Lightscattering Theory and Practice PDF

Light Scattering
Demystified
Theory and Practice
Lars Øgendal
University of Copenhagen 12th September 2017

ii
Contents
Preface 1
1 Introduction 3
1.1 What is light scattering? . . . . . . . . . . . . . . . . . . . . . . 3
2 Simple theory for static light scattering, SLS 9

2.1 The Rayleigh ratio, R . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 What is light? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Why and how do particles scatter light? . . . . . . . . . . . . . . 16
2.4 Dipoles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Scattering of light from one small particle . . . . . . . . . . . . . 18
2.6 Scattering from one small, composite object . . . . . . . . . . . . 24
2.7 Scattering from many identical particles . . . . . . . . . . . . . . 35
2.8 Comparisons with older literature . . . . . . . . . . . . . . . . . 38
2.9 Summing up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3 Determination of size and molecular weight 43

3.1 Concentration effects for small molecules . . . . . . . . . . . . . 44
3.2 Scattering from large molecules . . . . . . . . . . . . . . . . . . 46
3.3 The form factor . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Size determination . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5 Size and molecular weight. How to average in mixtures . . . . . . 56
3.6 Equivalence of the two definitions of Rg . . . . . . . . . . . . . . 62
3.7 Summing up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
iii
0 Contents
4 Simple theory for dynamic light scattering, DLS 69

4.1 71
Intensity fluctuations . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Making sense of fluctuations: The autocorrelation function . . . . 74
4.3 Data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5 Light scattering in practice 99

5.1 Sample preparation . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2 Apparatus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.3 Absolute calibration . . . . . . . . . . . . . . . . . . . . . . . . . 111
A Complex numbers 113
B Data fitting 117

Preface
This is an introduction to light scattering conceived as a measurement technique.

The actual physical phenomenon of light scattering is dealt with to an extent
which is necessary and hopefully sufficient to understand the inner workings of
the technique, to understand its strengths and limitations and which is relevant
to the interpretation of data. There seems to be a lack of simple and accessible
introductions to light scattering so the scientist or student who wants to get ac-
quainted with this field has the choice of either reading a few pages in a physical
chemistry textbook or embark on the the rather inaccessible specialist literature.
A third source of information is the sales prospects from the companies that sell
light scattering instruments but of course this material is influenced by its purpose:
to sell light scattering instruments. So don’t expect any scientific depth or honesty
regarding the limitations, hassles and pitfalls of the method. This introduction to
the field of light scattering is written with the intention of being relatively access-
ible, of course at the cost of mathematical rigour.
1
2 Preface
1
Introduction
1.1 What is light scattering?

In the following we shall treat light itself sometimes as photons, sometimes as
electromagnetic waves whichever gives the simplest description.
The very phenomenon that light that strikes a particle (e.g. a molecule) or a col-
lection of particles, thereby changing its direction, is called light scattering. This
is a somewhat loose definition of light scattering and it is also not quite true, as
we shall see later. For example it turns out that light can actually be scattered in
the forward direction, i.e. without changing direction at all! And even the notion
that light actually strikes anything turns out to be unfruitful.
To keep things simple one could say that light – like any other kind of electromag-
netic radiation – interacts with matter in mainly two different ways:
3
4 1. Introduction
1. Absorption (The photons disappear)
2. Scattering (The photons change their direction)

A third way of interaction exists: The light can change its state of polarisation.
This type of interaction we will disregard here. In the case of absorption it may
happen that light is being (re-)emitted at a different wavelength, i.e. as fluores-
cence or phosphorescence. This phenomenon is usually unwanted when using
light scattering as a measurement technique because it poses the problem to the
experimenter to make sure that the light measured is only scattered and not due
to fluorescence. Also, scattering can be subdivided into elastic and inelastic scat-
tering: If the scattered light has exactly the same wavelength as the incident light
meaning that the scattered photons have exactly the same energy as the incident
photons, the scattering is called elastic. If the photons come out of the scatter-
ing process with a changed energy the scattering process is termed inelastic. This
means that the scattered light will have either a longer or a shorter wavelength than
that of the incident light. Examples of inelastic light scattering are Raman scatter-
ing and Brillouin scattering. Neither of these type of scattering will be dealt with
here. One exception however is quasi elastic light scattering, termed so when the
scattered light has very nearly the same wavelength as the incident light. In prac-
tice this term is reserved to describe scattering of light on moving particles where
the movement of the particles changes the wavelength of the scattered light by the
Doppler effect. The technique called dynamic light scattering, DLS, is based on
this Doppler shift in wavelength and is also called quasi elastic light scattering or
QELS.
The three phenomena, reflection, refraction af diffraction are also also scatter-
ing phenomena but require the scattering particles to be highly ordered. In all of
the following we will assume that the scattering particles are present in a solution
free to move and tumble around by Brownian motion and thus inherently in an un-
ordered state. The miracle of light scattering is that contrary to what one would
think not all information about the particles is wiped out. It turns out to be pos-
sible to derive useful parameters about the dissolved particles like e.g. molecular
weight, size and to some extent, shape.
1.1. What is light scattering? 5
Both of the above mentioned main interaction phenomena have as a result that a
light beam is being attenuated when it passes through a solution1 (see figure 1.1).
The intensity of the light beam is weakened by either absorption or scattering (or
both). In both cases the transmitted intensity decreases exponentially with the
thickness x of the layer of material through which the light has to pass. In the
context of absorption the transmitted intensity I is usually written:
I = I0 · 10−αx
while in the context of scattering it is frequently written as:
I = I0 · e−τ x
The quantities α and τ are called the absorption coefficient and the turbidity, re-
spectively. The difference in the base of the exponential is just a matter of conven-
tion. Just as often the base 10 for the turbidity is used. As mentioned previously
I = I 0 ×10- αx
I = I 0 × e- τ x
Figure 1.1: The transmitted light beam is attenuated due to absorption (top) or scat-
tering (bottom).
we shall assume in the following that the system which scatters light is a solution
1 This also holds true for gasses and solids
6 1. Introduction
of the particles under investigation.
The principles of light scattering (e.e. the technique) is most easily discussed
on the basis of a sketch of a typical setup for the measurement of static light
scattering. One such is shown in figure 1.2. The sample is a solution in a cuvette
which is usually cylindric. A monochromatic light source (i.e. a light source
emitting light of only one wavelength, usually a laser) shines light on the sample.
The intensity or the power of the scattered light is being measured at some known
scattering angle by a detector which may be a photodiode or a photomultiplier
tube.
PC
A/D
con-
verter
Detector
Reference detector
Scattering angle
θ
Laser
Sample cuvette
Semi trans-
parent mirror
Figure 1.2: Sketch of a setup for the measurement of static light scattering, SLS. The
intensity (or the power) of the scattered light is being measured as a function of the
scattering angle θ, i.e. the angle of observation of the detector. The measured intensity
of the scattered light is divided by the intensity of the incident light which consequently
also has to be measured using a reference detector.
While absorption of light (e.g. in the detector) is most easily discussed in terms
of the photon concept, scattering of light is most easily understood in terms of a
wave description of light, i.e. considering light as electromagnetic radiation.
Scattering of light can explain many everyday phenomena like why the sky is
blue, why clouds are white and why they are visible at all while the water vapour
that creates them is invisible, why milk appears white even though it consists of
1.1. What is light scattering? 7
completely transparent components, etc. . . . . But as previously mentioned light

can also be used to gain information about molecules (usually macromolecules)
and particles in suspension.
On the following pages we will describe an discuss two measurement techniques,

namely static light scattering, SLS and dynamic light scattering, DLS. The tech-
niques will be described in that order because it is necessary to understand SLS in
order to understand properly DLS. At this point, however, we can already reveal
that the two techniques are complementary in the sense that they take advantage
of two different an completely independent features of the scattered light: SLS
measures the average intensity of the scattered light at many different scattering
angles and uses this to derive information about molecular weight, particle size,
particle shape and particle interactions. DLS, on the other hand, disregards the
intensity of the scattered light but uses the temporal fluctuations in the scattered
light to derive spectra of typical fluctuation times which in turn yield information
about the diffusion coefficient(s) of the particles and the indirectly about their size.
The difference is shown in figure 1.3.
10
8
Dynamic light scattering
t
6
Intensity
< I > Static light scattering

2
0
0 1 2 3 4 5
time (ms)
Figure 1.3: Light scattered from a solution of macromolecules has a mean intensity hIi
that reects the molecular weight of the particles while the uctuations in the intensity
have a characteristic uctuation time τ that reects the diusion coecient of the particles.
8 1. Introduction
So, in conclusion about the difference between static and dynamic light scattering
one can say:
SLS : Static light scattering employs measurement of the intens-

ity of the scattered light at many different angles (typically
10 – 100). The intensity is usually an average over approx-
imately one second or more. The information about the mo-
lecular weight and size of the molecule lies in the intensities
themselves at the different scattering angles.
DLS : Dynamic light scattering employs measurement of long

time series of mean intensity of the scattered light where
the averaging is over very short time intervals, typically
100 ns. Averaging over such short time intervals ensures
that the magnitude of the intensity fluctuations within the
time series become significant. The information about the
molecules (diffusion coefficient or size) lies in the typical
fluctuation times for the scattered intensity.
2
Simple theory for static light scat-
tering, SLS
Since the late 19’th century light has been used to determine the size of particles
(Tyndall 1869, Rayleigh 1871 and 1881) and later in the 20’th century to determ-
ine the molecular weight of macromolecules (Debye 1944 and 47). The technique
employed is called static light scattering or SLS. The method is based on the fact
that when light passes through a solution of molecules (particles) some portion
of the light will be scattered in all directions. If all of the light is scattered the
solution (or suspension) will appear completely opaque. This is the case with e.g.
milk which appears white because light is scattered by microscopic fat globules
and casein aggregates, called casein micelles although they are not micelles in the
physical meaning of the term. In other situations the scattered fraction of the light
9
10 2. Simple theory for static light scattering, SLS
is so small that the solution appears completely transparent to the naked eye. This
would be the case if one looked at a solution of BSA (bovine serum albumin), a
protein with a molecular weight of approx. 66400 g/mol at a concentration of,
say, 1 g/L. In this case it requires a very intense light source (i.e. a laser) to see
the scattered light. With a 10 mW laser the scattered light shows a glowing trace
in the solution (see figure 2.1.) It is this latter situation of weak scattering which
Figure 2.1: The laser to the right sends out an intense beam of light which - in principle
- is invisible when seen from the side. When the laser beam passes through a solution of
large molecules, like e.g. BSA, some of the light is scattered by the molecules and becomes
visible from the side as a glowing trace.
is relevant when light scattering is used to determine molecular weight and size of
molecules.
In order to be useful for quantitative light scattering the light itself has to fulfil
two conditions:
• The light has to be monochromatic, i.e. to have a well-defined wavelength.
• The light has to be in a collimated beam, i.e. the rays have to be parallel.
Both conditions are ideal ones: In practice one will have to accept some distri-
11
bution of wavelengths and some angular divergence of the rays. Ordinary white
light consists of all wavelengths between approximately 390 and 700 nm and is
therefore highly non-monochromatic. In the early days of light scattering one had
to employ colour filters, diffraction gratings or prisms to single out a narrow se-
lection of wavelengths from the light source. This of course meant that the useful
intensity of the light source was dramatically decreased. The more narrow the
selected spectrum of light the weaker the (remaining) light would be. As for the
collimation (i.e. the parallelity) this was achieved by having a very small light
source e.g. by screening a larger source with slits or diaphragms subsequently
making the rays parallel by a system of lenses. Again, the limiting factor for the
parallelism is the size (the extension) of the light source which in turn means that
a higher degree of collimation would necessarily be linked with a relatively small
and therefore effectively weaker light source. As the precision of light scattering
measurements is related to the monochromaticity and the collimation of the light
used1 it is seen that a trade-off has to be made because a weak light source in itself
impairs the precision of the measurements.
This dilemma is only of historical interest today because the present day light
sources are lasers which by their nature emit inherently monochromatic light at
a very high intensity. And laser light is also easily collimated without loss of
intensity. A third benefit from using lasers is that the beam can be made such
that it has the desired direction of polarisation. This is a technical advantage
as we shall see later but has the minor drawback that the equations used for the
interpretation of light scattering data have to be used with great care if they are
taken from literature prior to the advent of the laser (1960). Always check if the
equations pertain to un-polarised light or polarised light. We shall briefly mention
the old equations later ?? for reference and warning.
1 The monochromaticity and collimation are much more important if the purpose of
the light scattering measurement is size determination and not determination of molecular
weight.
2.1 The Rayleigh ratio, R
When a light scattering experiment is done on a solution (i.e. a suspension of

particles) the purpose is of course to deduce properties of the solution and not
have this mixed up with properties of the apparatus. It is evident that if the laser
is replaced by one of twice the power the intensity of the scattered light will be
doubled. And if the distance between the scattering solution and the detector is
doubled the measured intensity will be reduced by a factor of four as the observed
intensity is inversely proportional to the distance between the source and the re-
ceiver. In order to obtain a measure which is independent of apparatus character-
istics but only depends on the properties of the system one defines the Rayleigh
ratio, Rθ , for the solution under investigation:
Is (θ) · r2
Rθ = (2.1)
I0 · Vs (θ)
where the subscript θ indicates that the Rayleigh ratio is a function of the scat-
tering angle, Is (θ) is the intensity of the scattered light measured at the angle of
observation θ cf. figure 1.2, I0 is the intensity of the laser used, Vs (θ) is the scat-
tering volume (i.e. the volume which is illuminated by the laser while at the same
time being visible for the detector (see figure 2.2), and r is the distance from the
scattering volume to the detector. It turns out that the scattering angle θ, although
easy to envisage, makes the equations describing light scattering cumbersome to
work with. So, when we get to section 2.6 we abandon the scattering angle θ and
instead introduce the so-called scattering vector ~q (see later in this chapter and
figure 2.11) the length of which, q can be calculated as q = (4πn/λ) sin(θ/2).
When the scattering angle is effectively replaced by the length q of the scattering
vector it then becomes more natural to conceive the Rayleigh ratio as function of
q:
Is (q) · r2
R(q) = (2.2)
I0 · Vs (q)
We shall see in the following sections that R(q) is a function which depends on:
2.1. The Rayleigh ratio, R 13
on
visi
Apertures or slits of
a
re
ra
cto
te
De
d2
Laser beam d1 q
VS
Figure 2.2: Due to the apertures usually present in a detector it only sees light that enters
within a certain narrow eld of observation (approximately a cylinder) with a diameter
of d2 whereas the illuminated volume is approximately a cylinder of diameter d1 . The
intersection between these two volumes is called the scattering volume, Vs . The scattering
volume attains its minimum value, V0 , when the laser beam and the angle of observation
are perpendicular to each other, i.e. when θ = 90◦ . The frequently cited relationship
Vs (θ) = V0 / sin(θ) holds only approximately.
• the molecular weight of the suspended particles (molecules)
• the concentration of the suspended particles (molecules)
• the refractive index of the pure solvent
• the refractive index of the suspended particles (molecules)
• the size of the suspended particles (molecules)
• interaction forces between the suspended particles

2.2 What is light?

As mentioned at the beginning, the picture of light scattering as photons bouncing
off the particles in the solution is not fruitful. Instead, classic electromagnetic the-
ory (EMT) where light is treated as electromagnetic waves is much more straight-
forward. Light is described as an electromagnetic wave, an electric wave and a
magnetic wave mutually generating each other as they propagate with the speed
of light (see figure 2.3).
Figure 2.3: An electromagnetic wave propagates along the x-axis. The magnetic eld ~
B
~
and the electric eld E are perpendicular to each other and to the direction of propagation.
The electric eld is strictly vertical here, giving it only a z-component.
Light which propagates along the direction of the x-axis of some coordinate sys-
tem with the electric field oscillating in the direction of the z-axis is said to be
vertically polarised (i.e if the z−axis is in the vertical direction, of course). If the
vertically polarised light is of one colour, i.e. of one wavelength λ or equivalently
of one frequency ν, its electric field can be described by the following equation:
E(x, t) = Ez (x, t) = E0 cos(2πν(t − x/c) + φ) (2.3)
where ν is the frequency of the light, c is the speed of light and E0 is the amplitude
of the electric field component of the light. The quantity φ is just a constant which
defines the phase of the cosine function at time t = 0 and at x = 0. In light
2.2. What is light? 15
scattering we normally don’t bother about the magnetic field as this normally does
not interact with organic matter.
In order to simplify the appearance of equations it is customary to define the
wavenumber k as
2π
k=
λ
where λ is the wavelength of the light in the medium through which the light
propagates. The equation can also be written
2πn
k=
λ0
where λ0 is the wavelength of the light in vacuum and n is the refractive index of
the medium. In the case of light propagating in water the refractive index is just
n = 1.33. Also, for simplifying purposes, we define the cyclic frequency ω of the
light as
ω = 2πν
As ν/c = 1/λ we can finally rewrite equation 2.3 as:
E(x, t) = Ey (x, t) = E0 cos(ωt − kx + φ) (2.4)
This equation is easily generalised to propagation in any direction and a point of

observation not necessarily lying on the x-axis: First of all the wavenumber k
is replaced by the wavevector ~k defined as a vector pointing in the direction of
propagation and with a length |~k| = 2πn
λ0
. And the point of observation is now
given by a vector ~r instead of just an x-coordinate. The generalised version of
equation 2.4 then reads:
E(~r, t) = E0 cos(ωt − ~k · ~r + φ) (2.5)
According to EMT the intensity of the light is given by the time average of the
square of the electric field:
I0 (x) = 0 chE 2 (x, t)i (2.6)
where 0 = 8.85 · 10−12 F · m−1 is the so-called vacuum permittivity and c =

3.0 · 108 m · s−1 is the speed of light in vacuum. Inserting the expression for
E(x, t) we get:
I0 (x) = 0 chE 2 (x, t)i = 0 chE02 cos2 (ωt−~k ·~r)i = 0 cE02 hcos2 (ωt−~k ·~r)i (2.7)
1
As the mean value of cos2 (ωt − kx) calculated over one cycle equals 2
and
15 −1
ω ≈ 3 · 10 radians · s for a typical HeNe laser the averaging over even some
extremely short time interval like e.g. one nanosecond would mean averaging over
approximately half a million cycles. This means that even if this short averaging
time does not contain a whole number of cycles the result would still be extremely
close to 12 . Hence we get:
1
I0 (x) = 0 cE02 (2.8)
2
2.3 Why and how do particles scatter light?

In order to understand why and how particles or molecules can scatter light it turns
out that one needs to understand what an electric dipole is. Having touched upon
this subject it turns out to be a good idea to answer the questions in several steps:
First we shall look at how light is scattered from one very small (i.e. point-like)
particle. Next, how light is scattered from many small particles. And this latter
issue will be divided into a discussion on how light is scattered if the particles
are non-interacting (an ideal solution) and a discussion on how particles in non-
ideal solutions scatter light. Fundamentally, of course, light is scattered by the
same laws in both cases but the equations describing the scattering quantitatively
are different. After having looked at scattering from very small particles we turn
towards scattering from large particles.
2.4 Dipoles
An electric dipole is an object which has an asymmetrical distribution of charge,
usually such that the object has one end with positive charge and the other end
with negative charge (see figure 2.4). The dipole moment µ is defined – in the
simple case shown in the figure – as µ = d · Q, i.e. as the separation between
the positive +Q and the negative charge (−Q) multiplied by the magnitude of the
2.4. Dipoles 17
-Q + 300 V
-
-- -
d E
+ ++ +
+Q - 300 V
µ = d·Q
Figure 2.4: On the left side is shown an electric dipole. The dipole moment µ is dened
as the product of half the charge dierence (between the two ends) and their mutual
distance. On the right is shown how a particle becomes a dipole when under inuence of
an external electric eld.
charge. The dipole moment can be either permanent or induced, i.e. caused by
an external electric field. When it comes to scattering of light it is only induced
dipole momenta that have relevance. In the more general dipole case shown on
the right in figure 2.4 the dipole moment must be defined by an integral 2 . What is
interesting about the induced dipole moment is that its magnitude depends on the
strength of the external field E in a linear way:
µ = α · 0 · E
where α is the so-called polarisability of the particle. The vacuum permittivity,

0 , is included due to tradition3 . With this definition of the polarisabilityit has the
SI-unit m3 . The linearity holds at least as long as the electric field is not very
strong. Lasers used in light scattering do not exceed the limit of linearity even
when the beam is focussed. The polarisability α of the particle is a measure of
how easily an external field can move charges within the particle. The larger the
value of α the further the charges will move with a given external field strength.
This description of the behaviour of dipoles refers to particles in vacuum. But
2 The dipole moment is actually a vector, µ
~. For a distribution of n point charges Qi it
Pn RRR
can be calculated as µ
~ = i Qi r~i and for a continuous distribution as µ
~ = ~r ρ(~r)dV
where ρ(~r) is the charge density (coulomb per cubic meter). The integration is extended
over the body that contains the charge.
3 Sometimes the vacuum permittivity is excluded from the dening equation which then
reads : µ = α · E. This of course has the consequence that the polarisabilityof a given
material will have a dierent value and a dierent unit
light scattering experiments are usually carried out on particles in solution, i.e.
particles embedded in a medium. How to deal with this is described in figures 2.5
and 2.6. Suppose an electric field is present in a medium (figure2.5 a). Charges
may move in the medium under the influence of the electric field but no net charge
is generated anywhere. Now, take out a ”plug” of the medium and put it in the
electric field (figure2.5 b) without the rest of the medium. This plug will be polar-
ised in the electric field with the net result that some charge will be removed from
one end of the plug and put at the other end of the plug. The plug will have surface
charges at the ends. If we do the opposite, i.e. remove the plug from the medium
(figure2.6 a) leaving a hole in it, the hole will also be polarised but exactly oppos-
ite to the polarisation of the plug. It will have charges on the inner surface that
exactly match the surface charges on the plug. Because if the plug is put back
into the hole all charges will cancel as in figure figure2.5 a. Now, put a plug of
the same size and shape but made of a different material than the medium into the
hole in the medium. If this plug has a polarisability αplug which is different from
that of the medium αmedium the net dipole moment of the plug in the medium will
be µnet = αplug · 0 · E − αmedium · 0 · E or
µnet = (αplug − αmedium ) · 0 · E
The quantity αplug − αmedium is called the excess polarisability of the particle. It
is this quantity which governs the light scattering of particles in solution. When
we refer to the polarisabilityof particles from now on it should be understood as
the excess polarisability.
2.5 Scattering of light from one small particle

