Journal of Magnetism and Magnetic Materials 100 (1991) 261-271

North-Holland

Theory of itinerant electron magnetism

Tfru Moriya
Department of Physics, Faculty of Science and Technology, Science University of Tokyo, Yarnazaki, Noda 278, Japan

The present status of the theory of itinerant electron magnetism is outlined with particular emphasis on the developments in
the recent two decades, characterized by the substantial advances in the theory of spin fluctuations. The possible applicability
of the itinerant electron model to the high temperature superconductor oxides and the possible importance of the effects of
antiferromagnetic spin fluctuations are also discussed.

1. Introduction effective masses of the ferromagnetic transition

metals obtained by recent state-of-the-art band
Magnetism of itinerant electron systems, in calculations agree well with the observations.
particular, those including d-electrons and some- In the systems with weak magnetic long range
times even f-electrons, has been one of the central order, or those near the critical boundary for the
issues in condensed matter physics in this century appearance of magnetic long range order, the ef-
after the advent of quantum mechanics. This fects of electron correlations to the ground state
problem has been a difficult one because of the properties are more significant; the effective
effects of strong electron-electron correlations in masses are enhanced significantly beyond the val-
narrow energy bands which enforce us to go be- ues of the band calculations. However, the calcu-
yond traditional H a r t r e e - F o c k (HF) or mean field lated Fermi surfaces seem to agree fairly well with
theory. A number of approximation methods have the observations. In any case the band theory is
been developed for this purpose. believed to provide at least a good starting point
The problem m a y be divided into two parts. of the theory. In this article we will not discuss
One is to describe the magnetic ground state and this first part of the problem, referring the readers
the other is to describe the thermal excitations and to refs. [1,2].
the finite temperature properties of interacting The second part of the problem, i.e. the thermal
itinerant electron systems. excitations and the finite temperature properties
The first part, i.e., the ground state properties has been the more serious part of the problem,
are now believed to be described fairly well by the since the effects of electron-electron correlations
band theory which has advanced steadily and sub- are essentially important in this problem. In other
stantially in recent decades. In particular, the words, the simple Stoner theory based on the
ground state with a strong magnetic long range mean field approximation at finite temperature
order seems to be well described by the band turned out almost always to fail. The Stoner the-
theory. This may be more or less expected from ory takes account of the thermal excitations of
the following simple physical arguments; two elec- electron-hole pairs with opposite spins, each of
trons with opposite spins are spatially well sep- which moves independently of each other in the
arated in the systems with strong magnetic polari- mean potential field. This theory fails to explain
zations, and thus the effects of electron-electron the Curie-Weiss magnetic susceptibility observed
correlations are no longer very important there. in almost all ferromagnets far below their Fermi
As a matter of fact, the Fermi surfaces and the temperatures. Also, the calculated Curie tempera-

0304-8853/91/$03.50 © 1991 - Elsevier Science Publishers B.V. All rights reserved

262 T. Moriya / Theory of itinerant electron magnetism

ture Tc is at least an order of magnitude too high antiferromagnets. In other words, the theory con-
compared with experiment. This disagreement is tains a limited number of parameters (defined in
considered to be natural since the thermal excita- the k-space) which can be determined experimen-
tions in the Stoner theory are unrealistic, neglect- tally. When the number of independent experi-
ing the vitally important effect of the interaction mental methods to determine the parameter values
between the excited electron and the hole. When exceeds the number of parameters parameter-free
this effect is taken into account the excitations tests of the theory become possible. This type of
become collective modes with quite different exci- tests were carried out in these two decades and
tation energy spectra from the Stoner excitations were so successful as to establish the SCR theory
of free electron-hole pairs. When an electron-hole as the first fully satisfactory theory of itinerant
pair makes a bound state it corresponds to a spin electron magnetism for this class of materials. In
wave excitation and when the bound state is not other words, concerning the weak itinerant ferro-
formed the excitation becomes a collective mode and antiferromagnets we are in the same position
of dissipative nature, for example, like paramag- as concerned with the magnetic insulator com-
nons. Above the Curie temperature all the thermal pounds where the local moment theory was estab-
excitations are of dissipative nature. Thus it is lished with an innumerable number of tests by
essential to deal with the exchange-enhanced col- using the parameters defined in real space.
lective modes of magnetic excitations of spin den- Thus in 1970's the theory of magnetism was
sity fluctuations in the theory of finite tempera- advanced to the stage of having two established
ture properties of itinerant electron magnetism. regimes in the opposite extremes, the local mo-
Historically, the possible existence of local mo- ment systems and weak itinerant ferro- and anti-
ments in metals was the earlier subject of investi- ferromagnets. In the former regime the spin
gations in 1950's and 1960's. This is due to the fluctuations are localized in real space while in the
fact that the Curie-Weiss susceptibilities observed latter they are localized in k-space.
in almost all ferromagnets were long believed to Natural direction of research after these ad-
be explained solely by the local moment theory by vances seemed to develop a theory to fill in the
Langevin and Weiss. In 1960's the dynamical gap between the two extremes, or to develop an
Hartree-Fock theory or the random phase ap- interpolation theory for the intermediate regime.
proximation (hereafter abbreviated to RPA) was Such theories were advanced in late 1970's and
also developed to describe the collective excita- 1980's by using the functional integral method
tions in the k-space. However this theory did not within the static approximation. These theories
at all improve the Stoner theory for finite temper- were successful not only in explaining a number of
ature properties; the RPA theory calculates the so far unexplained experimental results qualita-
spin fluctuations around the Stoner equilibrium tively or semiquantitatively but also in providing a
state only. general perspective for the unified picture of mag-
In 1970's the self-consistent renormalization netism of the d-electron systems. As a matter of
(hereafter abbreviated to SCR) theory of spin fact this interpolation theory also enabled us to
fluctuations was developed [3-5]. This theory goes calculate the phase diagram for the Mott transi-
one step beyond the HF-RPA theory and deals tion in magnetic compounds for the first time on
with the coupled modes of spatially extended spin the basis of the Hubbard model. The dynamical
density fluctuations in a self-consistent fashion. spin fluctuation theory, however, is still to be
This theory gave a new mechanism for the Curie- developed for the intermediate regime.
Weiss magnetic susceptibility associated with the Since 1986, various high temperature supercon-
spatially extended modes of spin fluctuations in- ductor cuprates were discovered [6,7]. They are
stead of the local moments as spatially localized doped Mott insulators including charge transfer
spin fluctuations. Furthermore, the theory was insulators, and the relevant electrons are consid-
successful quantitatively in explaining various ered to form strongly correlated two-dimensional
physical properties of weak itinerant ferro- and systems in the CuO2 plane. Although various new
 T. Moriya / Theory o f itinerant electron magnetism 263

