HAL Id: jpa-00247364
https://hal.archives-ouvertes.fr/jpa-00247364
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.
Orbital Maghetism in Two-Dimensional Integrable Systems
E. Gurevich, B. Shapiro
To cite this version:
E. Gurevich, B. Shapiro. Orbital Maghetism in Two-Dimensional Integrable Systems. Journal de
Physique I, EDP Sciences, 1997, 7 (6), pp.807-820. �10.1051/jp1:1997194�. �jpa-00247364�
Orbital Magnetism in Two-Dimensional Integrable Systems
E. Gurevich and B.
Shapiro (*)
Department
ofPhysics,
Technion-Israel Institute ofTechnology,
Haifa 32000, Israel(Received
5January 1997,
received in final form iiFebruajj.
1997,accepted
19February1997)
PACS.75.20.-g Diamagnetism
and paramagnetismPACS.73.20.Dx Electron states in low-dimensional structures
(superlattices,
quantum well structures and
multilayers)
Abstract. We
study
orbitalmagnetism
of adegenerate
electron gas in a number of two- dimensionalintegrable
systems, within linear response theory. There are three relevant energy scales:typical
levelspacing /h,
the energy r, related to the inverse time offlight
across the system, and the Fermi energy SF-Correspondingly,
there are three distinct temperatureregimes:
microscopic (T
<hi, mesoscopic (A
« T £r)
and macroscopic(r
« T «eF).
In the first tworegimes
there arelarge
finite-size effects in themagnetic susceptibility
x, ~vhereas in the thirdregime
X approaches its macroscopic value. In some cases, such as a quasi-one-dimensional strip or a harmonicconfining
potential, it is possible to obtainanalytic
expressions for x in theentire temperature range.
1. Introduction
A
degenerate
electron gas, in the presence of a weakmagnetic field,
exhibits weak orbitalmagnetism ill (the
Landaudiamagnetism).
For a two-dimensional gas the value of the orbitalmagnetic susceptibility
isgiven by
XL =-e~ /12~Mc~,
where lzI is the electron mass.(Double degeneracy
withrespect
tospin
isimplied
in thisexpression,
as well as in allsubsequent formulae.)
This resultapplies
to a macroscopicsystem.
Whether a
sample
of agiven
size L can be considered asmacroscopic depends
on the tempera- ture T.Namely,
T(in
energyunits)
should becompared
with the characteristicsize-dependent energies
such as the mean levelspacing,
A ci2~h~/fi£L~,
or the inverse time offlight
acrossthe
sample,
r ci AUFIL
cikF
LA/2~,
where vF andkF
are the Fermivelocity
and wave number.Generally,
one shoulddistinguish
between three temperatureregimes:
ii)
For T < A(the "microscopic" regime)
discreteness of the energy levels comes intoplay
and thesample
can be viewed as agiant
atom. Themagnetic
response in this case can be verystrong
and includes such exoticpossibilities
asperfect diamagnetism
and Meissner effect [2].(ii)
Forhigher temperatures,
A « T$ r,
thesystem enters
themesoscopic regime (for
recent reviews see Refs.
[3, 4]).
Here thetypical
value of themagnetic susceptibility
is(* Author for
correspondence (e-mail: borisflphysics.technion.ac.il)
©
Les#ditions
dePhysique
1997of order
(kFL)°
XL and can have eithersign.
Theexponent
adepends
on thesample
ge-ometry.
For most two-dimensionalintegrable systems
a =3/2, although
other values are alsopossible
in somespecial
cases(see below).
Forcompletely
chaotic two-dimensionalsystems
a = I.(iii)
For stillhigher temperatures,
when T » r(but
smaller than the Fermi energyeF),
thesystem
can be considered asmacroscopic
and itsmagnetic susceptibility,
up to smallcorrections,
isgiven by
the Landau value XLThus,
atpresent
there is agood qualitative understanding
of thephenomenon
of orbitalmagnetism
in varioustemperature regimes
and for variousgeometries. However,
reliable quan- titative results are scarce. Most of such results refer to themesoscopic regime [4-6]
and are based on a semiclassicalapproximation
for thedensity
of states. Thisapproximation
becomesinadequate
both at very lowtemperatures,
T <A,
and athigh
temperatures, T > r.It, therefore,
seems useful to consider a fewsimple systems, starting
with an exactexpression
for thesusceptibility
x, and to observe the behaviour of x in the entiretemperature
range. This is done in the present paperby u8ing
a linear responseexpression
for x.In Section 2 we present several
equivalent expressions
for the orbitalmagnetic susceptibility
within the linear response
theory.
In Sections3, 4,
5 we considerspecific examples
of astrip, disc,
andsquare-geometries.
Section 6 is devoted to an electron gas confinedby
a two- dimensionalparabolic potential.
