HAL Id: jpa-00247371
https://hal.archives-ouvertes.fr/jpa-00247371
Submitted on 1 Jan 1997
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.
Persistent Currents in Interacting Systems: Role of the
Spin
Georges Bouzerar, Didier Poilblanc
To cite this version:
Georges Bouzerar, Didier Poilblanc. Persistent Currents in Interacting Systems: Role of the Spin.
Journal de Physique I, EDP Sciences, 1997, 7 (7), pp.877-887. �10.1051/jp1:1997207�. �jpa-00247371�
Persistent
Currents
in
Interacting
Systems:
Role
of
the
Spin
Georges
Bouzerar
(*)
and
Didier
Poilblanc
Groupe
dePhysique
Thdorique,
Laboratoire dePhysique
Quantique,
Universit6 Paul
Sabatier,
3106?Toulouse,
France(Received
5Jldy
1996, received i~1final form 27 March 1997,accepted
1Aprfl
1997)PACS.72.10.-d Theory of electronic transport;
scattering
mechanismsPACS.71.27.+a
Strongly
correlated electron systems; heavy fermionsPACS. 72.15.Rn
Quantum
localizationAbstract. Persistent currents
flowing
through
disorderedmesoscopic
rings
threadedby
amagnetic
flux areinvestigated.
Models of fermions with on-site interactions(Hubbard
model)
or models of spinless fermions with nearest
neighbor
interactions are considered on 2Dcylinders
with twisted boundary conditions in one direction to account for
a
magnetic
flux. Self-consistentHartree-Fock methods are used to treat the electron-electron interaction beyond first order. We show that the second harmonic of the current
(which
is relevant in the diffusiveregime)
isstrongly
suppresiedby
the interaction in thecase of spinless fermions while it is
significantly
enhancedin the Hubbard model. Our data also
strongly
suggest that the reduction(increase)
of this harmonic is related to a strong increase(reduction)
of the spatial fluctuations of thecharge
density.
The observations of
mesoscopic
currents in very pure metallic nano-structures was done inpioneering
experiments
[1-3].
In the first case theexperiment
dealt with the average current ofa system of
10~
disconnectedrings
in the diffusiveregime
while in the second asingle ring
wasused.
Although
the existence of suchpersistent
currents in small metallicrings
waspredicted
long
ago[4-6]
themagnitude
of the observed currents is still a realchallenge
to theorists.There is a
general
belief that the interactionplays
a crucial role inenhancing
the current.But,
so
far,
the role of the interaction in disorderedsystems
is still unclear.Treating
interactionand disorder on
equal
footings
is a difficult task. Previous work[7,8]
has shownby
exactdiagonalizations
(ED)
of small clustersthat,
forstrictly
1D systems ofspinless fermions,
the effect of arepulsive
interaction is to increase further the localization of the electrons and henceto decrease the value of the current.
Using
a Hartree-Fockapproach,
Kato et al.[16]
haveobtained a
qualitatively
good
agreement
with the exact calculations. On the otherhand,
Gia-marchi et al. [9] have
pointed
outthat,
for the 1D Hubbardmodel,
i-e- whenspin
isincluded,
the interaction enhances the
persistent
current. In this case, the increase of the current isclosely
related to the decrease of thespatial
fluctuations of thecharge
density
or,equivalently,
(*)
Author for correspondence(e-mail:
gb@thp.urd-koeln.de)
Present address: Institut ffiir Theoretische Physik, Universitit zu
K61n, Ziilpicher
Str. 77, 50937K6In,
Germany
to the
smoothing
out of thecharge density
as it occurs in the 1D Hubbard model withrepulsive
interaction. This
emphasizes
theimportant
role of thespin
in 1Dsystems.
In
higher
dimensions the role of thespin
is still unclear. First order calculations for whichspin
is irrelevant have shown that thepersistent
currents are increasedby
the interactions[11].