As is evident from the previous discussion of polarisabilityit is most convenient
to describe light as a propagating electromagnetic wave rather than beam of in-
dividual photons. The way the scattering comes about is that the electric field of
the incoming light moves the charges in the particles (molecules) thereby turn-
ing them into dipoles (see figure 2.7). The propagating electromagnetic wave
consists of an electric and a magnetic field that oscillate perpendicularly to each
2.5. Scattering of light from one small particle 19
+ 300 V
E
- 300 V
- - -
+ 300 V
- -
E ++ +
++
- 300 V
b
Figure 2.5: An electric eld is present in a medium in gure a. Charges may move in the
medium under the inuence of the electric eld but no net charge is generated anywhere.
A plug is taken of the medium and put it in the electric eld gure b while the rest of
the medium is removed. This plug will be polarised in the electric eld with the net result
that some charge will be removed from one end of the plug and put at the other end of
the plug, i.e. the plug will acquire surface charges at the ends.
+ 300 V
+++
+ +
E -
- - -
-
- 300 V
+-+--+
+ 300 V
+ -- ---+
-
E + +
+++++ -
-+-
+
--
- 300 V
b
Figure 2.6: Here the plug is removed from the medium (gure a) leaving a hole in it.
The hole will also be polarised but exactly opposite to the polarisation of the plug. It will
have charges on the inner surface that exactly match the surface charges on the plug that
was removed. Because if the plug is put back into the hole all charges will cancel as in
gure 2.5 a.
In gure b a plug of the same size and shape but made of a dierent material than the
medium is put into the hole in the medium. If this plug has a polarisabilityαplug which
is dierent from that of the medium αmedium the net dipole moment of the plug in the
medium will be µnet = αplug · 0 · E − αmedium · 0 · E because it reects the dierence in
the surface charge of the hole and of the new plug.
other and perpendicularly to the direction of propagation. The oscillations have

the frequency ν (and hence the cyclic frequency ω = 2πν). Only the electric field
is (usually) of importance for light scattering as this is what exerts a force on the
charges in the molecule (figur 2.7).
t=0
t = l/(2c)
Figure 2.7: The external, oscillating electric eld E displaces the electric charges in
the particle rendering it negative at one end and positive at the other end. Hereby the
particle becomes an oscillating dipole with a dipole moment that can be written as
µ = µ(t) = αE = E0 cos(ωt − kx) = µ0 cos(ωt − kx) where x is the position of the
molecule.
The charges in the molecule move in sync with the incoming electric field turn-
ing the molecule into an oscillating dipole, oscillating with the same frequency
as the light that shines on it. Not only do the charges in the molecule move in
sync with the external electric field they also move simultaneously in the same
direction within the molecule because the electric field has (practically) the same
magnitude everywhere in the molecule when the molecule is sufficiently small
compared to the wavelength of the light. Consequently, according to the theory
of electromagnetic radiation (EMT), the molecule emits electromagnetic radiation
into all directions (with an intensity which depends on the direction).
We shall here consider the situation where the incoming light is polarised 4 mean-
ing that the electric field always oscillates in the same plane (see figure 2.8)
Now, let’s make things quantitative. Remember that the molecule or particle is
4 The reason for looking at polarised light is that in practice the light source is now
always a polarised laser. In the earlier days of light scattering (up to the nineteen sixties)
light sources were generally unpolarised and consequently the equations describing light
scattering took this into account. This means that care should be taken if equations from
older literature is used to interpret light scattering data from modern equipment.
r
E0 Es,1(r)
f
y
q
Molecule
Figure 2.8: Light polarised in the xz−plane hits a molecule much smaller than the
wavelength of the light. The electric eld of the light shakes the electrons of the molecule
thereby turning it into an oscillating dipole which emits radiation in all directions. The
intensity of the emitted light depends on the direction of emission given by the two angles
θ and φ.
assumed to be much smaller than the wavelength λ of the incoming light. For
simplicity the particle is placed at the origo of our coordinate system. The elec-
tric field of the light that impinges on the molecule can be described as a wave
propagating along the x−axis according to equation 2.4.
The electric field strength Es,1 (r) of the radiation scattered from one particle
(hence the number 1 in the subscript) at the distance r from the molecule (cf.
figure 2.8) is according to EMT given by:
!
πα sin φ
Es,1 (r) = E0 · cos(ωt − kr) (2.9)
rλ2
where λ is the wavelength of the light (in the medium surrounding the molecule),
α is the polarisabilityof the molecule and r is the distance from the molecule to the
detector. Note that Es,1 (r) does not depend on the angle θ relative to the x-axis.
Using equations 2.9 and 2.6 we obtain the intensity of the scattered light:
2 2 2
1 2 π α sin φ
Is,1 (r) = 2
h0 cEs,1 (r)i = 0 cE0 · (2.10)
2 r2 λ4
or, applying also equation 2.8:
π 2 α2 sin2 φ
Is,1 (r) = I0 · (2.11)
r 2 λ4
In practice most light scattering instruments have the detector placed in the xy−plane
so from now on we shall assume that the angle φ = 90◦ . Hence we get the fol-
lowing expression, valid for the scattered intensity from a small molecule:
π 2 α2
Is,1 (r) = I0 · (2.12)
r 2 λ4
Here we see, not surprisingly, that the intensity of the scattered light at the pos-
ition of the detector is proportional to the intensity I0 of the incoming light and
inversely proportional to the square of the distance from the scattering molecule
to the detector. What seems perhaps more surprising is the fact that the scattered
intensity is inversely proportional to the fourth power of the wavelength of the
impinging light. This means that blue light (λ0 ≈ 480 nm is scattered approxim-
ately four times more effectively than red light with a wavelength of λ0 ≈ 690 nm.
This fourth power dependence on the wavelength is the well-known explanation
why the sky is blue.
This kind of scattering where the scattering particles are much smaller than the
wavelength of the light5 is called Rayleigh scattering and is characterised by be-
ing isotropic i.e. equally intense in all directions relative to the direction of the
incoming light (it is independent of the scattering angle θ c.f figure 1.2) and the
scattered light also has the same wavelength as the incoming light. Looking at
figure 2.9 it is intuitively clear that α must be proportional to the ”size” of the
molecule (i.e. volume or molecular weight) as putting two small particles each
with the molecular polarisability α1 next to each other will give a total dipole mo-
ment of µ = (2Q) · d whereas putting the two particles end-to-end will give a total
dipole moment of µ = Q · (2d). In both cases the total dipole moment is simply
twice the dipole moment of one particle.
Following the same line of thought is is evident that n particles have the dipole
moment µ = nµ1 whereby the compound particle gets the polarisability α = nα1 .
5 actually also the polarisabilityα needs to be small. We shall return to this point later
+ 300 V
-Q
-Q -Q d
E d
+Q +Q
d
+Q
-Q
+Q
- 300 V
Figure 2.9: Two small particles each have the dipolemoment µ0 = Q · d. Putting the two
particles side by side gives the dipole moment µ = (2Q) · d whereas putting them end to
end gives µ = Q · (2d), remembering that the charges +Q og -Q i the middle cancel out.
The result is in both cases that µ = 2µ1 . The argument can easily be generalised to n
particles.
As n = M/M1 where M is the molecular weight of the compound particle and

M1 molecular weight of the small particle we get that α = (α1 /M1 ) · M . It
therefore follows from equation 2.12 that the intensity of the light scattered from
a single molecule is proportional to the the square of the molecular weight. For
molecules that are large compared with the wavelength of the light this argument
does no longer hold as large molecules in general don’t have a well-defined state
of polarisation. This situation we will now begin to look at but will only finish the
discussion in section 3.3.
2.6 Scattering from one small, composite ob-

ject
We shall here see that the way light is scattered from a particle that is not infinitely
small is determined by the interference between scattered electromagnetic waves
originating from different points of the composite particle. For those familiar
with the use of complex numbers the calculations can be done quite simply and
elegantly. But first we shall see how this interference comes about without the use
of complex number. Then, in section 2.6, we shall see how things can be worked
out more simply with the use of complex numbers.
2.6. Scattering from one small, composite object 25
Without the use of complex numbers

This section can be skipped if one prefers the description based on the use of com-
plex numbers.
The terms ”particle” and ”molecule” are as usual interchangeable. What is crucial
for the calculations and derivations that follow is that we can think of a particle as
being composed of smaller ”point-like” particles all having the same polarisabil-
ity, α.
When it comes to the description of the inner workings of light scattering it turns
out that the scattering angle θ (cf. figures 1.2 and 2.2) is an inconvenient quant-
ity to work with. Instead one uses a less intuitive quantity, the previously men-
tioned scattering vector, ~q. The definition of the scattering vector is based on the
wavevector ~k which, as you may recall is a vector pointing in direction of propaga-
tion of the light (the electromagnetic wave). The length of the wavevector is, as
previously mentioned, |~k| = 2π/λ = 2πn/λ0 (see section 2.5) and is called the
wavenumber. When a photon with the wavevector ~kin hits a particle and emerges
from the collision with a different direction (i.e. it is scattered) its wavevector
changes accordingly into ~kout . The situation is depicted in figure 2.10.
The change in the wavevector is called the scattering vector:
~q = ~kout − ~kin (2.13)

The situation is illustrated in figure 2.10.
The length q of the scattering vector is easily calculated from figure 2.11. It is seen
that q = 2a where a = k · sin 2θ and the length of the wavevector is k = 2π
λ
. Hence
4π θ
we get the length of the scattering vector expressed as q = λ
sin 2
. Here, as usual,
λ is the wavelength of the light in the medium surrounding the particle. However,
it is customary to express the length of the scattering vector using the wavelength
of the light in vacuum λ0 (in practise the same as in air). The connection is that
λ = λ0 /n where n is the refractive index of the medium. You may remember
that the refractive index of a material expresses the factor by which the velocity
of light in the material is reduced relative to the velocity of light in vacuum:
cvacuum
n= (2.14)
cmedium
k in k in
q q
k out k out q
Figure 2.10: The gure on the left shows that the scattering of light on the particle
changes the wavevector from ~kin to ~kout . We assume that only the direction and not the
wavelength changes as a result of the collision. Hence the length of the wavevector remains
unchanged. The gure on the right shows the same situation in abstract vector form.
k
q /2
q /2
a
k a q
Figure 2.11: The gure shows the lengths of the vectors.It is seen that the length q of
the scattering vector is q = 2a a = k · sin θ2 . As the scattered light has the same
where
wavelength as the incident light we know that |~ kout | = |~kin | ≡ k , where the length of the
2π 4π θ
wavevector is k =
λ . Hence the length of the scattering vector is q = λ sin 2 . Note, that
λ is the wavelength of the light in the medium embedding the particle. Introducing the
vacuum wavelength of the light λ0 and the refractive index n of the medium the length of
4πn θ
the scattering vector can be written q =
λ0 sin 2
subunit
Finally we can write the length q of the scattering vector in its most common form:
4πn θ
q= sin (2.15)
λ0 2
where θ is the scattering angle (cf. figure 1.2)
Now, let us look at two small particles being hit by the same electromagnetic wave
(figure 2.12). At the position of the detector the scattered electric fields emerging
k in
r k out
R
E2
E1
Figure 2.12: Monochromatic light with the wavelength λ and the wavevector ~kin im-
pinges on two small particles separated by the vector ~r. The detector that detects the
scattered light is positioned at the distance R from the origo of the coordinate system.
The distance R is very much larger than the separation |~r| between the two particles. The
light scattered from the two particles has the electric eld strength contributions E1 and
E2 , respectively. These eld strengths have the same amplitude because the distance from
the each particle to the detector is essentially the same. However, it may happen that the
two elds do not have the same phase meaning that the don't attain their maximum value
at the same time.
from the two particles in figure 2.12 can be described as plane waves with the
same amplitude because the detector is at a very large distance from the particles.
In fact the scattered field emerging from each of the particles is a spherical wave
but at a sufficiently large distance from its centre and within a relatively small area
such a wave can be regarded as being plane:
~ · ~kout )
E1 = E0 cos(ωt − R (2.16)
~ · ~kout + ∆φ)
E2 = E0 cos(ωt − R
where the phase difference∆φ (not to be confused with the formerly used angle
φ between the direction of the scattered light and the z−axis of the coordinate
system) appears because the light scattered from the two particles travels different
distances in order to reach the detector. But not only does the light travel different
distances to reach the detector, the impinging light waves have also traveled dif-
ferent distances before they hit the particles. The total path length difference ∆s
can easily be shown to be ∆s = ~r · ~q/k, where k is the wavenumber and ~r is a
vector from one of the particles to the other. This difference in pathlength of the
light gives rise to the phase difference ∆φ between waves received by the detector
from the two particles:
∆φ = ~r · ~q (2.17)
The intensity on the detector Itotal of the total field E = E1 + E2 can the be
calculated using equations 2.16 and 2.6:
Itotal = 0 chE 2 i
~ · ~kout + ∆φ) 2 i

~ · ~kout ) + E0 cos(ωt − R
= h E0 cos(ωt − R
~ · ~kout ) + cos2 (ωt − R
= 0 cE02 hcos2 (ωt − R ~ · ~kout + ∆φ) +
~ · ~kout ) · cos(ωt − R
2 cos(ωt − R ~ · ~kout + ∆φ)i (2.18)
The two first terms in the final expression of equation 2.18 (the terms containing
1
cos2 ) both have the mean value 2
as explained earlier (page 16). The last term in
equation 2.18 can be calculated using a well known trigonometric identity:
1
cos x · cos y = · [cos(x + y) + cos(x − y)]
2
yielding:
~ · ~kout ) · cos(ωt − R
Itotal = 2 cos(ωt − R ~ · ~kout + ∆φ)
~ · ~kout + ∆φ) + cos(∆φ)
= cos(2ωt − 2R (2.19)
The mean value of the first term in the bottom line of equation 2.19 is 0. The
~ ~kout +∆φ) calculated
reason for this is, again, that the mean value of cos(2ωt−2R·
over one cycle equals 0 and ω ≈ 3 · 1015 radians · s−1 for a typical HeNe laser.
So averaging over even an extremely short time interval like e.g. one nanosecond
would mean averaging over approximately half a million cycles. This means that
even if this short averaging time does not contain a whole number of cycles the
result would still be extremely close to 0. The last term is constant and therefore
simply has the mean value cos(∆φ)
Summing up we get:
Itotal = 0 cE02 · (1 + cos(∆φ)) = 2 · Is, 1 · (1 + cos(∆φ)) (2.20)
Equation 2.20 shows that the intensity of the light scattered from two particles can
be anything from 0 to 4 times as high as the intensity scattered from one single
particle. This result we are now going to generalise to an arbitrary number of
particles:
Instead of assigning the field E2 the extra phase ∆φ in equation 2.19 one could just
as well have assigned E1 the phase φ1 and E2 the phase φ2 such that φ2 −φ1 = ∆φ.
We want to generalise equation 2.20 so, instead of looking at two small particles
we shall consider a ”molecule” (a compound particle) consisting of n subunits
(e.g. amino acids or monosaccharides). For simplicity we shall assume that
all subunits are identical. We number these subunits with an index i, such that
i = 1, 2, . . . , n, (see figure 2.13) At the detector position the total scattered elec-
tric field Es, n (from the n subunits) is then simply the sum of the contributions:
n
(2.21)
X
Es, n = E0 cos(ωt + φi )
i=1
i
rij
j
ri
z
rj
y
Figure 2.13: The molecule consists of n identical subunits and light is impinging on it
as shown in gure ref2particle. At the detector position the eld contributions from the
dierent subunits all have the same amplitude E0 but dierent phases φ1 , φ2 ,. . . , φn . The
positions of the individual subunits are given by the position vectors ~r1 ,~r2 . . . ,~rn
Using again equation 2.6 the total scattered intensity Is, n is then given by:
Is, n = 0 chEs,2 n i
n n
! !
X X
= 0 ch E0 cos(ωt + φi ) · E0 cos(ωt + φj ) i
i=1 j=1
n X n
2
X
= 0 cE0 h cos(ωt + φi ) · cos(ωt + φj )i
i=1 j=1
n X n
1
2
(2.22)
X
= 0 cE0 h [cos(2ωt + φi + φj ) + cos(φi − φj )]i
j=1 i=1 2
Here it is seen that the terms cos(2ωt + φi + φj ) have a mean value of 0 (see
top of page 29 for expalnation). The remaining terms are constant if the phases
φ1 , φ2 , . . . are time independent, which we shall assume for now. This just means
that the molecule lies still and doesn’t rotate. Hence we get:
n X
n
(2.23)
X
Is, n = Is, 1 cos(φj − φk )
j=1 k=1
where, as usual, Is, 1 = 0 cE02 denotes the intensity of the light scattered from
one single subunit of the molecule. If the overall size, d, of the molecule is very
much smaller than the wavelength of the light meaning that d · k = d · 2π/λ ≈ 0,
the the pase differences are |φj − φk | = |~rj · ~q − ~rk · ~q| = |~rjk · ~q| ≤ d · 2k ≈ 0.
Hence for all terms in equation 2.23 we get that cos(φj − φk ) ≈ cos(0) = 1. As
this double sum contains a total of n2 terms we get:
Is, n ≈ Is, 1 · n2 (2.24)
and, as n = M/M1 where M is the molecular mass and M1 is the molecular mass
of a subunit, we get:
Is, n ≈ (Is, 1 /M12 ) · M 2 (2.25)
This is basically the same result that was obtained in section 2.5, namely that the
”scattering power” of a single small molecule is proportional to the square of the
molecular mass. It is important to note that this result is valid when the molecule
is so small that the phase differences between light scattered from different parts
of the molecule are essentially 0. How to relax this restriction will be treated in
section 3.3 dealing with scattering from large molecules.
With the use of complex numbers

This section is an alternative to section 2.6. The description of scattering is based
on the complex number description of sinusoidal waves. The use of complex num-
bers leads to a strong simplification of the calculations leading eventually to the
same equations as we saw in section 2.6. Appendix A describes very briefly how
complex numbers are used. The present section has more or less the same wording
as section 2.6 and can thus be read independently from the previous section.
When it comes to the description of the inner workings of light scattering it turns
out that the scattering angle θ (cf. figures 1.2 and 2.2) is an inconvenient quant-
ity to work with. Instead one uses a less intuitive quantity, the previously men-
tioned scattering vector, ~q. The definition of the scattering vector is based on the
wavevector ~k which, as you may recall is a vector pointing in direction of propaga-
tion of the light (the electromagnetic wave). The length of the wavevector is, as
previously mentioned, |~k| = 2π/λ = 2πn/λ0 (see section 2.5) and is called the
wavenumber. When a photon with the wavevector ~kin hits a particle and emerges
from the collision with a different direction (i.e. it is scattered) its wavevector
changes accordingly into ~kout . The situation is depicted in figure 2.10.
The change in the wavevector is called the scattering vector:
~q = ~kout − ~kin (2.26)
The situation is illustrated in figure 2.10.

