STRUCTURED FLUIDS

This page intentionally left blank

Structured Fluids
Polymers, Colloids, Surfactants

T. W I T T E N
The University of Chicago,
Chicago, Illinois, USA
with
P. P I N C U S
University of California,
Santa Barbara, USA

1
3
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi
Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal
Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam

Oxford is a registered trade mark of Oxford University Press

in the UK and in certain other countries

Published in the United States

by Oxford University Press Inc., New York

© Oxford University Press, 2004

The moral rights of the author have been asserted

Database right Oxford University Press (maker)

First published 2004

First published in paperback 2010

All rights reserved. No part of this publication may be reproduced,

stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above

You must not circulate this book in any other binding or cover
and you must impose this same condition on any acquirer

British library catalogue in Publication Data

Data available
Library of Congress Cataloging-in-Publication Data
Data available
Typeset by Newgen Imaging Systems (P) Ltd., Chennai, India
Printed in Great Britain
on acid-free paper by
The Bath Press, Avon

ISBN 978-0-19-958382-9
10 9 8 7 6 5 4 3 2 1
For Molly
This page intentionally left blank
Preface
This book is about fluids containing polyatomic structures such as polymer
molecules or colloidal grains. Such structured fluids have come to be known
as soft condensed matter. The book takes a scaling approach, following
de Gennes’s important monograph Scaling Concepts in Polymer Physics. My
purpose is to provide a unified, pedagogical introduction to soft-matter phenom-
ena that embodies this scaling approach. The book focuses on how to account
in the simplest way for the distinctive length, time, and energy scales that char-
acterize each phenomenon. This approach allows unity and simplicity, but it
provides only an initial glimpse into the rich phenomena and science to be found
in soft condensed matter. Notably, I omit an important aspect of soft matter: the
broken symmetry of the liquid crystalline state and the distinctive responses
and structures that arise from it. This realm of soft matter has benefited greatly
from the modern tools of differential geometry and field theory, as discussed
elegantly in the advanced text Principles of Condensed Matter Physics by Paul
Chaikin and Tom Lubensky (Cambridge, 1995).
The book began life in the late eighties as a joint venture with Phil Pincus
of the University of California, Santa Barbara. The chapters developed over
several cycles of teaching a course on structured fluids at Chicago. The intended
audience is the advanced undergraduate in physical science or engineering
who is comfortable with elementary physics and has seen elementary statistical
mechanics. The needed knowledge of physics is at the level of e.g., D. Halliday
and R. Resnick’s Fundamentals of Physics (New York, Wiley 2000). I assume
the student has seen statistical physics at the level of Reif’s book, the second
item of the list below. A working knowledge of Fourier transformation will also
be helpful.
Important conceptual points are illustrated by problems interspersed through
the text. The serious student should work these problems since the logical
development remains incomplete without them. Almost all of the problems
have been assigned, graded, and refined at least once. The book also includes
a number of suggested experimental projects using household materials. These
are intended to show concretely the principles discussed. They also give the
student a sense of how well the idealizations treated in the book apply to real
liquids. Students in the structured fluids course were required to do one of these
projects or some other project of their own devising. The projects are meant to
open a line of inquiry; only some have been tested, and the student may well find
that great modifications are needed to make them work. To guide the student to
further information, the book includes references to more detailed work. I have
tried to suggest sources in each major subject area. In some cases I have cited
primary journal articles, but these references are far from being thorough or
balanced. Corrections and additions to the text will be available via the Oxford
University Press web site, http://www.oup.com/.
viii Preface
In teaching the course, I found that the text contains too much material to be
covered in a one-quarter term with 27 class hours. It could probably be covered
rapidly in a one-semester course.
In learning from this book, there are several useful reference works to keep
in mind. The first and most important is de Gennes’s monograph on polymer
physics, Item 1 on the list below. Polymers serve as the paradigmatic system in
the book. Most of the ideas have been developed using polymers as an example,
then applied to other structured fluids. For further depth or clarification of these
ideas, de Gennes’s book is the best single source of which we are aware. The
subject depends heavily on ideas of statistical physics. The book develops all the
needed ideas within the text and aims to keep their number to a minimum. Still,
more explanation or depth on statistical physics may be helpful; a good source
is Reif’s book cited in Item 2. For the discussion of colloids and surfactants,
Item 3 is useful. This practical book by Jacob Israelachvili is about liquids
generally but contains much information about colloids and surfactants. For
collective properties of surfactant assemblies, Sam Safran’s book (Item 4) is
useful.
1. P.-G. de Gennes, Scaling Concepts in Polymer Physics (Ithaca NY: Cornell,
1979).
2. F. Reif, Fundamentals of Statistical and Thermal Physics (New York:
McGraw-Hill, 1965).
3. Jacob N. Israelachvili, Intermolecular and Surface Forces, 2nd ed. (London:
San Diego, CA: Academic Press, 1991)
4. Samuel A. Safran, Statistical Thermodynamics of Surfaces, Interfaces, and
Membranes (Reading, MA: Addison-Wesley Publication, 1994).
The reader should know about other classic and recent works that cover
the same material. The polymer physics material in Chapters 3 and 4 is
well covered by other textbooks and monographs. Two of the best known
are Paul J. Flory’s Statistical Mechanics of Chain Molecules (New York,
Interscience Publishers, 1969) and Charles Tanford’s Physical Chemistry of
Macromolecules (New York, Wiley 1961). There is also an important property
handbook: Polymer Handbook, 4th edition, by J. Brandrup, E. H. Immer-
gut and E. A. Grulke (New York, Wiley, 1999). An important monograph
written from a statistical physicist’s point of view is Polymers in Solution by
G. Jannink and J. Des Cloizeaux (Oxford, Clarendon Press, 1992). Another
monograph gives an authoritative treatment of polymer motions. It is The
Theory of Polymer Dynamics by Masao Doi and S. F. Edwards (New York,
Oxford University Press, 1986). Two recent textbooks offer a pedagogical
treatment of polymers: Statistical physics of Macromolecules by A. Yu. Gross-
berg and A. R. Khokhlov (New York, AIP Press, 1994), and Polymer Physics
by M. Rubinstein, R. H. Colby (New York, Oxford University Press, 2003).
A recent text covering a variety of structured fluids is Ronald G. Larson’s The
Structure and Rheology of Complex Fluids (New York, Oxford University Press,
1999). It is more advanced and more comprehensive than this book. Richard
A. L. Jones’s new text Soft Condensed Matter (Oxford, 2002), is similar to the
present book in scope and level.
 Preface ix
By now the efforts of several people have improved the text greatly. The
book owes much to the generations of students who took the course, read early
drafts, and forced me to be more clear. Arlette Baljon, Alfred Liu, Keith Brad-
ley, Joe Plewa, and other students weeded out errors and confusing statements
in the text. Jung-ren Huang improved every chapter by his close reading and
thoughtful suggestions. He also compiled the data for the big table of semi-
dilute universal ratios in that chapter. Olgica Bakajin did the simulation to
make the dilation symmetry figure in the appendix of Chapter 3. Sidney Nagel
and other colleagues at the James Franck Institute took a lively interest in the
course, and the book benefited from their discussions and support. Oxford
editor Bob Rodgers provided early encouragement and obtained several very
useful reviews of an early draft of the book. Seven other reviewers looked at
the completed book last year and many improvements were made in response
to their wise suggestions. Phil Pincus inspired this entire project and wrote the
first draft of Chapter 5. He also spent many hours over several collaborative
visits to bring the book along. The book would not exist without the support
and encouragement of my wife Molly.

Chicago T. A. Witten
May 2003
This page intentionally left blank
Contents

1 Overview 1
1.1 Introduction 1
1.2 A gallery of structured fluids 2
1.2.1 Self-organization 2
1.2.2 Rheology 4
1.2.3 Scaling 5
1.3 Types of structured fluids 6
1.3.1 Colloids 6
1.3.2 Aggregates 7
1.3.3 Polymers 7
1.3.4 Surfactant assemblies 9
1.3.5 Association 10
1.4 The chapters to follow 11
References 11

2 Fundamentals 13
2.1 Statistical physics 13
2.1.1 Thermal equilibrium 13
2.1.2 Probability and work 17
2.1.3 Lattice gas 22
2.1.4 Approach to equilibrium 24
2.2 Magnitude of a liquid’s response 27
2.3 Experimental probes of structured fluids 31
2.3.1 Macroscopic responses 31
2.3.2 Probes of spatial structure 34
2.3.3 Probes of atomic environment 38
Solution to Problem 2.1 39
References 40

3 Polymer molecules 41
3.1 Types of polymers 41
3.1.1 Monomers 42
3.1.2 Architecture 43
3.1.3 Polymerization 45
3.2 Random-walk polymer 47
3.2.1 End-to-end probability 48
3.3 Interior structure 54
3.3.1 Scattering 56
3.4 Self-avoidance and self-interaction 61
3.4.1 Local and global avoidance 62
xii Contents
3.4.2 Estimating D 64
3.4.3 Self-interaction and solvent quality 67
3.4.4 Universal ratios 72
3.4.5 Polyelectrolytes 73
Appendix A: Dilation symmetry 75
Appendix B: Polymeric solvents and screening 79
References 82

4 Polymer solutions 83
4.1 Dilute solutions 83
4.2 Semidilute solutions 85
4.2.1 Structure 86
4.2.2 Energy 87
4.2.3 Concentrated solutions and melts 90
4.3 Motion in a polymer solution 90
4.3.1 Brownian motion of a sphere 90
4.3.2 Intrinsic viscosity 94
4.3.3 Polymer in dilute solution: hydrodynamic
opacity 96
4.3.4 Internal f luctuations 98
4.3.5 Hydrodynamic screening 98
4.3.6 Semidilute diffusion 99
4.3.7 Semidilute self-diffusion without
entanglement 102
4.3.8 Motion with entanglements 103
4.3.9 Stress relaxation and viscosity 105
4.4 Conclusion 109
Appendix A: Origin of the Oseen tensor 109
Solution to Problem 4.5 (Deriving permeability) 110
References 111

5 Colloids 113
5.1 Attractive forces: why colloids are sticky 114
5.1.1 Induced-dipole interactions 114
5.1.2 Solid bodies 116
5.1.3 Perturbation-Attraction Theorem 117
5.1.4 Depletion forces 120
5.2 Repulsive forces 122
5.2.1 Steric stabilization 123
5.2.2 Electrostatic stabilization 127
5.3 Organized states 132
5.3.1 Colloidal crystals 132
5.3.2 Lyotropic liquid crystals 133
5.3.3 Fractal aggregates 134
5.3.4 Anisotropic interactions 134
 Contents xiii
5.4 Colloidal motion 135
5.4.1 Electrophoresis 136
5.4.2 Soret effect 137
Appendix A: Perturbation attraction in a square-gradient
medium 137
Appendix B: Colloidal aggregates 139
References 149

6 Interfaces 151
6.1 Probes of an interface 151
6.2 Simple fluids 153
6.2.1 Interfacial energy 153
6.2.2 Contact angle 156
6.2.3 Wetting dynamics 158
6.2.4 Surface heterogeneity 159
6.2.5 Other interfacial flows 160
6.3 Solutes and interfacial tension 161
6.3.1 Fluid mixtures 162
6.4 Polyatomic solutes 163
6.4.1 Polymer adsorption 163
6.4.2 Concentration profile 165
6.4.3 Hard wall 167
6.4.4 Kinetics of adsorption 168
6.4.5 Surface interaction 168
6.4.6 Flow 168
6.5 Conclusion 171
References 171

7 Surfactants 173
7.1 Introduction 173
7.2 Mixing principles 174
7.2.1 Positivity 175
7.2.2 Additivity 175
7.2.3 Ordering: like dissolves like 176
7.2.4 Reciprocity 176
7.2.5 Transitivity 177
7.2.6 Effect of permanent dipoles: water 177
7.2.7 Effect of charges: ionic separation 177
7.3 Surfactant molecules 178
7.4 Surfactants in solution: micelles 180
7.4.1 Open aggregation: wormlike micelles 182
7.4.2 Open aggregation: two-dimensional micelles 185
7.4.3 Aggregation kinetics 185
7.5 Micelle interaction 186
7.5.1 Energy of two-dimensional micelles 188
7.5.2 Energy to confine a fluctuating membrane 191
xiv Contents
7.6 Mixing immiscible liquids: microemulsions 192
7.6.1 Interfacial tension 195
7.6.2 Emulsions and foams 197
7.7 Amphiphilic polymers 198
7.7.1 Micelle size 199
7.7.2 Other copolymers 201
7.7.3 Polymeric amphiphiles in solution 202
7.8 Dynamics and rheology 203
7.8.1 Wormlike entanglement and relaxation 204
7.8.2 Rheology of lamellar solution 206
7.8.3 Shear-induced restructuring 207
Appendix: Gauss–Bonnet Theorem 210
References 211

Index 213
 1
Overview

1.1 Introduction
As physical scientists, we are concerned with the behavior of matter in all
its forms. We want to know what matter does and why. This is our goal
in studying the primordial universe, the tenuous interstellar medium, the
gaseous atmosphere of the Earth, the ionized plasma of the Sun’s corona,
the mundane liquids and solids of our human surroundings, and the exotic
dense matter within a molecule, an atomic nucleus, or a proton. This book
is about a tiny subset of this vast range of forms of matter: structured fluids.
Structured fluids are liquids, i.e., condensed matter in which the atoms are
adjacent but freely mobile on a local scale. These liquids contain connected
polyatomic structures, such as solid grains or large molecules† . Though † By “polyatomic structure” we mean any

this is a small sampling of all forms of matter, it is a significant one. These assembly of many atoms that retains its
integrity over experimental time scales.
materials are important to our basic goal in studying matter: to explore the
This includes macromolecules and also
limits of what matter is capable of doing. Though they form a narrow range more weakly-bound structures.
of materials, structured fluids show a wide range of behavior. Indeed, they
show behavior not seen in other forms of matter, however exotic.
We may glimpse this distinctive behavior by recalling a few structured
fluids from everyday experience: pancake syrup, egg white, Silly Putty,
corn-starch-and-water paste, or toothpaste. Though each of these gooey
materials resembles a liquid, all respond to forces in a distinctive way that
distinguishes them qualitatively from simple fluids and from one another.
They owe their distinctive properties to connected structures much larger
than an atom, but much smaller than a macroscopic, classical body. They
belong to an intermediate, mesoscopic size regime. These structures give
raw egg white its springy consistency. They cause flowing corn-starch-
and-water paste to shatter like a brittle solid when struck. And they cause
toothpaste to flow out of its tube as a plug. These structures individually
confer distinctive properties to a liquid. More importantly, they interact
to create new forms of cooperative behavior and self-organization, as we
shall illustrate below. Some of these structured fluids have an additional
significance: they are important constituents of living cells or are important
technologically.
It was not possible until recently to study the molecular basis of
structured-fluid behavior. Now however experimental probes such as elec-
tron microscopy and dynamic scattering allow us to study structured fluids
with a resolution suited to the distinctive mesoscopic structures in them.
In addition present-day sophistication in chemical synthesis has made it
2 Overview

possible to make fluids whose polyatomic constituents are precisely known.

Such fluids serve as models for the types of behavior possible in structured
fluids.
In the following chapters we shall explore how the special properties of
structured fluids arise from their polyatomic structures. Our goal is to show
how a wide variety of important properties can be understood and quanti-
tatively predicted using a few simple principles. But before embarking on
this study, it seems good to give you a foretaste of the variety of behavior
that structured fluids can show in the laboratory. The next section shows a
few striking examples. The following section describes the broad classes
of structured fluid.

1.2 A gallery of structured fluids

1.2.1 Self-organization
Figure 1.1 shows the striking spatial self-organization that structured flu-
ids can give to matter. The figure is an electron micrograph of a fluid of
polyatomic molecules called block copolymers. Each polymer is a long,
flexible chain hydrocarbon molecule. Roughly one-third of each polymer
chain is a repeated sequence of monomers called isoprene. The other two-
thirds of each polymer are made of another type of monomer called styrene.
Once prepared and equilibrated, the fluid is frozen and cut into slices tens

Fig. 1.1
Transmission electron micrograph of the
bi-continuous copolymer domain structure,
reproduced from [1]. © 1987 American
Chemical Society. Repeat distance is about
100 nm. The minority species in black
forms two disjoint domains, labeled 1 and
2. The sketch at right, after [2], shows the
three-dimensional structure inferred from
such micrographs and from x-ray
diffraction studies. © American Physical
Society. The vertical dashed line gives the
line of sight for the micrograph. The
thickness of the two minority domains is
reduced to make the illustration clearer.
Domains 1 and 2 are shown with different
shadings to distinguish them. The
magnified sketch at bottom shows the
orientation of individual polymers in the
domains.
 A gallery of structured fluids 3

of nanometers thick. The two types of monomers can be made to scatter

electrons differently, so that when a beam of electrons passes through a
slice, regions denser in isoprene cast shadows; it is these shadows that form
the micrograph. Evidently the polymers have assembled themselves into a
periodic pattern. It is strikingly different from conventional crystals in both
shape and size. Half of the isoprene forms a network of connected nodes
in the so-called “gyroid” structure [3]. The other half forms an identical
network, displaced from the first by half a unit cell. Each spatial unit cell
is several tens of nanometers across and contains several hundred poly-
mer molecules. The same structure has now been seen in several types of
block copolymer fluids. How is it that these molecules choose to organize
themselves spontaneously into such a domain structure? What features of
the polymer molecule control the size and shape of the domains? In the
chapters to follow we shall develop the concepts needed to address such
questions.
Figure 1.2 shows another form of strong, mesoscopic organization,
caused by small surfactant molecules. It is an electron micrograph of a
liquid containing water, oil, and surfactants. Surfactants have the property
of segregating themselves at water–oil interfaces; they are roughly the size
of a water or an oil molecule. To make the micrograph, the liquid was rapidly
frozen, and the frozen liquid was fractured. Staining techniques were used
to reveal the topography of the fracture surface. We see that the fracture
surface is made up of rounded regions a few tens of nanometers in size.
Many studies of such mixtures support the view that these are regions of
unmixed water or oil. The surfactant has induced the limited mixing of the

Water

Octane

Fig. 1.2
Freeze-fracture micrograph of a
water–oil–surfactant microemulsion
reprinted with permission from [4].
50
© 1988 American Chemical Society. The
smooth hills and valleys are thought to
represent the boundaries of oil- and
water-containing regions.
4 Overview

oil and water into these mesoscopic domains. In later chapters we shall be
interested in how small surfactant molecules can induce oil and water to
mix at this coarse scale. We shall consider what features of the molecules
govern the size of the domains and discuss how the evident randomness of
the interfaces may be characterized quantitatively. Such questions are of
broad significance, because surfactant-mediated interfaces between liquids
are ubiquitous. In everyday life and in technology surfactants at interfaces
are used to control the mixing of oily and watery liquids. And living cells
use these surfactant interfaces as a fundamental structural building block.
Mesoscopic structures can create macroscopic self-organization. The
spatial scale of the stripes in Fig. 1.3 is a few millimeters. This is a pho-
tograph of a black droplet of “ferrofluid” sandwiched between two glass
plates a few millimeters apart. The plates and fluid are in a magnetic field
of several hundred Gauss, pointing out of the figure. As the field increases
from zero the ferrofluid droplet goes from a circular shape to the labyrinthine
Fig. 1.3 shape seen here. Evidently this fluid interacts strongly with magnetic fields.
Labyrinth pattern of a ferrofluid drop It consists of 10-nm bits of the magnetic mineral magnetite dispersed in
(courtesy of R. E. Rosensweig) [5]. A drop
of ferrofluid was put between two
kerosene. In this fluid the polyatomic structures of interest are these bits of
horizontal glass plates about a millimeter dispersed solid. Solids dispersed in this “colloidal” form often have strong
apart. The plates were then placed between interactions with external forces such as the magnetic field used here. As we
the poles of a magnet. As the (vertical) proceed, we shall want to account for these strong interactions and explore
field increases from zero, the circular their limits.
droplet of fluid develops a fringe of spikes,
which elongate and branch to fill the area
Just as colloidal particles interact strongly with external fields, they
between the plates with the pattern shown. can interact strongly with each other, leading to further forms of
The thickness of the channels is self-organization. They can repel each other so strongly that they form
comparable to the spacing of the plates. a crystalline arrangement like that of Fig. 1.4. Figure 1.5 shows what can
happen when dispersed colloidal particles are suddenly made to attract each
other strongly. The particles assemble into a complex branched aggregate.
The figure is a transmission electron micrograph of colloidal silica particles
in water that have aggregated and then settled on a carbon film. A drop of
the water was placed on a carbon film, dried, and then placed in the micro-
scope. The aggregate looks different from a bulk precipitate: it is wispy and
tenuous. Aggregates like this are used commercially to toughen rubber in
tires and to thicken liquids like ice cream. In the chapters to follow, we shall
consider what makes aggregates take this form and we shall explore ways
to characterize its wispiness quantitatively. We shall also investigate how
such wispy objects interact with the fluid and with other aggregates near
them.
10 μm

1.2.2 Rheology
Fig. 1.4
An emulsion of water, decane oil, and The self-organizing behavior of structured fluids illustrated above is only
surfactant in which the oil droplets were part of their distinctive behavior. Further striking behavior is revealed when
purified to have uniform size and then the equilibrium of a structured fluid is perturbed. A classic example is shown
concentrated [6], © American Physical in on the right side of Fig. 1.6, a photograph of a polymer solution in a beaker.
Society, courtesy J. Bibette. The droplet
diameter is 0.93 μm. The long parallel
A rotating rod has been inserted into the center. The polymer liquid climbs
rows of particles and the crosshatched this rod like bread dough in a mixer. A simple liquid behaves oppositely;
texture indicate an incipient periodic the rotating liquid drops in the center and rises up the outer wall because
lattice. of centrifugal force. In the sequel we shall identify the forces that cause
 A gallery of structured fluids 5

Fig. 1.5
Electron micrograph of aggregated
3.5-nm-radius particles of colloidal silica
from Lin et al. [7], courtesy of D. Weitz,
reprinted by permission from Nature
© 1989 Macmillan Publishers Ltd.
Aggregate was formed in water under
diffusion-limited conditions. The fraction
of the volume occupied by particles is
expected to vary as the sphere radius to
the −1.3 power whenever aggregation is
diffusion limited. This “universal” scaling
has been tested for a variety of materials.

Fig. 1.6
Rod-climbing liquid, reproduced from [8]
reproduced by permission of John Wiley &
Sons, Inc. (© 1977 John Wiley & Sons).
The control fluid in the left has the same
viscosity as the polymer solution on the
right, but is depressed in the center rather
than climbing the rod.

this rod-climbing behavior. We shall investigate what rate of stirring is (a) (b)
required to produce this effect. A less classical effect is sketched in Fig. 1.7.
This sketch depicts another polymer liquid. But the polymers in this liquid
contain surfactants on some of their monomers. When the liquid is properly
formulated, the act of quickly inverting its container produces a qualitative
change. The quiescent liquid has the consistency of motor oil; the disturbed
liquid is like half-cooled Jello. Phenomena like this show the potentialities
of these liquids for surprising behavior. We can only begin to sketch its
origins in what follows.

1.2.3 Scaling Fig. 1.7

Shear-induced gelation in an associating
When discussing the properties of a structured fluid like those shown polymer solution. (a) Moderate-viscosity
above, it is useful to imagine that the mesoscopic structures in the fluid state in gentle flow. (b) Gel state after
are indefinitely large relative to their small-molecule constituents. Thus rapidly inverting the jar.
6 Overview

various measured properties can be expected to show a simple asymptotic

dependence on the structures’ size. A fundamental question of this type
concerns the viscosity of a polymer liquid. It is natural to expect that this
viscosity may grow as a power of the molecular weight of the polymers.
Our focus will be to ask whether this asymptotic dependence is indeed a
power law and to predict the exponent. We shall keep our focus on such
fundamental scaling issues and avoid more detailed predictions. In this way
we can isolate the basic respects in which our structured fluids are distinc-
tive. We can also account quantitatively for a broad range of liquids from a
unified point of view.

1.3 Types of structured fluids

Having illustrated these examples of structured fluid behavior, we now
describe the major classes of structured fluids in more detail, with some
important examples for each. We begin with the simplest type, colloids.

1.3.1 Colloids
Colloids are fluids containing compact, polyatomic particles suspended
in a liquid solvent. A familiar example is black ink, which is made from
colloidal carbon. Colloidal particles give distinctive physical properties of
fluids. Thus the colloidal form of the carbon in black ink is what makes it
absorb light efficiently. These distinctive physical properties are incidental
to many applications. Instead the colloidal particles are often present to
give the fluid some chemical property of interest. Examples are the cells
in blood, manufactured particles which remove molecules from a liquid
by adsorbing them, the light-sensitive grains in photographic film, or the
pH-buffer particles in detergent motor oil. In other cases, such as paint and
rubber cement, the colloidal particles produce the desired property not in the
fluid but in the solid that forms when the solvent dries. Powder processing of
ceramics utilizes the fluid properties of suspensions of ceramic powders to
optimize packing in order to achieve low defect bulk ceramics for structural
materials.
There are colloids that are important for the physical properties they
impart to fluids. One such colloid is the dispersion of 10-nm particles in
the ferrofluid of Fig. 1.3. In moderate magnetic fields the magnetic energy
of the particles is strong enough to alter the fluid’s energy significantly. In
a similar way, colloidal particles, because of their bulk, respond strongly
to electric fields or to flow. In emulsions and foams the dispersed particles
are liquid or gaseous droplets. These can be concentrated so much that
the dispersed particles push against each other and give the dispersion the
solid-like consistency of mayonnaise or whipped cream. Figure 1.4 shows
an example of a concentrated emulsion.
The interaction energy of two colloidal particles in a given solvent is
also magnified because of their bulk. Consequently, small changes in the
solvent can have a large effect on the interaction energy. This makes it
possible to change the interaction between two colloidal particles abruptly
from an effective hard-core repulsion to an attraction whose strength is
 Types of structured fluids 7

many times the thermal energy kB T † . With such an attraction the particles † Here T is the absolute temperature and kB
must stick together when they encounter each other. The particles flocculate is the Boltzmann constant. The next chap-
ter reviews the significance of the thermal
or precipitate. This effect is exploited to sense small changes in a solution or
energy kB T .
to determine the presence or absence of a biological antigen—as in certain
early pregnancy tests.

1.3.2 Aggregates
The enhancement of particle–particle interaction between colloids makes
possible a form of self-organization not seen in simple molecular liquids—
namely colloidal aggregates like that of Fig. 1.5. Such aggregates form
when the attraction between two particles in contact is so strong that they
must stick together permanently. They cannot slide or roll around in order
to maximize the amount of contact. The result is the tenuous structure
shown in the figure. It is qualitatively unlike the dense-particle phase that
forms when small molecules precipitate from solution. The average particle
density within a radius r of a given particle decreases as r increases. Thus the
average density in an arbitrarily large aggregate becomes arbitrarily small:
the aggregates are “fractal” structures [9]. The origins and the consequences
of fractal structure will be a large concern in the chapters to follow.
Aggregated colloids show the enhancement effects discussed above for
dispersed colloids. In addition they have properties arising from their frac-
tal structure. Even though a tenuous aggregate occupies an arbitrarily
small fraction of the volume it pervades†† , it transmits forces efficiently †† The pervaded volume of an object is the
throughout the pervaded volume. An aggregate in a shear flow screens the spherical region of space that contains the
object.
surrounding solvent: the fluid is obliged to flow around rather than through
the aggregate. Since this screening inhibits flow, it enhances viscosity. Thus
the tenuous property of these aggregates makes them particularly effective
in increasing viscosity. Their fractal structure confers many other distinctive
properties to be explored in the course.

1.3.3 Polymers
Another way of producing a tenuous, polyatomic structure is to link small
molecules together into a flexible chain to form a polymer, as shown in
Fig. 1.8. The successive bonds between the monomers making up a flexible
chain have some randomness in their relative directions. Thus the directional
correlation between bonds more than a few bond lengths apart becomes
negligible; accordingly, a long polymer has the statistical properties of a
random walk. In some situations the statistical properties of a long polymer
are instead those of a self-repelling random walk. Like colloidal aggregates,
such polymers have the spatial scaling properties of a fractal: the average
monomer density of a self-repelling polymer of size R scales as R −4/3 .
This power for a simple random walk is −1. Chains whose pervaded-volume
fraction is a tenth of a percent or less can be readily produced. In a solution of
such polymers, there is room for a thousand chains in the volume pervaded
by one chain.
The flexibility of a tenuous polymer gives it properties that a tenuous
aggregate does not have. Unlike an aggregate, a polymer is not quenched
8 Overview

Fig. 1.8
Detail of a large polystyrene molecule as it
might appear in a good solvent such as
toluene. Each sphere represents a carbon
atom and one or two small hydrogen atoms
attached to it. The chemical bonds are
superimposed on one repeating unit or
monomer, and on a section of the chain
backbone. The backbone bonds may rotate
freely. A few successive backbone repeat
units are labeled 0, 1, 2, . . . . The vector a 1
for a four-monomer segment is shown.
This structure was generated by a Monte 2
Carlo computer simulation [Simulation by
M. Mondello, H.-J. Yang and R. J. Roe at 3
a
the University of Cincinnati using the Cray 4
Y-MP at the Ohio Supercomputer Center,
circa 1990, private communication], which
simulates the random rotations of the
bonds as they might occur in solution.
The simulated molecule has about 1/20 the
mass of those in a typical polystyrene
foam cup.
into some particular configuration but is free to explore the ensemble of
random bond directions. The randomness in each chain’s configuration thus
amounts to thermodynamic entropy, which may serve as a reservoir for heat
and work. The chain may be dramatically deformed by mild perturbations
without permanent effects. Thus the spontaneous thermal fluctuations in the
end-to-end distance of a flexible polymer are about as large as the average
end-to-end distance. Consequently, externally imposed energies as small as
the thermal energy kB T are sufficient to distort the shape of a polymer by
factors of order unity.
Unlike rigid colloidal aggregates, polymers may be concentrated to vol-
ume fractions up to unity. In this solvent-free limit called the melt state,
chains interpenetrate and entangle strongly. Each chain interacts directly
with hundreds of others and the forces thus communicated can produce
large, reversible deformation in each chain. The deformation of these ran-
dom chains is what produces the restoring force in a rubber band or a
plate of Jello. When a polymer liquid is abruptly deformed, it too responds
elastically over short times, like a rubber band. But over longer times the
chains disentangle and forget their initial distortions. The time scale for
disentanglement can easily be as long as seconds. By mixing polymers of
different lengths and architectures, one may produce liquids that behave
like a tough rubber on short time scales, like a weak rubber on longer
time scales and like a flowing latex at yet longer times [10]. This power
to control the storage of energy over time allows one to adapt polymer
solutions to the needs of a many-step manufacturing process, such as the
assembly of a car tire. Similarly, many everyday structured fluids such as
pancake syrup, shampoo, and paint are deliberately thickened to keep them
from flowing too much during application. The energy stored in a polymer
when it is deformed has striking effects on the flow properties of the liquid.
Elongating a chain produces spring tension along its length. This “normal
 Types of structured fluids 9

stress” combines with other applied forces to accelerate each small volume
of the fluid. The elastic energy stored in the polymers can easily exceed
the kinetic energy in the flowing fluid. This produces nonintuitive flow
properties like the rod-climbing behavior of Fig. 1.6 [8]. One such effect
of commercial importance is turbulent drag reduction. Trace amounts of
a polymer can substantially reduce the power required to push a turbulent
fluid through a pipe, in spite of the small increase in viscosity due to the
polymer [11].
The deformability of a polymer has dramatic consequences when elec-
trically charged species are attached along the polymer chain at a given
density. The electrostatic repulsion in a long enough chain is sufficient to
stretch the chain from a random-walk configuration to that of a rigid rod,
for which the end-to-end distance is proportional to the molecular weight.
Polyelectrolytes, as such polymers are called, can be controlled in a fashion
not possible for neutral polymers. When the interaction along the chain is
screened, by the addition of some salt or the presence of other chains, the
electrostatic repulsion is reduced and the polyelectrolyte chain shrinks in
size. This alters fluid properties such as the viscosity.
Another important consequence of the deformability of polymers is seen
in their behavior near an interface. Even when the binding energy of a
monomer to a surface or interface is much smaller than the thermal energy
kB T , the total binding energy of a polymer chain made up of such monomers
may be several times the thermal energy. Then the polymer chain can
increase its binding by flattening itself closer to the surface, with little
cost in deformation energy. The distinctive features of the adsorbed state
will be considered later.

1.3.4 Surfactant assemblies

Self-assembled surfactants like those at the interfaces in Fig. 1.2 are a further
major category of structured fluid. A surfactant molecule is amphiphilic: it
incorporates parts that if not connected would be strongly immiscible. A
common example is hexadecyl trimethyl ammonium bromide (CTAB)—a
hydrophobic 16-carbon chain with a bulky ammonium ion and a bromide
counterion at one end. Isolated surfactant molecules cannot exist in either oil
or water (except in extremely low concentrations). Instead, these molecules
assemble themselves into “micelles”. A micelle is a configuration in which
the molecules’ immiscible parts clump together, thus minimizing their con-
tact with the solvent. In water for example, the hydrophobic hydrocarbon
tails clump together, while the polar heads point outward towards the water.
In an oil the situation is inverted: the inverted micelles have their ionic parts
clumped together.
Micelles typically have diameters on the order of a few nanometers; they
constitute a colloid-like polyatomic species in their own right. Further-
more a micelle-containing fluid shows interfacial properties unlike anything
discussed above. Any water–oil interface clearly forms a very favorable
environment for a surfactant; at the interface the molecule’s hydrophilic
end can be in water and the hydrophobic tail can be in oil. Accordingly
surfactants assemble readily at these interfaces. The assembly’s energy is
10 Overview

lower than that of a random dispersion of surfactants in either oil or water.

The lowering of the energy amounts to a lowering of the interfacial free
energy or interfacial tension. Surfactants thus make water and oil more
nearly miscible.
In certain conditions a surfactant can reduce the interfacial tension vir-
tually to zero, allowing the interfacial area between oil and water to grow
spontaneously. The result is a microemulsion like that shown in Fig. 1.2—a
thermodynamically stable mixture of oil, water, and surfactant that is full
of fluctuating oil–water interfaces.

1.3.5 Association
The broad range of behavior discussed above is augmented even further
when we consider “association” of mesoscopic structures. By association
we mean a temporary joining together of the structures. The structures thus
joined can transmit forces and thus alter mechanical properties strongly. But
still these junctions are weak enough to break and reform over the time of an
experiment. Thus the associations alter themselves in response to the local
stress or flow in the liquid. We have already met the most prevalent form of
association: the joining together of surfactant molecules to make a micelle
fits the definition of association. Because these micelles are temporary,
the number and the average size of the micelles relaxes over time when
the micellar solution is subjected to e.g., a temperature jump. Association
behavior is common in colloidal aggregates. For example, silica aggregates
in water may be made to associate with one another via hydrogen bonds.
The result is a type of gel network whose links may be broken and readily
re-formed. Such a “network fluid” holds its shape in quiescent conditions
but flows under sufficiently strong shear. Upon removal of this shear, the
network is reestablished and the flow stops. These associating aggregates
are used to keep paint from running before it dries.
One can make polymers associate by attaching immiscible chemical
groups sparsely along the chains. Like surfactants, these groups assem-
ble themselves into micelles and form temporary crosslinks between the
polymers. Such polymer solutions can show a new form of response to
shear, namely reversible, shear-induced gelation. This is the phenomenon
sketched in Fig. 1.7. Another form of polymer association occurs in the
block copolymers of Fig. 1.1. Block copolymers, being amphiphilic, have
properties similar to those of surfactants. But the polymers’ entanglement
and deformability cause their resulting micellar microdomains to have some
distinctive properties not seen in surfactants. We discuss these in Chapter 7.
An example of commercial importance is Kraton, a rubbery polymer tipped
at each end with a small section of an immiscible, glass-forming polymer.
The ends of the Kraton polymer congregate into spherical micelles, each
containing many chain ends. The midsections of such a copolymer, being
attached at both ends, cannot disentangle themselves, and a strong, rubbery
material results. But when the material is heated above about 100◦ C the
spherical micelles at the ends melt and become more miscible, so that the
material flows. It can then be molded and processed. Such polymers, called
 References 11

thermoplastic elastomers, are used e.g., in adhesive coatings and for the
elastic stripes painted on disposable diapers.

1.4 The chapters to follow

Now that we have surveyed the varieties of structured fluids, we are ready
to try to understand them in the chapters to follow. Chapter 2 deals with
fundamental principles that underlie the treatment of all the systems to be
treated later. The first step is to review the principles of statistical physics
that form the basis of all the phenomena to follow. Next, we discuss the
orders of magnitude of the microscopic phenomena on which structured-
fluid properties are based. That is, we consider the energy, length, and
time scales characteristic of simple liquids. From these we infer the order
of magnitude of the viscosity of typical liquids. Finally, we discuss the
experimental techniques—some new, some old—by which structured fluids
are studied. We shall want to keep these concrete tests of our ideas in mind
as we try to make predictions about the fluids studied later.
In the rest of the book we discuss the classes of structured fluid in turn.
In Chapters 3 and 4 we start by treating polymers. We discuss the charac-
teristic length, energy, and time scales of single polymers and of strongly
interacting polymer solutions. In so doing we lay the groundwork for quan-
titative understanding of the distinctive responses of polymer liquids: their
elasticity and their high viscosity. In Chapter 5 we turn to colloidal fluids.
Many aspects of these fluids prove to be directly understandable using ideas
developed for polymers. But there are also new aspects. The chief one is
to understand why large colloidal particles are difficult to disperse in a sol-
vent, and how these difficulties may be overcome. Chapter 6 discusses
the interaction between fluids and interfaces: liquid–liquid, liquid–gas,
and liquid–solid. We discuss surface and interfacial tensions, statics and
dynamics of wetting, and other related topics including adhesion and forces
between colloidal particles induced by adsorption and depletion layers. In
Chapter 7 we consider surfactant assemblies. Here our central question is
how amphiphilic molecules are able to promote the mixing of immiscible
liquids. In understanding this puzzle, we are led to explore the statistical
properties of the fluctuating surfactant-coated interfaces.

References
1. H. Hasegawa, H. Tanaka, K. Yamazaki, and T. Hashimoto, Macromolecules 20
1651 (1987).
2. M. W. Matsen, Phys. Rev. Lett. 80 4470 (1998).
3. D. A. Hajduk, P. E. Harper, S. M. Gruner, and C. C. Honeker, Macromolecules
27 4063 (1994); M. W. Matsen and F. S. Bates, Macromolecules 29 1091 (1996).
4. W. Jahn and R. Strey, J. Phys. Chem. 92 2294 (1988).
5. See e.g. R. E. Rosensweig, in Physics of Complex and Supermolecular Fluids,
ed. S. A. Safran and N. A. Clark (New York: Wiley-Interscience, 1987), p. 699.
6. J. Bibette, D. Roux, and F. Nallet, Phys. Rev. Lett. 65 2470 (1990).
7. See e.g. M. Y. Lin, H. N Lindsay, D. A. Weitz, R. C. Ball, and R. Klein, Nature
339 360 (1989); R. Jullien and R. Botet, Aggregation and Fractal Aggregates
World Scientific (1987).
12 Overview

8. R. B. Bird, O. Hassager, R. C. Armstrong, and C. F. Curtiss, Dynamics of

Polymeric Liquids (New York: Wiley, 1977).
9. See L. M. Sander, Sci. Am. 255 94 (1987); B. B. Mandelbrot, The Fractal
Geometry of Nature (San Fransisco: Freeman, 1982).
10. For a review, see D. S. Pearson, Rubber Chem. Technol. 60 439 (1987).
11. R. B. Bird and C. F. Curtiss, “Fascinating Polymeric Liquids,” Phys. Today
January, (1984) p. 36.
 2
Fundamentals

This chapter assembles some underlying concepts that we will need through-
out our study of structured fluids. The first section reviews the main needed
ideas from statistical mechanics. For systems in equilibrium, we recall how
probabilities of states are determined, and the connection between probabil-
ity and the work to alter a system’s state. We also describe ways to estimate
the time for a system to reach equilibrium. The second section is about the
orders of magnitude common to the liquid environment of structured flu-
ids. In this section we justify the magnitude of a simple liquid’s viscosity.
The final section surveys the experimental probes that determine the useful
questions we can ask about structured fluids.

2.1 Statistical physics

2.1.1 Thermal equilibrium
All the structured fluids mentioned in the last chapter are intimately
controlled by the laws of thermodynamics. The conditions of thermal equi-
librium impose the dominant limitations on how these fluids can behave.
The bulk of the understanding achieved to date about these systems comes
from the laws of thermodynamics and statistical mechanics. In this section
we recall a few principles of statistical physics that will be important for the
rest of the course. We will discuss what determines the probability of ran-
dom states in our structured fluid. We’ll recall the meaning of temperature
and the ability of a statistical system to do work.
The structured fluids we’ll meet in the course contain many forms of
randomness, from the random bond angles of a polymer chain to the random
positions of the molecules of a simple liquid. But for this discussion let us
think of a simple and familiar example: a dilute monatomic gas. Each atom
of the gas has a random position and momentum. Nevertheless, not all
positions and momenta are equally likely.
To understand the relative likelihood of these random variables, we define
the notion of a configuration or microstate of the gas. The configuration is
the set of all the positions {ri } and momenta {pi } of all the N atoms. It will
save us from some awkwardness later to imagine that these positions and
momenta are discretely spaced at very fine intervals. This is natural since
the coordinates and momenta can only be known to limited precision in
any case. Thus all the randomness of our system is specified by telling the
relative probabilities f ({ri }, {pi }). In an arbitrary system, we shall denote
14 Fundamentals

a given configuration by the simple label c; any configuration c has a rel-

ative probability f (c). The absolute  probability p(c) is just obtained by
normalizing the f (c): p(c) = f (c)/( c f (c)). Since f is only a relative
probability, multiplying it by a constant (independent of c) does not change
its meaning at all. Any two f ’s whose ratio is independent of c are clearly
equivalent.
In principle a system such as our gas can be prepared in any specific
configuration c or with some initial probability distribution f0 (c). But in
general it does not stay in its initial distribution. In our gas for example the
coordinates and momenta of the particles are constantly changing because
of the motions of the atoms and collisions between them. Thus their relative
probabilities in general change from what they were initially as time goes
on, until a final steady distribution f (c) is reached. This final steady state
is called the equilibrium state.
A gas isolated in a container is a closed mechanical system. If it is initially
in a configuration c, it can evolve into many others as time goes on. But
there are many configurations it cannot evolve into. An immediate reason
for this is the conservation of energy. Since the total energy E in the gas
is conserved, the system is confined to configurations c with the same
energy as the original one c0 : E(c ) = E(c0 ). That is, f (c ) = 0, except
if E(c ) = E(c0 ). For many systems this is the only restriction on the
configurations c , however; all the configurations with the proper energy are
accessible. The hypothesis of statistical mechanics is that all these accessible
states in a closed and isolated system are equally probable. This relaxation
to this uniform probability distribution can be proven to occur under a
broad range of conditions; we shall assume that it occurs for the systems
we will study. That is, we will assume our system attains thermodynamic
† Throughout the book this font will equilibrium† .
flag each new term when it is formally The systems we will study consist of many degrees of freedom. In our
introduced.
gas, each cartesian coordinate or momentum of each atom is a degree of
freedom; in our polymer molecule each bond angle of each monomer is a
degree of freedom. We specify the configuration c of a system by specifying
the values of all its degrees of freedom.
This behavior of an isolated system leads to an interesting behavior of
parts of the system. In our gas system, for example, we can infer the prob-
abilities of a single atom’s coordinates and momenta. A subsystem, such as
a single atom, is a subset of the degrees of freedom of the overall system. A
single atom of mass m in a dilute gas has a fairly well-defined energy  that
depends on the configuration of its own degrees of freedom:  = 12 p 2 /m.
This energy is not conserved, since the atom occasionally transfers energy
by collisions. During a collision, the energy of an atom is not simply the
kinetic energy formula given above. Still, with increasing dilution, this
formula becomes more and more accurate.
In situations like this, we may infer the equilibrium distribution for sub-
systems like our atom, which are a small part of a large, isolated system
with fixed total energy Et . The subsystem has its own set of configurations
c, each with some relative probability f (c). We consider subsystems which
have a well-defined energy E(c); this energy is assumed to be little affected
by anything beyond the subsystem, as is the case for a dilute-gas atom. The
 Statistical physics 15

rest of the system beyond the subsystem has a configuration labeled by d;

this large set of degrees of freedom outside the subsystem of interest is
called the reservoir. Thus to specify the configuration of the whole system,
we must specify both c and d. First we consider two subsystem configura-
tions c and c with the same energy: E(c ) = E(c). The relative probability
f (c) is given by the total number of reservoir configurations d that are
compatible with that c. The only constraint on these d’s is that their total
energy must be the fixed total energy Et less the subsystem energy E(c).
All reservoir configurations with energy Et − E(c) are equally allowed.
This same set of reservoir configurations is equally allowed for the c con-
figuration. The number of overall configurations with sub-configuration c
is the same as the number with sub-configuration c; thus f (c ) = f (c).
We are led to the conclusion that configurations of a given subsystem with
the same energy have equal probability. Thus, f (c) can only depend on c
through its energy: f (c) = f (E(c)).
To determine how f depends on E we consider two separate subsys-
tems within a given larger reservoir (e.g. two atoms in our gas). The first
subsystem has configurations denoted c1 with energy E(c1 ); the second
one has configurations denoted c2 and energy E(c2 ). (The two subsystems
need not be the same size. For example, subsystem 1 could contain three
atoms, and subsystem 2 could contain 75 other atoms.) The only way our
separate subsystems have any connection is that they belong to the same
larger reservoir. Thus if one has more energy there is in principle less energy
available for the other. But the total energy Et is so much larger than the
energy of either subsystem that this effect is negligible. In the limit of a large
reservoir, the effect of one subsystem on the other must become negligibly
small. We now consider the subsystem made up of subsystems 1 and 2 com-
bined. Since the two have negligible effect on each other, the probability
f1+2 (c1 , c2 ) of the combined system must be simply the probability that † To see that f
1+2 has the same value of
subsystem 1 has configuration c1 times the probability that subsystem 2 has the constant e0 as f1 , we first suppose
configuration c2 : f1+2 (c1 , c2 ) = f1 (c1 ) × f2 (c2 ). As we have seen, these otherwise that f1+2 = (constant)e−E1+2 /e0



with e0 = e0 . Now we calculate the rel-
probabilities depend only on the respective energies: The energy E(c1 ) for
the first subsystem, E(c2 ) for the second subsystem and the total energy of ative probability for some configuration c̃1
of subsystem 1 using the f1+2 (c1 , c2 ) of
the two for the combined system: E(c1 , c2 ) = E(c1 ) + E(c2 ). The relative the combined system. Evidently we must
probabilities f must satisfy f1+2 (E1 + E2 ) = f1 (E1 ) × f2 (E2 ). sum the probabilities of all combined con-
One way to satisfy this requirement is f (E) = (constant)e−E/e0 , where figurations for which the simple subsystem
e0 is some constant. This f (E) in fact gives the equilibrium probability is in configuration c̃1 . Thus our desired

of a configuration c of any small subsystem whose energy is E(c). This probability f (c̃1 ) = c2 f (c̃1 , c2 ). Using
our Boltzmann formula for f1+2 (c1 , c2 ) this
probability distribution is called the Boltzmann distribution; deriving it is means
a major topic in statistical mechanics courses. By assuming this form one
readily verifies that f1+2 (E1 + E2 ) = f1 (E1 ) × f2 (E2 ); moreover, the f1 (c̃1 ) = exp(−E(c̃1 )/e0 )
constant e0 must be the same for subsystems 1, 2, and (1 + 2)† . Thus any 
× exp(−E(c2 )/e0 ).
two (small) subsystems of the same overall isolated system have the same c2
value for the coefficient e0 . Problem 2.1 gives another way of understanding
how the Boltzmann distribution arises. The sum factor is a mere constant independ-
ent of c̃1 and has no significance in our
2.1. Justifying the Boltzmann distribution It was stated above that the probability relative probability. Comparing this f1 (c̃1 )
that a system in thermal equilibrium is in a configuration c of energy E(c) varies with our original formula, exp(−E(c̃1 )/e0 ),
as exp(−E(c)/e0 ). This is true for any conserved quantity in a small subsys- we see that the two expressions will be the
tem of a closed random system. To illustrate this point, consider the following same only if e0 = e0 .
16 Fundamentals

Museum-of-Science-and-Industry example. Our system is a 1000 × 1000 grid

of U = 1,000,000 cells. We designate a section of V = 10,000 cells as our
subsystem. (The figure shows a typical configuration in a scaled-down ver-
sion. The inner rectangle is the subsystem.) Into the system a fixed total of
N = 1,000 identical balls are distributed at random. (The fixed number N of
balls corresponds to a fixed total energy. Just as different pieces of energy in
a system do not have an individual identity, neither do our balls. These cells
are allowed to contain more than one ball.) A specific k-ball configuration ck
of the subsystem is then specified by stating the number of balls in each of its
V cells. We define p(ck ) as the probability of a configuration ck of the sub-
system with k balls. It is the fraction of all system configurations in which the
subsystem has the configuration ck . We define the configuration of the outer
 of the system (which has N − k balls) as dN−k . Then p(ck ) = (constant)
part
[ {dN−k } 1]: the probability of a given single subsystem configuration is given
by the number of system configurations dN−k compatible with it. We denote
this number as zN−k . (Clearly any configuration with k balls has the same
probability as any other.) For simplicity in what follows, you may assume that
k  N (subsystem is small) and N  U (system is dilute). (a) Given an m-
ball configuration of the outer part of the system (the “reservoir”), how many
m + 1 ball configurations can be made from it by adding one ball? Denote
this number as b(m). (b) Given an m + 1 ball configuration of the reservoir,
now many different m-ball configurations could have led to this one by adding
one ball? Is zm+1 = zm × b(m)? (c) Show that the ratio p(ck+1 )/p(ck )(=
zN −k−1 /zN −k ) is independent ofk. (This can be done without calculat-
ing the (horrendous) zN−k ≡ [ dN−k 1] explicitly.) What is this ratio?
(d) Show that p(ck ) has the form p(k = 0) exp(−k/constant), and find the
constant.
See the end of the chapter for the solution.

The coefficient e0 is clearly an important parameter in the distribution.

Clearly, a configuration c with energy E(c ) more than a few times e0
greater than E(c) has a relatively insignificant probability. By analyzing
how e0 affects thermodynamic properties of the system, one can infer that
e0 is essentially the absolute temperature T : e0 = kB T , where kB is the
Boltzmann constant mentioned Chapter 1. One way to make the connection
between e0 and T is along the lines of Problem 2.3. The Boltzmann constant
kB is merely a conversion factor between conventional temperature units
(defined before the connection between temperature and energy was known)
and conventional energy units. We may avoid this constant by simply
quoting temperatures directly in energy units. For our purposes in this
course, the temperature will nearly always be room temperature—300 K.
This temperature in conventional energy units is 1/40 eV or 0.6 kcal/mol.
These conclusions were first drawn by Ludwig Boltzmann in the nine-
teenth century. We may summarize them as follows. Any subsystem which
interacts weakly with a very large “reservoir” system necessarily has at
equilibrium a “temperature” T , characteristic of the reservoir. The rel-
ative probability of a configuration c of that subsystem has the form
f (c) = exp(−E(c)/T ). We shall refer often to this Boltzmann principle
henceforth.

2.2. Sinking colloids It is possible to make colloidal particles in water out of

metallic iron, whose density is 10 times that of water. If such particles were
large enough, they would settle under gravity to the bottom of their container.
(a) For what mass of particle would the mean height be 1 cm in water solution?
 Statistical physics 17

Express your answer in atomic mass units (6.023×(1023 )−1 g). (b) What would
the particle radius be in this case?
An important part of statistical mechanics is the relationship between
the temperature of a subsystem and its ability to do work. To recall this
relationship, we consider a specific example: an atom in a container in a
uniform gravitational field. This field exerts a constant force mg on the atom
and gives it a potential energy mgz where z is the vertical distance above
(say) the bottom of the container. According to Boltzmann’s principle, we
may take this atom as our subsystem. Its relative probability to be at height
z is then given by f (z) = exp(−mgz/T )† . From this one may immediately † Here our subsystem consists of only the
find the mean height height coordinate z of the atom. Since
we know the energy associated with a
 ∞  ∞
dzf (z)z given height without knowing the other
z = dzz p(z) = ∞ . configuration variables (such as the atom’s
0 0 0 dzf (z) momentum), we are allowed to use the
height coordinate as a subsystem. Of course
Simple integration yields z = T /(mg). it is also correct to find the probability f (z)
The mean height is evidently influenced by the temperature and the grav- starting from the full configuration prob-
itational field g. By increasing g slightly, we would do work −mg δz on ability f (x, y, z, px , py , pz ). But as in the
the system. This fluctuating work is difficult to give experimental meaning, (c1 , c2 ) note above, the effect of the other
variables is simply to multiply f (z) by an
but the average work δW ≡ −mgδz is well defined. unimportant constant.
 gThus the work W done by an external agent to change g from g1 to g2 is
g1
2
δW . Expressing δz as (dz /dg)δg, or −(T /mg)δg/g, we conclude
 g2      
T dg g2 z1
W = mg = T log = T log . (2.1)
g1 mg g g1 z2
The point of this example is to remind you that (a) a system in thermal equi-
librium exerts forces on its surroundings due to its thermal fluctuations,
(b) it can be made to do work, and (c) the amount of work can be calcu-
lated using the Boltzmann probabilities f (c). Since the work is evidently
recoverable (by returning g to its initial value), the work done on the system
represents a form of stored energy.

2.1.2 Probability and work

The connection between probability and work is an important aspect of
statistical mechanics. Below, we describe how the work done by a gen-
eral change in a system in thermal equilibrium can be expressed in terms
of the Boltzmann probabilities f (c). We first discuss the work associated
with changing a degree of freedom from one definite value to another.
This work is related to the probability that the degree of freedom has vari-
ous values when it is allowed to fluctuate. Next, we relate this work to
the work associated with altering the system’s energy, as in the gravity
example above. Finally, we show the use of these ideas via an example, the
lattice gas.
The probabilities that influence work done on the system are more
involved than the simple Boltzmann probabilities f (c). In itself f (c) gives
the relative probability of a particular configuration c. Often we need to
know a probability that involves many configurations. For example in a
18 Fundamentals

liquid containing two colloidal particles, we often wish to know the prob-
ability that these are separated by a displacement r. In general the liquid has
many configurations in which the two particles have this separation. And
the number of configurations typically depends on the distance r. Solvent
molecules are nestled and packed around each colloidal particle, altering the
molecular arrangements near each particle. When two particles approach,
the “spheres of influence” around them overlap. In the overlap region the
solvent molecules must accommodate to two colloidal particles instead
of one. Thus the number of arrangements or configurations of the liquid
particles is altered. Figure 2.1 shows a schematic example of this effect. To
find the probability that the particles have separation r, we must add the
probabilities for all those configurations in the liquid with this separation.
We denote this subset  of configurations by cr . Thus except for normaliza-
tion this probability is cr f (cr ). When the two particles are very far away,
the liquid near one is unaffected by the other, so that the probability is a
constant. It is convenient to normalize our probability so that this constant
is unity, the resulting relative
 probability
 is called the “pair distribution
function” g(r): g(r) ≡ cr f (cr )/ c∞ f (c∞ ).
Depending on the nature of the colloidal particles and the solvent, the pair
distribution function g(r) may increase or may decrease from unity as the
particles approach each other from infinity. To understand the meaning of
Fig. 2.1
these changes, we compare our fluid with the simplest possible one-particle
Schematic picture of two colloidal spheres system: an imaginary particle alone in an empty space and experiencing a
surrounded by small solvent molecules. potential U (r). The particle is weakly coupled to some thermal reservoir at
Only the solvent molecules closest to the temperature T . We shall take a potential that vanishes at infinity. To find
spheres are shown. In the upper picture, the g(r) for this simple system is an easy task. This is because there is only
the colloidal spheres are far apart, and one
sphere has little effect on the solvent
one configuration at position r. (We take the subsystem to be the coordinate
molecules surrounding the other. In the of the particle; so its momentum need not be considered.)  Thus the only
lower picture, the colloidal spheres are r-dependent quantity in f (cr ) is e−U (r)/T . And thus cr f (cr ) = e−U /T
close together and the arrangement of the
and g(r) = e−U (r)/T .
solvent molecules is disrupted.
One way to interpret the probabilities g(r) for our colloidal system is to
compare with the g(r) of the simple system. In particular, one can find a
potential U (r) such that the g(r) of the simple system matches that of the
real system. Comparing the g(r)’s of the two systems, it is clear that they
are equal if U (r) is given by
 
 
U (r) ≡ −T log(g(r)) = −T log f (cr ) + T log f (c∞ )
cr c∞

of the real system. With this choice of U the simple system is equivalent to
the real one as far as probabilities are concerned. Now we may give meaning
to large or small values of g(r). When g(r) increases from one, the U of
the system is negative: the pair acts as though there were an attractive
interaction. Conversely, if g(r) is smaller than 1, the two particles act as
though they repel each other. In Chapter 5 to follow, we shall see how such
effective interactions behave in practice.
Remarkably, this effective attraction or repulsion describes not only the
probabilities of the two systems, but also their ability to do work. In the
 Statistical physics 19

simple system the work required to bring the particle from infinity to r is
evidently U (r)† . The analogous work in the real system is not obvious to † It may not be so clear how this work is
deduce, since many parts of the system change when r changes. To evaluate measured. One way is to subject our particle
to an external potential Vx (r) with a deep
this work, we resort to an indirect method. We apply an external force F
minimum that holds the particle at a spe-
in the x direction to pull the two particles apart. (This indirect method is cific r = x. Then we may move our particle
easiest to imagine for a case where U (r) is attractive and much larger than from infinity to r0 by manipulating V so
T , so that there is a large probability that the particles are close together.) that its minimum moves from infinity to
We may do recoverable work on the system by increasing F slightly and r0 . We shall move slowly enough that the
thus increasing x . We shall apply the same force to the real system and particle stays in equilibrium with the res-
ervoir. The work required is the work done
the simple system and show that the work done on the two systems is the by the external force F = −dV /dr. At
same. Thus we can express the work done on the real system in terms of each point x en route, the particle is sit-
the U of the simple system. ting arbitrarily near the minimum of U +
When the force F changes, the resulting work δW is simply F δx , as V , so that dU /dr = −dV  r0/dr = F .
in the gravity example above. To know the work for a given shift of F it is  r0 work W done by F is ∞ F (r) dr =
The
∞ dU /dr dr = U (r0 ) − U (∞).
sufficient to know the change in x . The force affects the system through
the microscopic energy E(c). This E(c) has the form E0 (c) − F x, where
E0 (c) was the energy of configuration c before the force was applied.
To compute x we must average over all c. It is convenient to perform this
average in two stages, first summing over all c’s for a given displacement
r and then summing (integrating) over r:
 3 
d r c f (cr )x
x =  3  r . (2.2)
d r cr f (cr )

We may write f (cr ) as exp(−E0 (cr )/T ) exp(F x/T ). Further,

 
exp(−E0 (cr )/T ) = g(r) exp(−E0 (∞)/T ).
cr c∞

Using these facts we can readily express the x in terms of the pair
distribution function g(r):
 3
d rxg(r) exp(F x/T )
x =  3 .
d r g(r) exp(F x/T )

Remarkably, the needed x can be found purely in terms of g(r). Thus

any two systems with the same g(r) must have the same x and thence the
same external work stored by a given force F .
Since our simple system has the same g(r) as the real system, it must store
the same energy under a given perturbation. In other words, the real system
 as though it were a simple system with potential U (r) =
stores energy
−T log( cr f (cr )) and no other degrees of freedom. In particular the work
required to bring the real system from a separation r1 to a separation r2 is
the same as in the simple system, viz.
 
 
U (r2 ) − U (r1 ) = −T log f (cr2 ) + T log f (cr1 )
cr2 cr1

This U (r) is called the potential of mean force. It is a form of “free energy.”
20 Fundamentals

In this example we have determined the work to change a degree of

freedom from one definite value to another e.g. from r1 to r2 . We have
shown that this work
 can be determined using the relative probabilities of
the variable e.g. cr f (cr ) when r is free to change in thermal equilibrium.
The same reasoning can be carried out in general. We thus define the free
energy F of a thermodynamic system as



F ≡ −T log f (c) . (2.3)
c

The quantity in parentheses is called the partition function. The work

required to change a system variable that is fixed for all c’s in the sum
is simply the change of F .
Often one performs work on a system by perturbing its energy rather
than by changing a variable from one definite value to another. In the
above examples, the agent that changes the gravitational force g or the
colloidal force F does work on the system, but a bit of care is needed in
determining the work done. Let us take the colloidal system as an example.
We shall imagine that the external force is exerted by a spring. We shall
change the force F exerted on the colloidal particles by displacing the other
end of the spring. Clearly, we are changing a system variable (the posi-
tion of the spring) from one definite value to another. According to our
reasoning above, the work done to displace the spring is the change in
F . However, only part of this work is done on the colloidal particles and
the fluid surrounding them. The rest of it is done on the spring. In order
to determine the work performed on the colloidal system by changing F ,
we must subtract the change in spring energy. Happily, there is a way to
infer the work on the system in terms of F itself. We will discuss this in
general terms.
We now consider a system with two contributions to the configuration
energy E(c). The first is the system’s intrinsic energy, denoted E0 (c).
The second is the energy associated with the external influence that is
to be altered to produce work. We shall write this external energy in the
form bB(c), where b is the parameter (analogous to F above) to be var-
ied and B(c) is some variable of the system, analogous to x above. Thus
E(c) = E0 (c) + bB(c). When we increase b by a small amount δb, we
make configurations with large B more energetic and thus less probable.
This means that B decreases. The work done on the system as a whole is
δ F ; the work done to change the external energy bB is (on average) δbB .
The remainder δ F − δbB is the work done on the intrinsic system. This
is simply the change in the quantity (F − bB ).
We may verify that this expression for the work agrees with our directly
determined value F δx from above. For this, we note that δ F is related
to B :
d
δ F = δb F.
db
 Statistical physics 21

Using the definition of F ,

 
d d 
F= −T log e−(E0 +b B)/T
db db c

−1/T c B(c) e−(E0 +b B)/T
= −T  −(E +b B)/T .
ce
0

The temperature factors cancel, and the remaining fraction is precisely B .

Thus
δ F = δbB .
The change in our quantity δ(F − bB ) can now be readily found:

δ(F − bB ) = δbB − (δbB + bδB ) = −bδB .

In our colloidal example, the external variable was the coordinate x and the
parameter b was −F . Thus −bδB → F δx = δW , as anticipated.
The free energy F is much used to discuss systems in thermal equilib-
rium. It is worthwhile to note how our discussion is related to standard
treatments.
1. One might worry that F is ill-defined, since it is based on the relative
probabilities f (c), whose normalization is arbitrary. If this normalization
were changed by multiplying all f (c) by a constant factor ea , one readily
sees from Eq. (2.3) that F is increased by an amount a. This ambiguity
is harmless; as with any energy, only changes in F are significant. Still,
it is conventional to fix the normalization by using f (c) = e−E(c)/T
with no other prefactor. Then, for a system with a single configuration
c, F is simply equal to E(c).
2. In macroscopic systems one is often interested in that part of the work
done on a 
 system which does not increase its internal energy E ≡
E f
c c c / c fc . In an isolated system the work done must be equal to
the change in E , in order to conserve energy. However a subsystem
which can exchange energy with a reservoir may transfer energy to the
reservoir when work is done on it. The energy transferred to an energy
reservoir in this way is heat. The amount of heat transferred, denoted
δQ, is evidently given by δQ = δW − δE . It can be shown that this
heat, like the free energy, can be expressed in terms of the normalized
probabilities pc . Specifically, δQ = −T δS, where the entropy S has
the form 
S=− pc log pc ,
c

and pc is the normalized probability fc /( c fc ). The entropy is a meas-
ure of the randomness of the system. Many of the concepts in this section
can be expressed in terms of maximizing this entropy. However, in this
book, issues of heat are not essential and thus the notion of entropy will
play a minor role.
22 Fundamentals

3. In thermodynamics textbooks, one usually distinguishes different types

of free energy depending on what parts of the energy are considered as
intrinsic (e.g. E0 (c)) and what parts (e.g. F x) are considered external.
The different free energies are denoted by different symbols. In this
convention, our F is called the “Helmholtz free energy.”
2.3. Ideal gas free energy A one-atom ideal gas of mass m is trapped in a cyl-
inder of height h. The atom can exchange energy with the outside world,
which is at temperature T . Gravity is negligible. The height can be reduced
by moving a piston. The relative probability f (c) of this system is evidently
f (x, y, z, px , py , pz ). (a) What is the function f (x, y, z, px , py , pz )? Use the
Boltzmann normalization: f (c) = e−E(c)/T , since this is the proper one for
computing a free energy, as done below. (b) How does f change if the height
is expanded from h/2 to h? (Does the dependence  on px , py , pz change?)
(c) The free energy F was defined as −T log( c f (c)). Find the change in
free energy upon changing  the height from h/2 to h. This can be expressed
as −T log[ ch f (ch )/ ch/2 f (ch/2 )]. (d) The change in F is supposed to
be the work done in changing the height. Thus there must be a force on the
piston given by −d F /dh. Express this force in terms of T and h. (e) On the
other hand, we can obtain this force using the conventional ideal gas law for
pressure p: pV = N T (here N = 1). Compare the force from this pressure p
to that obtained in (d). You may assume the piston has cross-sectional area A.
This problem shows the connection between the conventional temperature as
defined by the ideal gas equation of state and the T appearing in Boltzmann’s
Principle. Now a gravitational energy mgz is added to the energy E(c). The
work to change h is still given by the pressure at the top times the area. But the
density and thus the pressure at the top are now reduced by gravity. (f ) Verify
that the work δW to change h is still given by δ F .

2.1.3 Lattice gas

To illustrate the use of free energy, we discuss a model that will prove useful
later: the “lattice gas.” A lattice gas is a set of N particles that occupy the
V discrete sites of a lattice. To specify a configuration, we list which lattice
sites are occupied. No site may be occupied more than once. (Thus N
must be no greater than V .) We characterize the density of particles by the
“volume fraction” φ ≡ N /V . The simplest lattice gas has no energy. Thus
all the arrangements of the Nparticles on the V sites are equally likely. The
free energy is thus −T log( c 1). To count the possible arrangements c is
a tedious problem in combinatorics. We may avoid it by using the indirect
method of the last subsection. We alter our system by applying an external
energy that allows us to change φ. By keeping track of the work to change
φ we can infer the desired free energy.
We focus on a single site of the lattice, which is filled with probability φ,
and consider the work that would be required to empty the site. In order to
empty it, we introduce an external energy: E = μφ. (The field parameter
μ is often called a chemical potential.) We may calculate the free energy
of a single site in the presence of the μ energy. If the site is empty, this
energy is zero. If the site is occupied, the energy is μ. Summing over both
configurations, we have a free energy Fμ = −T log(1 + e−μ/T ). As we
have seen previously, the derivative of Fμ must give the average value
of φ: φ = d Fμ /d(μ). The intrinsic work required to change φ (not
 Statistical physics 23

counting the unwanted contribution from the μφ energy) must be given
by the change of (Fμ − (μ)d Fμ /dμ) = (Fμ − μφ ).
We now calculate these quantities explicitly:

d Fμ e−μ/T 1
φ = = −μ/T
= . (2.4)
dμ 1+e 1 + eμ/T
Thus the free energy is given by

Fμ − μφ = −T log(1 + e−μ/T ) + μ/(1 + eμ/T ).

We may express this free energy in terms of φ , by solving Eq. (2.4) for μ:
eμ/T = 1/φ −1 and μ = T log(1/φ −1) = T log(1−φ )−T log(φ ).
Similarly Fμ can be expressed in terms of φ :

Fμ = −T log(1 − 1/[φ −1 − 1]) = T log(1 − φ ).

Thus,

Fμ − μφ = T [log(1 − φ ) − [log(1 − φ ) − log(φ )] φ ] .

This simplifies to

Fμ − μφ = T [φ log(φ ) + (1 − φ ) log(1 − φ )].

This free energy gives the work to change φ in the one-site system. In a
system with V sites, the reasoning is equivalent. The sum on c for each site
is independent, since the work is simply the sum of site free energies.
Combining these results, the work per site done in changing the average
volume fraction from 0 to φ is given by T [φ log(φ )+(1−φ ) log(1−
φ )]. If the number of sites V is large, the actual volume fraction, φ must
be arbitrarily close to this average. Thus, in a large lattice gas the work
W per site required to change to a definite φ from φ = 0 is evidently
T [φ log(φ) + (1 − φ) log(1 − φ)]. But, as we have seen, the work to change
a coordinate of a system is simply the change in its free energy F .

F /V = T [φ log(φ) + (1 − φ) log(1 − φ)]. (2.5)

Clearly the lattice gas differs significantly from an ideal gas. In an ideal gas
each particle may occupy any position independent of the other particles.
In a lattice gas the particles are obliged to avoid sites occupied by others. In
the dilute limit of a single particle in a large volume V the volume fraction
φ becomes 1/V and the lattice gas free energy of Eq. (2.5) has the limiting
form
F = −T log(V ),
like the ideal gas treated in the preceding Problem. In this limit the second
term in Eq. (2.5) is negligible. But it becomes increasingly important as the
volume fraction increases, making the lattice gas free energy larger than
the ideal gas free energy. In the opposite limit, as the φ → 1, the lattice
gas free energy becomes small again. A nearly full lattice is equivalent to a
dilute gas of vacancies, since F (1 − φ) = F (φ).
24 Fundamentals

2.1.4 Approach to equilibrium

These relationships between work, probability and temperature presup-
pose that our many-particle system has attained a final equilibrium state.
Whenever the system is altered by a change in the environment, it leaves
equilibrium, and requires time to return. A fluid is never in equilibrium in
all respects. For example, the organic molecules in the fluid are typically not
in their most stable state, and given sufficient time they would decompose.
On a given experimental time scale—say, a minute—many features of the
system, like the profile of local composition, have ample time to reach an
equilibrium state, i.e., a state that is independent of the most recent change
in the environment. Other features, like the total number of molecules of
a given type, retain their initial values, and make essentially no progress
towards equilibrium. Degrees of freedom that reach their equilibrium stat-
istical state are often called annealed variables; those that keep their initial
values are called quenched variables. In studying a given fluid, one must
have a sense of which features of the fluid are quenched and which are
annealed. That is, one must have some way of estimating the timescale for
a given feature to relax from an initial state to one of thermal equilibrium.
Relaxation can take place in a variety of ways. The cooperative modes
of relaxation distinctive to structured fluids are an important subject to be
discussed in later chapters. Here we deal with a more generic and widespread
† The term “activated” comes from the con- cause of slow relaxation called activated hopping† . Activated relaxation
text of chemical reactions. Here the system accounts for the gradual dissociation of a population of molecules or the
at x = x ∗ is thought of as a chemically act-
condensation of a supercooled vapor to form droplets of liquid. In general,
ive state, which is free to pass to more than
one distinct final state. activated hopping occurs when two objects interact via a potential of mean
force like that of Fig. 2.2.
As the graph suggests, it is quite improbable for the separation x to take
U*
on values near x ∗ , since such separations have high (free) energy, even
in equilibrium. For example, if U ∗ − U0 = 10T 14 eV, the probability
U(x) that the separation is x ∗ is e10 20,000 times smaller than the probability
that it is 0. Since visits to x ∗ are rare, it is natural that passage from x
U0
near zero to x > x ∗ is slow. A quantitative understanding of this slow
T process was first developed by Eyring, Kramers [1], and many others [2].
x* If the two objects are initially at a separation x near 0, our problem is to
x find the average time required for it to cross the barrier at x ∗ as a result
of random thermal fluctuations. The answer clearly depends on the energy
Fig. 2.2
profile U (x). In addition we need information about the time scale for
Schematic plot of potential of mean force
U (x) versus separation x for two objects.
the fluctuations. The two molecules of our example move in response to
If the two objects are initially together, collisions from neighboring molecules. Thus x changes as a result of many
with x = 0, with energy U0 , they cannot independent, small, random increments. To be sure, the potential U affects
move apart without passing through a these random moves. Since lower values of U are more likely than higher
region of large potential U ∗ near the point ones, random steps that decrease U are more likely than those that increase
x ∗ . The thermal energy T is marked on the
vertical scale. Evidently a free energy
it. Thus the separation x tends to be pushed away from the energy barrier
much larger than T is needed if the objects and back towards the region around x = 0.
are to move apart or together. We may imagine the behavior of this pair of objects if an additional
artificial force were applied which neutralized the effect of the potential
U . For example, if our two molecules were part of a detailed molecular-
dynamics simulation, we could determine the potential of mean force U (x)
explicitly and then impose an additional force F (x) that would nullify U (x),
 Statistical physics 25

say, for all x < x ∗ . (Thus in the fictitious system, U (x) = U (0).) Without
a biasing potential the random moves of x can easily take it from its initial
position to x ∗ . We denote the time to reach x ∗ with no potential as τ0 .
Specifically, τ0 is that time for which x has reached x ∗ with probability 12 .
The inverse of this τ0 is called the attempt rate. The Eyring–Kramers theory
describes how the barrier increases the escape time from τ0 .
If the barrier is high, the objects must make many “attempts”—returns to
x = 0—before reaching x ∗ . As time passes, the (un-normalized) probability
f (x) that the separation is x, approaches its equilibrium value for a greater
and greater range of x around its starting point x = 0. After sufficient time,
this range reaches x ∗ . Then f (x ∗ )/f (0) approaches its equilibrium value
of exp[−(U ∗ −U0 )/T ]. Whenever x does reach x ∗ , it may move to the right
or to the left without bias. Thus with probability of roughly 12 , it crosses the
barrier.
To see the effect of U , we first imagine the motion from the initial x = 0
state for a time τ0 in the fictitious system with U nullified by the artificial
force F . In a population of such molecules one encounters many possible
motions. As noted above, this time τ0 is sufficient for typical motions
to reach x ∗ . One way to account for the effect of U is to form a new
ensemble of motions starting from the motions of the fictitious system. One
removes some of these motions from the fictitious ensemble to account for
the time they spend in regions of high energy. These removals must assure
that the overall probability for the separation to be x is the equilibrium
probability f (x), in the potential U . In the fictitious system all x’s have
the same equilibrium probability f0 . But in the real system, x occurs with
equilibrium probability f (x) ∝ exp(−U (x)/T ). Evidently f (x)/f0 =
exp[−(U (x) − U (0))/T ]. In equilibrium one may thus recover the proper
probabilities by simply throwing away all but a fraction f (x)/f0 of all
fictitious systems that end up at x. This same prescription is approximately
correct even for our ensemble of motions in a finite time t. The probability
that x has reached x ∗ during the motion is that of the fictitious system (viz. 12 )
times the probability that the motion has survived the removal process.
This probability is f (x ∗ )/f (0) = exp[−(U (x ∗ ) − U (0))/T ]. Since the
separation has reached x ∗ in time τ0 with this probability, the average rate
at which the separation crosses x ∗ is thus 12 τ0−1 exp[(−U (x ∗ ) + U (0))/T ].
The pair of objects stay together for an average time that is the inverse of
this rate. Over times much shorter than this escape time, the two objects
may be considered as permanently bound together.
Clearly the most important influence on the escape time is the activation
energy U ∗ − U0 . Escape over a barrier is exponentially slow if its height
is much more than T . Increasing this height by an amount T increases the
escape time by a factor e. The escape time also depends on the attempt
time, but much more weakly. The Eyring–Kramers formula is applicable
whenever the barrier height is much greater than T . Remarkably, the shape
† To check for activated kinetics one con-
of the U (x) function is immaterial in this limit. The formula has been
ventionally plots the log of the measured
derived under rather general conditions, as discussed in [2]. It has also been
rate of e.g., dissociation against inverse tem-
experimentally verified for a wide range of processes† . Though we have perature 1/T . Activated kinetics gives a
illustrated this behavior in the context of dissociation, it is applicable to any straight line on this “Arrhenius plot;” the
quantity x with a region of high free energy. Activated hopping is not the slope is the activation energy U ∗ − U (0).
26 Fundamentals

only way a quantity like x can pass through a high-energy region. Another
process is quantum tunneling. This requires no thermal fluctuations, but
only the positional uncertainty all particles have by virtue of the wave
nature of matter. Such tunneling phenomena are not generally important
for the liquids we will consider, and we will ignore this and other quantum
effects from here on.
Using the Eyring–Kramers theory, one may get a rough idea of what parts
of the system can or cannot relax over the timescale of a given experiment.
For example, the energy required to break a carbon–carbon bond is 2 eV, i.e.,
80 T . The attempt time τ0 is in the range of 10−12 s. or more, as discussed
in the next section. From this one obtains an estimated escape time of some
1015 years. We shall need the notion of activated lifetimes when we discuss
colloidal stability in Chapter 5.
2.4. Metropolis dynamics: a concrete kinetic model It is useful to have a specific
model to examine phenomena like dissociation and to gauge the validity of
the Eyring–Kramers kinetic picture in specific cases. The Metropolis algorithm
[3] used for Monte-Carlo simulations provides a simple example that is easy
to implement on a programmable calculator or computer. Like the motion
of real molecules, the Metropolis dynamics consists of a sequence of small
steps. And after a long time, the Metropolis algorithm leads to the proper
equilibrium Boltzmann distribution f (x). Applied to our dissociation problem,
the Metropolis algorithm works as follows. Time is divided into discrete steps
of length t. The reaction coordinate changes by discrete, random increments
of a fixed magnitude x  x ∗ . In the absence of a potential U , the motion is a
sequence of random steps. At timestep t, the current x(t) is changed by ±x,
with the + or − sign being chosen at random. In the presence of a potential U (x),
a given step results in a change of potential U . The dynamics must be modified
to account for the potential if it is to lead to the proper equilibrium distribution
f (x) ∝ exp(−U (x)/T ). The modification is that after each random step the
change of energy U is computed. If the step is found to decrease U , the
stepping process proceeds: the step is “accepted.” But if the step is found to
increase U , the step is “rejected” with probability 1 − exp(−U /T ). That is,
the position x is reset to its value before the last step. To implement this on a
computer, one chooses a random number between zero and one. If this number
is larger than exp(−U /T ), one resets the x coordinate; otherwise one retains
the last increment of x and proceeds to the next random step.
If the range of x is limited, the entire range is eventually visited, and after
a long time the probability f (x) that the coordinate is at x reaches a constant
distribution, independent of time t. One can show what this f (x) must be,
using the dynamical rule above. In general, one can express f (x, t) in terms of
f ’s at the previous timestep. In the absence of the potential U , if the coordinate
is x at timestep t, it must have been x − 1 or x + 1 at the previous timestep. If
it was at x − 1, then it moves to x with probability 12 . This 12 is the conditional
probability that the coordinate goes to x given that it was at x − 1. Thus f (x, t)
is the sum of two contributions for the two distinct states from the previous
timestep that can reach x:

f (x, t) = 12 (f (x − 1, t − t)) + 12 (f (x + 1, t − t)).

