HAL Id: hal-01896145
https://hal.archives-ouvertes.fr/hal-01896145
Submitted on 16 Oct 2018
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Critical phenomena near the antiferromagnetic quantum
critical point of heavy fermions
M. Lavagna, Catherine Pepin
To cite this version:
M. Lavagna, Catherine Pepin. Critical phenomena near the antiferromagnetic quantum critical point of
heavy fermions. Physical Review B: Condensed Matter and Materials Physics (1998-2015), American
Physical Society, 2000. �hal-01896145�
Critical phenomena near the antiferromagnetic quantum critical point of heavy fermions
M. Lavagna1,*and C. Pe´pin2
1
Commissariat a` l’Energie Atomique, DRFMC /SPSMS, 17, rue des Martyrs, 38054 Grenoble Cedex 9, France 2Department of Physics, University of Oxford, 1 Keble Road, Oxford OX1 3NP, United Kingdom
共Received 18 January 2000; revised manuscript received 27 April 2000兲
We present a paper of the critical phenomena around the quantum critical point in heavy-fermion systems. In the framework of the S⫽1/2 Kondo lattice model, we introduce an extended decoupling scheme of the Kondo interaction, which allows one to treat the spin fluctuations and the Kondo effect on an equal footing. The calculations, developed in a self-consistent one-loop approximation, lead to the formation of a damped collective mode with a dynamic exponent z⫽2 in the case of an antiferromagnetic instability. The system displays a quantum-classical crossover at finite temperature depending on how the energy of the mode, on the scale of the magnetic correlation length, compares to kBT. The low-temperature behavior, in the different regimes separated by the crossover temperatures, is then discussed for both two- and three-dimensional systems.
I. INTRODUCTION
A central issue in the thermal properties of heavy-fermion
compounds is the discovery of a non-Fermi liquid 共NFL兲
behavior in systems close to an antiferromagnetic quantum
phase transition.1 Such a behavior has been discovered in a
series of compounds containing cerium or uranium, for
example, CeCu6⫺xRx (R⫽Au,Ag),
2
CeIn3, CePd2Si2, 3
CeNi2Ge2,4 U1⫺xYxPd3,5 and CexLa1⫺xRu2Si2.6 In
CeCu5.9Au0.1, the specific heat C depends on T as C/T
⬃⫺ln(T/T0), the magnetic susceptibility as⬃1⫺␣
冑
T, andthe T-dependent part of the resistivity as⌬⬃T 共instead of
C/T⬃⬃const and ⌬⬃T2 in the Fermi-liquid state兲. Most interestingly, the breakdown of the Fermi-liquid behavior
can be tuned by alloying共chemical pressure兲 or by applying
a hydrostatic pressure or a magnetic field. The origin of this NFL behavior is a problem of current debate.
Different scenarios have been proposed in order to
eluci-date the NFL behavior at the quantum critical point 共QCP兲.
On the one hand, there are scenarios involving strong disor-der to account for the non-Fermi liquid properties, including
共i兲 a distribution of the Kondo temperature P(TK) 共Ref. 7兲
induced by disorder following a distribution of the local density-of-states, and共ii兲 the possibility of the formation of a
Griffiths phase at the quantum critical point.8,9On the other
hand, starting from a clean or rather weakly disordered
sys-tem, there is a scenario 共iii兲 referring to the proximity of a
quantum critical point in a theory of itinerant
antiferromagnetism,10–17 in which the Fermi surface is
coupled to critical collective antiferromagnetic excitations.
An alternative possibility has been recently put forward18on
the basis of an analysis of the neutron scattering
measure-ments in CeCu5.9Au0.1. The unusual spin dynamics that has
been observed suggests 共iv兲 the existence of critically
screened local magnetic excitations due to the formation of
local moments. Finally, 共v兲 a variant of the quadrupolar
Kondo effect19 in which the interactions of the rare-earth
共actinide兲 internal degrees of freedom with the conduction
electrons are reduced to a single impurity multichannel
Kondo effect has been proposed. As it stands, the situation remains highly controversial.
In this paper, we adopt the point of view共iii兲 of the
prox-imity to a quantum critical point. The effect of a nonzero temperature on quantum critical points is a long-standing problem encountered in itinerant magnetism. We refer to the
papers of Hertz20 and Millis11 using renormalization-group
techniques, and of Moriya21 introducing a self-consistent
renormalization 共SCR兲 theory of spin fluctuations. These
theories yield a rich phase diagram with, in addition to the
ordering temperature TN, a series of crossover temperatures
separating different regimes of behaviors. They are well adapted for the treatment of spin fluctuations as present in the single-band Hubbard model but essentially do not ac-count for the Kondo effect.
Indeed, the model that is believed to describe the physics
of heavy fermions is the Kondo lattice model22derived from
the periodic Anderson model in a given limit. The large N
expansions,23which have been carried out for these models
共where N represents simultaneously the degeneracy of the
conduction electrons and of the spin channels兲 give a good
description of the Kondo effect. But, unlike the above-mentioned spin-fluctuation theories, they fail to account for the spin fluctuations since the Ruderman-Kittel-Kasuya-Yosida interactions only appear at the order 1/N2.24
Hence our motivation to set up an approach of the S
⫽1/2 Kondo lattice model (N⫽2) that enlarges the standard
1/N expansion theories up on the spin-fluctuation effects.
Our previous paper25 presented such an approach by
per-forming a generalized Hubbard-Stratonovich transformation on the Kondo interaction term of the Hamiltonian, which
makes three types of fields 共magnetization densities and
Kondo type兲 appear on an equal footing. The dynamical spin
susceptibility was derived in a one-loop expansion associated with the Gaussian fluctuations of the magnetization density fields around their saddle-point values.