2. Linear
Response Theory
for theMagnetic Susceptibility
We consider an electron gas, confined to some domain in the
xv-plane
andsubjected
to a weakmagnetic
field B in the z-direction. Thejrand-canonical potential
Q
=
/ dEp(E) In[I
+efl~"~~)], (l)
fl
where ~t is the chemical
potential, fl
= I
IT
andp(E)
is thesingle-particle density
of states.Sometimes it is
useful, by integrating by parts twice,
to rewriteequation (I)
as:Q =
/ dEQ(E)$ (2)
where
f(E)
=
(I
+exp[fl(E ~)])~~
is the Fermi function and thequantity
Q(E)
=/~ dE' /~ dE"p(E") (3)
-c~
-~
has the
meaning
of agrand-canonical potential,
for the samesystem,
at zerotemperature
and with the chemicalpotential equal
to E.The
density
of states can be written asPIE)
=
-)Im
TrG(E)
=-)Im
TrlE
+i~ H)~~
,
14)
where
G(E)
is the retarded Green'sfunction,
at energy E. The full(single-particle)
Hamilto- nian H issplit
into theunperturbed part, Ho,
and theperturbation
~
&Ic~
~ ~ 2fifc~ ~ ' ~~~which describes the effect of the
(static) magnetic
field. It is assumed that the vectorpotential
A satisfies the condition div A
= 0.
Expanding G(E)
in terms of theunperturbed
Green'sfunctions, Go (E)
=
(E
+i~ Ho)~~,
one
obtains,
up to secondorder,
G =
Go
+GOVGO
+Go VGOVGO, (6)
which,
on substitution intoequation (4),
leads to thefollowing expression
for the correctionbp(E)
to thedensity
of states:The first
and the econd termdescribe,
spectively,the para-
lugging
nto
(I)
nd rating byparts
gives:
IQ =
This
expression
isproportional
toB~,
so that thesusceptibility,
in the B ~ 0limit,
is x =-2bQ/B~A,
where A is thesample
area.For further calculations it is useful to have an
expression
for x in terms ofxo(E)
which is thesusceptibility
of thesystem
at T=
0,
~= E
(compare
withEq. (2)):
X "
/ dEXo(E)$
191
Equation (8)
is written in an abstract operator form. Forpractical
calculations it is useful to use aparticular representation.
Forexample,
if onecomputes
the tracesusing
as a basis theeigenstates
ofHo (with
theappropriate boundary conditions),
oneobtains,
after somealgebra:
AB2
~
bet
~~
i
~~~~~~~"
~~~
~~'~~ ~~~~
Here the summation is over all states of the
unperturbed
HarniltonianHo,
witheigenenergies
e~. The
energies e(
andel'
denote the first(I.
e.proportional
toB)
and the second(proportional
to
B~)
corrections to theunperturbed energies
ei. The first correction can existonly
for levels which aredegenerate
in the absence of B.Another useful
expression
for x is obtainedby writing equation (8)
in the coordinate represen- tation. This results in:x =
~
~ ~
Im
/
dEf(E) / d~r / d~r'
~AB fi£c
x
(-h2 A(ri )Go (r, r'; )j A(r') )Go (r',
r;Eij
+Miir r'iGo jr, r'; E1421ri), iii)
where
Go(r,r'; E)
is theunperturbed (retarded)
Green's function in the coordinate represen- tation. Since thisrepresentation
isusually
the mostappropriate
formaking
various approx-imations, equation ill)
is agood starting point
in many cases. It wasused,
forinstance,
in thestudy
ofmesoscopic
effects in disorderedsystems
[7](in
this caseGo
includes the randompotential
ofimpurities).
It alsohelps
to prove in the most direct waythat,
for T >r,
the sus-ceptibility approaches
itsmacroscopic
value XL,independently
of thesample geometry. Indeed,
the Fermi function
f(E)
haspoles
at valuesEn
= ~ +
i(~ lpi (2n +1).
Therefore theintegral
over E in
equation (11)
can bereplaced by
a sumcontaining Go (r, r'; En).
This function in an infinitesystem decays
with distance asexp(-kF (r r'[ ~(2n+1) /fl~).
It is therefore clear that for a system with size Llarger
thanfl~/kF,
i.e. for T >r,
thesusceptibility
ceases todepend
on
sample
size or on itsgeometry. Therefore,
for T > r linear responsetheory
isapplicable
aslong
as thecyclotron
energyhwc
is smaller than T.However,
for T < r thesusceptibility-
xdoes
depend
onsample
size and itsgeometry,
and the linear response conditionrequires
thathwc
is smaller than the levelspacing
£h(I.e.
themagnetic
flux ibthrough
thesample
is smallerthan the flux
quantum
ibo =2~hcle).
A useful
approximate expression
for x is obtained uponusing
the semi-classicalapproxima-
tion [8] for the Green's functions inequation (11).
Let usbriefly
outline the derivation(details
are
presented
elsewhere[9]).
First, one derives a semi-classicalapproximation
for the function to(E).
This is doneby rewriting
the Green's functions in terms of thepropagators K(r, r', t),
approximating
thepropagators by
their semi-classicalexpressions
[8] andperforming
the inte-grals
within asaddle-point approximation.