Table 1
Anomalous temperature dependences of physical quantities around the critical phase boundary for itinerant ferro- and antiferromag-
netism predicted by the SCR theory. XQ: magnetic susceptibility, R: electrical resistivity, TI: nuclear spin-lattice relaxation time,
Cm: magnetic contribution to the specific heat, C.W. (Curie-Weiss): T + 0
3-dimensi0nal 2-dimensional Normal
ferro (Q = 0) a.f. ferro (Q = 0) a.f. Fermi liquid
XQ 1 T 4/3 ~ C.W. T 3/2 --', C.W. T 4/5 ~ C.W. T+ 0 a + bT 2
R T 5/3 T 3/2 T 4/3 T T2
r? 1 TX TX~ 2 rX 3/2 TXo
Cm/T - I n ]1 - a u ] I1 - au 1-1/2 - I n 11 - a s I

approaches from the strong correlation limit have metallic magnetism based on the local moment
been developed for this problem, in particular, to picture, in particular their applicability and limita-
explain various anomalous physical properties tions. In section 5 we discuss, in view of the
above the critical temperature T~, it seems that unified picture of magnetism, on the possible
these anomalous properties are well explained in physical reasons why the itinerant electron model
terms of the antiferromagnetic spin fluctuations in with antiferromagnetic fluctuations can be applied
two-dimensional itinerant electron systems as de- to high T~ oxides as doped Mott insulators. Finally
scribed by the SCR theory [8]. This fact seems to a brief summary is given in section 6.
indicate the importance of studying magnetism
and superconductivity simultaneously in strongly
correlated itinerant electron systems. 2. Self-consistent renormalization (SCR) theory of
As was mentioned briefly in the above, there spin fluctuations and weak itinerant ferro- and
have been quite interesting developments in these antiferromagnetism
two decades in the field of itinerant electron mag-
netism and to make a brief survey of these devel- The discovery of weak itinerant ferromagnetism
opments is the purpose of the present article. In in various intermetallic compounds, such as
view of the fact that we already have several ZrZn 2, Sc3In, Ni3A1, MnSi, etc. in late 1950's and
review articles which cover substantial part of 1960's and successful explanations of their physi-
these development [3-5,9], we will confine our- cal properties in terms of the SCR theory of spin
selves in this short article to explain the outline of fluctuations in 1970's have opened a new horizon
the main streams of these developments without in the theory of itinerant electron magnetism.
going into details for which we refer the readers to These investigations established a new category of
the previous review articles and recent papers. magnetism in the opposite extreme to the long
Thus no intention is made to cover all the im- familiar local moment systems.
portant research activities or to give a balanced First of all, the discovery by the SCR theory of
list of references. the new mechanism for the Curie-Weiss (CW)
In section 2 we discuss the present status of the susceptibility due to the interacting extended
self-consistent renormalization theory of spin modes of spin fluctuations has released us from a
fluctuations for weak itinerant ferro- and antifer- long-dominant prejudice of connecting the ob-
romagnetism which established a new regime of served CW susceptibility directly with the ex-
magnetism in the opposite limit to the local mo- istence of local moments. Another important point
ment regime. Recent application of the theory to of this development is the quantitative success of
high Tc cuprates will also be discussed. Before the SCR theory just in the same sense as the
discussing on a unified picture of magnetism based successful establishment of the local moment the-
on the interpolation theory in section 4, we briefly ory in magnetic insulator compounds [10]. Thus
mention in section 3 about early theories of the famous controversy over itinerant vs. localized
264 T. Moriya / Theory of itinerant electron magnetism