With the aim to describe the critical phenomena around the quantum critical point of the heavy-fermion systems, we
propose to push the treatment of the S⫽1/2 Kondo lattice
model in a self-consistent one-loop approximation in
com-PRB 62
plete analogy with the SCR theory of spin fluctuations
de-veloped for the Hubbard model.21,26,27As in itinerant
mag-netism, we will show how the system displays a quantum-classical crossover at finite temperature depending whether the temperature is lower or higher than the characteristic en-ergy scale of the collective mode associated with the mag-netic instability. The low-temperature behavior in the differ-ent regimes essdiffer-entially depends on the value of the dynamic exponent z of the mode and on the dimensionality d of the problem. In the case of an antiferromagnetic instability, z is found to be equal to 2. Heavy-fermion systems are usually believed to be three dimensional. However, a recent
proposal28 based on the neutron-scattering data obtained in
CeCu6⫺xAux, stipulates that the critical magnetic
fluctua-tions, which the quasiparticles are coupled to, are simply two
dimensional. So both situations d⫽2 and d⫽3 will be
con-sidered.
The rest of the paper is organized as follows. In Sec. II, we introduce the self-consistent one-loop approximation method for the Kondo lattice model using a generalized Hubbard-Stratonovich transformation on the Kondo interac-tion term. The approach combines both aspects of the 1/N expansion and of the self-consistent renormalization theory of spin fluctuations. The calculations, presented in the func-tional integral formalism, closely follow the presentation
made by Hertz and Klenin27for the Hubbard model. Hence,
we extract the general expression of the dynamical spin
sus-ceptibility extending the results of the previous paper25 to a
self-consistent treatment. Section III gives a discussion on the nature of the damped collective mode associated to the proximity of the magnetic transition with a dynamic
expo-nent z⫽2 in the antiferromagnetic case. In Sec. IV, we
de-rive the resulting quantum-classical crossover. This occurs at a finite temperature T, which depends on how the energy of
the mode, on the scale of the magnetic correlation length
compares to kBT. The low-temperature behavior of the
sys-tem in the different regimes, separated by the crossover sys- tem-peratures, is then discussed in Sec. V. Concluding remarks follow in Sec. VI.
II. SELF-CONSISTENT ONE-LOOP APPROXIMATION TO THE KONDO LATTICE MODEL
This section extends the approach introduced in our
pre-vious paper25 to a self-consistent one-loop approximation.
Our starting point is a path-integral treatment of the Kondo lattice model Hamiltonian
H⫽
兺
k k ck†ck⫹J兺
i Sf i•兺
⬘ci † ⬘ci⬘ ⫺冉
兺
i ci†ci⫺ncNS冊
, 共1兲where Sf i represents the spin (S⫽1/2) of the impurities
dis-tributed on the sites共in number NS) of a periodic lattice; ck †
is the creation operator of the conduction electron of
momen-tum k and spin quanmomen-tum number characterized by the
en-ergy ⑀k⫽⫺兺具i, j典ti jexp(ik•Ri j); are the Pauli matrices
(x,y,z) and0the unit matrix; J is the antiferromagnetic
Kondo interaction (J⬎0). The Hamiltonian is written in the
grand canonical ensemble with the introduction of a
chemi-cal potential in order to guarantee an average number of
conduction electrons per site nc⬍1. The spin of the
impuri-ties is treated within the Abrikosov pseudofermion represen-tation Sf i⫽兺⬘fi
†
⬘fi⬘ where the local constraint Qi ⫽兺fi⫹fi⫺1⫽0 is implemented in the usual way, by
in-troducing a time-independent Lagrange multiplier i. The
local Kondo interaction is decoupled using a Hubbard-Stratonovich transformation, which introduces four auxiliary fields ⌽i(), ⌽i*(), and fi(), ci(). We are then left
with the effective action
Se f f⫽
冕
0
d关L0共兲⫹HJ共兲兴,
whereL0() is the Lagrangian in the absence of the
interac-tion J and HJ共兲⫽
兺
i⬘ 关ci †共兲f i † 共兲兴 ⫻冉
⫺JS ifi共兲•⬘ JC⌽i*共兲⬘ 0 JC⌽i共兲 ⬘ 0 ⫺J Sici共兲•⬘冊
⫻冉
ci⬘共兲 fi⬘共兲冊
⫹JC兺
i ⌽i *共兲⌽i共兲 ⫹JS兺
i fi共兲•ci共兲, 共2兲with JS⫽J/4 and JC⫽J/3. Details on the saddle-point
solu-tion and the Gaussian fluctuasolu-tions can be found in our
pre-vious paper.25The saddle-point solution transposes to N⫽2
the large-N results obtained within the slave-boson mean-field theories. It gives rise to the formation of a Abrikosov-Suhl resonance pinned at the Fermi level and split by a
hy-bridization gap defining two bands ␣ and. The one-loop
approximation is then obtained by considering the Gaussian fluctuations of the bosonic fields around their mean-field val-ues. As in the conventional approaches to critical phenom-ena, the approach consists to integrate out the fermion fields and thereby reduce the problem to the study of an effective bosonic theory describing fluctuations of the ordering fields.
But contrary to Hertz20and Millis,11who considered a single
field corresponding to the magnetization density of one type of electron, the approach, starting from a different micro-scopic model, extends the description to the case of several fields, i.e., the Kondo fields on the one hand, and the f- and c- magnetization densities on the other hand.
