fluid en- 13
Foster, B., Verschueren, N., Knobloch,
closed and that outside, and to a uniform tension. The study of E. and Gordillo, L. Pressure-driven
this minimisation problem has a long history, with early work in the wrinkling of soft inner-lined tubes. New
J. Phys., 24:013026, 2022
1960s12 and continuing to the present.13
The equilibrium condition for the problem at hand is the normal

component of the force on the filament vanish. Expressed in terms
of the curvature κ the governing equation is

1 3
A κss + κ − γκ − Π = 0, (5.75)
2
where γ is the tension and Π is the pressure difference. We will

study the stability of a circular ground state of some radius R0 , and
rescale the various quantities using R0 and A via
γR20 ΠR30 s
K = κR0 , T= , P= , η= , (5.76)
A A R0
where η ∈ [0, 2π ], to obtain
1
Kηη + K3 − TK − P = 0, (5.77)
2
or, equivalently,
1
θηηη + θη3 − Tθη − P = 0. (5.78)
2
Given the curvature as a solution to (5.77), we obtain the θ via
Z η
θ (η ) = K ( η ′ ), (5.79)
0
and use θ to obtain the coordinates of the curve as

Z η Z η
x ( η ) = x0 + dη ′ cos θ (η ′ ), y ( η ) = y0 + dη ′ sin θ (η ′ ), (5.80)
0 0
where the location (x0 , y0 ) of the origin is arbitrary.

To make progress with this problem, we assume that the solution
θ (η ) as well as the tension T and pressure P can be expanded in
powers of a parameter ϵ ≪ 1 that serves as a bookkeeping device
to keep track of orders. We can assume that system breaks circular
symmetry into a shape with m lobes, with m ≥ 2 (recalling that
m = 1 corresponds to an infinitesimal translation, and m = 0 is a
dilation. The requirement of symmetry under rotations by half a
wavelength dictates that both T and P have only even terms in ϵ,
θ (η ) = θ (0) (η ) + ϵθ (1) (η ) + ϵ2 θ (2) (η ) + · · ·

T = T (0) + ϵ 2 T (2) + · · ·
P = P (0) + ϵ 2 P (2) + · · · . (5.81)
Direct substitution yields the zeroth order result
1
θ (0) (η ) = η, T (0) = − P (0) , (5.82)
2
a circle of unit radius (K (0) = 1), with a base tension determined

by the inward pressure P(0) . The fact that T (0) < 0 for P(0) > 1/2
indicates the possibility of a buckling instability. At order ϵ we find
Lθ (1) (η ) = 0, (5.83)
where the operator L is

L = ∂ηηη + P(0) + 1 ∂η ,
L = ∂ηηη + m2 ∂η . (5.84)
This has the simple solution
θ (1) (η ) = sin(mη ). (5.85)
We L as the derivative with respect to η of the operator that occurs

in the Euler buckling problem (4.63), showing further that this mem-
brane problem is related to a buckling instability.
Moving to order ϵ2 the second order perturbation must solve
3
Lθ (2) = P(2) + T (2) − m2 cos2 (ms), (5.86)
2
(1) 2
where the final term arises from a contribution (3/2)θη . Here we
must be careful to recognise that since cos2 (ms) = (1 + cos(2ms))/2,
the r.h.s. of (5.86) has a contribution in the nullspace of the operator
L, which involves the functions θ (η ) = 1, cos(ms), sin(ms). For there
to be a bounded solution at this order, the coefficients of all such
terms on the r.h.s. must vanish. Here, that condition requires the
constant term to vanish, and thus
3 2
P (2) + T (2) = m . (5.87)
4
At this order we do not have separate information on P(2) and T (2) .

The inhomogeneous solution to (5.86) is then
1
θ (2) ( η ) = sin(2mη ). (5.88)
8m
At cubic order in ϵ we note that the r.h.s has the terms

1 (1) 2 (1) (2) (1)
− θη + 3θη θη + T ( 2) θ η . (5.89)
2
Using the previously obtained results and now requiring the absence
of a term proportional to cos(ms) we find
3 2
T (2) = m +1 , (5.90)
8
so P(2) = (3/8) m2 − 1 . This completes the analysis to O(ϵ2 ).

The shapes that result from the function
ϵ2
θ (η ) = ϵ sin(ms) + sin(2ms) (5.91)
8m
for m = 2 are shown in Fig. 5.12 for increasing values of ϵ. At the

largest value, the calculated shape is remarkably similar to the cross-
section of the biconcave discoid shape of the red blood cell.
3 2:
1
2
K(2)
3(2)
1 : 0
Y
0
-1
-1 0 -1 0 1
0 : 2: 0 : 2: X
2 2
Figure 5.12: Shapes of two-dimensional
vesicles under pressure. Panels show
(a) scaled curvature K (η ), (b) tangent
5.3.2 Tether formation angle θ (η ), and (c) shape of vesicle
for three values of ϵ (ϵ = 0 (black),
A classic problem in the biological physics of membranes is the for- 0.5 (blue) and 1 (red). Obtained with
Figure_512_2dvesicles.m.
mation of "tethers" under the action of a localized force. There are
many examples of this phenomenon in living cells, typically associ-
14
Fygenson, D.K., Marko, J.F. and
ated with cytoskeletal filaments pushing against the cell membrane. Libchaber, A. Mechanics of micro-
A controlled example of this was achieved some time ago14 using an tubule-based membrane extension.
optical trap to pull a tether from a vesicle (Fig. 5.13). In what fol- Phys. Rev. Lett., 79:4497–4500, 1997
Figure 5.13: Equilibrium shapes of a bi-

layer lipid vesicle as the force exerted
by an optical trap is increased (left to
right) and a tether is produced. From
Ref. 14.
lows we outline the basic features of a model for tether formation in

15
Evans, E. and Yeung, A. Hidden
elastic membranes subject to large tension, as is the case in these and dynamics in rapid changes of bilayer
earlier experiments involving tethers extracted from vesicles.15 shape. Chem. Phys. Lipids, 73:39–56, 1994
Here we work through a calculation of perhaps the simplest model
for tether formation,16 in which the object that pokes the membrane 16
Powers, T.R., Huber, G. and Gold-
is abstracted to a ring of small radius that provides a suitable bound- stein, R.E. Fluid-membrane tethers:
minimal surfaces and elastic boundary
ary condition for the surface. Mathematically the problem is defined layers. Phys. Rev. E, 65:041901, 2002
as the search for energy minimisers of a functional that combines the
bending elasticity of fluid membranes and a tension associated with
contact of the membrane with a reservoir at some specified chemical
potential, given boundary conditions at a large loop of radius R that
represents a far-field lipid vesicle, for instance, and a much smaller
loop of radius r0 .
Within this picture of a tether, it is necessary first to understand
catenoids that span two loops of differing radii. We consider a setup
in which a ring of radius R lies in the plane z = 0 and a smaller one
of radius r0 is in the plane z = d. This calculation requires that we
generalise the up-down symmetric catenoid solution (5.15) to include
a second parameter, which we write in the form
z−c

ζ (z) = a cosh . (5.92)
a
The boundary condition at z = 0 implies that c/a = cosh−1 ( R/a),

which, when combined with that at z = d yields
p
d 2 2
d
r0 = R cosh − R − a sinh . (5.93)
a a
Although formidable in appearance, this is simply a quadratic equa-

tion in x = exp(d/a) whose solution can be put in the form
p !
1 β ± β2 − α2
D± = α ln √ , (5.94)
2 1 − 1 − α2
where, as before D = d/R and α = a/R, and now β = r0 /R is the

size ratio of the loops. A graphical representation of these solutions
0.5
is shown in Fig. 5.14 in which we may interpret the curves as follows:
0:8 1
D
0:6
for each chosen loop ratio β the parameter α necessary to match the 0:2
0:4
boundary conditions varies in the range (0, β) along the lower branch 0: 0
5
of each loop as the rings are separated from D = 0 up to some 0

0 0.5 1
maximum spacing indicated by a circle. That maximum spacing is ,
very nearly linear with β, with a slope close to unity; we conclude Figure 5.14: Trajectories of (5.94)
in the D − α plane for various
that catenoids spanning two different sized loops exist only up to values of the ring radius ratio
a height that is approximately the radius of the smaller loop. A β, as indicated. Obtained with
Figure_514_and_515_asymcatenoid.m.
montage of such asymmetric catenoids is shown in Fig. 5.15
Figure 5.15: Catenoids of maxi-

mum height for a range of asym-
metry parameters β. Obtained with
Figure_514_and_515_asymcatenoid.m.
The problem of interest is to find the minimum of the functional
1
Z Z
E= kc dS(2H )2 + σ dS (5.95)
2
for an axisymmetric surface with boundary conditions ζ (0) = R,
ζ (d) = r0 , H (0) = 0 and the tangent plane at z = d is parallel to
the plane. The Euler-Lagrange equation for the Helfrich functional
is an amusing exercise in differential geometry which we leave to
the student. The result can be expressed in a compact form that has
obvious similarities to the equivalent result for planar vesicles (5.75),

k c ∇2 H + 2H 3 − 2HK − σH + p = 0, (5.96)
where p is the pressure difference between the the inside and outside
of the tether. Here ∇2 is the covariant Laplacian. We shall ignore any
pressure difference in this problem, as it can be shown to introduce
only minor corrections to the main results.
Rescaling all lengths by the bottom hoop radius R, we obtain

ϵ ∇2 H + 2H 3 − 2HK − H = 0, (5.97)
where
kc
ϵ= . (5.98)
σR2
We shall be interested in the large-tension limit ϵ ≪ 1, as this gives
narrow tethers. Written this way, with the small parameter multiply-
ing the highest derivative in the problem, it is clear that a boundary
layer is likely to form, as the introduction of bending energy is a
singular perturbation. The limit ϵ = 0 is just the minimal surface
problem, so the choice H = 0 at the bottom loop is consistent with
that limit, and no boundary layer will form there. But the tangent
condition at the upper loop is inconsistent with H = 0 and a bound-
ary layer will form there. Such a layer will be such that the elastic
and surface tension contributions balance (ϵ∇2 H ∼ H, giving a (di-
√ √
mensionless) length scale ∼ ϵ (dimensionally, R ϵ).
To understand the boundary layer problem, we first work in the
Monge representation that is suitable for shallow surfaces slightly
deformed from the disc at z = 0, so D ≪ 1. A useful pair of coor-
dinates (ξ 1 , ξ 2 ) to describe the surface involves the arclength s along
the meridian and the angle φ around the z-axis, so that points on the
surface are given by
x(s, φ) = r (s)êr + z(s)êz , (5.99)
with rs2 + z2s = 1, as s is arclength. The first fundamental form is then
gij dξ i dξ j = ds2 + r2 dφ2 , (5.100)
while the second fundamental form is
Kij dξ i dξ j = (zs rss − rs zss ) ds2 − rzs dφ2 . (5.101)
With the previous convention for raising and lowering indices and
gik gkj = δji one finds the mean curvature

1 ij 1 rss zs
H= g Kij = − , (5.102)
2 2 zs r
Gaussian curvature
rss
K=− , (5.103)
r
and Laplacian
1 √ 1 d d
∇2 = √ ∂i gij g∂ j = r . (5.104)
g r ds ds
Thus, if the surface is described by the height function z(r ) above the
plane z = 0, then to leading order s ∼ r, H ∼ (1/2)∇2 z, and the
shape equation becomes
ϵ∇4 z − ∇2 z = 0, (5.105)
where
21 d d
∇ = r . (5.106)
r dr dr
To complete the specification of this problem we enumerate the bound-

ary conditions. At r = 1 we have z = 0 and ∇2 z = 0, while at r = r0
we have z = d and dz/dr = 0.
We solve (5.105) order by order in ϵ. At O(ϵ0 ) we are dealing with
the "outer" solution, for which ∇2 zout (r ) = 0, which is solved by
zout (r ) = b1 + b2 ln r. (5.107)
The condition at r = 1 sets b1 = 0, but the zero moment condition

does not fix b2 ; it is determined at higher order. This outer solution
diverges at the inner ring in the limit r0 → 0, an unphysical feature
corrected by the inner solution. To find the inner solution we rescale
the radial coordinate to balance the elastic and tension terms,
r
ρ= √ , (5.108)
ϵ
which leads to the amusing equation

1 d d 1 d d
ρ ρ zin + zin = 0. (5.109)
ρ dρ dρ ρ dρ dρ
This is solved by
zin
= c1 + c2 ln ρ + c3 I0 (ρ) + c4 K0 (ρ), (5.110)
d
where I0 and K0 are modified Bessel functions. The coefficient c3
vanishes due to the divergence of I0 as ρ → 0. The boundary condi-
tions at r = r0 yield two constraints, from which we deduce

zin ρ
= 1 + c4 K0 (ρ) − K0 (ρ0 ) + ρ0 K1 (ρ0 ) ln , (5.111)
d ρ0
√
where ρ0 = r0 / ϵ. Matching zin and zout , the constant terms of zin
must vanish and the coefficient of the log term matches b2 , yielding
zout ρ0 K1 (ρ0 ) ln r
= (5.112)
d ρ0 K1 (ρ0 ) ln r0 + K0 (ρ0 )
and
zin ρ0 K1 (ρ0 ) ln r + K0 (ρ)
= (5.113)
d ρ0 K1 (ρ0 ) ln r0 + K0 (ρ0 ) .
1 Figure 5.16: Shapes of poked mem-

branes in the Monge representation.
Curves are for β = 0.1 and ϵ = 0.005.
Solid blue curve is full numerical solu-
tion of (5.105), dashed red is asymptotic
z(r)=2d
solution zcomposite (r ) and black curve is

0.5 the β → 0 limit (5.114). Obtained with
Figure_516_and_517_tethers.m.
0
-1 -0.5 0 0.5 1
r
The composite solution is obtained by adding the two and sub-

tracting the common part, which leads to the simple result zcomp =
zin . It is then possible to take the limit r0 → 0 and obtain
√
zcomp ln r + K0 r/ ϵ
lim = √ , (5.114)
r0 →0 d −γ + ln 2 ϵ
where γ = 0.5772 . . . is Euler’s constant. Figure 5.16 compares the
full, composite, and limiting composite solutions.
To go beyond this linearised problem and address the formation of
a true tether with the observed property of a constant radius along
its length, it is generally necessary to invoke numerical methods,
particularly to connect such a solution to that obtained above in the
small-amplitude limit. Figure 5.17 shows examples of such a cal-
culation, illustrating that indeed a constant-diameter tether emerges
from the catenoidal solution at the supporting ring.
1 Figure 5.17: Numerical solution of

full tether problem for ϵ = 0.001
and r0 = 0.1 for various values of
the ring separation d. Obtained with
Figure_516_and_517_tethers.m.
0.5
z
-1 -0.5 0 0.5 1
r(z)
Analytical results can be obtained if we focus on the fully-developed

tether itself. First, in a straight tether, the equilibrium condition
(5.97) simplifies to 2ϵH 3 − H = 0, and since in our rescaled units
H = − R/2a for a cylinder of radius a, we deduce the tether radius
r r
ϵ kc
a= R= , (5.115)
2 2σ
which is indeed ≪ R in the high-tension limit.
Second, it is straightforward to develop a stability analysis around
this cylindrical solution, writing r (z) = a + u(z) and expanding
Z Z
ϵ
E= dS(2H )2 + dS (5.116)
2
p
to O(u2 ), using dz/ds = 1 + r ′ (z)2 . One finds
au2
Z
h ϵ ϵ
E =2π dz a + + 1− 2 u+ z
2a 2a 2
ϵa

u 2 i
+ u2zz + 4 + · · · , (5.117)
2 a
where a term that is a total derivative has been omitted. We observe

that the first bracketed term in (5.117) is minimised at the equilibrium
value of a, at which the second bracketed term vanishes identically,
and that the third bracketed term involves two contributions that
are each manifestly positive. Thus, at quadratic order the energy
is larger than the straight tether and therefore the tether is stable.
Thus, while a tether without any bending energy that is longer than
its circumference is unstable to the Rayleigh instability, one that is
at the equilibrium size determined by the balance of tension and
elasticity is stable for arbitrary lengths!
6
Cellular Motion (Lectures 17-20)
In this section we continue to move up the scale of lengths in biolog-

ical physics to consider the motion of uni- and multicellular organ-
isms.1 After an overview of the locomotion of flagellated prokaryotic 1
E. Lauga. The Fluid Dynamics of Cell
and eukaryotic cells we review historically important models devel- Motility. Cambridge University Press,
Cambridge, UK, 2020
oped to understand the link between flagellar motion and swim-
ming, as well as models that illustrate how hydrodynamic coupling
between those appendages can lead to phase synchronisation. This
motivates a discussion of the relationship between cell swimming
and the elementary force singularities of Stokes flow—the stokeslet
and the stresslet—and how those singularities can be used to un-
derstand a number of basic experimental observations. We conclude
with a discussion of some collective effects involving active matter.
6.1 Cell swimming
Cell motion can take many forms, from crawling on solid substrates
by means of membrane extensions ("filopodia" or "lamellipodia"), mi-
gration within confluent tissues, to swimming through fluids through
the action of moving appendages. Within this course we focus on
swimming, as it provides a context that is rich in phenomenology
and has proven to be amenable to substantial quantitative analysis.
250 μm
Figure 6.1: Examples of microswim-

mers. Left to right: E. coli, Parame-
The broad classification of living organisms into prokaryotes (cells cium bursara, Chlamydomonas reinhardtii,
without a membrane-bound nucleus) and eukaryotes (possessing a Volvox carteri.
nucleus) serves also to distinguish typical means of swimming seen
among unicellular and small multicellular organisms. As shown in
Fig. 6.1, prokaryotes consist of a cell body whose dimensions are
typically 1 − 10 µm, from which emanate one or more helical flag-
ella several times the body length or more. These are rotated at
frequencies that can reach 102 − 103 s−1 by the action of rotary mo-
tors embedded in the cell wall, powered by a flux of proteins passing
through them. Eukaryotes, as exemplified by uni- and multicellular
algae, choanoflagellates, etc. achieve motility through the undula-
tions of flagella driven by the action of motor proteins that slide the
underlying filaments within the flagellum past each other through
conformational changes powered by ATP consumption. These wave-
forms consist typically of a "power stroke", with a nearly straight
flagellum sweeping through a large angle, followed by a "recovery
stroke", in which a tightly curled flagellum is brought back to its
starting point by moving close to the cell body. Both large unicellular
organisms such as Paramecium and certain multicellular organisms
composed of tens, hundreds, or even thousands of flagellated cells
are often termed "ciliates", and the collective beating of their flag-
ella may exhibit perfect frequency and phase locking, or frequency
locking and phase modulations termed "metachronal waves".
The regime of Reynolds numbers inhabited by various swimming
organisms is summarized in Table 6.1; there is a clear distinction
between the Stokesian regime of the flagellated organisms in Fig. 6.1
and that of the larger swimmers familiar to us on a human scale.
organism length speed frequency Re

bacterium 10 µm 10 µm/s 100 Hz 10−4
spermatozoon 100 µm 100 µm/s 10 Hz 10−2
ciliate 100 µm 1 mm/s 10 Hz 10−1
tadpole 1 cm 10 cm/s 10 Hz 103
small fish 10 cm 10 cm/s 10 Hz 104
penguin 10 m 1 m/s 1 Hz 106
sperm whale 10 m 1 m/s 0.1 Hz 107
Table 6.1: Reynolds numbers for vari-

ous swimming organisms.
6.1.1 The general swimming problem

From a mathematical perspective, our goal is to solve the Stokes
equations for incompressible flow,
U+Ωxr
2
∇ p = µ∇ u, ∇ · u = 0, (6.1)
where p is the pressure, with a boundary condition that u on the

body equals the specified swimming "gait" uS , as in Fig. 6.2. With uS
uS known, the unknowns are the translational velocity U and the Figure 6.2: A generic swimmer with
angular velocity Ω. We assume that the fluid is Newtonian, with some reference shape (dashed) de-
formed by an actuation velocity uS .
viscosity µ and stress tensor

σ = − pI + µ ∇u + ∇uT , (6.2)
In the absence of any body forces and torques acting on the swim-
mer, the unknowns are found by demanding that the total force and
torque on the swimmer vanish,
ZZ
F( t ) = dS σ · n̂ = 0, (6.3a)
Z ZS
T( t ) = dS r × (σ · n̂) = 0, (6.3b)
S
cellular motion (lectures 17-20) 105
where n̂ is the outward normal to the body.

It is often of interest to calculate the power generated by propul-
sive movements. The rate of working of the swimmer is
ZZ
Ẇ = − dS u · σ · n̂. (6.4)
S
Standard manipulations using the divergence theorem allow us to

re-express this result as the volume integral
ZZZ
∂
Ẇ = dV uσ . (6.5)
V ∂x j i ij
Using the incompressibility of the fluid and the symmetry of the

strain rate Eij = (1/2) ∂ j ui + ∂i u j , we obtain the familiar result 2
Drescher, K. and Goldstein, R.E. and
ZZZ Michel, N. and Polin, M. and Tuval,
Ẇ = 2µ dV Eij Eij . (6.6) I. Direct Measurement of the Flow
V Field around Swimming Microorgan-
isms. Phys. Rev. Lett., 105:168101, 2010
Thus, we can ascertain the dissipation by measuring the fluid veloc- 3
Guasto, J.S. and Johnson, K.A. and
ity field outside of the body,2 , 3 or by tracking flagella movements.4 Gollub, J.P. Oscillatory Flows Induced
by Microorganisms Swimming in Two
Dimensions. Phys. Rev. Lett., 105:168102,
6.1.2 Rate independence and scallop theorem 2010
4
Brumley, D.R., Wan, K.Y., Polin, M.
There are two important properties of low Reynolds number propul- and Goldstein, R.E. Flagellar synchro-
nization through direct hydrodynamic
sion. The first concerns the dynamics of surface deformations:
interactions. eLife, 3:e02750, 2014
The distance travelled by a low-Reynolds number swimmer as its sur-
face changes from one configuration to another does not depend on
the rate at which the surface deformation occurs, but only on its ge-
ometry. For any periodic deformation of a swimmer, the total distance
travelled per period is therefore only a function of the sequence of t1
shapes displayed by the swimmer during one period.
shapes rs(t)
To establish this, we appeal to Fig. 6.3 where we depict a sequence
of shapes rs (t) that runs from some time t0 to t1 , along with a second tʹ1
sequence r′s (t′ ) that involves visiting the same shape sequence as rs (t)
but with a different time parameterization. Thus, as indicated by the
t0
dashed arrows, there exists some mapping between any two shape
shapes rsʹ (tʹ)
along the sequences. The swimming velocity U(t) can be expressed
as a surface integral of deformation velocity of the shape, tʹ0
ZZ Figure 6.3: Two different paths in shape
U( t ) = dSH(rs ) · ṙs , (6.7) space.
S
for some tensor H(r) that does not depend explicitly on time and is
found as a solution to the swimming problem.
The total distance travelled is
Z t
1
∆= dt U(t), (6.8)
t0
and the distance along the primed trajectory is

Z t′ Z t′ Z Z
dr′

1 1
∆′ = U′ (t′ )dt′ = dS′ H(r′s ) · s′ dt′
t0′ t0′ S′ dt
Z t
1
Z Z ′
dr dt
′
= dt′ dSH(r′s ) · s′ dt
t0 S′ dt dt
Z t Z Z
1 drs
= dSH(rs ) · dt = ∆, (6.9)
t0 S′ dt
thus proving the rate independence.

