Landau Theory
Landau Theory
Landau Theory
b InstituutLorentz,
Abstract
An introductory survey of the theoretical ideas and calculations and the experimental results which depart from Landau Fermi-liquids is presented. Common
themes and possible routes to the singularities leading to the breakdown of Landau
Fermi liquids are categorized following an elementary discussion of the theory. Soluble examples of Singular Fermi liquids include models of impurities in metals with
special symmetries and one-dimensional interacting fermions. A review of these is
followed by a discussion of Singular Fermi liquids in a wide variety of experimental
situations and theoretical models. These include the effects of low-energy collective
fluctuations, gauge fields due either to symmetries in the hamiltonian or possible
dynamically generated symmetries, fluctuations around quantum critical points, the
normal state of high temperature superconductors and the two-dimensional metallic
state. For the last three systems, the principal experimental results are summarized
and the outstanding theoretical issues highlighted.
Contents
Introduction
1.1
1.2
Landaus Fermi-liquid
2.1
2.2
11
2.3
17
1 February 2008
2.4
22
2.5
Modern derivations
27
2.6
29
34
3.1
34
3.2
38
3.3
39
3.4
Orthogonality Catastrophe
40
3.5
41
3.6
43
3.7
45
3.8
47
3.9
51
55
4.1
57
4.2
59
4.3
62
4.4
Thermodynamics
63
4.5
Correlation Functions
64
4.6
65
4.7
67
4.8
69
4.9
72
76
76
79
5.1
79
5.2
82
84
6.1
85
6.2
88
6.3
96
6.4
101
6.5
103
104
7.1
105
7.2
108
7.3
115
7.4
Microscopic Theory
117
123
8.1
124
8.2
126
8.3
130
8.4
Finkelstein Theory
135
8.5
137
8.6
Experiments
138
8.7
144
150
8.8
Acknowledgements
152
References
152
Introduction
In the last two decades a variety of metals have been discovered which display thermodynamic and transport properties at low temperatures which are
fundamentally dierent from those of the usual metallic systems which are
well described by the Landau Fermi-liquid theory. They have often been referred to as Non-Fermi-liquids. A fundamental characteristic of such systems
is that the low-energy properties in a wide range of their phase diagram are
dominated by singularities as a function of energy and temperature. Since
these problems still relate to a liquid state of fermions and since it is not a
good practice to name things after what they are not, we prefer to call them
Singular Fermi-liquids (SFL).
The basic notions of Fermi-liquid theory have actually been with us at an
intuitive level since the time of Sommerfeld: He showed that the linear low
temperature specic heat behavior of metals as well as their asymptotic low
temperature resisitivity and optical conductivity could be understood by assuming that the electrons in a metal could be thought of as a gas of noninteracting fermions, i.e., in terms of quantum mechanical particles which do
not have any direct interaction but which do obey Fermi statistics. Meanwhile Pauli calculated that the paramagnetic susceptibility of non-interacting
electrons is independent of temperature, also in accord with experiments in
metals. At the same time it was understood, at least since the work of Bloch
and Wigner, that the interaction energies of the electrons in the metallic range
of densities are not small compared to the kinetic energy. The rationalization
for the qualitative success of the non-interacting model was provided in a masterly pair of papers by Landau [144,145] who initially was concerned with the
properties of liquid 3 He. This work introduced a new way of thinking about the
properties of interacting systems which is a cornerstone of our understanding
of condensed matter physics. The notion of quasiparticles and elementary excitations and the methodology of asking useful questions about the low-energy
excitations of the system based on concepts of symmetry, without worrying
about the myriad unnecessary details, is epitomized in Landaus phenomenlogical theory of Fermi-liquids. The microscopic derivation of the theory was
also soon developed.
4
Fig. 1. Schematic phase diagram near a Quantum Critical Point. The parameter
along the x-axis can be quite general, like the pressure or a ratio of coupling constants. Whenever the critical temperature vanishes, a QCP, indicated with a dot
in the figure, is encountered. In the vicinity of such a point quantum mechanical,
zero-point fluctuations become very important. However when Tc is finite, critical
slowing down implies that the relevant frequency scale goes as |Tc T |z ,
dwarfing quantum effects; the standard classical critical methodology then applies.
An example of a phase diagram of this type for M nSi is shown in Fig. 34 below.
ature properties may be discussed in the language of FLT has led in turn
to the formulation and solution of impurity models with singular low energy
properties. Such models have a QCP for a particular relation between the
coupling constants; in some examples they exhibit a quantum critical line.
The thermodynamic and transport properties around such critical points or
lines are those of local singular Fermi-liquids. Although the direct experimental relevance of such models (as of one-dimensional models) to experiments
is often questionable, these models, being soluble, can be quite instructive in
helping to understand the conditions necessary for the breakdown of FLT and
associated quasiparticle concepts. The knowledge from zero-dimensional and
one-dimensional problems must nevertheless be used with care.
A problem which we do not discuss but which belongs in the study of SFLs
is the Quantum Hall Eect problem. The massive degeneracies of two-dimensional electrons in a magnetic-eld leads to spectacular new properties and
involves new fractional quantum numbers. The essentials of this problem were
solved following Laughlins inspired variational calculation. The principal reason for the omission is rstly that excellent papers reviewing the developments
are available [199,71,107] and secondly that the methodology used in this problem is in general distinct from those for discussing the other SFLs which have
a certain unity. We will however have occasions to refer to aspects of the quantum Hall eect problem often. Especially interesting from our point of view is
the weakly singular Fermi-liquid behavior predicted in the = 1/2 Quantum
Hall Eect [113].
6
One of the aspects that we want to bring to the foreground in this review is the
fact that SFLs all have in common some fundamental features which can be
stated usefully in several dierent ways. (i) they have degenerate ground states
to within an energy of order kB T . This degeneracy is not due to static external
potentials or constraints as in, for example the spin-glass problem, but degeneracies which are dynamically generated.(ii) Such degeneracies inevitably lead
to a breakdown of perturbative calculations because they generate infra-red
singularities in the correlation functions. (iii) If a bare particle or hole is added
to the system, it is attended by a divergent number of low energy particlehole pairs, so that the one-to-one correspondence between the one-particle
excitation of the interacting problem and those of the non-interacting problem, which is the basis for FLT, breaks down.
On the theoretical side, one may now view Fermi-liquid theory as a forerunner
of the Renormalisation Group ideas. The renormalisation group has led to a
sophisticated understanding of singularities in the collective behavior of manyparticle systems. Therefore these methods have an important role to play in
understanding the breakdown of FLT.
The aim of this paper is to provide a pedagogical introduction to SFLs, focused on the essential conceptual ideas and on issues which are settled and
which can be expected to survive future developments. Therefore, we will not
attempt to give an exhaustive review of the literature on this problem or of all
the experimental systems which show hints of SFL behavior. The experimental
examples we discuss have been selected to illustrate both what is essentially
understood and what is not understood even in principle. On the theoretical
side, we will shy away from presenting in depth the sophisticated methods
necessary for a detailed evaluation of correlation functions near QCP for
this we refer to the book by Sachdev [210] or for an exact solution of local impurity models (see, e.g., [118,212,244]). Likewise, for a discussion of the
application of quantum critical scaling ideas to Josephson arrays or quantum
Hall eects, we refer to the nice introduction by Sondhi et al. [232].
Landaus Fermi-liquid
The low energy and long wavelength excitations, as well as the correlation
and the response functions of the impossible problem bear a one-to-one correspondence with the simpler problem in their analytic properties. This leaves
xing only numerical values. These are to be determined by parameters, the
minimum number of which is xed by the symmetries. Experiments often provide intuition as to what the right simple problem may be: for the interacting
electrons, in the metallic range of densities, it is the problem of kinetic energy of particles with Fermi statistics (If one had started with the opposite
limit, just the potential energy alone, the starting ground state is the Wigner
crystal a bad place to start thinking about a metal!). If we start with
non-interacting fermions, and then turn on the interactions, the qualitative
behavior of the system does not change as long as the system does not go
through (or is close to) a phase transition. Because of the analyticity, we can
even consider strongly interacting systems the low energy excitations in
these have strongly renormalized values of their parameters compared to the
non-interacting problem, but their qualitative behavior is the same of that of
the simpler problem.
The heavy fermion problem provides an extreme example of the domain of
validity of the Landau approach. This is illustrated in Fig. 2, which shows the
specic heat of the heavy fermion compound CeAl3 . As in the Sommerfeld
model, the specic heat is linear in the temperature at low T , but if we write
Cv T at low temperatures, the value of is about a thousand times as
large as one would estimate from the density of states of a typical metal, using
the free electron mass. For a Fermi gas the density of states N(0) at the Fermi
energy is proportional to an eective mass m :
N(0) =
m kF
,
2h
2
(1)
with kF the Fermi wavenumber. Then the fact that the density of states at
the chemical potential is a thousand times larger than for normal metals can
be expressed by the statement that the eective mass m of the quasiparticles
is a thousand times larger than the free electron mass m. Likewise, as Fig. 3
shows, the resistivity of CeAl3 at low temperatures increases as T 2 . This also
is a characteristic sign of a Fermi-liquid, in which the quasiparticle lifetime
at the Fermi surface, determined by electron-electron interactions, behaves as
1/T 2. 2 However, just as the prefactor of the specic heat is a factor
thousand times larger than usual, the prefactor of the T 2 term in the resistivity
is a factor 106 larger while scales linearly with the eective mass ratio
m /m, the prefactor of the T 2 term in the resistivity increases for this class
2
Fig. 2. Specific heat of CeAl3 at low temperatures from Andres et al. [28]. The
slope of the linear specific heat is about 3000 times that of the linear specific heat
of say Cu. But the high-temperature cut-off of this linear term is smaller than that
of Cu by a similar amount. The rise of the specific heat in a magnetic field at low
temperatures is the nuclear contribution, irrelevant to our discussion.
Fig. 3. Electrical resistivity of CeAl3 below 100 mK, plotted against T 2 . From
Andres et al. [28].
of Fermi-liquids as (m /m)2 .
It should be remarked that the right simple problem is not always easy to
guess. The right simple problem for liquid 4 He is not the non-interacting
Bose gas but the weakly interacting Bose gas (i.e. the Bogoliubov problem
[44,146]). The right simple problem for the Kondo problem (a low-temperature
local Fermi liquid) was guessed [184] only after the numerical renormalization
group solution was obtained by Wilson [273]. The right simple problem for
two-dimensional interacting disordered electrons in the metallic range of
densities (section 8 in this paper) is at present unknown.
For SFLs the problem is dierent: usually one is in a regime of parameters
where no simple problem is a starting point in some cases the uctua10
Fig. 4. Bare-particle distribution at T = 0 for a given spin-direction in a translationally invariant Fermi-system with interactions (full line) and without interactions
(dashed line). Note that the position of the discontinuity, i.e. the Fermi wavenumber
kF , is not renormalized by interactions.
h
2k2
2m
(2)
moving a particle excitation below the Fermi energy means that we add a hole
excitation). As sketched in Fig. 5(a), for the non-interacting system, A0 (k, )
is simply a delta-function peak at the energy k , because all momentum states
are also energy eigenstates:
A0 (k, ) = ( (k )) , for > ,
1
1
1
= ImG0 (k, ) .
= Im
(k ) + i
(3)
(4)
Here is small and positive; it reects that particles or holes are introduced
adiabatically, and is taken to zero at the end of the calculation for the pure
non-interacting problem. The rst step of the second line is just a simple
mathematical rewriting of the delta function; in the second line the Greens
function G0 for non-interacting electrons is introduced. More generally the
single-particle Greens function G(k, ) is dened in terms of the correlation
function of particle creation and annihilation operators in standard textbooks
[182,4,208,159]. For our present purposes, it is sucient to note that it is related to the spectral function A(k, ), which has a clear physical meaning and
which can be deduced through Angle Resolved Photoemission Experiments :
G(k, ) =
dx
A(k, x)
.
x + i sgn( )
(5)
1
(k, ) .
(6)
(7)
With these preliminaries out of the way, let us consider the form of A(k, )
when we add a particle to an interacting system of fermions.
Due to the interaction (assumed repulsive) between the added particle and
those already in the Fermi-sea, the added particle will kick particles from below
the Fermi-surface to above. The possible terms in a perturbative description
of this process are constrained by the conservation laws of charge, particle
number, momentum and spin. Those which are allowed by the aforementioned
conservation laws are indicated pictorially in Fig. 6, and lead to an expression
of the type
1
1/2
N +1
|k
i = Zk ck | N i +
3/2
k1 ,k2 ,k3 1 ,2 ,3
k,k1 k2 +k3 (; 1 , 2 , 3 )| N i + .
(8)
Here the ck s and ck s are the bare particle creation and annihilation operators,
and the dots indicate higher order terms, for which two or more particlehole pairs are created and (; 1 , 2 , 3 ) expresses conservation of spin under
vector addition. The multiple-particle-hole pairs for a xed total momentum
can be created with a continuum of momentums of the individual bare particles
and holes. Therefore an added particle with xed total momentum has a wide
distribution of energies. However if Zk dened by Eqn. (8) is nite, there will
be a well-dened feature in this distribution at some energy which is in general
dierent from the non-interacting value h
2 k 2 /(2m). The spectral function in
such a case will then be as illustrated in Fig. 5. It is useful to separate the
13
well-dened feature from the broad continuum by writing the spectral function
as the sum of two terms, A(k, ) = Acoh (k, ) + Aincoh (k, ). The singleparticle Greens function can be expressed as a sum of two corresponding
terms, G(k, ) = Gcoh (k, ) + Gincoh (k, ). Then
Gcoh (k, ) =
Zk
,
k + i/k
(9)
which for large lifetimes k gives a Lorentzian peak in the spectral density
at the quasiparticle energy k k . The incoherent Greens function is
smooth and hence for large k corresponds to the smooth background in the
spectral density.
The condition for the occurrence of the well-dened feature can be expressed
as the condition on the self-energy (k, ) that it has an analytic expansion
about = 0 and k = kF and that its real part be much larger than its
imaginary part. One can easily see that were it not so the expression (9)
for Gcoh could not be obtained. These conditions are necessary for a Landau
Fermi-liquid. Upon expanding (k, ) in (12) for small and small deviations
of k from kF and writing it in the form (9), we make the identications
1
= Zk Im(kF , = 0) ,
k
k = k Zk Zk ,
(10)
where
Zk = 1
!1
1
Z = 1 +
vF k
=0,k=kF
(11)
=0,k=kF
Zk
= hkN +1 |ck | N i .
(12)
i.e. charge, spin and momentum and which may be dened simply by the
creation operator k,
:
|kN +1 i = k | N i .
(13)
(14)
as follows:
G = G0 +
1 X
1 X
fkk , nk nk + (15)
(k )nk +
V k,
2V 2 kk ,
15
kk kF
X
l=0
k
) .
Fls,aPl (k
(16)
From the expression (15) one can then relate the lowest order so-called Landau
coecients F0 and F1s and the eective mass m to thermodynamic quantities
like the specic heat Cv , the compressibility , and the susceptibility :
Cv
m
=
,
Cv0
m
= (1 + F0s )
,
0
m
= (1 + F0a )
.
0
m
(17)
Here subscripts 0 refer to the quantities of the non-interacting reference system, and m is the mass of the fermions. For a Galilean invariant system (like
3
He), there is a a simple relation between the mass enhancement and the
Landau parameter F1s , and there is no renormalization of the particle current
j; however, there is a renormalization of the velocity: one has
j = k/m ,
v = k/m ,
m
Fs
= 1+ 1
m
3
(18)
contains more than one type of particle with dierent characteristic frequency
scales.
X
k
fk+q fk
.
(Ek+q Ek ) + i
17
(19)
(a)
p1
p1 + q
k+q
(b)
p2
p1
p2 + q
p2
p1 p2
+k + q
p1 + q
(c)
p1
p2 + q
p2
p1 + k
p1 + q
p2
p2 + q
Fig. 7. The three second-order processes in a perturbative calculation of the correction to the bare interaction in a Fermi-liquid
q vk
df
.
q vk + i dk
(20)
qvF
(21)
qv
F
Re(q, ) = g N(0) 1 +
ln
qvF
+ qvF
2
(22)
which gives a constant (leading to a nite compressibility and spin susceptibility) at small compared to qvF . For diagram 7(b), we get a term
This behavior implies that this scattering contribution is a marginal term in the
renormalization group sense, which means that it affects the numerical factors, but
not the qualitative behavior.
18
(a)
(b)
Im
on
sm
pla
(q; !)
allowed
parti
le-hole
ex
itations
2kF
qvF
(Ep1 p2 +k+q Ek ) in the denominator. This term is always nite for general
momenta p1 and p2 , and hence the contribution from this diagram can always
be neglected relative to the one from 7(a). Along similar lines, one nds that
diagram 7(c), which describes scattering in the particle-particle channel, is
irrelevant except when p1 = p2 , when it diverges as ln .
Of course, this scattering process is the one which gives superconductivity.
Landau noticed this singularity but ignored its implication 5 . Indeed, as long as
the eective interactions do not favor superconductivity or as long as we are at
temperatures much higher than the superconducting transition temperature,
it is not important for Fermi-liquid theory.
Let us now look further at the absorptive spectrum of particle-hole excitations
in two and three dimensions, i.e., we examine the imaginary part of Eq. (19).
When the total energy of the pair is small, both the particle and the hole have
to live close to the Fermi surface. In this limit, we can make any excitation with
momentum q 2kF . For xed but small values of q, the maximum excitation
energy is qvF ; this occurs when q is in the same direction as the main
momentum k of each quasiparticle. For q near 2kF the maximum possible
energy is = vF |q 2kF |. Combining these results, we get the sketch in Fig.
8(a), in which the shaded area in the -q space is the region of allowed particlehole excitations 6 . From this spectrum one can calculate the polarizability, or
the magnetic susceptibility.
The behavior sketched above is valid generally in two and three dimensions
(but as we will see in section 4, not in one dimension). The important point
5
Attractive interactions in any angular momentum channel (leading to superconductivity) are therefore relevant operators.
6 In the presence of long-range Coulomb interactions one gets in addition to the
particle-hole excitation spectrum a collective mode with a finite plasma frequency
as q 0 in d = 3 and a q behavior in d = 2.
19
= g 2 N(0)
(k, )
EF
2
(23)
(24)
Thus in a perturbative calculation of the eect of interactions the basic analytic structure of the Greens function is left the same as for non-interacting
fermions. The general proof of the validity of Landau theory consists in showing that what we have seen in second order in g remains true to all orders in g.
7
This quantity has been precisely evaluated by Galitski [100] for the model of a
dilute Fermi gas characterized by a scattering length.
20
The original proofs [4] are self-consistency arguments we will consider them
briey in section 2.4. They assume a nite Z in exact single-particle Greens
functions and show eectively that to any order of perturbation, the polarizability functions retain the analytic structure of the non-interacting theory,
which in turn ensures a nite Z.