Then one substitutes xo(El
intoequation (9)
andintegrates
overE, making
use of theapproximation
/
dE E"CDS(F(Eii $
QS /L" CDS
(F(/L)I
l~i~F'(/LlT) (121
where
F(E)
is some smooth function of E(I.e.
does notchange appreciably
within an interval of orderT)
and11(x)
exl
sinh x. Theresulting
semi-classicalexpression
for x isHere I labels families of
primitive periodic
orbits forintegrable systems
or isolated orbits for chaotic ones. r is thewinding number,
TX is theperiod
of aprimitive periodic
orbit andTt "
h/~T.
Factorsd>.r(~)
are related to theoscillating
part of the(unperturbed)
semi-classical
density
of states[8,10]:
Posc
(Ill
"
~j d>,r(/~). (14)
A,r
For
integrable system (A()
is an orbit areasquared
andaveraged
over thefamily
~. For a chaoticsystem
it issimply
thesquared
area of an isolated orbit.Averaging
over afamily
amounts to
integration
over one of theangle
variables [4](All
=
) / ~~ A~(9)d9. (15)
The semi-classical
expression
for x,equation (13),
coincides with the one derived in refer-ence
[4],
in the B ~ 0 limit. This fact demonstrates that it does not matter which of the twoapproximations,
I.e. linear response or semi-classics. is done first.3.
Strip Geometry
In this section we consider electrons confined to a
strip -L~/2
< x <L~ /2, -Ly/2
< y <Ly/2.
Periodic and zero
boundary
conditions are assumedalong
x and y directionsrespectively,
andthe limit
L~
~ oo is taken. It is convenient to choose the Landau gauge:A~
=
-By,
Ay
=Az
= 0.
Stationary
states are labeledby
twoquantum numbers,
-cc < k < +cc andn =
1, 2,.
Theeigenfunctions
~§n,klx, vi
=e~k~un,k (VI,
whereu[,k
satisfies:1/M )~
~2~f ~~~
~~~~~~ ~"'~~~~ "~'~~~~'
Treating
themagnetic
field as aperturbation,
one obtainsill]
that the first order correction£[~
= 0 and the second order correction,,
L(
eB ~ 6k~L( ~2
151
~"~ 24M c ~2n2 ~ ~4 n2 n4 ~~~~
Thus, only
the first term inequation (10)
is present, and the zero-temperaturesusceptibility (including spin degeneracy) xo(E)
isgiven by
x0(El
=
/~~ /)) () f £i~o(E ~n~i, j17)
n=1
where £nk
=
(h~k~/2M)
+(h~~~n~/21zIL()
are theunperturbed
energy levels.Next,
weintegrate
overk, apply
the Poisson summation formulaill
to the sum over n, and insert theresulting expression
for xo(El
intoequation (9).
Theintegral
over E is thenperformed, using
the
approximation (12),
which amounts toneglecting
small terms of orderT/~t.
The finalexpression
for x is:~
= l +
fi~
~°~~~~(~~
~ ll
~~~)
+
Oil /fi), (18)
xL ~
~~
f
~~
r
where
kF
=@,
r=
h~kf/fifL~
=
hvf/Ly
and thefunction11(x)
has been defined above. Inequation (18)
we wrote downonly
theleading oscillating term,
of orderfi,
and the term
I, responsible
for the Landaudiamagnetism.
There exist also smalloscillating corrections,
of order(kFLy)~~/~
andsmaller,
which are not written downexplicitly,
eventhough they
are calculable [9]. Let usonly
mentionthat,
in addition tooscillating corrections,
there is also a smallnon-oscillating paramagnetic correction,
9[XL(/8kFLy,
to the Landau value xL.Thus,
thestrip geometry provides
a rareexample
for which it ispossible
to obtain anessentially
exact(up
to small corrections of orderT/~) expression
for the linearsusceptibility
x,
including
allsize-dependent
terms. One can observe thechange
of x with T in the entiretemperature
range, from T= 0 and up to T » r when x becomes
equal
to itsmacroscopic
value XL-
Equation (18)
isquite
similar to thecorresponding expression
for the case of aparabolic
confinement[12].
This fact demonstrates that the nature of confinement(I.e.
hard walls or "soft"confinement)
is immaterial for thephenomenon
ofsize-dependent
fluctuations.In
fact,
theleading oscillating
term in x can beobtained,
within a semi-classicalapproxi- mation,
for anyconfining potential
and forarbitrary magnetic
field[9].
For smallfields,
the oscillations aregiven by
anexpression
similar toequation (18),
up to an overall factor of orderunity
and an extraphase
in theargument
of the cosine.L~
should be understood as someeffective width of the
confining potential.
12.
~
©~
~
7.m
~
~
~ 2
20 22.5 25 27.5 30 32.5 35 37.5
k~R
Fig.
1. Linearmagnetic susceptibility
in a disc geometry, calculated at T = 0. lb and normalizedto
(XL((kFR)~.
4. Circular
Geometry
The electron gas is confined to a disc of radius R. The
unperturbed,
I.e. zero-fieldstationary
states are
given,
up to a normalizationfactor, by exp(im9)Jm(~mnr/R),
where~mn
is the n-thzero of a Bessel function
Jm(x).
