HAL Id: jpa-00247103
https://hal.archives-ouvertes.fr/jpa-00247103
Submitted on 1 Jan 1995
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.
Level Spacing Statistics of Bidimensional Fermi Liquids.
II. Landau Fixed Point and Quantum Chaos
R. Mélin
To cite this version:
R. Mélin. Level Spacing Statistics of Bidimensional Fermi Liquids. II. Landau Fixed Point and Quantum Chaos. Journal de Physique I, EDP Sciences, 1995, 5 (7), pp.787-804. �10.1051/jp1:1995169�.
�jpa-00247103�
Classification Physics Abstracts 05.20-y 05.30Fk
Level Spacing Statistics of Bidimensional Fermi Liquids.
Il. Landau Fixed Point and Quantum Chaos
R. Mélin
CRTBT-CNRS, BP 166X, 38042 Grenoble Cedex 09, France
(Received
10 November 1994, revised 27 February 1995, accepted 29 March1995)
Abstract. We investigate the presence of quantum chaos in the spectrum of trie bidimen- sional Fermi hquid by means of analytical and numencal methods. This model is integrable m a
certain lirait by bosomzation of the Fermi surface. We study the effect on trie level statistics of
the momentum cut-off A present in the bidimensional bosonization procedure. We first analyse
the level spacing statistics in trie.A-restricted Hilbert space in one dimension. With g2 and g4 interactions, the level statistics are found to be Poissonian at low energies, and G-O-E- at
higher energies, for a given cut-off A. In order to study this cross-over, a finite temperature is introduced as a way of focussing, for a large inverse temperature fl, on the low energy many-
body states, and driving the statistics from G-O-E- to Poissonian. As for as two dimensions
are
concerned, we diagonalize trie Fermi liquid Hamiltonian with a small number of orbitals. Trie level spacing statistics are found to be Poissonian
m the A-restricted Hilbert space, provided trie diagonal elements are of trie same order of magnitude as the off-diagonal matrix elements of trie
Hamiltonian.
1. Introduction
The ideas of quantum chaos bave
recently
beenapplied
to trie field ofstrongly
correlated electrontheory (1-3].
These methods allow for anon-perturbative description
of trie statisticalproperties
of a Hamiltonian ofstrongly
correlatedelectrons,
and may be a tool to extract some information from finite size systems. Trie atm of this article is to understand whether the methods of quantum chaos can shed some newlight
on trieproblems
ofstrongly
correlatedbidimensional Fermi systems. Trie
question
whether trie bidimensional Hubbard model is a Fermiliquid
or net is controversial. From a theoretical point of view, Anderson (4] suggests that theground
state of the bidimensional Hubbard model is similar to that of the one-dimensionalLuttinger liquid,
with Z= o in the
thermodynamic limit,
whereasEngelbrecht
and llmnderia [Si argue that theFermi-liquid theory
is notviolated,
so that thisquestion
is controversial [6].Numerical
computations
also lead to controversial answers. For instance,Dagotto
et ai.ii,
8]observe
quasiparticles
in trie bidimensional t Jmodel,
whereas Sorella [9]emphasizes
trieLuttinger liquid
behaviour. From trie point of view of quantum chaos, one bas to answer trie question: does levelspacing
statistics contain information about trie Fermiliquid
behaviour or© Les Editions de Physique 1995
IOURNALDEPHYSIOUEL-T.s.NO7.JIJLY iras
non Fermi
liquid
behaviour of trie t J model? Beforeanalysing
trie levelspacing
statistics of trie t J model(loi,
weanalyze
in finis paper models with a well-establishedphysical
content,that
is,
trie Fermiliquid.
As we shall see, we can answeronly partially
trieprevious
question:one can detect
integrable
modes at low energy, but quantum chaosby
itself does notgive
information as to whether these modes arequasiparticle
modes orLuttinger liquid
modes.However,
there is one case in which one can condude from quantum chaos: in trie absence ofintegrable degrees
of freedom at low energy, one can conclude to trie absence of a Fermiliquid
at low energy. Notice here trie dilference between trie
approach
of referenceiii
and Durpoint
of view. In reference[Ii,
ail trie energy levels of trie t J model areanalyzed
on anequal footing,
whereas we shall focus
essentially
on trie low energydegrees
of freedom of trie Fermiliquid.