In one dimension, phase space restrictions on the possible excitations are crucially dierent 8 , since the Fermi surface consists of just two points in the
one-dimensional space of momenta see Fig. 10. As a result, whereas in
d = 2 and d = 3 a continuum of low-energy excitations with nite q is possible, at low energy only excitations with small k or k 2kF are possible. The
subsequent equivalence of Fig. 8 for the one-dimensional case is the one shown
in Fig. 11. Upon integrating over the momentum k with a cut-o of O(kF )
the contribution from this particle-hole scattering channel to Re (q, ) is
Z
dk
1
ln[( + qvF )/EF ] .
+ (2k + q)vF
(25)
(Note that (25) is true for both q << kF and |q 2kF | << kF .) This in turn
leads to a single particle self-energy calculated by the process in Fig. 9 to
be Re (kF , ) ln giving a hint of trouble. Also the Cooper (particleparticle) channel has the same phase space restrictions, and gives a contribution also proportional to ln . The fact that these sngular contributions are of
the same order, leads to a competition between charge/spin uctuations and
Cooper pairing uctuations, and to power law singularities. Also, the fact that
instead of a continuum of low energy excitations as in higher dimensions, the
width of the band of allowed particle hole excitations vanishes as 0, is
the reason that the properties of one-dimensional interacting metals can be
understood in terms of bosonic modes. We will present a brief summary of
the results for the single-particle Greens function and correlation functions in
section 4.9.
In special cases of nesting in two or three dimensions, one can have situations
that resemble the above one-dimensional case. When the non-interacting Fermi
surface in a tight binding model has the square shape sketched in Fig. 12(a)
this happens for a tight-binding model with nearest neighbor hopping on
a square lattice at half-lling a continuous range of momenta on opposite
sides of the Fermi surface can be transformed into each other by one and the
same wavenumber. This so-called nesting leads to log and log2 singularities
for a continuous range of k in the perturbation theory for the self energy
(k, ). Likewise, the partially nested Fermi surface of Fig. 12(b) leads to
charge density wave and antiferromagnetic instabilities. We will come back to
8
It might appear surprising that they are not different in any essential way between
higher dimensions.
21
Fig. 11. Phase space for particle-hole excitation spectrum in one dimension. Compare with the same in higher dimensions, Fig. 8. For linearized single-particle kinetic
energy k = vF (k kF ), particle-hole excitations are only possible on lines going
through k = 0 and k = 2kF .
Fig. 12. (a) The nested Fermi surface on obtained in a tight binding model on a
square lattice with nearest neighbor hopping; (b) A partially nested Fermi surface
which leads to charge density wave or antiferromagnetic instabilities
The basic idea is that due to kinematic constraints, any perturbative process
with n particle-hole pairs in the intermediate state gives contributions to the
polarizability proportional to (/EF )n . Therefore the low energy properties
can be calculated with processes with the same skeletal structure as those
in Fig. 7, which have only one particle-hole pair in the intermediate state. So
one may concentrate on the modication of the four-legged vertices and the
single-particle propagators due to interactions to all orders. Accordingly the
theory is formulated in terms of the single particle Greens function G(p) and
the two-body scattering vertex,
(p1 , p2 , p1 + k, p2 k) = (p1 , p2 , k)
(26)
Here and below we use, for sake of brevity, p, etc. to denote the energymomentum four-vector (p, ) and we suppress the spin labels. The equation
23
d4 q
, (27)
(2)4
where (1) is the irreducible part in the particle hole channel in which Eq.
(27) is expressed. In other words, (1) can not be split up into two parts by
cutting two Greens function lines with total momentum k. So (1) includes
the complete vertex in the other (often called cross-) particle-hole channel.
The diagrammatic representation of Eq. (27) is shown in Fig. 13. In the simplest approximation (1) is just the bare two-body interaction. Landau theory
assumes that (1) has no singularities 10 An assumption is now further made
that G(p) does have a coherent quasiparticle part at |p| pF and 0,
G(p) =
Z
+ Ginc ,
p + i sgn(p )
(28)
2iz 2
vq k
()(|q| pF ) + (q) ,
vF vq k
(29)
for small k and , and where and ( + ) are frequencies of the two Greens
functions. Note the crucial role of kinematics in the form of the rst term which
comes from the product of the quasiparticle parts of G; (q) comes from the
scattering of the incoherent part with itself and with the coherent part and is
assumed smooth and featureless (as it is indeed, given that Ginc is smooth and
featureless and the scattering does not produce an infrared singularity at least
perturbatively in the interaction). The vertex in regions close to k kF and
0 is therefore dominated by the rst term. The derivation of Fermi-liquid
theory consists in proving that the equations (27) for the vertex and (28) for
the Greens function are mutually consistent.
9
To second order in the interactions the correction to the vertex in the two possible
particle-hole channels has been exhibited in the first two parts of Fig. 7.
10 The theory has been generalized for Coulomb interactions [194,184,4].The general
results remain unchanged because a screened short-range interaction takes the place
of (1) .This is unlikely to be true in the critical region of a metal-insulator transition,
because on the insulating side the Coulomb interaction is unscreened.
24
d4 q
.
(2)4
(30)
k0
= (p1 , p2 ) .
(31)
This is now used to write the equation for the complete vertex in terms of
:
(p1 , p2 , k) = (p1 , p2 )
Z
Z 2 p2F
vq k
+
(p1 , q)(q, p2, k)
dq ,
3
(2) vF
vq k
(32)
where in the above |q| = pF and one integrates only over the solid angle q .
Given a non-singular , a non-singular is produced (unless the denominator
in Eq. (32) produces singularities after the indicated integration-the LandauPomeranchuk singularities discussed below). The one-particle Greens function
G can be expressed exactly in terms of see Fig. 14. This leads to Eq. (28)
proving the self-consistency of the ansatz with a nite quasiparticle weight Z.
The quantity Z 2 is then a smooth function and goes into the determination
of the Landau parameters.
The Landau parameters can be written in terms of the forward scattering amplitude. In eect they parametrize the momentum and frequency independent
scattering of the incoherent parts among themselves and with the coherent
parts so that the end result of the theory is that the physical properties can
be expressed purely in terms of the quasiparticle part of the single-particle
Greens function and the Landau parameters. No reference to the incoherent
parts needs to be made for low energy properties. For single-component translational invariant fermions (like liquid 3 He) even the quasiparticle amplitude
Z disappears from all physical properties. This last is not true for renormalization due to electron-phonon interactions and in multi-component systems
25
Fig. 14. Diagram for the exact single-particle self-energy in terms of the exact vertex
and the exact single-particle Greens function.
(p, ) ,
0, q 0 (p; q) =
lim
q
m p
!
(p, )
lim
0, q 0 4 (p; q) = 1 +
,
q
p
d
q
0, 0 (p; q) =
(p, ) ,
lim
m dp
d(p, )
q
0, 0 4 (p; q) = 1 +
.
lim
( = 1, 2, 3) (33)
(34)
( = 1, 2, 3) (35)
(36)
A relation analogous to (34) is derived for elds coupling to spin for the case
that interactions conserve spin. The total derivative in (35) and (36) [rather
than the partial derivative in (33) and (34)] represents that d/d is the
variation in when is changed to + d together with to + d, and
d/dp represents the variation when the momentum p as well as the Fermisurface is translated by dp .
26
Eq. (36) is an expression of energy conservation, and Eq. (34) of particle number conservation. Eqs. (33) and (34) together signify the continuity equation.
Eq. (35) represents current conservation 11 .
In Landau theory, the right hand sides in Eqs. (33)-(36) are expressible in
terms of the Landau parameters. These relations are necessary to derive the
renormalization in the various thermodynamic quantities that we quoted in
Eqs. (17) and (18) as well as the Landau transport equation. Needless to say,
any theory of SFL must also be consistent with the Ward identities.
The Ward identity Eq. (35) does not hold for impure system where the Fermisurface cannot be defined in momentum space. Since energy is conserved, a Fermi
surface can still be defined in energy space, and hence the other Ward identities
continue to hold. This point is further discussed in section 7.
27
(p1 , p2 , k)
k
kF
!d1
f (p1 , p2 , k) ,
(37)
(38)
In our discussion Hsimple was the non-interacting Hamiltonian. The non-interacting Hamiltonian is actually a member of a line of xed-point Hamiltonians
H all of which have the same symmetries but dier in their Landau parameters Fls,a etc. The Fl s, obtained from the forward scattering in Landau theory
are associated with marginal operators and distinguish the properties of the
various systems associated with the line of xed points. Landau fermi-liquid
theory is rst of all the statement of the domain of attraction of this line of
xed points. The theory also establishes the universal low temperature properties due to the irrelevant operators generated by Hrest due to scattering
in channels other than the forward channel. Landau theory does not establish (at least completely) the domain of attraction of the critical surface
bounding the domain of attraction of the Fermi-liquid xed line from those of
other xed points or lines. If Hrest were to generate a relevant operator
i.e., eective interactions which diverge at low energies (temperatures) the
scheme breaks down. For example, attractive interactions between fermions
generate relevant operatorsthey presage a transition to superconductivity, a
state of dierent symmetry. But if we stay suciently above Tc , we can usually
continue using Landau theory 12 .
12
28
From Landaus phenomenological theory, one can only say that the theory
breaks down when the physical properties specic heat divided by temperature, 13 compressibility, or the magnetic susceptibility diverge or when the
collective modes representing oscillations of the Fermi-surface in any harmonic
and singlet or triplet spin combinations become unstable. The latter, called
the Landau-Pomeranchuk singularities, are indeed one route to the breakdown
of Landau theory and occur when the Landau parameters Fls,a reach the critical value (2l + 1). A phase transition to a state of lower symmetry in then
indicated. The new phase can again be described in Landau theory by dening
distribution functions consistent with the symmetry of the new ground state.
The discussion following Eq. (8) in section 2.2 allows us to make a more general
statement. Landau theory breaks down when the quasiparticle amplitude Zk
becomes zero; i.e. when the state ck | N i and |kN +1 i are orthogonal. This can
happen if the series expansion in Eq. (8) in terms of the number of particlehole pairs is divergent. In other words, addition of a particle or a hole to the
system creates a divergent number of particle-hole pairs in the system so that
the leading term does not have a nite weight in the thermodynamic limit.
From Eq. (11) connecting the Z s to s, this requires that the single-particle
self-energy be singular as a function of at k kF . This in turn means that
the Greens functions of SFLs contain branch cuts rather than the poles unlike
Landau Fermi-liquids. The weakest singularity of this kind is encountered in
the borderline marginal Fermi-liquids where 14
c
(kF , ) ln
+ i|| .
(39)
faster than linear in , the quasiparticle energy. Thus is the first power for
which this is not true. The ln(c /) term in Eqn. (39) is then dictated by the
Kramers-Kroning relation.
29
the interaction vertices are actually more divergent than the single-particle
self-energy.
Yet another route to SFLs is the case in which the interactions generate
new quantum numbers which are not descriptive of the non-interacting problem. This happens most famously in the Quantum Hall problems and in onedimensional problems (section 4) as well as problems of impurity scattering
with special symmetries (section 3). In such cases the new quantum numbers
characterize new low-energy topological excitations. New quantum numbers
of course imply Z = 0, but does Z = 0 imply new quantum numbers. One
might wish to conjecture that this is so. But there is no proof of this 15 .
In the nal analysis all breakdowns of Landau theory are due to degeneracies
leading to singular low-energy uctuations. If the characteristic energy of the
uctuations is lower than the temperature, a quasi-classical statistical mechanical problem results. On the basis of our qualitative discussion in section
2.3 and the sketch of the microscopic derivation in section 2.4, we may divide
up the various routes to breakdown of Landau theory into the following (not
necessarily orthogonal) classes:
(i) Landau-Pomeranchuk Singularities: Landau theory points to the possibility
of its breakdown through the instability of the collective modes of the Fermisurface which arise from the solution of the homogeneous part of Eqn.(32).
These collective modes can be characterized by the angular momentum of
oscillation of the Fermi-surface and whether the oscillation is symmetric s
or anti-symmetric a in spin. The condition for the instability derived from
the condition of zero frequency of the collective modes are [194,36]
Fs (2 + 1) ,
Fa (2 + 1) .
(40)
30
stabilities and the critical properties near such instabilities have not been well
investigated, especially for fermions with a lattice potential.
It is also worth noting that some of the instabilities are disallowed in the
limit of translational invariance. Thus, for example, time-reversal breaking
states, such as the anyon-state [148,59] cannot be realized because in a
translationally invariant problem the current operator cannot be renormalized
by the interactions, as we have learnt from Eqs. (18), (33).
(ii) Critical regions of Large Q-Singularities: Landau theory concerns itself
only with long wavelength response and correlations. A Fermi-liquid may
have instabilities at a nonzero wave-vector, for example a charge-density wave
(CDW) or spin-density wave (SDW) instability. Only a microscopic calculation
can provide the conditions for such instabilities and therefore such conditions
can only be approximately derived. An important point to note is that they
arise perturbatively from repeated scattering between the quasiparticle parts
of G while the scattering vertices (irreducible interactions)are regular. The superconductive instability for any angular momentum is also an instability of
this kind. In general such instabilities are easily seen in RPA and/or t-matrix
calculations.
Singular Fermi-liquid behavior is generally expected to occur only in the critical regime of such instabilities [112,156]. If the transition temperature Tc is
nite then there is usually a stable low temperature phase in which unstable
modes are condensed to an order parameter, translational symmetry is broken, and gaps arise in part or all of the Fermi-surface. For excitations on the
surviving part of the Fermi-surface, Fermi-liquid theory is usually again valid.
The uctuations in the critical regime are classical, i.e. with characteristics
frequency f kB Tc .
If the transition is tuned by some external parameter so that it occurs at
zero temperature, one obtains, as illustrated already in Fig. 1, a Quantum
Critical Point (QCP). If the transition is approached at T = 0 as a function
of the external parameter, the uctuations are quantum-mechanical, while if
it is approached as a function of temperature for the external parameter at its
critical value, the uctuations have a characteristic energy proportional to the
temperature. A large region of the phase diagram near QCPs often carries
SFL properties. We shall discuss such phenomena in detail in section 6.
(iii) Special Symmetries: The Cooper instability at q = 0, Fig. 7(c), is due to
the nesting of the Fermi-surface in the particle-particle channel. Usually indications of nite q-CDW or SDW singularities are evident pertubatively from
Fig. 7(a) or Fig. 7(b) for special Fermi-surfaces, nested in some q-direction
in particle-hole channels. One-dimensional fermions are perfectly nested in
both particle-hole channels and particle-particle channels (Figs. 7(a)-(c)) and
31
Fig. 16. A model band structure for a solid with more than one atom per unit
cell. Actually in CaB6 where excitonic singularities have been invoked to produce
a ferromagnetic state [280] there are three equivalent points in the Brillouin zone
where the conduction band minima and the valence band maxima occur.
the excitonic resonance and will in general be singular if the resonance is near
zero-energy.
Such singularities require interactions above a critical magnitude and are
physically and mathematically of an unfamiliar nature. In a 2-band onedimensional model, exact numerical calculations have established the importance of such singularities [237,234].
Recently it has been found that CaB6 or SrB6 with low density of trivalent
Eu or quadrivalent Ce ions substituting for (Ca, Sr) is a ferromagnet [280].
The most plausible explanation [288,33,35] is that this is a realization of the
excitonic ferromagnetism predicted by Volkov et al. [263]. The instability to
such a state occurs because the energy to create a hole in the valence band and
a particle in the conduction band above the fermi-energy goes to zero if the
attractive particle-hole (interband)interactions are large enough. This problem
has been investigated only in the mean-eld approximation. Fluctuations in
the critical regime of such a transition are well worth studying.
Excitonically induced singularities in the irreducible interactions are also responsible for the Marginal Fermi-liquid state of CuO metals in a theory to
be discussed in section 7.
The Kondo problem is at the same time one of the simplest and one of the
most subtle examples of the eects of strong correlation eects in electronic
systems. The experiments concern metals with a dilute concentration of magnetic impurities. In the Kondo model one considers only a single impurity; the
Hamiltonian then is
H=t
X
hiji
ci cj + JS c0 c0 ,
34
(41)
For the reader unfamiliar with reading a renormalization group diagram like
that of Fig. 17(b) or 18, the following explanation might be helpful. The flow in a
renormalization group diagram signifies the following. The original problem, with
bare parameters, corresponds to the starting point in the parameter space in which
we plot the flow. Then we imagine integrating out the high energy scales (e.g.
virtual excitations to high energy states); effectively, this means that we consider
the system at lower energy (and temperature) scales by generating effective Hamiltonians with new parameters so that the low energy properties remain invariant.
The length along the flow direction is essentially a measure of how many energy
scales have been integrated out typically, as in the Kondo problem, this decrease
is logarithmic along the trajectory. Thus, the regions towards which the flow points
signify the effective parameters of the model at lower and lower temperatures. Fixed
points towards which all trajectories flow in a neighborhood describe the universal
low temperature asymptotic behavior of the class of models to which the model
under consideration belongs. When a fixed point of the flow is unstable, it means
that a model whose bare parameters initially lie close to it flows away from this
point towards a stable fixed point; hence it has a low-temperature behavior which
does not correspond to the model described by the unstable fixed point. A fixed line
usually corresponds with a class of models which have some asymptotic behavior,
e.g. an exponent, which varies continuously.
17 A particularly lucid discussion of the renormalization procedure may be found
in [143]. Briefly, the procedure consists in generating a sequence of Hamiltonians
with successively lower energy cut-offs that reproduce the low-energy spectrum. All
terms allowed by symmetry besides those in the bare Hamiltonian are kept. The
coeffcients of these terms scale with the cut-off. Those that decrease proportionately
to the cut-off or change only logarithmically, are coefficients of marginal operators,
those that grow/decrease (algebraically) of relevant/irrelevant operators. Marginal
operators are marginally relevant or marginally irrelevant. Upon renormalization the
35
The interaction between conduction electrons and the localized electronic level
is not a direct spin interaction. It originates from quantum-mechanical charge
uctuations that (through the Pauli principle) depend on the relative spin orientation. To see this explicitly it is more instructive to consider the Anderson
model [19] in which
H=t
X
hiji
ci cj + d
(42)
k,
The last term in this Hamiltonian is the hybridization between the localized
impurity state and the conduction electrons, in which spin is conserved. In the
particle-hole symmetric case, d = U/2 is the one-hole state on the impurity
site in the Hartree-Fock approximation and the one-particle state has the
energy U/2.
Following a perturbative treatment in the limit Vt , VU 1 the Anderson model
reduces to the Kondo Hamiltonian with an eective exchange constant Jef f
(V 2 /t)2 /U.
The Anderson model has two simple limits, which are illustrated in Fig. 17:
(i) V = 0: This describes a local moment with Curie susceptibility 2B /T .