Theunperturbed energies
are £mn=
(h~ /2MR~)~$~.
Apair
of states
[m, n)
and m,n)
have the same energy.The
perturbation
term in theHamiltonian,
due to themagnetic field,
is:iheB b
~
e~B~
~2
(19)
V~@@ 8Mc~'
and the first and second-order energy corrections are:
~~"
2Mc'~~'
~~" 8Mc~ ~'~~'~~~ ~'~~'~~' ~~~~Note that the first-order term in V does not contribute to the correction
£$~
which is thereforepurely diamagnetic.
It now follows fromequation (10)
that1 1 i(mnr~mni/(£mm
+
Ii m~ Al
(211
We
analyze
first thelow-temperature regime,
T < A e2h~ /MR~.
Since(mn[r~ [mn)
ciR~,
the first
(diamagnetic)
term is of order of the total number ofelectrons,
N cf(kFR)~
=4~ IA.
The second
(paramagnetic)
term exhibitssharp peaks
every time when the chemicalpotential
~ coincides with an energy level £mn.
Indeed,
for ~= £mz~, the function
of /b£mn
isequal
to-1/4T,
and itrapidly
decreases when ~ deviates from £mz~by
a few T. Since thequantum
number m, for states near ~, istypically
of orderkFR,
theheight
of theparamagnetic peaks
is of order
(kFR)~A/T
ct~/T. Thus,
in thelow-temperature regime
thesusceptibility
x(normalized
to the Landau value[xL() Plotted
as a function of ~,displays
adiamagnetic background,
of order~/A,
withsharp paramagnetic peaks
ofheight ~/T
and width T. Anexact numerical
computation
of theexpression (21)
confirms thisqualitative picture (Fig. Ii.
Next,
we considertemperatures
in the range A « T£
r e AUF/2R.
For suchtemperatures
theparamagnetic peaks get
smeared and thediamagnetic background,
of order(kFR)~,
cancelswith the
corresponding paramagnetic
term. The net effect is anoscillating
term, of order(kFR)~/~
The calculation is based on a Poisson summation of the double sum inequation (21)
and on a semi-classical
approximation
for theunperturbed energies,
or~mn,
whichsatisfy
113]:
m2 m
arccos(m/~mz~)
=
~(n
+~). (22)
Let us outline the calculation of the
paramagnetic
termx~~~/(xL(
"
-(6h~/MR~) ~ m~(bf/b£mn).
As
usual,
it issimpler
to consider first thezero-temperature
case and then to useequation (9).
At T
=
0,
bf/b£
=
-b(~
El andxj~~ (E)
= 12 XL~j m~b(~$~ i~), (23)
m,n
where
i~
=
4~/A.
Afterapplying
the Poisson summationformula,
m and n go over into continuous variables: m ~ x,(n
+~)
~ y. Theintegral
over y isimmediate,
due to the )-function. Theintegral
over x is done in thesaddle-point approximation, using
thelarge
parameter i and the continuous version of
equation (22).
Theresulting expression
forxj~~ (E)
is:
x~~~
(E)
= [XL
~
7~
+i~/~ ~j #(K~, K~)
+O(1)
,
~
(24)
K~,Ky
where the sum runs over I <
K~
< cc and 0 <2K~
<K~,
and#(K~, K~)
=
~~
cos~ l~~~) sin~/~ ~~~~)
cos21K~
sin~~~~) ~~K~
+j (25)
/~ KY KY KY
2 4A similar treatment of the
diamagnetic
term results in a term-31~/4 plus oscillating
correc-tions of order
@. Thus,
thelarge non-oscillating
terms cancel and the net result for xo(El
isgiven by
the second term inequation (24). Finally, using equations (9, 12),
we find:x =
lxL i~/~ ~j 4(Kz, Ky)
7~ ~(~~
sin~ )) (26)
K~,K~ Y
Thus,
in the temperature range A « T <r,
thesusceptibility
xis, typically,
of order(kFR)~/~
and can have either
sign.
Itoscillates,
as a function of ~, with aperiod
of order(~ A)~/~
ci r.In
Figure
2 we present x as a function of i. as obtained fromequation (26) (solid line).
For
comparison,
dotsrepresent
the result of a numericalcomputation,
based on the exactequation (10),
with £[ and £I'given by equation (20).
These oscillations reflect the structure of thedensity
of states smoothed over energy intervals AE£ r,
in the same way as thesharp peaks
of thelow-temperature regime
reflected the exact(discrete)
spectrum of the system.Equation (26)
can be obtained from thecorresponding expression
of reference[4],
in the B ~ 0 limit. The derivation in reference [4] was based on a semi-classicalapproximation
for thedensity
of states
[8,10],
with asubsequent
introduction of themagnetic
field via the classical action.In
contrast,
we have started with an exactexpression
for the linearsusceptibility
and used(in
themesoscopic
temperatureregime)
a semi-classicalapproximation
for the energy levels of thesystem.
In thisrespect
our derivation is similar to that ofDingle [13].
