HAL Id: jpa-00247071
https://hal.archives-ouvertes.fr/jpa-00247071
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.
Density of States in the t’-Hubbard Model: Exact Diagonalisation Results
W. Fettes, I. Morgenstern, T. Hußlein, H.-G. Matuttis, J. Singer, C. Baur
To cite this version:
W. Fettes, I. Morgenstern, T. Hußlein, H.-G. Matuttis, J. Singer, et al.. Density of States in the
t’-Hubbard Model: Exact Diagonalisation Results. Journal de Physique I, EDP Sciences, 1995, 5 (4),
pp.455-464. �10.1051/jp1:1995139�. �jpa-00247071�
J. Phys. I Hauce 5
(1995)
455-464 APRIL 1995, PAGE 455Classification Physics Abstracts 74.20
Density of States in trie t'-Hubbard Model: Exact Diagonalisation
Results
W.
Fettes,
I.Morgenstern,
T.Hufllein,
H.-G.Matuttis,
J. M.Singer
and C. BaurUniversitàt Regensburg, Fakultàt Physik, 93040 Regensburg, Germany
(Received
18 November 1994, received in final form 5làecember
1994, accepted 14 December1994)
Abstract. We calculated the density of states in a Hubbard mortel with nearest and next
nearest neighbour hopping. Using an exact diagonalisation technique, we examined this mortel
in a wide area of parameters. The density of states was calculated for dilferent fillings and for
positive and negative interaction U. The additional hopping to the next nearest neighbour sites has dilferent elfects on the gap between the particle and hole spectrum compared to the pure
Hubbard mortel.
1. Introduction
There are several
theories,
which areactually favoured,
toexplain
thehigh-temperature
su-perconductors [1-4].
One of them is the van Hove scenario [4]. It states that the maximum of the critical temperature Tc is, where the Fermi energy and a van Hovesingularity
in thedensity
of states are at the sonneplace.
From theexperiments
we know [Si that thedoping
with maximum Tc is trot athalf-filling
for thehigh
T~superconductors.
On the other hard, thesingle-baud
Hubbard model hasonly
asingularity
in thedensity
of states athalf-filling
[6].Because of this reasons the
single-baud
Hubbard model should be extended. Onepossibility
is to allow, in addition to thehopping
to the nearestneighbours,
ahopping
to the next nearestneighbours t').
This t'-model is describedby
the Harniltoman:c~~cj~ '
~
<l,j> ~
~ ~
~ C$C
~
+ u
£
~~
«~'J»1«
~
~
'~~i'~~" ~i~
where t is the
hopping
parameter to the nearestneighbours (in
thefollowing
wealways
set t =1),
t' is thehopping
parameter to_the next nearestneighbours,
<1,j
> denotes the sumover ail nearest
neighbours
and < 1,j
» is the sum over the next nearestneighbours.
Il (s the interaction between thepartiales
andc$
creates apartiale
at the lattice site with spin aand nw
=
c$cw
is thepartide
number operator. In this model a van Hovesingularity
exists at t'= 1
(ni,
if the interaction is zero, U= 0 [7].
© Les Editions de Physique 1995
We use an exact
diagonalisation technique
to salve the model(1)
see Section 2).
Because of theexponential growth
of basis states with trie system size, it is trotpossible
to deal withlarge
system sizes. This means that thedensity
of states has individualpeaks
and is trot a continuousspectrum. SO the
density
of states cannot showsingularities,
which we wouldperhaps
obtainin infinite system sizes. Nevertheless we cari
study
the eoEects of theadditionally
t'hopping
to thedensity
of statescompared
to the pure Hubbard model.2. Calculation of trie
Spectral Density
and trieDensity
of StatesAs numerical
technique
we use the exactdiagonalisation
with the Lanczos method [8, 9]. One of theadvantages
of the Lanczos method is that one carieasily
calculate the Green functionswith this method
[loi.
Within the Lanczos scheme it is trot necessary to calculate ail excited states of the model. Tue Green function to tue state (flft) cari be evaluated
by
tue continued fractionexpansion [loi
(ÎPii(i H)~~
ilYi) =~~ ,
(2)
~ °~
~-a2-fi
wuere o~ and fl~ are tue
diagonal
andoOE-diagonal
elements of tuetridiagonal
matrix Tm from tue Lanczosalgoritum.
Tue initial vector of tue Lanczosalgoritum
is tue state(flft).