This limit is the correct point of departure for an investigation for the high
temperature regime. As noted one soon encounters the Kondo divergences.
(ii) U = 0: In this limit the impurity forms a resonance of width V 2 /t
at the chemical potential which in the particle-hole symmetric case is halfoccupied. The ground state is a spin singlet. This limit is the correct starting
point for an examination of the low temperature properties (T TK ). A
temperature independent contribution to the susceptibility and a linear contribution to the specic heat ( N(0)T /) are contributed by the resonant
state. Hence the name local Fermi-Liquid.
The conceptually hard part of the problem was to realize that (ii) is the correct stable low temperature xed point and the technically hard problem is to
derive the passage from the high-temperature regime to the low-temperature
regime. This was rst done correctly by Wilson [273] through the invention of
the Numerical Renormalization Group (and almost correctly by Anderson and
flow is to the strong coupling J = fixed point, see Fig. 17. The terms generated
from t 6= 0 serve as irrelevant operators at this fixed point; this means that they do
not affect the ground state but determine the measurable low-energy properties.
36
Fig. 17. (a) Hartree-Fock excitation spectrum of the Anderson model in the two
limits of zero hybridization, V = 0 and zero interaction, U = 0. (b) Renormalization
group flow of the Kondo problem.
Yuval [22,23] by analytic methods). The analysis showed that under Renormalization Group scaling transformations the ratio (J/t) increases monotonically
as illustrated in Fig. 17(b) continuous RG ows are observed from the high
temperature extreme (i) to the low temperature extreme (ii) and a smooth
crossover between between the two regimes occurs at the Kondo temperature
TK t exp(t/2J).
(43)
Cv = T fc (T /TK ),
(44)
/
Cv /Cv
37
(45)
(, )n ( ) + ... .
(46)
About a stable xed point the energy dependence is analytic near the chemical
potential ( = 0), so that we may expand
(, ) = + ... .
0 () = 0 + + ... ,
(47)
Just as the Landau parameters are expressed in terms of symmetric and antisymmetric parts, we can write
= = s + a ,
, = = s a .
(48)
(49)
= s = /N(0) .
(50)
Thirdly, one may borrow from Wilsons solution that the xed point has 0 =
/2. This expresses that the tightly bound spin singlet state formed of the
impurity spin and conduction electron spin completely blocks the impurity
site to other conduction electrons; this in turn implies maximal scattering
and phase shift of /2 for the eective scattering potential [184]. In other
words, it is a strong-coupling xed-point where one conduction electron state
is pushed below the chemical potential in the vicinity of the impurity to form
a singlet resonance with the impurity spin. One may now calculate all physical
properties in term of . In particular, one nds Cv /Cv = 2/(V N(0) and a
similar expression for the enhancement of ,such that the Wilson ratio of 2.
H =t
hiji,
ci cj +
Xh
k,k
J S + ck ck + S ck ck +
i
(51)
(J+ = J ) .
(52)
Fig. 18. Renormalization group flows for the Kondo problem, displaying the line of
critical points for the ferromagnetic problem and the flow towards the fixed point
= J for the antiferromagnetic problem.
J
z
modynamic limit: The single particle wave functions are not orthogonal. It
is only the overlap between the ground state formed by their Slater determinants 18 which vanishes as N tends to innity.
More quantitatively, if the injection of the additional particle produces an
s-wave phase shift 0 for the single particle wave functions (all N of them),
(kr) =
sin kr
sin(kr + 0 )
kr
kr
(53)
hN |N
i N 0 / .
(54)
Here |N i is the determinant Fermi sea wave function for N particles and
|N
i is the wavefunction of the system after undergoing a phase shift by the
local perturbation produced by the injected electron 19 .
Quite generally such an orthogonality (Z = 0) arises also if two N particle
states of a system possess dierent quantum numbers and almost the same
energy. These new quantum numbers might be associated with novel topological excitations. This is indeed the case in the Quantum Hall Liquid where
new quantum numbers are associated with fractional charge excitations. The
SFL properties of the interacting one-dimensional fermions (Section 4) may
also be looked on as due to orthogonality. Often orthogonality has the eect of
making a quantum many-body problem approach the behavior of a classical
problem. This will be one of the leitmotifs in this review. We turn rst to a
problem where this orthogonality is well understood to lead to experimental
consequences, although not at low energies.
The results hold true also for interacting fermions, at least when Fermi-liquid
description is valid for both of the states
19 Through the Friedel sum rule / has a physical meaning; it is the charge that
0
needs to be transported from infinity to the vicinity of the impurity in order to
screen the local potential [78].
41
Fig. 19. Absorption lineshape for transitions between a lower dispersionless level
and a conduction band, (a) for zero interaction between the conduction electrons
and local level, and (b) for small interaction.
D , as sketched in Fig. 19. In this case, a Fermi edge reecting the density
of unoccupied states in the conduction band is expected to be visible the
spectrum.
However, when a hole is generated in the lower level, the potential that the
conduction electrons see is dierent. The relevant Hamiltonian is now
H = d d d
X
1X
1
1
+
k ck ck +
V (k, k ) ck ck
2
L k,k
2
k
d d
1
2
, (55)
where spin indices have been suppressed. The operators (d, d) annihilate or
create holes in the core level, which is taken to be dispersionless. The rst two
terms in the Hamiltonian represent the unperturbed energies of the core hole
and the free electrons. The last term depicts the screened Coulomb interaction
between the conduction electrons and the hole in the core level.
As a consequence of the interactions, the line shape is quite dierent. This is
actually an exactly solvable problem [183]. There are two kinds of eects, (a)
excitonic the particle and the hole attract, leading to a shift of the edge
and a sharpening of the edge singularity and (b) an orthogonality eect of
the type just discussed above, which smoothens the edge irrespective of the
sign of the interaction. This changes the absorption spectrum to that of Fig.
19(b) in the presence of interactions. The form of the singularity is [158,183]
0
2
0
A() (
D )2 + 2 .
(56)
of the dispersion of the hole band. If the hole moves only in one dimension,
the singularity is not removed.
3.6 A Spinless Model with Finite Range Interactions
A model, which is a generalization of the Ferromagnetic Kondo problem and in
which the low-energy physics is dominated by the orthogonality catastrophe,
is given by the following Hamiltonian :
H=
X
k,l
t X
k k,l
k,l +
(k,0d + h.c.)
L k
1X
1
1
+
d d
.
Vl k,l k,l
L k,k
2
2
(57)
J , 0
t
,
2a
1
Sz ,
2
d d
(58)
produces
H=
X
1
Jz,l Sz sz,l .
k ck,,l ck,,l + J,0 (S s
l + h.c.) +
2
l
k,,l
X
(59)
akin to having spin up or spin down. Similarly, the kinetic hybridization term
transforms into a spin ip interaction term of the form (S d + h.c.). As Vl
couples to the occupancy of the impurity site,we might anticipate Jz to scale
with Vl . The additional correction (2vF ( 2 1)l,0) originates from the
subtle transformation taking the original fermionic system into an eective
spin model.
This problem has been solved by renormalization group methods, See Fig. 20.
But simple arguments based on the X-ray edge singularity, orthogonality and
recoil give the correct qualitative physics. When t = 0, the problem is that
of the X-ray edge Hamiltonian (with d = 0). When t is nite, the charge
at the impurity orbital uctuates (the impurity site alternately empties and
lls). This generates, in turn, a uctuating potential. The X-ray absorption
spectrum is the Fourier transform of the particle-hole pair correlator
() h (t)d(t)d (0)(0)i .
(60)
This quantity, which is none other than the hybridization correlation function,
should display the X-ray edge characteristics for large frequencies ( > ef f )
where the eect of recoil is unimportant:
() 0 (/W ) ; =
20 X l2
+
.
2
(61)
(62)
= 1/(1 ) ,
(63)
as W .
(64)
so that for
<1,
ef f 0
actual value of determines the singular properties at low energy or temperature. In the single channel problem such a scenario occurs if the potential V0
is suciently attractive. On mapping to the spin problem we nd that this region corresponds to the singular Fermi-liquid Ferromagnetic Kondo problem.
The scaling of the hybridization to zero corresponds, in the spinmodel, to
J 0. In a renormalization group language, the ows will impinge on the
line of xed points (J = 0, Jz < 0). In this regime, we recover, once again, a
continuous set of exponents.
If the number of channels is large enough, the orthogonality catastrophe associated with the change in the number of particles on the impurity site is
suciently strong to drive the hybridization to zero even for the case of repulsive interactions V or antiferromagnetic Jz,l .
In the singular regime various correlation functions may be evaluated [103].
For instance, the Greens function of the localized impurity
Gd ( ) = < T d( )d (0) > 0 e
(65)
(66)
H=
k,,l
k,k ,l
Vkkl
1
nd
2
X
"X
45
d ck0 + h.c.
ckl ck l
1 .
(67)
Fig. 20. The Renormalization flow diagram for the model with finite range interactions according to [103]. The initial values are Jx,0 = 0, J,0 = 0.5, and Jx,l varies
from 0 to 1 in increments of 0.05. When Jz,l becomes large enough (i.e. when V is
large enough) the flow veers from the usual Kondo fixed point to a zero hybridization
(J,0 = 0) singular Fermi liquid.
In this model both spin and charge may be altered on the impurity site. This
enhanced number of degrees of freedom implies that the states need to be
specied by more quantum numbers. This also allows, a priori, for higher
degeneracy. In the following N screening channels, all of equal strength V ,
will be assumed. In the U limit the spectrum of the impurity site may
be diagonalized. The two lowest states are
d
,
4
2
1/2
!2
N
V
d
|0i = |, 0i, with energy E =
+ t2 ,
2
4
(68)
where in the bras, the rst entry is the charge and spin of the impurity spin
and the second one is the compensating
charge in the screening channels.
Other states are elevated by energies O(V / N). The states satisfy the Friedel
There is a singularity only in the local charge response at the impurity, not in the
magnetic response. In this respect, the results of ([229]) are not completely correct
46
yields a self-energy
[ ln + isgn()] + O[( ln )2] .
(69)
n X
X
=1 i<j
() ()
()
()
ci cj + JS c0 c0 .
(70)
Here is the channel index. Degeneracy, the key to SFL behavior, is enforced
through equal antiferromagnetic coupling J > 0 for all the channels. When
the couplings to the various channels are not all the same, at low enough
temperatures a crossover to local Fermi-liquid behavior in the channel with
the largest J always occurs [7]. This crossover temperature is in general quite
large compared to TK because channel asymmetry is a relevant perturbation
about the symmetric xed point. Therefore in comparing this theory with SFL
behavior in experiments, one should ensure that one is above the crossover
temperature.
The simple Kondo problem is the case 2S = 1 = n. In this case at low
temperatures a singlet state is formed of the impurity state and the conduction
electron electrons in the appropriate channel. In the general multi-channel case
in which 2S = n, the physics is essentially the same, since there are exactly
the right number of conduction electron channels to compensate the impurity
spin at low temperatures. Thus, at low temperatures an eective spin 0 state
47
Since it is hard to imagine that the angular momentum states of the impurity are
larger than of the conduction electron states about the impurity, such models may
be regarded of purely theoretical interest. See however Sec (6.5)
48
Fig. 22. Flow diagram for the degenerate over-screened Kondo problem exhibiting
a critical point.
eective interaction, and so on. Of course, in reality one does not get a succession of antiferromagnetic Kondo problems the net eect is that a new
stable nite-J xed point appears. As sketched in Fig. 22, the renormalized
eective interaction ows to this xed point both from the strong-coupling as
well as from the weak-coupling side. One can understand this intuitively from
the above picture: if one starts with a large initial value of J, then in the next
order of perturbation theory about it, the interaction is smaller, since in perturbation theory the eective interaction due to virtual excitations decreases
with increasing J. This means that J scales to smaller values. Likewise, if we
start from small J, then initially J increases due to the Kondo scaling, but
once J becomes suciently large, the rst eect which tends to decrease J
become more and more important. The nite-J xed point leads to nontrivial
exponents for the low-temperature behavior of quantities like the specic heat,
Cv
T
T
TK
n2
n+2
(71)
For n = 2, the power law behavior on the right hand side is replaced by a ln T
term.
Another way of thinking about the problem is in the spirit of Wilsons renormalization group: Consider the problem of two channels interacting with a
S = 1/2 impurity. The conduction states can be expressed as linear combinations of concentric orbitals of conduction electrons around the impurity. These
successive orbitals peak further and further away from the impurity. Consider
rst the exchange coupling of the orbitals in each of the two channels peaking
at the impurity site. Each of them has S = 1/2. Only one linear combination
of the two channels, call it red, can couple, while the other (blue) does not. So
after the singlet with the impurity is formed, we are left with a S = 1/2, color
blue problem. We must now consider the interaction of this eective impurity
with the next orbital and so on. It is obvious that to any order we will be
left with a spin 1/2 problem in a color. Conformal eld theory methods rst
showed that the ground state is left with 1/2 ln 2 impurity. A nice application
of the bosonization method [83] identies the red and blue above as linear
combinations of the fermions in the two channels so that one is purely real,
the other purely imaginary. The emergence of new types of particles the
Majorana fermions in this case often happens at Singular Fermi-liquids.
As a detailed calculation conrms [7], the J1 = J2 xed point is unstable,
and the ow is like sketched in Fig. 24. This means that the J1 = J2 xed
49
S = 1=2
S = 1=2
S = 1=2
S = 1=2
Fig. 23. Effective shells in Energy space in Wilsons method. The first shell integrates
over a fraction of the top of the band, the next shell of the rest, and so on. In
the two-channel S = 1/2 problem, a S = 1/2 effective impurity is left at every stage
of interpretation.
i2
(J1 J2 )/J
T = O TK (J)
(72)
Fig. 24. Flow in temperature-anisotropy plane for the two-channel Kondo problem
with coupling constant J1 and J2 .
51
Fig. 25. (a) Differential conductance G(V, T ) as a function of voltage V in measurements on metal point contacts by Ralph et al. [205], for various
temperatures ranging
from 0.4 K (bottom curve) to 5.6 K (upper curve). Note the V type behavior developing as the temperature decreases. (b) The zero bias (V = 0) conductance as
a
function of temperature for three different samples shows a G(0, T ) G(0, 0) T
behavior. The scaling behavior as a function of voltage and temperature is consistent
with two-channel Kondo behavior [205].
usually 1, this is unlikely for S = 1/2 problems, but for large S, as encountered typically in rare earths and actinides, it is possible in some cases. But
the vast majority of rare earths and actinide compounds show magnetic order
and no Heavy fermion magnetic behavior.
The two-Kondo-impurity problem serves as a rst step to understanding some
of the physics of heavy fermions. We will however only consider the S = 1/2
problem and work with unrealistic JN(0) so that the competition between
the Kondo eect and RKKY is possible.
In this subsection we will summarize the results [128,129,8,9,229] for the two
Kondo impurity problem; like the models we discussed above, this one also has
a quantum critical point at which SFL behavior is found. However as for other
impurity problems, an unrealistic symmetry must be assumed for a QCP and
attendant SFL behavior.
52
X
k,
(73)
In this form, the problem has a symmetry with respect to the midpoint between the two impurity sites r1 and r2 , and hence one can dene even (e)
and odd (o) parity states relative to this point. In the approximation that
the k-dependence of the couplings is neglected [128] the two Kondo impurity
Hamiltonian can then be transformed to
H = Hnoninteracting +
+
X
kk
X
kk
(74)
The coupling between the spin and orbital channels generates an eective
RKKY interaction
HRKKY = I0 (Je , Jo , Jm ) S1 S2
(75)
2
between the two impurity spins, with I0 = 2 ln 2(Je2 + Jo2 2Jm
) for t = 1.
(76)
unless I0 is very small, |I0 /TK | 1. Since for uncorrelated impurity spins
hS1 S2 i = 0, (76) expresses that although in the RG language the RKKY
interaction is an irrelevant perturbation, it is quite important in calculating
physical properties due to large corrections to scaling. Another feature of the
solution in this regime is the fact that the phase shift is /2 in both channels.
This means that at the xed point the even-parity channel and the odd-parity
channel have independent Kondo eects, each one having one electron pushed
below the chemical potential in the Kondo resonance. As discussed below, this
is due to particle-hole symmetry assumed in the model without it, only the
sum of the phase shifts in the two-channels is xed.
(ii) There is no Kondo eect for I0 < 2.24 TK . In this case, the coupling
between the impurities is so strong that the impurities form a singlet among
53
themselves and decouple from the conduction electrons. There is no phase shift
at the xed point. Also, in this case the total spin Stot = 0, but the impurity
spins become only singlet like, hS1 S2 i 3/4 for very strong coupling,
I0 2.24 TK . So again there are important corrections to scaling.
(iii) The point I0 = 2.24 TK is a true critical point, at which the staggered
susceptibility h(S1 S2 )2 i/T diverges. Moreover, at this point the specic heat
has a logarithmic correction to the linear T dependence, Cv T ln T , while
the impurity spin correlation function hS1 S2 i becomes equal to 1/4 at this
value.
Although the approximate Hamiltonian (74) has a true quantum critical point
with associated SFL behavior, we stress that the analysis shows that this critical point is destroyed by any cke cko coupling. A coupling of this type is not
particle-hole symmetric; since the approximate Hamiltonian (74) is particlehole symmetric, these terms are not generated under the renormalization
group ow for (74). The physical two-Kondo-impurity problem (73), on the
other hand, is not particle-hole symmetric. Therefore, the physical two Kondo
impurity problem does not have a true quantum critical point in other
words, when the two-Kondo-impurity Hamiltonian (73) is approximated by
(74) by ignoring k-independent interactions, relevant terms which destroy the
quantum critical point of the latter Hamiltonian are also dropped. This has
been veried in explicit analysis keeping the symmetry breaking terms [229].
An illuminating way to understand the result for the two-impurity Kondo
problem, is to note that the Hamiltonian can be written in the following form:
H=
even
HS=0
0
Hmix
0
odd
HS=0
0
Hmix
even
Hmix HS=1
Hmix
0
0
odd
HS=1
(77)
the S = 0 and S = 1 states are degenerate, and where SFL behavior occurs.
At this critical value, the impurity spin is a linear combination of a singlet and
triplet state with hS1 S2 i = 1/4 (i.e. a value in between the singlet value
-3/4 and the triplet value 1/4) and the singular low-energy uctuations give
rise to the anomalous specic heat behavior.
Within this scenario, the fact that the susceptibility is divergent at the critical point signals that a term H (S1 S2 ) lifts the spin degeneracy. Moreover,
the leading irrelevant operators about the xed point are all divergent at the
critical point of course, this just reects the breakdown of the Fermi-liquid
description.
The reason for Hmix 0 is as follows. It arises from the last term in (74) which
is not particle-hole symmetric because under even-odd interchange both the
spin term and the fermion electron terms change sign. At the Kondo xed
point, the leading operators must all be biquadratic in fermions. An Hmix in
that case would be of the form cke ck o and and such a term by itself would
break particle hole symmetry.