In trie first article of this series
[iii,
we bave established that trie levelspacing
statistics of trie bidimensional Landau Hamiltonian is Poissonian. In finismodel,
triequasipartides
are ina one-to-one
correspondence
with trienon-interacting
gas ofspinless
electronexcitations,
and thus occupy orbitals labelledby
trie same quantum numbers k as in trie case of trie gas. Triequasiparticles
interact among themselves in adiagonal
manner~~~~~~
~Î =~
£kôYlk +~
k
LD
Îkk'ôYlkôn~,
~~'~
l
A numerical
computation proved
that trie levelspacing
statistics of trie Hamiltonian(1)
isPoissonian in two
dimensions,
and close to Poisson in onedimension,
with our truncation of trie Hilbert space. Trie notion of level spacing statistics seems to be relevant for quantumfluids,
and one bas todistinguish
between one and two dimensions. Trie link between triebreakdown of trie Fermi
liquid picture
in one dimension and trie levelspacing
statistics basalready
been studied (13]. As for as two dimensions areconcerned,
we showed in paper Iil Ii,
that trie Landau
liquid
was characterizedby
itsgenerical integrability, namely by
Poisson levelspacing
statistics in two dimensions.Again,
trie levelspacing
statistics are agood
trot to see whether trie Fermi system is at trie Landau fixed point or net. For trieliquid
to bea Fermi
liquid,
one should be able to generatequasiparticles
of trieinteracting
systemby
aswitching
onprocedure.
In trie framework of trie Landautheory,
trie success of trieswitching-
on
procedure (14, lsj
suggests the conservation of the number of conservedquantities
at Iow energyduring
theswitching-on procedure, namely
that the level spacing statistics of the gas and of theinteracting liquid belong
to the sameuniversality
class.Typically, adding
a non-diagonal perturbation
to equation(1)
andobtaining
a GaussianOrthogonal
Ensemble(G.O.E.)
level
spacing
statistics would mean adeparture
from the Landau fixed point. To be at the Landau fixedpoint,
one must exclude strong correlations between levels. Thecorresponding
generic level
spacing
statistics is Poissonian.However,
the case of the one-dimensional Landau Hamiltonian shows that the statistics may not beexactly Poissonian,
but close to a Poisson law.Poisson or close to Poisson level
spacing
statistics are not a sullicient condition for the systemto exhibit
quasipartides
since one couldimagine
a situation in which trie same mechanismas in one dimension for trie breakdown of trie Fermi
liquid froids,
so that trie Ievelspacing
statistics would remain Poissonian even in trie non-Fermi
liquid
case.However,
trie adiabaticprocedure,
asemphasized by
Anderson [14], isperformed
within a finite characteristic timeIle,
and triethermodynamic
limit is taken for a finite e which is then muchlarger
than trietypical
levelspacing.
Thus trie adiabatic continuation does not generate trueeigenstates (for
this to be trie case, one should take trie limit e - o
first,
before triethermodynamic limit).
This is
why
trie notion of adiabatic continuation advocated to introduce Fermiliquid
is weaker than a similar requirement for ail trieeigenstates
takenseparately. Therefore,
this iswhy
weshould ask whether a Fermi
liquid
follows Poisson or G-O-E- statistics.The atm of the present paper is to
study
the levelspacing
statistics of the Hamiltonian ofspinless
electronsH =
H°+H~ (2)
H°
=
£e(k)c(ck (3)
k
~fl
~
j~ f~~,
~+ ~+ ~~,~~j~)
~v >q k+q k'-q
k,k',q
where V is the volume of the system. The Harniltonian
(1)
describes a Fermiliquid
for times smaller than thedecay
time of thequasiparticles,
and leads togood thermodynamical predictions
[15].However,
triephenomenon
ofdecay
of triequasipartides
is not describedby
the Landau form
Il).
Sincequasipartides
are trueeigenmodes
ofIl), they
have an infinite tife-time.By
contrast,(2)
takes into account thedecay
of triequasiparticles
and it is indeedpossible
to calculate triedecay
rate of triequasipartides
from trie Hamiltonian(2) (16].
Eventhough
trie Hamiltonian(2)
is notdiagonal,
it can bebrought
to adiagonalform by using
someassumptions
andby bosonizing
trie Fermi surface [18]. We shall review trie mainassumptions
and trie bosonization of
(2)
in triethermodynamic
limit and in trie Iimit of a zero curvature of trie Fermi surface. One of trieingredients
of trie solution via bosonization is trie existence ofa momentum A which determmes trie
single partide
states whichparticipate
in trie formationof bosons in trie
vicinity
of trie Fermi surface. A is a necessaryingredient,
because of trie curvature of trie Fermi surface, but is also a source ofdifliculty
since trieabjects
with a true bosonic character in trie limit A - +co are nolonger exactly
bosonic in trie limit of a finitemomentum cut-off A.
