This article was downloaded by: [University of Guelph]

On: 01 May 2012, At: 12:49

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number:
1072954 Registered office: Mortimer House, 37-41 Mortimer Street,
London W1T 3JH, UK

Advances in Physics
Publication details, including instructions for
authors and subscription information:
http://www.tandfonline.com/loi/tadp20

The mean free path of

electrons in metals
E. H. Sondheimer
Available online: 08 Nov 2010

To cite this article: E. H. Sondheimer (2001): The mean free path of electrons in
metals, Advances in Physics, 50:6, 499-537

To link to this article: http://dx.doi.org/10.1080/00018730110102187

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/

terms-and-conditions

This article may be used for research, teaching, and private study
purposes. Any substantial or systematic reproduction, redistribution,
reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any
representation that the contents will be complete or accurate or up to
date. The accuracy of any instructions, formulae, and drug doses should
be independently verified with primary sources. The publisher shall not
be liable for any loss, actions, claims, proceedings, demand, or costs or
damages whatsoever or howsoever caused arising directly or indirectly in
connection with or arising out of the use of this material.
 Advances in Physics, 2001, Vol. 50, No. 6, 499 ±537

The mean free path of electrons in metals*

E. H. Sondheimer{
Royal Society Mond Laboratory, Cambridge

Contents page
1. Introduction 499
2. Conduction in thin ®lms and wires 505
3. Magnetic e€ ects in thin conductors 515
4. The anomalous skin e€ ect 524
Downloaded by [University of Guelph] at 12:49 01 May 2012

Acknowledgement 536
References 536

1. Introduction
1.1. The foundations of the modern electron theory of metals were laid at the
beginning of the present century, when the existence of a gas of free electrons was
postulated by Drude in order to explain the conducting properties of metals; the
behaviour of the electrons was subsequently analysed by Lorentz by means of the
statistical methods of the dynamical theory of gases. The chief success of the Drude±
Lorentz theory was the prediction of the Wiedemann±Frenz law connecting the
electrical and thermal conductivities, but later developments revealed an increasing
number of serious di culties, outstanding among them the inability of the theory to
explain why the conduction electrons do not contribute appreciably to the speci®c
heat of a metal. This paradox was not resolved until the advent of quantum
mechanics, when Pauli and Sommerfeld applied the Fermi±Dirac statistics to the
free electrons in a metal, showing that in this way most of the contradictions could
be reconciled.
The Drude±Lorentz±Sommerfeld theories are essentially formal in character.
They involve as arbitrary parameters the number n of free electrons per unit volume,
which is assumed to be of the same order as the number of atoms per unit volume,
and the mean free path l of the electrons which is to be determined from a
comparison of theory and experiment. So far as the electrical conductivity ¼0 is
concerned, the results of the Sommerfeld theory are summarized by the formulae
3
8p m·
v
nˆ ; …1a†
3 h
n°2 l
¼0 ˆ ; …1b†
m·v

* This article was the ®rst to appear in Advances in Physics, originally published in volume 1,
1952. It had attracted 1030 citations by October 2001, and is ranked 5 in the index of articles attracting
more than 100 citations.
{ Fellow of Trinity College, Cambridge.
Advances in Physics ISSN 0001±8732 print/ISSN 1460±6976 online # 2001 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080 /0001873011010218 7
 500 E. H. Sondheimer

in which h is Planck’s constant, ¡° is the charge and m is the mass of an electron, and
v· is the velocity of an electron at the surface of the Fermi distribution. ¼0 therefore
depends on n, l and fundamental constants only. In order to obtain the correct order
of magnitude for the conductivity and the correct temperature variation, it is
necessary to assume that l is of the order of several hundred interatomic distances
at ordinary temperatures, and increases rapidly in a pure metal towards very low
temperatures.
Such long free paths which vary with temperature are very di cult to explain on
classical theory, but they can be understood on the basis of the recent developments
in the theory of metals. These were initiated by Bloch and elaborated by a great
many authors; they have been concerned with a detailed quantum-mechanica l
analysis of the motion of electrons in a crystal latice, and they have made it possible
to give precise meanings to the two fundamental concepts of the `number of free
electrons’ and the `mean free path’ and to obtain numerical estimates of these
Downloaded by [University of Guelph] at 12:49 01 May 2012

quantities in certain cases. The electrons in a metal are regarded as distributed over a
number of energy bands, ®lling most of them completely. All the electrons are free to
move through the lattice, but only those which are contained in incompletely ®lled
energy bands can contribute to the resultant current and are to be regarded as free
electrons for the purposes of conduction theory. The number of free electrons in a
metal is of the same order as the number of atoms: the precise number, however,
depends on the detailed con®guration of the energy bands and need not be a simple
multiple or submultiple of the number of atoms. The variation of n with temperature
is negligible, since at ordinary temperatures the free electrons form a highly
degenerate Fermi±Dirac gas.
An electron can move freely through a perfect and rigid crystal lattice and there is
no resistance. In a pure metal a ®nite free path is caused by the thermal vibrations of
the lattice and is of the same order as the average wavelength of the sound waves in
the metal, which is large compared with the interatomic distance and is increased by
lowering the temperature. The free path does not increase inde®nitely, however, as
the temperature is lowered, and at very low temperatres it tends to a constant
`residual’ value lr which is determined by static lattice imperfections such as the
presence of impurity atoms, and which is of the order of the distance between the
impurities.
The following theoretical formula may be deduced for the electrical conductivity
of a metal, subject to many simplifying assumptions concerning the interaction
between the electrons and the lattice vibrations (see, for example, Wilson 1936,
Chapter VI):
±m²1=2 5 … ³=T
1 m·v 9ph2 C 2 T z5 dz
ˆ 2 ‡ : …2†
¼0 n° l r 2 2
8n D° MkY± 3=2 Y 0 …ez ¡ 1†…1 ¡ e¡z †

Here k is Boltzmann’s constant, ± is the Fermi energy level (± ˆ 12 m· v2 †, Y is the

Debye temperature, M is the mass of an atom, D is the volume of the unit cell, and C
is a constant which determines the interaction between the electrons and the lattice.
According to this formula the `ideal’ and `residual’ resistances are additive, and the
ideal resistance is proportional to T at high and to T 5 at very low temperatures. (For
commercially pure metals the residual resistance can be neglected at ordinary
temperatures.) The temperature variation of the resistance predicted by (2) is,
generally speaking, in good agreement with observation, particularly for the
 Mean free path of electrons in metals 501

monovalent metals. The absolute value of the free path at any temperature can be
estimated by combining (1) and (2) and substituting reasonable values for the
parameters. The chief uncertainty concerns the magnitude of the interaction
constant C, estimates of which can be obtained by numerical integration if the wave
functions of the conduction electrons are known; it is of the same order as the Fermi
energy ±. The estimates of the free path are of the right order of magnitude, but
precise numerical values cannot be obtained in this way.

1.2. An important conclusion to be drawn from the detailed quantum-mechan -

ical theory is that the simple Sommerfeld treatment of the conduction phenomena
remains correct within certain limits. In the Sommerfeld theory the electrons are
regarded as perfectly free, their energy being proportional to the square of the
velocity; this remains approximatel y true for electrons moving in a lattice, but the
mass m which appears in equations (1) and (2) must be regarded as an e€ ective mass
Downloaded by [University of Guelph] at 12:49 01 May 2012

which is of the same order of magnitude as, but not necessarily equal to, the mass of
a free electron. This model of `quasi-free’ electrons, which is implicit in the derivation
of (2), applies most closely to the monovalent metals in which the conduction
electrons are all contained in a single energy band; for these metals, moreover, the
number of conduction electrons should be precisely one per atom. In multivalent
metals, in which the electrons occupy more than one band, the model may still be
used to give a semi-quantitativ e description of the simpler conduction phenomena,
but quantities such as n and m must then be regarded as representing certain
averages of the numbers of electrons and the e€ ective masses of the electrons in the
various bands, and the precise numerical values have no immediate physical
signi®cance. It would, of course, be possible to consider more complicated models,
for example one in which the conduction electrons are contained in two overlapping
energy bands. In the present state of the theory of metals, however, it is impossible to
work out a theory which fully takes into account the electronic structure peculiar to
any particular metal, and instead of introducing a large number of parameters of
doubtful physical signi®cance it is best to work with the simplest model which gives
reasonable results. For more complicated conduction phenomena, however, par-
ticularly those which are associated with the presence of a magnetic ®eld or with
anisotropy e€ ects, the free-electron model is entirely inadequate (it leads, for
example, to a zero magneto-resistanc e e€ ect), and even a qualitative theory can
only be obtained by using a model in which the energy surfaces do not form a singly-
connected set of spheres.
In the Sommerfeld theory the free path is most conveniently introduced through
the time of relaxation ½ , which is de®ned as follows. Let v ˆ …vx ; vy ; vz † be the
velocity of an electron, and let 2…m=h†3 f …v; r† dr dv be the number of electrons in
the volume element dr ˆ dx dy dz which have their velocities in the range
dv ˆ dvx dvy dvz ; f is the distribution function as usually de®ned in Fermi±Dirac
statistics. Suppose that some non-equilibrium distribution function is set up by a
system of external forces which are suddenly removed; the rate of approach
to equilibrium under the in¯uence of collisions alone is then supposed to be
given by
µ ¶
@f f ¡ f0
ˆ¡ ; …3†
@t coll ½
 502 E. H. Sondheimer

where f0 is the equilibrium distribution function. The time of relaxation need not be a
constant and may depend, for example, on the velocity; if v is the mean velocity of
those electrons to which ½ refers, the corresponding free path l is de®ned by l ˆ v½ .
In general ‰@f =@tŠcoll takes the form of an integral operator which does not reduce to
the simple form (3); it is then impossible to de®ne a free path in any natural way. The
detailed theory of the conduction mechanism in metals shows, however, that a free
path does exist in the above sense under certain conditions for quasi-free conduction
electrons; thus it can always be de®ned for scattering by randomly distributed
impurity atoms, and it also exists for scattering by lattice vibrations if the
temperature is above the Debye temperature. So far as the electrical conductivity
is concerned, it is a reasonble approximation to assume that a free path can be
de®ned for all temperatures, so that equation (1) holds and the temperature variation
of the free path is the same as that of the electrical conductivity. An approximate
theoretical expression for l in terms of the atomic constants of the metal can be
Downloaded by [University of Guelph] at 12:49 01 May 2012

obtained by combining equations (1) and (2). It must be borne in mind, however,
that the free path associated in this way with the electrical conductivity is not
necessarily the same as the free path associated, for example, with the thermal
conductivity.

1.3. With all the provisos mentioned, then, the simple Sommerfeld picture of the
free electrons in a metal and their mean free path retains its validity and may be used
to discuss the conduction phenomena. It is therefore an important problem to
determine the various parameters of the theory by as many independent methods as
possible; such estimates are of interest in themselves in providing information about
the electronic structure of metals, and they serve as a valuable check on the
consistency of the free-electron theory. Furthermore, since the calculation of the
free path from fundamental principles is highly complicated and involves many
drastic approximations , it is desirable to have methods by which l may be estimated
directly from observational data.
The simplest procedure (Mott and Jones 1936) is to compare equation (1) with
the observed electrical conductivity, but this gives only n2=3 l and l cannot be
obtained unless n can be estimated independently.{ It is, however, possible to obtain
l directly by measuring the conductivity under conditions where the free path may be
compared with some other characteristic length in the metal. E€ ects of this type have
attracted much attention in the last few years, and it is with them that we shall
henceforth be concerned.
The most obvious method is to use a thin ®lm or wire and to arrange that the free
path is comparable in magnitude with the thickness or diameter of the specimen; the
arti®cial limitation of the free path by the boundaries of the specimen causes an
increase in the resistivity above its value in the bulk metal, and this may be used to
deduce the ratio of free path to thickness or diameter. This topic is reviewed in
section 2.
In section 3 we consider the more complicated e€ ects which occur when a thin
specimen is placed in a magnetic ®eld. The ordinary bulk magneto-resistanc e e€ ect
depends in a complicated way on the binding of the electrons in the lattice and is zero

{ It should be remarked that the value of l de®ned by Mott and Jones and tabulated on p. 268 of
their book is twice the free path de®ned here; the latter is the physically relevant quantity.
 Mean free path of electrons in metals 503

for quasi-free electrons; in thin specimens, however, where boundary scattering of

electrons is important, the alteration of the electron trajectories in a magnetic ®eld in
general leads to a non-zero change of resistance even if the electrons are regarded as
free. These curious `geometrical’ e€ ects, being classical in nature, are entirely
di€ erent from the bulk e€ ect and are essentially simpler to understand. The details
of the phenomena vary with the shape of the specimen and with the relative
con®gurations of specimen, current and magnetic ®eld; their analysis in all cases
involves a new quantity with the dimensions of length, namely the radius
r0 ˆ m· vc=°H of a free-electron orbit in a magnetic ®eld H. The experiments therefore
give, in addition to the free path, a direct estimate of the momentum m· v of the
electrons at the surface of the Fermi distribution, and hence of the number of free
electrons according to equation (1a). The method is, however, severely restricted in
practice by the disturbing e€ ect of the bulk magneto-resistanc e phenomenon.
In section 4, ®nally, we consider the so-called anomalous skin e€ ect in metals.
Downloaded by [University of Guelph] at 12:49 01 May 2012

This is a more sophisticated side e€ ect in which the free path is compared, not with
the physical dimensions of the specimen, but with the distance to which a high-
frequency electric ®eld penetrates into the metal. Experiments on the high-frequency
skin resistance of metals allow values of l=¯ to be deduced, where ¯ is the classical
skin penetration depth, and since ¯ depends only on the frequency and the d.c.
electrical conductivity, the free path can again be obtained directly from experi-
mental magnitudes.