In the two-Kondo impurity problem, one encounters again the essentials of degeneracy for quantum critical points and the need for (unphysical) constraints
to preserve the singularity. Once again new types of quantum numbers can be
invoked in the excitations about the QCP.
From the point of view of understanding actual phenomena for problems with
a moderate concentration of impurities or in reference to heavy fermion compounds, the importance of the solution to the two-Kondo impurity problem is
the large correction to scaling found in the Wilson-type solution away from the
special symmetries required to have a QCP. These survive quite generally and
must be taken into account in constructing low-energy eective Hamiltonians
in physical situations.
23
Even a Fermi-liquid displays distinct energy scales for charge and spin (particlehole) fluctuations because of the difference in the Landau parameters in the spinsymmetric and spin-antisymmetric channels. The phrase spin-charge separation
should therfore be reserved for situations, as in one-dimension, where the singleparticle excitations separate into objects carrying charge alone and which carry
spin alone.
56
(78)
k ck ck .
(79)
In one dimension the particles move in two branches, either to the left or the
right. Dene the charge density operators r and the spin density operators
Sr for the two branches, r = ,
r = =1 r,
r, ,
z
Srz = , r,
,
r, ,
2
(80)
(81)
(82)
is independent of k in one dimension (provided the band cuto is taken to innity). States created by r, (q) are linear combinations of individual electronhole excitations all of which have the same energy and are therefore eigenstates
of H0 . It follows that for q > 0, r=+, (q) [r=, (q)] is a raising [lowering] operator. Consequently H0 may be expressed in terms of the density operators:
H0 =
2vF X X
r, (rq)r, (rq) .
L r= q>0
(83)
Upon separating the densities on a given branch into charge and spin pieces,
1
r (x) = [r (x) + Srz (x)] ,
2
57
(84)
the free Hamiltonian may be expressed as a sum in the spin and charge degrees
of freedom:
H0 =
X
r
H0 [r ] +
X
r
H0 [Srz ].
(85)
It follows that in the noninteracting problem spin and charge have identical dynamics and propagate in unison. Once interactions are introduced, the
electron will fractionalize and spin and charge dynamics will, in general,
dier.
We now discuss the various interaction terms in the Hamiltonian. The piece
describing the forward scattering events (kF , ; kF , ) (kF , ; kF , )
and (kF , ; kF , ) (kF , ; kF , ) reads
Hf orward = H2 + H4 ,
(86)
with
1 X X
g +, (q), (q) ,
L q 2
1 X X
g [+, (q)+, (q) + , (q), (q)] .
H4 =
2L q 4
H2 =
(87)
(88)
X
k
cr,,k cr,,k+q .
(89)
XX
q
g1 +
(q) (q) .
(90)
1 X X +
g [ (q)+
(q) + (q) (q)] .
2L q 3
(91)
gi = gi|| + gi .
58
(92)
H1
H3
H2
kF
H4
kF
As the terms H1|| and H2|| describe the same process we may set g1|| = 0 with
no loss of generality. Umklapp processes are important only when 4kF is a
reciprocal lattice vector so that all scattering particles may be near the Fermi
points. If the one dimensional electronic system is suciently incommensurate
then these processes may be neglected.
~ = 0 reads
The condition for spin rotation invariance [H, S]
g2 g1 = g2|| g1|| .
(93)
A guide to the behavior of the Hamiltonian (78) is obtained from the perturbative Renormalization Group ow equations. In one dimension, there are no
truly ordered phases, but at T = 0 there is algebraic long range order. One
may thus determine a phase diagram according to which susceptibilities diverge as T 0: the one associated with singlet superconductivity (SS), triplet
superconductivity (T S), charge density wave (CDW ) state at 2kF , and spin
density wave (SDW ) state at 2kF . The expressions for these susceptibilities
are given in Eq. (111) below, and the resulting phase diagram is shown in
Fig. 27.
4.2 The Tomonaga-Luttinger Model
With forward scattering alone in (78) and after linearizing the kinetic energy
we obtain the Tomonaga-Luttinger (T-L)model. In terms of fermion eld op
erators r, (x) and r,
(x), and density and spin operators r (x) and Srz (x) in
real space, the T-L model is
HT L =
+
r=
1X
dx vF
rr,
ix r, +
g2,c r (x)r (x)
2 r=
r,=
X
r=
z
g2,s Srz (x)Sr
(x)
59
(94)
Fig. 27. Phase diagram for one-dimensional interacting fermions in the (Kc , g1 )
plane with g1|| = 0 in Eqn. (92) and Kc given by Eq.102 . SDW and (CDW )
in the upper left quadrant indicates that both the spin density wave susceptibility
and the charge density wave susceptibility diverge as T 0, but that the spin
density wave susceptibility diverges a factor ln2 T faster than the charge density
wave susceptibility see Eq. (111). Other sectors are labeled accordingly. From
[218].
where
gic =
gi|| + gi
,
2
gis =
gi|| gi
.
2
(95)
Note that the g2,s term is the only term which break SU(2) spin symmetry.
The T-L model is exactly solvable. After all, as previously noted, the (Diraclike) kinetic energy Hamiltonian H0 is also quadratic in the density operators.
So the Hamiltonian is readily diagonalized by a Bogoliubov transformation
whereupon the Hamiltonian becomes a sum of two independent (harmonic)
parts describing noninteracting charge and spin density waves: the charge and
spin density waves are the collective eigenmodes of the system.
The simplest way to describe the TL model and to explicitly track down these
collective modes is via the bosonization of the electronic degrees of freedom.
The reader should be warned that many dierent conventions abound in the
literature. The bosonic representation of the fermionic elds proceeds by writ60
ing [109,180,210]
exp[ir(kF x + r (x))]
Fr ,
a0
2a
r, (x) = lim
(96)
(97)
The Klein factors Fr are chosen such that the proper fermionic anticommutation relations are reproduced. The exponential envelope exp[ir (x)] represents the slow bosonic collective degrees of freedom which dress the rapidly
oscillating part Fr exp[ikF x] describing the energetic particle excitations near
the Fermi points.
may be written in terms of the bosonic elds c,s and their conjugate momenta x c,s
r, =
/2[(c rc ) + (s rs )] .
(98)
In terms of the new variables, the charge and spin densities read
(x) =
r (x) =
2
x c ,
S (x) =
X
r
Srz (x)
1
x s . (99)
2
T L
v
(x )2
dx
K (x )2 +
HsT L + HcT L . (100)
2
K
=c,s
X
"
The velocities of the collective charge and spin modes are easily read of by
analogy to a harmonic string:
vc,s =
g2c,s 2
g4c,s 2
(vF +
) (
) .
(101)
Likewise, the moduli determining the power-law decay of the correlations are
Kc,s =
v
u
u vF
t
+ g4c,s g2c,s
.
vF + g4c,s + g2c,s
61
(102)
In section 4.7 we shall show how the above expressions for the spin and charge
density wave velocities follow simply from the conservation of left and right
moving particles in the T-L model.
As previously noted, the charge and spin velocities are degenerate in the noninteracting model. When interactions are introduced, the charge and spin velocities (vc and vs ) as well as the energy to create spin and charge excitations
((vs /Ks and vc /Kc respectively) become dierent. The charge constant Kc is
less than 1 for repulsive interactions, which elevates the energy of the charge
excitations, while Kc is greater than 1 for attractive interactions.
4.3 One-particle Spectral functions
The zero temperature spectral functions of the T-L model are
A(k, ) ( vc (k kF ))2c 1/2 | vc (k kF )|c 1/2 (vc > vs )
A(k, ) ( vs (k kF ))c 1/2 | vs (k kF )|2c 1/2 (vc < vs ). (103)
These spectral functions are sketched in Fig. 28 for the case g2 = 0 and in
Fig. 29 for the general case. Notice that the quasiparticle pole in A(k, ) in a
Landau Fermi-liquid is smeared out to produce a branch cut extending from
the spin mode excitation energy to the charge mode excitation energy. These
branch cuts split into two in an applied magnetic eld, see Fig. 30. These
results are exact for small and small |k kF |. Another manifestation of the
SFL behavior is the behavior of the momentum distribution function
nk nkF const sign(k kF )|k kF |2c ,
(104)
where
c,s =
1
1
Kc,s + Kc,s
2 .
8
(105)
In contrast to a Fermi liquid, the expression for nk does not exhibit a step-like
discontinuity at the Fermi points. Note that the exponent 2c is non-universal
(as usual, an outcome of a line of critical points).
The single particle density of states
N() ||2c ,
vanishes at the Fermi surface.
62
(106)
Fig. 29. The generic (g2 6= 0, g4 6= 0) zero temperature spectral function. Note the
broader range of non-trivial singularities near vs k and vc k. Here both the effect of
spin-charge velocity difference and the emergence of non-trivial exponents is visible.
After Voit [262]
4.4 Thermodynamics
The contributions of the independent charge and spin modes appears independently in most physical quantities.
The specic heat coecient is found to be
/0 =
vF
2
63
1
1
.
+
vc vs
(107)
Fig. 30. The energy distribution curve (the spectrum A(k, ) at fixed k) as a function
of in the presence of a magnetic field. The dashed line is the zero-field result of Fig.
28. The magnitude of the Zeeman splitting is enhanced with respect to (vc vs )k
for clarity. From Rabello and Si [204].
/0 = vF Kc /vc
(108)
2vc
/0
=
,
/0
vc + vs
(109)
deviates from its Fermi Liquid value of unity by an amount dependent on the
relative separation between the spin and charge velocities.
cos(2kF (x x )) 3/2
Kc
ln
|x x |
+
B
1,c
((x x ))2
|x x |1+Kc
cos(4kF (x x ))
+ ...
(110)
B2,c
|x x |4Kc
cos(2kF (x x )) 1/2
1
~
~ )i
ln |x x | + ...
+
B
hS(x)
S(x
1,s
((x x ))2
|x x |1+Kc
h(x)(x )i
at asymptotically long distances and Ks = 1. For not too repulsive interactions, so that Kc < 1, the 2kF uctuations are dominant. We have previously
seen that such a CDW/SDW instability may arise due to the special 2kF
nesting wavevector in one dimension. The 4kF modulations are due to Umklapp scattering. The amplitudes {Bi,c } and {Bi,s } are non-universal while the
exponents are determined by the stiness of the free charge and spin elds.
While the above expressions are for Ks = 1, in the general case Ks 6= 1 the
spin correlator decays asymptotically with the exponent (Ks + Kc ).
At nonzero temperatures, it is found that the Fourier transforms of these
correlation functions scale as
CDW T Kc 1 | ln T |3/2 ,
SS T Kc 1 | ln T |3/2 ,
SDW T Kc 1 | ln T |1/2 ,
T S T Kc 1 | ln T |1/2 .
(111)
I4kF T 2Kc 1 .
(112)
The NMR probe couples to the spin degrees of freedom and the theoretically
computed nuclear relaxation time scales as
T1 T Kc .
(113)
Fig. 31. Left panel: Momentum distribution curves at = 0 (i.e., the spectrum at
fixed = 0 as a function of k) for a spin rotationally invariant Tomonaga-Luttinger
liquid, plotted as a function of vs k/T . Right panels: Energy distribution curves at
k = 0 (the spectrum at fixed k = 0) as a function of /T . In both panels, vc /vs = 3
and c = 0 in (a), c = 0.25 in (b), and c = 0.5 in (c). From Orgad [188].
dx g1
r=1
r,=+1
r,=1
r,=1 r,=1
(114)
vs
(x s )2
2g1
dx
Ks (x s )2 +
+
cos(
8s )
2
Ks
(2a)2
"
(115)
with rescaled values of the spin and charge velocities and stiness constants.
When g1 > 0 (repulsive interactions), g1 is renormalized to zero in the long
wavelength limit. Since along the RG ow trajectories Ks 1 g1 /(vs ), this
means that Ks renormalizes to 1. The physics corresponding to this case is in
the Tomonaga-Luttinger model that we just discussed.
When the backscattering interactions are attractive, g1 < 0, spin excitations
66
1/(22Ks )
(116)
vs
.
s
(117)
(118)
Luther and Emery observed that at the point Ks = 1/2 the Hamiltonian in
terms of these new spin elds becomes that of noninteracting free fermions
having a mass gap s = g1 /(2a).
Hs =
dx
[ivs rr x r + s r r ] ,
(119)
r=1
2s + vs2 (k kF )2 .
(120)
(121)
All of the currents are trivially invariant under this transformation as the
elds transform with opposite phases. Physically this corresponds to the
conservation of the number and net spin of left and right moving particles.
As discussed by Metzner and Di Castro [167], these separate conservation laws
for the left and right moving charge and spin currents lead to Ward identities
which enable the computation of the single particle correlation functions.
In the absence of Umklapp scattering, spin and charge are conserved about
each individual Fermi point. The net total charge density + + and
charge density asymmetry + in the Tomonaga-Luttinger Hamiltonian satisfy the continuity equations
= [H, ] = qj ,
= [H, ] = qj ,
(122)
j = uc [+ + ] ,
(123)
where
j(q) = uc [+ ] ,
and where the velocities are given by
uc = vF +
g4c g2c
,
uc = vF +
g4c + g2c
.
(124)
qL
2
(125)
Let us illustrate simply how many of the results derived via bosonization
may also be directly computed by employing these conservation laws. The
existence of gapless charge modes is a direct consequence of the right-left
charge conservation laws. The two rst-order continuity equations given above
lead to
[2 + uc uc q 2 ] = 0 ,
(126)
(127)
may be found for the spin velocity vs which in general is dierent from vc .
This spin-charge separation also becomes clear from the explicit form of the
expectation values of the charge and spin densities:
h0|r (x0 )r (x, t)r (x0 )|0i = (x x0 rvc t) ,
h0|r (x0 )Srz (x, t)r (x0 )|0i = (x x0 rvs t) ,
(128)
(129)
(130)
(131)
Thus the initial hole surrounded by two spins of the same polarization has
broken into a charge excitation (holon or chargon a hole surrounded
by antiferromagnetically aligned spins) and a spin excitation (spinon) composed of two consecutive parallel spins in an antiferromagnetic environment.
The statistics of the localized spinons and holons in this model must be such
that their product is fermionic.
The feasibility of well dened spin and charge excitations hinges on the commuting nature of the right and left kinetic (hopping) operators TRight , TLef t
which are the inverse each other. Any general term of the form
Lef t
(TLef t )n1
Right
(TRight )n1
Lef t
(TLef t )n2
Right
(TRight )n2
= (TLef t TRight )
NR
...
NL NR
TLef
t
NL NR
= I TLef
t
(132)
70
(133)
TDown TRight
TU p TLef
t we arrive at | i
-
(134)
2
2
a state which obviously diers from TDown Tup TRight
TLef
t |i = |i. Unlike the
one dimensional case, damage is not kept under check. Note the extended
domain wall neighboring the hole, enclosing a 2 2 region of spins of the
incorrect registry. Note also that the hole is now surrounded by a pair of antiferromagnetically aligned spins along one axis and ferromagnetically aligned
spins along the other. A path closing on itself does not lead to the fusion of the
a holon and spinon like entities back into a simple hole. As the hole continues to further explore both dimensions damage is continuously compounded.
2
2
+ +
(135)
As seen, the moving electron leaves a string of bad magnetic bonds in its
wake. The energy penalty of such a string is linear in its extent. It is therefore
expected that this (magnetic string) potential leads, in more than one dimension, to a conning force amongst the spin and charge degrees of freedom. As
this caricature for the single hole makes clear, the notion of localized spinons
and holons is unlikely to hold water for the U Hubbard model in more
than one dimension. The well-dened SFL solutions for special models with
nested Fermi-surface in two-dimensions should however be noted [94].
A certain form of spin-charge separation in two dimensions may be sought in
the very special hole aggregates (or stripes) that have been detected in some of
the cuprates [242] and the nickelates [58]. Here, holes arrange themselves along
71
lines which concurrently act as antiferromagnetic domain walls (i.e. behave like
holons) in the background spin texture. Charge and spin literally separate and
occupy dierent regions of space. In eect the two-dimensional material breaks
up into one-dimensional lines with weak inter-connections 24 . Related behavior
is also found in numerical work [163] in the so called, t -t-J model 25 . An
important question for such models is the extent to which the interconnections
between stripes are irrelevant i.e. the coupled-chains problem, which we
briey allude to in section 4.10.
(136)
2m i=1 x2i
2m x20
i=1
24
(137)
This observation has led to a line of thought which is of some interest in the
context of the issues discussed here. If one focusses on the quantum mechanics of a
single line of holes by formulating it as a quantum mechanical lattice string model
[85], the string traces out a two-dimensional world sheet in space-time. Quantum
mechanical particles in one dimension, on the other hand, trace out world lines in
space-time. It is claimed that one can recover most of the power law correlation
functions of one-dimensional interacting fermions from the classical statistical mechanics of fluctuating lines, and along these lines approach stripe formation as some
form of spin-charge separation in two dimensions [285].
25 In this paper it was further observed that the kinetic motion of single holes may
scramble the background spin environment in such a way that, on average, the
holes may become surrounded by antiferromagnetically ordered spins on all sides
(i.e. both along the horizontal and along vertical axis) this is claimed to be a
higher dimensional generalization of the holons encountered thus far.
72
The calculation for the small s-wave phase shifts for all single particle states
{i (xi )} is relatively straightforward. The quasiparticle weight Z reduces to
an overlap integral between two (N + 1)-particle Slater determinants, and one
nds
2
Z N 2(F /)
(138)
(139)
with
Assume a simple featureless nal state potential V , and consider rst the case
without recoil. The relevant quantity to calculate is the transient propagator
for the scatterer,
G(t) = h0|deiHt d |0i ,
(140)
Zt
0
Zt
0
d V 2 g(0, )g( ) ,
(141)
where g(0, ) is the free conduction electron propagator at the origin. For
large times, one has g(0, ) iN(0)/ , and when this is used in (141) one
immediately nd that for large times
C(t) = V 2 N(0)2 ln t ,
G(t) 1/tn ,
(142)
where
n = V 2 N(0)2 = 2 / 2 + 1
(143)
is the phase shift exponent due to the orthogonality eect, compare Eq. (56).
A power law decay of G(t) 1/tn at long times corresponds to power law
2
2
dependence ( D )n1 = ( D ) / just above the absorption edge.
If we now take into account the recoil eect; the dispersion of the lower band
implies that the hole in this band can hop from site to site. The propagator
G(t) is then obtained as a sum over all trajectories R( ) of the scattering hole
which begin and end at R = 0. For a given history, we can extend the above
analysis to lowest order by replacing the propagator g(0, ) by g(, ), where
( ) = R( ) R( ). For positive time dierence, we can then write
g(, ) =
eik ik .