Of course, numerical
computations
cannot beperformed
in triethermodynamic
limit sinceone can
only diagonalize
matrices of size 2000by
a Jacobi method. This technical limitation imposes to work with small systems and to impose a drastic cut-offA,
so that trie Hamiltonian(2)
is nolonger integrable
in trie framework of trie Hilbert space of trie numericaldiagonal-
isations. In trie case of trie presence of a
cut-off,
and for finite size systems, one may thus expect trie universal Poisson Ievelspacing
statistics to bereplaced by
trie universal G-O-E- level spacing statistics of trie formP(s)
=~s
exp
~~
(5)
2
~
It is necessary to determine trie importance of trie cut-off A on trie level
spacing
statistics and whether trie statistics evolves towards G-O-E- levelspacing
statistics as A is reduced. This bas netonly
anumerical,
but aise aphysical
interest. Forinstance,
lattice models bave a natural cut-off A+~
1la,
where a is trie latticespacing.
To answer thisquestion,
we came back to one-dimensionalspinless
fermionic systems, but with a cut-off A andstudy
trie levelspacing
statistics for gz and g4 interactions. As we shall see, trie levelspacing
statistics isdrastically
affected
by
trie presence of trie cut-off since trie bosonicsuperpositions
ofpartide-noie pairs
which guarantee trieintegrability,
do not survive in trie presence of trie cut-off. If trie cut-off A is fixed and trie size L of trie systemincreases,
one expects for a crossover from a G-O-E- Ievelspacing
statistics m triehigh
energy part of trie spectrum to a Poisson levelspacing
statistics at low energy, which shows that finite size effects are drastic in trie presence of trie cut-off. Inorder to determine trie crossover scale between trie Poisson
regime
and trie G-O-E-regime,
weintroduce
temperature-dependent
levelspacing
statistics. The statistics at finite temperatures is found to pass from anon-integrable
statistics athigh
temperatures to Poisson statistics at low temperatures.As far as two-dimensional systems are
concemed,
we carry out numericaldiagonalisations
of a small system of
electrons,
with a small number of orbitals. Under theseconditions,
the Hamiltonian(2)
is nonintegrable.
We found Poisson levelspacing
statistics even for a small system. We attribute this property to the fact that the interactions generate extradiagonal
matrix elements m trie Hamiltonian which are in
competition
withdiagonal
interaction matrixelements,
but trie interaction connectsonly
a small number of states,leaving
a lot of zero matrix elements in trie Hamiltonian. Provided trieoff-diagonal
matrix elements are of triesame order as trie
diagonal
ones, trie statistics is dominatedby
triediagonal
matrix elements.However, by keeping only off-diagonal
interaction matrixelements,
we were able to exhibit G-O-E- levelspacing
statistics.This paper is
organized
as follows. We first treat the one-dimensional case, in the Hilbert space restrictedby
the momentum cut-off A. The results established in one dimension arehelpful
in two dimensions since the presence of the momentum cut-off is aise a source of nonintegrability.
In a second step, we came to the bidimensional Fermihquids,
with thestudy
of the level spacing statistics of the Hamiltonian(2).
2. Level
Spacing
Statistics of lDSpinless
FermionSystems
in trie A-RestrictedHilbert
Space
2.1. BOSONIZATION AND LEVEL SPACING STATISTICS IN THE UNRESTRICTED HILBERT
SPACE
(LUTTINGER LiouiD). Using
the bosonizationprocedure
of Haldane [22], we carisalve the one-dimensional
spinless Luttinger liquid
with a lineardispersion
relation and g2 and g4 interactions. The Hamiltonianreads,
in term of fermionsH
= VF
~lak kF)
:
cÎ,ack,« +) ~ ~
qlg4qôoe«> +g~qôa,-«, )pq,ap-q,a,,
16)k,a a,a,
where the label a indexes the branch
a=R(ight),L(eft)=
+1, -1. The interaction g4 describes the diffusion of two fermions on the sameright
or leftbranch,
whereas g2 descnbes the diffusion of two fermionsbelonging
to theright
and left branch. Thedensity
operators are defined asPQ,a "
~ cÎ+q
rock,a.,
Ii)
k
and
obey
bosonic commutation relationswhich allows the definition of boson operators
~~ ~/Îl
~~~~Î
~~~~~~~~~~ ~~~and make it
possible
todiagonalize
the Hamiltoman(6)
via aBogoliubov
transformation:where
b)
= coshçJqa)
sinh çJqa-q(11)
g2q
~~~~
~~~~Î) = ÎÎÎj-19~~~)~[(l~' ~~~
~~ ~ ~j~~ + g~o)
-
g~o)vN "
vs exp
The interaction functions
g2(q)
andg4(q)
aresupposed
to tend to a constant in the limit q - 0, and to zero in the limit q - +co. Their decrease is controlledby
theimpulsion
scale27r/R,
where R is a
given length
scale. In thethermodynamic
limit(L
»R),
one can define an effective low energy Hilbert space. To do so, we assume that wq m wo for ail the wave vectorsq. Of course, this approximation is
only
valid if the temperature is lowenough.