If U is present, the conditional probabilities are modified. We suppose that at

timestep t − t the coordinate is x. Then the probabilities q− , q+ , q0 of going
to x − 1, going to x + 1 or staying at x are:

q− = 12 , q+ = 1
2 exp(−U /T ), q0 = 1−q− −q+ = 12 [1−exp(−U /T )].
 Magnitude of a liquid’s response 27

Here we define U (x) ≡ U (x +1)−U (x). We have supposed for definiteness

that U is increasing with x near the x in question. (If U is decreasing, the rules
for q+ and q− are reversed, since we always accept a candidate move that goes
downhill.) Combining the conditional probabilities from x − 1, x, and x + 1,
we obtain the full probability f (x, t).

f (x, t) = q+ (x − 1)f (x − 1, t − t) + q0 (x)f (x, t − t)

+ q− (x + 1)f (x + 1, t − t). (2.6)

This equation describes the “flow” of probability during one timestep. In an

ensemble of many copies of the system, the (relative) net number of systems
passing from x to x + 1 in a timestep is called the probability current j (x + 12 ).
Evidently j (x + 12 ) = f (x, t − t)q+ (x) − f (x + 1, t − t)q− (x + 1).
(a) Express the change of f during a timestep, f (x, t) − f (x, t − t) in terms
of the currents j (x − 12 ) and j (x + 12 ) for any q(x)’s that have q− + q0 +
q+ = 1. (b) Find a relationship between f (x), f (x + 1) and the q’s such
that j (x + 12 ) = 0. (c) When the system reaches equilibrium, f (x) becomes
independent of time. This can only happen if j (x + 12 ) = constant for all x.
In the Metropolis process, this constant is zero. We may define U (x = 0) = 0
and f (x = 0) = 1. Find the distribution f (x) for which j = 0 everywhere.
(d) (Harder) Write a program to implement the Metropolis algorithm for the
potential U (x) ≡ T ((x − 20)/3)2 . Let x range from 1 to 40 in steps of 1.
Start x someplace between 18 and 22. Define an array N (1), . . . , N (40), one
element for each x. At each timestep (whether the move is retained or not)
add one to the N (x) for the current value of x. Thus N (x) gives the number
of timesteps when the system was at x. Plot this experimental distribution
N (x) after 10,000 timesteps and after 107 timesteps, and compare it with the
equilibrium distribution exp[−(x − 20)2 /9]. Find the average time for the
system to reach x = 26 and compare with the Eyring–Kramers prediction.
Begin by turning off U and determining τ0 for some chosen starting value of
x. Then restore U and measure the time using this same starting value.
2.5. Equipartition theorem A thermodynamic system consists of a variable x and
many other variables, labeled collectively c . The energy E(c) of a configur-
ation has the form E0 (c ) + E1 (x). (Thus the x variable may be considered
as a subsystem.)
 Further,the x energy is quadratic: E1 = bx 2 . (a) Show that
E1 ≡ ( c E1 f (c))/( c f (c)) is simply 12 T . (Note that the sum on c can
be immediately reduced to a mere sum on x.)

2.2 Magnitude of a liquid’s response

In this section we discuss the quantitative behavior of liquids, including
structured fluids, at the most primitive level. We account for the order of
magnitude of the basic response properties in liquids. For example, we
know that most simple liquids in everyday experience—water, gasoline,
alcohol—have roughly similar viscosities, though they differ markedly in
other respects. This impression is confirmed by a look at a property-table
handbook. There one finds that most small-molecule liquids have viscosities
of around 0.001 Pa s, i.e. 0.001 SI unit† . This size viscosity is a rough lower †The cgs unit is called a poise; a poise is
limit: those viscosities that differ greatly from this value are on the high 10 Pa s.
side. Structured fluids like liquid polymers often have high viscosities. To
understand the significance of this, we must understand how viscosity arises
and what controls its magnitude.
28 Fundamentals

The viscosity of a liquid may be defined by an experiment with the liquid

between two parallel plates. If the top plate is slid over the bottom one
with some (small) velocity v, the fluid moves parallel to the plates, with
a velocity that increases linearly from zero at the bottom plate to v at the
top. The velocity gradient dv/dz, or shear rate is constant throughout the
liquid. Viscosities are typically measured at shear rates of 10–100/s. This
is the order of shear rate encountered in pouring a cup of coffee. Shear
rates larger than about 105 /s are difficult to achieve in controlled, steady,
laboratory flows.
It is useful to think of this velocity gradient in terms of the strain on a
fluid element, a tiny cubical volume in the fluid. As the flow proceeds,
the cube distorts into a parallelepiped. The amount of distortion of the
parallelepiped—the lateral displacement divided by the height—is the strain
γ which the fluid has undergone. This strain evidently increases at a constant
strain rate γ̇ , which is identical to the velocity gradient.
A steady force on the top plate is needed to maintain the motion; this
implies a stress in the fluid. Since the acceleration of every part of the fluid
is zero, the forces on each fluid element of the fluid must sum to zero.
If there is a force acting to the right at the top of a fluid element, there
must be an equal force acting to the left at the bottom. The force on a small
element must be proportional to its cross-sectional area. The force-per-unit-
area or stress σ must be uniform from top to bottom since the fluid is not
accelerating. Typically, the shear rate in each fluid element produces a stress
σ proportional to the shear rate γ̇ . The constant of proportionality is the
viscosity η: σ = ηγ̇ .
The coefficient η also measures the dissipation rate. The work done per
unit time to maintain the flow in a fluid element is the force times the
velocity:
dW /dt = F · v = (σ x y)(γ̇ z),
where x, y, z, are the dimensions of the element. Thus the dissipation
rate per unit volume ẇ is σ γ̇ or

ẇ = ηγ̇ 2 . (2.7)
Evidently, the dimensions of η are energy per unit volume times time.
Why should a fluid’s dissipation be governed by a law like Eq. (2.7)?
What determines the size of the parameter η? To discuss these questions,
it is useful to imagine applying the flow between our plates in a different
way. We imagine not a smooth motion at a fixed velocity, but a motion in
sudden, tiny steps. We apply a small step strain, wait for a moment, and
then apply another. These steps may be made so small that the dissipation
they produce is the same as in the steady flow. But this stepwise picture
gives us a way to examine the molecular basis of viscosity. For simplicity,
we shall also imagine the simplest of fluids, a monatomic liquid like liquid
Argon. (We should not need to treat too-specific features of the fluid, since
Fig. 2.3
The isotropic liquid at the left is suddenly
viscosities depend but little on the molecular nature of a simple fluid.)
subjected to a 20% shear on the right, thus Figure 2.3 shows a fluid element before and immediately after a step
pushing the molecules together from strain. The sketch attempts to show an affine deformation of the centers—
upper-left to lower-right. that which would occur if these were embedded in rubber and then distorted.
 Magnitude of a liquid’s response 29

Evidently the isotropic equilibrium arrangement of the atoms is distorted

and made anisotropic by the strain. Producing this non-equilibrium state
requires work; it costs a free energy proportional to the number of atoms in
the fluid element. It is this stored energy that gives the energy scale in the
viscosity, as we shall discuss below.
After the anisotropic state has been produced, it relaxes back to equi-
librium. After some characteristic relaxation time the state of the liquid in
the strained element is the same as it was before the strain was applied.
The time required is the time for the distances between nearest neighbors to
become the same in all directions. This is a few collision times, i.e., a few † One shows in statistical mechanics that it
times the time required for an atom to travel between nearest neighbors. is essentially the speed of sound in a gas
made of these atoms.
With this scenario in mind we can estimate the magnitude of the viscosity.
To find the order of magnitude of the transit time, τ we need to know the †† We shall use the symbols and ∼

typical speed v of the atoms. We may focus on one component of the throughout this book. Our use of these sym-
velocity, say vx . The energy E(c) of our system depends on vx in a simple bols is exactly that of DeGennes’ book on
way: E(c) = E(c ) + E(vx ), where the variables c are independent of polymers, op. cit. These weak forms of
equality are very useful for describing the
vx and the E(vx ) is independent of the c . In such cases, as we have seen scaling behavior we shall explore. The state-
above, the vx variable may be considered as an independent subsystem. In ment A B means that (in the regime or
our case the (kinetic) energy E(vx ) is quadratic in the variable. Then one limit under discussion) A/B is a numerical
can readily calculate that E(vx ) = 12 T , as shown in Problem 2.5. This constant. This constant need not be close
equipartition theorem is true for any variable whose energy is quadratic. to 1. For example, a rod of length L and
mass m has a moment of inertia I mL2 .
Using the equipartition theorem, the typical speeds of atoms in our simple The statement A ∼ B is weaker; it means
liquid are evidently given by mv 2 T or v (T /m)1/2 , where m is the that A/B remains finite in the regime or
atomic mass. For the light atoms found in air or common liquids, it is† limit of interest. For example, rods of differ-
a few hundred m/s. The distance between collisions is roughly an atomic ent length L and width a have moments of
diameter a—i.e., roughly 0.2 nm. We thus estimate that the relaxation time inertia I ML2 ∼ L3 when L  a. Gen-
erally we use ∼ rather than the stronger ∝;
τ is of order a/v 10−12 s †† . A ∝ B indicates strict proportionality, with
For the tiny step strains in our experiment the stress σ must be pro- no implication of an asymptotic regime or
portional to the strain, as in any weakly deformed material. Accordingly, limit.
we define the “step-strain modulus” G0 as the ratio of initial stress to the
magnitude of the step strain: σ ≡ G0 γ .
The stored energy per unit volume w is the work done per unit volume
by the stress force. We may compute it by integrating the force σ xy
onγ the top face with the displacement dx = γ z: wxyz =
[σ (γ )xy]d[γ  z]. Expressing σ in terms of the modulus G we
0 0
have w = 2 G0 γ . We see that G0 has dimensions of energy per unit
1 2
volume, like the dimensions of w, so that G0 is roughly the free energy cost
extrapolated to a strain of unity. This free energy is at least partly interaction
potential energy between the atoms; the free energy per atom is roughly the
cost of removing an atom. Such energies can be found from thermodynamic
measurements on the liquid. But they can be crudely estimated on funda-
mental grounds. Our atoms condense into a liquid phase (in equilibrium
with gas) because they attract each other. To form a phase, surface atoms
moving away from the surface under thermal motion must be opposed by
this attraction. The potential energy must thus be at least of order of the
kinetic energy, which in turn is of the order of T . Were this is not true,
the atoms could freely leave the liquid and the liquid would evaporate. But
the interaction energy cannot be too much greater than T ; otherwise the
30 Fundamentals

kinetic energy of the atoms would be insignificant. They would take on

an arrangement of minimum potential energy, and form a crystal. Thus the
mere knowledge that our atoms are in a liquid phase tells us that their inter-
action energy is not too different from T . The modulus G0 is thus of order
T /a 3 . In this estimate we neglected the non-potential part of the free-energy
of distortion. This “entropic” part arises because fewer configurations
 are
available to the atoms after the step strain, thus reducing c f (c). A unit
of strain reduces the number of configurations
 by a finite factor for each
atom, so that the free energy −T log( c f (c)) changes by an amount of
order T per atom. Its contribution to the modulus G0 is comparable to that
of the potential energy.
With these numbers in hand we may now estimate the rate of dissip-
ation ẇ. Each time a step strain is made, the energy density w stored is
2 G0 γ T /a γ . This energy is dissipated irrecoverably as the liquid
1 2 3 2
relaxes to its equilibrium state in time τ . The dissipation rate ẇ over this
time is of order T /a 3 γ 2 /τ . If we repeat the step strain at intervals τ , we
obtain a steady dissipation of this order. The shear rate γ̇ is then γ /τ , and
ẇ T /a 3 τ γ̇ 2 . This has the proper form for viscous dissipation; by com-
paring with Eq. (2.6), the viscosity η may be read off. Using (1/40) eV
(= 1.6 × 10−19 /40 J) for T , 10−12 s for τ and 0.2 nm for a, this yields
an estimate of 10−3 Js/m3 for η. That is, η = 0.001 in SI units, in agree-
ment with the measured values quoted at the beginning of this section.
(In view of the crudeness of our estimates, this agreement is somewhat
fortuitous!)
The mechanism for viscosity sketched above is somewhat altered for
molecular liquids. Let us imagine what happens to the viscosity as we pass
from the simplest hydrocarbon, methane, to the two-carbon chain ethane
to the three-carbon chain propane, and so forth. As we do this, we may
have to change the temperature so as to maintain our sample in the liquid
state. For definiteness we shall maintain a temperature, say, 1% above the
melting temperature. This temperature rises as the molecules get larger,
but only to a limited extent: it rises from roughly 200–400◦ K. The liquid
density is also roughly constant; the atoms stay roughly at the separation that
minimizes their interaction potential. This potential is virtually independent
of temperature. Thus, it remains roughly true that the interaction energy per
atom is roughly T . Accordingly, the energy scale of the viscosity should
not be greatly altered as the molecules become larger. The timescale for a
distortion to relax is also not greatly altered. Much of the distortion of the
hydrocarbon fluid can be relaxed by few-atom local motions of the flexible
hydrocarbon chains. (This flexibility is necessary to maintain a liquid phase;
if the molecules are too rigid, they pass directly from a solid to a gaseous
state.) Since each atom moves at thermal velocities as calculated above,
the time τ to move an interatomic distance is not qualitatively changed
as the molecules grow. We conclude that the order of magnitude of the
viscosity of a molecular liquid should not be greatly different from that of
an atomic liquid, as is observed. For very long hydrocarbon chains, the
assumptions made in this paragraph begin to break down and large changes
in the viscosity occur. The origin of these changes is an important subject
in the chapter on polymers.
 Experimental probes of structured fluids 31

2.6. Rubber-band modulus The postal service rubber bands you find on the side-
walk have a cross-sectional area of 5 mm2 . They consist of hydrocarbon poly-
mers with a molecular weight per atom of about 5. The density is about
1000 kg/m3 (1 g/cm3 ). Under a 0.200 kg load, such a rubber band elongates by
a factor of 2. (a) What is the modulus of this rubber band? (b) What is the modu-
lus expressed in atmospheres of pressure? (c) Under a factor-of-two elongation
(unit strain) using the density, Avogadro’s number and the average mass per
atom, how many atoms store energy T ? Order-of magnitude estimates are fine.
There is a major effect that influences the viscosity of small-molecule
liquids, which we have not discussed. This is the so-called glass transition,
and is most important in molecules of irregular shape. As these liquids are
cooled, the viscosity grows too large to be measurable as the liquid cools
towards a characteristic temperature Tg . One can certainly not account for
this divergent viscosity by the arguments discussed above. However, one
may still ask whether the large viscosity is due to a large modulus or a long
relaxation time. The answer is the latter. A liquid near its glass transition
cannot equilibrate from an imposed distortion in a few atomic collision
times. The molecules have so little room to move and are so interlocked
that stress takes exceptionally long to relax. This phenomenon is important
for us in two ways. First, many of the background solvents of our structured
fluids have experimentally important glass transitions. This often provides
a useful means of controlling their viscosity. Second, large fluid structures
often themselves lead to high viscosities. There are some parallels between
this type of high viscosity and the glass transition, as we will discuss in due
course.
In accounting for the viscosity we have defined the fundamental response
of liquids. We’ve also identified the lengths, times, and energies that char-
acterize liquid structure and dynamics on the atomic scale. These conditions
set the backdrop against which the distinctive behavior of structured fluids
occurs.

2.3 Experimental probes of structured fluids

Physicists are increasingly interested in structured fluids in large part
because the collective structures in them have become increasingly access-
ible to experimental probes. Before discussing specific phenomena in
particular fluids, it seems good to have in mind how the predicted behavior
might be observed. We shall see that though some of these experimental
methods are common to condensed matter studies, many are distinctive to
structured fluids.

2.3.1 Macroscopic responses

As in all condensed matter, the most apparent properties of structured-fluid
systems are their macroscopic responses: we perturb a macroscopic sample
in some global way and measure something that happens throughout the
sample. The primary case-in-point for structured fluids is the so-called
dynamic modulus G(ω). As discussed in the previous section, the modulus
is the ratio of stress to strain. The dynamic modulus is the ratio of stress to
32 Fundamentals

107
G⬘

G ⬘() and G ⬙() (dyn/cm2)

G⬙
106
Fig. 2.4
Complex dynamic moduli of a
high-molecular-weight polymer melt, 105
from [4]. Reprinted with permission
purchased from Rubber Chemistry and 104
Technology © 1987 Rubber Division,
American Chemical Society, Inc. G labels
the in-phase (storage) modulus; G labels 103
out-of-phase (loss) modulus. The angular
frequency ω is multiplied by a 10–2 100 102 104
temperature-dependent factor aT to show
aT (rad/s)
the behavior that would occur at 27 ◦ C.

strain when an oscillating strain at frequency ω/(2π ) is applied. In gen-

eral the stress is not in phase with the strain, so that the dynamic modulus
is a complex number. In a simple liquid with a characteristic step-strain
modulus G0 and relaxation time τ , both G0 and τ may be inferred directly
from the dynamic modulus function. Figure 2.4 shows a classic example:
the dynamic modulus of a polymer melt. Many structured fluids we shall
meet have several important frequency regimes. The dynamic modulus
gives complete information about the response to a (small) arbitrary time-
dependent strain. Another equivalent response function is often discussed:
the step-strain modulus. This is the stress remaining at time t after a small
step strain has been applied. Dynamic moduli are routinely measured in the
range from 0.01 to tens of hertz. In many structured fluids the characteristic
relaxation rates may be brought into this range e.g. by varying the tem-
perature. Alongside this distinctive dynamic modulus is the more generic
dielectric relaxation—i.e., the complex dielectric function (ω). It is used
to probe the characteristic relaxation timescales in a structured fluid, with
great dynamic range but without great specificity.
2.7. Jello modulus Straight, unflavored gelatin is mostly a flexible biopolymer
called collagen. It has an average molecular weight per atom of 10 or so. The
point of this problem is to see how effective these molecules are at giving elasti-
city to water. We want to know how much gelatin molecule (what molecular
weight) stores about T of energy under a unit of strain. The idea is to see if
there is about T per atom, T per 10,000 atoms or what. To estimate this, one
may use the following information. Five grams of dry gelatin in 0.5 kg of water
makes a gel of the consistency of normal Jello. A 5 cm cube of such a gel on a
plate oscillates at about 1 cycle/s. For simplicity, we can assume that the amp-
litude of the strain is about unity—i.e., the top of the cube goes back and forth
5 cm from its resting position. From this frequency we may estimate the stored
energy in the gel. Since the piece of Jello is a harmonic oscillator, the stored
energy is the same as the kinetic energy 12 mv 2 . The typical velocity is 5 cm/s,
grossly speaking. (a) From this data estimate the number of atoms of gelatin
responsible for storing each T of elastic energy. (b) (Harder) The oscillations
of this Jello stop in a few cycles. That means a substantial fraction of the initial
kinetic or potential energy is dissipated in one second. This dissipation should
somehow arise from viscous drag in the water. Using the viscosity of water
(10−3 SI units), calculate the dissipation rate in a 5-cm cube of pure water at a
 Experimental probes of structured fluids 33

shear rate of 1/s. Comment on whether this estimate accounts for the damping
of the Jello.
In solutions, the modulus generalizes in an interesting way. A solution
containing dispersed objects resists compression of these objects into a
smaller volume of solvent. This osmotic pressure is evident if one encloses
the concentrated solute in a membrane which can pass the solvent, but not
the solute: solvent flows into the enclosure increasing the pressure there.
Present day osmometers use more sensitive methods, such as e.g. the reduc-
tion in vapor pressure of the solvent or the depression of the freezing point
due to the presence of the dispersed objects [5]. This measurement provides
a way of inferring how many solute particles are present, and of verifying
that they are well dispersed. Other systems, like polymer solutions, have
distinctive predicted behavior of their osmotic pressure, which can be tested
by this means.
2.8. Osmotic sensitivity Many commercial osmometers for water solutions work
by lowering the temperature until ice crystals begin to form. The solute has
the effect of depressing the freezing point, and this depression allows one to
measure the osmotic pressure. The solute particles are virtually all excluded
from the ice crystals. Thus when ice forms, the fluid volume accessible to the
solute is reduced. Each molecule in the water, displaces a volume vw . When it
transforms into ice, it reduces the water volume by vw . The work per molecule
done against the osmotic pressure vw . In order for the molecule to be in
equilibrium with ice, its free energy in the ice must increase by this amount.
For this purpose we may estimate that the free energy is of order T , so that
the needed shift in temperature T required to provide the needed increase
in temperature is given roughly by T vw . (a) Find vw in nm3 given its
mass density and molecular weight. (A useful benchmark is that 1 g/cm3 =
3
0.6023 amu/Å .) (b) If an instrument is capable of measuring T = 0.001◦ ,
what is the largest volume per solute particle V that it can detect?
One of the most important macroscopic properties in a structured fluid
is almost trivial to measure. One simply determines whether or not the
system is stable as a single phase. Phase stability is by no means a foregone
conclusion in a solvent with polyatomic solutes. In some microemulsion and
colloid systems the conditions for single-phase stability and the nature of
the coexisting phases when a single phase is not stable are central questions
for study. We shall see that phase stability reflects distinctive conditions in
the large-scale spatial structure.
Recently the measurement of mechanical response of a fluid has been
extended in space as well as time. Forces can be measured across just a
tenths of a nanometer of a fluid using the Tabor/Israelachvili surface forces
apparatus (Fig. 2.5). This bold and simple device relies on the flatness of
cleaved crystals of mica. Such a surface can be atomically flat over square
millimeters of surface. Thus two such surfaces may be brought to e.g. a
nanometer spacing over many square microns. The surface forces appar-
atus is simply a pair of spring-mounted surfaces immersed in a fluid with
optical means for measuring the distance between them. With this device it
has proven possible to measure normal and shear forces at distances from
1 to 100 nm separations. The technique is sensitive: it routinely measures
surface interaction energies amounting to less than T per (30 nm)2 area.
34 Fundamentals

Force/radius (mN/m)
1 Linear alkanes

Branched alkanes
–1

vdW
–2

Fig. 2.5
Data from Jacob Israelachvili et al. from –3
the surface forces apparatus [6]. The liquid
between the mica plates is tetradecane, an
–4
unbranched hydrocarbon chain 14 carbons
long. Straight, rising segments of the curve
0 1 2 3 4 5 6 7
are unstable regions of the spring balance,
and contain no measured data. Distance, D (nm)

The apparatus was used to exhibit molecular-layer oscillations in forces

in simple fluids. It has also shown forces due to polymers and surfactants
adsorbed on the mica plates. Many of the triumphs of this technique are
reported in the recommended text by Israelachvili.

2.3.2 Probes of spatial structure

Several physical probes explore specifically the arrangement of parts of
a fluid in space. They all involve sending particles or waves through the
fluid and monitoring how they are diffracted, reflected, or absorbed. There
are three basic types: microscopes, scattering experiments, and what one
may call depth probes. Much of the available qualitative information about
structure comes from electron microscopy. As we have seen in the previous
chapter, these can yield direct evidence of structure down to 10 nm and less.
The principle is the same as an ordinary transmission optical microscope.
A beam of particles from a point source is passed through a thin, partially
absorbing sample, and onto a screen or detector. The more absorbing parts
of the sample thus cast shadows on the screen. But both obtaining electron
microscope images and interpreting them is a tricky art. This trickiness
is well epitomized by two ironic laws of electron microscopy from the
microscopists’ folk tradition: (i) It is impossible not to get an image, and
(ii) Not everything you see is an artifact.
The requirements of electron microscopy are essentially incompatible
with those of the liquid state. Structured fluid samples are typically prepared
by rapidly cooling them to cryogenic temperatures. The samples thus frozen
 Experimental probes of structured fluids 35

are sliced in an ultramicrotome for transmission studies. In order to enhance

the absorbing contrast, it is usually necessary to “stain” the sample with
chemicals that absorb strongly and that are attracted to certain molecules
in the sample. Or they may be fractured and the resulting fracture surface
treated to reveal its topography. Often (as in the microemulsion pictures
in the Overview chapter) feats of ingenuity are needed to cool the liquid
fast enough to freeze in the undisturbed liquid structure† . Another problem † Slow freezing invites crystallization and
often encountered is damage by the electron beam, especially when high associated disruption of the original struc-
ture. A slowly frozen sample may resemble
magnification is needed.
the original liquid no more than slowly
The discovery of scanning probe microscopy has enabled new ways to refrozen ice cream resembles fresh ice
examine structured fluids on surfaces. The classic example of this technique cream.
is scanning tunneling microscopy (STM) [7]. In STM one breaks a piece of
metal to produce a sharp asperity and then scans this “tip” over the surface
to be analyzed. One applies a voltage to the tip; thus a “tunneling” current
flows when the tip comes within a few tenths of a nanometer of the substrate.
The current is a very sensitive function of the distance of closest approach.
By adjusting the height of the probe above the surface as it moves across
the surface, one may keep this current constant, so that the distance above
the surface is constant. Thus the probe traces out the height of the surface as
a function of position. The vertical and lateral resolution may be as small as
a tenths of a nanometer. One may use the principle of the STM even when
the substrate is not conducting, as in most surfaces involving hydrocarbon
liquids. In the atomic force microscope (AFM), the tip is made to respond to
the mechanical pressure of contacting the substrate rather than to a current.
The ability to see structured fluid surfaces with such scanning tip probes is
rapidly being developed (see, e.g., Fig. 2.6).
A significant limitation of scanning tip microscopy is its lack of tem-
poral resolution. An image containing the information of a single television
frame requires many seconds to scan. This is very slow relative to the

Fig. 2.6
AFM image (20 nm by 20 nm) of a
three-monolayer-thick film of the
surfactant cadmium aracidate (CdA2 ) on
mica, courtesy of J. Zasadzinski. Lighter
colors correspond to higher areas, and the
peak-to-valley height modulations about
0.2 nm. The two-dimensional fourier
transform is inset. See [8].
36 Fundamentals

relaxation times of liquids, as we have seen. Since the measurement requires

mechanical movement of the macroscopic tip, even local measurements
without scanning are limited to acoustic frequencies. Another problem is
that these microscopes are best suited to measuring a topography—a height
as a function of lateral position. Adsorbates like polymers and surfactants
encountered in structured fluids have a richer structure, not well expressed
as a topography. Using scanning tip microscopy for such a system is rather
like trying to infer the structure of a forest by landing blimps on the treetops.
A complement to microscopy is the statistical information about spa-
tial structure provided by scattering. A plane wave (electromagnetic or
Shrödinger) is sent through the sample. Any inhomogeneities in the inter-
action between the matter and this wave result in spherical scattered waves
propagating outward from the inhomogeneities. The spherical waves add
together to give a net wave intensity in directions away from the incident
beam direction. Inhomogeneities as large as the sample itself do not appear
to the beam as inhomogeneities at all, and the incident beam propagates
through it without scattering. Conversely, isolated pointlike inhomogeneit-
ies produce an isotropic spherical scattered wave. Scatterers of intermediate
size produce scattered waves of finite angular spread.
The angular width θ of the scattered waves is related to the size R of the
scatterers by the well-known diffraction law studied in freshman physics:
R λ/ sin( 12 θ ), where λ is the wavelength of the waves. This simple
relationship between size of objects in the sample and the characteristic
angles of scattered waves can be refined into a powerful probe. We will have
much more to say about the interpretation of scattering in coming chapters.
To illustrate this power, Fig. 2.7 shows a diffraction peak obtained by scat-
tering x-rays from a surfactant–oil solution. The spacing of the diffraction
peaks and the dependence of their position on concentration shows that the
surfactants are arranged in a stack of regularly-spaced sheets about 6 nm
apart. The shape of the diffraction peak shows a more subtle thing: it shows
how the sheets fluctuate away from perfect order. The falloff of intensity
with distance from the peak is a power law. The power varies continuously
and predictably with concentration.

100 Water dilution

w = 0.47
w = 0.65
w = 0.77
Intensity I(q)/I(G)

Fig. 2.7
Scattering data obtained at National
Synchrotron Light source by Didier Roux 10–1
and Cyrus Safinya [9] on a lamellar phase
of 20% surfactant, oil, and water. The
ordinate I (q) is the intensity of scattered
radiation, in arbitrary units. The quantity q
on the abcissa means 2 sin(θ/2)/λ, where
10–2
θ is the deflection angle of the x-rays.
Three sets of data are for three different
amounts of water. The solid curves show –4 –2 0 2 4
the predicted power-law dependence on q. (q – G) 1000 (Å–1)
 Experimental probes of structured fluids 37

2.9. Scattering assay A HeNe laser beam (wavelength about 500 nm) passes
through a glass container of unknown solution and onto the wall. We observe
that there is a faint glow of laser light around the central spot. The angular width
of this glowing region is about 5◦ . From this we infer that there are objects in
the solution which are much larger than atoms. (a) About how large are they?
(b) A second solution, similar to the first, shows the same type of scattering,
◦
except that the angular width is now only 2 12 . By what factor are these objects
larger or smaller than the ones in (a)?

If the scatterers move, the scattered intensity fluctuates in time. By

analyzing these intensity fluctuations, one may infer something about how
the scatterers move. The principle is similar to that behind diffraction itself.
If in a given time the scatterers move much less than one wavelength in the
medium, the phase of its scattered waves on the detector is changed by a
only a small amount, and the net intensity is little affected. The intensity is
virtually constant over such a short time. Given longer time intervals, the
scatterers may move as much as a wavelength. Then the resultant intens-
ity is not constant but fluctuates. Thus the correlation time over which the
intensity stays constant measures the time for the scatterers to move about a
wavelength. Thus dynamic scattering is a powerful probe of motion as well
as structure [9].
The waves most commonly used in scattering studies on structured fluids
are visible light, x-rays, and neutrons. Visible light is of limited usefulness,
because most structures of interest are smaller than optical wavelengths.
Still it can give limited information about overall sizes and long-distance
diffusive motions. Dynamic visible-light scattering can probe correlation
times from a microsecond out to (at least) several seconds. X-rays have
typical wavelengths of roughly a tenth of a nanometer—smaller than the
structures of interest. But the high quality of the best x-ray beams permits
scattering measurements at very small angles, corresponding to distances
λ/ sin( 12 θ ) of 103 nm. (It is not trivial to use such beams; one must go to a
national synchrotron source.) X-rays are not routinely used to infer motion.
The reason lies in the quantization of the waves. For a quantized wave to
sense motion by the means discussed above, more than one quantum should
scatter off each scatterer during the correlation time. For x-rays this would
ordinarily entail the deposition of too much energy and momentum for the
sample to sustain† . † Recently it has proven possible to perform
Modern neutron sources have wavelengths of a few tenths of a nanometer. dynamic scattering with x-rays in favorable
cases, despite the disadvantages noted in the
The neutrons are produced by a nuclear chain reaction (in a nuclear reactor)
text [10].
or a reaction with a charged particle beam from an accelerator†† . The neutron
radiation thus produced is incoherent thermal radiation, like the light from †† The apparatus is called a spallation

a light bulb. It is far from the angular or energetic purity of a laser beam or source.
a synchrotron source. Thus the neutron beam must be collimated and made
mono-energetic by great sacrifices in intensity. Only a few facilities in the
world have enough intensity to do adequate scattering studies. Despite these
obstacles, neutrons offer some unique advantages. The main one is that they
permit isotopic labeling. One may significantly alter the scattering power
of an atom by neutrons if one substitutes a different isotope of the atom.
Such a change has virtually no effect on its behavior in a fluid. Thus, by
judicious attachment of different isotopes to different parts of a structured
38 Fundamentals

fluid, e.g. a colloidal core, the surfactants around it, and the solvent, one
may get much information about the spatial correlations of each component
and about cross-correlations between components.
Neutrons also give dynamic information. If a neutron scatters off a mov-
ing object, its kinetic energy and speed are altered. The distribution of
speeds in the scattered beam can be elegantly and sensitively measured by
sending this beam through a magnetic field. The magnetic field causes the
magnetic moment of the neutrons to oscillate with a specific frequency.
The neutron spin echo technique uses these oscillations to infer motions of
the scatterers in this frequency range. Frequencies between 106 and 109 /s
or so can be measured this way.
Waves and particles are used in another way to probe spatial structure
near surfaces. Specifically, they can sense the profile of some atomic spe-
cies with depth. The simplest of such method is to sense the reflectivity of
a wave incident on the surface. The profile of scatterers results in a profile
of index of refraction seen by the wave. This in turn shapes the reflectivity
as a function of angle, especially near the total-internal-reflection angle.
The method senses depths as small as a tenth of a wavelength and depths
as large as 100 wavelengths. It is not trivial to infer the index-of-refraction
profile from the reflectivity spectrum. One must assume various profiles
and calculate their spectra until a spectrum matching the measured one is
obtained. Sometimes two rather different profiles can give similar spec-
tra so that interpretation of the data is ambiguous. A less ambiguous but
more brute-force probe of surfaces is called Rutherford backscattering [11].
There the probe is a beam of massive particles like alpha particles, whose
wavelength is negligible. When a mono-energetic beam is directed onto a
surface, the particles backscatter elastically from the nuclei in the substrate,
losing a specific amount of energy that depends on the mass of the nucleus
hit. One may infer the depth of a backscattering event by the further energy
loss of the emerging particle on its way in and out. This method requires a
small room-sized particle accelerator. The method is routinely used to sense
the profile of composition in solid samples to depths of the order of 10 nm.

2.3.3 Probes of atomic environment

There is more to be learned about a fluid than simply where the atomic
species are and how fast they are moving. For many purposes it is valuable
to know how parts of a molecule are oriented, how fast this orientation
changes, what atomic species are near a given atom and how near. A vari-
ety of nuclear and electronic resonance methods are available to sense these
things. Magnetic resonance senses the frequencies of rotation of atomic-
scale magnetic moments in a large static magnetic field. When a small
oscillating electric or magnetic field of the same frequency is applied, res-
onant absorption can be detected. Nuclear magnetic resonance (NMR) [12]
senses the electric and magnetic field at a specific type of nucleus. The res-
onant frequencies are shifted or split by static fields. Time-varying internal
fields can also limit the lifetime of the resonant state. An important example
of NMR used in practice is to replace hydrogen by deuterium at specific
points on chain molecules. Then by measuring the effect on the resonance
 Solution to Problem 2.1 39

line, one may sense the average orientation of that part of the chain in space. Free chains ( = 0.92)
As Fig. 2.8 shows, the changes in the resonance can reveal subtle changes 100 Hz

in the atomic orientations due to overall deformation of the fluid. One can
sense the effect of a 10% stretch of a rubber in this way; likewise, one  = 1.0
can sense the alignment of specific carbon–deuterium bonds in a surfactant
monolayer.
Another class of local probes involves optical fluorescence of small dye
molecules [14]. The dye molecule is chemically attached to the macro-
 = 1.22
molecule to be probed. For example, fluorescence may be excited with
a pulse of light and its intensity or polarization are monitored with time.
The change of polarization senses the dye molecule’s rotation from its ini-
tial orientation. The time dependence of intensity is of interest because
the fluorescent decay can be strongly affected by the presence of quench-  = 1.40
ing molecules. From the intensity decay one may infer the rate at which
excited molecules encounter quenching molecules. By attaching the dye
and quencher at suitable places, one may use this as a probe of mobility, and
of where the dye molecule likes to sit. A variant of the quencher molecule  = 1.55
is an excimer-former—usually a second dye molecule. Excimers are bound
states of two dye molecules. The proximity of the second molecule shifts Fig. 2.8
the fluorescence in a specific way. Fluorescence probes of this kind are Nuclear magnetic resonance spectrum of a
very sensitive and flexible in their time scale. Temporal responses have rubber after [13]. The rubber is a polymer
been measured on scales ranging from fractions of nanoseconds to many made of hydrogen, carbon, oxygen, and
silicon (poly dimethyl siloxane) in which
seconds. the hydrogen atoms have been replaced by
deuterium atoms. The plot shows
absorption versus frequency of oscillation.
A strip of the rubber was stretched in the
Solution to Problem 2.1 direction of the static field by various
The outer part of the system (the “reservoir”) has N − k ≡ 
m balls in U − V ≡ v elongation factors λ shown at the left. The
sites. The number of distinct configurations with m balls {dm } 1 will be called splitting of the resonant peak indicates a
resulting slight anisotropy in the
the “partition function” zm . We wish to infer zm+1 given zm . Any m + 1-ball
neighborhood of the deuterons. The
configuration may be made from some m-ball configuration by adding one ball.
splitting is proportional to the amount of
Thus to count all m + 1 configurations we may consider each m-ball configuration
stretching. The frequency of the unsplit
and ask how many m + 1-ball configurations can be made from it.
peak is 13 MHz. The bottom curve shows a
(a) Whatever m-ball configuration we start from, there are exactly v m + 1-ball splitting of about 100 Hz.
configurations that can be made from it, since every possible site to add the
new ball results in a distinct configuration.
(b) Each of these m + 1-ball configurations could have been made from m + 1
different m-ball configurations. If we remove any one of the m + 1 balls, we
get one m-ball configuration. If v  m, the chance of two balls at a site is
negligible and so essentially all these m-ball configurations are distinct.
(c) Not all the configurations found in (a) are distinct: one can arrive at the same
m + 1-configuration starting from many different m configurations. In fact, if
we generate all the m + 1-configurations separately from each m configuration,
we will end up generating each m + 1 configuration m + 1 times—once for
every m configuration that could have given the m+1-configuration in question.
Thus of vzm configurations generated, a factor m + 1 are redundant: they are
duplicates of others. The number of distinct m + 1 configurations zm+1 is thus
(v/(m + 1))zm . We recall that the desired probability p(ck ) = (const)zN−k .
Thus we have found that p(ck+1 )/p(ck ) = zN −k /zN −k−1 = (N − k)/v. Since
N  k, this ratio becomes N /v and is independent of k, as we wished to show.
(If the reservoir is not dilute, the ratio p(ck+1 )/p(ck ) changes. It becomes a
function of the volume fraction N /U of the reservoir, because the number of
40 Fundamentals

m-ball configurations obtained from an m + 1-ball configuration is no longer

m + 1. Still, the p(ck+1 )/p(ck ) remains independent of k.)
(d) p(ck ) = p(ck−1 )(N/v) = p(ck−2 )(N/v)2 = · · ·
= p(0)(N/v)k = p(0)ek ln(N /v) .

This can be written p(ck ) = p(0)e−k/K where K = 1/ ln(v/N). Note that

p(ck ) is the probability of a given configuration of the subsystem. To find the
overall probability that the subsystem has k particles, one must add up the
probabilities for all the k-particle configurations of the subsystem.

References
1. See e.g. N. G. van Kampen, Stochastic Processes in Physics and Chemistry
(Amsterdam, New York: North-Holland, 1992).
2. S. Glasstone, K. J. Laidler, and H. Eyring, The Theory of Rate Processes,
(New York: McGraw Hill, 1941); H. A. Kramers, Physica 7 284 (1940);
S. Chandrasakhar, Rev. Mod. Phys. 15 1 (1943).
3. See e.g. H. Gould and J. Tobochnik, An Introduction to Computer Simulation
Methods (New York: Addison Wesley, 1988).
4. D. S. Pearson, Rubber Chem. Technol. 60 439 (1987).
5. C. Morris and H. Coll, Determination of Molecular Weight (New York: Wiley
and Sons, 1989).
6. Reprinted from Jacob N. Israelachivili, Intermolecular and surface forces, 2nd
ed. (London; San Diego, CA: Academic Press, 1991). p272. © 1991 with
permission from Elsevier.
7. R. Weisendanger and H. J. Guntherodt, Theory of STM and Related Scanning
Probe Methods, Springer Series in Surface Sciences, V 29 (1993).
8. J. A. Zasadzinski, R. Viswanathan, L. Madsen, J. Garnaes, and D. K. Schwartz,
Science 263 1726 (1994).
9. B. J. Berne and R. Pecora, Dynamic Light Scattering: With Applications to
Chemistry, Biology, and Physics (New York: Wiley, 1976); Dynamic Light
Scattering: Applications of Photon Correlation Spectroscopy, ed. R. Pecora
(New York: Plenum Press, 1985).
10. A. Q. R. Baron, H. Franz, A. Meyer, R. Ruffer, A. I. Chumakov, E. Burkel, and
W. Petry, Phys. Rev. Lett. 79 2823 (1997).
11. For this and related methods, see T. P. Russell, Ann. Rev. Materials Sci. 21 249
(1991).
12. See e.g. V. D. Fedotov and H. Schneider, Structure and Dynamics of Bulk
Polymers by NMR-Methods, NMR Basic Principles and Progress series, Vol. 21
(Heidelberg: Springer-Verlag, 1989).
13. B. Deloche and A. Dubault, Europhys. Lett. 1 629 (1986).
14. See e.g. C. E. Hoyle and J. M. Torkelson, eds., Photophysics of Polymers, ACS
Symposium Series Vol. 358 (Columbus, Ohio: American Chemical Society,
1987).
 3
Polymer molecules

In this chapter we have our first in-depth encounter with a specific fluid
structure: the flexible chain molecules known as polymers. The basis for
much of their distinctive behavior lies in their flexibility and randomness.
This chapter focus on the properties of individual molecules that embody
this flexibility and randomness. The next chapter focuses on the further
properties that emerge when large numbers of polymers interact in a liquid.
There we will also treat the basic motions of a polymer solution: diffusion
and flow. The Preface notes a number of other texts covering these polymer
phenomena.
We begin this chapter with an informal introduction to some common
monomers and the ways they can be joined to make a polymer chain. Next
we use the idealized example of a random-walk polymer to demonstrate
scaling and renormalization. The following section describes the interior
structure of polymers, emphasizing their fractal properties and dilation
symmetry, and showing how these features can be revealed by scattering
experiments. Next, we study the impact of self-avoidance on the structure
of a polymer. We find that self-avoiding polymers have fractal structure
but the scaling exponents are altered by the self-avoidance constraint. We
discuss another form of self-interaction that occurs when the polymer is
electrically charged; this leads to dramatic changes in its structure.

3.1 Types of polymers

The simplest polymer one can imagine is a chain of carbon atoms each
holding two hydrogen atoms: CH 3 − −[CH 2 −
−]n CH 3 . This polymer, poly-
ethylene, is also one of the most ubiquitous. The degree of polymerization
n is typically 103 –105 . This is the material of the cheapest plastic bags.
Polyethylene, more than other types of plastic, feels waxy. This is no
co-incidence: polyethylene differs from paraffin wax only by having a larger
value of n.
This polymer is the simplest in a rich variety of polymer types. In this
section we survey the various types of polymers that exist, to give a sense
of the range of molecules that can be made. First, I will discuss several
common monomers and the properties associated with them. Then I will
describe the main architectural variations in how monomers may be put
together. Finally I will discuss the two main types of chemical reactions
that produce polymers.
42 Polymer molecules

3.1.1 Monomers
†For reference, carbon in organic molecules The simplest and most common polymers are hydrocarbons, i.e., molecules
has a valence of four; i.e., it makes four consisting entirely of carbon and hydrogen. It is useful to classify the differ-
bonds to adjacent atoms. Two of these may
ent types according to their polarity, i.e., the degree of electric polarizability.
go to the same atom, forming a double bond.
Silicon also has a valence of four. Nitrogen The least polarizable hydrocarbons are the saturated hydrocarbons, where
has a valence of three. Oxygen and sulfur each carbon is bonded to four different atoms (i.e., there are no double
have a valence of two. Hydrogen has a bonds)† . The bonds in a saturated hydrocarbon are very stable, symmetrical,
valence of one. and difficult to deform. Small saturated hydrocarbons are oils and waxes.
All these species mix well with each other but less well with the more polar
†† We return in Chapter 7 to a more quant- species to be described below. They mix poorly with water†† . Polypropyl-
itative discussion of molecular mixing. ene is the main example besides polyethylene. It has a CH 3 group replacing
one of the hydrogens on every second CH 2 group of polyethylene††† . Poly-
††† A polymer is generally named after the
ethylene and polypropylene are structurally simple enough to crystallize
monomer from which it is made. For readily. This contrasts with another very common polymer, polystyrene,
example the ethylene monomer has four
the material of cheap plastic spoons and styrofoam cups. Figure 1.8 shows
hydrogens surrounding two double-bonded
carbons. In polymerization, the second a typical configuration of polystyrene. It is related to polypropylene by
carbon–carbon bond of ethylene becomes replacing each CH 3 group of polypropylene by a benzene C6 H5 ring. The
attached to the end of the polyethylene bulky rings make crystallization difficult; thus polystyrene solidifies in a
chain. Sometimes a polymer has two differ- glassy rather than a crystalline state.
ent names, because either of two monomers
Polystyrene is an unsaturated hydrocarbon. That is, it contains double
can be used to make it.
bonds (between some carbon atoms and their neighbors). Some of these
redundant bonds could be used to attach additional hydrogen atoms, thus
“saturating” the molecule with hydrogen. The redundant bonding in poly-
styrene resides in the delocalized electrons in the benzene rings. These, like
other double bonds, are more polarizable than simple bonds. This means
that polystyrene (at high molecular weight) is not miscible in saturated
polymers. Two other important unsaturated polymers are natural rubber
or polyisoprene, and the synthetic rubber polybutadiene. Several common
hydrocarbon polymers are sketched in Fig. 3.1.
Polyethylene The inclusion of elements beyond carbon and hydrogen makes for asym-
Polyisoprene
metric bonds with carbon, and often increases polarity. For example, one
may make a carbon chain with oxygen at every third position. The simplest
case is [−−CH 2 −−CH 2 − −O −−]n —polyethylene oxide§ . Polyethylene oxide
is sufficiently polar to be soluble in water. Polymer backbones can be made
Polypropylene Polystyrene with no carbon at all. The most common case is polydimethyl siloxane.
This is the material of silicone rubber, sealants, and lubricants. It has
Fig. 3.1 a backbone of alternating silicon and oxygen atoms. Each silicon has in
Four common hydrocarbon polymers. addition two methyl (CH 3 ) groups. Polyesters such as DacronTM have the
Black dots represent carbon atoms. Solid
lines represent chemical bonds. A
generic structure [− −R −−COO− −]n , where the R is an arbitrary molecular
hydrogen atom is at the end of each bond substructure called a residue and COO is a carbon with a double-bonded
where no carbon atom is shown. Dashed oxygen and a single-bonded oxygen taking three of its four available bonds.
lines represent bonds from the preceding The last bond is with the R group. The double-bonded oxygen may form
monomer in the chain. Variants of weak hydrogen bonds with any covalently-bonded hydrogen present, e.g.,
polybutadiene are shown in Fig. 3.3.
in the R group. Another example is the polypeptides; all proteins are
polypeptides. Their backbone has the form [− −N −−C −−C − −]n , with an H
§ also known as poly-oxy-ethylene, or poly- taking the third bond of the nitrogen, an H and a variable R taking the
ethylene glycol (PEG). remaining two bonds of the middle carbon, and a double-bonded O taking
the two bonds of the last carbon. In many polypeptides (with favorable
 Types of polymers 43

R groups) this oxygen forms a hydrogen bond with the hydrogen sev-
eral monomers down the chain, thus forming a helical structure called the
alpha helix. The alpha helix is one example of weak, cooperative bonding
within a chain which favors a certain configuration or crosslinking topo-
logy. This type of bonding is called secondary structure (in proteins) or
association (in synthetic polymers). Secondary structure can often be dis-
rupted reversibly without disrupting the primary, covalent bonds of the
polymer.
Other types of monomers respond more strongly to electric fields than can
be described by a polarizability. Conjugated polymers like polyacetylene
have alternating single and double bonds along the backbone. Moreover,
translating all the double bonds forward by one unit along the structure res-
ults in an identical structure. The position of the double bonds represents a
broken symmetry. By doing this translation, one has transported an electron
in each double bond one step along the chain. This means that such poly-
mers may readily be made into electrical conductors. This requires doping,
the removal or addition of a small fraction of the electrons. Conjugated
polymers can have electron mobilities comparable to those found in metals.
This conductivity can be made to occur in the liquid state and in crosslinked
solutions [1].
Polymers can be synthesized with ionic side groups. A common example
is polystyrene sulfonate. This derivative of polystyrene has an SO− 4 ion
attached to each benzene ring. The neutralizing counterion is typically a
metal like Na+ . In a polar solvent like water, the electrostatic binding energy
of the counterions is smaller than the thermal energy T , and accordingly
most of the ions dissociate and move freely in the solvent. The polymer is
left with a large net charge. Such polymers are called polyelectrolytes. We
will see later in this chapter that the charge has a dramatic effect on their
shape and properties.

3.1.2 Architecture
In polymerization monomers like those catalogued above are joined end
to end. The result is generally a flexible chain molecule, in which each
single bond can rotate. That is, the bond acts like a rigid rod, with the
atoms at the two ends attached as if on freely rotating bearings. The bonds
of a carbon atom point symmetrically outward at fixed angles of about
109◦ . A double bond may be regarded as a pair of rods; it does not allow
rotation. Even in a flexible polymer chain certain aspects of the structure
are locked in and unchangeable, over and above the fixed structure of the
constituent monomers. This structure is determined at the moment each
monomer joins the chain. Two types of locked-in structure are important:
isomerism and tacticity. Isomerism concerns the orientation of the two ends
of the monomer along the chain. Each monomeric link is attached to the
chain at its two ends. In general the two are not equivalent, so that monomers
may be attached “head to tail” in the same orientation, or “head to head”
in opposite orientations. How much of this isomeric irregularity occurs
depends on the details of the polymerization reaction. Of the polymers
shown in Fig. 3.1, only polyisoprene has isomeric variation of this type.
44 Polymer molecules

Fig. 3.2 R R R
H H H
A generic polymer chain with a C2 repeat
unit along the backbone. Bonds directed
out of or into the backbone plane are shown
as triangles. The residues R are at the
isotactic positions; they are all on the same H H
side (the front) of the backbone plane. H H

Cis-trans isomerization refers to the atomic configuration around a pair of

double-bonded carbons. An example is shown in Fig. 3.3.
Tacticity is a second way in which the structure of a polymer chain may
be irregular. To demonstrate what tacticity means, we consider a polymer of
the form [− −CH 2 −−CHR− −]n , like the polypropylene of Fig. 3.1. We rotate
the bonds so that the carbon backbone lies in the plane of the paper from
left to right. Since the successive bonds are at 109◦ , tetrahedral angles, the
backbone makes a zigzag line. We suppose the isomerism is head-to-tail, so
that every other carbon—say the upper ones—have an R group. As shown
in Fig. 3.2 the C −−R and C − −H bonds must stick obliquely out from and
back from the page. An isotactic structure is one in which all the R groups
lie on the same side of the backbone—i.e., all out of or all into the page. In a
syndiotactic structure the R groups alternate along the backbone: every other
one points out of the page. Atactic structures have a random positioning of
the R’s. Evidently tacticity is a permanent property of the chain structure; it
cannot be changed by rotating bonds. Tacticity has a noticeable effect on the
miscibility of polymers and on their ability to crystallize. It can be controlled
to some degree by the polymerization conditions like temperature, choice
of catalyst, etc.
More profound variations along the chain are those in which the
monomeric precursors are chemically different. A common example is
polyethylene–propylene, made by copolymerizing ethylene and propylene
monomers. The sequence of monomers obtained is quite disordered, though
not completely random. The disorder is sufficient to prevent the crystal-
lization that occurs in ethylene or propylene homopolymer. Thus these
copolymers remain liquid well below room temperature. Liquid polymers
when crosslinked have the springy deformability of rubber. Polyethylene–
propylene is one of the cheapest polymers and is used in great quantities
† The commonest and cheapest rubber is as a rubber† . Nowadays it is possible to control the sequence of monomers
styrene butadiene copolymer. in a copolymer in some ways. An important case is the diblock or trib-
lock copolymer—two or three homopolymer chains joined end to end.
When the two copolymerized species are mutually immiscible, dramatic
self-organization like that shown in Chapter 1 can occur. We postpone our
discussion of these phenomena to the “Amphiphilic polymers” section in
Chapter 7.
Many variants of the linear-chain topology occur. Liquid polymers are
converted to rubber by crosslinking: adding chemical groups (like sulfur)
that join two chains together. Often chains branch as they polymerize.
Nowadays, using favorable monomers, it is possible to build polymers
with virtually any desired topology. This includes rings, stars with up to
60 arms, and H-shaped molecules [2].
 Types of polymers 45

3.1.3 Polymerization
To create a polymer, one must find a chemical reaction mechanism that
joins the monomers together. The organic chemistry of these reactions
[3] is similar to that of small molecules, and we shall not discuss it here.
But some important features of polymerization can be appreciated without
organic chemistry. For example, statistical fluctuations are important in
the many-body process of polymerization. These produce variations in
isomerization and tacticity along the chain. More centrally, they produce
variations of molecular weight from chain to chain. We discuss three types
of polymerization to illustrate the possibilities.
The least variability occurs in biosynthesis, the natural process by which
proteins are synthesized in living organisms. By recombinant DNA techno-
logy the natural processes can now be commandeered to produce peptide
copolymers with the precise sequence desired [4]. These are virtually
identical in structure and molecular weight from chain to chain. This kind of
synthesis is still experimental as a means of producing synthetic polymers
and is restricted to polypeptides and DNA-like polysaccharides.
Some of this molecular uniformity occurs in the important type of poly-
merization called addition polymerization. In this method a catalyst initiates
polymerization in a solution of monomers. This turns a small fraction of
the monomers into a chemically active form such as a free radical. An
example is the polymerization of styrene monomers CH2 = CH R. In
the active form the double bond is broken and made reactive. This act-
ive form is able to combine with a second monomer. The activity of the
first monomer propagates to the second. Thus it is able to react with
a third, and so forth. Ideally many monomers are joined in sequence
before the activity of the chain end is lost. Finally the reaction is termin-
ated by deliberately adding a chemical that inactivates the active ends.
Another example of addition polymerization is shown in more detail
in Fig. 3.3.
The successive addition of monomers to the chains is a statistical process;
still, it results in relatively uniform chains when these chains are long. The
distribution of lengths P (n) becomes relatively narrower as the mean chain
length n gets larger. This results from the law of averages. We may see how
this works by considering the time T required to make an n-mer. Evidently
T is the time t2 required to attach the second monomer to the first, plus the
time t3 for the third monomer, and so on up to tn . Of course each of these
ti varies widely from chain to chain, owing to the statistical fluctuations in
the system when the ith monomer was being added. These variations are
reflected  in the total time T , by the familiar law of propagation of errors:
T 2 = ti2 . The variations of the ti are roughly the same for all i,
so that T 2 ntn2 . Thus T T n−1/2 ; the relative variation in T
becomes progressively smaller as n increases.
This increasing uniformity contrasts with the results of condensation
polymerization. Here instead of each chain having a single active end that
reacts only with monomers, many chain ends may react with one another.
An example is the polyamide nylon, in which any two chain ends may join,
producing a water molecule as a by-product (Fig. 3.4).
46 Polymer molecules

Fig. 3.3
M M
Addition polymerization of butadiene. *
Proper bond angles are not shown. The
initiation of the reaction by a metal atom *
M produces an unpaired electron, denoted
by a star. Under certain reaction
conditions, this unpaired electron resides
preferentially on the second carbon atom.
This unpaired electron then initiates M M
bonding with the end carbon of a second *
butadiene. It then moves to the second
*
carbon, as before. Additional monomers
are incorporated in the same way. The
resulting “1,2 polybutadiene” has
two-carbon side groups. Each may extend
either out of or into the plane of the
backbone, so that the chain has tacticity as
illustrated in Fig. 3.2. Alternatively, under
Cis Trans
different reaction conditions the unpaired
electron moves to the carbon atom farthest
from the metal initiator. This electron
induces bonding with the end carbon of
The kinetics of this reaction involves many processes contributing to the
another monomer, leaving an unpaired
electron that moves again to the farthest change of the n-mer concentration [n]. For simplicity we assume that all
carbon atom. Successive addition of n-mers are equally reactive with one another. Each reaction of i-mer with
monomers in this way produces “1,4 j -mer to produce n-mer is a second-order reaction, whose rate is propor-
polybutadiene.” Double bonds do not tional to the amount of each reactant present, i.e., to [i][j ]. Considering all
rotate freely like single bonds. Thus, the
these reactions that affect the number of n-mers, we have:
four backbone atoms around any double
bond can be in either the planar “cis” or

1 
“trans” isomeric states shown in ovals. n−1
d[n]
=K [k][n − k] − [n] [k] . (3.1)
dt 2
k=1

[n] increases owing to the reaction of (n − 1)-mers with 1-mers,

(n − 2)-mers with 2-mers, etc. [n] also decreases owing to the reaction
of existing n-mers with anything to produce longer chains. We have sup-
posed that all these reactions occur at some uniform rate governed by the
single constant K. This assumption is justified under certain conditions.
It happens that this formidable-looking set of coupled rate equations has
a simple solution, as one may readily verify. It is exponential for long times
and large n : [n] → (constant) exp(−n/n ). Thus the relative width of
the molecular weight distribution does not become narrower with time, in
contrast to addition polymerization.
A third category of polymerization is called living polymerization. Here
the polymerization reaction is reversible, so that the various chains are free
to exchange monomers amongst themselves. This freedom to exchange res-
ults in a broad distribution of chain lengths, much like that of condensation
polymers. We shall return to living polymers in Chapter 7, in our treatment
of wormlike micelles. In the sequel we shall encounter many reasons why
the molecular weight distribution is important.
3.1. Polycondensation In a polycondensation reaction, the reaction rate of k-mers
with l-mers was proportional to the product of their concentrations [k][l]. Thus,
 Random-walk polymer 47

H H H H Fig. 3.4
C C C C O H A polycondensation reaction. An amine
H O C
NH 2 end reacts with an alcohol COH end
O H H H H by eliminating the circled atoms to form a
H H H H H H H H
water molecule, thus joining the two
N C C C C C C N
H H chains together. The polymers thus
H H H H H H produced are called nylon 6-6.

considering all the reactions which increase or decrease the number of n-mers,
the rate of increase of the concentration [n] obeys Eq. (3.1) in the text, where K
is an overall rate constant. In analyzing this equation, you may assume that n, k,
and l are everywhere large enough that they may be considered as continuous
variables running from 0 to ∞  instead of discrete variables running from 1 to
∞. (a) Show that the total mass n[n] is independent of time, as it must be. (b)
 [n](t) = [0](t) exp(−nc(t)), where2 c
Show that there is a solution of the form
and [0] are some function of t. (Since n[n] = [0]/c2 , [0] must be  (const)c .)
How does c vary with t? (c) How is c related to n ≡ ( n[n])/( [n])?

3.2 Random-walk polymer

All of the chain polymers described in the last section have some ran-
domness in their spatial arrangements. This randomness is least in a
self-associating species like a protein. Proteins are often frozen into a spe-
cific configuration with a specific size and shape. The randomness in each
atom’s position then consists of small excursions from its average position,
like the vibrational fluctuations in a solid. Much greater configurational
randomness occurs when a non-self-associating molecule like polystyrene
is dispersed in a good solvent like toluene. In this section we explore the
results of this randomness. Here there is no specific size and shape; instead,
a statistical description is necessary. We shall also explore the characteristic
energy scales of these polymers—i.e., the work required to deform them
from their natural random state.
We may begin to explore the spatial randomness of a polymer by looking
at the smallest polymer: a four-carbon segment of polyethylene. A fun-
damental geometric property to examine is the distance between the two †In fact not all rotations are equally likely
terminal carbons. This distance is not definite because the carbon–carbon because of subtle chemical interactions
bonds are free to rotate;† the only restriction is that they maintain the 109◦ between adjacent bonds. But such subtleties
may be ignored for our purposes.
angle between successive bonds. We may assume that the first two bonds
lie in the plane of the paper and that the first bond is horizontal. Then the
third bond does not in general lie in this plane. As this bond swings around
into its various possible positions, the end-to-end distance changes. The
maximum end-to-end distance occurs when the third bond is parallel to
the first in a trans configuration; the minimum occurs when this bond is
rotated half-way around, into a cis configuration Fig. 3.5. Thus there is a
probability distribution in the end-to-end distance r. There is also a distri- Fig. 3.5
bution of angles between the first and third bonds, ranging from 0 degrees Two configurations of a four-bond
to 2 × 109◦ .
hydrocarbon polymer with maximal and
minimal end-to-end distances. The
A fourth bond adds to this randomness. For each configuration of the left-hand chain is all trans; the right-hand
three-bond chain, the fourth bond may swing around a complete circle of chain is all cis.
48 Polymer molecules

angles as the third one did. Here the last bond may point in any direction
relative to the first, and these directions are nearly equally likely. With
the addition of another bond or two, the directional randomness of the last
bond is nearly complete. To discuss the randomness in a long polymer, it
is convenient to imagine that its building blocks are segments which are
several atoms long, so that the directions of the beginning and the end of
† In some cases this directional randomness each segment may be assumed statistically independent† .
cannot be assumed. An important one is the Another feature appears when our hydrocarbon chain exceeds five bonds
case when the backbone is rigid rather than
in length: it may hit itself. There are some rotational configurations of the
flexible. Important examples are DNA and
RNA helices and straight-chain hydrocar- bond angles that would place one carbon atom on top of another. Clearly
bons like polyacetylene. Here each segment these configurations are absent in real polymers: the polymer must avoid
has nearly the same direction as its pre- itself. This requirement is clearly separate from the requirements on the
decessor. The direction can only become successive bond angles or local structure. Local structure can be expressed
random by the addition of many of these ran-
in terms of random variables (e.g. our bond angles) that are statistically
dom small increments over a long length of
chain. Chains shorter than this persistence independent for each bond or small segment. The self-avoidance constraint
length but much longer than a monomer cannot be expressed in this way. A major topic of this chapter is to analyze
have a universal statistical behavior more the effects of self-avoidance. For the moment, we shall ignore it. We shall
complicated than that of the random-walk depart from the real world and imagine phantom polymers which are free
polymers treated in the text. Rigid chains in
to intersect themselves. After their behavior is understood, we will be in a
this regime are called wormlike chains. [5].
position to attack the constraint of self-avoidance.
3.2. Law of Averages for r 2 A polymer in one dimension consists of n vectors
a1 , a2 , . . . , an of length 1 that point randomly left or right. Each vector is
independent of the others. (a) Find  a5 ,  · a5 , and 
a5  a4 · a5 . (b) Using
averages such as these, and noting that r = i ai , find r and (r )2 as a
function of n.

3.2.1 End-to-end probability

Although the end-to-end distance of our chain varies over a wide range, there
must be some way to characterize it statistically, even for very long chains.
We are thus led to look at the limiting behavior of the end-to-end probability
distribution p(n, r), defined as the probability that the end-to-end vector
has some particular
 value r in a chain of n segments. We normalize p(n, r)
according to d 3 r p(n, r) = 1, so that p is a probability-per-unit-volume.
We may define how this p(n + 1, r) must behave for all r if we know
p(n, r) for all r. This is because any n + 1-segment chain can be thought
of as an n-segment chain with a single segment joined to its end. Here as
indicated above, we shall think of each segment as a sequence of several
atoms. We suppose that the sequence is long enough so that the directions of
its two ends are independent. That means that the segment itself has an end-
to-end probability p0 (r1 ) that depends only on the magnitude of r1 . It also
means that this p0 (r1 ) is unaffected by the arrangement of the n-segment
chain attached to it. In that case, we may regard the occurrence of an n + 1
chain with vector r as the joint occurrence of two independent events: (1)
last segment has some vector r1 , and (2) the first n-segment part of the chain
has the vector r − r1 . The probability of this joint occurrence is p(n, r −
r1 )p0 (r1 ). To obtain the full p(n + 1, r) we must add the probabilities for
 Random-walk polymer 49

all occurrences that give the vector r:


p(n + 1, r) = d 3 r1 p0 (r1 )p(n, r − r1 ). (3.2)

This is a clear-cut prescription for determining p for arbitrary n given

p of a single segment. Without some specific input regarding p0 (r) this
seems to be as far as we can go. On the other hand, intuition tells us that
we should be able to say something further about p(n, r) by asking for
only part of the full behavior implicit in Eq. (3.2): namely, the limiting
behavior for arbitrarily large chain-length n. This problem is archetypical
of the quantitative issues in this book. We will be repeatedly interested in
how some quantity behaves when some parameter measuring the system
size goes to infinity.
Accordingly, we try to simplify Eq. (3.2) in the limit of large n. The
equation tells the effect of incrementing n by 1. When n becomes large, the
effect on p must become small. We may examine the small change in p
directly by subtracting p(n, r) from each side of Eq. (3.2).

p(n + 1, r) − p(n, r) = d 3 r1 p0 (r1 )[p(n, r − r1 ) − p(n, r)]. (3.3)

The small difference on the left can evidently be treated as a derivative.

The same should be true for the spatial difference of p’s in brackets on
the right. For large n we expect that the typical distances r for which p is
significant are much greater than the distances r1 for which the integrand has
a significant contribution. Thus the difference in [. . .] involves relatively
small spatial displacements r1 . Over such small distances, we expect p(n, r)
to be gently varying. Thus we use a Taylor expansion to represent this
variation:

p(n, r − r1 ) = p(n, r) − r1 · ∇p(n, r) + 16 r12 ∇ 2 p(n, r) + · · · (3.4)

Using these derivatives, Eq. (3.3) becomes

 
dp 1 2
= −∇p · d r1 r1 p0 (r1 ) + ∇ p(n, r) d 3 r1 r12 p0 (r1 ) + · · · .
3
dn 6
(3.5)

We note that the first integral is the average of r for a one-segment chain:
r 1 and the second is r 2 1 .
We have taken a case where the step distribution p0 (r) is isotropic; thus
the first integral must vanish. The same is true with all odd powers and their
associated odd-order derivatives. Thus our equation for p reduces to

dp/dn = 16 r 2 1 ∇ 2 p(n, r) + const. “∇ 4 ”p + · · · . (3.6)

Here we have indicated the form of the first nonvanishing term beyond the
∇ 2 term; the “∇ 4 ” denotes various fourth-order derivatives.
50 Polymer molecules

We can deduce much about this equation without solving it, by invest-
igating its asymptotic behavior as n → ∞. But it will not do simply to
evaluate p(∞, r). We would find that this is zero for all r. We are in the
paradoxical situation of taking n arbitrarily large, but not infinite. Since
it is arbitrarily large, the specific value of n cannot be important. In the
regime of interest the behavior of p is the same whether we consider n,
n/1000, or 1000n. That is, p should show the same behavior when n is
multiplied by an arbitrary factor λ. By assuming such a limiting behavior,
we are implicitly postulating some kind of symmetry; namely, invariance
under multiplicative scaling of n.
We anticipate that a change of n by some factor λ preserves the shape
of p(n, r), but not its specific value. That is, p(λn, r) should differ from
p(n, r) by a mere change of scale on the r axis and the p axis. We express
the r scale factor as μ(λ) and the p scale factor as η(λ). Then our anticipated
λ variation takes the form p(n, r) = η(λ)p(λn, μ(λ)r).
We may use Eq. (3.6) to decide whether p(n, r) indeed has an asymptotic
behavior of this form. Substituting the proposed form,
dηp(λn, μr) 1
= r 2 1 ∇ 2 ηp(λn, μr) + const. “∇ 4 ”ηp(λn, μr) + · · · .
dn 6
(3.7)
To see the effect of λ and μ on the equations, we define the rescaled λn as
ñ, and the rescaled μr as r̃. Similarly, ηp ≡ p̃. We may readily express
Eq. (3.7) in terms of these renormalized variables. Evidently, d/dn =
λd/d ñ and ∇r2 = μ2 ∇r̃2 . Likewise, any fourth-order derivative satisfies
∇r4 = μ4 ∇r̃4 . In these new co-ordinates,

d p̃ 1 μ2 μ4 4
= r 2 1 ∇r̃2 p̃(ñ, r̃) + const. “∇r̃ ”p̃ + · · · . (3.8)
d ñ 6 λ λ
We may now freely take n to be arbitrarily large; we simply select a
compensating scale factor λ so that ñ remains finite. Evidently, λ → 0
in this limit. In order for Eq. (3.8) to remain finite in this limit, μ must
evidently change with λ. If the ∇ 2 term is to remain finite, we must have
† Technically, the n derivative on the left
μ2 /λ = constant. With this choice, the ∇ 4 term has a factor μ4 /λ = λ.
side of our equation is also an approxima- The ∇ 4 term is arbitrarily small, so that it can be neglected. Such quantities
tion. Like the spatial derivatives on the right
that become arbitrarily small when an asymptotic limit is taken are termed
side, it should be replaced by a full Taylor
expansion. But the higher terms not shown irrelevant variables. We may clearly extend this reasoning to show that all
here may be readily shown to be irrelevant higher orders in ∇ are irrelevant in this same way† . Likewise, just as the ∇ 2
like the higher spatial derivatives. term involved a moment r 2 1 , one can readily show that the ∇ 4 terms have
factors of the form r 4 1 . By our reasoning these moments of p0 (r) are also
†† One may ask why we chose μ(λ) to make irrelevant for the asymptotic p(n, r). Indeed, the only feature of p0 (r) that is
the first term in our equation finite. By relevant is r 2 1 †† . By similar reasoning we
 may determine the scale factor
choosing μ ∝ λ1/4 we can make the second η for p. A simple method is to note that d 3 rp(n, r) = 1: the probability
term finite instead. But this choice does not that the end monomer is somewhere is a certainty. Inserting the scaling form
 
lead to a consistent equation. Now the first
and noting that d 3 r = μ−3 d 3 r̃, we infer η = μ3 = λ3/2 .
term has a factor λ−1/2 . Instead of being
irrelevant, the first term diverges, domin- The requirement that n and r must scale together means that the asymp-
ating the one we have rendered finite. We totic p(n, r) is really only a function of one variable. For any given n
must thus reject this choice of μ. and r, we are free to set λ = 1/n. Then recalling that μ ∝ λ1/2 ,
 Random-walk polymer 51

p(n, r) = n−3/2 p(ñ, r̃)|ñ=1,r̃=rn−1/2 † . By use of the scaling symmetry, † Here p(1, r̃) means the asymptotic func-
we have reduced our unknown function of two variables, p(n, r), to an tion p evaluated for ñ = 1. It does not mean
the monomeric p0 (r) from which p(n, r)
unknown function f of a single variable:
was constructed.

p(n, r) = n−3/2 f (rn−1/2 ).

For any r in the n step polymer, the corresponding r in the λn step polymer is
λ1/2 times as big. From this scaling law, we can readily determine how any
average r characterizing our polymer must scale with n. The most common
average to consider is the second moment r 2 ≡ d 3 rr 2 p(n, r). Using
the scaling law we can treat the arbitrary p’th moment:

r n = d 3 rn−3/2 r p f (rn−1/2 ),
p


= d 3 (rn−1/2 )[(rn−1/2 )p np/2 ]f (rn−1/2 ),

= np/2 d 3 r̂ r̂ p f (r̂), (3.9)

where the dummy variable r̂ stands for rn−1/2 . Since the integral is inde-
pendent of n, we infer that any measure of the polymer size of the form
[r p ]1/p scales the same with n:
[r p n ]1/p = Kp n1/2 . (3.10)
The constants Kp are numerical factors related to the r̂ integral in Eq. (3.9).
This same type of scaling appears whenever we do mathematics with
dimensioned quantities like mass, length, and time. Here we know that the
number representing, say, the time in our formulas has no significance in
itself. The formulas must give the same physical result whether we use one
unit of time or some other unit λ times smaller. Thus the number repres-
enting time can be multiplied by an arbitrary factor λ without changing the
physical result of the formula. This is the familiar requirement of dimen-
sional consistency: the answer must be unchanged when all quantities are
scaled in accordance with their time dimensions by an arbitrary factor λ.
Mathematically, this is a requirement of homogeneity. If we express our
formula in the form f (t, a, b, c, . . .) = 0, where t is the time variable, then
dimensional analysis amounts to the following invariance property: for each
physical quantity a, b, c, etc. in our formula there are exponents α, β, γ ,
etc. such that
f (λt, λα a, λβ b, λγ c, . . .) = 0. (3.11)
The exponents, usually integers or simple fractions, are the time dimensions
of the quantities a, b, c, etc. Though n in our polymer has no physical dimen-
sions, it has something in common with a dimensioned variable: namely,
the invariance of the results under changing the scale factor λ. The resulting
symmetry is the same homogeneity familiar in dimensional analysis: thus
asymptotic quantities like n behave as though they had dimensions. This
will prove true in the more subtle forms of scaling to be treated later.
52 Polymer molecules

Returning to our polymer, the asymptotic equation now has a simple and
familiar form. It is the heat equation, the equation that describes how a
† The polymer scaling is familiar in the con- diffusing substance spreads out with time† .
text of diffusion. It is the counterpart of the
fact that the distance a diffusing substance dp/dn = 1/6r 2 1 ∇ 2 p(n, r), (3.12)
spreads is proportional to the square root of
the spreading time. Here we have set λ = μ = 1 now that they have served their purpose. One
may readily verify that the solution is a Gaussian:
 
2 −3/2 3 2 2
p(n, r) = [2π nr 1 /3] exp − r /(nr 1 ) . (3.13)
2

U 3.3. Toy renormalization In the text we deduced a way of inferring polymer beha-
vior by analyzing the scaling properties of the equation without solving it.
This problem is meant to show the same procedures in the familiar context
of classical mechanics. Here we wish to find the limiting properties of a sys-
x tem’s motion as the energy of a particle goes to zero. The potential energy
0 function U (x)
 of a point particle has the form of two semicircles side by side:
U (x) = U0 1 − (1 − |x|)2 . A particle with energy E  U0 is trapped in the
region near x = 0. It is obliged to oscillate with some period τ . The object
is to find how the period τ varies as the energy E → 0. (a) Write conserva-
tion of energy √for the particle in the region x  1. Notice that in this region
U (x) U0 2|x|(1 − 14 |x|). (b) In order to find the limiting behavior as
E → 0, define Ẽ = λE so that Ẽ will stay finite when E → 0 when expressed
in terms of the ˜ variables. We wish to find a corresponding time scale t˜ = μt
and distance scale x̃ = ηx such that the equation of motion will stay finite
when λ → ∞. (c) Restate the conservation of energy from (a) in terms of the
˜ variables. (d) How must μ and η vary with λ in order for the terms in the
equation to stay finite? (e) How do the maximum displacement in x and the
temporal period τ vary with E as E → 0? ( f) If μ and η vary with λ as found
in (b), does the correction in 14 |x| become small or become large?
This function describes the end-to-end distance of long polymers.
Remarkably, the microscopic structure of the chain, given by p0 (r), has
little impact on p(n, r); the only relevant feature of this p0 (r) was its
mean-square average r 2 1 . Since virtually any p0 (r) leads to the same
asymptotic p(n, r), we say that the Gaussian p(n, r) is universal.
It is worth noting that this Gaussian is not the asymptotic p(n, r) in every
sense. To see this, we might consider our hydrocarbon polymer for very
large extensions r. The polymer has a maximal extension where the back-
bone bonds form a coplanar, zigzag line. Beyond this distance p(n, r) is
strictly zero, so that the Gaussian form is qualitatively wrong for such val-
ues of r. This rmax  is evidently proportional to n, so that it scales entirely
differently from r 2 n . This illustrates the point that asymptotic scal-
ing behavior depends on the question asked. In determining the Gaussian
scaling, we asked for the behavior of p(n, r) for “typical” r values with
appreciable probability. The rmax is extremely atypical; the probability that
r is any finite fraction of rmax is exponentially small in n, as one may
verify using Eq. (3.13). For our purposes in this book, we shall be inter-
ested in typical polymer configurations, so that the Gaussian scaling is the
right regime to consider. But there are some questions of physical import-
ance where the Gaussian treatment is not sufficient. The exercise on the
coil-stretch transition gives an example.
 Random-walk polymer 53

3.4. Short lattice chains A four-step polymer in one dimension consists of left-
and right- directed bonds of unit length. (a) How many configurations of the
chain have the end four units to the right of the beginning? How many have
the end three steps to the right? Two? One? Zero? You may find it easiest to
draw a picture of all 24 configurations. (b) Find r 4 /r 2 2 for this polymer
and compare with 3, the corresponding ratio for a one-dimensional Gaussian
chain.
Knowing the probability p(n, r) gives us information about the energy
needed to distort the polymer. We recall that any system with a given relative
probability f (r) had a free energy given by −T log f . We saw that this free
energy is equivalent to a potential energy: if r is changed from one definite
value to another the work required is the change in this free energy. For our
polymer, the free energy

3 T r2
U (r) = −T log(p(n, r)) = . (3.14)
2 r 2

The polymer stores energy like an ideal spring with zero unstretched length.
As with the spring, the force required to elongate is proportional to the
distance. In magnitude, the energy U is of order T for typical extensions
r r 2 1/2 . This spring energy appears in many forms in polymer liquids. It
is responsible for the rod-climbing behavior of Fig. 1.6. It is also responsible
for the microphase separation patterns seen in block copolymers, like that
of Fig. 1.1.
3.5. Coil-Stretch Transition An elongational flow field may be made by aiming
two round nozzles at each other underwater, then sucking water out through
both nozzles at the same rate. Any polymers at the center of this contraption
are pulled towards both nozzles at once. The effect is roughly equivalent to a
force pulling on each end, of magnitude bz2 , where z is the distance from the
center along the nozzle axis. The coefficient b is proportional to the elongation
rate γ̇ . One can detect the amount of elongation z induced in the polymers by
the flow. (a) What energy W (z) would a free particle starting at the origin gain
if it moved to z under the action of this force? This energy is part of the energy
of any polymer ending at z in the presence of the flow. (b) Denote the z2 of
the polymer without flow as Z 2 . Treating the polymer as a spring, as explained
in this section, sketch the work U (z) required to elongate the chain to length
z as a function of z for small b, counting both the elastic free energy of the
polymer and energy in (a), and write its functional form. (c) This energy U (z)
attains a maximum for some elongation z∗ and energy U ∗ (relative to its z = 0
value). How does the height of this maximum vary with b, when b is small? If
the height U ∗ is adjusted to be the thermal energy T , what is the corresponding
position z∗ of this maximum in terms of Z? (We have seen that this maximum
amounts to an activation barrier. When it is higher than about T , passage over
the barrier becomes very slow.) (d) If this polymer were in thermal equilibrium
at temperature T in the presence of this energy U (z), find z to lowest order
in b. In this case U ∗  T and |W (z)|  T for all z’s that matter. According to
this U (z), once z exceeds z∗ , it will increase to infinity. In reality z increases
only to the maximum extension rmax of the polymer. This maximum extension
is not accounted for in the Gaussian p(n, r), as the text points out.
3.6. Series and parallel polymers Two identical random-walk polymers are
fastened together end to end, to make a chain of double the  length of each.
(a) What is the root-mean-squared end-to-end distance r 2 of the two
free ends relative to that of a single polymer? (b) The same two polymers
54 Polymer molecules

are fastened together at both ends to make a double chain. What is the
root-mean-square end-to-end distance between the two ends in this parallel
configuration? Hint: One may construct the ensemble of double chains by
letting the two chains end anywhere, and then discarding all configurations
except those in which the two chains end at the same point. By this proced-
ure, one can relate the p(n, r) of the double chain to that of the two single
chains.