Tue Green function
G(+)(k,
uJ) of tue statec[~J(flfo)
witu tue norm 1 is defined as:G(+) (k,
uJ)= lim (flfo(ck,a(uJ + iô
H)~~c(~
flfo)(3)
à-o ,
G(~)(k,
uJ)= lim
(flfolcla
(L~ + iôHi ~~ck,«
Îflfo)(4)
wuere (flfo) is tue norm
groundstate
and à is a small real number.The
spectral density A(+)(k,uJ)
is connected with the Green functionG(+)(k,uJ)
of the same k-vectorthrough
the equation:A(+)(k,~J)
=-n(G(+)(k,o~)) (à)
Here U is the
imaginary
part of the Green function. Thespectral density A(+)
is the partide part of the spectrum andA(~) gives
the hole contribution to the spectrum.The
partide /
hole(+ /-
part of thedensity
of statesN(+)
is thesum over ail allowed momenta k of the
spectral
densitiesA(+)(k,
uJ):Nl+)(uJ)
=£A(+)(k,uJ) (6)
k
and tue
complete density
of state is tue sum of thepartide
N~ and noie N+ part of trie spectrum.N(uJ)
=
N~+1(uJ)
+N~~~(uJ) (7)
In trie simulation it is of course not
possible
to determme trie limits à - 0, therefore we takea small value of à
= 0.05 or 0.10.
N°4 DENSITY OF STATES IN THE t'-HUBBARD MODEL 457
ij
i j.
Il=6.0 !',
j', j',
ii
_
U=4,0 'j jj
~
-
, J i i
z m
i
Î
U=3.0 iià j , i
~
, ,i
~
U=2.0
"
o~
j'
j
o
~i
u=i.o »
/
'
U=0.0
',
J
-4 0 4 8
Energy
wFig. 1. Density of states N+
(w)
of the square 8 point system with the filling(n)
= 1.00, t'
= +0.22 and the imaginary part à = 0.10 for several interactions U. Partiale spectrum N~: solid fine, hole
spectrum N+: dashed fine, All curves, except for U = 0, are shifted upwards for clarity.
3. Results
We have studied the t'-Hubbard model over a wide range of parameters. As the van Hove
theory
suggests the value oft',
to be neon thedoping
[4], we takealways positive
values for t' andfillings
less orequal
halffilling.
In the square8-point
system(see
[11] we studied thethree dioEerent
fillings (ni
= 1.00,
(ni
= 0.75 and(ni
= 0.50, which
belong
to the number ofpartides
ni = ni = 4, 3 and 2. The third parameter that wechanged
is the interaction U.We calculated the
density
of states for values between U= -10 and U
= 20. But
only
in therange U
= -6 to 6 we see
changes
in the evolution of the gap between thepartide
and hole spectrum.In
Figure
1 the evolution of thedensity
of states is shown for t'= +0.22 for positive values of U = 0,1,.
.,
6. The spectra for U > 0 are shifted up for
darity.
The spectrum for U= 0 is
the spectrum that is
expected by solving
thenon-interacting
modelexactly:
e(k~, ky)
=
-2t(cos(k~)
+cos(ky))
+4t'cos(k~) cos(ky) (8)
and
N(uJ)
=
/
dk~/ dkye(k~, ky) (9)
no
The delta
peaks
of the energy-axis are broadenedby
the finiteimaginary
part à.For the interaction U > 0 the gap between the
partide
solid fine and holedash-pointed
fine spectrum does not open
immediately
with interaction U#
0 in contrast to thesingle-
baud Hubbard model.
Spectra
of the pure Hubbard model(t'
= 0 cari be found in[iii.
t'= +.38 .
~ ii
t'= +.30
,
~ ii
/ ,
1 [ l
9 f. 1
Ql ' '
a a
a~
'
a , Î
c '
Q~ '
' ' o
t'"
' l'
j '~
j .
~'
;
[ ',
Fig. 2. - nsity of tates
U
=Partiale solid
fine, hole pectrumN+:ashedfine, All except for t' = 0, are
hifted for clarity.
There is obviously up to U = 2 no or only
a very small ap the partiale
hole N(+) pectrum. Only
constant
U
and arious alues off' seeFigure 2
with
U = 2.0), one sees thatfor
t/
=
othere
is a gap of about1-fi, hich reduces for creasing t'
(
t' = +o.lo and t'The gap anishes almost completely for still larger
t'
( t' =+o.22
to
t' = +o.38 ).otting
the aps for and
ioEerent values of t' (see Fig.
3)
wesee
that the gap for
the
simple Hubbard
model
( idine) at
fixed U is he
Increasing t' it ecomes smaller ( dashed fines ). Onthe otherhand
if
thevalue
ofgap for small interactions
U
is
about zero dincreases at ertain value Ug(t')The
alue of Ug(t') mcreases th the value of t'. For large interactions ( U > 6 ) hereis
a nearly linear dependence between U and t' with about thesame lope for ail
values of
We can also
see
this transition in the pectral ensity.Therefore
let
us wonsider igure4.