Morerecently,
Ramin et al.[13]
havenumerically
shown that the first order Hartree-Fock(HF)
correction to the second harmonic of the
persistent
current was inagreement
with theanalytical
treatment
[11]. Hence,
for both thespinless
fermion model withnearest-neighbor
interactionsand the Hubbard
model,
the second harmonic is enhanced.However,
in this treatment it isfound that a nearest
neighbour
interaction tends to decrease the value of thetypical
currentwhile a
repulsive
extended Hubbard interaction enhances it. In someprevious work,
ExactDi-agonalizations
(ED)
calculations have beencompared
to a Self-consistent Hartree-Fock(SHF)
treatment of the interaction. For the small clusters
(4
x 4clusters)
which could be handled we have found agood
agreement
between the two sets of data[19].
This directcomparison
with the exact results has therefore established some
degree
ofreliability
of the self-consistent Hartree-Fockapproximation
at least in the diffusiveregime.
In this paper, we use both HF and SHF treatments of the interaction betweenparticles
which enables us to treat muchlarger
systems.
The SHF treatment takes into accounthigh
order terms in the interactionand,
simul-taneously,
deals withquantum
interference effects due to the disorder somehowexactly.
It isimportant
to note that this method is different from the usualperturbative
approach
ii ii
wherethe corrections to the current due to the
interacting
term are calculatedperturbatively.
Incon-trast to their
approach,
ourprocedure
includes a resummation ofhigher
order termsthrough
aself-consistency
relation,
which turns out to become essential at moderate interaction.Never-theless,
an exact connection with somediagrammatic
expansion
is a tediousproblem
and thisissue
probably
deserves furtherstudy.
Note that ourstudy
alsoapplies
inprinciple
tosingle
or
multi-ring
experiments.
From a theoreticalpoint
ofview,
the differencesimply
relies in theabsence or presence of
particle
number fluctuations. In thefollowing,
we shall assume that thenumber of electron of the
ring
is fixed.This paper is
organized
as follows:first,
we compare, for bothspinless
and Hubbardmodels,
the first order correction
(in
theinteraction)
and the SHF correction to the second harmonic ofthe
persistent
currents. We showthat,
in the case ofspinless fermions,
the two methodsstay
in
good
agreement
only
for rather small values of the interactionparameter
but acomplete
disagreement
appears for moderate values. In contrast, in the Hubbard case, theagreement
between the two
approaches
is rathergood.
Secondly,
wepi~esent
evidences that this decrease(increase)
of thepersistent
currents isdirectly
related to the increase(decrease)
of thespatial
fluctuations of the
charge
density
from site to site as the interaction is switched on.Thirdly,
the effect of the electron
repulsion
on the second harmonic is shown to increase with thesystem
size.
The Hamiltonian is defined on a L x L lattice with
periodic
boundary
conditions in onedirection
(e.g.
xdirection)
and reads:i~
"
i~K
+flint
+rides.
(1)
7iK
"
£~,jt~jc)cj
is the usual kinetic partcontaining
the fluxdependence,
t~j =
)exp(i~f~(i~
j~))
if I andj
are nearestneighbor
sites and 0 otherwise.7iint
isthe
interacting
part
and7ides
is the term due to thedisorder,
i~des
"
~
Wi n~
(2)
1
where n~ is the local
density
operator
at site I and w~ are on-siteenergies
chosenrandomly
in[-W/2, W/2].
Whenspin
isincluded,
ni = nit +ni ~
sum over the
spin
indices. In thespinless
fermion case the electronrepulsion
isgiven by
~int
~§
~3
~~ ~3' ~~)~
~
i-j
where
l~,j
ofstrength
V=
((j(
only
connects nearestneighbor
sites(screened interaction).