In thislimit,
the Hamiltonian
(10) depends only
on two velocities vN and vj:H =
Eo
+£lvNvJ)~/~lqlblbq
+)lvNN~
+VJJ~) [ii)
q#o
In one dimension, the hnk between the breakdown of the Fermi
liquid
picture and the level spacing statistics is well understood. The absence of a Fermi surface in onedimension,
in the presence oflong-range
g2 and g4 interactions was known toDzyaloshinskii
and Larkin in the 70's[17j.
The breakdown of the Fermiliquid picture
isalready
present at the level of the staticcorrelation
functions,
and in the Hilbert space of low energy. The usual infrareddivergencies
are
govemed by
the q - 0 limit ofg2(q)
andg4(q),
and are related to theorthogonality catastrophe
and the absence of a Fermi surface. The infrared spectrum is describedby
theHamiltonian
(Ii).
If the totalcharge
and current quantum numbers aregiven,
the q#
0 excitations are bosons with a lineardispersion relation,
but with a renormalizedvelocity.
Thedeparture
from the gas behaviour is measuredby
the anomalous exponents which appear in the static Green functions. The interaction energy scale associated with this static breakdown of the Fermiliquid
isgiven by
[13]L -1/2
g]~~~ +~ vF In
(18)
27rR
However,
asemphasized
in reference [13], the static part of the Green function does not contain ail thephysics
of the breakdown of the Fermihquid theory.
From a calculation of the two- point Green function, and from a calculation of theswitching-on procedure,
we found in (13]a
dynamicai
breakdown of the Fermiliquid
picture, controlledby
the interaction scale~~~~ ~
(kL~ÎÎΰ)1/2'
~~~~where a is defined
by
g2q " g2011
lqR)~). (20)
The static process of breakdown of the Fermi
liquid
picture does notmodify
the levelspacing
statistics, which remamsingular,
as m the case of the gas, smce thedispersion
relation of the bosons remams linear.By
contrast, thedynamical
breakdown of the Fermiliquid
appears to be related to a cross-over in the energy levelspacing
statistics,namely
tramsingular
levelspacing
statistics to
generical
Poisson level spacing statistics. As far as symmetries are concemed, the gas case and theLuttinger liquid
possess the conformai symmetry. The static breakdown of theFermi
liquid
preserves the conformai invariance. Thetheory
is theantipenodic-antiperiodic
sector of the
compactified boson,
with aninteraction-dependent
radius ofcompactification.
The
dynamical
breakdown of theLuttinger liquid corresponds
to a massless loss of conformai invariance (12].2.2. FOURIER TRANSFORM OF THE CAS SPECTRUM AND TEMPERATURE-DEPENDENT LEVEL
SPACING STATisTics. We first wish to characterize the spectrum of the gas m the absence of interactions, a momentum cut-off A and a finite size L. The idea is to find a criterion to detect when a boson of
given
wave vector q is present or net in the system. Since g2 and g4 interac- tions arediagonal
on the basis of bosonic excitations, this tool is agood
way to characterize thedegree
ofintegrability
of trie mortel with a momentum cut-off. We consider a one-branchmortel,
and trie number of quantum states issimply
trieinteger
part of2AL/27r.
Trie spec-trum is made up of levels with an
equidistant separation
27rvFIL.
In order to characterize trie spectrum, we use its Fourier transform+m
fL(T)
=
/
e~~~£ à(w E,)
=
£ e'~~~ (21)
~" lE,j jE~)
In trie
thermodynamic limit,
1
(23)
l~~~~~É ~°(fm(T)Î
"~
2(smvfTq/21
q>o
This function presents
potes
which are characteristic of trie bosonic modes for Tn,q = 27rnIv
fq,with n an integer. If we rescale T
by
a factor 27r and choose vi = 1, we obtain apote
for Tn,q=
n/q,
that is for every rational number. Weplotted foc(T)(
for 5 bosonic modes inFigure
1. One caneasily recognize
trie dilferent bosons in trie sequence ofpotes.
