HAL Id: jpa-00208544
https://hal.archives-ouvertes.fr/jpa-00208544
Submitted on 1 Jan 1976
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, estdestinée 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.
Discussion on the onset of antiferromagnetism in the hubbard model
F. Brouers, F. Ducastelle, J. Giner
To cite this version:
F. Brouers, F. Ducastelle, J. Giner. Discussion on the onset of antiferromagnetism in the hubbard model. Journal de Physique, 1976, 37 (12), pp.1427-1436. �10.1051/jphys:0197600370120142700�.
�jpa-00208544�
1427
DISCUSSION ON THE ONSET OF ANTIFERROMAGNETISM
IN
THEHUBBARD MODEL
F. BROUERS
Laboratoire de
Physique
desSolides,
UniversitéParis-Sud,
91405Orsay,
FranceF. DUCASTELLE
Office National d’Etudes et de Recherches
Aérospatiales,
92320Chatillon,
France and J. GINERInstitut de
Physique,
Université deLiège,
SartTilman,
4000Liège, Belgium (Reçu
le 31 mai1976, accepté
le30 juillet 1976)
Résumé. 2014 Nous calculons la réponse à un champ magnétique périodique de la phase désordonnée du modèle de Hubbard dans l’approximation
d’analogie d’alliage.
Nous étudions le comportement de la susceptibilité et en particulier nous recherchons une éventuelle divergence de la susceptibilitépour un vecteur d’onde
correspondant
à un ordreantiferromagnétique.
Nous ne trouvons aucuneinstabilité pour une bande à moitié
pleine
ou presqu’à moitié pleine.Abstract. 2014 We calculate the response to a periodical external magnetic field of the magnetically
disordered phase of the Hubbard model in the alloy
analogy
approximation. We investigate thebehaviour of the
q-dependent
susceptibility and in particular we look for apossible divergence
of thesusceptibility
for a wave-vectorcorresponding
to anantiferromagnetic
ordering. We do not find anyinstability for a half-filled or nearly half-filled band.
LE JOURNAL DE PHYSIQUE TOME 37, DTCEMBRE 1976,
Classification Physics Abstracts
8.516
1.
Introduction.
-Recently
several papers[1-6]
have been devoted to the discussion of the
possibility
of
ferromagnetic
andantiferromagnetic
solutionsand their
stability
with respect to theparamagnetic
solutions of the Hubbard model used in the discussion of the
magnetic
andconductivity properties
of narrowband materials.
It is now well established that the Hubbard III
approximation [7]
does notprovide
a stable ferro-magnetic
solution. As was shown inVelicky et
al.[8],
the Hubbard III solution which includes
only
thescattering
corrections is identical to the coherentpotential approximation (CPA)
of thetheory
ofalloys.
Fukuyama
and Ehrenreich[1]
have shown that the staticspin susceptibility
of the Hubbard model calculated in the CPA remains finite for anysimple
band model and carrier number
except possibly
in the case of a
precisely
half filledsplit
band. Brouersand Ducastelle
[2]
have shown that this statement is moregeneral
and should betrue if , the
Hubbardmodel could be solved
exactly within
thealloy
ana-logy approximation.
This statement is therefore true for any extension of theCPA,
for instance the mole- cular CPA.In
particular
since Weller and Gobsch[9]
haveshown that for a
half-filled
band and in the parama-gnetic phase
the full Hubbard III solutionincluding
resonant correction can be written as a solution of the CPA with a continuous distribution function for the stochastic
variables,
the conclusions of ref.[2]
are also valid in that case.
Schneider and Drchal
[4]
have shownby using
energy considerations that in the strong correlation limit the Hubbard III
scattering
correctionapproxi-
mation
(CPA) yields
an unstable satured ferroma-gnetic
state fornearly
half-filled band. For n 1only
theparamagnetic
solutionexists,
in agreement with theprevious
results.Although
it is not known if the exact solution of the Hubbard Hamiltonianyields
any stable ferro-magnetic
solution for any electron bandmodel,
the results of the
alloy analogy approximation
ofthat model are in agreement with Van Vleck’s
[10]
Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:0197600370120142700
statement that itinerant
ferromagnetism
is due essen-tially
to the intra-atomicexchange
interaction between electrons ofdegenerate
d-orbitals. Brouers and Ducas- telle[2]
have considered a model with orbitaldege-
neracy and
analyzing
the behaviour of the staticspin susceptibility
have shown that in that case an insta-bility
of thesusceptibility
and therefore a ferroma-gnetic
solution waspossible.