We now propose to include the self-consistent effects in the one-loop approximation. We do it in a variational way and replace the full expression Se f f⫺Se f f
(0) 共where S e f f (0)
and F(0)are the zeroth order effective action and the related free energy兲 by a trial quadratic form ⌬S˜e f f
(2)⫽S˜ e f f (2)⫺S e f f (0) in the fields␦rq,, ␦q,, ␦fq,, and␦cq,working in the radial
gauge for the Kondo parameters29 using Fourier transform
notations
⌬S˜e f f (2)⫽1  q,i
兺
冋
共␦ rq,␦q,兲D˜C⫺1共q,i兲冉
␦r⫺q,⫺ ␦⫺q,⫺冊
⫹共␦f q, z ␦ cq, z 兲D˜ S 储⫺1共q,i 兲冉
␦$xizf⫺q,⫺ ␦c⫺q,⫺ z冊
⫹共␦⫹f q,␦cq,⫹ 兲D˜⬜⫺1S 共q,i兲冉
␦f⫺⫺q,⫺ ␦c⫺⫺q,⫺冊
⫹共␦⫺f q,␦cq,⫺ 兲D˜⬜⫺1S 共q,i兲冉
␦f⫹⫺q,⫺ ␦c⫹⫺q,⫺冊册
, 共3兲where the indices q and , respectively, stand for the wave
vector q and the bosonic Matsubara frequencies i
⫽2i/. Note that Eq.共3兲 is written using the radial gauge
in which the phase of the fields ⌽i and⌽i* is incorporated
into the Lagrange multiplieri, which turns out to be a field.
The elements of the matrices D˜C⫺1(q,i), D˜S储⫺1(q,i)
and D˜⬜⫺1S (q,i) are determined variationally by minimiz-ing the free energy or, more precisely, the upper bound to the free energy according to the Feynman variational principle F⭐F˜e f f(2)⫹1/
具
Se f f⫺S˜e f f(2)
典
where F˜e f f(2) represents the free en-ergy related to S˜e f f(2) and
具
A典
means the average of the quan-tity A over the distribution of fields exp(⫺⌬S˜ef f(2)), i.e.,具
A典
⫽冕
DriDiDfiDciA exp共⫺⌬S˜e f f (2)兲冕
DriDiDfiDci)exp共⫺⌬S˜e f f (2)兲 . 共4兲After minimization, one can show that关DC⫺1(q,i)兴ris
equal to (1/2)2Se f f/(rq,⫺q,⫺) averaged over the
mentioned distribution of fields 共and equivalent expressions
for the other elements兲. Explicitly we obtain,
D˜C⫺1共q,i兲⫽
冉
JC关1⫺JC共
具
¯2共q,i兲典
⫹具
¯m共q,i兲典
兲兴 ⫺JC具
¯1共q,i兲典
⫺JC具
¯1共q,i兲典
⫺具
¯f f共q,i兲典
冊
, 共5兲
D˜S储⫺1共q,i兲⫽
冉
JS2
具
cc储 共q,i兲典
JS关1⫹JS具
储f c共q,i兲典
兴JS关1⫹JS
具
c f储 共q,i兲典
兴 JS2具
f f储 共q,i兲典
冊
,and an equivalent expression for the transverse spin part D˜⬜⫺1S (q,i). The different bubbles
具
(q,i)典
in the latterexpressions represent the susceptibilities (q,i) 共cf.
appendix兲 averaged over the distribution of fields
exp(⫺⌬S˜ef f(2)). At that point, the problem is highly complex and cannot be solved exactly. Therefore, we introduce a local and instantaneous approximation that is equivalent to
aver-aging over Gaussian distributions of local fields
(ri,i,f
i,ci) with variancesr, , f, and cdefined
as f 2⫽3
兺
q,i 关D ˜ S 储共q,i 兲f f兴, 共6兲and similar expressions for the three other variances. Equa-tion 共6兲 is obtained from Eq. 共3兲 by letting the summation
over (q,i) act on the matrix elements only, hence defining
the local and instantaneous fluctuations of the various de-grees of freedom. One expects that at low temperatures, the fluctuations of ri, i, andci only bring T2 corrections to
their saddle-point values. Those corrections are negligible compared to the T dependence brought by the fluctuations of
fiand will not be considered here. So we will neglect, in the
following, the fluctuations of ri, i, and ci, and focus on
the fluctuations offi only. Then
具
(q,i)典
is simply givenby
具
共q,i兲典
⫽冉
冕
Dfi共q,i兲exp关⫺fi 2 /共22f兲兴冊
冉
冕
Dfiexp关⫺fi 2/共2 f 2兲兴冊
. 共7兲Equations 共6兲–共7兲 are the basic equations of the paper
from which the dynamical spin susceptibility is extracted and the critical phenomena around the QCP are studied. Let us note that, when the averages acting on the various bubbles
(q,i) in Eq.共5兲 are omitted, one recovers the standard
results of the random-phase approximation. As pointed out
by Lonzarich and Taillefer,30 the corresponding
self-consistent renormalized spin fluctuation SCR-SF procedure can be analyzed within the Ginzburg-Landau formalism in which the local free energy is expanded in terms of a small
and slowly varying order parameter m(r) as f (r)⫽ f0
⫹1 2a兩m(r)兩 2⫹1 4b兩m(r)兩 4⫹1 2c兺i兩“m i(r)兩2⫹O(6) and
the following approximation is made 兩m(r)兩4
⫽
具
兩m(r)兩2典
兩m(r)兩2. Hence具
兩m(r)兩2典
is evaluated via thefluctuation-dissipation theorem.