In trie presence ofinteractions,
triepotes
aredeplaced.
We also studied trie Fourier transform of trie spectrum in trie case of a finite size system. Trie result isdepicted
inFigure
2. We can see that triepotes
are not so well defined. However, we canrecognize
trie formation ofpeaks
whichreplace
triepotes,
and attribute well defined boson wave vectors to somepeaks.
Triecorresponding
truncated
density
operators are definedby
pA(q)
=~
ÙA(k)ÙA(k
+q)c(~~ck, (24)
k
where ÙA(k) = Ù(A (k
kf().
We can ask under which condition apeak corresponding
toa truncated boson
pA(q)
appears on trie modulus of trie Fourier transform of the spectrumfL (T)
(. To answer thisquestion,
we assume that trie number nq of bosonic excitations of wavevector q
appearing
in the presence of a cut-off is such that qnq +~2A,
so thatfL(Ti,q)
z~J
2A/q
for trie
peak
at T = ri,q. A boson of wave vector q is well definedprovided fL (ri,q)
» 1, that is if A »q/2.
For agiven cut-off,
trie statistics isexpected
to be Poissonian at lowenergies,
and G.O.E. athigher
energies. This is due to trie fact that truncated bosons are createdessentially
at low energy. In order to test this
idea,
we introducetemperature-dependent
level spacingstatistics,
which is defined as follows. If(E,)
is trie fuit spectrum and(e,
are trie energy levelsio
9
8
7
6
s
4
3
2
1
0
0 1 2 3 4 5 6 7
Fig. 1. Modulus of the Fourier transform of the spectrum m the thermodynamic hmit
(fco(r)(
for 5 basons. Each pale corresponds to one or several definite boson excitations. We have chosen vi= 1
and L
= 2x. A given boson q > 0 generates a sequence of pales at r»,q =
27rn/q.
The pales aredenoted by their corresponding rational number
n/q.
20
15
io
5
0
0 1 2 3 4 5 6
Fig. 2. Finite size effects of the modulus of the Fourier transform of the spectrum
(fL(r)(.
Wehave chosen vi = 1 and L
= 2x, A
= 11. As shown on the Figure, we con attribute boson quantum numbers to some peaks. We recognize trie fractions 1/5, 1/4,
1/3,2/5,1/2,3/5,2/3,3/4, 4/5.
after the
smoothing procedure
[23], thedensity
of level spacingP(s)
is-i
Ppis)
=
(~
exp-,o~~+~~+ ~) L
exp-à~~+~~+
~~ois ie,+i e,)) 125)
~ ,
Moreover, for the statistics to be
comparable,
one needs to scalePp (s)
and s such that(Pp(s
=1, and
(sPp(s))
= 1. The reasonwhy
we have to rescale thesequantities
is thatthey
wereequal
to
unity
after thesmoothing procedure,
which was carried ont in the zero temperature limit.This property is no
longer
valid with the statisticalweights
of equation(25).
For an infinite temperature(fl
=0),
we recover the usual level spacing statistics. As the inverse temperaturefl
increases, the low energy levels carry more and more statisticalweight.
As we saw from thedescription
in terms of the truncated bosons(24),
the spectrum isexpected
to beintegrable
at low
energies,
so that the levelspacing
statistics should evolve from G-O-E-statistics,
or at least intermediate statistics(that
is, with 0 <P(0)
<1),
to Poisson levelspacing
statistics asfl
increases.2.3. LEVEL SPACING STATISTICS ÀÀÎITH g2 INTERACTIONS IN THE À-RESTRICTED HILBERT
SPACE. The numerical method to compute the level
spacing
statistics of twc-branch models with a momentum cut-off A and g2 interactions consists intabulating
the states of the Hilbert spaceby generating
ail the dilferentfillings
of the two branches, the number ofpartiales
on each branchbeing kept
constant. The second step consists incomputing
ail the matrix elements of the Hamiltonian anddiagonalising
the Harniltomanby
the use of the Jacobi method. Theinteractions are of the form
~~~ ~~° ~~~
lÎÎ~
~~~~The scale of the interactions R is choseu
equal
toA,
so that ail the matrix elements of the interaction are important. The evolution of the spectrum as a function of g20 isplotted
inFigure
3. Thisfigure
is to becompared
withFigure
2 of reference[13j,
where we haveplotted
the evolution of the energy levels as a function of g20 but in the absence of cut-off. The obvions dilference between the two
plots
is the presence of level repulsion with a cut-offA,
and theexistence of level
crossings
m the absence of cut-off. The level spacing statisticscorresponding
to a spectrum
analog
to that ofFigure
3 areplotted
inFigure
4.They
are ingood
agreement with G-O-E- levelspacing
statistics. In ourcomputation, AL/27r
= 3, so that the criterion À »27r/L
is not verified. ~Ve could not go tohigher
values ofAL/27r
becauseincreasing
the ratio
AL/27r
increases the size of the Hilbert space and numericaldiagonalizations
areno
longer possible.