The length q of the scattering vector is easily calculated from figure 2.11. It is seen
that q = 2a where a = k · sin 2θ and the length of the wavevector is k = 2π
λ
. Hence
4π
we get the length of the scattering vector expressed as q = λ
sin 2θ . Here, as usual,
λ is the wavelength of the light in the medium surrounding the particle. However,
it is customary to express the length of the scattering vector using the wavelength
of the light in vacuum λ0 (in practise the same as in air). The connection is that
λ = λ0 /n where n is the refractive index of the medium. You may remember
that the refractive index of a material expresses the factor by which the velocity
of light in the material is reduced relative to the velocity of light in vacuum:
cvacuum
n= (2.27)
cmedium
Finally we can write the length q of the scattering vector in its most common form:
4πn θ
q= sin (2.28)
λ0 2
where θ is the scattering angle (cf. figure 1.2)
Now, let us look at two small particles being hit by the same electromagnetic wave
(figure 2.12). At the position of the detector the scattered electric fields emerging
from the two particles in figure 2.12 can be described as plane waves with the
same amplitude because the detector is at a very large distance from the particles.
In fact the scattered field emerging from each of the particles is a spherical wave
but at a sufficiently large distance from its centre and within a relatively small area
such a wave can be regarded as being plane:
(2.29)
~~
E1 = E0 ei(ωt−R·kud )
~ ~kud +∆φ)
i(ωt−R·
E2 = E0 e
where the phase difference∆φ (not to be confused with the formerly used angle
φ between the direction of the scattered light and the z−axis of the coordinate
system) appears because the light scattered from the two particles travels different
distances in order to reach the detector. But not only does the light travel different
distances to reach the detector, the impinging light waves have also traveled dif-
ferent distances before they hit the particles. The total path length difference ∆s
can easily be shown to be ∆s = ~r · ~q/k, where k is the wavenumber and ~r is a
vector from one of the particles to the other. This difference in path length of the
light gives rise to the phase difference ∆φ between waves received by the detector
from the two particles:
∆φ = ~r · ~q (2.30)
The intensity on the detector Itotal of the total field E = E1 + E2 can the be
calculated using equation 2.6 plus the fact that |E|2 = E · E ∗ , where E ∗ denotes
the complex conjugate of E:
1
Itotal = 0 c|E|2
2
1
= 0 cE · E ∗
2
~~ ~~ ~~ ~~
= (E0 ei(ωt−R·kud ) + E0 ei(ωt−R·kud +∆φ) ) · (E0 e−i(ωt−R·kud ) + E0 e−i(ωt−R·kud +∆φ) )
1
= 0 cE02 (1 + 1 + ei∆φ + e−i∆φ )
2
1
= 0 cE02 (2 + 2 cos(φ)) (2.31)
2
Equation 2.31 can also be written:
Itotal = 0 cE02 · (1 + cos φ) = 2 · Is, 1 · (1 + cos(∆φ)) (2.32)
which is the same result obtained in the previous section, equation 2.20.
Equation 2.32 shows that the intensity of the light scattered from two particles can
be anything from 0 to 4 times as high as the intensity scattered from one single
particle. This result we are now going to generalise to an arbitrary number of
particles:
Instead of assigning the field E2 the extra phase ∆φ in equation 2.19 one could just
as well have assigned E1 the phase φ1 and E2 the phase φ2 such that φ2 −φ1 = ∆φ.
We want to generalise equation 2.20 so, instead of looking at two small particles
we shall consider a ”molecule” (a compound particle) consisting of n subunits
(e.g. amino acids or monosaccharides). For simplicity we shall assume that all
subunits are identical. We number these subunits with an index i, such that i =
1, 2, . . . , n, (see figure 2.13). At the detector position the total scattered electric
field Es, n (from the n subunits) is then simply the sum of the contributions:
n
E0 ei(ωt+φj ) (2.33)
X
Es, n =
j=1
Using again equation 2.6 the total scattered intensity Is, n is then given by:
1
Is, n = 0 cEs, n Es,∗ n
2  
n n
!
1 X i(ωt+φj ) 
X
−i(ωt+φk )
= 0 c E0 e E0 e
2 j=1 k=1
n X n
1
0 cE02 ei(φj −φk )
X
=
2 j=1 k=1
n
n X
ei(φj −φk )
X
= Is, 1
j=1 k=1
Xn X n
= Is, 1 (cos(φj − φk ) + i sin(φj − φk ))
j=1 k=1
n X n
(2.34)
X
= Is, 1 cos(φj − φk )
j=1 k=1
where Is, 1 denotes the (mean)intensity of the light scattered from one subunit
of the molecule and where to obtain the last identity it is used that sin(φj −
φk ) = − sin(φk − φj ), meaning that the sine terms cancel pairwise. If the over-
all size, d, of the molecule is very much smaller than the wavelength of the light
meaning that d · k = d · 2π/λ ≈ 0, the the pase differences are |φj − φk | =
|~rj · ~q − ~rk · ~q| = |~rjk · ~q| ≤ d · 2k ≈ 0. Hence for all terms in equation 2.23 we
get that cos(φj − φk ) ≈ cos(0) = 1. As this double sum contains a total of n2
terms we get:
Is, n ≈ Is, 1 · n2 (2.35)
and, as n = M/M1 where M is the molecular mass and M1 is the molecular mass
2.7. Scattering from many identical particles 35
of a subunit, we get:
Is, n ≈ (Is, 1 /M12 ) · M 2 (2.36)
This is basically the same result that was obtained in section 2.5, namely that the
”scattering power” of a single small molecule is proportional to the square of the
molecular mass. It is important to note that this result is valid when the molecule
is so small that the phase differences between light scattered from different parts
of the molecule are essentially 0. How to relax this restriction will be treated in
section 3.3 dealing with scattering from large molecules.
2.7 Scattering from many identical particles

E shall now consider the situation where light is scattered from a number N of
molecules that are present in a volume V which is evenly lit by a laser beam and
which are at the same time visible to the detector (This is the previously mentioned
scattering volume). Furthermore we assume that the molecules are all identical as
discussed above (see figure 2.14). We can now reuse equation 2.23, where we now
think of the individual scatterers as whole molecules whose individual scattering
intensity is given by equation 2.25. The total scattering intensity Is, total from the
N molecules can the be written as:
N
N X
(2.37)
X
Is, total = Is, n cos(φj − φk )
j=1 k=1
As the thing of interest in static light scattering (SLS) is the mean scattered in-
tensity we shall calculate this:
N X
N
(2.38)
X
hIs, total i = Is, n hcos(φj − φk )i
j=1 k=1
For now we shall assume that the molecules move about independently of each
other. This means that the phases φj and φk will change independently and ran-
domly all the time. Consequently the phase difference φj − φk can assume any
value as long as j 6= k. Hence hcos(φj − φk )i = 0 if j 6= k whereas the N led
where j = k have a mean value of hcos(φj − φk )i = hcos(0)i = 1. Therefore
i
rij
j
ri
z
rj
y
Figure 2.14: N identical molecules are in an evenly lit volume of a size so that the light
scattered from all the molecules can reach the detector.
equation 2.38 can be written as:
hIs, total i = Is, n · N (2.39)
As the number, N , of particles in the volume V is given by the molar concentration

c of the molecules through the expression
N =V ·c
we can rewrite equation 2.39 with the aid of equation 2.25:
hIs, total i = Is, n · NA V c

= (Is, 1 /M12 ) · NA V c · M 2
= (Is, 1 NA V /M12 ) · c · M 2
= (Is, 1 NA V /M12 ) · C · M (2.40)
where NA = 6.022 · 1023 mol−1 er Avogadro’s number and C is the weight con-
centration (unit g · L−1 ).
2.7. Scattering from many identical particles 37
As the scattering intensity of one molecular subunit, Is, 1 , is proportional to the

intensity of the incident laser beam and inversely proportional to the square of
the distance r from the scatterers to the detector we can write Is, 1 NA /M12 =
K ·I0 /r2 where K is the constant of proportionality. With this we can now rewrite
equation 2.40 as:
Is, total = KI0 V CM/r2 (2.41)
or, with the definition 2.2 of the Rayleigh ratio, R(q):
R(q) = KCM (2.42)
Note that we have written the Rayleigh ratio as a function of q although it is in

fact constant in this case (i.e. when the scattering particles are very small). The
reason is of course that when the scattering particles are sufficiently large the
Rayleigh ratio does in fact depend measurably on q, i.e. on the scattering angle.
The constant K is called the optical contrast constant and can easily be shown6 to
be related to the refractive index of the solvent and the dissolved particles through
the expression:
4π 2 n20 (dn/dC)2
K= (2.43)
NA λ40
where n0 is the refractive index of the solvent, λ0 is the wavelength of the laser
beam in vacuum and dn/dC er the derivative of the refractive index of the solution
with respect to the weight concentration of the dissolved molecules. In some
older literature you may come across a slightly different definition of the optical
contrast constant with a factor 2 instead of 4. For an explanation see section 2.8.
The quantity dn/dC is called the differential refractive index increment. For most
proteins in water based buffers it has a value of 0.18 – 0.20 ml g−1 for most
polysaccharides it has a value of about 0,14 – 0.15 ml g−1 .
In equation 2.42 the Rayleigh ratio R(q) is a measured quantity. Also, the weight
concentration C is also a quantity under the control of the experimenter. Hereby
it seems quite straightforward to determine the molecular weight M because the
constant K can be determined using the defining equation 2.43. It just takes doing
6 See e.g. the derivation in Physical Biochemistry, second edition, K. E. van Holde,
Prentice Hall 1985
dn
an independent determination of dC
. Alternatively one can use either a tabulated
value for dn
dC
or simply assume that the value is 0.18 cm3 · g−1 for proteins and
0.145 cm3 · g−1 for polysaccharides. This will in most cases give a molecular
weight which is at most 10% off if the measurements are done otherwise correctly.
2.8 Comparisons with older literature

If you compare the equations in this set of notes with the equations found in other
literature, especially older literature, you may find some strange differences. They
are due to two things:
1. We have assumed the incoming light to vertically polarised. But origin-

ally light scattering experiments were done with unpolarised light, which
changes both the mathematical form of the basic equations and the defini-
tion of both the optical contrast constant K and of the Rayleigh ratio R(θ).
2. The unit system used in older literature was different. We have used ba-
sically the SI-system whereas the cgs-system (also called the Gaussian unit
system) was the system of choice in science. This affects the looks of even
fundamental equations of physics, like the equation for the force between
two charged particles. The consequence of this is that even with polarised
light the old and the new equations would look different. In particular the
vacuum permittivity 0 may show up in one set of equations but not in the
other. Also, a factor of 4π sometimes raised to the power of two will be
present in one set of equations but not in the other.
Scattering from one small particle is described by equation 2.11 which reads
π 2 α2 sin2 φ
Is,1 (r) = I0 ·
r 2 λ4
Note that scattering is dependent only on the angle φ with the z-axis, not on the
angle θ with the x-axis. This will look as indicated in figure 2.15 If the incoming
light is no longer polarised the z-axis looses its special meaning in relation to the
2.8. Comparisons with older literature 39
Figure 2.15: The incoming light is moving along the x-axis and is polarised in the
direction of the z -axis. The surface indicates the intensity of the scattered light in dierent
directions. Note that nothing is scattered in the direction of the z -axis, up or down
scattering. Averaging over all polarisations in the yz-plane gives the following
expression for the scattered intensity:
π 2 α2 1 + cos2 θ
Is,1 (r) = I0 · · (2.44)
r 2 λ4 2
which is now independent of φ but depends on the angle θ with the x-axis. What
it will look like is shown in figure 2.16. As the scattered intensity in the case of
unpolarised incoming light shows an angular dependency which has nothing to
do with the form factor of the scattering particle the Rayleigh ratio is in this case
defined so as to factor out this dependency:
With polarised light it was defined in equation 2.1 as
Is (θ) · r2
Rθ =
I0 · Vs (θ)
With unpolarised light the definition of the Rayleigh ratio is
Is (θ) · r2 1
Rθ = · (2.45)
I0 · Vs (θ) 1 + cos2 θ
thus making the Rayleigh ratio of the scattering from an infinitely small particle
independent of the scattering angle. This alternate definition of the Rayleigh ratio
Figure 2.16: The incoming light is moving along the x-axis and is unpolarised. The
surface indicates the intensity of the scattered light in dierent directions. Note, that now
light is scattered in the direction of the z -axis, both up and down. And the scattered
intensity is now dependent on the angle θ with the x-axis.
has a small impact on the definition of the optical contrast constant K which with
polarised light was defined in equation 2.43 as
4π 2 n20 (dn/dC)2
K=
NA λ40
With unpolarised light, due to the alternate definition (2.45) of the Rayleigh ratio
the optical contrast constant is now defined as
2π 2 n20 (dn/dC)2
K= (2.46)
NA λ40
differing from the old definition only by a factor of 2.
With these (old) definitions of the Rayleigh ratio and of the optical contrast con-
stant the higher level equations of light scattering (like equations 2.42, 3.2, 3.13, etc.)
are unchanged.
2.9. Summing up 41
2.9 Summing up
Here are the most important equations from this chapter:
Refractive index:
cvacuum
n=
cmedium
The optical contrast constant:
4π 2 n20 (dn/dC)2
K=
NA λ40
The scattering vector:

~q = ~kout − ~kin
The length of the scattering vector:

!
4πn θ
q = |~q| = · sin
λ0 2
The Rayleigh ratio:

Is (q) · r2
R(q) =
I0 · Vs (q)
3
Determination of size and molecu-
lar weight
Probably the most common reason to do light scattering measurements is that it

can be used to determine molecular weights. This is what equation 2.42 shows. If
two conditions are fulfilled
1. the molecules (or particles) under investigation are small enough (as a rule
of thumb, smaller than λ/20, one twentieth of the wavelength of the light),
and
2. their concentration is sufficiently low (difficult to set up a criterion here)
molecular weight determination is unproblematic, at least in principle because

equation 2.42 actually holds in this case. However, if these two criteria are not
43
44 3. Determination of size and molecular weight
met equation 2.42 is often an oversimplification. We shall deal with these two
possible complicating factors in turn:
3.1 Concentration effects for small molecules

The simple relationship (equation 2.39 or 2.42) that the scattered intensity being
proportional to the weight concentration of the dissolved molecules turns out to
be increasingly inaccurate as the concentration increases. The reason for this is
obvious as equation 2.39 was based on the assumption that the molecules move in-
dependently of each other. This can of course not be completely true for the simple
reason that two molecules cannot occupy the same portion of space. So, in this
trivial sense the position of one molecule is dependent on the positions of other
molecules in the solution. It is also evident that the chance of two molecules acci-
dentally ”trying” to occupy the same portion of space is low at low concentrations
and high at high concentrations. Therefore equation 2.39 is a good approximation
at low concentration becoming increasingly inaccurate as concentration increases
and less and less room becomes available to the individual molecules. More gen-
erally the mutual influence of the molecules positions is due to forces (repulsive
or attractive or either, depending on intermolecular distances) acting between the
molecules making some distances more probable than others. The aforementioned
trivial effect of crowding is called ”the excluded volume effect” and can be seen
as result of strong repulsive forces acting between particles at very short distances
(less than their size) thus preventing the particles from penetrating into each other.
The double sum in equation 2.38 or 3.1 divided by the number N of molecules in
the scattering volume Vs is called the static structure factor and is denoted S(q):
N X
N
S(q) = N −1 · (3.1)
X
hcos(φj − φk )i
j=1 k=1
The assumption that the molecules move independent of each other leads as shown
to the simple relationship that the double sum of equation 2.38 or 3.1 assumes the
value N , or in other words that S(q) = 1. If we don’t assume independence
between molecular positions or movements we can then generalise equation 2.42
3.1. Concentration eects for small molecules 45
relating the Rayleigh ratio R(q) and the concentration and molecular weight of
the molecules:
R(q) = KCM S(q) (3.2)
The generalization is easy, the problem of course being how to calculate S(q).
Before we proceed we rewrite equation 3.2 for comparison with equation 3.6.
Rearranging gives:
KC 1
= (3.3)
R(q) M S(q)
For small (isotropically scattering) particles it can be shown (see e.g. Phys-
ical Chemistry of Macromolecules, C. Tanford, OUT-OF-PRINT-BOOKS-ON-

DEMAND 1992) that the Rayleigh ratio is related to the osmotic pressure Π of
the dissolved molecules through the relationship:
!
KC 1 ∂Π
= (3.4)
R(q) RT ∂C T,P
where R = 8.31 J · K−1 · mol−1 is the universal gas constant, not to be confused
with the Rayleigh ratio R(q). The osmotic pressure has a concentration depend-
ence, normally written as a series expansion:
Π 1
= C + A2 C 2 + A3 C 3 + · · · (3.5)
RT M
where the constants A2 , A3 , . . . are called the second, the third, . . . virial coeffi-
cient,respectively. For a so-called ideal solution the virial coefficients are all 0.
With this (i.e. equation 3.5) we can write equation 3.4 as:
KC 1
= + 2A2 C + 3A3 C 2 + · · · (3.6)
R(q) M
For practical purposes it is customary only to include the second virial coefficient.
This quantity can sometimes be calculated theoretically but is usually one of the
things that is determined experimentally, i.e. by doing light scattering measure-
ments.
Now we seem to have two different ways to express non-ideality of the solution,
namely equation 3.3 and equation 3.6. But they just show that in the case of very
small particles the static structure factor can be written as:

1
S(q) = (3.7)
1 + 2A2 M C + 3A3 M C 2 + . . .
which shows that the structure factor is indeed dependent on the concentration but
independent of the scattering vector, q. The latter turns out to be a consequence
of the assumption that the particles are sufficiently small to scatter isotropically.
3.2 Scattering from large molecules

Next we shall first disregard concentration effects but instead see how to deal
with particles that are not so small that they scatter isotropically. This is also a
necessary step towards developing a method to determine molecular weights. One
bonus of solving this complicating circumstance is that with the solution comes a
way to determine the size, i.e. the physical extension of the molecules.
3.3 The form factor

If a particle or molecule scatters light isotropically the reason is, as previously
mentioned, that light contributions scattered from the constituent parts (subunits)
of the molecule) are in phase, i.e. have no phase difference. This will be the case
if the molecule is much smaller than the wavelength of the impinging light. If on
the other hand the scattering particles are not much smaller than the wavelength
of the light light will in general be scattered with an efficiency that decreases with
increasing scattering angle θ and hence with increasing q-value. This at least hold
for particles without pronounced symmetry like e.g. identical spheres and and
identical cylinders. Such high symmetry particles will have scattering intensities
exhibiting (local) maxima and minima but still with an overall tendency to scatter
light less efficiently with increasing scattering angle. If for a solutions of particles
with a definite shape and identical size we measure the intensity of the scattered
light, Is (q), as a function of q then we will define a function of q called the form
factor, P (q), for these particles as the normalised scattering intensity:
P (q) = Is (q)/Is (0) (3.8)

3.3. The form factor 47
where Is (0), the scattering intensity at zero scattering angle (i.e. zero q value)
normally has to be determined by extrapolation of Is (q) to q = 0, as this value
is not accessible to direct measurement because a detector placed at zero scatter-
ing angle would see mainly the direct laser beam. Another way to interpret the
form factor is to conceive it as the ratio between the scattering intensity of the
particles and the scattering intensity of some hypothetical particles with the same
mass (molecular weight) but being infinitely small. When the particles are very
small the form factor is practically constant, P (q) ≈ 1 even for the largest exper-
imentally accessible q values.
Example: If the particles are dissolved in water (refractive index n = 1.333)
and the light source is a HeNe laser with a vacuum wavelength of λ0 = 633 nm
the maximum value of q corresponds to a scattering angle of θmax = 180◦ . Hence

qmax = 4πn
λ0
· sin θmax
2
= 0.026 nm−1 . For larger particles one gets form factor
curves that qualitatively look as shown in figure 3.1 It should be mentioned that
sometimes P (q) is called the scattering function . We can calculate the form
factor using equation 2.23 and 2.24 remembering that φj − φk = ~rj · ~q − ~rk · ~q =
~rjk · ~q. As equation 2.23 holds in generel and equation 2.24 holds for infinitely
small particles of the same mass we get the form factor by dividing the former
equation with the latter subsequently averaging over all orientations1 of the mo-
lecule:
n X n n X n
1 X 1 X
P (q) = h cos(φj − φk )i = hcos(~rjk · ~q)i (3.9)
n2 j=1 k=1 n2 j=1 k=1
We calculate the average value of the double sum by introducing a coordinate

system with the z-axis in the direction of ~q and normal polar coordinates (i.e.
with the angle θ relative to the z-axis and the angle φ in the xy-plane). Note that
in this coordinate system we have ~rjk · ~q = q rjk · cos(θ) so the mean value of one
term in equation 3.9 is given by:
1 Z 2π Z π
hcos(~rjk · ~q)i = cos(q rjk · cos(θ)) · sin(θ) dθ dφ (3.10)
4π 0 0
1 As the form factor pertains to a solution of molecules with random orientations we
can just do the averaging for one molecule
1,20
1,00
Rg = 20 nm
0,80
Rg = 50 nm
P(q)
0,60
0,40
Rg = 100 nm
0,20 Rg = 400 nm
0,00
0 0,005 0,01 0,015 0,02 0,025 0,03
q (nm-1)
Figure 3.1: The form factor becomes more and more steeply decreasing the larger
particles it represents. The particle sizes are indicated on the graphs by their radius
of gyration). The wavelength of the light is assumed to be 633 nm in vacuum and the
particles are suspended in water with a refractive index of n = 1.333. Note that the form
factor always has the value 1 for q=0
3.3. The form factor 49
The calculation is easily performed using the substitution x = q rjk cos(θ), yield-
ing:
sin(q rjk )
hcos(~rjk · ~q)i = (3.11)
q rjk
Substituting this expression into equation 3.9 we get the general expression for
the form factor of any molecule consisting of n identical point-like subunits with
pair-wise distances rjk :
n X n
1 X sin(q rjk )
P (q) = 2
(3.12)
n j=1 k=1 q rjk
This generel expression for the form factor can of course be calculated (on a com-
puter) for a given molecule if the structure of the molecule is known. Sometimes it
is useful to approximate the molecules with simple geometric bodies like spheres,
ellipsoids, cylinders, boxes etc. and then calculate the form factor analytically, i.e.
to arrive at a mathematical expression of the form factor based on known math-
ematical functions and relevant geometric parameters of the chosen geometrical
bodies. Examples of analytical form factors can be found in section 3.7.
One peculiar feature of the form factor is that if one makes an approximation2 of
P (q) to the lowest non-trivial order in q the approximated form factor turns out
to be independent of the shape of the molecule. It only depends on some average
size parameter, called the radius of gyration. The relevance of this we shall see in
the next section.
With the introduction of the form factor it is not difficult to see that we can gener-
alise equation 2.42 to:
R(q) = KCM P (q) (3.13)
Sometimes one defines the apparent molecular weight as:
Mapp = M P (q) (3.14)
i.e. the apparent molecular weight is dependent on the scattering angle. For the
apparent molecular weight to be the true molecular weight M it is necessary that
P (q) = 1 which requires the scattering angle to be small enough (in fact equal to
2 i.e. a Taylor series expansion of equation 3.12
zero degrees). This is consistent with what we have seen before: The form factor
is practically 1 if q is small enough (which is very small if the particles are large
and not so small if the particles are small) as seen in figure 3.1.
If we include the form factor we can generalise also equation 3.6 which was valid
for small particles, noting that it can be rewritten:
KC 1 1
= +2A2 C +3A3 C 2 +· · · = (1+2A2 M C +3A3 M C 2 +· · ·) (3.15)
R(q) M M
We now simply substitute the expression for the apparent molecular weight for
the molecular weight in this expression (3.15) thereby getting:
KC 1
= (1 + 2A2 M P (q)C + 3A3 M P (q)C 2 + · · ·) (3.16)
R(q) M P (q)
Now, the principle of determining the molecular weight becomes apparent: For a
number of ever decreasing concentrations of the sample, C1 , C2 , C3 , . . . , where
Cn → 0, measure the Rayleigh ratio R(q) at a number of ever decreasing q values,
q1 , q2 , q3 , . . . , where qn → 0. Subsequently, for every q value extrapolate the val-
KC
ues of R(q)
to C = 0 thereby finding 1/(M P (q1 )), 1/(M P (q2 )), 1/(M P (q3 )) . . ..
Finally, extrapolate these values to q = 0. Hereby we obtain the true molecu-
lar weight M (or, actually 1/M ) as P (0) = 1. The extrapolation to q = 0 is
less trivial than it may seem and is in practice often performed with the aid of a
Guinier-plot as described in section 3.4.
3.4 Size determination