He calculated theorbital
magnetization
of a thincylinder
and a smallsphere,
and obtained anexpression
for x, which containedoscillating
terms as well as alarge steady
term. His result for theoscillating
terms appears to be
correct,
while thesteady
term is due to an error.Q~ ~
~
l%
~ ~
~-l
~ -2
-3
40 50 60 70 80
k~R
Fig.
2. Linearmagnetic susceptibility
in a disc geometry, calculated at T= 5zh and normalized to XL
((kFR)~/~. Analytic
result obtained fromequation (26) (solid line)
iscompared
to numerical one based on equation(35) (dots).
5.
Square Geometry
Here we discuss electrons within a square of size L. The
unperturbed energies
are£nm =
(~~h~/2fi£L~)(n~
+m~).
A statein, m)
isdegenerate
with the state[m,n) (there
can
be,
inaddition,
accidentaldegeneracies
if apair n',
m' has the same sum of squares as thepair
n,m).
Let us first consider low temperatures, T « A e
2~h~ /ML~,
and discuss theparamagnetic peaks
due to the second term inequation (10).
The first order correction£[~
is due tolifting
of the double
degeneracies by
themagnetic
field.(We
do not consider accidentaldegeneracies, although
the treatment isreadily
extended to include that case aswell.)
Thedegeneracy
is liftedonly
if n and m have differentparity
and in that case~~~ ~~"
~2 &Ic
(n2 m2)3
I
3~~
d
~
~
~
~
20 30 40 50 60 70
k~L
Fig.
3. Linearmagnetic susceptibility
in a square geometry, calculated at T = 0.lA and normalized tojxLiikfR)~/~
20 40 60 80 100
k~L
Fig.
4. Linearmagnetic susceptibility
in a square geometry, calculated at T= 5 A and normalized to (XL(.
Analytic
result obtained from equation(28) (solid line)
is compared to numerical one basedon equation
(35) (dots).
The
largest
corrections occur when n= m + I. In such cases
£[~
ci(heB/fi£c)kFL
and theheight
of thecorresponding peak
insusceptibility
is xmax ci XL(kFL)2A IT, just
as in the case of the disc.Note, however,
that in the squaregeometry
suchlarge peaks
are much more rare than in a disc. For a disclarge peaks
wereseparated by
a distance of order A. For a square suchpeaks
occur,roughly,
for eachpair (n,
n +Ii,
I.e. areseparated by
a distance of order n/h cikFL/h
ci r e AUFIL.
The difference between a square and a disc isclearly
seen, if one comparesFigure
3 toFigure
I.Except
for thelarge paramagnetic peaks
inFigure
3 one cansee an
oscillating background.
Thisbackground
comes form the first term inequation (10).
It will be shown below that this term can be either para- ordiamagnetic
and itstypical
value is of the order(kFL)~/~.
In the
mesoscopic temperature regime,
/h < T <r,
thesusceptibility
x is describedby
the semi-classicalexpression (13).
This case was studied in detail in reference [6] andparticularly
in reference [4] where
expressions
for the factorsd>,r
and(A()
can be found. Theresulting expression
for x is:X 8
~
~j3
/2f ~j
~~~~~~~ ~~~
~~~~~
(XL
5@
~~ ~ ~
~l/2
(q~2 +q~2)~/~
q~2q~2~
/d/
~ ~ ~ ~2~rfi~T
xll
,
(28)
F
where u~ and uy are
positive coprime integers,
which labelprimitive
orbits.Only
odd u~ anduy enter the sum in
equation (28),
since otherwise the area enclosedby
an orbit isexactly
zero [4].
As an
example,
inFigure
4 wepresent xl
[XL as calculated fromequation (28) (solid line).
We chose the same ratio
Tllh
= 5 as in reference
[4].
The result ispractically indistinguishable
from the numerical one
(dots),
based onequation (10) (Let
us note, that in reference [4] somedisagreement
betweenequation (28)
and numerics wasobserved).
There are somequalitative
differences between themesoscopic
oscillations in the squaregeometry (Fig. 4)
and those for a circle(Fig. 2):
in the circle oscillations are modulated on an energy scale which exceeds rby
an order of
magnitude.
Comparison
with an exact numericalcomputation
demonstrates thatexpression (28)
is valid down totemperatures
T m/h,
but fails for T < /h.Nevertheless,
it can be used forestimating
the aforementionedbackground,
due to the first term inequation (10).
Thepoint
is that thisterm ceases to
change
whentemperature
is lowered from T m /h down to T = 0. To obtainan estimate of
expression (28)
at lowtemperatures,
we square it and average out the fast oscillations. Thisgives,
for thetypical
value ofx~
in thebackground,
X)ack
~X~ i~ ~~~~~~ ~/~~
~~~~~~~~~~~
~~~~
The sum over
repetitions
r is estimatedby replacing
it with anintegral,
with an upper cutoffprovided by
the function 11. Thisgives
alogarithmic factor,
so thatiXbacki
C~ (XL((kFL)~~~@@. (30)
Let us mention that a similar
logarithmic
factor appears in chaoticbilliards,
at low temper-atures
[5].
The behaviour of x in that case is, of course,quite
different from the squaregeometry:
the enhancement factor iskFL,
instead of(kFL)~/~,
and thelarge paramagnetic peaks disappear.