models, started with the advent of the quantum couplings in a self-consistent fashion, X0 being the
mechanical theories of magnetism, was believed to susceptibility of non-interacting electrons:
be resolved, at least partly, into a well-defined
problem of characterizing the nature of spin den- 1 1
- - + }F1S2L(T), (2.3)
sity fluctuations in each substance. x(O)
For weakly and nearly ferro- and antiferromag-
t9 -2 t
netic metals the SCR theory predicted that the S~(T)=3 f&ocoth(-~)No Y', I m x ( q , ~ ) .
spin fluctuations give rise to various anomalous q

temperature dependences of various physical (2.4)

quantities different from the prediction of the
normal Fermi liquid theory. These anomalies oc- Here XrtF(0) is the H a r t r e e - F o c k result for the
cur above a certain crossover temperature which uniform susceptibility given by XHF = X0/( 1 -- a),
decreases and tends to zero as the phase boundary (1 - a)-~ being the exchange enhancement factor,
is approached. Below the crossover temperature and the second term in eq. (2.3) represents the
physical properties show the Fermi liquid-like be- effect of t h e ' m o d e - m o d e coupling or renormaliza-
haviors. We summarize some of these anomalous tion. Since eqs. (2.1) and (2.2) for x(q, o~) contain
properties in the left hand side of table 1. All of X(0), the eqs. (2.1)-(2.4) must be solved con-
these predictions were borne out by subsequent sistently for X(0). This is the key point of the SCR
experimental investigations.in 1970's [3,4]. theory and the second term in eq. (2.3) is essential
As for the quantitative test of the theory, the for the Curie-Weiss behavior.
SCR theory contains a small number of parame- One may notice that the structure of the dy-
ters in the weakly ferromagnetic limit. The param- namical susceptibility in eqs. (2.1)-(2.2) is the
eters may be chosen as follows: m(0), the ordered same as the RPA theory. However, the RPA uses
moment per atom or X(0,0), the magnetic suscept- XHF(0) for X(0), neglecting the second term in eq.
ibility at T = 0 K for substances with para- (2.3) or the effect of the m o d e - m o d e coupling in
magnetic ground state, F 1, the m o d e - m o d e cou- contrast with the SCR theory and thus fails to
pling constant for the long wavelength modes of improve the Stoner theory. Furthermore, in the
spin fluctuations which may be deduced from the SCR theory the parameters a, A and F0 are also
Arrott plot of the magnetization data, A, the renormalized in general, although the effect is not
exchange or transverse stiffness constant for the so dramatic as for X(0). In actual comparisons
spin density, and F0, the parameter characterizing between the theory and experiments we use the
the energy spreads of the spin fluctuation modes. experimentally measured values for the parame-
The dynamical susceptibility above To for exam- ters. m(0) or X(0,0) and F 1 are obtained directly
ple, is given in terms of these parameters as fol- from the static magnetization measurements, while
lows: F0 and A can be obtained directly from the neu-
tron scattering experiment and indirectly from
various other measurements. Thus the measured
x(q, o~) = x ( q ) / ( 1 -iW/Fq), (2.1) values for the parameters may be regarded as the
renormalized ones.
with Various physical quantities such as the Curie
temperature Tc, the magnetic susceptibility above
Tc, the specific heat, nuclear spin relaxation rate
x(q) = X ( 0 ) / ( 1 + qZ/K2), Fq= Foq(x 2 + q2),
and its magnetic field dependence, electrical and
thermal resistivities, magneto-resistance, are now
~2 = Xo(O)/Ax(O), (2.2) expressed practically in terms of these four param-
eters. Intensive and extensive experimental investi-
where x(O) is obtained by solving the following gations of these properties were carried out in
equation which takes account of the m o d e - m o d e 1970's and early 1980's with overwhelming success.
 T. Moriya / Theory of itinerant electron magnetism 265