(144)
k>kF
For small hopping rates and large times, the integration over the modulus is
dominated by the energy term, and this yields a term proportional to 1/
as in the recoilless case. The trajectory of the hole enters through the average
74
exp(ik )
FS
exp(ik )
FS
cos(kF )
J0 (kF )
sin(kF )
kF
d=1
d=2
(145)
d=3
In order to calculate the large time behavior of the Greens function, we nally
have to average the square of this result over the distribution function of the
trajectories for large times. Using the large- behavior of the expressions
found above, one then easily nds that for large times
g 2 (, )
*
exp(ik )
FS
2 +
1 2
N (0)/ 2
2
d=1
d=3
where typ is the typical distance the hole trajectory moves away from the
origin. In one dimension, we see that g 2 still falls of as 1/ 2 and hence in
analogy to (142) that G(t) has power law long time behavior: in the presence
of recoil, an edge singularity persisits but the exponent n is now only half that
in the absence of dispersion of the lower state, due to the averaging over .
1
Since typ ( ) grows diusively as 2 for large , the integrand in the expression
(141) for C(t) converges faster than 1/t2 , and hence C(t) converges to a nite
limit for large times. The singular X-ray edge eect is washed out in two and
three dimensions due to the recoil.
If the above argument is extended to arbitrary dimension by analytically continuing the angular average over the Fermi surface to continuous dimensions,
one nds that the orthogonality and concomitant singular behavior is destroyed for any dimension d > 1. Nevertheless, the subdominant behavior of
the integrals will contain noninteger powers of time, and this gives rise to subdominant nonanalytic terms in the spectrum for noninteger d. This behavior
is completely in agreement with an analysis of the dimensional crossover from
Luttinger liquid behavior to Fermi liquid behavior as a function of dimension
[50].
75
(147)
The fact that the exponent here is only half the value attained in the simple
Luttinger Liquid see Eq. (106) is due to the fact that there are two Luttinger Liquid-like bands present in the carbon nanotubes 26 . Only one linear
combination of the two associated charge modes attains a nontrivial stiness
Kc 6= 1 [133].
By contrast, at the tips of the nanotubes, surplus electronic charge can propagate in only one direction and, as a consequence, the tunneling is more restricted,
Ntip () tip
(148)
26
2)
i(
(0,0)+
(0,0))/
2)
i(
(x,t)+
(x,t))/
cR
sR
cR
sR
e
i. Note the factor
is
GR=+1 (x, t) = he
variables.
1/ 2 in the exponent, coming from the projection onto the proper bosonic
i
(x,t)/
2
i
(0,0)/ 2 i
cR
cR
This expection
value
can
be
written
as
G
(x,
t)
=
he
e
R=+1
heisR (x,t)/ 2 eisR (0,0)/ 2 i, which with the aid of the results of section 4.3 becomes GR=+1 (x, t) = |xv1t|1/2 |x2 v12 t2 |c |xv1 t|1/2 |x2 v12 t2 |s . From this result,
c
one immediately obtains the density of states NR() = dt G(x = 0, t) eit . A simple integration then yields N () = [vc2c vs2s ] dt t1 t2(c +s ) eit 2(c +s ) .
Since Ks renormalizes to 1 for repulsive interactions [see the remark just after (115)],
one usually has s = 0 and so in this case the density of states exponent is simply
2s . This is precisely (106). Now consider what one gets for the spinless Luttinger
liquid. In this case neither the bosonic spin modes nor the projection factor 1/ 2
are present in the above expression for G. This results in a spatial decay with an
exponent 2c instead of c , and hence an exponent 4c in the density of states. In
other words, for s = 0 the density of states exponent in the spinful case is half of
what it is in the spinless case. The same mechanism is at work in the nanotubes.
R
77
Fig. 32. Differential conduction dI/dV measured by Postma et al. [197] for carbon
nanotubes. At low voltages or temperatures, Coulomb blockade effects dominate,
but at higher temperatures or bias voltages, one probes the one-dimensional SFL
behavior. Panel (a) shows the differential conductance as a function of bias voltage
for various temperatures (note that these temperatures are relatively high, reflecting
the fact that the electronic energy scales of the nanotubes is high). The effective
exponent for the large V behavior is 0.48; since these data are for tunneling
between two nanotubes, = 2c , so c 0.24 and Kc 0.27. The predicted value
is Kc 0.2 [133]. The data in panel (b) show the differential conductance as a
function of temperature at fixed bias for two nanotubes which cross, as well as for
a single nanotube with a bend.
By Fermis golden rule, the relevant exponents for (tip-tip) or (bulk-bulk) tunneling are tt = (Kc 1)/2 and bb = 2c respectively. Bulk-bulk tunneling
is achieved by arranging the nanotubes according to the crossing geometry
depicted in the inset of Fig. 32 above. By extracting the value of the charge
stiness Kc from each of the independent measurements of tt , bb for the
two dierent geometries, a single consistent stiness Kc 0.27 was found
[197], in good agreement with theoretical prediction. This is a beautiful realization of Luttinger Liquid physics!
Other, older, canonical realizations of Luttinger liquids include the Quantum Hall edge states. These represent a chiral spinless Luttinger liquid. Here,
low lying energy states can only prevail at the edge of the sample, and, concurrently, disperse linearly about the Fermi energy. Edge states can attain
macroscopic linear (perimeter) extent, and the tunneling experiments between
such states [170,110] have observed several features predicted theoretically
[271,132]. We refer the interested reader to the review articles by Schulz et al.
[218] and Fisher and Glazman [96] for further details.
78
ij ki kj
,
c2 k 2 2 i
(149)
(150)
Dij1 = Dij0
1
M ij ki kj
1
(151)
1
4
n+
M(k, ) =
m
m
2 ][fpk/2 fp+k/2 ]
dd p [p2 (p k)
. (152)
(2)d
(pk/2 p+k/2 )
3n
4
M(k, )
m
kvF
79
2
+ 2i
kvF
(153)
Fig. 33. (a) The diagrams for M; (b) The diagram for the self-energy from the in
the discussion of the SFL effect arising from the coupling of the electrons to the
electromagnetic gauge fields; (c) Anharmonic interaction of fluctuations such as
(c) are non-singular in the SFL problem of coupling of the electrons in metals to
electromagnetic fields.
ij ki kj
,
i6n nkv
+ c2 k 2 i
F
(154)
ij ki kj
.
4i(k, ) + c2 k 2
(155)
at small frequencies generally goes as k 3 for a pure metal. For dirty metals, the fact that (k, 0) approaches a nite limit 0 at small wavenumbers, so
for dirty metals there is according to (155) a crossover to a behavior k 2
at small wavenumbers.
Gauge invariance of the theory requires that the photon cannot acquire a
mass (a nite energy in the limit k 0) in the interaction process with
the electrons, and hence the form of Eq. (154) remains unchanged. Thus the
anharmonic corrections to the photon propagator due to processes such as
shown in Fig. 33(c) do not change the form of Eq. (154). The self-energy
of the electrons due to photon exchange, Fig. 33(b), may now be calculated
with condence given the small coupling constant in lowest order. The leading
contribution in d = 3 is
(kF , ) ( ln + isgn)
(156)
(158)
27
This follows from the fact that in the quasi-elastic approximation, the transport
1
is related to the single-particle relaxation rate 1 () due to
relaxation rate tr
scattering through an angle near the Fermi-surface
1
=
tr
(157)
81
while for the entropy for the electrons interacting with the electromagnetic
eld one nds from the above results [120]
2 2 T
S
0
ln
N
3 ckF
T
(159)
with
0 =
cF
.
vF
(160)
Upon comparing these two results, one concludes that the SFL eects start to
become important for temperatures
<
T 0 e
3m ck2
B
2
h 2 kF
(161)
Since the numerical factor in the exponent is typically of order 105 1 for
ordinary metals, the temperature range one nds from this is of order 1015
Kelvin for values of of order unity, according to this estimate. Note however
that for pure ferromagnetic metals can be as large as 104 . Possibly, in some
ferromagnets, the eects can become real.
5.2 Generalized gauge theories
The example of coupling to the electromagnetic eld identies one possible
theoretical route to SFL behavior, but as we have seen the crossover temperatures that one estimates for this scenario are too small for observable physical
properties. The smallness of the estimated crossover temperature is essentially
due to the fact that the coupling to the electromagnetic eld is determined by
the product vF /c, where = 1/137 and typically vF /c = O(102 ). Motivated
by this observation, many researchers have been led to explore the possibility
of getting SFL eects from coupling to dierent, more general gauge bosons
which might be generated dynamically in strongly correlated fermions. For
a recent review with references to the literature, see [150,179,180]. The hope
is that if one could consistently nd such a theory in which the small factor
vF /c arising in the electromagnetic theory is replaced by a term of order
unity, realistic crossover temperatures might arise. Much of the motivation
in this direction comes from Andersons proposal [24] of spin-charge separation and resonating valence bonds in the high-temperature superconductivity
problem which we discuss in section 7.
The essence of approaches along these lines is most easily illustrated by considering electrons on a lattice in the case in which strong on-site (Hubbard-type)
82
repulsions forbid two electrons to occupy the same site. Then each site is either occupied by an electron with an up or down spin, or by a hole. If we
introduce ctitious fermionic creation and annihilation operators f and f for
the electrons and ctitious bosonic hole creation and annihilation operators b
and b for the holes, we can express the constraint that there can only be one
electron or one hole on each lattice site by
2
X
fi
fi + bi bi = 1
for each i .
(162)
=1
With this convention, the real electron eld i can be written as a product
of these fermion and boson operators
i = fi bi .
(163)
This expresses the fact that given the constraint (162), a real fermion annihilation at a site creates a hole. This way of writing the electron eld may be
motivated by the physics of the one-dimensional Hubbard model: there a local
excitation may indeed be expressed in terms of a charged spinless holon and
an uncharged spinon. In general, in a transformation to boson and fermion
operators as in (163) there is some freedom with which operator we associate
the charge and with which one the spin.
Whenever we split a single electron operator up into two, as in (163), then
there is a gauge invariance, as the product is unchanged by the transformation
fi (t) ei(t) fi (t) ,
(164)
We can promote this invariance to a dynamical gauge symmetry by introducing a gauge eld a and writing the Hamiltonian in the continuum limit
as
H=
dr
i + a
2mf
!2
dr b
f + ( f )f f +
i + a
2mb
2
b + ( b )b b .
(165)
In practice, the most common situation in which one observes a quantum critical point is the one sketched in Fig. 1, in which there is a low temperature
ordered state a ferromagnetic state, antiferromagnetic state or charge density wave ordered state, for instance whose transition temperature to the
disordered state or some other ordered state goes to zero upon varying some
parameter. In this case, the quantum critical point is then also the end point
of a T = 0 ordered state. However, sometimes the ordered state really only
exists at T = 0; for example in metal-insulator transitions and in quantum
Hall eect transitions [232,198]. A well-known example of this case in spin
models is in two-dimensional antiferromagnetic quantum Heisenberg models
with quantumness as a parameter g [55], which do not order at any nite
temperature, but which show genuine ordered phases at T = 0 below some
value g < gc .
Although the question concerning the origin of the behavior of high temperature superconductors is not settled yet, there are strong indications, discussed
in the next section, that much of their behavior is governed by the proximity
to a quantum critical point.
One of the rst formulations of what we now refer to as quantum critical behavior was due to Moriya [177,178] and Ramakrishnan [174], who did an RPA
calculation for a model of itinerant fermions with a Stoner-type instability to
a ferromagnetic state. In modern language, their approach amounts to a 1/N
expansion. Various other important contributions were made [37,77]. The now
standard more modern formulation, which we will follow, is due to Hertz [117].
A nice introduction can be found in the article by Sondhi et al. [232], and for
a detailed expose we refer to the book by Sachdev [210]. See also the review
by Continento [66].
A clear example of quantum critical behavior, and actually one for which
one can compare with theoretical predictions, is summarized in Figs. 34-37.
The gures show various data from [239] on the magnetic compound MnSi
[191,192,152,161,239]. Fig. 34 shows that for low pressures and temperatures,
this compound exhibits a magnetic phase whose transition temperature Tc
vanishes as the pressure is increased up to pc = 14.8 kbar. This value of the
pressure then identies the quantum critical point. Fig. 35 shows that when
the same data are plotted as Tc4/3 versus pressure, the data fall nicely on a
straight line except close to the critical pressure. This observed scaling of Tc4/3
with ppc away from pc is in accord with the behavior predicted by the theory
discussed below. Actually the transition is weakly rst order near pc ; so very
85
(166)
Fig. 36. T -dependence of ((T )T =0 ) near the critical pressure pc with and without
an external magnetic field. From [239].
Fig. 37. Evolution of 0 and A under pressure, when the temperature dependence
of the resistivity is fitted to ( = 0 + AT 2 ). From [239].
then both the residual resistivity 0 and the amplitude A are found to show
a sharp peak at pc as a function of pressure see Fig. 37. This behavior
is not understood nor is the fact that the exponent does not appear to
regain the Fermi-liquid value of 2 for signicant values of p > pc (at H = 0)
and in a temperature regime where the theory would put the material in the
quantum-disordered Fermi-liquid regime.
87
(167)
(168)
The combination of these two results shows that critical slowing down implies
that near any nite temperature critical point the characteristic frequency
scale c goes to zero as
c |T Tc |z .
(169)
Therefore near any phase transition c Tc , and as a result the phase transition is governed by classical statistical physics: the Matsubara frequencies are
closely spaced relative to the temperature, the thermal occupation of bosonic
modes is large and hence classical, etc.
In classical statistical mechanics, the dynamics is a slave to the statics; usually,
the dynamical behavior is adequately described by time-dependent LandauGinzburg type of equations or Langevin equations which are obtained by
building in the appropriate conservation laws and equilibrium scaling behavior
[112]. At a quantum critical point, the dynamics must be determined a priori
from the quantum-mechanical equations of motion.
The general scaling behavior near a T = 0 quantum critical point can however
be discussed with the formalism of dynamical scaling [112,156], just as near
classical critical points. Consider for example the susceptibility for the case of
MnSi that we considered above. The scaling ansatz for the singular part of
the susceptibility
(k, , p) = hMMi(k, , p)
(170)
implies that near the critical point where the correlation length and time scale
88
(171)
where now
|p pc | ,
t = z .
(172)
This is just like the classical scaling with T Tc replaced by p pc . The reason
for writing t instead of is that in quantum statistical calculations, the timewise direction becomes like an additional dimension, so that t plays the role
of a correlation length in this direction. However, the time-direction has both
a long-time cuto given by 1/kB T and a short-time cuto given by the high
energy cuto in the problem exchange energy or Fermi-energy, whichever
is smaller in the ferromagnetic problem. The short-time cuto has its analog
in the spatial scale. The long-time cuto, which determines the crossover from
classical to quantum behavior, plays a crucial role in the properties discussed
below. The crucial point is that when z 6= 1, there is anisotropic scaling
between the space and time-wise direction, and as we shall discuss below, this
implies that as far as the critical behavior is concerned, the eective dimension
of the problem is d + z, not d + 1.
The exponent dM in (171) reects that a correlation function like has some
physical dimension which often is inevitably related to the spatial dimension.
The dependence of critical properties on spatial dimensions must be expressible purely in terms of the divergent correlation length . Often dM is xed by
dimensional considerations (in the language of eld theory, it is then given by
the engineering dimension of the eld), but this may not be true in general.
It must be so, however, if is a correlation function of a conserved quantity 28 .
Let us now address the nite temperature scaling, taking again the case of
MnSi as an example. The various regimes in the T -p diagram discussed below
are indicated in Fig. 38. To distinguish these regimes it is necessary to dene
an additional quantity, the thermal length
LT
.
kB T
(173)
E.g., if we consider the free energy per unit volume at a classical transition, the
energy scale is set by kB T , and dM = d; likewise, if we consider the surface tension
of an interface, whose physical dimension is energy per unit surface area, dM = d1.
89
Fig. 38. Generic phase diagram and cross-overs for quantum critical points with the
various regimes indicated. is the correlation length at T = 0; LT is the thermal
correlation length.
90
corrected Hertzs results [117] on this point, and has found that the critical
temperature of the phase transition scales as Tc |p pc |z/(d+z2) . Estimates
of the classical critical region are also given. Results along similar lines may
also be found in [65,66].
If we include both the temperature and the parameter p, the scaling ansatz
for the imaginary part of becomes
(k, , p, T ) dM 1 k, z ,
(174)
(k, , p, T ) =
d /z
LT M 2
1/z
kLT , LT ,
LT
t
(175)
(k, , p, T ) = T
dM /z
1
,
,
T 1/z T T t
(176)
Moreover, the scaling of the free energy F can be obtained from the argument
that it is of the order of the thermal energy kB T per correlated volume d .
Moreover, since LT acts as a nite cuto for t in the timewise direction, we
then get the scaling
F T d T (t )d/z T 1+d/z .
(177)
By dierentiating twice, this also gives immediately the specic heat behavior
at low temperatures. In writing the above scaling forms, we have assumed that
no dangerously irrelevant variables exist, as these could change /T scaling
to /T scaling 30 .
30
Dangerously irrelevant variables are irrelevant variables which come in as prefactors of scaling behavior of quantities like the Free Energy [95]. Within the Renormalization Group scenario, the hyperscaling relation d = 2 is violated above the
upper critical dimension because of the presence of dangerously irrelevant variables.
Presumably, dangerously irrelevant variables are more important than usually at
QCPs, since the effective fluctuation dimension is above the upper critical dimension for d = 3 and z > 1. Some examples are discussed in [210].
91
In order to get the critical exponents and the cross-over scales, one has to turn
to a microscopic theory. The theory for this particular case of the quantum
critical point in MnSi is essentially a Random Phase Approximation and
proceeds along the following lines: (i) One starts with a model of interacting
fermions; (ii) An ordering eld M(k, ) is introduced; (iii) One assumes the
fermions can be eliminated near the critical point to get a free energy in terms
of M of the form
F=
d{}
(1 + 2 + 3 + 4 )(k1 + k2 + k3 + k4 ) + (178)
This has important consequences for a scaling analysis of numerical data, aimed
at determining the critical behavior. For it implies that the finite size scaling has
to be done anisotropically, with the anisotropy depending on the exponent z which
itself is one of the exponents to be determined from the analysis
92
1 (k, ) = (p pc ) + k 2 +
i
kvF
(179)
In the rst two terms we recognize the usual mean eld type behavior with a
correlation length that diverges as (p pc )1/2 , hence the critical exponent
= 1/2. The last term, which describes Landau damping of the spin wave
modes, is very special here as it arises from uctuations of magnetization, a
quantity which is conserved, i.e., commutes with the Hamiltonian. Therefore
the characteristic damping rate must approach zero as k 0. Since at the
critical point (179) leads to a damping k 3 , the critical exponent z = 3.
According to the theory described above, the critical behavior at the quantum
critical point (T = 0, p = pc ) is therefore of the mean-eld type for any physical
dimension d > 1, since the eective dimension d + z > 4 for d > 1.