These conclusions confirm and extendprevious
calculations of Schneideret al.
[11]
who reached similar conclusions in the Hubbard Iapproximation [12]
and in the strong correlation limit.The
possibility
of anantiferromagnetic
solutionof the
single-band
Hubbard model remains also acontroversial
question although
a certain number of results have been obtained and discussedrecently.
Mehrotra and Viswanathan
[5]
have shown that the Hubbarddecoupling
scheme is not able toyield
a stable
antiferromagnetic
state.By
contrastGupta
et al.
[6]
have obtained anantiferromagnetic
solutionby calculating
the order parameterself-consistently
within the two sublattice CPA
(Brouers et
al.[13])
of the Hubbard
alloy analogy approximation.
Meyer
and Schweitzer[14]
have introduced aspin dependent
band shiftfollowing
Roth’s formalism[15]
and found two
antiferromagnetic
solutions in a smallregion
around the half-filled band for s.c. and b.c.c.structures but the solutions are unstable toward
paramagnetism
andferromagnetism.
In this paper, we report the result of a calculation of the response of the
magnetically
disorderedphase
of the
alloy analogy
model to aperiodical
externalmagnetic
field. Weinvestigate
theproperties
of theq-dependent susceptibility
and inparticular
we lookfor a
possible divergence
of thesusceptibility
for awave-vector
corresponding
to anantiferromagnetic ordering.
We do not find anyinstability
for a half-filled
single
band.2. Model. - The model
employed
toinvestigate
the itinerant
magnetism
in narrow bandbinary alloys
is
represented by
the Hubbard Hamiltonianwritten in second
quantization using
a Wannier basis.The operator
a’
and ana are the creation and annihi- lation operators for an electron withspin
Q at thesite n and
fin, = a ’
a,,,. As was firstpointed
outby
Hubbard
[7],
the motion of electrons withspin s
= ± Qcan be described
approximately by
thefollowing alloy
Hamiltonianwhere t;na is
equal
to e’ = U or c,’0, depending
on whether an electron with
spin a
is at site nor not. At a fixed instant in time the electrons with
opposite spins
areregarded
asoccupying given positions
onrandomly
distributed lattice sites. Oneneglects
thedynamical
effects ofopposite spin
elec-trons and one
replaces
the time average which arises because of the continuous rearrangement ofspins by
aconfiguration
average in thealloy problem.
Hamiltonian
(2)
can therefore be viewed asdescribing
two
interdependent binary alloys A1-n-a Bn-a’
Herena is the average number of electrons per site with
spin a :
nna).
The relative concentration of atoms with energy levels9
is 1 - n-a or 1 - n,, and with energy level t!3 is n-a or n,,corresponding
to whetherthe electrons have
spin
Q or - Qrespectively.
It isassumed for the purpose of
calculating
the motion of Jspin
electrons that electrons ofspin -
Q arefixed at the lattice sites and vice-versa.
For a
given configuration,
one can define aGreen’s
function
and the average number of electrons with
spin
+ ais
given by
where N is the number of sites. The average Green’s function
being
a function of the concentration n-a isgiven by (at
T = 0K) :
The Fermi level is defined
by
a thirdequation
where n is the average number of electrons per site.
3.
Spin susceptibility.
- Theparamagnetic
staticspin susceptibility corresponding
to Hamiltonian(2)
has been discussed in ref.
[2].
Here we consider theq-dependent susceptibility.
Aperiodic
external field induces in theparamagnetic phase periodic
variationsin the number of up and down
spin
electrons. In thealloy analogy
model thiscorresponds
to concentrationvariations.
If we start from the disordered parama-gnetic phase,
we can calculate thespin susceptibility corresponding
toperiodic spin dependent
concen-tration nn -u in the
alloy analogy approximation.
This
corresponds
to concentration variations1
1429
induced
by
an infinitesimal external fieldThe
q-dependent spin susceptibility
readswhere J-lB is the Bohr magneton and
This
gives
forX(q) :
In the
alloy analogy
modeln-,(q) plays
the role of aconcentration. One has
6G,(q)
is the Fourier transform of6 ( n ) I Gin>
andsince
in the
alloy analogy approximation
where Z is theself-energy
operator which is sitediagonal
in the CPAOne has therefore
If we define the Fourier transform
The
q-dependent spin
enhancedsusceptibility
isdefined
by
X’(q)
is the unenhancedsusceptibility
andwhere
This allows a calculation of a
possible instability
of the
paramagnetic susceptibility
for agiven
q.To calculate
K(q),
weexpand
the self-consistentequation
as a function of the concentration fluctua- tionsbc,,
= cn -co
on each site n,c’ being
theconcentration in the
completely
disordered para-magnetic
state.To achieve that programme, we use an extension of the CPA to
inhomogeneous
systems discussed inappendix
1 and we vary the self-consistent equa-tions tn >
= 0 with respect to the concentration fluctuationsðCn (appendix 2).