1.4. These phenomena present interesting problems from both experimental and
theoretical points of view. Let us consider the orders of magnitude of the various
characteristic lengths. The free path in metals at room temperatures is of the order of
10¡5 cm or less, but in a pure metal at liquid-helium temperatures the high values of
the conductivity indicate that it may be as large as 10¡2 cm. It is clear, therefore, that
extremely thin ®lms would be required in experiments on the size e€ ect at normal
temperatures; such ®lms are di cult to prepare and usually show subsidiary resistive
e€ ects, often time dependent, which tend to obscure the pure geometrical limitation
of the free path with which we are alone concerned. In order to obtain results which
are free from ambiguity, it is therefore essential to carry out the experiments at very
low temperatures where relatively large specimens (of thickness 10¡3 cm) may be
used. The necessity for using specimens of this order of thickness becomes even
greater when we consider the e€ ects which take place in a magnetic ®eld. These
e€ ects show up when the orbit radius r0 is comparable with the thickness of the
specimen. Since v· is of the order of 10 8 cm/sec, a magnetic ®eld of reasonable
magnitude (several kilogauss) corresponds to r0 10¡3 cm; and since r0 is inversely
proportional to H, impracticably large magnetic ®elds would be required in the case
of ®lms much thinner than this. Finally, in the anomalous skin e€ ect l must be large
compared with ¯; the ratio l=¯ is proportional to l3=2 and to the square root of the
frequency, so that both very low temperatures and high frequencies are required. For
a pure metal at liquid-helium temperatures and for microwave frequencies, l=¯ is of
the order of 100.
One may say, therefore, that the size e€ ects are essentially low-temperature
phenomena. Since the free path at low temperatures varies from specimen to
specimen and is not a characteristic property of the metal, it is usual to measure
(or estimate in some way) the value of the bulk conductivity ¼0 which corresponds to
the free path l, and to express the results of the experiments in terms of the ratio ¼0 =l.
 504 E. H. Sondheimer

This ratio determines directly the number of conduction electrons per unit volume,
since, according to equations (1a) and (1b),
1=3 2 2=3
¼0 n°2 8º ° n
ˆ ˆ ˆ 7:1 107 n2=3 : …4†{
l m·
v 3 h

Note that this relation involves only the assumption that the electrons are quasi-free,
but not that their e€ ective mass is necessarily equal to the mass of a free electron.

1.5. Many approximate theoretical treatments have been given of the various
free-path phenomena, but we shall not refer to these (except where they are
particularly relevant to the discussion), as they have been superseded by the strict
statistical analysis based on the Boltzmann equation for the distribution function of
the conduction electrons.{ This equation is formed by equating the rate of change in
Downloaded by [University of Guelph] at 12:49 01 May 2012

f due to external ®elds to the rate of change due to the collision mechanism, which is
assumed to be given by equation (3). In the presence of an electric ®eld E and a
magnetic ®eld H, the Boltzmann equation for quasi-free electrons takes the form
° 1 f ¡ f0
¡ E‡ v H grad v f ‡ v grad r f ˆ ¡ ; …5†
m c ½

which is purely classical except that the mass m is to be regarded as an e€ ective mass,
while the equilibrium distribution function is the Fermi±Dirac function
1
f0 …E† ˆ : …6†}
e…E¡±†=kT ‡ 1
The time of relaxation ½ is supposed to depend on the absolute value of v only.
f is a function of the velocity vector v and the space vector r, and the analysis of
the size e€ ects depends essentially on the appearance in equation (5) of the term
involving the space derivatives of f , which is required in the bulk metal only if
temperature gradients are present. This term takes into account the non-uniform
distribution in space of the conduction electrons which is characteristic of the size-
e€ ect phenomena. A general feature of the analysis is that boundary conditions
have to be imposed on the solutions of the Boltzmann equation; these are
determined in each case by the nature of the scattering at the physical boundaries
of the specimen.
In treating the various problems in this way we may apply the results directly to
the alkali metals, in which the electrons are most nearly free, and with somewhat
lesser con®dence to the noble metals; for more complex metals the model will
provide only a qualitative picture, but should give the orders of magnitude correctly.
In the presence of a magnetic ®eld, corrections of uncertain amount must be applied
to the classical formulae to allow both for the bulk magneto-resistanc e e€ ect and,
possibly, for the e€ ect of the quantization of the electron orbits in a magnetic ®eld.
These re®nements are di cult to take into account in any satisfactory way; the most

{ Gaussian units are used throughout the present article.

{The use of the Boltzmann equation in this connection appears to have been ®rst suggested by
Peierls.
}E here denotes the energy of an electron, but elsewhere in this article, E stands for electric ®eld
strength.
 Mean free path of electrons in metals 505

reasonable procedure, therefore, is to neglect them entirely and to restrict the

experiments to metals in which they are of least importance (see section 3.1).
The main problem, then, is to solve equation (5) for the various cases of interest,
and to use the solution to calculate the current density J by means of the usual
formula
± m² 3 …
J ˆ ¡2° vf dv: …7†
h
1.6. Objections have been raised from time to time against the whole basis of the
theory of metallic conductivity: thus the uncertainty principle has been used to argue
that the uncertainty in the energy of a single electron, which is determined by the
magnitude of the time of relaxation, may be so large in some cases as to render
meaningless any description of the situation by means of a distribution function
which obeys the classical Boltzmann equation (for a detailed discussion of this
Downloaded by [University of Guelph] at 12:49 01 May 2012

somewhat oversimpli®ed statement compare, for example, Peierls 1934). These

objections have never been satisfactorily met, and it must be admitted that there is
no entirely rigorous quantum theory of conductivity. So far as the e€ ects considered
here are concerned, however, it may be argued that the free path appears as a
parameter which need not itself be determined by the theory and which determines
the phenomena in an essentially classical way; since the physical concepts involved
are classical, it is not perhaps surprising that the e€ ects can be explained by a theory
which employs essentially classical methods. In any case, the great success of the
theory in explaining the experimental facts strongly encourages belief in the basic
validity of its assumptions .

2. Conduction in thin ®lms and wires

2.1. Thin ®lms
The strict analysis of the conductivity of a thin metallic ®lm was given by Fuchs
(1938). The method is typical of all strict calculations on the various size e€ ects and
will therefore be presented in detail.
Consider a metal ®lm of thickness a and suppose that the z-axis is perpendicular
to the plane of the ®lm, the surfaces of the ®lm being the planes z ˆ 0 and z ˆ a. The
problem is essentially a one-dimensional one, and the distribution function of the
electrons may be written in the form
f ˆ f0 ‡ f1 …v; z†; …8†
where the function f1 which has to be determined depends on the space variables
only through z. The electric ®eld E is supposed to be in the x-direction, and, if we
neglect the product of E with f1 (which is permissible so long as we are not interested
in deviations from Ohm’s law), the Boltzmann equation reduces to
@f1 f1 °E @f0
‡ ˆ : …9†
@z ½ vz mv z @vx
The general solution is easily written down, and is
» ¼
°½ E @f0 z
f1 …v; z† ˆ 1 ‡ F …v† exp ¡ ; …10†
m @vx ½ vz
where F …v† is an arbitrary function of v.
 506 E. H. Sondheimer

2.11. To determine F…v† we have to introduce the boundary conditions at the

surfaces of the ®lm. The simplest assumption is to suppose that every free path is
terminated by collision at the surface, so that the scattering is entirely di€ use. The
distribution function of the electrons leaving each surface must then be independent
of direction; equation (10) shows that this can only be satis®ed if we choose F …v† so
that f1 …v; 0† ˆ 0 for all v such that vz > 0 (that is, for electrons moving away from
the surface z ˆ 0), and f1 …v; a† ˆ 0 for all v such that vz < 0: There are therefore two
distribution functions, f ‡ ¡
1 for electrons with v z > 0 and f 1 for electrons with vz < 0;
they are
» ¼ 9
°½E @f0 z
‡
f 1 …v; z† ˆ 1 ¡ exp ¡ …vz > 0†; >
>
>
=
m @vx ½ vz
» ¼ …11†
°½E @f0 a¡z >
>
f¡ …v; z† ˆ 1 ¡ exp …v z < 0†: >
;
1
m @vx ½ vz
Downloaded by [University of Guelph] at 12:49 01 May 2012

2.1.1. We can now proceed with the calculation of the current density J…z†.
Combining (7), (8), (11), introducing polar coordinates …v; ³; ¿† in the v-space (with
vz ˆ v cos ³), and remembering that f0 depends on v ˆ jvj only, we obtain
… … 2p
2°2 m2 E 1 @f0
J…z† ˆ ¡ dv d¿ ½ v3 cos2 ¿
h3 0 0 @v
"…
p=2 n ± z ²o
sin3 ³ 1 ¡ exp ¡ d³
0 ½ v cos ³
…p #
n ± a ¡ z ²o
3
‡ sin ³ 1 ¡ exp d³ : …12†
p=2 ½v cos ³

The integration over ¿ is trivial, and the integration over v is carried out by means of
the formula
…1
@f0
¡ Á…v† dv ˆ Á…·v†; …13†
0 @v
which holds for a degenerate electron gas. This gives, after some rearrangement ,
… p=2 » ± ¼
4p°2 m2 ½ v·3 3 a ² a ¡ 2z
J…z† ˆ E sin ³ 1 ¡ exp ¡ cosh d³; …14†
h3 0 2l cos ³ 2l cos ³
where l ˆ ½ v· is the free path of the electrons at the surface of the Fermi distribution.
Equation (14) gives the current distribution across the thickness of the ®lm. For
comparison with experiment, however, we require the apparent overall conductivity
of the ®lm, and we must therefore average the current density over all values of z
from 0 to a. Carrying out the integration over z, we obtain for the e€ ective
conductivity
… " … #
1 a 3l p=2 3 n ± a ²o
¼ˆ J…z† dz ˆ ¼0 1 ¡ sin ³ cos ³ 1 ¡ exp ¡ d³ ; …15†
Ea 0 2a 0 l cos ³

¼0 being the conductivity of the bulk metal, which is given by equation (1b). The
ratio of the resistivity 1=¼ of the ®lm to that of the bulk metal 1=¼ 0 is most
 Mean free path of electrons in metals 507

conveniently written in the form, obtained from (15) by means of a simple

substitution,
¼0 F…µ†
ˆ ; …16†
¼ µ
where µ ˆ a=l and where
…1
1 1 3 3 1 1 ¡µt
ˆ ¡ ‡ ¡ e dt: …17†
F…µ† µ 8µ2 2µ2 1 t3 t5

Repeated integration by parts gives the alternative expression

1 1 3 1 3 5 1 µ ¡µ
ˆ ¡ 1 ¡ µ2 Ei…¡µ† ¡ 2 …1 ¡ e¡µ † ¡ ‡ ¡ e ; …18†
F…µ† µ 4 12 8µ 8µ 16 16
Downloaded by [University of Guelph] at 12:49 01 May 2012

where
…1
e¡t
¡Ei…¡u† ˆ dt; …19†
u t

which is a tabulated integral; this form of the result is particularly convenient for
computation.
The limiting form for large µ (thick ®lms) is

¼0 3
ˆ 1‡ …µ ¾ 1†; …20†
¼ 8µ
and for very thin ®lms we have

¼0 4
ˆ …µ ½ 1†: …21†
¼ 3µ log…1=µ†

These formulae are discussed further in section 2.3.