Herewe
see thespectral nsity Al+)(k,uJ) for the
momentum
k
=(o,7r) at ioEerent
interactions U. At small nteractions ( U <
2 )
we seesome huge peaks
in
~(k,uJ) and only
relatively small
peaks in the
holespectrum A+(k,uJ). or an
interaction
larger than U = 3 ietuation ere th trie partiale A~
and trie
noiespectrum
A+
bave
an quivalentpeak.
A similar hange in the weightsof
thespectral density
sonne
interaction Ug, where the gap opensfaster
( compare Figs.l
and 4 ). This change inweights of
the
spectral ensity isseen
for ailallowed
omenta k.For the
k-vectors,which lie
at the teraction U = o completely in e noie spectrum (
for
mple k
=
7r,7r) ),N°4 DENSITY OF STATES IN THE t'-HUBBARD MODEL 459
8
~ ...U...£...
~~$
~ ""'
73 ,,'
~#
_',,'
h
~ '.~ÎÎ[
~ ,':~ ~_,
3
U , ,'
~
"
Ç~ ~
Éd :
~ ,';
,"
Î "'1"""""
0 2 4 6 8 10
Interaction U
Fig. 3. Gap between the partiale N~ and the hale N+ spectrum for the
square 8 point system
with the filling
(ni
= 1.00 for dilferent values of t': t'
= o-o
(-+),
t'= +0.10
(-+),
t'= +0.15
(-
-n)~ t' = +0.22(-
-x ), t' = +0.30(- -6),
t'= +0.38
(- -%).
ii
i
, ,
'p
1(4 U=4.0
,'
'~
j
~ j~
= Î
~
~ 'u md
u
~ ,u
~
~'
u=o.o
;
3
Fig.
nteractions U the imaginary
part à
= 0,10, trie extnearest neighbour hopping t' =+0.22 and
the -vector k = (0, 7r). solid fine, hole spectrum: dashed fine. Ail
curves, except
for U =
0, arehifted upwards for clarity.
3
,
-.-"
Z
2,5_,.JK"'
+ 2
...1...
__.--
~ --"~"
~ j -'
8
j
--j
""
__...;._...
0.5
...Î...
o 2 3 4 5 6
Interaction U
Fig. 5. Gap between the partiale N~ and the hole N+ spectrum for the
square 8 point system
with the filling
in)
= 0.75 for dilferent values of t': t'
= o-o
(-+),
t'= +0.15
(- -+),
t'= +0.22
(-
-x ), t'= +0.25 (-
-D),
t'= +0.30
(-
-6), t'= +0.50 (- -%).the distribution moves to the partiale spectrum.
A
possible explanation
for this behaviour isthat,
in the pure Hubbardmodel,
t'= o, the
Fermi surface is square and unstable
against
the metal insulator transition which occurs assoon as the Hubbard interaction U is turned on. As t' is
increased,
t' > o, the Fermi surface in the free system, U = o, is curved [4] and at small interaction U nolonger
suitable for trie metalinsulator transition.
Only
at alarger
value of U,depending
ont',
the interaction is strongenough
to mix the states and the metal insulator transition occurs.To
summarize,
a metal insulator transition occurs in the t'-model at halffilling,
but de-layed
with respect to U. This transition at halffilling
is characteristic for modelsdescribing high-temperature superconductors. Therefore,
tuis is anotuer indication that trie t'-mortel and tueunderlaying
van Hove scenario describes tuerigut puysics
of tueuigh-temperature
superconductors.
If we return to
Figure
2, we see for the pure Hubbardmodel,
t'= 0,
only
one centralpeak
in the
partiale
spectrum at uJ m -o.8 ). For t' > o thesepeaks spht
into two parts thepeaks
at uJ = -o.8 and uJ m -1.2 for t' m +o.lo
).
The distance of thesesplitted peaks
increases withincreasing
t'. These twopeaks belong
to two dioEerent momenta k. The lower one results from the spectral densities of k =(o,
7r) and k = (7r,o)
and the upper onebelongs
to the momenta k=
(7r/2, 7r/2)
anddegenerated
momenta.Now we reduce the
filling
and agamplot
the gap between thepartiale
and the hole spectrum.Here we must
distinguish
between tuefilling (ni
= o.75 and
(ni
= 0.50
(Figs.
5 and6).