The scattering of electrons o the long wavelength spin waves is dominated
by small angle scattering, and it is easy to calculate the resulting dominant
behavior of the self-energy of the electrons. Near the critical point the behavior
of is very similar to the electromagnetic problem we discussed in section 5.
Analogously, one also nds SFL behavior here: in d = 3, (kF , ) ln +
i|| while in d = 2 one obtains (kF , ) 2/3 + i||2/3 . Furthermore, for the
resistivity one nds in three dimensions T 5/3 this is consistent with the
behavior found in MnSi in the absence of a eld, see Fig. 36. Moreover, as we
mentioned earlier, according to the theory, near the critical point Tc should
vanish as |p pc |z/(d+z2) ; with d = z = 3 this yields Tc |p pc |3/4 . As we
saw in Fig. 35, this is the scaling observed over a large range of pressures,
except very near pc . ZrZn2 [111] is an example in which the Ferromagnetic
transition is shifted to T = 0 under pressure continuously. The properties are
again consistent with the simple theory outlined. There is however trouble on
the horizon [153]. The asymptotic temperature dependence for p > pc is not
proportional to T 2 , as expected. We will return to this point in sections 6.4.
For antiferromagnets or charge density waves the critical exponents are dif93
ferent. In these cases, the order parameter is not conserved, and the inverse
susceptibility in these cases is of the form
i
(k, ) = (p pc ) + (k k0 ) +
(180)
Fig. 39. Left panel: Temperature-pressure phase diagram of high-purity single-crystal CeIn3 . A sharp drop in the resistivity consistent with the onset of
superconductivity below Tc is observed in a narrow window near pc the pressure
at which the Nel temperature TN tends to absolute zero. Upper inset: this transition is complete even below pc itself. Lower inset: just above pc , where there is
no Nel transition, a plot of the temperature dependence of d(ln )/d(ln T) is
best able to demonstrate that the normal state resistivity varies as T 1.60.2 below
about 3 K. is the difference between the normal state resistivity and its residual
value (which is calculated by extrapolating the normal-state resistivity to absolute
zero). For clarity, the values of Tc have been scaled by a factor of ten. Right panel:
Temperature-pressure phase diagram of high-purity single-crystal CeP d2 Si2 . Superconductivity appears below Tc in a narrow window where the Nel temperature
TN tends to absolute zero. Inset shows the normal state a-axis resistivity above the
superconducting transition varies as T 1.20.1 over nearly two decades in temperature. The upper critical field Bc2 at the maximum value of Tc varies near Tc at a
rate of approximately 6T /K. For clarity, the values of Tc have been scaled by a
factor of three, and the origin of the inset has been set at 5 K below absolute zero.
Both plots are from Mathur et al. [161].
95
(181)
Fig. 40. The specific heat C/T of CeCu6x Aux versus log T . From L
ohneysen et al.
[266268].
sistent with any of the models that we have discussed. Information on the
magnetic uctuation spectra for CeCu5.9 Au0.1 is available through neutron
scattering experiments [215,236]. The data shown in Fig. 43 show rod-like
peaks, indicating that the spin uctuations are almost two-dimensional at
this composition. The neutron scattering data can be tted by an expression
for the spin susceptibility of the form
1 (k, ) = C [f (k) + (i + aT ) ]
(182)
Fig. 43. Neutron scattering data for CeCu5.9 Au0.1 , a compound which is close to
a QCP. The figure shows q-scans along three different crystallographic directions,
from top to bottom in the a, b and c directions for h
= 0.1meV . The figures shows
that there is only a weak q-dependence along the rods (q ), while transverse scans
(q|| ) shows well defined peaks with nearly the same line width. From Stockert et al.
[236].
98
tering,
f (k) = b(k )2 + c(kk )4 .
(183)
kk and k are the deviations from the AFM Bragg vector parallel and perpendicular to the c-axis in these tetragonal crystals. Further
a1,
= 0.74 0.1 .
(184)
At present, there is no room for this anomalous exponent within known theoretical frameworks. However, if one accepts this particular form of as giving
an adequate t, the observed specic heat follows: at the critical composition,
we expect the temperature scaling relation
(d1) 1
k
F T
T T (d1)/2 T /4 T 1+(d1/2)/2 ,
(185)
(186)
This ensures that the single-particle relaxation rate as well as the transport
relaxation rate is proportional to T . A major theoretical problem is why the
non-local or recoil terms in the interaction of itinerant fermions are irrelevant; i.e why is the eective Hamiltonian not
(187)
Fig. 44. The phase diagram of U Cu5x P dx . At low dopings and temperatures the
system is in an antiferromgnetic phase. In the undoped sample (x = 0): TN = 15 K
with a magnetization 1B . For doping x = 1 and 1.5 the specific heat Cv /T T
(for x = 1) while displaying weak logarithmic characteristics for x = 1.5. Similarly,
the susceptibility (T ) ln T and T 0.25 (for x = 1, 1.5 respectively). Courtesy
of M. C. Aronson.
T 1/3
(188)
The data for that show this power law behavior for UCu3.5 P d1.5 and UCu1 P d
are shown in Fig. 45. The fact that this is genuine scaling behavior is independently conrmed [29,30] from the fact that the frequency dependent susceptibility, measured by neutron scattering, shows a very good collapse of the
data with the scaling assumption
(, T ) = T 1/3
(189)
In passing, we note that this as well as the result for in CeCuAu are examples
100
Fig. 45. The temperature dependence of the static susceptibility of 0 (T ) for both
U Cu4 P d and U Cu3.5 P d1.5 , showing that for both compounds 0 (T ) has a low temperature divergence as T 1/3 . The measuring field is 1 Tesla. From Aronson et al.
[29,30].
(190)
Actually, the Harris criterion is derived in the form (190) for and d, not in terms
of the exponent . This is particularly important at QCPs, since as we discussed
in section 6.2 for QCPs one is often above the upper critical dimension where the
hyperscaling relation d 2 = breaks down.
103
(a)
Fig. 47. (a) Schematic structure of the copper-oxide ab-planes in La2 CuO4 . Ba
or Sr substitution for La in the parent compound La2 CuO4 introduces holes in
the CuO2 planes. The structure of other high Tc materials differs only in ways
which do not affect the central issues, e.g., it is oxygen doping in Y Ba2 Cu3 O6+x
that provides planar holes. The magnetic moments of the planar copper atoms
are ordered antiferromagnetically in the ground state of the undoped compounds.
From [127]. (b) The orbital unit cell of the Cu-O compounds in the ab plane.
The minimal orbital set contains a dx2 y2 orbital of Cu and px and a py orbital of
oxygen per unit cell.
About 105 scientic papers have appeared in the eld of high-Tc superconductors since their discovery in 1987. For reviews, see the Proceedings [164] of the
latest in a series of Tri-annual Conferences or [106]. Although no consensus
on the theory of the phenomena has been arrived at, the intensive investigation has resulted in a body of consistent experimental information. Here
we emphasize only those properties which are common to all members of the
copper-oxide family and in which Singular Fermi-liquid properties appear to
play the governing role.
The high-Tc materials are complicated, and many fundamental condensed matter physics phenomena play a role in some or other part of their phase diagram.
As we shall see, the normal state near the composition of the highest Tc shows
convincing evidence of being a weak form of a SFL, a Marginal Fermi Liquid.
Since the vertices coupling fermions to the uctuations for transport in the
normal state and those for Cooper pairing through exchange of uctuations
can be derived from each other, the physics of SFL and the mechanism for
superconductivity in the cuprates are intimately related.
104
Fig. 48. a) The crystal structure of La2 CuO4 . From [187]. Electronic couplings
along the c direction are weak. b) Schematic of the CuO2 plane. The arrows indicate
the alignment of spins in the antiferromagnetic ground state of La2 CuO4 . Speckeled
shading indicates oxygen p orbitals; coupling through these leads to superexchange
in the insulating state.
Fig. 49. Generic phase diagram of the cuprates for hole doping. The portion labeled
by AFM is the antiferromagnetic phase, and the dome marked by SC is the superconducting phase. Crossovers to other characteristic properties are marked and
discussed in the text. Not shown is a low-temperature insulating phase in region
II due to disorder.
plane. In the few compounds with electron doping which have been synthesized
properties vary with doping density in a similar way.
Antiferromagnetism disappears for x typically less than 0.05 to be replaced
by a superconducting ground state starting at somewhat larger x. The superconducting transition temperature is peaked for x typically between 0.15 and
0.20 and disappears for x typically less than 0.25. We will dene xm to be
the density for maximum Tc . Conventionally, copper-oxides with x < xm are
referred to as underdoped, with x = xm as optimally doped and with x > xm
as overdoped. Superconductivity is of the d-wave singlet symmetry.
The superconducting region in the T -x plane is surrounded by three distinct
regions: a region marked (III) with properties characteristic of a Landau Fermiliquid, a region marked (I) in which (marginally) SFL properties are observed
and a region marked (II) which is often called the pseudo gap region whose correlations in the ground state still remain a matter of conjecture. The topology
of Fig. 49 around the superconducting region is that expected around a QCP.
Indeed it resembles the phase diagram of some heavy fermion superconductors
(see e.g. Fig. 39) except that region II has no long-range antiferromagnetic order the best experimental information is that, generically, spin rotational
invariance as well as (lattice) translational invariance remains unbroken in the
passage from (I) to (II) in the Cu-O compounds.
The quantity (T ) Cv (T )/T and the magnetic susceptibility (T ), which are
temperature independent for a Landau-Fermi-liquid begin to decline rapidly
[154] in the passage from region I to region II, which we will call Tp (x), but
106
Fig. 50. The electronic contribution to the specific heat as a function of temperature
for underdoped, optimally doped and overdoped samples of Y0.8 Ca0.2 Ba2 Cu3 O7x .
For optimally doped and overdoped samples the heat capacity remains constant
as the temperature is lowered, then shows the characteristic features at the superconducting transition temperature Tc and rapidly approaches zero in the superconducting state. For underdoped samples, however, the heat capacity starts to fall
well above Tc as the temperature is reduced, and there is only a small peak at
Tc indicating much smaller condensation energy than the optimal and overdoped
compounds [154]. From [34].
107
Fig. 51. Signatures of the pseudogap in various transport properties for the underdoped compound Y Ba2 Cu4 O8 . At high temperatures the resistivity (solid line)
decreases linearly with temperature. In the pseudogap region it drops faster with
temperature before falling to zero at the superconducting transition temperature
(about 85K). Similarly the NMR relaxation time displays characteristics of the optimum doped compounds above about 200 K (squares on dashed line) but deviates
strongly from it in the pseudogap region. The NMR shift (top squares) also deviates from the temperature-independent behavior (not shown) below the inset of
the pseudogap. Note that the pseudogap expresses itself as a sharper change with
temperature in the transport properties compared to the equilibrium properties specific heat and magnetic susceptibility. [45,278,10]. From [34].
Fig. 52. Resistivity as a function of temperature for various high temperature superconductors. From [38].
(q, )
= o /T , for T ,
= o sgn() ,
for c > T .
(191)
Here o is the order of the bare single-particle density of states N(0) and
c is an upper cut-o. The uctuation spectrum is assumed to have only
a weak momentum dependence, except at very long wavelength, where a q 2
dependence is required for uctuations of conserved quantities like density or
spin (in the absence of spin-orbit interactions). A form which implements these
requirements for the conserved quantities with a rather arbitrary crossover
function to get the dierent regimes of /T is
(q, )
xq 2
,
( 2 + 2 x2 )
for vF q
x,
(192)
110
ln
i x
c
2
(193)
Fig. 53. (a) Diagram for the singular contribution to the one-particle self-energy
with the fluctuating (q, ). gs are the vertices which in microscopic theory [255]
is shown to have important momentum dependence, but which gives negligible momentum dependence to the self-energy. (b) For 0 (q, ) which is momentum independent, a total vertex may sometimes be usefully defined, which has the same
frequency dependence as Eq. (193) and which is also q-independent. The lines are
the exact single-particle Greens functions.
1 ln
x
c
1
(194)
1
(q ) (q, )
(195)
(196)
(197)
(q, )
1
.
[ (q ) (q, )]2 + [ (q, )]2
(198)
(199)
Fig. 55. (a) Momentum distribution curves for different temperatures. The solid
lines are Lorentzian fits. (b) Momentum widths of MDCs for three samples (circles,
squares, and diamonds). The thin lines are T -linear fits which show consistency with
Eqn.(198) and the MFL hypothesis. The resistivity (solid black line) is also shown.
The double-headed arrow shows the momentum resolution of the experiment. From
Johnson et al. [246].
Fig. 56. Fits of the MFL self-energy + h to the experimental data, according to
(199). Estimated uncertainties of 15% in and 25% in . (a) A scan along the
(1,0) direction, = 0.12, = 0.27; (b) A scan along the (1,1) direction, = 0.035,
= 0.35. From Kaminsky and co-workers, [131].
Fig. 57. Left panel: Fermi surface of the optimally doped Bi2212, measure in the
superconducting state. Indicated are the lines (1) to (5) on which the temperature
dependence is measured in the experiments by Valla et al. [247]. Typical spectra are
shown for line (2) in the normal (b) and superconducting state (c), as well as for
line (4), in the normal (d) and superconducting state (c). Right panel: momentum
widths as a function of temperature for different positions on the Fermi surface,
obtained by fitting the Momentum Distribution Curves with Lorentzian lineshapes
in the same experiments. Widths are measured at the Fermi level and at the leading
edge, in the normal and in superconducting (grey region) state, respectively.
114
(200)
Such a logarithmic correction has not yet been deciphered in the data presumably because the electronic specic heat in the normal state is less than
O(102 ) of the total measured specic heat and must be extracted by a subtraction procedure which is not suciently accurate.
The MFL self-energy, Eq. (193), has been veried in such detail in its (, T, q)
dependence that it is hard to see how any theory of CuO compounds can be
relevant to the experiments without reproducing it (or a very close approximation to it) in Region I of the phase diagram of Fig. 49. Such a scale-invariant
self-energy is characteristic of the quasi-classical regime of a QCP and indeed
the topological features of the phase diagram are consistent with their being
a QCP at xc near the composition for the highest Tc (Alternatively, a QCP in
the overdoped region where Tc vanishes is predicted in some approaches, like in
[220]). To date, no way has been found to obtain Eq. (192) except through the
scale-invariant form of uctuations which is momentum independent (z )
as in Eq. (191).
A consistent and applicable microscopic theory of the copper-oxides must show
a QCP with uctuations of the form (192). This is a very unusual requirement
for a QCP in a homogeneous extended system for at least two reasons. First,
the uctuations must have a negligible q-dependence compared to the frequency dependence, i.e. z and second, the singularity in the spectrum
should just have logarithmic form; i.e. there should be no exponentiation of
the logarithm giving rise to power laws. Such singularities do arise, as we discussed in section 3, in models of isolated impurities under certain conditions
but they disappear when the impurities are coupled; recoil kills the singularities. The requirement of negligible q/ dependence runs contrary to the idea
of critical slowing down in the uctuation regime of the usual transitions, in
which the frequency dependence of the uctuations becomes strongly peaked
at zero-frequency because the spatial correlation length diverges.
115
Another crucial thing to note is that any known QCP (in more than one
dimension) is the end of a line of continuous transitions at T = 0. Region II
(at least at T = 0) must then have a broken symmetry (This includes part of
Region III, which is also superconducting). Experiments appear to exclude any
broken translational symmetry or spin-rotational symmetry for this region 33
although as discussed in section 7.1, a sharp change in transport properties is
observed along with a four-fold symmetric diminution of the ARPES intensity
for low energies at T Tp (x). If there is indeed a broken symmetry, it is of a
very elusive kind; experiments have not yet found it.
A related question is how a momentum-independent (q, , T ) can be the
uctuation spectrum of a transition which leads to an anisotropic state as
in Region II. Furthermore, how can such a spectrum lead to an anisotropic
superconducting state because it is unavoidable that (q, , T ) of Eq. (192)
which determines the inelastic properties in Region I also be responsible for the
superconductive instability. After all, the process leading to the normal selfenergy, Fig. 53, the superconductive self-energy, and the Cooper pair vertex,
Fig. 58, are intimately related. Given Im(q, , T ) the eective interaction in
the particle-particle channel is
Vpair (k, k q) = g 2 Re (q, ) .
(201)
Re(q; ) is negative for all q and for all c c . So we do have a mechanism for superconductive pairing in the Cu-O problem given by the normal
state properties just as the normal state self-energy and transport properties
of say P b tell us about the mechanism for superconductivity in P b. In fact
given that the normal state properties give that the upper cut-of frequency is
of O(103 )K and the coupling constant g 2 N(0) is of O(1), the correct scale
of Tc is obtained. The important puzzle is, how can this mechanism produce
d-wave pairing given that (q, ) is momentum independent. How can one
obtain momentum-independent inelastic self-energy in the normal state and a
d-wave superconducting order parameter from the same uctuations?
We summarize in the next section a microscopic theory which attempts to
meet these requirements and answer some of the questions raised.
33
A new lattice symmetry due to lattice distortions or antiferromagnetism, if significant, would change the fermi-surface because the size of the Brillouin zone would
decrease. This would be visible both in ARPES measurements as well as in hall
effect measurements.
116
Fig. 58. The Cooper pair-vertex and the normal state self-energy, Fig. 53, are intimately related.
Fig. 59. Three bands result from the orbitals shown in Fig. 47(b) in a one-electron
calculation; two of these are shown. The chemical potential lies in the anti-bonding
band and is varied by the doping concentration. The other band shown is crucial
for the theory using excitonic effects as in [255].
An alternative idea from the microscopic characterization of these materials as doped Hubbard models is that a dilute concentration of holes in the
Hubbard model is likely to phase separate or form ordered one-dimensional
charge-density wave/spin-density wave structures [283,284,84]. There exists
both empirical [242] and computational [272] support for this idea at least for
a very dilute concentration of holes. For concentrations close to optimum compositions these structures appear in experiment to exist only at high energies
with short correlation lengths and times and small amplitudes in the majority
of copper-oxygen compounds. Their relation to SFL properties is again not
theoretically or empirically persuasive.
Fig. 60. The current pattern in the time-reversal breaking phase predicted for Region
II of the phase diagram.
but time-reversal symmetry and the four-fold rotation symmetry about the
Cu sites is broken. The ground state has a current pattern, sketched in Fig.
60, in which each unit cell breaks up into four plaquettes with currents in
the direction shown. The variation of the transition temperature with doping
x is similar to the line Tp (x) in Fig. 38, so that there is a QCP at x = xc .
Experiments have been proposed to look for the elusive broken symmetry
sketched in Fig. 60 [259].
The long-range interactions in the model also favor other time-reversal breaking phases which also break translational invariance. This is known from calculations on ladder models [189]. Such states have also been proposed for
copper-oxide compounds [140,121,57].