2.1.2. A somewhat more general theory, which does not assume that the
scattering at the surface of the ®lm is entirely di€ use, can be obtained as follows.
We assume that a fraction p of the electrons is scattered elastically at the surface with
reversal of the velocity component vz , while the rest are scattered di€ usely with
complete loss of their drift velocity. p is supposed to be a constant independent of the
direction of motion of the electrons. This is of course a highly arti®cial model, which
in e€ ect merely interpolates between the extreme cases of perfectly di€ use re¯ection,
considered above, and perfectly specular re¯ection, for which the conductivity is
unaltered. However, in the absence of any detailed theory of the nature of the surface
scattering mechanism it is best to work with the simplest possible assumptions .
The distribution function of the electrons leaving the surface z ˆ 0 is now
given by

f0 ‡ f ‡ ¡
1 …vz ; z ˆ 0† ˆ pf f0 ‡ f 1 …¡vz ; z ˆ 0†g ‡ …1 ¡ p† f0 ; …22†

and similarly, at z ˆ a,

f0 ‡ f ¡ ‡
1 …vz ; z ˆ a† ˆ pf f0 ‡ f 1 …¡vz ; z ˆ a†g ‡ …1 ¡ p† f0 : …23†
 508 E. H. Sondheimer

Table 1. The resistivity of thin metallic ®lms divided by the resis-

tivity of the bulk metal.

¼0 =¼
µ
(thickness/free path) pˆ0 p ˆ 12

0.001 182 73.5

0.002 100.4 41.5
0.005 46.6 20.0
0.01 26.5 11.8
0.02 15.3 7.1
0.05 7.69 3.87
0.1 4.72 2.62
0.2 3.00 1.91
0.5 1.90 1.402
1 1.462 1.206
Downloaded by [University of Guelph] at 12:49 01 May 2012

2 1.221 1.102
5 1.081 1.039
10 1.0390 1.0191
20 1.0191 1.0095
50 1.0076 1.0038
100 1.0038 1.0019

These equations are su cient to determine F …v†, and instead of (11) we obtain for
the distribution function
» ¼ 9
°½ E @f0 1¡p z >
>
f‡1 …v; z† ˆ 1 ¡ exp ¡ …v z > 0†; >
=
m @vx 1 ¡ p exp…¡a=½vz † ½vz
» ¼ …24†
°½ E @f0 1¡p a¡z >
>
¡
f 1 …v; z† ˆ 1¡ exp …vz < 0†: ;>
m @vx 1 ¡ p exp…a=½vz † ½vz
The current density is calculated as before, and the result is that the function F…µ†
de®ned by (17) must be replaced by Fp …µ†, where
…1
1 1 3 1 1 1 ¡ e¡µt
ˆ ¡ 2 …1 ¡ p† ¡ dt: …25†
Fp …µ† µ 2µ 1 t3 t5 1 ¡ pe¡µt

This reduces to (17) when p ˆ 0, and to the bulk-metal value 1=µ when p ˆ 1. It is, of
course, again possible to write down alternative forms of (25) more convenient for
numerical work; compare, for example, equation (31). Instead of (20) and (21), we
have
¼0 3
ˆ 1 ‡ …1 ¡ p† …µ ¾ 1† …26†
¼ 8µ
and
¼0 4 1 ¡ p 1
ˆ …µ ½ 1†: …27†{
¼ 3 1 ‡ p µ log…1=µ†

Values of ¼0 =¼ for p ˆ 0 and 1 are shown in Table 1.

2

{ The corresponding formula given by Fuchs (equation (23) of his paper) is incorrect.
 Mean free path of electrons in metals 509

2.2. Thin wires

The corresponding analysis for the case of wires has been carried out by
MacDonald and Sarginson (1950) , who considered a wire of square cross-section,
and by Dingle (1950) for the more important but more di cult case of a wire of
circular cross-section. The method is essentially the same as that for a ®lm, but the
details of the analysis, for which the reader is referred to the original papers, are
more complicated because the problems are now two-dimensional and the Boltz-
mann equation takes the form of a partial di€ erential equation in the space variables.
The result for a cylindrical wire of diameter a is,{ for p ˆ 0,
¼0 C…µ†
ˆ ; …28†
¼ µ
where µ ˆ a=l,
…1
1 1 12
…1 ¡ t2 †1=2 S4 …µt† dt;
Downloaded by [University of Guelph] at 12:49 01 May 2012

ˆ ¡ …29†
C…µ† µ pµ 0
and
…1
Sn …u† ˆ e¡ut …t2 ¡ 1†1=2 t¡n dt: …30†
1

The result for p 6ˆ 0 may be derived from that for p ˆ 0 by means of the simple
relation
¼ X1
¼
ˆ …1 ¡ p†2 npn¡1 ; …31†
¼0 µ;p nˆ1
¼0 nµ;pˆ0

this holds incidentally also for the case of a thin ®lm, as may be shown by expanding
(25) in ascending powers of p and comparing with (17).
For very thick and very thin wires the results reduce to
¼0 3
ˆ 1 ‡ …1 ¡ p† …µ ¾ 1† …32†
¼ 4µ
and
¼0 1 ¡ p 1
ˆ …µ ½ 1†: …33†
¼ 1‡pµ
(32) holds also for a square wire (µ being a=l as before where a is now the side of the
wire), but (33) must be replaced by
¼0 1 ¡ p 0:897
ˆ …µ ½ 1†: …34†
¼ 1‡p µ
Numerical values of ¼0 =¼ for a cylindrical wire are shown in Table 2.

2.3. Discussion of the formulae

It will be noted that, although the general theory is complicated, relatively simple
formulae are obtained for specimens which are either thick or thin compared with
the free path. The results for these limiting cases in fact possess a simple interpret-
ation.

{ Note that Dingle denotes the radius of the wire by a.

 510 E. H. Sondheimer

Table 2. The resistivity of thin wires divided by the resistivity of the

bulk metal.

¼ 0 =¼
µ
(thickness/free path) pˆ0 p ˆ 12

0.001 1000 337

0.002 503 170
0.005 202 69.3
0.01 102 35.7
0.02 52.1 18.7
0.05 21.6 8.31
0.1 11.45 4.88
0.2 6.33 3.02
0.5 3.14 1.84
1 2.04 1.422
Downloaded by [University of Guelph] at 12:49 01 May 2012

2 1.475 1.208
5 1.172 1.080
10 1.081 1.038
20 1.0390 1.0191
50 1.0152 1.0076
100 1.0076 1.0038

2.3.1. The results given above for thick specimens are particular cases of a
formula applicable to wires with cross-sections of arbitrary shape, the general form
of which follows from simple dimensional considerations (Dingle 1950). When the
thickness is large compared with the free path, the deviation in the current density
from that which would exist in the bulk metal for the same applied ®eld is
appreciable only in a region close to the surface of the specimen. The deviation in
the current is proportional to the perimeter P of the cross-section, whilst the mean
current is proportional to the cross-sectional area S. Hence
¼0 P
ˆ1‡ constant ˆ 1 ‡ ClP=S; …35†
¼ S
where C is a dimensionless constant independent of the shape of the specimen. For
a thin ®lm of thickness a, P=S ˆ 2=a, and comparison with (26) shows that
C ˆ 3…1 ¡ p†=16. Therefore, for a wire of arbitrary cross-section,
¼0 3 lP
ˆ 1 ‡ …1 ¡ p† : …36†
¼ 16 S
In the notation of section 2.2, P=S ˆ 4=a for both square and circular wires, and we
regain (32).

2.3.2. We may de®ne an `e€ ective’ free path leff for conduction in thin
specimens by writing the e€ ective conductivity in the standard form
¼ ˆ n°2 leff =m·
v. For wires of diameter small compared with the free path, we see
from equations (33) and (34) that leff ˆ a, apart from a constant of the order of
unity. This result is intuitively obvious; for a very thin ®lm, however, we ®nd that
leff a log…l=a†, so that leff tends to in®nity with l. The qualitative di€ erence between
the behaviour of thin ®lms and of wires may at ®rst sight seem surprising; its origin
 Mean free path of electrons in metals 511

becomes more apparent if we consider a simple qualitative treatment of conduction

in thin ®lms due to Lovell (1936). We assume that all free paths start at the surface,
and that the e€ ective free path is the average of all free paths of a given electron, the
free path being taken as the distance to the next intersection with the surface or the
ordinary free path, whichever is least. Assuming further that the density of electrons
travelling inwards from the surface is unifom (di€ use scattering), we have, if ³ is the
angle between the direction of motion and the normal to the ®lm surface,
… ³0 … p=2
a
leff ˆ sin ³ d³ ‡ l sin ³ d³;
0 cos ³ ³0

where cos ³0 ˆ a=l. This gives

1
leff ˆ a log ‡ 1 ; …37†
a
Downloaded by [University of Guelph] at 12:49 01 May 2012

which is in qualitative agreement with the exact result when l ¾ a. Examination of

the integrals shows that the main contribution to leff when l ¾ a comes from values
of ³ close to ³0 ; the dominant contribution to the current in ®lms, however thin
compared with the free path, therefore arises from electrons which, after re¯ection at
the surface of the ®lm, move in directions so nearly parallel to the surface that their
free path remains of the order of the bulk free path. In the case of wires the
constriction of free paths due to the geometry is more severe, and such electrons do
not play any appreciable part.

2.4. Comparison with experiment

It was known already at the end of the nineteenth century that very thin ®lms of
metal exhibit a higher electrical resistivity than the same metal in bulk, and J. J.
Thomson (1901) was the ®rst to suggest that the e€ ect arose from the limitation of
the free path of the electrons and to give an approximate theory; he derived a
formula similar to (37). Since then an enormous amount of experimental work has
been carried out on the electrical properties of thin ®lms; we mention in particular
only Lovell’s careful measurements on ®lms of the alkali metals (Lovell 1936), and
refer the reader for further information and references to the critical review by
Chambers (1951). However, as we have already emphasized, if the measurements are
to give information about the electronic parameters characteristic of the bulk metal,
great care must be taken to eliminate all e€ ects which may be peculiar to thin
specimens other than the pure geometrical limitation of the free path with which the
theory is alone concerned. In particular, both strain and impurity may be expected to
play an increasing part in determining the phenomena as the size of the specimen is
reduced. In fact it was not until recently that Andrew, working with relatively thick
®lms at liquid-hydroge n and liquid-helium temperatures, obtained data which could
be compared with any con®dence with the theoretical predictions (Andrew 1949).
For thin wires there is much less experimental work than for ®lms. The most
important earlier experiments in this ®eld are those of Eucken and FoÈrster (1934) and
succeeding workers, and data on wires were also obtained by Andrew in his recent
work. To illustrate the kind of information that can be obtained from the experiments,
Andrew’s results on both ®lms and wires are compared with the theory below.
There are two ways in which the ratio a=l may be varied in carrying out the
experiments. We may ®rstly use a single specimen and vary l by changing the
temperature. Since the bulk conductivity ¼0 cannot be measured on the specimen
 512 E. H. Sondheimer

itself, it must then be assumed that the temperature variation of ¼0 , and hence of l, is
the same for a thin specimen as for a bulk specimen. Alternatively (and this was
Andrew’s procedure) ¼ may be measured at one or more ®xed temperatures for a
number of specimens of di€ erent thickness or diameter; it is then important to ensure
that the bulk free path is the same for all specimens.

2.4.1. Figure 1 shows Andrew’s results on the resistance of rolled foils of tin at
3.88K. The resistance, expressed as a fraction of the value at 2918K, is plotted
logarithmically against the foil thickness, which ranges from 3 to 2000 m.
In comparing these results with the theory, it is necessary to assume a value for
the surface re¯ection coe cient p. Such evidence as exists mostly tends to show that
the scattering is di€ use, so that p ˆ 0 (compare sections 2.4.1.1 and 4.6.4 below), and
this was the value adopted by Andrew; the theoretical curve of ¼0 =¼ for a ®lm as a
function of a=l can then be ®tted directly to the experimental results at constant
Downloaded by [University of Guelph] at 12:49 01 May 2012

temperature, and the full curve in ®gure 1 has been drawn in this way. The rather
large scatter of the observations about the theoretical curve is probably due to a
variation in residual resistivity between the specimens, but on the whole the
agreement between theory and experiment is satisfactory. The process of ®tting
gives directly the value of the free path l corresponding to the temperature of the
experiment and the ratio of the bulk conductivity ¼0 to the (known) room-
temperature value. The results are that l ˆ 9:5 10¡3 cm at 3.88K, and ¼0 =l ˆ
4:5 1022 gaussian units. Measurements at higher temperatures con®rm that the
value of ¼0 =l is constant, and with this value equation (4) gives n ˆ 1:6 1022 .
The number of atoms per unit volume, na , is 3:7 1022 for tin, and thus n=na ˆ 0:43.
Such a value for the number of electrons per atom is entirely reasonable, although in
a complicated metal such as tin the precise ®gure has little signi®cance.