The second property of interest is know as the "scallop theorem": 5
Purcell, E.M. Life at low Reynolds
If a sequence of shapes displayed by a swimmer deforming in a time number. Am. J. Phys., 45:1–11, 1977
periodic fashion is identical when viewed under a time-reversal trans-
(a)
formation, the swimmer cannot move on average.
As first stated by Purcell,5 this has a key consequence:

(b)
A single degree of freedom cannot swim at low Reynolds numbers since
one-degree-of-freedom kinematics is necessarily reciprocal.
This is illustrated in Fig. 6.4. Neither (a) a pair of spheres that cycli-
cally move together and apart, nor (b) a hinged object that cyclically
(c)
opens and closes (i.e. a scallop) can achieve net locomotion. But, (c)
a three-link swimmer can swim by following a cycle of shapes.
The need to break time-reversal invariance to achieve locomotion
explains the ubiquitous appearance of helical prokaryotic flagella ro-
tated along their long axis or undulating eukaryotic flagella with
either sperm-like sinusoidal traveling waveforms (as we saw in our Figure 6.4: Swimming sequences and
the scallop theorem.
study of wiggling elastica in Sec. 4.3.2), or two distinct phases of
beating with very different waveforms. Two classic examples of the
latter is shown in Fig. 6.5, taken from the very early study by James
Gray of the ciliated gills of the blue mussel Mytilus edulis6 and very 6
J. Gray. The mechanism of ciliary
movement. Proc. R. Soc. Lond. B, 93:104–
recent work on the green alga Chlamydomonas.7 The first phase of
121, 1922
a beat cycle is the "power stroke", in which a nearly straight cilium 7
Leptos, K.C. and Chioccioli, M. and
roughly pivots around its attachment point to the underlying tis- Furlan, S. and Pesci, A.I. and Gold-
stein, R.E. Phototaxis of Chlamydomonas
sue or cell, followed by a "recovery stroke" in which the waveform is
arises from a tuned adaptive pho-
highly curled and returns to the original position close to the surface. toresponse shared with multicellular
Volvocine green algae. Phys. Rev. E, 107:
014404, 2023
Figure 6.5: Ciliary waveforms. Left
panel: sketch of the waveform in the
mussel Mytilus edulis. (a) Power stroke.
(b) Recovery stroke. From Ref. 6.
Right panel: flagellar waveforms of
the two flagella of Chlamydomonas rein-
hardtii. Cis (red) and trans (blue) flagella
are shown during one complete cycle.
From Ref. 7.
6.2 Flagella and the physics of low-Re propulsion
First we study the simplest models with which to understand self-

propulsion in Stokes flow, including Taylor’s "waving-sheet" model,
to the approximation known as Resistive Force Theory, Lighthill’s
analysis of undulating waveforms and the use of RFT to quantify
bacterial swimming by rotating helical flagella. 8
Rothschild, L. Measurement of Sperm
Activity before Artificial Insemination.
6.2.1 Taylor’s model a single waving sheet Nature, 163:358–359, 1949
9
Taylor, G.I. Analysis of the swimming
Inspired by Lord Rothschild’s observation8 that nearby swimming of microscopic organisms. Proc. R. Soc.
A, 209:447–461, 1951
sperm cells synchronised the beating of their flagella, Taylor in 19519
developed a highly-simplified model for low Reynolds number swim-

y
ming that consists of a two-dimensionally infinite material sheet in
x
the configuration of a wave of deformation traveling along one di- z
rection, as in a carpet with periodic "rucks" (Fig. 6.6). These defor- U c

mations are a consequence of unspecified internal forces. (xs,ys)
Let xs = ( xs , ys ) be the coordinates of material points on the sheet,

Figure 6.6: Geometry of Taylor’s wav-
which supports undulations with wavevector k and frequency ω, ing sheet model.
moving at speed c = ω/k. We expect the sheet is propelled to the
left with velocity U = −U êx , where U is as yet unknown. In the
laboratory frame, the fluid is at rest as y → ±∞, but in the reference
frame moving with the swimming speed the fluid flow far above
or below the sheet is U∞ = U êx , the x coordinate of a given point
will not change, while the y coordinate will simply oscillate up and
down as ys = B sin(kx − ωt). We assume the long-wavelength limit
ϵ ≡ Bk ≪ 1, and use ϵ as the small parameter in a perturbative ap-
proach. The frequency and wavevector of the traveling wave serve
as natural spatial and temporal scales for nondimensionalising the
problem, so the rescaled sheet coordinates are xs = x and
ys = ϵ sin( x − t). (6.10)
The problem at hand is to find u in the half space y ≥ ys (the

solution for y < ys follows from it trivially), subject to two boundary
conditions. The first is a connection between the Eulerian description
of the flow and the Lagrangian description of the material,
∂xs
u xs
= (6.11)
∂t
Second, the fluid velocity must asymptote to u = U êx as y → ∞.
Assuming translational invariance along z, we have a two-dimensional
problem ( x, y), and introduce a stream function ψ( x, y) such that
∂ψ ∂ψ
ux = , uy = − , (6.12)
∂y ∂x
which guarantees the incompressibility condition ∇ · u = 0. (Note
this definition of ψ has a minus sign relative to Taylor’s choice). The
vorticity ω = ω êz is out of the plane and its magnitude satisfies
ω = −∇2 ψ. Together with the relation ∇2 ω = 0 from the Stokes
equation, we obtain the biharmonic equation for the stream function,
∇4 ψ = 0. (6.13)
We must now solve (6.13) subject to four boundary conditions,
∂ψ( xs , ys )
u x ( xs , ys ) = = 0, (6.14a)
∂y
∂ψ( xs , ys ) ∂ys
uy ( xs , ys ) = − = = −ϵ cos ϕ, (6.14b)
∂x ∂t
∂ψ( x, y → ∞)
u x ( x, y → ∞) = = U, (6.14c)
∂y
∂ψ(y → ∞)
uy ( x, y → ∞) = − = 0, (6.14d)
∂x
with the shorthand notation ϕ = x − t. While the small parameter ϵ

appears explicitly in (6.14b), it also appears implicitly in both (6.14a)
and (6.14b) by virtue of the condition applying at ys = ϵ sin ϕ.
To solve the above we assume that both ψ and the unknown swim-
ming speed U have regular expansions in ϵ,
U = ϵU (1) + ϵ2 U (2) + · · · , ψ = ϵψ(1) + ϵ2 ψ(2) + · · · , (6.15)
and note that ∇4 ψ(n) = 0 ∀n. When expanding the boundary con-
ditions evaluated on the sheet we make use of relationships familiar
from the boundary perturbation theory used in Sec. 2.4.2, such as
∂ψ(1)
ψ(1) ( xs , ys ) = ψ(1) ( xs , 0) + ϵ sin ϕ ( x s , 0) + · · · . (6.16)
∂y
Collecting terms, we find at O(ϵ) that
∂ψ(1) ∂ψ(1)
( x, 0) = 0, ( x, 0) = cos ϕ,
∂y ∂x
∂ψ(1) ∂ψ(1)
( x, ∞) = 0, ( x, 0) = U (1) . (6.17)
∂x ∂y
The general solution of the biharmonic equation consists of a sum

of polynomial terms plus an infinite series of separable solutions,
ψ =ψ0 + ky + Gy2 + Hy3

∞
∑ (Cn + Dn y) e−ny + ( En + Fn y) eny einϕ .

+ (6.18)
n =1
The constant ψ0 is irrelevant, as ψ is only defined up to an arbitrary

constant. The term Gy2 is associated with a shear flow at infinity,
which is absent. Absent as well is the Poiseuille-like flow associated
with the term Hy3 . The linear term produces a uniform flow at
infinity, and we identify k = U (1) . In the present case, the forcing
from the boundary conditions (6.17) is 2π-periodic, so only the n = 1
terms in (6.18) are relevant. Ignoring the growing exponential terms
and imposing the boundary conditions (6.17), we find
ψ(1) = (1 + y) e−y sin ϕ, U (1) = 0. (6.19)
There is no swimming speed at this order.

At O(ϵ2 ) we encounter boundary conditions such as
∂ψ(2) ∂ 2 ψ (1)
( x, 0) = − sin ϕ ( x, 0), (6.20)
∂y ∂y2
which are proportional to sin2 ϕ. This and the other boundary con-
dition at ( x, 0) force the system at its first harmonic, leading to
ψ(2) = U (2) y + (C2 + D2 y) e−2y e2iϕ . (6.21)
Rather than solve directly for the unknown coefficients, we observe

that if we average ∂ψ(2) /∂y over one wavelength of the traveling wave
(and denote the average by angular brackets), the periodic terms give
no contribution and thus
(2)
(2) ∂ψ
U = , (6.22)
∂y
a relation that holds at any value of y. We are free to choose a conve-

nient one, namely y = 0 where the boundary condition holds,
∂ψ(2) ∂ 2 ψ (1)

U (2) = ( x, 0) = − sin ϕ ( x, 0 ) . (6.23)
∂y ∂y2
A simple calculation yields U (2) = 1/2, or, in original units,

1 1
U= c ( Bk )2 = B2 kω. (6.24)
2 2
This is a central result of Taylor’s calculation. We conclude: (i) loco-
motion is possible without inertia, (ii) the swimming speed scales as
the square of the wave amplitude, and (iii) to motion is "retrograde",
opposite the travelling direction of the wave.
y
6.2.2 Taylor’s model; two coupled sheets z

x
Taylor was not only interested in the question of propulsion at van-

2h
ishing Reynolds number, but also in the synchronization of nearby
undulating filaments. Following his analysis, we extend the calcula- 2ϴ
tion to the case of two nearby sheets separated by a mean distance Figure 6.7: Two nearby waving sheets
separated on average by 2h and with a
2h and offset by a phase angle 2Θ, as in Fig. 6.7. Here we quote the
phase shift Θ.
essential results, leaving algebraic details to the reader.
Having rescaled x by the wavevector k in (6.10), if the two sheets
are separated on average by 2h, then we define the dimensionless
δ = kh, (6.25)
so in these units the sheets are located at
y = δ + y1 = δ + ϵ sin (ϕ + Θ) ,
y = −δ + y2 = −δ + ϵ sin (ϕ − Θ) , (6.26)
where ϕ = x − t, so the total phase shift is 2Θ. The generalisation

of the stream function form (6.18) to the case at hand (for the flow
between the sheets) is
ψ(1) = ( A1 y sinh y + B1 cosh y) cos Θ sin ϕ

+ ( A2 y cosh y + B2 sinh y) sin Θ cos ϕ. (6.27)
Imposing the boundary condition ∂ψ(1) /∂y = 0 at y = ±δ yields

B1 B2
= − (δ coth δ + 1) , = − (δ tanh δ + 1) . (6.28)
A1 A2
Similarly, imposing the condition ∂ψ(1) /∂x = cos(ϕ ± Θ) at y = ±δ

yields the pair of conditions
A1 δ sinh δ + B1 cosh δ = 1, A2 δ cosh δ + B2 sinh δ = 1. (6.29)

Solving (6.28) and (6.29) together gives the final results

sinh δ δ cosh δ + sinh δ
A1 = − , B1 = ,
sinh δ cosh δ + δ sinh δ cosh δ + δ
cosh δ δ sinh δ + cosh δ
A2 = − , B2 = . (6.30)
sinh δ cosh δ − δ sinh δ cosh δ − δ
We focus on the rate of energy dissipation in the system, for which
we need the pressure p, which can be obtained by integrating the
Stokes equations ∇ p = µ∇2 u. Assuming that the pressure has the
same form of expansion as the stream function (6.15), namely p =
ϵp(1) + · · · , a bit of algebra leads to
p (1) sinh y sinh δ cos Θ cos ϕ cosh y cosh δ sin Θ sin ϕ

= − , (6.31)
2µk sinh δ cosh δ + δ sinh δ cosh δ − δ
which, on the upper surface at y = δ, takes the value
(1)
p1
= α cos Θ cos ϕ − β sin Θ sin ϕ, (6.32)
2µk
where
sinh2 δ cosh2 δ
α= , β= . (6.33)
sinh δ cosh δ + δ sinh δ cosh δ − δ
The instantaneous rate Ẇ at which the upper sheet does work on the
fluid is − p(∂y/∂t), and since both p and y are O(ϵ), we expect an
expansion of Ẇ of the form Ẇ = ϵ2 Ẇ 2 + · · · , with
(1) ∂y1
Ẇ (2) = − p1 . (6.34)
∂t
If we user an overbar to denote an average over a temporal cycle
of oscillations, and define the (scaled) mean rate of working W as
µkW (2) = Ẇ (2) , then we obtain
W (Θ, δ) = α + ( β − α) sin2 Θ. (6.35)
Since it is evident from (6.33) that β > α, ∀δ, we deduce that W (Θ, δ)
is minimised for Θ = 0; the in-phase synchronised state has the
lowest dissipation rate.
Intuitively, since the dissipation rate is a spatial integral of the
squares of fluid velocity gradients, it stands to reason that the con-
figuration in which the separation between the sheets is nearly con-
stant throughout the waveform will be less dissipative than states 1
with large thickness variations. This can be made concrete by con-
sidering the ratio R(δ) = W (0, δ)/W (π/2, δ)) of dissipation rates
for in-phase and out-of-phase configurations. We find
R
0.5
sinh δ cosh δ − δ

R(δ) = tanh2 δ , (6.36)
sinh δ cosh δ + δ
which is plotted in Fig. 6.8. This ratio tends to unity at large separa- 0
0 1 2 3 4
tion, indicating in that limit that the two sheets do not interact and / = kh
the dissipation is independent of the phase shift. But for δ < 2 (i.e. Figure 6.8: Dissipation ratio in Tay-
lor’s calculation, as a function of
h/λ < 1/π ∼ 0.3), R drops to small values, signalling a significant scaled sheet separation. Obtained with
streamlining effect associated with synchronisation. Figure_608_Taylordissipation.m.
Thus we conclude that if nature chooses to minimise the dissipa-

tion, then Rothschild’s observations are explained. But we should
emphasise that while minimisation of dissipation may often hold in
dynamical systems, it is not a general principle by which to derive
results. Indeed, Taylor’s calculation does not actually include any
dynamics in the sense of allowing the phase shift itself to evolve
under the forces in the system. More recent work10 shows that intro- 10
Elfring, G.J. and Lauga, E. Synchro-
ducing elasticity into this problem can lead to an evolution toward nization of flexible sheets. J. Fluid Mech.,
674:163–173, 2011
synchrony. In Section 6.3 we return to the problem of synchronisa-
tion and study in detail a separate class of models that gives insight
into synchrony of objects such as cilia and flagella.
6.2.3 Resistive Force Theory and drag-based thrust

In our discussion of elastohydrodynamics in Sec. 4.2.1 we introduced
the basic ideas of Resistive Force Theory that lead to the general
relationships (4.23) and (4.99) between the shape of a filament (given
by its tangent vector), its local velocity relative rt to any background
flow u, and forces acting on it, restated here as a relation for the local
force f acting on the fluid,

f = ζ ∥ t̂t̂ + ζ ⊥ I − t̂t̂ · (rt − u) ,

(6.37)
where we will typically assume the large aspect ratio limit ζ ⊥ = 2ζ ∥ .

A heuristic argument for drag anisotropy involves recognizing
that the action of a slender filament on the fluid velocity can be
decomposed into the contributions of a distribution of point forces
along the filament.11 The flow field u that arises from an elementary 11
Keller, J.B. and Rubinow, S.I. Slender-
point F is termed a "stokeslet", and the two are linearly related by body theory for slow viscous flow. J.
Fluid Mech., 75:705–714, 1976
u(x) = G(x − x′ ) · F, (6.38)
where the "Oseen tensor" G(r) is

1
G(r) = (I + r̂r̂) (6.39)
8πµr
where r = |r| and r̂ = r/r is a unit vector. The intrinsic anisotropy of

the stokeslet flow field is seen directly from (6.39) and Fig. 6.9; the
flow to the side of a force (where F · r̂ = 0) has magnitude F/8πµr,
whereas that in front of the force is 2 × F/8πµr. Furthermore, if we
integrate along a filament to obtain the velocity, the relationship will Figure 6.9: Origin of drag anisotropy of
slender bodies. (a) Forces along a fila-
schematically be ment. (b) Forces perpendicular to the
F L1
Z
long axis of the filament.
u∼ ds, (6.40)
µ a r
where the filament radius a and length L provide natural cutoffs.
One then finds u ∼ ( F/µ) ln( L/a), and hence drag coefficients F/u
of the schematic form ζ ∼ µ/ ln( L/a), consistent with RFT.
We are now in a position to understand the crucial role of drag
anisotropy in the generation of net propulsion by both prokaryotic
and eukaryotic flagella. In the case of bacteria, propulsion is typi-
cally achieved by the rotation of one or more rigid helical flagella by
motor proteins embedded in the cell wall. For eukaryotes like sper-
matozoa and choanoflagellates, the flagellum has a sinusoidal shape
deformed by the sliding action of motor proteins along the length of
the filament. As shown in Fig. 6.10, in both cases we find segments
of the flagellum that are moving down when tilted upward from left
to right, and moving up when tilted downward from left to right.
Figure 6.10: Forces due to rotating

prokaryotic and undulating eukaryotic
flagella.
Consider the case shown in Fig. 6.11, and project the upward
speed U onto the tangent and normal directions, giving U∥ and U⊥ .
The drag forces F∥ and F⊥ are as indicated. The vectorial sum of par-
Figure 6.11: Normal and tangential
allel and perpendicular forces necessarily has a component (shown forces on a moving filament segment.
in green) that is to the left, opposite to the direction of the propaga-
tion of the traveling wave of deformation along the filament.
We can verify that this conclusion holds for any inclination angle
of the segment by considering a straight segment oriented as in Fig.

6.12. Here, the tangent vector is t̂ = cos θ, sin θ êy and the segment
velocity is u = (0, u). A simple calculation yields the force

f = ζ ⊥ − ζ ∥ u sin θ cos θ êx
h i
+ −ζ ⊥ u + ζ ⊥ − ζ ∥ u sin2 θ êy , (6.41) Figure 6.12: A tilted rod segment mov-
ing upwards.

and thus f · êx = ζ ⊥ − ζ ∥ u sin θ cos θ. We deduce that if u > 0
and sin θ < 0 we obtain r · êx < 0 and if u < 0, sin θ > 0 and again
r · êx < 0; the swimming force component is independent of the sign
of u and propulsion is directed forward everywhere. For a helix
every point is the same... for a waving flagellum they are different,
but all contribute in the same sense.
6.2.4 Propulsion by a traveling wave of deformation

Building on the analysis in the previous section, we next seek to 12
M.J. Lighthill. Mathematical Biofluiddy-
understand how drag anisotropy links to the propulsion of flagellar namics. Society for Industrial and Ap-
plied Mathematics, Philadelphia, 1975
waveforms that exhibit a traveling-wave dynamics, and follow the
analysis presented by Lighthill.12
In its simplest form the analysis pertains to an infinitely long
filament that is characterized by its RFT drag coefficients, the fre-
quency ω and wavelength λ of the traveling waves, as in Fig. 6.13.
In the laboratory frame there is an unknown swimming speed U =
F (ζ ⊥ , ζ ∥ , ω, λ, shape), where "shape" refers to the detailed waveform.

As there is only one natural speed in the problem, c = λω, we expect
y
U (a) lab frame U∞=0 x

= F (ρ, shape), (6.42) c
z
c
U
where ρ = ζ ∥ /ζ ⊥ and we leave open the possibility (discussed be-
low) that ρ may differ from the usual value 1/2. (b) wave frame U∞=U-c
The key to the analysis is a switch to a frame moving at speed
�
U − c, in which the undulating filament appears stationary. In this �
frame points along the curve are moving, and this is possible only Figure 6.13: Geometry of Lighthill’s cal-
if the system is like a conveyor belt in which points move along the culation of propulsion by a slender fila-
ment. (a) In the lab frame the filament
curve, with the velocity u(s along the tangent. Because the filament is has traveling waves to the right and
inextensible, the tangential speed must be the same at all s, and is the swims to the left. (b) In the wave frame
the filament is stationary and material
wave speed c multiplied by the ratio of arclength to wavelength along points move along the tangent vector.
one period, u(s) = −c(Λ/λ)t̂(s). Translating back to the laboratory
frame, the velocity of points on the filament is
Λ
u(s) = (c − U )êx − c t̂(s). (6.43)
λ

Since the force/length on the filament is ζ ⊥ I − ζ ⊥ − ζ ∥ t̂t̂ · u,
the local force component on the filament will be
2
f · êx = ζ ⊥ (c − U ) − ζ ∥ Q t̂ · êx − ζ ⊥ − ζ ∥ (c − U ) t̂ · êx , (6.44)
where Q = cΛλ. To enforce the condition of zero net force, by inte-

grating (6.44) over one period, we encounter the arclength integrals
2
of t̂ · êx and t̂ · êx . Since t̂ · êx = dx (s)/ds, the first integration triv-
ially yields Λ, but the second integral depends on the precise shape
2
of the filament. Since t̂ · êx ≤ 1, we can write the integral as βΛ
for some β ≤ 1. Solving for the speed we obtain

U ζ ⊥ − ζ ∥ (1 − β ) (1 − ρ ) (1 − β )
= = . (6.45)
c ζ − ζ −ζ β 1 − (1 − ρ ) β
⊥ ⊥ ∥
The is precisely the form (6.42) anticipated by dimensional analy-

sis, and shows that U > 0 for ρ < 1 as is the case within RFT; the
swimming motion is opposite the wave propagation direction. For
typical waveforms one finds U/c ∼ 0.2 − 0.3, implying a consid-
erable amount of "slip" relative to what one might term the "wood
screw" speed c (imagining turning a screw into a piece of wood,
where the advancement speed would be precisely the wave speed).
If, by some means, ρ > 1 then the propulsion speed changes sign.
This can happen when a flagellum has fine hairs orthogonal to the
main filament, termed "mastigonemes" which at to greatly increase
the effective longitudinal drag coefficient.
If we set y(s) = A sin ks as in Taylor’s calculation, and assume
( Ak)2 ≪ 1, then 1 − β ≃ ( Ak)2 /2 and
U 1
≃ ( Ak)2 , (6.46)
c 2
in complete agreement with Taylor’s result (6.24).
6.2.5 Bacterial locomotion

The analysis above provides a relationship between the speed of a
traveling wave of deformation moving along a filament and its swim-
ming speed. In this subsection we build on this result to understand
the swimming of a bacterium that rotates a helical flagellum. The
setup shown in Fig. 6.14 involves a cell swimming at speed U, whose
flagellum is rotated by a molecular motor embedded in the cell wall,
Figure 6.14: Kinematics of bacterial
and whose body rotates in the opposite direction, consistent with a swimming.
system with no net torque. If we let the body angular speed be Ω
and that of the flagellum be Ω + ω, we seek a relationship between
the helical geometry, these frequencies, and the swimming speed U.
We begin by specifying the helix properties as in Fig. 6.15; the fil-
ament has diameter 2a, describes a helix of wavelength λ and radius
R, rotating at some angular speed Ω. Because all points on the helix
Figure 6.15: Dynamics of a rotating he-

lix.
are inclined at the same angle θ with respect to the x −axis, we can
use the results derived in the context of Fig. 6.12, where the upward
speed of the segment shown is u = −ΩR, to deduce that the total
drag force acting along x is

F = f x L = − ζ ⊥ − ζ ∥ sin θ cos θRLΩ, (6.47)
while the total torque along x is

T = − f y RL = − ζ ⊥ cos2 θ + ζ ∥ sin2 θ R2 LΩ. (6.48)
For transition along x at speed U the translational drag is

F = − ζ ⊥ sin2 θ + ζ ∥ cos2 θ LU. (6.49)
These results can be summarized in a linear relationship between

forces and torques on the one hand, and speed and angular speed
on the other, of the form
! ! !
F A B U
=− , (6.50)
T B D Ω
helix helix
where

A = ζ ⊥ sin2 θ + ζ ∥ cos2 θ L, (6.51a)

B = ζ ⊥ − ζ ∥ sin θ cos θRL, (6.51b)

D = ζ ⊥ cos2 θ + ζ ∥ sin2 θ R2 L. (6.51c)
Clearly, A, D > 0, and if we have a left-handed helix then θ > 0, so

B > 0. We adopt the common convention that angular speeds are
taken to be positive when they appear as clockwise rotations when
viewed along the direction of angular velocity. Thus, if Ω < 0 then
the propulsive force is in the positive x direction.
The motion of the cell body itself will have a linear relationship of
the form (6.51), but with purely diagonal terms in the matrix,
! ! !
F A0 0 U
=− . (6.52)
T 0 D0 Ω
body body
Next we adopt the simplifying assumption that the dynamical prop-

erties of the composite system cell body + flagellum can be obtained
by adding the forces and torques that we have separately computed.
Apart from the assumptions of RFT, this superposition is also an
approximation because it does not take into account the near-field
contributions of each part to the flows around the other (this has
been done by Higdon.13 Nevertheless, this simplification has great 13
Higdon, J.J.L. The hydrodynamics of
heuristic value. Within this analysis, we take the angular speed of flagellar propulsion: helical waves. J.
Fluid Mech., 94:331–351, 1979
the helix to be Ω + ω, and the requirement that the total force and
torque vanish can be written as
! ! ! ! !
A0 0 U A B U 0
+ = , (6.53)
0 D0 Ω B D Ω+ω 0
or, equivalently,
! ! !
A0 + A B U Bω
=− (6.54)
B D0 + D Ω Dω.
This is readily solved to yield the swimming speed, body rotation

frequency and flagellum rotation frequency,
BD0
U= ω, (6.55a)
B2 − ( A0 + A) ( D0 + D )
D ( A0 + A ) − B2
Ω= ω, (6.55b)
B2
− ( A0 + A) ( D0 + D )
D ( A0 + A )
Ω+ω = 2 ω. (6.55c)
B − ( A0 + A) ( D0 + D )
The fact that the energy dissipation in (6.6) is positive leads to the
requirement that the matrices in (6.51) and (6.54) be positive definite,
and thus AD − B2 > 0 and ( A0 + A)( D0 + D ) − B2 > 0. Exper-
imental observations are that ω < 0 which implies U > 0, with
Ω + ω < 0 and Ω > 0, and we therefore conclude that the flagellum
and cell body counter-rotate.
The results in (6.55) exhibit an interesting dependence on the flag-
ellar length L. If we consider for simplicity a spherical cell body
of radius Rc , with A0 = 6πµRc and D0 = 8πµR3c , and assume
ζ ⊥ = 2ζ ∥ = 4πµ/c, where c is the logarithmic term in (4.21), and
rescaling cell radius and flagella length on the helix radius R via
U Ω Rc L
U= , Ω̃ = , ρc = , ℓ= , (6.56)
−ωR −ω R cR
then the key results in (6.55) can be expressed as
4ℓρ3c sin θ cos θ

Ũ = ,
[3ρc + ℓ(1 + sin2 θ )][4ρ3c + ℓ(1 + cos2 θ )] − ℓ2 sin2 θ cos2 θ
(6.57a)
ℓ(1 + cos2 θ )[3ρc + ℓ(1 + sin2 θ )] − ℓ2 sin2 θ cos2 θ
Ω̃ = .
[3ρc + ℓ(1 + sin2 θ )][4ρ3c + ℓ(1 + cos2 θ )] − ℓ2 sin2 θ cos2 θ
(6.57b)
Figure 6.16 displays these results for the case θ = π/6 and various
0.2 1 0
~ !1
(a) (b) (c)
body rotation frequency +
~
.agellar rotation frequency +

swimming speed U
;c = 2
0.1 0.5 -0.5

;c = 1
;c = 0:5
0 0 -1
0 10 20 0 10 20 0 10 20
` ` `
Figure 6.16: Swimming dynamics of
a model bacterium, from Eqs. (6.57).
values of ρc . For each cell body size there is an optimum flagel- Panels display the rescaled (a) swim-
lar length for maximum swimming speed, followed by a slow decay ming speed, (b) cell body rotation fre-
quency, and (c) flagellar rotation fre-
in the speed with increasing length. As ℓ increases, the body ro- quency in lab frame. Obtained with
tation frequency tends toward the negative of the flagellar rotation Figure_616_bacterialswimming.m.
frequency, and the net frequency in the laboratory frame vanishes,

while at the length of optimum propulsion speed we see Ω/(−ω ) ∼
0.3 − 0.6 for the range of ρc considered. As a typical radius R is on
the order of a micron, and the factor c in the RFT drag coefficients is
∼ 5, the peak length for ρc = 1 corresponds to a flagellar length of
∼ 15 µm, which is comparable to that observed in nature.
In summary, this analysis shows how a combination of the linear-
ity of Stokes flow and the requirements of vanishing force and torque
provide the necessary ingredients to determine the swimming and
rotation speeds in a model of microorganism locomotion.
6.3 Models of synchronisation
In Section 6.2.2 we saw a calculation that suggested that synchro-

nised beating of flagella could arise through a minimum-dissipation
principle, but that argument did not provide a dynamical evolution
to the state of synchrony. In this section we discuss the pedagogic
example of "bead-spring" models that have been developed to study
this problem. The analysis leads to a model for the evolution of the
phase difference between flagella, and provides interpretable results
that can be compared to experiment.
6.3.1 Bead-spring model