Lastly,
the Hubbard interaction is definedby
li~t
" U
~
~St ~SI'
(4)
1
In the SHF
approximation,
the interactionpart
of thespinless
fermion model reduces to~Int
"~
~~SJ~l
~J +~
~~°~ ~~~
vJ
((~llOl~Jl0
l~l
~~lO~)
(~)
S,J S S,Jwhere
()~
stands for theexpectation
value in theground-state
wavefunction. Theexpression
above contains two terms: an Hartree term
proportional
to dw~ =£~
~~l~j
(nj)~
which comesout as an extra on-site disorder
potential
and a Fock termproportional
todt~j
=l~j (c)
c~)~
which is an extra
hopping
amplitude.
Thequantities
(nj)~
and(c)
ci)~
are calculated selfconsistently
so that the SHF Hamiltonian itselfdepends
on thefilling
factor,
the disorderstrength
etc.Similarly,
in thisapproximation
7if~
becomes~~t
~~~~'~St~0
~~~ ~ ~~S~~o ~~t~~s~~o~~st~o~.
~~~i
Since we are interested in the
paramagnetic
phase
we shall assume(ni~)~
=(nit
)~.
In thiscase the
spin f
and(
aredecoupled
so that one can write7if~
=£~
7i[~
where~~t
~~
~~?~~0~"
~~~"~~'
~~~i
The current is defined as the derivative of the total energy versus flux
aE(~)
~i~)=-a~ ~~)
where
E(4l)
is the total energy. The current is aperiodic
function of 4l ofperiod
1(4l
ismeasured in units of
4lo)
and thus can beexpanded
as a Fourierseries,
1(4l)
=
~j
Ihn
sin(27rn4l)
(9)
where
Ihn
are the harmonics of the current. The solution of theprevious
set of non-linearself-consistent
equations
are obtainednumerically
on small clustersby
an iterativeprocedure
forarbitrary
values of 4l andarbitrary
disorder realizations. The variousphysical
quantities
arethen
averaged
over disorder. The disorder average denotedby
()~.~
in thefollowing corresponds
typically
to an average over at least 1000configurations
of theiisorder.
It is now well established that the ensemble average
(over filling
ordisorder)
suppresses thefirst harmonic of the current
[18,19].
This fact was indeed observed inmulti-ring
experiments
~.~~~~ ~#~ la) _~~~ ' 5 O.O025
~fl
~~~ o ooi 5 o.ores O.0005 0.25 °'~~ ~ ~~ 4~/4~0Fig.
1. Totalaveraged
currentI(#I
versus #(non interacting
fermions)
for a 8 x 8 system and fordifferent values of the disorder parameter W, Ne is the total number of electrons. An average over
1000 disorder
configurations
has been done.o ooo8
Spinless Fermions SpInless Fermions
la)
~
o ooo8 0 0004 jb) fl fl o ooo4@
o ooo3 , ~@
~ o o002 o ooo2 o oooi ~~~~~00 02 04 06 08 10 ~00 02 0A 06 0 V lfFig.
2.(Ih2)
~ as afunction of V calculated within the SHF method for spinless fermions on a
8 x 8
cylinder.
A~
average over loco disorderconfigurations
has been done. a) Comparison betweenfirst order HF
(open symbols)
and SHF results(full
symbolsl
athalf-filling-
Circlescorrespond
toW
= 3 and squares to W = 4.
b)
SHF results athalf-filling
(full
symbols)
and quarterfilling
(open
symbols).
Circles correspond to W= 3, squares to W = 4 and
triangles
to lV = 5.harmonic is
suppressed
forsufficiently large
disorder. We show inFigure
1 thedependence
ofthe total
averaged
current(1(4l))~~~
versus flux for different. values of the disorderparameter.
We observe from that
figure
that the current is4lo/2
periodic
forsufficiently
large
values ofthe disorder
parameter
(W/t
> 4 for asystem
of size 8 x8).
This is asimple
way to estimatethe
cross-overfrom the
ballistic
to thediffusive
regime.
As noted
previously,
theself-consistency
relations can take care ofhigh
order effects in theinteraction. It is therefore necessary, as a
preliminary
study,
toinvestigate
the role of theself-consistency by comparing
the SHF results to thesimple
first order calculations(referred
to hereafter asHF).