We conclude that the statisticsdepends drastically
on thelength
of the system, if the momentum cut-off is fixed. If weapply
the ideas oftemperature-dependent
statistics of level
spacings
to the system, we find that the statistics of levelspacings
is driven from a G-O-E- law at fl= 0 to a Poisson law as
fl
increases, as depicted inFigure
5. This evolution shows the fact that"nearly integrable" degrees
of freedom exist at low energy. These bosonicdegrees
of freedom areexactly integrable
m the absence of cut-off. In term of classicaltrajectones (provided
one is able to find a classicalphase
space for the Fermiliquid !),
this situationcorresponds
to the existence of conserved tori at low energy, which transform into chaotictrajectones
as the energy increases. The cross-over temperature scale will be derived later.25
io
s
~ _jj__j~~'~~-'---_
--.---______
~~ ~h'
àÙÎ~))Tlfifi==.
tm_jmmm-=- --.."~=
_ __ ---i =fi=+1
--= ==_=-
o ---. ..- -~= zm=- ~~~
~~~
~-
~
~
-5
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Fig. 3. Evolution of the energy levels in trie presence of g2 interactions with a momentum cut-off.
Trie parameters of trie model are ~i = 1 and L
= 2x. Trie cut-off is chosen equal to 3, so that we bave 3 fermions on each branch, for 6 available quantum states. Trie interactions are of trie form g2q =
g20exp(-q/R).
with R= 6. Trie momentum sector is 1, leading to 45 states m trie Hilbert
space. Because of the partiale-hale symmetry, trie spectrum is symmetric, but with no level crossings.
2.4. LEVEL SPACING STATISTICS WITH g4 INTERACTIONS IN THE À-RESTRICTED HILBERT
SPACE. We now consider the case of g4 interactions. The interaction Hamiltonian is
H~
=
£ £ g4qc)~~c$_~ck,ck (27)
q#o k,k,#k+q
The interest of the g4 term is that we can use
only
a one-branch model, and we can reachhigher
values of the ratioAL/27r
withoutincreasing
the size of the Hilbert space. We couldreach
AL/27r
= 12 in a sector of total momentum P =
24.27r/L.
P is theimpulsion
withrespect to the fundamental. As
long
as P <AL/7r,
one hasgenerated
thecomplete
Hilbert space in the absence of interactions. This fact motivates the choice of P=
AL/7r.
The conditionAL/7r
= 24 » 1 is
respected. However,
we did not find Poisson statistics, butours is intermediate between a Poisson law and G-O-E- statistics.
Namely,
the value of thedensity
of normahzed zerocrossings
is not 1 but o-à- The statistics isplotted
inFigure
6. Thiscomputation gives
an idea of the extension of the crossover as a function of AL/27r,
since weobtain intermediate statistics for
ÀL/27r
= 24. We alsocomputed
thetemperature-dependent
level spacing statistics. We found a cross-over to Poisson statistics as the inverse temperature
mcreases, as
plotted
inFigure
7. This cross-over is much morerapid
than in the case ofthe g2 interaction.
However,
the values of the cut-off in the g2 case and the g4 case are notcomparable.
2.5. CRoss-OVER SCALES. In order to get an idea of the temperature cross-over between the G.O.E. and Poisson regime, we look for the energy scale
kBT*
below which the energy levels1
Poisson G-O-E-
0. 8
0 6 ",
',, '
d
',,
',~
',,
~.