Our primary interest here is how the mechanism of transition to such a phase
produces the particular SFL uctuation spectrum (192) in Region I of the
phase diagram. The driving mechanism for the transition is the excitonic singularity, due to the scattering between the states of the partially lled conduction band c and the valence band v of Fig. 59. This scattering is of course
what we considered in sections 3.5 and 4.9 for the problem of X-ray edge
singularities for the case that the interband interaction V in Eq. (55) is small
and the valence band is dispersionless (i.e., the no recoil case). Actually, the
problem is exactly soluble for the no-recoil case even for large V [64]. For large
enough V the energy to create the exciton, ex , is less than the v-c splitting
. The excitonic lineshape is essentially the one sketched in Fig. 19(b) and
given by Eq. (56) for > ex , but is now the phase shift modulo which
is the value required to pull a bound state below . The excitonic instability
arises when ex 0.
120
Fig. 61. Sketch of the development of the particle-hole spectra in the microscopic
model for the cuprates.
The eect of a nite mass or recoil on the excitonic spectra is to smoothen the
edge singularity on the scale of the valence band dispersion between k = 0 and
k = kF . The phase shift or the interaction energy V no longer determines
the low-energy shape of the resonance. V does determine its location. The
low-energy uctuation spectra is determined by the following argument: Let
us concentrate on q = 0 which is obviously where Imex (q, ) is largest. The
absorptive part of a particle-hole spectra must be odd in frequency:
Im (q, ) = Im (q, ) .
(202)
(203)
For any nite |ex |, Re is nite and there is no instability. Only for |ex | 0,
i.e. Im(0, ) sgn, Re(0, ) is singular ln || and there is an instability.
Thus in an excitonic instability of a Fermi-sea with a dispersive valence band,
the zero-temperature spectrum has the form (, 0) ln || + i sgn at the
instability. Given a parameter p such that the instability occurs only at pc , i.e.
ex (p pc ) 0, the generalization for nite temperature T and momentum
q and p 6= pc is
i
c
(q, ) =
+ ln
max(, T, ex(p))
max(, T, ex(p))
+a2 q 2 + (pc (T ) p)
121
i1
!1
(204)
Fig. 62. Singularity of interaction aaaa between states a near the chemical potential generated by the excitonic singularity between the states a of the conduction
band and states b of the valence band. The excitonic singularity is indicated by
the shaded block.
Here c is the frequency cut-o of O(). Since the binding energy is O(1eV),
the size of the exciton, a, is of the order of the lattice constant. The q dependence of (204) is negligible compared to the frequency dependence. The
exponent z is eectively innite. At p pc to logarithmic accuracy, the above
expression (204) is identical to the phenomenological hypothesis (192).
The eective low-energy interaction for states near the chemical potential,
which is sketched in Fig. 62, is singular when the excitonic resonance is at
low frequency. Here is an example of the mechanism mentioned under (v) in
section 2.6 where the irreducible interaction obtained by integrating over nonperturbed high-energy states is singular. This is of course only possible when
the interactions represented by the shaded block in Fig. 62 are large enough.
In relation to some of the questions raised about the phenemenology at the end
of the last subsection, the momentum dependence of the vertex coupling the
low energy fermions to the uctuations in. Figs. 53 and 58 has been evaluated
[258]. It is non-local with a form depending on the wave functions of the
conduction and valence bands and the leading result is
(205)
ory. This would be accompanied by a feature in the specic heat unlike the
observations. 34
The microscopic theory reviewed above reproduces the principal features of
the phase diagram Fig. 49 of the copper-oxide superconductors, and of the SFL
properties. It also gives a mechanism for high temperature superconductivity
of the right symmetry. Further condence in the applicability of the theory to
the cuprates will rest on the observation of the predicted current pattern of
Fig. 60 in region II of the phase diagram 35 .
The distinction between metals and insulators and the metal-insulator transition has been a central problem in condensed matter physics for seven decades.
Despite the accumulated theoretical and empirical wisdom acquired over all
these years, the experimental observation made in 1995 of a metal-insulator
transition in two-dimensions [141] was a major surprise and is a subject of
great current controversy. The theoretical work in the 1980s [11,9092,149,14]
on disordered interacting electrons pointed to a major unsolved theoretical
problem in two dimensions. Infrared singularities were discovered in the scattering amplitudes which scaled to strong coupling where the theory is uncontrolled (The situation is similar to that after the singularities in the one- or
two-loop approximations in the Kondo problem were discovered revealing a
fascinating problem without providing the properties of the asymptotic low
temperature state). However the problem was not pursued and the eld lapsed
till the new experiments came along.
The 1980s theoretical work shows that this problem naturally belongs as a
subject in our study of Singular Fermi-Liquids. We will rst summarize the
principal theoretical ideas relevant to the problem before the 1995 experiments. We then briey summarize the principal results of these and subsequent experiments. Reviews of the experiments have appeared in [3,16,17].
There are two types of theoretical problems raised: the nature of the metallic
state and the mechanism of the metal-insulator transition. We will address
the former and only briey touch on the latter.
34
One might appeal to disorder to round off the transition, but this appears implausible quantitatively. More likely, the nature of the transition is strongly affected
by the Fluctuation spectra of the form of Eq. (191) and is unlikely to be of the Ising
class.
35 As already mentioned in section 2.6, ferromagnetism in some compounds has an
excitonic origin. The dynamics near such a transition should also exhibit features
of the edge-singularity as modified by recoil.
123
Fig. 63. Sketch of a MOSFET. Holes (or electrons) are trapped at the interface of
the semi-conductor and the insulator due to the band gap difference between them,
the dipole layer and the applied electric field. Two-dimensional electrons (holes)
may also be found by layered structures (heterostructures) of semi-conductors with
different band gaps such as GaAs and AlGaAs.
We consider an electron gas with a uniform positive background with no complications arising from the lattice structure this is how the many-electron
problem was originally formulated: the Jellium model. This situation is indeed
realized experimentally in MOSFETS (and heterostructures) in which an insulator is typically sandwiched between a metallic plate and a semiconductor
see Fig. 63. By applying an electric eld a two-dimensional charge layer
accumulates on the surface of the semiconductor adjacent to the insulator,
whose density can simply be changed by varying the eld strength (For details see [26]).. Similar geometries have been used to observe the Quantum
Hall Eects and the metal-insulator transition by varying the density 36 .
Typically we will be interested in phenomena when the average inter-electron
distance is O(10)O(102) nm. The thickness of the insulating layer is typically
more than 100 nm, so that the positive (capactive) charge on the insulator
provides a uniform background to a rst approximation. In Si samles, surface
roughness is the principle source of disorder at high densities, while at low
densities (in the regime where the transition takes place) ionized impurity
scattering dominates due to the fact that there is much less screening. In
GaAs, remote impurity scattering dominates, and this scattering is mostly
small-angle. This is the main reason the mobility in these samples is large.
36
124
4
h2 n
(206)
Here n is the electron density, m the band mass, and the background static
dielectric constant. We can also write
rs2 a2o =
1
,
n
(207)
which expresses that rs is the radius of the circle whose area is equal to the
area per conduction electron, measured in units of the Bohr radius a0 . For a
two-valley band structure, as on the 110 surface of Si, the kinetic energy is
reduced and rs is twice that dened by Eqs. (206) and (207).
For rs 1, (the dense electron limit) the kinetic energy dominates and metallic behavior is expected. For rs 1, the potential energy dominates and a
crystalline state (Wigner crystal) is expected. The best current numerical estimates put the transition to the crystalline state at rsc 37 [238]. The entropy
at the transition is tiny, indicating that the radial distribution function for
the liquid state at low densities is similar to that of the crystal for distances
up to a few times rs a0 .
It is important to note for our subsequent study that magnetism is always
lurking close by. Reliable numerical calculations show that the magnetic state
in the Wigner crystal near the critical density is determined by multipleparticle exchanges [46]. On the metallic side the energy of the ferromagnetic
state is only a few percent above the unpolarized metallic or crystalline states
for rs rsc [238]. Disorder is expected to make the metal-insulator state
continuous. On the insulating side at T 0, the disordered Wigner crystal
is expected to be glassy and have low-energy properties of a Coulomb glass
[224]. On the metallic side uctuations in the local density of electrons might
be expected to lead to locally polarized magnetic states or possibly to some
unusual frustrated magnetic states [56]. The perturbative calculations with
disorder and interactions, already alluded to [9092,53] also hint at formation
of magnetic moments in the metallic state. It is the interplay of such magnetic
uctuations with itinerancy which is one of the principal theoretical problem
in understanding the metallic state.
125
g = G/ e2 /h .
(208)
Next, he argued that g for a box of linear size 2L may be obtained from the
properties of a box of size L and the connection between two of them. The
conductance of a box of size L itself increases with the transition amplitude t
between energy levels in the two boxes and decreases with the characteristic
width of the distribution of the energy levels in the boxes W (L) due to the
disorder,
g(L) f
t(L)
W (L)
(209)
For weak-Gaussian disorder, the bandwidth may be expected to be proportional to the square-root of the number of impurities in the box, so W (L)
Ld/2 . The transition amplitude t is obtained by the hopping between nearneighbors near the surface of the boxes of size L. It is therefore proportional
to the surface area Ld1 . Thus 37
g(L) = f L(d2)/2 .
(210)
Now, in three dimensions the conductivity should approach a constant for large
L (Ohms law!), and hence the conductance should scale as L. This implies
37
126
that the scaling function f (x) should go for large L as f (x) x2 . Note that
while for d > 2 the g therefore increases with increasing L while for d < 2 the
large L behavior is determined by the small argument behavior of the scaling
function, while d = 2 is the marginal dimension.
In a very inuential paper, Abrahams et al. [1] analyzed the -function of the
RG ow
(g) d(ln g)/d(ln L) ,
(211)
1 1
,
2 g
(212)
where the rst part comes from Eq. (210) with f (x) x2 .
For small enough g (i.e. for large disorder) we expect exponential localization
g(L) eL/ , where is the localization length, so that (g) (L/). The
smooth connection between the perturbative result (212) for large g and the
exponentially localized solution at small g is shown in Fig. 64. While for d = 3
(or any d > 2), a critical disorder gc is required for localization, in d = 2 states
are asymptotically localized for any disorder for non-interacting fermions. The
characteristic value of the localization length in d = 2 is estimated from the
perturbative solution:
1
L
g(L) = go 2 ln
(213)
kF
2
(214)
(215)
Fig. 64. The scaling function for non-interacting electrons with disorder deduced by
Abrahams et al. [1].
(216)
replacing L then gives the nite temperature scaling behavior to which experiments may be compared.
The characteristic temperature Tx at which weak-localization eects become
prominent may be estimated in a manner similar to (213),
Tx (Tx ) = exp (kF ) .
(217)
Fig. 65. Interfering (time-reversed) parts in elastic scattering off a fixed set of impurities. The probability for the particle to arrive at B is reduced because of the
enhance probability for the particle to arrive back at A, as a result of interference.
X 2
ai
X
i
|ai |2 +
ai aj ,
(218)
i6=j
where ai is the amplitude of the i-th path. The second term in Eq. (218) is
non-zero only for classical trajectories which cross, for example at the point
O in Fig. 65. The probability of nding a particle at the point O is increased
from 2|ai |2 to
|a1 |2 + |a2 |2 + 2 Re a1 a2 = 4|a1 |2 ,
(219)
because the two paths are mutually time-reversed. Increasing this probability
of course leads to a decrease in the probability of the particle to arrive at B,
and hence to a decrease in the conductivity.
This argument makes clear why the interfering paths must be shorter than the
phase relaxation rate due to inelastic processes and why magnetic impurities
or a magnetic-eld which introduces phase-shift between two otherwise timereversed paths suppress weak-localization. In two-dimensions [14]
(H, T ) (O, T ) =
e2
1 1
+ ln x ,
+
2
2 h
2 x
(220)
(221)
(222)
Dq 2
,
i + Dq 2
q 1 and 1 .
(223)
k (k, )
m
k
(224)
holds because after impurity averaging momentum is conserved. But microscopic calculations show that at least when Fermi-Liquid theory is
valid (Cf. section 2.4)
Lim0 Limq0 impure
= Limq0 Lim0 impure
.
(225)
Indeed, if this were not so, one would not get a nite d.c. conductivity
at T = 0 for a disordered metal in d = 3. An argument for this is
as follows: Normally, we calculate the conductivity by rst taking the
limit q 0 and then the limit 0, as on the left side of (225). In
practice, however, even when we apply a homogeneous eld to a system,
the electrons in a disordered medium experience a eld which varies on
the scale of the distance between the impurities, and so the physically
relevant limit is the one on the right hand side of (225), where the limit
0 is taken rst. But the validity of (225) appears not to extend to
the case of Singular Fermi Liquids, at least for the present case where
the singularities are q-dependent. This is one of the important diculties
in developing a consistent theory for disordered interacting electrons in
d = 2.
The diusive form of the density correlation function and spin-density correlation is the genesis of the singularities due to interactions in two-dimensions.
For example the elementary eective vertex in Fig. 66 due to a bare frequencyindependent short-range interactions in two-dimensions is
v
1
Z
o
1
dq q
i + Dq 2
v 2 N(0) ln( ) .
(226)
The singularity arises because (, q) = f (/Dq 2). Recall that for pure electrons (, q) = f (/vF q) leading to a logarithmic singularity for the secondorder vertex in one-dimension and regular behavior in higher dimensions. Sim131
Fig. 66. Elementary processes important in the problem of two-dimensional disordered interacting electrons and referred to in the text. (a) Representation of the diffusion propagator due to impurity scattering vertices and corresponding self-energy.
The particle lines and hole lines should be on opposite sides of the chemical potential.
(b) Singular second-order interactions. (c) Singular vertex in the density channel
(and in the spin-density channel for the spin-conserving problem). (d) Singular (irreducible) first-order interactions. (e) Elementary singular polarization propagator.
i + Dq 2
,
=
o
(227)
Fig. 67. Effective interactions can be split up into singlet and triplet channels.
In the singlet-only channel (a), the density-density interaction is screened by the
Coulomb interaction and is universal at long wavelengths. In the triplet channel and
in the singlet channel for large momentum, the screened density-density interaction
appears only in the cross channel and is therefore non-universal.
states with one line above (particle) and the other below (hole) the chemical
potential contribute as they alone dene the physical density. This leads to
the rst order irreducible interaction and the polarization graph shown in Fig.
67 to be logarithmically singular.
For the small q of interest for singular properties, one need consider interactions only in the s-wave channel. One then has two interaction parameters,
one in the singlet channel and the other in the triplet channel.
Consider the problem with Coulomb interactions. Then the eective interaction in the singlet channel sums the polarization bubbles connected by
Coulomb interactions. Using (223) for the polarization bubble, it is shown [14]
that for small momentum transfer the interaction in the spin singlet (S = 0)
channel, Fig. 67, becomes
Vsinglet = 2 .
(228)
Fig. 68. Simplest processes contributing to the singular self-energy. (a) Exchange
process. (b) Hartree process.
1
N
,
=
ln | | ln |
|
N
4F
(Ds2 )2
C
3
1
1 F ln |T | ,
=
C
F
2
3
1
2 F ln |T | .
=
4 2
2
(229)
(230)
(231)
134
Although the theory breaks down in the strong coupling regime, this situation is
somewhat comparable to the hints that the weak coupling expansion gave in the
135
Fig. 69. Schematic renormalization group flow for the disordered interacting electron
problem according to the Finkelstein theory. dashed lines represent the effect on the
solid lines on applying a magnetic- field which couples to spins alone.
In making such a guess, the SFL properties towards which the Finkelstein
solution ows should be kept in mind:
(i) The conductivity ows towards a nite value in the theory as T 0.
(ii) The density of single-particle states ows towards zero
N() .
(232)
(iii) The magnetic susceptibility diverges at a nite length scale (the eect of
a diverging z) indicating the formation of local moments.
The last point appears to be crucial. As may be seen from Eq. (239) below,
the growth of the triplet scattering overrides the exchange processes which
favor the insulating state. Indeed if the triplet divergence is suppressed by an
applied magnetic eld, the theory reverts to the perturbative form of Eq. (220).
The scale of the magnetic-eld for this eect is given by the temperature. The
early phase of the work on the Kondo problem: these weak coupling expansions
broke down at temperatures comparable to the Kondo temperature, but did hint at
the fact that the low temperature regime was a strong coupling regime.
136
pure
(233)
Here pure is the compressibility for zero disorder, 0 = N(0), and 0 is the
Rydberg. In the Hartree-Fock approximation
0
= 1 ( 2/)rs
pure
137
(234)
The best available numerical calculations also give 0 / varying slowly enough
with rs that the correction term (233) dominates for rs of interest near the
metal-insulator transition even for modest disorder; for example for 0 10,
the disorder contribution in Eq. (233) is larger than the pure contribution
for rs 10. This has an important bearing on the Metal-Insulator transition
because the screening length s is given by
s/s0 = 0 / ,
(235)
where s0 = a0 /2. Strictly speaking s is the screening length for an external immobile charge and the screening of the electron-electron interactions is
modied from (233) due to vertex renormalizations. But in this case they do
not change the essential results. From Eq. (233) it follows that the screening
length s() > , the mean-free path, for
rs 3(0 ) .
(236)
Suppose the condition s() > L > is satised. Here L again is the size of the
box for which the calculation is done, dened through DL2 T . The assumption of screened short-range interactions, with which perturbative corrections
leading to results of Eqs. (229), (230), and (231) are obtained, is no longer
valid. In this regime the calculations must be done with unscreened Coulomb
interactions. The correction proportional to F in Eqs. (229), (230), and (231)
is not modied but the singlet contributions are more singular (due to the
extra q 1 in the momentum integrals). For instance, Eq. (231) is modied to
L
/ = ( 2/ 2 )rs .
(237)
8.6 Experiments
Soon after the publication of the theory of weak localization its predictions
were seemingly veried in experiments on Si-MOSFETS [47,245]. The experiments measured resistivity on not very clean samples of high density with
138
Fig. 70. Resistivity data on a fine scale for the two highest densities in Fig. 75 below,
showing correspondence with the theory of weak localization at such high densities.
From Pudalov et al. [202].
resistivity of O(102h/e2 ). In a limited range of temperature the predicted logarithmic rise in resistivity with decreasing temperature with about the right
prefactor was found [40]. In view of the perturbative results of Altshuler and
Aronov [14] and the knowledge that electron-electron interactions alone lead
to a Wigner insulator at low densities, one was led to the conviction that the
metallic state does not exist in two dimensions. It was expected that samples will larger rs will simply show logarithmic corrections to the resistivity
at a higher temperature and pure samples at a lower temperature. Not too
much attention was paid to Finkelsteins results which pointed to the more
interesting possibility of corrections in the opposite direction.
The more recent experiments on a variety of samples on a wider range of density and of higher purity than earlier have re-focused attention on the problem
of disorder and interactions in two dimensions and, by implication, in three dimensions as well. Several reviews of the experiments are available [3,16,17]. We
will present only a few experimental data to highlight the theoretical problems
posed, and will focus on the behavior of the data as function of temperature.