InFigure
2a(Ih2)~.
forspinless
particles
isplotted
as a function ofis of course Iinear in V. As
expected,
we observe for small values of the interaction aperfect
agreement
between the SHF and the HF calculations.Indeed,
theslope
at V= 0 of the SHF
results is
actually given by
the HF data.However,
forincreasing
V the SHF calculation showsa
strong
reduction of the current whilst the HFpredicts
an increase. We also observe thatthe
region
ofagreement
between HF and SHF is reduced as the disorder increases. In otherwords,
this meansthat,
as thestrength
of the interactionincreases,
the effect ofhigher
orderterms becomes more and more dominant and thus a first order calculation is not sufficient.
We will see later on that this reduction is
related,
as in the 1D case, to an increase of thespatial
fluctuations of thecharge
density.
The influence of thefilling
factor and of the disorderstrength
on(Ih2)~.
versus V is shown inFigure
2b. Thisfigure
shows that therepulsive
interaction is
detri£ental
to thepersistent
currents which aredrastically
suppressed
for a widerange of
parameters.
Note that, atquarter
filling, (Ih2
)~.
startsimmediately
to decrease withV
contrary
to the behavior observed athalf-filling.
~~Similar conclusions can also be reached from a direct
investigation
of thecomplete
distribu-tions of the first and second harmonics obtained in the SHF method
(Figs.
3a andb)
Notethat these distributions contain more information than the above
expectation
values whichcorrespond
only
to their first moments. In absence ofinteraction,
weclearly
observe thatthe distribution of the first harmonic is
perfectly
symetric
((Ihi)~~~
"
0)
inagreement
withFigure
1. The effect of thespinless
interaction onIhi
consistsonly
onreducing
the width ofthe distribution. That is
why
we focus on theconjugated
effect of interaction and disorderon
Ih2
only.
The case of the second harmonic iscompletely
different.Indeed,
in absence ofinteraction the distribution is
strongly
asymetric,
(Ih2)~
is finite(current
is4lo/2 periodic).
When we switch on the
interaction,
the effect isdraniat~~,
forV/t
= 0.8 the distributiongets
perfectly
symetric
((Ih2)
~.is
suppressed
by
theinteraction)
and the width of the distributionis
significantly
reduced. ~~In contrast to the
spinless model,
the Hubbard model exhibits acompletely
differentbehav-ior,
as seen inFigure
4showing
the relative increase of(Ih2)~~~
as a function of the Hubbardrepulsion
U.Interestingly
enough,
both SHF and HF calculationspredict
an increase of thecurrent. It is also
interesting
to notice that the first order calculationgives
reliable resultsregarding
the effect of the interaction.However,
the first order calculationsalways
predict
larger
values of the currents. InFigure
3c,
we also show the effect of the Hubbard interaction on the distribution ofIh2
due to the interaction(SHF).
Theshape
of the distribution remainsasymetric
and(Ih2)~~~
isincreasing
with the interaction as shownpreviously.
At this
point,
these resultsalready
suggest
that the nature of the interactionplays
a crucialrole: in the
spinless
case the currents arestrongly
reducedby
the interaction in contrast toa Hubbard
repulsive
interaction which enhances the currents.Secondly,
we have shownthat,
in the
spinless
fermions case,higher
order terms becomerapidly
dominant even forrelatively
small values of the interactionstrength.
Hence a first orderafiproach
is not sufficient.Let us now
try
todevelop
aphysical
picture
that couldhelp
to understand the role of thespin.
We shall argue that the enhancement(reduction)
of thepersistent
currents is related tothe reduction
(increase)
of the fluctuations of thecharge density
ni from site to site. One wayof
observing
this effect on thecharge density
consists inplotting
the distribution of the localdensity
(ni)~.