3.3 Interior structure

For our future work we need to know much more about the polymer than its
end-to-end distance. We need to characterize the whole distribution of the
monomers in space. The most fundamental quantity is the average density ρ̄
within the chain. If R is the radius of a sphere containing on average say half
of the chain, then the average density is roughly n/R 3 . We expect that R is
1/2
of the same order as a typical end-to-end distance such as r 2 n . Using this
estimate, we see that ρ̄ ∼ n/n3/2 ∼ n−1/2 ∼ R −1 . Remarkably the aver-
age density becomes smaller and smaller as the chain length increases: the
polymer is tenuous. A major issue in the pages to follow is whether such
† The pervaded volume of an irregular a tenuous object can have appreciable effects on the pervaded volume† .
object means the spherical region enclos- We will be concerned with effects of solvent flow and of interactions with
ing that object, or the volume of that region,
other polymers. We shall see that these tenuous objects have remarkably
as noted in Chapter 1.
strong effects. There is clearly room within the pervaded volume for arbit-
rarily many polymers, provided each is long enough. Thus we expect these
polymers to be able to interpenetrate and entangle strongly. Another issue
we face is to characterize this entangled state spatially, energetically, and
dynamically.
We may begin to see how the chain comes to have the average density ρ̄
by examining the local density around a given point on the chain. For this
purpose we define ρ(r) as the monomer number density at r in a given con-
figuration. For polymers passing through the origin, there is some average
local density ρ(r) 0 at distance r. The subscript 0 stands for the restric-
ted average in which only chains passing through the origin are included.
We expect that this local density falls off as the distance from the origin
increases. Using the asymptotic behavior of p(n, r) we may readily find
this dependence.
If a monomer passes through r, it must be some distance i along the chain
from the monomer at the origin. The probability that this monomer is at r
is p(i, r). The average density at r is justthe sum of these probabilities for
the possible monomers i: ρ(r) 0 = 2 i p(i, r). The factor 2 accounts
for monomers i steps ahead or i steps behind the origin monomer. The
r dependence may be readily found by using the scaling property of p:
p(i, r) = i −3/2 f (r 2 /i) This density is finite even in an infinite chain. For
large i and r where the asymptotic behavior is reached, we may replace the
sum on i by an integral:

 ∞  
−3/2 r2
ρ(r) 0 = 2 di i f . (3.15)
0 i
 Interior structure 55

Defining a scaled variable ĩ as i/r 2 , we find

 ∞
2 −3
ρ(r) 0 = 2r r d ĩ ĩ −3/2 f (ĩ −1 ). (3.16)
0

The integral is finite: at small ĩ, the integrand is exponentially small because
f is small for large argument; at large ĩ f is a constant, but the integrand
still falls off as an integrable power of ĩ. The integral is simply a constant
independent of r, so that ρ(r) 0 = (const)/r.
This power-law dependence of the local density means that our random-
walk polymer is a fractal set† in the sense of Mandelbrot [6]. That is, †That is, the polymer approaches the fractal
the average mass (number of monomers) M(r) within distance r of an behavior as one approaches asymptotic con-
ditions of large chains and long distances.
arbitrary point of the set varies as some power r D . Then this power is
called the fractal dimension. This M(r) can be found for the polymer by
integrating the local density: M(r) = r  <r d 3 r  ρ(r  ) 0 = (const)r 2 .
A random-walk polymer has a fractal dimension of two, like a uniform
surface. Clearly, this property generalizes: any fractal with dimension D
has a local density ρ(R) 0 that falls off as R D−d in any spatial dimension
d > D. This is discussed in Problem 3.7.
3.7. Why the fractal dimension D is called a dimension It was said in this section
that in a fractal object, the number of points of the object within a given distance
R of a given point grows as R D . Show that this property holds for a straight
line or a plane in three-dimensional space. (a) What is the fractal dimension D
for these two objects? (b) What happens to the two D values when the line and
the plane are embedded in four dimensions?

A fractal set of points necessarily contains large empty spaces. If the M

particles of a fractal of radius R were uniformly dispersed in that volume,
the result would be a cloud of particles of volume per particle R d /M. The
distance from an arbitrary point to the nearest particle is roughly the volume-
per-particle to the 1/d, i.e., R/M 1/d ∼ R 1−D/d . But a fractal arrangement
of points is far from uniform and the average distance from an arbitrary
point to the fractal is much larger than this distance. Indeed, the average
distance is of order R. This openness of fractal structures will prove very
important in how they influence their environment.
This openness is an obvious property of the simplest fractal, a straight line
segment. If we enclose the line in a sphere, and then choose a point within
the sphere at random, the distance to the line is typically a sizeable fraction
of the radius R. (The average distance is (3π/16)R in three dimensions.)
To see that this is true for general fractals, we consider the volume within
distance r of a fractal with mass M. For this purpose we enclose all the
points of the fractal within spheres of radius r. Each sphere contains on
average M(r) points. To enclose all the points requires roughly M/M(r)
spheres. The volume V (r) enclosed by the spheres is of order r d (M/M(r)).
Since M/M(r) = M(R)/M(r) = (R/r)D , we conclude that V (r)
R D r d−D . The volume containing all the points lying within a distance r
from the fractal is larger than V (r) by only a finite factor. Thus the volume
of the region that is a distance between r and r + dr from the fractal is
roughly dr(dV /dr). Using this fact, we can express the average distance
56 Polymer molecules

r between some arbitrary point and the fractal as

R
dr r dV /dr R d−D+1
r a R R.
drdV /dr R d−D
a

This is what we wanted to show. Fractal structure implies tenuousness, but

tenuousness does not imply fractal structure. Our cloud is as tenuous as our
fractal, yet the cloud has qualitatively smaller open spaces.
The fractal property of polymers is the sign of a deeper and more powerful
property called dilation symmetry. Appendix A of this chapter discusses
this symmetry. Further geometrical implications of fractals in solution are
discussed in [7].

3.3.1 Scattering
The behavior of the local density ρ(r) 0 is directly observable by
† The subject of scattering from condensed scattering† . We saw in Chapter 1 that scattering senses the size of an object:
matter is discussed in detail in e.g. Stephen small objects diffract the incoming waves like a slit diffracts light. Scat-
W. Lovesey Theory of neutron scattering
tering senses not only the overall size of a structure but also the internal
from condensed matter (Oxford, Clarendon
Press, 1984). The book by Jannink and Des distribution of matter within it, as we now show.
Cloizeaux mentioned at the beginning of In a scattering experiment a plane wave or beam of scattering particles
this chapter also has an extensive discussion (photons, neutrons, etc.) impinges on an unknown object, such as liquid con-
of neutron and light scattering. taining polymers. For definiteness we will imagine that our beam particles
are neutrons: the waves are then complex scalar Schrödinger waves. The
beam wave has the form A exp(i k · r − iωt). The constant A is the amp-
litude of the wave; its square is the number of neutrons per unit volume in
the beam. The angular frequency ω is proportional to the kinetic energy of
the neutrons. The wave vector k is proportional to their momentum. Evid-
ently, one can change k and ω together by varying the speed of the neutron
beam. But one cannot vary k and ω independently.
When this wave encounters an atomic nucleus, such as a carbon nucleus in
our solvent liquid, the great majority of it continues unperturbed. But a tiny
fraction radiates outward from the nucleus in a spherical wave of the form
B exp(i |k| r)/r. Of course doubling the number of neutrons in the beam
doubles the number scattered in this spherical wave. Thus B is proportional
to A; the proportionality factor is the (complex) scattering amplitude f . A
detector receives the wave far away in some particular direction (Fig. 3.6).
The part of this spherical wave entering the detector is a plane wave with
some wave vector k of magnitude k. The detector measures the intensity of
the wave as a function of the direction of k . The r −1 factor in the scattered
wave is virtually the same distance for all the scatterers, since the detector
is far away from the sample.
The detector senses the scattered waves from all the M scatterers j
in a given volume. The amplitude from a particular scatterer at rj is
B/r exp(k ·(r −rj )). As we have noted, the prefactor B must be proportional
to the wave function of the incident wave: B = f A exp(i k · rj − iωt). The
wave entering the detector is the sum of these contributions. The intensity I
is proportional to the absolute square of this wave. Since the scatterers in our
 Interior structure 57

k⬘

Fig. 3.6
Schematic view of neutron scattering from
a solution containing a polymer. The
detector is shown and the wavevectors k
and k are indicated.

liquid are generally in motion, the intensity I fluctuates in time. To obtain

a time-independent characterization of the system, we consider the aver-
age intensity. This intensity grows with the amount of sample; to remove
this uninteresting size dependence, we consider the average intensity per
scatterer I /M ∝ |ψ|2 /M.
 2 

I   
2 1 
exp(i(k − k) · rj ) | exp(i k · r − iωt)|2 /r 2 .
  † Evidently q may be changed simply by
∝ |A f | 
M  varying the angle θ between the incident
M
j 
beam and the scattered rays that are detec-
(3.17) ted. We are considering a case where the
The last factor, with the r and t dependence, is an unimportant constant, as scattering particles lose no energy in the
is the prefactor giving the dependence on beam intensity and intrinsic scat- (elastic) scattering process, so that |k  | =
|k|. Then some simple geometry shows that
tering power of the nuclei. The rest gives information about the structure. |q| = 2|k| sin(θ/2).
Evidently it depends on k and k only through the difference k − k ≡ q† .
This remaining factor is called the structure factor S(q)†† . We may †† Sometimes it is necessary to consider the
simplify it by expressing the squared sum as an explicit product of two q dependence of the scattering amplitude
sums: f from each elementary scatterer. This q
dependence reflects the distribution of mat-
1  ter within an elementary scatterer and is
S(q) = exp(i q · rj ) exp(−i q · rk ) , called the form factor. In some discussions
M
j ,k of scattering, the elementary scatterers are
1 
taken to be an entire polyatomic structure
= exp(i q · (rj − rk )) . (3.18) such as a polymer. In that case, much of
M the information in the S(q) above would
j ,k appear in the form factor. The separation
of the intensity of Eq. (3.17) into form
One important property of the scattering intensity is immediately appar- and structure factors is evidently somewhat
ent from this expression. The spatial information in S(q) is strongly arbitrary. In our discussion we include all
blurred. The degree of blurring can be expressed in terms of the scattering spatial information in the structure factor.
58 Polymer molecules

wavelength λ defined as 2π/q. To see this, we imagine that the scatterers

i were moved by arbitrary small displacements uj . The effect would be to
add a new factor exp(i q · ( uj − uk )) in Eq. (3.18). But if all the uj were
much smaller than the scattering wavelength λ, these new factors would all
be nearly unity, and S(q) would hardly change. That is, S(q) cannot detect
anything about the positions of the scatterers to a precision of much less
than a wavelength λ. We could for example gather all the scatterers lying
within a tenth of a wavelength of one another into bunches and the effect on
S(q) would be minor. Conversely, it is immaterial whether the scattering
occurs from discrete scatterers at specific points rj or from a smeared distri-
bution of scattering spread over a small regions around each rj —provided
the region is much smaller than 2π/q. Scattering is myopic within a scale
2π/q. If 2π/q is larger than the entire collection of scatterers, all the phase
factors in Eq. (3.18) are unity and S(q) → M.
We may use our density language to simplify the expression in Eq. (3.18).
The density ρ(r) is the probability per unit volume that one of the scatterers
is at r. If the scatterers areat positions {rj } the density is  given by ρ(r ) =
 3
j δ (r − rj ). Inserting d r1 δ (r1 − rj ) (= 1) and d r2 δ (r2 − rk ),
3 3 3 3
we may rearrange Eq. (3.18) to read
 
1
S(q) = d r1 d 3 r2 ρ(r1 )ρ(r2 ) exp[i q · (r1 − r2 )].
3
(3.19)
M

The absolute location of r is clearly immaterial in the ρρ . Both points

may be shifted by r2 so that ρ(r1 )ρ(r2 ) = ρ(r1 − r2 )ρ(0) . Now we
note that the integrand depends only on r1 − r2 , which we denote by r:
 3 
d r1
S(q) = d 3 rρ(r )ρ(0) exp[i q · r]. (3.20)
M

The quantity ρρ is called the density correlation function. This ρρ is
simply related to the local density defined in the previous section. We may
think of it as the joint probability that a monomer is at the origin and that
another is at r. This is the overall probability that a monomer is at the origin
times the conditional probability that a second monomer is at r given that
one is at 0. The probability that a monomer
 is at the origin (or any other
point in the sample) is ρ = M/( d 3 r). The conditional probability is
† The density correlation function also con- the same as the local density ρ(r) 0 . Thus ρ(r )ρ(0) = ρ ρ(r) 0 † . The
tains the same information as the pair correl- relation between S(q) and the local density is simple:
ation function g(r) introduced in Chapter 2.
We recall that g(r) means the average dens- 
ity of particles at displacement r from S(q) = d 3 rρ(r) 0 exp(i q · r). (3.21)
a particle, relative to the average dens-
ity ρ . Thus the connection is g(r) =
ρ(r) 0 /ρ = ρ(r)ρ(0) /ρ 2 . Of the The reduced scattering intensity S(q) is related to the local density by a
three quantities, the local density is most simple integral transform—the Fourier transform. This means that if one
convenient for discussing fractal structure,
expresses the scattering density ρ(r ) as a sum of plane waves of all different
because it avoids unimportant factors of
average density. wavevectors, the scattering intensity and S(q) arise from those waves that
have wavevector q. If the local density is constant the S(q) is simply zero
for q > 0.
 Interior structure 59

In principle each nucleus in the fluid sample produces scattering. But in

practice the scattering from a pure fluid is very weak for wavevectors q of
interest here. We argued in the last chapter that structure at a given spatial
scale R causes scattering at wave vectors q of order 1/R. Thus we shall be
interested in very small wave vectors relative to the size of an atom. We have
seen above that the scattering from these atoms would not be appreciably
changed if they were replaced by smeared-out densities several atoms in
width (since such smearing is much less than 1/q). The result in a uniform
fluid is an essentially constant density. We have also seen that a perfectly
constant density causes no scattering. It is thus not surprising that a simple
fluid causes little scattering† . † At these small wave vectors the scatter-
Any departure from this uniform density of scatterers gives rise to scatter- ing from the fluid is independent of q and
can readily be subtracted. As discussed in
ing intensity. The polymers in the fluid do this: their atoms have scattering
Problem 4.2, the amount of scattering can be
amplitudes f that differ from those of the solvent, so that there is a contrast shown to be proportional to the compress-
between the scattering per unit volume from regions that contain monomers ibility of the fluid relative to an ideal gas
and regions that do not. Thus, we may treat the scattering as though only at the same density. All familiar liquids are
the polymers were present. Whenever a liquid contains independent large relatively incompressible; this leads to weak
scattering.
objects of typical size R, the behavior of S(q) depends on the size of q rel-
ative to the R. Evidently, the local density falls to zero for distances r  R.
When qR  1 then exp(iq · r) → 1 for all r < R. Thus,
 R
S(q) → d 3 rρ(r) 0 1 → M. (3.22)

That is, the internal structure is completely invisible to S(q) and the scat-
tering is about the same as though all the scatterers were concentrated at
a point. Our polymers show this behavior like any other large object. It is
straightforward to see how S(q) begins to vary when q grows comparable
to 1/R; this is discussed in an exercise.
In the complementary limit where qR  1 the S(q) senses the internal
structure of the object. If this object is a fractal made of units of size a,
as our polymer is, we may readily infer how it scatters in this limit. As
discussed above, a fractal of dimension D in three-dimensional space has a
local density of the form ρ(r) 0 = c r D−3 for a  r  R. For such power
laws it is easy to perform the integral in Eq. (3.21) by defining r̃ ≡ |q|r .
This yields
 |q|R
−D
S(q) = c|q| d 3 r̃ r̃ D−3 exp(i q · r̃/|q|). (3.23)

The integral in r̃ is independent of |q|. It cannot depend on the direction of

q since ρ(r) 0 is independent of direction. Less obviously, it is finite for
|q|R → ∞ as long as D − 3 < 0 (it can be expressed in terms of Euler
 functions involving D). Thus the integral is a numerical constant, and
all the q dependence comes from the power in front. The same is true for
any spatial dimension d > D. We recall that the average number of points
M(r) within distance r of a fractal grows as r D . Thus in scattering from
a fractal object S(q) ∼ M(λ), where λ is the scattering wavelength 2π/q.
60 Polymer molecules

The scattering is the same as though the fractal were chopped up into pieces
the size of a scattering wavelength.
1.0 Figure 3.7 summarizes the expected scattering from a fractal. As q
S (q) increases from 1/R to the inverse scatterer size 1/a, the scattering intensity
M drops by a factor of M, the number of scatterers in the fractal. Remarkably,
the S(q) from a random-walk polymer can be calculated explicitly as a
0.1 –D
check on these ideas.
3.8. Scattering from a random-walk polymer It was shown in the text that
the  structure factor S(q) of a set of n scatterers at positions rj is
0.01 1/n nj,k=1 exp[iq · (rj − rk )] . Recall that the probability distribution p(i, r)
0.1 1.0 10.0
q RG
for a segment of length i to have its ends at displacement r can be expressed
in the product form with three factors for the x, y, and z directions:
p(i, r) = A(i) exp[−x 2 /(2ix 2 1)] × [x → y] × [x → z], where the pref-
Fig. 3.7 actor A assures normalization: r p(i, r) = 1. (a) From this and the fact
Schematic log–log plot of S(q) versus q

that dx exp(−x 2 ) exp(iqx) = π 1/2 exp(−q 2 /4), derive an expression for
for a fractal set of M  1 scatterers with S(q) in closed form. Note that the number of monomer pairs separated by
fractal dimension D. (In fact, this is the a distance m monomers is not the same for all m. There is only one such
exact structure function for a random-walk pair for m = n, but there are n − 1 pairs when m = 1. If one has a suf-
polymer, as requested in Problem 3.7. The ficient range of q accessible experimentally, one may discern two scaling
wave vector q is given in the combination
regimes in this S(q): for q 2 x 2 n  1, S(q) ∼ q 0 ; for q 2 x 2 n  1,
qRG , where RG is the radius of gyration
discussed in Problem 3.11.) S(q) ∼ q −2 . To get a feeling of what range of q is needed, plot log(S(q))
1/2 1/2
versus log(q) for (b) 0.3 < qx 2 n < 3, (c) 0.03 < qx 2 n < 30,
1/2
(d) 0.001 < qx n < 1000. The scales should be adjusted so that the three
2
plots are of roughly equal size. What range of q would be needed in your judg-
ment to discriminate between the expected q −2 scaling and a hypothetical q −1.8
scaling?
3.9. Scattering from a line The text shows that the scattering intensity at large
wave vector q from a fractal object was proportional to k −D . (a) Find explicitly
the structure factor S(q) from a line of length 2L oriented along the z-axis
between −L and L. The scattering wave vector q makes an arbitrary angle θ
with the line. (b) For a line at random orientation, show that the scattering is
proportional to q −1 for large q, as it should be for a one-dimensional fractal.
Note
 1 that in three dimensions an angular average of a function f (θ) amounts
to 0 d(cos θ )f (θ )/(4π ).
3.10. Scattering from a Gaussian Cloud Since the monomers in a polymer chain
are randomly arranged throughout a volume of limited size R, it is sugges-
tive to approximate the polymer as a mere cloud of n monomeric scatterers
confined to a region of this size. A simple realization would be a Gaussian
cloud, whose local density ρ(r) is given by nCR −3 exp(−(r/R)2 ). How
does the scattering intensity from such a cloud vary with wave vector
q? Does it have the proper fractal behavior at large q? A useful integral
∞
is −∞ dx exp(iqx − x 2 ) = π 1/2 exp(−q 2 /4).
3.11. Radius of gyration The structure factor S(q) of any object composed of
n scatterers may be expanded in powers of the wave vector q: S(q) =
n(1 + aq + bq 2 + · · · ). (a) What are the coefficients a and b, expressed in
terms of the positions rj ? Assume that the scatterers are isotropically arranged
so that the scattering is the same in all directions. Use the power series expan-
sion for ex : ex = 1 + x + 12 x 2 + · · · . (b) The radius of gyration of any set
of n particles is defined as 
the root-mean-square distance between an arbitrary
pair of them: RG 2 ≡ n−2 1 rj − rk )2 . Find an expression for RG in terms
2 j k (
of the scattering coefficients a and b. (c) For a random-walk polymer, how is
the radius of gyration related to the mean-squared end-to-end distance r 2 n ?
 Self-avoidance and self-interaction 61

(d) How is RG related

 to the root-mean-square distance of a particle from the
center of mass, n−1 j rj2 ?

3.4 Self-avoidance and self-interaction

Our treatment of polymer configurations to this point has ignored the liquid
environment of the molecules. A polymer coil must nestle intimately among
the small molecules of its solvent. Likewise, the polymer chain must not
intersect itself. The valid, self-avoiding polymers we want to study are
of course a subset of the random-walk polymers of the last section. Our
interest is to learn how much the properties of this subset differ from those
of the full set. We shall focus on the spatial distribution of monomers,
as exemplified by the mean squared end-to-end distance r 2 . The sub-
set of self-avoiding polymers is obtained from the full set of random-walk
polymers by examining each member of the full set and discarding each con-
figuration that has self-intersections. (This is a rather crude simplification of
how equilibrium polymer configurations occur in practice. Certainly chains
having two carbon atoms within 0.05 nm of each other are not present. But
other closely self-approaching configurations may be partly excluded or
even enhanced, depending on the details of the monomer and solvent inter-
actions. We shall return to such refinements later. For the moment, we shall
regard the self-avoidance as an all-or-nothing proposition. For example, we
might exclude all configurations in which two atoms are separated by less
than 0.1 nm.)
This discarding process can of course affect the end-to-end probability
p(n, r). The discarding must be quite dramatic in order to have a qualitative
effect on the size of the chains remaining. Suppose we remove all but a
fraction t(r) of the chains ending at r. If t is a constant, the average size
of the remaining chains is of course affected not at all: t must depend on r.
If the self-avoiding chains are to be much longer than the random walks,
the fraction remaining must decrease strongly as r decreases. We might
imagine, for example, that the t fraction goes as a positive power of r:
t(r) ∼ r a . Then,
 
r 2 SA ∼ d 3 rt(r)p(n, r)r 2 ∼ d 3 rr a+2 p(n, r). (3.24)

Since we know the gaussian form of p(n, r), this integral can be worked
out explicitly (along with the needed normalization). Since p(n, r) falls
off faster than any power for large r, all moments r a 1/a are of order
1/2
r 2 RW . (The reasoning is the same as for the P (M) distribution discussed
in Appendix A.) Thus any power-law dependence of the remaining frac-
tion t would leave the average size of the self-avoiding chains basically
unaffected. They would increase by only a finite factor. Conversely, if
the self-avoiding chains are to be arbitrarily larger than the random-walk
chains, t must vary more strongly than a power law. Clearly, the overall
fraction remaining from the initial random walks must be indefinitely small.
62 Polymer molecules

Our aim in this section is to understand what types of contacts are important
and how much they alter the shape of the polymer.

3.4.1 Local and global avoidance

It is instructive to look at several different types of contacts in turn, start-
ing with local contacts. By local contacts I mean contacts between nearby
monomers along the chain, within some fixed distance i. It is not hard to
deduce the asymptotic effect of such local contacts. We imagine a random-
walk chain in which only these local contacts are excluded. The asymptotic
p(n, r) for such chains can be found for any exclusion range i. To show
what happens, we may simplify a bit further, and divide our chain into
blocks of length i. We then exclude only contacts within a given block. We
can immediately treat this model using Eq. (3.2). We need only build up
our chain from blocks of length i rather than from single segments. Since
successive blocks are statistically independent, we may again express the
overall probability p(n, r) in the form of Eq. (3.2). Our former reasoning
holds and leads to the conclusion that r 2 ∼ n. We may also estimate how
much the local contacts reduce the number of allowed configurations. The
remaining fraction ti in each block is necessarily finite. For a two-block
chain any surviving configuration from the first block may be combined
with any from the second, so that the surviving fraction is the product of the
fractions from the two blocks. For n monomers or n/i blocks the overall
remaining fraction tn is evidently given by: tn = (ti )n/i . The surviving
fraction is exponentially small in n. The same conclusions hold if instead
of using our self-avoiding blocks we exclude all contacts within a fixed
distance i of one another. Local contacts can change the chain size by only
a finite factor, and thus they cannot change the scaling behavior of chain
size with chain length.
At the other extreme are global contacts. By these we mean contacts
whose distance along the chain is a fixed fraction of the total length n. For
example, we may consider the effect of contacts between the first third of the
chain and the last third. These two mutually avoiding tails act to elongate the
midsection of the chain. To see the potential importance of these contacts,
we consider chains in which the two tails occupy the same pervaded volume:
two spheres drawn around each tail interpenetrate significantly. We now ask
r whether the two tails are likely to have mutual contacts, leading to discarding
x these configurations. Contacts between two fractals such as these will be
important in several contexts. In the general case we have two fractals A
and B of radius R placed randomly in the same volume. A contact means a
point of fractal A that lies within some small sphere of volume v surrounding
some point of fractal B.
We may find the average number of contacts by using the local dens-
ity of points within each. We pick some point from each fractal, and let
one of these be at the origin. The displacement x between the two chosen
Fig. 3.8
points is of order R. The probability per unit volume of a contact at some
Scheme used in Eq. (3.26) for counting
contacts between two fractals in a sphere point r is then ρA (r) 0 vρB (r − x) 0 , as shown in Fig. 3.8. The average
of radius R. The fractals are the first and number of contacts MAB is then the integral of this probability over the
last third of a random-walk polymer. pervaded volume. We recall that the local density ρA (r) 0 ∼ r DA −d , in d
 Self-avoidance and self-interaction 63

dimensions. Thus,

MAB ∼ d d rr DA −d (r − x)DB −d . (3.25)

Again we simplify the integral using a scaled variable r̃ ≡ r/R to give

 1
DA +DB −d
MAB ∼ R d d r̃ r̃ DA −d (r̃ − x/R)DB −d ∼ MA MB /R d .
0
(3.26)
The integral is finite, since the divergence at r̃ = 0 and r̃ = x/R are
integrable. It is thus a harmless constant, which we may safely ignore. Thus
the average number of contacts varies as a power of the size R of the region.
This power may be either positive or negative. For our two random-walk
polymer tails DA = DB = 2, and the power is negative whenever these
walks live in a spatial dimension d larger than four. Then the average number
of contacts between two such polymers placed randomly into the same † The probability of contact between two
volume becomes indefinitely small as the polymers are made large. The mutually opaque fractals behaves in a dis-
probability of contact goes to zero. We say that two such polymers are tinctive way that is different from (a) ran-
mutually transparent in spatial dimensions larger than four. Evidently any domly distributed clouds of particles and
two fractals are mutually transparent when DA + DB < d. One may readily (b) transparent fractals or clouds. If r̃ ≡
x/R remains fixed as R → ∞, the prob-
confirm this result for simple fractals like lines or planes placed at random ability of no contacts goes to a finite limit
in a d-dimensional box. For our two tails in spatial dimensions greater between 0 and 1.
than four, the fraction of configurations thrown away because of contacts This distinctive behavior occurs because
becomes negligible, even when the two tails occupy the same volume. These the points of a fractal are statistically correl-
contacts thus have negligible impact on the behavior of such polymers. ated. For definiteness, we consider r̃ = 12 .
For any given r̃ and R there is a probability
For d < 4 the two tails are mutually opaque; the average number of distribution of number of contacts: Pr̃ (M).
contacts grows indefinitely with R. Thus the likelihood of a contact between The probability of no contacts is Pr̃ (0). The
the two tails becomes significant† . We may think of each tail as a sphere of average M r̃ is dMMPr̃ (M). This M r̃
radius R which repels the other sphere. Clearly this prevents the two ends can be expressed as (1 − Pr̃ (0))M r̃ . The
of the middle third from coming as close together as they would otherwise. new average M r̃ is the average number of
contacts for all configurations where there
Still, for a significant fraction of positions of these two ends, the tails do is at least one contact. This average can be
not overlap and their mutual repulsion does nothing. Thus the tails increase found by choosing e.g., the contact closest
the average end-to-end distance of the midsection by a no more than a finite to the center of the R sphere as the origin
factor. Global contacts are not sufficient to make a qualitative change in the and then integrating ρA (r) 0 ρB (r) 0 . This
chain size. is legitimate because the tails A and B are
completely arbitrary except that they touch
In fact it is necessary to consider all types of contacts, from the most at the origin. The result is just what we mean
local to the most global, to account for the expansion due to self-avoidance. by M r̃ with r̃ = 0. Thus our equation for
To get a feeling of how this comes about, we imagine a simple chain on M r̃ amounts to
a lattice with no self-avoidance. For convenience we take its chain length
to be a power of two: n = 2k . We first divide the chain into n/2 blocks M r̃ = (1 − Pr̃ (0))M 0 .
of length two, and discard configurations with contacts within a block. For r̃ = 0 the Pr̃ (0) must vanish, as the for-
Next, we group pairs of these blocks to make n/4 blocks of length four, mula shows. But as r̃ increases towards 1,
and discard contacts within each of these. Then we do the same for blocks M r̃ decreases by a finite factor, while
of length eight, and so on. Before each stage of this process we have a M 0 does not. Thus, Pr̃ (0) must be finite
random chain of many self-avoiding blocks. We then impose mutual avoid- for r̃ fixed and less than 1.
If the fractal particles are dispersed into
ance between each block and one of its neighbors. Considering these two two uniform clouds, there are no correla-
blocks in isolation, the contacts involved are mostly global ones. As we tions, M r̃ → M r̃ and thus Pr̃ (0) → 0
have seen, these increase the size of the pair of blocks by finite factor b. as R → ∞.
64 Polymer molecules

When all the blocks of a chain swell by a factor b the overall size of the
chain must increase by the same factor, since the chain is still a random
walk of the expanded blocks. We have argued above that the expansion
factor b must be finite irrespective of the number of monomers involved.
Thus the expansion factors for all stages k of our process are essentially the
† The initial blocks, since they involve short same† . Therefore the self-avoiding size RSA expands by a factor of order
segments, may have different expansion bk relative to the random-walk size, RRW . Recalling that RRW ∼ n1/2
factors b from the majority. But the longer
and noting that bk = 2k log2 b = nlog2 b , we conclude that RSA ∼ nν ,
the chain is and the more stages k of subdivi-
sion are present, the less these initial blocks where ν = 12 + log2 b is a power larger than 12 . This argument says that
can matter. self-avoidance alters the scaling relationship between chain length and
size. The exponent ν describing the new scaling is called the Flory swelling
exponent.
In random-walk polymers, the scaling of the overall size R reflected a
pervasive dilation symmetry for the internal mass distribution. The same
thing proves to be true of self-avoiding polymers. First, the end-to-end
distribution function p(n, r) should have a scaling property analogous to
that of the random walk, since it represents a limit of indefinitely large n.
††The radius of gyration grows by a factor The scaling form must give average end-to-end distances proportional to
of about 10%, compared with the same nν , and must preserve normalization. Thus p(n, r) = n−νd f (r/nν ). If we
chain without tails: cf. [8], Chapter XV.4,
consider a finite piece of length n within an infinite self-avoiding polymer,
Table 8.
its mean size is of the same order as that same piece in isolation. That
is, the attachment of an infinite tail to each end of the piece extends it
†††It is natural to ask whether a self- by only a finite factor [8]†† , as with the finite tails discussed above. This
avoiding polymer has the full dilation sym- scaling law can be used to infer the local density ρ(r) 0 , in the same
metry of the random walk, as discussed in way done for the simple random-walk polymer in Eq. (3.16). The result
Appendix A. The answer is yes, though the is ρ(r) 0 = (constant) r 1/ν−3 . Since ρ(r) 0 falls off as a power of r,
proof is too complicated to give here [8]:
M(r) is also a power of r. Repeating our reasoning for the random walk,
ρ(r1 ), . . . , ρ(rk ) 0 we find M(r) = (constant) r 1/ν . Thus the self-avoiding polymer is a
fractal whose fractal dimension D = 1/ν ††† .
= λ−k(D−d) ρ(λr1 ), . . . , ρ(λrk ) 0 .

The case k = 1 gives the local dens-

ity we have just analyzed. Here we may
3.4.2 Estimating D
set λ = 1/r and infer ρ(r) 0 =
r (D−d) ρ(1) 0 . Happily, this agrees with It is possible to calculate D systematically using sophisticated methods of
the result we just derived. The density
renormalized field theory [9]. These methods rely on a deep correspondence
products ρ(r1 ), . . . , ρ(rk ) 0 behave simil-
arly to those of a random-walk polymer. between a polymer chain and a liquid at a second-order phase transition.
We may find the scaling properties of this The books of De Gennes and of Jannink and Des Cloizeaux deal extensively
product by looking at chains that pass with this correspondence. In practice, the systematic calculations require
through the labeled points in a specific a detailed accounting of contacts between segments of random walks, and
order, such as the order listed. If we ignore
they convey little insight about how the exponent D arises.
self-avoidance except within the 0-to-r1
segment, the r1 -to-r2 segment, etc. we can A much simpler argument to estimate D is based on a qualitative
express the density in the product form comparison of free energies involved in expanding a chain. It was first

i1 , ..., ik p(i1 , r1 ), . . . used for the random propounded by the eminent mid-century polymer chemist Paul Flory [10].
walks. Using the scaling form of the p’s, The Flory approach considers the work done on the system in imposing
one finds a scaling property under dila-
the self-avoidance constraint. If this work depends on the size R of the
tion analogous to that for the random walk.
If one now includes the remaining effects chain, that means the probabilities of various sizes R will be altered. For
of avoidance among these segments, the definiteness, we shall take R to be the root mean square end-to-end dis-
densities are reduced, but only by a finite tance: R = (r 2 )1/2 . If a chain is to expand to a size R  RRW ,
factor [8]. work must be done to increase its elastic free energy. One can achieve
 Self-avoidance and self-interaction 65

the needed expansion by pulling on the chain ends. Then the work S
required is the free energy of extension, calculated in the last chapter:
S = 32 T R 2 /RRW
2 = 32 R 2 /(a 2 n), where a is the size of a monomer. Such
expansion is favored because it reduces the number of contacts. To account
for these contacts, we imagine that contacts are allowed, but at a cost of a
small energy U per contact. That is, we replace the mutually self-avoiding
monomers by weakly self-repelling ones. As we have seen, local contacts,
despite their large number, are not important for the chain expansion. The
number of nonlocal contacts may be found by imagining that the chain is
a dispersed cloud of n monomers confined to the region of size R. The
number of contacts is n times the probability that a given monomer has
a contact. This probability is the monomer volume v times the monomer
density n/R d . Combining these facts, the overall energy V due to these
contacts is given by V vU n2 /R d . Asymptotically (if d < 4 and the
chain is opaque to itself) there are indefinitely many contacts and the con-
tact energy grows much larger than the thermal energy T ; this is enough
free energy to account for substantial elastic stretching.

3.12. Simulating a self-avoiding polymer The structure of self-avoiding polymers

is not known analytically. Often the most direct way to get quantitat-
ive information about such polymers is simply to simulate them. Because
the asymptotic properties of these polymers are independent of their
local construction, one can readily simulate a self-avoiding polymer using
simple lattice methods. A simulation is a computational procedure in which
a system is repeatedly subjected to small changes. These changes are devised
so that the probability of a given configuration of the system is the same
as in an equilibrium system. In this problem we develop such a simulation.
The polymer will consist of a sequence of nearest-neighbor steps on a square
lattice. Each monomer k occupies the lattice site (i(k), j (k)), where i and
j are integers. To perform a simulation, one first needs to define an initial
configuration that is a self-avoiding chain. We may for example use the fol-
lowing one: i(k) = [[99k/100]] + 1, j (k) = [[(k − 1)/100]] + 1. Here
[[x]] means the integer part of x. Thus as the first few monomers will be
at (1, 1), (2, 1), (3, 1), . . . , (98, 1), (99, 1), (100, 1), (100, 2), (101, 2), . . . This
chain is stretched in the horizontal direction with a slight negative slope. It
will serve as a convenient starting configuration.
In our self-avoiding chain, any self-avoiding configuration is supposed to
be as likely as any other. (This is the content of the Boltzmann Principle here,
since all the allowed configurations have the same energy (namely, 0) and
thus the same probability.) In our earlier simulation problem we showed, that
any two configurations A and B will be equally likely if the probability to go
from A to B in one step is equal to the probability to go from B to A. We shall
devise a scheme that has this property.
One way to change from one self-avoiding walk to another one is to remove
a monomer from one end and add it to the other. For example, we might take A B
a monomer from the beginning of the chain (k = 1) and add one to the end
(k = n) on the right side. Let us assume that this position is not blocked by
another monomer. Further, let us assume that the removed monomer was just
above its successor. We made our choice of where to place the end monomer by
selecting one of the four possible directions at random. Thus the new monomer
is placed on the right side with probability 14 . That means the probability of
going from A to B is 14 in this case. Now suppose the system is in configura-
tion B. We again choose one of the four step directions at random, remove the
monomer from the end and attach it to the beginning in the chosen direction.
The probability that the upward direction was chosen is 14 . Thus the probability
66 Polymer molecules

to go from B to A is 14 , like the probability of going from A to B. Since this

is true for all allowed moves, then all sampled configurations have equal
probability.
To assure that the two probabilities are really equal, we must be a bit
careful how we proceed when the chosen move is forbidden. We must select
the step direction at random, test whether the chosen step is allowed, make
the step if it is allowed and leave the configuration unaltered if it is not. This
procedure will assure that the probability of any allowed change is the same
as the probability of the reverse change. Whenever a change is made, the list
(i(1), j (1)), . . . , (i(n), j (n)) must be updated accordingly.
One efficient way to test for whether a candidate site is occupied is to make
an image matrix M(i, j ). The image covers a range 1–L by 1–L which is large
enough to accommodate (almost) all the polymer configurations. For every
occupied site (i, j ) in the lattice M(i, j ) = 1. For the other sites M = 0.
Thus if we wish to check whether a site (i, j ) is available, we have only
to test the value of one element of M, namely M(i, j ). We do not need to
check the whole list of monomers to see if one happens to be at (i, j ). The
use of an image matrix does take some extra work. When the configuration
is changed, M(i, j ) must be set to 0 for the monomer that was removed and
M(i, j ) must be set to 1 for the site that was added. (One might hope that if
we know M(i, j ) we can construct the list (i(1), j (1)), . . . , (i(n), j (n)). This
is possible on some special lattices, but not on the square lattice. For example,
if the configuration of a four-monomer chain happens to be a square, there is
no way to know from the occupied sites which monomers are the ends of the
chain.)
The straightforward way to update the list of monomer positions when the
beginning of the list is removed, is simply to shift each item in the list to the
previous position. But this requires n operations for every successful move.
A way to avoid this is to allow the list to start with an arbitrary element.
A separate number K keeps track of which element of the list is the first
monomer. Then the first monomer has position (i(K), j (K)). The second
monomer is at (i(K + 1), j (K + 1)), etc. We calculate this index mod(n), so
that the monomer after (i(n), j (n)), is at (i(1), j (1)), continuing until the n’th
monomer at (i(K − 1), j (K − 1)). Now when the first monomer is removed,
we just change set (i(K), j (K)) to be the new position of the added monomer.
Then we increase K by 1 to indicate that the beginning of the chain is now
what was formerly the second monomer.
(a) Implement this simulation for a 64-step chain on a 100 by 100 lattice.
After every 5000 steps, plot the chain to show that it is a connected chain
and is self-avoiding. Do this for a dozen chains. Store the image matrix
for each of these chains.
(b) Using the image matrices from part (a) calculate M(r), the number of
monomers within a distance r of an arbitrary monomer. This means count-
ing the number of occupied sites within a circle of radius r of every site
and taking the average. You may use a square instead of a circle; it does
not change the scaling properties. Verify that M(r) ∼ r D . Compare the
D you obtained with the 43 calculated by the Flory argument in the text.
It is convenient to work with the average local density M(r)/r 2 , which
should vary as r D−2 . This density shows up the fractal behavior more
clearly than M(r).

3.A Suggested experiment: Simulated polymer properties. Use the simulation

developed in the preceding problem to show that the fractal dimension does not
depend on the degree of self-avoidance. Scale up the simulation to 128 or 256
beads. Try to obtain an accurate value for D by measuring M(r)/r 2 , and correct-
ing for distortions due to monomer-scale and chain-scale effects by extrapolation.
Then modify the simulation so that every other bead is invisible, thus reducing the
 Self-avoidance and self-interaction 67

self-repulsion. (The invisible beads are freely allowed to sit at occupied sites. This
10–1
reduces the average b2 by a factor of 4.) Observe the resulting changes in M(r)/r 2 .
Determine how much the asymptotic D changes as a result. Compare the radius R I –1(q)
10–2 1.67
with the case of full self-avoidance and compare with the prediction of the Flory 2.00
formula.
10–3
The polymer is free to choose its size R. By the above estimates, the A
potential of mean force associated with size R is S + V . We thus expect
10–4 B
that the typical size will be that which minimizes S + V . Since S and V are
both powers of R, the minimum occurs when S V : C
10–5
10–2 2 5
10–1
R 2 /n ∼ U n2 /R d , or R 2+d ∼ U n3 , or n ∼ R (2+d)/3 . (3.27) q Å–1

We are led to the conclusion that for sufficiently large n, the fractal Fig. 3.9
dimension Log–log plot of the inverse scattered
intensity as a function of the scattered wave
D = (2 + d)/3. (3.28) vector q, confirming the internal fractal
structure, after [11, Figure 1], © American
Physical Society, courtesy G. Jannink. The
In three dimensions D = 53 . We note that the D value is independent of the full points are experimental data for a
strength U assumed for the contact interaction: for sufficiently long chains, polymer in a good solvent, curve A, a theta
even an arbitrarily small U still leads to a strong expansion. solvent, curve B, and the bulk polymer,
This prediction for D is consistent with estimates based on systematic curve C. The behavior of curves B and C
will be discussed later in the chapter. The
calculations, experiments and computer simulations. One experimental polymers used were polystyrene and
confirmation is shown in Fig. 3.9. Moreover, in two and in four dimen- deuterated polystyrene with molecular
sions, D is known exactly and it agrees with the Flory prediction. Despite weight about 106 . The solvents were
the success of the Flory argument, it has not proved possible to justify it† . carbon disulfide (curve A), deuterated
Because of the success of the Flory formula, efforts continue to find a simple cyclohexane at 40◦ C (curve B), and
undeuterated polystyrene (curve C). The
way to understand the scaling of self-avoiding polymers [12]. dashed lines are reference straight lines
3.13. A lower bound on D for a self-avoiding polymer According to the Flory with slopes of 2.00 and 53 showing the
argument, the constraint of self-avoidance lowers the fractal dimension D. power-law behavior expected for D = 2
But it must not be too much lower: if D were too low, the expanded polymer and D = 53 . The full lines are calculated
would not be opaque to itself, and the contact energy would be insignificant. curves incorporating the expected scaling.
(a) From this consideration, give a lower limit on D in d dimensions [13]. Is
this bound consistent with the Flory formula for D? †
3.14. Triple contacts In some situations polymers obey a modified type of The Flory reasoning can easily lead one to
self-avoidance: intersections of two monomers (of radius a) are freely wrong conclusions. For example, the argu-
allowed, but intersections involving three monomers are forbidden. (a) For ment appears to be equally justified for a
n monomers in a region of radius R how does the number of triple intersec- chain where only the first and last third repel
tions vary with n and R. Your estimate should be in the same spirit as the each other. Yet we have shown that such
Flory estimate for double contacts cited above. (b) Assuming that each triple avoidance can only expand a chain by a
contact costs a given energy U , apply the Flory reasoning to estimate the factor of order unity.
fractal dimension of the chain in three dimensions, and in two dimensions.

3.4.3 Self-interaction and solvent quality

Up to this point we have treated the self-avoidance constraint in a simple
way. We have simply removed configurations whose atoms came closer
than some threshold distance. To treat this constraint more realistically,
we must investigate how monomers interact in solution. As before, we
shall consider our chain to be composed of segments long enough to be
small polymers in their own right, yet much smaller than the overall chain.
The mutual avoidance between these segments is essentially the same
68 Polymer molecules

whether the segments are attached to each other in sequence to form a

polymer or whether the segments are floating free in solution. We thus
consider two such detached segments of some nominal diameter a and
ask how free they are to approach each other. We select an atom on
each segment to label its position. We then examine the relative prob-
ability g(r) that the two segments are separated by displacement r. We
introduced this g(r) in Chapter 2. There we noted that this g(r) is pro-
portional to the Boltzmann-weighted sum of all solution configurations
with the two segments separated by r. If r  a two segments have no
influence on each other; the configurations of solvent near the one seg-
ment are the same irrespective of the position of the other. The same holds
for the various configurations of the segment itself. By convention, we
define this constant to be one; then g(r) is called the pair distribution
function.
As the distance r decreases, the solvent molecules around one segment
may also be influenced by the other segment, so that g departs from
unity. Whether g becomes larger than or smaller than one is different in
different solvents. The g can be greater than one, for example, if the
solvent molecules pack more easily around two nearby segments than
around two distant ones. If g > 1 then the potential of mean force
−T log(g) is negative, so that the segments experience a mutual attrac-
tion. For smaller distances r  a, configurations where the segments
avoid each other become rare, so that the probability g(r) becomes small.
The same local interactions that exist for free-floating monomers also
influence segments along the chain. Without this interaction, two seg-
ments which are far apart along the chain fall at displacements r with
uniform probability for r  R. But when the interaction is taken into
account these segments too must have a pair distribution function g(r)
similar to that of the free-floating segments considered above. In order
to have a chain which behaves correctly, the polymer configurations
must be weighted so as to reproduce the correct g(r) for all pairs of
segments.
To see the impact of these interactions, we consider n free-floating
segments confined to a large sphere of radius R and volume . Initially, we
allow these to occupy any relative positions indifferently. But we must attain
a final state where relative positions are governed by the proper g(r) for
these segments and this solvent. For definiteness, we consider a case where
the true g(r) is everywhere less than unity. In general work is required to
accomplish this change in g(r); it is this work that governs how important
the interactions are. We produce the necessary change in g(r) by intro-
ducing a potential of mean force U (r) between our segments. First we
consider the effect of introducing U (r) between just one of the segments
and all the others. We do this in stages, starting from the smallest r val-
ues. Thus for example, when we have reached a generic distance r0 we
turn on a potential U (r0 ) of sufficient strength to exclude all but a frac-
tion g(r0 ) of the original segments at that distance from the chosen one.
This exclusion requires work, since it reduces the volume in which the
other n − 1 segments may live. Like any dilute gas, these segments resist
compression with an effective pressure  = T (n − 1)/. This  is the
 Self-avoidance and self-interaction 69

osmotic pressure encountered in the last chapter. The volume within r0 will
be denoted v0 . To exclude segments completely from volume v0 would
require work v0 ; to exclude a fraction (1 − g(r0 )) requires a proportion-
ate amount of work, since the work is proportional to the number excluded:
(1−g(r0 ))v0 . This process has produced the correct g(r) for one segment
out to a distance r0 . We may extend it to produce a correct g(r) out to some
slightly larger radius r1 . The new volume affected will be denoted v1 . The
associated work is evidently (1 − g(r1 ))v1 . Continuing in this way, we
may impose the full g(r) between the chosen segment and the others, with
an associated work
 
[T (n − 1)/ ] vi (1 − g(ri )) = [T (n − 1)/ ] d 3 r(1 − g(r)). (3.29)
i

Fig. 3.10 illustrates the first two terms of this sum.

We next choose a second segment and perform the same treatment on it,
by turning on interactions between  it and the n − 2 remaining ones. The
work required is [T (n − 2)/ ] d 3 r(1 − g(r)). We proceed in this way
g(r)
until all segments have been chosen. The total work V is given by

V =T d 3 r(1 − g(r))[1 + 2 + · · · + n − 1]/ . (3.30)

The sum in brackets is n(n − 1)/2 → n2 /2. Thus when n  1, r0 r 1

   r
1
V = (T n2 / ) d 3 r(1 − g(r)) . (3.31) Fig. 3.10
2
Removal of segments near segment 1 to
We see that this V has the same form as in the Flory estimate above. create the proper g(r). Initially segments
occur equally at all displacements r. The
But instead of assigning an ad hoc energy U to nearby segments, we may
two shaded bars show fraction of segments
express the free energy associated with segment interactions in a general that must be removed from r < r0 and
and rigorous way, insofar as the segments can be viewed as a dilute solution. r0 < r < r1 . The volumes of these bars are
The quantity in [. . .] in Eq. (3.31) arises often in solution theory. It evidently the first two terms in Eq. (3.29).
has dimensions of volume and is positive when g(r) < 1. It is called the
excluded volume and we shall denote it b2 [14]:

1
b2 ≡ d 3 r(1 − g(r)). (3.32)
2
This is the quantity that determines the effect of interactions between seg-
ments widely separated along the chain. Up to now we have assumed that
g(r) < 1 so that the segment interactions must carry a positive free energy
cost. Whenever b2 is positive, this is true, and we expect the chain to expand
in accord with the reasoning above. But the excluded volume may also be
zero or negative if g(r) is greater than 1, i.e., if U (r) is attractive.
This same b2 measures the degree of repulsion in a polymer chain. Any
configuration c of a random-walk polymer has a free energy U (c) owing
to self-interaction. It is the potential of mean force between all the pairs
of segments. To account for the interaction’s effect on e.g., r 2 , one must
include the Boltzmann weight e−U (c)/T for each configuration. If U is suf-
ficiently small, one may account for these interactions systematically as a
70 Polymer molecules

power-series expansion in U . Since the important interactions are between

segments that are typically separated by distances much larger than a seg-
ment, the effect of the interactions is similar to that in a dilute solution of
isolated segments. Accordingly, one finds that the effects of U enter through
the segment excluded volume b2 .
We now suppose that b2 is slightly negative, so that the effective inter-
action between segments is attractive. Now the Flory reasoning above
does not work: both the interaction energy and the elastic energy favor
contraction of the chain, and no minimum-energy state can be found.
To find the equilibrium size R, we must consider other effects of increasing
concentration. As the concentration of segments within the chain increases,
the dilute-solution formula for V must be corrected. One may include these
effects as a concentration-dependent excluded volume. In terms of the con-
† In discussing this collapsed state we con- centration n/ , b2 → b2 + b3 n/ + O(n/)2 . In most solutions the third
sidered only interaction energy and neg- virial coefficient b3 is positive. Thus when b2 is negative, there is a dens-
lected the work required to contract a
ity at which the contact energy V = [T n2 / ](b2 + b3 n/) is minimal:
random-walk polymer, analogous to S
above. Problem 3.15 shows how to estim- n/  −b2 /b3 . The chain elastic energy becomes irrelevant and the chain
ate this energy, and shows that it is much collapses to a density governed by the local segment interactions. The col-
smaller than the energy of interaction dis- lapsed chain is no longer fractal: the number of segments per unit volume
cussed above. This is why we are justified n/  −b2 /b3 is independent of chain length† .
in neglecting it.
An intermediate case occurs when b2 is zero. Then it is the b3 interaction
energy that must balance the elastic energy S. Setting these equal, we con-
††This case may be interpreted in terms clude R 2 /n ∼ b3 n3 /R 2d , so that R 2+2d ∼ n4 . Thus again, n ∼ R D ,
of the triple-contact exclusion treated in where D = (d + 1)/2†† . In three dimensions this is just the dimension
Problem 3.14
of a random walk. We conclude that a chain with b2 = 0 behaves as a
simple random-walk polymer despite the remaining interaction effects of
†††When one examines the effect of b3 b3 , b4 , etc.††† .
in detail one finds subtle differences from In order to verify the above predictions completely, one would need to
an ideal random walk. For example,
measure changes in b2 and observe the resulting changes in the polymer size.
there are logarithmic corrections to the
random-walk size: R → n1/2 (1−37/363× But it is difficult to measure b2 explicitly. This would require measuring the
log(n)−1 + · · · ) [15]. osmotic pressure of a dilute solution of segments. However, the segments
are altered when they are bonded together to form a chain; so even such
a measurement would not give the b2 appropriate for the chain segments.
Still, we expect to see the three classes of behavior outlined above: as the
effective interaction between segments is made more attractive, one passes
from self-avoiding scaling with D = 53 through random-walk scaling with
D = 2 to the collapsed state with D = 3. This behavior is well known
experimentally, as shown in Fig. 3.9. Polymers showing the self-avoiding
scaling are said to be in good-solvent conditions. Those showing contraction
are said to be in poor-solvent conditions. The intermediate state between
these two is denoted the theta condition. A well-known theta state occurs
for polystyrene in cyclohexane at about 36◦ C. At a few degrees higher
temperature, the chain expansion is discernible. At a few degrees lower
temperature, it is difficult to discern the anticipated contraction, but one
sees other effects of mutual attraction, to be explored below.
As we have seen, even the smallest positive b2 will change the scal-
ing properties from random-walk to self-avoiding behavior if the chain
length n is long enough. But for a fixed n the chain’s behavior cannot change
abruptly; it must change smoothly as b2 increases from zero. We can readily
 Self-avoidance and self-interaction 71

judge the magnitude of b2 necessary to perturb a finite polymer significantly

from its random walk state. If b2 is small enough, the chain may be treated
as a random-walk chain with radius R R0 an1/2 (a being the segment
size). Thus the Flory interaction energy U b2 T n2 /R0d b2 /a d n2−d/2 .
We expect the effect of U to be minor if U is much smaller than the
elastic energy S. For the random-walk chain S T . Thus b2 is negli-
gible when U  T or b2  a d nd/2−2 . In three dimensions b2 must be less
than or about a 3 n−1/2 . As n grows bigger, the chain becomes increasingly
sensitive to b2 .
For the most part the magnitude of b2 is a matter of chemical detail and
our scaling methods can give us little insight. But there is one important
systematic effect on b2 . This occurs when the solvent molecules have a
spatial extent larger than the monomers. If the solvent molecules are con-
nected into chains of length k, this reduces the b2 of our polymer segments
by a factor k. Appendix B discusses how this reduction comes about.
From the preceding paragraphs it is clear that the interaction between
monomers is more complicated than simply forbidding monomers from
coming within a certain distance of each other. And yet, one type of inter-
action is well described by this picture, namely the good-solvent regime
where b2 is positive. One way for two monomers or segments to have
a given positive b2 is for them to be completely excluded from distances
less than (3b2 /(4π ))1/3 and completely allowed beyond this distance. Such
monomers would have the same free energy of interaction V and hence the
same expansion effects as any other monomers with the same b2 . The b2
of the real solution thus plays the role of the volume parameter v in our
initial, simple model of self avoidance. We conclude that a polymer in a
good solvent is well described by a self-avoiding random walk.

3.15. Confining a polymer Forcing a random-walk polymer to expand to a size R

much larger than its natural size R0 requires work S ∼ (R/R0 )2 , as we have
seen. It also requires work to confine a random-walk polymer into a volume
of radius R much less than R0 . This work is not simply the work of bringing
the two ends close together; that would just make our polymer into a big loop.
We need to confine the whole chain. One strategy to convert an ideal chain
into a chain of average size (say, RG ) much smaller than R0 is to constrain
a few of the random steps in the polymer. If we look at a small subsegment
of our polymer, there is some length g  n whose natural radius R0 (g) is
just the desired confined radius R. We divide our polymer into segments of
length g, and then manipulate it so that it has the proper radius, as follows.
Imagine some arbitrary configuration of the polymer. Imagine a model of
it, made out of wire. Now take the end of the first segment and bend the
wire at that point until the second segment overlaps the first. Now go to the
end of the second segment and make a second bend so that the first three
segments overlap. (We need not worry about intersections, since our random-
walk polymer is invisible to itself.) We continue in this way until we have
bent the wire at each of its n/g segments. The resulting wire clearly has a size
of order R. This shows that we can satisfy the constraint by restricting a small
fraction n/g of the random steps in the polymer. A step that was formerly
free to take on all orientations is now restricted
 to a subset of orientations. We
have restricted the configuration sum c for this step by a finite factor—say
K. For the whole chain, the number of configurations has been reduced  by
a factor K n/g . Thus the free energy of our polymer F (R) ≡ −T log c 1 is
F0 + T (n/g) log K.
72 Polymer molecules

(a) Use this expression for F (R) to find the work to confine as a specific
function of R. This estimate gives the correct scaling of the confinement
energy, agreeing with more rigorous arguments. It also can be extended
to the confinement of self-avoiding polymers, as the rigorous arguments
cannot.
(b) Compare the work of confinement found in (a) to the interaction energy
of a collapsed chain discussed in the text. Is it important or not for long
chains?

3.4.4 Universal ratios

We have seen that random-walk polymers have a single asymptotic p(n, r)
independent of the details of how the random walk was constructed. Now
we have seen that self-repelling polymers also have a common behavior
independent of the details of the repulsion, provided b2 is positive. This
means that all sufficiently long polymers in good solvents are quantitatively
identical to each other in the same way all random-walk polymers are. For
example, they have the same function p(n, r) describing their end-to-end
distribution. We say that these properties are universal. As with random-
walk polymers, and other asymptotic functions, there is some amount of
arbitrary choice in the universal functions. These choices amount to scale
factors for the various independent variables, e.g., the λ and μ of the
“Random-Walk Polymer” section above. Thus, in a random-walk poly-
mer, the ratio r 2 /n is arbitrary, since the random-walk polymer rescaling
changes it by a factor of μ2 /λ. However, certain quantities characterizing
the asymptotic behavior clearly have values that are independent of such
† It is natural to ask what is the reason scale factors. An example is r 4 /r 2 2 , encountered in a problem above.
for the strong parallels between the asymp- The numerator and denominator individually acquire factors of μ4 when
totic scaling of the good-solvent polymers
rescaled. But the quotient is constructed to be independent of such scale
and that of the random-walk polymers. In
both cases we used a scale factor for chain factors. Thus this quotient depends only on the universal asymptotic func-
length and a second scale factor for geo- tion p(n, r) and not on the arbitrary scale factors involved with taking the
metric distance. Does this resemblance arise asymptotic limit. This ratio (3 for a random walk polymer in one dimension)
from a deep mathematical property of self- is called a universal ratio.
avoiding polymers that we have glossed
For good-solvent polymers the asymptotic behavior is also universal.
over? No. It arises because we sought sim-
ilar information in the two cases. We asked Here again we seek the asymptotic spatial properties (such as the distribution
for the asymptotic behavior in a limit in function ps (n, r) in the limit n → ∞). Here again we are led to use rescaled
which a single variable (n) was taken to variables ñ ≡ λn and r̃ ≡ μr defined so that p̃s (ñ, r̃) is a finite function.
infinity. The behavior we sought concerned The specific values of μ and λ are arbitrary, as for the simple random walk.
only the spatial distribution of monomers
Still, one can readily construct universal quantities which are independent
on length scales that remain finite fractions
of the size of the polymer when the asymp- of these scale factors† . For example, the scattering intensity I (q, n) can
be written as I (0, n)S̃(qRG ), where RG 2 = 3(dI (q, n)/dq 2 )/I (0, n)|
totic limit is taken. If we had asked for other q=0
information, such as the number of config- is the radius of gyration cf. Problem 3.11. Changing the arbitrary scale
urations of n monomers, additional scale factors cannot change the dimensionless wave vector qRG . Once qRG is
factors would have been needed. The same
is true if we had asked for spatial behavior fixed, S̃ ≡ I (q, n)/I (0, n) cannot be changed by scale factors, either. Thus
on smaller scales than that of the overall the function S̃ is the same for all asymptotic good-solvent polymers. Since
polymer. For example, this analysis did not good-solvent polymers are fractals, the function S̃ must behave at large q
according to the fractal law S̃ → h (qRG )−D , as sketched in Fig. 3.7. The
tell us that the large q (small distance) beha-
vior of the chain was that of a fractal. We
determined this from a specific analysis of coefficient h gives an important quantitative connection between the short-
how short segments of a chain resemble distance structure and the overall size of a polymer. Scattering experiments
longer ones. give a value for h of about 1.14 ([8] using Eq. (XV–3.54)). Thus if the low-q
 Self-avoidance and self-interaction 73

scattering is known for a given good-solvent polymer, its high q scattering

intensity is completely determined. Clearly such quantities as h give great
predictive power in describing polymer behavior.

3.4.5 Polyelectrolytes
It was mentioned in the first section that polymers in solution can be made
with an ionic charge every few bonds along the chain. The electrostatic
repulsion between these ions tends to expand the chain, and the degree of
expansion proves to be more extreme than that of neutral chains. To see
this, we imagine turning on the electrostatic repulsion from the most global
progressively down the most local scales, inverting the procedure of the last
section. We begin by considering the repulsion only between the first half of
the chain and the second half. When this is done for a neutral chain, it causes
the two halves to swing away from each other; this reduces the number of
contacts between the halves, and thus its interaction energy, dramatically.
The result of this global interaction by itself is to increase the radius by a
factor of order unity. At that point the energy of interaction between the two
halves is of order of the thermal energy T † . †For this statement to be strictly true, we
When we do the same thing with a polyelectrolyte chain, we must must also turn off the interactions involving
monomers near the center one.
consider the electrostatic repulsion between the first and last halves. If
the entire chain has n monomers, each bearing, say, one charge, the repul-
sion is roughly that between two charges of magnitude en/2 at a separation
R as large as the polymer. This energy is roughly e2 (n/2)2 /(R), where
 is the dielectric constant of the solvent. It may be expressed in units of
the thermal energy T as T (n/2)2 (/R). The Bjerrum length  ≡ e2 /(T )
is roughly 0.7 nm in water at room temperature. If R were the size of a
random-walk chain, so that R ∼ n1/2 , this energy clearly grows indefin-
itely large with n. Unlike the short-ranged energy, it cannot be much reduced
by simply making the two halves swing out of each other’s way. This global
energy would cause a large expansion of the chain—not just a finite-factor
expansion as in the neutral case. Balancing this global Coulomb repulsion
against the elastic energy T R 2 /(a 2 n), the two energies are comparable
when R ∼ n. We are led to the conclusion that the swelling exponent
ν = 1; this is as large as ν can be for a connected chain. The repulsion
is strong enough to stretch the chain to within a finite factor of its full
extension, as though the chain were a rigid rod.
To confirm this picture and to see how rigid this rod is, we are led to model
the chain as a row of evenly spaced charges at separation R/n, connected
1/2
by random-walk polymers of unstretched length r 2 1 ≡ a. The preferred
distance between charges can be found by considering the tension in the
polymer near the center. This tension F is equal to the repulsive force
between the left and the right halves. Adding up the forces on each charge
on the left half, one finds†† †† Here we have approximated the sum over

charges by an integral. The approximation

F T (n/R)2 log(n). (3.33) is valid when n  1.

The force increases indefinitely as n → ∞, even with fixed spacing n/R.

This force must be balanced by the elastic tension in the chain. Since
74 Polymer molecules

random-walk chains store energy like an ideal spring, they exert a force
proportional to their elongation. Specifically, F = 3T (R/n)/a 2 . Equating
these tensions leads to the preferred spacing: R/n a( log(n)/a)1/3 . If the
charges were placed sparsely along the chain, a  , and the spacing R/n
would be much smaller than the unperturbed size a. Thus locally, the effect
of the charges can be weak. Still, the overall effect is strong. Even a weakly
charged chain, if long enough, would stretch the monomers out to their
full extension, because of the logarithmic factor. We shall assume in what
follows, though, that the charges are sparse enough so that the monomers
are not strongly distorted: R/n  a.
The real chain must fluctuate around this regular rod state by stretching
and by bending. If the chain stretches by a small amount γ R, the asso-
ciated energy must be proportional to the square of this small change. It
is convenient to express the constant of proportionality as ER so that the
stretching energy is ERγ 2 . To estimate the modulus E we consider strains γ
of order unity, so that the length R is doubled. Such a stretching must change
both Coulomb and elastic energies by factors of order unity; hence the total
energy is changed by such a factor. This means the stretching energy is of
the order of the original Coulomb energy: ERγ 2 |γ =1 nT ( log(n)/a)2/3 ,
or E a −1 ( log(n)/a)1/3 . Except for the weakly varying log(n) factor,
the modulus E is independent of n, as in a solid, macroscopic rod. One
may now estimate the amount of strain γT typically present because of
thermal fluctuations by setting ERγT2 T . Thus γT ∼ 1/R 1/2 if we
again neglect the log n so that E is independent of size. The log(n) factor
only reduces γT further. As R becomes large, the thermal fluctuations in R
become increasingly insignificant.
R A polyelectrolyte is also rigid with respect to bending, e.g., into an arc
of radius B. One can readily find the Coulomb energy needed to bend a
uniformly charged rod in this way. Any small bending must cost an energy
of the form ẼR(R/B)2 . But a bend of, say, 90◦ with B R changes
the distance between two arbitrary charges by a factor of order unity as
B illustrated in Fig. 3.11. Thus the Coulomb energy increases by a similar
factor: ẼR n2 T /R. The coefficient Ẽ is of the same order of mag-
nitude (except for the log(n)) as the modulus E. The amount of thermal
Fig. 3.11 bending R/BT , like the thermal strain γT , falls off as the square root of the
A charged rod of length R is bent into an chain length n. We conclude that a polyelectrolyte chain has little of the
arc of radius of curvature B R, thus fluctuations that random-walk or self-avoiding chains have. The assumed
shortening the distance between a typical state of a uniformly charged line is affected but little by thermal fluctuations
pair of charges (heavily drawn) by a finite
factor.
in stretching or bending.
In this chapter we have seen that connecting arbitrary small molecules to
form a flexible chain leads to remarkable structures. These structures are
very different from their compact constituents. They are indefinitely tenu-
ous and deformable. In order to think about these objects quantitatively, a
new concept is needed. This is the notion of fractal scaling. Using it, we
found that chains made of different constituents nevertheless converge to a
common quantitative behavior, provided they are long enough. This com-
mon asymptotic behavior is embodied in scaling laws and scaling functions
like the reduced scattering function S̃(q) defined above.
 Appendix A: Dilation symmetry 75

In the next chapter we exploit and extend these scaling laws to explain the
striking properties of polymer solutions: their springiness, their viscosity,
and their tendency to absorb further solvent.
3.16. Screened polyelectrolyte: the Odijk length In our discussion of polyelec-
trolytes we ignored the countercharge that neutralized the ions on the chain
before it was put into solution. The interaction between the polyelectrolyte
charges is modified by the countercharges. In addition, the solvent always
contains further ions, not related to the polymers. We shall see in Chapter 5
that the effect of these ions is to change the interaction potential between two
unit charges to the form (e2 /r) exp(−κr): the potential falls off exponentially
for distances larger than the screening length 1/κ. This κ depends only on the
concentration of ions and is independent of the chain length n. Thus when
n is large, the interaction is relatively short-ranged: κR  1. Still, even such
screened chains are remarkably rigid. By extending the methods sketched in
the text, Odijk [16] found that that screened chains can be rigid on length
scales much greater than the screening length 1/κ.
Consider a length R of chain much longer than 1/κ, bent into a gentle arc of
radius B  R. As in the text, the chain is treated as a line of uniformly spaced
charges separated by a distance a. (a) Find the change in Coulomb energy of a
given charge in the limit of weak bending. It must be of order 1/B 2 . The total
energy cost of bending the rod is evidently this energy per charge times the
number of charges n. For sufficiently long R, the cost of bending the chain
through a unit angle, with B = R, becomes of order T . This length R is called
the electrostatic persistence length, or Odijk length L0 . Chains shorter than the
Odijk length behave as rigid rods, even with screening; chains much longer
than the Odijk length coil randomly. (b) The quantity L0 κ is the persistence
length relative to the screening length. In a limit where the distance between
charges a is much smaller than the screening length 1/κ, how does L0 κ scale
with κa? Is the chain flexible or rigid on the scale of 1/κ?

Appendix A: Dilation symmetry

The fractal property reflects a pervasive symmetry of the polymer: a general
invariance under spatial dilation. Dilation invariance means that these poly-
mers are statistically essentially identical when all distances are multiplied
by an arbitrary constant λ. To define this invariance operationally, we ima-
gine a random set of points in space such as those occupied by our polymer,
represented by a dot for each monomer. We study this object by making a
large sample of images of its interior. The scale of these images is to be
much smaller than the object as a whole, and much larger than any ultimate
microstructure such as a single monomer. The images have limited resolu-
tion, so that the individual points of the set are not distinguished. Instead the
object appears as a cloudy blur spanning the picture. It is natural to select
images that have substantial density at the center. Similarly, it is natural
to regulate the overall contrast of the images so that the set is clearly vis-
ible. Figure 3.12 shows a sample of such images. Each image represents a
density ρ(r) distributed in some way over a finite region of space. Dilation
invariance is a property of this sample of images relative to another sample
taken at a different spatial magnification. Evidently, if there is no objective
way to distinguish the one set of images from the other, the random object
is invariant under dilation. The figure illustrates this property: the different
magnifications shown appear indistinguishable. One example of a dilation
invariant set is a vertical line. The images at any magnification show a
76 Polymer molecules

Fig. 3.12
Eight randomly sampled images of a long
self-avoiding random-walk polymer. (We
show self-avoiding polymers because
simple random-walk polymers are difficult
to illustrate in two dimensions.) Half of the
images are at a 2.5-fold magnification
relative to the other half. The monomer
positions are blurred to a fixed fraction of
the picture size. In each case the picture
was centered on a monomer. The two
magnifications are interspersed randomly.
The reader is invited to guess which are the
magnified images. (Polymers were
generated via a Monte-Carlo simulation
originally written by Betsy Weatherhead
and drawn by Olgica Bakajin, 1995) For
the solution see the end of this section.

blurry line of the same darkness passing through the center. There is no
way to distinguish which has the larger magnification. An example lacking
dilation symmetry would be a soap froth in a pan, sampled by preparing the
froth repeatedly and then photographing it. Pictures at different magnifica-
tions are clearly distinguishable because the typical bubbles have different
sizes at different magnifications.
To state this invariance mathematically, we may define a set of statistics
that describe the random density ρ(r). One way to characterize the vari-
ations of ρ over the images is by means of the correlation functions
ρ(r1 )ρ(r2 ) . . . ρ(rk ) 0 . The 0 subscript indicates that the pictures in our
sample are selected to have the origin occupied. The local density discussed
in the main text is the simplest example of these correlation functions. The
points r1 , . . . , rk are specific points on each image. For example, r1 could
be 1 cm inward from the upper left corner. Any of these correlation func-
tions can be measured to arbitrary precision using a large enough sample of
images. The same process performed on images magnified by a factor λ−1
gives ρ(λr1 )ρ(λr2 ) · · · ρ(λrk ) 0 . Dilation invariance implies that the two
statistics are the same:

ρ(r1 )ρ(r2 ) · · · ρ(rk ) 0 = μk (λ)ρ(λr1 )ρ(λr2 ) · · · ρ(λrk ) 0 . (3.34)

The μ prefactor accounts for a possible overall change in “contrast” upon

dilation. Many of the structures that occur in complex fluids have dilation
invariance of this type. An important source of recent progress in under-
standing these fluids is the recognition and exploitation of this symmetry.
We may show the dilation symmetry in an infinitely long random walk
polymer by using again the scaling property of the end-to-end probability
p(n, r). We first consider a typical configuration that contributes to
ρ(r1 ) · · · ρ(rk ) 0 . Any chain that contributes must pass through each of the
points r1 , . . . , rk as well as through the origin. The chain may pass through
these points in any order, and with any arbitrary number of monomers in
 Appendix A: Dilation symmetry 77

each segment between these points. Of these cases, we consider first the one
where the points are visited in order: the chain goes from the origin to r1 ,
then to r2 , etc. We further suppose that the number of monomers between
0 and r1 is i1 , and that the succeeding segments have i2 , . . . , ik monomers.
Once all this has been specified, the probability of such a chain can easily
be stated: it is the product p(i1 , r1 )p(i2 , r2 − r1 ) · · · p(ik , rk − rk−1 ). The
probability takes this simple form because each of the chain segments is
statistically independent. To find the joint probability for all chains that pass
through the points in this order, we must simply add all the contributions
from the different possible i1 , etc.:
⎡ ⎤ ⎡ ⎤
 
⎣ p(i1 , r1 )⎦ [ ] · · · ⎣ p(ik , rk − rk−1 )⎦ .
i1 ik

This sum is itself dilation invariant. To see this we use the scaling form
p(i, r) = i −3/2 f (r 2 /i), and replace the monomer sums by integrals. The
same sum dilated by a factor λ is
   
−3/2 2 −3/2 2
di1 i1 f ((λr1 ) /i1 ) [ ] · · · dik ik f (ik , (λrk − λrk−1 ) /ik ) .
(3.35)

We may scale out the λ factors by defining new ĩ variables as ĩ ≡ λ−2 i.

Each integral may be converted to the ĩ variables by converting the di’s and
the i −3/2 factors. This results in a factor of λ2 λ−3 for each integral. Thus
our sum may be written
   
−1 k −3/2 2 −3/2 2
(λ ) d ĩ1 ĩ1 f ((r1 ) /ĩ1 ) [ ] · · · d ĩk ĩk f ((rk − rk−1 ) /ĩk ) .
(3.36)

Except for the λ−k prefactor, this is the same as the undilated sum. Thus
the sum is dilation invariant. We may repeat the reasoning for any order
of passage among the points 0, . . . , rk . All the corresponding sums have
dilation symmetry with the same prefactor. To obtain the overall probability
of passage through these points, we must simply add these sums together
for all possible orders of passage. The total evidently has the same dilation
symmetry as each part. Finally the density of monomers at r is evidently the
same as the probability density for a monomer to be there. That means that
our joint probability is simply ρ(r1 ) · · · ρ(rk ) 0 . We have thus shown that

ρ(r1 ) · · · ρ(rk ) 0 = λk ρ(λr1 ) · · · ρ(λrk ) 0 . (3.37)

Comparing this result with the general definition in Eq. (3.34), we see that
the μ prefactor takes a simple and suggestive form: it is a power of λ which
is proportional to the number of density factors ρ. The density behaves as
though it had dimensions of length−1 rather than its standard dimensions
of length−3 . This latter is the scaling we would find if we multiplied all
lengths, including the size of a monomer, by a factor λ. By contrast, our
78 Polymer molecules

dilation invariance applies to the polymer with no change in the monomers.