We will not show details here referring to refs. anomalous behaviors characteristic to spin
[3-5,9] for interested readers. fluctuations. The anomalous properties observed
One apparent difficulty with the SCR theory in high Tc oxides reminded us of these previous
has been associated with the reported discovery of results in weak itinerant ferro- and antiferromag-
temperature-induced ferromagnetism in Y2Ni7 nets. Thus the SCR theory was extended to the
about ten years ago. Although the Stoner theory two-dimensional itinerant electron systems.
reproduced this result with the use of a certain Such an extension of the theory is straightfor-
model for the density of states, the SCR theory ward and the results contain many interesting
showed from a general point of view that the features. We discuss here only the antiferromag-
effect of spin fluctuations make such a phenome- netic case. The first of all it is shown that there is
non hardly possible [11]. This serious experimental no antiferromagnetic long range order at finite
challenge to the SCR theory, however, disap- temperatures even if the ground state is antiferro-
peared recently; the previous result of measure- magnetic. When the ground state is antiferromag-
ments was reported to be an artifact caused by netic, the staggered magnetic susceptibility di-
gadolinium impurities after careful recent investi- verges toward T---0 exponentially; we have X cc
gations [12]. exp[KM~(O)/T]/T at low temperatures, where
As an important recent topic we next discuss MQ(0) is the staggered magnetization at T = 0 and
the possible relevance of the SCR theory of anti- K is a constant. This behavior is similar to that in
ferromagnetic spin fluctuations to the physical the two-dimensional Heisenberg antiferromagnets.
properties of high temperature superconductor When the ground state is paramagnetic (nearly
cuprates. Intensive investigations after the dis- antiferromagnetic) we have XQ(0) -- XQ(T) ec T 2
covery of the high T~ oxides have shown that their at low temperatures. At elevated temperatures
physical properties such as electrical resistivity, XQ(T) obeys the CW law in both of the above
Hall effect, optical conductivity, nuclear spin- cases. We will not go further into the details of the
lattice relaxation rates, etc. are quite anomalous theory, but show in the right-hand side of table 1
above the critical temperature T~. These anoma- the SCR predictions for various anomalous behav-
lous properties led many people to suspect the iors in the two-dimensional system around the
possible breakdown of the Fermi liquid picture. critical phase boundaries. We find that these re-
Since these high T~ cuprates are doped Mott insu- suits for antiferromagnetic case just explain the
lators where the electron-electron correlations are reported anomalous behaviors of high T~ cuprates
expected to be quite strong and the conducting above T~.
electrons or holes are believed to form two-dimen- Furthermore, the theory was successful even
sional systems in the CuO2 plane, novel ap- quantitatively. Electrical resistivity, specific heat,
proaches from the strong correlation limit in two- nuclear spin-lattice relaxation rates of 63Cu and
dimensional models were developed and extended infrared conductivity were consistently explained
intensively [6,7]. with the use of only four parameters. In particu-
From a rather different point of view, the anti- lar, good parameter-free comparisons were re-
ferromagnetic spin fluctuations in two-dimen- ported for the infrared conductivity, after fixing
sional itinerant electron systems were recently in- all the parameter values from the other experi-
vestigated by using the SCR theory [8]. As was mental results. Thus we expect that the SCR the-
seen in table 1, the SCR theory for the three- ory applies well to two-dimensional itinerant elec-
dimensional magnets predicted quite anomalous tron systems and the nearly antiferromagnetic
temperature dependences for various physical two-dimensional itinerant electron system pro-
quantities near the critical boundary for the mag- vides a good model for the high Tc cuprates. For
netic long range order. In this case the behaviors further details we refer to the original papers [8].
in the low temperature limit are just normal Fermi After further extended discussions on the itinerant
liquid-like and there is a certain crossover temper- electron magnetism and on a unified picture of
ature above which we have a wide range of magnetism in sections 3 and 4, we will come back
266 T. Moriya / Theory of itinerant electron magnetism

to this problem in section 5 and discuss about model, he discussed the magnetism of d-metals
possible reasons why this model can apply to high and their alloys with qualitative and semiquantita-
Tc cuprates as doped Mott insulators. tive success [18]. The same type of arguments were
extended also to the Wolff tight-binding model
with essentially the same results [19]. The func-
3. Early theories based on the local moment tional integral theory was later applied to the two
picture impurities problem and the local saddle point
approximation turned out to correspond to the
It has been well known for many years that the above local H a r t r e e - F o c k theory [16]. The re-
local moment theory of magnetism was well markable success, though qualitative, of these early
established for magnetic insulator compounds and theories were then considered to be an encourag-
in most of the rare earth metals. As was men- ing signal for further extension of this line of
tioned in the previous section we now have the approach. As a matter of fact Hubbard [20], for
other established theories for weak itinerant ferro- example, used a similar local moment idea much
and antiferromagnets, which constitute the oppo- later in his calculation of the Curie temperature of
site limit to the local moment systems. It is thus iron by using the result of band calculations. In
important to fill in the gap between these two order to estimate the effective Heisenberg ex-
extremes. Before going into this subject in the next change constant he calculated the energy cost of
section let us discuss here some earlier attempts at turning a local moment from the direction of the
the theory of metallic magnetism based on the ferromagnetic moment in the ground state of iron
local moment picture and their limitations when as a function of turning angle. Incidentally, the
applied to d-metals. Such a review of rather old functional integral theory for the Hubbard model
but initial theories may be worthwhile to be given was first discussed by Cyrot by using the coherent
here briefly, since more detailed calculations along potential approximation and the local saddle point
this line, based on the results of band calculations, approximation [21]. It seems important to notice,
have been pursued around this decade. however, that these theories are of qualitative or
The importance of electron-electron correla- semiquantitative significance and should by no
tions in d-band metals was emphasized by Van means be regarded as establishing the local mo-
Vleck who seemed to conceive a possible local ment picture in metallic magnets.
moment picture in d-metals [13]. The theory of Exactly speaking, the local moment theory or
local moment in metals was first developed for the Heisenberg model, for d-electrons are estab-
dilute alloys [14,15]. The condition for the ap- lished only for the Mott insulators, where the
pearance of the local moment was explicitly dis- energies of spin slip excitations or bound elec-
cussed by Anderson with the use of the Anderson tron-hole pair excitation are well separated from
model within the H a r t r e e - F o c k approximation the excitation energies of unbound electron-hole
[15]. Anderson correctly connected this result to pairs. In magnetic d-metals these two types of
the Curie-Weiss susceptibility. This corresponds, excitation spectra overlap in general. Thus the
as a matter of fact, to use a saddle point ap- concept of local moment cannot be well-defined
proximation in the functional integral formalism and have only approximate significance at best.
as was discussed later by Schrieffer et al. [16]. As will be discussed in the following section, the
Alexander and Anderson extended the theory to amplitude variation of the local spin density should
two impurity problems and discussed the effective generally be taken into consideration when we
exchange coupling constants [17]. The present deal with metals. This fact should be kept in mind
author generalized this theory to include all kind when we discuss theories for the intermediate reg-
of pairs of neighboring moments and obtained ime.
simple rules relating the sign of the effective ex- To be more explicit, we expect that most of the
change constant and the occupied fraction of the magnetic d-metals are in the intermediate regime.
d-shell. Then switching to the effective Heisenberg Some of them are clo~e to the local moment limit
 T Moriya / Theory of itinerant electron magnetism 267