In tue first case tuere is a gap between tuepartiale
and uole spectrum in tue free system, because tue Fermi energy lies between tue two contributions at uJ = -4t' and uJ = 0 in thedensity
ofstates. One cari see this gap of the free system U
= o m the lowest curve in
Figure
1 gap between thepeaks by
uJ = -o.88 and uJ= 0
).
In the second case the Fermi energy lies at uJ =4t',
and not in a finite size gap of the free system. In both cases tue gap for 0 < t' < 0.45 islarger
than for tue pure Hubbard model. At tuefilling (ni
= 0.75 this increase of the gap witu t' is
mainly
causedby
theincreasing
limite size gap of tue free systemFig.
5).
Also in tue case of tue
filling (ni
= 0.50
Fig.
GI tue gap mcreases wituincreasing
t'. ButN°4 DENSITY OF STATES IN THE t'-HUBBARD MODEL 461
0.6
~
0.5md
fi
o4 ...j...+
Î _~§---.-,,j---j---
0.3
',
""( 8
0.2...i.... '
JJ ,'
~
o j ..l... ..l....__ .Îj_.___.I__..__..._____.]_,O
0
0 2 4 6 8 10
Interaction U
Fig. 6. Gap between trie partiale N~ and the noie N+ spectrum for the
square 8 point system with the filling
(ni
= 0.50 for dilferent values of t': t'
= 0.0
(-+),
t'= +0.15
(- -+),
t'= +0.22
(- -n),
t'= +0.30
(-
-x ), t' = +0.38(- -6),
t'= +0.45
(- -%),
t'= +0.50
(- -o),
in contrast to tue
filling (n)
= 0.75 tuere is no limite size gap at tue Fermi level. Moreover tue gap tends to a constant value for
uuge
interactions U e-g- at t'= +0.22 tue gap tends to
0.36,
dasued fine witu boxes inFig. 61.
Tue situationchanges dramatically
for t' > +0.45 andlarge
interactions U > 4. At t'= +0.45 tue gap bas its maximum at U
= 4 and decreases for
larger
U. If we increase t' to t'=
+0.50,
tuen we see a sudden breakdown of tue gap between U = 5.0 and U = 5.3. For U > 5.3 there is nolonger
a gap between tuepartiale
N~ and halespectrum N+ of the
density
of states. The reasons for this sudden breakdown of the gap are not yet clear.There are some
possible explanations:
the breakdown may be a result of thedegeneration
of the k-vectors k
=
(0,7r),
k= (7r,
o)
and k= (7r,7r) for t' = +0.50 or of the special system
size. Another
possibility
is, that the accumulation of dioEerent k-vectors at the Fermi energyas in the van Hove scenario causes this breakdown.
Dagotto
et ai. found a similar breakdown for asuperconducting
order parameter in the t J model with the samefilling
[12]. Theirexplanatioii
is aphase
transition.In addition to the
positive
U case, we have aise calculated thedensity
of states and the gap between thepartiale
and the hole spectrum of thedensity
of states fornegative
interactions U.Trie
resulting
gap for severalfillings
areplotted
inFigure
7. Here we see the gaps for the threepossible fillings, in)
=1.oo 7(a) ), (ni
= o.75 7(b)
and(ni
= o.50 7
(c) ).
In trie casesof
(ni
= 1.oo and(ni
= o.50 tue gap decreases witu increasing t'. In tue case of
(ni
= 0.75
tue situation is a little more
complicated.
As above, tuere is a finite size gap between tuepartiale
and tue uole spectrum in tue free system, U =0,
witu t'#
0. Tuerefore we see mFigure
7(b)
tuat tue gap increases wituincreasing
t' for small interactions 0 > U > -1. But atlarge
interaction U > 4 tue gap shows trie same behaviour at triefillings in)
= 1.00
(a)
andin)
= o.50(c).
So foruuge
interactions tuere is a decrease of tue gap wituincreasing
t' for ailfillings.