6. Harmonic Confinement
Our last
example
is adegenerate
electron gas confinedby
a harmonicpotential [14-17].
Theproblem
of orbitalmagnetism
in this case was consideredpreviously by
a number ofauthors,
and some
analytical
and numerical results have been obtained for various temperatures and fields. Below we shall derive ananalytical expression
for the linearsusceptibility
x, as a function of temperature[18].
Thisexpression
demonstrates the crossover fromstrong magnetic
effectstowards the weak Landau
diamagnetism,
under increase of temperature.We consider a
slightly anisotropic
harmonicpotential. U(x, y)
=
(M/2) (Hi x~
+Q2 y~ ),
withHi
close toQ2, namely [Hi Q2(
+ AR <hQ( /~.
In the absence of amagnetic
field the spectrum isgiven by
Enm =
hoi (n
+))
+hQ2 (m
+)),
where n, m=
1, 2,.
For theisotropic
case,Hi
=Q2
"
Q,
energy levels can be labeledby
asingle integer
I=
1,2,.
,
and the f-th level is t-fold
degenerate.
A smallanisotropy
~Qsplits
thedegenerate
levels into narrow"multiplets".
The width of the n-thmultiplet
is of orderthan,
which is~AQ IQ
formultiplets
near the energy ~. The above formulated
condition,
AR <hQ~/~,
isjust
therequirement
forhaving
well definedmultiplets.
It is clear onphysical ground,
and is verifiedby
the calculationbelow,
that such weakanisotropy
can affect thephysical properties
of the systemonly
attemperatures,
T <~AQ IQ.
When a weak
magnetic
field isapplied,
energy levelsacquire
a second order correction,,
1 eB ~
(n
+))Qi
-.
(m
+))Q2
~"'~
2
~
Mc
Q] -.Q]
~~~~Since for any finite
anisotropy
the first order correction is zero,equation (10) gives:
fi
"
~$
Qj Qj ~ (~l
+))fli (nl
+jlfl2j f(£nm). (32)
n,m
The effective radius R is determined from
MQ~R~/2
= ~, where Q
=
(Hi
+Q2)/2
mHi Applying
to the double sum inequation (32)
the Poisson summationformula,
we obtain for thezero-temperature susceptibility xo(E):
X°(~)
12j~
~~~i/> ,->y
[XL1
7(1 ~2j (~~) dY dx(z ~y)e~~~(lY+k~) ~~~j
I,k=-cu
where ~
=
Q2/Qi
and i =~t/hQ
=
kFR/2.
Calculating
theelementary integrals
andmaking
use ofequations (9, 12),
one can obtain a finalexpression
for thesusceptibility x(T).
Thisexpression
is rather cumbersome and will not begiven
here[19].
Fortemperatures
T »ihAQ
andanisotropy
AR « QIi
itsimplifies
to:~
= -l +
21~ f
cos(2~ri)11 ~~~
~(34)
(XL
~~~
F
where r
=
hvf/2R
=
hQ/2. Except
for theleading oscillating term,
of orderi~,
there are smalleroscillating
terms which are not written inequation (34).
Thisequation
does not containanisotropy,
which demonstratesthat,'for
T muchlarger
than themultiplet
widthihAQ,
theanisotropy
does not come intoplay (up
toexponentially
smallcorrections).
For T »r,
the oscillations inequation (34)
arenegligible
and x = XL For T£ r,
one cankeep only
the first term in the sum, which results in anoscillating
term of orderi~
ci(kFR)~.
ForihAQ
« T «r,
many terms contribute to the sum. The
resulting expression
exhibitsparamagnetic peaks,
ofheight (kFR)~r/16T
and widthT,
on adiamagnetic background
of order-(kFR)~.
For T <
ihAQ, equation (34)
ceases to beapplicable.
Ananalysis
of the fullexpression [19]
for
X(T)
shows that it matches theequation (34)
at T mihAQ
and thatonly
minorchanges
inx(T)
occur when T is lowered belowihAQ.
This means that at lowtemperatures,
T <ihAQ,
the width of the
paramagnetic peaks
becomesihAQ
and theirheight
is of orderkFRr/hAQ.
This result is a manifestation of a
general
rule. If there is a cluster ofnearly degenerate
levels about some energy £c(the
width of the cluster b is much smaller than thetypical
levelspacing /h),
then for T > b the cluster behaves as asingle degenerate
level: itgives
rise to aparamagnetic peak
ofheight g(kFL)~/h/T,
where g is the number of non-zeroeigenvalues
of the matrix(I[iz[j)
in thesubspace
of the cluster. For T <b,
the width of the cluster becomesrelevant and the
peak
saturates at a value of orderg(kFL)~/h lb.