while there are some others close to the weakly stantial short range order remains above Tc the
ferro- and antiferromagnetic limit. For example, magnetic susceptibility should deviate from the
Heusler alloys such as Pd2MnSn, where magnetic Curie-Weiss law significantly [22]. Since almost
Mn atoms are well separated from each other, are all the metallic ferromagnets exhibit good Curie-
practically in the local moment regime. FePt 3 are Weiss susceptibilities the local moment picture
expected to be close to the local moment regime (the constant amplitude of the local spin density)
while Fe3Pt are not quite so, according to the and the short range magnetic order are not com-
neutron scattering experiments. Neutron scatter- patible in them.
ing experiments on Fe and Ni show that their spin On the other hand the long wavelength compo-
wave energies extend to much higher energies than nents of spin fluctuations are predominant in weak
kBTc, indicating that they can hardly be described itinerant ferromagnets even far above the Curie
consistently by the local moment theory. For fur- temperature, i.e., strong short range order remains
ther discussions on various magnetic metals see even far above Tc. This is the result of the SCR
for example ref. [4]. Thus, the remarkable qualita- theory which gives the new mechanism for the
tive success of the early theories based on the local Cur/e-Weiss susceptibility compatible with the
moment picture should mean that in describing short range magnetic order, and was confirmed in
some of the physical properties of certain metallic MnSi by neutron scattering experiment [23]. Thus
magnets the spin fluctuations may fairly well be the consideration of the short range magnetic order
approximated by those arising from a set of local should be another important factor to be taken
moments. The limitation of the local moment pic- into account in the unified theory.
ture as discussed in the above should be kept in The third point of contrast is the energy spread
mind in extending more realistic theories based on of the spin fluctuations compared with kBTc. Now
the calculated band structures. Fq denotes the energy spread of the Fourier q-
component of the spin fluctuation above Tc. In
the local moment systems the q-dependence of Fq
4. A unified picture of magnetism - an interpola- is relatively weak except for the range of small q
tion theory and we have (Fq) = kBTc. While in weak itinerant
ferromagnets Fq increases significantly with q and
Let us now compare the characteristic features w e have (ffq) >> k B T c.
of spin fluctuations in the two opposite extremes. It is now desired to make a theory of general
The first point is concerned with the local am- spin fluctuations which gives the already known
plitude of the spin density which should be kept correct results in the opposite two extremes and
constant in the local moment systems but not in smoothly interpolate between them. It seems to be
itinerant electron systems in general. As a matter a very hard task to make such a unified theory for
of fact the mean square local amplitude of the dynamic spin fluctuations which should give a
spin density SL2 varies significantly with tempera- final resolution of the controversy over localized
ture in weak itinerant ferromagnets and its in- vs. itinerant pictures in metallic magnetism. As a
crease with temperature above Tc is considered to step toward this goal various spin fluctuation the-
give the origin of the new mechanism of the ories were developed within the adiabatic or static
Curie-Weiss susceptibility. Thus the variable na- approximation by using the functional integral
ture of the amplitude of the spin density must be formalism.
taken into account in the unified theory. Since earlier functional integral approaches to
The second point is concerned with the spatial this problem employed the local saddle point ap-
spin correlations or the short range magnetic order proximation which resulted in a constant ampli-
above the transition temperature Tc. It is well tude of the local moment or the local spin density
known that in the usual Heisenberg local moment above Tc, we had clearly to proceed beyond this
ferromagnets the short range order decays rapidly approximation, so that the variation in amplitude
with increasing temperature above Tc. When sub- of the local spin density was taken into account.
268 T. Moriya / Theory of itinerant electron magnetism