The case k = 1 shows that the polymer is a fractal object. It says

ρ(r) 0 = λρ(λr) 0 . (3.38)

Setting λ = constant/r, this amounts to ρ(r) 0 ∝ 1/r: The density falls

off as in a fractal with D = 2, confirming our previous conclusion. In
general a density profile with dilation symmetry is a fractal density.
The dilation invariance amounts to a form of statistical regularity which
is important in predicting properties. To illustrate this regularity, we return
to the fractal analysis applied above to our polymer. We consider the
mass M(r) of the atoms lying within a distance r of an arbitrary monomer.
We have seen that the scaling of the average M with r is that of a fractal.
But the value of M obtained in a given instance fluctuates about this aver-
age. Using the dilation invariance property above, we can see how much M
fluctuates. Different physical systems fluctuate by different amounts. For
example, the mass in a given macroscopic volume of a solid object fluc-
tuates very little: the percentage spread in the sampled M values becomes
indefinitely small as the size r is made larger. In other cases, the fluctuations
of M are wild. For example, if we dropped our restriction that a monomer
be at the center of our sphere and studied spheres placed arbitrarily in the
pervaded volume of the polymer, we would almost always obtain M = 0,
and only rarely encounter volumes containing monomers. The distribution
of M’s would be heterogeneous. It is important to know how wildly M
fluctuates for monomer-centered spheres. The less wild they are, the more
predictable the polymer behavior is.
One way to analyze the fluctuations of M is to consider moments
like M k . We can readily express such moments in terms of our
ρ(r1 )ρ(r2 ) . . . ρ(rk ) 0 . Using the dilation invariance property of the
ρ . . . ρ ’s we can readily infer how the moments of M scale with the
sampling size r: M(r) = r  <r d 3 r  ρ(r  ), so that
 
3
M(r) =
k
d r1 · · · d 3 rk ρ(r1 )ρ(r2 ) · · · ρ(rk ) 0 . (3.39)
r1 <r rk <r

Using the dilation invariance property of the ρ · · · ρ ’s we can readily

infer how the M moments scale with the sampling size r: M(λr)k =
λ2k M(r)k . This scaling implies regularity, because it says that ratios
of different characteristic masses are independent of r: [M(r)k ]1/k =
Ck M(r) , where the constants Ck are independent of r. The mean mass,
root-mean-square mass, etc. are all of the same order, so that knowing any
one of these averages determines them all approximately. Another way to
express this regularity is in terms of the probability distribution of sampled
M values: Pr (M). We may then express the normalized k-moment as

M̃ k = d M̃ M̃ k Pr (M̃), (3.40)

where M̃ ≡ M/M . Since the moments scale the same, they are all finite
for a given r, however large. Thus the P (M̃) must fall off faster than any
 Appendix B: Polymeric solvents and screening 79

power as M̃ → ∞. By further analysis, one may show that P (M̃) falls

log(具M典 P (M ))
off exponentially with M̃. The distribution also falls off for small M̃, as
exp(−1/M̃). It thus has the form sketched in Fig. 3.13. The function P (M̃)
must also be universal, since it can be calculated from the universal p(n, r).
Compared to a solid body, the fluctuations in M(R) for our random-
walk polymer are large as Fig. 3.13 suggests. The relative width of the
–2 0 2
distribution, e.g., (M − M )2 /M 2 , does not grow with r, but neither log M/具M典
does it go to zero† . Real, good-solvent polymers are dilation-symmetric
[8],with μk (λ) = (λ3−D )k . Their P (M) distributions are thought to be Fig. 3.13
universal and to resemble Fig. 3.13 [17]. Other fractal objects, such as Qualitative sketch of the mass distribution
colloidal aggregates are also thought to have this form of dilation symmetry. P (M) as a log–log plot. The M variable
has been scaled with M ; the P variable
has been scaled with M −1 . With this
Appendix B: Polymeric solvents and screening scaling the curve is universal. The
downward curvature for large and for small
In this Appendix we discuss how linking solvent molecules together results M indicates that P (M) falls off faster than
in a weakening of interaction between segments of a polymer dissolved any power for large or for small M.
in that solvent. This is easiest to understand if we place the solvent and
polymer on a lattice, with each segment occupying a lattice site. The solvent † Some known sets of points called multi-
molecules, being larger than the segments, occupy k > 1 sites within some fractals have scale invariance of a more
radius Rs . For concreteness, we can consider the k-site solvent molecules to general type. An example of a multifractal
be small polymers of the same species as the large polymer being studied. is the charge distribution on a conduct-
Thus in principle all the issues of fractal structure arise for both the solvent ing fractal that has been charged to some
potential V . Multifractal densities have μ
and the polymer, as discussed in the chapter. However, the weakening of powers that are not simply proportional to
interactions we wish to explain is independent of the details of this fractal the number of ρ factors. Thus multifractals
structure. Accordingly, we will ignore questions of fractal structure here. have mass moments that diverge at differ-
At first glance, our change to k-site solvent molecules seems unimport- ent rates as r → ∞. For a multifractal one
ant. For example, if these solvent molecules occupied a compact cube of must determine individually the scaling of
all these moments in order to characterize
k sites, our system would amount to a change to a coarser lattice whose the fluctuations of M.
sites had a linear size of k 1/3 instead of 1. But we wish to consider tenuous
solvent molecules that can interpenetrate. The k sites of each molecule per-
vade a volume shared by many other such molecules. This is the situation
when the solvent itself consists of polymers. But it is equally true when the
solvent consists of any tenuous structures, whose volume fraction decreases
indefinitely as k increases.
150x150 Polymer 2

150x150 Polymer 3
60x60 Polymer 1

60x60 Polymer 4
150x150 Polymer 4

150x150 Polymer 1

60x60 Polymer 2

60x60 Polymer 3

Fig. 3.14
The polymers of Fig. 3.12 without
blurring.
80 Polymer molecules

The assumption of interpenetration makes it easier to see what happens

when a long polymer occupying N  k sites is dissolved in such a solvent.
The main effect of the large solvent molecules is to reduce the pressure, as
we now explain. Our system consists of many k-site solvent molecules on
a lattice, such that the average number of monomers on a site is about 1.
Pressure is the work required per site volume to empty a site. We used
this notion of pressure above in our explanation of excluded volume. In
a real lattice gas, no more than one monomer is allowed on a site: the
monomers are impenetrable. This impenetrability constraint acts in a similar
way whether the solvent consists of single sites or k-site molecules. For the
moment we ignore it. Instead we shall allow all system configurations in
which the average volume fraction φ is about 1. That means, for an N site
lattice, there must be about N /k solvent molecules. But their positions are
arbitrary. Thus our solvent forms a noninteracting, ideal gas. Now it is clear
why the k-site solvent has a reduced pressure. The pressure in an ideal gas
is T times the density of molecules. For our system this amounts to T /k
per site-volume. The pressure is reduced by a factor k relative to that of a
single-site (k = 1) solvent.
This reduced pressure has an important effect on the interaction between
objects in the solution. We now consider two designated molecules and
investigate their mutual excluded volume. To find this excluded volume,
we place one of our solute molecules at the origin, and then determine the
work required to place the second solute molecule into the same region.
For this purpose it is convenient to identify some compact region of that
contains all the sites of a given molecule. We may imagine it to be a sphere of
radius Rs and volume V sites. We can suppose that all the solvent molecules
have roughly the same size, so that we may take the enclosing region to be
the same size for all the molecules, including our two designated molecules.
The number of cells V in this region is the pervaded volume of a molecule.
We shall see that the precise definition of this region is unimportant. Our
assumption that the solvent molecules are tenuous implies that V is much
larger than k. The molecule added at the origin causes the average volume
fraction nearby to increase slightly, by an amount k/V .
We now consider the second designated molecule, and focus on
configurations in which it lies in the pervaded volume of the first molecule.
We compare these with configurations where the second molecule lies at
some fixed position far from the first molecule. Work is required to take
the second molecule from the far-away position into the pervaded volume.
Since the pervaded volume has a slightly higher density, we may regard
it as having slightly higher pressure. The work is the volume k of the
molecule times the change of pressure p. This p = (T /k)φ =
(T /k)(k/V ) = T /V . Note that this excess pressure is a factor k smal-
ler than it would have been in a simple one-site solvent. The work W
required to bring the second molecule to the origin is thus kp = T k/V .
We see that this work is much smaller than T . Thus there is little hindrance
for the second molecule to approach the first. Nevertheless, this hindrance
reduces the probability that the second molecule lies at the origin, by a
factor e−W /T → (1 − W /T ). This reduction factor is precisely the pair
distribution function for our two designated molecules: g(r)|r=0 . The same
 Appendix B: Polymeric solvents and screening 81

reduction factor applies roughly for typical points r in the pervaded volume:
1 − g(r)|r∈V W /T .
Now we can find the mutual excluded volume b̃2 of our pair of molecules:
1
b̃2 = (1 − g(r)) W /T V . (3.41)
2 r

Using our estimate for W , we find b̃2 k. Using this result, we can infer
the self-interaction energy of an n-site polymer, containing n/k of the k-site
molecules. Repeating the reasoning of Eq. (3.31), the self-interaction energy
V of the polymer is given by

Vpolymer T (b̃2 /R 3 )(n/k)2 T b2 /R 3 n2 /k. (3.42)

Here b2 is the mutual excluded volume of the original segments with
k = 1. The interaction has been reduced by a factor k, as announced in
the text. We say that the large solvent molecules have screened the inter-
action between the two solute molecules. Because of screening, the chain
is but little perturbed, and R R0 an1/2 so long as Vpolymer  T ,
i.e., 1  (b2 /a 3 )n1/2 /k. The solvent chains can be much shorter than the
polymer and still prevent it from expanding significantly.
This subtle notion of screening plays a central role in complex fluids.
Screening occurs whenever two interacting objects also interact similarly
with many surrounding objects that are free to move. The prime example
is electrostatic screening, to be considered in Chapter 5 to follow. The
profound effect of polymer screening was first recognized by Flory [10, 18],
and justified by a combinatorial lattice gas argument. The unity of polymer † Another way to justify the use of our
and electrostatic screening was elucidated by Edwards [19]. ideal system is to argue that in the tenu-
The above treatment of screening used the worrisome assumption that ous limit the idealized system nearly sat-
the dense solution of solvent molecules could be treated as an ideal gas, isfies the constraints of the real system.
ignoring all the packing constraints that should affect the system profoundly. For definiteness, we suppose that our sys-
tem has an average volume fraction φ
In our ideal system, we constrained only the average volume fraction of the
of 0.99. We consider the statistical fluctu-
system to be some value slightly less than 1. In the real system, there can ations of φ within the pervaded volume V
be no configuration where a site has a volume fraction exceeding one. Such of a molecule. On average V /k molecules
a treatment is possible because we have treated the solvent molecules as share this volume. In the ideal system these
arbitrarily tenuous, with the pervaded volume of each shared by indefinitely molecules enter and leave independently,
their number fluctuates by an amount equal
many others. Because the molecules were tenuous, we could find their
to the square root of the average: (V /k)1/2 .
mutual excluded volume by considering only the slight change of pressure Thus in this region φ fluctuates over a
required to increase the volume fraction by an arbitrarily small amount. An range of ±(V /k)1/2 /(V /k) or ±(k/V )1/2 .
actual liquid made of large solvent molecules has a pressure determined If this fluctuation amounts to much less
by many detailed effects, including packing constraints and the mutual than 1%, the occurrence of φ > 1 becomes
rare. To satisfy this condition would require
attraction of the atoms. However, one can create slight changes in density
V > 104 k. If this condition is satisfied,
in a region simply by translating solvent molecules into the region, without the ideal system nearly obeys the con-
appreciable changes in their local packing or attraction. Therefore it is straints of the real system without further
mainly the free energy associated with the translation of the molecules that intervention. Thus the ideal system is a rea-
affects the small changes of concentration that are important for our system. sonable approximation to the real system.
Real liquids have an average volume frac-
This translational free energy can be treated without considering packing
tion substantially lower than their maximum
or other interactions between solvent molecules—i.e., we may consider the volume fraction. Thus the requirements on
solvent to be an ideal gas† . V /k are less stringent.
82 Polymer molecules

References
1. See e.g., D. W. Van Krevelen, Properties of Polymers: Their Correlation with
Chemical Structure; Their Numerical Estimation and Prediction from Additive
Group Contributions, 3rd ed. (Amsterdam: Elsevier, 1990) p. 333.
2. P. Rempp, E. Franta, and J.-E. Herz, in Polysiloxane Copolymers/Anionic Poly-
merization, Advances in Polymers Science series, no. 86 (Heidelberg: Springer,
1986), p. 147
3. G. G. Odian, Principles of Polymerization, 3rd ed. (New York: Wiley, 1991).
4. K. P. McGrath, M. J. Fournier, T. L. Mason, and D. A. Tirrell, J. Am. Chem.
Soc. 114 727 (1992).
5. H-G. Elias, Macromolecules, 2nd ed. (New York: Plenum Press, 1984), p. 124.
6. B. B. Mandelbrot, The Fractal Geometry of Nature (San Francisco: Freeman,
1982).
7. T. A. Witten, Rev. Mod. Phys. 70 1531 (1998).
8. G. Jannink and J. Des Cloizeaux, Polymers in Solution (Oxford: Oxford, 1992).
9. J. C. Le Guillou and J. Zinn-Justin, Phys. Rev. Lett. 39 95 (1977).
10. P. Flory, Principles of Polymer Chemistry (Ithaca, New York: Cornell, 1971),
Chap. XII.
11. J. P. Cotton, D. Decker, B. Farnoux, G. Jannink, R. Ober, and C. Picot, Phys.
Rev. Lett. 32 1170 (1974)
12. M. A. Moore and A. J. Bray, J. Phys. A: Math. Gen. 11 1353 (1978); S. F.
Edwards and P. Singh, J. Chem. Soc. Faraday Trans. 2, 75 1001 (1979); M. K.
Kosmas and K. F. Freed, J. Chem. Phys. 68 4878 (1978).
13. J. Des Cloizeaux, J. Phys. 31 715 (1970).
14. R. K. Pathria, Statistical mechanics (Oxford: Pergamon Press, 1972).
15. B. DuPlantier, Europhys. Lett 1 491 (1986); G. Jannink and J. Des Cloizeaux
op. cit., Chap. XIV.
16. T. Odijk, J. Polym. Sci. 15 477 (1977).
17. B. M. Sterner, to be published, demonstrates numerically the universality
of P (M) for two-dimensional self-avoiding polymers and obtains values for
several of the numerical coefficients Ck for positive and negative k.
18. M. Huggins, J. Phys. Chem. 46 151 (1942); M. Huggins, Ann. NY. Acad. Sci.
41 1 (1942); M. Huggins, J. Am. Chem. Soc. 64 1712 (1942).
19. S. F. Edwards, Proc. Phys. Soc. 88 265 (1966).
 4
Polymer solutions

The previous chapter has shown how a polymer chain behaves when it is
by itself in a solvent. But the properties seen in a macroscopic fluid result
from many polymers interacting in the solvent. In this chapter we turn to
a survey of how polymer structure, energy, and motion are altered by their
mutual interactions. We first describe the energy associated with interac-
tion in dilute solution. The following section discusses the new structural
and energetic features that appear when the polymers interpenetrate. Then
we turn to dynamics. The first subsection discusses the Brownian motion
of a sphere, the starting point for understanding how objects move in a
liquid. We show how spontaneous thermal motions are related to the work
needed to create motion. We discuss the viscosity added to a liquid by
inserting particles or polymers into it. Next we consider how these proper-
ties are altered when polymers interpenetrate. We describe how flow fields
are modified by hydrodynamic screening. We distinguish between self-
diffusion and cooperative diffusion. Finally we discuss how entanglement
affects both diffusion and the relaxation of stress.

4.1 Dilute solutions

The interactions between polymers are easiest to treat when the polymers
are far apart in solution and their mutual influence is weak. The main effect
of this influence is on the free energy. The work required to add an additional
polymer to the solution is altered by the presence of the others. To evaluate
this effect, we start with an initial state in which the monomers of a given
chain interact with each other but those of different chains do not. Even with
no interaction between chains, one must do work to confine N polymers
into a solution volume . We saw this in Chapter 2 when we considered the
work to compress an ideal gas. Interactions modify this work, as explained
in the previous chapter under “Self-interaction and Solvent Quality.” The
work to change the volume available to the polymers by an amount d is by
definition − d, where  is the osmotic pressure. This work and thus 
are altered when the interactions between the monomers of different chains
are turned on. The work V required to turn on the interaction must have the
same form as that found in the preceding chapter: V = T (N 2 /)B2 , where
B2 is the mutual excluded volume of two polymer chains. This excluded
volume must be expressible in terms of the polymers’ gp (r) just as in the
84 Polymer solutions

preceding chapter:

1
B2 = d 3 r(1 − gp (r)). (4.1)
2
The presence of V means that each increment of compression d requires
an extra work −(dV/d) d. Thus V adds an amount −dV/d to the
osmotic pressure :

 2 
dV NT N N
 = 0 − = 1 + B2 + O . (4.2)
d   

This formula for osmotic pressure is valid irrespective of the type of solute
molecule; the distinctive polymer features show up in how the excluded
volume B2 depends on molecular weight n and size R. We may infer how
B2 behaves by considering the pair correlation function gp (r). For r  R,
two polymers are out of each others’ range of influence, so that gp (r)
should be nearly 1. But for r < R there is the possibility that gp will be
strongly influenced. To gauge the magnitude of such an influence, we con-
sider our polymers to be in a good solvent, i.e., one in which the monomers
have b2 > 0. We have seen that the effect of the b2 may be reproduced
by excluding all configurations in which two monomers approach closer
1/3
than some small distance of order b2 . To see the effect of this exclusion,
we take two polymer configurations at random, move them to a separation
r < R, and then discard the result if there are any intersections. The gp (r)
is the probability that the result is retained, i.e., the probability that there
are no intersections. As we saw in Eq. (3.26), the average number of inter-
† Here and in the following we shall use sections MAB (r) ∼ r 2D−d ∼ r 1/3† , and the probability of no contact is
the Flory value for D, keeping in mind significantly less than 1 for r < R. For example, gp (r) < 12 for all r smaller
that the actual number may well be slightly
than some fixed fraction αR as R → ∞. (This was a consequence of two
different.
polymers being mutually opaque.) This sets a lower bound on B2 :
  
1 3 1 1 4π 3
B2 = d r(1 − gp (r)) > (αR) . (4.3)
2 2 2 3
On the other hand B2 cannot grow faster than R 3 : polymers cannot exclude
each other from regions arbitrarily larger than their size. There must thus
be some constant C for which B2 < CR 3 ; this is an upper bound. In
view of these two bounds, B2 /R 3 must remain finite (and greater than 0)
as R → ∞. The polymers increase the pressure  as though they were
mutually impenetrable spheres whose radius is roughly the geometric radius
R. This is a consequence of their mutual opacity.
4.1. Osmotic pressure of hard-sphere gas A dilute gas of hard spheres of radius
R has a g(r) which is zero for r < 2R and 1 beyond. (a) What is the excluded
volume B2 for such a gas? (b) For what volume fraction φ = N 43 π R 3 /  of
spheres is the B2 term in the osmotic pressure equal to the ideal-gas term?
(Some amusing points of comparison: a close-packed crystal has a volume
fraction of 74%; random close packing has a volume fraction of 64%.)
An important further consequence of this opacity is implicit in the finite-
ness of B2 /R 3 as R → ∞. We have argued that it is finite whenever the
 Semidilute solutions 85

monomer excluded volume b2 is positive. This means that there is a finite

limit for B2 /R 3 , when b2 → 0. More precisely, for any fixed b2 , no matter
how small, some finite B2 /R 3 is ultimately reached for large enough R. The
limiting B2 /R 3 is evidently independent of b2 . As long as b2 is small enough,
it can be varied by a large factor without changing B2 /R 3 significantly. But
we have argued that all solvent effects enter only through b2 (as long as
it is positive). Thus any self-repelling flexible polymer must achieve the
same value of B2 /R 3 . The limiting value is universal. Though one can
show this convincingly [1], one can only calculate its value very crudely.
Thus our best estimates of the ratio come from experimental data like those
4.9
of Fig. 4.1. Taking R to be the radius of gyration RG measured in scatter-
ing experiments (see Problem 3.11), one finds B2 /RG 3 = 4.75 ± 0.5 [2].

Another way to express this universality is to define Rt as the radius of B2

the hard sphere with the same B2 as the polymers. B2 ≡ 4( 43 )πRt3 . Then RG3
Rt /RG is also a universal ratio. Table 4.2 summarizes this and other such
ratios to be introduced below. 4.5
As the polymer concentration increases from zero, the interactions make 50 250
RG (nm)
an increasingly important contribution to the osmotic pressure . Ulti-
mately the B2 term in Eq. (4.2) becomes as large as the ideal solution
pressure 0 . The concentration N/  = 1/B2 where this occurs is an Fig. 4.1
3
The ratio B2 /RG versus RG , as reported
important one in polymer solutions. At this concentration the distance in [2], confirming that the ratio is
−1/3
between chains (/N )1/3 is about equal to B2 , which as we have seen independent of chain length. Data is for
is of order R. If polymers were put at random into the solution at this polystyrene chains of increasing molecular
weight in benzene.
concentration, a sizeable fraction of them would overlap.
The corresponding monomer concentration nN/  = n/B2 is called the
overlap concentration ρ ∗ (pronounced “rho star”). This ρ ∗ is of order n/R 3 ,
the average monomer density within a given chain volume. We have seen
that this density goes as R D−d ∼ R −4/3 : ρ ∗ becomes indefinitely small as
R → ∞. Evidently if ρ  ρ ∗ , polymer interactions play a small role in
the osmotic pressure or the energy of the solution. Conversely, if ρ  ρ ∗ ,
interactions play an important role. The interaction energy, like  at this
overlap concentration is evidently of order of a thermal energy per chain
volume:  20 T R −3 .

4.2 Semidilute solutions

For long polymers, if ρ ρ ∗ the solution is still almost all solvent. One
may thus remove some large fraction of the solvent and arrive at a solution
much more concentrated than ρ ∗ which still has an arbitrarily small fraction
of solvent. Such solutions are termed semidilute. There is no counterpart
of this semidilute regime in small-molecule solutions. Small molecules
begin to interact only at some finite density ρ. Thus one may only raise
the concentration from ρ ∗ by a finite factor before all the solvent has been
removed. In order for a semidilute regime to exist, the solute must be tenuous
as our fractal polymers are.
As one removes more and more solvent, one arrives at concentrated
solutions and then a liquid of pure polymers, with all solvent removed. This
liquid is called a polymer melt, with monomer density ρmax . An important
measure of concentration is the density relative to that of the melt ρ/ρmax ;
86 Polymer solutions

this is called the volume fraction and is denoted φ. The overlap volume
fraction φ ∗ is clearly much smaller than the melt volume fraction—viz. 1.
Volume fraction is an unambiguous way to measure concentration, since it
does not depend on our definition of the monomer building block for the
chain. For typical hydrocarbon polymers, with molecular weight of 105 , φ ∗
is roughly 1%.

4.2.1 Structure
As the concentration rises above φ ∗ the spatial structure of the poly-
mers is significantly altered. We may express this structure in terms of
the local volume fraction analogous to the local density ρ 0 : φ(r) 0 ≡
ρ(r) 0 /ρmax . The r dependence of φ(r) 0 for r  R can be written
(A/r)d−D ; the length A depends on the type of polymer and solvent but
not on its length n. Thus e.g., for polystyrene in benzene A 1 nm. If
φ  φ ∗ , there is some distance r  R for which φ(r) 0 = φ. This dis-
tance, called the correlation length ξφ = Aφ −1/(d−D) , plays an important
role in describing the structure of the solution. If we consider a region
around a given monomer much closer than this ξφ , the local density is
much greater than the average density. The average density can thus have
little effect on the self-avoidance interaction. We infer that the portion of
the polymer in this region is but little affected by the other chains. However,
for distances larger than ξφ , the original φ(r) 0 is much smaller than the
average volume fraction φ: the chain passing through the origin contributes
only a tiny fraction of the overall density. Thus the other chains may alter
the self-avoidance effects strongly. The segment of chain within a distance
ξφ is often called a blob. Each chain thus may be thought of a sequence of
blobs, each having an internal structure that is relatively unaffected by the
other chains.
To see how self-avoidance works on scales much larger than ξφ , we con-
sider global contacts between two blobs that are far apart along their chain.
As shown in the preceding section, the swelling of a chain depends on
the mutual excluded volume b2 of its constituent monomers. Here we may
take our monomers to be blobs. To find the b2 of two blobs, we imagine a
dilute admixture one-blob chains in the semidilute solution of k-blob chains.
Because interpenetration of the blobs is significantly suppressed, this situ-
ation is similar to a gas of monomers in a lattice, with k-site molecules
filling all the other sites. We discussed the effect of such multi-site solvents
in Appendix B of the preceding chapter. There we found that the extended
solvent molecules reduce the b2 between single sites by a factor of k.
To gauge the importance of this reduced b2 , we consider the number
of effective contacts within a given chain in the absence of swelling. This
chain has a radius R of order ξ k 1/2 and a volume of order (ξ k 1/2 )3 . Each
of the k blobs has an effective intersection with roughly kb2 /R 3 other
blobs of the same chain. Thus the number of effective intersections is
k 2 b2 /R 3 . Accounting for the k dependence of R and b2 , this gives an
effective number of contacts of k 2−1−3/2 ∼ k −1/2 . The effective interac-
tion energy between the monomers is thus of order T k −1/2 —much smaller
than the thermal energy. The interaction energy is not sufficient to expand
 Semidilute solutions 87

the polymer significantly, since such expansion requires work of the order 1
of T . We infer that a chain has the structure of an unperturbed random
walk beyond the scale ξφ : it is like a random walk of blobs. The local

log ((r))
– 4/3
density thus behaves as shown in Fig. 4.2. The density from the chain at
the origin falls off as in a self-avoiding walk within the blob radius, and
like a simple random walk beyond. Monomers from other chains have a 
+0.8
suppressed local density near the origin. These dominate at distances bey- –1
ond the blob radius. The suppression as a function of distance also obeys a a  log(r)
power law ([1], chapter 13). Calculations suggest that its numerical value
is about 0.8.
By scattering one can verify the behavior anticipated above. One may
isotopically label a dilute fraction of the chains to measure the density profile Fig. 4.2
of individual chains in the solution. As anticipated, on scales smaller than ξφ Local volume fractions of monomer φ(r)
they have the fractal dimension of a self-avoiding chain (viz. 53 ), while over at distance r from an arbitrary monomer in
a semidilute solution of very low
distances much larger than ξφ they have the fractal dimension of a simple concentration φ. Solid curve: monomers
random walk (viz. 2) [1], as shown in Fig. 3.9. Each random-walk chain from the chain going through the origin.
has a self density of order R −1 . It is arbitrarily small as the chains are made Dot-dashed curve: volume fraction of
longer, with fixed total density. To achieve the given density with chains other chains. Scales are logarithmic, so
of increasing length there must be more and more chains in the volume R 3 that power laws appear as straight lines.
Within a blob size ξ the total volume
pervaded by one chain. The chains must interpenetrate. They interpenetrate fraction is dominated by the chain passing
more and more as φ increases above the overlap concentration φ ∗ . This through the origin. Beyond the distance ξ
interpenetration will become important when we come to consider how this the total volume fraction is dominated by
solution can flow. other chains. Simple dashed curve shows
excess density φ(r) 0 − φ(∞) 0 , which
decays exponentially to zero with a decay
4.2.2 Energy length ξs .

The blob picture is also useful for describing the interaction energy and the
osmotic pressure in the semidilute state. We saw above that interactions
between pieces of a chain larger than ξφ were unimportant relative to inter-
actions between different chains. Chemically distant parts of a given chain
might as well be different chains as far as the interactions are concerned.
This means that the interaction energy would be not much changed if the
chains were cut into pieces the size of a blob. But we can readily estimate
the energy of such cut-up chains, for these chains are at the overlap con-
centration φ ∗ . The distance between them is of the order of their size ξφ .
As we have noted at the beginning of this section, the interaction energy at
φ ∗ is of order T per cut-up chain piece—i.e., T per blob. The same is true
for the osmotic pressure:  T ξφ−3 . The interaction energy and pressure
are oblivious to the chain length in the semidilute regime.
According to this reasoning the quantity ξφ3 /T is predicted to be a
finite number of order unity regardless of the quantitative goodness of the
solvent, e.g., b2 /a 3 . In this sense it is like the quantity B2 /R 3 encountered
in the last section. And for similar reasons, this ratio too is universal. It has
been measured via scattering experiments for several polymers and several
solvents [3]. One finds /T = (3.2ξφ )−3 . Recalling that φ = φ(ξφ ) 0 =
(A/ξφ )−4/3 , one finds

/T = (3.2A)−3 φ 9/4 . (4.4)

88 Polymer solutions
100

Fig. 4.3
Reduced osmotic compressibility d /d ˜ φ̃
versus reduced volume fraction φ̃ = φ/φ ∗ Polyisoprene in cyclohexane
for various polymers and various good ~ ~ M = 640,000
d Π/d
solvents. For each polymer sample the ~
˜ φ̃ values are normalized so that they
d /d 5/4
approach 1 at low concentration. The φ̃
˜ φ̃ 2
values are normalized so that d 2 /d
approaches 2 at low concentration. Data
10
are shown for one sample of polyisoprene Polystyrene
in cyclohexane from [3] and three [Noda]
molecular weights of polystyrene from [4].
Theory [5] predicts that the pressure thus
normalized should be a universal function
of the normalized concentration. Theory
˜ φ̃ should scale as
also predicts that d /d
φ̃ p at high concentration, where the
exponent p is close to 54 . Thus
1
˜ φ̃)/φ̃ 5/4 should approach a (nearly)
(d /d 0.01 0.1 1 10 100
constant value at large φ̃. Both predictions ~
are well confirmed. 

Some experimental confirmation of this universality is shown in Fig. 4.3.

The figure also shows that the whole function (φ) is universal when
properly scaled.
4.2. Compressibility sum rule When scatterers are dispersed in a liquid, the small q
limit of S(q) has a general interpretation, even when the scatterers are inter-
acting strongly. In this limit, the scattered wave sees only the large-scale
modulation of the scatterer density. At such large scales the scatterers can be
regarded as a uniform density with slight independent random variations from
point to point. As we have noted, a uniform density cannot produce scattering,
but the slight variations can. According to Eq. (3.20)

  
d 3r
S(q → 0) = lim d 3 r  ρ(0)ρ(r  ) exp(i q · r ) .
M q→0

Here M is the number of scatterers in the sample. The limit as q → 0 is not

the same as q = 0. If q = 0 all the forward scattering along the original beam
is mistakenly counted and S(0) appears proportional to the sample size. To
eliminate this spurious contribution, we note that as long as q  = 0, the scattering
cannot be affected by adding a constant to ρ(r) (since (constant)e iq·r = 0).
 3
Accordingly, we replace ρ(r) by ρ(r) − ρ̄, where ρ̄ = M/ d r is the average
density of scatterers. With this choice, the limit q → 0 can be taken:
  
1
S(q → 0) = (ρ(0) − ρ̄) d 3 r  (ρ(r  ) − ρ̄) .
ρ̄

We consider the scattering from some large volume within the sample. The
number of scatterers N in this volume is evidently  d 3 rρ(r); its average is
 Semidilute solutions 89

N̄ = ρ̄. In terms of N ,
   
S(q → 0) = 1/ρ̄ (ρ(r) − ρ̄)(N − N̄ ) = (N − N̄ )2 /N̄ .

The scattering at q → 0 measures the mean-squared fluctuations of the average

density. (We have yet to show that this expression is independent of the volume
 chosen!)
These fluctuations can be related to thermodynamic quantities. We recall
that an average like (N − N̄ )2 can be expressed
   dN exp(−F (N )/T )(N − N̄ )2
(N − N̄ )2 =  ,
dN exp(−F (N )/T )

where F (N ) is the free energy of the fluid constrained to have N particles

in the volume . (It includes the work of pushing particles out of  into
the surrounding fluid. Thus it is not the free energy of an isolated sys-
tem of N particles in volume .) This F (N ) must be proportional to the
volume : F (N ) = f (ρ); since  is large, only slight fluctuations of N
around N̄ will be significant. For a small range of N around N̄ F (N) can
be expressed in the form F (N ) = (f (ρ̄) + 12 f  |ρ̄ (ρ − ρ̄)2 ), or equival-
ently, F (N ) = (a + b(N − N̄ )2 / 2 ). (There is no term proportional to f  ,
because (N − N̄ ) = 0.) (a) Express (N − N̄ )2 in terms of the coefficient b
and the temperature T . It is convenient to use the equipartition theorem from
Chapter 2. (b) Express the osmotic pressure (ρ) = −∂ F /∂ at constant N
in terms of f (ρ), f  (ρ), etc. for general ρ. (c) Relate d/dρ to f and its
derivatives. (d) Combining the results of (a) and (c), show that S(q → 0) =
T (d/dρ)−1 .

4.A Suggested experiment: Turbidity of polymer solution If a medium has a

compressible density of solute, it will scatter light, as the previous problem shows.
The data in Fig. 4.3 was obtained using this connection. As we have shown, the
compressibility of a polymer solution goes down when the concentration goes up.
This means that it should scatter less light. One can see this scattering by looking
through a polymer solution. It is not completely transparent, but is a little milky or
turbid. This turbidity shows up in another way when you shine the beam of a laser
pointer through the solution. The beam is visible from the side. The intensity of the
side-scattered light is proportional to S(q), where 1/q is a hundred nanometers or
more. Since this 1/q is much larger than the correlation lengths ξ in the solution, we
can treat q as being essentially 0. Thus the side-scattered intensity is proportional
to the osmotic compressibility of the solution. Specifically, if ρ is the number of
scatterers per unit volume and  is the volume of the beam, the number of scatterers
is ρ and the scattered intensity I ∝ (ρ)S(q). If we shine a fixed beam through
solutions of greater and greater concentration ρ, S(q) = T /(d/dρ) ∼ ρ −5/4 ,
as shown in Fig. 4.3. Thus I ∼ ρ −1/4 : the scattered light gets weaker as the
concentration increases.
You could use an inexpensive laser pointer and gelatin solutions of several con-
centrations to verify this predicted behavior. You could make up several gelatin
solutions of widely varying concentrations. (Keep them warm so they do not gel.
Gelation causes extra scattering not accounted for in the theory.) To measure the
intensity you could use a light meter from a camera. You could also place the two
samples side by side so that the laser beam shone through both of them equally. One
will scatter more strongly than the other. Then reduce the beam intensity through the
strongly scattering medium until the two scattered intensities look equal. You can
reduce the beam by masking off a known fraction of it. Another method is to look at
the light from different angles. Though S(q) should be independent of angle, beam
looks brighter from lower angles because you are looking at more scatterers in a
90 Polymer solutions

given projected area. That is, the intensity seen within the beam goes as 1/ sin(θ),
where θ = 0 is the beam direction. You could find a way to illuminate the two
samples and view the two beams at different angles until their intensities looked
the same.

4.2.3 Concentrated solutions and melts

As the concentration is increased further and further above φ ∗ the correlation
length shrinks progressively. Ultimately there comes a point where the blobs
may no longer be regarded as asymptotically long self-avoiding chains. This
is the regime of concentrated solutions. If the monomer excluded volume
b2 is small, a chain must reach a certain size called ξT before it shows the
self-avoiding properties. This size generally depends on the temperature
† The thermal blob size can be inferred because b2 does; accordingly it is called [6] the thermal blob† . Even in a
from b2 by requiring that its interaction good solvent with large b2 , our asymptotic picture of the blob breaks down
energy be significant relative to the elastic
on small length scales. The number of independent steps in the random walk
energy of the unperturbed blob—i.e., 1:
T n2 b2 /ξT3 T . Here n is the number of is no longer large. And the concentration within the blob is no longer small
monomers in the thermal blob: ξT2 = na 2 . enough to treat the monomers as a dilute solution. At these concentrations
The result is ξT a 4 /b2 . This same cri- the general and universal behavior of semidilute solutions breaks down
terion was used in the preceding chapter in somewhat. As the volume fraction of solvent becomes small, the osmotic
the discussion of theta solvents to determine
pressure required to remove the remaining solvent rises dramatically, until
when the b2 of a whole chain was sufficient
to swell it appreciably. in the melt state the liquid becomes essentially incompressible. But the lack
of self-avoidance continues to hold: these chains remain simple random
walks on all but the smallest length scales.

4.3 Motion in a polymer solution

A large fraction of the distinctive practical behavior of a structured fluid
involves the way it flows. It is now time to discuss the molecular motion that
gives rise to these distinctive flow phenomena. We must first understand the
molecular motions in a quiescent fluid in equilibrium. Before considering
polymers we treat the simpler question of how a colloidal sphere of radius
R moves.

4.3.1 Brownian motion of a sphere

An object in a fluid moves in response to the thermal fluctuations of the
fluid around it. In this section we describe the amount of motion in terms
of the viscosity of the solvent. Since the random motions of the solvent
are statistically independent over long times, the long-time motion is a
random walk. From our discussion of random-walk polymers, we have
seen that the long-time behavior depends only on the mean-squared dis-
placement r 2 1 of the individual steps. Accordingly, we are free to take
these steps in a convenient, schematic way. We imagine
 that the particle
moves in discrete steps on a cubic lattice of spacing r 1 ≡ b. For many
2

purposes it is convenient to define the probability current density j(t, r):

††We encountered this current density in the average number of walkers crossing an imaginary surface at r per unit
the problem on Monte Carlo sampling in area in a time step t †† . For definiteness we imagine a horizontal plane
Chapter 2. lying between two lattice planes. We imagine some smooth distribution
 Motion in a polymer solution 91

b3 p(t, r) giving the probability that the walker is at position r on the lat-
tice. If the walker is just above our imaginary plane, it has a probability 16
of passing down through the plane. The overall probability that this step
happens is 16 b3 p(t, r). On the other hand if the walker is just below the
plane, at r − b, the net probability that a particle moves up through the
plane is 16 b3 p(t, r − b). The probability current b2 jz moving through the
plane at this point is the difference of these two probabilities. Since p is
assumed to vary smoothly on the scale of a lattice step, we can express
b2 jz = − 16 (b3 /t)b∂p/∂z. Considering the current in arbitrary directions,
we have
† A comment
  about the connection between
j = − 16 b2 /t ∇p.
 (4.5) mean-squared displacement and ζ is in
order. Our random walker takes indepen-
dent steps of length b at every time interval
The coefficient in [ ] is called the diffusion constant ζ † . t. Thus the mean-squared displacement
To characterize the brownian motion further, we now ask what controls r 2 (t) = b2 t/(t). But a diffusing particle
the size of ζ . We can now consider the brownian motion of a simple particle, with diffusion constant ζ by definition has
a colloidal sphere of radius R. The solvent around such a sphere may be a mean-squared displacement in a given
direction of x 2 (t) = 2ζ t. We may also
treated as a continuum fluid with a velocity field v(r , t). The fluid velocity
recover the familiar diffusion equation for
on the boundary of the sphere must equal that of the sphere, but elsewhere the probability p(t, r), by noting that the
it may fluctuate because of thermal excitation: any configuration v(r ) has change of p at r during a time step is the sum
a probability given by the Boltzmann principle discussed in Chapter 2. of probability currents j onto site r during
Fortunately, one may determine the diffusion constant ζ without treat- that timestep: p = t∇ · j = tζ ∇ 2p.
This equation has the same form as the
ing these fluctuations explicitly, by means of a trick discovered by Albert
random-walk probability Eq. (3.12). See
Einstein [7]. He noted that there is a rigorous connection between ζ of Table 4.1.
any object and the drag coefficient  describing the force required to pull
it along at some small speed v †† . To understand the Einstein relation, we †† This Einstein relation is a special case
imagine that our brownian particle is subjected to a weak gravitational of the fluctuation-dissipation theorem relat-
ing any small perturbation requiring work
field that exerts a force of magnitude F . As a result of this field we antic-
to a fluctuating quantity in the unperturbed
ipate that the particle gradually moves downward with an average speed system in thermal equilibrium [8].
v proportional to F ; this proportionality defines the drag coefficient :
F = v . Thus if the particle is at r with probability p(r), this motion
gives rise to a downward current density j (r) = p(r)v = −p(r)F / .
After the gravitational force has acted long enough for the colloidal parti-
cles to come to equilibrium, p(r) becomes a constant distribution given by
the Boltzmann Principle: p(r) = Z −1 exp(−F z/T ). Here F represents the
magnitude of the force, and z the height; the Z is a normalization constant.
Since this p(r) is not uniform, it gives rise to a diffusing upward current
jζ = −ζ ∇p, as discussed above. If gravity were suddenly turned off, one
could immediately observe this current: jζ = ζ F /Tp. But in the equi-
librium state with gravity acting, there must be no net current (otherwise
p(r) would not be constant in time): jζ + j = 0, or ζ F /Tp = pF / .
We conclude that (a) the induced speed v is indeed proportional to F as
anticipated, and (b) its magnitude is given by  = T /ζ .
One may calculate the drag coefficient  using conventional hydrody-
namics [9] and thus account for both drag and diffusion. But the basic
behavior of  can be understood without hydrodynamic formalism. We
first recall the relationship between force and velocity v(r ) in a simple shear
flow, as discussed in Chapter 2. Here the flow was in the x direction but
varied only in the z direction. As we discussed, such flows transmit forces
92 Polymer solutions

between adjacent fluid elements. We may think of a fluid element as a tiny

cube of side b. A surface of a fluid element perpendicular to z experiences
a force per unit area in the x direction σzx = ηs ∂vx /∂z. Each fluid element
has momentum b3 ρs v, where ρs is the mass density of the fluid. Any net
force on the element causes this momentum to change: ∂ρs vx /∂t = ∂σ/∂z.
In terms of the viscosity this gives ∂vx /∂t = [ηs /ρs ]∂ 2 vx /∂z2 . This is a
diffusion law like that describing the motion of a random walker. It says
that momentum diffuses in directions perpendicular to the momentum itself;
the diffusion constant ηs /ρs is proportional to the viscosity and is called the
kinematic viscosity.
4.3. Sedimentation Polystyrene has a density a few percent greater than that of
toluene. Assume it is 5% denser. This means that a polymer in toluene is pulled
downward by gravity. The point of this problem is to see how fast. (a) Write a
formula for the sedimentation rate v in terms of the hydrodynamic radius Rh ,
the acceleration of gravity, the viscosity ηs of toluene, and the buoyant mass
Mb of the polymer. You may assume, as will be shown later, that Rh is a fixed
fraction of the radius of gyration Rg from Chapter 3. How does this rate scale
with molecular weight? Do bigger polymers fall faster or slower? (b) Do a
numerical estimate using 10-million molecular weight polymers. These each
displace a volume Vd of about 17 million Å3 or 17,000 nm3 in the absence of
solvent. Using the formula for local volume fraction φ(r) = (0.7 nm/r)4/3 ,
estimate the radius of the chain. (c) Roughly how long would it take such
a chain to travel 1 cm in an ultracentrifuge where gravity is effectively
multiplied by 105 ?
The transmission of momentum is more complicated when it originates
at a point as with our colloidal particle. To analyze this case, we con-
sider a small downward impulse applied at the origin at time t = 0. Thus
the velocity field is initially a delta function in space. The momentum
must propagate out to infinity. It is convenient to think of this momentum
by
 expressing the velocity field as a sum of plane waves: v(r, t) =
v (t)e ik·r . For transverse components, with v  we may use the
k ⊥ k,
k k
momentum diffusion equation of the last paragraph: ∂vk /∂t = ηs /ρs k 2 vk .
However, these transverse components are not sufficient to produce our ini-
tial impulsive disturbance. Part of this disturbance is longitudinal, having
pieces with vk  k.  Longitudinal waves clearly cause local compression
of the liquid. To know how such a disturbance spreads, we would have to
know what forces result from compression. That is, such spreading depends
on the compressibility of the fluid; viscosity is not sufficient to describe it.
But we know from elementary physics what happens when a medium
is locally compressed: sound waves are generated. Thus two qualitatively
different types of propagation occur in response to our impulse. The longi-
tudinal part of the disturbance propagates out at the speed of sound. Sound
waves can carry appreciable momentum if the source moves at speeds near
the speed of sound. But for the gentle motions of interest here, the pro-
duction of sound, and the compression of the fluid, are insignificant. The
fluid may be treated as incompressible. The transverse part propagates by
diffusion. As with all diffusion it covers a distance r in a time proportional
to r 2 . In the initial delta-function disturbance all wave components vk are
equal. For each k there are two transverse amplitudes and one longitudinal
one, only the transverse ones survive in an incompressible fluid.
 Motion in a polymer solution 93

We now consider the velocity field from a particle subjected to a constant

downward force F in a quiescent fluid. One can determine this field by
adding all the transverse wave contributions as described in the previous
paragraph. Instead, we may consider the constant force as a sequence of
many impulses. The momentum from a given impulse reaches r in a time
t r 2 /(ηs /ρs ). Thus the momentum contained in a sphere of radius r is of
order F t = F r 2 ρs /ηs . (Over longer times the momentum begins to flow
out of the sphere as fast as it is injected
 within it.) In terms of the velocity
field, this momentum is of order r  <r d 3 r  ρs v(r  ). Taking derivatives with
respect to r, we infer F rρs /ηs r 2 v(r), or v(r) F /(ηs r). The exact
form of v(r) is known as the Oseen tensor [5].
|F |    
v(r) = F̂ + F̂ · r̂ r̂ . (4.6)
8π ηs r

The F̂ and r̂ are unit vectors in the F and r directions. The factor in ( )
depends only on the angle the vector r makes with v and with F . It is a
consequence of removing the longitudinal part of the v field. If we consider
only the downward velocity in response to a downward force, it is twice as
strong directly above and below the source as it is on the sides. There are
also velocities directed perpendicular to the force F . The fluid flows around
the particle. We shall need the Oseen tensor to analyze how the solvent
flows near tenuous objects like polymers. A mathematical explanation of
the Oseen tensor is given in Appendix A.
To find the drag force, we may equate the rate of work done by the force
with
 3 the power P dissipated in the viscous flow. This latter has the form
d rηs γ̇ 2 . Here γ̇ is the shear rate introduced in Chapter 2; in a simple
shear flow, it is the derivative of the velocity. In more general flows it is a
combination of derivatives of the velocity. In the Oseen field γ̇ dv/dr
F /r 2 . Thus the power density

ẇ(r) = ηs γ̇ 2 F 2 /(ηs r 4 ).
The power dissipated beyond some radius R in the Oseen field is of order
 ∞
d 3 r ẇ(r) F 2 /(ηs R).
R

The power is dissipated predominately at short distances R of the order

of the object size. Since the force does work at a rate F v = F 2 / , we
infer F 2 /  = P F 2 /(ηs R), so that 1/  1/(ηs R). For a sphere, one
may calculate the dissipation rate exactly to obtain  = 6π ηs R. Using the
Einstein relation described above gives the Stokes formula for the diffusion
constant [9]
ζ = T/(6π ηs R). (4.7)
The diffusion constant scales inversely with the sphere size. This scaling is
an inevitable consequence of dimensional analysis, once we recognize that
only the temperature, the viscosity, and the radius can enter the formula.
Most simple liquids have approximately the same viscosity at room tem-
perature, as discussed in Chapter 2. Thus all particles of a given size diffuse
94 Polymer solutions

at about the same rate in most solvents. The following benchmark is a

convenient way to remember this rate. A sphere of radius R = 11 nm
in water diffuses its own radius in 1 μs. In this time, its mean-squared
displacement is R 2 .
This section has shown that diffusive motion of a particle follows the
same quantitative laws as a random-walk polymer. Moreover, the flow of
momentum away from a source is also governed by a diffusive law, if one
properly considers only flow transverse to the momentum itself. Thus each
aspect of random-walk polymers that we have a studied tells a corresponding
aspect of diffusive motion and of flow. I have compiled a dictionary of
corresponding properties in Table 4.1.

4.3.2 Intrinsic viscosity

A large object in a liquid influences motion in another way: by increasing
the viscosity of the solution. As we saw in Chapter 2, a shear flow is char-
acterized by the shear rate γ̇ = dv/dz, as sketched in Fig. 4.4. With no

Table 4.1 Three types of diffusion in d dimensions

Quantity Random-walk Diffusing Diffusing

polymer particles momentum

Dependent variable Probability p(n, r) Particle density ρ(r) Momentum densitya ρm v

Independent variable Monomer number n Time t Time t
r 2 1
Material constant Diffusion constant ζ Kinematic viscosity ν
2d
Equation ∂p/∂n = (r 2 1 /2d)(∇ 2 p) ∂ρ/∂t = ζ ∇2ρ ∂v⊥ /∂t = b ν∇ 2 v⊥
Mean squared distance from r 2 1 n 2d ζ t 2d νt
point source
Local density (d = 3) Monomer density at distance r Particle density at distance r Velocity at distance r from
from an arbitrary monomer from a steady point source of i particle dragged with constant
3/(πr 2 1 r) particles per unit time i/(2π ζ r) force F
(ρm |F |/8π νr)[F̂ + (F̂ · r̂)r̂]

Notes:
a ρ is mass density of the fluid.
m
 composed of wave vectors
b The symbol v has to be interpreted using the Fourier transform method discussed in Appendix A. It is that part of v
⊥
transverse to v.

Fig. 4.4
Extra viscous dissipation from a sphere in
a liquid. Arrows show unperturbed
velocity field, with shear rate γ̇0 . Graph at
right shows dissipation rate along the
vertical dotted line through the sphere. Far
from the sphere the rate ẇ(z) has the
0
unperturbed value ẇ0 . Near the sphere the z
shear rate and dissipation rate are 2R
enhanced. Within the sphere, they vanish.
The
 extra dissipation due to the sphere,
(ẇ(r) − ẇ0 ) d 3 r is suggested by the ·
shading.
w(z) w· 0
 Motion in a polymer solution 95

solute present, the shear has a uniform value γ̇0 everywhere and a uniform
dissipation rate ẇ0 = ηs γ̇02 . A solid object such as a sphere perturbs this
flow and increases the dissipation. With such objects present, greater dissip-
ation occurs for a given macroscopic flow rate: the macroscopic viscosity is
increased. This increase is a major way of sensing the presence of structures
in a fluid. To understand how this perturbation works, we consider a single † In a viscous liquid, and many other sys-
colloidal sphere of radius R in a shear flow between two horizontal sliding tems where the forces are proportional to
plates. The solvent is obliged to flow around the sphere. This increases the the velocities, the energy dissipation rate
shear rate γ̇ outside the sphere. The sphere alters the shear flow from its ini- must be a quadratic function of the velocit-
tial uniform state of minimum dissipation; thus it increases the dissipation† . ies. For such systems the steady-state velo-
The dissipation per unit volume ẇ(r) is zero inside the sphere. But ẇ(r) city field is that which minimizes this energy
dissipation rate. This can be proven from the
must be increased just outside the sphere. For example, the integral of the equations of hydrodynamics, which amount
shear rate γ̇ along the vertical line through the center is the difference in to Newton’s laws of motion applied to these
speed of the two plates, as for any other vertical line. Since the shear rate linear forces ([10], Section 4.6).
vanishes within the sphere, it must be greater than γ0 above and below in
order to compensate. The perturbation in velocity v(r) caused by the sphere
resides mostly near the sphere. It can be thought of as the sum of many
Oseen tensors centered at each point of the sphere. Since these produce a
velocity field that falls off as 1/r, the resulting shear rate γ̇ (r) = dv/dr
falls off as 1/r 2 as discussed above in connection with the drag force. In
order to make up the deficit in shear rate within the sphere, γ̇ /γ̇0 must be
significantly greater than 1 for distances r R. In this same region the
dissipation ẇ(r)/ẇ0 is also significantly greater than 1. Thus the effect of
the sphere is to increase the dissipation by a finite factor
 in a region of the
order of the sphere size. That is, the extra dissipation d 3 r(ẇ(r) − ẇ0 ) is
roughly ẇ0 R 3 .
By accounting for the excess dissipation, we may find the increased
viscosity. On the one hand, the viscosity is defined by ẇ = ηγ̇ 2 , where γ̇
is the average velocity gradient. On the other hand if the fluid has volume ,
     
3 ẇ(r) 3
ẇ = ẇ0  + (ẇ(r) − ẇ0 ) d r = ẇ0  + −1 d r ,
ẇ0
Here the first term is the dissipation due to the unperturbed flow throughout
the fluid. The second term is the excess dissipation near the sphere. The
integrand is significant over a volume comparable to the sphere volume
V = 43 π R 3 . The integral can be found explicitly in terms of the known

velocity field around the sphere [9,11]. The result is (ẇ(r)/ẇ0 −1) d 3 r =
5
2 V . The dissipation from a sphere is the same as if the dissipation were
doubled over a volume equal to 52 that of the sphere. Comparing the two
expressions for the dissipation, we infer
 
5V
η = ηs 1 + .
2
If the solution contains N spheres far from each other, the dissipation is the
sum of that due to each, so that

 2 
5 NV N
η = ηs 1 + +O . (4.8)
2  
96 Polymer solutions

The quantity in [. . .] is the volume-fraction of spheres φ. The relative

increase in viscosity measures the fraction of volume in which flow is sig-
nificantly perturbed. The viscosity senses only the total perturbed volume,
and cannot discern the number of solute objects. The effect of a solute on
† Here we depart slightly from the standard the viscosity is characterized in general by the intrinsic viscosity [η]† :
definition used in physical chemistry books.
In the standard definition the volume frac-
tion φ is replaced by the concentration in η ≡ ηs (1 + [η]φ + · · · ). (4.9)
mass per unit volume.
For an object other than a hard sphere, one may define a viscometric radius
Rv as the radius of a sphere whose effect on the viscosity is equal to that of
the object. Comparing the definition with that of Rv , we find
 
3
[η] = 5
2
4
3 π Rv /Vd , (4.10)

where Vd is the volume displaced by the object (the increase in total

volume upon adding one object). Thus [η] measures the ratio of the
hydrodynamically perturbed volume around an object to its displaced
†† Analogous behavior occurs for a scalar volume Vd †† .
diffusing field, such as charge carriers in
a conductor. Here the current j is driven
by the electric potential gradient E. If a 4.3.3 Polymer in dilute solution: hydrodynamic opacity
perfectly conducting sphere is embedded in
a slab of material, there is increased cur- Flow near a polymer arises in the same way as that near a hard sphere. But
rent density j and dissipation around it, the tenuous structure and deformability of a polymer make it unclear how
and the average conductivity σ defined by
ẇ(r) = σ E 2 , increases. The overall con-
much the polymer flow will resemble the hard sphere flow. We find below
ductivity increases by a factor which is a that polymers act remarkably solid in their hydrodynamic interaction with
universal constant times the volume frac- the solvent.
tion of embedded spheres. This behavior A polymer, like a hard sphere, experiences a drag force if it is pulled
depends on the dimension of space. In two through the solvent. If the pulling speed is slow enough, the drag force will
dimensions, the effect of an embedded disk
is not strongly confined to the disk region
be much smaller than the force T/R required to deform it, and there
and the conductivity due to it is much lar- will be negligible deformation. For concreteness it is helpful to model the
ger than its area fraction. The same is true polymer as a string of n beads, each with radius b and each separated from
for flow in two dimensions. [The author its predecessor by a phantom bond of length a  b. The phantom bonds
thanks J. Wyman and I. Cohen for a useful do not interact with the solvent, but each bead acts like a point source of
discussion on this point.]
momentum, as we have discussed above. To estimate the drag force, we
shall use a reference frame moving downward with the average speed of
the polymer. In this frame the fluid has some asymptotic (upward) speed
v∞ . We begin by supposing that the polymer is transparent to the flow, so
†††
This assumption is called the Rouse that the flow passes through the polymer coil with little perturbation††† .
model [12] for flow near a polymer. Then the speed near each bead is of order v∞ and the downward forces F
transmitted into the fluid are all comparable.
Under this supposition we may readily find the amount of backflow
induced by the beads. Each bead i induces a velocity field vi (r) given
by the Oseen tensor. In order of magnitude vi (r) F /(ηs (r − ri )). The
upward velocity at some point r within the polymer is thus of order

 
F
v∞ − vi (r) v∞ − d 3 r0 ρ(r0 ) .
ηs (r − r0 )
i
 Motion in a polymer solution 97
Table 4.2 Universal ratios of length dimensions of a dilute polymer solutiona Notes:
a Adapted from [14], Table VI, using their
Ratiosb RG /Rh  /R
RG Rt /Rh Rv /Rh
h original data and earlier data they compiled
from various sources.
Hard sphere (3/5)1/2 = 0.775 1 1 1 b R is the hydrodynamic radius—the radius
h
Ideal chain (theory)c 1.48 1.91 0 1.23 of a hard sphere with the same diffusion
Self-avoiding chain (theory)c 1.56 2.01 1.04 1.14 constant as the polymer. RG is the radius
Polyisoprene in cyclohexaned 1.39 1.79 0.95 1.11 of gyration, as defined in Problem 3.11.
Polystyrene in benzened 1.51 1.94 1.01 1.03 RG  ≡ ( 3 )−1/2 R is the radius of a uniform
5 G
Poly(α-methylstyrene)d 1.55 2.0 1.05 1.11 sphere with the same RG as the polymer.
Rt is the thermodynamic radius, that of hard
spheres having the same excluded volume
B2 as the polymers. Rv is the viscometric
If we take as our origin some bead of the polymer, then on average ρ(r0 ) = radius, that of hard spheres having the same
ρ(r0 ) 0 ∼ r0 D−3 , where D is the fractal dimension of the polymer. Thus intrinsic viscosity as the polymers.
c These theories estimate the given ratios by
 R treating  ≡ (d − 4) as a small parameter
v(r) − v∞ ∼ d 3 r0 r0 D−3 (r − r0 )−1 . and calculating the given ratio to first or to
second order in .
d These experiments used scattering and

The integration is now of the same form as Eq. (3.26) for MAB , the viscometry to measure the various R values
for different molecular weights. The lim-
number of intersections of two fractals with dimensions D and 2 in three- iting ratios for high molecular weights are
dimensional space. For all fractals with D > 1, including our polymer, the reported.
integral goes as a positive power of R. We say that such fractals are opaque
to the flow. We note that opacity is attained for large enough R no matter
how small the individual beads were or how weakly they interacted with
the fluid.
The result of this flow opacity is to invalidate our supposition that each
bead feels the unperturbed speed v∞ . For this leads the conclusion that
v(r) − v∞ diverges with R. In that case the backflow would be stronger
than the asymptotic speed v∞ . To avoid this absurd result, we conclude that
the fluid speed within the fractal is actually much smaller than v∞ , so that
the forces Fi on each bead are also smaller than an isolated bead would feel.
Thus, the fluid cannot move transparently through the fractal; the flow must
go around it. The flow speed throughout the fractal volume is suppressed
by at least a finite factor.
Because a fractal screens out the flow within itself qualitatively, it also
alters the flow far away. The far-field velocity must extrapolate to a finite
fraction of v∞ at r = R. This means that the far field is that of a hard sphere
whose radius is of order R † . Thus a fractal with D > 1 in a uniform flow †This hard-sphere behavior was first pro-
causes dissipation like that of a hard sphere whose radius is of order R. posed by Zimm, and is called the Zimm
model [13].
Accordingly, its drag coefficient  or hydrodynamic radius Rh are equal to
R up to a finite factor independent of the microscopic structure. A polymer
must also perturb a shear flow like a hard sphere of comparable radius.
Thus its effect on the viscosity is like that of a hard sphere whose radius Rv
is of order R. Like other asymptotically finite ratios we have encountered
above, the Rh , Rv as well as the thermodynamic Rt defined above are uni-
versal multiples of the radius of gyration RG for any self-avoiding polymer.
Table 4.2 summarizes the experimental data on these ratios. Remarkably
these universal ratios are not far from 1.
4.4. Diffusive analog of hydrodynamic opacity We saw in this section that tenuous
fractals can interact strongly with each other by causing the surrounding solvent
to move. We suggested that this interaction could be understood in terms of
the mutual opacity of two fractals. This analogy is clearest if we consider the
98 Polymer solutions

diffusive analog of hydrodynamics. Suppose a region has a uniform density

of diffusing particles (random walkers), of density u in three dimensions. A
fractal of radius R and dimension D is placed at the origin, as shown above. This
fractal has the property of absorbing any diffusing particle that touches it. In the
presence of this absorption, the density u is no longer constant: u(r) < u(∞).
The picture shows the tracks of many random walkers. All tracks that intersec-
ted the fractal have been removed, leaving a depleted region around the fractal.
The relative density u(r)/u(∞) is simply the probability that an (indefinitely
long) random walk ending at r within the fractal does not intersect the fractal.
(a) Using the formula for the intersection of two fractals, show that the relative
density at the origin approaches one as R → ∞ if the dimension d of space is
high enough. How high must d be? For such cases, we may say that the fractal
is transparent to a diffusing substance. In the complementary case, the relative
density is reduced within the fractal by a factor of order unity (at least). In such
cases, we say that the fractal is opaque to a diffusing substance. (b) Consider a
point r  R. A walker at r has a small probability of intersecting the absorber,
and thus a small probability of being removed. This probability is proportional
to its local density at the origin. From this fact, find how u(∞) − u(r) varies
with r.

4.3.4 Internal fluctuations

In addition to their overall diffusive motions, polymers have spontan-
eous internal motions. These motions can readily be seen by dynamic
scattering. If one scatters from dilute polymers at a scattered wavelength
λ ≡ (2π )/q  ξφ , one observes a time-dependent relaxation, as we will
discuss in the semidilute diffusion subsection. The characteristic relaxation
time τ (q) for such scattering is roughly the time for a given scatterer to
move a distance λ via internal Brownian motion. The dominant motion for
scales of size λ is for the scatterer to move in concert with a segment of
size λ. The time τ (q) is thus similar to the time for a sphere of size λ to
diffuse its own diameter, i.e.,

τ (q) λ2 /ζλ ηs λ3 /T ηs q −3 /T . (4.11)

This predicted dependence is well confirmed by experiments [15].

4.3.5 Hydrodynamic screening

The suppression of flow within a polymer coil shown above is an example of
hydrodynamic screening. This screening is essential for describing motion
in a semi-dilute polymer solution. As we have seen above, hydrodynamics
describes the flow of a conserved quantity: transverse momentum. If some-
thing is present in the fluid to absorb this momentum, screening results.
We now illustrate this effect with a quantitative example. We imagine a
standard shear flow between two horizontal plates at separation h, when
the lower one slides from left to right with a speed v(0). In an ordinary fluid
this leads to a horizontal velocity v(z) = v(0)(1 − z/h). Now we place
a number of tiny spheres in the fluid and hold them stationary by some
means. The fluid is not trapped at all by these spheres, but their drag forces
take momentum from the fluid. A small volume b3 experiences a drag force
proportional to v(r) and to the number of spheres inside. The force can be
 Motion in a polymer solution 99

expressed as −b3 ρ2 v(r), where ρ2 is the density of spheres. The other
force on this volume element is the viscous force ηs b2 dv/dz on the bottom
and the top of the element. In steady state, if the fluid is not accelerating,
these forces must balance: ηd 2 v/dz2 = ρ2 v. This familiar equation has
solutions exponentially growing with z or exponentially decreasing. The
physical solution is the decreasing one† : v(z) = v(0) exp(−z/ξh ), where † Strictly, some of the exponentially grow-
ξh−2 = ρ2 /ηs . The flow is now confined to a region of size ξh near the ing solution is needed to satisfy the bound-
ary condition v(h) = 0. Here we restrict
bottom plate. Most of the momentum is absorbed before reaching the top ourselves to a thick layer with h  ξh , so
plate. The same hydrodynamic screening length ξh appears when many that this part is negligible.
such one-dimensional flows are combined to make more general flows. For
example the Oseen tensor describing flow around a moving sphere acquires
a damping factor exp(−r/ξh ). Thus if a small sphere is moved through a
fluid containing stationary spheres, the flow around it is not much perturbed
for distances r  ξh . But for distances r ξh , the velocity is reduced by
a substantial factor.
If a small sphere is moved through a semidilute polymer solution, hydro- †† You may wonder what the polymers do

dynamic screening also occurs, since the polymers absorb momentum being with the momentum they have absorbed.
The polymers are not attached to the con-
supplied by the moving sphere†† . If the small sphere is inserted at random
tainer, so they cannot absorb momentum
into the fluid, the typical distance to a monomer is the correlation length without starting to move themselves. We
ξφ introduced above. Thus if one examines the flow at distances r  ξφ , will consider the fate of the momentum
there are essentially no monomers there to impede it; accordingly, the flow absorbed by the polymers in Problem 4.8.
is unscreened. However, at distances r ξφ the polymers alter the flow.
If our sphere were near an isolated polymer of radius ξφ , the flow would
be damped appreciably within the coil volume because of its opacity, as
discussed above. The same is true for a solution of polymers of size ξφ at
their overlap concentration φ ∗ . It is still true if these polymers are connected
together to form a semidilute solution with the same φ and ξφ . At distances
of order ξφ from the sphere, there is an appreciable chance that monomers
are present to impede the flow speed by a finite factor. Since screening
reduces the flow by a finite factor at r ξφ , we infer that the screening
length is of order ξφ . At distances much larger than ξφ , the polymer solution
may be regarded as a more-or-less uniform mass of momentum-absorbing
blobs. Thus exponential damping of the flow is expected, as in the last
paragraph, with screening length ξh ξφ .

4.3.6 Semidilute diffusion

With this knowledge of hydrodynamic screening we may understand the
large-scale thermal motion in a polymer solution. In a semidilute solution,
one may distinguish two different kinds of motion. One is the overall relative
motion between the polymers and the solvent. The other is motion of a given
polymer relative to the solution. We treat the overall motion first.
Whenever the monomer concentration profile ρ(r) in the solution is not
uniform, a flow of polymer and solvent occurs so as to restore uniformity.
This flow is described by a current density of monomers j (r) analogous to
the probability current defined above in the discussion of brownian motion
of a sphere. If the nonuniformity is small, we expect the current to be
proportional to the departure from uniformity, so that j = −ζc ∇ρ, again
like the probability current. Thus the amount of motion is described by a
100 Polymer solutions

diffusion constant ζc ; this is called the cooperative diffusion constant. It is

called cooperative since it describes the collective motion of many chains.
As with the simple diffusion discussed above, the cooperative diffusion can
be described in terms of the local concentration ρ or the volume fraction φ
alone:
∂φ/∂t = (−∇ · j )/ρmax = ζc ∇ 2 φ. (4.12)
We now ask how ζc should behave as concentration increases from the
dilute limit. If a large-scale density gradient is produced in dilute conditions,
the relaxation of the gradient occurs by the independent brownian motion of
the polymers. Then cooperative diffusion is no different from the diffusion
of the individual polymer chains. This remains true as long as each polymer
moves independently of the others. But as φ approaches φ ∗ their interactions
become appreciable. The osmotic pressure variation caused by the gradient
is increased by the polymer interactions by a factor of order unity (in a
good solvent). The interaction produces a stronger average force on each
polymer away from regions of high density. This effect tends to speed up the
diffusion. However, the motion of each polymer is inhibited somewhat by
its neighbors. The thermal motion of the fluid around each one is damped by
hydrodynamic screening from these neighbors. Still, both of these effects
can only change the cooperative motion by a finite factor near φ ∗ . Thus

ζc (φ ∗ ) ζ T/(ηs R). (4.13)

From this fact we can infer the behavior of ζc far above the concen-
tration φ ∗ . We may attain this semidilute regime starting from a solution
near φ ∗ by joining chains together, as we have done previously. Then each
of the original chains becomes a blob of the solution, and each original
chain’s size R is now of the order of the blob radius ξφ . The joining of these
polymers does not change the distribution of monomers in space very much;
accordingly the solvent is free to flow around and through the blobs about as
much as it could with the original solution before joining. Thus a gradient of
density produces a current of monomers of about the same size whether the
polymers are joined or unjoined. This means that the cooperative diffusion
constant ζc is about the same in the two cases:

ζc T/(ηs ξφ ) ∼ φ 3/4 . (4.14)

Significantly, ζc , like the correlation lengths and the osmotic pressure, is

independent of chain length in the semidilute regime.
Cooperative diffusion can be measured by imposing a macroscopic con-
centration gradient and monitoring its decrease over time by some means.
But it can also be observed directly by scattering from the equilibrium
solution. Any instantaneous density profile ρ(r) can be expressed as a
superposition of plane waves ρ(r) = ρ̃
q q e iq·r , for some given com-

plex amplitudes ρ̃q . Applying the diffusion law of Eq. (4.12) to ρ(r) we
find an equation for the time derivative of ρ̃(q): ∂/∂t ρ̃(q) = −q 2 ζc ρ̃(q).
The amplitude evidently decays exponentially with a decay time τ (q) =
(q 2 ζc )−1 . As shown in the last chapter, waves scattered at wavevector q
have an intensity I (q) proportional to the square of this amplitude; thus,
 Motion in a polymer solution 101
Table 4.3 Universal semidilute length ratios inferred from polystyrene solutionsa Notes:
a These ratios were inferred by Huang

Quantity Theta solventb Carbon disulfidec Toluened and Witten [18] using scattering and
cooperative-diffusion data from the litera-
ture.
b These values were inferred from data on
A (nm)e 0.32 (± 7%) 0.28 (± 7%) 0.32(± 7%)
ξφ /ξs f 0.61(± 9%) 1.23 (± 8%) 1.23 (± 9%) polystyrene in the theta solvent cyclohex-
ξ /ξs g 2.97 (± 5%) – 3.81 (± 6%) ane at 36 ◦ C [19–22].
c These values were inferred from data on
ξf /ξs h – 1.27(± 1%) –
perdeuterated polystyrene in the good sol-
ξζ /ξs i 4.29 (± 10%) – 1.65 (± 7%) vent carbon disulfide at 20 ◦ C [23–25].
r2 j 3 (± 22%) 1.2 (± 14%) 1.1 (± 23%) d These values were inferred from data on
r4 48 (± 28%) 4.5 (± 21%) 4.3 (± 29%) ordinary polystyrene in the good solvent
toluene at 25 ◦ C [26–28].
e Fractal amplitude defined by φ(r) =
0
(A/r)D−d , as in the subsection “Semidilute
I (q, t) ∝ |ρ(q, t)|2 = I (q, 0)e−2t/τ (q) . The decay time is roughly the time Solutions” above.
f ξ is the correlation length inferred from
for the density to diffuse a wavelength 2π/q. s
scattering: S(q) = const.(1 − (qξs )2 +
If many initial states were prepared at random with proper Boltzmann- O(q 4 )). ξφ is the distance from a monomer
weighted probabilities, and the intensity I (q, t) monitored for each, each at which the local volume fraction φ(r) 0
experiment would show a different initial intensity, but the same decay of a large single chain inferred from high-
q scattering is equal to the overall volume
time τ (q)/2. One may measure this decay conveniently by measuring the fraction φ: φ(r) 0 |r=ξφ = φ, as described
correlation function I (q, t + t  )I (q, t  ) t  , where  t  is the average over in the subsection “Semidilute Solutions.”
initial times t  . Under the diffusive law, this correlation function has the g ξ
 is the length scale implicit in the
−3
form [16] osmotic pressure :  = T ξ .
h ξ is defined by S(q) = S(0)(qξ )−D for
f f

I (q, t + t  )I (q, t  ) t  = (I (q) − I (q) )2 e−2t/τ (q) + I (q) 2 . (4.15) q in the fractal regime.
i ξ is the radius of a sphere whose stokes
ζ
diffusion constant is the same as the coop-
The scattering from an equilibrium fluid behaves in this way, as well. The erative diffusion constant. It is rigorously
random thermal fluctuations have the effect of preparing initial states of related to ξ and the permeability length
ξp , as discussed in the text.
random amplitudes, which then decay according to the diffusion equation. j r is the reduced moment of the con-
n
At any given moment the intensity is the cumulative effect of many partially centration profile defined by rn ≡
∞
decayed random fluctuations. 0 dxx (φ(xξφ ) 0 /φ − 1). The r2 and r4
n

Careful scattering measurements have been done to test the predicted determine the ratios ξs : ξφ : ξ : ξf , as
explained in [18].
behavior of τ (q). The data [17] show good consistency with the φ depen-
dence predicted above. They lead to the estimate ζc = T/[6π ηs (αξφ )],
where α = 1.3 ± 10%, as inferred from Table 4.3.
4.5. Cooperative diffusion and permeability The cooperative diffusion constant
ζc tells the flow of concentration in response to a concentration gradient. One
can readily infer from this the flow of fluid in response to an osmotic pres-
sure gradient. The ratio of fluid velocity to a pressure gradient is defined as
the permeability P : v = j/ρ = −(P /η)∇p.  The permeability is defined for
flow of any fluid; no polymers or other solute need be involved. Here η is the
viscosity of the fluid and p is the pressure, j is the current density of fluid and
ρ is the particle density of the fluid. Evidently the permeability has the dimen-
sions of a length squared. (a) Find the permeability P relating the average
velocity to the pressure gradient in a long circular pipe of radius R. Recall that
the velocity profile in the pipe is parabolic while the pressure is constant over
the cross-section of the pipe. You can find the pressure gradient by equating
the power required to maintain the pressure with the viscous energy dissipation
in the fluid. (b) If monomers with number density ρn are flowing to the right
with a current density jn , what is the average speed of monomers with respect
to solvent? Assume monomers have the same volume as solvent molecules.
Considering the monomers as fixed, what is the solvent current js in terms
of jn and volume fraction φ = ρn /(ρn + ρs )? (c) A semidilute solution has
a monomer density ρn and solvent viscosity ηs . Relate its permeability P to
its cooperative diffusion constant ζc and the osmotic compressibility d/dρn .
102 Polymer solutions

(d) Recognizing that  is a power of ρn and can be related to the correlation

length ξ by T /ξ 3 , relate the permeability to ξ and the hydrodynamic cor-

relation length ξh defined by ζc = T /(6π ηs ξc ). (e) What diameter pipe has
the same permeability as a semidilute solution whose scattering correlation
length is ξs ? For this estimate, you may assume that the various ξ ’s are equal:
ξh ξ ξs . You may ignore numerical factors.
The solution is given at the end of the chapter.
Though the diffusive law for ρ is valid over sufficiently long times and
distances, other effects can modify the relaxation of ρ over shorter times.
The chief one is elastic stress in the polymer chains. We shall see below
that such stress relaxes in a time that is independent of the length scale
of observation. Thus for sufficiently great observation scales, this stress
relaxation is always much faster than diffusion. Then the stress relaxation
effects have a negligible effect on the observed density relaxation.

4.3.7 Semidilute self-diffusion without entanglement

The diffusion of a single chain behaves quite differently from that of the
overall density. As the concentration rises towards φ ∗ the chain’s motion
becomes inhibited by its neighbors and it diffuses slower. Again the self-
diffusion constant ζs changes by a finite factor at φ φ ∗ . And again we can
use the φ ∗ behavior to infer that for semidilute chains. We begin by selecting
k chains at random at φ ∗ and considering the diffusion of their center of mass.
Each chain diffuses because of the random motions of the fluid around it.
Different chains are typically several hydrodynamic screening lengths from
† This is a good starting assumption, but each other because of hydrodynamic screening† . The random fluid motion
it is not rigorously valid, as explored in is separate and independent for such chains.Accordingly their brownian
Problem 4.8.
motions are independent. After a time t the center of mass displacement
r̄(t) can be expressed in terms of the displacements ri (t) of the chains i:

k r̄(t) = ki=1 ri (t). The mean-square displacement r̄ 2 can be written

2 

k
k 2 r̄ 2 = ri (t)
i=1


k
= ri (t)rj (t) .
i,j

Insofar as the individual chains move independently, there is no correlation

between ri and rj , so that ri rj = 0 unless i = j . Also, when i = j ,
ri ri = r 2 is independent of the chain chosen. Thus k 2 r̄ 2 = kr 2 .
Thus the diffusion constant ζcm = r̄ 2 (t) /t ∼ ζ /k, where ζ is the diffusion
constant of one of the chains.
4.6. Damping of jello In Problem 2.7 we calculated the damping expected for
a jiggling cube of Jello owing to the viscous dissipation of the water. We
assumed that the flow was just that of pure water undergoing the overall
motion of the block of Jello. We found that this dissipation appeared too
little to account for the damping of real Jello. In the previous section we
have recognized that relative flow of solvent through a polymer solution like
Jello produces an extra dissipation. This problem revisits the Jello question to
 Motion in a polymer solution 103

see whether we can account better for its damping. If a block of Jello were
stretched to double its natural length, it would eventually expel a finite fraction
of the water inside. The stretched water wants to conserve its volume when
stretched, while the polymer network wants to increase its volume. Thus when
the Jello is distorted, there is a current of water through the polymer network.
From this we can estimate the dissipation of Jello. From the oscillation fre-
quency and density of Jello, we know its modulus G is roughly 4000 J/M3 (or
40,000 erg/cm3 ). This modulus is of the same order as the osmotic pressure.
(a) From this information, estimate the blob size ξ . A block of Jello is rapidly
subjected to a shear strain of magnitude γ . This strain produces an osmotic
pressure difference of order Gγ between the center and the outside. (b) Esti-
mate the rate of energy dissipation per unit volume caused by the resulting
flow in a cube of side b. (c) Assuming this dissipation has a small effect on
the motion, estimate the damping rate for a jiggling cube of Jello 5 cm on a
side.

4.B Suggested experiment: Jello elasticity The theory of polymer elasticity

makes some clearcut predictions about Jello that you can check. There is a clear
predicted dependence of the modulus and hence the oscillation frequency as one
changes the concentration. Try checking this by making up some Jello samples at
different strengths, ranging from half strength to quadruple strength. It is probably
better to use unflavored gelatin instead of Jello. Measure the oscillation frequencies
and damping rates for samples of the same size and shape but different concentra-
tions. You could gauge the effect of sliding motion against the bottom by using
bottoms of different consistency, including a thin layer of more concentrated Jello.
From these measurements infer the change of modulus. Does the modulus change
with concentration as predicted for a semidilute solution? Note that the modulus
depends on cooling time, so cool for a long time.

We now form a semidilute solution from the original φ ∗ solution by join-

ing up all the chains in groups of k chains to make long, interpenetrating
ones. We compare the motion of the center of mass of one k chain with
the center of mass motion of the last paragraph. The motion of this con-
nected chain is now different in many ways from that of the disconnected
chains; for example, the constituent chains cannot now move far apart.
These differences arise from the forces between the k connected subunits
of the k chain, formerly absent. However, such internal forces cannot affect
the motion of the center of mass. Accordingly, we infer that the long chain
diffuses like the separated subchains; its diffusion constant is reduced by a
factor k relative to that of one subchain. The subchains, defined to be at their
φ ∗ have size ξφ and diffusion constants ζ roughly as large as they would
have been in dilute solution. This ζ is that of cooperative diffusion, as we
have just seen. The number of monomers M(ξ ) in one subchain (or blob)
is of order ξ D . The molecular weight M is evidently kM(ξ ). Combining,
we may obtain the scaling of the self-diffusion constant ζs :
ζs ζc /k ∼ M −1 ξ D+2−d ∼ M −1 φ (D+2−d)/(D−d) . (4.16)
For polymers in a good solvent ζs ∼ ξ 2/3 /M ∼ φ −1/2 /M.