As we have just seen the form factor reflects the shape and size of the particles. If
the particles are not so small that their form factor can be considered essentially
constant (with the value 1) it is necessary to measure the intensity of the scattered
light at several angles and subsequently perform an extrapolation of the intensity
down to a scattering angle of θ = 0◦ (i.e. q = 0). This procedure may seem
cumbersome which it is, especially if on e is only interested in the molecular
weight of the particles. On the other hand, if the form factor is not equal to one at
all scattering angles it actually contain information about the size and sometimes
3.4. Size determination 51
the shape of the particles. In many cases one may be content with a coarse measure
of the size of the particles and we shall now see how this can be accomplished.
This coarse measure of size is called the radius of gyration Rg . We shall see how
to arrive at this quantity through the calculation of some useful approximations of
the form factor 3.12.
The Guinier approximation

If q rjk < 1, the form factor 3.12 can be approximated by its Taylor series ex-
pansion to the smallest non-trivial order which turns out to be second order. The
easiest way to do this by approximating the sine function by its Taylor series ex-
pansion: sin(x) = x − 3!1 x3 + 5!1 x5 − · · · leading to the approximation
sin(x) x − 3!1 x3 + 5!1 x5 − · · · 1 1

≈ = 1 − x2 + x4 − · · ·
x x 3! 5!
Keeping the first two terms only we then obtain

n X n
1 X 1
P (q) ≈ 2 (1 − (q rjk )2 ) (3.17)
n j=1 k=1 6
This approximation can be written in a simpler and more useful form which is
seen by expanding the right hand side of equation 3.17:
n X n
1 X 1
2
(1 − (q rjk )2 ) =
n j=1 k=1 6
n X n n X n
1 X 1 X 1
2
(1) + 2
(q rjk )2 =
n j=1 k=1 n j=1 k=1 6
n X n
1 2 1 2 1 X
· n + q · r2 =
n2 6 n2 j=1 k=1 jk
1 2 2
1− q Rg
3
where Rg is the aforementioned radius of gyration defined as:

n X n
1 X
Rg2 = r2 (3.18)
2 n2 j=1 k=1 jk
The radius of gyration needs af few supplementary comments which can be found
at the end of this section (page 55). We thus arrive at an expression for the form
factor approximation which reads:
1 2 2
P (q) ≈ 1 − q RG (3.19)
3
The approximation 3.19, called the the Debye approximationen is sometimes used
”as is” but suffers from a flaw: It drops very fast down to 0 and even worse it
becomes negative over a certain q value (the graph is a parabola intersecting the
q-axis at q = 3/Rg ). A negative value for the form factor is of course meaningless
but even the fast drop down to zero is a bad representation of most measurements
which usually drop much more gently to low values never becoming zero. This
undesired feature of the Debye approximation can easily be remedied, however
arbitrarily: Just choose a function which has a Taylor series expansion whose first
terms coincide with the Debye approximation but which decays more gradually
and stays positive. The most well-known of such approximations is the Guinier
approximation:
(3.20)
1 2 Rg2
P (q) ≈ e− 3 q
whose Taylor series expansion is 1− 1!1 ( 31 q 2 Rg2 )+ 2!1 ( 13 q 2 Rg2 )2 − 3!1 ( 13 q 2 Rg2 )3 +. . . and
thus coincides with the Debye approximation on the two first terms. As it is based
on a Taylor series expansion it requires q to be sufficiently small or rather that q Rg
being small enough. The criterion most frequently used to ensure the validity of
the Guinier approximation is that q Rg < 1. How good the approximation actually
is depends on the shape of the particles.
The Guinier approximation can be used to determine the size of particles but is
also an intermediate step when determining the molecular weight of large particles:
One measures the scattered intensity at several q values i.e. at several scattering
angles. Assume that at each scattering angle the scattered intensity has been meas-
ure at a concentration C low enough that concentration effects can be neglected.
Then the Rayleigh ratio can be calculated for each q value. With the Guinier
approximation (3.20) we can write equation 3.13 as:
(3.21)
1 2 Rg2
R(q) ≈ KCM · e− 3 q
Taking the logarithm on both sides we get:

1 2 2
ln R(q) ≈ ln(KCM ) − q Rg (3.22)
3
It is seen that if ln R(q) is plotted against q 2 one gets a straight line with a slope
coefficient a given by
1
a = − Rg2 (3.23)
3
and an intercept with the vertical axis of ln(KCM ) If the only information wanted
from the light scattering experiment is the radius of gyration it is not necessary to
calculate the Rayleigh ratio. As the Rayleigh ratio is proportional to the scattered
intensity Is (q) we get, in analogy with equation 3.22:
1 2 2
ln Is (q) = ln Is (0) − q Rg (3.24)
3
A plot of measuered Rayleigh ratios or scattered intensities, ln R(q) or ln(Is (q))
vs. q 2 is called a Guinier plot. In practice the slope a is found by linear regres-
sion which will also give either ln(KCM ) or ln(Is (0)) from which the molecular
weight may be calculated. However, before accepting the results of the linear re-
gression one should make sure that the Guinier approximation was in fact valid.
This amounts to check if the radius of gyration found Rg and the largest value
of q used for the linear regression (qmax ) satisfy the condition qmax Rg < 1. If
the criterion is not met the linear regression should be confined to only smaller
values of q, i.e. using a smaller value of qmax . In practice there is a limit to how
small scattering angles can be used in a measurement and thus to how small q val-
ues are experimentally accessible. This puts a practical upper limit to how large
particles it is possible to measure. Note, that if the particles are so large that it is
impossible to fulfil the criterion qmax Rg < 1 then neither the radius of gyration
nor the molecular weight can determined. But possibly a lower bound to both
can be established (see the worked example below). There is also a limit to how
small particle sizes it is possible to determine: If the particles are very small the
slope in the Guinier plot a = − 31 Rg2 becomes so small that the linear regression
can not determine it with meaningful accuracy. Due to inevitable scatter of the
measured points the slope may in practice sometimes become negative in which
case no radius of gyration satisfies equation 3.23. The lower limit is not a the-
oretical but only a practical one. It is a matter of the statistical scatter (noise) of
the measured scattering intensities. In literature the limit of isotropic scattering is
often set at a particle size of λ/20, meaning that below this size you cannot de-
termine what the size actually is. The criterion, however, is neither a strict one nor
is it very precise. It is often not specified whether the wavelength is the vacuum
wavelength of the source or the ensuing wavelength in the medium surrounding
the scattering particles. Likewise it is usually not specified what is meant by the
size of the particle. In practice the smallest radius of gyration measurable today
(2013) is about Rg ≈ λ0 /60, where λ0 is the wavelength of the laser employed.
Comments on the radius of gyration

Note that any object can be assigned a radius of gyration. This is an
exact property of any object. The approximation lies in just keeping
the first few terms in the Taylor series thus apparently disregarding
details about the particle structure and shape and only keeping one
aspect of the particle ”structure”: its radius of gyration.
The squared radius of gyration is thus half the average value of the
squared distances between the constituent parts of the particle. The
half originates from the fact that we are defining what we call a radius
not a diameter.
As an example of this consider a ”particle” consisting of two identical
subunits, 1 and 2, separated by the distance d. The radius of gyration
for this particle would be (with n = 2):
n X n
1 X
Rg2 = r2
2 n2 j=1 k=1 jk
1 2 2 2 2
= · (r11 + r12 + r21 + r22 )
2 · 22
1
= · (02 + d2 + d2 + 02 )
8
!2
1 2 d
= ·d =
4 2
which shows that the radius of gyration assumes the ”natural” value
of half the separation between the two constituent subunits.
Alternate definition of the radius of gyration

There is another definition of the radius of gyration which may be
easier to envisage than the average of n2 intraparticle distances: The
radius of gyration is related to the (only) n distances from the particle
subunits to the particle’s centre of mass.
n
1X
Rg2 = rj,2 c.m. (3.25)
n j=1
where rj, c.m. denotes the distance from the j’th subunit to the particle’s
centre of mass. The two definitions of Rg turn out to equivalent but we
shall put off the proof to section 3.6.
3.5 Size and molecular weight. How to average

in mixtures
Suppose you have a solution containing a mixture of different kinds of particles.
These may be entirely different molecular species or could be different oligomeric
states of the same basic molecular unit, i.e. it could be monomers, dimers, trimers,
tetramers, etc. of the same molecule. These will be treated as different types of
particles. Suppose we have n different types of particles with molecular weights
M1 , M2 , . . . , Mn , radii of gyration R1G , R2G , . . . , RnG and which are present at
weight concentrations C1 , C2 , . . . , Cn . Furthermore we will assume that the light
scattering contributions from the different types of particles can simply be ad-
ded together. This essentially means that the concentrations are so low that the
particles can be considered non-interacting. Thus we can write for the total light
scattering for the mixture:
n n
(3.26)
X X
Rtotal (q) = Ri (q) = K Ci Mi Pi (q)
i=1 i=1
How to calculate averages

Suppose you do a light scattering measurement on such a mixture in the same
manner as you would do if measuring on a solution of only one species, (i.e.
measure at many different angles and possibly at many different dilutions). If you
then do a data analysis of the measurement results assuming that the solution only
contained one species then you would arrive at an average molecular weight and
possibly an average radius of gyration. In order to be able to understand, interpret
or use these average values it is important to realise how the individual molecular
weights and radii of gyration are weighted relative to each other. In other words,
what type of average you get from a light scattering measurement. This we shall
now look at:
First, if we want to calculate the average molecular weight hM i, the corresponding
weight concentration must be the total weight concentration of the species present
in the solution (disregarding the solvent molecules themselves). This means that
3.5. Size and molecular weight. How to average in mixtures 57
the total light scattering can be written as:

n
!
(3.27)
X
Rtotal (q) = K Ci hM ihP (q)i
i=1
If we set equal the two expressions for the total light scattering Rtotal from equa-
tions 3.26 and 3.27 we get
n n n
!
(3.28)
X X X
K Ci hM ihP (q)i = Ri (q) = K Ci Mi Pi (q)
i=1 i=1 i=1
If, furthermore, we set q = 0 (extrapolate the measurements to zero scattering

angle) all form factor assume the value 1, we get that the average molecular weight
can be calculated as:
n
X n
X
Ci Mi Ci Mi
hM iw = i=1
n = i=1
(3.29)
X Ctotal
Ci
i=1
This type of average (indicated by the subscript w by the angle bracket) is called
the weight average molecular weight as opposed to other possible averages, like
e.g. the number average molecular weight where the individual molecular weights
would be multiplied by the molar concentrations instead of the weight concentra-
tions and the total concentration would be the total molar concentration. Note that
this number average weight concentration cannot be determined by light scatter-
ing. Only the weight average molecular weight can. Note that we have tacitly
assumed that the contrast constant K is the same for all molecular species. This
is often the case but not always. If the different molecular species have different
values of the contrast constant the simple equation 3.29 does not hold.
In a similar way we can determine an average radius of gyration for the molecules
in the solution. To this end we use the Debye-approximation 3.19 for the form-
faktor. Substituting into equation 3.26 we get:
n n
1
Ci Mi (1 − Rg2 i q 2 ) (3.30)
X X
Rtotal (q) = Ri (q) = K
i=1 i=1 3
where Rg i denotes the radius of gyration of the i’th type of molecule. If again
we assume that the scattering is due to some ”average” particles with the aver-
age molecular weight hM iw , an average (squared) radius of gyration hRg2 i and a
Pn
concentration Ctotal = i=1 Ci we can write the total Rayleigh ratio as:
1
Rtotal (q) = KCtotal hM iw (1 − hRg2 iq 2 ) (3.31)
3
When we compare equations 3.30 and 3.31 equating the terms containing Rg we
get the expression for the average radius of gyration:
n
Ci Mi Rg2 i
X
hRg2 i = i=1
(3.32)
Ctotal hM iw
which means that the individual radii of gyration (squared) are weighted by the
light scattering contribution of the same species. Using the expression for the
average molecular weight 3.29 we can rewrite equation 3.32 into its more common
form:
n
Ci Mi Rg2 i
X
hRg2 i = i=1
n
X
(3.33)
Ci Mi
i=1
This average value is called the z−average of hRg2 i and the square root of this
quantity is called the z−average radius of gyration, hRg iz .
A word of caution: The average quantities obtained by light scatter-

ing, hM iw and hRg iz can be hard to interpret. They are both strongly
biased towards the influence of high molecular weight species. Ima-
gine a solution of protein where the pure, monomeric species has a
molecular weight M1 . Chemists will tell that the solution is 99% pure
monomer plus 1% aggregates of the monomer. The purity may be
stated in weight percent or in mole percent. Assume first that we are
talking about weight percent, e.g. 9.9 g · L−1 of the monomeric species
and 0.1 g · L−1 of the aggregates giving a total weight concentration of
10.0 g · L−1 . What if the aggregates are really big consisting of, say,
100 monomeric species? The aggregates would then have a molecu-
lar weight of 100M1 . How would light scattering judge the molecu-
lar weight? Inserting into equation 3.29 gives the average molecular
weight:
n
X
Ci Mi
i=1
hM iw =
Ctotal
9.9 · M1 + 0.1 · (100M1 ) 9.9M1 + 10.0M1
= =
10.0 10.0
= 1.99M1
or almost twice the monomer molecular weight. Had the aggregate

been made up of 1000 monomeric units but still only at a weight
concentration of 1% the average molecular weight would have been
almost 11 times higher than the monomer molecular weight. It is
thus evident that it is extremely important to do everything possible
to measure on clean well filtered samples in extremely pure buffers.
Even then the usefulness of the weight average molecular weight will
depend on how ”well behaved” the sample is. Also one should note
that in the above example specifying the purity by molar concentration
would make things seem even worse: In the case where the aggregates
have a molecular weight of 100M1 a weight fraction of 1% would cor-
respond to a molar fraction of approximately 0.01%, i.e. the solution
would be 99.99% pure by molar fraction but still the average molecu-
lar weight would be wrong by a factor of two!
Similar arguments hold regarding the z−average of the radius of gyr-
ation. So again these averages can be hard to interpret and could be
strongly dependent on details of sample preparation.
Detailed determinations
The ambiguities of measuring in principle only average values of the molecular
weight and the radius of gyration can sometimes be circumvented by combining
light scattering with size exclusion chromatography. Instead of using a classic
light scattering setup (see figure 1.2) another setup is used where the light scatter-
ing instrument measures the scattered intensity at several angles simultaneously.
The the combined technique is called SEC-MALS (Size Exclusion - Multi Angle
Light Scattering). A sketch of the setup is shown in figure 3.2.
Buffer reservoir
HPLC pump
Sample
injection Size exclusion
column
Light scattering
Measures MCP(θ )
Buffer stream Separated

molecular sizes
Mixed molecular Refractive index
sizes Measures C
Waste
Figure 3.2: The HPLC-pump creates a constant ow of liquid (buer) through the gel
column. The sample, typically a 100 µL solution, is injected into the buer ow through
a special port (valve). As the molecules of the sample are ushed through the gel column
the larger species will pass through quickly whereas the smaller species will come out later.
The smaller the molecules the later they elute. The dierent size classes of molecules then
pass through the light scattering instrument where the Rayleigh ratio (KCM P (q)) is
determined. Next the molecules are carried though a dierential refractometer (or, altern-
atively, a UV-detector) which determines the concentration C of the molecules. Having
measured both KCM P (q) at several angles and C the taking the ratio the molecular
weight M can be calculated and also the radius of gyration if the scattering is suciently
anisotropic. This can be done for every species eluting from the column.
The principle of the measurement is that the light scattering measurements are
done on the molecules while they are flowing through a measurement cell and
subsequently flowing through another instrument which measures the concentra-
tion og the molecules. The sample solution is injected into the buffer flow which
continuously flows through the gel column and subsequently trough the light scat-
tering cell and the refractometer cell. The gel column separates the molecules in
the buffer flow in such a way that the larger molecules make a quick passage
whereas the smaller molecules are retained longer before they eventually elute.
When the molecules leave the gel column they pass through the light scattering
instrument which determines the Rayleigh ratio R(q) = KCM P (q) at several
scattering angles (several values of q) at the same time. Remember that K de-
notes the optical contrast constant and P (q) is the form factor for the molecular
species. As the Rayleigh ratio is measure at many angles at the same time it is
usually possible to perform the extrapolation to zero q so that R(0) = KCM can
be determined. After leaving the light scattering instrument the molecules flow
through a differential refractometer which measures the difference in refractive
index between the sample solution and the pure buffer, n − n0 . This difference
can be written (neglecting higher order terms in the concentration)
dn
n − n0 = ·C (3.34)
dC
dn
As the quantity dC
called the refractive index increment can be determined by
separate measurements (or be found in a table) a measurement of the refractive
index difference is a way to determine the weight concentration of the molecules.
As the light scattering instrument in principle measures M C and the refracto-
meter measures C the molecular weight can be found as the ratio between the two
measurements.
More details about this method can be found in chapter 5.
As the light scattering instrument measures at many angles at the same time the
form factor P (q) is determined for each of the molecular species that flow through
the instrument separated by the column. It is therefore possible to determine the
radius of gyration for the different species if the q−values (scattering angles) of
the instrument make it possible:
1. If the particles are very small the q−values of the instrument are not suffi-
ciently large for the form factor to decrease measurably at even the largest
q−value. In this case only an upper bound can be set for the radius of gyra-
tion.
2. If the particles are very large even the smallest scattering angles of the in-
strument gives q−values that are too big for the criterion q · Rg < 1 to be
fulfilled so none of the approximations used to determine Rg are valid.
3.6 Equivalence of the two definitions of Rg

We still need to prove that the two definitions of the radius of gyration are equi-
valent:
First, remember that the position ~rc.m. of the centre of mass is defined as
n
1X
~rc.m. = r~j (3.35)
n j=1
where ~r1 , ~r2 , ~r3 , . . . are the positions of the particle’s subunits.
First let us start with the new definition expanding the expression:
n
1X
Rg2 = r2
n j=1 j, c.m.
n
1X
= (~rj − ~rc.m. ) · (~rj − ~rc.m. )
n j=1
 
n n n
1 X 2
X
2
X
= r + r −2· ~ri · ~rc.m. 
n j=1 j j=1 c.m. i=1
 
n n n
1
rj2 + 2
X X X
= rc.m. − 2~rc.m. · ~rj 
n j=1 j=1 j=1
 
n
1
rj2 + nrc.m.
2
X
= − 2~rc.m. · (n~rc.m. )
n j=1
 
n n
1 2 2  1X 2
rj2 − rc.m.
X
= rj − nrc.m. =
n j=1 n j=1
Next, let us expand the original definition 3.18 of Rg :

n X n
1 X
Rg2 = r2
2 n2 j=1 k=1 jk
3.6. Equivalence of the two denitions of Rg 63
n X n
1 X
= (~rj − ~rk ) · (~rj − ~rk )
2 n2 j=1 k=1
 
n X n n X n n X n
1 X 2
X
2
X
= · rj + rk − 2 · ~rj · ~rk 
2 n2 j=1 k=1 j=1 k=1 j=1 k=1
 
n n n n
1  X 2
X
2
X X
= · n · r j + n · rk − 2 · ( ~
r j ) · ( ~rk )
2 n2 j=1 k=1 j=1 k=1
 
n n
1  X 2
rj2 − 2 · (n~rc.m. ) · (n~rc.m. )
X
= · n · r j + n ·
2 n2 j=1 j=1
 
n n
1  X 2 2 2  1X
= · 2n r j − 2n rc.m. = rj2 − rc.m.
2
2 n2 j=1 n j=1
It is thus seen that the two expressions for Rg can be rewritten into the same form
and thus are equivalent.
3.7 Summing up
Here are the most important equations from this chapter:
The form factor definition:
Is (q)
P (q) =
Is (q = 0)
Apparent molecular weight:
Mapp = M P (q)
Radius of gyration:
n X n
1 X
Rg2 = r2
2 n2 j=1 k=1 jk
or
n
1X
Rg2 = r2
n j=1 j, c.m.
The Debye approximation:
1 2 2
P (q) ≈ 1 − q Rg for Rg · q ≤ 1
3
The Guinier approximation:
1 2 2
P (q) ≈ e− 3 q Rg
for Rg · q ≤ 1
General equation for static light scattering:
KC 1
= 1 + 2A2 M P (q)C + 3A3 M P (q)C 2 + · · ·
R(q) M P (q)
3.7. Summing up 65
Radii of gyration that can be expressed analytically
Examples:
Massive sphere of radius r:

s
3
Rg = r
5
Ellipsoid with semi-axes a, b og c:

s
a2 + b 2 + c 2
Rg =
5
Spherical shell (infinitely thin) with radius r:
Rg = r
Cylinder with radius r and length L:

s
r2 L2
Rg = +
2 12
Box with sides a, b og c :

s
a2 + b 2 + c 2
Rg =
12
Note, that the last equation also gives the radius of gyration for an
infinitely thin rectangular plate (c = 0) and for an infinitely thin
rod (b = 0 and c = 0)
Special cases:
Static light scattering for very dilute solutions:

KC 1
=
R(q) M P (q)
Static light scattering for very small particles:

KC 1
= + 2A2 C + 3A3 C 2 + · · ·
R(q) M
Static light scattering for very dilute solutions of very small

particles:
KC 1
=
R(q) M
3.7. Summing up 67
Form factors in closed analytical form

Many ad hoc form factors exist to describe empirically the
”shape” of scattering data. However, only few form factors can
be derived analytically and expressed in a closed mathematical
form. Here we just mention the form factor of a massive,
homogeneous sphere and the form factor of a random coil.
Massive sphere with radius R:

!2
3 · (sin(qR) − qR cos(qR))
P (q) =
q3
Random coil with radius of gyration Rg :

2
P (q) = · (exp(−(qRg )2 ) − 1 + (qRg )2 )
(qRg )4
How these two form factors may look is shown in figure 3.3
1,2
0,8
P(q)
0,6
0,4
0,2 Sphere Random coil
0
0 0,005 0,01 0,015 0,02 0,025 0,03
q (nm-1)
Figure 3.3: The two form factors correspond to the same radius of gyration: The
sphere radius is set at 250 nm and the random coil radius of gyration is set at the value
q
3
Rg = 5 · 250 nm. Note that having the same radius of gyration the two form factors co-
incide at small q -values, as expected. The q−value corresponding to q · Rg = 1 is 0.005 nm.
The whole q -range corresponds to a vacuum wavelength λ0 = 633 nm (a HeNe laser) and
a range of scattering angles from 0◦ to 180◦ .
Note also that the sphere form factor drops to exactly zero and then rst rises and later
drops again. This damped oscillatory behaviour is only found with highly symmetric
particles like spheres, ellipsoids and cylinders.
4
Simple theory for dynamic light
scattering, DLS
Dynamic light scattering (DLS) is a relatively new technique dating back, in prin-
ciple, to 1964. The technique is also known as quasi elastic light scattering
(QELS) or photon correlation spectroscopy (PCS). The technique has gained in-
creasing popularity ever since the mid-seventies. The reason why this technique
came about so late is that in many ways it is technologically more demanding than
static light scattering: The light source need to be very monochromatic, intense
and coherent (so the light can be focussed to a very narrow bundle of rays). All
of this essentially means that the light source has to be a laser (invented in 1960).
Furthermore the detector system has to be very sensitive (being able to detect
single photons) and relatively fast. And last, the post-processing of the detected
69
70 4. Simple theory for dynamic light scattering, DLS
signals requires fast, specialised electronics with hig computing capacity.