0 4 ~~~~~~~~ "
0 2
"""""
0
0 0.5 1 1,5 2 2.5 3 3.5 4
normalized level spacing s
Fig. 4. Level spacing statistics of ID spinless fermions with g2 interactions and a cut-off. The parameters are ~i = 1, L = 17r. In order to eliminate the particle-hale symmetry present in Figure 3,
we chose a non-symmetric cut-off for particles and hales: the number of fermions on the right branch
is 4, and the number of quantum states is 9. On the left branch, the number of fermions is 5 for 9 quantum states. The Hilbert space contains 1052 states in the sector of momentum 1. The level
spacing statistics is found to be well fitted by G-O-E- statistics.
are uncorrelated. To do so, we make the
approximation
that trieonly
effect of interactions is to compress trie spectrumby
an amount((1+ g4/ui)~ (g2/vf)~))~/~,
where g2 and g4 aretypical
energy interactions. This
compression
of trie spectrum means that trie effective mass increases if g2 islarge compared
to g4. This approximationcorresponds
totaking
local interactions intrie real space. Then we see that trie
required integrable
modes are such thatvi (q( <
1
+ ~~ ~ ~~
)
~~~kBT.
(28)
Vi Vi
These modes are
integrable provided
thatthey
lead to well-defined bosons. We say that a boson is well-definedprovided
that a sufficient number ofpartiele-noie
excitations enter trie summation(24),
that is, if 2À (q( >27ra/L,
where a is a dimensionlesscoefficient,
whichcounts trie number of non~zero terms in
(24),
and is thus a measure ofintegrability.
We getkBT
t 12À27ra/L) (il
+g4/vf)~ ig2/vf)~)~~~ 129)
The energy scale kBT* thus increases as trie cut-off A increases, and decreases if trie interaction
strength
g2 increases.1
Poisson
0.8 ~'~'~'
Î ,'
Î
--~
Î
j ~
, ,
0.6 '
: ,
',
: .,.
"
w '-.
£ ". ,
", ,,
., ,
0,4 ',
', N
j .,
, ',
' ...
1
, .,
1 '
0
level s
Fig. 5. -
emperature. Trie are vi = 1, L = 27r. In
order
to eliminate
trie symmetryresent in Figure 3, we chose a non ut-off for particles and noies: trie number of fermions on trie right ranch is 4,
and the number of uantum states is 9. On the left branch,
trie number
is 5 for 9 quantumstates. Trie ilbert space contains
1052
tates in the sector of omentum 1. Trie
level
tatistics is plotted for inverse temperature equal to fl = 0, 0.1, 0.5 A cross-over is found
from G-O-E- statistics for fl = 0 to Poisson tatisticsas fl
3. Level
Spacing
Statistics of a 2DSpinless
FermionSystem
in trie A-Restricted HilbertSpace
3.1. BOSONIZATION OF THE BIDIMENSIONAL FERMI SURFACE. TÎ1e bosonization of tÎ1e
Fermi
hquid (19-2 Ii
involves acovering
of trie Fermi surfaceby spheres
of radius A. Each smallsphere
is labelledby
aninteger
a and one assumes that trie Fermi surface is flat at trie scale A.Assuming
that A » 27rIL,
trie commutation relations of trie operatorspq,a =
£ 9«lk
+q/2)9alk q/2)nqlk), 130)
Wiill
'lq(~) C~-q/2~k+q/2>
(~~~must indude a
Schwinger
term,leading
to trie commutation relationlpq,~, p(,~>l
=ôa,~,ôq,q, £ c~jk, q)jnl+
j
nl- j), 132)
k
where trie constraint
C~
is written asCoElk>~)
" ÙoElk +))ÙoElk ))ÙoElk'+ ( )ÙOElk'~ ( ).
133)1
Poisson
", G.O.E.
0 8 1,
~, "..
', ." ".
',
," ",
~,, ." "
0 6 ,~'
"_
Îi ".
iÎ '.
0 4 ,/
~~
~.
0 2
"'
0
0 0.5 1 1-S 2 2.5 3 3.5 4
normalized spacing s
Fig. 6. Level spacing statistics of ID spinless fermions with g4 interactions and with a momentum cut-off. The parameters are chosen such that ~F
= and L
= 2x. The value of the momentum cut-off A is 12 and the total momentum is 24. The Hilbert space contains 1185 states. The statistics
is found to be intermediate between
a G-O-E- statistics and
a Poisson statistics.
To
simplify
expression(32),
one makes theassumption
of a flat Fermi surface in eachsphere
of radius A. In order tospecify
thiscondition,
oneimposes
that no bosonic excitations withan
angle
>7r/2
exist, where denotes theangle
between kf and q. The maximalangle
is such thattan
Ù 1)
§), (34)
which leads to
A 1
l~~~
)~/~ 135)Equation (35)
means that the curvature isnegligible
in asphere
of radiusA,
and iscompatible
with the condition
A » 27r
IL (36)
provided
the hnearlength
islarge enough:
L »7r/kf.