2.4.1.1. Andrew’s measurements on wires were carried out using mercury, and
the results for 3.58K and 2.58K are shown in ®gure 2. The exact analysis of
conduction in thin wires had not been given at the time of Andrew’s experiments,
and the full curves in ®gure 2 were obtained on the basis of an approximate theory

Figure 1. Experimental results on the resistivity of tin foils at 3.88K, plotted logarithmically
against the foil thickness. The full line is the theoretical curve for di€ use scattering
…p ˆ 0†.
 Mean free path of electrons in metals 513
Downloaded by [University of Guelph] at 12:49 01 May 2012

Figure 2. Experimental results on the resistivity of mercury wires at 3.58K and 2.58K,
plotted logarithmically against the wire diameter. The full lines are theoretical curves
for di€ use scattering.

similar to Lovell’s. Dingle (1950) later re-examined Andrew’s results and found that
the exact theory did not substantially alter the conclusions.{ Dingle considered in
particular the question of the value to be assigned to p; he found that, if a non-zero
value is assumed, agreement with experiment could only be obtained by supposing
that p increases slightly with temperature. This is not plausible on theoretical
grounds, since the various mechanisms which may lead to di€ use surface scattering,
such as the thermal vibration of the surface atoms or the number of mobile surface
defects of various kinds, would be expected to become more e€ ective the higher the
temperature, causing p to decrease. The value p ˆ 0 is therefore the most likely; it
must be emphasized, however, that more experimental and theoretical work is
required before the question can be regarded as de®nitely settled (compare also
section 4.6.4.1). Adopting the value p ˆ 0 and disregarding the results obtained for
the thinnest wires used in the experiments, which do not ®t in with the theoretical
predictions, it is found that l ˆ 5:6 10¡3 cm at 2.58K and a0 =l ˆ 2:5 1022 , which
corresponds to 0.15 electrons per atom in mercury.

2.4.2. Corresponding experiments have been carried out by MacDonald and

Sarginson (1950) on thin wires of very pure sodium; such experiments are of
particular interest, since for an alkali metal quantitative agreement with the free-
electron theory is to be expected. MacDonald and Sarginson compare their results
with the theory developed by them for a square wire, using values of l estimated on
the basis of the free-electron theory from the observed conductivity of a bulk sample.
(l 4:5 10¡3 cm at 4.28K.) The agreement between theory and experiment is only
fair, and MacDonald and Sarginson conclude that the hypothesis of entirely di€ use
scattering at the boundaries is not generally ful®lled for their specimens, and that the

{ It is worth noting that the simple formula ¼0 =¼ ˆ 1 ‡ 1=µ gives values of ¼0 =¼ which di€ er
from the exact values (for p ˆ 0) by less than 5% over the whole range of µ. This formula (or a slight
generalization of it) is associated with the name of Nordheim, and was used in the evaluation of the
earlier experimental work on thin wires (Nordheim 1934, Eucken and FoÈrster 1934).
 514 E. H. Sondheimer

degree of elastic scattering is a function of the specimen size. Further experiments on

sodium are considered in sections 3.2 and 3.3.

2.5. The `kinetic theory’ solution of thin-conducto r problems

Chambers (1950a) has recently given an elegant formulation of the theory of
thin-conductor problems, in which he writes down the solution of the Boltzmann
equation in a form suggested by simple kinetic theory considerations.
In the presence of an electric ®eld E in the x-direction, and with f ˆ f0 ‡ f1 …v; r†,
the Boltzmann equation takes the form
E° @f0 f1
v · grad r f1 ¡ ˆ¡ : …38†
m @vx ½
A particular solution is
» ¼
°½E @f0 jr ¡ r0 j
f1 …v; r† ˆ 1 ¡ exp ¡ …39†
Downloaded by [University of Guelph] at 12:49 01 May 2012

;
m @vx ½v
where r ¡ r0 is a vector parallel to v. This solution applies to conductors of arbitrary
shape and, if r ˆ r0 is taken to be a point on the surface of the conductor, it satis®es
the boundary conditions corresponding to di€ use scattering at the surface.
Equation (39) may be derived from ®rst principles as follows. Consider electrons
passing through a point r in the metal in the direction r0 ¡ r. The probability that an
electron will travel a distance s and then su€ er a collision between s and s ‡ ds is
e¡s=l ds=l, but electrons which arrive at r0 will certainly collide there if the scattering
is di€ use. The mean distance travelled by an electron without collision after passing
through r is therefore
… jr¡r0 j …1
l0 ˆ se¡s=l ds=l ‡ jr ¡ r0 j e¡s=l ds=l ˆ l…1 ¡ e¡jr¡r0 j=l †; …40†
0 jr¡r0 j

also, for electrons travelling in the opposite direction r ¡ r0 , the mean distance
travelled without collision before reaching r is given by the same expression. The
mean drift velocity acquired by these electrons in the electric ®eld is therefore
°El 0 °½E
Dvx ˆ ˆ …1 ¡ e¡jr¡r0 j=l †; …41†
mv m
and f1 which is the change due to the presence of the ®eld in the number of electrons
travelling in the direction r ¡ r0 with velocity v, is given by
@f0 °½ E @f0
f1 …v; r† ˆ Dvx ˆ …1 ¡ e¡jr¡r0 j=l †; …42†
@vx m @vx
in agreement with equation (39).

2.5.1. The current density may now be calculated exactly as in section 2.1.1,
except that the appropriate polar coordinate system (v; ¯; $) to be used here has the
x-axis as polar axis, this being the only preferred direction which occurs in the
present general formulation of the theory. The e€ ective conductivity ¼ is obtained by
integrating the current density over the cross-sectional area S of the conductor, and
the ratio of ¼ to the bulk conductivity ¼0 may be written in the compact form
… … 2p …p
¼ 3
ˆ 1¡ dS d$ d¯ sin ¯ cos2 ¯ e¡jr¡r0 j=l : …43†
¼0 4pS S 0 0
 Mean free path of electrons in metals 515

To evaluate ¼=¼ 0 for any shape of conductor it is necessary to express jr ¡ r0 j in

terms of the position of the point r and the angles ¯ and $ which de®ne the direction
of r ¡ r0 , and to carry out the integrations. For example, the expressions (15) and
(28) given above for a thin ®lm and a thin wire may readily be derived in this way; for
a ®lm, however, it is convenient ®rst to change back to the polar coordinates …v; ³; ¿†
used in section 2.1.1. A more complicated problem, which well illustrates the power
of the present method, is provided by the case of a thin wire placed in a longitudinal
magnetic ®eld; this is discussed in section 3.2.
The case of partially elastic scattering has also been considered by Chambers,
who derived the appropriate generalization of (43); in this case, however, it is
impossible to obtain a simple formula applicable to conductors of arbitrary cross-
section.
Downloaded by [University of Guelph] at 12:49 01 May 2012

3. Magnetic e€ ects in thin conductors

3.1. In considering conduction in thin specimens when a magnetic ®eld is
present, we shall con®ne attention to the ®lm and the circular wire, and to purely
longitudinal or transverse ®elds. There are then altogether ®ve possible arrange-
ments: these fall into three groups, depending on whether the magnetic ®eld is
parallel or perpendicular to the electric current, and whether the Hall electric ®eld
which is set up in the latter case is or is not uniform across the specimen.
(A) H parallel to J. No Hall ®eld is produced, and this considerably simpli®es the
analysis. There are two cases in this group:
(A1) the wire with H and J parallel to the axis (section 3.2), and
(A2) the ®lm with H and J parallel to each other and in the plane of the
®lm.
(B) H perpendicular to J: Hall ®eld uniform. The Hall ®eld is set up in the
direction perpendicular to both H and J to prevent any current ¯ow in this
direction. If it is developed across a `thin’ dimension of the specimen it will
vary in magnitude across the specimen and its determination as a function of
position presents considerable di culties (see section 3.5). In one case,
however, namely that of a thin ®lm with H perpendicular to the plane of
the ®lm, the Hall ®eld (being in the plane of the ®lm) is developed across a
`long’ dimension of the specimen and is consequently constant. This case is
in fact very simple to analyse (section 3.4).
(C) H perpendicular to J: Hall ®eld non-uniform . The two remaining cases fall
into this group: they are
(C1) the ®lm with H in the plane of the ®lm but perpendicular to J, and
(C2) the wire with H perpendicular to the axis.
An approximate treatment has been given of case (C1) (section 3.5); case (C2) has
not yet been considered, and indeed is of such complexity as to be virtually insoluble.
Exact treatments have so far been given only of cases (A1) and (B), and these will
be considered in some detail below. Experimental data are so far available only for
wires (cases (A1) and (C2)); the case (A1) of a thin wire in a longitudinal magnetic
®eld is thus of particular interest as being the only case at present for which there
exist both an exact evaluation of the theory and experimental results with which to
compare it.
 516 E. H. Sondheimer

A thin-conductor e€ ect of the present type was ®rst observed by MacDonald

(1949). The ordinary bulk magneto-resistanc e e€ ect always leads to an increase in
resistance; MacDonald found, however, that the resistance of a thin sodium wire
decreased with H in a longitudinal ®eld, whilst in a transverse ®eld it increased
initially, passed through a maximum and then decreased. Sodium has also been used
in subsequent investigations (MacDonald and Sarginson 1950, Chambers 1950a);
this metal (and possibly pure lithium and potassium) appears unfortunatel y to be the
only metal suitable for the experiments in view of the higher inherent bulk magneto-
resistance of all other metals, including even the heavier alkali metals (MacDonald
1950). The bulk e€ ect is greatest under just the same conditions of low temperatures
and high ®elds as are required to observe the free electron phenomena, and in general
it obliterates the latter entirely.

3.2. The conductivity of thin wires in a longitudinal magnetic ®eld

Downloaded by [University of Guelph] at 12:49 01 May 2012

MacDonald, at the time of his discovery of the e€ ect, correctly interpreted its
physical origin, and in particular he explained the simple decrease in resistance which
occurs in a wire in a longitudinal ®eld as being due to the lessened infuence of
scattering at the walls of the wire when the electrons are forced to pursue spiral paths
around the lines of force of the magnetic ®eld.
An exact analysis of this case (A1) has been given by Chambers (1950a),
assuming di€ use scattering at the boundaries. When the electric and magnetic ®elds
are parallel, the magnetic force on the electrons is always perpendicular to the
electric force; we can then regard the electric ®eld alone as producing a drift current
in the usual way, and the magnetic ®eld simply as modifying the electronic
trajectories. The solution of the Boltzmann equation is therefore formally the same
as in the absence of a magnetic ®eld (equation (42)), and the conductivity is still given
by (43), provided that jr ¡ r0 j is replaced by the distance s from the point r to the
point on the surface r0 measured along the curved trajectory of the electrons.

3.2.1. The integral (43) was evaluated by Chambers for this case by means of an
ingenious series of transformations . With a magnetic ®eld H along the x-axis (the
axis of the wire), electrons travelling at an angle ¯ to the x-axis will move in helical
paths whose projections on the yz-plane are circles of radius
r ˆ m·
vc sin ¯=°H ˆ r0 sin ¯: …44†{
If now, while an electron is travelling from r0 to r, the projection of its path on the
yz-plane traverses an angle Á around such a circle, then the projection of the distance
s on the yz-plane is Ár0 sin ¯, so that s ˆ Ár0 . Equation (43) therefore becomes
… … 2p …p
¼ 3
ˆ 1¡ dS d$ d¯ sin ¯ cos2 ¯ e¡r0 Á=l ; …45†
¼0 4pS S 0 0

where Á ˆ Á…y; z; ¯; $†.