Such a theory was first presented in late 1970's and in the weakly ferromagnetic limit, though
[24]. The spirit of this theory is to consider the within the adiabatic approximation.
following form of the free energy as a function of As a direct application of this formalism; the
spin variables: phase diagram in the ( U / W ) - T plane of the
Hubbard model with a half-filled band, showing
~'~'~-- -- E Vq(82) [ S q [ 2 "~ Not(82) the Mott transition, was actually calculated for the
q first time [3,25]. There are a number of numerical
q calculations performed so far based on the band
= - NoL(S ), structures of magnetic transition metals. They are
j,l mostly successful semiquantitatively, although
some of them assume a constant local amplitude
with
of the spin density and the others either neglect or
S~ = N o 2 E l S q l 2, E Vq(S~) = 0, (4.1) empirically parametrize the short range order [26].
q q In view of the fact that the established theories
in the both extremes, in practice, are those includ-
where No is the number of magnetic atoms and ing parameters, it is also desirable to make a
the functional forms of Vq(S~) or Vj,(SL2) and parametrized theory for the intermediate regime.
L(S 2) are to be calculated from a given band Although it is difficult to find any valid expansion
structure. The partition function is then calculated parameters in the intermediate regime, it should
by taking an average of e -'WkBT with respect to be useful if the two extremes are continuously
both direction and amplitude of the spin variables. connected in terms of certain parameters even
This model may be regarded as a generalization of qualitatively.
the Heisenberg model to include the variation in For this purpose the longitudinal stiffness con-
amplitude of the local moments. However, the stant for the spin density was introduced. When
above discussion, using classical spin variables, this parameter is large enough we are in the local
shows only the spirit of the theory and in actuality moment regime with fixed S L, 2 while small values
the Stratonovich-Hubbard technique was em- for the parameter allow SL2 to change with temper-
ployed to deal with the quantum mechanical sys- ature. Introduction of this parameter and the
tems. Confining ourselves to the static approxima- transverse stiffness constant (corresponding to the
tion, we calculate qJ[~], a free energy of the non- familiar exchange stiffness constant) makes it pos-
interacting system under the influence of the mag- sible to discuss general features of various physical
netic field (¢rUkBT)1/2~j at the site j, etc., and quantities in the intermediate regime interpolating
then take an average of exp( - ,~Ej ~ - qJ[ ~ ]/k BT) between the extremes. For example, a general
over the field variables { ~i }, leading to the parti- expression for the magnetic susceptibility was ob-
tion function exp(-F/kBT). The charge density tained, interpolating between the two extremes
fluctuations can also be taken into account in this characterized by different mechanisms of the
scheme introducing a charge field variable ~/j. Curie-Weiss law.
Various methods of calculating ~b[~, r/] from a One of the most important features of the
given band structure and several methods of unified picture is the temperature variation of S 2
evaluating the functional integrals were presented. which depends primarily on the band structure
We refer to ref. [4] for these developments and and the electron occupation. This new degree of
just mention here that a closed form expression freedom enabled us to explain various old enigmas
for ~[~, ~/], corresponding to eq. (4.1) was given or so far unexplained phenomena. For example
[3,25]. This expression takes care of the arbitrary various kinds of anomalous temperature depen-
variation of the amplitude of local spin density dences of the magnetic susceptibilities in FeSi,
and the pairwise spatial spin correlations and re- COS2, CoSe2, MnAsl_~Px, etc. were explained
duces to the known correct results in the local primarily in terms of the specific temperature vari-
moment limit of the half-filled Hubbard model ations of SL2 in them [27]. The Invar phenomenon
 T. Moriya / Theory of itinerant electron magnetism 269