One obtains for
bc(q)
theexpression
in terms ofcomplete
disorderquantities
where
and
This
gives
4. Calculation of
K(q).
- Thisquantity
has beencalculated for two models :
a)
Thesimplest
one, a. one-electron band with nearestneighbour
interactions on a BCC(similar
to ref.
[6]) CICs ;
b)
The modeldensity
of statessimulating
the BCCstructure
already
introduced for the calculation ofordering
energy[16, 17].
In that case, we calculateK(q)
for the wave vectorBy introducing
a secondneighbour
interaction one can avoidpathological
difficulties related to theoccurrence of a gap in model
a).
The first case is
trivial,
one needsA( q, E)
whichis obtained from its definition
(16)
If one divides the sublattice into two sublattices a
and fl
and if n is on sublattice a :For
with
I k,, >
is the Bloch state built on the sublattice a ; kbelongs
to the Brillouin zone of the s.c. lattice.In the first
model,
the Green’s function isgiven by
where a a(p) is the
self-energy corresponding
to sublatticerx(P)
and determined from the self-consistent equa- tion(A8)
ands(k)
is thedispersion
relationcorresponding
to the BCC structure. One getsimmediately
in thecompletely
disordered casewhere ax
= upwhere we have used the definition of the
density
of statesIf one notices that the
diagonal
Green’s functionone has
In the second
model,
the band structure is charac- terizedby
the values of the energylevels 8;
and thevalues of the
hopping integrals
between atoms of thesame
(Wma,nP)
and of different sublattices(Wma,nP).
In the BCC structure, the transfer
integrals
betweenfirst
Wap
and secondWem
nearestneighbours
are of thesame order of
magnitude
and determinemainly
the band structure of the pure metals. We
keep only
these two
integrals
and we do not take into accountthe
degeneracy
of the d-states.If we
neglect
the transferintegrals W,,fl
whichcouple
bothsublattices,
we obtain twotight binding
bands centered on the
energies 8; (i
=A, B)
whoseenergies
andeigenstates
arerespectively given by
8«(k) (or ap(k)) and I qi,,,(k) > (°r I ql,6(k) >).
The com-plete
Hamiltonian is thengiven
the basis{ t/J a’ t/J p } by
where the summation is extended over the Brillouin
zone of the cubic
crystal.
Moreover for the numericalcomputation,
we willreplace
thecoupling
functionby
an average valueindependent
of k.1431
This kind of
approximation
hasalready
been usedin the case of s-d
mixing
effects[18] ;
the correspon- dance is established if wereplace
aand fl by
s and d.In that model
A(q)
is obtained from its defini- tion(17).
We consider the wave-vectorTo calculate
equation (26)
we need
From the definition of the Green’s function
one gets
immediately
and
In the
complete
disordered case a,, = a, and we assumeUsing
the definition of thedensity
of states(29),
onegets
The function
K(q)
has been calculated in both modelsusing
forgo(E) a semi-elliptic
function fora half-filled band and for
nearly
half-filled band.We have found
numerically
thatK(q*)
is such that- 1
K(q*)
0. In the half-filledband,
one canshow
analytically (Appendix 3)
thatK(q*) --+ -
1when U --+ oo. In the same limit
x(q*)
tends to 0as
U-’.
In
figure
1 we have drawnx(q*)
in terms of U forthe first model. It
diverges
at U = 0 due to the pre-sence of a gap in the BCC one-band and decreases
slowly
with U.From our
calculations,
one can conclude that aperturbative
treatmentstarting
from the parama-gnetic
solution of the CPAalloy analogy approxi-
mation Hubbard model does not exhibit any
tendancy
towards
antiferromagnetism.
This result contredictsa recent self-consistent calculation of
Gupta et
al.[6]
who claims that an
antiferromagnetic
solution ispossible
in thenon-degenerate
Hubbard model within thealloy analogy approximation
when U islarger
than some critical value. As our calculations are not
incompatible
with apossible
first order transitionwe have redone their calculations and we do not find any
antiferromagnetic
solutioliq We discuss thispoint
in ref.[19].