This is best seen if £( and £I' inequation (10)
are writtenexplicitly using
thesymmetric
gauge. This leads to:X
_
~~
~(i)r~)I) f(£1)
+~ '~~'~'~~~ /~
~~~'~~~
~~~~~~~l
~~~~j~~j
A~
~
iJ °
When b < T < /h the
paramagnetic
contribution of thecluster,
asgiven by
the second term inequation (35),
isapproximately -(3~/AM) f'(£c)
trL(,
where the trace is within thesubspace
of the cluster.Estimating
the trace asgh~(kFL)~
leads to theexpression g(kFL)~/h IT
for theparamagnetic peak.
Let us,
finally,
mention that when theanisotropy
ARapproaches
the value QIi ii.
e.neigh- bouring multiplets
start tooverlap),
the wellpronounced paramagnetic peaks disappear,
al-though
the oscillations are still of order(kFR)~.
7. Conclusions
We have calculated the linear
magnetic susceptibility x(T)
for several two-dimensional inte-grable systems. Generally,
there are three distinct energy scales: levelspacing /h,
inverse time offlight
r and the Fermi energy £F.For T <
/h,
thesusceptibility
is sensitive to the detailed structure of the energyspectrum
as well as the matrix elements of the
angular
momentumoperator.
Inparticular, degeneracies
lead to
large paramagnetic peaks
of orderg(kFL)~/h IT,
gbeing
the leveldegeneracy.
At such low temperatures thesample dimensionality
is of noimportance
and similarpeaks
exist also in three-dimensional systems withdegenerate
levels[20].
For apair
ofnearly degenerate levels,
when the levelseparation
b </h,
theheight
of thecorresponding peak
saturates at T m b.The
typical
value of x between thepeaks
is also non-universal: forsystems
with rotationalsymmetry
it is of order XL(kFL)~
whereas for a square it is of order [XL((kFL)~/~ ln(kFL)
and can be either para- or
diamagnetic.
For the
mesoscopic temperatures,
/h « T£ r,
the orbitalmagnetic susceptibility
remainslarge.
In units of[xL(,
it is of order(kFL)"
andoscillates,
as a function of the chemicalpotential
~t, with aperiod
A~t ct r. Forgeneric integrable systems
a =3/2 (compare
toa = I for chaotic
systems [3]).
Forquasi-one-dimensional
systems(a strip)
a=
1/2
and for aharmonic
confining potential
a = 2.Thus,
the harmonicpotential
is a veryspecial
case evenamong the
integrable systems.
Thepoint
isthat,
for theisotropic
case and at zeromagnetic field,
the two-dimensional harmonicpotential
exhibits "accidental"degeneracies,
which are not related to the rotational symmetry(alike
the "accidental"degeneracies
in thehydrogen atom).
Finally,
for r « T « ~t, alllarge
orbitalmagnetic
effectsdisappear
and x becomesequal
to the Landau value XL-
Thus,
for themacroscopic
limit to beachieved,
it is not sufficient to have T »/h,
as onemight naively expect.
A much morestringent condition,
T »r,
isneeded. We close the paper with the
following
remarks:(ii Although
our calculations have been done for thegrand-canonical ensemble,
one canimmediately
infer about thepicture
for the canonical ensemble. For T «/h,
the sus-ceptibility
can be very sensitive to the exact number ofparticles
in thesystem, provided
that the last
occupied
level isdegenerate.
If such a level ispartially occupied,
the sys- tem isparamagnetic
with x Cfc[xL((kFL)~/h/T,
where the factor c accounts for thedegeneracy
and occupancy of the level. For afully occupied level,
and in the presence of rotationalsymmetry,
the response isdiamagnetic,
with x cixL(kFL)~. (The
discus-sion can be
generalized
to the case of a cluster ofnearly degenerate levels,
as was done above for thegrand-canonical ensemble.)
Formesoscopic temperatures,
/h « T£ r,
the standard transition from the
grand-canonical
ensemble to the canonical one ap-plies,
I.e.xcan(N)
=
xgr(~t(N)). Indeed,
the thermal fluctuation IN ctfi
in the number ofparticles
is much smaller than thechange
in the number ofparticles, (AN)
ci(A~t)(bN/b~t)
ciTllh, corresponding
to asignificant change
in x.(iii
So far we have discussedonly
the orbitalmagnetic susceptibility.
The Zeemansplitting,
due to the electron
spin,
also contributes. Within the linear response its contribution xs issimply
added to the orbitalpart
of thesusceptibility.
This isclearly
seen form equa-tion
(10),
if the Zeemansplitting +~tBB
is included into the first order energy correction£(.
Thisgives
xi=
A~~~t[(bN/b~t),
where N=
£~ f(£i).
At lowtemperatures,
T </h,
xs exhibits
paramagnetic peaks
xs ci [XL/h/T,
which isjust
the Curieparamagnetism
due to the last
occupied
level. Fortemperatures
T »/h,
xs isgiven by
the Pauli param-agnetism,
xs"
3[xL(,
up to small corrections.Thus,
in this case, there is noappreciable
mesoscopic
effects due to the electronspin.
(iii)
Themagnetic
field in this paper was treated as agiven (homogeneous)
externalfield, Bext.
Forsufficiently large
x,however,
the orbitalmagnetic
currentsflowing
in thesample produce
a fieldBcurr
which iscomparable
toBext.