The purpose of doing dynamic light scattering experiments is usually to meas-
ure the diffusion coefficient of molecules in solution but may also be to measure
characteristic relaxation times in e.g. gel systems. We shall here only deal with
measurements of diffusion processes. Through the measurement of the diffusion
coefficient for particles in suspension it is possible to determine indirectly the size
of the particles. The size measure obtained in this way is called the hydrodynamic
radius or sometimes the Stokes radius of the particles because the determination
is based on the Stokes-Einstein relationship which relates the diffusion coefficient
of suspended spheres with radius r with their diffusion coefficient D
kB T
D= (4.1)
6πηr
where T is the absolute temperature of the suspension, η is the viscosity of the
liquid and kB = 1.38 · 10−23 J · K−1 is Boltzmann’s constant. Strictly speak-
ing equation 4.1 is valid only for non-interacting particles to whose surface the
surrounding liquid sticks. If the diffusion coefficient Dexp of some particles is ex-
perimentally determined the diffusion coefficient can be inserted into the Stokes-
Einstein equation 4.1 which can then be solved for the radius. The radius thus
obtained the hydrodynamic radius rh and is defined as:
kB T
rh = (4.2)
6πηDexp
where Dexp is the measured diffusion coefficient of the suspended particles. If the
particles are not spheres the hydrodynamic radius is nevertheless a rough measure
of the particle size in the same way that the radius of gyration is. The two numbers
are usually not exactly equal. If the particles are non-spherical their hydrodynamic
radius is often approximately the radius of a sphere with the same volume as the
particle. This hold as long as the particles do not have extreme geometry like
being very long and thin or being wide and flat. As a rule of thumb expect the
hydrodynamic radius to increase by 5 – 7% relative to the radius of a same volume
sphere when the axial ratio increases by one.
The diffusion coefficient is – like the Rayleigh ratio – determined thermodynam-
ically though the osmotic pressure Π. Equation 4.1 is a special case of the gener-
4.1. Intensity uctuations 71
alised Stokes-Einstein equation:

!
M ∂Π
D= (1 − φ)2 (4.3)
NA f ∂C T,P
where φ denotes the volume fraction of the suspended (dissolved) particles and
f is the so-called frictional coefficient. The frictional coefficient for spheres with
the liquid sticking to their surface has been calculated by Stokes as f = 6πηr. Un-
fortunately this case is more complicated than the determination of the molecular
weight when it comes to the influence of concentration. In his former case it was
sufficient to make a series expansion (a virial expansion) of the osmotic pressure
to arrive at the virial coefficients. In the present case not only the osmotic pressure
but also also the frictional coefficient f depends on the concentration which tends
to make things a lot more complicated. A theoretical exposition of the influence
of concentration on the diffusion coefficient is therefore difficult. However, it is
of course possible to write the concentration dependence as a Taylor series
D = D0 (1 + kD C + · · ·) (4.4)
where kD is a coefficient for the first order dependence of the diffusion coefficient
on the concentration. A number of different approaches to the calculations of
this coefficient exist but we shall not be concerned with these here. It is however
evident that to obtain the diffusion coefficient of some particles it is necessary
to do the measurements at several concentrations that make it possible to do an
extrapolation to zero concentration.
In the following we shall see hoe diffusion coefficient can be measured using
dynamic light scattering.
4.1 Intensity fluctuations

When performing dynamic light scattering measurements the setup has some sim-
ilarity to the setup used for static light scattering (figure 1.2). There are, however
some important differences (see figure 4.1). The laser beam is focussed down to a
1
diameter of approximately 10
mm and the scattered photons are detected by a pho-
tomultiplier tube (PMT) or sometimes be an avalanche diode detector which are
PC
Autocorrelator
D
PA
T
PM
Pinholes
Imaging lens
Focussing lens
q
Laser
Vat
Sample cell
Figure 4.1: Setup for measuring dynamic light scattering (DLS). The detector is usually
a photomultiplier tube (PMT) of the type that can detect single photons but can also be
based on an avalanche diode. The detection is done through a number of pinholes where
the smallest one is nearest to the detector. It usually has a diameter of 100 − 200 µm. Also
the laser beam is focused in the centre of the measurement cuvette down to a diameter
of about 100 µm. Furthermore the measurement cuvette is placed in a vat embedded
containing a thermostatted liquid as the measured diusion coecient is temperature
dependent.
able to detect single photons. Each detected photon creates an electrical voltage
pulse which is subsequently sent to a pulse amplifier and discriminator (PAD)
which only lets through pulses of a sufficient voltage (indicative of them origin-
ating from photons an not internal noise in the detector) and then shapes these
pulses into a standard voltage (5 V) and a standard duration (e.g. 20 ns). These
standardised pulses are the fed into a digital autocorrelator, a piece of hardware
which calculates the autocorrelation function for the detected photons. The inner
workings of the digital autocorrelator is discussed below and the autocorrelation
function is described in section 4.2.
As we saw in section 2.7 the intensity of the light scattered from N particles can
be written (see equation 2.37):
N X
N
(4.5)
X
Is, total (t) = Is, n cos(φj (t) − φk (t))
j=1 k=1
Here the time dependence of the intensity is indicated explicitly with the t in
the bracket. The reason why the intensity depends on time is that the particles
4.1. Intensity uctuations 73
move about due to diffusion (Brownian motion). These random movements of

the particles make the individual phases φj (t) and φk (t), and thereby the phase
differences, change with time in a random way. The diffusion coefficient depends
on two things, the size (and shape) of the particles and the viscosity of the embed-
ding liquid. If, e.g., the particles are big or if the liquid has a high viscosity the
particles will move about slowly Consequently the phases and thus φj (t) − φk (t)
will change slowly and Is, total (t) becomes a slowly fluctuating of function of time.
The rapidness or the average frequency of the fluctuations thus becomes a measure
of how rapidly the particles move about by diffusion or, in other words, a measure
of the diffusion coefficient D.
How to measure fluctuations

If you look at equation 4.5 two things are to be noted:
1. The scattered intensity is measured at a point
2. If the sum contains many terms the fluctuations of the individual terms will
tend to average out
In a practical DLS setup the laser beam is focussed to a very narrow diameter,
typically 1/10 mm ensuring that not too many particles contribute to the sum. If
a screen is set up next to a scattering solution the scattered light shows on the
screen in the form of speckles moving rapidly around. In order for the naked
eye to actually see this phenomenon one needs large, slowly diffusing particles
and a very narrow laser beam. The size of the speckles is larger the smaller the
diameter of the laser beam. In order to have large intensity fluctuations on the
detector it needs to see only one speckle at a time, i.e. the effective detector area
should match the size of the speckles: If the detector area is much smaller than the
speckle size the fluctuations will still be large but little light is collected, leading
to poor statistics. If the detector area is made very large in order to collect a lot
of light the detector will see many speckles (but nearly the same number) all the
time and the fluctuations will be weak.
4.2 Making sense of fluctuations: The autocor-

relation function
Although intuitively clear that a rapidly fluctuating scattered intensity corres-
ponds to a large diffusion coefficient (i.e. small particles) and slowly fluctuating
scattered intensity corresponds to small diffusion coefficient (i.e. large particles)
it is not immediately evident how to obtain a precise, quantitative measure of the
fluctuations. This is of course a necessary first step before the fluctuations can be
used to make precise numeric statements about the diffusion coefficient. In the
early days of DLS it was customary to use the spectrum of characteristic fluctu-
ation frequencies to quantitate the fluctuations. But soon this approach was super-
seded by the use of the so-called autocorrelation function G2 (t). The two methods
are equivalent but the electronic circuitry and the mathematical treatment of the
acquired data differ. The autocorrelation function for the scattered light is defined
by the equation:
1ZT
G2 (τ ) = lim Is, total (t) · Is, total (t + τ ) dt (4.6)
T →∞ T 0
meaning that the integral contains scattered intensities recorded with a time dif-
ference τ . The function G2 (τ ) is called the intensity autocorrelation function for
the scattered light because it correlates intensities. One can define a so-called
field autocorrelation function G1 (τ ) in a similar manner substituting the intensit-
ies with the magnitude of the electric field instead:
1ZT
G1 (τ ) = lim Es, total (t) · Es, total (t + τ ) dt (4.7)
T →∞ T 0
We shall come back to the relation between the two autocorrelation functions
shortly. In practice the intensity is measured at discrete times. It is therefore
more straightforward and precise to count how many photons are detected within
a given time frame ∆t called the sample time. Instead of the continuous quantity
Is, total (t), the total scattered intensity one measures a sequence of photon counts
n(0 · ∆t), n(1 · ∆t), n(2 · ∆t), n(3 · ∆t), . . ., all recorded during the time period
∆t. The sample time ∆t can vary, typically from 0.1 to, say 50 µs. It is (was)
4.2. Making sense of uctuations: The autocorrelation function 75
a quantity set by the experimenter. In the early days of DLS (i.e. before 1985)
autocorrelators simply had a dial switch used to set the sample time. Newer cor-
relators are so-called multiple tau correlators which are essentially an assembly
of connected autocorrelators which operate at different sample times. We shall
not describe the construction of such multiple tau correlators but only look at the
classic linear tau correlators which are easier to describe. When the autocorrela-
tion function is measured at discrete times we define the autocorrelation function
at these times τ = 0 · ∆t, τ = 1 · ∆t, τ = 2 · ∆t, τ = 3 · ∆t, τ = 4 · ∆t, . . . as:
N
1 X
G2 (0 · ∆t) = n(j · ∆t) · n([j + 0] · ∆t)
N j=1
N
1 X
G2 (1 · ∆t) = n(j · ∆t) · n([j + 1] · ∆t)
N j=1
N
1 X
G2 (2 · ∆t) = n(j · ∆t) · n([j + 2] · ∆t)
N j=1
..
.
N
1 X
G2 (k · ∆t) = n(j · ∆t) · n([j + k] · ∆t) (4.8)
N j=1
..
.
or put differently:
G2 (k · ∆t) = hn(t) · n(t + k · ∆t)it (4.9)
The time k · ∆t is called the correlation time or the lag time. This is simply a
multiple of the sample time ∆t. Usually k has a maximum value of 128 or 256
effectively limiting the maximum value of the lag time. The 128 or 256 values
of the autocorrelation function G2 (k · ∆t) are stored in memory registers called
channels. Every time the time ∆t has elapsed a new term is added to each of the
256 sums defining the autocorrelation function. As these terms are products of
the form n(j · ∆t) · n([j + 2] · ∆t) 256 additions and 256 multiplications1 have
to be done during the time ∆t. If ∆t = 0.1 µs it requires a calculation speed
1 Alternatively the whole sequence n(0 · ∆t), n(1 · ∆t), . . . , n(N · ∆t) could be store in
computer memory whereupon the sums of 4.8 could be calculated once the measurements
of 2.56 billion multiplications + 2.56 billion additions per second. So far (2013)
no desktop computer is fast enough to perform this task so instead specialised
hardware is used to calculate the autocorrelation function a so-called digital auto-
correlator usually comprising 256 small, primitive integrated computers running
in parallel. Autocorrelators can be made as expansion boards for desktop PS’s or
as stand-alone boxes. Figure 4.2 shows a selection.
(a) (b) (c)

Figure 4.2: Three dierent digital autocorrelators from companies ALV Brookhaven
Instruments: (a) ALV 5000, (b) ALV 5000/EPP, (c) TurboCorr
Returning briefly to the multiple tau correlators the general principle is simple:
If a sample contains particles with only one diffusion coefficient the sample time
∆t can be set byt he experimenter so that the exponential decay effectively takes
place over the first half of the channels and the remaining half of the channels are
used to establish the baseline. Thus it may take a few tries before a good value of
∆t is found. If the sample contains particles with very different diffusion coeffi-
cient with perhaps a hundredfold ratio it is not possible to choose a sample time
that will enable the capture of the two ensuing exponential decays at the same
time. Also, establishing the baseline may cause some problems and at least the
autocorrelation function will have to be ”patched” together from different meas-
were completed. This of course would set an upper limit to the length of measurements,
a limit which would be of the order minutes rather than hours even with several GB of
memory to store the intermediate photon counts.
urements. With a sample of unknown composition the work involved in selecting

proper sample times and the subsequent treatment of the different autocorrelation
functions can be quite time consuming. But how many channels do you actu-
ally need to determine a single exponential decay? Not very many, say four +
four to establish the baseline, so the 256 channels may seem like a lot. But if
the sample time is chosen so that some small species produces an exponential de-
cay within the first 8 channels (4+4) the smallest diffusion coefficient measurable
with the sample time chosen would be 256/8 = 32 times smaller (corresponding
to particles 32 times bigger). If the sample contains species which differ in size
by more than a factor of 32 more measurements would be necessary. Now, since
the determination of one diffusion coefficient can be done using 8 channels if the
sample time is chosen optimally why not combine a number of correlators operat-
ing at fixed sample times that grow exponentially from one correlator to the next?
So the lag times would be, e.g.:
sub-correlator 1 100 ns 200 ns 300 ns ··· ··· ··· ··· 800 ns

··· 409600 ns ··· ··· ··· ··· ···
··· ··· ··· ··· ··· ···
In this case the sample time increases by a factor of 16 from one sub-correlator
to the next. Other schemes are of course possible, e.g. letting the sample time
increase by a factor of only 8 enabling an overlap between the lag time of the last
channel of one correlator and the first channel of the next correlator.
It should be noted that the autocorrelation function is calculated quantity which
depends on statistically fluctuating photon count rates. The ”precision” or the
quality of the autocorrelation function thus depends on the precision of the terms
in the sums of equation 4.8 as well as on the number of terms. The individual
photon counts can be assumed to have a precision (statistical spread) inversely
proportional to square root of the photon count itself (as is usually the case when
counting random events). Thus, the higher the photon count during the sample
time ∆t the higher the precision of the counts themselves. This goes to show, that
to have an accurately determined autocorrelation function it is desirable to have as

many photon counts as possible during the sample time. As the mean count rate
is proportional to the intensity of the scattered light it is seen (equation 2.41) that
the photon counts are proportional the laser power, the molecular weight and the
weight concentration. The only way to compensate for low molecular weight, low
concentration or low laser intensity is to have more terms in the sums og equa-
tion 4.8, i.e. to measure for a longer time. As only the concentration and the laser
power is under the control of the experimenter and as high sample concentrations
are generally not desirable it is obvious that a high laser power is usually desirable
(as well as high efficiency detectors).
The number of photons n(t), detected during the time ∆t (i.e. from time t to time
t + ∆t) fluctuates for a number of reasons:
1. The individual scattering processes are inherently random
2. The scattering particles move randomly relative to each other (Brownian

motion) creating time varying interference (see equation 4.5)
3. The number of particles in the scattering volume may vary randomly (i.e.
the number N in equation 4.5 is not constant)
4. The individual particles may exhibit varying scattering efficiency depending

on their internal structure and/or their orientation both of which may change
at random
Here only Brownian motion is normally of concern:

If only the statistical nature of the individual scattering processes was present the
autocorrelation function would be constant as all photon counts would be inde-
pendent.
The fluctuating number of scatterers in the scattering volume is usually very small
in relative terms (a typical scattering volume in DLS is 10−3 mm3 but still the aver-
age number of particles at realistic concentrations would be of the order N ≈ 109 .
√
This would give number fluctuations of the order 109 thus giving relative num-
ber fluctuations of the order 10−5 − 10−4 ).
The effect of internal motions are usually of no concern except in gelling sys-
tems and the effect of scattering efficiency being dependent on particle orient-
ation is only of importance if the scattering particles are very large an have a
pronounced dissymmetry (like long rods) because rotational diffusion in all other
cases is so fast that time dependent anisotropic scattering is averaged out during
normal sampling times.
The only mechanism of concern is therefore traslational diffusion (Brownian mo-
tion). What then will the autocorrelation function look like, typically? To answer
this we note that the photon count n(t + k · ∆t) is close to being the same as
n(t), if the time separation k · ∆t between the two counts is very short because
the particles will not have had time to move and change the phases φi . On the
other hand, if the time separation is very long then there will be no connection
(correlation) between the number n(t + k · ∆t) and n(t). Therefore the autocor-
relation function will have a value of hn2 (t)it at short lag times and a value of
hn(t)i2t at very long lag times. The former is always a larger number than the
latter. At intermediate lag times there will be a gradual decay from the high to
the low value as the particles randomly redistribute themselves. A typical plot
of G2 will therefore look like 4.3 The decay of the autocorrelation function can
be envisaged as a gradual loss of memory in the system of particles or indicating
for how long the particles retain their relative positions. The decay looks like an
exponential decay towards an asymptotic value but is only so when the particles
responsible for the scattering all have the same diffusion coefficient D. But now
we shall briefly return to the previously mentioned (p. 74) field autocorrelation
function.
The autocorrelation function

Looking again at figure 4.1, a sketch of a typical DLS-setup, you may have
wondered why the laser beam is focussed and why there are pinholes in front
of the detector. Equation 4.5 seems to offer a perfect explanation why the meas-
ured scattered intensity exhibits time dependent fluctuations: The phases φj and
φk change randomly due to Brownian motion of the scattering particles. But one
2
<n <
< n <2
Figure 4.3: The autocorrelation function for the intensity of light scattered from a
solution of particles undergoing Brownian motion always decays from a value of hn2 i
down to hni2 at very long correlation times, n being the number of scattered photons
counted within a xed time frame ∆t
should note that equation4.5 describes the intensity at a point. And no detector
measures only at a point. The phases φj and φk are the phases of the electric fields
scattered from particles number j and k as observed at a point on the detector.
If the detector (including the lens and the pinholes) in figure 4.1 is replaced by
a screen facing the scattering particles one will see areas of constructive interfer-
ence between the electric fields alternating with areas of destructive interference
(figure 4.4). The bright spots are called speckles. These speckles move about at
random on the screen because the phase differences φj −φk constantly change due
to the Brownian motion of the scattering particles. It turns out that the diameter
of the speckles depends on the diameter of the laser beam: The smaller the dia-
meter of the laser beam the larger the diameter of the speckles. As the principle
of dynamic light scattering is to utilise the fluctuations of the scattered intensity it
is evident that in order to maximise the recorded fluctuations the detector should
see only one speckle at a time: If the detector area is much larger than the area of
one speckle the recorded intensity fluctuations will be small because approxim-
ately the same number of speckles will fall on the detector area all the time. Or, in
Figure 4.4: The speckles on the screen are areas where the scattered electric elds have
the (approximately) the same phase and consequently interfere constructively creating a
high intensity. The dark areas between the speckles are where the scattered electric elds
are in counterphase thus eliminating each other. Positioning the screen further away from
the scatterers will make the speckles larger. Reducing the diameter of the laser beam by
focussing will also make the speckles larger.
statistical terms, the number of speckles falling on the detector area is fluctuating
around a mean value N according to a Poisson distribution which has a statistical
√
standard deviation of N . This means that the relative magnitude of the fluctu-
√ √
ations is N /N = 1/ N , i.e. smaller the more speckles the detector sees. Thus
the ideal detector will have an area which matches the area of the speckles: Either
one or no speckle on the detector. Making the detector area even smaller would
be a disadvantage because then less light would be detected. So why not make the
speckles as large as possible by moving the detector further away from the scat-
tering particles? The answer to this is of course that making the speckles larger
in this way also makes them weaker, so it does’t help. Then why not make the
speckles as large as possible by focussing the laser beam as narrowly as possible?
Here the answer is less straightforward. First of all there is a theoretical limit to
how narrowly a laser beam can be focussed. But more importantly more narrow
focussing requires a shorter focal length focussing lens. The shorter focal length
the closer the lens has to be to the sample. But there are practical limits to how
close the lens can be to the sample. And with a narrow laser beam optical align-
ment of the instrument becomes more critical. And finally, the more narrow the
laser beam becomes the more divergent it also becomes making the initial direc-
tion of the incoming photons more uncertain. This means that narrow focussing
creates an uncertainty in the magnitude of the (length of the) scattering vector q.
And, as we shall see soon, an uncertainty in q directly translates into an uncer-
tainty in the measured diffusion coefficient or equivalent hydrodynamic radii. So,
in practice most DLS instruments make the same trade-offs: The laser beam is
focussed down to a diameter to about 1/10 mm and the diameter of the pinhole in
front of the detector is also about 1/10 mm (the diameter of the speckles).
Now, let us return to the two autocorrelations functions, the intensity autocorrel-
ation function G2 (τ ) (equation 4.6) and the field autocorrelation function G1 (τ )
(equation 4.7). It is customary to normalise the functions so that the normalised
intensity autocorrelation function g2 (τ ) has the asymptotic value of one and the
normalised field autocorrelation function g1 (τ ) has the initial value of one (as the
asymptotic values is zero and hence cannot be used for normalisation):
g1 (τ ) = G1 (τ )/G1 (0) (4.10)

g2 (τ ) = G2 (τ )/G2 (∞) (4.11)
The relevance of the field autocorrelation function g1 is that it is directly related

to the fundamental physical processes creating the fluctuations: The interaction
of the electric field with the particles undergoing random motions and the inter-
ference between scattered fields. The relevance of the intensity autocorrelation
function g2 is that it is related to what is directly observable (the fields are not
observable as the oscillate by far too rapidly). Fortunately, and not surprisingly,
there exists a relationship between the two autocorrelation functions, the so-called
Siegert relationship:
g2 (τ ) = β[g1 (τ )]2 + 1 (4.12)
where β is the spatial coherence factor a number which is always less than or
equal to one. It is approximately one divided by the average number of speckles
that the detector sees. Here again it is seen that the use of a large detector area
does not produce better data because the informative part of g2 which is [g1 (τ )]2
is reduced by the factor β which becomes smaller the larger the detector area,
eventually wiping out all information in g2 . So, again, it is seen that the DLS setup
should be optimised: The diameter of the focussed laser beam an the diameter of
the pinholes should match.
So, what does the intensity autocorrelation function look like in practice? With a
good deal of qualitative reasoning one can argue what the analytical form of the
autocorrelation function should be: Look at figure 4.5 Let us consider diffusion in
ctor
D ete
Figure 4.5: Light that reaches the small entrance to the detector will exhibit intens-
ity uctuations. The reason for this is that the scattering particles create constructive
interference in some directions and destructive interference in other directions and these
directions change when the particles move relative to each other. As a result intensity
maxima and minima sweep across the detector entrance due to the Brownian motion of
the scatterers. In order to have large intensity uctuations on the detector the entrance
hole must be small enough that only one interference maximum can fall on it at a time.
one dimension. The squared distance a particle will move from its initial position
by diffusion during time t is given by the well-known expression:
hx2 i = 2Dt (4.13)