4.3.8 Motion with entanglements

The above picture reduces the motion of a chain in solution to that of
an object whose parts are subjected to independent random forces. This
104 Polymer solutions

simple model of motion is called the Rouse model [5]. The Rouse model
describes self-diffusion reasonably well when the concentration is not too
far above φ ∗ . But for solutions far above φ ∗ we must consider another
difference between our original solution of subchains and the final solution
of connected k-chains: entanglement. The forbidding task of accounting for
entanglements quantitatively was reduced to a simple and intuitive model
by Edwards and by deGennes [5, 10] in the 1970s. Our discussion below
recounts these ideas.
Intuitively, entanglements between chains impose constraints that go
beyond mere hydrodynamic drag. We may begin to anticipate how they
influence a given chain by replacing the other chains with fixed, regular
obstacles, such as a regular lattice of bars of spacing ξ . If a random walk
polymer threads through such a “jungle-gym” lattice, it certainly cannot
diffuse according to the Rouse model discussed above. If a random force
pushes the chain against one of the bars, the bar exerts a restoring force
that prevents the chain from crossing it. The chain is confined. If each
end of the chain were to be fixed, this confinement would be permanent;
the chain must forever follow the same path through the jungle gym that
it followed initially. But if the ends are free to move, there is one type
of motion it can perform without a restoring force from the bars. This
is for each monomer to move in the direction of its successor along the
chain. The result of such a motion is equivalent to removing a bit of the
chain at one end and adding it on to the other end. This motion is called
reptation [5].
Of all the different small random motions of our chain only a reptational
motion results in no restoring force from the bars. Thus for this motion
the random thermal forces are free to act unopposed. This means that the
chain is free to do random small steps along its own path: it performs
curvilinear brownian motion. The displacement along the path s(t) obeys
s 2 (t) ∝ t. The constant of proportionality is the curvilinear or tube dif-
fusion constant ζt . This diffusion motion is the same as it would be if the
contorted sequence of blobs were straightened out into a straight path of
length L ξ k. Each section of the chain of size ξ is subjected to independ-
ent random forces, as in the Rouse model above. Accordingly the diffusion
constant for the overall motion of a chain containing k such sections is
reduced by a factor of k. The curvilinear diffusion constant is simply the
Rouse diffusion constant derived above, viz. ζc (M(ξ )/M). Under this dif-
fusion, the chain ultimately moves a curvilinear distance equal to its path
length L. At this point the chain is no longer constrained by its original
entanglements. The time τrep required is given by L2 /τrep ζc (M(ξ )/M).
This τrep plays a central role in many phenomena of entangled polymers.
Using L = ξ(M/M(ξ )), we may infer the scaling of τrep :

τrep ∼ M 3 ξ d−3D ∼ M 3 φ (3D−d)/(d−D) . (4.17)

For polymers in a good solvent D 53 and τrep ∼ M 3 φ 3/2 . The reptation

picture predicts the self diffusion of entangled polymers. In a time τrep the
chain has moved a spatial (as distinguished from curvilinear) distance of
the order of its own size R. Each subsequent interval τrep leads to another
 Motion in a polymer solution 105

independent translation of order R. Thus the long-time motion is diffusive,

with a diffusion constant ζrep R 2 /τrep . Using R 2 ξ 2 (M/M(ξ )) one
finds in good solvents that

ζrep ∼ M −2 φ −7/4 . (4.18)

It is wise to verify this important result.
Real entanglements between polymers differ from our jungle-gym entan-
glements in two basic ways. First, the polymers are not a regular array of
lines, but are random coils. This means, for instance that the test chain
encounters a given chain at qualitatively more places than it would encounter
a given bar in the jungle gym. Nevertheless, experiments and computer
simulations of the diffusion of a mobile chain among frozen or immob-
ilized chains confirm the reptation picture [10]. They support the view
that the entanglement constraints confine each chain to its original path.
A second defect of the jungle-gym model is that it neglects the motion
of the other polymers. This motion raises the possibility that the chains
can avoid the entanglement constraints by some type of collective motion.
While we cannot rule out this possibility, we can argue that the reptation
hypothesis is self-consistent even when the other chains move. If all the
chains move by reptation, a test chain may release an entanglement con-
straint imposed by a given bar in two ways. The first way, treated above,
is to reptate away from the constraining bar. The second is for the chain
represented by that bar to reptate so that it no longer constrains the test
chain. The time required for a given constraint to be released in this second
way is the time for the constraining chain’s end to move to the constraining
point. Since the constraining chain is also assumed to move by reptation,
this time is roughly the reptation time τrep . Thus in the time τrep required
for the test chain to escape by reptation, a finite fraction of the entan-
glements have also been released by their own reptation. This allows the
test chain to diffuse slightly further, but does not alter our simple scaling
picture.
The reptation model for diffusion has recently been stringently tested by
experiments and simulations [29]. In addition to measuring the reptation
time, one may also check further predicted details of the motion and check
predicted changes in the diffusion when different chain lengths coexist in
the solvent. Overall these tests give strong confirmation to the reptation
model and its recent refinements.
4.7. Diffusion in a theta solvent If polymers are dissolved in a theta solvent in the
semidilute regime, the cooperative and self-diffusion constants are modified
because the blobs are random walks with D = 2 instead of self-avoiding walks
with D = 53 . How do ζc , ζs , and ζrep scale with molecular weight M and volume
fraction φ in theta solvents? That is, do they change faster or slower than with
good solvents?

4.3.9 Stress relaxation and viscosity

Entanglement also controls how elastic energy is stored and dissipated in
a polymer solution or melt. We have studied the work required to com-
press polymers into a smaller volume of solution; it is given by the osmotic
106 Polymer solutions

pressure . The polymers also store energy when the solution is distorted,
even with no change of volume. We discussed how this energy is measured
in the Experimental Probes section of Chapter 2. There we told how rheo-
meters apply a known, oscillating shear strain and measure the stress that
oscillates in phase or in quadrature with that strain.
In strongly interpenetrating polymer solutions, this stress occurs because
of entanglement. To understand the effect, we return to our jungle-gym
model. We impose a step shear strain like that shown in Fig. 2.3. This
amounts to tilting the jungle-gym lattice so that each square of the lattice
distorts (affinely) into a parallelogram, like the sample as a whole. This
distorts the shape of the polymer and its internal density profile. If we con-
sider the end-to-end vectors of each blob of the chain, they are initially
isotropically distributed. But after the distortion, the vectors become aniso-
tropic. They are more concentrated along the long diagonal of the sheared
square and less concentrated along the short diagonal. The anisotropy has
a strength of order unity under one unit of shear. Each of these blob distor-
tions costs energy. To estimate this energy, we imagine that the anisotropic
distribution of vectors were produced by an external potential U (θ ). The
distribution of angles f (θ) = exp(−U (θ)/T ). In order to produce an f (θ )
that varies over a factor of order unity, U (θ ) must be of order T . The work
required to turn on this potential must also be of order T , since the energy
U affects most of the possible angles θ. This means that the work required
to make the shear distortion is of order T for every cell of the lattice, or for
every blob of the chain.
4.C Suggested experiment: Rubber elasticity and temperature The elastic
restoring force of rubber is essentially the restoring force of an elongated random
walk. This means the force should be proportional to the absolute temperature,
provided the basic local structure of the rubber polymers does not depend on tem-
perature. If this is true, one should observe a roughly 30% increase in the spring
constant as the temperature is changed from 0 ◦ C (273 K) to 100 ◦ C (373 K). Devise
an experiment with rubber bands to test this prediction. You might use weights, a
ruler, boiling water, and ice water.
Since a unit of shear strain stores energy at a density T /ξ 3 , the solution
has a step-strain modulus G0 of this order, owing to entanglements. This
G0 is independent of molecular weight and is of the order of the osmotic
† The ratio of G0 to  is predicted to be pressure † . The solution retains this stored energy until the entangle-
universal in the semidilute regime in good ment constraints are released. If each chain is prevented from reptating, by
solvents. But this universality has not been
immobilizing its ends or by crosslinking, the stored energy lasts indefinitely
verified experimentally. Instead,  appears
to vary more strongly with solvent quality and the sample is an elastic solid—a gel. Similarly a polymer melt immob-
than G0 does. As concentration increases ilized by a few crosslinks makes an elastic solid—a rubber. The order of
above the semidilute regime or solvent qual- magnitude of this modulus has been mentioned in Problems 2.6 and 2.7.
ity diminishes, the predicted universality is The highest-modulus rubbers are those with the highest density of entan-
expected to be compromised. These limita-
glements. Empirically the largest moduli for a weakly crosslinked rubber
tions may account for why the universality
has not been observed. are of the order of 1 atm, or 105 J/m3 .
4.8. Stokes diffusion in semidilute solution In treating the self-diffusion of a poly-
mer in semidilute solution, we have pretended that the neighboring chains
completely remove any momentum in the fluid, so that the flow around a point
source of momentum is completely screened out. But this can not be completely
true, since a chain that absorbs the momentum can only give it to the adjacent
solution; the chains do not remove the momentum entirely. This means that if
 Motion in a polymer solution 107

Fig. 4.5
How reptation leads to stress relaxation.
Top picture shows a chain confined in a
jungle-gym representing the other chains.
Bottom left picture shows the effect of a
horizontal stretching: the whole structure
is distorted and each section of the chain
becomes anisotropic. Bottom right picture
shows the chain after some time has
passed. The heavy section on the left
shows the part of the chain that has
reptated out of the initial constraints. This
part is isotropic and holds no stress. The
dashed section on the right shows the part
of the initial chain that has reptated away.
This section is equal in length to the heavy
section on the left.

a chain is gradually dragged through a semidilute solution, the drag coefficient

 must obey Stokes law valid for any viscous liquid:  = 6π ηRh , where η is
the viscosity of the solution. The hydrodynamic radius Rh must be of the order
of the geometric radius R of the chain or smaller. Thus the corresponding self-
diffusion constant ζs = T/  must not be too small: ζs  T/(6π ηR). In the text
we have derived scaling laws for both the viscosity η and the self-diffusion con-
stant ζs , without regard to this Stokes Law constraint. This problem explores
whether the two approaches are mutually consistent. (a) Using the scaling of η
and R for an entangled, semidilute solution, find the scaling expected for the
self-diffusion constant according to Stokes’ law. Compare with the scaling for
self-diffusion according to the reptation model.

If the chains are free to reptate, the stored energy can relax (Fig. 4.5). In
our jungle-gym model, a given cell of the lattice continues to store energy
until the chain end has reached that cell and relaxed the constraint. The
same is true for a real entangled solution. The time required to relax a finite
fraction of these constraints is evidently the reptation time τrep . By extending
this reasoning, one can make detailed predictions of the fraction of the
initial stress relaxed as a function of time. These predictions agree well
with experiment [10] (Fig. 4.6).
4.9. Partial stress relaxation function in polymer melt If the volume fraction of
polymer is increased above the semidilute regime, the correlation length shrinks
to the size of a monomer. This is the concentrated solution regime. When all
the solvent is removed, the polymer liquid is called a melt. In this regime most
of what we derived about the semidilute regime remains valid. The point of
this problem is to calculate how the stress G(t)γ begins to fall below its initial
value G(0)γ in a strongly entangled solution or melt. The stress falls over time
because part of each chain has reptated out of its initial (distorted) tube and into
a new (isotropic and stress-free) set of entanglements, as shown in Fig. 4.5.
In effect, part of the initial distorted tube has disappeared. The lost stress is
just the initial stress times the fraction of initial tube which has disappeared.
This disappearance occurs as the chain executes a random walk forward and
backward in its tube. At a given time t the amount of tube removed from the
left is the maximum curvilinear distance the chain has traveled to the right in
that time. This maximum is roughly the root mean square curvilinear distance
108 Polymer solutions

Fig. 4.6
Storage modulus G versus oscillation PS 160 °C
8
frequency ω for narrow-distribution
polystyrene melts, reprinted with
permission from [30, Fig. 2]. © 1987 7
American Chemical Society. Molecular
weight ranges from 8.9×103 (curve L9) to 6
5.8×105 (curve L18). The modulus drops

log G⬘ dynes/cm2
off below a characteristic frequency on
each curve. This frequency is the inverse 5
of the stress relaxation time. The high
molecular weight samples have a large 4
range of frequencies where the modulus is
independent of frequency. Over this
plateau region, the liquid behaves like an 3 L34
L27 L37
elastic solid. The associated modulus is the L18 L15
step-strain modulus discussed in the text. L14
2 L19
(To achieve the huge reported range of L5 L22
oscillation frequencies, the experimenters L16 L9
L12
use a trick. They cool the short polymers to 1
slow their motion and heat the long ones to –6 –5 –4 –3 –2 –1 0 1 2 3 4
speed up their motion. The temperature log ωaT s–1
effect can be separately quantified. The
reported frequencies have been multiplied
by a factor aT , determined separately for
each sample, to show what the moduli
would be at a fixed temperature of 160 ◦ C.)
traveled. In a given time, about the same amount is lost on each end. (a) Derive
a function for the fractional stress lost as a function of time t after the initial
strain in terms of the curvilinear diffusion constant ζt . Do not worry about late
times when the amount of tube lost is nearly the entire tube. And do not worry
about numerical prefactors.

Knowing the step-strain modulus and the relaxation time, we can determ-
ine the viscosity η of the entangled solution. As we saw in Chapter 2, the
viscosity scales as the modulus times the relaxation time. If one increases
molecular weight M at a given concentration φ, the reptation time increases
as M 3 , while the modulus remains constant. Thus the viscosity should
increase as M 3 . Experimentally, the viscosity agrees better with M 3.4 than
with M 3 . After many years of puzzlement it appears that this disagreement
is not fundamental. It appears that in the case of viscosity it is particu-
larly difficult to attain the asymptotic M  M(ξ ) behavior. Several effects
alter the viscosity when M is not asymptotically large. One may estimate
these effects quantitatively and they seem to explain why an apparent M 3.4
behavior is observed in the experimental range of M [10].
If a liquid undergoes step strains that are more frequent than the relaxa-
tion rate, the stress from each step progressively builds up. Thus, if a steady
shear rate exceeds the inverse relaxation time, substantial elastic stress
can build up. In particular, normal stress builds up as the polymer chains
stretch along the streamlines of the flow. This stress from stretched poly-
mer chains is the origin of the unusual flow behavior pictured in Fig. 1.6.
The appearance of steady-state deformation and its attendant change in
the stress at shear rates exceeding the relaxation rate is known as the
Cox–Merz rule [31]. Cox–Merz behavior is observed in many structured
 Appendix A: Origin of the Oseen tensor 109

fluids; this is natural since the large structures found in such fluids make for
long relaxation times. Thus the Cox–Merz criterion for shear-rate dependent
viscosity can be met at moderate shear rates.

4.4 Conclusion
In this chapter we have recounted the basic behavior of large, flexible
chain molecules in solution. The qualitative elements of randomness,
self-repulsion, and entanglement combine to make distinctive spatial struc-
ture, energy storage, and dynamic response. We have seen that many of
these properties can be understood using primitive geometric notions such
as the number of contacts between two fractals. These geometric properties
allow one to predict all the asymptotic scaling behavior of long polymers
as a function of molecular weight and concentration. In many cases the
common geometric basis of these polymer properties allows one to predict
one property like the diffusion constant in terms of a completely different
property like the osmotic pressure, as exemplified in e.g., the universal
ratios of Table 4.2. In the coming chapters we explore to what degree such
generality can be extended to other forms of structured fluid.

Appendix A: Origin of the Oseen tensor

The velocity field from an external force-per-unit-volume f(r, t) obeys
Newton’s second law. The component α of the velocity field obeys

ρ∂vα /∂t − ∂β σβα = fα ,

for the small velocities considered here. (Repeated indices are to be

summed.) The ∂β means the spatial derivative in the β direction. The stress σ
arises from two origins. The first is viscous stress proportional to the velo-
city gradient tensor (∂β vα + ∂α vβ ). The second is the stress that arises from
compressive forces. These forces generate a scalar pressure field p.

σβα = η(∂β vα + ∂α vβ ) − δβα p

For an incompressible fluid the pressure p must be such as to prevent

vβ from having any compressional part, i.e., ∂β vβ = 0. Since we wish to
consider steady velocity fields, the time derivative will vanish. Our equation
for vβ now reads
−η∂β ∂β vα + ∂α p = fα .
(A term in ∂β ∂α vβ has dropped away because ∂β vβ = 0.)
To find vβ around a point source, we Fourier transform in space, defining

v(r) = k ṽk exp(ik ·r). The incompressibility condition becomes k · ṽk =
0. The equation for the ṽ is thus

ηk 2 ṽα + ikα p̃ = f˜α .

Thus
ηṽα + ikα p̃/k 2 = f˜α /k 2 .
110 Polymer solutions

We can find p̃ by imposing the incompressibility condition kα ṽα = 0:

0 + i p̃ = kα f˜α /k 2 .

Thus
1
(fα − kα (kβ f˜β /k 2 )).
ηṽα =
k2
For a localized force F at the origin f̃k = F /(4π ). The desired Oseen
tensor is thus given by
|F |  1  
v(r) = 2
F̂ − k̂(k̂ · F̂ ) eik·r .
4π η k
k

The first term must integrate to F /(ηr) up to a numerical factor. The second
term must be in the radial r̂ direction and must be such that ∂β vβ = 0. The
result is the Oseen tensor given in the text.

Solution to Problem 4.5 (Deriving permeability)

(a) The speed at distance r from the axis is given by v(r) = v0 (1 − (r/R)2 ). The
average speed v̄ is 12 v0 , as can be readily found by integrating. The source of
the pressure does work in a small length L at the rate Ẇ = force × velocity,
or Ẇ = −Lv̄(πR 2 )∇p  = − 1 Lv0 (πR 2 )∇p.  Here p is the pressure and z is
2
the coordinate along the axis. This work goes into viscous dissipation. The
local rate of dissipation per unit volume ẇ is η(dv/dr)2 = (ηv02 /R 4 )(2r)2 .
Integrating,
  R2  
1
Ẇ = L π dr 2 ẇ(r) = (4π)L(ηv02 /R 4 ) dr 2 r 2 = Lηv02 (4π ) .
0 2

Combining, we find 12 ∇pv  0 π R 2 = ηv 2 (2π ), or R 2 ∇p

 = 4ηv0 . Now we
0

can find the permeability P , using v̄ = −(P /η)∇p, or v0 = −2(P /η)∇p. 

Comparing with our other relation between v0 and ∇p, we conclude P = R /8. 2
(b) Monomers with density ρn and speed v have a current density jn = ρn v so
v = jn /ρn . Viewing the monomers as fixed, we then have solvent flowing at
speed v, and current js = vρs = jn ρs /ρn = jn (1 − φ)/φ.
(c) The cooperative diffusion constant ζc relates the monomer current jn to the
gradient of monomer density ∇ρ  n : jn = −ζc ∇ρ
 n . The permeability P gives
a like relation between solvent current js and solvent density ρs : js =

ρs P ∇p/η s . The pressure we are considering here is osmotic pressure , so we
shall use  for p henceforth. To compare these we must relate changes of ∇ρ  n
  
to ∇. We can say ∇ρn = (dρn /d)∇. Combining, we have
     
dρn  ρn ρn P 
jn = −ζc ∇ = js =− ρs ∇,
d ρs ρs ηs

so that
     
dρn P 1 dρn 1 dφ
ζc = ρn ; P = ηs ζc = ηs ζc .
d ηs ρn d φ d
 References 111

(d)  varies as a power of φ:  = φ q , where the exponent q = 3/(3 − D). Thus

d/dφ = q/φ and
3
P = ηs ζc /(q) = ηs ζc ξ /(qT ).

Expressing ζc in terms of ξζ ≡ T /(6π ηs ξζ ), we have


1 3 3 − D ξ 3
P = qξ = .
6π ξζ 18π ξζ

Thus the solvent flows through the polymer network as though it were a bunch
of pipes of radius R ξ .
(e) Specifically, R 2 /8 = (3 − D/18π )ξ 2 or using D 53 , R 0.4ξ . We can get
a more refined estimate using the data in Table 4.3: ξ = 3.81ξs ; ξζ = 1.65ξs .
Then
R (2.5 ± 10%)ξs ,
where ξs is the scattering correlation length defined by S(q) = S(0)
(1 − (qξs )2 + · · · ).

References
1. G. Jannink and J. Des Cloizeaux, Polymers in Solution (Oxford, UK: Oxford
University Press, 1992).
2. N. Nemoto, Y. Makita, Y. Tsunashima, and M. Kurata, Macromolecules 17 425
(1984).
3. M. Adam, L. J. Fetters, W. W. Graessley, and T. A. Witten, Macromolecules 24
2434 (1991).
4. I. Noda, N. Kato, T. Kitano, and M. Nagasawa, Macromolecules 14 668 (1981).
5. P. G. deGennes, Scaling Concepts in Polymer Physics (Ithaca, NY: Cornell,
1979).
6. M. Daoud and G. Jannink, J. Physique 37 973 (1976).
7. A. Einstein, Ann. Physik 17 549 (1905); 19 371 (1906).
8. R. K. Pathria, Statistical Mechanics (Oxford: Pergamon Press, 1972), chap. 13.
9. J. Happel and H. Brenner, Low Reynolds Number Hydrodynamics With Spe-
cial Applications to Particulate Media, 2nd rev. ed. (Leiden: Noordhoff
International Publishing, 1973).
10. M. Doi and S. F. Edwards, The Theory of Polymer Dynamics (Oxford: Oxford
University Press, 1986).
11. A. Einstein, Ann. Phys. 34 591 (1911).
12. P. E. Rouse, J. Chem. Phys. 21 1273 (1953).
13. B. Zimm, J. Chem. Phys. 24 269 (1956).
14. N. S. Davidson, L. J. Fetters, W.G. Funk, N. Hadjichristidis, and
W. W. Graessley, Macromolecules 20 2614 (1987).
15. B. Ewen and D. Richter, Adv. Polymer. Sci. 134 1–129 (1997).
16. See e.g. B. J. Berne and R. Pecora, Dynamic Light Scattering: With Applications
to Chemistry, Biology, and Physics (New York: Wiley, 1976).
17. A. Z. Akcasu, G. C. Summerfield, C. C. Han, C. Y. Kim, and H. Yu, J. of Polym.
Sci. Part B Polym. Phys. 18 863 (1980).
18. J. R. Huang and T. A. Witten, Macromolecules 35 10225 (2002).
19. M. Adam and M. Delsanti, Macromolecules 18 1760 (1985).
20. P. Stepanek, R. Perzynski, M. Delsanti, and M. Adam, Macromolecules 17 2340
(1984).
21. J. P. Cotton, M. Nierlich, F. Boué, M. Daoud, B. Farnoux, G. Jannink,
R. Duplessix, and C. Picot, J. Chem. Phys. 65 1101 (1976).
22. J. Roots and B. Nyström, Macromolecules 13 1595 (1980).
112 Polymer solutions

23. M. Rawiso, R. Duplessix, and C. Picot, Macromolecules 20 630 (1987).

24. J. des Cloizeaux and G. Jannink, Polymers in Solution, Their Modelling and
Structure (Oxford: Oxford University Press, 1990).
25. B. Farnoux, Ann. Phys. 1 73 (1976).
26. Y. Higo, N. Ueno, and I. Noda, Polym. J. 15 367 (1983).
27. F. Hamada, S. Kinugasa, H. Hayashi, and A. Nakajima, Macromolecules 18
2290 (1985).
28. D. W. Schaefer and C. C. Han, in Dynamic Light Scattering: Application of
Photon Correlation Spectroscopy, ed. R. Pecora (New York: Plenum Press,
1985).
29. K. Kremer, G. S. Grest, J. Chem. Phys. 92 5057 (1990); D. Richter, B. Farago,
L. J. Fetters, J. S. Huang, B. Ewen, and C. Lartigue, Physical Rev. Lett. 64 1389
(1990).
30. S. Onogi, T. Masuda, and K. Kitagawa, Macromolecules 3 109 (1970).
31. W. P. Cox and E. H. Merz, J. Polymer Sci. 28 619 (1958).
 5
Colloids∗

Colloidal dispersions† , are homogeneous suspensions of solid particles in a †Historically colloids referred to any sticky,
fluid. Most inks and paints are colloidal dispersions. The particle size ranges gelatinous material. Here, we follow the
more recent usage.
from nanometers to microns. Such a particle is large enough to approach
its bulk properties on the atomic scale but small enough that its thermal
energy dominates its gravitational energy†† . As discussed in Chapter 1, †† This means that gravity is not strong

colloidal suspensions or dispersions have a variety of direct applications as enough to drive the particles to the bottom or
top of the container, as discussed in Problem
well as serving as intermediates in various processing technologies, such as
2.2.
ceramics. Colloidal particles tend to attract each other and thus aggregate
together. If this aggregation grows unchecked the aggregates eventually
migrate to the top or bottom of the container, thus destroying the desired
dispersed state. Much of colloid science is concerned with maintaining the
dispersed state. This is the problem of colloidal stabilization. Stabilization
is achieved by modifying the particles’ surfaces to prevent aggregation.
Thus understanding and control of interfacial forces is central to colloid
science and to this chapter.
In this respect, colloidal dispersions resemble emulsions and foams. As
noted in Chapter 1, emulsions differ from colloids in that the dispersed
objects are liquid droplets rather than solid particles. In a foam the dispersed
objects are in the gas phase. Most of our reasoning about colloidal stability
applies equally to emulsions and foams.
In the next section we discuss the main causes of the attractive interac-
tions that compromise the stability of colloidal dispersions. The following
section deals with means of combating these attractions via repulsive forces.
The last section discusses consequences of strong colloidal interaction: the
particles take on various forms of spatial organization: crystals, liquid crys-
tals, fractal aggregates, and magnetic states. Appendix A treats one of the
major mechanisms of attraction in detail. Appendix B explains the various
forms of fractal aggregation mentioned in Chapter 1.
How are colloids made? There are two general approaches to the pro-
duction of colloidal-sized particles: dispersion††† and condensation. These ††† “Dispersion” is used to denote both a

are reviewed in Everett’s [2] book on colloid science. Dispersion means the process for making colloids and the result-
ing state in which particles are dispersed in
breakup of larger-sized particles into smaller ones, usually by some type of
a liquid.
rough mechanical treatment, such as grinding or shaking. It is difficult by
such means to apply sufficiently strong stresses to achieve the lower range
of colloidal dimensions. More importantly, dispersion generally results in

* Thischapter is heavily based on an earlier draft by Phillip A. Pincus, itself based on [1].
The organization of that draft, some text, and several figures survive in the present version.
114 Colloids

a broad distribution of sizes, shapes, and compositions; this is awkward for

quantitative experiments and applications.
Condensation refers to various methods where dispersed molecules are
induced to come together to form the colloidal particle. These often involve
† A familiar form of nucleation is the forma-
a nucleation process† in the carrier fluid. If all the nucleation can be initiated
tion of a cloud of water droplets when moist in a short time period, then subsequent growth proceeds in a fairly uniform
air is cooled.
manner, leading to a narrow distribution of sizes. Silver halide suspensions
for photographic applications and polymer latex systems used in paints may
be prepared this way.

5.1 Attractive forces: why colloids are sticky

Colloidal particles have a strong tendency to aggregate into large clusters,
which may be compact or tenuous (Fig. 1.5), under the influence of ubi-
quitous attractive forces, which become increasingly potent as the particles
become larger. Stabilization refers to processes whereby this precipitation
is prevented from occurring, resulting in a homogeneous distribution of the
particles throughout the volume of the fluid. Colloidal stabilization gen-
erally requires the modification of the particle surfaces in order to create
repulsive interactions which may compensate the attractive forces. The sub-
ject of surface forces is reviewed in detail in Israelachvili’s [3] book. In this
section we discuss some of these interactions.

5.1.1 Induced-dipole interactions

†† Induced-dipole interactions are vari-
Van der Waals or induced-dipole†† interactions are ubiquitous interac-
ously called van der Waals interactions, tions which have important consequences in condensed matter. They exist
Keesom forces, charge-fluctuation inter-
between all atoms and molecules and do not depend on whether the species
actions, London interactions, and disper-
sion forces. These different names arose are electrically charged. Indeed, the shape-specific forces which govern
from considering different mechanisms for enzymatic processes in biochemistry rely on these interactions. The van der
charge polarization to be induced. The name Waals interaction is, in detail, rather complicated and arises from fluctu-
“dispersion forces” does not refer to col- ating atomic and molecular electrical dipoles. It is treated in depth in the
loidal dispersions, but rather to optical
excellent book by Mohanty and Ninham [4] and also in the briefer dis-
dispersion, the slowing of light waves
in matter. The electronic susceptibility of cussion by Israelachvili [3]. We present a simple classical argument which
the material causes both this slowing and demonstrates the origin of these interactions.
induced attraction between its molecules. In The interaction energy of an atom (or molecule) of electric dipole moment
this chapter we shall deal only with the basic  with an electric field E is
μ,
features common to all these mechanisms,
and will think of them as generalized van
U = −μ 
 · E. (5.1)
der Waals interactions.
Even if the atom does not have a permanent dipole moment, it must nev-
ertheless have a nonzero μ owing to its polarizability α: μ = αE. Then the
interaction energy between the atom and the field is given by

U = − μ(E)   = − 1 αE 2 ,
· dE (5.2)
2
where the factor of one half arises because the electric field must first
polarize the electronic structure of the atom before the two fields can couple.
Suppose now that in the absence of any external E field, there are two
 Attractive forces: why colloids are sticky 115

identical polarizable atoms separated by a displacement r which is large

compared to atomic dimensions, so that we can treat them as point dipoles.
At some instant there may be a thermal or quantum fluctuation causing one
of the atoms (say, atom 1) to have a nonzero moment μ1 . If μ  1 is pointing
toward atom 2, i.e., parallel to r, the electric field at atom 2 is E2 = 2μ  1 r −3 .
Then, using Eq. (5.2), there is an interaction between the two atoms given by

U1,2 = −2αμ21 r −6 . (5.3)

5.1. Fluctuating dipoles We assumed above that the instantaneous dipole was
directed parallel to r. Recalculate U1,2 , relaxing this assumption.
This result merits some discussion. The stability of the unpolarized atom
implies α ≥ 0. Then the interaction energy U1,2 is negative definite; the
interaction is attractive independently of whether the fluctuating moment
 1 is pointing toward or away from atom 2. We can now average U1,2 over
μ
all fluctuations of atom 1; in Eq. (5.3), the factor μ21 is then replaced by
its average μ21 , which is nonzero even though μ1 = 0. The r −6 force
law is characteristic of the unretarded van der Waals interaction [4]. Let us
try to estimate the order of magnitude of U1,2 . The mean square value of
the fluctuating dipole moment, μ21 , is related to the elementary atomic
dipole given by the electronic charge e multiplied by an atomic length, a,
i.e., μ21 ∼ (ea)2 ; the atomic polarizability scales with the atomic volume,
α ∼ a 3 [3]. The interatomic coupling may be generally expressed as

U1,2 = −C(a/r1,2 )6 , (5.4)

where C is a constant with dimensions of energy and magnitude comparable

to the binding energy of an electron—several electron volts. The length a is
an atomic size scale and is thus a few tenths of a nanometer in magnitude.
As r1,2 decreases from infinity, the van der Waals attraction increases, but
it is ultimately overwhelmed by short-range repulsion when the electron
clouds of the two atoms begin to overlap. The repulsion becomes dominant
when r1,2 is 2–3 times a. Thus at the most favorable separation, U1,2 is only
a small fraction of C. It is typically a few percent of the atomic ionization
energy and is thus comparable to the thermal energy T at room temperature.
This U1,2 is reduced further when the two atoms are not in free space but
in a homogeneous, isotropic medium. If the medium consists of atoms just
like atoms 1 and 2, the attractive force between 1 and 2 vanishes, since the
attraction is the same in all directions. Thus it is not surprising that when
the two atoms are embedded in a solvent, the constant C grows with the
difference of polarizabilities between the atoms and the solvent. This solvent
effect may reduce C from its vacuum value significantly. However, it is not
generally possible to adjust the solvent to cancel out the van der Waals
attraction completely. Even if the solvent had the same static polarizability
as the atoms, this would not cancel out the attractions. To achieve this
cancellation, the solvent would have to respond like the two atoms at all
frequencies where fluctuations occur, since fluctuations at all frequencies
contribute to the van der Waals attraction.
116 Colloids

5.1.2 Solid bodies

Using Eq. (5.4) and assuming binary interactions, the van der Waals interac-
tion between bodies containing many atoms can be calculated. The mutual
energyV  of two thick slabs 1 and 2 at separation h is then given by
V = r1 r2 C(a/r1,2 )6 . Here r1 is the position of an atom in slab 1 and r2
is the position of an atom in slab 2. Supposing that each atom occupies a
volume , this sum can be approximated by an integral:

V(h) = Ca 6 −2 dx1 dy1

   
0 ∞
× dz1 dxdy dz2 [x 2 + y 2 + (z1 − z2 )2 ]−3 ,
−∞ h
(5.5)
where x and y are the transverse coordinates of atom 2 relative to atom 1.
Now, the integral in [. . .] is clearly independent of x1 and y1 , so that the
x1 and y1 integrals give simply the surface area A. The remaining [. . .]
integrals are finite and depend only on the separation h. For dimensional
consistency, [. . .] must be a constant times h−2 . This is conventionally
written
V(h) = −H A(12π h2 )−1 , (5.6)
where the coefficient H is called the Hamaker constant. The Hamaker
constant is an energy characteristic of the material and the intervening
medium.
5.2. Van der Waals interaction between layers (a) Find the Hamaker constant
H for two thick slabs consisting of identical atoms interacting via Eq. (5.4),
expressing H in terms of the energy C, the atomic length a and the volume per
atom . (b) How does V (h) depend on h if the two infinite slabs are replaced
by films of thickness d in the limit d  h?
Using Eq. (5.6), it can be shown (Problem 5.3) that the mutual interaction
energy of two spheres has the form
Vb (h) −kH(b/h), (5.7)
where b is the sphere radius and b  h. This is an example of the Derjaguin
Approximation [3]. It is important to note that when two particles touch,
i.e., h in Eq. (5.7) becomes a molecular dimension a, the interaction energy
between the particles becomes arbitrarily larger than the Hamaker constant
and thus becomes arbitrarily large compared to T if the particle size b
is large enough. We may estimate the time τ for two initially contacting
spheres separated by a fraction of a nanometer to escape from each other
using the Eyring formula from Chapter 2 in the section on Approach to
Equilibrium: τ τ0 exp(U /T ). Since U is large, the typical time for
escape τ may become long even on laboratory timescales. In effect, the
attraction is irreversible. As time goes on, the particles aggregate. The
aggregates may eventually become large and dense enough so that gravity
may cause settling (sedimentation) to the bottom of the container or floating
to the top (creaming).
 Attractive forces: why colloids are sticky 117

For simplicity we have discussed the van der Waals interaction in the
context of single atoms and atomic media. But this mechanism of attraction
is much more general. It arises simply because materials are polarizable
and the charge within them fluctuates. Since all condensed matter has
these properties, van der Waals attractions operate in all condensed matter,
atomic or molecular, insulating or metallic. Because the van der Waals
energy has the same distance dependence for all materials, all like surfaces
separated by another medium have characteristic Hamaker constants, and
attractive interactions of the form of Eq. (5.6). But these attractions are
just a special case of an even more general attractive mechanism, to be
discussed next.
5.3. Derjaguin approximation The Derjaguin approximation relates the force
between curved bodies to the interactions between flat surfaces composed of
the same materials. This approximation can be used to find the force between a
sphere and a plane, equivalent to the interaction between two spheres. Consider
a sphere of radius b at a distance of closest approach h from a half space of
the same material, with b  h. Suppose that the interaction energy per unit
area between to half spaces separated by a distance z is known to be v(z). (a)
Show that if v(z) decays sufficiently rapidly with z, then the force between the
sphere and the plane is F (h) = 2π bv(h). (b) Using the results of Problem 5.2,
show that the van der Waals interaction potential energy between a sphere and
a plane is of the form Vb (h) = −kH (b/h). What is the constant k?

5.1.3 Perturbation-Attraction Theorem

Most attractive colloidal interactions arise from the property of the sur- † The case of van der Waals forces is some-
rounding fluid that its equilibrium structure may be altered in response to what different, since these have no expo-
nential cutoff: ξ → ∞. Instead they show a
a disturbance, in this case the presence of a colloidal particle. The van der
power law decay of the perturbation rather
Waals force, discussed in the previous paragraph, may be thought of as than an exponential fall off. This differ-
resulting from a distortion of the fluctuating electric and magnetic fields ence is associated with the fact that photons,
in the space between the particles. The following simple argument demon- the quanta of electromagnetic radiation, are
strates why these interactions are generally attractive. Consider the insertion massless.
of a particle into the solvent. The solvent may respond by a redistribution of
the structure of the internal degrees of freedom, such as a distortion of the
average packing density of the solvent ( cf. Fig. 2.1) or concentration varia-
tions in a mixed solvent. If the particle alters at least one such internal degree
of freedom, a work W is required to deform the internal degree of freedom
away from its equilibrium value in the absence of the particle. Fig. 5.1 illus-
trates such a deformation in the simple case of an elastic medium such as a
slab of foam rubber. A heavy object placed on the slab produces a deforma- Fig. 5.1
tion of the surface nearby. The displacement of the surface u(r) at a distance The Perturbation-Attraction Theorem.
r from the object dies off exponentially with distance: u(r) ∼ e−r/ξ at large Top left: two perturbing particles in a
deformable medium, illustrated as a slab
distances from the object. The decay length ξ of the exponential depends compressed by the particles’ weight.
only on the properties of the medium, and not on the perturbing object. A Top right: two particles which share the
particle inserted into a fluid often perturbs the fluid in an analogous way. The same deformed region make a deeper
analogous decay lengths may range from a few tenths of a nanometer up to depression, and thereby lower the energy
of the system. Bottom left and right:
microns† .
particles’ weight now expands the slab.
Now let us try to insert another particle, identical to the first one, into the Still the deformation is enhanced when
medium, as illustrated in Fig. 5.1. There are two situations, depending on they are close; interaction is again
whether the distance between the particles r is less than or greater than the attractive.
118 Colloids

decay length ξ . If r  ξ , the two particles are effectively decoupled and

the work required is equal to that of the first: viz. W . However if r ≤ ξ ,
the second particle profits, because some of the needed deformation has
already been done. Thus the deformation energy required to add the second
particle (r ≤ ξ ) is less than W . This results in an effective attractive inter-
action between the two particles, of range comparable to ξ . Remarkably,
the two perturbing objects attract each other regardless of the direction of
deformation, as illustrated in the bottom half of Fig. 5.1.
This reasoning suggests that whenever our inserted particle perturbs
its surroundings as described above, two similar particles experience an
induced attraction. We refer to this statement as the Perturbation-Attraction
Theorem. We can verify the theorem under certain conditions: (i) the
solvent has sufficient time to reach thermal equilibrium in response to the
perturbing particle i.e., the solvent is annealed; (ii) the solvent deformation
may be described as a scalar field; and (iii) the perturbation of the solvent
falls off monotonically with distance from a perturbing particle. We now
confirm this conclusion with a mathematical argument for the case where
the perturbations are weak. We suppose that the medium is characterized by
some (scalar) quantity ψ(r) whose equilibrium value is zero. For example,
ψ could be the concentration of some solute in the solution measured from
its equilibrium value. We wish to characterize the work required to insert
two particles at separation x. We imagine inserting these two particles in
sequence, first at the origin, and then at the point x. The first insertion
requires a work denoted W1 .
We may think of W1 as the sum of three pieces. The first piece is the work
required if ψ were fixed at its equilibrium value of 0. We call this piece W0 .
However, by hypothesis, ψ does not stay fixed. This is because there is an
interaction between the particle and the medium that depends on ψ. This
interaction (free) energy alters W1 . It adds an energy of the form W  ψ. Here
we have assumed that the perturbation of ψ is small and have thus expanded
the interaction energy to lowest order in ψ. The coefficient W  may be of
either sign. Finally, perturbing ψ at the origin alters the medium, and this
alteration requires work. The (free) energy of the medium is increased if ψ
changes from its equilibrium value. We can express this change as a function
of the perturbed value at the particle: ψ(0). If the perturbation is small, we
may again express this energy to lowest order in the perturbation: 12 χ ψ(0)2 .
Since the medium’s (free) energy must be minimum when ψ(0) = 0, the
coefficient χ must be positive. Summarizing, the insertion energy W1 has
the form
W1 (ψ(0)) = W0 + W  ψ(0) + 12 χ ψ(0)2 . (5.8)

When we insert the particle, ψ(0) adjusts itself to minimize W1 . One readily
verifies that ψ(0) = −W  /χ and W1 = W0 − 12 W 2 /χ.
Our first particle has perturbed the quantity ψ in the whole vicinity of
the origin. The perturbed ψ at x can be expressed f(x)ψ(0). The simplest
and most common behavior is for f(x) to decay monotonically to zero
as x increases, as in our foam-rubber example above. We shall assume
that f(x) behaves in this way. Later we shall identify conditions when
this monotonic decrease occurs. In Problem 5.4 we will note a contrasting
 Attractive forces: why colloids are sticky 119

case where f(x) oscillates instead. This f(x) can be related to the net
work W1 (ψ(0), ψ(x)) to change both ψ(0) and ψ(x). This W must be
minimal when ψ(x) = f(x)ψ(0). Also, since ψ is assumed small, we take
W to be quadratic in both ψ’s. One can easily deduce a W1 (ψ(0), ψ(x))
that is consistent with Eq. (5.8) and which gives ψ(x) = f(x)ψ(0) when
∂W1 /∂ψ(x) = 0:

W1 (ψ(0), ψ(x))
!
= W0 + W  ψ(0) + 12 χ ψ(0)2 + ψ(x)2 − 2f(x)ψ(0)ψ(x) .

We now add the second particle at x. This adds a work W0 + W  ψ(x) to

the total work W :

W (ψ(0), ψ(x)) = 2W0 + W  ψ(0) + W  ψ(x)

!
+ 12 χ ψ(0)2 + ψ(x)2 − 2f(x)ψ(0)ψ(x) . (5.9)

The free energy must now accommodate to both particles:

∂W
0= = W  + χ(ψ(0) − f(x)ψ(x)),
∂ψ(0)

and

∂W
0= = W  + χ(ψ(x) − f(x)ψ(0)).
∂ψ(x)
Since the two ψ’s play equivalent roles in these linear equations, they have
equal values:
−W 
ψ(0) = ψ(x) = .
χ (1 − f (x))
Since f was positive, both ψ’s increase owing to their proximity. Moreover,
at this equilibrium value of the ψ’s, The free energy W is given by

W 2
W = Wmin = 2W0 − . (5.10)
χ (1 − f )
The nonlocal response f lowers the energy and thus induces an attractive
interaction between the two particles, as claimed.
Our proof covers the case where the perturbation on the medium is
weak. But perturbation leads to attraction in some cases of strong perturb-
ation, provided further conditions are met. Appendix A treats an important
category of strong perturbations.
One further category of perturbation not treated in our discussion is the
case where the perturbed quantity ψ is not a scalar. An important example is
the case where the particles induce a vector dipole moment in the molecules
around the particle. Such polarization often occurs when the solvent is water.
We suppose that the polarization vectors tend to point towards particle 1
and then estimate the energy to add particle 2. The dipoles nearest particle 1
120 Colloids
† 2,6-Lutidine or 2,6-dimethyl pyridine is a are now oriented away from where particle 2 is to be inserted. This adds to
small organic molecule whose chemical the W2 rather than diminishing it. Thus non-scalar perturbations need not
formula is C7 H9 N. Lutidine and water mix-
produce attraction.
tures undergo a continuous phase transition
into a lutidine-rich and a water-rich phase An example of the Perturbation-Attraction Theorem occurs when col-
at a convenient composition and temper- loidal particles are suspended in a mixed solvent, composed of a miscible
ature. At such a transition fluctuations of binary mixture such as lutidine and water† . Normally the particle surface
composition occur that have spatial extents prefers one of the solvents; this preference alters the concentration near
arbitrarily larger than the size of a molecule.
the particle. The concentration gradients extend a correlation length ξ from
Lutidine–water mixtures are often used for
studying the effects of phase separation in a the particle; when the composition and temperature approach the conditions
solvent. of phase separation, ξ becomes large. We refer the reader to a text on phase
transitions such as the book by H. E. Stanley [5]. Other particles are then
attracted to the regions of high favorable solvent concentration. As the The-
orem suggests, colloidal particles in mixed solvents tend to aggregate when
the solvent is near the threshold of demixing [6].
5.4. When perturbation causes repulsion Generally two like objects that perturb
a medium experience a mutual attraction. But sometimes they do not. Here
is a simple and important counterexample. The medium is a membrane with
a bending rigidity: the membrane resists curvature. Such a membrane whose
heightabove some reference plane is ψ has a free energy W [ψ] of the form
x W = dx Aψ 2 + B(d 2 ψ/dx 2 )2 provided ψ varies only with x. If dψ/dx is
h not too large, the second term is simply the square of the curvature. This is a
quadratic free energy and can be readily solved, though there is an awkward
integral to do. One finds that a perturbation ψ0 induced by a phantom vertical
plane at x = 0 with interaction energy V = λψ(0) has the form ψ0 (x) =
(a) Cλe−κz [cos(kx) + δ sin(kx)] for positive x. Evidently, ψ0 (−x) = ψ0 (x).

(a) Find the prefactor C and the coefficients δ, κ and k, that minimize W [ψ] +
V (ψ(0)). Note that d/dxψ|x=0 = 0 by symmetry.
(b) Now a second plane is inserted at distance x = h from the first. Using the
discussion in the text, find the energy W of the medium with ψ(0) and
ψ(h) fixed.
(b) (c) How does the total energy V + W vary with separation h? You can ignore
the overall magnitude of this energy. Are there separations where W + V
is larger than at h = ∞? If so, where?

5.1.4 Depletion forces

Fig. 5.2 A further example of the Perturbation-Attraction Theorem is provided by
Depletion attractions. Two plates immersed
in a suspension of small particles. (a)
depletion attractions between colloidal particles. Such attraction occurs in
When the separation between the plates situations like that pictured in Fig. 2.1. The large colloidal particles are in
exceeds the particle diameter, particles a solvent containing smaller particles that cannot penetrate the colloidal
may enter the interstitial region with only particles. This is an important case of the mixed-solvent situation we just
minor modifications of the particle discussed. The theorem tells us that attractive interaction is expected when
distribution. Thus the internal and external
osmotic pressures balance and there is no
the separation between the surfaces is small enough to sense the structure
interplanar force induced by the particles. of the solvent, viz. the diameter of the small particles. For this case, we can
(b) The separation is smaller than a particle confirm this prediction by a more primitive argument, that directly gives
diameter; particles cannot enter. The the magnitude of the attraction.
external osmotic pressure is not balanced We first consider Fig. 5.2, which depicts two surfaces immersed in a sus-
and there is a net attractive force per unit
area between the plates which is equal to
pension of smaller colloidal particles. These planes may be viewed as an
, the osmotic pressure of the reservoir. approximation to the surfaces of much larger colloidal particles immersed
in a suspension of smaller ones as in Fig. 2.1. When the particles are not
 Attractive forces: why colloids are sticky 121

excluded from the interplanar region, the pressures on each wall are bal-
anced. However, when the particles are excluded by their size from entering
the channel between the plates, the osmotic pressure  on the outer surfaces
which is associated with the suspended particles is not counterbalanced,
leading to an effective attractive force per unit area between the plates of
magnitude . The range of this interaction is approximately the particle
diameter.
What happens when the colloidal particles of Fig. 5.2 are replaced by
flexible polymer coils? There are two distinct cases: the polymers are either
attracted to or repelled from the solid substrate. Which case is operative in h
any particular situation depends upon the three media—polymer, substrate,
and solvent—and the interactions between them. Polymer adsorption due
to attraction to the surfaces often leads to repulsive colloidal forces and Fig. 5.3
will be discussed in the next section. The non-adsorbing situation leads to Two dimensional representation of a
depletion attractions. Indeed, when the solution is dilute and the polymer polymer coil confined to the region
coils are well separated from one another, the system is quite analogous to between two plates for h < RG . The
that depicted in Fig. 5.2. In fact the osmotic pressure which drives the plates circles denote blobs of diameter h.
together at separations smaller than the polymer radius of gyration, RG , is
simply  = (c/M)T , where c is the monomer concentration of chains
having degree of polymerization M. Of course, polymer coils are different
from colloidal particles in that polymers are deformable. Thus the chains
may distort and squeeze into the interplanar space even if the separation,
h, is small, h ≤ RG , as in Fig. 5.3. The free energy associated with this
deformation can be estimated (in a good solvent) as T per blob, as studied
in Problem 3.15. We discuss this confinement energy more carefully in
Chapter 6 to follow. Totalling over all the blobs in the channel yields a free
energy per polymer F given by

F (M/M(h))T M(a/h)5/3 T . (5.11)

As in Chapter 4 M(h) is the number of monomer units per blob of size h,

while a is the size of a monomer. The number of coils in the interstitial region
is suppressed by the Boltzmann factor e−F /T ; this decreases exponentially
with 1/h and the attractive force between plates will approach  = (c/M)T
exponentially with 1/h. This occurs for h RG . Thus there is only a
quantitative, not qualitative, difference between the polymer and small rigid
particle situations.
When the polymer solution is semidilute, the coils interpenetrate and form
the transient network discussed in Chapter 4. The solution of volume frac-
tion φ may be considered as a melt of blobs of size ξ , with M(ξ ) (ξ/a)5/3
monomers per blob. When h ≥ ξ and there are many blobs between
the surfaces, the polymer-induced interaction is weak. However, as the
plates approach one another and h < ξ , the polymers are squeezed out of
the channel and as h → 0, there is an osmotic pressure  φ 9/4 a −3 T
pushing the surfaces together. Therefore, even in a semidilute solution
there exists an attractive force between surfaces induced by depletion
effects. In that case, the range of the negative disjoining pressure is the
correlation length, ξ . Note that as the polymer solution becomes increas-
ingly concentrated, (i) the range of the disjoining pressure decreases
122 Colloids

from RG in dilute solutions to ξ aφ −3/4 in semi-dilute solutions;

(ii) the strength of the attractive force increases as φ 9/4 in the semidilute
range.

5.2 Repulsive forces

We have seen that there are many mechanisms which induce attraction and
potential aggregation of colloidal particles. These may be van der Waals
forces which are parametrized by the Hamaker constant H or indirect forces
associated with nonlocal deformation of the solvent matrix. Furthermore, as
the colloidal particles become larger, the binding energy between touching
dimers increases. Therefore, in the absence of opposing forces, Brownian
colloidal particles will eventually aggregate into sufficiently large clusters
that they sediment or cream out of suspension. Colloid stabilization then
requires one or more mechanisms to provide repulsive forces to counter-
act the attractive interactions. These repulsions are generally supplied by
elastic effects; usually some kind of deformable “bumper” is attached to
the particles. The bumper distorts as two particles approach one another.
This deformation results in an elastic restoring force, tending to keep the
particles apart. In practice, it is often the case that the bumper elasticity is
of entropic origin, so that its modulus scales like T . This is indeed the situ-
ation when the stabilization is carried out (a) by flexible macromolecules,
via steric stabilization or (b) by electric charge, via electrostatic stabilization.
Below we discuss each of these cases in turn.
Both of these methods give repulsive forces that have some character-
istic range R and that fall off rapidly for larger distances. Thus at very long
distances, the van der Waals attraction dominates, and the potential of inter-
action resembles that of Fig. 5.5. Under these conditions we can already
deduce some general requirements for colloidal stability. Since one cannot
compensate for the van der Waals attraction at all distances, one must at
least assure that it is weak enough to be harmless in those regions where it
dominates. As we showed in Chapter 2, two particles have a substantially
increased likelihood of being together whenever their attraction has strength
of T or more. If two isolated particles occupy a volume , and have attrac-
tion of strength V0 over volume ω their net probability of being together
is of order exp |V0 /T | (ω/). Thus when V0  T log(/ω), the particles
are likely to be together. In a solution of many such particles, any particle
is likely paired under the same conditions, where now  is the volume per
particle. This strong association creates phase separation.
To avoid this phase separation we must have |V0 |  T . (Here we have
ignored the weak log  dependence.) Since the van der Waals energy is a
substantial part of the total near this minimum, we may use Eq. (5.7) to
estimate V0 . If h0 is the separation at the minimum, stability requires that
−kH (b/h0 )  T , or
h0  b(H /T ).
For many particles and solvents, H happens to be of order T . This tells us
that we must provide a repulsion sufficient to push the attractive minimum
out to a separation h0 of the order of the particle radius b. Accordingly,
 Repulsive forces 123

the range of the repulsion must be of order b or more. A shorter-ranged

repulsion will not do.
5.5. Depletion binding of colloidal particles Consider two identical spherical col-
loidal particles of radius b which are just touching each other while immersed
in a semi-dilute, good solvent polymer solution of volume fraction φ and
correlation length ξ . The polymers do not adsorb on the particles. Thus the
concentration is depleted at small distances z from the surface: φ(z) → 0 for
z  ξ and φ(z) → φ for z  ξ . If b  ξ , estimate the binding energy
of the dimer arising from the depletion forces i.e., the work required to sep-
arate the particles. [Hint: Use the concept of the Derjaguin approximation of
Problem 5.3.]

5.2.1 Steric stabilization

We have seen in the last section that polymers which do not adhere to
surfaces immersed in a solution induce attractive forces between them.
This interaction arises from the depletion of polymers (compared to the
bulk solution) in the channel between the colloidal grains. If the polymers
were prevented from escaping from the interstitial region, their compression
as the particles approach each other would lead to an osmotic repulsion. This
may be seen by again referring to Fig. 5.3 and the associated discussion for
dilute polymer solutions. The free energy cost in deforming a polymer coil
trapped between two surfaces is given by Eq. (5.11). If there exist N of these
chains per unit area of surface (where N is the number of nonoverlapping
trapped polymers per unit area), the corresponding free energy penalty per
unit area F is given by F = N F , and there is a disjoining pressure,
dF 5  a 5/3
d = − MN h−1 T , (5.12)
dh 3 h
if the chains are not permitted to leave the interstitial region. This trapping
may be engendered by the attachment of the polymers to the surface by,
e.g., formation of covalent chemical bonds between an end monomer and Fig. 5.4
Sketch of a colloidal particle with
the surface, as in Fig. 5.4. chemically end-bonded polymers which
Such bonds are generally sufficiently strong to prevent the polymers are non-adsorbing. The polymer hairs act
from leaving the surface except under the harshest treatments. Is this as bumpers opposing aggregation.
disjoining pressure strong enough to prevent binding of the particles under
the influence of van der Waals forces? This question may be addressed by
combining Eqs. (5.6) and (5.11). The form of the interaction free energy Free energy
between the surfaces is sketched in Fig. 5.5. The absolute minimum is at
contact between the surfaces, but there exists a barrier to overcome in order
h
for the two surfaces to feel the strong binding at contact. If the barrier height
is much greater than T , the adhesive contact is effectively unattainable dur-
ing any laboratory experiment. Note that there is a secondary minimum at
larger separations (h ≥ RG ), where the polymeric bumper is not deformed
but there still remains van der Waals attraction. If the depth of this mini- Fig. 5.5
mum is several T , the particles may aggregate in this minimum, making Sketch of the disjoining free energy versus
loose aggregates called flocs. The maxima and minima are determined by the interplanar separation taking into
the disjoining free energy per unit area multiplied by an appropriate area; account the repulsions from dilute grafted
hairs and van der Waals attraction. The
for example, for the flat plates this is the area of the surfaces. For spheres absolute minimum is at contact with a
of radius b, large compared to the nearest interparticle separation, in the barrier and a secondary minimum is at
Derjaguin Approximation this area is proportional to b. larger separations.
124 Colloids

The strength of the polymer-induced interactions depends on how much

polymer is in the system under study. For example, if the barrier (Fig. 5.5) is
not sufficiently high to prevent aggregation with dilute hairs, the concentra-
tion of polymer that is grafted to the surfaces may be increased (with some
chemical ingenuity) into the regime where their radii of gyration start to
overlap; this is called the brush regime and is depicted for a planar surface
d in Fig. 5.6.
In order to analyze the contribution of the polymer brush to the disjoining
pressure, we first give a basic description of the height L (Fig. 5.6) of the
L isolated brush due to Alexander and deGennes [7]. The excluded volume
repulsions between the overlapping polymer coils provides an internal pres-
Fig. 5.6 sure, pushing the chains into the solution so as to reduce the internal
Sketch of a polymer brush grafted to a solid monomer concentration.
substrate. The height of the brush is L; the
average distance d between grafting points
The equilibrium height is then determined by a balance between the
is less than the radius of gyration, RG , of excluded-volume repulsions and the stretching elasticity of the polymers.
isolated coils. Thus in good solvents, To estimate these, we may use the blob scaling arguments of the Chapter 4.
adjacent chains are strongly interacting. Chains of M monomers of size a are grafted to a surface at separation
d  a but substantially smaller than the chain size. In that case the local
environment is semidilute, and we may describe it in terms of blobs of size ξ ,
with k blobs per chain. Now we can express the interaction energy U of each
chain (the energy required to insert it against the local osmotic pressure) in
the form U T k. This energy favors few blobs per chain and thus dilute
conditions. The opposing stretching energy S may be found by treating
the chain as a random walk of blobs. Each chain must stretch to a height of
order of the thickness of the polymer layer L. Thus S T (L/ξ )2 (1/k). The
thickness L in turn determines the concentration of blobs: 1/ξ 3 k/(d 2 L).
Thus L kξ 3 /d 2 . We may now express the stretching energy S in terms
of k and ξ :
S T (kξ 2 /d 2 )2 (1/k) T k(ξ/d)4 .
We see that both energies U and S are proportional to k; thus the optimal
volume fraction is independent of M. At the optimal volume fraction S
U , so that 1 (ξ/d)4 . Thus ξ d. The blob size and concentration are
independent of the chain length M; thus the height of the layer is simply
proportional to M; the chains become stretched as anticipated. The force
per unit area needed to compress the layer is of the order of its osmotic
pressure:
 T /d 3 . (5.13)
The effective barrier against van der Waals-induced aggregation may thus
be augmented by increasing the grafting density, which decreases d in
Eq. (5.13). Within this Alexander–de Gennes approximation for the grafted
brush, the effective disjoining pressure is given by Eq. (5.13) for h ≤ L and
drops abruptly to zero for h ≥ L. In fact the dropoff is gentle, because not
all the chains stretch to the same height L. The chains gain both stretching
and osmotic free energy if some stretch more and some less. The optimal
concentration profile, is close to a parabola [7]. The resulting disjoining
pressure has been calculated [8] and is shown in Fig. 5.7. The Alexander–
de Gennes scaling form for the brush height is preserved but the detailed
concentration profile eliminates the spurious abrupt drop-off.
 Repulsive forces 125
100,000

10,000
Force/radius (μNm–1)

1000

Fig. 5.7
Measured forces between two surfaces
100
coated with grafted polystyrene chains
about 1400 monomers long in toluene,
after [9]. The curves show predicted force
using free-solution properties of
10 polystyrene in toluene, together with the
0 200 400 600 800 1000 1200 1400 parabolic-profile theory described in the
Separation (Å) text, after [8].

In these last paragraphs, we have discussed how polymers which are

intrinsically repelled by solid surfaces may stabilize colloidal dispersions b
when the chains are firmly end attached to the substrates. However it would
be more economical to use homopolymers, since no special chemistry would
need to be performed to get the polymers to attach to the surfaces. Indeed,
this can be accomplished by using polymers which adsorb onto the surfaces
of the solids to be stabilized. A cartoon of this situation is given in Fig. 5.8.
Then the polymer loops can act as bumpers keeping the particles out of
adhesive contact. This scheme may indeed work, but it is quite delicate [1]. Fig. 5.8
A sketch of two interacting colloidal
In fact several conditions must be satisfied for homopolymer stabiliz- particles with adsorbed polymer. Note the
ation. These include: (i) irreversible adsorption; (ii) sufficient polymer deformation of the polymer distribution in
to saturate the surfaces; (iii) a sufficiently thick polymer corona. These the interstitial region.
conditions will now be discussed quantitatively.
Let us consider the picture, Fig. 5.8, in more detail. The “bumper” view
is that the coronae, which extend a finite distance, L, into the solvent in
a series of loops and tails when compressed by two particles approaching
one another, cause an osmotic repulsion, as discussed in the first paragraph
of this section. The range of this repulsion, L, must be such that the van
der Waals attractive energy between the particles when the coronae are
just touching is less than about T . Thus the minimal stabilizing corona
thickness L should be proportional to the particle size, L ∝ b, as noted
at the beginning of this section. However, for the adsorbed polymer to
produce an osmotic repulsive force between two surfaces, the free energy
of the interstitial chain must be increased. (This is because the disjoining
pressure is d = −∂F /∂h. Eq. (5.12).) Such an increase in free energy
per chain would take these polymers out of thermodynamic equilibrium
with the bulk solution. Thus the polymers would desorb, reducing any
repulsion. This is another example of the Perturbation-Attraction Theorem.
Consequently, to achieve the strongest repulsive forces, the polymers should
126 Colloids
† The conditions for the Perturbation- be out of equilibrium with the polymer dissolved in the solvent† . We discuss
Attraction Theorem as stated in Appendix A below why this might indeed occur for adsorbing polymers. Condition (ii),
are not strictly met in the polymer system.
concerning surface saturation, arises from the fact that the polymers are
Each tail can transmit some force directly to
the surface, as with grafted polymers. Thus adsorbing; when the two surfaces are within a distance that can be spanned
we cannot conclude rigorously that surfaces by one coil, a given chain can adsorb on both surfaces to form a bridge.
with equilibrium-adsorbed polymers must Such bridging will contribute a negative term to the polymer free energy
attract. Recently a careful treatment [10] has and thus an attractive component to the force between surfaces. If, however,
suggested that equilibrium-adsorbed poly-
the surface is effectively covered by polymer, adsorbing sites will not be
mers repel weakly at long distances.
available and this bridging will be screened out. Thus, effective stabilization
requires well-coated surfaces.
L Some of the energetics of adsorbed polymers may be understood by a
simple argument given by de Gennes [11] for the geometry of a single
adsorbed chain. Referring to Fig. 5.9, the adsorption free energy of the
polymer may be approximated as

F /T (R0 /L)2 − Mx, (5.14)

where R02 = N a 2 is the characteristic size of a Gaussian chain, T is the

adsorption energy gained per segment in contact with the surface, and x
Fig. 5.9 is the fraction of segments which contact the surface. The first term is the
Sketch of a polymer chain adsorbing on a free energy required to confine the chain to a region of thickness L. This
flat surface. Note that the coil is confinement energy was first encountered in Problem 3.15. The second
constrained in the direction normal to the
surface and spreads out along the surface.
term is the local gain in free energy when monomers touch the surface, or
adsorption energy. Making the simple assumptions that the monomers are
uniformly distributed throughout the slab and that x a/L, and minimiz-
ing f with respect to L, we obtain a/L . Thus, provided  > M −1/2 ,
the polymer is strongly deformed from its Gaussian conformation. Typic-
ally,  ≈ 1. The inequality is easily fulfilled for M ≥ 100. (In the presence
of excluded volume interactions, each term in Eq. (5.14) is modified but
the result for L remains essentially unchanged. We discuss adsorption in
more detail in Chapter 6.) The adsorption free energy per chain is M 2 T
which may be substantially greater than T . This is essentially the mag-
nitude of the energy barrier in the Eyring formula in Chapter 2 for the
kinetics of polymer desorption. If M 2  1, the desorption rate decreases
exponentially in time; this means that often, especially for high molecu-
lar weight polymers, the polymers cannot jump off the surface during an
experimental observation time. Under such conditions, the chains may be
considered to be irreversibly adsorbed, i.e., out of thermal equilibrium
with the dissolved unadsorbed coils. This explains why the adsorbed poly-
mers may be practically considered to be trapped on the surfaces and the
Perturbation-Attraction Theorem fails to be applicable.
Thus adsorbing high molecular weight polymers are often irreversibly
adsorbed because the overall barrier height for desorption is proportional to
M. However, the layer thickness, L, is only a few nanometers (for  10 1
)
and independent of molecular weight. This would not be a thick enough
layer to stabilize the colloidal particles against van der Waals-induced
aggregation. Fortunately, one may readily achieve thicker layers whose
surfaces are saturated with polymers. If the surfaces are not saturated with
polymers these gain several T in free energy by adsorbing. Accordingly
 Repulsive forces 127

their probability of being desorbed in solution is exponentially small. Thus,

even for quite dilute solutions, adsorbing surfaces can be saturated and each
adsorbed chain will have many chains nearby. Then, even for dilute solu-
tions, the local concentration in the interfacial region may be sufficiently
high that it is in the semidilute regime. Different chains will be compet-
ing for adsorption sites on the surface, leading to a reduced number of
surface contacts per polymer. This, in turn, induces chain swelling and a
corresponding thickening of the surface layer relative to the single chain
situation.
How thick will the layer become? In Chapter 4, it was shown that the
characteristic length over which there are significant concentration vari-
ations in a polymer solution is the correlation length, ξφ , which varies from
the polymer radius, R, in dilute solutions to ξφ ∝ φ −3/4 in semidilute solu-
tions. This suggests that, in solutions with sufficient polymer to saturate
the surfaces, the coating thickness corresponds to ξφ . In Chapter 6, it will
be shown that this result indeed follows from a more careful calculation of
the adsorption profile. Then the interaction range, L, should be L ξφ .
In dilute solutions where ξφ R, this implies that the optimal polymer
molecular weight for stabilization of colloidal particles of size b should be
given by M (b/a)D where D is the fractal dimension relating chain size
to chain length, R M 1/D a. The molecular weight of adsorbed polymers
which is required to stabilize colloidal particles increases with nearly the
square of the particle dimension. This is in contrast to grafted chains, for
which the molecular weight only grows linearly with particle size. Thus
more polymer is required in the adsorption case, but less chemistry is
necessary to synthesize specific grafting bonds. Notice that increasing the
polymer concentration in the solution is deleterious to colloid stabilization.
This is because the increased osmotic pressure of the solution squeezes the
adsorbing layer and reduces the thickness from R to ξφ as the concentration
increases. Thus the corona thickness L becomes molecular weight inde-
pendent and decreases with increasing concentration as φ −3/4 . Too much
polymer destabilizes the suspension! Thus colloidal control with adsorbing
homopolymers is delicate and requires attention to the molecular weight
and concentration for the given particulate system to be controlled.

5.2.2 Electrostatic stabilization

In polar solvents where hydrocarbon polymers are rather difficult to dis-
solve, colloidal particles are often stabilized by attaching electrically
charged groups to the surfaces, creating locally unbalanced charges and
consequently Coulomb repulsion between the particles. The high dielec-
tric constant of such solvents permits charge separation and ionization, as
explained in Chapter 7. Nevertheless global neutrality is generally main-
tained, and the nature of the electrostatic interactions must be considered
in this context.
Let us first consider a salt solution in which monovalent salt molecules,
e.g., NaCl are dissolved. Assuming that all the molecules are dissociated,
there is then a concentration, c, of ions, half Na+ cations and half Cl−
anions. What is the effective potential experienced by a test charge at r
128 Colloids

arising from, e.g., a Na+ ion at the origin? The Na+ ion attracts Cl− ions
and repels other Na+ ions. Therefore if the test charge is at a large dis-
tance (compared to the average interionic spacing d ≡ c−1/3 ) from the
origin, Gauss’ Theorem would yield a reduction in the effective electro-
static potential: (r ). This reduction is called Debye–Hückel screening.
The form of the Debye–Hückel screening may be derived directly from
Maxwell’s electrostatic equation (written in cgs units)

∇ 2 (r ) = −4πρ(r ), (5.15)

where  is the dielectric constant (80 for water),  is the electrostatic poten-
tial and ρ(r ) = ρ+ (r ) − ρ− (r ) is the charge density at r; ρ± (r ) are the
anion and cation concentrations. In a solution, the charges are mobile and
may adjust to the local forces that they experience; the position of each ion
is governed by its Boltzmann probability owing to the local potential that
it experiences.
0
ρ± (r ) ρ± exp(∓ e(r )/T ), (5.16)

where ρ± 0 are constants that adjust the total number of ions of each species

to correspond to the concentration of ions dissolved, i.e., c = (ρ+ +
ρ− )(r ) d 3 r, where  is the total volume of solution. We now suppose that
the potential seen by an ion is that arising from the mean charge density
ρ(r) of Eq. (5.15). Combining Eqs. (5.15) and (5.16) generates a nonlinear
second order partial differential equation for the self-consistent electrostatic
potential, (r ):
    
2 0 −e(r) 0 e(r)
∇  = 4π −ρ+ exp + ρ− exp . (5.17)
T T

This is called the Boltzmann–Poisson Equation. It is not exact, because it

neglects correlations between the ions, as well as their fluctuations. Each
ion is taken to move independently in the average potential generated by
the other ions.
If the electrostatic potential energy range is smaller than T , the
Boltzmann–Poisson equation may be linearized by expanding the expo-
nentials to first order in a Taylor series, yielding the Debye–Hückel
equation:
∇ 2 (r ) = κ 2 (r ), (5.18)
where κ −1 is the Debye screening length and is given by

κ 2 = (4πe/T )(ρ+
0 0
+ ρ− ). (5.19)

For the monovalent salt solutions, ρ+ 0 = ρ 0 = 1 ec and the Debye length

− 2
may be expressed as κ 2 = 4π c, where  ≡ e2 /T is the Bjerrum length
encountered in our discussion of polyelectrolytes ( 0.7 nm for water).
A 1 mM salt solution corresponds to a Debye length of approximately 10 nm.
If a positive ion is assumed to be at the origin, the solution of Eq (5.18) is
 Repulsive forces 129

the screened Coulomb potential,

e(r) = T/re−κr , (5.20)

which is of the same form as the Yukawa potential that is familiar from nuc-
lear physics. This is a reasonable approximation if there are many screening
ions in a screening length, which corresponds to /d  1, as may be verified
using the formula for κ.
5.6. Debye–Hückel screening (a) Show that the condition that there are many
screening ions in a Debye length κ −1 is equivalent to /d  1 for monovalent
salts, where d is the mean spacing between ions in solution and  is the Bjerrum
length. (b) What is the corresponding result for multivalent salts? (c) For a
CaCl2 solution in water, what is the range of concentrations for which the
Debye–Hückel approximation is valid?
A natural extension of Eq. (5.20) to the case of a colloidal particle of size
b and having Z ionizable groups firmly anchored to the surface is
 −κ(r−b) 1
e(r) = ZT e . (5.21)
r κb + 1
However this is typically an overestimate of the strength of the repulsive
potential between charged colloidal particles. This can be seen by consid-
ering the potential energy of a negative counterion at the surface of the
particle, e(b) = −4π bσ T , where σ is the number of anchored charges
per unit area. For b 1 μ and σ 10−4 / nm2 , this gives |e(b)| 104 T .
There is therefore a strong attractive force driving the counterions to recon-
dense on the particles, so that the particle is incompletely ionized. This arises
because the strong, generally covalent anchoring of the surface charges
leads to very high local electric fields. This results in a reduction of the
effective charge, Z ∗ , of the particle as experienced by a distant test charge.
We can readily estimate Z ∗ for colloidal particles whose radius b is much
larger than the screening length κ −1 , the typical case. The surface charge
gives an enhanced density of counterions near the surface. The Boltzmann
enhancement factor is exp(e(b)/T ), where e(b) Z ∗ T /b is the sur-
face potential. The enhancement acts over a volume of order b2 /κ. The
probability that the charge is in the enhanced region is the Boltzmann-
weighted ratio of the enhanced volume to the volume per ion 1/c. This
ratio is about 1 when eφ(b) −T ln(cκ/(b2 )) or

Z ∗ −(b/) log(cκ/b2 ). (5.22)

5.7. Debye–Hückel potential of a sphere Derive Eq. (5.21) from the Debye–
Hückel equation (5.18).
For the numerical example above, this gives Z ∗ /Z 10−3 ; the effective
charge is only about 0.1% of the nominal charge. The effective electro-
static potential is then given by Eq. (5.21) with Z replaced by Z ∗ . There
exists some evidence for this reduction in effective charge in mixed micellar
systems [12] and charged colloids [13].
Is the electrostatic repulsion between charged colloidal particles suffi-
cient to stabilize them against the flocculation induced by the van der Waals
130 Colloids

attractions of Eq. (5.7)? In terms of the effective charge, the Coulomb

repulsion dominates the van der Waals attraction (at short distances) when
Z ∗2  (b/)(H /T ). However, at interparticle separations exceeding about
κ −1 the van der Waals force, which only decays as a power law, wins, giving
a force curve qualitatively similar to Fig. 5.5. Flocculation in the secondary
minimum is prevented if T  H (κb). If we consider the Debye length,
κ −1 , to be the corona thickness in this case, as with polymer stabilization,
the corona thickness must grow in proportion to the particle size to prevent
destabilization. If too much salt is added to a charge stabilized suspension,
it separates; this phenomenon is called salting out.
5.A Suggested experiment: Salting out Skim milk is a colloid whose largest
particles are spherical casein micelles about 100–200 nm in diameter [14]. These
casein micelles are stabilized by a short brush of charged polymers. According
to the discussion in the text, it should be possible to make the micelles aggregate
by adding salt so that the screening length is only a small fraction of the radius.
The project consists of testing this hypothesis. A suggested approach is first to
dilute the milk with distilled water so that the micelles are far apart and so that the
solution is just vaguely cloudy. Try to determine the micelle size by shining a laser
through the suspension. As implied in the Chapter 4 under “semidilute diffusion”
the flickering time is the time for a micelle to diffuse a distance λ/θ where θ is
the (small) angle between the beam and the observed speckle point. Determine
the amount of salt needed to make the Debye screening length equal to the radius.
Make salt solutions with different amounts of salt added, ranging from two orders
of magnitude
√ too much to two orders of magnitude too little. Make eight samples,
each 10 more dilute than the last. Then add 5–10% full-strength skim milk to
each sample and mix. (Adding milk to salt water is better than adding concentrated
saltwater to milk. You do not want the milk to see saltier water than it should,
even for a moment before it gets mixed.) Shine the laser pointer through each
sample and look for signs of scattering intensity increasing with time, indicating
aggregation. Leave the samples in the refrigerator over night and see whether any
samples undergo macroscopic precipitation. If so, play with the precipitate and try
to determine whether the particles are stuck together tightly or loosely. Weigh the
wet precipitates, let them dry and weigh them again, to determine what fraction
was water. A smaller Debye screening length should lead to a denser precipitate.

The reduction in the effective charge is sometimes called charge renor-

malization. When it is expressed as an effective surface charge density, σ ∗ ,
we find σ ∗ ln φ/(4π b). Note that σ ∗ → 0 as b → ∞. This implies that
all the counterions are bound for large particles, and therefore the use of
the Debye–Hückel approximation is suspect. This result can be understood
fairly easily in terms of the nonlinear Boltzmann–Poisson equation. Con-
sider the special case of a flat half space with a surface charge density σ
(of immobilized charges) that is in contact with pure water, i.e., there is
no dissolved salt. The only ions in solution are the counterions associated
with the surface charged groups. If the surface is negatively charged, the
Boltzmann–Poisson equation can be written

d 2φ
− = (2λ2 )−1 e−φ , (5.23)
dz2

where z is the coordinate perpendicular to the surface, φ(z) ≡ e(z)/T is

the potential measured from the surface and λ is given by λ2 ≡ T /(8π eρ 0 ).
 Repulsive forces 131

With this definition of λ, ρ 0 is the free charge density adjacent to the sur-
face. One readily verifies this by evaluating Eq. (5.23) at the surface, and
recognizing Poisson’s equation with charge density ρ 0 . We must solve the
equation in order to find this ρ 0 . This one dimensional Boltzmann–Poisson
equation has the solution,

φ = 2 ln[1 + (z/2λ)]. (5.24)

This may easily be verified by inserting (5.24) into (5.23). We note that
φ → ∞ as z → ∞. The boundary condition which is satisfied at the surface
is given by Gauss’ Theorem and is ∂φ/∂z|z=0 = 4π σ . The corresponding
charge distribution is

ρ = ρ 0 [1 + (z/2λ)]−2 . (5.25)

The constant amplitude

∞ ρ 0 is computed from the condition of charge neut-
rality, eσ = 0 ρ(z) dz. This results in λ = (4π σ )−1 . The result that
ρ(∞) = 0 implies that all the counterions are bound to the surface and
there is no ionization, in agreement with the charge renormalization argu-
ment given above. The majority of the counterions are contained in a layer
of thickness λ adjacent to the interface.
The length scale λ, known as the double-layer thickness or the Gouy–
Chapman length is an important measure of surface charge. It is evidently
the distance from a surface of charge density σ in empty space whose poten-
tial energy φ differs from that at the surface by an amount T . Alternatively,
it may readily be seen to be the Debye screening length associated with a
slab in which the ions are self-consistently confined, i.e., λ−2 4πρ 0 /e,
with ρ 0 = eσ λ. Thus the charge renormalization discussed above for
spherical particles arises from the nonlinearity of the Boltzmann–Poisson
equation, which is neglected in the Debye–Hückel approximation. For a
typical charged surface with an area per charge of approximately 0.5 nm2 ,
the double layer thickness, λ, is of order 1 nm. This small value for the
characteristic length scale over which the counterions are confined, brings
into question the validity of the continuum Boltzmann–Poisson equation.
Numerical simulations [15] have shown that for monovalent counterions
these results are relatively accurate. However for multivalent ions (see
Problem 5.8), λ is much smaller and the Boltzmann–Poisson approximation
is even qualitatively inaccurate [16]† . Here we restrict our attention to the † When ions have a valency of more than
monovalent case for simplicity, bearing in mind that the important situations
one, the Coulomb energy of two adjacent
with polyvalent counterions must be treated more microscopically. ions is often substantially greater than T ,
5.8. Multivalent counterions (a) Determine the Debye screening length κ −1 for a and the ionic positions become highly cor-
related. One consequence of these correla-
solution of m different ionic species each having a valency zm and concentration
tions is overcharging: Counterions continue
ρm . (b) What is the double layer λ thickness when the counterions have valency
to be attracted to the surface even after
z? Estimate λ for a surface charge density corresponding to an area per charge
it has been neutralized. Overcharging can
of 5 nm2 . thus reverse the net charge on a surface. It
The disjoining pressure between two identical charged surfaces in the can cause two like-charged surfaces to be
bound together, with a layer of counterions
absence of added salt is considered in Problem 5.9. However, an excel- in between. Ions with a valency of two or
lent estimate may be obtained from the single surface solutions to the three appear to play an important role in
Boltzmann–Poisson equation treated above. If the separation between the binding macro-ions like DNA together.
132 Colloids

two surfaces is 2h, then by symmetry there is no electric field on the mid-
plane, i.e., ∂φ/∂z|z=h = 0. Then the only force that can be transmitted is
associated with the osmotic pressure  of the counterion “gas”:

 = T ρ(h)/e. (5.26)

Within the Boltzmann–Poisson approximation, the calculation of the dis-

joining pressure is then reduced to the determination of the counterion
concentration on the mid-plane. For h  λ, this may be approximated
by twice the value arising from a single surface (the sum of the contribu-
tions from each interface) concentration at h, i.e., ρ(h)/e (πh2 )−1 , and
a disjoining pressure,
 T (π h2 )−1 . (5.27)
In the absence of added salt, the repulsive electrostatic force between the
surfaces decays through a power law. With added electrolyte, the pressure
decays asymptotically as e−κh . For pure pH 7 water without added electro-
lyte, the Debye length, κ −1 , is a few microns; this sets a practical upper limit
for the effective range of the electrostatic interactions in aqueous solvents.
Quantitative experiments, e.g., using the surface force apparatus [3] have
validated the basic ideas embodied in this treatment.
5.9. Disjoining pressure between charged interfaces Consider two identical
charged surfaces of charge densities σ separated by a distance 2h in water.
There is no added salt and the neutralizing counterions are monovalent. (a) By
solving the Boltzmann–Poisson equation, determine the counterion distribu-
tion, i.e., ρ(z). (b) Find an expression for the disjoining pressure and compare
to Eq. (5.27).