3.2.2. If we consider a ®xed value of the azimuth angle $ and perform the
integrations over S and ¯, it is evident by symmetry that the result must be

{ r and r0 in this equation must not, of course, be confused with the magnitudes of the position
vectors r and r0 .
 Mean free path of electrons in metals 517
Downloaded by [University of Guelph] at 12:49 01 May 2012

Figure 3. Construction for the determination of P…®; Á0 † (see text).

independent of $; we may therefore restrict ourselves to a particular value of $, say

$ ˆ 0. We may furthermore con®ne ourselves to values of ¯ in the range …0; 12 p† and
write (45) in the form
… …
¼ 3 p=2 2
ˆ 1¡ d¯ sin ¯ cos ¯ dS…e¡r0 Á=l †$ˆ0 : …46†
¼0 S 0 S

The integrand of (46) refers to electrons which, since colliding with the wall, have
turned through angles between Á and Á ‡ dÁ and are, at the instant considered,
travelling in the direction (¯; $ ˆ 0†. If we denote the proportion of the total cross-
section of the wire occupied by these electrons by p…®; Á† dÁ (where ® is de®ned
below), we may rewrite (46) as
… p=2 … 2p
¼
ˆ 1¡3 d¯ sin ¯ cos2 ¯ p…®; Á† e¡r0 Á=l dÁ: …47†
¼0 0 0

It remains to ®nd an expression for p…®; Á†, which is a function only of Á and of the
ratio of the radius 12 a of the wire{ to the radius r ˆ r0 sin ¯ of the projection of the
orbit, i.e. of ® ˆ 2r=a ˆ 2 sin ¯=­ , where ­ ˆ a=r0 .
Chambers has given the following construction for the determination of p…®; Á†
(®gure 3). Suppose, for example, that r 4 12 a (® 4 1), and consider electrons which
traverse the projection of their orbit in the anti-clockwise direction and which are
travelling vertically downwards at the instant considered. Let O be the centre of the
wire, and let R be chosen such that OR is horizontal and of length r. With centre R,
draw the circles OPQ of radius r and BDE of radius 12 a ¡ r. The circle BDE is then
such that, for the group of electrons considered, no point within it can be reached by
a trajectory starting from the wall. Supose that OQ subtends an angle Á0 at R, and
with centre Q and radius 12 a draw the circular arc ABC which touches the circle BDE
at B; it is easily seen that this is the locus of the positions of all electrons of the kind
considered which have turned through an angle Á0 since colliding with the wall. The
portion BC must be excluded, however, since it corresponds to electrons whose

{ Note that Chambers uses a for the radius of the wire.

 518 E. H. Sondheimer

trajectory lies partly outside the wire; if, instead, we continue with the arc BD of the
circle BDE, then the shaded area between ABD and the surface of the wire contains
R Á have turned through angles Á 4 Á0 . The shaded area is
all those electrons which
therefore P…®; Á0 † ˆ 0 0 p…®; Á† dÁ. p…®; Á† may be obtained from this by di€ erentia-
tion or, alternatively, (47) may be expressed directly in terms of P…®; Á† by partial
integration.
Analytical expressions for p…®; Á† may be obtained by considering the geometry
of ®gure 3 and the corresponding ®gure for ® > 1. It is readily shown that, for ® 4 1,
pp…®; Á† ˆ ®…1 ¡ ®† ˆ ® 2 cos 2 12Á ‡ ® cos 12Á…1 ¡ ®2 sin2 12Á†1=2 …0 4 Á 4 2p†;
…48†
and, for ® 5 1,
)
pp…®; Á† ˆ 2® cos 12Á…1 ¡ ® 2 sin2 12Á†1=2 …0 4 Á 4 2 sin¡1 …1=®††;
…49†
Downloaded by [University of Guelph] at 12:49 01 May 2012

ˆ0 …2 sin¡1 …1=®† < Á 4 2p†;

where the positive value of the square root is to be taken.
Note that, for ­ 5 2, ® 4 1 for all ¯, while for ­ < 2, we have ® < 1 for small ¯
and ® > 1 for large ¯… 12p†. Physically this means simply that for strong enough
®elds …­ 5 2† all electronic trajectories are curved into paths of radius less than the
wire radius; for smaller ®elds, those electrons travelling at small angles to the axis
and therefore having small transverse velocities will still follow such paths, but
electrons moving at greater angles to the axis will follow paths of radius greater than
the wire radius.

3.2.3. Values of ¼=¼ 0 may now be computed by graphical integration using

®gure 3, or by approximate analytical methods using equations (48) and (49). The
®nal results, as is clear from (47), depend only on the two dimensionless parameters
­ ˆ a=r0 and µ ˆ a=l, which are measures of the strength of the magnetic ®eld and of

Figure 4. Theoretical conductivity of thin wires in a longitudinal magnetic ®eld.

 Mean free path of electrons in metals 519

the free path respectively. In ®gure 4 theoretical values of the resistivity ratio ¼0 =¼
are shown plotted against 12 ­ , for various values of µ. The values for ­ ˆ 0 are those
given by Dingle (section 2.2), and, as is to be expected on physical grounds, the
resistance decreases steadily as ­ increases and tends to the bulk value as ­ tends to
in®nity.

3.2.4. Comparison with experiment

If we plot µ¼0 =¼ against 12 ­ for various µ, then, since µ ˆ a=l and ¼0 =l is
independent of the temperature, the ordinates are proportional to the resistivity 1=¼,
and, since 12 ­ ˆ °aH=…2m· vc†, the abscissae are proportional to the applied magnetic
®eld. The theoretical curves of µ¼0 =¼ against 12 ­ , plotted logarithmically, are
therefore directly superposable on the experimental curves of 1=¼ against H at a
number of temperatures, and the proportionality constants give values of ¼0 =l and
m·v directly. The advantage of using a magnetic ®eld is that all the required
Downloaded by [University of Guelph] at 12:49 01 May 2012

information can be obtained from one specimen only, and it is not necessary to
assume, as in zero-®eld experiments, that the free path in a thin specimen is the same
as in a bulk specimen.
Figure 5 shows experimental results obtained by Chambers on a sodium wire of
diameter 30 m at various temperatures; also shown are the theoretical curves for
µ ˆ 0:5, 1 and 2, adjusted to ®t as well as possible. It has, of course, been assumed
that the hypothesis of di€ use re¯ection at the surface is correct, and that the bulk
magneto-resistanc e e€ ect is negligible. The departure from theory in high ®elds is, in
fact, due to the onset of the bulk e€ ect, but this does not interfere seriously with
the comparison between theory and experiment. The results of the comparison are
that, in sodium, ¼0 =l ˆ 8:1 1022 and m· v ˆ 9:1 10¡20 ; these values provide two
independent estimates of the number of conduction electrons, and agree reasonably
well with the values
¼0 =l ˆ 6:3 1022 and m·
v ˆ 9:7 10¡20
obtained from (4) and (1a) assuming one conduction electron per atom.

Figure 5. Experimental results on a 30 m sodium wire in a longitudinal magnetic ®eld at

various temperatures, with theoretical curves.
 520 E. H. Sondheimer
Downloaded by [University of Guelph] at 12:49 01 May 2012

Figure 6. Experimental results on sodium wires of various diameters in a transverse

magnetic ®eld. Diameters: 1, 20 m; 2, 30 m; 3, 66 m. Temperature ˆ 4:2 K.

3.3 Similar experiments have been reported by MacDonald and Sarginson

(1950) on sodium wires in both longitudinal and transverse magnetic ®elds, and
some typical results for transverse ®elds are shown in ®gure 6; note particularly that
in this case the resistance may increase initially with H. Detailed theoretical analysis
of this arrangement (C2) is prohibitively di cult, and MacDonald and Sarginson
compared their results with an approximate theory developed by them for the
somewhat simpler thin-®lm arrangement (C1) (see section 3.5), with which they are
in qualitative agreement.

3.4. The conductivity of thin ®lms in a transverse magnetic ®eld

The case (B) of a thin ®lm in a magnetic ®eld which is perpendicular to the plane
of the ®lm has been analysed by Sondheimer (1950) . In this case it is not immediately
obvious from simple physical arguments how the presence of the magnetic ®eld
a€ ects the conductivity at all. The formal analysis, however, presents no di culties
and is a straightforward generalization of the treatment given in section 2.1 for a ®lm
in zero magnetic ®eld; it leads to the interesting result that the conductivity is an
oscillatory function of the strength of the applied magnetic ®eld.

3.4.1. We use the notation of section 2.1, but suppose now that the ®lm is
subjected to an electric ®eld …Ex ; Ey ; 0† in the plane of the ®lm and a magnetic ®eld
…0; 0; H†. For this particular arrangement, the condition curl E ˆ 0 ensures that
the electric ®eld components are constant across the thickness of the ®lm. The
Boltzmann equation takes the form
@f1 f1 °H @f1 @f1 ° @f0 @f0
‡ ¡ vy ¡ vx ˆ Ex ‡ Ey ; …50†
@z ½ vz mcv z @vx @vy mv z @vx @vy
 Mean free path of electrons in metals 521

which is the generalization of equation (9). To solve this equation we put

@f0
f1 ˆ …vx c1 ‡ vy c2 † ; …51†
@v
where c1 and c2 do not depend explicitly on vx and vy . With this form for f1 , equation
(50) leads to two simultaneous equations for c1 and c2 ; if we introduce the complex
quantities g ˆ c1 ¡ ic2 , F ˆ Ex ¡ iEy , we may write these equations in the compact
form
@g 1 i°H °F
‡g ‡ ˆ ; …52†
@z ½vz mcvz mvv z
and the problem is now formally identical with that of section 2.1.

3.4.2. The rest of the calculation is entirely analogous to that of section 2.1, and
Downloaded by [University of Guelph] at 12:49 01 May 2012

the result may be expressed as follows. We de®ne a complex conductivity ¼c by

means of the relation J· ˆ ¼c F , where J· ˆ J·x ¡ iJ·y is the complex current density
averaged in the usual way across the thickness of the ®lm. We further introduce the
parameters µ ˆ a=l, ­ ˆ a=r0 (a being the thickness of the ®lm), and the complex
variable s ˆ µ ‡ i­ . We then have, as the generalization of (16),
¼0 F…s†
ˆ ; …53†
¼c µ
where the function F is given as before by (17) when p ˆ 0 and by (25) when p 6ˆ 0,
but is now a function of a complex variable. (We may remark here that the present
problem is the only one of the magnetic-®eld problems which can easily be solved for
the case p 6ˆ 0.)

3.4.3. Equation (53) contains all the results required for comparison with
experiment. The ordinary electrical conductiviy ¼, for example, is measured by
applying an electric ®eld in, say, the x-direction and observing the current in this
direction, no electric current being allowed to ¯ow in the transverse direction. We
therefore have ¼ ˆ R…J·†=R…F †, with I …J·† ˆ 0, where R and I denote the real and
imaginary parts respectively; eliminating the transverse electric ®eld, we readily ®nd
that
¼0 1
ˆ ¼0 R ˆ RfF…s†g=µ: …54†
¼ ¼c
The Hall coe cient is de®ned by AH ˆ Ey =H J·x , where Ey is the transverse electric
®eld set up under the above experimental conditions, and we easily obtain
AH =AH;0 ˆ I fF…s†g=­ ; …55†
where AH;0 ˆ ¡1…n°c† is the Hall coe cient of the bulk metal.

3.4.4. In the limit of zero magnetic ®eld …­ ˆ 0†, equation (54) for the
conductivity reduces to equation (16), and (55) gives a corresponding expression
for the Hall coe cient of a thin ®lm in a vanishingly small magnetic ®eld. In this
limit AH =AH;0 depends only on µ, and is shown graphically in ®gure 7 for three
values of p; it is seen that the Hall coe cient of a thin ®lm shows an increase above
 522 E. H. Sondheimer
Downloaded by [University of Guelph] at 12:49 01 May 2012

Figure 7. Theoretical Hall coe cient of thin metallic ®lms (in small magnetic ®elds) divided
by the Hall coe cient of the bulk metal.

the bulk value analogous to the increase of the electrical resistivity. For very small µ,
we ®nd that
AH 41 ¡ p 1
ˆ ; …56†
AH;0 3 1 ‡ p µflog …1=µ†g2
which shows that the Hall coe cient increases less rapidly than the resistance
(compare (27)).