The expression of the dynamical spin susceptibility is then derived in the usual way, by adding a source term
⫺2Sf•B to the Hamiltonian. In the Matsubara
f f共q,i兲⫽
具
f f共q,i兲典
1⫺JS2
冋
具
f f共q,i兲典具
cc共q,i兲典
⫺具
f c共q,i兲典
2⫺2
JS
具
f c共q,i兲典
册
. 共8兲
In the low-frequency limit, one can easily check that the dynamical spin susceptibility may be written in terms of intra- and interband suceptibilities corresponding,
respec-tively, to particle-hole pair excitations within the lower ␣
band and from the lower␣ to the upper band
f f共q,i兲⫽
具
¯␣␣共q,i兲典
⫹具
¯␣共q,i兲典
1⫺JS2
具
␣␣共q,i兲典具
¯␣共q,i兲典
. 共9兲The expressions of the susceptibilities ␣␣(q,i) and
¯
␣(q,i), corresponding to intra- and interband
particle-hole excitations are given in the Appendix. The latter expres-sion is reminiscent of the behavior proposed by Bernhoeft
and Lonzarich31to explain the neutron scattering observed in
U Pt3 with the existence of both a ‘‘slow’’ and a ‘‘fast’’
component in
⬙
(q,)/due to spin-orbit coupling. Also ina phenomenological way, the same type of feature has been suggested in the duality model developed by Kuramoto and
Miyake.32 To our knowledge, the proposed approach
pro-vides the first microscopic derivation from the Kondo lattice model of such a behavior. Expanding the various
suscepti-bilities up to the second order in fi, and making use of
Eq.共7兲, one can draw from Eq. 共9兲 the following expression
of the dynamical spin susceptibility 共taking the analytical
continuation i→⫹i␦) f f共q,兲⫽ ␣␣共q,兲⫹¯␣共q,兲 1⫺JS 2 ␣␣共q,兲¯␣共q,兲⫹f 2, 共10兲
where is a constant which can be evaluated from the
ex-pansion of
具
␣␣(q,i)典
and具
¯␣(q,i)典
up to the second order in2f. The variancef2defined by Eq.共6兲 can as well be expressed as a function off f(q,i)f 2⫽
兺
q,i f f共q,i兲⫽ 3 兺
q冕
0 ⫹⬁ coth 2 f f⬙
共q,兲d. 共11兲The two last equations 关Eqs. 共10兲–共11兲兴 provide a
self-consistent determination of2f and hence of the spin
suscep-tibility. We will successively present now共i兲 in Sec. III, the
result for the dynamical spin susceptibility in the
random-phase approximation共RPA兲 giving information on the nature
of the collective mode that is involved;共ii兲 then in Sec. IV,
the implications of that mode in the critical phenomena around the quantum critical point.
III. RPA-DYNAMICAL SPIN SUSCEPTIBILITY NEAR THE ANTIFERROMAGNETIC WAVE VECTOR:
DISCUSSION ON THE NATURE OF THE COLLECTIVE MODE
As pointed out in Sec. II, the RPA corresponds to the absence of any averaging in the different equations. Thereby the expression of the dynamical spin susceptibility at the
RPA level is given by Eq. 共10兲 takingf2⫽0. We now
dis-cuss the q and dependence of the RPA dynamical spin
susceptibility around the antiferromagnetic wave vector Q.
The bare intraband susceptibility ␣␣(Q⫹q,) is well
approximated at 兩q兩Ⰶ兩Q兩 by a Lorentzian ␣␣(Q⫹q,)
⫽␣␣/(1⫺i/⌫0⫹bq
2) where
␣␣⫽␣␣
⬘
(Q,0) and ⌫0 isthe relaxation rate of order 兩yF兩⫽TK. As expected for the
antiferromagnetic case, the relaxation rate remains finite when q goes to zero reflecting the fact that the fluctuations of the order parameter are not conserved. The bare interband susceptibility¯␣(Q⫹q,) can be considered as purely real
and frequency independent equal to ␣ in the
low-frequency limit. Due to the presence of the hybridization gap in the density-of-states, the continuum of the interband
particle-hole excitations displays a gap ranging from 20,
the value of the direct gap for the wave vector 0, to 2兩yF兩,
the value of the indirect gap at Q. The RPA dynamical spin susceptibility is given by f f R PA⬙共Q⫹q,兲⫽f f
⬘
共Q⫹q兲⌫共q兲 2⫹⌫共q兲2 共12兲 with f f⬘
共Q⫹q兲⫽ ␣␣⫹␣ 共1⫺Iq兲 , ⌫共q兲⫽⌫0共1⫺Iq兲,where Iq⫽I⫺bq2 and I⫽JS
2
␣␣␣ 共of order 1 near the
antiferromagnetic instability兲.
The bulk staggered susceptibility f f
⬘
(Q) diverges at I⫽1. The frequency dependence of f f R PA⬙
(Q⫹q,) is a
Lorentzian with a vanishing relaxation rate⌫(0) at the
anti-ferromagnetic transition. The corresponding excitation can be analyzed as an antiferromagnetic paramagnon mode that
softens at the magnetic transition. Equation共12兲 provides us
with the dispersion of that mode. The corresponding dynami-cal exponent z is found to be equal to 2 due to the
overdamp-ing of the mode when enters the continuum of the ␣-␣
particle-hole excitations. We will show how the dynamic ex-ponent z strongly affects the static critical behavior. This is
due to the fact that, contrary to the T⬎0 phase transitions,
statics and dynamics are intimately linked in quantum (T
⫽0) phase transitions. We turn now to the resulting
quantum-classical crossover occurring at finite temperature.