When a eukaryotic flagellum moves through its beat cycle it continu-
ally exerts a distributed force on the fluid, setting it in motion. That
motion in turn creates a distributed force an a second, nearby flagel-
lum, possibly deforming it. A very simple caricature of this dynam-
ics involves representing the flagellum as a single sphere moving
on an orbit, exerting a localised force on the fluid and responding
to forces from other nearby moving spheres. This idea lies behind
a class of “bead-spring" models that has been used to study the
problem of synchronisation of flagella. The simplest of these14 is 14
Niedermayer, T., Eckhardt, B. and
analytically tractable and interpretable in terms of the mechanism of Lenz, P. Synchronization, phase lock-
ing, and metachronal wave formation in
synchronisation, and provides a context within which to study the ciliary chains. Chaos, 18:037128, 2008
simplest example of phase dynamics.
The model involves a microsphere of radius a moving along an
orbit with instantaneous radius R by the action of a constant tangen-
tial force F while connected to the origin by a spring with rest length
R0 . Force balance in the tangential and radial directions gives the
equations of motion
ζR φ̇ = F, ζ Ṙ = −λ ( R − R0 ) , (6.58)
where λ is the spring constant. The dynamics (6.58) has a trivial
fixed point R = R0 and ω ≡ ϕ̇ = F/ζR0 , and we can trade the force
F for the product ζR0 ω in the following.
In the bending of an elastic filament by a localised force at its end,
dimensional analysis shows that there is a spring-like response with
λ ∼ A/L3 . The drag coefficient would be just that of a sphere in
Stokes flow, ζ = 6πµa. Together, ζ and λ define a relaxation time
ζ
τλ = . (6.59)
λ
If we take a ∼ 0.1 µm, then ζ ∼ 2 × 10−3 pN·s/µm, and with A ∼
4 × 10−22 N·m2 and L = 5 µm (i.e. half the length of a flagellum) we
have λ ∼ 3 pN/µm, giving τ ∼ 0.6 ms. This can be compared to the
beat period of ∼ 20 ms for Chlamydomonas flagella, showing that we
expect elastic relaxation to be fast compared to the orbital motion.
This separation of time scales can be used to simplify the analysis.
2a
Figure 6.17: Two coupled bead-spring
F oscillators in the NEL model. Each
bead (red circles) is pushed around an
orbit by a constant tangential force F
R1 and is held by a spring (green) to the
φ� φ�
R2 equilibrium orbit radius R0 .
F
R0 R0
x l
Now we generalise the model to two coupled systems whose center-

to-center separation is ℓ, as shown in Fig. 6.17. In the coordinate
system shown, the position of each of the beads (i = 1, 2) and the

associated basis vectors are

ri = Ri (t) cos φi (t)êx + sin φi (t)êy + δi2 ℓêx , (6.60)
êri = cos φi (t)êx + sin φi (t)êy , (6.61)
ê φi = − sin φi (t)êx + cos φi (t)êy . (6.62)
The coupled dynamics of the two oscillators generalises (6.58) to ac-

count for the fluid flow produced at the sphere 1 due to 2,

ζ R1 φ̇1 − ê φ1 · v12 = ζR0 ω1 , (6.63)

ζ Ṙ1 − êr1 · v12 = −λ ( R1 − R0 ) , (6.64)
and likewise for particle 2. Here, we take the flow to be the far-field
flow due to a moving sphere,
s · (I + n̂12 n̂12 )
v12 = +··· (6.65)
|r12 |
where the vector r12 points from 2 → 1 and
3a 3a
s= ṙ ≃ R2 φ̇2 ê φ2 , (6.66)
4 12 4
the latter approximation holding under the assumption of small dis-
placements from a circular orbit.
In the "weak-coupling" regime ℓ ≫ R the terms simplify:
r12 ≃ −ℓêx , |r12 | ≃ ℓ, n̂12 ≃ −êx , (6.67)

3a
ê φ1 · s ≃ R2 φ̇2 cos( φ1 − φ2 ), ê φ1 · n̂12 ≃ sin φ1 , (6.68)
4
3a
s · n̂12 ≃ R2 φ̇2 sin φ2 , ê φ1 · v12 ≃ −ρR2 φ˙2 J ( φ2 , φ2 ), (6.69)
4
êr1 · v12 ≃ ρR2 φ˙2 K ( φ2 , φ2 ), (6.70)
where
J ( φ1 , φ2 ) = −3 cos( φ1 − φ2 ) + cos( φ1 + φ2 ), (6.71)

K ( φ1 , φ2 ) = 3 sin( φ1 − φ2 ) − sin( φ1 + φ2 ), (6.72)
and ρ = 3a/8ℓ is a scaled measure of the orbital separation. Substi-

tuting into the governing equations, we obtain
R0 R2
φ̇1 = ω1 − ρJ ( φ1 , φ2 ) φ̇2 , (6.73)
R1 R1
Ṙ1 = −τλ−1 ( R1 − R0 ) + ρR2 K ( φ1 , φ2 ) φ̇2 , (6.74)
and likewise for particle 2. If the relaxation time τλ is small we can

adopt a quasi-steady approximation in which Ṙ1 ≃ 0, yielding an
algebraic relationship for R1 ,
R1 ≃ R0 + ρτλ R2 K ( φ1 , φ2 ) φ̇2 . (6.75)
This and the equivalent relation for R2 can now be substituted into
the angular dynamics and the result simplified under the assumption
that the natural frequencies ω1 and ω2 are weakly perturbed by the

coupling ρ. The resulting coupled phase dynamics are
φ̇1 ≃ ω1 − ρω̄ J ( φ1 , φ2 ) − ρτλ ω̄ 2 K ( φ1 , φ2 ), (6.76)

2
φ̇2 ≃ ω2 − ρω̄ J ( φ2 , φ1 ) − ρτλ ω̄ K ( φ2 , φ1 ), (6.77)
where in the coupling terms we have approximated each ωi by the

mean value ω̄ = (ω1 + ω2 )/2. The results (6.76) and (6.77) show how
the model naturally leads to in-phase synchronisation. Consider the
dynamics of the phase difference θ = φ1 − φ2 . As J ( φ1 , φ2 ) is sym-
metric under the exchange 1 → 2, as is the second term in K ( φ1 , φ2 ),
the dynamics for θ has a particularly simple form. If we rescale via
T = ω̄t and define
∆ω
δ= , and γ = 6ρω̄τλ , (6.78)
ω̄
then
θ T = δ − γ sin θ. (6.79)
6.3.2 The Adler equation

Equation (6.79) is the fundamental result of the NEL model, and
is known as the Adler equation after Robert Adler who first devel-
oped it in the context of coupled electronic oscillators.15 It occupies 15
Adler, R. A Study of Locking Phe-
a prominent place in the general study of phase synchronisation in nomena in Oscillators. Proc. IEEE, 61:
1380–1385, 1973
a broad range of physical and biological systems, from neurons to
Josephson junctions. Let us first examine it as a dynamical system.
ϴT Figure 6.18: Graphical representation of

the Adler equation. Curves indicate the
r.h.s. of (6.79) for increasing values of
δω. Fixed points within the first two
periods of sin θ are shown by circles.
� �� ϴ
As shown in Fig. 6.18, the r.h.s of (6.79) supports fixed points at

θ = nπ for δω = 0, and continues to support two within each in-
terval [2nπ, (2n + 1)π for positive δω up to a critical value δ∗ = γ,
beyond which there are no fixed points. Staying within the first pe-
riod [0, 2π ], the fixed points θ ∗ obey θ ∗ = sin−1 (δ/γ). Let us term
the smaller of the two θ− ∗ and the larger θ ∗ , with θ ∗ = π − θ ∗ . Small
+ + −
deviations ψ± from each of these obey the linearised dynamics
q
∂ T ψ± = ± γ2 − (δω )2 ψ± + · · · , (6.80)
∗ rep-
and thus the smaller root is stable and the latter is unstable; θ−
resents stable synchrony. For the special case δ = 0 one can solve
exactly the initial value problem to find

−1 θ0 −γT
θ ( T ) = 2 tan tan e , (6.81)
2
from which we infer that the synchronized state is an attractor for all
initial conditions. From the form of γ we see that the time to achieve
synchrony can be significantly longer than an orbital period if the
oscillators are far apart.
slows down Figure 6.19: Elastohydrodynamic

mechanism of synchronisation in the
NEL model.
lags leads
speeds up
The mechanism of synchronisation can be seen in the diagram in

Fig. 6.19, where we consider the effect of the fluid flow produced by
particle 1 on particle 2. If 1 lags behind 2, then the flow it produces
V(ϴ)
deflects 2 to a larger orbit. But under the constraint that the internal
force F is constant, motion at a larger radius involves a smaller angu-
lar velocity φ̇ to keep the drag force constant, and thus 2 slows down,
reducing the lag. Similarly, if 1 leads 2, then the fluid flow pushes
ϴ
2 to a smaller orbit with a higher angular velocity and 2 speeds up,
reducing the lead. Thus, synchronisation in this model involves both
hydrodynamical coupling and "waveform compliance".
The Adler equation (6.79) is a gradient flow θ T = −dV (θ )/dθ in Figure 6.20: Effective potential in the
Adler model.
the "tilted washboard" potential
16
Polin, M., Tuval, I., Drescher, K., Gol-
V (θ ) = −δθ − γ cos θ, (6.82) lub, J.P. and Goldstein, R.E. Chlamy-
domonas swims with two ‘gears’ in a eu-
pictured in Fig. 6.20. This connection is significant in light of exper- karyotic version of run-and-tumble lo-
comotion. Science, 325:487–490, 2009
imental results shown in Fig. 6.21 on the dynamics of the flagella of
17
Goldstein, R.E., Polin, M. and Tuval,
Chlamydomonas reinhardtii. In these experiments,16 , 17 the phase dif- I. Noise and synchronization in Pairs of
ference is found from the time series of beating flagella by tracking Beating Eukaryotic Flagella. Phys. Rev.
the pixel intensity in small interrogation regions on either side of the Lett., 103:168103, 2009
cell body (Figs. 6.21(a-l)). In the simplest analysis, the peaks in inten-
sity as flagella move through the regions define the endpoint of each
cycle, with the phase angle linearly interpolated in time between.
The data in Fig. 6.21(m) on the scaled phase difference ∆(t) =
( φ1 (t) − φ2 (t))/2π shown in blue reveals that within periods of syn-
chrony there are significant fluctuations away from the mean value
of ∆ (red). A simple calculation shows that the fluctuations observed
are too large to arise from timing jitter due to purely equilibrium
thermal fluctuations of elastic flagella. The data also show a transi-
tion over the course of ∼ 1 s of the running mean value of ∆ by one
unit, followed by resynchronisation at that new value. This transition
corresponds to the loss of one complete cycle of motion for flagellum
2 relative to flagellum 1. In the language of coupled oscillators this is
termed a "phase slip", and corresponds to a transition from one min-
imum to an adjacent one of the effective potential V (θ ). This kind of
Figure 6.21: Experimental data on flag-

ellar synchronisation in the green alga
Chlamydomonas, from Ref. 15.
event is not possible within the purely deterministic evolution (6.79)

when the potential has truly local minima; it requires some kind of
stochastic forcing to provide the energy to hop over the potential
energy barrier separating the adjacent minima, as discussed next.
6.3.3 The stochastic Adler equation

In light of these observations, we consider a generalisation of the
Adler equation to include random forcing, the Langevin equation
dV (θ )
θT = − + ξ ( t ), (6.83)
dθ
where the noise has the typical properties we discussed in (3.45) of
Sec. 3.3, ⟨ξ (t)⟩ = 0 and ⟨ξ (t)ξ (t′ )⟩ = 2k B Teff δ(t − t′ ), where Teff is
some effective temperature. This stochastic Adler equation allows
for a heuristic view of synchronisation, in that we can view (6.83)
as the overdamped equation of motion of a particle with coordinate
θ on the "landscape" defined by V (θ ), subject to noise. Periods of
oscillator phase synchrony then correspond to sojourns in one of
the minima of V, and the noise induces quasi-thermal fluctuations
in a harmonic potential (like the optical trap considered in (5.41)
of Sec. 5.2.3 on flicker phenomena of membranes). The linearised
dynamics (6.80) around the stable minimum θ− ∗ implies that there is
an effective harmonic potential there of the form

1
q
Veff (ψ) = γ2 − δ2 ψ2 . (6.84)
2
It follows from equipartition that the autocorrelation function of ψ is
k B Teff ′ 1
⟨ψ(t)ψ(t′ )⟩ = p e−|t−t |/τ , τ= p . (6.85)
γ2 − (δω )2 γ2 − (δω )2
Phase slips like that shown in the middle panel of Fig. 6.21 are
thermally-assisted hops from one minimum of V to an adjacent one,
an "activated process" that is well-studied within nonequilibrium sta-

tistical physics. Quite generally the rate r of these events is propor-
tional to the Boltzmann factor of the barrier height separating the
beginning and final states. Thus, the ratio R± ≡ p+ /p− of rightward
and leftward hops is independent of the proportionality constant and
depends only on the energy difference of the two states, which can
be read off from (6.82) as 2πδω, and thus
R± = e2πδ/Teff . (6.86)
Thus, measurement of (i) the amplitude, (ii) the decay time of the
autocorrelation function, and (iii) the relative probability of left-right
hopping, provides three relations involving the three parameters of
the model (γ, δ, Teff ), allowing all to be determined from the data.
Finally, we note that the NEL model makes testable predictions
regarding the coupling strength due to hydrodynamic interactions.
For example, it shows that the coupling constant should vary with 18
Goldstein, R.E., Polin, M. and Tuval,
flagellar length as γ ∼ L3 ; experiments that take advantage of the I. Emergence of Synchronized Beat-
ing during the Regrowth of Eukaryotic
ability of Chlamydomonas to shed and then regrow flagella have been Flagella. Phys. Rev. Lett., 107:148103,
used to confirm this prediction.18 Second, the dependence of the 2011
coupling on the oscillator separation, γ ∼ 1/ℓ, a consequence of the 19
Brumley, D.R., Wan, K.Y., Polin, M.
hydrodynamical coupling due to stokeslet flows, has been confirmed and Goldstein, R.E. Flagellar synchro-
nization through direct hydrodynamic
by examining the interaction between flagella on two separated cells interactions. eLife, 3:e02750, 2014
interacting only through the fluid between them.19
6.4 Swimming Cells
In this section we discuss the representation of swimming cells by

force singularities in Stokes flow, setting the stage to discuss the basic
features of interactions between swimmers and surfaces.
6.4.1 Stokeslets
We begin by recalling the flow field induced by a point force, u(x) =
G(x − x′ ) · F, with the Oseen tensor
1
G(r) = (I + r̂r̂) , (6.87)
8πµr
where r = |r| and r̂ = r/r is a unit vector. We refer to this flow field
as a "stokeslet". In terms of components, we have
Fj ri r j

δij
ui = + 3 , (6.88)
8πµ r r
so for a force along z, the flow in the x − z plane has components
Figure 6.22: Streamlines of a stokeslet
u x = ( F/8πu) xz/r3 and uz = ( F/8πµ)(1/r + z2 /r3 ), giving rise to
at the origin, pointing upwards in
the streamlines shown in Fig. 6.22. the x − z plane. Obtained with
A biophysical realisation of this flow is seen with an organism like Figure_622_stokeslet.m.
Volvox in Figs. 1.2 and 6.1, as its density ρV is slightly greater than
ρw , that of the water it swims through. It is therefore acted on by a
downward (negatively) buoyant force
4π
Fb = − δρw gR3 ez , (6.89)
3
where the fractional density offset δ = (ρV − ρw )/ρw ≪ 1. Inserting

typical numbers (δ = 0.002, R = 200 µm, and with ρw = 1 g/cm3 and
g = 10 m/s2 , we find |Fb | ∼ 6 × 102 pN. This force will balance the
viscous drag force 6πµRusettle to determine the settling speed
20
Drescher, K., Leptos, K.C., Tuval, I.,
2 δgR2 Ishikawa, T., Pedley, T.J. and Goldstein,
usettle = , (6.90) R.E. Dancing Volvox: Hydrodynamic
9 ν Bound States of Swimming Algae. Phys.
Rev. Lett., 102:168101, 2009
where again ν = µ/ρw is the kinematic viscosity of water, leading to
usettle ∼ 200 µm/s, consistent with that found experimentally.20
Figure 6.23: Experimentally deter-

mined flow field around the alga V. car-
Figure 6.23 shows experimental results on the stokeslet flow around teri algae. From Ref. 2.
Volvox colonies from Ref. 2. These include (a) (i) the measured veloc-
ity field in the laboratory frame, in terms of streamlines and a heat
map of the speed, (ii) a theoretical fit to the stokeslet field and (iii) the
very small residuals of the fit. Panels in (b) show the near-field flow,
including (i) the magnitude, vector field, and streamlines after sub-
traction of the stokeslet contribution and (ii) the velocity magnitude
along a horizontal cross section through the colony centre. An av-
erage stokeslet (green dashed line) follows the observed decay (blue
circles) averaged over the sample set (black dots). Deviations from
pure stokeslet appear at distances less than ∼ 5R and are accounted
for by adding higher-order singularities (red solid line). (iii) Vertical
section of u through the centre of a colony, with symbols as in (ii).
6.4.2 Stresslets and Rotlets

If there are no body forces such as gravity acting on an organism,
the flow around it can not contain any stokeslet contribution and
will fall off as 1/r2 (or faster), corresponding to force dipoles. From
the illustrations in Fig. 6.24, we see that there are two basic types
of dipoles. For bacteria and spermatozoa, the head pushed forward
on the fluid as the tail pushes backward, while for algae the leading
flagella push backwards as the trailing body pushes forward. These

are, respectively, "pusher" and "puller" dipoles.
Figure 6.24: Examples of the two types

of force-free swimmers described by a
force dipole.
To understand the flows produced by pushers and pullers, we

consider two forces separated by the vector ϵℓd, where ϵ is a small
order-counting parameter. We adopt a temporary notation in which
G(r; e) is the flow at r due to a force along e. If we choose the origin O
O as indicated in Fig. 6.25, then the induced fluid flow is
u(r) = FG(r − ϵℓd; e) − FG(r; e)

Figure 6.25: A force dipole.
≃ −ϵℓ F (d · ∇) G(r; e) + O(ϵ2 ), (6.91)
where the second relation follows from a Taylor expansion. Setting

P = ϵℓ F to be the dipole strength, we may write
udipole (r; e, d) = −P d · ∇G(r; e). (6.92)
A simple calculation shows that

 
 
1  (d × e) × r (d · e)r 3(e · r)(d · r)r 
∇G(r; e) =  − +  , (6.93)
8πµ 
| r3
{z } | r
3
{z r
5
}

antisymmetric: symmetric:
e↔d e↔d
where the symmetry of the terms under exchange of e and d are

indicated. separating terms with differing symmetries, we identify
antisym P (d × e) × r
udipole (r; e, d) = , (6.94a)
8πµ r3
P 3(e · r)(d · r)r (d · e)r

sym
udipole (r; e, d) = + . (6.94b)
8πµ r5 r3
We analyze the contributions (6.94) in turn. Recall that the Stokes

flow around and viscous torque due to a sphere of radius a rotating
with angular velocity Ω are
a3
u= Ω × r, and T = 8πµa3 Ω. (6.95)
r3
If we substitute the second relation into the first to eliminate refer-
ence to the sphere, we obtain the "rotlet" flow uR (r; T), defined as the
flow due to a point torque T,
1 T×r
uR (r; T) = , with T = P d × e. (6.96)
8πµ r3
Rotlet flows arise when external torques act on swimmers, such as

when bacteria with internal magnetic particles swim in the earth’s
magnetic field, but free-swimming cells have no associated rotlet.
Instead, as we have seen in Sec. 6.2.5 in the analysis of the com-
posite system of a bacterial cell + flagellum, the requirement of zero
torque leads to counter-rotation of the two components. This feature
is captured by a rotlet dipole, whose flow field falls off as 1/r3 .
The symmetric contribution in (6.94) is a particular case of the
most general far-field flow21 due to a torque-free and force-free body, 21
Batchelor, G.K. Slender-body theory
for particles of arbitrary cross-section in
3 (r · S · r) r 1 Stokes flow. J. Fluid Mech., 44:419–440,
uS (r; S) = − 5
∼ 2, (6.97) 1970
8πµ r r
where S is symmetric and trace-free. Here, the "stresslet" tensor is

1 1
S=P (d · e) I − (ed + de) . (6.98)
3 2

In the symmetric limit e = d, we have S = P 13 I − ee and
!
P 3 (e · r)2 1
uS (r ) = − 3 r, (6.99a)
8πµ r5 r
P 3 cos2 θ − 1

= êr (6.99b)
8πµ r2
where the second form is expressed in terms of the polar angle θ
from the dipole axis. P > 0 for pushers and P < 0 for pullers,
and as expected, the flow falls off as r −2 in all directions. Figure
Figure 6.26: Streamlines and vec-

tor field of stresslet flows for (a)
pushers and (b) pullers. Red ar-
rows indicate directions of forces com-
prising the dipoles. Obtained with
Figure_626_stresslets.m.
6.26 shows the flows fields for pushers and pullers. The streamline
geometry is the same in the two cases, only the direction of motion
along them switches. A characteristic feature of the flows is the nodal
lines at angles θn such that cos2 θn = 1/3 or θn ≃ ±54.7◦ .
These results can be compared to experimental measurements of 22
Drescher, K., Dunkel, J., Cisneros,
L.H., Ganguly, S. and Goldstein, R.E.
the flow field around individual freely-swimming E. coli bacteria Fluid dynamics and noise in bacte-
shown in Fig. 6.27.22 These data were obtained by tracking the rial cell-cell and cell-surface scattering.
motion of micron sized tracer spheres around individual cells swim- Proc. Natl. Acad. Sci. USA, 108:10940–
10945, 2011
ming in the focal plane of the microscope to obtain short traces of
the fluid velocity field, and then averaging over a very large number
of traces rotated to align the cellular axis in a common direction. The

mined flow field around E. coli. (A)
flow field displays the characteristic topology—forward of the cell Flow field in the frame of reference of
the swimming cells. (B) Theoretical
body, backward from the flagella, and inward on each side—as well stresslet flow field and (C) residuals of
as the 1/r2 falloff in magnitude expected from a stresslet. Fitting the the data after subtracting it from the
data, showing now systematic trends.
data to the theoretical result of two oppositely-directed point forces
(D) Decay of the magnitude of the ve-
of magnitude F and separation ℓ yields the values F = 0.42 pN and locity field in three different directions
ℓ = 1.9 µm, which are consistent with the cell geometry. away from the cell. From Ref. 22.

mined flow field around the alga C.
Similar methods can be applied to understand the flow field around reinhardtii. From Ref. 20.
a puller organism like Chlamydomonas. Figure 6.28 shows the results
of such a study, where by the nature of the algorithm the flow that is
captured is time-averaged over the flagellar beating. As expected for
a puller stresslet, the far-field flow is inward toward both the front
and the back of the organism and is outward from the sides, just the
opposite of the bacterial flow field. But closer to the cell the near-
field is significantly different, with a stagnation point near the front
and eddy-like circulation due to the flagella. The overall topology
of the flow field is captured by a superposition of three stokeslets,
one pointing forward at the cell body, and one each at the middle of
the flagella, pointing backwards (with the vectorial sum of the three
vanishing). The flow magnitude falls off as 1/r2 .
6.5 Microswimmers in External Fields and Flows
In this section we consider examples of microswimmers in external

fields and flows. Moving beyond the simplest effect of gravity, which
leads to a net force on a swimmer, we consider the torque associated
with unequal mass distribution in an organism, and then combine
this with net swimming motions and the effects of fluid vorticity
and shear to arrive at a general model for swimmers in flows.
6.5.1 Bottom-Heavy Swimmers

23
Kessler, J.O. Gyrotactic buoyant con-
Many of the organisms of interest in biological fluid dynamics have vection and spontaneous pattern for-
a complex internal distribution of material, such that, as shown in mation in algal cell cultures. In M.G.
Velarde, editor, Nonequilibrium Cooper-
Fig. 6.29, the centre of buoyancy of the organism and its centre of ative Phenomena in Physics and Related
gravity are not coincident, like a ship with a heavy keel. Recognition Fields, pages 241–248. Plenum Press,
New York, 1984
of the consequences of this feature in algal systems was first made
by Kessler23 and studied in detail by Pedley and Kessler.24 24
Pedley, T.J. and Kessler, J.O. The ori-
As in the figure, we introduce the unit vector p that points along entation of spheroidal microorganisms
swimming in a flow field. Proc. R. Soc.
the major axis of the cell. In the case of an alga like Chlamydomonas, Lond. B, 231:47–70, 1987
p lies between the two flagella. The displacement of the centre of
gravity G from the centre of buoyancy C is taken to be −hp, where
h is a small distance. If the colony is tilted from the vertical by an
angle θ, then the torque about the centre of mass of the colony is
Vρc hg sin θ, where V = 4πR3 /3 is the colony volume. Balancing this
torque against the rotational drag on the sphere, with drag coefficient p
ζ r = 8πµR3 , we obtain the equation of motion
1
θ̇ = − sin θ, (6.100)
τbh
C
where the bottom-heaviness relaxation time is G -hp
6µ 6ν
τbh = ∼ . (6.101)
ρc gh gℓ
In the second relation we have used the fact that the colony density is
mg
typically very close to water, and thus µ/ρc is close to the kinematic
Figure 6.29: Geometry of a biflagel-
viscosity ν = 106 µm2 /s of water. As g = 107 µm/s2 in our favourite lated bottom-heavy swimmer. Centre
system of units, we see that a submicron scale of ℓ yields relaxation of gravity (G) differs from centre of
buoyancy (C).
times of a few seconds, as seen experimentally. The angular dynam-
ics (6.100) can be recast as an equation for p,
1
ṗ = p × (ez × p ) . (6.102)
τbh
a formulation that preserves p as a unit vector.
6.5.2 Swimmers in flows

Now we ask how a swimming cell responds to ambient fluid flows.
Cells can of course respond passively through their shape, and actively
by changing the actuation of their appendages in response to the
local flow. Here we consider only the former, and assume that the
cell is characterised by its swimming speed U in the direction set
by its instantaneous axis vector p and ignore any intrinsic rotational
motion. In the regime of low Reynolds number (and implicitly in the
regime of low Stokes number, where particle inertia is sufficiently
small that they follow streamlines), we expect that the center of mass
of the cell is also advected by the flow u(r) at the location r(t), and
the equation of motion of the particle would be
ṙ = Up + u(r), (6.103)
where u(r ) is the flow evaluated at the centre of mass of the particle.
The subtlety in this approach is that the presence of the particle will
disturb the flow, so even in the simple situation where we specify
a flow field u∞ (r) far away from the cell it is unclear the level of
approximation entailed in setting u(r) = u∞ (r).
A systematic approach to this is given by Faxén’s laws,25 whose 25
Faxén, H. Der Widerstand gegen
derivation we summarise below (see also Ref. 1, Sec. 10.1). We die Bewegung einer starren Kugel
in einer zähen Flüssigkeit, die zwis-
assume that the flow u∞ is that specified in the absence of the swim- chen zwei parallelen ebenen Wänden
mer, and that in the absence of the flow a rigid spherical swimmer eingeschlossen ist. Ann. Phys., 373:89–
119, 1922
moves with velocity Us and angular velocity Ωs For a rigid swimmer,
the principle of superposition is valid, so that if U f and Ω f are the
velocities due to the flow, then the total velocities are
U = Us + U f , (6.104a)
Ω = Ωs + Ω f . (6.104b)
We seek to express U f and Ω f in terms of u∞ .