For that purpose, we shall consider here a 10 x 10system
and assume that the(ni)~
(where
thesubscript
k labels the various realizations of thedisorder)
areindependent
variables. The distribution of the
charge
densities can then be defined asP(p)
= ~~j
~
d((ni)~
p)
(10)
NdisL
~~
~016 O.14 ~'~ ~ W=4 V=O W=4 V=O.8 ~ ~ la) T 08
f
O.06 o.04 0.02~'~i.020 ,OOIO OOOO O.O10 0020.O.0~0 .O.O10 OOOO O.010 O020
0 20 W=4 V=0 W=4 V=O.8 O.lS 16) « ~ °~° r o.05 ~'~~ 002 0 001 0 0 001 0 002 ~ 002 0001 0 0 00> 0 002 o 30 W=4 U=0 W=4 U=0 4 ic)
~
O.20 j f 2 o.io~'~i.O04 .O.0O2 O.COO O.002 0.004 -O.004 -O.002 0.000 0002 O.004
Fig.
3. Distribution of the first and second harmonic calculated for a 8 x 8 system athalf-filling-The disorder parameter is fixed
(W/t
= 4) We have considered loco
configurations
of disorder.a)
Ihiin the
spinless
fermion case(V/t
= o and
o.81, b)
Ih2 in thespinless
fermion case(V/t
= o ando.8),
c)
Ih2 in the Hubbard case(U/t
30
%fl
m 2.2 o E $ r 18 1.4 al 02 03 04 O-S UFig.
4. Ratio(Ih2)~
(U,
Wl/(Ih2)~
(o,W)
as a function of U calculated in the Hubbard modelwithin both HF
(open
~ymbols)and
SiF
(full
symbols)
methodson a 8 X 8 cylinder. An average
over loco disorder realizations have been performed. Circles correspond to W = 3 at
quarter-filling,
squares to W = 3 at
half-filling
andtriangles
to W = 4 athalf-filling-o s v=o 8 Spinless fermions
° °
fl£
*~~ W"° S ~"° ° °fi£
~~~hi
~~~~@
(a) (b)~
0.05 °OS o-~ W=0 5 V=0 ~~~00 020 030 04 P PFig.
5. DistributionPip)
in the case ofspinless
fermions. The calculations have been done atquarter
filling
n = o.25(al
and n = o.4(b)
on a lo X locylinder.
We haveaveraged
over 30configurations
of the disorder. The values of V and W aregiven
in thefigure.
where L is the
length
of thesystem
(L
=10)
andNdis
is the numberof disorder
configurations.
As
usual,
the local densities(n~)~
are calculatedself-consistently.
Note that, for asufficiently
large
system, weexpect
the distribution to becomeindependent
of the choice of the disorderconfigurations.
For
spinless
fermions,
we haveplotted
inFigures
5a,
b the distributionPip)
calculated on a 10 x 10cylinder
for two differentfillings.
Weclearly
observe that, with theinteraction,
both the
shape
of the distributionchange
and the width of the distribution increases. Theseeffects are
strongly
emphasized
inFigure
5b in which twopeaks
appear,showing
atendency
towards the formation of a
charge density
wave as weapproach
commensurability
in
=0.5).
P(P)
%~
- ~=~ ~=~S o.io(
@
a-as ~~~ 04 o PFig.
6. DistributionP(p)
in the Hubbard case. The calculations have been done athalf-filling
(n
= 1) on a lo X 10 cylinder. We have considered 30
configurations
of the disorder. The values of Uand W are
given
in thefigure.
by
preventing
perfect
nesting
of the Fermi surface(for
example, by including
hopping
terms atlarger
distances).
However,
the effect of the interaction to broaden the distributionPip)
seems to be
independent
of theprecise
details of the band structure and isgeneric
for anyfilling.
Incontrast,
in the case of the Hubbard model(Fig. 6),
we observe theopposite
effect:the distribution shrinks around the average value as the interaction is switched on.