5.3 Organized states

The interactions between colloidal particles may be sufficiently strong to
induce various forms of long range ordering analogous to self-organization
of atoms and molecules into crystals or rigid nonsymmetric molecular
liquids into liquid crystals. In this section, we discuss briefly several such
cases.

5.3.1 Colloidal crystals

One of the best studied of the ordered colloidal systems is that of charged
latex particles in low ionic-strength water [17]. These are spherical plastic
particles generally in the range of from 10 nm to tens of microns in dia-
meter. They are often prepared by polymerization of species such as styrene
in emulsion droplets which are charge stabilized in water by sulfonic acid
containing surfactants [18]. This results in quite uniformly sized spherical
particles with surfaces decorated with ionizable groups. These particles
interact principally via a repulsive Yukawa potential, Eq. (5.21), which
may be strong enough to induce crystallization. This occurs when the inter-
action energy gained by localizing the particles on a lattice (and hence
maximally far from one another) overwhelms the translational free energy
associated with the particle delocalization into a liquid. Assuming highly
 Organized states 133

charged particles, so that the charge renormalization condition is applicable,

the scaling form for the freezing may be written as b2 /a > k, where k
is a constant of order 10 and a is the mean interparticle spacing. Note that
this implies a minimum particle size kb 5 nm, below which there can
be no freezing under the action of electrostatic forces. Generally freezing
occurs when the Debye screening length is close to the interparticle separ-
ation. A more complete phase diagram has been given by [19]. Often the
colloidal crystals have lattice constants which are comparable to the particle
dimensions, which is often in the range of the wavelength of visible light.
Thus diffraction may be easily observed. These colloidal crystals are quite
soft, and because of the ease of imagining the building units, they are often
studied as a paradigm for real crystals and their properties, such as defect
structures and other metallurgical aspects. The motion is, however, rather
different because the “atoms” are embedded in a viscous fluid.

5.3.2 Lyotropic liquid crystals

Suppose the colloidal particles are not spherical but, e.g., rod-like, with
length L and radius b, with L  b, then there is an additional degree of
freedom, namely the orientations of the rods. (An example of such relatively
rigid rods is provided by various viruses such as Tobacco Mosaic Virus (a)
which is 300 nm long and 18 nm in diameter.) This allows the possibility
of another state of order in which the rods throughout the liquid tend to
point in a common direction. Such an arrangement, similar to logs in a log
jam, is called a nematic liquid crystal. The liquid spontaneously chooses
an alignment direction, and thus not all directions are equivalent as they (b)
are in an ordinary liquid. This loss of directional equivalence amounts to a
spontaneously broken symmetry.
The reason for nematic symmetry breaking was explained by Onsager
[20]. In Chapter 4, the excluded volume parameter, b2 , was defined, and
in Problem 5.10 it is determined that for an isotropic distribution of rigid
rods, b2 ∝ bL2 , which is L/b larger than the actual rod volume. This
Fig. 5.10
result suggests the possibility of a phase transition to a state of parallel rods
(a) Sketch of an isotropic solution of rods.
where the effective excluded volume is simply the geometric volume. This (b) Weakly nematically ordered rods.
provides the mechanism for the formation of lyotropic liquid crystals as a
function of increasing concentration of rods; they can simply pack more
efficiently. The disordered and nematically ordered states are shown in
Fig. 5.10. With ever increasing concentration even more types of order may
develop. For example, a smectic liquid crystal is a layered structure in which
there is no center of mass order in the layers, as illustrated in Fig. 5.11. Thus
each layer is a two dimensional fluid. An excellent, detailed treatment of Fig. 5.11
liquid crystals is given in the book by de Gennes and Prost [21]. A smectic liquid crystal.

5.10. Rod excluded volume A slender rod of length L and radius b  L is centered
at the origin and pointed along the z axis. A second identical rod lies along
the x axis at a distance x from the first. Its orientation is arbitrary.

(a) What is the probability that the second rod will intersect the first? Ignore
numerical prefactors and consider the dependence on L, b, and x. How
small must x become in order for the probability to be of order unity?
134 Colloids

(b) Using the results of (a), estimate the mutual excluded volume of two such
rods if they are dispersed in a solution. Again, ignore numerical prefactors.
(c) Repeat parts (a) and (b) with the rods replaced by disks of diameter L
and thickness b. Compare with the result for the rods. Compare with the
result for hard spheres of diameter L found in Chapter 4.

5.3.3 Fractal aggregates

Much of the discussion of this chapter has centered on the methods to pre-
vent colloidal aggregation and suspension instability. But there are cases
where aggregation is desirable. Here the strong tendency to aggregate can be
turned to advantage to produce distinctive and useful aggregated structures
like those of Fig. 1.5. These aggregates are different from ordinary precip-
itates that form when small molecules attract one another in a solvent. The
difference arises because the attraction between the particles of Fig. 1.5
is very strong. It is so strong that particles once joined cannot move or
unstick. The randomness in these aggregates is thus of a different type than
the thermal randomness we have encountered up to now. Accordingly, a
different approach is needed to analyze it. These aggregates, like polymers,
are fractal structures. Their fractal dimension depends on the conditions of
aggregation. When aggregation occurs as fast as possible and the rate is
limited only by the Brownian movement of the aggregates, the process is
said to be diffusion limited, and the fractal dimension D is observed to be
about 1.7. When the joining of particles is strongly inhibited by an activa-
tion barrier (as discussed in Chapter 2) the aggregation is said to be reaction
limited. Then the fractal dimension D is observed to be about 2.1.
Many properties of fractal aggregates are analogous to those of polymers.
For example, fractal aggregates have hydrodynamic radii comparable to
their geometric sizes. Some properties of these aggregates are qualitatively
different from those of polymers, because aggregates are rigid structures
that hold their initial shape. If an external force distorts an aggregate, it
responds with an elastic restoring force. This force arises from the joining
interaction that holds the constituent particles together. The scaling of this
restoring force with the aggregate size can be derived in terms of geometric
scaling properties: viz. the fractal dimension D and another geometrically
defined exponent. In Appendix B we lay the groundwork for a quantitative
description of fractal aggregates and their properties.

5.3.4 Anisotropic interactions

In the previous sections, the forces which exist between spherical colloidal
particles have been assumed to depend only on the distance between them.
Anisotropic effects, e.g., liquid crystalline ordering, were associated with
nonspherical objects. However, even spherical particles may have angular
dependences in their interactions if there is an internal direction defined
within the particles. For example, the particles may be ferromagnetic, hav-
ing a permanent magnetic dipole moment, or they may be electrically or
magnetically polarizable, and respond to external electric or magnetic fields.
In these cases, the interaction between the particles depends on the angle
that the interparticle vector makes with the external field. Suspensions
 Colloidal motion 135

of ferromagnetic particles, like magnetite, iron, and cobalt, are called

ferrofluids, and have applications to fluid seals, printing, etc. [22]. Elec-
trorheological fluids, e.g., starch suspensions, are dispersions of particles
having large electric polarizabilities, which develop electric dipole–dipole
interactions in the presence of externally applied electric fields [23]. These
systems are being investigated for clutches and self-adjusting suspensions
in automotive applications.
The new feature to be considered in a fluid of dipoles is the dipolar
energy. We consider a suspension of spherical particles which possess
identical net dipole moments, μ. For concreteness, they will be assumed
to be ferromagnetic with magnetization M and of radius b. In a typical
ferrofluid, M 100 Oe and b 10 nm, so that the dipole moment
μ Mb3 4 × 10−16 cgs units. The dipolar interaction energy between
two particles separated by a vector, r (r > b), is given by

U1,2 = r −3 [μ
1 · μ
 2 − 3(μ  2 · r)r −2 ].
 1 · r)(μ (5.28)

The strength of the dipolar interaction between particles may be expressed

in terms of a dimensionless coupling constant, λ ≡ μ2 /4b3 T . At ambi-
ent temperatures, λ 1, which implies that the dipolar interaction may
significantly influence the organization of the colloidal suspension. When
λ ≥ 1, the anisotropic nature of the dipolar energy favors a head-to-tail con-
figuration of the dipoles and may lead to chain formation, If the strength of
the dipolar interaction is sufficient, some form of phase separation might
be expected, but the complete phase behavior in the zero external field has
yet to be determined.

5.4 Colloidal motion

A colloidal suspension shows the same kinds of time-dependent properties
as a polymer solution. The particles diffuse and sediment. The fluid as a
whole flows in response to stress. Colloidal particles are essentially solid
spheres. The motion of hard spheres was explained in Chapter 4. Thus we
need only recall those previous results here.
In the dilute state, colloidal particles undergo Brownian motion, charac-
terized by a self-diffusion constant. This is the Stokes diffusion constant
of Eq. (4.7), varying inversely with the particle size. This same diffusion
constant tells the speed of motion in response to a given force, such as sed-
imentation under gravity, according to the Einstein relation. Particles with
a stabilizing corona of polymers have a hydrodynamic radius large enough
to include most of the corona, since this polymer layer is hydrodynamically
opaque. This is true whether the corona is a brush or an adsorbed layer.
Colloids stabilized by charge may in principle suffer an analogous effect.
The countercharge in the Debye screening cloud around a moving charged
particle feels electrostatic forces tending to drag it along with the particle.
These forces in turn tend to entrain the fluid in the Debye zone. These
effects are subtle and cannot readily be summarized. In any case, they slow
the diffusion by at most a factor of order unity. Such “electroviscous effects”
are treated, e.g., in [24]. When colloidal particles repel each other strongly,
136 Colloids

Brownian motion is inhibited, as with polymers. Each particle becomes

entrapped in the cage formed by its neighbors. This glassy effect occurs
with small molecules as well, and we mentioned it in our discussion of vis-
cosity in Chapter 2. In such systems, as with polymer solutions, diffusion
is characterized by a slowed self-diffusion and a speeded-up cooperative
diffusion.
When strong interaction leads to marked changes in structure such as
colloidal crystallization, it must also lead to storage of mechanical energy.
Thus colloidal crystals must have elastic moduli and associated viscosity.
Strong crystalline order occurs when the interaction energy per particle is a
few hundred T . Micron-scale colloidal particles thus have an energy density
measured in T per μm3 . The corresponding energy density in a semidilute
polymer solution is T per blob volume, where the blob size is a few nano-
meters. Evidently the colloidal modulus is weaker than the typical polymer
modulus. One generally cannot sense the mechanical moduli without spe-
cial apparatus. Analogous interaction effects occur in emulsions and foams,
as discussed in Chapter 7. There the underlying forces come from interfacial
tension, and the resulting forces are stronger.

5.4.1 Electrophoresis
Charged colloidal particles move in a distinctive way when an electric field
acts on them. The contrast with conventional forcing is greatest when there
are many free ions and the screening length is short. In Chapter 4, we
analyzed the response to external forces by tracing the added momentum as
it flowed outward from the forced particle. The long-ranged flow field of the
surrounding fluid leads to hydrodynamic drag. However, when a charged
particle is forced by an electric field, the situation is radically different.
In any unit of time, the momentum added to the particle is balanced by
opposite momentum added to the counterions in the oppositely charged
screening layer. For a uniformly charged body that is smooth on the scale of
a screening length, the two momenta cancel nearly completely, and virtually
no momentum is transferred to the fluid. Thus no long-range flow occurs.
The particle crawls through the liquid rather than swimming through it. All
the dissipation leading to the retarding force occurs in the narrow screening
zone. Thus the velocity of each bit of surface is determined independently of
the rest. The electrophoretic mobility depends on the surface charge density
and the screening length, but not on the size of the particle.
When the charge on a particle is nonuniform, the phenomenon of electro-
phoresis becomes startlingly rich and complex, as recent discoveries have
shown [25]. If two regions on the particle have different charge densities,
the electrophoretic force wants to pull them at different speeds. Since the
two regions are obliged to move at the same speed, one region must exert a
force on the other. This means that each region experiences a net unbalanced
force, which it must give to the surrounding fluid, producing hydrodynamic
backflow. An opposing momentum is injected at the other region. A net
momentum current comparable to the internal force is injected into the
fluid, and long-range Stokes flow occurs. One may control the nature of
the induced motion in striking ways by controlling the shape of the particle
 Appendix A: Perturbation attraction in a square-gradient medium 137

and the placement of charge on it. One may induce motion at right angles
to the electric field or rotation with no translation, for example [25].

5.4.2 Soret effect

Other perturbing influences create motion analogous to simple electro-
phoresis. A particle inserted into a fluid acquires interfacial energy because
of the interaction of the surface with adjacent fluid molecules, as we will
learn in Chapter 6 to come. If the fluid is inhomogeneous, the interfacial
energy can be different on different parts of the particle. For example, a
temperature gradient induces a gradient in the interfacial energy. Typic-
ally it is reduced at higher temperature. Thus by translating to a region of
higher temperature, the particle can lower its interfacial energy. The gradi-
ent of this energy is a force. It causes motion of the particle. This motion in
response to a gradient of interfacial energy is called the Soret effect [26]. As
in the case of simple electrophoresis, this motion does not add momentum
to the fluid. Thus it does not produce hydrodynamic drag. One may imagine
inhomogeneous surfaces that produce unbalanced Soret forces analogous
to the unbalanced electrophoretic forces discussed above. It would seem
that analogous drag phenomena would occur.

Appendix A: Perturbation attraction in

a square-gradient medium
As we have suggested above, two like objects that perturb a fluid tend to
attract each other. In this appendix we show that this attraction is inevitable
under a wide range of conditions [27]. We consider two identical, parallel
surfaces of area L2 in a fluid at some small separation 2h. We suppose
that each surface perturbs the fluid by altering some scalar property ψ of
the fluid immediately adjacent to the surface. This ψ may be e.g., the
local concentration of polymer or colloid, or the chemical composition, as
in the examples in the text. Naturally, if ψ is perturbed at the surfaces,
it is perturbed nearby as well. Thus there is a perturbed profile ψ(z) at all
distances z from the surfaces. To describe this profile quantitatively, we must
consider the free-energy cost of perturbing ψ in the liquid. This cost depends
on the whole profile ψ(z): it is a functional of the function ψ. We denote it
as W [ψ]. We may also treat the effect of the each surface as a contribution
to the free energy: V (ψs ). Here ψs is the value of ψ at the surfaces.
The ψ(z) profile is that which minimizes the total free energy functional
F /L2 ≡ W [ψ(z)] + 2V (ψs ). The first term is the work required to alter
ψ from its equilibrium value ψ0 . When ψ(z) = ψ0 , we shall set W = 0.
Any change of ψ from ψ0 requires positive work. Naturally the free energy
W [ψ] + 2V (ψs ) depends on the separation of the surfaces 2h. The work
required to separate the surfaces is precisely the change in this free energy.
For many fluids, the free energy W for a nonuniform profile has a simple
form, provided ψ(z) varies gradually enough. Then only the difference
between ψ(z) and ψ in adjacent regions matters. That is, the cost for each
region depends only on the gradient of ψ there. Except for this gradient
effect, the work to perturb ψ in a region is the same as though ψ were
138 Colloids

constant in space. In other words, W [ψ] has the form

  
h 1 dψ 2
W [ψ] = 2 dz w(ψ(z)) + m(ψ(z)) . (5.29)
0 2 dz

Here the w(ψ) is the free energy per unit volume required to make a
uniform change from ψ0 to ψ . We have restricted the integration to the
left half of the system, noting that the contribution from the right half must
be identical. We shall suppose that the variation of ψ is gentle, so that we
† There is no piece linear in this gradient. may neglect higher powers of the gradient dψ/dz† . Its coefficient m may
Such a piece would not be invariant under depend in general on the local value of ψ †† .
a change of coordinate system that replaces
Two identical surfaces immersed in such a square-gradient medium must
z by −z; it is thus ruled out by the require-
ment that the free energy density must be attract each other at long distances. Further, the attractive force per unit
independent of coordinate system. area is simply the local energy density at the midplane wm . To see this, we
††
separate the surfaces by a small amount 2h and examine the change in free
As in the polymer solution, any fluid
with a free energy of this form has spatial
energy F . We perform the separation in two stages. First we separate the
correlations in the fluctuating equilibrium surfaces without allowing a change in the ψ profile. We simply extend its
ψ(z). These correlations die off exponen- midpoint ψm over the extra interval h on each side of the system. Second
tially in space with a decay length ξ given we allow the ψ field to relax to minimize F .
by ξ 2 = d 2 w/dψ 2 |0 /m(ψ0 ). In the first stage F changes by an amount 2hL2 wm . Since ψ is constant
here, the gradient energy is zero. In the second stage, ψ changes in the pre-
existing region and in the newly created gap region. The resulting change
of F in the gap region is second order; it is the product of the small change
ψm and the small interval 2h. The profile in the old region from 0 to h
was that which minimized the free energy F with the constraint that dψ/dz
vanish at h. This derivative vanished because h was at the midpoint of the
full system. To examine how this profile changes, we first determine the
optimal profile between 0 and h without constraints. We allow dψ/dz(h) to
vary freely. But the optimal value of dψ/dz(h) is zero in a square-gradient
medium. We may check this by dividing the interval up into many equal,
discrete intervals ending at z1 , z2 , . . . , zk = h. The corresponding ψ values
are ψ1 , . . . , ψk . The free energy F is the sum of contributions from each
interval. To minimize F in the i’th interval we need only adjust ψi and
ψi−1 . The only interval that involves ψk is the last one. The free energy
contribution to it is
 −2
h h h
w(ψk−1 ) + m(ψk−1 )(ψk − ψk−1 )2 .
k k k

Since ψk appears only in the gradient term, the optimal value for it is
that which makes the gradient vanish. Minimizing F automatically makes
the gradient vanish at h. The vanishing of the gradient is not an additional
††† The author is grateful to Alexei constraint††† . When we allow ψ(z) to relax to the new, wider interval, there
Tkachenko for pointing this redundancy out is no proportionate change in F . The original ψ(z) already minimized F
to him.
in the old region 0–h with respect to arbitrary variations. This means any
small change ψ can make at most a second-order change in F in that
region. Thus the only change in F is the contribution from the gap region,
 Appendix B: Colloidal aggregates 139

proportional to wh . The increase in F implies an attractive force per unit area

d(F /L2 )
= wh .
d(2h)
Though we have assumed our surfaces to be identical and flat, this force
law holds more generally. If two surfaces are sufficiently similar, there must
be a point between them where the gradient of ψ vanishes. Then the above
argument can be extended to show that the force per unit area at that point
is attractive and equal to w.
While this perturbation attraction occurs quite broadly, there are import-
ant cases where the mechanism does not apply. One is the case of surfaces
with grafted polymers treated in the chapter. Here polymers are chosen
which are attracted to the surfaces only at one end. The attraction results
in a polymer-enriched layer near each surface. When the two surfaces are
brought together so that the layers interpenetrate, a repulsion occurs, not
the attraction implied by the Perturbation-Attraction Theorem. Why is the
theorem not applicable? The reason is that the free energy does not have the
square-gradient form, as in Eq. (5.29). Instead, the energy cost of a nonuni-
form concentration is nonlocal. The cost at position z does not depend
merely on ψ at z and on ψ at neighboring z. It depends on ψ throughout
the layer. Thus e.g., if some chains were removed from the layer, the free
energy at the midpoint h would be affected—even if some external agent
forced ψ to be fixed in the intervening layer. Thus the free energy cost of
nonuniformity cannot be found by adding up gradient contributions from
the whole layer. Nonlocal differences such as ψs − ψ(h) also contribute
directly to the free energy.
Another case where perturbation does not lead to attraction is the
deformation of a membrane with bending rigidity, as treated in Problem 5.4.

Appendix B: Colloidal aggregates

We have discussed several mechanisms of colloidal attraction and sev-
eral ways of countering such attractions with compensating repulsions.
Repulsions of sufficient strength can induce the colloidal particles to form
ordered states: the particles form a periodic lattice in the solution. Attrac-
tions of sufficient strength can also induce self-organization. The simplest
effect of attraction is the precipitation or creaming mentioned above. Rather
than dispersing throughout the solution, precipitating colloidal particles
prefer to assemble into compact masses, in which their mutual attrac-
tion energy may be maximized. As noted in Chapter 2, such precipitation
happens when the probability of two particles being near to each other is
substantially larger than the probability that they are far apart. As we saw,
this amounts to saying that the interaction free energy of two particles is
attractive by an amount of order T or greater. If the mutual attraction is
strong enough to bring two particles together, it is ordinarily more than
strong enough to bring pairs or larger groups together. Thus the process
of coming together feeds on itself; its ultimate result is that the particles
have assembled into a compact, macroscopic mass or “dense phase.” The
140 Colloids

nature of the dense phase and the amount of attraction or colloid concen-
tration required to produce it are described by the statistical mechanics of
phase transitions [5]. This well-developed branch of theoretical physics
gives powerful, specific, and well-verified predictions about how mutually
attracting particles undergo phase separation. We encountered this phe-
nomenon briefly in Chapter 4, in discussing the “collapse” of a polymer in
a poor solvent. For the most part the theory of phase transitions applies as
well to colloidal particles as it does to small molecules. Thus the nature of
precipitation and creaming is virtually the same as, e.g., the precipitation of
steam or the liberation of carbon dioxide from a freshly opened soft drink.
But in some cases the distinctive attraction of colloidal particles results in
qualitatively new phenomena.
Phase transition theory applies only when the particles have come to
thermodynamic equilibrium, at least locally. As we saw in Chapter 2, equi-
librium is the state attained after a sufficiently long time. For any change
of configuration the system has made, it must also have time to make the
reverse of that change. For example, any given pair of particles that are
together should have had enough time to separate and recombine several
times in order to attain equilibrium. With small molecules, phase separation
is usually slow enough that each local region (containing a few molecules) is
close to thermal equilibrium. But in colloids, one encounters a new regime
(a) of irreversible attraction, far from equilibrium.
(b)
As we have noted, most forms of colloidal attraction grow in strength with
the particle size. For large particles, the attractive energy may far exceed
the thermal energy T . Then the attraction becomes for practical purposes
irreversible, as discussed in Chapter 2. Two particles that stick together
(c) (d) have virtually no chance to unstick in the time the system is observed:
the strength of the attraction prevents even local equilibration. Thus phase
transition theories cannot describe the assembly process; a new approach
is needed.
The assembly process in this irreversible regime is called kinetic aggrega-
tion (Fig. 5.12). Kinetic aggregation of a colloidal solution ultimately
Fig. 5.12 produces large aggregates like the one shown in Fig. 1.5. These are strik-
Nonequilibrium aggregation. (a) Typical ingly different from the compact droplets or grains of normal precipitation:
contacting configuration of three solid the aggregates are open, tenuous assemblies. The nature of these assem-
colloidal particles. The angle between the blies has been much studied in the last two decades and has been treated in
three is arbitrary. (b) Configuration of
comprehensive reviews [29, 30].
lower attraction energy. In thermal
equilibrium this state is much more The colloidal particles of Fig. 1.5 were initially dispersed in a very dilute
probable than (a), but in irreversible state. Four-nanometer charge-stabilized silica particles were suspended in
aggregation it is no more probable. water at a volume fraction of roughly 10−6 [31]. Then a large amount
(c) Typical contacting configuration of of salt was added, screening out the repulsive Coulomb barrier between
three triangular clusters of liquid droplets.
the particles nearly completely, and exposing the particles’ strong van der
(d) Compact configuration. Like
configuration (b), this state is unreachable Waals attraction. A few minutes after the salt was added, the aggregate had
in the available time. Liquid drops can go formed. When two particles encountered each other in the course of their
from state (a) to state (b) with no energy ordinary Brownian motion, they stuck together permanently. The resulting
barrier, but not from state (c) to state (d) pair continued its Brownian motion until it encountered another particle or
[28]. To pass from (c) to (d) the shaded
cluster. These two clusters stuck together permanently in their contacting
ball would have to break the contact
indicated by the arrow and overcome the configuration and continued to move. In this way the clusters grew to the
associated energy barrier. size shown. One may readily simulate this process on a computer [29], to
 Appendix B: Colloidal aggregates 141
1
Gold
Fig. 5.13
Silica Structure factors S(q) from colloidal
Polystyrene aggregates like that pictured in Chapter 1,
0.1
obtained from light scattering, after [31].
The upper three data sets marked DLCA
DLCA were from aggregates grown under
diffusion-limited conditions using 7.5 nm
I (q)

0.01
gold particles, 3.5 nm silica particles, and
RLCA 20 nm polystyrene spheres. Straight lines
have the slopes expected for a fractal
0.001 structure with D 1.84. The lower three
data sets marked RLCA were from
aggregates grown under reaction-limited
conditions made from the same three
0.0001 dispersed colloids. Straight lines have
0.002 0.005 0.01 0.02 0.05 slopes expected for a fractal structure with
q (nm–1) D 2.1.

produce an ensemble of representative clusters. These simulated clusters

have the same treelike structure and the same fractal dimension seen in
experiments [31] (Fig. 5.13).

Simplest aggregation model

The “Sutherland’s Ghost” model of Ball [32] gives a simple way to under-
stand why the structure is open. In this model we imagine that many single
particles have combined to form dimers. Then all the dimers combine to
form tetramers, then all these combine to form octomers, and so forth. When
two n-clusters combine, they stick together as randomly as possible. We Fig. 5.14
choose a particle on the first cluster and a particle on the second cluster at One aggregation step in the Sutherland’s
random. Then we choose a direction at random. Finally we combine the Ghost model. Left: a (shaded) particle from
two clusters by connecting the two chosen particles in the chosen direction, two arbitrary dimers is selected at random,
and the bond direction indicated by the line
as shown in Fig. 5.14. (There is no need to rotate the clusters, since this
segment is chosen at random. Right: the
assembly process automatically produces all rotations of a given cluster.) selected particles are joined by the bond.
In general, the combined cluster is not self-avoiding: its particles intersect.
We freely allow such intersections: hence the name “ghost.”
The average size of a large Sutherland’s Ghost aggregate can be readily
estimated. To define an average size, we select two particles in a large 2n-
cluster at random. These two are not in general connected directly: one must
traverse some number b of connecting bonds to reach the second particle
from the first. The two unshaded particles in Fig. 5.14 are separated by
three bonds. (We do not count accidental intersections as connections.) The
average of the bond distances b in an n-cluster will be called Bn .
Luckily, the average bond distance B2n for a 2n-cluster can be expressed
simply in terms of the Bn of the n-clusters. Any 2n-cluster is made from
two subclusters of size n. If two points on the 2n-cluster are chosen at
random, there is one chance in four that the two belong to the first subcluster.
Since the points on this subcluster are arbitrary, the average bond distance
in this case is simply Bn . It is the same if both chosen points had been on
the second subcluster. But there is one chance in two that the two chosen
142 Colloids

points belong to different subclusters. In that case the bond distance is a

bit harder to find. Each subcluster has one particle that is joined directly to
the other subcluster. To reach the second chosen particle from the first, we
must first travel on the first subcluster to its joining particle, then bridge to
the joining particle of the second subcluster, and finally travel to the second
chosen particle. The bond distance is the sum of these three distances.
Now, the average distance in the first subcluster is simply Bn , since the
first particle and the joining particle were chosen at random. Likewise, the
average distance in the second subcluster is also Bn . The distance between
the joining particles is one bond. Thus the average bond distance between
particles on different subclusters is just 2Bn + 1.
We may now find the overall average bond distance B2n by combining
this distance with the single subcluster distance, each with its appropriate
probability:
B2n = 14 Bn + 14 Bn + 12 (2Bn + 1), (5.30)
or
B2n = 32 Bn + 12 .
Since we know B2 = 1, we can readily find any Bn by direct calculation.
!log(n)/ log(2)
If n is large, we may ignore the 12 ; then Bn 23 32 , or
 
3
log / log(2)
Bn ∼ n 2
.

The exponent is roughly 0.58.

From this Bn we may readily find the average geometric distance between
two arbitrary points in terms of the particle diameter a. Clearly this geo-
metric distance depends not only on the number of bonds traversed but also
on their directions. In our model, these directions are completely random.
Each bond along the connecting path was chosen randomly and independ-
ently of all the others. Thus the connecting path is a simple random walk,
and the mean-squared distance r 2 between two points separated by b bonds
† We have only shown that the overall mass is simply a 2 b. The mean-square distance Rn2 between two arbitrary particles
n scales with overall size according to a is evidently a 2 Bn . Thus Rn ∼ nlog(3/2)/(2 log(2)) , or
fractal law. The same fractal dimension D
is believed to describe the average mass
n ∼ RD , where D = 2 log(2)/ log (3/2) 3.4.
n(R) within some local region of radius R
around an aggregate particle. This is clearly
true if one takes the sphere to enclose one
The mass scales with size like a fractal with dimension D.
of the subclusters making up the aggregate. This structure is very dense in three dimensional space, but in four or
(Incidental incursions into this sphere from more dimensions, D is smaller than the spatial dimension and the Suther-
other subclusters do not contribute qual- land’s Ghost structures are tenuous fractals† . Thus the model shows how the
itatively to the enclosed mass if d > D.) simplest features of kinetic aggregation can lead to an open fractal structure.
Since we do not expect the mass for a
given size R to vary widely with the pos-
ition of the sphere on the aggregate, we Effects of polydispersity and self-avoidance
expect that n(R) for an arbitrary sphere is of
the same order as that for spheres covering While the Ghost model gives a qualitative account of the open structure of
subclusters. Such arguments, together with kinetic aggregates, it contains two gross simplifications of potentially great
computer experiments, lead to the accepted
belief that these aggregates are fractals at all
importance. The first of these is the assumption that all clusters are made
scales between that of the particles and that from two equal-size subclusters. In reality, when a dimer encounters another
of the aggregate [29]. colloidal particle, this particle may be a free particle, another dimer, or a
 Appendix B: Colloidal aggregates 143

larger cluster. The two joined clusters are typically unequal, and they may be
greatly unequal. The second drastic simplification is that self-intersections
are ignored. We may see the effect of both of these simplifications by simple
modifications of the model.
It is easy to modify the Ghost model to account in a primitive way for
unequal clusters. This modification is called the “fixed-ratio Sutherland’s
Ghost” model [32]. This model is like the original model, except that the
reacting clusters are forced to have a particular mass ratio, such as 3 : 1.
One may readily modify the above reasoning to determine the average
bond distance, B4n , in terms of those of its constituents, Bn and B3n . Now
when two particles are chosen at random to compute the combined bond
distance, B4n , the probability that both particles are on the smaller cluster is
now reduced (to 1/16); the other probabilities are also altered. For a general
ratio r : 1, the probability that an arbitrarily picked point belongs to the
small cluster is 1/(r + 1). The probability for the larger cluster is r/(r + 1).
Eq. (5.30) becomes
 2  2
1 r r
B(r+1)n = Bn + Brn + 2 (Bn + Brn + 1).
r +1 r +1 (r + 1)2 † The probability P (n, t) that a given cluster
has mass n at time t is called the mass
This equation also allows power-law solutions of the form Bn ∼ nx . distribution. To account for the change of
Substituting into the above equation, one finds an implicit equation for x. this distribution as the aggregation pro-
ceeds, one must consider all processes
 2  2 which may increase or decrease the num-
1 r r
(r + 1) =
x
+ rx + 2 (1 + r x ). (5.31) ber of n-clusters. The joining of an i cluster
r +1 r +1 (r + 1)2 with an (n − i)-cluster increases P (n, t).
The joining of an n-cluster with any other
The exponent x decreases with increasing ratio r. For small r, Eq. (5.31) cluster decreases P (n, t). The rate of join-
reduces to ing of i-clusters with j -clusters is evidently
proportional to their numbers, P (i, t) and
x = 2(1 + r x ).
to P (j , t). But two different mass pairs
As in the simple Ghost model, the geometric path between any two sites on a with the same cluster numbers P need not
cluster is still a random walk, so that n ∼ R D with D = 2/x. Evidently the have the same joining rate. A given pair
of clusters with masses i and j has a spe-
nominal fractal dimension D goes to infinity as the ratio r → 0. It is possible cific joining rate, denoted K(i, j ). Adding
to extend this treatment to include a complete distribution of aggregating together all the processes that change the
sizes, not merely a fixed ratio [33]. For the expected mass distributions† , D number of n-clusters, one obtains the so-
increases moderately from the 3.4 of the simple Sutherland’s Ghost model. called Smoluchowski equation [34]:
Clearly, to account for D of kinetic aggregates quantitatively requires a ∂P (n, t) 1 
realistic treatment of the relative masses of the aggregating clusters. = K(i, j )P (i, t)P (j , t)
∂t 2
i,j
The second glaring defect of the model is its neglect of self avoidance. We
may evaluate the importance of this defect in the same way we have done ×[δi+j ,n −δi,n −δj ,n ].
for polymers in Chapter 3. We modify the process to assure self-avoidance.
The Smoluchowski equation gives several
Pairs of clusters are joined as in the original model. Then the combined different types of P distributions, depend-
cluster is checked for self-intersections. If any are found, the combined ing on the specific joining rate K. The
cluster is discarded. If this discarding probability approaches unity for large equation itself is an approximation, for
clusters, the average properties of the remaining ones may be qualitatively it accounts for only the masses of the
affected. But if the discarding probability does not approach unity, the clusters and ignores other variables, such
as the cluster shapes and their positions
average properties cannot be qualitatively affected and scaling exponents in space. Nevertheless, it is believed to
such as D must be unchanged. be well-justified for three-dimensional col-
By this reasoning, we may see that self-avoidance has no impact on D loidal solutions. These subjects are treated
in sufficiently high spatial dimensions d. We saw in Chapters 3 and 4, that at length in [34].
144 Colloids

when two fractals of size R are placed at random in the same volume, the
average number of intersections between them within distance R goes as
R D1 +D2 −d . When this exponent is negative, the number of intersections
goes to zero as R → ∞. Thus our two aggregates with dimension D have
no intersections in spatial dimension d > 2D. Even if the two aggregates
are connected at the origin, the number of intersections is limited and does
not grow indefinitely with R. Thus in these high spatial dimensions the
probability of self-avoidance is finite and does not go to zero as the cluster
sizes go to infinity. Then the aggregation process is not qualitatively affected
by the self-avoidance constraint, and the D of the aggregates is unchanged.
For Sutherland’s Ghost aggregates with D 3.4, self-avoidance has no
affect on D above d = 6.8. If the aggregation process were carried out on
a computer in a virtual space of seven or more dimensions, the resulting D
should be 3.4 even with self-avoidance imposed. We have encountered the
analogous property for polymers: their self-avoidance does not affect their
fractal dimension in more than four spatial dimensions.
In lower spatial dimensions, self-avoidance becomes relevant. In poly-
mers the discarding process has the greatest impact on the more compact
configurations. The remaining, self-avoiding, ones are generally increased
in size, and their D is reduced. The same qualitative behavior is expected
for aggregates. D should decrease progressively as the spatial dimension d
is reduced.

Diffusion-limited and reaction-limited

The Ghost model has another unrealistic aspect. It gives no account of how
the clusters must move in order to join together. At first sight, this omission
does not appear serious. For polymers any questions of motion are irrelev-
ant to their fractal dimension. This is true because the configurations of a
polymer form an equilibrium ensemble: the probability of a given configur-
ation is prescribed by the Boltzmann principle. This Boltzmann probability
depends only on the energy of that configuration and not on any motion. But
our aggregates are not an equilibrium system. The configurations and their
probabilities are defined by an irreversible, kinetic process. Accordingly,
we must consider how various types of motion may affect the configura-
tions. We imagine a particular joining configuration of two clusters from
the Ghost model and consider the probability that this configuration occurs.
In a real colloidal suspension, two clusters that encounter each other move
by Brownian motion. Instead of simply being placed next to each other as
in the Ghost model, the two clusters move together by a random walk. The
walk influences the probability that a particular contact will be made. One
Fig. 5.15 limit of interest is when the two clusters stick together on first contact. Then
A candidate joining configuration in only those random walks that happen to reach the joining configuration with
diffusion-limited aggregation. The no intervening contact between the clusters are possible. This constraint is a
previous steps in the random walk of the form of self-avoidance constraint. We can phrase it in geometric terms. For
cluster on the right are shaded. One
each cluster in our ensemble, we construct a typical random-walk history
particle of this cluster is filled to indicate
its history. Since the history of this cluster by moving the cluster in a random walk and including all the spatial points
intersects the other cluster many times, this thus traversed, as shown in Fig. 5.15. (For simplicity we consider only
configuration must be discarded. translational motion in this random walk; rotational motion also occurs but
 Appendix B: Colloidal aggregates 145

does not affect the conclusions.) The resulting object is a fractal whose
dimension is two greater than that of the cluster itself. We now choose a
random point on this cluster and on another cluster. We then join these
according to the Ghost model. However, for a valid joining configuration,
the history must not intersect the other cluster. If it does, it must be discarded.
The remaining configurations are valid joining configurations, in which the
one cluster has avoided the history of the other† . † In making this construction we have
This constraint is again that of the mutual avoidance of two fractals. treated one cluster as stationary and the
other one as moving in a random walk. This
The Brownian-motion history of a fractal of dimension D is a fractal of
amounts to working in the frame of refer-
dimension D + 2. The history of each particle of the fractal lying in a ence of one cluster. This is equivalent to
sphere of radius R is a random walk, with roughly R 2 particles in the sphere. the actual situation in which both clusters
The number of particles leaving these random-walk trails is the number of execute random walks.
particles of the fractal—roughly R D . The total number of particles in the
history within the sphere is thus roughly R 2 R D = R 2+D . As before, it must
be irrelevant in sufficiently high spatial dimensions d. If the clusters have
dimension D, then the avoidance has no effect on D provided (D + 2) +
D − d < 0. For our Ghost aggregates, this condition holds for d  8.8.
Evidently this random-walk avoidance has a stronger effect than the simple
avoidance discussed previously. D begins to decrease from its asymptotic
value for higher d (8.8 versus 6.8), and at a given d we would expect it
to decrease further. When clusters move by Brownian motion and stick
on contact, the aggregation is called diffusion-limited. This form is very
prevalent in colloidal solutions. The aggregate of Fig. 1.5 was produced in
diffusion-limited conditions.
Sometimes diffusion is not the rate-limiting aspect of the aggregation
process. It can happen that particles stick irreversibly with very low prob-
ability even when they are adjacent. Indeed this is the case for stabilized
colloidal particles. If the stability is diminished somewhat, e.g. by adding a
little salt, the sticking occurs fast enough to observe. Still the sticking rate
may be much slower than the rate of encounters via Brownian motion. This
regime is called reaction-limited aggregation. In the reaction-limited regime,
two clusters which join at a particular moment have had ample opportunity
to visit each other’s neighborhood. Thus all configurations that do not inter-
sect are equally likely, as in equilibrium. (Here we ignore the local repulsion
that inhibits very close approaches; we implicitly count such approaches
as intersections.) Now the joining proceeds as in the self-avoiding Suther-
land model discussed above: the joined configurations are taken at random
from all contacting configurations. As we have seen, this model is less
constrained than the history-avoiding diffusion-controlled case. Thus we
expect the resulting aggregates to have a fractal dimension D closer to the
large D = 3.4 of the unconstrained model.
The diffusion-limited and reaction-limited regimes are the most import-
ant types of aggregation seen in colloidal suspensions. But other types of
motion are also of interest. In ballistic aggregation, the joining clusters
follow straight-line paths instead of random walks. One form of ballistic
aggregation is sedimentation aggregation, in which the joining clusters are
drifting downwards in a gravitational field. They join when a heavier cluster
overtakes a lighter one [35]. Large, composite snowflakes are made by this
form of aggregation.
146 Colloids
4

D=d D = d/2
Not Transparent
tenuous to self
D

×o + + + + +

o? +◊
◊?
+

1 +
o Reaction-limited aggregates, after [31]
◊ Diffusion-limited aggregates, after [31]
+ Good-solvent polymer
Fig. 5.16
× Randomly branched polymers
Fractal dimension D of colloidal
aggregates and other fractals versus spatial 2 4 6 8
dimension d. d

We have seen that a number of aspects of colloidal aggregation have a

profound effect on the structure and thence on the fractal dimension D.
To account for all these aspects—mass distribution, reaction rates, self-
avoidance—and thus obtain a realistic estimate of D is a complicated job.
This accounting may not yield great insight. But the qualitative nature of
these effects can be simply understood, as we have seen. In Fig. 5.16
we summarize the known information about D, based chiefly on simu-
lations. The agreement between these simulations and experiment gives
confidence that the essential features responsible for the fractal structure are
recognized.

Properties of aggregates
The fractal structure of colloidal aggregates makes them broadly useful, as
noted in Chapter 1. The aggregates are used like polymers to thicken liquids
and solidify them by producing a gel network. They are used to reinforce
rubber and make it tougher. The origin of these special mechanical prop-
erties can be readily understood, adapting the reasoning used in Chapter 3
and 4.
A small volume fraction of colloidal aggregates in a liquid increases its
viscosity markedly. The reason is the same as in a polymer solution. The
aggregates, like polymers, are opaque to flow since they have D > 1: a
shear flow goes around rather than through them. Thus they impede flow as
though their pervaded volume were essentially filled in with solid material
instead of being nearly pure solvent.
Aggregates like polymers produce osmotic pressure in dilute solution.
Since these fractals are opaque to each other, their mutual excluded volume
 Appendix B: Colloidal aggregates 147

is proportional to their pervaded volume, as with polymers. The osmotic

compressibility produces detectable effects on the scattering from aggregate
solutions. But this pressure is weak: since aggregates are generally larger
than polymers, their concentration in dilute solution is small.
At higher concentrations the distinctive features of the aggregates begin
to appear. Aggregates are rigid structures that hold their shape. Thus as
solvent is removed the aggregates cannot readily interpenetrate as polymers
can. Instead, they bend or break under the pressures from their neighbors.
The macroscopic osmotic pressure opposing further removal of solvent is no
longer due to thermal fluctuations. It is due to the microscopic elasticity of the
structure. When the structure is compressed, the bonded-together colloidal
particles are deformed. The deformation produces stress, as it would in a
macroscopic chain of glued-together beads. The resulting pressure can be
expressed in terms of an effective elastic modulus for the aggregates.
The elasticity of aggregates has been measured in several ways [36, 37],
though its scaling has not been well established. Nevertheless, the measured
elasticity appears to be consistent with a simple theoretical argument first
published by Brown and Ball [36]. One way to measure this elasticity is to
trap an aggregate between two parallel plates and then compress it slightly.
This causes the few longest arms of the aggregate to bend. Only a small
part of the structure stores energy in this process. By construction, a per-
fectly rigid, irreversible aggregate has no loops. Simulations allowing some
flexibility and some rearrangement also have no loops except at the scale
of a few particles. Thus an aggregate is a treelike structure with spanning
arms and many side branches. Under deformation only the spanning arms
bend and store energy. Since these grow more slowly than the aggregate as
a whole, they can be made to constitute an arbitrarily small fraction simply
by taking a larger aggregate.
A spanning arm displaced by a distance u exerts a restoring force propor-
tional to u. The constant of proportionality K(R) depends on the shape
of the arm, just as it does for a macroscopic contorted wire. Simula-
tions suggest that spanning arms of colloidal aggregates are themselves
fractals, with a fractal dimension C somewhat lower than the overall fractal
dimension. Thus the number of particles Ma in an arm of length R obeys
Ma (R) (R/a)C . The C of diffusion-limited aggregates is roughly 1.26
[38]. The C value of reaction-limited aggregates is consistent with unity.
For both types of aggregate, the horizontal and vertical dimensions of the
overall structure remain comparable. Given these facts, we may estimate
the force constant K(R) [39].
When the end of the arm is displaced, the dominant distortion is bending

of the arm. The direction between successive beads is altered slightly, as

shown in Fig. 5.17. The energy Ei stored at a given bead i is proportional to
the square of the bending angle θi : Ei = kθi2 . When the arm is bent, the θi ’s
adjust
 so as to produce the required displacement u with the least energy Fig. 5.17

i E i . Each θi , acting on its own, would produce a displacement ui = θi ri . Local bending of two particles in a
spanning arm caused by displacement of
Here ri is the distance from the bead i to the displaced end. Clearly, beads
one end. The undistorted configuration is
further from the end produce the displacement more efficiently. The bulk of shown in light shading. The two lines,
the bending will involve beads at distances of order R from the end. All such drawn transverse to each bead, were
beads will have a comparable share of the bending and comparable angles parallel in the unbent configuration.
148 Colloids

θi . This θ will be such that u i θ R. The number of beads involved
is roughly the number of beads Ma in the arm. Thus θ u/(Ma R). The
resulting energy E is given by

E= Ei kMa θ 2 ku2 /(Ma R 2 ).
i

This expression defines the desired spring constant K(R), the coefficient of
u2 . Using the fractal law for Ma , we find

K(R) kR −C−2 .
From this basic elastic constant K(R), we may infer the bulk elasticity of
a mass of aggregates at volume fraction φ. When an aggregate suspension is
compressed to produce this elasticity, the individual aggregates have begun
to press against each other. The aggregates are at their overlap concentration:
φ = φi (R/a)D−3 . The longest spanning arms of a typical aggregate
are deformed by these external contacts. If a small strain γ is now applied,
then these longest arms are displaced by an amount u γ R. Each stores
an energy
E = Ku2 kR −C−2 γ 2 R 2 .
Each aggregate has only a few of these spanning arms: their number does
not increase with R. Thus the energy stored per aggregate volume is of order
E/R 3 . Combining, we find that the overall strain γ produces an energy per
unit volume of order kR −C−3 γ 2 . The coefficient of γ 2 defines an elastic
modulus G for the gel. This G, expressed in terms of volume fraction, obeys

G kφ (C+3)/(3−D) a −3 . (5.32)
Figure 5.18 shows a comparison of the predicted power and
experimental data.

0.1
Modulus (GPa)

0.01
Fig. 5.18
Modulus versus concentration in two
different aggregate materials, after [40]
(silica aerogel, filled dots) and [41], 0.001
(fumed silica, open rectangles). These
experiments approach the percolation
threshold at low concentrations; this is
0.0001
expected to depress the moduli below the
predicted power law. The straight lines,
with slopes 3.2 and 3.55, indicate the
power law predicted by Eq. (5.32), 0.00001
including the experimental uncertainties in 0.01 0.1 1
D and C [38, 42]. Volume fraction
 References 149

Aggregates, like polymer solutions, have elastic properties that scale in

predictable ways in terms of the geometric scaling properties represented
by D and C. Like polymer solutions, the aggregate gels have moduli for
bulk, shear, or other types of strain that scale with the same power of φ.
Our reasoning leading to this scaling did not depend on the particular type
of distortion imposed.

References
1. P. Pincus, in Lectures on Thermodynamics and Statistical Mechanics: XVII
Winter Meeting on Statistical Physics, ed. A. E. Gonzalez and C. Varea
(Singapore: World Publishing Co., 1988).
2. D. H. Everett, Basic Principles of Colloid Science (London: Royal Society of
Chemistry, 1988).
3. J. N. Israelachvili, Intermolecular and surface forces, 2nd ed. (London: San
Diego, CA: Academic Press, 1991).
4. J. Mohanty and B. W. Ninham, Dispersion Forces (London: Academic Press,
1976).
5. H. E. Stanley, Introduction to Phase Transitions and Critical Phenomena (New
York: Oxford University Press, 1971).
6. F. Brochard and P. G. de Gennes, Ferroelectrics 30 33 (1980).
7. For an excellent review of polymer brushes see A. Halperin, M. Tirrell and
T. Lodge, Adv. Polym. Sci. 100 (1991).
8. S. T. Milner, Europhys. Letts. 7, 695 (1988).
9. H. J. Taunton, C. Toprakcioglu, L. J. Fetters, and J. Klein, Nature 332 712
(1988).
10. A. N. Semenov, J.-F. Joanny, A. Johner, and J. Bonet-Avalos, Macromolecules
30 1479 (1997).
11. P. G. de Gennes, Scaling Concepts in Polymer Physics, (Ithaca: Cornell
University Press, 1979).
12. S. Bucci, C. Fagotti, V. Degiorgio, and R. Piazza, Langmuir, 7 824, (1991).
13. S. H. Behrens and D. G. Grier, in Electrostatic Effects in Soft Matter and Bio-
physics, ed. C. Holm, P. Kekicheff, and R. Podgornik (Dordrecht: Kluwer,
2001).
14. M. Alexander, L. F. Rojas-Ochoa, M. Leser, and P. Schurtenberger, J. Colloid
Interface Sci. 253 35 (2002).
15. R. Kjellander, T. Akesson, B. Jonsson, and S. Marcelja, J. Chem. Phys. 97 1424
(1992).
16. L. Guldbrand, B. Jonsson, H. Wennerstrom, and P. Linse, J. Chem. Phys. 80
2221 (1984).
17. P. Pieranski, Contemp. Phy. 24 25 (1983).
18. See e.g., M. Constantinos, Paleos ed., Polymerization in Organized Media
(London: Taylor and Francis, 1992).
19. M. J. Stevens, and M. O. Robbins, J. Chem. Phys. 98 2319 (1993).
20. L. Onsager, Ann. N.Y. Acad. Sci. 51 627 (1949).
21. P.G. de Gennes, and J. Prost The Physics of Liquid Crystals, 2nd ed. (Oxford :
Clarendon Press, 1993).
22. R. E. Rosensweig, M. Zahn, and R. Shumovich, J. Magn. Magn. Mater. 39
127(1983); R. E. Rosensweig, Ferrohydrodynamics (New York: Cambridge
Univ. Press, 1985).
23. A. P. Gast and C. F. Zukoski, Adv. Colloid Interface Sci. 30 153 (1989).
24. E. J. Hinch and J. D. Sherwood, J. Fluid Mech. 132 337 (1983).
25. D. Long and A. Ajdari, Phys. Rev. Lett. 81 1529 (1998).
26. J. L. Anderson, Ann. Rev. Fluid. Mech. 21 61 (1989).
27. This treatment extends P. G. de Gennes, Macromolecules 15 492 (1982); 14
1637 (1981).
28. P. Poulin, J. Bibette, and D. A. Weitz, Eur. Phys. J. B 7 277 (1999).
150 Colloids

29. P. Meakin, in Phase Transitions and Critical Phenomena, Vol. 12 C, eds. Domb,
J. L. Lebowitz (New York: Academic, 1988) p. 335.
30. See e.g. R. Jullien and R. Botet, Aggregation and Fractal Aggregates
(Singapore: World Scientific, 1987).
31. M. Y. Lin, H. M. Lindsay, D. A. Weitz, R. C. Ball, R. Klein, and P. Meakin,
Nature 339 360 (1989).
32. R. C. Ball and T. A. Witten, J. Stat. Phys. 36 873 (1984).
33. R. Botet, J. Phys. A: Math. Gen. 18 847 (1985).
34. P. G. J. VanDongen and M. H. Ernst, J. Phys. A: Math. Gen. 18 2779 (1985).
35. P. Meakin, Rev. Geophys. 29 317 (1991).
36. R. Buscall, P. D. A. Mills, J. W. Goodwin, and D. W. Lawson, J. Chem. Soc.,
Faraday Trans. I 84 4249 (1988).
37. T. A. Witten, M. Rubinstein, and R. H. Colby, J. Phys. (Paris) II 3 367 (1993).
38. P. Meakin, I. Majid, S. Havlin, and H. E. Stanley, J. Phys. A 17 L975 (1984).
39. Y. Kantor and I. Webman, Phys. Rev. Lett. 52 1891 (1984).
40. T. Woignier, J. Phalippou, Revue de Phys. Appliquee 24 C4-179 (1989).
41. J. Forsman, J. P. Harrison, and A. Rutenberg, Can. J. Phys. 65 767 (1987).
42. P. Dimon, S. K. Sinha, D. A. Weitz, C. R. Safinya, G. S. Smith, W. A. Varady,
and H. M. Lindsay, Phys. Rev. Lett. 57 595 (1986).
 6
Interfaces

Most of our experience with liquids arises not from the bulk of the liquid but
from its interfaces with other media. As we gaze onto the ocean or into our
cup of coffee, it is the surface of the liquid that catches our eye. The property
that distinguishes a liquid from a gas is the existence of this surface. This
chapter deals with such interfaces—how they are structured, how much
energy they have, and how fast they move and relax. Most importantly, it
deals with how solute molecules, both simple and complex, affect these
properties.
A great deal of the importance of structured molecules in liquids arises
from their affect on interfaces. Molecules that migrate to an interface and
affect its properties are called surfactants. Often such molecules are added
to fluids like motor oil, cosmetics, cleaning products, and food, in order to
achieve desired interfacial properties. In rubber and plastic composites like
automobile tires and appliance cases, the strength of the material hinges
on the properties of the interface between the components making up the
composite. Discoveries involving the interface between water and fabric or
solids and internal body tissues have had recent technological impact. But
the most dramatic property of surfactants is their ability to cause spontan-
eous production of interfaces. Such surfactant-generated interfaces are the
subject of the next chapter.
We begin this chapter by describing what the quantitative properties mean
in experimental terms. We survey ways of measuring energy, structure, and
motion associated with interfaces. Next we examine how interfacial energy
comes about for simple liquids. We discuss wetting of an interface by a
liquid, including the motion of the liquid in the process of wetting. Next we
turn to the effect of added solute molecules on an interface. Such molecules
can alter the energy of an interface and create structure there. Finally we turn
to structured fluids like polymers. These create structures at the interface
with distinctive scaling properties extending over distances as large as the
polymers.

6.1 Probes of an interface

In this chapter we shall lay the groundwork for the next one by describing
the general properties of liquid interfaces. Before we begin, let us survey
the ways by which interfacial properties are measured. The most basic
property of an interface is its interfacial energy—the work per unit area
required to create the interface from a bulk material of the same composition.
152 Interfaces
(a) (b) (c)

Microscope observes drop

Plate
Air

Injects light liquid

Tube of
Light source Drop of
heavy liquid
light liquid
Liquid

Fig. 6.1
Devices for measuring interfacial properties. (a) contact angle micrometer after [1]. Schematic light bulb in foreground illuminates the black
droplet on the white sample plate from the side. Lens system behind the sample plate projects the shadow of the drop. Contact angle is measured
from this shadow. (b) Spinning drop tensiometer, after [2], used for measuring small interfacial tensions. Horizontal tube at center rotates about
horizontal axis. Centrifugal forces elongate the black drop of lower-density fluid. The microscope at the top views the distorted drop shown and
measures its dimensions. (c) Wilhemeny plate, pulling upwards on a liquid surface. The support F is attached to an analytical balance to measure
the force on the plate. A motor, not shown, raises or lowers the fluid so that the fluid film meets the plate vertically.

The simplest way to learn something about an interfacial energy is to meas-

ure the contact angle. Figure 6.1(a) shows an optical device for measuring
contact angles. We will explain the connection between this angle and inter-
facial energies below. The spinning drop tensiometer of Fig. 6.1(b) measures
Fig. 6.2
this energy, as does the Wilhemeny plate of Fig. 6.1(c). These devices pro-
Sketch of a Langmuir trough. Insoluble duce controlled increases in the interfacial area and measure the associated
surfactant molecules are trapped on the work. A related device of great importance is the Langmuir trough, Fig. 6.2.
left side of the barrier. They exert excess It is used to study surfaces with a fixed amount of surfactant trapped on
pressure against the slider bar. Arrows
them. The slider bar compresses the surfactants; a Wilhemeny plate mon-
show the force needed to counter this
pressure. A motor, not shown, compresses itors the associated decrease in interfacial energy. Langmuir troughs are the
the slider bar, while a Wilhemeny plate, standard apparatus to prepare a controlled air–water interface containing
not shown, measures the surface tension surfactant. It forms the basis of many of the microscopic probes described
inside. below.
A more fine-scale view of interfacial forces can be gained using the
surface forces apparatus of Fig. 2.5. It measures the tiny forces exerted
by a 10 μm2 surface on another such surface, one or more nanometers
away. Often the forces arise from molecules bound to the surfaces. Thus
the apparatus probes these interfacial molecules. It can sense how deeply
they extend into the gap between the two surfaces, for example.
We have seen in Chapters 2, 3, and 4 that scattering is a powerful means
of measuring microscopic structure. Such probes have been ingeniously
adapted to the study of interfaces. Ingenuity is needed to assure that the
signal comes from the molecules near the interface and not from the far
Fig. 6.3 more numerous molecules in the bulk of the liquid. The standard tech-
Photograph of an ellipsometer from [3]. nique of ellipsometry Fig. 6.3, senses the amount of relative surfactant
Laser at left illuminates a thin-film sample
on a surface by measuring the phase shift of a light beam reflecting from
at the center of the black stage. Detector
at right measures the intensity and it, relative to that from a bare surface. To sense the spatial distribution
polarization of the reflected light. Incident of the adsorbate, ellipsometry is not enough. If the distances involved
and reflected angles are controllable. are many tens of nanometers, evanescent-wave fluorescence can be used.
 Simple fluids 153

The idea is sketched in Fig. 6.4. The incident light is arranged to send an
evanescent wave along the surface in question. The light intensity decays
exponentially with distance from the interface, and the decay length may be
controlled within limits. This light excites the adsorbate molecules, which
have to be labeled=† with a fluorescent dye. By measuring the intensity
of the fluorescence as the evanescent decay length is varied, one may †That is, a fluorescent chemical group must
obtain information about how the fluorescent molecules are arranged in be attached to the molecule to be probed.
space [4].
Often one wants to know about structure at the single nanometer
scale. Here x-ray and neutron scattering are used. To infer the profile
of adsorbate density with distance from the surface, one may meas-
ure reflectivity near conditions of total internal (or external) reflection.
As the incident beam angle is increased, the fraction reflected begins
to drop from 100% as transmission becomes possible. If the sample is
a thin film, the reflectance variation with angle can be used to infer
the composition as a function of depth. The inference is not straight-
forward, though. One must guess the profile and keep improving the
Fig. 6.4
guess until the calculated reflectivity versus angle matches the measure- Selective detection of molecules near a
ments. surface via evanescent wave fluorescence,
Quite often molecules are not spread uniformly over an interface, but after [4]. Beam of light hits the interface
are heterogeneous or patterned in some way. This kind of spatial struc- at a low angle and is totally internally
reflected, entering the liquid only in the
ture can make a big difference in the properties of the interface, yet
form of an evanescent wave near the
this in-plane structure is not apparent from the depth profiles treated interface. Any fluorescent molecules in the
above. To find such information from scattering, one must probe scattered evanescent layer emit light of a lower
wave vectors with components parallel to the surface. One must detect frequency. The intensity of fluorescent
not just the reflected waves, but waves scattered at an azimuthal angle light is measured by a photodetector, not
shown.
to the incident beam Fig. 6.5. Such experiments can detect moving
inhomogeneity as well as static patterns, using the methods of dynamic
scattering sketched in Chapter 4. Naturally, since the scattering occurs
from only a small number of molecules at the surface, the scattering is
weak and hard to detect. Thus these measurements require large-scale
neutron or x-ray facilities that are based on nuclear reactors and particle
accelerators.
Another way to see irregular coverage of an interface is to use microscopy,
as we previously discussed in Chapter 2.

6.2 Simple fluids

Fig. 6.5
6.2.1 Interfacial energy
Schematic view of evanescent-wave
In a liquid the attraction between molecules is generally of order T or less, scattering. The gray rectangle is the
as explained in Chapter 2 in the discussion of viscosity. This suggests that interface of interest, which is often an
air–water interface in a Langmuir trough.
The beam hits the interface at grazing
incidence (from the left), so that practically
all of it is reflected (darker beam). If the
scatterers are nonuniformly spread over
ϕ the interface, some of the incident beam is
scattered, and may be detected at various
azimuthal angles φ from the reflected
beam (lighter beam).
154 Interfaces

Fig. 6.6
A snapshot of a fluid near a hard wall. Left
picture shows a “phantom” wall that does
not alter the particle arrangements. In the
center picture, local configurations in
which particles intersect the wall have
been removed. In the right hand picture a
hard slab has been inserted into the liquid,
distorting the particle configurations.

the interfacial energy per molecule in a liquid is also of order T or less. But
interfacial energy is significant even when there is no attraction between the
atoms. The boundaries of any fluid have a spatial structure and an energetic
cost. If the fluid fills a box, then these fluid boundaries are at the walls of
the box. We may account for the effect of these walls using the statistical
mechanics principles introduced in Chapters 2, 3, and 4. We consider the
simplest possible wall, a plane which adds an infinite energy cost to any
particle intersecting it. We call this a hard wall. The effect of such a wall
is to remove any configurations in which wall intersects particles. The
result is something like the middle of Fig. 6.6. Clearly the wall disturbs the
arrangement of particles in its vicinity. The density near the wall is reduced.
The disturbance extends roughly a particle diameter into the liquid.
To add a hard wall to a liquid requires work. As we know from Chapter 2,
the work can be found by considering properties of the pure liquid. In a pure
liquid there is probability pA that a given region A has no particles passing
through it. As told in Chapter 2, the work required to remove the particles
from such a region is −T log pA . If the region happens to be the slab shown
on the right side of Fig. 6.6, then this is the work required to insert the slab.
If two slabs were inserted at separate places in the fluid, the probability that
each region is empty is pA , independent of the other region. The probability
that both regions are empty is pA 2 , and the work required is just double that
for one of the slabs. That is, the work is proportional to area of the slabs. The
same is true if two large slabs are joined to form a single slab of twice the
area. Then almost all of each slab is far from the other slab, and independent
of it. Thus the work to insert the slab has the form α(area), where (area)
means the area (on both sides) of the slab. The coefficient α is called the
interfacial energy or interfacial tension. If the liquid is a hard sphere liquid
like the one pictured, the probability pA is some number of order unity if the
slab is no larger than a particle radius a. That means the interfacial tension
α T /a 2 for such a liquid.
Real liquids generally have interfacial energies of this order, as well. The
fact that the molecules attract or repel each other somewhat does not change
the picture above qualitatively. Likewise, if the interface has some attrac-
tion or repulsion for the molecules, or some mild curvature or roughness,
this does not change the interfacial energy greatly. Even the free surface
between a liquid and its vapor is qualitatively similar. Like a hard wall, a free
surface is a surface where the density must be a small fraction of the liquid
density. Thus the disturbance it causes is qualitatively as great as that of a
hard wall.
 Simple fluids 155

3 Fig. 6.7
Surface energies for the liquids tabulated
in [5]. Horizontal scale is the surface
energy, in millijoules per square meter (or
2.5
Scaled dynes per centimeter). Molecules with NH
surface or OH groups are denoted by squares;
energy others are denoted by diamonds. Vertical
2 axis is the scaled surface energy
α/(T /δA). This δA is the area per flexible
segment of the molecule. It is calculated by
1.5 first determining the volume of a flexible
segment from its mass m and the liquid
density. Then δA is the projected area of a
sphere containing that volume. For small,
1
rigid molecules, the flexible segment is the
whole molecule. For flexible molecules, it
is the smallest freely jointed unit larger
0.5 than a C2 H2 unit. For cases where the
flexible unit is taken as smaller than the
Surface energy  whole molecule, an arrow extends upward
0 to indicate the range of values resulting
0 20 40 60 80 from choosing larger pieces of the
molecule as the flexible unit. Upper limit
of the range is indicated by a ceiling on its
arrow. Arrows with no ceiling are
Figure 6.7 shows some representative interfacial energies. As argued polymers. The lowest α’s shown are for
above, liquid interfacial energies are of the order of T per surface degree liquid helium at 4◦ K, m = 4 and liquid
of freedom. For small rigid molecules, this amounts to T per molecular helium at 1.6◦ K, m = 4. The remaining
area. For large flexible molecules, there is a characteristic area for each liquids, at room temperature, are, from left
to right, n-pentane, C5 H12 , m = 72.15,
independently moving piece† , which displaces a characteristic area at the polytetrafluoroethylene (Teflon),
surface. Then the interfacial energy is roughly T for each such area. Within (C2 F4 )n , m = 100, n-octane,
this generalization, polar molecules such as water tend to have more energy C8 H18 , m = 29, ethanol,
than nonpolar ones. Solids as well as liquids have interfacial energy. To C2 H5 OH, m = 46, methanol,
create new surface (e.g. by breaking a block of the solid in half ) requires CH3 OH, m = 32, acetone,
CH3 COCH3 , m = 58, n-dodecane,
work. One must overcome the attraction that binds the molecules of the CH3 (CH2 )10 CH3 , m = 28,
solid together. Thus the surface energy is of order of the binding energy cyclohexane, C6 H11 OH, m = 100,
of a molecule per area occupied by a molecule on the surface. The bind- n-hexadecane, CH3 (CH2 )14 CH3 , m = 28,
ing energy of a solid is generally much larger than the thermal energy T . benzene, C6 H6 , m = 78, carbon
Thermal fluctuations are minor and each atom is fixed at a position that tetrachloride, CCl4 , m = 153, polystyrene,
(CH2 CH(C6 H5 ))n , m = 52, polyvinyl
minimizes its potential energy (cf. the argument in Chapter 2 for viscosity). chloride, C2 H3 Cl, m = 62, ethanediol,
Thus the surface energy of a solid is often much more than T per surface HOCH2 CH2 OH, m = 62, formamide,
atom. Among solids, metals and ionic crystals have high surface energies HCONH2 , m = 45, glycerol,
(which moreover depends on the orientation of the surface with respect to HOCH2 CH(OH)CH2 OH, m = 31, water,
the crystal axes). Hydrocarbons have low surface energies. Rare-gas solids H2 O, m = 18, hydrogen peroxide,
H2 O2 , m = 34. For this wide range of
have the lowest ones liquids, the scaled surface energy remains
The interfacial energy shows up concretely whenever a liquid can change near 1.
its interfacial area. If a drop of liquid is suspended in space, it minimizes its
interfacial energy by adopting a spherical shape. If an external agent alters †For a polymer this would be a few back-
this shape, the work it does is α times the increase in area. If the fluid is dis- bone bonds and their attached atoms.
torted and then released, it begins to move so as to decrease its interfacial
area, converting interfacial energy into kinetic energy. The kinetic energy
is equal to α times the reduction in area (less any loss from viscous dissipa-
tion). If a spherical drop of radius R could reduce its volume by an amount
156 Interfaces

δV = 4π R 2 δR, the interfacial area would decrease by 8πRδR = 2δV/R.

Thus the interfacial energy would decrease by (2α/R)δV . Since energy is
gained by decreasing the volume, the interior of the drop experiences a pres-
sure 2α/R. This pressure, called the Laplace pressure, is present generally
inside a liquid with a curved surface.
When electric or magnetic fields are present, the energy cost of deforming
a surface can be greatly altered. A familiar example occurs when one draws
a dielectric fluid up into the gap of a parallel-plate capacitor by applying a
voltage to it. The force holding up the fluid acts at the interface. The strongly
paramagnetic fluids called ferrofluids experience analogous forces in a mag-
netic field, as noted in Chapter 1. These interfacial forces are important for
describing the behavior of some fluids in external fields. To characterize
what these forces do is difficult in general, since one must determine the
field profile and the interfacial profile simultaneously. The energy asso-
ciated with the interface depends in a nonlocal way on the shape of the
interface, in contrast to the surface energy, which is simply proportional to
the area.
6.1. Characteristic sizes of droplets.
(a) A drop of water of radius R has an interior Laplace pressure of one atmo-
sphere, 105 ergs/cm3 . Water has an interfacial energy of α of 70 ergs/cm2 .
How big is R?
(b) Another drop of water lies on a neutrally wetting surface and thus has a
hemispherical shape in the absence of gravity. Find the change of grav-
itational energy if this hemispherical drop is flattened onto the surface.
For what radius R is this gravitational energy equal to the original surface
energy?
(c) (Harder) What is the actual shape of a drop with this radius, accounting
consistently for gravitational distortion. (To find the shape of least energy
requires knowledge of Euler–Lagrange and Lagrange multiplier methods.)

6.2.2 Contact angle

The forces associated with interfacial energy are readily seen when two
interfaces meet. The most common example is a drop of liquid sitting on
a solid surface. If the drop is smaller than a few millimeters in diameter,
the effects of gravity are not important. The interface between the solid
and the liquid generally has a different interfacial energy αs from the air
interfacial energy α. Moreover, the solid–air interface has its own interfacial
energy αs0 . The body of liquid must adopt a shape which minimizes the
total interfacial energy: this is no longer a spherical shape with no contact
with the surface. Instead, the liquid may spread into a film that covers the
solid surface. This occurs when the interfacial energy of each area δA of
† In this comparison we have not considered the liquid-covered surface is smaller than that of the same area uncovered.
changes in the liquid interface area away The uncovered surface has energy δAαs0 . The covered surface has two
from the δA region being considered. In fact interfaces: the liquid–solid interface with energy δAαs and the liquid–air
there is no need to do so. When the liquid interface with energy δAα. Thus the spreading occurs when (αs + α)−
spreads to cover an additional area δA of
αs0 < 0† . From this condition we note two things. First since the α’s must be
bare surface, the necessary fluid can be sup-
plied by an overall thinning of the film. This positive, αs must be smaller than αs0 . The liquid molecules must be attracted
thinning does not change the interfacial area to those of the solid. Second this attraction must be sufficiently strong to
elsewhere. pay for the cost of extra free surface, with its energy α. This situation is
 Simple fluids 157

known as complete wetting. The liquid spreads to cover all the exposed area
of the solid surface (as much as the supply of liquid will allow). Complete
wetting tends to occur on high-energy solid surfaces, where αs0 is large. For
example, hydrocarbon oils typically wet metal surfaces completely. Often
a liquid is attracted to such a high-energy surface by enough to justify
complete wetting [6].
At the other extreme, it can happen that any contact between the solid
and the liquid costs energy. This occurs when the energy of an air–liquid
plus an air–solid interface is lower than that the of the solid–liquid interface:
α + αs0 < αs . Clearly this can only happen when αs > αs0 , so that the
surface repels the liquid. In this situation, a spherical drop in grazing contact
with the surface would separate from the surface (in the absence of gravity).
Such a surface is said to be completely nonwetting for the liquid.
For both complete wetting and complete nonwetting, α is smaller than
|αs − αs0 |. If instead α is larger than this difference, partial wetting occurs.
The liquid surface meets the solid surface at contact angle θ. Figure 6.8
shows how interfacial energy is related to forces in this familiar case. If
a section of the liquid of length L advances, a work −Lδxαs0 is done in
decreasing the bare surface. This is just the work that would be done if the
solid exerted a tensile force Lαs0 on the liquid trying to advance it. The
interfacial energy thus can be viewed as a force per unit length. This is why
interfacial energies are often called interfacial tensions. The liquid-covered
surface exerts an opposing tension Lαs . Finally, the advance would create
free liquid surface of amount Lδx cos(θ ), so that the free liquid exerts a
force Lα cos(θ ) opposing the advance. The sum of these forces must be
zero; otherwise the fluid would advance or retreat spontaneously. Thus

α cos(θ ) = αs0 − αs . (6.1)

This is Young’s law of partial wetting [6]. Young’s law determines the con-
tact angle; for a circular drop the contact angle in turn determines the
radius R. The free surface of a drop must have a constant spherical curvature
everywhere; otherwise, the Laplace pressure would be different in different
places. When the solid neither attracts nor repels the liquid, αs = αs0 . This
is the case of neutral wetting. Here the contact angle θ is 90◦ for any value
of α. In general αs is different from αs0 . Then if α is made smaller, the Fig. 6.8
contact angle decreases to zero to produce complete wetting or increases to Left: a partially wetting droplet on a
180◦ to produce complete nonwetting. horizontal surface, viewed from the side.
The contact angle θ is indicated. The liquid
The surface tension force of a liquid against its supporting structures can surface meets the solid surface at the
be mechanically significant. Such capillary forces can make fluid rise to contact line. As θ approaches 0 the droplet
macroscopic heights in a thin tube. Fluid transport in plants is accomplished approaches the state of complete wetting.
largely by capillary forces. In fine granular materials with large surface to As θ approaches π , the droplet approaches
the state of complete nonwetting (center).
Right: motion of the contact line by an
amount dx along the surface stretches the
s0 liquid surface by an amount dx cos θ. The
s vectors show the forces on a unit length of

the contact line, including the bare solid
dx cos
surface tension αs0 , the liquid-covered

solid surface tension αs , and the liquid–air
dx surface tension α.
158 Interfaces

volume ratios, these capillary forces can distort their supporting structures
or bind them together. Thus small amounts of water in the air can readily
form wetting layers on fine particles, turning a free-flowing powder into a
solid cake.
6.2. Capillary solid A person standing on a cube of wet sand exerts a stress of
104 N/m2 . This stress must be supported by sand grains whose size is a.
(a) Roughly how much force does a given grain support for a given size a?
On some of the grains this force is a tensile force trying to separate two
adjacent grains. Treat these two grains as cubes of length a and suppose
that the cube faces are separated by a thin water film. Separating the cubes
increases the thickness of this film and reduces its cross-sectional area to
maintain constant volume.
(b) For what size a is the interfacial tension equal to the applied force? Grains
smaller than this ought to support the weight of the person.

6.2.3 Wetting dynamics

If a drop of completely wetting fluid is placed on a surface, the interfacial
tensions discussed above are not balanced and the droplet begins to spread.
An extended treatment of this spreading dynamics is given in a big review
article by de Gennes [7]. The contact angle θ decreases with time. The speed
of advance v(t) must be such that the rate of decrease of interfacial energy
is equal to the rate of viscous dissipation. When a section of the drop moves
outward by an amount δx, the energy decreases by Lδx[αs −αs0 +α cos(θ)].
The quantity in [. . .] is called the spreading pressure S. For small θ it takes
the form S = S0 + 12 αθ 2 . Here S0 is the spreading pressure at θ = 0: S0 ≡
αs − αs0 + α. The nonzero wetting angle increases the spreading pressure.
Energy is lost at a rate LvS. The driving force for spreading is that which
would occur in a flat film plus an additional force due to the nonzero contact
angle. The dissipation opposing this driving force can involve thin film flow
of the liquid near the surface, which should not depend on θ, and also the
flow in the wedge shaped region beyond the contact line, which does depend
on θ. In order to sidestep the thin film effects we focus on the case where
the flat spreading pressure is nearly zero, so that S 12 αθ 2 .
The dissipation rate depends on the velocity gradient inside the drop. At
a distance x from the contact line, the drop has thickness h = x sin θ. The
average velocity gradient γ̇ is of order v/h = v/(x sin θ ) (since the velocity
goes to zero at the solid surface). The dissipation rate per unit volume w for
a fluid with viscosity η is ηγ̇ 2 or w ηv 2 /(x sin θ )2 . To obtain the total
dissipation rate W , we must integrate over position dx and multiply by the
R
thickness h at each x: W Lηv 2 b dx/(x sin θ ). Here we have taken
a minimum value of x to be that where the continuum description of the
fluid breaks down; b is roughly the size of a fluid molecule. The maximum
distance is of the order of the droplet radius R. Evidently, the integral is
a logarithm. For small θ, W Lηv 2 log(R/b)/θ. Equating the energy
loss rate with the dissipation rate, we find 12 αθ 2 = ηv log(R/b)/θ, so that
v log(R/b) (α/η)θ 3 . This speed v is simply the time derivative dR/dt.
The geometric properties of the drop give a further relation between v
and θ. The drop is a spherical cap of constant volume V . For small θ,
 Simple fluids 159

V R 3 θ. We can now combine this with the spreading-pressure relation

above for small θ to obtain
   
dR R α V 3
log
dt b η R3

This equation readily tells how R varies with time t, if we neglect the slowly
varying log(R/b):
R [(V 3 α/η)t]1/10 . (6.2)
This is the well-known Tanner law of wetting [8]. It says that the spreading
slows in a universal way with time in the marginal case where S0 0.
Remarkably, Tanner’s law is observed even when the flat spreading pres-
sure S0 is large and should dominate the full spreading pressure. Instead
of advancing rapidly in response to the full spreading pressure, the macro-
scopic contact line advances gradually as though there were no flat spreading
pressure. As anticipated above, the extra spreading pressure is opposed by
an extra dissipation mechanism, not considered above, that is independent
of the angle θ. As explained by de Gennes and Hervet [9], the fluid spreads
far ahead of the macroscopic contact line in a thin precursor film or foot. The
larger the flat spreading pressure is, the faster this film advances. The flat
spreading pressure is balanced by dissipation in the precursor film. Only the
residual spreading pressure 12 αθ 2 remains to be balanced by the dissipation
in the macroscopic drop.

6.2.4 Surface heterogeneity

Real solid surfaces are not completely uniform. Impurities and irregularities
cover the surface in a way that varies from place to place on the atomic scale.
Thus at some places the αs0 − αs is bigger than average, so the local contact
angle is smaller. At other places the local angle is larger than average. If the
contact line is to advance for a macroscopic distance, it must be able to move
over even those places with the largest contact angle. If it is to recede, it must
detach from even those places with the smallest contact angle. If the contact
angle lies between these two angles it can neither advance nor recede; it can
only make local readjustments. The contact line is pinned. This pinning
means that the contact angle is not uniquely determined but depends on
how the fluid was placed on the surface. The wetting is hysteretic.
Wetting hysteresis is readily observable on most surfaces. Even nomin-
ally smooth and uniform surfaces show a several-percent spread between
advancing and receding contact angles. Deliberately nonuniform surfaces
can pin a contact line over a wide range angles, from nearly vertical to
nearly zero [10].
An important form of surface heterogeneity is surface roughness. Micro-
scopic roughness increases the surface area relative to the projected area
by some factor β larger than 1. Thus the energy per unit projected area αs0
is increased by a factor β relative to a flat surface. So is the liquid–solid
energy αs . The contact angle θ is now given by α cos θ = β(αs0 − αs ).
We see that roughness amplifies the wetting preference of the surface: it
160 Interfaces

increases the departure from the neutral conditions where θ is 90◦ . In

addition to this effect roughness causes wetting hysteresis.