A useful starting point is the "reciprocal theorem" of Stokes flow,
which states that if we have two solutions ū and û of the incompress-
ible Stokes equations with no body forces inside the fluid, associated
respectively with stress tensors σ̄ and σ̂, then the virtual rates of
work of one flow on the other are equal, and with n the outward
sphere normal,
Z Z
dS ū · σ̂ · n = dS û · σ̄ · n, (6.105)
so long as there are no contributions at infinity. This theorem has

broad applicability in Stokes flow; the art of using it centres around
the choice of the two solutions. Here, we choose the solution (û, σ̂ ) to
be that of a rigid body moving with instantaneous velocity Û and an-
gular velocity Ω̂, associated with boundary conditions on the swim-
mer (where r = rs ),
û(rs ) = Û + Ω̂ × rs . (6.106)
Next, choose ū = u − u∞ , namely the perturbation away from the ex-

ternal flow. This conveniently vanishes at infinity and has the bound-
ary condition
ū(rs ) = Û f + Ω̂ f × rs − u∞ (rs ). (6.107)
As û is the flow associated with the swimmer translating and rotat-

ing, the right-hand-side of (6.105) us just Û · F̄ + Ω̂ · T̄, with F̄ and
T̄ the net forces and torques on the sphere governed by the ū flow.
These are identically zero, as is therefore the r.h.s. of (6.105).
Turning to the l.h.s. of (6.105), and noting that for a sphere of
radius a in rigid-body motion
3µ
σ̂ · n = − Û − 3µΩ̂ × n, (6.108)
2a
one finds from the arbitrariness of Û and Ω̂ that dS ū = 0 and

R
R
dS n × ū = 0. From the boundary conditions and the identities
dS nn = (4πa2 /3)I we then obtain

R R
dS n = 0 and
1
Z
Uf = dSu∞ , (6.109a)
4πa2
3
Z
Ωf = dSn × u∞ . (6.109b)
8πa3
These very powerful results express the kinematics of a rigid moving
sphere in terms of the imposed flow (not the actual flow), evaluated
on its surface. Importantly, that imposed flow is defined everywhere
in space, including inside the sphere.
Recalling that we wish to relate the motion of the sphere to the
value of the fluid velocity at its centre, while (6.109a) requires it on
the sphere surface, we Taylor expand the velocity around the centre.
By symmetry the relevant terms involve higher and higher order
contributions of the form ∇2n u∞ , which all vanish for n ≥ 2 by the
Stokes equation, leaving a simple relation between the translational
velocity and the external flow evaluated at the particle position. This
leads leads to the positional dynamics
a2 2
ṙ = Us + u∞ (r) + ∇ u∞ (r ). (6.110)
6
The final term is generally of order ( a/L)2 , where L is the length

scale for gradients of the fluid velocity, and can be neglected in most
applications where the confining geometry is large compared to the
particle size. We shall henceforth ignore it.
The final result for the motion of a spherical particle concerns
its angular velocity as given by (6.109b). Applying the divergence
theorem to the surface integral and noting that the vorticity ω =
∇ × u is harmonic in Stokes flow, the mean value theorem for the
Laplacian (the mean value of a harmonic function over a sphere is its
value at the centre) leads to the result
1
Ωf = ∇ × u∞ (r ). (6.111)
2
We conclude that the evolution equation for the orientation vector of
a spherical particle, the partner relation to (6.110), is
1
ṗ = (∇ × u∞ (r)) × p. (6.112)
2
Already in our discussion of the instability of semiflexible fila-
ments under compressional flows (Fig. 4.19) we saw that elongated
objects can rotate in flows that lack vorticity but instead have a finite
rate-of-strain. This implies that there must be an additional contri-
bution to the evolution equation for p beyond that in (6.112). It also
raises the question of whether for elongated self-propelled swim-
26
Hohenegger, C. and Shelley, M. J. Dy-
mers there is an additional contribution beyond (6.110). A careful namics of complex biofluids. In New
analysis26 of these questions is summarized below. Trends in the Physics and Mechanics of Bi-
To begin, we define xc (t) to be the position of the centre of a rod ological Systems: Lecture Notes of the Les
Houches Summer School: Volume 92, July
of length L, parameterized by arclength s ∈ (− L/2, L/2) relative to 2009. Oxford University Press, 2011
that centre. Choosing the orientation vector p to lie along the rod’s
tangent, a point on the rod is at xc (t) + sp and its velocity is
urod (s, t) = ẋc + sṗ. (6.113)
It must also be the case that this velocity is expressible in terms of

the values of the flow-induced velocities U f and Ω f via
urod (s, t) = U f + Ω f × (sp). (6.114)
Our goal is to find ẋc (= U f ) and ṗ(= Ω f × p).

We assume that the imposed velocity varies linearly in space,
u∞ (r) = A · r, (6.115)
where the constant tensor A is trace-free. The motion of a self-

propelled rod is described within resistive force theory as discussed
in (4.23) and (4.99) in the context of filament motion, so that the hy-
drodynamic force density along the filament is
h i
f(s, t) = − ζ ∥ pp + ζ ⊥ (I − pp) · [urod (s, t) − u∞ (xc + sp)] .
(6.116)
Given our assumptions, the final bracketed term in (6.116) is
ẋc − A · xc + s (ṗ − A · p) . (6.117)

R L/2
The requirement that the rod is overall force-free is − L/2 f = 0,
and since the drag tensor is constant, we can see that the term linear
in s in (6.117) integrates to zero, and thus
U f = ẋc = A · xc , (6.118)
or equivalently U f = u∞ (xc ), exactly the same result as a sphere.

This result, along with (6.117), therefore implies that the relative ve-
locity between the rod and fluid varies with arclength as urod − u∞ =
s(ṗ − A · p), but since all the arclength dependence is known, we can
express the force density as f(s, t) = sf̃(t), where f̃ only depends on
time. The condition that the filament be torque free then becomes
L3
Z L/2
ds sp(t) × sf̃(t) = p(t) × f̃(t) = 0, (6.119)
− L/2 12
which shows that f̃ ∥ p, and we may therefore write f̃(t) = Γ(t)p(t)
for some function Γ.
The calculation is completed by noting that the relation (6.116) can
be inverted (as we did in deriving (4.99)) to obtain
1
urod − u∞ = − (I + pp) · f, (6.120)
ζ⊥
and thus, with f simplified as above, we must have
sΓ 2sΓ
s (ṗ − A · p) = − (I + pp) · p = − p. (6.121)
ζ⊥ ζ⊥
Hence, ṗ − A · p = −(2Γ/ζ ⊥ )p. Taking the dot product with p and
noting that p · ṗ = 0 since p is a unit vector, we obtain
ζ⊥
Γ= p · A · p, (6.122)
2
and finally
ṗ = (I − pp) · (A · p) . (6.123)
With some further vector calculus manipulations one can show that
A can be expressed in terms of the rate-of-strain tensor E and vor-
ticity ω such that the final evolution equation for p takes the form
1
ṗ = Ω f × p, Ωf = ω + p × (E · p) . (6.124)
2
The result (6.124) has the very same vorticity contribution as the rigid
sphere and an additional contribution from the rate-of-strain tensor.
This second contribution describes the alignment effect we saw in
experiments on actin filaments discussed in Chapter 4.
Thus far, we have obtained the contribution to the angular velocity
from the rate-of-strain tensor for spheres, where there is no contri-
bution, and for extremely elongated rods, where we found (6.124).
These results are special cases of a more general result due to Jef-
fery,27 who computed the contribution for spheroids Combining his 27
Jeffery, G.B. The motion of ellipsoidal
result with bottom-heaviness and the vorticity contribution, we ob- particles immersed in a viscous fluid.
Proc. R. Soc. Lond. A, 102:161–179, 1922
tain "Jeffery’s equation",
1 1
ṗ = p × (ez × p) + ω × p + αp · E · (I − pp) , (6.125)
τbh 2
where, α is a parameter of the ellipsoid given in terms of the ratio r
of the major to minor axes,
r2 − 1
α= , (6.126)
r2 + 1
with α = 0 for a sphere and α → 1 for a highly elongated object.
There are many interesting and biologically relevant results that
follow from Jeffery’s equation (6.125) applied to a range of flow ge-
ometries and particle shapes. In the following two subsections we
illustrate a few key results for swimmers far from surfaces.
6.5.3 Gyrotaxis in Poiseuille flow

As a first example, we consider spherical (α = 0), bottom-heavy
swimmers in pressure-driven channel flow. Assuming a channel
width of 2L, the standard parabolic flow profile is
x2

u = U0 1 − 2 ez , (6.127)
L
where U0 is the maximum flow speed, achieved along the centreline
x = 0. The vorticity of this flow is
2U0 x
ω= ey , (6.128)
L2
and with p = (cos ϕ(t), 0, sin ϕ(t)), we obtain the coupled dynamics
ẋ = U cos ϕ, (6.129a)
x2

ż = U sin ϕ + U0 1 − , (6.129b)
L2
xU0 1
ϕ̇ = − + cos ϕ. (6.129c)
L2 τbh
A natural set of rescalings uses the channel half-width L to scale

space and the relaxation time τbh to scale time, and we introduce
x z t
X= , Z= , T= , (6.130)
L L τbh
leading to
XT = β cos ϕ, (6.131a)

ZT = β sin ϕ + β 0 1 − X 2 , (6.131b)
ϕT = − β 0 X + cos ϕ, (6.131c)
where
U0 τbh Uτbh
β0 = , and β= (6.132)
L L
are the ratios of the bottom-heaviness relaxation time to the time
necessary for the maximum flow or the swimming organism to move
the channel half-width. Note that β > 0, while β 0 can be of either
sign; β 0 > 0 is an upward channel flow, β 0 < 0 is downward.
The dynamical system (6.131) allows us to understand the phe-
nomenon of gyrotaxis, the tendency of bottom-heavy organisms to
swim into regions of vorticity. If β 0 < 0 (downward flow), then we
can see from (6.131c) that ϕ will tend to increase when X > 0, mean-
ing that the cell rotates to point toward the flow centreline. Like-
wise, ϕ decreases when X < 0 leading again to motion toward the
center of the parabolic flow. For a flow with constant vorticity it is
possible for there to be a fixed inclination angle ϕ, but in the case
of Poiseuille flow the trajectories eventually reach the centreline with
the cell pointing upwards, but being advected downward at the same
time (if β 0 + β < 0), as depicted in Fig. 6.30. This leads to the phe-
nomenon of gyrotactic focusing of algae, first described by Kessler,
in which a thin concentrated plume of algae forms at the centreline.
:
(a) (b) Figure 6.30: Gyrotaxis of a bottom-
heavy organism in channel flow. (a)
2 Trajectory for downward parabolic
flow, showing motion to centreline,
with β = 2.5, β 0 = −3, starting at
X = 0.7, Z = 0, and ϕ = π/4 (red) and
X = −0.5, Z = 0, and ϕ = 5π/8 (blue).
(b) Corresponding tilt angles. Obtained
0 :=2 with Figure_630_gyrotaxis.m.
Z
-2
0
-1 0 -1 0 5 10 15
X T
6.5.4 Jeffery Orbits in a flow with constant shear rate

An important application of the general formula for the motion of
swimming cells in flows concerns the behaviour of elongated cells in
the simplest shear flow, that with constant shear rate. Let us therefore
consider a velocity field
u = γ̇yex , (6.133)
where γ̇ is the shear rate. If we restrict ourselves to motion in the
x − y plane and write the orientation vector as p = (cos ϕ(t), sin ϕ(t),
then we obtain from our general equation of motion (6.103)
ẋ = γ̇y + U cos ϕ(t), (6.134a)

ẏ = U sin ϕ(t). (6.134b)
The associated vorticity and rate of strain tensor are

!
γ̇ 0 1
ω = −γ̇ez , E = . (6.135)
2 1 0
Writing ṗ = Ω × p and Ω = Ωez , and noting that ϕ̇ = Ω, some

algebraic manipulations lead to the evolution
γ̇ 2 2 2

ϕ̇ = − cos ϕ + r sin ϕ . (6.136)
r2 + 1
The substitution f = r tan ϕ reduces the equation to
f˙ γ̇
=− , (6.137)
1+ f 2 r + r −1
leading to the recognition of an arctan relation, and finally to

1 tγ̇
tan ϕ(t) = − tan , (6.138)
r r + r −1
where t = 0 is the time when ϕ = 0. This result indicates that ϕ(t) is

periodic in time with orbital period
2π
Torbit = r + r −1 . (6.139)
γ̇
Figure 6.31(a) shows the time evolution of ϕ for three values of r. For
0 1
Figure 6.31: Dynamics of the orienta-
tion angle in Jeffery orbits for various
aspect ratios of the cell. Curves are an-
gle ϕ in (6.138) for r = 1 (black), r = 2
(blue) and r = 10 (red). Obtained with
cos ?
-2: 0 Figure_631_and_632_Jefferyorbits.m.
?
(a) (b)
-4: -1
0 1 2 0 1 2
t=Torbit t=Torbit
r = 1 we have uniform rotation with ϕ = γ̇t/2, but for larger r the

function develops plateaus of nearly constant value, connected by
rapid transitions to the next plateau. This is seen even more clearly
in Fig. 6.31(b) where we plot the projection cos ϕ. Note the important
feature that ϕ(t) does not depend on the vertical position y(t), sd the
vorticity and rate-of-strain are constant for this linear shear profile.
The swimming trajectories in this shear flow follow from the gen-
eral equation of motion (6.103) with the given ϕ(t). If we introduce
the scaled time T = t/Torbit and scale lengths with the quantity
ℓ = UTorbit , giving X = x/ℓ and Y = y/ℓ, then
XT = 2π (r + r −1 )Y + cos Φ( T ), (6.140a)
YT = sin Φ( T ), (6.140b)
where, in these rescaled units, we have

−1 1
Φ( T ) = tan − tan (2πT ) . (6.141)
r
In this form, we see that for a given value of r the dynamics is other-
wise parameter-free, so the trajectories are in a sense "universal", but
they do depend on the initial conditions. As the equation of motion
has no explicit dependence on X, the dynamics are independent of
the value of X (0), but the explicit appearance of Y in (6.140) leads to
a one-parameter family of solution parameterized by Y (0). Numeri-
cally integrating the equations of motion (6.140) reveals that there is
a spatially-dependent competition. dependent on Y (0), between the
tendency for approximately closed orbits and large-scale advection
with the shear flow. Fig. 6.32 shows this for the case r = 5.
0.4
Figure 6.32: Jeffery orbits in the
laboratory frame for r = 5. Curves
0.2
correspond to different initial po-
sitions as denoted by the solid
circles. Vertical scale is exagger-
Y
0 ated for clarity. Obtained with

Figure_631_and_632_Jefferyorbits.m.
-0.2
-5 0 5 10 15
X
h
6.6 Surface-Mediated Interactions
We close this chapter on cellular motion with a discussion of the basic zy

aspects of the interactions between microswimmers and surfaces, in x
which we adopt the singularity description of the swimmers. The
two most important flows of interest are those due to stokeslets or
stresslets near no-slip and stress-free surfaces. Figure 6.33: Geometry of stokes singu-
larities near a free surface. Singularity
is located at red circle, image is located
6.6.1 Singularities Near Surfaces at blue circle on opposite side of the
surface z = 0. For the vertical stokeslet
Consider first a stokeslet near a stress-free surface, such as an air- the image is of the opposite sign, while
for the horizontal stokeslet it points in
water interface. In the geometry of Fig. 6.33, suppose an upward- the same direction.
pointing force is at z = h. Placing a mirror-image stokeslet at z = −h
leads, by symmetry, to a vanishing of the vertical fluid velocity at

z = 0 and the vanishing of tangential derivatives of the horizontal
velocity. If a stokeslet due to a point force with components Fj (for
j = 1, 2, 3) has the flow (6.88), then the image system yields
Fj ri r j Ri R j

δij δij
u i (x) = + 3 − + 3 , (6.142)
8πµ r r R R
where, if the singularity is located at ( xs , ys , h) and the location of
the flow is ( x, y, z), then the quantities r and R in (6.142) are
h i1/2
r = ( x − x s )2 + ( y − y s )2 + ( z − h )2 , (6.143a)
h i1/2
R = ( x − x s )2 + ( y − y s )2 + ( z + h )2 . (6.143b)
Suppose first that the force points upward, away from the surface, so
Fj = Fδj3 . If the lateral position of the singularity is (xs = ys = 0),
and it is at zs = h, the (i = 1, 3) components of the flow are
x (z − h)

F x (z + h)
u⊥x = − , (6.144a)
8πµ r3 R3
( z − h )2 ( z + h )2

⊥ F 1 1
uz = − + − . (6.144b)
8πµ r R r3 R3
It is easily seen that uz vanishes identically at z = 0, where r = R,
as does ∂u x /∂z. Figure 6.34(a) shows the fluid velocity vector field
and heat map of velocity magnitude. While the fluid velocity falls
Figure 6.34: Stokeslet flow fields near

for (a) vertical and (b) horizontal
off with distance as 1/r around a stokeslet in free space, the decay is
stokeslets near a stress-free surface,
altered by the presence of the surface. For example, in the case of a and (c) vertical stokeslet near a no-
vertically-oriented force we find from (6.144) that directly above the slip surfaces. Red circles indicate
location of singularities at z = 1,
singularity (x = 0) the vertical flow decays as uz (0, z) ∼ h/z2 . This where surface is at z = 0. Colors
can be compared to the situation with a point force along x, whose indicate magnitude of fluid velocity
on a logarithmic scale. Obtained with
image points in the same direction, giving velocity components
Figure_633_and_634_stokesletsurf.m.
( z − h )2 ( z + h )2

∥ F 1 1
ux = + + + , (6.145a)
8πµ r R r3 R3
x (z − h)

∥ F x (z + h)
uz = + . (6.145b)
8πµ r3 R3
the streamlines of which are shown in Fig. 6.34(c). In this case, the
∥
parallel flow u x far above the singularity retains its 1/z falloff.
The problem of a stokeslet near a no-slip surface is much more

involved, as all three components of the velocity u must vanish on
the surface. Blake was the first to solve this analytically in his 1971
PhD thesis.28 , 29 The method of solution involves a two-dimensional 28
J.R. Blake. Note on the image system
Fourier transform of the governing equations along the directions of for a stokeslet in a no-slip boundary.
Math. Proc. Camb. Phil. Soc., 70:303–310,
the surface, followed by solution of the resulting ODEs along the per- 1971
pendicular direction, and then inversion of the transforms. Written 29
Blake, J.R. and Chwang, A.T. Fun-
damental singularities of viscous flow.
in a form consistent with the notation in (6.142), the solution is
Part I: The image systems in the vicin-
ity of a stationary no-slip boundary. J.
Fj ri r j Ri R j

δij δij
ui = + 3 − + 3 Eng. Math., 8:23–29, 1974
8πµ r r R R

∂ hRi δi3 R i R3
+ 2h δjα δαk − δj3 δ3k − + (6.146)
∂Rk R3 R R3
where α = 1, 2. The quantity δjα δαk − δj3 δ3k is nonzero only when
j = k and is +1 for j = 1 or 2 and −1 for j = 3. The solution
involves terms with prefactors proportional to various powers of h.
That with no prefactor is just the image stokeslet as in (6.142). The
term proportional to 2hF has the interpretation of a force dipole (as it
is the derivative of terms associated with a force), while that scaling
as 2h2 F is a source dipole (the derivative of a source).
For the same vertically-oriented force (Fj = Fδj3 ), the flow is
x (z − h) x (z − h) 6hxz(z + h)

F
u⊥
x = 3
− − , (6.147a)
8πµ r R3 R5
(z − h)2 z2 + h2 6hz(z + h)2

F 1 1
u⊥
z = − + − − . (6.147b)
8πµ r R r3 R3 R5
This flow field is shown in Fig. 6.34(c), where we observe the new
feature of closed streamlines. These do not appear when the force is
parallel to the surface. For the perpendicular case, the existence of
recirculating flows is thought to have implications for filter-feeding
of organism that attach themselves to surfaces with a slender stalk.30 30
Higdon, J.J.L. The generation of feed-
The no-slip surface has a strong effect on the far-field behaviour of ing currents by flagellar motions. J.
Fluid Mech., 94:305–330, 1979
the flow. For example, for a vertical stokeslet, the flow falls off as
1/r3 , while u ∼ 1/r2 for a force parallel to the surface. The slower
fall-off for forces oriented parallel to the surface suggests, for exam-
ple, that in the action of cilia at the surface of organisms the move-
ments parallel to the surface are more important for propulsion and
fluid transport than those perpendicular to the surface.
The general result that gradients of elementary flow fields yield
higher-order singularities can be used to understand the flow field
generated by a bacterium swimming parallel to a no-slip surface.
Using the general formulation (6.146), we first obtain the flow field
in the x − y plane associated with a point force along x,
1 x2 x2 3x2 z

∥ F 1 z
ux = + 3 − + 3 − 2h + ,
8πµ r r R R R3 R5
(6.148a)

∥ F xy xy 6hxyz
uy = − 3+ . (6.148b)
8πµ r3 R R5
The streamlines of this flow field are shown in Fig. 6.35(a), and we
observe from (6.148) that u ∼ 1/r2 in the far field. The gradient
of (6.148) is shown in Fig. 6.35(b), which is proportional to the flow
field of a stresslet near a no-slip surface. Again we see the generation
of strong recirculation regions.
Figure 6.35: Stokeslet and stresslet

flows near a no-slip surface. (a)
In-plane flow due to a point force
along x. (b) Gradient of the flow in
(a), giving a stresslet flow field. Colors
indicate magnitude of fluid velocity
on a logarithmic scale. Obtained with
Figure_629_and_630_stokesletsurf.m.
The techniques used to determine the bacterial flow field in Fig.