An intuitive characteristic distance scale in this context would be the wavelength
of the light because destructive interference between two sources of scattered light
occurs when their separation is a multiple of half wavelengths. But the inter-
ference not only depends on the distance between the scatterers but also on the
scattering angle. Therefore the simplest ”characteristic” distance in the context
of light scattering is 1/q (q being the length of the scattering vector) because ac-
cording to equation 2.30 light scattered from a particle separated the distance π/q
from another particle will interfere destructively with light scattered from the first
particle. Combining this with equation 4.13 we can ”derive” a characteristic time
τ0 for the diffusion process: !2
1
= 2Dτ0 (4.14)
q
whereby τ0 = 1/(2Dq 2 ). Of course this approach does not account for the precise
numeric constants in equation 4.14. It might actually seem more natural to write
π
q
instead of just 1q . But if we leave equation 4.14 as it is it does not seem unnatural
to guess that the autocorrelation function in this simple case (of just one diffusion
coefficient involved) would be an exponential decay with the characteristic time
constant τ0 , i.e. G2 would be of the form:
G2 (τ ) = Ae−τ /τ0 + B
(4.15)
2
= Ae−2Dq t + B
Miraculously, this result turns out to be correct and we shall not give a rigorous
derivation here2 . In the more general case where n different types of particles are
present each type having different diffusion coefficient D1 , D2 , D3 , . . . , Dn , we
can generalise equation 4.15 without proof to:
(4.16)
2τ 2τ 2
G2 (τ ) = (A1 e−D1 q + A2 e−D2 q + . . . + An e−D1 q τ )2 + B
where the amplitude factors A1 , A2 , . . . , An are proportional to to the light scat-

tering contribution from each class of particles through the relationship
Aj ∝ Pj (q)Cj Mj (4.17)
2 For those who do not believe in miracles a more rigorous derivation can be found in
chapter 10 in Dynamic Light Scattering, R. Pecora, Ed., Plenum Press 1985 especially if
combined with chapter 10 in Biophysical Chemistry, C. Cantor & P. Schimmel, Freeman
and Company 1980
4.3. Data analysis 85
where Pj is the form factor for the j’th species, Cj is the weight concentration and
Mj is the molecular weight.
We can normalise the measured autocorrelation function G2 in the usual way so
that it attains the asymptotic value of one for infinite lag time (equation 4.11).
With the expressions (4.17) for the amplitude factors the normalised autocorrela-
tion function can then be written:
(4.18)
2τ 2τ 2
g2 (τ ) = β(a1 e−D1 q + a2 e−D2 q + . . . + an e−D1 q τ )2 + 1
where a1 , a2 , . . . , an denote the fractions

P P P
P1 (q)C1 M1 / Pi (q)Ci Mi ,P2 (q)C2 / Pi (q)Ci Mi , . . . , Pn (q)Cn / Pi (q)Ci Mi
and β denotes the spatial coherence factor. The sum in the bracket is simply the
field autocorrelation function. As discussed in connection with the Siegert rela-
tionship (equation 4.12) this is a number which is smaller than one and which
depends on the detector geometry (and is thus an apparatus dependent constant).
The more interference maxima the detector sees at the same time the smaller the
coherence factor becomes. In practice the coherence factor also depends on the
amount of uncorrelated light (or actually detector signal) the autocorrelator re-
ceives: Stray light from the surroundings entering the detector will diminish the
coherence factor as will noise from the detector (noise being ”signal” generated
by the detector itself and not by actual light).
The expression 4.18 is usually the basis for the analysis of dynamic light scattering
data on diffusing particles3 .
4.3 Data analysis

How to analyse autocorrelation data is a comprehensive subject as we shall see in
the following, just scratching the surface.
3 Equation 4.18 only describes diusion of comparatively dilute suspensions of particles.
And the particles should not exhibit internal motions, like bending or stretching on the
length scale of 1/q or at least do so rapidly that the characteristic time scale is short
compared to the shortest lag time of the experiment. Also, equation 4.18 does not de-
scribe internal motions in the system which are not related to diusion processes like e.g.
vibrational modes in gels.
The methods of data analysis can be divided into direct methods and indirect meth-
ods:
The problem is usually the following: Assume that a normalised autocorrelation
function g2 (t) has been measured. How can this measured autocorrelation func-
tion be written in the form of equation 4.18? What are the values of the different
diffusion coefficient and how many (n) are there? Or stated in a slightly more
general way: If we substitute the sum in equation 4.18 with an integral:
Z ∞ 2
g2 (τ ) = β A(D)e −Dq 2 τ
dD +1 (4.19)
0
where A(D) is the distribution of diffusion coefficient which ”generates” the

measured intensity autocorrelation function, how do we find the distribution A(D)
which gives us the measured autocorrelation function when substituted into equa-
tion 4.19? Note that the integral in the bracket represents the field autocorrelation
function: Z ∞
(4.20)
2
g1 (τ ) = A(D)e−Dq τ dD
0
So we may equally well look for the distribution A(D) which gives us the meas-
ured field autocorrelations function g1 when substituted into equation 4.20. The
field autocorrelation function g1 can in principle be obtained from the measured
intensity autocorrelation function g2 through the Siegert relationship 4.12.
There exits a mathematical method called an inverse Laplace transformation which,
at least in principle, can calculate the function A(D) in equation 4.20. This would
be a direct method of data analysis. As we shall see shortly this direct method
does not work in practice. It leads to solutions that are unrealistic meaning that
they usually show a mixture of features some describing the system on which the
measurement was done and some features that are simply artifacts of doing a direct
inversion on a measured (and therefore noise containing) autocorrelation function.
We are therefore left with only indirect methods. An indirect method means that
you assume that you know the mathematical form of the autocorrelation function,
that you can write an expression for it containing some adjustable parameters. The
indirectness of the method is that you then adjust the parameters of this function
so that it best fits the measures autocorrelation function. Finding the best fit means
to minimise the deviation between the measured autocorrelation function and the
model function. This deviation is measured by the quantity called the chi-square,
χ2 . The smaller the value of χ2 the better the fit resembles the measured autocor-
relation function. How this works is described in appendix B. But how does one
choose the fitting function?
For a start one could try to simplify the problem e.g. based on an assumption that
you know something about your system. The simplest assumption would be that
the system contains only type of particles, i.e. there is only one diffusion coeffi-
cient involved. Such systems are called monodisperse whereas systems containing
more than one type of particles are termed polydisperse. We shall look at these
two possibilities in turn.
Monodisperse solutions
If the autocorrelation function is measured on a system with only one diffusion
coefficient the normalised autocorrelation function g2 can – using equations 4.15
and 4.18 – be written as:
(4.21)
2τ
g2 (τ ) = β · e−2Dq +1
where β is the aforementioned coherence factor. If we now take the natural log-
arithm on both sides of equation 4.21 we get:
ln(g2 (τ ) − 1) = ln β − 2Dq 2 τ (4.22)
so, if ln(g2 (τ ) − 1) is plotted versus q 2 τ we get a straight line with the slope −2D
(which of course is determined by linear regression). If, on the other hand a plot
of ln(g2 (τ ) − 1) versus q 2 τ does not produce a straight line this of course means
that more than one diffusion coefficient is present in the system, i.e. the system is
polydisperse.
Polydisperse solutions
In real life it is rare that solutions only contain one type of particles so they can be
represented by only one diffusion coefficient. Again one can try to simplify the
problem by assuming that the autocorrelation function has a specific form based
on e.g. some knowledge or presumptions about the system. This could be that the
system is composed of only two different species or it could be that the system
contains particles with a narrow size distribution so that the diffusion coefficient
can be described by a Gaussian distribution or are distributed according to some
other known distribution function given by a relatively small number of paramet-
ers. In all of these cases the parameters are found through some type of regression
analysis (see Appendix B). The details of the fitting procedure depend on the pre-
cise assumptions one has made about the system and thus about the mathematical
form of the fitting function or about the distribution A(D) of the diffusion coeffi-
cients.
We shall now look at a few of the analysis methods:
Multiexponential analysis
This is straightforward in principle: Simply assume that the measured autocorrel-

ations function can be written as in equation 4.18:
2τ 2τ 2
g2 (τ ) = β(a1 e−D1 q + a2 e−D2 q + . . . + an e−D1 q τ )2 + 1
Then, by non-linear regression (see Appendix B) have a computer adjust the para-
meters β, a1 , a2 , . . . , an and D1 , D2 , . . . , Dn until the calculated g2 deviates as
little as possible from the measured autocorrelation function. In fact only 2n and
not 2n + 1 as it might seem need to be adjusted because the a’s, being fractions of
light scattering contribution are not independent: a1 + a2 + · · · + an = 1.
This seemingly simple procedure sometimes works well but nevertheless has a
number of flaws:
1. You must decide in advance how many terms to include, i.e. the value of n
2. The best fit is not necessarily a good fit. The goodness of the fit is determ-
ined on the basis of the value of the so-called reduced χ2 (see Appendix B).
It should ideally be close to one. Another test is to look at a residual plot.
The residual plot is a plot of the difference between the measured autocor-
relation function and the fit, i.e. a plot of g2, measured (τ ) − g2, fit (τ ) vs. τ . The
ideal is that this difference looks completely random being sometimes pos-
itive, sometimes negative with no particular system. This situation indicates
that the residual simply represents the ”noise” in the measured autocorrela-
tion function. But sometimes (often) the residual plot shows a pronounced
structure like an exponential or perhaps a sine-like function. This indicates
that there is something apart from noise in the measured autocorrelation
function that the fit can not account for: The fitting function has too few
terms that you can or perhaps the wrong kind of terms. In any case, clear
structure in the residual plot shows that you have the wrong model (fitting
function).
3. You will always get a better fit the more terms you include in the fitting
function because you get more parameters to adjust.
4. The drawback of adding more terms is that the parameters found become
more and more irreproducible the more parameters you include: If you
make, say, ten repeat measurements on the same system to get ten meas-
ured autocorrelation functions then the fitting procedure will not give you
the same values for the fitted parameters for each of these measured auto-
correlation functions. Some may not be very different but others may differ
so much that it is impossible to say what set of diffusion coefficient best
represents the system. The catch here is that using few parameters will
sometimes generate a bad fit but the parameters will be reproducible upon
repeat measurements. Increasing the number of parameters to improve the
quality of the fit will produce increasingly irreproducible and thus meaning-
less parameter values.
How do you decide how many terms to include? Sometimes the autocorrelation
function can actually be fitted well by using two terms that also give reproducible
parameter values. One term rarely works well at least because there may be trace
amounts of relatively large impurities present in the sample or the laser may pro-
duce light with an intensity which is not 100% constant. Random and relatively
slow fluctuations of the laser intensity may be interpreted as a result of (usually)
large particles in the sample so a term to account for this is nearly always needed
in the fitting function. But even with three terms reproducibility begins to be an
issue and at four terms the limit is reached for the practical application of this
procedure.
Cumulant analysis
We saw in section 4.3 that in the case of monodisperse solutions the analysis
of DLS data was simple. If the solution is not monodisperse the relation 4.22
of course no longer holds. Still, the philosophy behind cumulant analysis (also
called cumulants analysis) is that if the solution is not monodisperse but has a
rather narrow Gaussian-like distribution of sizes then a plot of ln(g2 (τ ) − 1) vs.
q 2 τ should produce if not a straight line then at least look like a polynomial. So, let
us assume that diffusion coefficient (not the sizes) follow a Gaussian distribution.
Then the function A in equation 4.19 can be written:
(D−D0 )2
A(D) = A0 e− 2σ 2 (4.23)
where D0 is the mean diffusion coefficient, σ is the standard deviation of the

R∞
distribution, and A0 is a normalisation constant ensuring that 0 A(D)dD = 1.
With this equation 4.19 assumes the form:
Z ∞ (D−D0 )2
2
(4.24)
2
g2 (τ ) = A0 e− 2σ 2 e−Dq τ dD +1
0
which, after few integral substitutions, becomes:

!2
√ Z ∞
g2 (τ ) = e −D0 q 2 τ +σ 2 q 4 τ 2 /2
A0 2σ √ √ e −x2
dx +1 (4.25)
−D0 / 2σ+σq 2 τ / 2
√ √
If τ is so small that D0 / 2σ σq 2 τ / 2 the integral in equation 4.25 is practic-
ally constant (independent of τ ) and we can rewrite (approximate) equation 4.24:
(4.26)
2 τ +σ 2 q 4 τ 2
g2 (τ ) ≈ C · e−2D0 q +1
where C is the constant given by the expression:

!2
√ Z ∞
C = A0 2σ √ e −x2
dx (4.27)
−D0 / 2σ
It is seen that ln(g2 (τ ) − 1) = −2D0 q 2 τ + σ 2 q 4 τ 2 = −2D0 · (q 2 τ ) + σ 2 · (q 2 τ )2 i.e.

a second order polynomial in the variable q 2 τ . The coefficient of the first order
term is the mean diffusion coefficient 2D0 og and the coefficient of the second
order term is the variance σ 2 of the Gaussian distribution. Thus a parabolic fit to
ln(g2 (τ ) − 1) will thus give both the mean diffusion coefficient as the variance og
the distribution. Both under the assumption that only data points corresponding
to relatively small values of τ are used. This is also clear for another reason:
The autocorrelation function is a decreasing function whereas the approximating
parabola will eventually diverge at large values of τ .
The quantity (σ/D0 )2 is usually called the polydispersity or the the polydispersity
index of the distribution. If we set the polydispersity (or σ) equal to zero we see
that equation 4.26 reduces to the earlier equation 4.21 for monodisperse solutions.
This method of making a parabolic fit to ln(g2 (τ ) − 1), called second order cumu-
lant analysis, must be used with some caution. The validity of the method requires
that the distribution of diffusion coefficient has at least some resemblance with a
Gaussian distribution, i.e. it should be unimodal (monomodal) and not multimodal
(having some clearly separated size classes). Sometimes fitting with polynomials
of higher order than two is used in which case you talk about third order (using
a third order polynomial) or fourth order cumulant analysis. In all cases it is im-
portant not to extend the fitting procedure to τ values that are too large: The poly-
nomial used will always diverge at large τ and thus become a bad representation
of the autocorrelation function which is always asymptotically decreasing. On the
other hand one should make sure to use enough data points to make a meaningful
fit. Usually one does not go beyond third order cumulant analysis because the
extra parameters are normally poorly determined, i.e. they vary a lot when the
analysis is performed on autocorrelations functions obtained by simply repeating
a measurement. The parameters of a second order cumulant analysis (mean diffu-
sion coefficient and polydispersity) are usually quite reproducible which explains
why this method is so popular. One should note, however, that the individual
data points are weighted differently in a cumulant analysis than in an analysis
employing a direct fit to the measured autocorrelation function. This means that
when employing direct fitting procedures the mean diffusion coefficient and the
polydispersity may come out with slightly different values than when cumulant
analysis is used.
On can wonder what it entails that the diffusion coefficient are described by a
Gaussian distribution and how the diffusion coefficient of the individual species
are weighted when the cumulant analysis calculates a mean diffusion coefficient
D0 . To investigate this further note that at short correlation times the autocorrela-
tion function 4.26 can be written as
(4.28)
2
g2 (τ ) ≈ C(e−D0 q τ )2 + 1
meaning that ln(g2 (τ ) − 1) plotted vs. q 2 τ is a straight line for sufficiently small
values of τ . If we now make a Taylor series expansion to first order of both
equation 4.28 and equation 4.18 and subsequently equate the results we get:
n
ai (1 − Di q 2 τ ) = 1 − D0 q 2 τ (4.29)
X
i=1
leading to
n
(4.30)
X
D0 = ai D i
i=1
or, with the expressions for the fractions ai written out:
n
X
Pi (q)Ci Mi Di
D0 = i=1
n
X
(4.31)
Pi (q)Ci Mi
i=1
This type of mean value is called the z-average of the diffusion coefficients:
D0 = hDiz (4.32)
This expression is seen to be exactly analogous to the expression for the mean
radius of gyration 3.33.
Inverse Laplace transformation
The determination of the distribution of diffusion coefficient is mathematically

equivalent to doing an inverse Laplace transformation of the previously mentioned
field autocorrelation function g1 (τ ). To make a Laplace transformation of some

function f (Γ) means to calculate the function g(t) (the Laplace transform of f )
given by the expression:
Z ∞
g(τ ) = f (Γ)e−Γt dΓ (4.33)
0
Note that if t is time then Γ has the dimensions of inverse time. From the meas-
ured autocorrelation function one can define the field autocorrelation function as
q
g1 (t) = |g2 (t) − 1|. It is necessary to take the absolute value under the square
root because noise in the measured autocorrelations function g2 may give occa-
sional values of the baseline below one. If one then makes a change of variables,
Dq 2 = Γ and defines the function B(Γ) describing the distribution of Γ:
1
B(Γ) = A(D/q 2 )
q2
then it is seen that equation 4.20 is equivalent with:
Z ∞
g1 (τ ) = B(Γ)e−Γt dΓ (4.34)
0
As g1 (τ ) is determined experimentally (through g2 (τ )) all that is needed is to

perform a so-called inverse Laplace transformation of g1 in order to find the func-
tion B(Γ) and thereby the distribution of diffusion coefficient A(D). This is a
mathematically well-defined operation. But even though there exists computer al-
gorithms that can perform inverse Laplace transformations it turns out that the
result of this operation is highly ambiguous in the sense that if one repeats a
measurement by simply letting the sample stay in the instrument an run a new
measurement the ensuing distribution function A(D) of diffusion coefficient (or,
equivalently the distributions B(Γ)) can be very different. In fact, the results of
inverse Laplace transformations can deviate arbitrarily much from the ”true” dis-
tribution. The source of this ambiguity is ”noise” in the experimentally determ-
ined function g1 (τ ). One says that the inversion problem is ”ill conditioned” or
”ill-posed”. It is not difficult to see how this comes about: Let us assume that
for a given field autocorrelation function g1 (τ ) we have found a corresponding
distribution of the gammas, B(Γ) which of course can be translated into an equi-
valent distribution of diffusion coefficients. Now, to this distribution function add
a term b · sin(Γτ1 ), where b is an amplitude constant and τ1 is a time constant.