The scaling of the data as a function of the electron density n nc or eld
E will not be discussed; there is a considerable body of data on nonlinear
E-dependence (see e.g. [223] and references therein) but the signicance of
the data is not clear at present.
The rst thing to note is that results consistent with the earlier data [47,245]
are indeed obtained for high enough densities. Fig. 70 shows the resistance
versus temperature in Si for rs O(1). The magnitude of the temperature
dependence is consistent with the predictions of weak localization corrections.
As we will show below in the same region of densities, the negative magnetoresistance predicted as correction to weak-localization, discussed above, is also
139
10
10
(h/e )
10
10
10
2
(h/e )
10
10
1
0.5
10
1.5
T (K)
10
10
10
10
T (K)
Fig. 71. Resisitivity as a function of temperature for a wide range of densities (and
Fermi-energy) in a disordered Si MOSFET. The inset shows accurate measurements
of (T ) close to the speparatrix for another sample. From Sarachik and Kravchenko
[211,3].
observed.
Fig. 71 shows the resistivity as a function of T over a wide range of densities.
Similar data from [142] over a large small of densities is shown in Fig. 72,
and data over a large range of densities are plotted as a function of T /EF
in Fig. 73. The resistivity clearly shows a change of sign in the curvature as
a function of density at low temperatures. The resistivity at the cross-over
density as a function of temperature is shown down to 20mK in the inset to
Fig. 72 and is consistent with temperature independence. The true electron
temperature in these samples is a question of some controversy [16,3], but more
recent experiments, whose data are shown in Fig. 74, have corroborated these
results by studying this issue very carefully down to 5mK. In the high density
region the resistivity does rise with decreasing temperature logarithmically,
consistent with earlier measurements. The consistency of these data sets for
two very dierent types of samples therefore gives strong evidence that these
are genuine eects in both types of systems.
The data shown in Figs. 71-73 is for Si-MOSFET samples. The data for
140
Fig. 72. Resistivity versus temperature at five different electron densities in the
experiments of Kravchenko and Klapwijk [142]. The inset shows that the middle
curve (ns = 7.25 1010 cm2 ) changes by less than 5% in the entire temperature
range.
Fig. 73. Plot of the resistivity as a function of the scaled temperature T /EF . The
encircled region indicates the range of parameters explored in Fig. 72 and in [142].
The dash-dotted vertical line depicts the empirical temperature TQ = 0.007EF
below which the logarithmic temperature dependence like that of Fig. 70 sets in.
From Prinz et al. [200].
for n nc is shown in Fig. 77 and agrees quite well with the theoretical
curves as shown; similar results for n close to nc are also reported [114]. More
recent low temperature data in GaAs heterostructures[172] is reproduced in
Fig. 78 for densities n > nc but close to nc . A magnetoresistance two orders
of magnitude smaller than the weak localization theory is estimated although
the width of the negative magneto-resistance region is not in disaccord with
the weak-localization correction.
Compressibility measurements Compressibility () measurements [79,122] in
the region around n = nc show a rapid change in 1 from the negative value
characteristic of high rs metallic state to positive values see Fig. 79. These
are very important measurements which show that a thermodynamic quantity
has a very rapid variation near n nc . We have already discussed that such
changes were predicted [226] to occur through perturbative corrections due
to disorder to the energy of interacting electrons. Some more recent measurements [276] of the local electrostatic potential show that in the region n nc
143
Fig. 75. Results for resistivity versus temperature and versus magnetic field applied
in the plane for a few densities on either side of nc . The magnetic field is shown
on the upper axis and the data is taken at the lowest temperature for some of the
densities shown in the resistivity versus temperature plots. From Pudalov et al.
[201].
large scale density uctuations (puddles) occur with very weak connections
between them. Completely isolated puddles evidently have 1 = 0.
8.7 Discussion of the Experiments in light of the theory of Interacting Disordered Electrons
In comparing the experimental results with the theory, it is necessary to separate out the eects due to customary-physics for instance electron-phonon
interactions, creation of ionized impurities with temperature [15], change of
screening from its quantum to its classical form as a function of temperature
[72] change of single particle wavefunctions with a magnetic elds [137], intervalley scattering [277], etc. from the singular eects due to impurities and
interactions. The separation is at present a matter of some debate. However
it seems that the following features of the experimental data in relation to the
theoretical ideas summarized in sections 8.3, 8.4 are especially noteworthy.
144
Fig. 76. Plot of the magnetoresistance. In (a) the T dependence of in the zero
field metallic phase is shown on a semilog plot for a hole density 3.7 1010 cm2 for
varying B|| values. As B|| increases from zero, the strength of the metallic behavior
measured by the total change in from about 1K to 50mK weakens progressively,
and for B|| Bc d/dT becomes negative (i.e. the system becomes insulating).
An alternate way of demonstrating the existence of a well defined B||c is to plot
against B|| at several different temperatures. In (b) is plotted vs B|| at a hole
density 1.5 1010 cm2 . B||c is read off the crossing point marked by the arrow. In
(c), the differential resistivity dV /dI is measured at 50mK across the B|| induced
Metal-Insulator transition is shown at magnetic field strengths similar to those in
(a). From Yoon et al. [279].
145
Fig. 78. Variation of the longitudinal resistance with perpendicular magnetic field
for 2D sample at T = 9mK and at various indicated densities. The weak-localization
correction is estimated to be O(102 ) larger than the observations at these densities.
From Mills et al. [172].
146
These must be read bearing in mind our earlier discussion that most of the
interesting experiments are in a range of rs and disorder where the theoretical
problems are unresolved and only hints about the correct form of a theory are
available.
At rs O(1) and kF 1 a logarithmic increase in resistance with decreasing T consistent with weak-localization as well as with the perturbative interaction correction is observed. A positive magneto-resistance consistent with
the latter is also observed. Also observed is the correction to weak-localization
due to phase-breaking of backscattering in a perpendicular magnetic eld. The
latter yields sensible values and temperature dependence for the phase relaxation rate given by the theory. It appears that at high enough density, the weak
localization theory supplemented by the perturbative theory of interactions is
in excellent agreement with the experiments in the range of temperatures
examined.
As rs is increased (and kF decreased) the logarithmic resistance is lost in the
observed temperature range, whereas weak-localization theory predicts that
the coecient of such terms (as well as the onset temperature for their occurrence) should increase. For rs not too large the decreased logarithmic term
147
The data also led to suggestions for a superconducting ground state on the metallic side [193], and to an anyonic state [286]!
149
,
T
T
(238)
This means that the transition from the metallic state to the insulating state
can be driven by a magnetic eld. It appears that the metallic state owes
its existence to low-energy magnetic uctuations which are quenched by a
magnetic eld. This is in line with Finkelstein theory and the ow diagram of
Fig. 69 yet the existence of a scale Hc is not anticipated by the calculations
of Finkelstein (nor, of course, is the mere existence of nc ). As Hk is further
increased d/dT approaches the insulating behavior characteristic of n > nc
at Hk = 0. At a xed temperature the resistivity saturates for gB Hk EF ,
i.e. for a fully polarized band.
For small perpendicular elds, negative magneto-resistance of the form of (220)
continues to be observed at least for Si for n nc . In GaAs this contribution
at least in the range nc n 2nc is negligible.
The Hall coecient RH is continuous across the transition, obeying the
kinetic theory result RH 1/n. On the metallic side this is not surprising.
On the insulating side this is reminiscent of the properties of Wigner glasses
[60,105]
rs
Metal
Insulator
Impurity density
Fig. 80. A tentative phase diagram at T = 0 for two-dimensional disordered electrons
with interactions.
(239)
so that for small initial value to at L = one gets t (L) = to +(kF ) ln(L/).
As L = , t to + 1. The boundary between the metallic and the insulating regions on this basis is linear at small rs and small g 1, as shown in
Fig. 80. This is highly conjectural but the existence of the phase boundary at
the point rs 0 and 1/g 0 is more robust.
This scenario can be tested in high density, low disorder samples by mea151
Acknowledgements
References
[1] E. Abrahams, P. W. Anderson, D.C. Licciardo and T. V. Ramakrishnan, Scaling
Theory of Localization: Absence of Quantum Diffusion in Two Dimensions,
Phys. Rev. Lett. 42, 673 (1979).
152
153
154
155
[51] C. Castellani, C. Di Castro, P. A. Lee and M. Ma, Interaction-driven metalinsulator transitions in disordered fermion systems, Phys. Rev. B 30, 527
(1984).
[52] C. Castellani, C. Di Castro, G. Forgacs, and E. Tabet, Towards a Microscopic
Theory of the Metal-Insulator Transition, Nuclear Phys. B 225, 441 (1983).
[53] C. Castellani, C. Di Castro, P. A. Lee, M. Ma, S. Sorella, and E. Tabet Spin
fluctuations in disordered interacting electrons Phys. Rev. B 30, 1596 (1984).
[54] A. H. Castro Neto, G. Castilla, and B. A. Jones, Non-Fermi Liquid Behavior
and Griffiths Phase in f-Elecron Compounds, Phys. Rev. Lett. 81 3531 (1998).
[55] S. Chakravary, B. I. Halperin and D. Nelson, Low temperature behavior of twodimensional quantum antiferromagnets, Phys. Rev. B 39, 2344 (1989).
[56] S. Chakravarty, S. Kivelson, C. Nayak, and K. Voelker, Wigner glass, spin
liquids, and the Metal-Insulator Transition, Phil Mag. B 79, 859 (1999).
[57] S. Chakravarty, R. B. Laughlin, D. K. Morr, and C. Nayak, Hidden Order in
the Cuprates, cond-mat/0005443.
[58] S-W. Cheong, H. Y. Hwang, C. H. Chen, B. Batlog, L. W. Rupp, Jr., and
S. A. Carter Charge-ordered states in (La, Sr)2 N iO4 for hole concentrations
nh = 1/3 and 1/2, Phys. Rev. B 49, 7088 (1994).
[59] Y. H. Chen, F. Wilczek, E. Witten, and B. I. Halperin, Int. J. Mod. Phys. B 3
1001 (1989).
[60] P. Chitra, T. Giamarchi, and P. Le Doussal, Dynamical Properties of the pinned
Wigner Crystal, Phys. Rev. Lett. 80, 3827 (1998).
[61] P. T. Coleridge, R. L. Williams, Y. Feng, and P. Zawadzki, Metal-insulator
transition at B = 0 in p-type SiGe, Phys. Rev. B 56, R12764 (1997).
[62] P. T. Coleridge, A. S. Sachrajda, and P. Zawadski, Weak localisation, interaction
effects and the metallic phase of SiGe, cond-mat/9912041.
[63] P. G. Collins and P. Avouris Nanotubes for Electronics Sci. Am., Dec. 2000, 38.
[64] M. Combescot and P. Nozi`eres, Infrared Catastrophy and Excitations in the
X-ray Spectra of Metals, J. de Physique, 32, 913 (1971).
[65] M. A. Continentino, Universal behavior in heavy fermions, Phys. Rev. B 47,
11587 (1993).
[66] M. A. Continento, Quantum Scaling in Many Body Systems, Phys. Rep. 239,
179 (1994).
[67] J. Corson and J. Orenstein, Normal State Conductivity Varies as Fractional
Power of Transport Lifetime in a Cuprate Superconductor, cond-mat/0006027.
[68] D. L. Cox, Quadrupolar Kondo effect in uranium heavy-electron materials?,
Phys. Rev. Lett. 59, 1240 (1993).
156
A.M.Polyakov
and
P.wiegmann,
Phys.letters,127A,112
157
158
159
160
161
162
163
164
165
166
[236] O. Stockert, H. v. L
ohneysen, A. Rosch, N. Pyka, and M. Loewenhaupt, Twodimensional fluctuations at the quantum-critical point of CeCu6x Aux , Phys.
Rev. Lett. 80, 5627 (1998).
[237] A. Sudbo, C. M. Varma, T. Giamarchi,, E. B. Stechel, and R. T. Scalettar,
Flux quantization and paring in one-dimensional copper-oxide models, Phys.
Rev. Lett. 70, 978 (1993).
[238] B. Tanatar and D. M. Ceperley, Ground state of the two-dimensional electron
gas, Phys. Rev. B 39, 5005 (1989).
[239] C. Thessieu, J. Floquet, G. Lapertot, A. N. Stepanov, and D. Jaccard,
Magnetism and Spin Fluctuations in a Weak Itinerant Ferromagnet: MnSi,
Solid State Comm. 96, 707 (1995).
[240] D. J. Thouless, Electrons In Disordered Systems And The Theory Of
Localization Phys. Rep. 13, 93 (1974).
[241] D. Thouless, Percolation and Localization, in Ill condensed matter, Eds. R.
Balian, R. Maynard and G. Toulouse (North-Holland, Amsterdam, 1979).
[242] J. M. Tranquada, B. J. Sternlleb, J. D. Axe, Y. Nakamura, and S.
Uchida Evidence for stripe correlations of spins and holes in copper oxide
superconductors, Nature 375, 561 (1995)
[243] O. Trovarelli, C. Geibel, S. Mederle, C. Langhammer, F. M. Grosche, P.
Gegenwart, M. Lang, G. Sparn, and F. Steglich, Y bRh2 Si2 : Pronounced NonFermi-Liquid Effects above a Low-Lying Magnetic Phase Transition, Phys. Rev.
Lett. 85, 626 (2000).
[244] A. M. Tsvelik, Quantum field theory in condensed matter physics (Cambridge
University Press, Cambridge, 1995).
[245] M. J. Uren, R. A. Davies, and M. Pepper, The observation of interaction and
localisation effects in a two-dimensional electron gas at low temperatures, J.
Phys. C 13, L985 (1980).
[246] T. Valla, A. V. Fedorov, P. D. Johnson, B. O. Wells, S. L. Hulbert, Q. Li,
G. D. Gu, and N. Khoshizuka, Evidence for Quantum Critical Behavior in the
Optimally Doped Cuprate Bi2 Sr2 CaCu2 O8+ , Science 285, 2110 (1999).
[247] T. Valla, A. V. Fedorov, P. D. Johnson, Q. Li, G. D. Gu, and N. Koshizuka,
Temperature Dependent Scattering Rates at the Fermi Surface of Optimally
Doped Bi2212, cond-mat/0003407.
[248] C. M. Varma, Mixed Valence Compounds, Rev. Mod. Phys. 48 219 (1976).
[249] C. M. Varma and A. Zawadowski, Scaling in an interacting two-component
(valence-fluctuating) electron gas, Phys. Rev. B 32, 7399 (1985).
[250] C. M. Varma, Phenomenolgical Aspects of Heavy Fermions, 55, 2723 (1985).
[251] C. M. Varma, S. Schmitt-Rink and E. Abrahams, Charge Transfer Excitation
and Superconductivity in Ionic Metals, Solid State Comm. 62 681 (1987).
167
168
[269] H. v. L
ohneysen, T. Pietrus, G. Portish, H. G. Schlager, A. Schr
oder, M. Sieck,
and T. Trappmann, Non-Fermi-Liquid Behavior in a Heavy-Fermion Alloy at
a Magnetic Instability Phys. Rev. Lett. 72, 3262 (1994).
[270] F. J. Wegner, in Anderson Localisation, H. Nagaoka and H. Fukuyama, Eds.
(Springer, Berlin, 1982).
[271] X. G. Wen, Chiral Luttinger liquid and the edge excitations in the fractional
quantum Hall states, Phys. Rev. B 41, 12838 (1990).
[272] S. R. White and D. J. Scalapino Density Matrix Renormalization Group Study
of the Striped Phase in the 2D t J model, Phys. Rev. Lett. 80, 1272 (1998)
[273] K. G. Wilson, The Renormalization Group: Critical Phenomena and the Kondo
Problem, Rev. Mod. Phys. 47 , 773 (1975).
[274] N. S. Wingreen, B. L. Altshuler, and Y. Meir, Comment on 2-channel Kondo
scaling in conductance signals from 2-level tunneling systems, Phys. Rev. Lett.
75, 769 (1995).
[275] B.Wuyts,V.V.Moschalkov and Y.Bruynseraede, Resistivity and hall Effect
of metallic oxygen-deficient Y Ba2 Cu3 Ox films in the normal state,
Phys.Rev.B53,9418 (1996).
[276] A. Yacoby, private communication to CMV.
[277] Y. Yaish, O. Prus, E. Buchstab, S. Shapira, G. B. Yoseph, U. Sivan, and
A. Stern, Interband Scattering and the Metallic Phase of Two-Dimensional
Holes in GaAs/AlGaAs, Phys. Rev. Lett. 84, 4954 (2000).
[278] H Yasuoka 1997 Hyperfine Interactions 105 27 (1997).
[279] J. Yoon, C. C. Li, D. Shahar, D. C. Tsui and M. Shayegan, Parallel Magnetic
Field Induced Transition in Transport in the Dilute Two-Dimensional Hole
System in GaAs, Phys. Rev. Lett. 84, 4421 (2000).
[280] D. R. Young, D. Hall, M. E. Torelli, Z. Fisk, J.L. Sarrao, J. D. Thompson,
H.-R. Ott, S. B. Oseroff, R. G. Goodrich, and R. Zysler, High-temperature weak
ferromagnetism in a low-density free-electron gas, Nature 387 ,2219 (1999).
[281] Y. Yuval, O. Prus, E. Buchstab, S. Shapira, G. D. Yoseph, U. Sivan, and A.
Stern, Interband scattering and the metallic phase of two-dimensional holes
in GaAs/AlGaAs, Phys. Rev. Lett. 84, 4954 (2000).
[282] J. Zaanen, G.Sawatsky, and J.W. Allen, Band gaps and electronic structure of
transition-metal compounds, Phys. Rev. Lett. 55, 418 (1985).
[283] J. Zaanen and O.Gunnarson, Charged domain lines and the magnetism of highTc oxides, Phys. Rev. B 40, 7391 (1989).
[284] J. Zaanen, M. Horbach, and W. van Saarloos, Charged domain-wall dynamics
in doped antiferromagnets and spin fluctuations in cuprate superconductors,
Phys. Rev. B 53, 8671 (1996).
169
[285] J. Zaanen, Order out of disorder in a gas of elastic quantum strings in 2+1
dimensions, Phys. Rev. Lett. 84, 753 (2000).
[286] F.-C. Zhang and T. M. Rice, An Anyon Superconducting Groundstatein Si
MOSFETS?, cond-mat/9708050.
[287] G. Zarand and A. Zawadowski, Theory of tunneling centers in metallic
systems: Role of excited states and orbital Kondo effect, Phys. Rev. Lett. 72,
542 (1991).
[288] M. E. Zhitomirsky, T. M. Rice, and V. I. Anisimov, Magnetic Properties:
Ferromagnetism in the hexaborides, Nature 402, 251 (1999).
170