Equation-Of-Motion Impurity Anderson Model With Finite Coulomb Repulsion B
Equation-Of-Motion Impurity Anderson Model With Finite Coulomb Repulsion B
15 OCTOBER 1995-I
Equation-of-motion treatment of the impurity Anderson model with a finite on-site Coulomb repulsion
Department
Kicheon Kang and B. I. Min of Physics, Pohang Uniuersity of Science and Technology, Pohang 790-78$, Korea (Received 19 May 1995)
The impurity Anderson model with a finite Coulomb repulsion U has been studied by using the conventional equation-of-motion treatment for the f-electron Green's function. To get a closed One is to neglect, in a large U limit, equation, two types of approximations are incorporated. higher order Green s functions which involve three f-electron annihilation operators. The other is to linearize the equation of motion by using a projection operator method. The average valency nt and the spectral density of states (DOS) of f electrons are evaluated from the self-consistent equation for the Green's function. Results for ny are in good agreement with the exact results of the Bethe ansatz method, and the obtained spectral DOS exhibits a well-known three-peak structure. EA'ects of finite U and finite temperature on the spectral DOS are discussed.
The impurity Anderson model has been extensively studied in the past decades, in relation to the Kondo efFect, valence fIuctuations, and heavy fermion phenomband systems. ' Due to the ena observed in narrow complexity of the model Hamiltonian, an infinite U is assumed in most of theoretical works for the Coulomb electrons in the impurity site. A rerepulsion between alistic Coulomb repulsion U is, however, typically of the 6 order of 5 eV, and it has been pointed out that finite-U efFects are important in determining spectroscopic, thermodynamic, and transport properties. ' ' have studied finiteGunnarsson and Schonhammer U efFects on spectroscopies, using the zero-temperature variational approach in the large Kt limit (Ny denotes the spin-orbit degeneracy of the orbital). The generalization to finite temperature was also done by includIn ing doubly occupied configurations of electrons. these approaches, methods of the functional integral or the generalized noncrossing diagrammatic approximation were employed with the double occupancy slave bosons. On the other hand, finite-temperature properties can be easily described by using the conventional equationof-motion scheme for the single-particle Green's function. Indeed, this method was applied by several authors to the infinite-U Anderson model system, but there are not many works available for the finite-U system. Among has exploited a decoupling approximathose, Czycholl tion suitable for the finite-U impurity Anderson model in the large %y limit. The purpose of this paper is to reexamine the equationof-motion treatment for the f-electron Green's function in the finite-U impurity Anderson model. Our approach is more or less similar to other works using the equationof-motion schemes. However, compared to other approaches, the following aspects in the decoupling procedure of higher order Green's functions are difFerent: (i) to obtain the f-electron Green's function, we consider a large U limit in the sense that higher order Green's functions, which involve three f-electron annihilation operators are neglected, (ii) we incorporate the projection operator method to linearize the equation of motion and
In this way, we have obtained a self-consistent equation for the f-electron Green's function, from which the valency and the spectral density of states (DOS) of the electrons are evaluated numerically. We consider the finite-U impurity Anderson Hamiltonian in a partial wave representation:
'R
) e(k)ctcg
k, m,
+et
ft f
cg
k, m.
+)
k, m
v(k)
(ct
+ ft
) +u m'Pm n, n
,
1, 2, Nf) denote spin-orbit degeneracies. The operator c& (ck ) creates (annihilates) a conduction electron in the state (k, m, ), with an energy dispersion of e(k) = e(k), k = ~k~. The magnetic impurity with electrons has a degenerate level, cy, where the spin-orbit and multiplet splittings are neglected. The corresponding creation (annihilation) operator and the number operator are t (f ) and n = t respectively. The matrix element V(k) represents the hybridization between conduction and impurity electrons. We assume that V(k) is independent of
I, rn'(=
f,
((f; f
where z = ber. Here,
))=
i
e"'((f
b is
is the
(t),
ft
)f) dt,
(2)
an infinitesimal positive numdenote the anticommutator and statistical average of operators, respectively. The equation of motion for the f-electron Green's function yields
~ + i b, and
j, ) and
()
Egm
0163-1829/95/52(15)/10689(4)/$06. 00
10 689
1995
10 690
BRIEF REPORTS
52
with a higher order Green's function on the right-hand side. Successive equations of motion, for the Green's functions on the right-hand side of (3), give rise to
(4)
closed form of the equation of motion for the f-electron Green's function. This is the second approximation employed in our scheme. Note that linearization by using the projection operator method is known to give better results than that by using a simple decoupling method. Then we derive the following equation:
= (nl) +
V(k) (((f('&k-
f'
))
(5)
z
+ Fp(z)
S(z) =
) .V(k) V(k')
k, k'
cs, , ct, t)
(12)
This equation is self-consistent, because the statistical averages on the right-hand side, such as (nt), (ft cat) in
in
self-
((f
f t ))
p Z
where zq
= z U, z = 2cf + U z, sf
This is essentially equivalent to the result by Czycholl, oo limit. The locawhich is known to be exact for Nf tion of the pole of Eq. (13), with respect to the Fermi energy, corresponds to the Kondo temperature TK. Assuming a square DOS for conduction electrons [p(c) = 1/W for s W/2] and constant hybridization matrix W/2 elements [V(k) = V], we get the following expression for TK
( (
=
k~TIc
where
1
exp
7r.