Doing a Laplace transformation on this sum of functions will produce a result
R∞
which is is 0 (B(Γ) + b · sin(Γτ1 )) · e−Γτ dΓ. The result is (look in a table of
bτ1
integrals): g1 (τ ) + τ 2 +τ12
. The last term of this expression is the deviation from
the experimentally determined field autocorrelation function g1 . This term has the
b
maximum value of τ1
(for τ = 0). This maximum value can be made as small as
you wish just by choosing τ1 large enough. This means that you can add a term
to the distribution B(Γ) with an arbitrarily large amplitude, b but with no visible
consequence to the field autocorrelation function. This is so because the term ad-
bτ1
ded to g1 is τ 2 +τ12
which can be made arbitrarily small by choosing τ1 sufficiently
large. A similar argument shows that adding a cosine component b·cos(Γτ1 ) to the
distribution function B(Γ) will also have negligible impact on the ensuing Laplace
transformed function. This argument shows that you can add any number of arbit-
rarily large terms to any found distribution B(Γ) and still reproduce the measured
field autocorrelation function within the limits of experimental precision. Adding
such sine and cosine terms with different time constants τ1 , τ2 , . . . has the general
impact on the distribution function B(Γ) to give it more ”structure”, more spikes,
more detail. This means that when doing an inverse Laplace transformation on a
measured autocorrelation function you actually never know what features of the
distribution function to trust. What features reflect properties of your system and
what features are accidental artifacts due to noise in the measured autocorrela-
tion function in combination with the inverse Laplace transformation algorithm?
So, the problem can be stated in another way: Apparently there is no limit to the
amount of ”detail” or ”information” you can get in your solution. But how much
of this information is real? There are a number of different ways to cope with this
problem, as we shall see.
ILT
One of the first data analysis methods to address this problem was ILT (Inverse
Laplace Transformation, Ostrowsky m.fl. (1981)). The idea behind this method
is to fit the experimentally measured field autocorrelation function g1 (t) with an
expression of the form:
f (τ ) = a1 e−Γ1 τ + a2 e−Γ2 τ + . . . + aN e−ΓN τ (4.35)
where the relaxation constants Γi are spaced exponentially: Γi = ϕΓi−1 where ϕ

is a constant determining the separation between consecutive relaxation constants.
This way of distributing the relaxation constants means that they will be spaced
equidistantly on a logarithmic scale. What is determined when fitting method the
expression 4.35 to the field autocorrelation function is only the constants a1 , a2 ,
. . . , aN while the ”fineness of the exponential grid”, ϕ, is determined in advance
based on a knowledge of the noise in the measured g1 . Also, the the smallest and
largest relaxation constants Γ1 and ΓN are set in advance.
CONTIN
Yet another way to limit the number of possible distribution functions B(Γ) is to
impose constraints on the solution. The idea is to somehow minimise the amount
of structure in the solution without impairing too much how well the solution
reproduces the measured autocorrelation function when Laplace transforming it.
The data analysis method CONTIN (Provencher 1982) uses regularisation to min-
imise the amount of structure in the solution: Instead of finding a solution B(Γ)
which produces the best fit to the measured autocorrelation function, i.e. giving
the smallest possible value of χ2 the procedure attempts to minimise the expres-
sion Z Γmax
2
χ +α 2
(B 00 (Γ))2 dΓ (4.36)
Γmin
The term B 00 (Γ) is a measure of the curvature of the function B. If the solution
B has a sharp peak it will have a high curvature near the top. If the solution has
several closely spaced peaks it will have high (but negative) curvature near the
top of each peak but also near the bottom between peaks. So, taking the square
of B 00 makes sure that positive and negative curvature don’t cancel each other.
The integral measures the overall ”spikiness” through the amount of curvature
of the solution. And a lot of this type of detail is usually unwanted because it
is unrealistic. Therefore the fitting procedure tries at the same time to make χ2
small (in order to make a good fit) and to make the integral of the expression 4.36
small in order to make a the solution have a realistically small amount of structure.
Which of the two terms count the most depends on the value of the regularisation
parameter α. If e.g. α = 0 only χ2 counts and the best fit is found irrespective of
its possible unrealistic amount of detail. If on the other hand α is very large χ2
counts very little and a solution B(Γ) with very little detail is found. It will also
reproduce the measure autocorrelation function poorly, giving a large value of χ2
but still, if α is large enough this doesn’t matter. The problem which CONTIN
solves is to set a sensible value for α. In terms of computing time CONTIN is
somewhat slower than ILT. The difference between the solutions found by the two
method is that CONTIN always produces wider and sometimes fewer peaks than
ILT. One can say that CONTIN gives you the least detailed solution (distibution
of diffusion coefficients) which is not in serious disagreement with the measured
autocorrelation function.
Other methods
Within the last 30 years a number of different method to analyse autocorrelation

functions have been developed. Cumulant analysis is one of the oldest but still has
widespread use despite its obvious limitations, especially with modern correlators
which operate with lag times that are spaced exponentially instead of linearly as
was customary in the early days of dynamic light scattering and which was de-
scribed in section 4.2. The newer methods are usually related to CONTIN and
ILT in the sense that they use regularisation of some kind. Examples are REPES
og RNONLIN which both analyse g2 directly (without calculating g1 first). Also
MEM (maximum entropy method) is sometimes used but has not yet gained wide-
spread usage. The principle of MEM is to assign to each of the possible solutions
(B(Γ)) an ”entropy” (probability) and then choose the solution with the highest
entropy.
A method often used is to represent the data as a squared sum of exponentials as

shown in equation 4.18 and then do a non-linear least squares fit the the meas-
ured field autocorrelation function, g2 , to extract the parameters a1 , a2 , . . . og
D1 , D2 , . . .. It is necessary to limit the number of terms in the sum to e.g. 3
to avoid strong coupling (also called ”cross correlation”) between the parameters.
If the method works it allows you to find three sharply separated diffusion coef-
ficient (or particle sizes). But often it doesn’t work because there are more than
three particle classes. If they have widely separated diffusion coefficient more
terms can be added but the method usually doesn’t give useful and reproducible
results with more than four terms, at the most. Sometimes the failure of fitting
with a small sum of the type 4.18 is due to particle diffusion coefficient not falling
into sharply defined classes but the classes exhibit polydispersity. It is possible to
take this into account in an ad hoc way by fitting instead with a function of the
Kohlrausch type:
(4.37)
2 τ )β1 2 τ )β2 2 τ )βn
g2 (τ ) = β(a1 e−(D1 q + a2 e−(D2 q + . . . + an e−(D1 q )2 + 1
where β1 , β2 , . . . , βn are all numbers lying between 0 and 1. If all β exponents

are 1 we get the same type of functions as before (equation 4.18) but the closer
βi is to 0 the wider the distribution is belonging to the corresponding term. These
individual terms are normally referred to as ”stretched exponentials”. Their main
use in the analysis of autocorrelation functions for gelling systems, i.e. systems
where the correlated intensity fluctuations are not necessarily related to diffusion
processes in a solution but more to coupled motions in the gel.
5
Light scattering in practice
We shall here look at some of the ways in which light scattering measurements
are done in practice.
5.1 Sample preparation

In light scattering sample preparation is crucial. Of course this is always so, but
in light scattering special precautions need to be taken. The reason is dust. The
light scattering power of a species is proportional to M C, molecular weight times
weight concentration. Or put another way: The light scattering power of indi-
vidual particles is proportional to M 2 (see equations 2.25 and 2.41). This means
that even the tiniest amounts of dust in a sample can be devastating to measure-
ments. The same goes for aggregated material. The precaution against this is
99
100 5. Light scattering in practice
thought cleaning of buffers, sample cells and filtering of finished samples. Filter-
ing the sample through a 0.2 µm syringe filter is standard procedure. Often putting
the filtered sample into the syringe again filtering it once more through the same
filter works wonders. The reason is that standard syringe filters often release small
amounts of ”dust” from the membrane during the filtering procedure. This dust is
the retained by the filter during the second round of filtration.
5.2 Apparatus
It was promised earlier, in section 3.5, that a more thorough description would
follow of how detailed measurements of molecular weights can be done. The
principle of the method is to separate the different molecular species in the solu-
tion so that the measurements are done on one class of molecules at a time and
not on a mixture of species. One way this can be done is by passing the solutions
through a size exclusion column which is usually connected in-line to the light
scattering instrument and a concentration detector. The in-line setup is of course
not crucial to the principle but is a major convenience. The in-line setup requires
that the concentration detector and the light scattering instrument be equipped
with flow cells. For concentration detector this has been the standard mode of
operation for many years but for light scattering it is a mode of operation offered
by only a few companies. A light scattering flow cell from Wyatt Technology is
shown in figure 5.1. In this flow cell the laser beam passes trough in the direction
of the sample flow. Detectors are place around the flow cell in the instrument al-
lowing the measurement of the Rayleigh ratio at several angles simultaneously.
The separation principle of a size exclusion column is that larger molecules pass
through the column faster than smaller molecules. The mechanism behind this
seemingly ”backward” behaviour is that the column material is porous and spongy
on many different size scales, small and large pores are interconnected. The lar-
ger molecules can only pass though the mesh of large interconnected pores in the
column material whereas the small molecules can get into also some of the smal-
ler pores where they will spend some time. Also, the speed of the carrier liquid is
5.2. Apparatus 101
Figure 5.1: A light scattering ow cell from Wyatt Technology. Other companies use
ow cells of other designs.
lower in the small pores due to viscous drag. So, consequently the molecules are
sorted according to their size rather than their molecular weight.
It should be mentioned that size exclusion chromatography is not the only way to
separate particles according to size. In recent years field flow fractionation (FFF)
and asymmetric flow field flow fractionation (A4F) have gained widespread use
especially for systems where traditional size exclusion columns have difficulties
doing the separation.
In order to see how separating the different molecular species in a sample helps
retrieve information about the system we can combine equations 3.16 and 3.14 to
get
R = KCMapp. (5.1)
where we have assumed that the concentration is low enough that we can disregard
the virial coefficients of equation 3.16. If we measure the Rayleigh ratio R and
the concentration C of the suspended, separated molecules we can calculate the
apparent molecular weight:
R
Mapp. = (5.2)
KC
where we assume the optical contrast constant K is known. In practice one often
circumvents the step of knowing the contrast constant. We shall return to this issue
shortly. The Rayleigh ratio is usually measured at several angles at the same time
so that a true molecular weight can be obtained by extrapolation to zero scattering
angle.
How to determine concentration

One of the steps in the calculation of the molecular weight is to measure the con-
centration of the molecules. This is frequently done by measuring the UV ab-
sorption of the sample, but this of course only works if the suspended molecules
absorb UV radiation. This will be the case if the molecules under investigation are
proteins or amino acids but not if they are polysaccharides. Instead one can meas-
ure the refractive index n of the solution because its magnitude will depend on the
amount of dissolved material and of course also on the refractive index of the pure
solvent. In fact, up to quite high concentrations (more than 15% by volume) there
is a linear relationship between the weight concentration of the dissolved material
and the difference in the refractive index n0 of the pure solvent and the refractive
index n of the solution. Therefore the concentration can be determined by a meas-
urement of n − n0 which is conveniently done with a differential refractometer.
This instrument contains a double flow cell where one of its compartments holds
the pure solvent and the other has the solution running through it. The instrument
then outputs a voltage which is proportional to the difference in refractive index
n − n0 . If you make up a number solutions of different concentration of the pro-
tein under investigation, e.g. 0 g/L, 1.5 g/L, 2.0 g/L, 2.5 g/L, 3.0 g/L and 3.5 g/L
and then send these solutions through the diffractometer sequentially the output
of the instrument might look something like figure 5.2. Next, the the measured
refractive index differences are plotted vs. the concentration of the solute. The
result is shown in figure 5.3
What this means is that as long as the linear relationship holds we can write:
dn
n − n0 = ·C (5.3)
dC
dn
Thus, a knowledge of the value of dC
enables us to determine the concentration
dn
through a measurement of n − n0 . It turns out that the value of dC
depends
5.2. Apparatus 103
0,0006
Refractive index difference n - n0
3.5 g/L
0,0005
2.5 g/L
0,0004
2.0 g/L
0,0003
1.5 g/L
0,0002
1.0 g/L
0,0001 0.5 g/L
0,0000 0 g/L
0 2 4 6 8 10
minutes
Figure 5.2: The refractive index of a solution depends on the refractive index of the
solvent, on the refractive index of the pure solute and on the weight concentration C
of the solute. The instrument measures the dierence n − n0 between the refractive
index n0 of the solvent and the refractive index n of the solution. Solutions of increasing
concentration are pumped through the instrument. Solution of unchanged concentration
is pumped through the instrument until the output for that concentration has reached a
stable value, ie.e a plateau. Next to each plateau is shown the corresponding concentration
of the solute.
0,0006
Refractive index difference n - n0
0,0005
0,0004
0,0003
0,0002
0,0001
0,0000
0 0,5 1 1,5 2 2,5 3 3,5
Concentration C (g/L)
Figure 5.3: Refractive index dierence as a function of the weight concentration of the
solute. The points are taken fron the plateau values of gure 5.2. The relationship is
dn
linear (in most cases up to quite high concentrations). The slope of the regression line
dC
is characteristic of the combination of solute and solvent.
mainly on the type of molecule dissolved and on the solvent. For water-based
solvents (buffers) the value is usually around 1.85 · 10−4 m3 · kg−1 for proteins
and approximately 1.45 · 10−4 m3 · kg−1 for (poly)saccharides. The precise value
will be somewhat dependent on the refractive index of the solvent and on other
properties of the solute/solvent system (like, e.g. whether volumes are additive
upon mixing). But in most cases the two quoted values will be accurate to within
5%.
dn
Notice, that looking up dC
values in the literature can be confusing as different
authors often disagree on the values. If a precise value is of importance to the
measurements it is necessary to do one’s own determination preparing a number
of different concentrations as described above. This will be especially important
if the solvent is in some way ”unusual”, e.g. containing also apolar components,
like alcohol or acetone.
Just for the sake of argument rather than for its usefulness it is possible to get a
dn
rough estimate of the dC
value of a solute in a given solvent just by considering
the difference in refractive index between the pure solute and the pure solvent.
When the solute is present in its pure form (i.e. without being dissolved) it has a
weight concentration which is the same as its density (look at the units, both mass
pers unit volume). this means that we can make the estimate
dn ∆n nrent stof − nbuffer
≈ = (5.4)
dC ∆C ρrent stof
because ∆C = Cmax − 0 = ρpure solute . The refractive index of pure, water-
free proteins is typically about 1.60 and for (poly)saccharides about 1.56 whereas
the refractive index of most water-based buffers is around 1.33. The density of
water-free proteins is typically ρ ≈ 1.4 · 103 kg · m−3 and for polysaccharides
ρ ≈ 1.58 · 103 kg · m−3 . Using equation 5.4 we obtain the following estimates:
!
dn 1.60 − 1.33
≈ 3 −3
= 1.93 · 10−4 m3 · kg−1 (5.5)
dC protein
1.4 · 10 kg · m
!
dn 1.56 − 1.33
≈ 3 −3
= 1.44 · 10−4 m3 · kg−1 (5.6)
dC polysac
1.58 · 10 kg · m
Both estimates lie within 5% of typical values.

5.2. Apparatus 105
Inline measurement of concentration in conjunction with

SEC
When one makes a determination of the molecular weight of dissolved molecules

using a setup as shown i figure 3.2 it is seen that the solution passes through the
light scattering instrument before it passes through the concentration detector (the
RI detector). Nevertheless we shall discuss the RI detector first.
In order to do a measurement a small volume (typically 100 µL) of the dissolved
molecules is injected into the flow of buffer which is constantly being pumped
through the column and the rest of the setup. Let us for simplicity assume that
only one kind of solute molecules are present in the solution. When these mo-
lecules pass through the RI detector it will temporarily detect a higher refractive
index difference n − n0 . If the molecules went through the column in exactly the
same time the RI detector signal would suddenly jump to a higher value, the stay
constant for a short while as the molecules passed, and then suddenly drop down
to its original value again when all the molecules had passed (If the RI detector
is ”nulled” perfectly the signal will be zero when no solute molecules pass). But
all molecules don’t take exactly the same path through the column, so they will
elute at slightly different times. So, instead of this ”square” like output from the
RI detector the signal will look like shown in figure 5.4. The RI detector meas-
ures the difference in refractive index between the solution being pumped through
its measurement cell and a liquid in a reference cell. The other liquid is usually
just the pure solvent used to dissolve the molecules. In principle the RI detector
should produce a zero signal when just pure buffer is running through the meas-
urement cell but in practice this is usually not quite so. There will be an offset for
a number of reasons e.g. small differences in the optical system of the measure-
ment cell and of the reference cell, and small differences in the composition of the
solvent used to make up the solution and the solvent running through the system1 .
1 RI measurements are extremely sensitive. Usually the running buer which is in a

reservoir is fed through the measurement cell and the reference cell at the beginning of
a measurement. Then after some minutes the reference cell is shut o so that the buer
only runs through the measurement cell. A dierence in refractive index between the two
cells may develop if evaporation of water from the reservoir takes place thereby increasing
0.50
0.40
0.30
URI (volt)
0.20
0.10
URI, baseline
0.00
0 5 10 15 20 25 30 35
Elution time (minutes)
Figure 5.4: The signal from the refractive index detector (RI detector) lies at a constant
value, the baseline, as long as only buer is running through the measurement cell. The
baseline value is ideally zero but often has a slight oset, which is of no importance. As
the dissolved molecules pass through the measurement cell the signal rises and then falls
back to its baseline value. In practice only the deviation from the baseline value is used
which is exactly why the baseline oset is of no concern.
Finally, there are offset errors in the electronic system which amplifies the signals
from the two cells.
All in all this means that the electrical voltage which is the output from the RI
detector can be written
URI = URI, Baseline + kRI · (nsolution − nbuffer ) (5.7)
or, applying equation 5.3:
dn
URI = URI, Baseline + kRI · ·C (5.8)
dC
The constant kRI depends, among other factors, on the amplification of the elec-
tronics in the RI detector but often has a value so that a 0.01 change in the refract-
ive index produces a change in output the voltage URI of 1 volt. The precise value
of kRI has to be known, of course (it can be found in the instrument manual), in
dn
order to use it to determine concentrations or dC
values.
the concentration of salts in the buer. Also air may dissolve in the buer during the day
which will also change its refractive index!
5.2. Apparatus 107
Inline measurement of light scattering in conjunction with

SEC
We still imagine a small amount of solution containing identical solute molecules

having been injected into the buffer stream. When the molecules have passed
through the size exclusion column they first pass through the flow cell of the light
scattering instrument producing a signal from one of the light scattering detectors
as shown shown in figure 5.5. When the instrument measures the intensity of
3.00
2.50
2.00
ULS (volt)
1.50
1.00
0.50
ULS, baseline
0.00
0 5 10 15 20 25 30 35
Elution time (minutes)
Figure 5.5: The signal from the light scattering instrument lies at a constant baseline
value as long as only the pure buer is running through the ow cell. When the dissolved
molecules pass the ow cell the signal rises and subsequently subsides back to the baseline
value when all the solute molecules have passed. Compare this signal with the RI signal
(concentration signal) of 5.4. Note, that the RI signal is delayed approximately three
minutes compared to the light scattering signal. This is simply the time it takes for the
molecules to run through the tubing that connects the light scattering instrument with
the RI detector. This time depends, of course, on the ow rate, the length of the tubing
and the inner diameter of the tubing. And three minutes is, by the way, unrealistically
long.
the scattered light the same problem arises as with the measurement of refractive
index: The signal is not zero when just buffer and no solute molecules run through
the measurement cell. There are a number of reasons for this: First of all the
electronics of the measurement system usually has an output voltage offset which
means that the output voltage is different from zero even when the laser is turned
off. Secondly, when the laser is turned on but only pure solvent runs through
the flow cell some stray light will be scattered into the detector system from glass
surfaces, microscopic scratches and dents and dirt particles on the flow cell and its
entrance windows. And finally, even a pure liquid will scatter light to some extent
due to local spontaneous statistical fluctuations in the liquids refractive index.
Adding up all these factors means that the output signal of the light scattering
instrument for a given scattering angle or corresponding q value can be written
ULS = ULS, Baseline + kLS · R(q) (5.9)
where R(q) is the Rayleigh ratio at the given q value. The value of the constant
kLS depends on the detector efficiency, the detector area and the gain in the elec-
tronics of the instrument. If the instrument has multiple detectors each of them
has its own value of the constant kLS . With the help of equation 5.1 we can write
equation 5.9 as
ULS = ULS, Baseline + kLS · KCMapp. (5.10)
4π 2 n20 (dn/dC)2
where K = NA λ40
is the previously mentioned optical contrast constant
defined in section 2.7. If, for simplicity, we write the contrast constant as K =
k1 · (dn/dC)2 we can rewrite equation 5.10 once more to read
ULS = ULS, Baseline + kLS · k1 · (dn/dC)2 · CMapp. (5.11)
Note the similarity between this signal (equation 5.11) and the concentration
signal 5.8. The RI signal is, however, delayed a certain time, in this case by three
minutes. The delay is the time it takes for the molecules in the light scattering
flow cell to move to the RI detector flow cell. This of course depends on the
flow rate and the volume of the tubing connecting the two instruments. So, when
comparing the light scattering signal and the concentration signal they should be
compared with this time delay in maind, i.e. compare ULS (t) with URI (t + ∆t)
where ∆t is the time delay between the two signals. In this example 3 minutes.
The precise value of has to be determined experimentally for a given flow rate by
5.2. Apparatus 109
comparison of positions of the peak values of the two signals. If we then combine
equations 5.8 and 5.11 we can calculate Mapp. as
(ULS − ULS, Baseline ) · kRI

Mapp. = dn (5.12)
(URI − URI, Baseline ) · k1 · kLS · dC
We may just as well combine the three constants kRI , k1 and kLS into one constant
(for each detector):
kRI
kdet. =
k1 · kLS
With this we can write the apparent molecular weight (at some q value)
kdet. ULS (t) − ULS, Baseline

Mapp. = dn · (5.13)
dC
URI (t + ∆t) − URI, Baseline
where we have indicated explicitly the time delay between the two signals.
Looking at equation 5.13 it might look as if the molecular weight is a function of
time. Elution time, that is. In principle this is true. But if there are no appreciable
concentration effects, i.e. the second virial coefficient is sufficiently small or the
concentration is sufficiently small then the light scattering is simply proportional
to the concentration. In this case the numerator and the denominator of equa-
tion 5.13 will be proportional at all times and hence they will give the same value
of Mapp. at all times. This, of course, is theory. In practice it is important to make
sure that URI is not too close to its baseline value because then one would divide
with a number close to zero in equation 5.13. The consequence of this would be
to produce a very ”unstable” value of Mapp. . So, in practice the molecular weight
is determined using values not too far from the peak values of the light scattering
and RI curves. If these precautions do not produce a constant molecular weight
across the central part of the peak it may be a sign that the second virial coeffi-
cient cannot be neglected. The molecular weight across the concentration peak
may look ”parabolic” with either upward or downward turning branches. Which
of the two is the case depends on the sign of the second virial coefficient (with
upward turning branches indicating a positive second virial coefficient and vice
versa).
How to determine the molecular weight, and possibly the

radius of gyration
We shall now assume that we can disregard concentration effects. Whether this is
justified can be checked, as previously mentioned, by checking if the molecular
weight calculated at the peak of the concentration curve is the same as the mo-
lecular weight calculated slightly off the peak (where the concentration is lower).
What remains to actually determine molecular weights is to establish the value
of the constant kdet. in equation 5.13, or in fact, all of the values of this constant
i.e. one for each detector of the instrument. The value is determined by injecting
into the apparatus a solution of a solute of known molecular weight and known
dn
value of dC
. It should be a solute consisting of molecules small enough that their
form factor has the constant value of one at all scattering angles of the instrument.
This ensures that the apparent molecular weight of the solute, Mapp. is actually
the true molecular weight at all scattering angles. Such a molecule could be BSA
(bovine serum albumin) which in its monomeric form has a molecular weight of
MBSA = 66400 g · mol−1 . When running a solution of BSA through the column
a number of peaks will show up both in light scattering and in the RI detector.
Usually four peaks are encountered the first, pertaining to the largest species, cor-
responds to aggregates of BSA and possible impurities. Then next come three
more peaks corresponding to oligomers (mainly tri- and tetramers), dimers and
monomers. The monomer peak is last and largest and is used to establish the
value of the detector constants:
!
dn URI, top − URI, Baseline
kdet. = MBSA · · (5.14)
dC BSA
ULS, top − ULS, Baseline
Note that the value of the detector constant(s) does not depend on the calibration
solute used as long as its molecular weight and dn/dC value are known. But
still the solute used for calibration should be a small enough molecule to scatter
isotropically. This is not a major limitation as anisotropic scattering is usually not
detectable for proteins with a molecular weight below 500, 000 g · mol−1 . Once
the detector constants have been established they can be used in equation 5.13
to determine the apparent molecular weight of the protein one is studying. The
5.3. Absolute calibration 111
apparent molecular weight is calculated for each detector (q value) and a Guinier
plot can then be used to determine the true molecular weight by extrapolation to
q = 0. If the slope of the Guinier plot is sufficiently high the radius of gyration
for the molecules can be determined as well. How small a radius of gyration can
be determined depends on the quality of the data and on the precision with which
the detector constants have been determined. In practice Rg values down to about
10 nm can be determined under optimal conditions.
5.3 Absolute calibration