3.4.5. The ®eld variation of the resistivity may be computed from (54), and, as
mentioned above, it is found that the resistance oscillates with the strength of the
applied magnetic ®eld. Figure 8 shows some typical curves for p ˆ 0 and various
values of µ. In general the resistance increases initially with H and reaches its ®rst
maximum when ­ is approximately unity; the higher oscillations are of roughly
constant spacing but decrease rapidly in amplitude, and in very strong ®elds the
resistance tends to the constant asymptotic value
±¼ ² 3
0
ˆ 1 ‡ …1 ¡ p†; …57†
¼ ­ ˆ1 8µ
which, for small µ, is greater than the value in zero ®eld. The oscillations die out and
the resistance tends towards the bulk value as µ becomes large or as p tends to unity;
the position of the maxima and minima, however, is almost independent of µ and p,
and ­ may thus be estimated unambiguously by observing the value of H for which,
for example, the ®rst maximum is reached (assuming, however, as usual that the bulk
magneto-resistanc e e€ ect is negligible).
The Hall coe cient, on the other hand, does not oscillate with H, and decreases
steadily towards the bulk value as H increases to in®nity.
 Mean free path of electrons in metals 523
Downloaded by [University of Guelph] at 12:49 01 May 2012

Figure 8. Theoretical variation of resistance of thin metallic ®lms with magnetic ®eld
perpendicular to surface of ®lm, assuming di€ use surface scattering.

3.4.6. The resistance oscillations have been explained by Chambers (1950a) as

being essentially due to the oscillations in speed of an electron moving in
perpendicular electric and magnetic ®elds. The speed at time t is a trigonometric
function of °Ht=mc ˆ v·t=r0 , and this causes the distribution function f1 …v; z† to be a
¯uctuating function of z=r0 . The e€ ect occurs only in thin ®lms because the presence
of the metal surface is required to provide a ®nite limit to the distance from which
electrons can come to contribute to the current at z; when the thickness is large
compared with the free path, the elementary oscillating contributions to f1 …v; z† add
up to give a non-¯uctuating total, and in fact f1 is then independent of z.
Equation (57) shows that, in a strong magnetic ®eld, the e€ ective free path of the
electrons (see section 2.3.2) is of the order of the thickness of the ®lm. We therefore
have the physically plausible result that a strong transverse magnetic ®eld eliminates
the contribution to the current of those electrons moving nearly parallel to the
surface which are responsible in the absence of a ®eld for the logarithmic dependence
of leff on the ratio l=a.

3.5. MacDonald and Sarginson (1950) have given an elaborate discussion of a

thin ®lm for the case (C1) where the magnetic ®eld is in the plane of the ®lm but
perpendicular to the current. For the details of the analysis the reader is referred to
the original paper. The Hall ®eld is now in the z-direction (perpendicular to the plane
of the ®lm) and varies across the thickness of the ®lm. It can, in principle, be
evaluated as a function of z and H by solving a complicated integral equation which
expresses the condition that the current density in the z-direction must vanish
everywhere. MacDonald and Sarginson have evaluated the Hall ®eld approximately
for the case of small magnetic ®elds, and have found that it increases rapidly near the
edges of the ®lm but is approximatel y constant over most of the cross-section, as
shown in ®gure 9. In their calculation of the conductivity MacDonald and Sarginson
 524 E. H. Sondheimer

Figure 9. Theoretical variation of Hall ®eld Ez across a thin ®lm of thickness

a ˆ 0:64 free path, assuming di€ use scattering at the boundaries
Downloaded by [University of Guelph] at 12:49 01 May 2012

were forced, however, to assume for simplicity that the Hall ®eld was constant over
the whole cross-section. In view of this assumption the results must be treated with
caution; they indicate that there are no oscillations, but that, for su ciently thin
®lms, the resistance increases initially with H, passes through a single maximum and
then decreases steadily towards the bulk metal value.

3.6. The only magnetic-®eld experiments on thin ®lms so far reported (Mac-
Donald and Sarginson 1950) have been carried out on foils of silver, gold and tin.
Although small indications of the e€ ects discussed above were observed, it was
found that the bulk magneto-resistanc e e€ ect was the dominating factor in all cases,
and no de®nite conclusions could be drawn. Experiments on thin ®lms of sodium in
magnetic ®elds both in and perpendicular to the plane of the ®lm are highly desirable
in order to test the theoretical predictions; at the same time the theory should be
rounded o€ by a discussion of the third case (A2) where the magnetic ®eld is in the
plane of the ®lm and parallel to the current.

4. The anomalous skin e€ ect

4.1. H. London (1940) discovered that the resistivity of metals at high
frequencies and very low temperatures was much greater than is predicted by the
usual theory of the skin e€ ect. He suggested that a free-path phenomenon was
involved, but did not elaborate his idea. The problem was later taken up by Pippard
(1947), who studied it in detail both experimentally and theoretically; using mainly
physical arguments he succeeded in elucidating all the essential features involved.
The quantitative theory of the e€ ect is based on the methods discussed in the present
article and is due to Reuter and Sondheimer (1948).

4.2. The normal skin e€ ect

The basic concepts are best introduced by considering the theory of the skin e€ ect
in metals in its simplest form. Let the surface of a semi-in®nite metal be the xy-plane
and let the positive z-axis be directed towards the interior of the metal. The electric
®eld E…z†ei!t is taken to be in the x-direction, ! being the angular frequency, and the
magnetic ®eld H…z†ei!t is in the y-direction; the factor ei!t which occurs in all time-
dependent quantities need not be written down explicitly and will in future be omitted.
 Mean free path of electrons in metals 525

Neglecting the displacement current, Maxwell’s equations take the form

¡H 0 …z† ˆ 4pJ…z†=c; E 0 …z† ˆ ¡i!H…z†=c; …58†
where J…z† is the current density. Eliminating H, we obtain
E 00 …z† ˆ 4pi!J…z†=c2 ; …59†
and, combining this with Ohm’s law in the form J ˆ ¼0 E, we have
E 00 …z† ˆ 4pi!¼0 E…z†=c2 ; …60†
so that
p







E…z† ˆ E…0†e¡kz ; with k ˆ …1 ‡ i† 2p!¼0 =c ˆ …1 ‡ i†=¯; …61†
p







¯ ˆ c= 2p!¼0 …62†
being the depth of penetration of the ®eld (the `classical skin depth’).
Downloaded by [University of Guelph] at 12:49 01 May 2012

In high-frequency experiments the quantity most directly accessible to meas-

urement is not the resistivity of the metal in the ordinary sense, but the surface
resistivity R, which is the real part of the surface impedance Z. Z is de®ned as the
ratio of the electric ®eld at the surface of the metal to the total current per unit area
of surface, and thus ¿ …1
Z ˆ R ‡ iX ˆ E…0† J…z† dz; …63†
0

X being the surface reactance. Equations (59) and (61) give

p 








R ˆ X ˆ 2p!=c2 ¼0 ; …64†
p
and R and X therefore
p vary directly as ! for a specimen at a given temperature,
and inversely as ¼0 for a given frequency.

4.2.1. These formulae are in good agreement with observation on metals at room
temperature for all wavelengths in the radio-frequency region. Pippard found,
however, that as ¼0 is increased by lowering the temperature, the surface con-
ductivity 1=R increases less rapidly than is predicted by equation (64) and that, as ¼0
tends to in®nity, 1=R tends to a constant value which is di€ erent for each metal; the
typical form of the curves is shown in ®gure 11 below. This is the phenomenon
known as the anomalous skin e€ ect, and it evidently involves a breakdown of the
simple theory outlined above.

4.2.2. In the above form the theory is essentially phenomenological in character

and makes no direct reference to electron theory; it is based only on Maxwell’s
equations and the assumption that Ohm’s law may be used to relate the current at a
point in the metal to the electric ®eld at the same point. It is well known that Ohm’s
law must be generalized by introducing a complex conductivity when the frequency
is so high that the period of the applied ®eld becomes comparable with the time of
relaxation of the electrons (Drude 1904, compare also Wilson 1936, p. 124).
Relaxation e€ ects of this nature, however, play no signi®cant role in the anomalous
skin e€ ect except at infra-red frequencies; they are considered brie¯y in section 4.7.1,
but may be disregarded for the present. The generalization of Ohm’s law which
concerns us here is of a di€ erent type, and is connected with the very long free path
 526 E. H. Sondheimer

of the conduction electrons when the conductivity is high; when the free path is
comparable with the skin depth ¯, it is not permissible to assume that an electron
moves under the in¯uence of a constant ®eld between collisions, and the current at a
point in the metal will depend on values of the electric ®eld at other points.
Mathematically, the departure from Ohm’s law is a consequence of the
appearance of the usual term v · grad r f in the Boltzmann equation. The present
problem di€ ers from most of the previous ones, however, in that the electric ®eld in
the metal is not given a priori, but must itself be determined by the theory. When the
relation between the current and an arbitrary ®eld E…z† has been found by solving
the Boltzmann equation, it must be combined with Maxwell’s equation (59) to give
an equation from which E…z† can be calculated; the surface impedance can then be
obtained from the equation
4pi! E…0†
Zˆ¡ ; …65†
c2 E 0 …0†
Downloaded by [University of Guelph] at 12:49 01 May 2012

which follows directly from (59) and (63).

4.3. The fundamenta l equation

So long as we are only interested in the surface impedance of the metal (and not,
for example, in the transmission coe cient of a thin ®lm), the penetration depth of
the ®eld may always be treated as small compared with the linear dimensions of the
specimen, and we may continue to consider a semi-in®nite slab of metal. The
Boltzmann equation then takes the same form (9) as in the discussion of the d.c.
conductivity of a thin ®lm, namely
@f1 f1 ° @f0
‡ ˆ E…z†; …66†
@z ½ vz mvz @vx
where the electric ®eld is now, however, a function of z. (The in¯uence of the high-
frequency magnetic ®eld may be neglected in writing down equation (66).) The
general solution is now
» … ¼
z ° @f0 z t
f1 …v; z† ˆ exp ¡ F …v† ‡ E…t† exp dt : …67†
½ vz mvz @vx ½vz
The value of F …v† when vz > 0 is determined in the usual way by the nature of the
scattering at the surface of the metal, which is described by a re¯ection coe cient p;
when vz < 0, F …v† must be chosen so that f1 does not become exponentially large as
z ! 1. De®ning E…¡z† ˆ E…z† for z < 0, the solution which satis®es the boundary
conditions may be written in the form
» …z 9
° @f0 z t >
f‡ …v; z† ˆ exp ¡ p E…t† exp dt >
>
1
mvz @vx ½ vz ½ vz >
>
¡1 >
>
…z ¼ >
>
t =
‡…1 ¡ p† E…t† exp dt …vz > 0†; …68†
0 ½ vz >
>
>
>
…1 >
>
° @f0 z t >
>
¡
f 1 …v; z† ˆ ¡ exp ¡ E…t† exp dt …vz < 0†: >
;
mvz @vx ½ vz z ½ vz
The ®rst of these expressions shows that the e€ ect of partially elastic scattering may
be described formally by replacing the semi-in®nite metal slab by an in®nite metal in
 Mean free path of electrons in metals 527

which the ®eld in the region z < 0 is an image of the ®eld in the region z > 0 and in
which a fraction p of the electrons moves through the image ®eld.

4.3.1. The calculation of the current density is now straightforward, and the
result is
» …1 ± …1 ± ¼
3¼0 z ¡ t² z ¡ t²
J…z† ˆ p k E…t† dt ‡ …1 ¡ p† k E…t† dt ; …69†
4l ¡1 l 0 l
where
…1
1 1
k…u† ˆ ¡ e¡sjuj ds; …70†
1 s s3

it is seen that the current density at z involves the values of the electric ®eld at all
points in the metal. The two integrals on the right-hand side of (69) represent the
Downloaded by [University of Guelph] at 12:49 01 May 2012

contributions to J of the electrons which su€ er specular and di€ use re¯ection
respectively at the surface of the metal. We may note here that in the anomalous
skin e€ ect the value of p plays only a subordinate part in determining the surface
impedance (see section 4.6.1); there is thus a basic di€ erence between the present
phenomenon and the d.c. thin-conductor free-path e€ ects which are in general
entirely due to the presence of di€ use scattering at the boundaries.

4.3.2. The equation for E is obtained by combining (59) and (69). It is convenient
to introduce dimensionless coordinates x ˆ z=l, y ˆ t=l, and to write E…lx† ˆ f …x†;
the fundamental equation of the problem is then obtained in the form
» …1 …1 ¼
00
f …x† ˆ i¬ p k…x ¡ y† f …y† dy ‡ …1 ¡ p† k…x ¡ y† f …y† dy ; …71†
¡1 0

where
¬ ˆ 3p!l 2 ¼0 =c2 ˆ 32 l 2 =¯ 2 : …72†

For the evaluation of the surface impedance it is only necessary to know f …0†=f 0 …0†
(see equation (65)), and f …x† is not required explicitly.
It is easily shown that the results of section 4.2 are regained in the limit where
¬ ½ 1.