Let us now turn to more
qualitative
results. The width of the distribution(assuming
inde-pendent
variables for differentconfigurations
of thedisorder)
dp
isgiven by
dp
=
)
(
j
~iiniii
n)2
iii)
~_~ ~
where n =
N~/L2
is thefilling
(iV~ is the number ofelectrons).
We have studied the effect ofthe interaction on
dp
as a function of V(spinless
fermions)
or U(Hubbard model)
for differentvalues of the disorder
strength
W and differentfillings.
InFigure
7adp(V, W)
isplotted
athalf-filling
as a function of theparameter
V(spinless fermions).
Here weclearly
observe an increaseof the width
dp
of the distribution forincreasing
V.Very
crudely,
this effect seems todepend
only
on the combinedparameter
V/W.
In contrast, in the Hubbard case shown inFigure
7b,
we observe a reduction ofdp(U, W)
as U increases. InFigures
8a,
b we haveplotted
therelative variations of the width of the distribution
dp(V,
IV/dp(0,
W)
anddp(U. W) /dp(0, W)
as a function of the interaction parameters V and U for various densities. We observe that the
effects described above become
stronger
forlarger fillings
i.e. when the interaction between theparticles
becomes more effective. Note howeverthat,
in the case of thespinless
fermion model.this behavior is unrelated to the
commensurability
since densities n = 0.25 and n = 0.4 havebeen
chosen,
clearly
away fromhalf-filling
in
=
0.5).
InFigures
8 we seethat,
in thespinless
case, the width has increased
by
almost a factor 4 for V = 0.8 andN~
= 40. In the Hubbardcase the reduction of
dp(U, W)
is25%
larger
compared
tohalf-filling.
Our presentstudy
thenstrongly
suggests
that a reliableexplanation
of the observedlarge
persistent
currents mustsomehow take into account the
spin
of theparticles.
We finish our discussion
by
astudy
of the influence of thesystem
size on our results. In~ ~~ Splnless fermions
%
~~~%%
@~
~@
la) ~~~~i
j~
~ ozo 3p3p
oio __~~~ oio -w=1 2 oos ~~~00 0.2 04 06 08 ~~100 020 040 lf UFig.
7. Effect of the interaction on the width&p(I§
W)
athalf-filling
as a function of the interactionparameter
(U
orVi.
The calculations have been done on a lo X 10cylinder
and 30configurations
ofthe disorder have been considered. The values of W are
given
in thefigure.
a)Spinless
fermion model;b)
Hubbard model.~
i coo ~ ~ *~$[~
(]~~~%%
~ °~~° (b) ~ ~~ Spinless Fermionsj~
o 900 ~ 10X10 c~ j~
~~° la) ill o850~c ~ zoo ~~~~ w-w=05 n=05 w- w=o 5 n=i 150
@
0 04 0.6 08 lo ~~~~00 o-lo 020 030 040 050 lf UFig.
8. Ratio&p(V
Wl/&p(o,
W)
as a function of the interaction parameter(U
orV)
for differentfilling
factors(the
filling
is indicated on theplot)
and values of W. The calculations have been doneon a 10 X lo cylinder with 30
configurations
of the disorder.a)
Spinless
fermion model(n
= o.25 andn =
o-s)
b)
Hubbard model(n
= o-s and n =1).
versus U at fixed
density
but for different system sizes. Thisfigure clearly
indicatesthat,
as the size of the systemincreases,
the effects inducedby
the interaction becomestronger.
Wealso
expect
that this is also true when theconnectivity
of the lattice increases(i.e.
going
form2D to real 3D
rings).
Although
asystematic
accuratestudy
as a function of thesample
sizeis not
feasible,
our data for the Hubbard modelfor,
let say U =0.4,
are not inconsistent withthe
magnitude
of the currents observed in theexperiments.