6.27 have been applied to the situation when a bacterium swims par-
allel to a nearby no-slip surface. Figure 6.36 shows the measured
flow field, its comparison with the ideal stresslet field, and the falloff
of the velocity magnitude with distance, decaying as 1/r3 .

mined flow field around the bacterium
E. coli. From Ref. 22.
6.6.2 Attraction and reorientation of stresslets by surfaces
When a microswimmer approaches a surface there are certain key
hydrodynamic interactions between them and the surface that are
important in various phenomena within biological physics. The gen-
eral protocol for calculating the effects of a surface on a swimmer
involves (i) finding the image system appropriate to the swimmer,
(ii) calculating the vorticity and rate-of-strain tensor of the singular-
ity+images, and (iii) determining the translation and rotation of the
swimmer via Jeffery’s equation. Since this can become algebraically
involved, we focus here on two examples involving stress-free sur-
faces, here where the image system is the simplest and defer more
complex calculations to the Example Sheet.
The first example concerns the interaction of a stresslet oriented
parallel to a stress-free surface. Here we can appeal to the flow field
of a stresslet in free space (6.99) to write the solution for the entire
flow field for a stresslet oriented along ex at position (0, 0, h) with
image at (0, 0, −h) near a free surface in the x − y plane as
2 2
P

∥ 3x 1 3x 1
u x ( x, z) = − x + − x , (6.149a)
8πµ r5 r3 R5 R3
2 2
P

∥ 3x 1 3x 1
uz ( x, z) = − ( z − h ) + − ( z + h ) ,
8πµ r5 r3 R5 R3
(6.149b)
p p
where r = x2 + (z − h)2 and R = x2 + (z + h)2 . It is straight-
∥ ∥
forward to verify that uz vanishes at z = 0 as does ∂u x /∂x there, as
required. If we now evaluate this flow at the stresslet location (0, h)
we find the wall-induced velocity is
∥ P
uz (0, h) = − . (6.150)
32πµh2
This is an attractive interaction for a pusher stresslet P > 0, and a
repulsion for a puller. A nearly identical result holds for a no-slip
surface, with only a change in the numerical factor in the denomi-
nator. Experiments31 indicate that this is likely the dominant effect 31
Berke, A.P. and Turner, L. and Berg,
leading to surface accumulation of bacteria and spermatozoa. H.C. and Lauga, E. Hydrodynamic At-
traction of Swimming Microorganisms
Next we consider a stresslet oriented at some angle ϕ with respect by Surfaces. Phys. Rev. Lett., 101:038102,
to the stress-free surface, for which the velocity components are 2008
P (3x cos ϕ + (z − h) sin ϕ)2

ϕ 1
u x ( x, z) = − 3 x
8πµ r5 r

(3x cos ϕ + (z + h) sin ϕ) 2 1

+ − x ,
R5 R3
(6.151a)
P

(3x cos ϕ + (z − h) sin ϕ) 2 1

ϕ
uz ( x, z) = − 3 (z − h)
8πµ r5 r
(3x cos ϕ + (z + h) sin ϕ)2

1
+ − 3 (z + h) ,
R5 R
(6.151b)
A direct calculation shows that the vorticity contribution (1/2)ω

gives rise to the angular dynamics
3P
ϕ̇ = − sin ϕ cos ϕ, (6.152)
64πµh3
with additional contributions when the particle in non-spherical. This
leading contribution has the feature that for pullers there are stable
fixed points at ϕ = 0, π, while for pushers the stable fixed points are
at ϕ = ±π/2. We conclude that pushers will be oriented parallel to
the wall but pushers will either swim into or away from the wall.
6.6.3 Surface-mediated attraction of stokeslets

In a final illustration of the key features of flows near surfaces, we
32
Squires, T.M. Effective pseudo-
study the interaction of two organisms near a no-slip surface, follow- potentials of hydrodynamic origin. J.
ing an analysis first given by Squires.32 As shown in Fig. 6.37, we Fluid Mech., 443:403–412, 2001
consider two spherical colonies of radius R that by some means are

maintained at an upper, no-slip wall, with gravity acting downwards.
Assuming that each can be represented by a stokeslet of strength F, x
from (6.147) we read off the lateral flow at some distance x from one
colony in isolation, evaluated at an offset distance R from the surface,
R
3R3 F x
ux (x) = − . (6.153)
2πµ ( x2 + 4R2 )5/2 g
This has the important feature that when the force is pointing away Figure 6.37: Two spherical microswim-
mers near an upper no-slip surface,
from the wall (either pointing upwards at a bottom wall or pointing
with gravity pointing downward.
downwards at an upper wall) the lateral flow is inwards towards the
singularity. If a second such particle is also similarly near the wall,
they will mutually advect each other so that their separation r evolves
with twice the speed u x . If we rescale the separation as X = x/R and
time as T = Ft/µR2 , then we obtain the equation of motion
dX 3 X 0
=− . (6.154)
dT π ( X 2 + 4)5/2
There are several interesting points about this dynamical law. First,
V (X)
it is a gradient flow in the sense that dX/dT = −dV ( X )/dx, with
1 1
V (X) = − , (6.155)
π ( x2 + 4)3/2 !1=8:
0 2 4 6
a potential illustrated in Fig. 6.38. We can therefore interpret the X
Figure 6.38: Effective potential for hy-
fluid-mediated interaction as a direct mechanical interaction between drodynamic interaction between two
the spheres, characterised by a potential energy function. stokeslets near a no-slip surface. Con-
tact at occurs at X = 2 (dashed line).
Figure 6.39: Infalling dynamics of two

Volvox colonies. From Ref. 19.
7
X
5
1
-4,000 -2,000 0
T
A second result of the rescaled dynamics (6.154) is that it is totally

parameter-free, and therefore the infalling trajectories of spheres to-
ward contact (X = 2) should be universal. Figure 6.39 shows the
solution to this compares remarkably well to data on the "infalling"
of Volvox colonies as they approach each other toward contact, form-
ing what is termed a "bound state".
7
Kinetics & Pattern Formation (Lectures 21-24)
The problem of understanding biological patterns —whether at the

scale of a bacterium or a forest—has occupied a central role in most
discussions of mathematical biology for decades. The seminal work
of Alan Turing in 19521 led to a profound change in our understand- 1
Turing, A. The chemical basis of mor-
ing of the potential for systems of reacting and diffusing chemicals phogenesis. Phil. Trans. R. Soc. London
B, 237:37–72, 1952
to produce spatio-temporal patterns, and is a natural focus in this
chapter. In building up to that study, we first cover some of the most
basic aspects of chemical kinetics in order to understand the origins
of particular kinds of generic nonlinearities found in these systems,
and to introduce the important role of separations of time scales in
problems of this type.
7.1 Michaelis-Menten kinetics
Many of the key concepts in chemical kinetics can be understood

using the Michaelis-Menten model2 of enzyme kinetics. In general, 2
Michaelis, L. and Menten, M.L. Die
Kinetik der Invertinwirkung (English
an enzyme is a molecule that accelerates a chemical reaction without
translation). Biochem. Z., 49:333–369,
itself being consumed. In the MM scheme, an enzyme E interacts 1913
with a substrate S to form an "enzyme-substrate complex" C, which
dissociates into the enzyme and the reaction product P. In standard
notation, the reaction scheme is written as
1 k 2 k
E+S −
↽−
−⇀− C −−→ E + P (7.1)
k−1
where k1 , k −1 , k2 are the rate constants for the formation of C, the

breakup of C and dissociation of C, respectively. In this scheme we
have neglected the back reaction that would convert the product to
the complex. In translating this scheme into dynamical equations, we
imagine a fluid volume within which are the various species E, S, . . .,
whose average concentrations serve to determine the rates of reac-
tion. In using these concentrations to define the kinetics, there is an
implicit assumption that the system is "well mixed", as in a so-called
"continuously stirred tank reactor". If we use our previous notation
for chemical concentrations and let s = [S], e = [ E], c = [C ], and
p = [ P], then the law of mass action gives the differential equations
for the concentrations

de
= −k1 es + k −1 c + k2 c (7.2a)
dt
ds
= −k1 es + k −1 c (7.2b)
dt
dc
= k1 es − k2 c − k −1 c (7.2c)
dt
dp
= k2 c . (7.2d)
dt
Note that there are no nonlinearities higher than second order in this
scheme. The dynamics of p are completely decoupled from the other
reactions, so it is completely known once c(t) is known,
Z t
p(t) = k2 c(t′ )dt′ ,
0
and we need not consider it any further.

In the typical situation, the initial state of the system has a spec-
ified amount of substrate and enzyme, but the reaction has not yet
begun, and thus we set
s (0) = s0 , c(0) = 0, e ( 0 ) = e0 , p(0) = 0. (7.3)
By the structure of the reaction scheme the total amount of enzyme

is fixed, and one readily verifies that e + c must be conserved, since
de dc
+ = 0, (7.4)
dt dt
and hence e + c = e0 at all times. The system of three remaining
reactions after eliminating that for p can thus be reduced to two by
setting e = e0 − c, yielding
ds
= − k 1 e0 s + ( k 1 s + k −1 ) c (7.5)
dt
dc
= k 1 e0 s − ( k 1 s + k −1 + k 2 ) c (7.6)
dt
As in all problems of this type, we now nondimensionalise the
equations. The enzyme is typically present at much lower concen-
trations than the substrate, so we identify the small parameter ϵ =
e0 /s0 . Let the rescaled time be τ = k1 e0 t, and set
s(t) k2
u(τ ) = , λ= , (7.7)
s0 k 1 s0
c(t) k + k2
v(τ ) = , k= 1 , (7.8)
e0 k 1 s0
Then we have the final equations for the substrate and complex,
du
= −u + (u + k − λ)v, u(0) = 1, (7.9a)
dτ
dv
ϵ = u − (u + k)v, v(0) = 0. (7.9b)
dτ
In this form, with ϵ multiplying the highest (time) derivative in
the problem, it is clear that a finite ϵ is a singular perturbation to the
kinetics & pattern formation (lectures 21-24) 143
1 Figure 7.1: Solutions to the Michaelis-

numerical solution Menten dynamics. Full numerical
u(= )(!)
u0 (= )(!!) solution of (7.9a) and (7.9b) (solid
lines) for ϵ = 0.1, k = 0.5 and
λ = 1, compared to steady-state ap-
proximation (dashed). Obtained with
u; v
0.5 Figure_701_and_704_MichaelisMenten.m.
v(= )(!)
v0 (= )(!!)
0
0 1 2 3
=
ϵ = 0 problem and the solutions should exhibit a kind of boundary

layer in time. This can be seen in Fig. 7.1(a), where we plot with
solid lines the numerical solution of (7.9a) and (7.9b); the variable
v(τ ) climbs very rapidly from its initial value to a peak at a time
of order ϵ, and then decays much more slowly. The behaviour of
both u and v at times much larger than ϵ matches quite closely a
steady-state approximation (dashed) discussed next.
7.1.1 Steady-state approximation

If we simply ignore the possibly singular nature of ignoring the time
derivative in the v equation and set dv/dτ = 0, we obtain an algebraic
relationship between u and v that is trivially solved as
u0
v0 = , (7.10)
u0 + k
where the subscript indicates the order of ϵ. If we return to the u
equation (7.9a) and substitute the steady-state result (7.10) we obtain
the single dynamics
du0 λu0
=− . (7.11)
dt u0 + k
Note that having reduced the full MM dynamics, consisting of two
first order ODEs, to a single first order ODE, we have lost the ability
to impose the two initial conditions in the problem (u(0) = 1, v(0) =
0). This is a typical feature of problems with singular perturbations.
It is simple enough to solve this equation to obtain the implicit form
τ = (1 − u0 − k ln u0 ) /λ. (7.12) 1
R/Vmax
This satisfies the initial condition u0 (0) = 1, but the partner relation
(7.10) does not satisfy v0 (0) = 0. Nevertheless, the comparison in
Fig. 7.1 between u0 and v0 and the full numerical solutions shows
that they agree well beyond times of order ϵ. 1 s/Km
Returning to the fundamental result (7.10), we see in dimensional Figure 7.2: Michaelis-Menten reaction
rate versus substrate concentration.
units that the rate of reaction R = dp/dt would be
s
R = Vmax , (7.13)
s + Km
where Km = (k −1 + k2 )/k1 is known as the Michaelis constant and
Vmax = k2 e0 . As shown in Fig. 7.2, this rate law exhibits a linear
dependence at small substrate concentrations that rolls over to satu-

rate at the maximum rate Vmax at large s, reaching Vmax / at s = Km .
The Michaelis constant has the interpretation of the substrate con-
centration at which the reaction reaches half its maximum rate. The
constants Km and Vmax can be determined from a plot of 1/R versus
3
Lineweaver, H. and Burk, D. The
1/s, as 1/R = (1 + Km /s)/Vmax . This is the so-called Lineweaver- Determination of Enzyme Dissociation
Burk plot,3 as shown in Fig. 7.3. Constants. J. Am. Chem. Soc., 56:658–
The result (7.10) occupies a central place in the quantitative de- 666, 1934
scription of biochemical reactions, and is used widely in the mod- 1/R

eling of systems for which there is only partial information on the
complete kinetics, for it incorporates the nearly universal feature of
saturation of rates due to the presence of fast variables that are im- Km/Vmax
plicitly present. A particularly important feature of this result is its
1/Vmax
strong nonlinearity despite the original reaction scheme having only
quadratic nonlinearities. 1/s
-1/Km
7.1.2 Matched asymptotic approximation

Figure 7.3: The Lineweaver-Burk plot
To find an approximate analytical solution to the Michaelis-Menten used to determine parameters of
Michaelis-Menten dynamics.
dynamics for small ϵ we will use standard methods well-described
elsewhere.4 The solution we have found by setting ϵ = 0 is to be 4
viewed as the "outer" solution, but importantly that of u0 derives

from (7.11) but without yet imposing boundary conditions:
u0 + k ln u0 = −λτ + C, (7.14)
where C will be determined by matching to the inner solution.

To find the inner solution, we must identify the relevant inner time
scale, which we can do by defining a scaled variable T as
τ
T= , (7.15)
δ(ϵ)
where δ(ϵ) is the as yet unknown inner time that we will find on
the basis of balancing terms in the rescaled dynamics. By direct
substitution into (7.9a) and (7.9b) we obtain
1 dU
= −U + (U + k − λ)v, U (0) = 1, (7.16a)
δ dT
ϵ dV
= U − (U + k)V, V (0) = 0. (7.16b)
δ dT
where the inner solution is (U, V ). It is now clear that in order that
the time derivative balances the terms on the r.h.s. of (7.16b) we must
have δ(ϵ) ∝ ϵ and we are free to set the proportionality constant
to unity and thus find T = τ/ϵ. In other kinds of problems the
balancing of terms can be more subtle, but as ϵ only appears in the
time derivative, here it is simple. We therefore find that the equations
to be solved to find the inner solution are
dU
= ϵ [−U + (U + k − λ)v] , U (0) = 1, (7.17a)
dT
dV
= U − (U + k)V, V (0) = 0. (7.17b)
dT
Now, ϵ appears on the r.h.s. of (7.17a) and we can expect there to be a

regular perturbation expansion of the solutions in the form U ( T, ϵ) =
U0 ( T ) + ϵU1 ( T ) + . . . and V ( T, ϵ) = V0 ( T ) + ϵV1 ( T ) + . . .. Direct
substitution shows that the zeroth order problem is
dU0
= 0, U0 (0) = 1, (7.18a)
dT
dV0
= U0 − (U0 + k)V0 , V (0) = 0. (7.18b)
dT
This is easily solved to yield
1 h i
U0 ( T ) = 1, V0 ( T ) = 1 − e−(1+k)T . (7.19)
1+k
Here we note that the argument of the exponential is (1 + k )τ/ϵ,

which reveals the singular nature of the perturbation as ϵ → 0.
Next we match the inner and outer solutions to find C in (7.14),
and introduce an intermediate time scale Ti = τ/ψ(ϵ) with
ψ(ϵ)
lim ψ(ϵ) = 0, lim = ∞. (7.20)
ϵ →0 ϵ →0 δ ( ϵ )
The matching condition is then simply that the limit of the outer
solution at some fixed Ti as ϵ → 0 equals that of the inner solution,
h i
lim u0 (τ )|τ =ψTi = lim U0 ( T )| T =Ti ψ/δ , (7.21)
ϵ →0 ϵ →0
which implies u0 (0) = 1 since U0 = 1. Thus, C = 1 in (7.14) and

therefore u0 + k ln u0 = −λτ + 1. One readily verifies that the match-
ing condition on the complex concentration is
i
1 h u0 (0)
v0 (0) = lim 1 − e−(1+k)Ti ψ/δ = , (7.22)
ϵ →0 1+k u0 (0) + k
and that this is automatically satisfied.
1 Figure 7.4: Matched asymptotic

U0 (= )(!!) solutions to the Michaelis-Menten
matched asymptotics dynamics. Inner and uniform so-
lutions to the matched asymptotic
problem (dashed) compared to the
V0 (= )(!!) numerical solutions. Obtained with
u; v
0.5 Figure_701_and_704_MichaelisMenten.m.
vunif orm (= )
0
0 1 2 3
=
A uniform approximation ũ0 (τ )is obtained by a standard recipe: add

the inner and outer approximations and subtract their common part.
In this case, we have
τ
ũ0 (τ ) = u0 (τ ) + U0 − 1 ⇒ ũ0 (τ ) = u0 (τ ), (7.23)
ϵ
since U0 ≡ 1. The uniform approximation for the complex is
τ 1
ṽ0 (τ ) = v0 (τ ) + V0 − (7.24a)
ϵ 1+k
u0 ( τ ) e−(1+k)τ/ϵ
= − , (7.24b)
u0 ( τ ) + k 1+k
These results are shown in Fig. 7.4b, where we see excellent agree-
ment with the full numerics over the entire time range.
7.1.3 Cooperativity in Reaction Rates
The Michaelis-Menten scheme is the simplest example of a biochem-

ical kinetics displaying a saturating reaction rate, and has the note-
worthy feature that at small substrate concentrations s, the rate is lin-
ear in s. Such is not always the case in biochemical systems. One well
known example is provided by the binding of oxygen to hemoglobin,
the protein carrier of oxygen in the blood. Each of the fours sub-
units of hemoglobin can bind a molecule of oxygen. The analogue of
the enzyme-substrate complex concentration (7.10) is then the frac- 1
fractional binding
tional occupancy of binding sites on hemoglobin. As a function of
the partial pressure of oxygen in the blood, the occupancy exhibits a
characteristic "sigmoidal" shape shown in Fig. 7.5. This arises from
the feature of "cooperativity", in which the binding of one oxygen
pO2
molecule triggers a conformational change in the hemoglobin that
Figure 7.5: Schematic sigmoidal bind-
facilitates the binding of subsequent molecules. Such behaviour is ing curve of oxygen to hemoglobin, as
also known as allosterism and hemoglogin is said to be "allosteric". a function of the partial pressure of oxy-
This produces a very large change in absorbed oxygen over a rel- gen.
atively small shift in oxygen concentration. Thus, the high partial

pressure of oxygen in the lungs loads the hemoglobin nearly com-
pletely, while in the tissues of the body the lower oxygen concen-
tration leads to off-loading from hemoglobin. Thus, cooperativity
produces a response akin to a switch. The first treatment of this
problem was given by Monod, Wyman and Changeux.5 5
Monod, J., Wyman, J. and Changeux,
Sigmoidal reaction kinetics can arise also from an enlarged kinetic J.-P. On the Nature of Allosteric Transi-
tions - a Plausible Model. J. Mol. Biol.,
scheme with multiple complexes. For example, consider an allosteric 12:88–118, 1965
enzyme E that reacts with a substrate S to produce a product P,
k1
−−
E+S ↽−⇀
− C1 (7.25a)
k−1
2k k3
C1 + S −−→ C2 −−→ C1 + P, (7.25b)
where C1 and C2 are enzyme-substrate complexes and the ks are the

corresponding rate constants. It is a straightforward exercise to write
down the evolution equations for e, s, c1 , c2 and p that follow from the
law of mass action, and to verify that conservation of enzyme implies
the constancy of c1 + c2 + e. If we then assume initial conditions
s(0) = s0 , e(0) = e0 , c1 (0) = c2 (0) = p(0) = 0, and rescale with
ϵ = e0 /s0 , τ = k1 e0 t, u = s/s0 , and vi = ci /e0 , the nondimensional
reaction mechanism can be reduced to the form

du
= f (u, v1 , v2 ), (7.26a)
dτ
dv
ϵ 1 = g1 (u, v1 , v2), (7.26b)
dτ
dv2
ϵ = g2 (u, v1 , v2). (7.26c)
dτ
where the functions f , g1 and g2 depend on the three dimensionless
quantities α = k −1 /k1 s0 , β = k2 /k1 and γ = k3 /k1 s0 . Invoking the
steady-state approximation, one then finds the rate of production
dp u2
=A , (7.27)
dt α + u + ( β/γ)u2
where A is a constant. This functional form is a simple example of a
sigmoidal relationship. It is one example of a more general form
sn
(7.28)
Km + sn
which is known as the Hill equation, with n being the Hill coefficient.
The larger is n, the more sigmoidal and cooperative the reaction. 6
P. Cluzel, M. Surette and S. Leibler. An
ultrasensitive bacterial motor revealed
Systems with n as large as 10 or more are known, producing "ultra- by monitoring signaling proteins in sin-
sensitive" responses.6 gle cells. Science, 287:1652–1655, 2000
7.1.4 Slaving
The appearance of strong nonlinearities in the effective dynamics of
slow variables in systems with a strong separation of time scales can
be seen within a very simple model with only quadratic underly-
ing nonlinearities. Suppose that there are two degrees of freedom
p, q, where q is a slow variable (as in the Michaelis-Menten substrate
concentration) and p is fast (as in the MM complex concentration),
dq
= αq − βpq (7.29a)
dt
dp
ϵ = γp − δq2 . (7.29b)
dt
We interpret the terms αq and γp as representing autocatalysis. In
the steady state approximation, p = δq2 /γ and the slow dynamics is
dq βδ
≃ αq − q3 . (7.30)
dt γ
Importantly, the r.h.s. of (7.30) has the form −dV (q)/dq, where
1 βδ 4
V (q) = − αq2 + q . (7.31)
2 4γ
Thus, just as in our study of Euler buckling, we have a potential
capable of producing a pitchfork bifurcation as a function of α. This
corresponds to the physical scenario of overdamped dynamic, and in
fact produces a bifurcation as a function of α, from a function with
a single minimum for α < 0 to a double well. The key point to take
away from this simple example is that by slaving fast variables to
slow ones we can obtain very strong nonlinearities in the effective
equations of motion for the slow degrees of freedom, and these can
exhibit bifurcations and even bistability.
7.2 Excitable media: Neurons and the FHN model
One of the most important developments in biological physics in

the 20th century was the quantitative understanding of electrical im-
pulse propagation along neurons. This was achieved through a long
sequence of experimental and theoretical work, culminating in the
celebrated work of Hodgkin and Huxley (HH),7 which gave a com- 7
Hodgkin, A.L. and Huxley, A.F. A
quantitative description of membrane
plete description of pulses of electrical activity known as action po-
current and its application to conduc-
tentials that travel down axons at speeds that can reach m/s. Here tance and excitation in nerve. J. Physiol.,
we present a very brief summary of the key features of neurons that 117:500–544, 1952
are captured by mathematical models of their dynamics.

Axons are long, slender cylindrical structures that extend from
neurons. Their diameter is typically in the range of 1 − 20 µm, al-
though there are a few notably large exceptions such as the "giant
squid axon" that can exceed 1 mm in diameter. As shown schemati-
cally in Fig. 7.6, the axon itself is defined by a biological membrane
water + ions (high Na+, low K+, ...)

Figure 7.6: Schematic of an axon.
ion channels
ion pumps
water + ions (low Na+, high K+, ...)
membrane
a few nanometers thick that separates the fluid inside from that sur-
rounding the axon. Within the membrane are specialized protein
structures that produce and regulate a large difference in concentra-
tions of certain ions in those fluid regions. These include "pumps"
that actively and selectively transfer ions across the membrane as
they consume chemical energy, and "channels" that can open and
close to regulate the passive flow of ions across the membrane. Na+
ions may have a concentration of 150 mM outside and 15 mM inside,
while K+ is 5 mM inside and 150 mM outside. As mentioned in the
introduction, the Nernst equation relates the electric potential differ-
ence across a membrane to the concentrations on the two sides

kB T Cout
V= ln (7.32)
e Cin
where the prefactor has the value 25 mV. The typical concentration
ratios yield voltages on the scale of 60 − 70 mV.
One of the key developments of HH, and Cole & Cole before
them,8 was to take advantage of the large size of the squid axon 8
K.S. Cole. Membranes, Ions and Im-
to insert an electrode inside it, oriented parallel to the long axis. This pulses. A Chapter in Classical Biophysics.
University of California Press, Berkeley,
allowed measurements of the electrical properties without any spa- 1968
tial effects (a "space clamp"), so that the unusual properties of the ion
channels could be studied in isolation. The basic phenomenology of
electrical impulse generation in an axon that was found is depicted
in Fig. 7.7. In the absence of any stimulus, there is a "resting poten-
tial" difference of ∼ 70 mV across the membrane. If a perturbation
is made, say by injecting current into the system, that voltage can
be made more positive to see the reaction of the axon. Provided
voltage (mV) Figure 7.7: Schematic of an action po-

tential.
40
0 t (ms)
0 1 2 3 4 5
sation
repola
depolari
risatio
n
-55 threshold
-70 resting potential
refractory
period
hyperpolarisation
that perturbation is sufficiently small, it will relax back to the rest-

ing potential on the scale of ∼ 1 s. However, a perturbation beyond
a threshold leads to a very large positive spike that corresponds to
rapid "depolarization" (associated with an influx of Na+ , followed by
a "repolarisation" process that slightly undershoots the resting poten-
tial ("hyperpolarisation"), associated with an outflow of K + ions, and
finally a slower return to the resting potential. During that return the
axon is resistant to further perturbations and is said to be in a "re-
fractory period". An important feature of the action potential is the
strong separation of time scales between the fast dynamics of sodium
influx and the slow dynamics of potassium outflow.
A key component of the mathematical model of HH is its treat-
ment of the nonlinear features of ion channels. In modern termi-
nology, they are said to be "voltage-gated", meaning that their con-
ductance is dependent on the voltage across the membrane. Thus,
beyond the threshold, as the membrane depolarizes the conductance
increases, allowing further depolarisation, and further increased con-
ductance. The HH model combines this feature with the physics of
current transport along the fluid-filled interior of axons to arrive at a
complete model of action potential generation and propagation.
While able to capture the experimental data on voltages and cur-
rents during action potentials, the HH model is extremely cumber-
some and not easily amenable to analytical study. A decade after
their work came the papers by FitzHugh9 and Nagumo, et al.10 re- 9
FitzHugh, R. Impulses and physiolog-
porting vast simplifications of the dynamics to two coupled dynamical states in theoretical models of nerve
membrane. Biophys. J., 1:445–466, 1961
ical variables. Now known as the FitzHugh-Nagumo (FHN) model, 10
Nagumo, J. and Arimoto, S. and
this formulation is in fact remarkably rich and capable of exhibiting Yoshizawa, S. An active pulse transmis-
many properties beyond those associated with neurons, particularly sion line simulating nerve axon. Proc.
IRE, 50:2061–2070, 1962
in spatially-extended systems. We focus on aspects of this model
with particular pedagogical value.
As a model of the nonlinear dynamics of voltages across a neu-
ronal membrane, the FHN model consists of two variables, u and v
that obey the coupled ODEs

1 3
u̇ = c v + u − u − V0 , (7.33a)
3
1
v̇ = − (u − a + bv) , (7.33b)
c
where V0 is some external voltage, 0 < b < 1, 1 − 2b/3 < a < 1, and
c ≫ 1. The large value of c creates a separation of time scales such
that the variable u is fast and v is slow.
The FHN model has an interesting variational structure that gives
insight into its dynamical regimes. If we rewrite (7.34) and (7.35) as

1 3
u̇ = c u − u − V0 + cv, (7.34)
3
1 1
v̇ = − (− a + bv) − u, (7.35)
c c
then we observe that a dynamic that incorporates only the first,
bracketed terms on each of the right hand sides of (7.34) and (7.35)
has the variational form
∂E ∂E
u̇ = − , v̇ = − , (7.36)
∂u ∂v
where
1 1 1
E = − u2 + u4 + V0 u + bv2 − av, (7.37)
2 12 2
and this truncated dynamics would drive E downward in time. On
the other hand, including only the final terms in each equation yields
1
u̇ = cv, v̇ = − u, (7.38)
c
which is just the Hamiltonian dynamics u̇ = dH/dv, v̇ = −dH/du of
a harmonic oscillator with
1 2 1 2
H= u + cv . (7.39)
2c 2
Equivalently, ü = −u, so this structure naturally leads to oscilla-
tions at unit frequency. It is the competition between the dissipative
and oscillatory tendencies of the dynamics that leads to interesting
behaviour.
Finally, note that if we rescale time in (7.34) and (7.35) as t = cτ,
then they take the form
1
ϵu′ = v + u − u3 − V0 , (7.40a)
3
v′ = − (u − a + bv) , (7.40b)
where ′ = d/dτ and ϵ = 1/c2 . Thus, ϵ ≪ 1 for c ≫ 1 and u is the

fast variable while v is slow, a separation of time scales analogous to
that in the Michaelis-Menten dynamics (7.9a) and (7.9b).
Next we examine the FHN dynamics in detail, considering first
the case V0 = 0. As is typical for dynamical systems analysis, we
first establish fixed points, which must simultaneously satisfy u̇ = 0
and v̇ = 0, where the curves so defined in the u − v plane are termed

nullclines. In this case, they are
2
1
v = u3 − u, (7.41a)
3 1
a−u
v= . (7.41b)
b
v
0
These are plotted in Fig. 7.8. The slope of the cubic nullcline at
u = 0 is −1, whereas the second nullcline slope is −1/b, and thus -1
the restriction 0 < b < 1 guarantees the latter is larger and there is
-2 -1 0 1 2
only a single intersection of the two nullclines, and therefore only u
one fixed point (FP) (u∗ , v∗ ) and it can easily be shown that u∗ > 1. Figure 7.8: Nullclines of the FHN
If we write the generic two-component dynamics as model for V0 = 0. Obtained with
Figure_708_to_711_FitzHughNagumo.m.
u̇ = f (u, v), (7.42a)

v̇ = g(u, v), (7.42b)
then it is the Jacobian

!
fu fv
J= , (7.43)
gu gv
where f u = ∂ f /∂u|u∗ ,v∗ that determines the stability of the FP. Per-
turbations of the form u = u∗ + δu, v = v∗ + δv will grow as eσt with
the growth rate σ found as the solution to the condition
fu − σ fv
= 0, (7.44)
gu gv − σ
or σ2 − Trσ + Det, where Tr = f u + gv and Det = f u gv − f v gu , giving
1n p o
σ± = Tr ± Tr2 − 4Det . (7.45)
2
For stability, the real part of both roots must be negative, which re-
quires Tr < 0. If 0 < Det < Tr2 /4 the square root is real but less
then |Tr| and the roots are negative, while for larger Det the square
root yields an imaginary contribution, and still the real part of σ is
negative. Thus, stability requires Tr < 0 and Det > 0. In our case,
! !
c (1 − u ∗2 ) c − +
J = ∝ (7.46)
z =0 −1/c −b/c − −
and indeed Tr < 0 and Det > 0, so the FP is linearly stable.