4.4. Solution of the fundamental equation

In the two limiting cases p ˆ 1 and p ˆ 0 equation (71) can be solved by standard
methods based on the theory of Fourier integrals (Titchmarsh 1937). When p ˆ 1,
for example, the equation reduces to
…1
f 00 …x† ˆ i¬ k…x ¡ y† f …y† dy; …73†
¡1

where f …x† is, by de®nition, an even function of x. It is easy to see that f 0 …x† is
discontinuous at x ˆ 0, and tends to a limit · (say) as x ! ‡0 (¡· as x ! ¡0†. The
formal solution of (73) now proceeds as follows. Let
…1 …1
¡ixt
¿…t† ˆ f …x†e dx ˆ 2 f …x† cos xt dx …74†
¡1 0
 528 E. H. Sondheimer

and
…1
µ…t† ˆ k…x†e¡ixt dx: …75†
¡1

From (74), integrating by parts twice, it follows that

…1
f 00 …x†e¡ixt dx ˆ ¡t2 ¿…t† ¡ 2·; …76†
¡1

on the other hand, (73) implies that

…1 …1 …1
00 ¡ixt ¡ixt
f …x†e dx ˆ i¬ e dx k…x ¡ y†f …y† dy
¡1 ¡1 ¡1

ˆ i¬µ…t†¿…t† …77†
(inverting the order of integration). Hence, from (76) and (77),
Downloaded by [University of Guelph] at 12:49 01 May 2012

¿…t†ft2 ‡ i¬µ…t†g ˆ ¡2·;

and the inversion formula for Fourier transforms gives the ®nal result
…
2· 1 cos xt dt
f …x† ˆ ¡ : …78†
p 0 t2 ‡ i¬µ…t†
For x ˆ 0, in particular, it follows that
…1
f …0† 2 dt
ˆ¡ : …79†
0
f …0† p 0 t2 ‡ i¬µ…t†
The explicit form of µ…t† is readily obtained from (70) and (75), and is
µ…t† ˆ 2t¡3 f…1 ‡ t2 † tan¡1 t ¡ tg: …80†

4.4.1. When p ˆ 0, equation (71) is of the same general type as the well-known
integral equation of Milne which has been much studied in the theory of radiative
transfer and in neutron di€ usion problems. The solution in this case is based on the
method which has been applied to Milne’s equation by Wiener and Hopf; for details
of the analysis the reader is referred to Reuter and Sondheimer’s paper. The results
are complicated but lead to conclusions which are generally similar to those found
for p ˆ 1 (see section 4.6).

4.5. The electric ®eld

The electric ®eld f …x† is not itself required for the evaluation of the surface
impedance, but its behaviour is of interest. It is clear from (78) that f …x† is not of
exponential form, and in general its evaluation is complicated; for large values of x,
however, it may be shown (Reuter and Sondheimer 1948, appendix III) that, when
p ˆ 1,
f …x† ˆ C1 e¡s1 x ‡ C2 e¡x =x2 ; …81†
where C1 and C2 are constants, and where s1 is a complex parameter whose real part
Rs1 increases steadily with ¬ and equals unity when ¬ ˆ ¬0 2:63.
The physical signi®cance of this result is best understood by considering the case
of a metal which is unbounded in all directions (Pippard, Reuter and Sondheimer
1948). In this case, when ¬ 4 ¬0 , the electric ®eld is found to be the simple damped
 Mean free path of electrons in metals 529

exponential wave e¡s1 x , the penetration depth of which is l=Rs1 and is thus equal to l
when ¬ ˆ ¬0 .{ For ¬ > ¬0 there are no solutions at all, which means that an
exponential wave which is attenuated by a factor e in a distance smaller than the free
path cannot exist: it would correspond to a situation in which the contribution to the
current at a point is greater the further the electrons contributing to the current have
travelled in the wave direction, so that the expression for the resultant current
density at any point diverges.
It is therefore clear that, in order to obtain solutions for all values of ¬, it is
essential to take the boundary conditions at the surface of the metal speci®cally into
account in the calculation; it is to be expected, furthermore, that there will be a
fundamental di€ erence in the nature of the electric ®eld for values of ¬ less or greater
than the critical value ¬0 . This is in accordance with the form of equation (81). When
¬ < ¬0 , the ®rst term is dominant and the ®eld in the metal, at su cient distances
from the surface, is una€ ected by the presence of the surface, being similar in form to
Downloaded by [University of Guelph] at 12:49 01 May 2012

the solution for an in®nite metal; when ¬ > ¬0 , however, the second term is
dominant, and the form of the excitation is now conditioned at all points by the
presence of the surface and never approximates to a simple exponential form, being,
in fact, a true surface excitation. In the latter case the electric ®eld, though largely
con®ned to the surface, has a long `tail’ of small amplitude which is e€ ectively
damped out in a distance of the order of l and which may be regarded as being
transmitted into the metal by electrons which move in the interior under the
in¯uence of no forces except their collisions with the lattice.

4.6. The surface impedance

The surface impedance can be obtained from (65) and (79) (or the corresponding
expression for p ˆ 0), and in general the integrals have to be evaluated numerically.
Explicit expressions for Z can, however, be obtained in the limiting case (¬ ¾ 1)
when the free path is very large.

4.6.1. The asymptotic expressions for Z

The limiting value of f …0†=f 0 …0† when ¬ ¾ 1 is obtained from (79) by replacing
µ…t† by its approximate value p=t for large t. The integral is then readily evaluated
and leads to the following limiting value of Z when p ˆ 1:
p 










 p 

Z1 ˆ 89 … 3p!2 l=c4 ¼0 †1=3 …1 ‡ 3i†: …82†

The corresponding result when p ˆ 0 is

p 




 




 p


Z1 ˆ … 3p!2 l=c4 ¼0 †1=3 …1 ‡ 3i†: …83†

These expressions are to be contrasted with the result (64) which holds when ¬ ½ 1.
We see that Z is independent of l when the free path is very large, in accordance with
the experimental results; further, Z varies with frequency as !2=3 , and the surface
resistivity and reactance are related by
p 



X1 ˆ 3R1 …84†

{ We recall that x is the distance measured in units of l.

 530 E. H. Sondheimer

under extreme anomalous conditions. These results hold both for p ˆ 1 and for
p ˆ 0, and it is evident that although the precise nature of the surface scattering plays
some part in determining the surface impedance, the e€ ect is not considerable.

4.6.2. The `ine€ ectiveness concept’

The physical principles underlying these results are not readily apparent from the
detailed calculation, and it is therefore worth while to outline a simple qualitative
theory which was developed by Pippard (1947) before the exact solution had been
obtained. When the free path is large compared with the penetration depth, the
electrons may be divided into two groups, those moving at small angles to the
surface which have a reasonable chance of colliding in the surface layer, and the rest
whose chance of such a collision is small. Pippard gave reasons for supposing that
only the former group of electrons contributes e€ ectively to the resistance, and the
proportion of e€ ective electrons may thus be written ­ ¯0 =l, where ¯0 is the (unknown)
Downloaded by [University of Guelph] at 12:49 01 May 2012

e€ ective penetration depth and where ­ is a numerical constant of order unity. It is

now supposed that the ine€ ective electrons may be entirely neglected, and that the
e€ ective electrons move in a constant ®eld during a free path, so that the theory of
the normal skin e€ ect applies to them. The e€ ective conductivity is then ¼0 ˆ ­ ¯0 ¼0 =l,
and ¯0 is given by an expression of the form (62) with ¼0 replaced by ¼0 . We thus
obtain an implicit equation for ¯0 , which, when solved, gives

¯0 ˆ …c2 l=2p!­ ¼0 †1=3 : …85†

Inserting ¼0 instead of ¼0 in (64) and using (85), we obtain the surface resistivity in
the form
R1 ˆ …4p2 !2 l=­ c4 ¼0 †1=3 ; …86†

in qualitative agreement with (82) and (83).

Later work on the interpretation of the theory (Marcus, to be published) has
shown that the physical picture underlying the ine€ ectiveness concept must not be
taken too literally, but it does provide a useful guiding principle which is of
particular value when considering more complicated situations for which no exact
theory exists (compare, for example, section 4.7.2).

4.6.3. Since E…0†=E 0 …0† ˆ lf …0†=f 0 …0†, equations (65) and (79) show that Z
depends on the free path both through l and ¬. It is convenient to express Z as a
function only of ¬ and of quantities independent of l; using equations (1b) and (72)
to eliminate l, we ®nd that
q 
Z ˆ ¡i 83A¬1=3 f …0†=f 0 …0†; …87†
where
p
± p! ²2=3 m·
v 1=3
Aˆ 6 2 : …88†
°c 3n

A is constant for a given metal and a given frequency and has been chosen such that,
when ¬ ½ 1,

Z ˆ A¬¡1=6 …1 ‡ i†; …89†

 Mean free path of electrons in metals 531
Downloaded by [University of Guelph] at 12:49 01 May 2012

Figure 10. The theoretical high-frequency surface impedance of metals at low temperatures.

as is easily veri®ed. When ¬ ¾ 1, we obtain

p 
4 2 p

Z1 ˆ 1=3 A…1 ‡ 3i …p ˆ 1† …90†
9p
and
A p


Z 1 ˆ p



…1 ‡ 3i …p ˆ 0†; …91†
2p1=3
equations (89), (90), and (91) are of course equivalent to (64), (82), and (83). The
complete theoretical curves are shown in ®gure 10, where A=R and A=X have been
plotted against ¬1=6 .

4.6.4. Comparison with experiment

The experimental method used by Pippard to measure R involves the construc-
tion of a resonator of the metal to be studied and the determination of its selectivity
Q; for a given resonator, Q is proportional to the surface conductivity 1=R. The
p
results for any metal are exhibited by plotting 1=R against ¼0 at constant
p p
frequency. Since ¼0 and ¬1=6 both vary as l, the experimental points can be
scaled directly to ®t the theoretical curve if it is assumed that p equals either 1 or 0. It
is then possible, by reading o€ the value of ¬ corresponding to any pair of values of
R and ¼0 , to estimate ¼0 =l for any metal; according to equations (62) and (72),
p
¼0 =l ˆ …3p!¼30 =c2 ¬†.
Pippard’s (1947) results were in qualitative agreement with the theory, but were
not su ciently detailed to indicate which value is to be chosen for p. Chambers
(1950b) has recently carried out more extensive experiments, in which the tempera-
ture was varied between 28K and 908K; the frequency (as in Pippard’s experiments)
was 1200 Mc/s, corresponding to a free-space wavelength of 25 cm. Some of
Chambers results for copper, silver, and tin are shown in ®gure 11, scaled (I) to ®t
the theoretical curve to p ˆ 0, and (II) in an attempt to ®t the curve for p ˆ 1. The
agreement with curve I is really excellent, and it thus appears that the re¯ection at
the surface is completely di€ use at all temperatures for all specimens.
 532 E. H. Sondheimer
Downloaded by [University of Guelph] at 12:49 01 May 2012

Figure 11. Experimental results on the surface resistance at low temperatures.

(I) Experimental points ®tted to p ˆ 0 curve. (II) Attempt to ®t to p ˆ 1 curve.

4.6.4.1. These experiments, as Chambers has pointed out, suggest strongly that
the value of p which describes the situation in real metals is zero in all cases,
including the d.c. size e€ ects (compare sections 2.4.1 and 2.4.1.1). It should be noted
that the appropriate value of p need not necessarily be the same for all the di€ erent
problems studied: although p has been de®ned in the same way in each case, it
probably represents some kind of average value of some more complicated quantity,
and there is no reason to suppose that this average should have the same value in all
cases. If Pippard’s ine€ ectiveness concept is taken literally, the value of p in the
anomalous skin e€ ect is determined by the behaviour on re¯ection of electrons
travelling at very small angles to the surface. If specular re¯ection occurs at all, it
seems most likely to occur for these electrons, and since it is not observed for them
(i.e. since p ˆ 0 in the anomalous skin e€ ect), it is reasonable to expect that p ˆ 0
also for the d.c. case, where electrons travelling at all angles to the surface are
`e€ ective’.

4.6.4.2. Values of ¼0 =l deduced from the experiments are shown in table 3,

together with the corresponding values of n=na , the number of conduction electrons

Table 3. Estimates of ¼0 =l and n=na from

observations on the anomalous skin effect.