was associated with the band structures which give ties of the system, not to speak of its poor com-
rise to strong temperature variations of S 2. parison with experiment.
The phenomena of coexistence of ferro- and In order to make any consistent descriptions of
antiferromagnetism and the transition between the physical properties at finite temperatures we
them are reasonably described with the use of this must use more advanced theories of spin fluctua-
degree of freedom or the variable nature of S 2 [4]. tions in each regime; the SCR theory for weak
Thus the interpolation theory, though within the itinerant ferro- and antiferromagnets, the local
adiabatic approximation, was rather successful in moment theories for the Mott insulators, and the
giving a unified picture for a broad range of interpolation theory (yet to be worked out beyond
magnets and in elucidating various old problems the adiabatic approximation) for the itinerant
that remained without explanation. magnets in the intermediate regime.
In the above discussions we are dealing with
the itinerant magnets and the Mott insulators on
the same footing. This point of view made it
5. High T~ oxides as doped Mott insulators possible to describe the phase diagram including
the Mott transition in section 4. In some previous
As was mentioned in section 1 the magnetic strong correlation theories of the Mott transition
ground states with strong spin polarizations are [28,29], the antiferromagnetic nature of the ground
expected to be described primarily by the state is neglected and thus the theories practically
H a r t r e e - F o c k or band theory. This applies to the deal with the relatively high temperature phase
antiferromagnetic ground state of the Mott insula- although the temperature does not explicitly come
tor, as represented by the Hubbard model with a into the theories. The meanings of these theories
half-filled band, where even the spin waves as should thus be taken with care; they may be
elementary excitations are correctly described by useful for very qualitatively understanding of the
the RPA, i.e, the RPA results for large U/t ratio nature of this problem but by no means indicate
agree precisely with those obtained from the cor- the breakdown of the H F or band theory in the
responding Heisenberg model with the Anderson antiferromagnetic ground state.
kinetic superexchange interaction [4]. Now let us consider the high T~ oxides as doped
These situations are the same as in the itinerant Mott insulators. In most of the doped cuprates of
electron magnets where the ferromagnetic ground our interest, antiferromagnetic long range order is
states and the spin waves are fairly well described reduced rapidly with doping and superconductiv-
by the HF-RPA scheme with certain correlation ity seems to appear only in the range of small or
corrections. vanishing antiferromagnetic long range order and
Since the Hartree-Fock-RPA scheme is known a substantial number of charge carriers whose
to be correct in the weak coupling regime, it is effective masses do not seem to be much en-
natural to expect that the ground state and the hanced.
elementary excitations from the ground state are It may be pertinent to note here about the
fairly well described by this scheme in practically effect of random impurity potential in this prob-
the whole range of interaction. At least this scheme lem. This effect is expected to be important when
should give a good starting point of the theory the number of carriers is small and the Anderson
and the many body corrections, which are ex- localization takes place. When the number of car-
pected to be most significant around the boundary riers is substantial, as in the high Tc regime, the
of magnetic phase transition, are expected to give impurity potential is expected to be screened so
rise to no serious qualitative change of the physi- that we may neglect it in the first place. The
cal picture. screening effect is also expected on the electron-
At finite temperatures, however, the H F - R P A electron interactions and the effective interaction
scheme fails almost completely; it no longer gives may be reduced from the value in the insulator
any consistent description of the physical proper- phase.
270 T. Moriya / Theory of itinerant electron magnetism

Now in view of the general arguments in the 6. Conclusions

earlier part of this section and in the preceding
section, the high Tc oxides seem to be well re- As was outlined in the above sections our un-
garded as weakly or nearly antiferromagnetic derstanding of itinerant electron magnetism has
itinerant electron systems and thus their normal advanced significantly in these two decades, as far
states may be treated in terms of the SCR theory. as the d-metals and their compounds are con-
One should note here that the above arguments cerned. It has turned out that the ground state
may apply certainly to three-dimensional systems properties of the d-metals are fairly well described
but not quite evidently to two-dimensional sys- by the band theory with some many body correc-
tems of the present interest. However, the success- tions, while the thermal excitations and finite tem-
ful application of the SCR theory of two-dimen- perature properties should be treated by the ap-
sional itinerant antiferromagnets to interpret the propriate theories of spin fluctuations beyond the
anomalous physical properties of high T~ oxides, H F - R P A scheme.
as mentioned in section 2, seems to indicate that The old controversy over the itinerant and the
the SCR theory is applicable to two-dimensional localized pictures was concerned primarily with
itinerant electron systems and the two-dimen- the nature of magnetic excitations rather than the
sional nearly antiferromagnetic itinerant electron ground state properties and was resolved into the
system is a good model for the relevant electrons problem of the nature of spin density fluctuations
in high Tc oxides. in each substance which varies from material to
Since antiferromagnetic spin fluctuations are material, ranging between the local moment limit
expected to be responsible for the anomalous and the extended moment limit (localized in k-
physical properties, including the electrical resis- space).
tivity, of high Tc cuprates above T~, it may be We now have established theories of spin
natural to expect the same spin fluctuations to fluctuations in the two extremes, the local moment
play an important role in the mechanism of super- regime and the weakly ferro- and antiferromag-
conductivity. netic regime. As for the intermediate regime, where
Antiferromagnetic spin fluctuations are known most of the d-metals and their compounds are
to be a possible origin of the anisotropic supercon- considered to belong the theories at present are
ductivity of d-wave character [30]. This mecha- based on the adiabatic approximation for the spin
nism is believed to be realized in some of the fluctuations. Although the interpolation theory not
heavy electron superconductors. The antiferro- only gave a general picture of this regime but also
magnetic spin fluctuations in high Tc cuprates as succeeded in explaining many old enigmas or un-
evaluated from the SCR analyses discussed in explained phenomena, the dynamical theory of
section 2, were employed to study this mechanism spin fluctuations is still to be developed in the
of superconductivity in high Tc oxides within the intermediate regime.
weak coupling theory [8]. The result showed that Aside from the above mentioned traditional
the superconducting order was of Big type and T~ problems of itinerant electron magnetism, recent
was estimated to be of the right order of magni- discovery of high temperature superconductor
tude. From experimental point of view it seems oxides seems to have introduced a challenging
that, there is no generally accepted conclusion at problem in this field. We have strongly correlated
present as to the type of superconductivity in high itinerant d-electrons in these systems, doped Mott
T~ oxides. The above theoretical analysis is still insulators, where the antiferromagnetic spin
too simple to be convincing. However, in view of fluctuations seem to play an important role. This
the success of the above mentioned analysis above fact makes us to expect that magnetism and super-
Tc we expect the antiferromagnetic spin fluctua- conductivity should somehow be related strongly.
tions to play an important role in the mechanism This problem will provide us with one of the most
of superconductivity. challenging area of investigation in coming years.
 T. Moriya / Theory of itinerant electron magnetism 271

References [14] J. Friedel, Nuovo Cimento suppl. 7 (1958) 287.