6.2.5 Other interfacial flows

The simple flow represented by a spreading drop is just one of a rich vari-
ety of interfacial flow phenomena. As we saw, the driving force for this
spreading was the surface interfacial energy; such flows are termed capil-
lary flows. This same force can drive the reverse phenomenon of dewetting.
Dewetting occurs when a surface is coated with a film of liquid that does
not wet that surface in equilibrium [11]. It may be prepared by letting a
thick layer of liquid evaporate. Such a film is metastable. If a small dry
spot forms, it spontaneously spreads to form a growing circular dry patch.
The liquid from the dry region collects near the border faster than it can
spread through the film, thus forming a thick ring of spreading liquid. Since
the interfacial energy gained is dissipated in the adjacent ring, the spread-
ing velocity becomes independent of the ring’s radius. Both spreading and
dewetting flows are qualitatively different and faster if the solid substrate
is replaced by a viscous fluid.
Even without dewetting, evaporation causes flow when the thickness of
a fluid layer is not uniform. For example, a thin circular droplet with a
pinned contact line must thin further without contracting as it evaporates.
The thickness at a given place must thus decrease at a rate proportional
to its local thickness. But the local evaporation rate does not depend on
thickness in this way. Thus to maintain the proper droplet shape, lateral
flows are necessary. These flows are strong enough in practice to bring
material from the center to the contact line during the drying time [12].
Another important driving force occurs when a liquid is nonuniform in
composition or temperature. In such cases the interfacial energy is in gen-
eral nonuniform, as well: part of the surface could have a lower energy than
an adjacent part. For example, one part of the surface could have a higher
concentration of some solute molecule, leading to a lower interfacial energy
there. We shall explore this solute effect in the next section. Whatever the
cause of the nonuniform interfacial energy, the nonuniformity results in a
driving force to expand the region of lower interfacial energy, just like that
which causes spreading of a wetting liquid on a surface. The resulting flow
is called a Marangoni flow [13]. The classic example of Marangoni flow is
readily seen in a freshly poured glass of wine. “Tears of wine” climb up the
sides of the glass. The upward flow that forms the “tears” results from a sur-
face tension gradient. The top of a tear has a higher surface tension because
more alcohol has evaporated there than elsewhere. The wine does not climb
the glass in a uniform ring, but rather climbs higher in some regions than in
others, thus forming the tears. The reason is that a uniform ring is unstable.
The higher a climbing region rises, the further it gets from the rest of the
wine. This accentuates the evaporation and allows that region to climb still
higher. Such instabilities are common features of Marangoni flows.
Gradients of temperature or composition on a solid surface can also
give distinctive features to liquids spreading over them. The flow can even
advance and retreat many times before coming to rest [13]!
 Solutes and interfacial tension 161

6.3 Solutes and interfacial tension

As we noted above, molecules dissolved in a liquid generally influence its
interfacial tension. In this section we investigate how large this influence is.
A very simplified approximation gives the right order of magnitude. We
imagine a spherical droplet of liquid with Ns solute molecules. We further
imagine that these molecules have a strong attraction for the surface—so
strong that they are all at the surface. We suppose that Ns is small enough so
that the molecules are far apart on the surface and have negligible interac-
tions. We now distort the shape of the interface, thus increasing its surface
area by an amount δA. Without the solute, this distortion requires work. If
the liquid’s surface tension is α0 , then the work is α0 δA. The Ns solute
molecules alter this work. These molecules are confined to an area of
A/Ns apiece. Increasing the area allows each molecule to take on more
configurations c and thus decreases its free energy; for all the molecules
this free energy F is a constant minus Ns T log(A/Ns ). This form and
the rationale for it were discussed in a three-dimensional context in Prob-
lem 2.3). The work δW is δA∂ F /∂A = −δAT (Ns /A). Not surprisingly,
this work is the same as we would encounter in an ideal, two-dimensional
gas: the surface pressure is T times the number density Ns /A.
As we see, the solute acts to reduce the surface tension by means of
an opposing surface pressure. Generally, this opposition is weak. We have
argued above that most simple liquids have a surface energy of the order of
T per molecular area. If the solute molecules are dilute, the area per solute
molecule is necessarily much bigger than that of the base liquid. Thus these
solute molecules can only reduce the surface tension by a small fraction.
To attain a big change of surface tension, the adsorbed molecules must be
strongly concentrated at the surface and must interact strongly. We shall
explore such strongly interacting adsorbates in the next chapter.
We may improve the primitive picture above by relaxing our assump-
tion that the solute molecules are confined to the interface. The effect of
the solute molecules can be treated systematically by first looking at how
one solute molecule affects the free energy of the droplet. The main part
of the free energy has the form F1 − T log , where  is the volume of
the drop and the constant F1 gives the work required to insert the molecule
at some given place (removing solvent molecules so that the volume stays
fixed). Now we consider the effect of the surface. In general our molecule
interacts with the surface and thus it has a nonuniform probability density.
We may represent the local probability density as ρ0 f (z), where ρ0 = 1/ 
is the density of solute molecules, z is the distance to the interface and f (z)
is the relative probability that the molecule is at a particular position at
distance z from the interface. Thus f (z) is an un-normalized configuration
probability, as discussed in Chapter 2. In that chapter, we showed how the
free energy of such a system could be expressed in terms of f :
 
F2 = F1 − T log 3
d rf (z(r))

We notice that if the interface has no effect on the solute, f (z) is a constant
and the expression for F2 reduces to our initial expression F1 − T log().
162 Interfaces

Our interest is in a macroscopic interface with little or no curvature. By

comparison, any nonuniformity in f (z) is a short-ranged enhancement or
suppression. In this situation we can simplify F2 . We  shall first add and
subtract 1 from f (z) to obtain F2 = F1 − T log( + d 3 r[f (z(r)) − 1]).
The last integral vanishes
 except near the surface and becomes in the limit
of small curvature A dz[f (z) − 1]. For sufficiently large , the first term
of the log must always be much larger than the second. Thus the log can
be expanded: F2 → F1 − T log  − A [f (z) − 1]/ . The nonuniform
f (z) creates a contribution to F proportional to the interfacial area. Now
we can readily write the overall free energy F . It has a contribution F0 (A)
in the absence of solute molecules, plus a second contribution N F if there
are N noninteracting solute molecules:

f (z) − 1
F = F0 (A) + N F2 = F0 (A) + N F1 − T N log  − T N A

As we have explained above, the liquid without solute has its own interfacial
energy α0 , so that F0 (A) = F00 + Aα0 . We notice that N /  is simply the
average density of solute molecules ρ0 . We thus see that F has the form
constant + Aα, where

α = α0 − T ρ0 [f (z) − 1] dz

The ρ0 [f (z) − 1] dz has a simple physical interpretation. It is the average
number of solute molecules per unit area near the surface in excess of those
†In normal usage  is defined as the excess present with no surface interaction. It is called the surface excess  † . We see
mass of solute per unit area. Thus if the that whenever the surface excess is positive, the surface tension is reduced.
solute molecules have mass m, the usual
The reduction is just the pressure that would be exerted by an ideal surface
surface excess is m times our .
gas of areal (mass) density , as in our initial example above. Conversely,
if the solute is repelled from the surface, the interfacial energy is increased.

A 6.3.1 Fluid mixtures

The above picture remains qualitatively true when the solute concentration
Distance from wall becomes substantial, so that our liquid is a two-component mixture. In
general, one of the two components will be attracted to the interface. We
Fig. 6.9 denote this attracted component as A, and denote the other as component B.
Concentration profile of a fluid mixture From the above discussion adding more A to the mixture should lower
near an attracting wall. φA is the volume the interfacial tension. A further phenomenon can happen if A and B are
fraction of species A. Lower curve shows immiscible. Then when enough A is added, the liquid partitions into two
profile for when the two constituents mix;
the attraction increases the interaction,
phases, as sketched in Fig. 6.9. The phase richer in A is then attracted
which increases the concentration of A to the interface and sits next to it. If the two phases have a high interfacial
near the wall. Upper curve shows the tension between them, then the A-rich phase forms partially wetting droplets
profile after more A is added to the on the external interface. One important limiting case is a critical mixture, in
mixture so that it demixes into an A-rich which the two phases become nearly identical in composition. It has been
and an A-poor phase. Now the A-rich
phase condenses onto the wall and extends
shown [14] that such a critical mixture always completely wets any external
a macroscopic distance from it. If the A–B interface: the A-rich phase coats the interface.
interfacial energy is large, the A-rich phase It appears that adding immiscible solute molecules should be a good
may form droplets on the wall. strategy for lowering interfacial energy: the immiscible molecules are
 Polyatomic solutes 163

driven to the surface, thus lowering the surface energy. Making the solute
more immiscible in the liquid, amounts to increasing the free energy cost
F1 defined above. But in fact, making the solute immiscible is generally not
a good strategy for lowering surface tension. Immiscible solute molecules
generally experience an effective attraction for each other owing to their
mutual repulsion from the solvent. Such attractions are present even when
the solute molecules are driven to the surface. Thus instead of spreading
over the surface and producing surface pressure, the solute molecules may
phase separate on the surface, thus cutting the desired surface pressure.
A way around this dilemma is to use amphiphilic solute molecules. Such
molecules have a part that by itself would be insoluble and another part
that by itself would be highly soluble. As we shall see in the next chapter,
such amphiphilic molecules can segregate strongly to the interface without
phase separating there.

6.4 Polyatomic solutes

Up to now we have dealt with small-molecule solutes. What new inter-
facial phenomena appear when the solutes are large, polyatomic objects:
colloidal particles, colloidal aggregates, or flexible polymers? For colloidal
particles, the situation is much the same as for small solute molecules. The
main difference is that interactions are strong. A colloidal particle at a free
surface can deform it, thereby inducing attractive interaction with neigh-
boring particles, like Cheerios on the surface of milk, or like the mattress
effect of Fig. 5.1. The colloidal aggregates treated in the Appendix to the
Chapter 5 experience similar capillary forces; these are often strong enough
to crush an aggregate.

6.4.1 Polymer adsorption

In comparison to the cases above, the adsorption of polymers is rich and
subtle. Chapter 5 gave us a hint of this subtlety. The subtlety arises because
polymers are deformable, so that small increments in adsorption strength
can lead to significant changes in the polymer’s shape. Polymers adsorb
because their monomers have a short-ranged affinity for a fluid boundary.
We may represent this affinity by a monomer interaction potential u(z),
where z is the distance from the surface. Since this affinity arises from
the local liquid structure, its range a is no larger than a monomer and is
much shorter than the polymer coil size R. The attractive potential u(z)
is sometimes called a phantom attraction. Real interfaces have additional
features, to be revealed below.
How big must u be in order to cause significant adsorption? We wish to
enhance significantly the equilibrium probability that the polymer will be
found next to the adsorbing surface. This means that the total free energy
of the polymer at the surface must be smaller than its free energy far away
by an amount T or more. To estimate the u necessary, we consider a pen-
etrable or phantom surface, rather than a hard wall. Further, we suppose
initially that u is so weak that the polymer does not get distorted by the
adsorption. Then the adsorption energy U is simply u times the number
164 Interfaces

of monomers within a distance a of the surface. To find this number, we

note that it is the number of intersections MAB of two fractals, as often
encountered in the Chapters 3 and 4. One of the fractals is the surface,
with dimension DA = 2; the other is the polymer, with DB 53 . Accord-
ingly for a polymer of radius R having n ∼ R DB monomers the number of
intersections is given by (const)R DA +DB −3 ∼ R 2/3 ∼ n2/5 . Evidently in
order to have an adsorption energy of T , one needs u ∼ n−2/5 . The energy
umin needed to adsorb a polymer is far smaller than that needed to adsorb
a monomer.
We now consider the effect of increasing u above the value umin needed
for adsorption. It is easily possible for u to be much larger than umin and
still much smaller than T . This means that short sections of the polymer
are affected hardly at all by the surface, but long sections have large energy
and are affected strongly. We have seen that the energy needed to make a
significant distortion in a polymer coil is only of order T . Thus we must
anticipate that adsorption energies much larger than T should cause large
distortions.
We anticipate that the polymer will be compressed towards the attract-
ing surface, so that its typical thickness is some ξu  R. We may find an
optimal value of ξu by considering the two forms of energy that depend on
ξu . To find the absorption energy U , we note that the compression mul-
tiplies the density of monomers at the surface by R/ξu . Thus the energy
is uMAB R/ξu un(a/ξu ). By itself, this adsorption energy U favors ξu
being as small as possible. But opposing U is the energy of confinement;
let us call it C. We touched on this confinement energy in Chapter 5 in
the Section on repulsive forces. We considered another form of confine-
ment in Problem 3.15. Paraphrasing that discussion, we imagine confining
polymer to a slab of width ξu . To achieve this confinement we divide the
initial unconfined chain configuration into k blobs of size ξu . Finally, by
manipulating one degree of freedom in each blob, we twist the chain so that
it fits into the slab. We have thereby constrained k degrees of freedom and
thus increased the free energy C by roughly T k. Since k(ξu /a)5/3 n, we
conclude C T n(a/ξu )5/3 . Naturally, C resists confinement; it increases
as ξu decreases.
The balance of the opposing effects of C and U gives an optimal ξu .
For the optimal ξu , C U , as with all such optimizations (cf. the Flory
argument in Chapter 3). We note that both C and U are proportional to the
number of monomers n: doubling n (at fixed ξu ) doubles the number of
attracting monomers; it also doubles the number of confined blobs. Thus
n is a common factor in C and U , and the optimal ξu cannot depend on n
(or R). Comparing C and U , we infer T (a/ξu )5/3 ua/ξu , or ξu
a(u/T )−3/2 . The balance of energy occurs blob by blob, with each blob
having an attracting energy and a confinement energy of order T .
Having found the width ξu , we can now ask what is the lateral dimension
R of the adsorbed chain. The adsorbed blobs repel one another just as
unadsorbed blobs do. Thus they form a self-repelling random walk on the
surface. The self-repulsion causes lateral expansion as in three dimensions.
As we saw in Chapter 3, the scaling law for this expansion is R ∼ n3/4
 Polyatomic solutes 165

Since this power is larger than its three-dimensional counterpart 35 , R must

become larger on adsorption due to increased self-repulsion effects.
If there are other chains in the solution, there is plenty of room for these to
adsorb on the surface along with the single chain discussed above. The entire
surface can be covered densely with adsorbed blobs with little interference
from other chains. The surface is then a two-dimensional semidilute solution
of these adsorbed blobs. It has an osmotic pressure owing to the mutual
repulsion of the blobs of order T per blob—the same magnitude as the
adsorption energy U and the confinement energy C. This osmotic pressure
reduces the net binding energy, but only by a factor of order unity. The result
is to increase the size of the adsorbed blobs by a finite factor. Thus we arrive
at a picture of how polymers adsorb from dilute solution in the presence of
weak attraction. There is a dense layer of the thickness of one adsorption
−1/3
blob of size ξu . This layer contains (ξu /a)D /ξu 2 ∼ ξu monomers per
unit area. This surface excess produces a surface osmotic pressure T /ξu 2 ,
opposing the surface tension of the bare surface. We have seen that bare
surfaces have energies of order T /a 2 or more; thus, the polymers in the
weak-adsorption regime have little impact on the overall surface energy.

6.4.2 Concentration profile

This picture describes the dominant energies and length scales in an
adsorbed polymer layer. In this simple picture, the monomer concentra-
tion falls rapidly to zero beyond distances of order ξu . For some purposes
it is good to know just how rapid this falloff is. The problem is to find
the optimal concentration profile, given that the surface concentration is
(ξu /a)D /ξu3 . In our simple picture, there are k blobs per chain, and thus
1/kξu 2 chains per unit area. One possible way to optimize the concentra-
tion profile is to adsorb more chains but with fewer adsorbed blobs per
chain. Attracting additional chains costs free energy. On the other hand, all
the adsorbed chains can be less confined in this situation, thus reducing the
confinement energy. Our problem is to find the best tradeoff between these
two effects.
We may analyze this tradeoff most simply by supposing that our polymers
have an enormous molecular weight, so that the bulk polymers are much
larger than ξu . From the reasoning above, we expect the surface to be
saturated with blobs of size ξu . We now consider a chain that passes through
a point at some height z  ξu . In the absence of the adsorption, this chain
would intersect the wall at a distance of order z from the starting point.
However, with the attraction and its associated layer of blobs, our chain is
perturbed. We now consider a sequence of a few blobs on our chain near
the surface, as shown in Fig. 6.10, top. Since the surface is assumed to be
saturated with blobs, our chain cannot crowd into the layer without removing
some other blobs. However, if blobs from our chain simply exchange with
those of the adsorbed chains, as in Fig. 6.10, bottom, there is no change
in the adsorption or confinement energy. Thus the new configuration is
roughly as probable as the old one. Thus any segment of our chain that
166 Interfaces

ξu

Fig. 6.10
Chains adsorbed on a phantom surface.
Top: a single chain adsorbs in a thickness
ξu as described in the text. Middle: a chain
from the bulk solution approaches a
saturated adsorbed layer. Bottom: blobs of
the bulk chain exchange with those of the
surface, thereby binding the bulk chain and
weakening the attachment of the
pre-adsorbed chain.

approaches the wall has a finite probability of being adsorbed, displacing

some of the previously adsorbed chains. The displacement process has little
effect on the density near the wall. Still, the tightly adsorbed chains of our
initial state evolves into a greater number of chains that are more loosely
adsorbed and that extend further from the surface. We now ask how many
such chains are present at a height z  ξu .
When the adsorbed layer has reached equilibrium, chains at any height
z feel no net force. We consider a segment of chain of size z that is at
height z above the surface. This segment is part of a much longer chain,
thus it is typically attached to the wall. There is some tensile force in the
attaching ends tending to pull the segment towards the wall. Our arguments
above lead us to suspect that these chains are weakly bound, with an energy
of order T . This would mean that there is not enough energy to distort
the segments substantially. Accordingly we shall suppose that the adsorbed
segment keeps its unperturbed size. Then we will argue that this state is
self-consistent. The tensile force is then of order T /z. If one chain can
be marginally bound at this height, others can as well. Chains congregate
at height z until their mutual repulsion prevents them from congregating
further. At this point their repulsive interaction energy is equal to their
adsorption energy, which in our picture is of order T . These loops of height
z are thus in the same situation as the original ξu blobs. They form a layer
of loops at a separation of order z. The concentration at this height z is thus
roughly the average concentration within our segment, viz φ 0 (z) ∼ zD−3 .
Our assumption of weakly distorted segments is self-consistent. Now we
further define our segment to be that part of a chain which extends from the
original height z to z/2. This segment experiences a gradient of osmotic
 Polyatomic solutes 167

pressure, tending to push it away from the surface. Since the diameter of our
segment is roughly z, the force that it feels is roughly ((z/2) − (z))z2 .
Now, the blobs at height z have a size that is also roughly z according to
our picture. Thus (z) T /z3 . The force is thus of order (T /z3 )z2 or
T /z. This is of the same order as the tensile force we supposed above. Thus
under our assumption, the forces acting on any typical segment are of the
same order. Any adjustment required in order to bring the segments to an
equilibrium state of balanced forces is a minor adjustment that does not
alter the scaling behavior we have deduced.
This picture of the height concentration profile, originally postulated by
de Gennes [15], has been supported by several calculational approaches.
It has also been confirmed by realistic computer simulations [16]. Direct
experimental confirmation is difficult because the profile involves only
a tiny fraction of the adsorbed monomers. Indeed, the above argument
says that φ(z) ∼ z−4/3 , so that the total surface excess  ∼ dzφ(z)
is dominated by the smallest z’s—i.e., z ξu . Thus  ξu φ(ξu ), as
argued above. Still, scattering investigations of the profile show behavior
consistent with the de Gennes picture [17].

6.4.3 Hard wall

This picture describes the main features of adsorbed polymer structure. But
these features are modified subtly by additional effects. Up to now we have
considered adsorption by a penetrable, phantom surface. Real adsorbing
surfaces are normally impenetrable. Impenetrable surfaces without adsorp-
tion are not neutral; they actively suppress close approaches of the polymer
coils. Any monomer that is adjacent to the surface suffers extra constraints:
the subsequent monomer may not pass through the surface. Thus, its angular
freedom is restricted. This amounts to a free energy cost of order T for each
such monomer. A weak attractive energy u is not sufficient to overcome this
repulsion. Instead, u must be of the order of T if its range is a statistical seg-
ment of length a. Adsorption begins to occur only if u exceeds a threshold
of order T . Weak adsorption with small surface excess then occurs when
u exceeds this threshold by a small fraction of T . In this regime, one may
define the adsorbed blob size ξu and discuss the concentration profile for
distances larger than ξu as above.
Over heights z smaller than ξu , the impenetrability has an interesting
effect. For a phantom surface, the concentration is virtually independent of z
in this proximal region. For segments much smaller than the blob, the poten-
tial energy of adsorption U of the segment is much smaller than T . Thus it
can be little affected by the surface. But for real surfaces there is an interplay
between the impenetrability constraint and the self-repulsion of the chain.
This leads to new effects [18]. Even a single chain constrained to touch an
impenetrable surface with a large number of its monomers has a power-law
falloff of concentration near the surface varying z−p , where p 13 . This
extra concentration at the surface arises because the adsorbed monomers
occur not independently but in correlated trains. These correlations can
arise because the adsorption energy per monomer for an impenetrable wall
must be significant on the scale of T . When adsorption is very weak, this
168 Interfaces

proximal power law can lead to measurable effects on the surface excess
and the concentration profile. The reasons for the proximal power are dif-
ficult to account for in simple terms, and direct evidence for it has proved
difficult to establish experimentally [19]. There is in particular no simple
connection between the proximal power and the fractal dimension D of a
self-avoiding polymer.
Our discussion of the density profile assumed that the chains had indefin-
itely long length. When their length is finite, the power-law regime is limited
to distances z smaller than the unperturbed size of the chains. In addition, the
finite-length chains appear to show a second power-law regime that implies
a falloff of concentration slightly weaker than the z−4/3 found above for
sufficiently large z, owing to tails of chains adsorbed at only one end [20].

6.4.4 Kinetics of adsorption

As we have seen above, the attainment of an equilibrium adsorbed layer
requires the cooperative motions of many chains as the initial tightly
bound chains are gradually displaced by additional loosely bound chains.
A new unabsorbed chain must overcome an activation barrier (discussed in
Chapter 2) in order to be adsorbed. It must penetrate into regions of increas-
ingly high concentration against an osmotic pressure gradient before any
of its monomers can experience the attracting surface. If the adsorption is
strong, the region near the surface is often so crowded with monomers that
they are virtually immobilized as in a glass. In such cases the relaxation to
equilibrium can take hours [21].

6.4.5 Surface interaction

As we saw in Chapter 5, when surfaces perturb the liquid around them, there
is an induced attraction or repulsion between those surfaces. The adsorption
of polymers induces interactions between two surfaces in this way. If the
polymers are irreversibly adsorbed, the osmotic pressure at the midpoint
increases as the two surfaces approach. Thus external work is required to
decrease the separation and the interaction is repulsive. Even though most
of the monomers live at distances of order ξu from the surfaces, the range
of the significant repulsion may be much greater than ξu . If two colloidal
particles have radii much larger than the polymer size R, the repulsive
energy at separation R is many times T . If the adsorption is reversible and
reaches equilibrium, two surfaces with overlapping adsorbed layers may
release polymers, since one surface may now use chains from the other
surface in order to attain its optimal coverage. Now the forces crossing the
† The recent theory of Semenov and Joanny midplane come not only from the repulsive osmotic pressure, but also from
[20] predicts a net repulsion at larger separa- the tension in chains that bridge between the two surfaces. The net effect is
tions. Such a repulsion need not violate the attractive for sufficiently small separation† .
attraction theorem discussed in Chapter 5.
That theorem concerns of perturbed fields
whose energy senses nonuniformity only 6.4.6 Flow
through local gradients. However, polymer
solutions can in principle have energies that An adsorbed polymer layer has an important effect on the flow past a surface.
depend nonlocally on the concentration. Near a bare, solid surface, the velocity goes to zero linearly as the surface is
 Polyatomic solutes 169

approached, yielding a uniform shear rate and a uniform viscous stress. But
flow through an adsorbed layer need not have a uniform viscous stress.
Instead, force can be transmitted from the fluid to the polymers, and thence
directly to the surface. We saw in Chapter 4 that fluid velocity decays expo-
nentially in a semidilute solution of blob size ξ , with a decay length of the
order of ξ . Similar decay occurs in an adsorbing layer. We may picture the
layer as having an outermost sublayer of thickness R/2, extending from
height R/2 to height R and having blobs of size R/2. Then the adjacent
denser sublayer has thickness R/4 and blob size R/4, followed by suc-
cessively denser sublayers. We suppose that the flow velocity at the outer
boundary of the layer is v0 . Then on traversing the outer sublayer, the
velocity must decay to a finite fraction of v0 , say v0 /χ. If this χ is larger
than 2, the velocity decreases faster than the height z. After k sublayers,
z = R × 2−k and v = v0 × χ −k . Solving for −k yields

log z/R log v/v0

=k= ,
log 2 log χ

so that v/v0 = (z/R)log χ / log 2 .

This χ must indeed be larger than 2. If χ were 2, the velocity would
interpolate linearly between 0 and v0 as one traverses the layer. This is
exactly the velocity profile near the bare layer, with only viscous drag.
Now, the polymer layer certainly has substantial drag in addition to the
viscous drag, and it must decrease the velocity within the layer by at least a
finite factor. The only way to attain this decrease is for χ to be bigger than 2.
We conclude that the flow velocity extrapolates to zero at some height of
order R, greater than zero. Thus as far as the flow is concerned, the layer
behaves as though it had a thickness t of order R rather than ξu , as shown
in Fig. 6.11. Our reasoning depends on the flow being weak, so that it does
not disturb the adsorbed layer appreciably. The effect of adsorbed layers
on flow has been tested using the surface-forces apparatus [22]. These
experiments confirm our conclusion that the hydrodynamic thickness of
order R, though they have not demonstrated the form of the velocity profile.
The factor χ depends on the screening properties of polymers in a semidilute
environment. We have seen that these are characterized by a screening
length of order ξ , independent of the chemical structure of the polymer or
the solvent. We argued that the screening length is a universal multiple of the
structural correlation length. Similar reasoning led to our conclusion that the
χ factor above is a finite number for any adsorbed layer in a good solvent,
independent of such structural factors. Thus we expect χ to be universal,
and accordingly the exponent log χ / log 2 for the velocity profile should be
universal, as well. To our knowledge this universal exponent has not been
calculated or measured.
Polymers influence flow past a surface in an important non-adsorbing
situation. Here we consider an entangled polymer solution or melt flow-
ing past a smooth wall. Now the stress reaches the wall via two different
processes in series. Within the fluid, the stress is transmitted through entan-
glements and is governed by the large viscosity of the polymer solution.
But there are no entanglements connecting the wall to the adjacent fluid.
170 Interfaces

Fig. 6.11
Top: sketch of a saturated adsorbed
polymer layer in equilibrium with a
solution. Layer consists of loops at various
heights such that the distance between
loops at height z is comparable to z.
Speed v
Bottom: illustration of the procedure in the
text for accounting for the velocity profile.
Beyond the adsorbed layer, the velocity
v0
grows linearly with distance from the
surface. Within the adsorbed layer, the
velocity decreases by more than a factor 2
for each halving of the distance, resulting v0/2
in a power-law increase of velocity whose v0/χ
exponent is larger than 1. The velocity at
large distances extrapolates to a height t, t
the effective hydrodynamic thickness. It is Height z
comparable to the layer thickness R. R/4 R/2 R

Stress

Fig. 6.12
Flow of an entangled polymer solution past Velocity
a wall, sketched at top. Upper curve shows
the (uniform) shear stress. Lower broken
curve shows the velocity profile required to
attain the uniform stress. The extrapolation Height z
length b is shown on the horizontal axis. b

Here the stress is conventional fluid stress, caused by fluid and polymer
atoms hitting wall atoms. It can be characterized by the viscosity of the
solvent or of disconnected monomers. Though the means of transmitting
stress in these two regions are very different, the amount of stress must be
the same. The stress must pass from the bulk of the fluid to the container
via the wall layer. To attain the same stress as the bulk, the wall layer must
have a much larger velocity gradient, as sketched in Fig. 6.12. The velocity
extrapolates to zero beyond the wall, at a distance b called the slip length.
For simple fluids the slip length is of the order of an atomic size a. But if the
 References 171

polymer viscosity is larger than the solvent viscosity by a large factor, the
slip length exceeds a by an equally large factor. Such large slip lengths are
readily observed, and they may reach the scale of microns [23]. Sometimes
one wants to avoid this slipping behavior. One way to achieve this is to
fasten polymers to the wall. Then the fluid polymers must disentangle from
the wall polymers in order to flow. One may envisage all kinds of exotic
behavior when these flows are fast enough to perturb the fastened poly-
mers or to inhibit re-entanglement. The engineering literature has no lack
of examples of such exotic behavior. Perhaps the best known are the “shark-
skin” patterns caused by stick-slip flow past a wall under these nonlinear
conditions [24].

6.5 Conclusion
In this chapter we have accounted for the energy scales associated with
liquid surfaces and interfaces. We have seen how these energies create geo-
metric and kinetic effects e.g. in wetting phenomena. Polymers and colloidal
suspensions interact strongly with surfaces and this interaction can alter the
interaction between two surfaces. It can also alter flow past a surface. With
these interfacial phenomena in mind, we are ready to explore the distinctive
phenomena caused by surface-loving molecules, or surfactants.

References
1. Tantec contact angle micrometer Tantek Co. 630 Estes Avenue, Schaumburg
IL 60193 http://www.tantecusa.com/camplus2.html (2002).
2. Pamphlet “Spinning Drop Tensiometer Site 04” KRUESS USA, Instruments
for Surface Chemistry 9305 Monroe Road, Suite B, Charlotte, NC 28270 -
1488 http://www.kruss-usa.com/public_pdf/Site04e.pdf
3. SE400: Sentech Discrete Wavelength Ellipsometer http://www.sentech.
com/ellipsometer.htm SENTECH Instruments GmbH Carl-Scheele-
Strae 16, 12489 Berlin, Germany.
4. F. Rondelez, D. Ausserre, and H. Hervet, Ann. Rev. of Physical Chemistry 38
317 (1987).
5. J. Israelachvili, Intermolecular and Surface Forces (New York: Academic,
1985) Table XVII, p. 158.
6. J. Israelachvili, op. cit. Chapter 14.
7. P. G. de Gennes, Rev. Mod. Phys. 57 827 (1985).
8. L. Tanner, J. Phys. D. 12 1473 (1979).
9. H. Hervet and P. G. de Gennes, Comptes. Rendus Acad. Sci. 299II 499 (1984).
10. L. Leger and J. F. Joanny, Rep. Prog. Phys. 55 431 (1992).
11. C. Redon, F. Brochard-Wyart, and F. Rondelez, Phys. Rev. Lett. 66 715 (1991).
12. R. D. Deegan, O. Bakajin, T. F. Dupont, G. Huber, S. R. Nagel, and T. A. Witten,
Nature 389 827 (1997); Phys. Rev. E. 62, 756–765 (2000).
13. S. H. Davis, Ann. Rev. Fluid Mech. 19 403 (1987).
14. D. Jasnow, Rep. Prog. Phys. 47 1059 (1984).
15. P. G. de Gennes, Adv. Colloid Interface Sci. 27 (1987).
16. J. de Joannis, R. K. Ballamudi, C. W. Park, J. Thomatos, and I. A. Bitsantis,
Europhysics Lett. 56 200 (2001).
17. L. Auvray and J. P. Cotton, Macromolecules 20 202 (1987).
18. E. Eisenriegler, J. Chem. Phys. 79 1052 (1983).
19. J. M. DiMeglio and C. Taupin, Macromolecules 22 2388 (1989).
172 Interfaces

20. A. N. Semenov, J. F. Joanny, A. Johner, and J. BonetAvalos, Macromolecules

30 1479 (1997).
21. K. Kremer, J. Phys-Paris 47 1269 (1986).
22. J. Klein, Ann. Rev. Materials Sci. 26 581 (1996).
23. F. Brochard and P. G. de Gennes, Langmuir 8 3033 (1992).
24. R. G. Larson, Rheol. Acta 31 213 (1992).
 7
Surfactants

7.1 Introduction
In the previous chapter we saw that the boundary of a liquid is respons-
ible for significant, controllable forces—capillary forces. In this chapter
we encounter an enormous expansion of that control through the use of
“amphiphilic” molecules. An amphiphilic molecule contains a part that by
itself would be soluble in the liquid and another part that would be insol-
uble in it. The insoluble parts have a more favorable free energy when they
are away from the liquid; thus these molecules tend to concentrate at the
liquid boundary. Because they influence surfaces and other boundaries,
these molecules are called surfactants.
In a sense a surfactant incorporates a bit of interface within itself. The
two dissimilar parts of the molecule, having opposite affinities for the
liquid, generally have a repulsive interaction with each other. This repul-
sion amounts to a free energy that was stored in each surfactant molecule
when it was made. The boundary between the two parts of the molecule
often plays the role of an interface. To minimize the repulsive energy
inherent in the amphiphilic molecules, they often organize to create real
interfaces throughout the liquid, as shown in Fig. 1.2. The resulting struc-
tures can dramatically alter the forces within a liquid. In this way capillary
forces come to control the bulk properties of the liquid, not just its surface
properties.
A second feature of surfactants is that they can cause immiscible liquids
to mix. Here one considers surfactant amphiphiles one of whose parts is
soluble only in the first liquid and the other of whose parts is soluble only
in the second liquid. Such surfactants are at equilibrium at the interface
between the two liquids. Adding more surfactants creates more interface.
With a sufficient amount of surfactant, this interface may grow so much that
any point in either fluid is only a microscopic distance from the interface.
Then the immiscible fluids are in effect mixed.
This chapter explores the special properties that amphiphilic molecules
impart to a liquid. As in previous chapters, we focus on the simplest level
of quantitative, predictive analysis. We identify the characteristic spatial
scales, energy scales, and time scales that make these fluids act the way they
do. We begin with a section on mixing; we recall the empirical rules that
describe the mixing of different molecules. The next section describes the
most common surfactant molecules, noting their resemblances and dif-
ferences. The following section examines how strong amphiphiles should
interact in a liquid, and why they should aggregate into micelles when the
174 Surfactants

concentration is sufficiently large. Then comes a section discussing how the

fluctuations of micelles lead to interaction and generate forces. The follow-
ing section consider how surfactants alter the properties of a fluid interface.
We will discuss how it is possible for surfactants to lower an interfacial
tension by orders of magnitude, and to induce spontaneous mixing of two
immiscible liquids, thus forming a microemulsion. A section on amphiphilic
polymers shows how the large size of polymers gives rise to scaling laws
that predict their micelle structure. Finally we touch on the way micel-
lar structures move and relax mechanically, and we mention the various
nonequilibrium structures that can result when surfactants are aggregated
in a fluid.

7.2 Mixing principles

The behavior of surfactants in solution hinges on the principles controlling
miscibility in liquids. In the previous chapter we saw that the free energy
of a fluid A changes when one adds a foreign molecule B at a specific
place. To be specific, we consider the work W required to bring molecule
B from a reference liquid of pure B to the chosen point in liquid A. Work
is required because the atomic environment of our B molecule changes
when it is transferred from a liquid of identical B molecules to a liquid of
foreign molecules A. If the work is negative, then liquids A and B must
spontaneously mix. Even if the work W is positive, the liquids still mix to
some degree. If the A and B liquids are in contact, B molecules still are
present in liquid A at some nonzero concentration. The probability that our
B molecule resides at a specific point in the A liquid is exp(−W /T ) relative
to the corresponding probability in the native B liquid. If W is much greater
than T , then the B concentration in A is small, the Bs’ in the A solvent
have little effect on each other, and the work to bring a B into the A solvent
in equilibrium with B is nearly equal to W . Then the equilibrium volume
fraction of A in B is exp(−W /T ) if A and B displace the same volume. On
the other hand, if the work W is no more than about T , then equilibrium
the volume fraction of B in A is substantial and thus B mixes well with A.
Various aspects of the fluid give rise to the work W . One such effect is
steric: the A molecules must arrange themselves differently around our B
molecule than did the fellow B molecules in the reference liquid (cf. the dis-
cussion of Fig. 2.1 and Fig. 6.6). A second difference arises from electrical
interactions. The molecules generally have a distribution of charge and each
atom has an electric polarizability, as discussed in Chapter 5 under “induced
dipole interactions.” To treat these interactions accurately is a complicated
enterprise that requires knowledge of many specific properties of the A and
B molecules. Still, the qualitative mixing properties of two liquids can be
understood in simple terms, by presuming that the interactions, both steric
and electrical, are local in range, affecting mainly adjacent atoms.
The limiting case of this local point of view is to ignore difference in
charge at different parts of the surrounding molecules, and to consider only
the polarizability of the surrounding environment. This amounts to repla-
cing the actual A solvent by a fictitious atomic solvent, whose electric
polarizability matches that of the real A solvent. We treat the B solvent
 Mixing principles 175

in the same way. By this approach one may justify five principles that are
generally borne out empirically. (1) Positivity: the work required to trans-
fer our B molecule is generally positive and increases with the molecular
size. (2) Additivity: the work W may be found by adding contributions for
each atomic constituent of the B molecule being mixed. (3) Ordering: the
work is greater when the two liquids differ more in their static dielectric
constant and their optical index of refraction. (4) Reciprocity: if A and B
are of comparable size and if A is miscible in B, then B is miscible in A.
(5) Transitivity: if A, B, and C are of comparable size and A and B are
miscible in C, then A is miscible in B. We discuss each of these principles
briefly below. These important aspects of physical chemistry are discussed
much more carefully in Israelachvili [1] and Denbigh [2].

7.2.1 Positivity
In the discussion of van der Waals forces in Chapter 5 we found that two
identical polarizable B atoms have an attractive interaction in vacuum or in
any polarizable medium. Now, the van der Waals interaction perturbs each
atom only slightly; thus the net interaction of many atoms is nearly the
sum of pairwise interactions. This means that a B atom near a boundary
between B and A liquids feels an attraction towards the B liquid—i.e.,
the work W is positive. Now if a few atoms are joined to form A and B
molecules, the atomic polarizability of each atom is altered by its bonding
environment. Still, each atom of B feels a net attraction for its counterpart
on other B atoms even in an environment of A. Thus two B molecules feel
a net attraction, and again the work W is positive. The removal of our B
molecule to an A environment also affects the packing of the neighboring
molecules around our B molecule. But for the small organic molecules we
encounter in practice, these differences in packing are usually not dominant.
Furthermore, the group of atoms is itself a polarizable entity that must
feel a net attraction towards similar entities in an environment of different
polarizability. Our small B molecule will in general have a nonspherical
distribution of charge, though it has no net charge. At large distances, this
nonspherical distribution leads to a dipole field. This dipole may be oriented
by electric fields; in other words it leads to a molecular polarizability. This
polarizability creates van der Waals attraction just as atomic polarizability
does. Thus it adds another positive contribution to the work W .

7.2.2 Additivity
Many molecules, such as hydrocarbon chains, are too large to be considered
as small, fixed clusters of atoms. How can we understand the work W for
carrying a hydrocarbon chain B from a liquid of similar chains into some
other liquid A? As we have seen, this work results from the altered interac-
tions between each hydrocarbon group and its immediate surroundings in
the A medium compared with its original B surroundings. The interactions
are local. Thus it is reasonable to approximate W as the sum of contrib-
utions W0 that each hydrocarbon group would experience on being carried
from the B liquid to the A liquid. Adopting this point of view leads us to an
176 Surfactants

important conclusion, already encountered in Chapters 3–5. We have noted

that W0 is in general positive. Thus the work W required to transfer a large
B molecule grows with its size. Ultimately the work becomes substantially
larger than T and mixing becomes virtually nil: large molecules resist mix-
ing. This conclusion holds even when the A solvent is nearly identical to
B, so that W0 is very small. Still, for sufficiently large B molecules, the
total work W must become large. For example, polymers with a molecular
weight of several million can demix strongly even when the only difference
between them is the replacement of the hydrogen atoms with their isotope
deuterium [3].
Sometimes the shape of the B molecules prevents additivity from work-
ing. Additivity requires that the environment of all small segments of the
B molecule be the same, independent of the number of segments in the
molecule. When B is transferred to the A medium, each segment is pre-
sumed to be surrounded by A. If B has a globular shape so that most of the
atoms of B are surrounded by other atoms of the same B molecule, additiv-
ity clearly is not valid. Indeed, the B molecule may adopt a globular shape
when inserted in the A medium because this globular shape reduces W .
Such changes of shape generally require work of at least T in themselves,
as we saw in Problem 3.15. Thus this work alone is enough to inhibit mixing
substantially.

7.2.3 Ordering: like dissolves like

Returning to our atomic A and B liquids, we see that the van der Waals
attraction must vanish when A and B become identical in atomic polarizab-
ility. Thus to mix well, molecules should have similar polarizability. Thus
B molecules with large dipole moments mix well in A solvents with large
dipole moments, such as water. Likewise hydrocarbon molecules with vir-
tually no dipole moment, mix well with hydrocarbon solvents. These dipole
moments control the static polarizability that produces the static dielectric
constant of the liquid . However, the molecular dipole moment is not the
only source of polarizability. Polarizability is the induced dipole produced
by an applied electric field of arbitrary temporal frequency. Polarizabil-
ity at any frequency gives rise to van der Waals attraction. Thus liquids
of identical static dielectric constant  may nevertheless be immiscible,
because they differ in polarizability at nonzero frequencies. Another meas-
ure of polarizability at the optical frequencies that characterize electronic
motion is the optical index of refraction. Thus two small-molecule liquids
with similar static dielectric constants and similar indices of refraction are
typically readily miscible.

7.2.4 Reciprocity
For atomic liquids of the same atomic size, only the contrast in polarizability
leads to attraction and immiscibility. If A and B atoms differ in polarizability,
this limits the miscibility of A atoms in B and B atoms in A to a comparable
degree. The same is expected when A and B are small molecules. This
reciprocity becomes less valid as the contrast between A and B increases.
 Mixing principles 177

Reciprocity breaks down completely if for example, B is a large molecule

and A is a small one, for then B is generally immiscible in A even though
A may be miscible in B.

7.2.5 Transitivity
The principle of transitivity says that two molecules that are miscible in a
third should be miscible in each other. This is another consequence of the
like-dissolves-like principle. We have argued that small-molecule solvents
that are similar in dielectric constant and in index of refraction should be
miscible in each other. If these properties are similar for solvents A and
C, and for B and C, they must be so for A and B. Thus A and B should
be miscible. If the molecules differ greatly in size, the reasoning breaks
down, and the transitivity property need not hold. The most obvious counter
example is the case where A and B are large hydrocarbon polymers and C is
a small hydrocarbon molecule.

7.2.6 Effect of permanent dipoles: water

The principles stated above treated the solvent as a weakly polarizable
A medium like a monatomic liquid. Then the effect of two different B
molecules is approximately the sum of their individual effects. This view
of the medium works badly in the case of a strongly polarizable solvent
like water. For water, the dielectric constant  is about 80. Thus the dipole
moments of the water molecules respond so strongly to an external field
that they reduce it by a factor of 80. Each water dipole lowers its energy
by favorable interactions with nearby dipoles; it loses part of this energy
if adjacent water molecules are replaced by nonpolarizable B molecules.
This mutual alignment means that water molecules orient in concert much
more than most solvents do. This alignment gives rise to longer range
interactions between B molecules than in simpler solvents. The unusual
properties of water as a solvent are discussed at length in Israelachvili’s
book [1].

7.2.7 Effect of charges: ionic separation

Our discussion up to now has considered only neutral A and B molecules.
But sometimes B molecules can dissociate in the A medium to form ions.
This ionization decreases the work W and thus promotes mixing. Now when
the B molecule is brought into the A medium, with one atom at a fixed posi-
tion, an ion is liberated. To find the Boltzmann weight exp(−W /T ) we must
now add a contribution for each possible position of this ion, using the defin-
ition of free energy from Chapter 2. Accounting for these possible positions
multiplies the Boltzmann weight by a factor proportional to the volume per
ion V . This amounts to a decrease in W by an amount T log V /V0 , where V0
is the volume per ion in the original B medium. (The log V /V0 is conven-
tionally called the translational entropy.) The accessible volume of the ion
is evidently much larger in the ionized state of the A medium, as compared
178 Surfactants

with the un-ionized state of the native B medium. Thus the ionization gives
a negative contribution to W and promotes mixing.
These benefits of ionization do not come without a cost: namely, the
electrostatic energy of separating each ion from its oppositely charged part-
ner(s) in the native B medium. The electrostatic cost must be smaller than
about T . Otherwise, there would be a net positive cost to ionization even
considering the entropy effect above. To avoid this cost, the mutual elec-
trostatic energy of the two opposite ions must be made small. This energy
U is reduced in proportion to the reduction of the electric field of, say, the
negative B ion; it is smaller than in vacuum by a factor . Thus large 
means small U and small electrostatic cost of separating. A useful measure
of this electrostatic energy is the Bjerrum length , defined in Chapter 3. It
is defined so that the mutual energy of two singly charged point ions in the
liquid at large separation r is given by U (r) = T (/r). Evidently  varies
as 1/. For water at room temperature  0.7 nm. Accordingly, the energy
cost to separate two ions from r = 0.7 nm to a large separation is just the
thermal energy T .
In water most small singly charged ions dissociate and mix readily. The
energy gained by bringing them to a distance of, say, 1 nm from infinity is
only a fraction of T , in accordance with the formula above. At this distance,
the organized layers of polarized water molecules around each ion begin to
interfere with each other. Bringing the ions closer disrupts these “solvation
shells” and costs free energy. The result is that little further net energy would
be gained by bringing the ions into contact. The total energy of separation is
of the order of T , and the ions may readily dissociate. For example, sodium
and chloride ions dissolve in water at volume fractions up to about 10%.
For volume fractions smaller than this, the gain in translational entropy is
more than enough to compensate for any loss in electrostatic and solvation
energy.
Evidently, the special properties of water are important in allowing ions
to dissociate. In most solvents dissociation is a minor effect. Even in
water dissociation is difficult for ions bearing more than one charge. Two
divalent ions must be further than 2.8 nm apart in water in order to have an
electrostatic energy smaller than T .

7.3 Surfactant molecules

As stated above, surfactant molecules have one part that is miscible in
the solvent and one that is immiscible in it. For water solutions, the most
common surfactants have ion pairs, like sodium sulfate, as their miscible
parts. In water these singly charged sodium ions readily dissociate, so that
such ions are strongly miscible. The immiscible part is a common molecule
with small polarizability: a hydrocarbon chain. This chain increases the free
energy of mixing W . Still, if the chain is short, the combined molecule is still
readily miscible in water. Adding more CH2 groups to the chain decreases
the miscibility (according to the additivity principle above). Thus one finds
that chains with as many as 12 carbons attached to a sodium sulfate group
mix only at volume fractions below about 14 %. Thus this molecule, called
 Surfactant molecules 179
–
Br
H3C
O O N
+
S CH3 CH3
– H3C CH3
O O
+
Na

CH3
O
O
CH3
CH3 K+ O S O
– O CH3
O
O O
O O
H3C O O CH3
+
N P – O
H3C CH O O
3

H3C

H3C O O O OH
O O

Fig. 7.1
Five common surfactant molecules from [4]. Top row: SDS, the cationic surfactant cetyl trimethyl ammonium bromide (CTAB). Bottom,
the phospholipid 1-palmitoyl-2-oleoylphosphatidylcholine (POPC), sodium bis(2-ethylhexyl)sulfosuccinate (AOT), pentaethylene glycol
monododecyl ether (C12 E5).

sodium dodecyl sulfate or SDS, is much less miscible than its miscible part.
Figure 7.1 shows a diagram of it.
SDS, also called sodium lauryl sulfate, is the leading ingredient in house-
hold cleaning products, like soap, detergent, and shampoo. Many other
surfactants are variations on this theme. The hydrocarbon part, called the
tail, may be lengthened by four or more carbons. The polar sulfate head
group SO− −
4 can be replaced by carboxylate, CO3 , or by phosphate, PO4 .
−

Finally, some head groups may have two hydrocarbon tails attached. As we
shall see below, the significant differences in these molecules lie not only
in the solubility of the different parts but also in the relative bulk of the head
and tails and in the extensibility of the tails.
The polar head is most commonly a negative ion like the ones above.
Then the molecule is called an anionic surfactant. Sometimes one chooses
180 Surfactants

a cationic group such as ammonium, NH+ 4 as the polar head as in the CTAB
molecule shown in Fig. 7.1. Naturally, cationic and anionic surfactants
have a strong mutual electrostatic attraction. The dissociating ion is most
commonly a singly charged metal ion like sodium. But this ion can be an
organic group as well. Likewise, the ions may be weakly dissociating. An
important example is an OH− H+ pair whose dissociation may be controlled
by the ambient H+ concentration or pH. Amine groups NH− are a second
important weakly dissociating group.
Some surfactants, called nonionic, have polar heads without dissoci-
ating ions. One example is a zwitter ion, in which an organic cation is
attached to an organic anion by e.g., a short hydrocarbon chain. A second
important example is the ethylene glycol group, whose polar oxygens asso-
ciate strongly with surrounding water molecules and lead to a negative
contribution to the work of mixing W .
To summarize, surfactants vary in strength from weak to strong. Weak
surfactants have parts with little difference in miscibility. Strong surfact-
ants have parts that differ greatly. Beyond this difference, surfactants
differ in the relative bulk and deformability of their polar and nonpo-
lar parts. Further, the polar parts differ in being anionic, cationic, or
nonionic.

7.4 Surfactants in solution: micelles

Surfactants, like any molecules, must mix in any host liquid A at some
nonzero concentration. If one mixes a small enough amount of surfactant in
the liquid, the surfactant molecules must disperse through the liquid. But if
one increases the amount, a point comes where this molecular dispersion no
longer works. Beyond this point some of the B molecules typically demix
into a separate B-rich phase. However, surfactant B molecules can find
another way to reduce the work W of mixing. This work W has two large
and opposite contributions. For definiteness we suppose that the solvent
is a polar one. Then the polar part of the surfactant contributes a negat-
ive W , but the nonpolar tails contribute a larger positive part. In a native
surfactant B phase, the mutually miscible nonpolar tails are adjacent to each
other. The positive part of W arises from the cost of replacing this miscible
environment with polar A solvent.
This cost can be avoided if the surfactants can be placed in the A solvent
without the tails leaving their environment of other tails. Thus a number of
surfactants tend to enter the A liquid together, with their tails adjacent to one
another. The resulting aggregated structure or micelle can take forms like
those shown in Fig. 7.2. If the surfactants enter the A solvent in the form of
micelles, the net cost Wm can well be negative. For example, SDS forms
spherical micelles which then mix with water at volume fractions that can
exceed 10%. At higher concentrations the micelles’ repulsive interaction
forces other forms of aggregation. This is much higher than the 10−4 volume
fraction that dispersed SDS achieves.
Any micelles present in the A solvent must be in equilibrium with
dispersed surfactant molecules. For the case of spherical micelles with
 Surfactants in solution: micelles 181

Fig. 7.2
Three types of micelles. Left to right,
spherical, cylindrical, and bilayer.

K surfactants B, this equilibrium is expressed by the chemical reaction

KB ←→ BK .

This equilibrium dictates a condition on how the concentration of indi-

vidual surfactants, [B] may depend on that of micelles [BK ]. (Here we used
the convenient notation where [X] means “number of species X per unit
volume.”) This law of mass action, derived in all physical chemistry texts [2]
is [B]K /[BK ] = constant. The constant is independent of concentration.
The well known outcome of this constraint is that at low concentration the
dispersed form dominates while at high concentration the aggregated form
dominates, as explained below.
In practice, the aggregation number K of micelles is sizeable: for
example, in SDS it is about 64 [5]. When K is large, one can understand the
equilibrium in a simpler way, using the work W defined above. If micelles
are present, one may define the work W to carry a surfactant from a micelle
to a given place in the A solvent. Then by our reasoning above, the volume
fraction of dispersed surfactant should be given by exp(−W /T ). We are led
to the conclusion that the micelles are in equilibrium with a fixed volume
fraction of dispersed surfactant. Since adding surfactants cannot increase
the dispersed concentration above this amount, all the added surfactants
must be in the form of micelles. Likewise, removing surfactants reduces the
number of micelles without changing the dispersed number. This continues
until there is no excess surfactant available to form micelles. This threshold
concentration of surfactants is known as the critical micelle concentration,
or CMC for short.
This fixed CMC is consistent with the conventional chemical equilibrium
picture mentioned above for large aggregation number K. If most of the
surfactant is in the form of micelles so that the total concentration C
K[BK ], then [B] = C − K[BK ]  C and we can write the equilibrium
formula as
[B]/(C − [B])1/K = constant
Neglecting the small [B] part of the denominator, we conclude that

[B] constant × C 1/K

For large aggregation number K, [B] grows very slowly with C—[B] is
practically constant.
What determines the aggregation number K? Evidently it has something
to do with the shape and deformability of the surfactant molecules. For
example, an aggregate of SDS with K = 2 would not be very favorable,
182 Surfactants

since the two tails could not avoid contact with the solvent. Thus nearly
as much work would be required to put the aggregate into the solvent as
to put the dispersed molecules there. But a spherical micelle with tens of
molecules can readily shield the tails from the solvent. Beyond this point
there is little to be gained by increasing the aggregation number. In addition,
there is something to lose: the charged head groups repel each other, and
enlarging the aggregation number clearly increases this repulsion.
7.1. Law of mass action Molecules A, and B in a dilutesolution can form a com-
bined species AB. The work required to separate an AB into its constituents
A and B at two given positions far apart is W . Show that in equilibrium the
concentrations [A], [B], and [AB] are related by

[A][B]/[AB] = constant e−W /T .

7.4.1 Open aggregation: wormlike micelles

Some surfactants are constructed such that they profit from aggregation
beyond that of small spheres. A spherical shape is no longer feasible beyond
a diameter of about twice the tail length, for the tails could not reach the
center if the sphere were larger than this. However, there are two other
shapes that permit arbitrarily large K: one-dimensional wormlike micelles
and two-dimensional bilayers. We consider each in turn.
In a solution of wormlike micelles, the aggregation number K may vary
over a wide range to form shorter or longer worms. Each of these species
must be in equilibrium with the others. These constraints control the relative
number of micelles of given size, and they control how the average size
depends on the overall concentration C, as we now show.
Wormlike micelles, like spherical micelles, have a minimal aggregation
number K0 , below which aggregation is not worth its energetic cost. Any
micelle of length aggregation number K > K0 may give free surfactant
molecules to the solvent, or may take them, with some cost or benefit
in free energy. That is, there is an equilibrium between the free surfactants
(micelles with K = 1) the K micelles and the K + 1-micelles: (1) +(K) ↔
(K + 1). The law of mass action mentioned above describes the resulting
relationship between the concentration of free surfactant molecules [1] and
the concentrations of the micelles [K] and [K + 1]:

[K][1]/[K + 1] = 1/V0 exp(−W /T ), (7.1)

where W is the work to carry a surfactant from the K + 1 micelle to the

solution and V0 is the reference volume per surfactant encountered above.
For long, wormlike micelles with K  K0 the work to remove a surfact-
ant is sometimes entirely due to local structure in the micelle and the solvent
adjacent to the molecule being removed. This is the case if the micelles are
dilute, rigid rods, or flexible chains without self interaction. Then the work
W is independent of K. In this case determining the concentrations [K] in
terms of the total concentration C is straightforward. First, we note that the
right hand side of Eq. (7.1) has dimensions of concentration. We denote it
 Surfactants in solution: micelles 183

as C0 . Then the equation can be written

[K + 1]/[K] = [1]/C0 .

The equation describes a geometric falloff with increasing micelle size:

[K] = C0 ([1]/C0 )K . For very small free surfactant concentration [1] the
right side is much smaller than 1. Thus the K concentrations fall off rapidly
with increasing K. But as [1] approaches C0 the falloff becomes slower
and slower, so that large micelles become more abundant. The surfactant
concentration in the form of K-micelles is K[K], so that the overall con-
centration C is K K[K]. It is clear that the total concentration C becomes
larger as the free surfactant concentration [1] increases. Indeed, if [1] = C0 ,
the concentration of K-micelles no longer decreases with increasing K, and
C becomes infinite. Since C must remain finite, [1] must always remain
smaller than C0 . This C0 plays a role similar to that of the critical micelle
concentration for spherical micelle-formers. We shall call C0 the limiting
free surfactant concentration or limiting monomer concentration.
To describe
 how [K] depends on C explicitly, it is convenient to define
f (x) ≡ ∞ x k . The series sums to f (x) = 1/(1 − x). Its derivative

k=0
f  (x) = x −1 ∞ k=0 kx = 1/(1 − x) . We may express the concentration
k 2

C in terms of f . Under the conditions where long worms are prevalent,
aggregates smaller than K0 are negligible and we can write
† We have seen that C only affects the con-
∞
 ∞
 centrations [K] via a scale factor. That is,
C KC0 ([1]/C0 ) = C0 x
K K0
Kx K−K0
, (7.2) [K] = p(C)g(K/q(C)). When q(C)  1,
K=K0 K=K0 the dominant [K]’s must occur for K  1.
Thus
where x ≡ [1]/C0 . Switching to a new sum index K  ≡ K − K0 , this  
amounts to K n [K] dKK n p(C)g(K/q(C)).
K
∞
  Defining the variable K̃ ≡ K/q(C), this
C C0 x K0 (K  + K0 )x K = C0 x K0 (xf  (x) + K0 f (x)). gives
K = 0
 
K n [K] q(C)n+1 p(C) d K̃ K̃ n g(K̃).
We focus on the regime where large micelles are dominant, so that δ ≡ K
1 − x  1. Then f (x) → δ −1 , f  (x) → δ −2 , and C → C0 /δ 2 . In the
same way we can find the average micelle size K ≡ K K[K]/ K [K]. The integral is now a constant, independent
of C. (The constant is finite, since our dis-
The numerator is C, while the denominator is C0 f (x) → C0 /δ. Combining,
tributions g(x) fall off faster than any power
K → 1/δ. for large x.) Now, any moment ratio Mmn
Combining this expression for K with the one for C above, we conclude of the [K] distribution has the form
that  
K → (C/C0 )1/2 . (7.3) Mmn = K n [K]/ K m [K].
K K

In terms of the concentration of micelles Cp ≡ C/K , this says K → Our scaling law for the sums shows that
Cp /C0 . Thus the micelle size can be controlled and predicted over a wide Mmn q(C)n−m . That is, all moments
range of concentration based on fundamental principles. behave as though they had dimensions of
This prediction of how concentration affects micelle size can readily be q(C)n−m . The averages discussed in the text
tested by scattering experiments. For example, are moments of this form: K = M0,1
 static scattering
 measures
and K w = M1,2 . They both grow with
the weight-averaged micelle size K w ≡ K 2 [K]/( K[K]). All such the same power of q(C), so that their ratio
averages scale with the same 12 power of overall concentration C † . Thus remains finite as either goes to infinity.
184 Surfactants

one may verify the square root law by measuring scattering intensity versus
concentration.
If the limiting monomer concentration C0 is small, we can create solu-
tions with very long worms. As these grow longer, they curve more and
more, eventually taking the form of random walks like the polymers stud-
ied in Chapters 3 and 4. In our study of these polymers we learned that
the interaction of the chain with itself affects its size in a qualitative way.
Repulsive interactions lead to the fractal properties of self-avoiding walks
rather than those of simple random walks. Repulsive interactions between
parts of a wormlike micelle must ultimately have the same effect. In par-
ticular, the work W needed to add a monomer chain now depends on the
chain size K. Expressed in terms of the equilibrium constant for K-mers
assembled from monomers, the K dependence has the form

[K] 1
= K−1 K γ −1 (7.4)
[1]K C̃0

The K γ −1 factor arises from the self repulsion, and the constant C̃0 plays
the role of C0 . Here γ is an universal exponent like the fractal dimension
D characteristic of any self-repelling polymer [6]. The value of γ is about
1.17 [6]. This factor alters the dependence of K on concentration, as we
may deduce by repeating the reasoning above. We first note that K ∼ 1/δ
as before. However, the overall concentration C has a different scaling. For
large K we may follow the scaling argument of the footnote to infer

C → constant δ −1−γ

Combining, we find a new dependence of K on C:

K → constant C 1/(1+γ ) . (7.5)

Now K varies more weakly with concentration: the power is reduced

by about eight percent owing to the repulsion. Micelle interaction also
influences K , as discussed below.
Wormlike micelles have internal degrees of freedom, as the above dis-
cussion implies. A given section of the micelle bends due to thermal
fluctuations. Longer sections bend through larger angles. Thus there is some
micelle length such that the mean squared bending angle is of order unity—
e.g., one radian. This length is called a persistence length and denoted p .
We encountered the notion of persistence length in the discussion of charged
polymers in Chapter 3. The persistence length has been measured by two
methods in cetyl pyridinium bromide, a close cousin of CTAB in water
with 0.8 M sodium bromide salt added [7]. At a concentration of 6 mM
surfactant and a temperature of 35◦ C, the persistence length was found to
be about 20 nm. The measured cylinder radius of the micelles was 2 nm.
In addition to this prosaic type of internal freedom, wormlike micelles can
branch. That is, three sections of wormlike micelle can join together. The
junction region requires distortion from the normal cylindrical arrangement
of the surfactants; thus the energy of a surfactant in the junction region is
 Surfactants in solution: micelles 185

larger than it is elsewhere in the micelle. Nevertheless, there is a nonzero

fraction of junctions and branches in equilibrium with the ordinary micelles.
In favorable cases this fraction is large enough to be observable. When the
micelles are above their overlap concentration, the branches act to crosslink
different micelles to form a network structure. Such structures have been
observed and quantitatively explained [8]. Certain surfactants prefer the
type of packing found at a branch junction. These naturally form branches
more easily.

7.4.2 Open aggregation: two-dimensional micelles

Sometimes surfactants have a preferred aggregation in the form of bilayers
of surfactant. That is, the surfactants form micelles in the form of two-
dimensional surfaces, as shown in Fig. 7.2. In principle we can extend
the same reasoning used above to predict how the micelle mass and size
depend on concentration. But in practice this is less worthwhile than for the
wormlike micelles treated above. One reason is that the energy to form a
two-dimensional micelle is not so straightforward as for a one-dimensional
micelle. Now the role of the endcap energy is played by the perimeter or
boundary energy. To know this, we must know the shape of the micelle:
flat, bowl-like, or saddle-like. This additional complication makes two-
dimensional aggregation much different from one-dimensional aggregation.
Less can be predicted with generality. In any case, it appears that when two-
dimensional micelles become large, the resulting properties can be predicted
without knowing their size or size distribution in detail [9].

7.4.3 Aggregation kinetics

It takes time for micelles to form and to adjust to their environment.
Generally these adjustments require surfactant molecules to pass through
energetically costly configurations. This leads to barrier-dominated time
dependence of the kind discussed in the Chapter 2. One can detect the time
for these readjustments by a so-called T-jump experiment: one suddenly
gives the solution a small pulse of heat and then monitors a property that
reflects the state of the micelles, such as light scattering [5]. Typically, the
solution relaxes exponentially to its new state. One generally observes two
characteristic relaxation times. To illustrate these, let us imagine giving a
10◦ heat pulse to a solution of SDS above its critical micelle concentration.
The new temperature alters the free energies of the variously sized micelles.
Typically the mean aggregation number is reduced. Surfactants must leave
micelles on the average to attain the new equilibrium distribution of sizes.
This requires a time of the order of microseconds for small surfactants like
SDS [5]. This time is called τ1 . But this exchange of surfactants is not
sufficient to attain full equilibrium. The total number of micelles must in † Since virtually all of the surfactants live
general change as well. Typically on heating, the number of micelles must
in the micelles and their average size
increase by some factor larger than 1† . Thus new micelles must nucleate, is reduced, more micelles are required
and grow from aggregation number 1 to the preferred aggregation number to accommodate the given amount of
of about 64. The necessary alteration of the system to assemble so many surfactant.
186 Surfactants

molecules is great and the energy barrier is high. Thus the micelle relaxa-
tion time—denoted τ2 —is many times longer than the surfactant relaxation
time τ1 . For SDS in typical conditions it is on the scale of milliseconds.
It may be much longer in very pure solution where there are few impurity
molecules to aid the nucleation of micelles [5].
These two characteristic times tell only part of the story of the dynamics
of micelles. In the rheology section below we shall encounter further forms
of relaxation that become important for interacting micelles.

7.5 Micelle interaction

When micelles form, they are often in such a dilute state that their interaction
is negligible. But as their concentration increases, interactions must become
important. Spherical micelles interact like a colloidal suspension: as the
volume fraction of spheres rises above 10–20%, enhanced osmotic pressure
and viscosity become apparent. One-dimensional, wormlike micelles reach
interpenetrating, semidilute conditions, just like polymer solutions. Their
osmotic pressure, cooperative diffusion, and correlation lengths can be
understood by adapting the methods developed for polymers and discussed
in Chapter 4.
The semidilute environment gives rise to a new effect in wormlike
micelles not seen with ordinary polymers. It alters the distribution of worm
lengths [K]. As discussed in Chapter 4, the effects of self-repulsion extend
only to a correlation distance ξ , since the solution is uniform beyond such
distances. These local repulsions also influence the concentrations [K]. We
first consider micelles whose geometric size is ξ or less. Such micelles have
a monomer number no greater than some g. These “blob micelles” feel the
full self repulsion, so that [g] follows Eq. (7.4) above:

[g] = C̃0 ([1]/C̃0 )g g γ −1 . (7.6)

Typical micelles in this semidilute solution have sizes K  g. We may

infer their concentrations [K] by considering the equilibrium (K) + (g) ↔
(K + g). The work to assemble the K + g micelle from its K and g pieces
includes the work done against mutual repulsion. But this repulsion is only
effective within a distance ξ , which is independent of K. Thus the work
itself must be independent of K. For this reason we can find the equilibrium
constant using
[K + g] [g + g]
.
[K][g] [g][g]

The micelles on the right are subject to the effects of repulsion and thus
obey Eq. (7.6)
[g + g]
C̃0−1 2γ −1 g 1−γ .
[g][g]

The repulsion suppresses the connection of the two pieces by a factor g 1−γ .
 Micelle interaction 187

Repeating the reasoning used above for non-repelling micelles, we can

deduce [K] whenever K is an integer multiple of g:
[K] [K − g]
[K] = · · · [g] = ([2g]/[g])(K/g)−1 [g].
[K − g] [K − 2g]
Using Eq. (7.6), this can be written as,

[K] ([g]g 1−γ /C̃0 )(K/g)−1 [g]

= C̃0 g γ −1 (([g]g 1−γ /C̃0 )1/g )K = C̃0 g γ −1 x K . (7.7)

Here x is defined as the contents of the large (. . .). The repulsion has added
a new feature: the g-dependent prefactor. Again [K] falls off exponentially
with the controlling parameter x, as it did without repulsion. The only
change is to replace C0 by C̃0 g γ −1 . This [K] varies smoothly with x, so
it is approximately valid even when K is not an integer multiple of g.
Repeating the reasoning after Eq. (7.2), we find

K → (Cg 1−γ /C̃0 )1/2 .

This K has a new concentration dependence, since g depends on concen-

tration. Since g is essentially the micelle mass whose geometric size is ξ ,
we may use the semidilute polymer in good solvent from Chapter 4 to infer
g ∼ ξ D ∼ C D/(D−3) . Thus [7]

K ∼ C y , where y = 12 [1 + (γ − 1)D/(3 − D)] 0.6. (7.8)

The self-interaction has changed the growth exponent from 0.5 to 0.6. This
subtle change in how micelles grow with concentration will prove important
when we consider the viscosity of these solutions below.
7.2. Worm interpenetration As noted above, the growth of wormlike micelles with
concentration leads to interpenetration.
(a) How does the overlap ratio φ/φ ∗ increase with concentration, assuming
(i) self-avoidance effects are negligible so that D = 2, and (ii) self-
avoidance effects are important, so that D = 53 ?
(b) Suppose that the cylindrical micelles have a radius of 2 nm and that each
surfactant occupies a volume of 0.4 nm3 . Suppose that the persistence
length is 20 nm, so that the persistent segments are long and slender. For
significant entanglement effects to occur, one should have φ/φ ∗ > 10.
But if the volume fraction of worms exceeds 10–20%, their packing leads
to nematic order as treated in Problem 5.10. Then the resemblance to a
polymer liquid diminishes. If nematic order to be avoided, φ < 0.1, so
that φ ∗ must be less than 0.01. Find the condition on the characteristic
concentrations C0 or C̃0 that permits this much entanglement. What is the
corresponding volume fraction of free monomers?
Wormlike micelles often have substantial rigidity, as noted above: they
remain straight over a length p that is several times their width. Thus at
sufficiently high concentration, the micelles may align to form a nematic
liquid crystal of the kind discussed at the end of Chapter 5. To understand
what concentration is necessary, it is useful to think of the micelles as being
188 Surfactants
(a) (b) (c)

Fig. 7.3
Sketch of three periodic bicontinuous
micellar structures discovered in
concentrated surfactant solutions, courtesy
John Seddon [12] with permission from
Elsevier. (c) is known as the plumber’s
nightmare [11]. a a a

cut up into straight rods of length p . Then one may adopt the reasoning
used in Chapter 5 section on organized states.
Surfactants whose shape favors branch junctions can at sufficiently high
concentration form structures that are branched everywhere. Such struc-
tures resemble a regular lattice of connected cylinders. The simplest such
structure is a simple cubic lattice resembling a tangle of pipes, known as
the plumber’s nightmare, as shown in Fig. 7.3. For example, 60 volume
percent glycerol monoolein (a C18 hydrocarbon chain with polar OH and
C== O groups at one end) in water is believed to form this phase [10]. Many
similar periodic network structures have been discovered. These are called
Luzzati phases [11].

7.5.1 Energy of two-dimensional micelles

As we have seen, surfactants often aggregate in the form of bilayers, so that
the micelles are two-dimensional surfaces. To understand their form and
their interactions, we must understand their fluctuations. Like random-walk
polymers, surfactant interfaces have important thermal fluctuations. The
first step in accounting for these fluctuations is to understand the energy cost
of deforming the layer. First we consider the energy of stretching the layer
and changing its area. For ordinary interfaces, this is the dominant form of
energy, as we saw in Chapter 6. But this stretching energy is unimportant for
the micelles we are considering here. Since the micelles are in equilibrium
with excess surfactant, the number of surfactants per unit area saturates to
a value that is set by the solution conditions (a typical density for common
surfactants like SDS is 0.25 nm2 per polar head). At such high densities, the
R1 energy to alter the surface density appreciably is large, and it is reasonable
to suppose that the amount of surfactant per unit area is fixed. Thus the only
important fluctuations are those involving curvature of the surface.
Our bilayer consists of two monolayers or leaflets. It will be useful to
consider each leaflet individually. At each point on a smooth surface there
are two perpendicular directions in which a straight line on the surface makes
R2
a planar arc in space. Their radii of curvature R1 and R2 orientation define
the state of curvature at that point, as shown in Fig. 7.4. The principal
Fig. 7.4 curvatures C1 and C2 are defined as the inverses of these radii. Clearly
A small section of a curved surface,
showing the two principal radii of
zero curvatures mean a flat surface. It is convenient to define two scalar
curvature R1 and R2 . The two principal quantities that specify the two curvatures. The mean curvature is defined
curvatures have opposite signs, so that the as 12 (C1 +C2 ), and we shall denote it by c. The Gaussian curvature is C1 C2 .
Gaussian curvature is negative. We note that these two curvatures have different dimensions.
 Micelle interaction 189

We may describe how the surface deforms in thermal equilibrium once

we know the energy. Nonzero curvature costs energy. If the curvature is
gentle—so that R1 and R2 are much smaller than the molecular thickness
of the layer—then we can express the energy as a Taylor expansion:

E= ds[AC1 + BC2 + DC12 + F C22 + κ̄C1 C2 + · · · ].

We can simplify this general form if the surface is isotropic:


E= ds[2κ[c − c0 ]2 + κ̄C1 C2 ]. (7.9)

Here κ, κ̄, and c0 are material constants of the interface. The constant κ is
called the bending modulus or bending stiffness. The κ energy is minimal
when the mean curvature has the value c0 ; this c0 is known as the spon-
taneous curvature. The term in κ̄ is not relevant for our purposes. Thermal
fluctuations do not affect this energy. Thereason is a geometric fact called
the Gauss–Bonnet Theorem. It says that dsC1 C2 over any smooth sur- R⬘

face with a fixed boundary is unaffected by continuous deformation of the S
surface. That is, the integral is a topological invariant. If the deformation R
increases the Gaussian curvature in one place, it must decrease it in another
so as to compensate [12]. Thus compensation is illustrated in Fig. 7.5 and
discussed in the Appendix.

Knowing the energetic cost of bending, we can now analyze the fluc-
tuating shapes that occur in equilibrium. First we consider sections of the
surface that are small enough to be nearly flat. This must be true for small Fig. 7.5
enough sections; otherwise we would not be able to characterize our micelle Bump construction on a flat surface used
as a smooth, two-dimensional surface. For such a nearly flat surface, there to demonstrate the Gauss–Bonnet theorem.
is a convenient way to describe the fluctuating shapes. We represent the A flat surface was deformed to make the
circular bump shown, with positive
surface by giving its local height h above some reference plane (x, y), that Gaussian curvature at the middle. In order
makes only a small angle with the surface everywhere. The smallness of to rejoin the flat surface smoothly, the
these angles means that ∂h/∂x and ∂h/∂y are much smaller than unity. This surface must have negative Gaussian
h(x, y) scheme is known as the Monge representation. curvature at the rim. The Appendix shows
The energy of Eq. (7.9) can be readily expressed in the Monge rep- that the Gaussian curvature averages
to zero.
resentation. Indeed, the mean curvature is simply 12 ∇ 2 h, so that E =
1
 2 2†
2 dx dy κ|∇ h| . Noting that the energy density is invariant under trans-
†This is the energy we encountered in
lation, we may simplify by expressing h as a sum of plane waves. We define Problem 5.4, with ψ corresponding to h.
h̃(q) via

h(x, y) = L−2 h̃(q)eiq·r . (7.10)
q

The wave vectors q are a complete set of waves needed to represent functions
on this square region. Then

1 −2  2 2
E= κL (q ) |h̃q |2 . (7.11)
2 q
190 Surfactants

The energy is a sum of normal plane-wave modes. Each fluctuates independ-

ently. Because their energy is quadratic, each has an average energy of 12 T by
virtue of the equipartition theorem of Chapter 2. From this fact, we can infer
the average squared amplitude of the surface: |h̃q |2 = T L2 /(κ(q 2 )2 ).
Using the behavior of |h̃q |2 we can understand how the typical heights
grow with lateral size of the surface. We consider a square of micellar
surface of length L. Using Eq. (7.10), the mean square height relative to its
average is given by

     1 2
T
h2 L = L−4 |h̃q |2 = L−2 .
q
κ q
q2

The summand falls off rapidly with

 increasing q; evidently it is dominated
by the lowest q’s in the sum: q 1/(q 2 )2 1/qmin 4 , where q
min is the
smallest nonzero q in the sum. For our patch of size L, this qmin is 2π/L.
Thus
h2 L (T /κ)L2 . (7.12)

The height of the patch is scales linearly with its width.

A second important consequence is to see how the orientation of the
surface fluctuates on the patch. The normal  to the surface has a vertical
component given by (1 − |∇h|2 )1/2 = (1 − q q 2 |hq |2 )1/2 . The typical

slope μ of the surface is given by μ2 = |∇h|  2 = L−4 q q 2 |h̃(q)|2 .
Since the thermal average of |h̃(q)|2 falls off as (q 2 )−2 , we infer

 T −2  −2
μ2 = L−4 q 2 |h̃(q)|2 = L q .
q
κ q

This sum falls off 

only slowly with increasing q; it can be approximated as
an integral, using q → (L/2π )2 d 2 q:

 2   
T L qmax d 2q T qmax
μ = L−2
2
log . (7.13)
κ 2π qmin q2 κ qmin

We recall that qmin is of order L−1 . As for qmax , 1/qmax is the smal-
lest wavelength at which significant fluctuations occur. This smallest
wavelength is roughly the thickness b of the membrane. Waves with length
shorter than the thickness are not adequately treated by the thin-membrane
formula (7.9). Such waves cost more energy than this formula suggests,
and their effect can be neglected. We conclude that the typical local slope
of a patch of surface of size L is of order (T /κ log(L/b))1/2 —it increases
logarithmically with the patch size. If the patch size is small, the slope is
smaller than unity. (Otherwise there would be no scale on which the surface
was smooth.) As one enlarges the patch, there comes point, L ≡ p where
 Micelle interaction 191

the typical slope becomes unity. For patches larger than this, our Monge
approximation breaks down. Still, we expect the slopes to continue to grow
with L. This means that for patches much larger than p , two arbitrarily
chosen points on the membrane will typically have completely independ-
ent orientations. Thus the orientations become uncorrelated over distances
longer than p . Thus p represents a persistence length like the persistence
length encountered above for wormlike micelles. The orientation at a given
point persists out to distances of order p . Since 1 = T /κ log(p /b), we
conclude
p b exp(κ/T ). (7.14)

The persistence length increases exponentially if we increase the bending

stiffness or reduce the temperature. The significance of the persistence
length was first discussed by DeGennes and Taupin in the 1980s [13].