…¼0 =l† 10 ¡22

Metal (gaussian units) n=na

Cu 13.9 1.0
Ag 8.3 0.68
Au 7.6 0.60
Sn 8.6 1.1
Hg 3.3 0.23
Al 5.5 0.37
 Mean free path of electrons in metals 533

per atom. The values for copper, silver, gold, and tin were obtained by Chambers,{
and are subject to probable errors of about 5%; the values for mercury and
aluminium have been deduced from Pippard’s measurements and have larger
probable errors.
The results for the monovalent metals are in fair agreement with the expected
value of one conduction electron per atom, although the ®gures for silver and gold
are somewhat too low. The ®gures for tin and mercury may be compared with the
corresonding ®gures (n=na ˆ 0:43 and 0.15) obtained from the thin-conductor
experiments; the agreement is probably as good as can be expected for such
complicated metals.

4.6.4.3. The present method of estimating the free path is free from the main
di culties which beset the thin-conductor experiments, such as the impossibility of
measuring the bulk conductivity directly on the specimen, or the disturbing in¯uence
Downloaded by [University of Guelph] at 12:49 01 May 2012

of the bulk magneto-resistanc e e€ ect. It is, however, very important to ensure that
the specimens have no surface layer with properties di€ erent from those of the bulk
metal: if, for example, there is a thin surface layer, of thickness comparable with the
skin depth, with an abnormally high residual resistance, the temperature variation of
the surface resistivity will simulate the behaviour in the anomalous skin e€ ect,
although in this case the high resistivity will not be a result of the long free path. This
is in fact what appears to have happened in early measurements by Pippard on wires
of the noble metals, which led to very low values of n=na . The process of drawing
wires of these metals produces an amorphous layer which is not easily removed by
annealing, but which, in Chambers’ experiments, was removed by electrolytic
polishing. In subsidiary experiments Chambers has con®rmed that the predominant
cause of abnormally high surface losses is the presence of a high-resistance layer, and
that possible alternative causes such as the roughness of the surface are generally of
smaller importance.
Similar disturbing in¯uences appear to have played a part in experiments on the
anomalous skin e€ ect carried out in the United States (Fairbank 1949, Maxwell,
Marcus, and Slater 1949).

4.6.4.4. Pippard (1950a) has recently made experiments on tin using a frequency
of 9400 Mc/s, corresponding to a wavelength of 3.2 cm. By comparing the results
with the 25 cm measurements, Pippard con®rmed that the asymptotic value of the
surface resistance varies with frequency as !2=3 for a given material; this law is found
to be accurately obeyed. At the same time Pippard was able to estimate the surface
reactance X by an indirect method involving the superconducting behaviour of the
metal. He found pthat X1 =R1 1:76 for tin, in quite as good agreement with the
theoretical value 3 as may be expected. These experiments are considered further in
section 4.72.

4.7. Extensions of the theory

We conclude by considering brie¯y various generalizations of the basic theory
given above.

{ These values di€ er from those given in the published note (Chambers 1950b); they are based on
more accurate values of the d.c. conductivity ¼0 .
 534 E. H. Sondheimer

4.7.1. Relaxation e€ ects

Up to now it has been assumed that the product !½ is small compared with unity.
In the 9400 Mc/s experiments this condition is no longer satis®ed, but we have seen
that the results are in good agreement with the non-relaxatio n theory. This is
explained by the extension of the theory to take account of relaxation, which may be
obtained by replacing the time of relaxation in the Boltzmann equation (66) by a
complex time of relaxation ½=…1 ‡ i!½†. It is then found (Reuter and Sondheimer
1948, section 6) that the integrals involve, instead of i¬, the generalized parameter
¹ ˆ i¬=…1 ‡ i!½†3 ; so long as j¹j ¾ 1, however, the ®nal expressions obtained for Z
are identical with (82) and (83), so that relaxation e€ ects are negligible. The criterion
j¹j ¾ 1 for neglecting relaxation is equivalent to the condition, which may be easily
understood physically, that the period 1=! of the applied ®eld shall be large
compared with, not ½ , but the time taken by an electron to traverse the skin depth,
which is very much smaller than ½ . As a result it is not to be expected that relaxation
Downloaded by [University of Guelph] at 12:49 01 May 2012

e€ ects will play any part at wavelengths greater than about 10 1 mm.

So long as the displacement current is negligible, the absorption coe cient A of a

metal for normally incident radiation, de®ned as the ratio of the intensities of the
absorbed and incident waves, is directly proportional to the surface resistivity.
p
According to the usual relaxation theory A varies as ! at low frequencies (this is
the Hagen±Rubens relation), but is independent of frequency when !½ ¾ 1. The
di€ erence in relaxation behaviour which the anomalous skin e€ ect introduces at low
temperatures is illustrated in ®gure 12, which shows the theoretical absorption
coe cient of a good silver specimen at liquid-helium temperatures and at frequencies
ranging from long radio waves to ultra-violet light. The anomalous e€ ects are most
marked in the long infra-red, while at higher frequencies the relaxation e€ ects tend to
restore the normal behaviour; the maximum value of A in ®gure 12 is 1:5 10¡3 and
occurs at a wavelength of 70 m.
The absorptivity of pure electropolished metals for infra-red radiation of
wavelength 14 m has recently been measured by Ramanathan (unpublished) at
liquid-helium temperatures. The observed behaviour is not in agreement with the

Figure 12. Theoretical absorption coe cient of silver at liquid-helium temperatures for
p ˆ 1, plotted logarithmically against the wavelength ¶.
 Mean free path of electrons in metals 535

theory: the absorptivit y in all cases exceeds the theoretical value by a factor of the
order of 10, and is in fact about as great as the value obtained by extrapolating
the !2=3 law which holds when relaxation e€ ects are negligible. The origin of the
discrepancy is not clear at present, but it is possible that the disagreement with the
relaxation theory is only apparent and is connected with the value of the surface
re¯ection coe cient p. p was assumed to be unity by Reuter and Sondheimer when
evaluating their formulae for high frequencies; it has been reported that in the infra-
red region the absorptivity for p ˆ 0 is much higher than for p ˆ 1 (Holstein,
unpublished calculations), but details of the computations are not yet available.

4.7.2. The anomalous skin e€ ect in anisotropic metals

In his experiments at 9400 Mc/s, Pippard (1950a) studied the surface resistance
of single crystals of tin as a function of orientation. In ®gure 13 1=R, measured under
Downloaded by [University of Guelph] at 12:49 01 May 2012

extreme anomalous conditions, is displayed as a function of ³, the angle between the

tetrad axis of the crystal and the axis of the (cylindical) specimen, which is also the
direction of current ¯ow.
The form of the anisotropy of 1=R is interesting, and di€ ers from the result to be
expected for a metal obeying Ohm’s law. In an anisotropic metal the usual form of
Ohm’s law, J ˆ ¼E, must be generalized by treating ¼ as a symmetric second-order
tensor. The theory of the normal skin e€ ect in such a metal has been considered by
Pippard, who has shown that, for the conditions of his experiments, it leads to a
monotonic variation of R with ³, which is at variance with the experimental results.
Pippard has pointed out that the reason for the anomalous behaviour is to be found
in the form of the equations connecting J and E in the normal and anomalous
theories. In the former the relation is a linear point-relation whether the metal is
isotropic or not, the value of J at any point being determined solely by the value of E
at that point. In the anomalous theory, on the other hand, the relation between J and
E, while still linear, is no longer a point relation; there is then no longer any quantity
analogou s to ¼ which may be simply generalized by the introduction of a tensor, and
as a consequence the mathematical formulation of the problem permits a much
greater freedom of behaviour in an anisotropic metal than is allowed by a point-
relation. It is clear, furthermore, that the free-electron model which we have always
employed hitherto will be entirely inadequate to deal with the present problem, and
the anisotropic behaviour will depend on the anisotropic form of the energy surfaces

Figure 13. Variation with orientation of surface conductivity of single crystals of tin.
 536 E. H. Sondheimer

in real metals. A ®rst discussion of the problem using the ine€ ectiveness concept has
been given by Pippard, but further investigation, both experimental and theoretical,
is required before the phenomenon can be said to be fully understood.

4.7.3. Application to superconductivit y

Most of the recent investigations on the high-frequency behaviour of metals at
low temperatures have been concerned primarily with the properties of the super-
conducting state, a topic which we have ignored hitherto and which is outside the
scope of the present article. It should be mentioned, however, that the methods used
in the theory of anomalous skin e€ ect ®nd an application in the theoretical analysis
of certain models of a superconductor, the so-called two-¯uid models. The essential
feature of these models is that the electrons in a superconductor may be thought of
as belonging to either of two groups, superconducting and normal respectively, and
that the current density may be written as the sum of separate contributions from
Downloaded by [University of Guelph] at 12:49 01 May 2012

each group, the supercurrent and the normal current. Under the in¯uence of an
electric ®eld the normal electrons in a superconductor are supposed to behave in the
same way as the electrons in a normal metal, and they will therefore exhibit similar
free-path phenomena at high frequencies. The theory has been worked out in detail
by Maxwell, Marcus and Slater (1949) for the particular model of a superconductor
which is due to H. London (1940), and a careful critical discussion of the whole
subject has been given by Pippard (1950b), whose paper should be consulted for
further details. It must be emphasized that, since there is as yet no accepted and fully
developed theory of superconductivity , the status of these calculations on super-
conductors is much more speculative than that of the theory which applies to normal
metals.

4.7.4. Magnetic e€ ects

We wish to point out, ®nally, that e€ ects analogous to those considered in section
3 are to be expected at high frequencies if a steady magnetic ®eld is superimposed on
the specimen. In particular, if the magnetic ®eld is perpendicular to the surface of the
metal, the surface impedance will vary with H in a manner which may be analysed
theoretically by combining the methods of section 3.4 with those of the present
section; at the same time there will be a high-frequenc y Hall electric ®eld, and at low
temperatures there will be an `anomalous Hall e€ ect’. The analysis of these
phenomena, though laborious, is straightforward in principle, but elaborate calcula-
tions are probably not worth undertaking until expeimental data have become
available.

Acknowledgement
My best thanks are due to Mr R. G. Chambers for several instructive discussions
and for reading and commenting on the article in manuscript.

References
Andrew, E. R., 1949, Proc. Phys. Soc. A, 62, 77.
Chambers, R. G., 1950a, Proc. Roy. Soc. A, 202, 378; 1950b, Nature, Lond., 165, 239; 1951,
PhD Dissertation, Cambridge.
Dingle, R. B., 1950, Proc. Roy. Soc. A, 201, 545.
Drude, P., 1904, Ann. Phys., 14, 936.
 Mean free path of electrons in metals 537

Eucken, A., and FoÈ rster, F., 1934, Nach. Ges. Wiss. GoÈttingen, 1, 43, 129. See also papers
by various authors in Ann. Phys., 1937±41.
Fairbank, W. M., 1949, Phys. Rev., 76, 1106.
Fuchs, K., 1938, Proc. Camb. Phil. Soc., 34, 100.
London, H., 1940, Proc. Roy. Soc. A, 176, 522.
Lovell, A. C. B., 1936, Proc. Roy. Soc. A, 157, 311.
MacDonald, D. K. C., 1949, Nature, Lond., 163, 637; 1950, Proc. Phys. Soc. A, 63, 290.
MacDonald, D. K. C., and Sarginson, K., 1950, Proc. Roy. Soc. A, 203, 223.
Maxwell, E., Marcus, P. M., and Slater, J. C., 1949, Phys. Rev., 76, 1332.
Mott, N. F., and Jones, H., 1936, The Theory of the Properties of Metals and Alloys
(Oxford: University Press), p. 268.
Nordheim, L., 1934, Act. Sci. et Ind., No. 131 (Paris: Hermann).
Peierls, R., 1934, Z. Phys., 88, 786; Helv. Phys. Acta, 7 (Supp.), 24.
Pippard, A. B., 1947, Proc. Roy. Soc. A, 191, 385; 1950a, Ibid., 203, 98; 1950b, Ibid., 203,
195.
Pippard, A. B., Reuter, G. E. H., and Sondheimer, E. H., 1948, Phys. Rev., 73, 920.
Downloaded by [University of Guelph] at 12:49 01 May 2012

Reuter, G. E. H., and Sondheimer, E. H., 1948, Proc. Roy. Soc. A, 195, 336.
Sondheimer, E. H., 1950, Phys. Rev., 80, 401.
Thomson, J. J., 1901, Proc. Camb. Phil. Soc., 11, 120.
Titchmarsh, E. C., 1937, Introduction to the Theory of Fourier Integrals (Oxford:
Clarendon Press), chapter 11.
Wilson, A. H., 1936, The Theory of Metals (Cambridge: University Press).