IV. CRITICAL PHENOMENA AROUND THE ANTIFERROMAGNETIC QUANTUM
CRITICAL POINT„AF-QCP…
The parameter which controls the temperature depen-dence of the thermodynamical variables near the AF-QCP is
f
2, which expresses the thermal local fluctuations of the
staggered magnetization. Using Eqs.共11兲 and 共12兲
f 2⫽
兺
q
S共Q⫹q兲⬃
冕
S共Q⫹q兲qd⫺1dq, 共13兲where S(Q⫹q) is the static form factor
S共Q⫹q兲⫽3
冕
0⬁
coth共/2兲
⬙
f f共Q⫹q,兲d. 共14兲Depending on the temperature scale considered, quantum or classical behavior is observed. In general, the thermal
fluctuations f2 of the magnetization should depend on the
total self-consistent renormalized dynamical spin susceptibil-ity
⬙
f f(Q⫹q,) as defined in Eq.共9兲. However, in practice,we will be content with its truncated form R PAf f ⬙(Q⫹q,) derived at the RPA level. The temperature dependence of the
static form factor S(Q⫹q) is given by
S共Q⫹q兲⫽6T
⬘
f f共Q⫹q兲tan⫺1冉
T
⌫共q兲
冊
. 共15兲A first crossover temperature TI⫽2⌫0(1⫺I) appears, which
separates the quantum from the classical regime.
共i兲 T⬍TI: quantum regime. One can show that the static
form factor S(Q⫹q) exhibits a quadratic temperature
depen-dence S共Q⫹q兲⫽6
⬘
f f共Q⫹q兲 ⌫0共1⫺Iq兲 T2, 共16兲The thermal local fluctuations f2 of the magnetization are
also quadratic as a function of the temperature. They can be
easily evaluated from Eq.共15兲. Taking advantage of the fact
that the integrand S(Q⫹q)qd⫺1 is peaked at q1
⫽
冑
(1⫺I)/b, the temperature dependence of f2 is also found to be quadratic f2⫽ 3 2 ⬘
f f共Q兲 ⌫0共1⫺I兲 T2, 共17兲The physics is quantum in the sense that the fluctuations on
the scale of the magnetic correlation length have energy
much greater than the temperature kBT.
共ii兲 T⬎TI: classical regime. In this regime, the static
form factor S(Q⫹q) shows a linear temperature
depen-dence. As long as q is smaller than the thermal cutoff q*
defined bymax(q*)⫽T, S(Q⫹q) takes the following form
S共Q⫹q兲⫽3
⬘
f f共Q⫹q兲T. 共18兲The thermal local fluctuations f2 of the magnetization are
deduced from Eq. 共15兲. The main contribution arises from
the integration over 关q1,q*兴. The result depends on the
di-mensionality d. One finds f2⬃T3/2 at d⫽3 and
f 2
⬃T ln T at d⫽2. This classical regime corresponds to the
case where the energy of the mode on the scale ofbecomes
less than kBT.
Table I recapitulates the expressions that we get for the
thermal local fluctuations 2f of the staggered magnetization
depending on the temperature scale and on the
dimensional-ity. Following Eq. 共10兲, the staggered spin susceptibility is
given by
Q
⬘
⫽f f 0⬘共Q兲
1⫺I⫹f2, 共19兲
where f f0⬘(Q)⫽␣␣⫹␣. Incorporating the values of the
thermal local fluctuations f2 reported in Table I, we are
able to identify a second crossover temperature TII. Above
TII, f2 becomes larger than (1⫺I) and all the physical
quantities are controlled by the temperature only. TII⬃(1
⫺I)2/3at d⫽3 and T
II⬃(1⫺I) at d⫽2. As expected in the
two-dimensional case, the T⫽0 transition is at its upper
critical dimension since z⫹d⫽4 in the antiferromagnetic
case characterized by a dynamic exponent z⫽2. That
corsponds to the marginal case for which the intermediate
re-gime TI⬍T⬍TIIis squeezed out of existence. The results for
the d⫽3 and d⫽2 cases are, respectively, summarized in
Figs. 1 and 2, which picture the different regimes of behav-iors reached depending on the values of the temperature T
and of the control parameter I⫽J2␣␣␣. When I⬎1,
long-range antiferromagnetic order occurs when T is smaller
than the Ne´el temperature TN. TN⬃(I⫺1)2/3 at d⫽3 and
TN⬃(I⫺1) at d⫽2.
V. DISCUSSION
The low-temperature physics is dominated by the pres-ence of an antiferromagnetic paramagnon mode that softens at the magnetic transition. The results depend crucially on
the value of d⫹z, where d is the spatial dimension and z is
the dynamic exponent associated to that mode. Due to the overdamping of the mode when it enters the continuum of intraband particle-hole excitations, the dynamic exponent z is found to be equal to 2 in the antiferromagnetic case. As far as the dimensionality is concerned, there is no doubt about the dimensionality associated with the heavy quasiparticles in those systems that are clearly of 3. However the question about the dimensionality of the critical magnetic fluctuations, which the quasiparticles are coupled to, is still an open problem. Up to now, it has been assumed that the critical
fluctuations of the quantum phase transition 共QPT兲 are
dominated by the three-dimensional antiferromagnetic
TABLE I. Predictions for thermal local fluctuations f 2
of the staggered magnetization depending on the temperature T and on the dimensionality d. TI⬃(1⫺I) represents the first crossover tempera-ture separating the quantum from the classical regime in the vicinity of the antiferromagnetic-quantum critical point共AF-QCP兲.
Quantum regime T⬍TI Classical regime T⬎TI d⫽3 f2⬃⬘f f(Q)T2/⌫0 f 2 ⬃T3/2 d⫽2 f 2⬃ f f ⬘(Q)T2/⌫0 f 2⬃T ln T
correlations. This would impose a description by a quantum
critical theory with d⫽3, z⫽2. Recently Rosch et al.28
pro-posed, on the basis of the neutron-scattering data in CeCu6⫺xAux, that the critical magnetic fluctuations are two
dimensional, which leads to a QCP with d⫽2 and z⫽2. Let
us summarize the different regimes of behaviors that we get
in each of those two cases (z⫽2, d⫽3 or 2兲 depending on
the values of the temperature T and of the control parameter I⫽J2␣␣␣. As one can see, the results that we obtain are
very similar to those established by Hertz20and Millis11
us-ing renormalization-group approaches in the spin-fluctuation theory.