Similar to tuepositive
U case and ualffilling
tue increase of tue gap is almost linear witu tue interaction U forlarge
interactions U > 4, for ailfillings
and for ail values of t'.Given
Orly
tuedensity
of statesNl+)
(uJ) and tuespectral density A(+) (k,
uJ), it isimpossible
~
---j---j---j---j
+
3
,
~~T~~~~~~~~~~
Z
25, , 1--- ---J--
~
,,j
~) g
2 -~--~ fi- --~---~--'
1.5
-~---~-b~jt~--
j~
Ô,~ l -r---~---t-
é
°.5~Î ~~~Î~~~Î~~~
,0
~ -5 -4 -3 -2 -1 0
Interaction U
Î
~~,
~~~T~~~T~~~~~~~~~~~
fi
35 L---L---J---~---' '
~~
3,
--t---~---~
b) (
~'( ~( "~ ~~j~~~
j
1.5~Î~~~Î~~~
,
~~~~
j -~---L---L--
Î
0.5---1---1---1---1---~
0
-6 -5 -4 -3 -2 -1 0
Interaction U
,
5
2~
]
4.5 --p---+---+---+---~-
~ 35 L---1---J
z
,~
3
,
-t---~---~-
~) g
2,5 --~- ---J---~-~
2~~Î~~~~ ~~~~~Î~
fl
1.5 --~---~---+ ~-it
l~~Î~~~Î~~~Î~~
o 0.5 --~---~---t--
0
-6 -5 -4 -3 -2 -1 0
Interaction U
Fig. 7. Gap between the
partiale
N~ and the hole N+ spectrum for tue square 8 point system for negative interaction U and dilferent values of t': t'= o-o
(-o),
t'= +0.15
(-
-+), t' = +0.30(- -D).
Sllbplot
a)
filling(n)
=1.00,b) in)
= 0.75 and c)
in)
= 0.50.
N°4 DENSITY OF STATES IN THE t'-HUBBARD MODEL 463
to
clarify
wuat is tueorigin
of tue gap. Tuerefore we must calculate additional observables like correlation functions.Projector
quantum Monte Carlo results [13] indicate that the t'-model in thenegative
Uregime
issuperconducting
away from halffilling.
4. Conclusions
Within the
positive
U range for halffilling
t' retards theopening
of the gap with respect to U. This retardation increases withincreasing
t'. So the metal insulator transition seems to bedelayed through
a finite value of t'. This transition supports the t'-model and the van Hove scenario aspossible explanations
for thehigh-temperature superconductors.
At the
filling (ni
= 0.75 themcreasing
value of the gap withincreasing
t' is dommatedby
the finite size gap in the free system, U = o. For that reason the gap increases with
increasing
t'. And at the nextpossible filling, (ni
=o.50,
there is ahuge magnification
withincreasing t',
until t' reaches the value t'= +o.45. Here we see a maximum of the gap at U = 4.
Finally
at t'
= +o.50 there is a sudden breakdown of the gap at the interaction U m 5. Tue
origm
of tuis beuaviour is not yet clear. Additional calculations are in progress.In the case of
negative
interactions U tuere are nolarge changes
m trie evolution of the gap for variousfillings
observed. Here the additionalhopping
t' decreases the gap withincreasing
t'. Theonly exception
is the case of thefilling (ni
= 0.75 with small interactions (U( < 1, therethe finite size gap for t'
#
0 dominates.In order to decide on the
origin
of this gap m thenegative
Uregime,
it is necessary to calculate other observables like correlation functions. But results obtained with theprojector
quantum Monte Carlo [13] indicate that thenegative
U t'-model issuperconducting
away from halffilling.
Acknowledgments
The authors would like to thank D. M. Newns for
helpful
discussions.References
iii
Monthoux P, and Pines D,, Phys. Rm. Lett. 69(1992)
961[2] Scalapino D.J., Loh E. and Hirsch J-E-, Phys. Rm. 835
(1987)
6694 [3] Markiewicz R-S-, Physica C217(1993)
381.[4] Newns D.M., Comments Gond. Mat. Phys. 15
(1992)
273.[Si Torrance J-B-, Tokura Y., Nazzal A.I., Bezinge A., Huang T.C. and Parkin S-S-P-, Phys. Rm.
Lett. 61
(1988)
1127.[6] Hirsch J-E-, Phys. Rm. B31
(1985)
4403.[7] Morgenstern I., Singer J-M-, Hufllein Th. and Matuttis H.G.,"Numerical Simulation of High Temperature Superconductors", Proceedings of the 20th Int. School of Cryst., Erice, 1994, E.
Kaldis, Ed.
(Klower).
[8] Golub G-H- and Van Loan C.F., Matrix Computations
(The
Johns Hopkins University Press, Baltimore and London, 2nd edition,1989).
[9] Cullum J. and Willoughby R-A-, Lanczos Algorithms for Large Symmetric Eigenvalue Computa-
tions (Birkhàuser, Boston, 1985) vol. I-II.
[10] Fulde P., Electron Correlations in Molecules and Solids (Springer Verlag, Heidelberg,
1991).
[11] Dagotto E., Joynt R., Moreo A., Bacci S. and Gaghano E., Phys. Rm. 841
(1990)
9049.[12] Dagotto E. and Riera J. Phys. Re~. Lett. 70
(1993)
682.[13] Singer J. M, et ai., unpublished.