7.5.2 Energy to confine a fluctuating membrane

These membrane fluctuations have an energetic effect as well as the struc-
tural effects described above. Often these surfactant membranes are spaced
closely enough so that they interfere with each other. In such situations a
given membrane cannot fluctuate freely without hitting other membranes.
The fluctuations are increasingly suppressed as the density of membranes
grows. This suppression requires work. Thus the fluctuations induce a
repulsive interaction between the membranes.
To understand the interaction quantitatively, we consider the simplified
case of a single large fluctuating membrane confined between two paral-
lel walls spaced by a distance d much smaller than the persistence length
p discussed above. Naturally constraints from the walls occur where the
membrane comes in contact with a wall. At such a point, the membrane is
obliged to curve away from the wall rather than fluctuating freely. A finite
fraction of the possible states of curvature are excluded at the point of con-
tact. The probability that the unconstrained system would happen to satisfy 2d
this constraint is thus some finite fraction. As we have often seen above, x
imposing such a constraint requires work of order T .
In a large membrane there are many points of contact, each of which Fig. 7.6
Section of a fluctuating membrane trapped
carries a free energy cost of order T . To find the net work needed to bring
between two walls. The spacing d and
the walls to the distance d, we need to know the number of contacts per the contact separation x are as defined in
unit area. Figure 7.6 illustrates the situation. Let us consider a patch of size the text.
L around some contact point whose width is a few times the thickness b
of the membrane. As we saw above, typical places on the patch have a
slope μ of order (T /κ log(L/b))1/2† . Since the patch size is much smaller † To see this more simply, we note that the
than the persistence length p , this slope is much smaller than unity. As we energy of our patch of width L is roughly
saw above, the typical slope depends only weakly on the size of the patch. κL2 c2 , where c is the mean curvature. Not-
Given such a slope, we must anticipate that in a distance x of order d/μ the ing that this energy is of order T , we infer
that (cL)2 (T /κ). But cL is just the
membrane will encounter the opposite wall, leading to another free energy difference in slope accumulated as one tra-
penalty of order T . The density of such contacts is roughly x −2 , or μ2 /d 2 . verses the patch. Thus the typical slope
This leads us to expect an energy per unit area of order T (T /κ)d −2 . One μ near the patch is of order (T /κ)1/2 , as
can calculate this energy explicitly using the Monge representation. The claimed.
192 Surfactants

energy E per membrane of area L2 is given by [14]

E = (3/π 2 ) (T 2 /κ) (L/d)2 . (7.15)

Interactions between adjacent fluctuating membranes evidently have the

same functional form. The energy falls off as the inverse square of the
distance, like the van der Waals interaction between surfaces. This universal
repulsion caused by fluctuating membranes was first identified by Wolfgang
Helfrich [14].
The Helfrich repulsion evidently becomes important when two-
dimensional micelles become concentrated so that the distance between
Fig. 7.7 them is smaller than about the persistence length p . If the micelles are
A micellar solution swollen with water. charged, this adds a further source of repulsion. At such concentrations the
Decreasing the curvature of the droplets micelles align into parallel layers or lamellae in order to avoid each other.
with fixed total area necessarily decreases The mutually oriented lamellae give an overall anisotropy to the fluid: it
the number of droplets and increases the
enclosed volume. Here the mean radius has
becomes a smectic liquid crystal. We encountered such liquid crystals in
!1/2 Chapter 5. This type of order shows up readily in scattering: the periodic
increased by 43 , so that the area per
droplet has increased by a factor 43 . Thus alternation of solvent and surfactant in space creates diffraction peaks like
the surface area in the original four those in Fig. 2.7. Knowledge of the power-law form of the Helfrich repulsion
droplets is only enough to cover three. The allows one to understand the distinctive power-law falloff of the scattering
volume of water has changed by a factor intensity away from the peak. This is explained in the paper from which
! !
4 3/2
!1/2
4 × 3
3
, i.e., 43 .
Fig. 2.7 is taken.

7.6 Mixing immiscible liquids: microemulsions

Up till now we have studied the behavior of surfactants in a single solvent
species A such as oil. In this section we ask what happens when we add
a second immiscible species B such as water. Before adding any B, our
surfactants are assumed to be well above the critical micelle concentration.
We now specify the nature of the micelles. We shall suppose that the micelles
are spherical and that the surfactants pack together with a fixed area per
surfactant. Thus in adding a given amount of surfactant, we have added a
specific amount of interfacial area, . It will prove convenient to measure
the amount of surfactant in terms of this area. We anticipate that adding
solvent B may cause some B to enter the micelles. A micelle acquiring B
solvent will grow in size and thus its (mean) curvature c will decrease, as
indicated in Fig. 7.7. We suppose that the energy of the surfactant layer
is described by a bending energy κ and a spontaneous curvature c0 , as in
the last section. E = 2κ ds(c − c0 )2 . The curvature ci of the micelles is
generally bigger than c0 ; we shall suppose that it is much bigger.
We now add a drop of B solvent and ask what happens. If the surfactant
were absent, this droplet would keep its integrity and not mix (appreciably)
with A. But with the micelles present, things are different. Let us transfer a
volume V of B solvent into the centers of the micelles, and note the work
†
required. The B solvent is now in the form of spheres surrounded by a
We shall suppose that the volume V is
surfactant layer. Since the area per surfactant is fixed, the total area of these
large enough so that R is much larger than
the original micelle size. Then the thickness layers remains at its initial value . Thus the typical radius R of the water
of the surfactant layer itself is negligible spheres is given by V / = 43 π R 3 /(4π R 2 ) = R/3. The curvature c of the
compared to R. surfactant layer has now decreased to 1/R † .
 Mixing immiscible liquids: microemulsions 193

7.3. Size variability of micelles In our example we supposed that the solvent
containing micelles all had the same radius R. To examine this supposition,
consider the energetic cost of changing this radius. Assume that the micelle
radius R = 1/c0 . For a given bending stiffness κ, how much may R decrease
from 1/c0 before the bending energy Eκ increases by T ?

Our surfactant interface has reduced its curvature towards the spontan-
eous curvature c0 . This has reduced the energy of our system. Eκ (V ) =
2κ(/(3V ) − c0 )2 . As long as the micelles have a radius much smal-
ler than 1/c0 , this energy decreases as the amount V of included solvent
increases:
Eκ (V ) 2κ 3 /(9V 2 ).

This makes a negative contribution to the work required. But there are other
contributions to consider. One such contribution is the interaction energy
of the transferred B molecules. Most of the B molecules transferred remain
surrounded by B, so that they feel no change in interaction energy. Only
the ones adjacent to the surfactant layer experience a change; the associated
work is simply proportional to the area , and is independent of V . We
shall return to this contribution below. There is a further form of energy
to be noted. Adding the B solvent reduces the number of micelles. This
number Nm is evidently /(4π R 2 ), or  3 /(36π V 2 ). This reduction in the † In more detail, each small increment of
number of micelles gives the solution fewer degrees of freedom. As we saw micelle number δNm costs a work δW =
T δNm log(/Nm ), where  is the volume
in the section on mixing principles, this requires a work T Nm log(v)+const of the system in units of the micelle volume.
where v is the volume fraction per micelle† . In terms of the solvent volume To find the total work, one adds (integrates)
V , this amounts to a free energy cost ET : the work δW for each increment of micelle
number. The resulting integral simplifies to
T Nm log(v)+const in the regime where the
ET (T  3 /36π V −2 ) (log v + const). volume fraction of micelles is small.

The two opposing contributions to the work have similar dependences

on the volume V ; both vary as V −2 . ET is proportional to T while Eκ is
proportional to κ. If κ is sufficiently larger than T , the gain of curvature
energy dominates, and the micelles swell with solvent. Concretely, if the
volume fraction v is 10−3 or more, the micelles swell if κ  T /3.6.
Well-defined surfactant interfaces have bending moduli greater than T ,
as noted above. Thus there is normally a strong tendency to swell. The
swelling continues until the micelles grow comparable to their preferred
size 1/c0 . They do not grow larger than this size, since then both the
bending energy and the reduction of the micelle number oppose further
swelling.
This simple example shows how surfactants lead to the dispersion of one
solvent into another immiscible one. This equilibrium dispersed state is
known as a microemulsion. One only needs surfactants that disperse in the
host solvent and prefer to be less curved than they are in the form of bare
micelles. Our analysis suggests that under these conditions a large volume
fraction of solvent can be dispersed, with an indefinitely small amount of
surfactant, simply by selecting a surfactant with a large κ and a small c0 .
The limitations on ability to disperse arise from the interaction between
the swollen micelles. Large micelles interact with each other in the same
194 Surfactants

ways as the large colloidal particles discussed in Chapter 5. The larger they
become, the larger their attractive interaction. Beyond some given size the
attraction leads the micelles to precipitate into their own micelle-rich phase
rather than dispersing [15].
In our example we assumed that κ was large enough that the thermal
fluctuations of the surfactant interface would be small. This meant that the
micelles had to be smaller than the persistence length p . If R becomes
larger than p then the micelles fluctuate strongly in shape. If the distance
between micelles is comparable to R, the micelles can even merge to make
larger irregular objects. A limiting case of these fluctuations is the situ-
lρ ation where c0 = 0, the volumes of A and B solvent are equal, and the
radius R p . Then the distinction between the inside and the outside
Fig. 7.8 of the droplets disappears. The microemulsion can then be viewed as a
Left: Schematic sketch of a symmetric, collection of cells of size p , each containing randomly A or B solvent,
bicontinuous microemulsion. Right: with a surfactant membrane between cells of different solvent, as shown
cellular model of the microemulsion in Fig. 7.8. Such bicontinuous microemulsions were first discussed in these
according to DeGennes and Taupin [13].
terms by DeGennes and Taupin [13]. Since that time, the so-called Winsor
The persistence length p is shown.
III microemulsions [16] known to chemical engineers were identified as
† Awell-known Winsor microemulsion con-
bicontinuous microemulsions [17]† (cf. Fig. 1.2).
sists of SDS, brine, and toluene, with a small The larger the typical size of the microemulsion droplets, the smaller
amount of butanol. The SDS has a concen-
the surfactant fraction becomes. One may ask how small the fraction of
tration of 6–10 g/l, the butanol has a weight
fraction of 3%, the brine contains 6.5 weight surfactant can become. Is there a limit to how little surfactant is needed
percent salt, and the toluene and water are in order to mix oil and water in an equilibrium dispersion? Many Winsor
in equal volumes within a factor 2. microemulsions achieve a surfactant volume fraction of as low as 10−4 or
10−5 . The typical size R̂ of the oil or water volumes appears in scattering
experiments as a peak at some q ∗ 1/R̃, as shown in Fig. 7.10. Such
measurements indicate typical sizes of the order of 100 nm [15]. But it
is difficult in practice to increase R̂ much beyond this scale, or to reduce
the amount of surfactant below this level. To increase R̂ further requires
that the bending stiffness κ be increased so that the persistence length p
increases. But in addition, the spontaneous curvature c0 must decrease so
that c0  1/R̂. Thus 1/c0 must become much larger than the surfactant
layer thickness b. However, slight changes in the A or B solvents can make
small changes in c0 and thus reduce R̂ significantly.
There is a great incentive to create stable microemulsions in order to
mix immiscible fluids. Accordingly, the structure and phase stability of
microemulsions is extensively studied. The simple scheme of random
water and oil volumes sketched above has given way to much more
systematic and powerful analyses [19]. A major focus of modern the-
ories is to explain the occurrence and the coexistence of phases in the
oil–water–surfactant systems. Figure 7.9 shows a typical phase diagram,
indicating the multiple phases that can occur and the involuted boundaries
separating them.
All of this discussion presupposes that the solvent B enters the bare
micelles. In order to enter, the first B molecules must create a surfactant-B
interface and pay any associated interfacial energy α, as noted above. If
this energy price is higher than the gain in curvature energy discussed above,
the energy to absorb the next B is unfavorable. For a favorable balance,
 Mixing immiscible liquids: microemulsions 195

AOT Fig. 7.9

Gibbs phase diagram for AOT, decane oil,
saltwater system from [20, figure 2],
H2
S © American Institute of Physics, courtesy
R. Strey. The saltwater or brine has a 0.6%
T = 40°C weight fraction percent of salt and is fully
I2
deuterated. The temperature is 40 ◦ C. Each
point on the diagram represents a
particular composition. Top vertex
B O indicates pure AOT surfactant, and the
other vertices are pure brine and pure oil.
L Percentage of each component is the
1
–2 fractional distance from the opposite edge.
C 2– CP As an illustration, the dot on the legend at
3 the right indicates the composition with
D2O/Nacl n-decane (weight) ratio 20 : 50 : 30 of oil to brine to
ε = 0.6 wt% surfactant. The region marked H2 consists
of densely packed cylinders of AOT. The
phase I2 is another surfactant-rich phase.
Lα labels a lamellar phase. The regions
indicated by 2 have coexisting
we need α  κci2 , or α  κci2 . The bare micelle radius 1/ci is of compositions. The triangular bottom
the order of the surfactant layer thickness b. As we have seen, the bending region has three coexisting compositions.
modulus κ is ordinarily of order T . Thus a favorable swelling energy implies One of these is the microemulsion
α  T /b2 . This is a rather strong requirement on the interfacial tension α. composition indicated by C. In the region
marked 3 this C composition coexists with
Typical interfaces have energy of the order T for every translational degree nearly pure oil and nearly pure water. The
of freedom at the interface. For our surfactants this amounts to T for each two eyebrow-shaped lines are the upper
surfactant head or tail area projected onto the interface. This area is normally boundary of the two phase regions. On the
smaller than the surfactant length b squared. Thus we need the B-surfactant right-hand boundary, the black dot marked
“cp” indicates a critical point where the
energy to be small on the scale of typical interfacial energies in order for the
two coexisting compositions coincide. The
micelles to take up the B solvent. This fact explains why one cannot readily two open circles on either side of “cp” are
make a microemulsion where solvent B is air, for example. No surfactant two other coexisting phases.
has a part that dissolves in air; thus the interface between surfactants and
air is not small.

7.6.1 Interfacial tension

The spontaneous mixing seen in microemulsions leads to very low sur-
face tensions. To understand this, we consider the interface of the simple
spherical microemulsion discussed in Fig. 7.11. The system has come to
equilibrium, so that the micelles have their preferred radius 1/c0 . Here the
amount of surfactant is limited, so the volume fraction of micelles is mod-
erate. The interface between the microemulsion and the excess B solvent is
naturally coated with surfactant. The interfacial tension is the work required
to add a unit of area to this interface. We may expand this interface by an
area 4π R 2 by transferring a micelle from the microemulsion to the interface.
This transfer requires the surfactant to depart from its preferred curvature
1/c0 and become flat, with an associated energy cost 4π R 2 κc02 or simply
4π κ. Removing the free micelle also reduces the number of translational
configurations of the system, with a free energy cost T log v as encountered
above. Since κ is generally of order T or more, our energy cost is typically
dominated by the curvature contribution. The associated interfacial tension
196 Surfactants

105
 = 6 wt%

104

I(q) [cm–1]
12

103
21

Fig. 7.10
Neutron scattering intensity for 102
bicontinuous microemulsions, reproduced
from [18, figure 17], with permission from
Elsevier. The microemulsion consists of
equal volumes of D2 O and n-octane. With 101
the indicated weight percent of C12 E5
surfactant, and closely resembles the
microemulsion of Fig. 1.2. The wave
vector q at the peak decreases as the 100
0 0.2 0.4 0.6 0.8
amount of surfactant decreases, indicating
larger droplets of oil and water. q [nm–1]

Fig. 7.11
Interfacial tension versus temperature in a
bicontinuous microemulsion, from [18, 32
figure 5b], with permission from Elsevier.
Solution contains three phases, with 30
overall composition of 58.5% water, cb
28 Tu
36.5% n-decane, and 5% C8 E5 nonionic
T (°C)

surfactant. As the temperature increases, 26

the proportion of the different components
changes. The three phases are denoted a 24
for the water-rich phase, b for the oil-rich ac+ cb
22
phase, and c for the microemulsion phase. ab
The microemulsion is close in composition 20
to that of Fig. 1.2. At the crossing point of
the two curves in the middle of the figure, 18
the microemulsion has a very low
16 Tl
interfacial tension with both oil and water. ac
(The very low tensions near the 14
temperatures Tu and Tl correspond to
phases that are becoming equal in 12
composition, so a vanishing tension is to 0.001 0.01 0.1 0
be expected.) Interfacial tension (mN/m)

is thus of order κ/R 2 . We recall that typical interfacial tensions are of the
order T per interfacial degree of freedom, or several T per square nanometer.
Our micelles have a diameter of many nanometers. Thus the interfacial ten-
sion can be orders of magnitude smaller than that of a simple-liquid interface
(cf. Fig. 7.11).
 Mixing immiscible liquids: microemulsions 197

7.6.2 Emulsions and foams

Surfactants can assist the mixing of oil and water even when they do not
make a microemulsion. By assembling at the interface they can lower the
interfacial tension significantly. This makes it easier to disperse e.g., water
(B) in oil (A). Without the curvature effects prescribed above, the bare
micelles dispersed in the A solvent need not absorb significant amounts of
B solvent. (For example, the common SDS does not disperse significant
amounts of oil in water.) However, by disturbing the two-phase fluid with
external forces, one can achieve significant dispersion. The most common
way to apply these forces is to stir the mixture, subjecting it to viscous
stress. A given large droplet of B in A of radius R is distorted by the shear.
This distortion is opposed by interfacial tension α that pushes the droplet
back towards the shape of least area: a sphere. To simplify our thinking we
may view the hydrodynamic force as exerting a tension between the two
halves of the droplet. This force is the hydrodynamic stress times the area—
roughly ηγ̇ R 2 , where η is the viscosity and γ̇ is the shear rate. The opposing
interfacial-tension pulls the two halves together with a force of order αR.
Evidently the shear force grows faster with R than the interfacial force. For
small R the interfacial force dominates and the droplet is scarcely affected
by the flow. But for large R, the shear force dominates and pulls the droplet
apart. Thus stirring a two-component fluid mixture at a shear rate γ̇ should
break it into droplets of size R α/(ηγ̇ ). This rule, called the Taylor
formula is roughly borne out in practice [21, 22].
Clearly anything that reduces the interfacial tension α, such as surfactants,
promotes finer dispersion. The surfactants have another helpful effect. They
keep the dispersed droplets, once formed, from coalescing, thus maintaining
the dispersed state. Often the surfactant-coated surfaces repel each other
owing to the Coulomb or steric repulsions treated in Chapter 5. Moreover,
if two surfactant-coated droplets touch, they cannot merge unless many
surfactant molecules reorganize. Thus it is that two adjacent surfactant-
coated droplets can coalesce only on a timescale of years [23]. These effects
explain the long shelf life of household emulsions like mayonnaise or cold
cream.
In a foam, one of the solvents to be dispersed is air. For most purposes
foams are just another type of emulsion. Often foams can be made without
external stirring. Instead the foam bubbles can be made to form by starting
with a saturated solution of the gas at high pressure and then suddenly
lowering that pressure. The solution of gas in liquid then finds itself super-
saturated, bubbles nucleate and grow, and the foam forms. This is what
happens when shaving cream is released from its aerosol can or a bottle of
beer is opened.
7.A Suggested experiment: Do not shake the soda can! If a can of carbonated
soda is shaken rapidly before opening, there is a violent release of bubbles and gas.
Even with no shaking, the opened liquid begins to bubble, because the release of
pressure makes the liquid supersaturated with gas. Normally, the gas is released
slowly because it cannot make a bubble without overcoming an energy barrier. The
energy cost is proportional to the interfacial tension times the bubble area. Thus
the gas escapes by creating bubbles at the walls (at places that cost less energy)
and by expanding existing bubbles. When the can is shaken, many new bubbles
198 Surfactants

are dispersed in the liquid. Then when the pressure is released, the gas may escape
by expanding these preexisting bubbles: there is no barrier to overcome. Thus the
release of gas is much faster. This project investigates this effect.
A 1 L can of 10 cm diameter is shaken with an amplitude of 10 cm at 10 times
per second. You can assume that the center of the liquid remains stationary while
the can shakes.
(a) Estimate the shear rate γ̇ inside.
(b) What is the order of magnitude of the viscous stress in the liquid, assuming it
has the viscosity of water?
The shearing force pushes the air at the top of the can into the fluid and breaks it
up into small bubbles. The bubble size is such that the interfacial tension pushing
bubbles towards a spherical shape is comparable to the viscous force tending to
elongate the bubbles.
(c) For the given shear rate, what is the size of bubble for which these two forces
are comparable? You may assume that the interfacial tension is the same as that
between water and air.
(d) Supposing that the 5% of the original 1 L volume that was air is dispersed into
these small bubbles, what is the distance between these bubbles compared to
the average distance to the wall of the can?
(e) Suppose the bubbles rise as though they were Stokes spheres as discussed in
Chapter 4. (Bubbles actually rise somewhat faster than solid spheres of the
same size. Also a cloud of bubbles rises faster than an individual bubble.) How
long will they stay in the liquid? This is the time you must wait before opening
the can in order to avoid a mess.

7.7 Amphiphilic polymers

In recent decades it has become possible to make polymers with the defin-
ing characteristic of an amphiphilic molecule: two mutually immiscible
parts are joined together. The simplest example is a diblock copolymer, in
which two immiscible polymer chains are joined end to end, as described
in Fig. 1.1. Melts or solutions of diblock copolymers often form micellar
structures, as surfactants do. The polymeric micelles are of course larger
than their small surfactant counterparts. But more importantly, the proper-
ties of polymeric micelles are far more predictable and controllable than
are conventional micelles. This is because these properties are controlled
by universal features of long-chain polymers. In this section we sketch how
the polymer chain properties lead to precise predictions about the size and
energy of a micelle. The micelles constrain the polymer chains into a state
of mutual stretching that is qualitatively different from the relaxed state.
This stretching gives rise to new scaling properties with molecular weight.
We survey a number of situations where this stretched state is important.
The theory of this stretched state of polymers has grown powerful, soph-
isticated and successful. A good introduction to the subject appears in two
recent surveys [24, 25].
We encountered a polymeric micelle structure in Fig. 1.1. The inter-
woven branches of this structure show the intricate structures that polymeric
amphiphiles can spontaneously make. We now describe the simplest form
of structure arising from the same principles. To this end, we consider a
liquid of symmetric diblock copolymers, in which the two joined chains are
 Amphiphilic polymers 199

the same in every way except that they are mutually immiscible. The chain
contains N monomers, with N /2 in each subchain or block. We label the
two blocks A and B. We first recall the criterion for being immiscible. We
denote the work to insert one isolated B monomer into a liquid of pure A
polymers (or the reverse) as χ T † . Most A and B monomers encountered in † In nonsymmetric polymers, χ T as con-
practice have χ parameters that are only a few times 10−2 or smaller. In our ventionally defined [26] is the average of
discussion below, we will make the simplifying assumption that χ  1. the work of inserting a B into A and the work
of inserting A into B. In our symmetric case
The χ parameter measures the immiscibility. As noted at the beginning these two works are the same.
of this chapter, two liquids remain miscible as long as the work to insert
remains below about T . The work to insert a random coil B block into pure
A is of order N χ T . Unless N is sufficiently large, this work is negligible,
and free A and B chains mix readily. However, as N increases, the work
increases; ultimately, phase separation occurs. The same is true when A and
B chains are attached to make copolymers. If χ N  1, the A and B blocks
must phase separate, so that virtually all the A monomers are surrounded by
A blocks and likewise for the B blocks. In the limit of large N any diblock
polymer liquid with a positive χ must become strongly phase separated in
this sense.
This strong phase separation imposes a serious constraint on the indi-
vidual chains. Each junction point between the two blocks is effectively
confined to the interface between A and B regions. To see why, we imagine
moving a junction point from the interface into an adjacent B region, for
example. This movement necessarily brings monomers from the A block
of that polymer into the B region. Each monomer thus pulled into the B
region requires a work of order χ T . Since χ N  1, even a small fraction
of the A block thus pulled requires a work exceeding T . This work must
decrease the probability that the junction point moves spontaneously into
the B region according to the Boltzmann Principle. The larger N is, the less
free the junction point is to explore the B region. Below, we shall treat the
junction points as being confined near the A–B interfaces.
Our liquid consists of regions of nearly pure A or pure B. Since this
is polymer melt, each point in an A region is uniformly occupied by A
monomers†† . The A monomers encountered at this point have a pecu- †† The cost of altering this uniform dens-
liar property not shared with ordinary polymer melts. Each such monomer ity, by e.g., mixing empty space into the A
region, can be viewed as a mixing process
belongs to a block that ends at the A–B interface. The same is true for the B
with its own χ parameter. This cost is pre-
monomers. This constraint limits the size of the A or B regions. If a region is sumed arbitrarily larger than the χ for AB
large, then blocks passing through the center must necessarily be elongated mixing. Thus we can neglect the effect of
in order to extend to the A–B interface as required. Such elongation costs A–B phase separation on the individual A
free energy and this cost makes large regions unfavorable. and B densities.

7.7.1 Micelle size

The combined effects of phase separation and polymer stretching determine
the size of the A and B regions in our symmetric copolymer liquid. Since
there is no difference between the A and B blocks, we expect no difference
in the size or shape of the A and B regions. It is thus natural to suppose that
the liquid consists of planar slabs of alternating A and B. We will denote
the thickness of each slab as 2h. If we impose a specific thickness h, we can
200 Surfactants

readily estimate how the free energy of the liquid depends on h for chains
of a given length. It is most convenient to measure chain length in terms of
the volume V that a chain displaces in the melt. Evidently V is proportional
to the number of monomers N . A free chain in the melt state is a random
walk, whose mean-squared end-to-end distance R 2 is also proportional to
V : R 2 = V /a. The proportionality constant a is a characteristic length
that depends on the chain structure and the packing of the monomers in the
melt state, but it is independent of the chain length. We call it the packing
length.
If we change the size h, one major effect is to alter the interfacial energy.
As with any liquid interface, the energy is proportional to the area of the
interface and is characterized by an interfacial energy α. The ultimate origin
of this energy is the immiscibility of the A and B blocks, measured by χ. For
our purposes, the connection between χ and α is unimportant. We need only
observe that α arises because of a thin mixing region that is independent of
† The mixing region consists of chain seg- chain length† . Thus α is independent of chain length for long enough chains.
ments of length k such that χ k 1. This The interfacial energy per unit volume is the interfacial area per unit volume
mixing region is the same as that at the
times α. The chains for a given area A of A–B interface occupy a volume
interface between A and B homopolymers.
2Ah. Thus the interfacial energy per unit volume is αA/(2Ah) = 12 α/h.
The interfacial energy per chain, denoted U , is thus given by U = 12 V α/h.
Evidently this U is smallest when h is largest. The interfacial energy favors
large phase-separated regions.
As noted above, there is an opposing cost to increase h. This is the cost of
the required chain elongation. If the slab thickness is h, chains must stretch
to lengths of order h. Indeed, any monomer at the center of a slab must
belong to a chain whose end-to-end distance is atleast h. Any monomer
picked at random lies at a distance of order h from the nearest interface.
Thus to increase h is to increase the end-to-end distances of the chains in
proportion to h. The free energy S required for this stretching is a basic
property of random-walk molecules. As discussed in Chapter 3, the work
needed to stretch a chain to an end-to-end distance h perpendicular to the
slab is given by
1 h2 ah2
S(h) = T 2 T .
2 R /3 V
Since most of the chains extend to a distance of order h, the chains have an
average stretching energy S of order S(h).
If the system is arranged into slabs of a given size h, the energy per chain
is S + U (h). If the system is free to adjust h, the relative probability of a
given h is e−(S +U (h))/T . The most likely value h∗ is that which minimizes
S + U (h). Evidently
∂
0= [S + U (h)]h∗ .
∂h
Since S varies as h2 while U varies as h−1 , we have

2S − U (h) 
0=  ∗. (7.16)
h h
 Amphiphilic polymers 201

This equation tells us three things. First it gives the scaling of the preferred
size h∗ with chain size V

aT (h∗ )2 /V S U (h∗ ) αV /h∗ ,

so that
h∗ ∼ V 2/3 (α/aT )1/3 . (7.17)
We note that h∗ becomes arbitrarily larger than the unperturbed size
R(∼ V 1/2 ) in the long-chain limit. This confirms that the main distortion
of the blocks resulting from the phase separation is indeed one of stretching,
as implicitly assumed above.
Second Eq. (7.16) tells us how the free energy per chain F ∗ grows with
the chain length V :

[S + U (h)]h∗ S(h∗ ) T a (h∗ )2 /V . (7.18)

Since h∗ grows arbitrarily bigger than the resting size R, this energy grows
indefinitely relative to T . Thus, the stable structure contains significant
energy. Even when the stretching energy is little more than T per chain,
the total stretching energy is large, because there are many chains in the
structural unit. The number of chains in a cube of size h∗ is (h∗ )3 /V ∼
V α/(aT ). Both α and a have length dimensions of atomic scale. But the
chain volume V is typically thousands of times the volume of an atom. Thus
even when the immiscibility is weak, the number of chains in a region of
size h∗ is typically hundreds, and the energy associated with the structure
is many times T .
This large energy means that the region size h is well defined and varies
little from h∗ . If h departs slightly from h∗ , the free energy per chain can be
found by expanding S + U (h) around its minimum. The free energy in a
volume (h∗ )3 thus has the form (h∗ )3 V F ∗ [1+(const)(h−h∗ )/h∗ ]2 , where
(const) is of order unity. Since F ∗ is large, as argued above, even a small
departure of h from h∗ costs more than T and is thus improbable. The size
of the regions is very well defined and becomes more so with longer chain
length. Likewise, significant distortions of the shape of the A or B regions
creates stretching energy of order S and is thus strongly suppressed.
Finally the equilibrium condition for h∗ of Eq. (7.16) gives the ratio of
stretching energy S to interfacial energy U at equilibrium. Though we do
not know the coefficient relating S to h2 , we do know that the power is 2.
Thus we know rigorously that dS /dh = 2S /h. U also varies as a known
power of h, and its derivative is rigorously given by dU /dh = −U /h. Thus † Such “virial relations” appear whenever
Eq. (7.16) does not have unknown coefficients. When it is satisfied, we must the free energy of a system is the sum of
have 2S = U . The interfacial energy must be twice the stretching energy two parts, each of which varies as a power
at equilibrium. In other terms, the stretching energy is 13 of the total energy† . of some variational parameter.

7.7.2 Other copolymers

The reasoning above applies far beyond the special case of symmet-
ric diblock copolymers treated above. If for example, the A block is
shorter than the B block, the A regions need not be the same size or
202 Surfactants

shape as the B regions. In order to accommodate the large B volume

with minimal stretching, it is natural that the interface curve away
from the majority blocks. As the A blocks become shorter and shorter
relative to the B blocks, there comes a point where the layered struc-
ture is no longer the most stable. Instead, the stable structure is a
lattice of parallel cylinders. With further reduction of the A blocks,
another transition to a lattice of spherical A micelles occurs. Further
morphologies are possible in principle. Indeed the complicated double-
gyroid structure pictured in Fig. 1.1 appears to be the equilibrium
structure in a narrow range of composition between the layered and
the cylindrical regimes [27]. More new structures may emerge in the
future.
The basic energies that give rise to these structures are the same as in
the symmetric case treated above. The region size is still set by a balance
between stretching energy and interfacial energy. The former scales as h2 as
before, while the latter continues to scale as 1/h. Only the numerical coef-
ficients depend on the specific geometry of the structure. Ignoring these
prefactors, we may repeat the arguments above and arrive at the same con-
clusions. The region size h scales as V 2/3 , the energy per region is large
and proportional to V , and the stretching and interfacial energies are in the
ratio 1 : 2. With care, these arguments can be extended to
• polymers with a range of molecular weights,
• mixtures of block copolymers and homopolymers,
• multiblock polymers in which several alternating blocks of A and B occur
on each chain,
• star polymers in which several A and B blocks emerge from a common
node,
• triblock copolymers with three distinct immiscible species.
It is not surprising that a wealth of self-forming structures can be created
using all these variations. Since these structures arise from the simple prop-
erties of immiscibility and chain randomness, the mathematical problem
of predicting the structures from the molecules is well-defined and con-
ceptually simple. For complicated structures, the practical calculation of
S becomes technically complicated. It is not simple in practice to find
the energy of the mutually and inhomogeneously stretched chains through-
out each region. Still, numerical means have been devised to accomplish
this task, and the numerical predictions match successfully with experi-
ments [24]. The usefulness of these structures is also broad. Many different
types of polymers can be accommodated and their structures predicted,
since it is easy to make polymers flexible and mutually immiscible without
constraining other desirable properties.

7.7.3 Polymeric amphiphiles in solution

The distinctive state of mutual stretching described above becomes only
stronger when one considers copolymers in solution. For example, one
may add a solvent for B to the symmetric diblocks discussed above. If
the solvent does not dissolve A, then the junction points are tethered to
 Dynamics and rheology 203

the A–B interface as before. As solvent enters the B regions, their volume
increases, and the B blocks expand and stretch to dissolve in the solvent as
much as possible. The result is strong stretching of the B blocks and great
asymmetry between the A and B regions. The calculation of the region size
h is modified, because the volume per chain is no longer fixed and the
quality of the solvent must be considered. These changes, though, prove
minor so that the same mathematical methods used for diblock melts can
be modified to describe these solutions. Now in addition to the structures
above, one can form free spherical or wormlike micelles in the solution.
One can also attach chains to solid surfaces to form the brush structures
discussed in Chapter 5. That chapter gives a hint of how the concentration
profile of brushes or other micelles arises.
One may wonder how well amphiphilic polymers function as surfactants,
dispersing immiscible solvents and reducing interfacial tension. In practice,
these polymers are little used for these functions. Part of the reason lies in
the large energy of these stretched polymer blocks. Many of the functions
of surfactants require their micelles to be flexible and deformable. For
example, the Helfrich repulsion between surfactant bilayers relies on this
flexibility. But polymer micelles have little of this flexibility, because their
deformation energy is large, as we saw above. For example, the bending
modulus of a lamellar layer is of the order of the energy within a cube
of thickness h as thick as the layer. As shown in this section, this energy
grows rapidly with polymer molecular weight. Thus the main interest in
amphiphilic polymers lies in the great range of ordered and controllable
structures that they make.

7.8 Dynamics and rheology

The time scales for motion and relaxation in surfactant solutions and
microemulsions are controlled by solvent hydrodynamics as in the polymer
solutions treated in Chapter 4. The forces transmitted across these solutions
arise from hydrodynamic stress and from the distortion of the surfactant
interfaces. From these basic facts, we can understand the viscosity of most
surfactant-structured fluids with compact micelles. If the volume fraction
of micelles is small, the viscosity is near that of the solvent. If the volume
fraction grows substantial, the viscosity grows, because then the micelles
interact and distort each other, thus storing free energy. For example, in
our spherical microemulsion, at volume fractions approaching the close-
packing density of about 60%, the spheres come into contact and distort
each other. Then a unit of shear distortion translates into a further distor-
tion of the spheres, increasing their area and mean curvature by a factor of
order unity. Their curvature is altered, and this increases the energy of each
sphere by an amount of order the bending stiffness κ. The associated stress
is of order the energy per unit volume, or κR 3 . In a simple (nonequilibrium)
emulsion the distorting work is done against the interfacial tension α. Then
if the droplets (or bubbles) have a typical radius R, the distortion energy for
a unit of shear of a droplet is of the order αR 2 . The associated stress is then
of order αR 2 /R 3 or α/R. Figure 7.12 shows the modulus of emulsions of
various droplet sizes, confirming this scaling with droplet size.
204 Surfactants

Fig. 7.12
101
Elastic modulus for emulsions made of
spherical droplets of uniform size 100
from [28], © American Institute of
Physics, courtesy T. Mason. Droplets of 10–1

Scaled modulus
silicone oil coated with SDS were
dispersed in water. Droplets had radii r of 10–2
0.25 μm (solid circles) 0.37 μm
10–3
(triangles), 0.53 μm (squares), and 0.74
μm (diamonds). The horizontal axis is the 10–4
volume fraction of the droplets, including
the volume displaced by the polar heads. 10–5
The vertical axis is the measured elastic
modulus divided α/r. The large circles 10–6
0.5 0.6 0.7 0.8 0.9 1
report measurements of osmotic pressure
for r 0.48, scaled in the same way. The Effective volume fraction
modulus and osmotic pressure are very
similar in magnitude for all but the largest
volume fractions. They rise sharply as the The relaxation of stress in these packed micelle or droplet structures
volume fraction increases so that the
droplets push against each other.
can be very slow. When the droplets are concentrated enough to touch, an
applied stress can only relax if the droplets can move past each other. To
do this, a droplet must increase its area by a nonzero fraction. This motion
thus requires an energy of order αR 2 , which can be much larger than T .
Thus the spontaneous relaxation of stress requires activated motion as is
exponentially slow. For many purposes the system can be regarded as a solid.

7.8.1 Wormlike entanglement and relaxation

Naturally wormlike micelles experience the same entanglement constraints
that polymers do, as discussed in Chapter 4. We consider worms that are
most like their polymeric cousins, with negligible branching and with a per-
sistence length p that is much smaller than the distance ξ between micelles.
On this basis, we would expect the following. (a) Entanglements should
have a density of about 1/ξ 3 . (b) A small applied step strain of magnitude
γ should elongate the randomly coiling segments between entanglements,
thereby reducing the randomness and requiring a work per unit volume
of order (T /ξ 3 )γ 2 . (c) To relax this applied strain, a worm should have to
reptate over a distance comparable to its length L. This should require a time
τrep τξ (L/Lξ )3 . Here τξ is the Zimm time of an isolated worm of size
ξ , and Lξ is the length of such a segment. Some of these expectations hold
true, but some are qualitatively wrong, as we now discuss. The relaxation
of wormlike micelles is discussed at length by Cates and Candau [7].
The density of encounters between worms is of order 1/ξ 3 as anti-
cipated in (a): this fact relies only on the structural form of the worms.
These encounters cannot obviously be regarded as impenetrable barriers
or entanglement constraints, however. In principle, the worms can merge
and pass through each other as polymers cannot. This kind of merging
behavior has been proposed to explain some experiments [7]. But most
wormlike micelles appear to resist merging sufficiently that merging can
be neglected. If merging is negligible, the constraints amount to entangle-
ments, and strain must elongate the worms. The energy cost again arises
 Dynamics and rheology 205

simply from the random-walk fluctuations common with polymers. Thus

one expects a stored energy per unit volume given by (b) above. This means
that the solution should have a plateau modulus of order T /ξ 3 as a polymer
solution has.
The relaxation of the stress occurs by whatever routes lead to a res-
toration of isotropic random-walk structure in the worms. These worms
have a channel for relaxation that polymers lack. They can break and
reconnect. We consider a network of worms that define the entanglement
constraints for a given segment of a micelle. We will take this segment to
be much larger than Lξ , but much smaller than the mean micelle length
L. After a step strain, the constrained micelle may reptate to and fro
along this network, but this motion in itself does not reduce the imposed
anisotropy of the micelle and does not reduce the stress. Only when a
micelle end passes through the network can it redefine the network by
penetrating through it along a new path. With polymers, the relaxation must
wait until the end of the chain has passed through the section of network
being considered. But with micelles, there is a continual risk of break-
age. Every breakage results in the creation of two ends, whose subsequent
motion creates relaxed pieces of micelle. Likewise, every reconnection
of two ends stops the relaxation of constraints that would otherwise have
occurred.
A typical micelle of length L̄ breaks and reconnects on a timescale τb
that depends on the local breaking kinetics. There is a fixed rate of breakage
 per unit length, i.e., 1/τb  L̄. If this τb is as long as τrep , then micelles
have time to relax much of their stress by reptation before they break. Thus
breakage does not play a crucial role. On the other hand, if τb  τrep , then
breakage is important. We now ask how fast the stress should relax in this
fast-breaking regime.
We can understand the relaxation by considering what fraction of the
stress should relax in a time τb . In this time, a typical micelle has reptated
a length  given by (/L̄)2 = τb /τrep . The micelle segments within a
length  of an end have disengaged from the network and their stress has
relaxed. This motion has released a fraction /L̄ of the stress. Thus if the
stress remaining after time t + τb is denoted σ (t + τb ), we can express the
decay by

σ (t + τb ) − σ (t) −σ (t)(/L̄) −σ (t)(τb /τrep )1/2 , (7.19)

or
dσ/dt −σ (t)(τb /τrep )1/2 /τb −σ (t)(τb τrep )−1/2 . (7.20)
This suggests that the stress decays exponentially with a relaxation time
τ (τb τrep )1/2 . Remarkably, the relaxation time is both much shorter than
the reptation time and much longer than the breaking time.
It is interesting to see how τ increases with the mean length L̄, with
everything else held constant, using τb ∼ 1/L̄ and τrep ∼ L̄3 , we find
τ ∼ L̄. It increases much slower with length than the reptation time does.
This kind of breaking-dominated relaxation has been confirmed in rhe-
ological experiments [7]. Its most important effect is on the viscosity η of
206 Surfactants

the fluid. As for any fluid, this viscosity is roughly the product of the step-
strain modulus G0 and the stress relaxation time. The step-strain modulus,
like the osmotic pressure is roughly T /ξ 3 , where ξ is the distance between
chains in the semidilute solution. The relaxation time is the τ discussed
above. Knowing the dependence of L, and ξ on concentration φ, we can
readily predict how the viscosity depends on concentration. Self-repulsion
effects appear important for these solutions [7], so we must use the scaling
exponents D and γ introduced above. For wormlike micelles, L̄ ∼ φ y ,
where y is given by Eq. (7.8). We have deduced the dependence of τrep on
ξ in Eq. (7.17): τrep ∼ L̄3 ξ 3−3D . Thus for ordinary reptating polymers in
these conditions,

η ∼ G0 τrep ∼ L̄3 ξ −3D ∼ φ 3D/(3−D) ∼ φ 3.9 . (7.21)

For the wormlike micelles, we combine τrep and τb ∼ 1/L̄ to infer

τ ∼ (τrep τb )1/2 ∼ (L̄2 ξ 3−3D )1/2 . Thus [29]

η ∼ ξ −3 L̄ξ (3−3D)/2 ∼ φ y φ (3D+3)/(6−2D) ∼ φ 3.72 . (7.22)

The micelles have virtually the same growth of viscosity with concentration
as a solution of ordinary polymers. The competing effects of increasing L̄
and the breaking kinetics nearly compensate.
As noted above, wormlike micelles can sometimes branch or join together
to form a network. The stress relaxation of branched polymers is a tricky
subject that depends on the state of branching, as noted in Chapter 4. The
crosslinks between micelles can inhibit stress relaxation altogether. We shall
not attempt to give a general account of how stress relaxes in these cases.
If the imposed shear rate is significant, the solution may be altered by the
shear. Some types of alteration were discussed in Chapter 4. The chief effect
is mutual alignment of polymers that reduces the degree of entanglement.
In a micellar solution, new forms of alteration appear [29]. The length
distribution of the worms can change, and can be different for worms with
different orientations. The breaking rates can also increase.

7.8.2 Rheology of lamellar solution

The response to strain in a lamellar solution depends on the direction of the
strain. We first consider the simplest case where parallel, flat lamellae are
made to slide over each other, as shown in Fig. 7.13. Such a solution has
little means to support stress or inhibit flow. Since most of the solution is
solvent, the shear occurs mostly in the solvent. The shear rate is only slightly
increased above the average shear rate, even if the surfactant layers allow
no shear at all. Thus the viscosity is little higher than that of pure solvent. If
we shear in a perpendicular direction, the surfactant layers themselves are
obliged to shear (Fig. 7.13(b)), or to bend and compress (Fig. 7.13(c)). The
layers resist such distortions strongly. Shearing a layer forces the awkwardly
shaped and densely packed surfactant molecules to slide past each other.
The stress required should be much higher than for similar shear in the
solvent. If one bends and compresses the layers as in Fig. 7.13(c) there is
 Dynamics and rheology 207
(a)

(b)

Fig. 7.13
The three principal orientations of a
(c) lamellar liquid under shear.
(a) Layer-sliding orientation,
(b) Layer-shearing orientation,
(c) Layer-stretching orientation. Shear
strain is in the plane of the paper in all
three pictures. The layer normal is
indicated by an arrow. In (a) a fluctuation
in two of the layers is shown to illustrate
the resulting compression under shear.

an energy cost even with no motion; the bending cost is controlled by the
bending stiffness κ; the compressional cost is controlled by the Helfrich
repulsion or by Coulomb repulsion. Here steady flow is impossible without
breaking or strongly distorting the layers.
If the shear is applied in a generic direction, one expects the lamellae
to reorient so as to allow easier flow. That is, the lamellae should adopt
orientation (a), and the viscosity should be little greater than that of the
solvent. This behavior is observed sometimes when the lamellae are care-
fully prealigned [30]. But typically a lamellar solution is disordered, with
different orientations in different places. Such solutions do not flow readily
at all. It is not merely that they show a high viscosity. They have a dynamic
modulus that varies roughly as the square root of frequency. This contrasts
with the behavior of a viscous liquid, where the dynamic modulus is pro-
portional to frequency. This square-root behavior may persist down to the
lowest frequencies measurable, implying a stress qualitatively stronger than
that of a liquid.
The reasons for this square-root behavior are ill understood. It is believed
that the lamellae are forced to store elastic energy even when the shear is
weak and slow. The amount of energy stored decreases as the frequency is
lowered because then the lamellae have a chance to relax over larger length
scales. These effects are seen most clearly when the surfactants used are
amphiphilic polymers. To date the justifications for this behavior remain
complicated and tentative [31].

7.8.3 Shear-induced restructuring

Shear in lamellar solutions causes mysterious forms of restructuring. In
weak shear, the lamellae show a tendency to reorient so as to reduce
208 Surfactants

the stress for a given strain rate, as anticipated above. Surprisingly, the
alignment direction is not always in orientation (a) above. Instead, orient-
ation (b) can occur. Sometimes orientation (a) occurs for the lowest shear
rates and orientation (b) occurs at higher shear rates. This strange behavior
can be understood if one looks further into the shape of the layers being
sheared.
Surfactant layers fluctuate in shape; it is these fluctuations that give
rise to the Helfrich repulsion discussed earlier in this chapter. These shape
fluctuations persist for a certain time. If the shear occurs faster than this time,
one may encounter a situation like Fig. 7.13(a) when one tries to shear in
orientation (a). Instead of minimizing stress, the shear drives the bumpy
lamellae into their neighbors and increases the stress. The orientation (b)
avoids this situation. The shear no longer causes rippled lamellae to run
into each other.
What governs the persistence time for these ripples? We first recall
that the fluctuations that control the local distance between bilayers have
wavelengths λ much larger than the layer spacing d. Thus to gauge the
persistence time, we imagine a thought experiment involving a large scale
distortion. We pinch the layers together over long distance λ so that in the
pinched region the spacing is reduced by a small amount A from its equi-
librium value d. We then release the pinching force and watch the spacing
return to its equilibrium value. In order to restore the spacing, solvent must
flow into the pinched region. The flow causes dissipation, which retards
the relaxation. We denote the typical shear rate in this region as γ̇ and the
pinched volume by . During this flow, the energy is dissipated at a rate
Ẇ η(γ̇ )2 , where η is the solvent viscosity.
The return of the spacing to its equilibrium value releases compressional
free energy. As we have seen, these fluctuating bilayers repel each other
because compression constrains their random fluctuations. According to
the Helfrich formula of Eq. (7.15) above, the energy per unit area u can be
expressed in terms of the temperature T and the bending stiffness κ: u
(T /d 2 )(T /κ). Thus the Helfrich energy U in the volume  is given by
U (/d)(T /d 2 )(T /κ). Even if there is additional attraction between
the layers that holds them at an equilibrium spacing, the Helfrich energy is
a substantial fraction of the energy and thus the overall energy is proportional
to this U .
We now express these energies in terms of our perturbed spacing d − A.
If A increases at a rate Ȧ, there must be an inward flow to supply the needed
volume of fluid. The flow enters a given lamellar gap at the boundary of
the pinched region, whose height is roughly d and whose circumference
is roughly λ. Thus the inward current I is roughly λdv. This current must
match the increase of volume λ2 Ȧ, so that v Ȧλ/d. Finally, the shear
rate γ̇ is the gradient of velocity, or γ̇ v/d Ȧλ/d 2 . Thus Ẇ
η(Ȧλ/d 2 )2 . To express the perturbation in the free energy, δU , in terms
of A, we imagine that d is close to an equilibrium, so that δU ∼ A2 . In
view of the order of magnitude of U given above, we conclude

δU (/d)(T /d 2 )(T /κ)(A/d)2 .

 Dynamics and rheology 209

Now we can estimate the relaxation rate Ȧ by requiring that the release rate
of stored energy, balance the dissipation rate:

η(Ȧλ/d 2 )2 Ẇ −U̇ −(/d)(T /d 2 )(T /κ)ȦA/d 2 . (7.23)

Simplifying this equation yields

Ȧ −[(T /η)1/(λ2 d)(T /κ)]A.

Thus A relaxes in a time τ 1/[. . .]. We can rearrange the factors to write

τ (λ/d)2 [6π ηd 3 /T ] (κ/T )/(6π ). (7.24)

This type of compressional relaxation is one of the standard hydrodynamic

relaxation modes of any smectic liquid crystal. It is known as the baroclinic
mode. We recognize the [. . .] factor as the Zimm time for a sphere of
radius d, introduced in Chapter 4. We recall that the Zimm time for a 6.1
nm radius sphere in water is 1 μs. From this time we may estimate τ for a
typical lamellar solution. For simplicity we consider lamellae with a spacing
of 6.1 nm. Typical bilayers have a bending stiffness κ of roughly 10 T . Thus
for λ in the range of 10 d, the expected relaxation time is in the range of
tens of microseconds. The persistence time for a random initial distortion
of the layers is of this same order.
If one increases the shear rate to approach this inverse persistence time,
the shear begins to influence the orientation of the lamellae. Increasing the
shear rate further leads to more dramatic changes. The most prevalent of
these is the rupturing of lamellae to form bilayer vesicles enclosing part of
solvent. When the equilibrium state is lamellar, the vesicles contain many
concentric bilayers in an onion or multilamellar vesicle morphology [33],
as shown in Fig. 7.14. Such onions have been observed in a wide range
of surfactant solutions from simple SDS to lipids. When they form, they
generally share these features:
Fig. 7.14
• The onions form a close-packed mass, as observed in freeze-fracture Freeze-fracture micrograph of onion
electron microscopy. The micrographs show the onions pushed together morphology from [32], © American
so that they deform each other. Chemical Society, 1996.
• The onions become very uniform in size after long exposure to shear.
Onions often contain hundreds of bilayers and are microns in size.
• This size decreases with increasing shear rate. The selected size appears
to have a baroclinic relaxation time comparable to the shear rate [33].
• If the shear is stopped, the onions can retain their integrity over months
or years. Thus they can be used as controlled-release containers.
• The viscosity of an onion-phase solution is typically somewhat lower
than that of the disordered lamellar phase from which it was made. This
is thought to be due to a compression of the layers within each onion and
a corresponding increase in the space between onions. Thus detergent
liquids are often processed so that they are in the onion morphology in
order to increase their concentration without excessive viscosity.
Quantitative understanding of the onion phase is currently very primitive.
210 Surfactants

40 μm

Fig. 7.15
Light micrograph of myelin structures
emerging from a drop of concentrated
C12 E5 surfactant in water, from [34],
courtesy Mark Buchanan

A second important morphology is the myelin morphology pictured in

Fig. 7.15. These wormlike myelin figures consist of many concentric, cyl-
indrical bilayers. They were first observed by biologists in the nineteenth
century [35], who likened them to the myelin structures surrounding nerve
fibers. A classic way to create myelin structures is to place a concentrated
lamellar solution under osmotic stress. For example, one can immerse a drop
or lump of concentrated SDS solution with pure water and then observe the
surface of the drop in a microscope. In the presence of this excess water, the
lamellae have an equilibrium spacing substantially larger than their initial
spacing. Thus the layers start to take in water and swell. Rather than swelling
uniformly, the solutions form bumps at their boundary, and these grow into
the micron-sized worms seen in the figure. Mysteriously, these worms are
often observed to twist upon themselves. They twist indifferently in either
direction. Quantitative understanding of myelins is even more primitive
than that of onions.

Appendix: Gauss–Bonnet Theorem

One can prove the Gauss–Bonnet Theorem elegantly using advanced meth-
ods of differential geometry [12]. One can prove it for nearly flat surfaces
by using the Monge representation introduced in Section 7.5.1. Finally,
one can show that any small deformation of a surface from s1 to s2 leaves
dsC1 C2 invariant. Here we explicitly verify the theorem for the case of
a small bump on a flat surface, as shown in Fig. 7.5.
The heavy solid line shows a vertical section through the bump. The
bump is small on the scale of any preexisting curvature, so that the unde-
formed surface can be considered flat. One can readily verify that the
integral of C1 C2 over the bump is zero. The integral consists of two parts.
The central part is constructed to be a section of a sphere with radius of
curvature R, so here C1 C2 = R −2 . The area of the spherical part is evidently
 References 211

R 2 d cos θ dφ = 2π R 2 (1 − cos θ ). Thus the integral over the spherical
cap of C1 C2 is just 2π(1 − cos θ ). We may allow the bump to be shallow,
so that θ is small and (1 − cos θ ) 12 θ 2 .
To complete the integration, we must include the annular rim region. We
shall choose a rim shape whose cross section is an upward curving arc with
radius of curvature R  . Evidently C1 = 1/R  . The other principal curvature
C2 lies in the direction perpendicular to the vertical section. The two lines
curve towards opposite sides of the surface as in Fig. 7.4, so that C1 C2 is
negative. We can easily find C2 at the inner edge of the rim. This inner edge
is part of the sphere and has C2 = 1/R. On the outer edge, the rim meets
a flat plane and C2 = 0. At intermediate points within the rim we can find
C2 by using the projected radius ρ. At the inner edge C2 = 1/R = sin θ/ρ
(Since ρ = R sin θ). As one goes outward on the rim, the radial distance ρ
changes negligibly, but the inclination angle α (not shown) varies from θ to
0. We may find C2 by imagining a cone tangent to the rim at the point of
interest. The cone’s C2 is the same as the rim’s. At the inner boundary of the
rim, we have found that C2 = sin α/ρ. The curvature of the cone depends
only on the cone angle α and the radial coordinate ρ. Thus throughout the
rim C2 = sin α/ρ α/ρ.
As we move a distance s from the outside of the rim, the inclination
angle α is s/R  . Thus
      †
θ 1 α −2π θ 2 The two integrals must clearly cancel even
C1 C2 = −2πρ dsC1 C2 −2πρ (R  dα) = . when the angle θ is not small if the Gauss–
rim 0 R ρ 2 Bonnet theorem is valid. One readily veri-
fies that for general θ the θ 2 appearing
This rim integral is independent of the rim curvature R  and is exactly the in the two integrals should be replaced by
negative of the integral over the spherical region† . Thus, the overall integral 2(1 − cos θ ). Thus the two integrals indeed
of C1 C2 is 0, as we aimed to show. cancel for large θ as well as small.

References
1. J. N. Israelachvili, Intermolecular and Surface Forces, 2nd ed. (London: San
Diego: Academic, 1992).
2. K. G. Denbigh, The Principles of Chemical Equilibrium: With Applications in
Chemistry and Chemical Engineering, 4th ed. (Cambridge: Cambridge, 1981).
3. D. L. Ho, R. M. Briber, R. L. Jones, S. K. Kumar, and T. P. Russell,
Macromolecules 31 9247 (1998).
4. These drawings are provided by the ChemIDplus
(http://chem.sis.nlm.nih.gov/chemidplus/) resource using
molecular data from the National Library of Medicine, 8600 Rockville Pike,
Bethesda, MD 20894 (http://sis.nlm.nih.gov).
5. J. Lang and R. Zana, in Surfactant Solutions: New Methods of Investigation,
ed. R. Zana (New York: Marcel Dekker, 1987).
6. P.-G. deGennes, Scaling Concepts in Polymer Physics (Cornell: Ithaca, 1979),
Chapter 1.
7. M. E. Cates and S. J. Candau, J. Phys.: Condens. Matter. 2 6869 (1990).
8. T. Tlusty and S. A. Safran, J. Phys.: Condens. Matter. 12 A253 (2000).
9. S. A. Safran, Statistical Thermodynamics of Surfaces, Interfaces, and Mem-
branes (Reading MA, USA: Addison-Wesley, 1994).
10. S. T. Hyde, in Handbook of Applied Surface and Colloid Chemistry, ed., Krister
Holmberg (New York: Wiley, 2001), Chapter 16.
11. V. Luzzati, P. Mariani, and T. Gulik-Krzywicki, in Physics of Amphiphilic
Layers, eds. J. Meunier and D. Langevin (Berlin: Springer, 1987).
212 Surfactants

12. F. David, in Statistical Mechanics of Membranes and Surfaces, eds. D. Nelson,

T. Piran, and S. Weinberg (Singapore: World Scientific, 1989), p. 174.
13. P. G. DeGennes and C. Taupin, J. Phys. Chem. 86 2294 (1982).
14. W. Helfrich, Z. Naturforsch. 33a 305 (1978).
15. M. Kotlarchyk, S.-H. Chen, J. S. Huang, and M. W. Kim, Phys. Rev. A 29 2054
(1984).
16. P. A. Winsor, Solvent Properties of Amphiphilic Compounds (London:
Butterworths, 1954).
17. S. H. Chen, Ann. Rev. Phys. Chem. 37 351 (1986).
18. M. Kahlweit, R. Strey, D. Haase, H. Kunieda, T. Schmeling, B. Faulhaber,
M. Borkovec, H.-F. Eicke, G. Gusse, F. Eggers, Th. Funck, H. Richmann, L.
Magid, O. Söderman, P. Stilbs, J. Winkler, A. Dittrich and W. Jahn, J. Colloid
Interface. Sci. 118 436 (1987).
19. T. Tlusty, S. A. Safran, R. Menes, and R. Strey, Phys. Rev. Lett. 78 2616 (1997).
20. S. H. Chen, S. L. Chang, and R. Strey, J. Chem. Phys. 93 1907 (1990).
21. G. I. Taylor, Proc. R. Soc. A 146 501 (1934).
22. R. G. Larson, The Structure and Rheology of Complex Fluids (New York:
Oxford 1999), see Chapter 9.2.2.
23. J. Bibette, D. C. Morse, T. A. Witten, and D. A. Weitz, Phys. Rev. Lett. 69 2439
(1992).
24. M. W. Matsen and F. S. Bates, Macromolecules 29 1091 (1996).
25. A. Halperin, M. Tirrell, and T. P. Lodge, Adv. Poly. Sci. 100 31–71 (1992).
26. P. J. Flory, J. Chem. Phys. 9 660 (1941); M. L. Huggins, J. Phys. Chem. 46 151
(1942).
27. H. Hasegawa, H. Tanaka, K. Yamazaki, and T. Hashimoto, Macromolecules
20 1651 (1987); M. W. Matsen, Phys. Rev. Lett. 80 4470 (1998).
28. T. G. Mason, J. Bibette, and D. A. Weitz, Phys. Rev. Lett. 75 2051 (1995).
29. M. S. Turner and M. E. Cates, J. Phys.: Condens. Matter. 4 3719 (1992).
30. D. Bonn, J. Meunier, O. Greffier, A. Al-Kahwaji, and H. Kellay, Phys. Rev. E
58 2115 (1998).
31. G. H. Fredrickson and F. S. Bates, Annu. Rev. Matter. Sci. 26 501 (1996).
32. T. Gulik-Krzywicki, J. C. Dedieu, D. Roux, C. Degert and R. Laversanne,
Langmuir 12, 4668 (1996).
33. P. Panizza, D. Roux, V. Vuillaume, C. Y. D. Lu, and M. E. Cates, Langmuir 12
248 (1996).
34. M. Buchanan, Dynamics of Interfaces in Surfactant Lamellar Phases, Ph.D.
thesis, Department of physics and astronomy (Edinburgh UK: University of
Edinburgh, 1999).
35. R. L. K. Virchow, Virchows Arch. B Cell pathol. 6 562 (1854) (Berlin, New York:
Springer-Verlag, 1968–79).
Index

activated hopping 24, 25 Bjerrum length 73, 128, 129, 178 condensation 24, 45, 46, 113, 114
activation barrier 53, 134 blob 86, 87, 90, 99, 100, 104, 105, 106, conjugated 43
activation energy 25 121, 124, 164–9, 186 conjugated polymers 43
addition polymerization 45 block copolymer fluids 3, 10 constant of proportionality 28, 74, 104,
additivity 175–6, 178 block of a copolymer 199 147
adsorbates 36, 152, 161 Boltzmann constant 7, 16 contact angle 152, 159, 171
adsorbed polymer 125–7, 165, 167, 168, Boltzmann distribution 15, 26 interfaces 156–8
170 Boltzmann factor 121 cooperative diffusion constant 100, 101,
affine deformation 28 Boltzmann normalization 22 110
aggregates 4, 5, 7, 113, 116, 123, 134, Boltzmann principle 16, 22, 65, 91, 144, cooperative diffusion and permeability
142, 145, 147 199 101
colloidal 8, 10, 79, 139, 140, 143, 146, Boltzmann probabilities 17, 128 copolymers 44, 201–2
147, 163 Boltzmann–Poisson Equation 128, 132 block copolymers 2, 10, 53, 202
diffusion-limited 134, 144, 145, 147 Boltzmann weight 68, 69, 177 diblock copolymers 44, 198
elasticity 147 bond angles see local structure peptide copolymers 45
fractal 113, 134 breaking-dominated relaxation 205 triblock copolymers 202
properties 146–8 Brownian motion 83, 90, 91, 98–9, 104, copolymerizing 44
reaction-limited 134, 141, 144, 145, 136, 140, 145 corona 125
147 of aggregates 134 correlation functions 76, 84, 101
silica 10 of the polymers 100 correlation length 86, 100, 102, 111, 120,
air–solid interface 157 of a sphere 99 121, 123, 127, 169, 186
air–water interface 152, 153 brush regime 124 correlation time 37
Alexander–de Gennes approximation Coulomb energy 74, 75, 131
124 Coulomb repulsion 73, 127, 130, 207
Alexander, M. 124 capillary flows 160 counterions 43, 129, 130, 131–2, 136
alpha helix 43 capillary forces 157, 163, 173, Cox–Merz rule 109
amphiphilic molecule 11, 163, 173, 198 capillary solid 158 creaming 116
amphiphilic polymers 174, 198–202, 207 carbon–deuterium bonds 39 critical micelle concentration (CMC)
as surfactants 203 cartesian coordinate 14 181, 183, 185, 192
anisotropic interactions 134 cationic surfactant cetyl trimethyl critical mixture 162
annealed variables 24 ammonium bromide (CTAB) 9, crosslinking 43, 44
Argon 28 179, 180, 184 curvilinear diffusion constant 104, 108
Arrhenius plot 25 chain swelling 127 Cyrus, S. 36
“association” of mesoscopic structures charge renormalization 130, 131, 133
10, 43 charge separation 127
asymptotic dependence 6 chemical potential 22 Debye–Hückel approximation 129,
asymptotic equation 52 Cis–trans isomerization 44 130–1
atactic structures 44 configuration energy 20 Debye–Hückel equation: 128, 129
atomic collision times 31 coil-stretch transition 52, 53 Debye–Hückel potential 129
atomic force microscope (AFM) 35 collision 14, 24, 29, 31 Debye–Hückel screening 128, 129, 135
atomic polarizability 115, 175, 176 colloidal aggregate 7–8, 10, 79, 139–46, Debye screening length 128, 130, 131–3
attempt rate 25 163 deformability
average local density 54 a form of self-organization 7 of polymers 9, 96
colloidal crystals/crystallization 132, 136 of rubber 44
colloidal dispersions 113, 125 of surfactants 180, 181
Bakajin, O. 76 colloidal motion 135 deformation of random chains 8
ballistic aggregation 145 colloidal particles 4, 113, 120 De Gennes, PG 29, 64, 104, 124, 126,
Ball, R.C. 141, 147 colloidal stabilization 113–14 133, 158, 159, 167, 191, 194
baroclinic mode 209 colloidal suspensions see colloidal degrees of freedom 14, 15, 19, 24, 117,
beam direction 36, 90 dispersions 164, 184, 193
bending modulus 189, 195, 203 complete nonwetting 157 density correlation function 58
bending stiffness 189, 193, 203, 207, complete wetting on high-energy solid depletion attractions between colloidal
208, 209 surfaces 157 particles 121
bicontinuous microemulsions 194, 196 compressibility 92 Derjaguin approximation 116, 117, 123
bilayers 182, 203, 208, 209 concentration profile 99, 101, 124, 162, Des Cloizeaux J. 56, 64
biosynthesis 45 165–7, 168, 203 dewetting 160
214 Index

diblock 44 excimers 39 hydrodynamic radius 97, 107, 135

diblock copolymer 198 excluded volume 69, 70, 80, 81, 83, 84, hydrodynamic relaxation mode 209
dielectric relaxation 32 86, 90, 97, 107, 124, 126, 133, 146 hydrodynamic screening 83, 98–9, 101,
diffusing momentum 94 Eyring–Kramers theory 25, 26, 27 102
diffusing particles 93, 98 hydrogen bonds 42
diffusion constant 91, 93, 101, 105, 107, far-field velocity 97 hysteretic wetting 159
110, 135 ferrofluids 135, 156
diffusion-limited aggregation 144–6, 147 ferromagnetic 134
diffusion-limited regimes 145 fictitious system 25 ideal gas 22, 23, 58, 80, 81, 83, 84
diffusive analog of hydrodynamic opacity “fixed-ratio Sutherland’s Ghost” model immiscible polymer chains 198
97 141–4, 145 immiscible solute molecules 162
diffusive law for ρ 94, 102 flocculation 130 incompressible 92
dilation invariance 75–8 flocs 123 independent random forces 103
dilation symmetry 56, 75–8 Flory approach 64, 67, 69, 81 index-of-refraction profile 38
dilute solutions 83–90 Flory, P.J. 64, 81 induced-dipole interactions 114–15, 174
2,6-dimethyl pyridine 120 Flory swelling exponent 64 initiates 45
dipole–dipole interactions 135 flow opacity 97 interfaces 151–71
disjoining pressure 123–7, 132 fluctuating dipole 115 interfacial energy 137, 151, 153–6, 160,
dispersed colloids 7, 140 fluctuating membranes 191 171, 195, 199–202
dispersion 113 fluctuation–dissipation theorem 91 interfacial tension 10, 152, 154, 157,
dispersions 113 fluid compressibility 92 158, 161–3, 174, 195–6
dissipation 27–30, 37, 91, 93–5, 97, 108, foams 6, 113, 136, 197, 198 intrinsic viscosity 94–6, 97
109, 110, 137, 156, 158, 159, 209 foot 159 ionic separation 177–8
dissociating ion 180 foot region 159 irrelevant variables 50
distortion 8, 27, 29, 30, 106, 117, 147, force and velocity: relationship between irreversible adsorption 125
148, 156, 161, 164, 184, 197, 200, 91 isomerism 43, 44
201, 203, 208, 209 form factor 57 isotactic structure 44
DNA-like polysaccharides 45 Fourier transform 58, 94, 109 isotopic labelling 37–8
double-layer thickness see fractal aggregates 113, 134 Israelachvili, J.N. 34, 114, 175, 177
Gouy–Chapman length aggregated structure 134, 181
downward impulse 92 fractal dimension 55, 59, 64, 87, 96, 127,
Jannink, G. 56, 64
drag coefficient 91, 97, 107 133, 134, 141, 142, 143, 145, 147,
Jello elasticity 103
drag forces 93, 96, 98 168, 184
Jello modulus 32
dynamic modulus 31, 207 fractal
jungle-gym model 105–6
ratio of stress to strain 31 transparent to a diffusing substance 98
dynamic scattering 1, 37 opaque to a diffusing substance 98
free energy 20 kinematic viscosity 92, 94
Edwards, S.F. 81, 104 kinetic aggregation 140, 142
effective charge 129–30 Gauss–Bonnet Theorem 189, 210, 211 Kraton polymer 10
Einstein, Albert 91 Gaussian curvature 188, 189
Einstein relation 91, 135 Gaussian chain, one-dimensional 53, 126
elastic stress 102, 108 Gaussian scaling 52 labeled 153
electron microscopy 34, 209 Gauss’ Theorem 128, 131 lamellae 192, 206–10
electrophoresis 136–7 gelation 5, 10, 89 lamellar solution 206, 209, 210
electrophoretic mobility 136 glass transition 31 rheology 206–7
electrorheological fluids 135 good-solvent polymer 72, 146 Langmuir trough 152, 153
electrostatic screening 81 good-solvent regime 72 Laplace pressure 156, 157
electrostatic stabilization 122, 127–32 Gouy–Chapman length 131 latex particles 132
electroviscous effects 135 grafted polymers 126, 139 lattice gas 17, 22–3, 80, 81
elongated random walk 106 “gyroid” structure 3, 202 free energy 22
elongational flow 53 law of mass action 182
emulsions and foams 6, 113, 136, 197–8 Hamaker constant 116–17, 122 leaflets 188
end-to-end distribution function 48, 63 Helfrich energy 208 like dissolves like principle 176, see also
end-to-end probability 48–53, 61, 76 Helfrich repulsion 192, 203, 207, 208 ordering
energy barrier 24, 126, 140, 185, 200 Helfrich, W. 192 limiting free surfactant concentration
energy dissipation rate: quadratic “Helmholtz free energy” 22 182, 183
function of the velocities 95 Hervet, H. 159 liquid polymers 28, 44
entanglements 10, 83, 103–7, 169, 187, hexadecyl trimethyl ammonium bromide liquid–solid interface 157
204, 205–6 (CTAB) 9, 179, 180, 184 liquids: quantitative behavior 27–31
entropy 21 history-avoiding diffusion-controlled living polymerization 46
equilibrium-adsorbed polymers 126 case 145 local structure 48, 106, 182
equilibrium probability 15, 25, 163 homogeneity 51 Lovesey, S.W. 56
equipartition theorem 27, 29, 89, 190 Huang, J.R. 101 2,6-lutidine 120
escape time 25–7 hydrocarbons 42, 48, 155 Luzzati phases 188
evanescent-wave scattering 153 hydrodynamic opacity 96, 97 lyotropic liquid crystals 133–4
 Index 215

macroscopic responses: properties of Oseen tensor 93–4, 94–5, 95–6, 99 living polymerization 46
structured-fluid systems 31–4 origin 109–10 polymers 7, 32–3, 86, 96, 41, 66
magnetic resonance 38 osmotic pressure 33, 69, 70, 83–5, 87, flexibility 41
Mandelbrot, B.B. 55 101–3, 109–11, 121, 124, 127, 132, in good-solvent conditions 70
Marangoni flow 160 146, 165, 168, 186, 204, 206 in poor-solvent conditions 70
mass distribution 63, 79, 143, 144 of hard-sphere gas 84 randomness 41
mean curvature 188, 189, 191, 192, 203 variation 100 random-walk 41, 47, 53, 60, 64, 69,
mean-squared displacement 91, 103 osmotic repulsion 123, 125 90, 94, 188
mean-squared fluctuations 89 osmotic sensitivity 33 spatial structure 86
melt state 8, 90, 200 overcharging 131 spontaneous internal motions 98
solvent-free limit 8 overlap concentration 85, 87, 99, 148, theta condition 70
membrane fluctuations 191–2 185 polymer screening 81
Metropolis algorithm 27 polymer solution 89–90, 99
Metropolis dynamics 27 packing length of monomers 200 turbidity 89
micelle 180 “pair correlation function” 84 motion in 90–109
micelle interaction 186–92 parallel polymers 53 polypeptides 42, 45
micelles 9, 10, 11, 130, 173, 180–5 partial wetting 157 polypropylene 42
non-repelling micelles 187 particle–particle interaction 7 polysaccharides 45
spherical micelles 130, 187 partition function 20, 39 polystyrene sulfonate 43
two-dimensional micelles 185, 188–91 pentaethylene glycol monododecyl ether polystyrene: unsaturated hydrocarbon 42
wormlike micelles 182, 184, 187–8, (C12 E5) 179 poor solvent 70
202, 204, 206 permanent dipoles 177–8 positivity: miscibility in liquids 175
microemulsion 3, 10, 34, 174, 190–5, permeability 101, 110 potential of mean force 19, 24, 67–9
203 persistence length 48, 75, 184, 187, 191, power-law regime 168
or equilibrium dispersed state 193 194, 204 principal curvatures 188
miscibility in liquids: principles perturbation–attraction theorem 117–19, probability current 28, 91
controlling 174–8 120, 126–7, 137–9 probability and work 17–20
modulus 74 pervaded volume 7, 54, 62, 78, 80, 81, probes 34, 38
Mohanty, J. 114 147 of atomic environment 38
molecular-layer oscillations 34 phantom attraction 163 of spatial structure 34–7
molecular polarizability 175 phantom polymers 48 of structured fluids 31
moments 78, 81 phase stability 33, 194
momentum-absorbing blobs 99 phase transition theory 140
quantum tunneling 26
Monge representation 189, 191, 210 phospholipid 1-palmitoyl-2-
quenched variables 24
Mondello, M. 8 oleoylphosphatidylcholine (POPC)
monomer-centered spheres 78 179
Monte-Carlo simulations 26, 76 pinned contact line 159 radius of gyration 60, 64, 72, 85, 97,
multifractals 79 plateau region 108 121, 124
multilamellar vesicle morphology 209 plumber’s nightmare 188 randomly branched polymer 146
multivalent counterions 131 polarity 42 random thermal fluctuations 24, 101
myelin morphology 210 polarizability 114 random-walk 184, 200, 205
polyacetylene 43, 48 avoidance 145
nematic liquid crystal 133, 187 polyamide nylon 45 chain 62, 74, 87
“network fluid” 10 polyatomic solutes 163–71 fluctuations 204
neutral wetting 157 polybutadiene 42 history 144
neutrons 37, 38, 56, 57 1,2/1,4 polybutadiene 42, 46 molecules 200
neutron scattering intensity 196 polycondensation 46, 47 properties 7
neutron spin echo technique 38 polydimethyl siloxane 42 self-repelling 7, 164
Ninham, B.W. 114 polydispersity 142 reaction-limited aggregates 145, 147
non-scalar perturbations 120 polyelectrolytes 9, 43, 73–5, 128 reaction-limited regimes 144–6
normalized probability 21 polyesters 42 real system 19
normal stress 108 polyethylene 41 reciprocity: miscibility in liquids 175–7
nuclear magnetic resonance (NMR) 38 polyethylene glycol (PEG) 42 recombinant DNA technology 45
polyethylene oxide 41, 42 renormalized variables 50
onion morphology 209 polyethylene–propylene 44 reptation 104–5, 107, 205
Onsager, L. 133 polyisoprene 42 leads to stress relaxation 107
opaque structures 63, 65, 67, 84, 96–9, polymer deformability 96 model 103, 104, 105
135, 146 polymer melt 32, 85, 106, 107, 199 repulsive forces 113, 122–31, 164
open aggregation 182–5 polymeric amphiphiles 198, 202 requirement of homogeneity 51
optical fluorescence of small dye in solution 203 reservoir configurations 15
molecules 39 polymeric solvents and screening 79–81 Roe, R.J. 8
ordering: miscibility in liquids 176–7 polymer-induced interactions 121, 124 Rouse diffusion constant 104
order of magnitude 11, 27, 30, 31, 96, polymerization, types 41–7 Rouse model 96, 104
115, 161, 208 addition polymerization 45 Roux, D. 36
oscillating strain 32 condensation polymerization 45–6 rubber 4, 8, 28, 39, 42, 44, 106, 146, 151
216 Index

rubber-band modulus 31 spreading pressure 158 total-internal-reflection angle 38

Rutherford backscattering 38 stabilization 114 trans configuration 45
staining techniques 3 transitivity: miscibility in liquids 175,
salting out 130 Stanley, H.E. 120 177
scaling properties 7, 52, 54, 64, 66, 134, starch suspensions 135 translational free energy 81, 132
151, 198 static polarizability 115, 176 transmission of momentum 92
scanning probe microscopy 35, 36 steady-state deformation 108 transparency of a structure 63, 89, 96–8
scaling regimes 60 step-strain modulus 28, 32, 106, 108, 206 transverse momentum 98
scanning tunneling microscopy (STM) steric stabilization 122–7 triblock copolymer 202
35 Stokes diffusion constant 102, 135 triple-contact exclusion 70
scattering 36, 37, 56–61, 73, 97, 153 in semidilute solution 106 turbulent drag reduction 9
intensity 57–8 Stokes formula 93
wavelength 58, 60 Stokes Law constraint 107
scattering assay 37 universal functions 72
strain 28 universal quantities 5, 52, 72, 79, 85,
screening length 75, 99, 102, 128–31, stress 28
136, 169 87–8, 90, 96–7, 106, 109, 140, 159,
stress relaxation and viscosity 105–7 169, 184, 192
secondary minimum 123, 130 structure factor 57, 60
secondary structure 43 universal ratio 72, 85, 97
styrene butadiene copolymer 44 unsaturated polymers 42
sedimentation 92, 116 styrene: sequence of monomers 2
self-avoidance constraint 41, 48, 64, 67, surface energy 155, 156, 161, 163, 165,
86, 143, 144, 214 171 van der Waals attraction 115, 122, 125
self-avoiding chains 61, 65, 74, 87, 90, surface excess 162, 167 van der Waals energy 117
97 surface heterogeneity 159–60 van der Waals forces 123, 175
self-avoiding polymers 41, 61–2, 64–6, surface interaction 33, 162, 168 van der Waals-induced aggregation 124,
72, 76 surface forces apparatus 33, 132 126
self-avoiding Sutherland model 145 surfactants 3, 10, 34, 36, 151, 173 van der Waals interaction 116, 192
self-diffusion 83, 104 assemblies 9 vesicles 209
of entangled polymers 104 interfaces 4, 11, 151, 188, 203 virial 70
self-diffusion constant 102, 135 layers 208 viscometric radius 96–7
self-repelling polymer 7, 72, 184 molecules 3–4, 178–80 viscometry 97
self-repelling random walk 7, 164 nonionic surfactants 180 viscosity 28
semidilute solutions 85–9, 99, 101, 127 in solution 180 viscous dissipation 30, 94, 102, 110, 155,
shear-induced gelation, reversible 5, 10 suspension instability 134 158
shear-induced restructuring 207–10 “Sutherland’s Ghost” model 141, 143–4 viscous stress 109, 169, 197, 198
shear rate 28 symmetric diblock copolymers 198 visible light 37
Singh, Pooran 82 syndiotactic structure 44 volume fraction 85–6
sinking colloids 16–17
slip length 170 Tabor/Israelachvili surface forces water–oil interface 3, 9
smectic liquid crystal 133, 192, 209 apparatus 33 wetting dynamics 158–9
Smoluchowski equation 143 tacticity 43–6 Wilhemeny plate 152
sodium bis(2-ethylhexyl)sulfosuccinate Tanner law of wetting 159 Witten, T.A. 101
(AOT) 179, 195 Taupin, C. 194 worm interpenetration 187
sodium dodecyl sulfate (SDS) 179–81, Taylor formula 197 wormlike chains 48
185, 188, 194, 197, 204, 210 tenuous structures 4, 7, 56, 74, 79–81, wormlike entanglement 204
solid–air interface 156 85, 93, 96–7, 114, 140, 142 relaxation 204–6
solid–liquid interface 157 thermal blob 90 wormlike micelles 46, 182, 191, 203–4,
solutes 161 thermal equilibrium 13–17, 15, 20–1, 24, 206
and interfacial tension 161–3 53, 91, 118, 126, 140, 189
polyatomic solutes 33, 163 thermodynamic radius 97
small-molecule solutes 163 theta solvent 67, 70, 90, 101, 105 Yang, H.-J. 8
solvent hydrodynamics 203 thin-membrane formula 190 Young’s law of partial wetting 157
Soret effect 137 time-dependent relaxation 98 Yukawa potential 129, 132
spallation source 37 time scale: for disentanglement 8
spontaneous curvature 189, 192, 194 Tobacco Mosaic Virus 133 Zimm model 97
spontaneously broken symmetry 133 topological invariant 189 Zimm time 204, 209