共i兲 Case d⫽3. A long-range antiferromagnetic phase
oc-curs when I⬎1 below the Ne´el temperature TN⬃(I⫺1)2/3.
A first crossover temperature TI⬃(1⫺I) separates the
quan-tum from the classical regime. In the regime I(I⬍1) and T
⬍TI), the physics is quantum in the sense that the energy of
the relevant mode on the scale of is greater than the
tem-perature kBT. Since d⫹z⬎4, the T⫽0 phase transition is
above its upper critical dimension and the various physical quantities depend upon the parameter I. The magnetic
corre-lation length diverges at the magnetic transition
⬃1/
冑
(1⫺I) and the staggered spin susceptibility behaves asQ
⬘
⫽f f 0⬘(Q)/(1⫺I⫹aT2). Regimes II and III are both
classical regimes characterized by large thermal effects since
the fluctuations on the scale of have energy much smaller
than kBT. In regime II关TI⬍T⬍TIIwith TII⬃(1⫺I)2/3兴, the
magnetic correlation length ⬃1/
冑
(1⫺I) is still controlledby (1⫺I) even though modes at the scale ofhave energies
less than kBT. On the contrary, the staggered spin
suscepti-bility is sensitive to the thermal fluctuations Q
⬘
⫽f f 0⬘
(Q)/(1⫺I⫹aT3/2). In regime III(T⬎TII), the thermal
dependence of physical quantities becomes universal and
both andQ
⬘
are controlled by the temperature: ⬃1/T3/4andQ
⬘
⫽f f 0⬘(Q)/T3/2.
共ii兲 Case d⫽2. Then, since z⫽2, z⫹d⫽4 and the T ⫽0 phase transition is at its upper critical dimension. The
physics is qualitatively similar to the case d⫽3 with stronger
fluctuation effects particularly in the classical regime. A
long-range antiferromagnetic phase occurs when I⬎1 below
the Ne´el temperature TN⬃(I⫺1). For d⫽2, the two
cross-over temperatures TI and TII coincide so that regime II of
Fig. 1 is squeezed out of existence. Regime I 关T⬍TI with
TI⬃(1⫺I)兴 is the quantum regime in which the thermal
fluctuations are negligible : ⬃1/
冑
(1⫺I) and Q⬘
⫽f f
0⬘(Q)/(1⫺I⫹aT2). Regime III (T⬎T
I) is the unique
classical regime characterized by very strong fluctuation ef-fects since kBT is larger than the energy of the relevant mode
on the scale of . Both the magnetic correlation length and
the staggered spin susceptibility are then controlled by the temperature:⬃1/
冑
T ln T andQ⬘
⫽f f0⬘(Q)/(T ln T).VI. CONCLUDING REMARKS
We considered the S⫽1/2 Kondo lattice model in a
self-consistent one-loop approximation starting from a general-ized Hubbard-Stratonovich transformation of the Kondo in-teraction term. The model exhibits a zero-temperature magnetic phase transition at a critical value of the Kondo coupling. The transition is usually antiferromagnetic but it may be incommensurate depending on the band structure considered. The low-temperature physics is controlled by a collective mode that softens at the antiferromagnetic transi-tion with a dynamic exponent z equal to 2.
A quantum-classical crossover occurs at a temperature TI
related to the characteristic energy scale of that mode. Heavy-fermion systems are usually believed to be three-dimensional. However, since some recent
inelastic-neutron-scattering experiments performed in CeCu6⫺xAux show that
FIG. 1. Phase diagram in the plane (T,I) for dimension equal to 3. The shaded region represents the long-range antiferromagnetic phase bordered by the Ne´el temperature TN. The unshaded region
marks the magnetically-disordered regimes I, II, and III, associ-ated to different behaviors of the system. Regime I is the quantum regime in which the energy of the relevant mode on the scale of is much greater than kBT. The magnetic correlation length is ⬃1/
冑
(1⫺I) and the staggered spin susceptibility is Q⬘ ⫽f f0⬘
(Q)/(1⫺I⫹aT2). Regimes II and III are both classical
re-gimes in which the thermal effects are important since the fluctua-tions on the scale of have energy much smaller than kBT. In
Regime II,⬃1/
冑
(1⫺I) is still controlled by (1⫺I) but the stag-gered spin susceptibility is sensitive to the thermal fluctuations:Q⬘⫽f f
0⬘
(Q)/(1⫺I⫹aT3/2). In Regime III, both and
Q
⬘ are con-trolled by the temperature:⬃1/T3/4andQ⬘⫽f f
0⬘
(Q)/T3/2.
FIG. 2. Phase diagram in the plane (T,I) for dimension equal to 2. The shaded region represents the long-range antiferromagnetic phase bordered by the Ne´el temperature TN. The unshaded region
marks the magnetically-disordered regimes I and III. In that d⫽2 case, the equivalent of regime II in Fig. 1 is squeezed out of exis-tence since TI⫽TII. Regime I is the quantum regime in which the
thermal fluctuations are negligible: ⬃1/
冑
(1⫺I) and Q⬘ ⫽f f0⬘
(Q)/(1⫺I⫹aT2). Regime III is the unique classical regime in which both the magnetic correlation length and the staggered spin susceptibility are controlled by the temperature:⬃1/
冑
T ln TandQ⬘⫽f f
0⬘
(Q)/(T ln T).
the critical magnetic fluctuations, which the quasiparticles are coupled to, are two-dimensional, we both considered the
cases d⫽2 and d⫽3 with z⫽2. The low-temperature
behav-ior of the system is deduced, with predictions for the tem-perature dependence of the physical quantities such as the
magnetic correlation lengthand the staggered susceptibility
Q
⬘
. A second crossover temperature TII appears abovewhich the thermal dependence of andQ
⬘
becomesuniver-sal and is uniquely controlled by the temperature. It would be very interesting now to study the temperature behavior of other physical quantities as the specific heat or the transport properties.