The calibration procedure described above is often replaced by a so-called abso-
lute calibration. The purpose is exactly the same, i.e. to establish the value of
the detector constants. What is used is a liquid which in itself scatters light. The
liquid most commonly used is toluene which scatters light due to spontaneous
density fluctuations (creating fluctuations in the refractive index). The advantage
of this is that you don’t have to make up a fresh solution of protein every time
the instrument has to be calibrated. Pure toluene is always the same, it doesn’t
change composition due to evaporation and it doesn’t degrade with age or if kept
at the wrong temperature. All liquids scatter light for the same reason, but toluene
is particularly good at it. Pure toluene scatters light as efficiently as a BSA solu-
tion with a concentration of 0.65 mg · mL−1 . For comparison pure water scatters
approximately 20 times less. The details of the calibration procedure will be dif-
ferent from those outlined above but commercial instruments come with detailed
instructions as to how calibration is done.
A
Complex numbers
Complex numbers are a very handy tool in certain types of computations that have
to do with the sine and cosine functions. The reason is that with complex numbers
it turns out that the sine and cosine functions are closely related to the exponential
function. And it is often much more straightforward to do calculations with the
exponential function than with trigonometric functions. The rules of calculation
with the exponential function are few and simple whereas the rules of manipulat-
ing the trigonometric functions are intricate and seem nearly endless in number.
Complex numbers are an extension of the well-known real numbers. The were in-
vented with the purpose of enabling solution of ordinary equations of any degree.
It turns out that the only thing required to accomplish this is the introduction of a
new ”number” defined as
√
i= −1
113
114 A. Complex numbers
This number is called the imaginary unit. With this new and strange number one
can define a more general type of number called complex numbers. A complex
number z is defined as
z =x+i·y
where x and y are real numbers. If y = 0 z will be a real number so complex
numbers are clearly a generalisation of real numbers. If, on the other hand, x = 0
the number z is said to be a pure imaginary number. An example of the use of the
complex numbers is the solution the algebraic equation of the second degree
z2 + 1 = 0
which has no solutions among the reals numbers but has the solutions i and −i
among the complex numbers. All the usual rules of algebra are valid for these
”extended” numbers and no further extensions are necessary to solve any degree
of algebraic equation. Not even if the coefficients of the equation are themselves
complex numbers. And it is possible to calculate e.g. square roots of not only
negative numbers but also of complex numbers. The result would just be another
complex number.
A complex number z cannot be represented on a one-dimensional line like the
reals numbers. Complex numbers require two dimensions, a coordinate system
called the complex plane.
One defines (as for vectors) the absolute value |z| of a complex number z as
q
|z| = x2 + y 2
One common operation on a complex number is called complex conjugation.

What this does is the change the sign of imaginary part of the complex num-
ber. The complex conjugate of a complex number z = x + i · y is denoted z ∗ and
is defined as
z∗ = x − i · y
The complex conjugate of z has the useful property (remember i2 = −1):
zz ∗ = |z|2
115
Proof: zz ∗ = (x + i · y) · (x − i · y) = x2 + i · xy − i · yx − i · y · i · y = x2 − i2 · y 2 =
x2 + y 2 = |z|2
Now comes the really useful part: The exponential function with a purely imagin-
ary argument i · y is defined as:
ei·y = cos(y) + i · sin(y) (A.1)
which shows that there is a connection between the exponential and the trigo-
nometric functions. The most important property of the ordinary exponential is
that it is its own derivative. How about this new definition? To check this we
differentiate equation A.1 with respect to y:
d i·y d
e = (cos(y)+i·sin(y)) = − sin(y)+i·cos(y) = i(cos(y)+i·sin(y)) = i·ei·y
dy dy
(A.2)
But note that the argument of the exponential is not y but i · y, so in fact we should
have differentiated with respect to iy, so with the above calculation we get
d i·y 1 d i·y 1
e = e = i · ei·y = ei·y (A.3)
d(iy) i dy i
i.e. the new exponential is indeed its own derivative.
More generally the exponential of a complex number z = x + iy is simply defined
by the usual rules for the exponential:
ex+iy = ex · eiy = ex · (cos(y) + i sin(y))
Note that the complex conjugate of the complex exponential is
(ex+iy )∗ =
(ex · (cos(y) + i sin(y)))∗ =
ex · (cos(y) − i sin(y)) =
ex · (cos(−y) + i sin(−y)) =
ex−iy
i.e. simply change the sign of the imaginary part in the exponent.
The usefulness of the complex exponential in physics lies in its use to describe
116 A. Complex numbers
wave phenomena. An electromagnetic wave propagating along the x−axis can be

described with a sine or a cosine function as e.g.
E(x, t) = E0 cos(ωt − kx)
With the complex exponential this can be written in an alternative form:
E(x, t) = E0 ei (ωt−kx)
In fact one gets too much here. Also an imaginary term (E0 i sin(ωt − kx)) is in-
cluded. The ”trick” is simply to disregard the imaginary part of the the expression
(it is ”unphysical”). It is seen that using the definition of the complex conjugate
(or the above expression for the complex conjugate of the exponential) that taking
the complex conjugate of E(x, t) gives
E ∗ (x, t) = E0 e−i (ωt−kx)
whereby
E(x, t)E ∗ (x, t) = E02 ei (ωt−kx) e−i (ωt−kx)

= E02 ei (ωt−kx)−i (ωt−kx)
= E02
which means that E(x, t)E ∗ (x, t) or equivalently |E(x, t)|2 is simply the squared
amplitude of the electromagnetic wave. If E(x, t) is a sum of many different
cosine terms (as the total electric field originating from the scattering of several
particles) it is a simple matter to compute the amplitude of this electromagnetic
wave. This is the reason for the use of the complex exponential in physics.
B
Data fitting
A frequently encountered problem in experimental physics is that of describing

some measured data as well as possible with a function. Assume one has done
a number of measurements of some physical quantity y as a function of some
other physical quantity x. It could be measurement of the pressure of an en-
closed gas as a function of the volume of the gas. The ensuing set of data is
y1 , y2 , . . . , yN measured as a function of the quantity x at the corresponding
values x1 , x2 , . . . , xN . The task is now somehow to determine a function f ,
i.e. y(t) = f (x) which best fits the measured data. The question is what ”best”
means. What is meant is usually that the sum of squared differences between the
measured values and the calculated ones is as small as possible. This quantity is
117
118 B. Data tting
called the chi-square of the fit and is assigned the symbol χ2 . It is defined as
N
2
X (yj − f (xj ))2
χ =
j=1 σj2
The denominators are the statistical variance (a statistically precise measure of

the uncertainty, squared actually) of the individual data points. The value of these
may be established e.g. by repeating the measurements or by theoretical consider-
ations. This is a fair definition of the goodness of the fit because data points with a
large uncertainty contribute less to the sum and hence have correspondingly little
influence on how to choose the function f (x). Now, the function normally has to
belong to a specific class of functions, e.g. fourth degree polynomials or sums of
six sine functions of different frequency. What the fitting procedure comes down
to is then to determine the optimum set of parameters (e.g. the five coefficients
in a fourth degree polynomial) that will make χ2 as small as possible. Of course
it is always possible to make a better fit if more parameters are included in the
fitting function, e.g. using a fifth degree polynomial instead of a fourth degree
polynomial. Note also that the magnitude of χ2 will grow if more data points are
included since this will increase the number of terms. Therefore, in order to judge
how good a fit actually is it is not sufficient to calculate χ2 it is also necessary
somehow to correct for the number of terms in the sum and for the number of
parameters used. To accomplish this task one defines the reduced chi-square χ2r
N
1 (yj − f (xj ))2
χ2 =
X
·
N − n − 1 j=1 σj2
where N is the number of data points (i.e. the number of terms in the sum) and
n is the number of parameters used to describe the function f (x). The sign of a
good fit that χ2r ≈ 1. If the reduced chi-square is much larger than one the fit is
poor, i.e. the parameters are not optimally chosen or the choice of function for the
fit is wrong so that even the minimum value of χ2r is substantially larger than one.
If, on the other hand the reduced chi-square is much smaller than one the reason
is usually that the variances σj2 have been overestimated.
As an example, say we have measured an autocorrelation function at 256 lag
119
times: g2 (tj ) where j = 0, . . . , 256 and we want to fit it with a model func-
2 2
tion of the form y(t) = (A1 e−D1 q t + A2 e−D2 q t )2 + 1. The parameters to adjust
are A1 , A2 , D1 , D2 , i.e. we have to minimise
256 2 2
2
X (g2 (tj ) − [(A1 e−D1 q t + A2 e−D2 q t )2 + 1])2
χ =
j=1 σj2
In practice this is done with a computer program which adjusts the different para-
meters until a minimum for χ2 seems to be found. In order to assess the goodness
of the fit the minimum value just found is divided by the number 256 − 4 − 1.
A number of different programs to do this search for optimum parameters ex-
ists but none of them can guarantee that an absolute, global minimum is actually
found. Nor do the programs normally guarantee that even a local minimum is
found, sometimes the search simply diverges. The way most programs of this
type work is by calculating the gradient of χ2 which is a vector in parameter
space pointing in direction where χ2 increases the most. The program (or the user
of the program) makes and initial guess about the value of the parameters, then
calculates the gradient at this parameter point. Next, the program finds a new and
hopefully better parameter point by moving a short distance in the exact opposite
direction of the gradient. Then the gradient is calculated in this new parameter
point and the procedure is repeated over and over again until the gradient is close
to zero (or actually ~0) indicating that a minimum has been found. Or a maximum
or a saddle point. Further analysis may be required to ensure that the parameter
point found is actually a minimum. This type of analysis is called a gradient
search and has the advantage of being fast when it works. But it frequently fails if
the initial parameter guess is far from a minimum point.
Sometimes more ”safe” types of fitting schemes can be employed. The fit de-
scribed above is so-called non-linear fit. An example of the opposite, a linear
fit would be if one would fit the field autocorrelation function g1 (τ ) with a model
2 2
function of the form f (t) = A1 e−D1 q t +A2 e−D2 q t where the diffusion coefficient
D1 and D2 were fixed so that only the amplitude parameters A1 and A2 had to be
optimised. In this case there is a direct method to calculate the optimum paramet-
ers with no risk of failure. Of course the problem then is what values of D1 and D2
120 B. Data tting
to choose in advance. One way to circumvent this problem is to use not two but
perhaps 100 terms with predefined diffusion coefficient D1 , D2 , D3 , . . . separ-
ated by a small factor, like e.g. 1.3, i.e. Di+1 = 1.3 · Di for i = 1, 2, 3, . . .. This,
however gives rise to other problems which are discussed on pages 94 and 95.
Index
dn/dC, 37, 41, 108 in RI detection, 106, 109

q−vector, see scattering vector in SLS, 107, 108
bovine serum albumin, see BSA
A4F, 101
Brillouin scattering, see scattering, Bril-
absolute calibration, 111
louin
absorption, 4
Brookhaven Instruments, 76
acetone, 104
Brownian motion, 4, 73, 78–80, 83
aggregate, 59, 110
BSA, 10, 110, 111
alcohol, 104
ALV, 76 chi-square, 87, 118
approximation reduced, 118
Debye, 52, 57, 64 reduced, 118
Guinier, 51–54, 64 coherence factor, 82, 85, 87
artifacts, 86, 94 coherent radiation, 69
autocorrelation function, 72, 74–97 complex conjugate, 33, 114–116
field, 74, 79, 82 complex numbers, 24
intensity, 74, 82 concentration effects
normalised, 82, 85 in static light scattering, 44
autocorrelator, 72, 76 CONTIN, 95
multiple tau, 75 contrast constant, 41
avalanche diode, see detector, avalanche correlation time, 75, 80
diode correlator, see autocorrelator
averaging, 56–60 Cumulant analysis, 90–92, 96
baseline data analysis, 50, 85

in DLS, 76, 77, 93 data anlysis
121
122 Index
dynamic light scattering, 85–97 electromagnetic wave, 3, 6, 14, 18, 21,

Debye, 9 24, 25, 27, 31, 32, 116
Debye approximation, see approxima- ellipsoid, 49, 65, 68
tion, Debye
detector FFF, 101
avalanche diode, 72 field flow fractionation, see FFF
photodiode, 6 fitting, 86, 88, 89, 91, 94–96, 117–119
photomultiplier, 6 linear, 119
differential refractive index increment, non-linear, 88, 97, 119

see dn/dC flow cell, 60, 100–102, 105–108
differential refractometer, 60, 61, 102 fluctuations, 7, 8, 73, 74, 78–83
diffusion in refractive index, 108
rotational, 79 measuring, 73
translational, 79, 83–85 in refractive index, 111
diffusion coefficient, 7, 8, 70–74, 76, intensity, 71
77, 79, 82, 84, 86–93, 97, 119, of laser intensity, 90
120 form factor, 46–50, 57, 64
for sphere, 70 in relation to DLS, 85
thermodynamic determination, 71 frictional coefficient, 71
dilute solutions, 66, 85 for spheres, 71
dipole, 14, 16
gain, 108
moment, 21, 23, 24
Gaussian distribution, 88, 90–92
dirt, 108
gel column, 60, 61, 100, 101, 105, 107,
discriminator, see PAD
110
DLS, see light scattering, dynamic
goodness of fit, 88, 119
DLS-setup, 79
gradient, 119
Doppler effect, 4
of chi-square, 119
dynamic light scattering, see light scat-
gradient search, 119
tering, dynamic, 85
Guinier
electric field, 14, 17–19, 21, 22, 27, 30, plot, 50
33, 74, 82, 116 Guinier approximation, see approxima-
Index 123
tion, Guinier data analysis, 85

data anlysis, 97
HPLC, 60
static, 6, 9–41
hydrodynamic radius, see radius, hydro-
in practice, 99
dynamic
macromolecules, 7, 9
ill conditioned problem, 93
maximum entropy, 96
ill posed problem, 93
MEM, see maximum entropy
ILT, 94
miracle, 4, 84
imaginary
molecular weight
number, 114
apparent, 49, 50, 64
unit, 114
determination of, 43, 50, 56
indirect methods, 86
monodisperse solution, see solution, mon-
intensity, 5
odisperse
relation E-field, 15
monomodal distribution, 91
interference, 24, 28, 33
multimodal distribution, 91
constructive, 80, 81, 83
destructive, 80, 83, 84 non-ideal solution, 16, 45
non-interacting particles, 16, 56, 70
lag time, 75
Laplace transformatio
osmotic pressure, 45
inverse, 92
osmotisc pressure, 45
Laplace transformation
Ostrowsky, 94
definition, 93
inverse, 86, 93, 94 PAD, 72
laser, 6, 10–13, 17, 35, 37, 47 parameter
focussing, 71–73, 79–81, 83 regularisation, 96
HeNe, 47, 68 parameters, 97
light adjustable, 86, 88, 89
interactionswith matter, 3 irreproducible, 89
what is it, 3 number of, 118
light scattering optimum number, 91
dynamic, 4, 7, 8, 69–85 search strategy, 119
124 Index
peak Rayleigh ratio, 12–108

delay, 109 and osmotic pressure, 45
in size distribution, 95, 96 refractive index, 25, 26, 32, 37, 41
Pecora, 84 regression
phase, 27 linear, 53, 54
definition, 14 regularisation, 95, 96
phase difference, 28–30, 32, 34, 36, 73 REPES, 96
photodiode, see detector, photodiode residual plot, 88, 89
photomultiplier, see detector, photomul- RNONLIN, 96
tiplier
sample preparation, 99
photon, 3, 4, 6, 18
scattering
scattered, 4, 14, 25, 31
anisotropic, 60, 79, 110
polarisability, 17, 18, 20, 22, 23, 25
Brillouin, 4
excess, 18
elastic, 4
polarised light, 14
inelastic, 4
polydisperse solution, see solution, poly-
isotropic, 23
disperse
quasi elastic, 4, 69
polysaccharide, 37, 102, 104
Raman, 4
polysaccharides, 104
scattering angle, 6–8, 12, 25
protein, 10, 59, 102, 110, 111
scattering function, 47
proteins, 104
scattering power, 30
Provencher, 95
scattering vector, 25, 26, 31, 41
radius length, 25, 27, 31, 32, 41, 46, 62
of gyration, 51 scattering volume, 12, 13
hydrodynamic, 70 scattering, isotropic, 46
of gyration, 48, 49, 51, 53–56 scatttering power, 34
radius of gyration, see radius, of gyra- SEC, 60, 101
tion SEC-MALS, 60
z-average, 92 second virial coefficient, see virial coef-
Raman scattering, see scattering, Raman ficient
Rayleigh raio, 13 Siegert relationship, 82, 85, 86
Index 125
size determination, 43, 50

size exclusion chromatography, see SEC
size exclusion column, see gel column
SLS, see light scattering, static, 9
solution
monodisperse, 87
polydisperse, 87
speckle, 73, 80–82
speckles, 81
sphere, 65
static light scattering, see light scatter-
ing, static
Stokes radius, see radius, hydrodynamic
Stokes-Einstein relationship, 70, 71
stretched exponential, 97
structure factor, 44
toluene, 111
virial coefficient, 45
viscosity, 70, 73, 101
wavenumber, 15
wavevector, 15, 25–27, 31
weight concentration, 36, 44
z-average, 58

Lightscattering Theory and Practice PDF

Uploaded by

Copyright:

Available Formats

Lightscattering Theory and Practice PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lightscattering Theory and Practice PDF

Uploaded by

Copyright:

Available Formats

Light Scattering

University of Copenhagen 12th September 2017

2 Simple theory for static light scattering, SLS 9

3 Determination of size and molecular weight 43

4 Simple theory for dynamic light scattering, DLS 69

5 Light scattering in practice 99

A Complex numbers 113

B Data fitting 117

This is an introduction to light scattering conceived as a measurement technique.

1.1 What is light scattering?

1. Absorption (The photons disappear)

2. Scattering (The photons change their direction)

while in the context of scattering it is frequently written as:

of the particles under investigation.

completely transparent components, etc. . . . . But as previously mentioned light

On the following pages we will describe an discuss two measurement techniques,

< I > Static light scattering

SLS : Static light scattering employs measurement of the intens-

DLS : Dynamic light scattering employs measurement of long

• The light has to be monochromatic, i.e. to have a well-defined wavelength.

2.1 The Rayleigh ratio, R

When a light scattering experiment is done on a solution (i.e. a suspension of

• the molecular weight of the suspended particles (molecules)

• the concentration of the suspended particles (molecules)

• the refractive index of the pure solvent

• the refractive index of the suspended particles (molecules)

• the size of the suspended particles (molecules)

• interaction forces between the suspended particles

2.2 What is light?

E(x, t) = Ez (x, t) = E0 cos(2πν(t − x/c) + φ) (2.3)

As ν/c = 1/λ we can finally rewrite equation 2.3 as:

E(x, t) = Ey (x, t) = E0 cos(ωt − kx + φ) (2.4)

This equation is easily generalised to propagation in any direction and a point of

E(~r, t) = E0 cos(ωt − ~k · ~r + φ) (2.5)

I0 (x) = 0 chE 2 (x, t)i (2.6)

where 0 = 8.85 · 10−12 F · m−1 is the so-called vacuum permittivity and c =

2.3 Why and how do particles scatter light?

where α is the so-called polarisability of the particle. The vacuum permittivity,

µnet = (αplug − αmedium ) · 0 · E

2.5 Scattering of light from one small particle

other and perpendicularly to the direction of propagation. The oscillations have

or, applying also equation 2.8:

As n = M/M1 where M is the molecular weight of the compound particle and

2.6 Scattering from one small, composite ob-

Without the use of complex numbers

~q = ~kout − ~kin (2.13)

Itotal = 0 cE02 · (1 + cos(∆φ)) = 2 · Is, 1 · (1 + cos(∆φ)) (2.20)

Is, n ≈ Is, 1 · n2 (2.24)

With the use of complex numbers

~q = ~kout − ~kin (2.26)

The situation is illustrated in figure 2.10.

Itotal = 0 cE02 · (1 + cos φ) = 2 · Is, 1 · (1 + cos(∆φ)) (2.32)

2.7 Scattering from many identical particles

equation 2.38 can be written as:

hIs, total i = Is, n · N (2.39)

As the number, N , of particles in the volume V is given by the molar concentration

we can rewrite equation 2.39 with the aid of equation 2.25:

hIs, total i = Is, n · NA V c

As the scattering intensity of one molecular subunit, Is, 1 , is proportional to the

I0 (x) = 0 chE 2 (x, t)i (2.6)

where 0 = 8.85 · 10−12 F · m−1 is the so-called vacuum permittivity and c =

µnet = (αplug − αmedium ) · 0 · E

Itotal = 0 cE02 · (1 + cos(∆φ)) = 2 · Is, 1 · (1 + cos(∆φ)) (2.20)

Itotal = 0 cE02 · (1 + cos φ) = 2 · Is, 1 · (1 + cos(∆φ)) (2.32)