FIG. l.
The results of this paper as well as the conclusions of ref.
[2]
indicateclearly
the limitations of thealloy analogy approximation.
It is a well known exact result of the half-filled band Hubbard model that in the limit of verylarge U,
itapproaches
the Heisen-berg
model. In thatregime
the interatomic electroncorrelation at
neighbouring
sitesproduces
the ten-dancy
towardsantiferromagnetism.
For that reason, it is
hardly expected
that theHubbard
alloy analogy approximation yields
ameaningful
result in thisregime
where localized moments are formed and the method of this paper should be restricted to the metallicregime.
As it isbelieved
generally
that aninstability
towards anti-ferromagnetism
will not occurprior
to the appearance of localized moments, the conclusions obtained in this paper are inagreement
with thisgeneral
expec- tation.Appendix
1 : Concentration fluctuations in disor- deredalloys.
- In the usual CPA forhomogeneous alloy
cn = c, and the energy levels areonly dependent
on the nature of the considered atom i ; the effective Hamiltonian
JCO
is then determinedby
a site inde-pendent
effectivepotential (or self-energy)
localizedon each site n
This.
self-energy
is determinedby
the condition that thedensity
of states per atom on site n and for the averagemedium g(E) >’
isequal
to the average of thepartial (or conditional)
densities of stateswhich are the densities of states
averaged
over allthe
configurations
for which the site n isoccupied by
an atom i :Z’ can also be obtained
by
the condition that the electrons of the average medium are not scatteredon the average
by
atomssitting
on n withrespective probabilities
c’(i
=A, B)
the t-matrix
elements ti
aregiven by
in
terms of the function Fn n >
whereUl 0) = (E - Jeeff) -1
is the Green’s function for theaveraged
medium andc(O) = n E (0) 1 n >.
Thisscheme can be extended to
spatially varying
concen-tration cn. The
density
of states for the averagemedium gn(E)
and the conditional densities of statesgn(E)
are now sitedependent
and(A3)
isreplaced by
similar conditionsThe effective medium is defined
by
sitedependent potentials
localized on thecorresponding
sites nThe
potentials a,,
are obtained in terms of thescattering
operatorsby
the conditionwhere the brackets mean that we
perform
the averageaccording
to(A6)
and where thescattering
operatorstn
are definedby
Finally Fn = n I Gin> in (A9)
is related tothe
Green’s function G =
(E - JCeff) 1.
The function G and therefor E andJeeff depend
on the energy E.The average
density
of states per atomg(E)
is obtained fromFn by
The conditional densities of states
gn(E)
are relatedto
Fn by
The
equivalence
between the conditions(A6)
and(A8)
can be
easily
obtained from(A 11).
The
previous
formal treatment can beapplied
tothe
study
ofpartially
ordered state. The state of anordered
crystal
is characterizedby m
different values of the concentrationcorresponding
to the m differentsublattices. The
quantities
a.and g.
are determinedby
theprevious equations
andby
the m self-consistentequations (A8).
This scheme has beenpreviously
used for models of
ordering
insimple
structureswhere m
corresponds
to aand fl
the two sublattices of an alternant lattice[13].
Appendix
2. - We calculate the variation to first order of thespatially varying
self-consistent equa- tion(A6)
due to concentration(order parameter)
variation
For a
binary alloy,
one hast(O) and c(O) refer to the
complete
disorder.1433
From
taking
account of the relationone obtains
immediately
and since
averaging
where we have
put Fn
=F,
thediagonal
Green’sfunction
corresponding
tocomplete
disorder.Introducing
the Fourier transformwe have for each wave-vector q :
where
and
G(R)
=GRO, 0 being
aparticular
site of the lattice.The Fourier transform of condition
(A 12) yields
or
i.e.
if we define
since for a
binary alloy dcn - - dct
=den
one haswith and
Appendix
3. - In thisappendix
we want to calculatethe limit
of x(q*)
for U --+ oo. To’this end we determine in this limit thequantities K(q*)
andx°(q*).
We usethe moment method discussed in the paper of
Velicky
et al.
[8]
which we first recallbriefly.
We consider a
binary alloy Ai-c B,
characterizedby
thepotentials
I;A = 0 SB = 6 in the limit 6 -+ oo(Fig. 2).