If A < 27rIL
and (q( «A,
theleading
term m
(32)
is of the form[pq,~,p(,
~,] =ô~,~,ôq,q,Va(q.n«).
We evaluate the corrections under the conditions A »27r/L anl
(q( < A. Under these assumptions, theleading
termrepresents the number of states m the
parallelogram
ofFigure
8. To obtain the corrections,one has to subtract the number of states contained in the small shaded
triangle.
The numberof states to be removed is
approximately equal
toÎÎ~
~~~~ ~°~ ~ ~~~ ~~~~'~~~'~~~kÎ~
~~~~
1. i
1
Poisson
0.9
0.8
o.7
0.6 f
~ 0.5
0 4 "._
0 3 ,,___,~~".._
0 2
"",,
o.1
0
0 0.5 1 1.5 2 2.5 3 3.5 4
normalized spacing s
Fig. 7. Level spacing statistics of ID spinless fermions with g4 interactions and with a momentum cut-off. The parameters are chosen such that ~F
= 1 and L
= 2x. The value of the momentum cut-off A is 12 and the total momentum is 24. The Hilbert space contains 1185 states. The statistics is plotted
for the inverse temperature equal to fl
= 0 and fl
= 0.3. The statistics for fl
= 0.3 is close to the Poisson statistics.
The commutation relations are thus of trie form
jpq,~, pj,,~,j
=ô~,a>ôq,q,Vajq.na)jl
+Oj'~ ~~')j. j38)
Followmg
reference(21j,
we defineaqlkf)
"
~ 4Allk kil)Inqlk)Ùlq.Vk)
+n-qlk)Ùl-q.Vk)1>
139)and
bq(kf)
=
(NAV(q.vkl)~~~~aq(kf), (40)
where ç§A is a
smearing
function such that #A - ôk,k~ if A - 0, and NA is trie localdensity
of states:NA "
) L i4Aiik kfi)~ôilL £k).
141)
Provided trie curvature of trie Fermi surface is
negligible,
that isprovided
condition(35)
issatisfied,
one can bosonize trie Harniltonian(2)
to obtainH =
~j £ (q.vk16( (kf)bq(kf)
+£ fk
k,
qn-q(k)nq(k'). (42)
~ ~
2V
~ ~,
' '
f q,q.vk>
, ,q
sphere
nsphere
n+IS
to be removed Fernù
surface
Fig. 8. Representation of a sphere ai trie Fermi surface. Trie commutation relation
(32)
is pro-portional
to the number of states contained in trie intersection of theparallelogram
and trie sphere n.Trie leading order term takes into account ail the states in the parallelogram.
Since trie interactions are
quadratic
in trie boson operators, one condiagonalize
them via ageneralized Bogoliubov
transformation. However, in numericalcomputations,
one cononly
treat a Hilbert space of size 2000. This means that trie cut-off A must be reduced
drastically,
aswell as trie Fermi wave vector
kf.
Inparticular,
conditions(35)
and(36)
are nolonger
valid andone cannot
diagonalize
trie Hamiltonian as described above. Two scenarios are candidates for trie appearance of chaos m trie spectrum of trie bidimensional Fermiliquid.
Trie first scenario is trie effect of curvature. In trie limit L - +co, and for a cut-offindependent
ofL,
condition(35)
is nolonger valid,
and trie model cannot be solvedby
bosonization. We shall notstudy
this effect in trie present paper. Trie second scenario comes from trie fact
that,
even with a flat Fermisurface,
condition(36)
may beviolated,
so that trie system is nolonger integrable by
bosonization. We shall
study
trie latter type of elfect in trie rest of trie paper.3.2. LEVEL SPACING STATISTICS OF A 2D SPINLESS FERMION SYSTEM IN THE À-RESTRIC-
TED HILBERT SPACE. We now tum to trie bidimensional case in trie presence of a momentum cut-off. Trie Fermi sea in its fundamental state is
pictured
in trie inset ofFigure
9. We treateda Fermi sea of 5 electrons for a total of 29 available quantum states. Trie Hamiltonian is
given by expression (2).
Trie interaction term can besplit
into two terms as followsH~
=
HI
+H) (43)
HI
=) ~j fk,k+q,qnk+qnk
144)k,q
HI
=
~~j ~j
fk,k,,qc(~~c(,_~ck,ck (45)
~
q#o k,k,#k+q
The term
HI
is of trie sonne nature as triediagonal
Landau interaction betweenquasiparticles.
Trie