Let us now examine the dynamics of the FHN model, starting
from an initial condition that is only slightly displaced from the FP
(u∗ , v∗ ) for V0 = 0. Obviously, if we keep V0 = 0 then both u and v
flow back to u∗ and v∗ . Next we turn on a small V0 (0.2). The v̇ = 0
nullcline does not change but the u̇ nullcline moves upward and the
system adjusts to the new fixed point with only small excursions in
both u and v from their initial values, as shown in Fig. 7.9. For
these parameter values, Tr2 − 4Det < 0 and the growth rates σ±
are complex, leading to the observed spiraling approach to the FP.
Figure 7.9: Dynamics of the FHN

model I. (a) Phase plane analysis
of trajectory (blue) starting from an
initial condition near the V0 = 0
fixed point, with a = 0.7, b = 0.8,
c = 3 and V0 = 0.2. Nullclines for
V0 = 0 are shown in black, that for
V0 = 0.2 in green. (b) Time courses of
u (blue) and v (red). Obtained with
This behaviour is analogous to the cases in Fig. 7.7 in which the

perturbation is sub-threshold.
Now we increase V0 to 0.3. As shown in Fig. 7.10, this leads to a
single large excursion in the phase plane that eventually converges
to the new fixed point, and corresponds to a single "firing" event.
Note that the variable u makes a large negative excursion, which can
be interpreted as the depolarisation spike in Fig. 7.7, followed by a
smaller positive peak that represents the hyperpolarisation epoch of
the action potential. The separation of time scales can be seen in the
slow decay of u to the FP relative to its initial very rapid change.
2
(a) Figure 7.10: Dynamics of the FHN
2 (b) u
model II. As in Fig. 7.9 except
1
for V0 = 0.3, producing a sin-
1
gle firing event. Obtained with
u; v
0
v
0
v
-1
-1
-2
-2 -1 0 1 2 0 10 20 30 40 50
u t
Finally, we increase V0 to 0.4 and find that the system enters a

regime of repeated large excursions in the phase plane, and thus a
"train" of pulses (Fig. 7.11).
2
(a) Figure 7.11: Dynamics of the
2 (b) u
FHN model III. As in Fig. 7.9
1
except for V0 = 0.4, leading to
1
periodic firing. Obtained with
u; v
0
v
0
v
-1
-1
-2
-2 -1 0 1 2 0 10 20 30 40 50
u t
We thus see that the FHN model in the slow inhibitor limit dis-
plays all the features of an "excitable medium" as exemplified by
neurons; small-amplitude response for stimuli below a threshold, a
single large-amplitude response beyond that threshold, and periodic
large-amplitude responses for even higher stimuli. An important les-
son from the study of this model is that the precise algebraic form
of the nullclines is less important than the overall geometry. The ex-
citable dynamics fundamentally requires nullcline intersection with
a single stable fixed point. In the next section we consider the situa-
tion with multiple fixed points and a fast inhibitor.
7.3 Front Propagation
The possibility that a system of chemical reactions can lead to com-

peting states, as discussed above, leads naturally to a scenario in
which those states occupy different regions of space, separated by
an identifiable boundary. To the extent that dynamics like the law
of mass action pertain not only to well-mixed chemicals but also to
populations of organisms, it is then not surprising that such "fronts"
may exist in contexts such as the spread of infectious diseases or in
population genetics. A classic example of the former is provided by
the spread of the plague in the middle ages,11 as shown by the ad- 11
Langer, W.L. The Black Death. Sci.
vancing contour lines in Fig. 7.12. We see the boundary of infected Am., 210:114–121, 1964
Figure 7.12: Examples of front propaga-

tion. Left: spread of the black plague in
regions steadily marching northward from the southern tips of Italy Europe during the Middle Ages, from
and France in late 1347 to Hungary and southern England by late Langer. Right: Interacting chemical
1348, Poland by late 1349 and beyond. The natural question that fronts.
arises from this steady motion is to understand first why there is a

sharp boundary and second, the factors that determine its speed.
An example of sharp boundaries in chemical systems was dis-
covered in 1993 in the ingenious experiments of Lee, McCormick,
Ouyang and Swinney.12 Using the ferrocyanide-iodate-sulfite system
that was know to exhibit bistability in a well-mixed context, they en- 12
Lee, K.J., McCormick, W.D., Ouyang,
gineered a two-compartment reaction vessel in which subset of the Q. and Swinney, H.L. Pattern For-
reactants were in each one, with a boundary between the two com- mation by Interacting Chemical Fronts.
Science, 261:192–194, 1993
posed of a porous medium. This allowed chemical reactions within
the medium, free of any contributions from fluid motion. Using a pH
indicator to visualise the different chemical states, they discovered
the patterns show in Fig. 7.12, in which the black-white boundaries
repel each other, leading to a space-filling, labyrinthine pattern.
There are many ways, both formal and informal, to begin to un-
derstand the mathematical principles that underlie front propaga-

tion. Here we adopt a particular class of models that allows for a
variety of behaviours and reveals the key techniques to determine
the front speed. Building on the class of dynamics seen in the pre-
vious section, of the variational type qt = −∂V/∂q, we include a
diffusive contribution and work in one spatial dimension to obtain
∂F
ut = mu xx + f (u) , where f (u) = − , (7.47)
∂u
F(u)
where we call the diffusion constant m for later convenience. By
construction, the dynamics of u is variational, with the form
Z ∞
δE

1 2
ut = − , E [u] = dx m (∇u) + F (u) . (7.48)
δu −∞ 2
As we are considering the possibility of bistability, a useful general 1 u

form of the function F (u), shown in Fig. 7.13, has two local minima,
potentially of distinct depths, separated by a local maximum at u = Figure 7.13: A potential F (u) with two
competing, locally stable states. As a
0. There can be a kink or front either between the two locally stable parameter is varied, the more stable
states, or between the unstable state and one of the stable ones. Our state switches from u = 0 (dashed) to
u = 1 (dotted).
goal is to find the speed of front propagation in these two cases.
A simple pedagogical model for f (u) involves a cubic nonlinearity
and a single control parameter r,
f (u) = − F ′ (u; r ) = −u(u − r )(u − 1), for 0 < r < 1, (7.49)
as shown in Fig. 7.14. There are three fixed points; u = 0, 1 and

u = r. Integrating this function we obtain the effective potential

1 2 2 1 1 2 1 3 1 f(u)
F (u; r ) = u (1 − u) + r − u − u − . (7.50)
4 2 2 3 12
In writing F this way, we have split out the first term, which repre-
sents the product of parabolas centred at u = 0 and u = 1, giving two
minimum with F (0; r ) = F (1; r ) = 0, and the second term that rep-
resents the contribution of the control parameter to changing those r 1 u
depths. The result is a function such that F (0; r ) = −(r − 1/2)/12
and F (1; r ) = (r − 1/2)/12 and the difference between the two is

1 1
∆F = F (1; r ) − F (0; r ) = r− . (7.51)
6 2
For r < 1/2 the state u = 1 is the more stable, and for r > 1/2 the Figure 7.14: The function f (u).
state u = 0 is more stable.
We first need to establish some terminology and methods that will
be useful for many subsequent discussions. The chief technique is a
linear stability analysis of a stationary state. Let us examine in turn the
three states ū that are zeros of f (u), and hence satisfy ut = 0. In each
case we write u = ū + û, linearize the dynamics in û, set û = eikx eσt
and find the growth rate σ (k) as a function of the wavevector k. If
σ < 0, ū is stable, while if σ > 0 ū is unstable. In adopting this
plane wave perturbation we are ignoring any lateral boundaries and
the boundary conditions (Dirichlet or Neumann) that might pertain
there. Such simplifications are not always justified (see Sec. 7.4.3),
but here they are helpfully simplifying. σ(k)
Near ū = 0, we find
ût = mû xx − r û, (7.52a)

2
σ = −r − mk . (7.52b) k
-r
As we see in Fig. 7.15, σ < 0 for all k, recalling that 0 ≤ r ≤ 1. Hence,
û = 0 is stable.
Near u = 1, we have
ût = mû xx − (1 − r )û, (7.53a) Figure 7.15: Growth rate for the stable
state ū = 0.
2
σ = −(1 − r ) − mk , (7.53b)
which is identical in form to the û = 0 case, except for the exchange

of r → 1 − r, a symmetry of the full problem. This state is also
therefore stable for all k. These results are intuitive, as from the form
of F the states u = 0 and u = 1 are clearly stable minima in the
absence of diffusion, and become even more so in its presence.
Near u = r, we find
ût = mû xx + r (1 − r )û, (7.54a)

2
σ = r (1 − r ) − mk . (7.54b)
Now the situation is qualitatively different (Fig. 7.16), as there is a σ(k)

p
band of modes extending from k = 0 to k = r (1 − r )/m in which
σ > 0, with stable modes at larger k. Intuitively, the maximum in r(1-r)
F at u = 0 is unstable to homogeneous perturbations, and diffusion
k
only restabilises perturbations at sufficiently small wavelengths.
Given the above, we now consider a front between the locally sta-
ble states at u = 0 and u = 1, with u → 1 as x → −∞, u → 0 as
x → ∞. By symmetry this front will not move if r = 1/2 because Figure 7.16: Growth rate for the state
ū = r, showing band of unstable modes
there is no energy difference between the states u = 0 and u = 1. extending to k = 0.
For r near 1/2 we expect the the front profile and speed can be ob-
tained perturbatively relative to that at r = 1/2. We start with the
governing PDE,

1
mu xx − u u − (u − 1) = 0. (7.55)
2
Multiplying through by u x and integrating we find
1
mu2x − F (u; 1/2) + C = 0, (7.56)
2
where the integration constant C can be seen to vanish from the
boundary conditions. As F (u; 1/2) has the simple form given in
(7.50), a straightforward integration yields

1 x
u= 1 − tanh √ , (7.57)
2 2 2m
a profile (Fig. 7.17) whose width is controlled by m. Looking back at
the energy functional E (7.48), we see that this front shape is a com-
promise between energetic penalties for gradients, which would be
156 biological physics and fluid dynamics u(x)
minimized by spreading out the transition region to infinite width,

making U constant, and for deviations from the minima of F, which
x
would be minimized by shrinking the transition region to zero width.
Figure 7.17: Stationary front shape for
To determine the behavior of the case r ̸= 1/2, we first derive a r = 1/2.
heuristic result and then proceed to a more systematic method. For
the general equation of motion
F(u)
ut = mu xx − F ′ (u), (7.58)
subject to boundary conditions as above at x = ±∞. The only as-

pect of F of relevance is that it exhibits two locally stable minima of
differing depth, as in Fig. 7.18.
After some initial transient we expect there to be a steady, uni-
formly moving solution, and thus seek a solution of the form 1 u
u( x, t) = U ( x − vt), (7.59)
Figure 7.18: Generic form of F (u).
for some unknown speed v. Substituting, we find
mUzz + vUz = −{− F (U )}′ , (7.60)

-F(u)
which we have written in a form that suggests a parallel with the
Newtonian dynamics. Eq. (7.60) is just the equation of motion of a t = -∞
particle of "mass" m, "position" U as a function of "time" z, with a
drag coefficient v and an effective potential − F (U ) that is an upside-
t = +∞
down version of the potential in Fig. 7.18.
1 u
In light of this mapping to a mechanical problem, we can view
the problem of the front shape to be equivalent to the mechanical
problem shown in Fig. 7.19: find the trajectory of a particle that
Figure 7.19: Effective potential in the
starts at time −∞ at position U = 1 and ends up as time goes to +∞
stable-to-stable traveling-wave solution.
at the local maximum of the effective potential at U = 0, as shown in
Fig. 7.19. Since the particle must lose a precise amount of energy to -F(u)
arrive at the lower potential and neither over- nor undershoot, there
t = -∞
is clearly a unique value of the friction coefficient v (front speed) to
achieve U → 0 as t → ∞.
The stable-to-stable front propagation problem can be contrasted t = +∞
with the stable-to-unstable one using the above mechanical analogy. 1 u
Now, the fictitious particle that starts at U = 1 must end up at the
intermediate minimum of − F at infinite time (Fig. 7.20). Clearly,
any damping coefficient greater than the critical value vc that al- Figure 7.20: Mechanical analogy in the
unstable-to-stable case.
lows the particle to asymptote to the lower maximum will suffice.
Those just above vc will lead to decaying oscillations of the particle
around the minimum, while those above a second critical value will
exhibit monotonic decay. It is quite often the case that the underly-
ing nonlinear dynamics does exhibit a unique front speed. Often, the
selected front speed is the critically-damped one, but the analysis of
this problem is beyond the scope of this course.
For the stable-to-stable front we can obtain a quantitative under-
standing of the front speed by taking the traveling wave result (7.60),
multiplying by Uz and integrating over the whole domain of z,
Z ∞ Z ∞
1 dF dU
mUz2 |∞
−∞ + v dzUz2 = dz = −∆F, (7.61)
2 ∞ −∞ dU dz
where the energy difference ∆F is given in (7.51) and the first term
vanishes by the boundary conditions at infinity. We can then for-
mally solve for the front speed as
−∆F
v = R∞ 2
. (7.62)
−∞ dzUz
It is intuitive that the front speed should be proportional to the

energy difference ∆F. The denominator in this expression can be
thought of as a kind of drag coefficient. In the spirit of a pertur-
bative analysis near r = 1/2, the leading order front speed can be
obtained by using the stationary front profile to evaluate the denom-
inator. Rather than do complicated integrals with hyperbolic func-
tions, we use the first integral (7.56) to obtain Uz2 = (1/2m)U 2 (1 −
U )2 so, with appropriate choice of sign for the square root, Uz =
√ √
−(1/ 2m)U (1 − U ) and thus Uz2 dz = Uz dU = 1/6 2m and
R R
thus for the front transitioning from u = 1 at the left to u = 0 at the

right the speed is
√ 1

v = − 2m r − . (7.63)
2
A consistency check is that when r < 1/2 the lower energy state is
indeed u = 1 and the front moves to the right. We will return to this
result in the context of a more systematic perturbation theory for a
two-component reaction-diffusion system.
7.4 Reaction-Diffusion Systems
7.4.1 Phenomenology
Systems of reacting and diffusing chemicals are ubiquitous in bi-
ological systems, essentially responsible for life’s processes inside
cells. While many biological pathways involve numerous coupled
reaction-diffusion systems, much of our intuition is built on mod-
els with just a few coupled species. In this section we review on σ(k)
a phenomenological level some considerations that indicate the im-
portance of examining at least two coupled species in order to obtain
pattern-forming dynamics consistent with observations. This leads
k
naturally into Turing’s analysis, covered in the next section.
Let us start from a generic structure of a nonlinear PDE that em-
bodies diffusion and nonlinear chemical kinetics. From our previous
discussions we expect the form
ut = Lu + N (u) , Lu = αu + Du xx . (7.64) Figure 7.21: Generic growth rate curves

for a single species reaction-diffusion
system. Green arrow indicates progres-
Here, we group both autocatalysis (coefficient α) and diffusion into sion of growth curves as a control pa-
the linear operator L. Linearising around u = 0 we find the growth rameter is increased.
rate σ(k ) = α − Dk2 , seen before in Fig. 7.16, but now with α a free
parameter. As we move from α < 0 to α > 0 the system goes from
stable at all wavelengths to unstable in a band of modes that extends
to k = 0, as in Fig. 7.21. This dynamical structure would not produce
patterns with an identifiable (finite) length scale.
A more interesting possibility is when both long and short wave- σ(k)
length are damped and σ(k ) is positive only in an interval of finite
k, as in Fig. 7.22. This band of unstable modes typically increases k
k*
in width as a control parameter is varied, starting from a critical k∗
when σ has a maximum just crossing the axis. The maximum of σ
defines the fastest growing mode, which often determines the wave-
length λ∗ = 2π/k∗ of the finite-amplitude patterns beyond onset.
In order to obtain growth curves like those in Fig. 7.22 with a sin- Figure 7.22: Growth curves for a
gle species we would require the linear operator L to have higher- pattern-forming system as a parameter
order derivatives than those from diffusion. Respecting parity sym- is increased (green arrow).
metry (x → − x) of the underlying physics, σ should be a function

only of even powers of k, and the simplest form would be
σ (k) ∼ α + βk2 − γk4 + . . . , (7.65)
corresponding to the linearised PDE
ut = αu − βu xx − γu4x + . . . . (7.66)
While mathematically acceptable, such higher-order derivative the-

ories for a single degree of freedom are generally hard to justify
a priori in a system of chemical reactions (although such terms do
arise in the theory of elasticity), except in the presence of nonlocality
(as discussed in Sec. 7.2). Instead, the simplest local mechanism to
produce this is to couple two diffusive systems together, for then the
linear stability problem involves a 2x2 matrix with entries at most of
order k2 , leading to a determinental equation of order k4 .
The FHN model considered in previous sections in the context
of electrophysiology and front propagation provides the basis for a
suitable two-variable reaction-diffusion system. In the language of
chemical reactions, this generalised FHN model has two species: u,
the activator, and v, the inhibitor. They evolve according to the PDEs
ut = D ∇2 u + f (u) − ρv (7.67a)
2
ϵvt = ∇ v + αu − βv, (7.67b)
in which we have scaled space and time in order that the diffusion
constant of the inhibitor is unity. The presence of the relative diffu-
sion constant D can lead to the key phenomenon of lateral inhibition
for D ≪ 1 as we discuss below. We may be interested in a whole
range of values for ϵ, not necessarily small. The various terms on
the right hand sides of these equations have the following interpre-
tation: The function f (u) embodies autocatalysis and bistability in
the manner we have discussed earlier in this chapter. The term −ρv
represents inhibition of the activator due to the presence of the in-
hibitor. In the inhibitor equation, αu represents stimulation of the
inhibitor due to the activator, while − βv describes self-limitation of
the inhibitor. Depending on the structure of f and the values of
the various coefficients, the FHN model can produce homogeneous
states, stripes and other periodic patterns, and even rotating spiral
waves. It is a "standard model" of pattern formation.
As in our discussion of the FHN model of electrophysiology, this

PDE version has a mixed variational structure.13 To align this dis- 13
Goldstein, R.E., Muraki, D.J. and Pet-
cussion with the literature, we adopt a slightly modified version that rich, D.M. Interface proliferation and
the growth of labyrinths in a reaction-
results in a more symmetric form by including an additional term diffusion system. Phys. Rev. E, 53:3933–
linear in u in the activator dynamics and setting α = β = 1, 3957, 1996
ut = D ∇2 u − u(u − r )(u − 1) − ρ(v − u), (7.68a)

2
ϵvt = ∇ v + u − v. (7.68b)
The model is then invariant under the simultaneous transformations

u → 1 − u, v → 1 − v and r → 1 − r. It is now straightforward to
verify that the model (7.68b) has the variational form
δE δE
ut = − − ρv, ϵvt = − + u, (7.69)
δu δv
with

1 1 1 1
Z
E= d2 x D (∇u)2 + F (u) − ρu2 + (∇v)2 + v2 . (7.70)
2 2 2 2
As before, terms involving the functional derivative of E are gradient
flows, and the remaining terms have a Hamiltonian form(ut = −ρv;
ϵvt = u). The FHN dynamics thus embodies a competition between
terms which are dissipative and would tend to reach a stationary
state and those which would produce oscillations. We focus on the
regime in which gradient flow contributions dominate.
The presence of the parameter ϵ in the FHN inhibitor dynam-
ics indicates that the fast inhibitor limit corresponds to ϵ ≪ 1, much
like in the Michaelis-Menten dynamics where the enzyme-substrate
complex dynamics are fast compared to those of the substrate itself.
Let us adopt the steady-state approximation in which we set ϵ = 0
without delving into the details of the short-time behaviour. As in
the MM case, this assumption leads to an instantaneous-in-time re-
lationship between u and v, except that here it is nonlocal in space,
(∇2 − 1)v = −u. (7.71)
This is mathematically equivalent to the Debye-Hückel problem we

considered in Chapter 2 (Eq. 2.43), except with a source term on the
r.h.s. The solution to (7.71) is given in terms of the Green’s function
G for the modified Helmholtz operator (∇2 − 1),
Z
v (x) = dx′ G (x − x′ )u(x′ ). (7.72)
In the two spatial dimensions we are considering here,

1
G (x − x′ ) = K0 (|x − x′ ), (7.73)
2π
where K0 is the modified Bessel function. Thus, in the fast inhibitor
limit we can rewrite the dynamics as a single nonlocal PDE for the
activator,
Z
ut = D ∇2 u + f (u) + ρu − ρ dx′ G (x − x′ )u(x′ ). (7.74)
The u dynamics is variational in this limit, ut = −δE /δu, where

1 1 2
Z
2
E = dx D |∇u| + F (u) − ρu
2 2
1
Z Z
+ ρ dx dx′ u(x) G (x − x′ )u(x′ ). (7.75)
2
Not surprisingly, the new contribution is like the electrostatic energy
of a system with a charge distribution u(x) within screened electro-
statics. This feature underlies the repulsion of fronts in the bistable
regime of the FHN model, as we shall see below.
The fast-inhibitor limit provides us with a particularly simple con-
text in which to understand a pattern-forming system instability,
since the dynamics is contained within a single PDE. While those
dynamics are nonlocal in real space they are local in Fourier space.
Thus, if we linearise the equation of motion (7.74) around the state
u = 0 and assume the usual plane wave perturbation, then the
growth rate of modes is simply
k2
σ (k) = − Dk2 − r + ρ . (7.76)
1 + k2
Figure 7.23a shows a plot of σ (k) for several values of ρ and r at fixed
D, illustrating how this system conforms to the anticipated curve
in our phenomenological discussion in Fig. 7.22. This is a particu-
larly easy problem in which to see that a system of reaction-diffusion
equations can lead to a finite-wavelength instability.
(a) 0:6 (b) (c) Figure 7.23: Stability analysis of FHN

model in the fast-inhibitor limit. (a)
0 1 1
Turing Turing Growth rate versus k forD = 0.01,
0:4 unstable unstable r = 0.35 and various values of ρ as in-
dicated. (b) Stability diagram around
<(k)
u = 0, using (7.77). (c) Stability di-

;
; = 0:2
-0.5 0.5 0.5 agram around u = 1. Obtained with
Turing Turing Figure_722_FHNgrowth.m.
stable stable
-1 1 1
0 5 10 0 0.5 1 0 0.5 1
k r r
Given the shape of σ (k) it is clear that the threshold for the onset of
a finite-wavelength instability occurs with σ (k) = 0 and dσ (k)/dk =
0 for some k c . The second relation may also be chosen as dσ/dk2 = 0
for convenience. One finds the threshold lies along a curve in the
r − ρ plane given by
√ √ 2
ρ c (r ) = r+ D , (7.77)
along which the critical wavevector is

r 1/4
kc = . (7.78)
D
By the symmetry of the model, the stability condition for the u =
v = 1 base state is ρ(r ) = ρc (1 − r ). These two boundaries are shown
in Figs. 7.23b&c.
7.4.2 Front Propagation II

The FHN model provides a useful context to understand the prob-
lem of front propagation beyond the heuristic argument presented
earlier. We begin by recalling from (7.57) that in the absence of in-
hibitor coupling the stationary front at r = 1/2 is

1 x
u( x ) = 1 − tanh √ , (7.79)
2 2 2D
√
and thus the width of the transition region scales as ∼ D. This
defines the length scale of the "inner" region of the problem, in the
sense of matched asymptotics, while the "outer" region is on an O(1)
scale. In addition to assuming D ≪ 1, we shall also take ρ and
r − 1/2 to be small to develop a systematic perturbation theory. We
use the experimental image in Fig. 7.12 as motivation, and assign
u = 0 to white and u = 1 to black.
In the outer region we rewrite the equations to put all small terms
on the r.h.s. Assuming a one-dimensional system, we have
u(u − r )(u − 1) = Du xx − ρ (v − u) − ut , (7.80a)

v xx − v + u = ϵvt , (7.80b)
where in addition to the terms in D, ρ, and r − 1/2, we have also

anticipated that the explicit time dependence will be small. As there
may be multiple interfaces between black and white, we write the
outer solution as

1 (black)
u ( x ) ∼ u0 ( x ) = (7.81)
0 (white),
and 
1 (black)
v0xx − v0 ∼ −u0 = − (7.82)
0 (white).
Thus, v0 solves a screened electrostatic Poisson problem with a source

given by the piecewise constant function u0 ( x ).
If r ̸= 1/2, the front positions Q j (t) will evolve in time, and it is
thus natural to examine the inner region of the front using the shifted
and rescaled variable
x − Q j (t)
η= √ , (7.83)
D
and to consider u as a function of η; u = U (η, t). Now define the
scaled operator from the r.h.s. of the governing PDE at r = 1/2 as
S[U ] = Uηη − U (U − 1/2) (U − 1) . (7.84)
Then a simple calculation shows that the inner equations are

j
Q
S[U ] = − √ t Uη − 6∆FU (U − 1) + ρ (V − U ) + Ut , (7.85a)
D
Vηη = D (V − U ) + ϵDVt . (7.85b)
The homogeneous solutions to the left-hand sides of (7.85a) and

(7.85b) are simply the stationary front solution

0 1 η
U (η ) = 1 ∓ tanh √ , (7.86a)
2 2 2

V 0 = v0 x = Q j . (7.86b)
Here, as η → ±∞ these match the outer solutions u0 and v0 at the

interface where x → ( Q j )± . The ∓ sign is fixed by the orientation
of the front; a negative sign corresponds to a left-to-right transition
from u = 0 to u = 1. Note that to leading order the inhibitor field is
constant within this inner region, much as the substrate concentra-
tion in Michaelis-Menten kinetics is constant in the inner region.
The leading-order front solutions are stationary, but they will be
set in motion by the effects of the r.h.s. terms in (7.85a) and (7.85b).
We assume the asymptotic balance
j
Q
√ t ∼ ∆F ∼ ρ ≪ 1. (7.87)
D
We thus assume that the solutions can be expanded in a regular
series in the small parameter ρ of the form
U (η ) ∼ U 0 (η ) + ρU 1 (η ) + · · · . (7.88)
Substituting into (7.85a) and (7.85b), and identifying the derivative

of the operator S as
′′
S′ [U 1 ] = Uηη
1
− F (U 0 ; 1/2)U 1 , (7.89)
we find the zeroth order solution to be simply the homogeneous

solutions already identified, and the first-order correction to satisfy
j
" #
′ 1 Qt
ρS [U ] = ± √ + 6∆F U 0 (U 0 − 1)
2D
h i
+ ρ v0 ( x = Q j ) − U 0 , (7.90)
√
where we used the identity 2Uη0 = ∓U 0 (U 0 − 1) to combine terms.
We have arrived at a key result in the theory of pattern formation,
for Eq. 7.90 is an inhomogeneous PDE in which the forcing on the
j
r.h.s. involves an unknown quantity (here, Qt ). As the full solution
to (7.90) is the sum of its homogeneous solution and a particular
solution of the inhomogeneous equation, the requirement that the
solution be bounded (a condition which itself can be deduced from
the original PDE formulation of the problem) requires that the r.h.s.
be orthogonal to the nullspace of the operator on the l.h.s. This is
the so-called Fredholm alternative. The nullspace of interest can be
deduced easily since the operator
′′
∂ηη − F (U 0 ; 1/2) (7.91)
is exactly that which appears if we differentiate the equation S[U ] =

0 with respect to η, implying that the nullspace is the function Uη0 .
This result can also be understood as a consequence of the trans-

lational invariance of the stationary front U 0 , and the fact that the
derivative ∂η is the infinitesimal generator of displacements.
With this result for the nullspace, and defining the inner product
for the orthogonality condition as
Z ∞
⟨Uη0 , •⟩ ≡ dηUη0 • . (7.92)
−∞
the orthogonality of the r.h.s. of (7.90) yields immediately
j √
1

Qt ∼ ∓6 2D ∆F + ρ v0 ( x = Q j ) − . (7.93)
2
This general result can now be applied to several important cases.
For the case of a single front, we consider the case of a pattern that
is white (u = 0) to the right of an interface at x = Q(t) and black
(u = 1) to the left,

1, x ≤ Q(t)
u0 ( x ) ∼ (7.94)
0, x ≥ Q(t).
The outer inhibitor field can be solved in a piecewise manner with

continuity at the black/white boundary, yielding

1 − 1 e+( x−Q) , x ≤ Q(t)
v0 ( x ) ∼ 2
(7.95)
 1 e−( x−Q) , x ≥ Q ( t ).
2
Here, we have imposed the condition that v0 → 0 at infinity. Since

v0 ( x = Q) = 1/2, there is no O(ρ) correction to the front speed,
√ √ 1

Qt = −6 2D∆F = − 2D r − , (7.96)
2
in complete agreement with the heuristic argument (7.63).
A second application of the one-dimensional front propagation
result (7.93) is the case of two fronts of opposite sense facing each
other, with 
1 − ∞ < x ≤ − Q ( t )


u0 ( x ) = 0 − Q(t) ≤ x ≤ + Q(t) (7.97)

+ Q(t) ≤ x < +∞.