We want to calculate the
integral
of aquantity
over the states of the A-subband. We can write
FIG. 3.
e is a contour around the cut « A » on the real axis
(Fig. 3).
We assume thatqJ(z)
has no othersingularity
than the cuts on the real axis and that
qJ(z)
tends tozero faster than Z-l when z - oo. One has therefore the
spectral representation
If one defines the
density ql(E) = - 1m E + ,
one can writeIf o y
( )
7EP( )
Expanding
both sides in z - ’ andidentifying equivalent
terms one getsthe moments of the total band in terms of the moments of the subbands
in the limit 6 --+ oo the moments are obtained
by
identification. One can obtainpg
which isequal
to theintegral
This
technique
can be used to calculate theasymptotic
form ofequation (23) integrated
over subband AOne has first to
expand
theintegrand.
SinceG(R)
calculated in the CPA is exact up toz-’,
thisapproximation
will
give
the first exact terms in anexpansion in 6 - ’.
By definition,
one hasdifferentiating
since
one can write
and Fourier
transforming
For the CICs model
q* - [1, 1, 1]
and weintegrate
over the band A.One knows that
and for
q*
or
with
This
gives :
if R1
are the sites of the sublattice 0 0.1435
If we
expand
we can write
keeping only
thehigher
terms in thepotentials
ER l In the same wayWe have thus :
and the summation on
Rl
in(A40)
isgiven by
theexpansion
where
pg
is the second moment of the ordered band.The moments of
(A40)
areUsing
nowequation (A29)
to define thepf and
in terms of /In.Solving
the set ofequations
up to /l5 for ð ---+- ooyields
the resultand
As we have taken
only
terms z-4 in theexpansion
ofG,
one can conclude that in this limit CPAgives
an exact result.
We can calculate
X’(q) by
the sametechnique
for 6 -+ oo . We start fromone has
and thus
We are interested in the first terms in
ð - 1,
we takeonly R,
for firstneighbours.
To thehighest
order of g in thez - 1
expansion,
we haveAveraging G2(Ri) gives
thefollowing expression
From which one determines the moment N and from the
equations (A29)
up to the seventh moment,we obtain for b - oo and
As
we have
and since here
c = 2
References
[1] FUKUYAMA, H. and EHRENREICH, H., Phys. Rev. B 7 (1973) 3266.
[2] BROUERS, F. and DUCASTELLE, F., J. Physique 36 (1975) 851.
BROUERS, F., Phys. Stat. Sol. (b) 76 (1976) 145.
[3] NOLTING, W., Phys. Stat. Sol. (b) 68 (1975) 243.
[4] SCHNEIDER, J. and DRCHAL, V., Phys. Stat. Sol. (b) 68 (1975)
207.
[5] MEHROTRA, C. and VISWANATHAN, K. S., Phys. Rev. B 7 (1973)
559.
[6] GUPTA, A. K., EDWARDS, D. M. and HEWSON, A. C., J. Phys. C 8 (1975) 3207.
[7] HUBBARD, J., Proc. R. Soc. A 281 (1964) 401.
[8] VELICKY, B., KIRKPATRICK, S. and EHRENREICH, H., Phys.
Rev. 175 (1968) 747.
[9] WELLER, W. and GOBSCH, G., Phys. Stat. Sol. (b) 60 (1973) 783.
[10] VAN VLECK, J. H., Rev. Mod. Phys. 25 (1953) 220.
[11] SCHNEIDER, J., HEINER, E. and HAUBENREISSER, W., Phys. Stat.
Sol. (b) 53 (1972) 553.
[12] HUBBARD, J., Proc. R. Soc. A 276 (1963) 238.
[13] BROUERS, F., GINER, J. and VAN DER REST, J., J. Phys. F4 (1974)
214.
[14] MEYER, J. S. and SCHWEIZER, J. W., AIP Conf. Proc. 10 (1973)
526.
[15] ROTH, L., Phys. Rev. 184 (1969) 451.
[16] GAUTIER, F., DUCASTELLE, F. and GINER, J., Phil. Mag. 31 (1975) 1373.
[17] GINER, J., BROUERS, F., GAUTIER, F. and VAN DER REST, J., to be published.
[18] BROUERS, F. and VEDYAYEV, A. V., Phys. Rev. B 5 (1972) 348.
[19] BROUERS, F., DUCASTELLE, F., TREGLIA, G., BRAUWERS, M.
and GINER, J., J. Physique Colloq. 37 (1976) C4-205.