(e'f
(14)
z, 2Fp(z) + F()(z)
+Ckr lCkm
)
Ckl
).
kk'
(C, f ,
((X;
C,Ck m
fl),
g m.
p is the Fermi energy, and A = mV2/W. The expression of TIc in Eq. (14) differs from the correct one by a factor of Nf/(Nf in the exponent. is is 1) &ng the Eq. (1.1.) for the f-electron Green's function, we calculate numerically the valency nf = (n ) and the f-electron spectral DOS. It is very hard to obtain fully self-consistent solutions of the equation. Thus, in this study, statistical averages in S(z) and B(z) are replaced by their average values with V(k) = 0 to demonstrate qualitative behaviors:
-P
Equation (6) still involves higher order Green's funcft )), and so the equation tions represented by is not of a closed form yet. To linearize Eq. (6), we introduce the projection operator P, which projects an such that arbitrary operator, 0, onto
k, k'
) . V(h)V(h')
R(z) =
operator
Counting only the projected component of higher order that is X -+ 'P X, we can get a X onto
(f,cst)
(16)
f,
These approximations
52
BR&EF REPQRTS
near p, since corrections b d I ()&=o are of higher order contributions of y' ls s 'ng Eq () mith Eq' ( 5) and (16), the . valency 7t m (+m) &s calcuiateA self-consistent]y for Nf = 2. 17 Ffigure 1(a) presents tl e results of n& a function o f & for infinite U which ~ cy with existing results. As ex u s. s expected, n deere shf up in energy N o t e t h at present r ilats b h h d h an t h ose from the lowes 2, 19 ~in the variational a j~j in the perturbational sch erne. Of course t iona sc
' "'
f
O
CL
0. 3a
') 0.,
0. 25
03
C)
0.25 0.2
0.2
0. 1'a
0. 1
~ 15 ~0.
CL
-0.00,
'&
0.1
0.05
tio
f th I/N
qui e simi ar to those of t igure 1 b s ows the resultssofny -U function of cy for 6ni't e- values. Fo r no s ow any significant de en is is ecause the contributi g2 configuration is ne le neg igi l in the lar e U 1
~
asa
)
-3
-2
ag
f electrons
)
for
Ko doli 't (i y && Ky A) atT = 1.16K. The f-electron K spectral DOS , p, u, is determined from the imaginary reen s unction. In the s ll-k o th ree-peakk structure is seen the
peak. T
= 1.16 K corresponds
ows
dependence
of t e Kondo resona to T T or ,
=,
~ = p, t h epeaknear P
en
emonstrated
p g ecomes smaller the the resonance shiftss t oward higher ener s i a h resonance tends to increase. b d tood i t erms of the renorm o in rmalization of the Kondo temperature. ra ure. imilar features were o imi th e large N treatment o f the finite-U And men h b t ofth e d oubly occu pied configuration of the orbital. ' ' Figure 3 presents the temperature de
s
e inset
of Fi
fth K d
fh
tk
0.5
d U h ave a very weak tern e emperature depenon o ea dence, the Kondo pea is smeared out a th roug T low T~. At lo temperature on ormed of single (T T~), thee Kond o resonance is o electron excit a t ions from the man -bo y singlet ground any- o state. For T & T~, the weight of the ground state dee r creases appreci ia bl y, an thus the ondo resonance disc
0.5
O
I
0.35 0. 3
0.'24
0. 2
-4
-3
-2
-1
0
Ef
(eV)'
FIG.