1

The solution of the outer inhibitor field is


+ x −∞ < x ≤ − Q(t)
1 − sinh Q e


v0 ( x ) = e−Q cosh x − Q(t) ≤ x ≤ + Q(t) (7.98)

1 − sinh Q e− x + Q(t) ≤ x < +∞.


and thus the inhibitor contribution to the front speed is

1
v0 ( x = Q ) = 1 − e−2Q , (7.99)
2
and the speed itself is
√
1

−2Q

Qt = 2D r− + 3ρe . (7.100)
2
5
(b) Figure 7.24: Dynamics of two nearby
(a) fronts in the FHN model. Param-
4
eters are D = 0.04, r = 0.4 and
u(x; t); v(x; t)
ρ = 0.15. (a) Activator (blue) and in-

3
Q(t)
hibitor (red) fields for two approach-
2 ing fronts, with time increasing up-
wards. Black circles indicate location
1 of u = 1/2, the midpoint of the fronts.
1 (b) Distance between fronts as a func-
0 0 tion of scaled time. Obtained with
-10 -5 0 5 10 0 5 10 Figure_724_FHNfronts.m.
p
x (1=2 ! r) 2Dt
Figure 7.24 shows the time evolution of a pair of fronts that start
at a large separation Q, with r < 1/2, so the u = 1 state is more
stable and the two fronts approach one another. As they do, the
leftward (negative) speed from the term r − 1/2 will be reduced as
Q decreases by the repulsive effect of the overlapping inhibitor fields,
eventually vanishing at a distance
1/2 − r

1
Q∗ = − ln . (7.101)
2 3ρ
This front repulsion is one of the best indications of the effects of

lateral inhibition and the separation of length scales between the ac-
tivator and the inhibitor. Although the predictions of this leading-
order asymptotic analysis typically require the parameter ρ to be as
small as ∼ 0.01 for a quantitative match, the phenomenology is well-
captured by this coarsest level of approximation.
Figure 7.25: Labyrinthine pattern for-

mation in the FHN model. Param-
eters are D = 0.01, r = 0.52,
and ρ = 0.15. Obtained with
Figure_725_labyrinth.m.
The results above show that there is a regime in parameter space in

which nearby approaching fronts repel each other sufficiently strongly
as to be prevented from crossing. This is the basic observation of Lee,
et al. seen in Fig. 7.12. The FHN model in the fast-inhibitor limit also
reproduces the labyrinthine patterns seen in experiment. Figure 7.25

shows the result of a particularly large-scale numerical study that
illustrates this scenario.
7.4.3 The Turing Instability

Now we move to a more general discussion of the possible pattern-
forming instabilities of a two-species model. We consider a model
with concentrations u(x, t) and v(x, t) in a domain D on whose bound-
ary ∂D there are the no-flux conditions n̂ · ∇u = n̂ · ∇v = 0.
∂D ∂D
The pair of reaction-diffusion equations is written in the general form
familiar from our discussion of the FHN model, Eqs. 7.42a and 7.42b,
only now including two possibly different diffusion constants,
ut = Du ∇2 u + f (u, v), (7.102a)

2
vt = Dv ∇ v + g(u, v), (7.102b)
where f and g are some smooth functions of their arguments, repre-

senting, for example, autocatalysis, feedback inhibition, etc.
The central result of Turing involves the possible instability of a
reaction-diffusion system under conditions in which the homoge-
neous (i.e. reaction-only) dynamics is stable. Thus, we assume that f
and g are such that there exists a uniform steady state (u∗ , v∗ ) such
that f (u∗ , v∗ ) = g(u∗ , v∗ ) = 0 whose Jacobian entries f u , f v , gu , gv at
(u∗ , v∗ ) are such that
Tr = f u + gv < 0 and Det = f u gv − f v gu > 0. (7.103)
If we perturb this homogeneous steady state with spatial-temporal

variations u = u∗ + p(x, t), v = v∗ + q(x, t), we obtain the dynamics
pt = f u p + f v q + Du ∇2 p, (7.104a)
2
qt = gu p + gv q + Dv ∇ q, (7.104b)
Next we observe that it will always be possible to expand a func-

tion in the domain D as an infinite series of eigenfunctions of the
Helmholtz equation
∇2 wk +k2 wk = 0 (in D), (7.105a)

n̂ · ∇wk = 0 (on ∂D). (7.105b)
For example, in d = 1 with D = [0, L], we have wk = cos(kx ) and

k = nπ/L for integers n. If we then write
p= ∑ p̂k eσ(k)t wk (x), (7.106a)

k
q= ∑ q̂k eσ(k)t wk (x), (7.106b)
k
then the 2 × 2 stability matrix takes the modified form

! !
fu fv − Du k 2 0
Jmod = + . (7.107)
gu gv 0 − Dv k 2
While the trace of Jmod is just the sum of the traces of the two matri-
ces in (7.107),
Trmod = f u + gv −( Du + Dv )k2 , (7.108)
| {z }
Tr<0
and is < 0 since f u + gv < 0 by hypothesis and the diffusive contri-
butions are manifestly negative. In contrast, it is generally the case
for two matrices M and N that
Det( M + N ) ̸= Det( M ) + Det( N ). (7.109)
Were this to be true, then Detmod would be positive definite since

Det > 0 and the determinant of the second matrix in (7.107) is
Du Dv k4 > 0 as well. In fact, the modified determinant is
Detmod = f u gv − f v gu − ( Du gv + Dv f u ) k2 + Du Dv k4 , (7.110)
| {z } | {z } | {z }
Det>0 ? >0
where we have indicated that it is the middle term whose sign is

unclear. For convenience, write (7.110) as Detmod = Ak4 − Bk2 + C,
with A, C > 0. Recalling that the general condition for stability is
a negative trace and positive determinant, the criterion for instabil-
ity can be summarized as follows: The reaction-diffusion system is
unstable if, for some k among the allowed wavevectors, Detmod < 0.
Given the known signs in (7.110), this can only happen if B > 0, or
Du gv + Dv f u > 0, (7.111)
Det(k)
for then, as shown in Fig. 7.26, the function Detmod (k) can dip below
the axis. If we let x = k2 , then any crossings of the axis must be
roots of the quadratic equation Ax2 − Bx + C = 0, namely x± =
√
(1/2)( B ± B2 − 4AC ), with B2 − 4AC > 0. The smallest value of B
is that which yields the tangent condition shown as the middle curve
√
in Fig. 7.26, namely B = 4AC.
k
We thus conclude that the onset of the instability occurs at k*
p
Dv f u + Du gv = 4Du Dv Det. (7.112)
Figure 7.26: Behaviour of the determi-
nant of the stability matrix.
If we define the diffusivity ratio
Du
d= , (7.113)
Dv
then the condition (7.112) is
√
f u + dgv = 2 dDet. (7.114)
Since the r.h.s. of this equation is clearly > 0, we conclude that the
conditions that the homogeneous fixed point be stable and that the
diffusive system be unstable require the simultaneous equations
f u + gv < 0, (7.115a)
f u + dgv > 0, (7.115b)
which clearly is not possible when d = 1. That is, a Turing instability

at equal diffusivities is not possible. Without loss of generality, we
can take Du > Dv , so d > 1. If both conditions (7.115b) hold then f u

and gv must have opposite signs, with f u < 0 and gv > 0. Then we
√
can view the onset condition (7.114) as a quadratic equation for d,
whose solution is
√
√
p
Det + Det − f u gv
d= . (7.116)
gv
This is the minimum possible value for the diffusivity ratio for there
to be a Turing instability. Finally, we can examine the typical length
scale of the instability. Since Detmod = Ax2 − Bx + C, where x = k2 ,
the condition of a double root of Detmod (k) = 0 implies a turning
point which yields the condition
Det 1/2
r
C
kc = = . (7.117)
A Du Dv
We may now summarize the fundamental features of the Turing
instability. A two-component reaction-diffusion system that has a
stable homogeneous FP can, given suitable conditions on the ratio of
diffusion constants of the two species, become unstable to a finite-
wavelength instability. As we have focused only on the results ob-
tainable from linear stability analysis we can say nothing definite
about the fate of a system beyond the instability, but generically one
expects the nonlinearities in the system to stabilize finite-amplitude
patterns with a characteristic wavelength close to that at onset. As
in many other pattern-forming systems,14 the precise form of those 14
Cross, M.C. and Hohenberg, P.C. Pat-
nonlinearities will determine the symmetries of the patterns, distin- tern formation outside of equilibrium.
Rev. Mod. Phys., 65:851–1112, 1993
guishing for example between stripes and spots.
Figure 7.27: Experimental observation

of the Turing instability in a reaction-
diffusion system, from Ref. 18.
Although Turing’s work appeared in 1952, it was many years be- 15

E.F. Keller. Making sense of life. Ex-
fore the biological community took notice of it15 in any significant plaining biological development with mod-
way, but many of the terms he introduced in describing his model, els, metaphors, and machines. Harvard
University Press, Cambridge, MA, 2003
such as "morphogen", have been adopted universally in biology (and
biological physics). It was not until the early 1990s that experimen- 16
Castets, V. and Dulos, E. and Bois-
sonade, J. and De Kepper, P. Experi-
tal demonstrations of his predictions were reported. These were re- mental evidence of a sustained stand-
ported by the Bordeaux group of Castets et al.,16 , 17 followed by the ing Turing-type nonequilibrium chem-
Austin group of Ouyang and Swinney.18 Each of these groups uti- ical pattern. Phys. Rev. Lett., 64:2953–
2956, 1990
lized experimental setups with two reservoirs, each containing a non- 17
De Kepper, P. and Castets, V. and Du-
reacting subset of chemicals, linked by a porous medium in which los, E. and Boissonade, J. Turing-type
chemical patterns in the chlorite-iodide-
reactions occur without the disturbing effects of fluid motion. The
malonic acid reaction. Physica D, 49:
Austin experiments in particular allowed for spatially extended pat- 161–169, 1991
terns in two dimensions without any imposed chemical gradients. 18
Ouyang, Q. and Swinney, H.L. Tran-
sition from a uniform state to hexago-
Figure 7.27 shows that patterns in the form of stripes and hexagonal
nal and striped Turing patterns. Nature,
arrays of spots were observed. 352:610–612, 1991
We close this section by noting that reaction-diffusion patterns are 19
K.S. Kiprijanov. Chaos and beauty
not restricted to the stationary ones we have focused on. Turing in a beaker: The early historyof the
Belousov-Zhabotinsky reaction. Ann.
himself in his 1952 paper established that there could be oscillatory
Phys. (Berlin), 3-4:233–237, 2016
patterns as well, and we know this is the case from studies of the
Belousov-Zhabotinski reaction,19 which is famous for the rotating
spiral waves and target patterns that it produces (Fig. 7.28). Similar
patterns occur in a range of living systems, from slime molds to the
heart, as we mention below. The remarkable history of the discovery
of oscillatory dynamics in the BZ reaction also provides a case study
in the resistance of science to new ideas.20
7.5 Chemotaxis and Instabilities

Figure 7.28: Patterns in the Belousov-
Zhabotinski reaction in a Petri dish.
7.5.1 Background
In previous sections we focused on systems of chemical reactions
described by reaction-diffusion equations consisting of nonlinear ki- 20
A.T. Winfree. The prehistory of
netic terms and spatial diffusion, of the general form (7.102a) and the Belousov-Zhabotinsky oscillator. J.
(7.102b). Of significance is the feature that the diffusive contribu- Chem. Educ., 8:661–663, 1984
tions are "diagonal" in the sense that the stability matrix (7.107) has
diffusive terms only on the diagonal and there is no spatial cross-talk
between the various components. Here we turn to the continuum
description of populations of organisms that includes their directed
motion in response to chemical cues, the phenomenon of "chemo-
taxis". We shall see that this leads to fundamentally distinct contri-
butions to the governing PDEs, with new kinds of instabilities.
7.5.2 Phenomenology of Dictyostelium discoideum

Chemotaxis is found in a broad range of organisms, from bacteria
to much larger and more complex eukaryotes. Among these many
organisms, one can distinguish between those that swim through a
fluid environment (like bacteria) and those that move along solid sur-
faces (like amoebae). In this section we focus on the historically im-
portant case of crawling cells of the species Dictyostelium discoideum,
whose striking behaviour motivated many important developments
in the theory of biological pattern formation.
The genus Dictyostelium consists of eukaryotic cells that are abun-
dant in the soil, typically feeding on bacteria. They are often referred
(a) (b)
Figure 7.29: Life cycle of Dictyostelium

discoideum.
to as cellular "slime moulds", and can exhibit both uni- and multi-
cellular lifestyles. as shown in Fig. 7.29. In the former the cells feed
on bacteria in the environment and have little if any communication
with each other. In the typical laboratory setting of a population of
cells on the surface agar in a petri dish, one observes that if the cells
are starved for some hours, they transition to a multicellular form
by a complex, multistage process that begins with the emission of
the chemical cyclic AMP (cAMP) by a small subset of cells dispersed
throughout the population. This diffuses to neighbouring cells, stim-
ulating them to release cAMP as well, eventually leading to process
in which the population produces a mixture of target patterns or ro-
tating spiral waves of cAMP (Fig. 7.30). Although it is difficult to
visualise the cAMP directly, as the waves pass over cells they trigger
cellular elongation that can be visualised.
As the wave pattern continues for some hours, cells begin to chemo-
tactically move toward the centers of the targets or spirals and to
form there a dense multicellular body. As shown in Fig. 7.31, that
Figure 7.30: Spiral and target patterns
body then can migrate as a whole and eventually differentiates into
in Dictyostelium discoideum.
an upright form with a base, stalk, and spores at the top. These
spores can persist in a dormant state for long periods of time, even-
tually carried away by wind or passing animals to regions with more
available food, where they germinate and produce again the unicel-
lular organisms, completing the life cycle.
7.5.3 The Keller-Segel Model

During the phase of the Dictyostelium life cycle in which chemotaxis
begins in response to the waves of cAMP, it is observed that while Figure 7.31: The streaming instability in
Dictyostelium discoideum.
the gradients of cAMP are in the radial direction with respect to the
centres of targets or spirals, the initially uniform distribution of cells 21
Keller, E.F. and Segel, L.A. Initiation
develops transverse modulations leading eventually to the formation of slime mold aggregation viewed as
of radial "streams" along which cells move toward the centres. This an instability. J. Theor. Biol., 26:399–415,
1970
transverse instability (Fig. 7.31) was the subject of the important
work by Keller and Segel in the early 1970s,21 , 22 which established 22
Keller, E.F. and Segel, L.A. Model for
the key continuum formulation of chemotactic dynamics. chemotaxis. J. Theor. Biol., 30:225–234,
1971
As Keller and Segel point out, a full description of chemical waves
and chemotaxis requires at least four variables; the cell concentra-

tion n(x, t), the concentration c(x, t) of cyclic AMP, the concentration
p(x, t) of phosphodiesterase, an enzyme that breaks down cAMP,
and the concentration i (x, t) of the cAMP-phosphodiesterase com-
plex that is the intermediate in the chemical reaction. The starting
point for their model is the assumption that cell division can be ne-
glected on the time scales of interest in the experiment, and thus
there is a conservation law
nt = −∇ · Jn , (7.118)
where the flux Jn of cells has both diffusive and chemotactic contri-
butions, written as
Jn = − Dn ∇n + nr ∇c. (7.119)
The first term is the standard Fickian form, where Dn is the cellu-
lar diffusion constant that arises from their random motions, and is
assumed constant. The chemotactic part embodies the fact that cells
climb up gradients in cAMP, and we can interpret the product r ∇c
as the typical chemotactic velocity uc , with the coefficient r (c) poten-
tially varying with the cAMP concentration. Keller and Segel suggest
a form r (c) ∼ 1/c that represents logarithmic sensing (uc ∼ ∇ ln c).
Thus, the variable n evolves according to the nonlinear equation
nt = Dn ∇2 n − ∇ · (nr ∇c) . (7.120)
Even if the response function r (c) is a constant, the system is quadrat-

ically nonlinear.
The time evolution of cAMP reflects not only production by the
amoebae at a rate n f (c), where f is the production rate per cell,
but also diffusion with diffusivity Dc and enzyme kinetics via the
reaction scheme
k1
2 k
−−
cAMP + PDE ↽−⇀
− C −−→ PDE + P, (7.121)
k−1
where the product is non-chemotactically active. Combining all the

contributions, and including diffusion of each species, we obtain the
cAMP, PDE, and complex dynamics
ct = Dc ∇2 c + n f (c) − k1 cp + k −1 i, (7.122a)
2
it = Di ∇ p + k1 cp − (k −1 + k2 ) i, (7.122b)
2
pt = D p ∇ p + ng(c, p) − k1 cp + (k −1 + k2 ) i. (7.122c)
To make progress with the full dynamics (7.120), (7.122a), (7.122b),

and (7.122c), KS invoke the steady-state assumption for the complex,
k1 cp − (k −1 + k2 ) i = 0, (7.123)
ignore diffusion of i, and utilise conservation of PDE via p + i = p0 ,

where p0 is the initial concentration of PDE. As in MM kinetics, this
removes two equations from the problem, leaving only the coupled
dynamics of the cell and cAMP concentrations, in the form
nt = Dn ∇2 n − ∇ · (rn∇c) , (7.124a)
2
ct = Dc ∇ c + n f (c) − k(c)c, (7.124b)
where
p0 k 2
k(c) = , (7.125)
Km + c
and Km = (k −1 + k2 )/k1 as in Michaelis-Menten theory.
There are obvious similarities between this two-variable system
and that of the Turing instability in the sense that the reaction terms
on the r.h.s. of (7.124b) embody production and degradation in fa-
miliar forms. But the key difference is the presence of an off-diagonal
spatial term from chemotaxis. To explore the consequences of this
term, we simplify the model further by taking the functions k (c) and
f (c) to be constant, and also assuming the chemotactic response co-
efficient r to be independent of c, yielding
n t = Dn ∇ 2 n − r ∇ · ( n ∇ c ) , (7.126a)
ct = Dc ∇2 c + αn − βc, (7.126b)
Noting that the n dynamics is of a flux form that conserved the total
number of cells, we choose the mean cell concentration n0 as a ref-
erence value for rescaling. The homogeneous system has an obvious
fixed point at
α
c0 = n0 . (7.127)
β
Setting r
n c α
ρ= , χ= , T = αt, X= x, (7.128)
n0 c0 Dc
we obtain
ρ T = d ∇2 ρ − γ ∇ · ( ρ ∇ χ ) , (7.129a)
2
χ T = ∇ χ + λ (ρ − χ) , (7.129b)
where
Dn rc0
d= , γ= . (7.130)
Dc Dc
With this simplified system we linearise around the fixed point
ρ = χ = 1 via
ρ = 1 + δρ, χ = 1 + δχ, (7.131)
to obtain the linearised system
δρ T = d∇2 δρ − γ∇2 δχ, (7.132a)

δχ T = ∇2 δχ + λ (δρ − δχ) , (7.132b)
The linear stability problem for perturbations of the form eik·X+σT is
−dk2 − σ γk2
2 = 0. (7.133)
λ −k − λ − σ
If we write this as σ2 + bσ + c = 0, with b = λ + (d + 1)k2 and

√
c = k2 [d(λ + k2 ) − λγ], then σ± = (−b ± b2 − 4c)/2. It is straight-
forward to show that
b2 − 4c = (1 − d)2 k4 + 2λ(1 − d + 2γ)k2 + λ2 , (7.134)
and for d < 1 is positive. With b manifestly positive, we deduce that

if c > 0 we have stability (both roots negative). In other words, the
condition for stability is
λ (γ − d) < dk2 . (7.135)
To understand the relevant values of k for this criterion to hold we

Taylor expand the root σ+ for small k, obtaining
1
σ+ (k) = (γ − d) k2 − 1 − d + 2γ2 k4 + · · · . (7.136)
λ
The same calculation shows that σ− is strictly negative as k → 0.
Thus, we see that as γ increases beyond d the first instability is at
k = 0, but there is a peak in the growth rate for γ > d, at k∗ ∼
p
γ − d. Figure 7.32 shows σ+ (k) as γ increases.
0.1 Figure 7.32: Growth rate σ+ (k ) ver-

sus k in the Keller-Segel model. Pa-
rameters are d = 0.1, λ = 1, and
γ = 0.1, 0.2, 0.3, 0.4. Obtained with
Figure_732_KSgrowth.m.
0
<(k)
-0.1
-0.2
0 1 2
k
This instability represents the transverse instability seen in the
Dictyostelium system (Fig. 7.31). The mechanism behind the instabil-
ity is the essential feedback between cAMP production and chemo-
taxis; a local fluctuation in cell density n leads to a local increase in
cAMP production, producing a gradient in cAMP the points toward
the peak in n, followed by chemotaxis toward the peak, further in-
creasing n. The potential runaway effect is tamed by both cell and
cAMP diffusion, along with cAMP degredation. Under certain cir- 23
Childress, S. and Percus, J.K. Non-
cumstances, in certain spatial dimensions, a true singularity does linear Aspects of Chemotaxis. Math.
Biosci., 56:217–237, 1981
arise from the KS equations.23
8
Bibliography
M. Abramowitz and I.A. Stegun. Handbook of Mathematical Functions

with Formulas, Graphs, and Mathematical Tables. Dover Publications
Inc, Mineola, N.Y., 1965.
B. Audoly and Y. Pomeau. Elasticity and Geometry: From hair curls to

the non-linear response of shells. Oxford University Press, Oxford,
UK, 2010.
K.S. Cole. Membranes, Ions and Impulses. A Chapter in Classical Bio-

physics. University of California Press, Berkeley, 1968.
B. Corti. Osservazioni Microscopische sulla Tremella e sulla Circolazione

del Fluido in Una Pianta Acquajuola. Appresso Giuseppe Rocchi,
Lucca, Italy, 1774.
P.G. de Gennes. Scaling Concepts in Polymer Physics. Cornell Univer-

sity Press, Ithaca, NY, 1979.
M. Doi and S.F. Edwards. The Theory of Polymer Dynamics. Clarendon

Press, Oxford, UK, 1986.
P.J. Flory. Principles of Polymer Chemistry. Cornell University Press,

Ithaca, NY, 1953.
H. Goldstein. Classical Mechanics. Addison-Wesley, Boston, MA, 2nd

edition, 1980.
van Leeuwenhoek, A. IV. Part of a letter from Mr. Antony van

Leeuwenhoek, concerning the worms in sheeps livers, gnats, and
animalcula in the excrements of frogs. Philos. Trans. R. Soc., 22:
509–18, 1700.
Brown, R. XXVII. A brief account of microscopical observations

made in the months of June, July, and August, 1827, on the parti-
cles contained in the pollen of plants; and on the general existence
of active molecules in organic and inorganic bodies. Philos. Mag.,
4:161–73, 1828.
Engelmann, T.W. Ueber Sauerstoffausscheidung van Pfalnzenzellen

im Mikrospektrum. Pflüger, Arch., 27:485–89, 1882.
Goldstein, R.E. Coffee stains, cell receptors, and time crystals:

Lessons from the old literature. Phys. Today, 71:32–38, 2018.
Berg, H.C. and Purcell, E.M. Physics of chemoreception. Biophys. J.,

20:193–219, 1977.
Goldstein, R.E. Are theoretical ‘results’ results? eLife, 7:e40018, 2018.
Maxwell, J.C. Remarks on the mathematical classification of physical

quantities. Proc. London Math. Soc., s1-3:224, 1869.
Purcell, E.M. Life at low Reynolds number. Am. J. Phys., 45:1–11,

1977.
Goldstein, R.E. Green algae as model organisms for biological fluid

dynamics. Annu. Rev. Fluid Mech., 47:343–375, 2015.
Einstein, A. Investigations on the theory of the Brownian movement.