A number of significant issues remain to be addressed. Among others, we will mention the possibility of getting a
different value of the dynamic exponent for instance z⫽1 if
nesting effects are considered corresponding to an absence of damping of the mode when located in the gap of excitations around the antiferromagnetic vector. Disorder is expected to play a crucial role in this problem. In that direction, some
recent works33,17 pointed out the formation of some ‘‘hot
lines,’’ i.e., points at the Fermi surface linked by the mag-netic Q vector. On the hot lines, the quasiparticle scattering rate is linear in temperature while, away from the hot lines, it
acquires the standard Fermi liquid form in T2. At low
tem-perature, the ‘‘cold’’ regions are shown to short circuit the
scattering near those lines, and induce a T2 behavior of the
resistivity. The disorder is found to emphasize the contribu-tion of the hot lines, eventually leading to a non-Fermi liquid behavior. It would be worthy to study the effects of the ‘‘hot lines’’ in the presence of disorder within the critical phenom-ena description presented in this paper.
ACKNOWLEDGMENTS
We would like to thank Nick Bernhoeft, Piers Coleman, Mucio Continentino, Jacques Flouquet, Patrick A. Lee, and Gilbert Lonzarich for very useful discussions on this work.
APPENDIX EXPRESSIONS OF THE DIFFERENT BUBBLES
The expressions of the different bubbles appearing in the
expression of the boson propagators 关cf. Eq. 共5兲兴 are given
here共with i⫽1, 2, m, or ff兲
¯
i共q,i兲⫽i共q,i兲⫹i共⫺q,⫺i兲, 共A1兲
1共q,i兲⫽⫺
1
 k,i
兺
nGc f0共k⫹q,in⫹i兲Gf f0共k,in兲,
2共q,i兲⫽⫺ 1  k,i
兺
n Gcc 0 共k⫹q,i n⫹i兲Gf f0共k,in兲, m共q,i兲⫽⫺ 1  k,i兺
n Gc f 0 共k⫹q,i n⫹i兲Gc f0共k,in兲, 储f f共q,i兲⫽⫺ 1  k,i兺
n Gf f 0 共k⫹q,i n⫹i兲Gf f0共k,in兲, cc储 共q,i兲⫽⫺ 1  k,i兺
n Gcc 0 共k⫹q,i n⫹i兲 ⫻Gcc0共k,in兲, f c 储 共q,i 兲⫽⫺ 1  k,i兺
n Gf c 0 共k⫹q,i n⫹i兲Gf c0共k,in兲, ⬜f f共q,i兲⫽⫺ 1  k,i兺
nG↑f f0共k⫹q,in⫹i兲G↓f f0共k,in兲,
⬜cc共q,i兲⫽⫺ 1  k,i
兺
n Gcc 0 ↑ 共k⫹q,i n⫹i兲 ⫻Gcc↓ 0共k,in兲, ⬜f c共q,i兲⫽⫺ 1  k,i兺
n Gf c 0 ↑ 共k⫹q,i n⫹i兲G↓f c0共k,in兲, where Gcc 0 (k,in), Gf f0(k,in), and Gf c0(k,in) are the
Green’s functions at the saddle-point level obtained by in-versing the matrix G0(k,) defined in Eq. 共2兲.
The different bubbles f f(q,i), cc(q,i), and
f c(q,i) can be expressed as a function of the Green’s
functions associated with the eigenoperators ␣k
† andk † Gf f共k,in兲⫽uk 2G ␣␣共k,in兲⫹vk 2G 共k,in兲, 共A2兲 Gcc共k,in兲⫽vk 2 G␣␣共k,in兲⫹uk 2 G共k,in兲, Gc f共k,in兲⫽Gf c共k,in兲
⫽⫺ukvk关G␣␣共k,in兲⫺G共k,in兲兴,
where G␣␣(k,in) and G(k,in) are the Green’s
func-tions associated to the eigenstates␣k†兩0
典
andk†兩0典
. In the low-frequency limit, one can easily check that the dynamical spin susceptibility may be written asf f共q,i兲⫽
␣␣共q,i兲⫹¯␣共q,i兲
1⫺JS2␣␣共q,i兲¯␣共q,i兲, 共A3兲 for both the longitudinal and the transverse parts.
␣␣共q,i兲⫽ 1 
兺
k nF共Ek⫺兲⫺nF共Ek⫺⫹q兲 i⫺Ek⫺⫹q⫹Ek⫺ ¯␣共q,i 兲⫽ 1 兺
k 共uk 2 vk⫹q 2 ⫹v k 2u k⫹q 2 兲 ⫻nF共Ek ⫺兲⫺n F共Ek⫹⫹q兲 i⫺Ek⫹⫹q⫹Ek⫺ .*Also at the Center National de la Recherche Scientifique共CNRS兲. 1For reviews see H. von Lohneysen, J. Phys.: Condens. Matter 8, 9689共1996兲; D. A. Gajewski, N. R. Dilley, R. Chau, and M. B. Maple, ibid. 8, 9793共1996兲; F. Steglich, B. Buschinger, P. Ge-genwart, M. Lohmann, R. Helfrich, C. Langhammer, P. Hell-mann, L. Donnevert, S. Thomas, A. Link, C. Geibel, M. Lang, G. Sparn, and W. Assmus, ibid. 8, 9909共1996兲.