In
conclusion,
we have shownthat,
in thespinless
fermionmodel,
the effect of a moderat.einteraction leads to a drastic reduction of the
magnitude
of the current,contrary
to what isexpected
from first order calculations. For the firsttime,
we have establishedthat,
in2D,
taken into account thespin
degrees
offreedom is crucial toexplain
the enhancement ofthe current due30 26
H
3 n=I 22j~
2 1.8j£
1.4o-lo O.20 O.30 O.40 O.50
U
Fig.
9. Ratio(Ih2)(U, W)d>s/(Ih2)(o,
W)d;s
as a function of U(Hubbard
case)
calculated withinthe SHF method. The calculations have been done on 6 X 6
(circlesl,
8 X 8(squares)
and 10 X 10(triangles)
cylinders
athalf-filling
(n
= 1) and for W = 3. loco realizations of the disorder have beenused.
deal with rather small 2D
clusters,
weexpect
that,
forlarger
systems
andhigher
connectivity
(e.
g. in3D)
and in the presence of an Hubbardinteraction,
theimpurity
scattering
will becomeeven less effective to localize the electrons.
Although
asystematic
study
as a function of thesample
size is still out of reach ofpresent
day
computers,
our data of the Hubbard model atintermediate U
(around 0.4)
are not inconsistent with themagnitude
of the current observedin the
experiments.
Ourstudy
suggests
that the SHFapproach provides
arelatively
good
toolto
study,
onequal
footings,
the effects of the interaction and of the disorder.Acknowledgments
We
gratefully acknowledge
stimulating
discussions with T. Giamarchi and G. Montambaux.D.P.
acknowledges
support
from the EEC HumanCapital
andMobility
Program
undergrant
CHRX-CT93-0332. Laboratoire de
Physique
Quantique
(Toulouse)
is UA No. URA505 duCNRS.
References
[1]
Levy L.P.,
DolanG.,
Dunsmuir J. and BouchiatH.,
Phys.
Rev. Lett. 64(1990)
2074.[2] Chandrasekhar
V.,
WebbR.A., Brady M.J.,
KetchenM.B.,
Galager
W.J. and KleinsasserA.,
Phys.
Rev. Lett. 67(1991)
3578.[3]
Mailly
D.,
Chapelier
C. and BenoitA.,
Phys.
Rev. Lett. 70(1993)
2020.[4] Hund
F.,
Ann.Phys.
(Leipzig)
32(1938)
102.[5]
Byers
N. andYang
C.N.,
Phys.
Rev. Lett. 7(1961)
46;
KohnW.,
Phys.
Rev. 133(1964)
A171.
[7] Bouzerar
G.,
Poilblanc D. and MontambauxG., Phys.
Rev. B 49(1994)
8258;
see alsoBerkovits
R., Phys.
Rev. B 48(1993)
14381.[8] Bouzerar G. and Poilblanc
D.,
J.Phys.
I France 4(1994)
1699.[9] Giamarchi T. and
Shastry
B.S.,
preprint
(1994).
[10]
Berkovits R. and AvishaiY., Europhys.
Lett.(in press).
[11]
Ambegaokar
V. and EckernU., Phys.
Rev. Lett. 65(1990)
381;
see also AltshulerB.L.,
Gefen Y. and
Imry Y., Phys.
Rev. Lett. 66(1991)
88.[12]
Kopietz P., Phys.
Rev. Lett. 70(1993)
3123.[13]
RaminM.,
Reulet B. and BouchiatH.,
preprint
(1994).
[14]
Haydock
R.,
Heine V. andKelly M.J.,
J.Phys.
C 8(1975)
2591.[15]
Kirkman P.D. andPendry J.B.,
J.Phys.
C17(1984)
4327;
Kappus
M. andWegner
F.J.,
Z.
Phys.
B 45(1981)
15.[16]
Kato H. and YoshiokaD.,
Phys.
Rev. B 50(1994)
4943.[17]
Bouchiat H. and MontambauxG.,
J.Phys.
France 50 2695(1989);
see also Ho-FaiCheung
et