In three-dimensionalsaniplu
this
happens
when themagnetization
M becomescomparable
toBe,t/4~.
orix
C~1/4~.
In two dimensions the
magnetization
is defined as themagnetic
moment per unit ar(>a(rather
than per unitvolume),
so that x has units oflength. Also,
since the thickness uf thesample (height
inz-direction)
is much smaller than its sizeL, Bcurr
differs very flinch from the volumemagnetization (demagnetization effect).
So one has to estimate the fieldBcurr
and compare it toBext.
The moststringent
limitation is setby
therequirement,
that the
largest possible
value the local fieldBcurr
can assume should be much smaller thanBext.
Thisyields
[9][x[kF In(kFL)
« I, or
[x/xL(
<In(kFL)/kfre
,
where re
=
e~/fi£c~
is the classical electron radius.
(Though,
in some cases, e.g. in a squaregeometry,
the condition is less severe,namely ix /xL(
« LIn(kFL) Ire.)
As anexample
consider a discgeometry,
where atparamagnetic peaks x/(xL(
Cf(kFL)~/h/T.
IflF
Cf10~~
cm andkFL
ct100, then,
aslong
as T »10~~/h,
the self-consistent treatment is not necessary.If the condition is not
satisfied,
one cannot assume anymore agiven
external field. but should solve the entireproblem self-consistently.
A self-consistent treatment leads to suchpossibilities
as Meissner effect and orbitalferromagnetism,
both in bulksamples
andrings [2, 3, 20-23].
Aninteresting possibility
appears to exist in thestrip geometry,
considered in Section 3. Here the linearsusceptibility, equation (18),
is xo Cf(xL(@~. However,
the
differential magnetic susceptibility,
xd, in a finite field B$ &IcvfleLy,
turns out to be of order [XL[(kFLy)~/~ (see
Ref.[12],
where astrip
with harmonic confinement wasanalyzed) [24].
If[xo[
«lF /4~ln(kFL)
but[xd[
>lF/4~ln(kFL),
then one encounters a situation similar to the one which can occur in the deHaas-van-Alphen
effect and leads to a break up of thesample
intomagnetic
domains[25].
Acknowledgments
This work was
supported by
the Fund for the Promotion of Research at the Technion.References
iii
Landau L-D- and LifshitzE-M-,
StatisticalPhysics,
3rdEdition,
Part1, (Pergamon Press, 1980).
[2]
Ginzburg V-L-, Uspekhi
Fiz. Nauk 48(1952)
25(German
translation in: Fortscht.Phys.
1
(1953) 101).
[3]
Shapiro B., Physica
A200(1993)
498.[4j Richter
K.,
Ullmo D. and JalabertR-A-, Phys. Rep.
276(1996)
1 and references therein.[5j
Agam O.,
J.Phys.
I France 4(1994)
697.[6] von
Oppen F., Phys.
Rev. 850(1994)
17151.[7] Akkermans E. and
Shapiro B., Europhys.
Lett. il(1990)
467.[8] Gutzwiller
M-C-,
Chaos in Classical andQuantum
Mechanics(Springer, 1990).
[9] Gurevich
E.,
Ms. Thesis(Technion).
[10] Berry
3l.V. and TaborM.,
Proc. R. Sac. Land. A349(1976)
101.[11]
FriedmanL., Phys.
Rev. 134(1964)
A336.[12] Hajdu
J. andShapiro B., Europhys.
Lett. 28(1994)
61.[13j Dingle R-B-,
Proc.Roy.
Sac. Land. A212(1952) 47;
see alsoBogachek
E-N- andGogadze G-A-,
Sov.Phys.
JETP 36(1973)
973.[14]
NemethR.,
Z.Phys.
B81(1990)
89.[15]
Yoshioka D. andFukuyama H.,
J.Phys.
Sac.Jpn
61(1992)
2368.[16]
MeirY.,
Entin-Wohlman O. and GefenY., Phys.
Rev. 842(1990)
8351.[17]
AzbelM.,
Solid State Commun. loo(1996)
195.[18]
Theanalysis
can be extended to anarbitrary magnetic
field(Ref. [19] ).
[19]
Gurevich E. andShapiro B., preprint (1997)
[20]
See e-g- BuzdinA-I-, Dolgov
O-V- and LozovikY.E., Phys.
Lett. looA(1984)
261.[21]
KulikI.O.,
Pisma Zh.Eksp.
Tear. Fiz. ll(1970)
407(JETP
Lett. ll(1970) 275).
[22]
WohllebenD.,
J. Less Common Metals138(1988) II;
Wohlleben D. etal., Phys.
Rev.Lett. 66
(1991)
3191.[23]
AzbelM.Ya., Phys.
Rev. 848(1993)
4592.[24]
This can beexplained
in terms of the semi-classicalpicture.
In thestrip geometry
theleading
contribution to xosc comes from the"bouncing-ball" trajectories (at
B= 0 these
are
self-retracing ones).
At zeromagnetic
field an area enclosedby
thetrajectories
isexactly
zero. This area increasesalong
with B and becomes of orderL(
at B ~mcvfleLy,
what