2H. von Lo¨hneysen, A. Schro¨der, M. Sieck, and T. Trappmann, Phys. Rev. Lett. 72, 3262共1994兲; H. von Lo¨hneysen, J. Phys.: Condens. Matter 8, 9689共1996兲; O. Stockert, H. V. Lo¨hneysen, A. Schro¨der, M. Loewenhaupt, N. Pyka, P. L. Gammel, and U. Yaron, Physica B 230-232, 247共1997兲.
3N. D. Mathur, F. M. Grosche, S. R. Julian, I. R. Walker, D. M. Freye, R. K. W. Haselwimmer, and G. G. Lonzarich, Nature
共London兲 394, 39 共1998兲, and references therein.
4F. Steglich, B. Buschinger, P. Gegenwart, M. Lohmann, R. Hel-frich, C. Langhammer, P. Hellmann, L. Donnevert, S. Thomas, A. Link, C. Geiber, M. Lang, G. Sparn, and W. Assmus, J. Phys.: Condens. Matter 8, 9909共1996兲.
5C. L. Seaman, M. B. Maple, B. W. Lee, S. Ghamaty, M. S. Torikachvili, J.-S. Kang, L.-Z. Liu, J. Allen, and D. L. Cox, Phys. Rev. Lett. 67, 2882共1991兲.
6S. Kambe, S. Raymond, L. P. Regnault, J. Flouquet, P. Lejay, and P. Haen, J. Phys. Soc. Jpn. 65, 3294共1996兲.
7E. Miranda, V. Dobrosavljevic, and G. Kotliar, J. Phys.: Condens. Matter 8, 9871共1996兲.
8A. H. Castro-Neto, G. Castilla, and B. A. Jones, Phys. Rev. Lett.
81, 3531共1998兲; M. C. de Andrade, R. Chau, R. P. Dickey, N.
R. Dilley, E. J. Freeman, D. A. Gajewski, M. B. Maple, R. Movshovich, A. H. Castro Neto, G. Castilla, and B. A. Jones, ibid. 81, 5620共1998兲.
9R. Narayanan, T. Vojta, D. Belitz, T. R. Kirkpatrick, Phys. Rev. B 60, 10 150共1999兲; T. R. Kirkpatrick and D. Belitz, Phys. Rev. Lett. 76, 2571共1996兲.
10B. Andraka and A. M. Tsvelik, Phys. Rev. Lett. 67, 2886共1991兲. 11A. J. Millis, Phys. Rev. B 48, 7183共1993兲.
12T. Moriya and T. Takimoto, J. Phys. Soc. Jpn. 64, 960共1995兲. 13M. A. Continentino, Phys. Rev. B 47, R11 587共1993兲. 14P. Schlottmann, Phys. Rev. B 59, 12 379共1999兲. 15
P. Coleman, Physica B 259-261, 353共1999兲.
16Q. Si, J. L. Smith, and K. Ingersent, Int. J. Mod. Phys. B 13, 2331
共1999兲.
17A. Rosch, Phys. Rev. Lett. 82, 4280共1999兲.
18A. Schro¨der, G. Aeppli, E. Bucher, R. Ramazashvili, and P. Cole-man, Phys. Rev. Lett. 80, 5623共1998兲.
19D. L. Cox, Phys. Rev. Lett. 59, 1240共1987兲; D. L. Cox and M. Jarrell, J. Phys.: Condens. Matter 8, 9825共1996兲.
20J. A. Hertz, Phys. Rev. B 14, 1165共1976兲.
21T. Moriya and A. Kawabata, J. Phys. Soc. Jpn. 34, 639共1973兲; H. Hasegawa and T. Moriya, ibid. 36, 1542共1974兲.
22S. Doniach, Physica B 91, 231共1977兲; C. Lacroix and M. Cyrot, Phys. Rev. B 20, 1969共1979兲; D. M. Newns and N. Read, Adv. Phys. 36, 799共1987兲; and for a recent review see H. Tsunetsugu, M. Sigrist, and K. Ueda, Rev. Mod. Phys. 69, 809共1997兲 and references therein.
23A. J. Millis and P. A. Lee, Phys. Rev. B 35, 3394共1987兲; A. Auerbach and K. Levin, Phys. Rev. Lett. 57, 877 共1986兲; M. Lavagna, A. J. Millis, and P. A. Lee, ibid. 58, 266共1987兲. 24A. Houghton, N. Read, and H. Won, Phys. Rev. B 37, 3782
共1988兲.
25C. Pe´pin and M. Lavagna, Phys. Rev. B 59, 2591共1999兲. 26P. Nozie`res共unpublished兲.
27J. A. Hertz and M. A. Klenin, Phys. Rev. B 10, 1084共1974兲; J. A. Hertz and M. A. Klenin, Physica B 91, 49共1977兲.
28A. Rosch, A. Schro¨der, O. Stockert, and H. von Lo¨hneysen, Phys. Rev. Lett. 79, 159共1997兲.
29N. Read and D. N. Newns, J. Phys. C 16, 3273共1983兲. 30G. G. Lonzarich and L. Taillefer, J. Phys. C 18, 4339共1985兲. 31N. R. Bernhoeft and G. G. Lonzarich, J. Phys.: Condens. Matter
7, 7325共1995兲.
32Y. Kuramoto and K. Miyake, J. Phys. Soc. Jpn. 59, 2831共1990兲. 33R. Hlubina and T. M. Rice, Phys. Rev. B 51, 9253共1995兲.