[15] P.W. Anderson, Phys. Rev. 124 (1961) 41.
[16] J.R. Schrieffer, W.E. Evenson and S.Q. Wang, J. de Phys.
[1] C. Herring, Magnetism IV, eds. G.T. Rado and H. Suhl
(Academic Press, New York, 1964). 31 (1971) C1.
[2] V.L. Moruzzi, J.F. Janak and A.R. Williams, Calculated [17] S. Alexander and P.W. Anderson, Phys. Rev. 133 (1964)
Properties of Metals (Pergamon, New York, 1978). A1594.
O.K. Anderson, O. Jepsen and D. G16tzel, in: Highlights [18] T. Moriya, Prog. Theor. Phys. 33 (1965) 157.
[19] M. Inoue and T. Moriya, Prog. Theor. Phys. 38 (1967) 41.
of Condensed-Matter Theory, eds. F. Bassani, F. Fumi
and M.R. Tosi (North-Holland, Amsterdam, 1985) p. 59. [20] J. Hubbard, Phys. Rev. B 19 (1979) 2626.
[3] For a review see: T. Moriya, J. Magn. Magn. Mater. 14 [21] M. Cyrot, J. de Phys. 33 (1972) 125; Phil. Mag. 25 (1972)
(1979) 1. Also see the following two references. 1031.
[4] T. Moriya, Spin Fluctuations in Itinerant Electron Mag- [22] For a current discussion see: B.S. Shastry, D.M. Edwards
netism, Springer Series in Solid State Sciences, vol. 56 and A.P. Young, J. Phys. C 14 (1981) L665.
[23] Y. Ishikawa, Y. Uemura, C.F. Majzak, G. Shirane and Y.
(Springer, Berlin, 1985).
[5] T. Moriya, Unified Picture of Magnetism, in: Metallic Noda, Phys. Rev. B 31 (1985) 5884.
Magnetism, Topics in Current Physics, vol. 42, ed. H. [24] T. Moriya and Y. Takahashi, J. Phys. Soc. Jpn. 45 (1978)
Capellmann (Springer, Berlin, 1987) p. 15. 397.
[6] For a recent review see for example: High Temperature [25] T. Moriya and H. Hasegawa, J. Phys. Soc. Jpn. 48 (1980)
Superconductivity, eds. K.S. Bedell, D. Coffey, D.E. 1490.
Meltzer, D. Pines and J.R. Schrieffer (Addison Wesley, [26] For a review see ref. [4] and the following references:
Reading, 1990). See also the following reference. Electron Correlation and Magnetism and Narrow Band
[7] Physics and Chemistry of Oxide Superconductors, eds. H. Systems, Springer Series in Solid State Sciences, vol. 29,
Yasuoka and Y. Iye (Springer, Berlin, 1991). ed. T. Moriya (Springer, Berlin, 1980); Metallic Mag-
[8] T. Moriya, Y. Takahashi and K. Ueda, J. Phys. Soc. Jpn. netism, Topics in Current Physics, vol. 42, ed. H. Capell-
59 (1990) 2905; in ref. [7]. mann (Springer, Berlin, 1987).
T. Moriya and Y. Takahashi, J. Phys. Soc. Jpn. 60 (1991) [27] References may be found in ref. [4] except for MnAs I _xPx
776. for which see: K. Motizuki and K. Kato, J. Phys. Soc.
[9] G.G. Lonzarich, J. Magn. Magn. Mater. 54-57 (1986) Jpn. 53 (1984) 735.
612. [28] J. Hubbard, Proc. Roy. Soc. A281 (1964) 401.
[10] P.W. Anderson; in: Solid State Physics, vol. 14, eds. F. [29] W.F. Brinkman and T.M. Rice, Phys. Rev. B 2 (1970)
Seitz and D. Turnbull (Academic, Press, New York, 1963) 4302.
p. 99. [30] K. Miyake, S. Schmitt-Rink and C.M. Varma, Phys. Rev.
[11] T. Moriya, J. Phys. Soc. Jpn. 55 (1986) 357. B 34 (1986) 6554.
[12] R. Ballou, B. Gorges, P. Mollo and P. Rouault, J. Magn. D.J. Scalapino, E. Loh, Jr. and J.E. Hirsch, Phys. Rev. B
Magn. Mater. 84 (1990) L1. 34 (1986) 8190.
[13] J.H. Van Vleck, Rev. Mod. Phys. 25 (1953) 220.