,
l.
are present d 1/K by sing y me expansion method respectively. Dashed l'znes with diamond s e marks denote exa t
= . e e,K. o=,a &=0471eV 2, and oo, e, = an and T = 0usin Solidlamest dotted lines the order resultss results t
I
0. 1
0.01
o,
&
EX'LOGY (eV)
FIG.. 3. Temperature
a e hne), respectively. Other parameters ar the same as those of (a~ e ers are h
e .
t er parameters
lg. 2.
BRIEF REPORTS
appears. These features were shown previously for both the infinite-U (Refs. 14 and 22) and the finite-U AnderThe above illustrations indicate that the son model. present, rather simple approximations lead to a proper equation-of-motion scheme, which describes well finitetemperature single-particle properties of the finite-U impurity Anderson model. To summarize, we have developed an equation-o fmotion treatment of the impurity Anderson model with a Rnite Coulomb repulsion. Two types of approximations One consists of neglecting higher order are employed. Green's functions, which include three f-electron annihilation operators. The other is based on the projection operator method in obtaining a closed form of the equation of motion for the single-particle Green's function. The results obtained using these approximations become oo limit for infinite U. Self-consistent exact in the Nf
evaluation of valencies yields values that are in good agreement with the exact results of Bethe ansatz method. = 2 exhibits Further, the f-electron spectral DOS for the well-known three-peak structure in the Kondo limit, including the strong temperature dependent Kondo resonance peak. Finally, this work should be extended to describe correctly the finite-U Anderson model with a the parameters of which are more realistic large Nf, for mixed valence or heavy fermion systems. Generalization to the lattice version of the Anderson model is also required.
i'
This work was supported by the POSTECH-BSRI program of the Korean Ministry of Education, and in part by the Korea Science Engineering Foundation through the SRC program of SNU-CTP.
P. W. Anderson, Phys. Rev. 124, 41 (1961). See, e.g. , O. Gunnarsson and K. Schonhammer, in Handbook on the Physics and Chemistry of Rare Earths, edited by K. A. Gschneidner, Jr. , L. Eyring, and S. Hiifner (North163, and refHolland, Amsterdam, 1987), Vol. 10, pp. 103
erences therein. N. Grewe and F. Steglich, in Handbook on the Physics and Chemistry of Rare Earths, edited by K. A. Gschneidner, Jr. and L. Eyring (North-Holland, Amsterdam, 1991), Vol. 10, 474, and references therein. pp. 343 N. E. Bickers, Rev. Mod. Phys. 59, 845 (1987), and references therein. 3. W. Wilkins, and K. G. Wilson, H. R. Krishna-murthy, Phys. Rev. B 21, 1003 (1980); 21, 1044 (1980). A. M. Tsvelick and P. B. Wiegman, Adv. Phys. 32, 453
3. Holm, R. Kree, and K. Schonhammer, Phys. Rev. B 48, 5077 (1993). C. Lacroix, J. Phys. F 11, 2389 (1981). G. Czycholl, Phys. Rev. B 31, 2867 (1985). T. A. Costi, J. Phys. C 19, 5665 (1986). A. E. Ruckenstein and S. Schmitt-Rink, Int. J. Mod. Phys. 3, 1809 (1989); Electron Correlations in Molecules and
Solids, edited by P. Fulde, Springer Series in Solid State Sciences Vol. 100 (Springer-Verlag, Berlin, 1991). The spectral DOS sometimes has a negative value near z = p, for a large Nf. It is considered that these unphysical results arise mainly from the neglect of higher order Green s functions, which involve three annihilation operators, since such approximations might not be justified for finite-U system with large Nf. So, in the present demon2 stration, we confine our discussion to Nf case, where fs f, = 0 (a g b g c) is always satisfied. P. Schlottmann, Phys. Rep. 181, 1 (1989). O. Gunnarsson and K. Schonhammer, Phys. Rev. B 28,
(1983).
O. Gunnarsson 4815 (1985).
and K. Schonhammer,
Phys. Rev. B
31,
69,
Th. Pruschke and N. Grewe, Z. Phys. B 74, 439 (1989). H. Keiter and Q. Qin, Physica B 163, 594 (1990); Q. Qin and H. Keiter, Z. Phys. B 84, 89 (1991). A. Schiller and V. Zevin, Phys. Rev. B 47, 9297 (1993).
4315 (1983). J. W. Rasul and A. C. Hewson, J. Phys. C 17, 3337 (1984). C. Lacroix, J. Appl. Phys. 53, 2131 (1982). N. E. Bickers, D. L. Cox, and 3. Wilkins, Phys. Rev. B 36, 2036 (1987).