Ann. d. Phys., 17:549–560, 1905.
Sutherland, W. LXXV. A dynamical theory of diffusion for non-

electrolytes and the molecular mass of albumin. Phil. Mag., 9:
781–785, 1905.
Haldane, J.B.S. On Being the Right Size, 1927.
Hodgkin, A.L. and Huxley, A.F. A quantitative description of mem-

brane current and its application to conductance and excitation in
nerve. J. Physiol., 117:500–544, 1952.
Hyman, A.A., Weber, C.A. and Jülicher, F. Liquid-liquid phase sepa-

ration in biology. Annu. Rev. Cell Dev. Biol., 30:39–58, 2014.
Mayer, J.E. Statistical Mechanics of Condensing Systems. I. J. Chem.

Phys., 5:67, 1937.
van der Waals, J.D. Over de Continuiteit van den Gas- en Vloeistoftoes-
tand (On the continuity of the gas and liquid state). PhD thesis, Leiden
University, 1873.
Holstein, B.R. The van der Waals interaction. Am. J. Phys., 69:441,
2001.
Herzfeld, K.F. On atomic properties which make an element a metal.

Phys. Rev., 29:701–705, 1927.
Bussi, Y., Shimoni, E., Weiner, A., Kapon, R., Charuvi, D., Nevo, R.,
Efrati, E., Reich, Z. Fundamental helical geometry consolidates
the plant photosynthetic membrane. Proc. Natl. Acad. Sci. USA,
116:22366–22375, 2019.
Sculley, M.J., Duniec, J.T., Thorne, S.W., Chow, W.S. and Boardman,
N.K. The stacking of chloroplast thylakoids. Quantitative analysis
of the balance of forces between thylakoid membranes of chloro-
plasts, and the role of divalent ions. Arch. Bioch. Biophys., 201:
339–346, 1980.
bibliography 175
Hamaker, H.C. The London-van der Waals attraction between spher-

ical particles. Physica, 4:1058–1072, 1937.
Verwey, E.J.W. and Overbeek, J.Th.G. Theory of the Stability of Lyopho-

bic Colloids. Elsevier Publishing Company, Inc., New York, 1948.
Debye, P. and Hückel, E. The theory of electrolytes. I. Lowering of

freezing point and related phenomena (English translation). Phys.
Zeit., 24:185–206, 1923.
Marcelja, S. and Radić, N. Repulsion of interfaces due to boundary

water. Chem. Phys. Lett., 42:129–130, 1976.
Lis, L.J., McAlister, M., Fuller, N., Rand, R.P. and Parsegian, V.A. In-
teractions between neutral phospholipid bilayer membranes. Bio-
phys. J., 37:657, 1982.
Canham, P.B. The minimum energy of bending as a possible expla-

nation of the biconcave shape of the human red blood cell. J. Theor.
Biol., 26:61–76, 1970.
Helfrich, W. Elastic properties of lipid bilayers: theory and possible

experiments. Z. Naturforsch. Teil C, 28:693–703, 1973.
Winterhalter, M. and Helfrich, W. Effect of surface charge on the

curvature elasticity of membranes. J. Phys. Chem., 92:6865–6867,
1988.
Goldstein, R.E., Pesci, A.I. and Romero-Rochín, V. Electric double

layers near modulated surfaces. Phys. Rev. A, 41:5504–5515, 1990.
Duplantier, B., Goldstein, R.E., Romero-Rochín, V. and Pesci, A.I.

Geometrical and topological aspects of electric double layers near
curved surfaces. Phys. Rev. Lett., 65:508–511, 1990.
Balian, R. and Bloch, C. Distribution of eigenfrequencies for the wave

equation in a finite domain: I. Three-dimensional problem with
smooth boundary surface. Ann. Phys. (N.Y.), 60:401–447, 1970.
Planck, M. Ueber das Gesetz der Energieverteilung im Normalspek-

trum. Ann. Phys., 309:553–563, 1901.
Brown, R. XXIV. Additional remarks on active molecules. Philos.

Mag., 6:161–66, 1829.
Ashkin, A. Acceleration and trapping of particles by radiation pres-

sure. Phys. Rev. Lett., 24:156–159, 1970.
Ashkin, A. Optical trapping and manipulation of neutral particles

using lasers. Proc. Natl. Acad. Sci. USA, 94:4853–4860, 1997.
Neuman, K.C. and Block, S.M. Optical trapping. Rev. Sci. Instrum.,
75:2787–2809, 2004.
Powers, T.R. Dynamics of filaments and membranes in a viscous

fluid. Rev. Mod. Phys., 82:1607–1631, 2010.
J. Gray and G.J. Hancock. The propulsion of sea-urchin spermatozoa.

J. Exp. Biol., 32:802, 1955.
Oberbeck, A. Ueber stationäre Flüssigkeitsbewegungen mit Berück-

sichtigung der inneren Reibung. J. Reine. Angew. Math., 81:62–80,
1876.
Keller, J.B. and Rubinow, S.I. Slender-body theory for slow viscous
flow. J. Fluid Mech., 75:705–714, 1976.
Goldstein, R.E. and Langer, S.A. Nonlinear dynamics of stiff poly-

mers. Phys. Rev. Lett., 75:1094–1097, 1995.
Wiggins, C.H., Riveline, D., Ott, A. and Goldstein, R.E. Trapping and
wiggling: elastohydrodynamics of driven microfilaments. Biophys.
J., 74:1043–1060, 1998.
Kantsler, V. and Goldstein, R.E. Fluctuations, dynamics, and the

stretch-coil transition of single actin filaments in extensional flows.
Phys. Rev. Lett., 108:038103, 2012.
Fygenson, D.K., Marko, J.F. and Libchaber, A. Mechanics of micro-

tubule-based membrane extension. Phys. Rev. Lett., 79:4497–4500,
1997.
Machin, K.E. Wave propagation along flagella. J. Exp. Biol., 35:796–

806, 1958.
Wiggins, C.H. and Goldstein, R.E. Flexive and propulsive dynamics

of elastica at low Reynolds number. Phys. Rev. Lett., 80:3879–3882,
1998.
Stokes, G.G. On the effect of the internal friction of fluids on the

motion of pendulums. Trans. Camb. Philos. Soc., IX, part II:8–106,
1851.
Gadêlha, H. and Gaffney, E.A. Flagellar ultrastructure suppresses

buckling instabilities and enables mammalian sperm navigation in
high-viscosity media. J. Roy. Soc. Interface, 16:20180668, 2019.
Camalet, S. and Jülicher, F. and Prost, J. Self-organized beating and

swimming of internally driven filaments. Phys. Rev. Lett., 82:1590–
1593, 1999.
Young, Y.-N. and Shelley, M.J. Stretch-coil transition and transport

of fibers in cellular flows. Phys. Rev. Lett., 99:058303, 2007.
Batchelor, G.K. Slender-body theory for particles of arbitrary cross-

section in Stokes flow. J. Fluid Mech., 44:419–440, 1970.
H. Manikantan and D. Saintillan. Buckling transition of a semiflexi-

ble filament in extensional flow. Phys. Rev. E, 92:041002, 2015.
Herrmann, G. and Bungay, R.W. On the stability of elastic systems

subjected to nonconservative forces. J. Appl. Mech., 31:435–440,
1964.
bibliography 177
De Canio, G., Lauga, E. and Goldstein, R.E. Spontaneous oscillations

of elastic filaments induced by molecular motors. J. R. Soc. Interface,
14:20170491, 2017.
Schneider, M.B., Jenkins, J.T. and Webb, W.W. Thermal fluctuations

of large quasi-spherical bimolecular phospholipid vesicles. J. Phys.,
45:1457–1472, 1984.
Helfrich, W. Size distributions of vesicles: the role of the effective

rigidity of membranes. J. Physique, 47:321–329, 1986.
Milner, S.T. and Safran, S.A. Dynamical fluctuations of droplet mi-

croemulsions and vesicles. Phys. Rev. A, 36:4371–4379, 1987.
Zhong-can, O.-Y. and Helfrich, W. Bending energy of vesicle mem-

branes: General expressions for the first, second, and third varia-
tion of the shape energy and applications to spheres and cylinders.
Phys. Rev. A, 39:5280–5288, 1989.
Helfrich, W. Steric interactions of fluid membranes in multilayer

systems. Z. Naturforsch. Teil A, 33:305–315, 1978.
Helfrich, W. and Servuss, R.-M. Undulations, steric interaction and

cohesion of fluid membranes. Nuovo Cim., 3D:137–151, 1984.
Marko, J.F. and Siggia, E.D. Fluctuations and supercoiling of DNA.

Science, 265:506–508, 1994.
Lipowsky, R. and Leibler S. Unbinding transitions of interacting

membranes. Phys. Rev. Lett., 56:2541–2544, 1986.
Brochard, F. and Lennon, J.F. Frequency spectrum of the flicker phe-

nomenon in erythrocytes. J. Physique, 36:1035–1047, 1975.
Wiener, N. Generalized harmonic analysis. Acta Math., 55:117–258,

1930.
Khintchine, A. Korrelationstheorie der stationären stochastischen

Prozesse. Math. Ann., 109:604–615, 1934.
Tadjbakhsh, I. and Odeh, F. Equilibrium States of Elastic Rings. J.

Math. Anal. Appl., 18:59–74, 1967.
Foster, B., Verschueren, N., Knobloch, E. and Gordillo, L. Pressure-

driven wrinkling of soft inner-lined tubes. New J. Phys., 24:013026,
2022.
Evans, E. and Yeung, A. Hidden dynamics in rapid changes of bilayer

shape. Chem. Phys. Lipids, 73:39–56, 1994.
Powers, T.R., Huber, G. and Goldstein, R.E. Fluid-membrane tethers:

minimal surfaces and elastic boundary layers. Phys. Rev. E, 65:
041901, 2002.
Drescher, K. and Goldstein, R.E. and Michel, N. and Polin, M. and

Tuval, I. Direct Measurement of the Flow Field around Swimming
Microorganisms. Phys. Rev. Lett., 105:168101, 2010.
Guasto, J.S. and Johnson, K.A. and Gollub, J.P. Oscillatory Flows
Induced by Microorganisms Swimming in Two Dimensions. Phys.
Rev. Lett., 105:168102, 2010.
Brumley, D.R., Wan, K.Y., Polin, M. and Goldstein, R.E. Flagellar

synchronization through direct hydrodynamic interactions. eLife,
3:e02750, 2014.
J. Gray. The mechanism of ciliary movement. Proc. R. Soc. Lond. B,

93:104–121, 1922.
Leptos, K.C. and Chioccioli, M. and Furlan, S. and Pesci, A.I. and
Goldstein, R.E. Phototaxis of Chlamydomonas arises from a tuned
adaptive photoresponse shared with multicellular Volvocine green
algae. Phys. Rev. E, 107:014404, 2023.
Rothschild, L. Measurement of Sperm Activity before Artificial In-

semination. Nature, 163:358–359, 1949.
Taylor, G.I. Analysis of the swimming of microscopic organisms.

Proc. R. Soc. A, 209:447–461, 1951.
Elfring, G.J. and Lauga, E. Synchronization of flexible sheets. J. Fluid

Mech., 674:163–173, 2011.
Higdon, J.J.L. The hydrodynamics of flagellar propulsion: helical

waves. J. Fluid Mech., 94:331–351, 1979.
Niedermayer, T., Eckhardt, B. and Lenz, P. Synchronization, phase

locking, and metachronal wave formation in ciliary chains. Chaos,
18:037128, 2008.
Adler, R. A Study of Locking Phenomena in Oscillators. Proc. IEEE,

61:1380–1385, 1973.
Polin, M., Tuval, I., Drescher, K., Gollub, J.P. and Goldstein, R.E.
Chlamydomonas swims with two ‘gears’ in a eukaryotic version of
run-and-tumble locomotion. Science, 325:487–490, 2009.
Goldstein, R.E., Polin, M. and Tuval, I. Noise and synchronization

in Pairs of Beating Eukaryotic Flagella. Phys. Rev. Lett., 103:168103,
2009.
Goldstein, R.E., Polin, M. and Tuval, I. Emergence of Synchronized

Beating during the Regrowth of Eukaryotic Flagella. Phys. Rev.
Lett., 107:148103, 2011.
Drescher, K., Leptos, K.C., Tuval, I., Ishikawa, T., Pedley, T.J. and
Goldstein, R.E. Dancing Volvox: Hydrodynamic Bound States of
Swimming Algae. Phys. Rev. Lett., 102:168101, 2009.
Drescher, K., Dunkel, J., Cisneros, L.H., Ganguly, S. and Goldstein,

R.E. Fluid dynamics and noise in bacterial cell-cell and cell-surface
scattering. Proc. Natl. Acad. Sci. USA, 108:10940–10945, 2011.
bibliography 179
Kessler, J.O. Gyrotactic buoyant convection and spontaneous pattern

formation in algal cell cultures. In M.G. Velarde, editor, Nonequi-
librium Cooperative Phenomena in Physics and Related Fields, pages
241–248. Plenum Press, New York, 1984.
Pedley, T.J. and Kessler, J.O. The orientation of spheroidal microor-

ganisms swimming in a flow field. Proc. R. Soc. Lond. B, 231:47–70,
1987.
Faxén, H. Der Widerstand gegen die Bewegung einer starren Kugel

in einer zähen Flüssigkeit, die zwischen zwei parallelen ebenen
Wänden eingeschlossen ist. Ann. Phys., 373:89–119, 1922.
Hohenegger, C. and Shelley, M. J. Dynamics of complex biofluids. In

New Trends in the Physics and Mechanics of Biological Systems: Lecture
Notes of the Les Houches Summer School: Volume 92, July 2009. Oxford
University Press, 2011.
Jeffery, G.B. The motion of ellipsoidal particles immersed in a viscous

fluid. Proc. R. Soc. Lond. A, 102:161–179, 1922.
J.R. Blake. Note on the image system for a stokeslet in a no-slip

boundary. Math. Proc. Camb. Phil. Soc., 70:303–310, 1971.
Blake, J.R. and Chwang, A.T. Fundamental singularities of viscous

flow. Part I: The image systems in the vicinity of a stationary no-
slip boundary. J. Eng. Math., 8:23–29, 1974.
Higdon, J.J.L. The generation of feeding currents by flagellar mo-

tions. J. Fluid Mech., 94:305–330, 1979.
Berke, A.P. and Turner, L. and Berg, H.C. and Lauga, E. Hydrody-
namic Attraction of Swimming Microorganisms by Surfaces. Phys.
Rev. Lett., 101:038102, 2008.
Squires, T.M. Effective pseudo-potentials of hydrodynamic origin. J.

Fluid Mech., 443:403–412, 2001.
Turing, A. The chemical basis of morphogenesis. Phil. Trans. R. Soc.

London B, 237:37–72, 1952.
Michaelis, L. and Menten, M.L. Die Kinetik der Invertinwirkung

(English translation). Biochem. Z., 49:333–369, 1913.
Lineweaver, H. and Burk, D. The Determination of Enzyme Dissoci-

ation Constants. J. Am. Chem. Soc., 56:658–666, 1934.
Monod, J., Wyman, J. and Changeux, J.-P. On the Nature of Allosteric

Transitions - a Plausible Model. J. Mol. Biol., 12:88–118, 1965.
P. Cluzel, M. Surette and S. Leibler. An ultrasensitive bacterial motor

revealed by monitoring signaling proteins in single cells. Science,
287:1652–1655, 2000.
FitzHugh, R. Impulses and physiological states in theoretical models

of nerve membrane. Biophys. J., 1:445–466, 1961.
Nagumo, J. and Arimoto, S. and Yoshizawa, S. An active pulse trans-

mission line simulating nerve axon. Proc. IRE, 50:2061–2070, 1962.
Langer, W.L. The Black Death. Sci. Am., 210:114–121, 1964.
Lee, K.J., McCormick, W.D., Ouyang, Q. and Swinney, H.L. Pattern

Formation by Interacting Chemical Fronts. Science, 261:192–194,
1993.
Goldstein, R.E., Muraki, D.J. and Petrich, D.M. Interface proliferation

and the growth of labyrinths in a reaction-diffusion system. Phys.
Rev. E, 53:3933–3957, 1996.
Cross, M.C. and Hohenberg, P.C. Pattern formation outside of equi-

librium. Rev. Mod. Phys., 65:851–1112, 1993.
Castets, V. and Dulos, E. and Boissonade, J. and De Kepper, P. Ex-

perimental evidence of a sustained standing Turing-type nonequi-
librium chemical pattern. Phys. Rev. Lett., 64:2953–2956, 1990.
De Kepper, P. and Castets, V. and Dulos, E. and Boissonade, J. Turing-

type chemical patterns in the chlorite-iodide-malonic acid reaction.
Physica D, 49:161–169, 1991.
Ouyang, Q. and Swinney, H.L. Transition from a uniform state to

hexagonal and striped Turing patterns. Nature, 352:610–612, 1991.
K.S. Kiprijanov. Chaos and beauty in a beaker: The early historyof

the Belousov-Zhabotinsky reaction. Ann. Phys. (Berlin), 3-4:233–
237, 2016.
A.T. Winfree. The prehistory of the Belousov-Zhabotinsky oscillator.

J. Chem. Educ., 8:661–663, 1984.
Keller, E.F. and Segel, L.A. Initiation of slime mold aggregation

viewed as an instability. J. Theor. Biol., 26:399–415, 1970.
Keller, E.F. and Segel, L.A. Model for chemotaxis. J. Theor. Biol., 30:
225–234, 1971.
Childress, S. and Percus, J.K. Nonlinear Aspects of Chemotaxis.

Math. Biosci., 56:217–237, 1981.
E.F. Keller. Making sense of life. Explaining biological development with

models, metaphors, and machines. Harvard University Press, Cam-
bridge, MA, 2003.
L.D. Landau and E.M. Lifshitz. Quantum Mechanics. Non-relativistic

theory. Pergamon Press, Oxford, UK, 2nd edition, 1965.
L.D. Landau and E.M. Lifshitz. Theory of Elasticity. Butterworth-

Heinemann, Oxford, UK, 3rd edition, 1986.
E. Lauga. The Fluid Dynamics of Cell Motility. Cambridge University

Press, Cambridge, UK, 2020.
bibliography 181
M.J. Lighthill. Mathematical Biofluiddynamics. Society for Industrial

and Applied Mathematics, Philadelphia, 1975.
C.C. Lin and L.A. Segel. Mathematics Applied to Deterministic Problems

in the Natural Sciences. Macmillan Publishing Co., Inc., New York,
1974.
S. Lipschutz. Differential Geometry. Schaum’s Outline Series. McGraw-

Hill Education, New York, 1969.
D.A. McQuarrie. Statistical Mechanics. Harper & Row, New York,

1976.
P. Nelson. Biological Physics: Energy, Information, Life. Chiliagon Sci-

ence, Philadelphia, PA, student edition, 2020.
R. Phillips, J. Kondev, J. Theriot, H.G. Garcia, and N. Orme. Physical

Biology of the Cell. Garland Science, Boca Raton, FL, 2nd edition,
1998.
J.S. Rowlinson and B. Widom. Molecular Theory of Capillarity. Oxford

University Press, Oxford, 1982.
C. Tanford. The hydrophobic effect: formation of micelles and biological

membranes. Wiley, New York, 1973.
D.W. Thompson. On Growth and Form. Dover Publications, Mineola,

N.Y., the complete revised edition, 1992.
9
Index
Chara corallina, 23 fast-inhibitor limit, 159 minimal surface, 83

Chlamydomonas, 106 Faxén’s laws, 128 minimum dissipation, 111
Dictyostelium discoideum, 168 FitzHugh-Nagumo model, 149 Monge representation, 56
Mytilus edulis, 106 flicker phenomenon, 88
Paramecium bursara, 103 Flory theory, 53 natural boundary conditions, 62
Volvox carteri, 22 fluctuating-dipole interaction, 28 Nernst equation, 23, 148
Fluctuation-Dissipation Theorem, 90 non-self-adjoint operator, 77
actin, 59, 64 follower force, 76 nullclines, 151
action potential, 148 Fredholm alternative, 162
activator-inhibitor model, 158 freely-hinged chain, 49 optical trap, 46
Adler equation, 119 freely-jointed chain, 50 Oseen tensor, 111
axon, 148 Frenet-Serret equations, 56
front propagation, 153, 161 persistence length, 59
bead-spring models, 116 phase dynamics, 119
Belousov-Zhabotinski reaction, 168 Gaussian curvature, 37 phase slip, 120
biharmonic eigenfunctions, 62 generating function, 41 pitchfork bifurcation, 67
bilayer membranes, 36 giant squid axon, 148 Poisson-Boltzmann equation, 32
bottom heaviness, 127 gyrotaxis, 132 polarisability, 29
boundary perturbation theory, 39
Hamaker constant, 31 rate independence of motion, 105
cAMP, 169 Helfrich repulsion, 86 Rayleigh dissipation function, 57
capillary length, 45 Hill coefficient, 147 reaction-diffusion systems, 157
catenoid, 83 Resistive Force Theory, 58
chemotaxis, 168 ion channels, 148
continuum limit, 44 scallop theorem, 106
cooperativity, 146 Jeffery orbits, 133 self-avoidance, 53
cutoff wavevector, 44 Jeffery’s equation, 131 singular perturbation, 98, 142
slaving, 147
Debye-Hückel, 32 Keller-Segel model, 169 steady-state approximation, 143
DNA, 59 stochastic Adler equation, 121
drag anisotropy, 72, 111 Lagrange multiplier, 61 Stokes Problems, 68
Langevin equation, 46 stokeslet, 111
elastic boundary layer, 98 Lineweaver-Burk plot, 144 stream function, 107
elastohydrodynamic length, 69 lubrication theory, 91 stresslet, 125
elastohydrodynamics, 68 stretch-coil transition, 74
entropic spring, 48 matched asymptotics, 98, 144
equipartition theorem, 42 mean curvature, 37 thylakoid membranes, 30
erythrocytes, 37 membrane tether, 96 time-reversal invariance, 106
Euler buckling, 65 metric tensor, 81 Turing instability, 165
excitable media, 148 Michaelis constant, 144 two-dimensional vesicles, 93
excitable medium, 152 Michaelis-Menten model, 141
excluded volume, 28 microtubules, 59 van der Waals, 27
virial coefficient, 25 waveform compliance, 120 Wiener-Khinchine theorem, 89

waving sheet model, 107

BPFD090623

Uploaded by

Copyright:

Available Formats

BPFD090623

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

BPFD090623

Uploaded by

Copyright:

Available Formats

T H E S C I E N T I S T D O E S N O T S T U D Y N AT U R E B E C A U S E I T I S U S E F U L ;

WORTH KNOWING, LIFE WOULD NOT BE WORTH LIVING.

published by department of applied mathematics and

First printing, June 2023

2 Microscopic Physics (Lectures 2-4) 25

3 Fluctuations & Brownian Motion (Lectures 5-7) 41

3.5 Polymer statistics and random walks 50

4 Biological Filaments (Lectures 8-12) 55

5 Surfaces & Membranes (Lectures 13-16) 81

6 Cellular Motion (Lectures 17-20) 103

6.2 Flagella and the physics of low-Re propulsion 106

7 Kinetics & Pattern Formation (Lectures 21-24) 141

7.5 Chemotaxis and Instabilities 168

1.1 Range of length scales and living systems of interest. 19

2.1 Temperature dependence of the second virial coefficient. 25

3.1 A string under tension. 42

4.1 Geometry of a curve in the plane. 55

5.1 A soap film in the form of a catenoid. 83

6.1 Examples of microswimmers. 103

6.25 A force dipole. 124

7.1 Solutions to the Michaelis-Menten dynamics. 143

1.1 Some physical quantities used in these notes. 20

3.1 Scaling exponent for self-avoiding walks in d dimensions. 53

4.1 The persistence lengths of various biological filaments. 59

6.1 Reynolds numbers for various swimming organisms. 104

Dedicated to John O. Kessler (1928-2022)

and Howard C. Berg (1934-2021), whose

enthusiasm for biological physics and deep

physical insight were a profound inspiration.

that end we have provided a series of simple Matlab programs for

R E Goldstein and E Lauga, July 2023

1.1 Length scales and methodologies

The field of biological physics covers phenomena spanning an enor-

Thermal ﬂuctuations Other forms of noise

proteins cilia hair ﬁsh

One of the key lessons of applied mathematics is the importance of

quantity symbol/definition value alternate units

Table 1.1: Some physical quantities

1.2 Scaling arguments

We now turn to several examples that illustrate the power of scal- 3

(proton + electron), the kinetic energy is p2 /2m, where p is the mo-

Here, rn∗ is n2 r B , where r B is the Bohr radius, and En = Ry/n2 , where

then we obtain the marvellously simple result

be its position during one-dimensional motion. Then the equation of

motion, with decay time τ.

If this is to be consistent with the Boltzmann factor of statistical

The second relation holds for water at room temperature, where ã is

∼ 107 s. Remembering that there are about 10π megaseconds in a

entrained by their motion, a phenomenon termed cytoplasmic stream-

2.1 Review of molecular physics

At the smallest scales of interest in these lectures we are concerned

canonical ensemble.2 We assume that classical statistical physics is 2

where pi is the momentum of particle i. Here, u is the intermolecular

non-interacting system (u = 0), where V is the volume of the system,

where uij = u(rij ) and Λ = (2πh̄2 /mk B T )1/2 is the thermal de

e− βuij = 1 + f ij with f ij = e− βuij − 1, (2.6)

and observe that f ij vanishes for a non-interacting system, and serves

the free energy F = −k B T ln Z, and then from F the pressure p =

cles around a given one (the radial distribution function) is uniform,

( p + aρ2 )(V − Nb) = Nk B T (2.13)