CRITICAL FIELDS OF SUPERCONDUCTING Sn, In, AND Ta 91

The deviation, D(t), of the critical field curve from

.02— parabolicity is defined as the difference between the
true critical field curve and a parabola drawn through
the experimental values of Ho and T, :

.OI— H, (g)
—(1 —gs)
I
II
Ig
0 Such curves are shown in Fig. 3 for the materials
discussed here as well as for lead and mercury. It is
clear that the diGerences among tin, indium, and
tantalum are small. The amplitudes of these curves
are sensitive to the choice of Ho and thus are affected
by the uncertainty in the extrapolation to O'K.
Measurements below 1'K will be required before the
".02— differences in amplitude of D(t) for these elements can
be considered to be clearly established. Further
measurements, if extended to suKciently low temper-
-.03- atures to yield reliable y values, will also permit the
deduction of the temperature dependence of C„(the
I I
super conducting electronic specific heat) from the
FIG. 3. Deviation of critical Geld curves of several super- shape of D(t).
conductors from parabolicity. For clarity the curves for individual
samples of tin and indium are not shown (since the di6'erences in IV. ACKNOWLEDGMENTS
D amount to 0.0014 or less).
Itis a pleasure to acknowledge the assistance of
found in dH, /d T' —
as temperature is lowered D. K. Finnemore in taking the measurements.
Whether this behavior will extend into the range Discussions with Professor C. A. Swenson, Dr. J. I.
below 1'K remains to be seen. " Budnick, and Dr. M. Garfinkel have been most helpful.
"It should be pointed out that the work of E. A. Lynton, data are extrapolated to O'K by means of an expression which takes
B. Serin, and M. Zucker on tin indicates that a value of y in explicit account of the exponential temperature dependence of
better agreement with calorimetric data results if critical Geld the superconducting electronic speciGc heat.

PH YSICAL REVIEW VOLUME 120, NUM BER 1 OCTOBER 1, 1960

Anisotropic Superexchange Interaction and Weak Ferromagnetism

TQRU MORIYA
Bell Telephone Laboratories, 3IIurray Hill, Eem Jersey
(Received May 25, 1960)

A theory of anisotropic superexchange interaction is developed by extending the Anderson theory of

superexchange to include spin-orbit coupling. The antisymmetric spin coupling suggested by Dzialoshinski
from purely symmetry grounds and the symmetric pseudodipolar interaction are derived. Their orders of
magnitudes are estimated to be (Ag/g) and (ng/g) times the isotropic superexchange energy, respectively.
Higher order spin couplings are also discussed. As an example of antisymmetric spin coupling the case of
CuC12 2H~O is illustrated. In CuCl~ 2H20, a spin arrangement which is different from one accepted so far
is proposed. This antisymmetric interaction is shown to be responsible for weak ferromagnetism in o.-Fe&03,
MnCO3, and CrF3. The paramagnetic susceptibility perpendicular to the trigonal axis is expected to increase
very sharply near the Neel temperature as the temperature is lowered, as was actually observed in CrF3.

INTRODUCTION of this phenomena based on an impurity effect, possibly

w KAK ferromagnetism of mainly antiferromag-

netic crystals, represented by cx-Fe203 and the
carbonates of Mn and Co, has been a controversial
magnetite. Many years later, Li' proposed a different
explanation based on antiferromagnetic domains with
Inagnetized walls. As he pointed out, however, the
problem for a decade. Neel' proposed an explanation formation of antiferromagnetic domains is not energeti-
'
*On leave of absence from Tokyo Metropolitan
Tokyo, Japan.
University,
' L. Weel, Ann. phys. 4, 249 (1949).
Y. Y. Li, Phys. Rev. 101, 1450 (1956).
 TORU MORI YA

cally advantageous and may be possible only when superexchange which is linear in the spin-orbit coupling
some imperfections of structural or chemical nature has an antisymmetric form as was briefiy reported. '
stabilize the domains. From these impurity mecha- In this paper we discuss the anisotropic superexchange
nisms, we expect the magnetic properties to change interaction to a fuller extent and the weak ferro-
from sample to sample depending on their purity. magnetism of o.-Fe203, the carbonates of Mn and Co
However, it seems to be natural to raise the following and CrF3.
question: Is weak ferromagnetism an intrinsic property
of e-Fe203 or not? An afFirmative answer was first given ANISOTROPIC SUPEREXCHANGE INTERACTION
by Dzialoshinski. ' His argument is phenomenological The anisotropic superexchange interaction is derived
and is as follows: He first showed that when the spins
by extending the theory of superexchange interaction
are perpendicular to the trigonal axis, an n-Fe203 to include the e8ect of spin-orbit coupling. We use
crystal with the antiferromagnetic spin arrangement Anderson's new formalism' of superexchange inter-
has the same symmetry as that with a canted spin
action. The basic one electron wave functions are those
arrangement which has a net magnetic moment localized at the positions of magnetic ions and are
perpendicular to the trigonal axis. The next step was orthogonal to each other. The one electron part of the
to write down the free energy of the system in terms of Hamiltonian is diagonalized within an ion, in this
spin variables, the magnetic moments of the four representation. These basic wave functions are denoted
sublattices. Among the. terms of the form allowed under
by
this crystal symmetry there was a term which favors
the canted spin arrangement rather than the antiferro-
f„t(r R), —g(r —R), f
magnetic one and therefore is essential to the weak where f and g mean the spin almost up and almost
ferromagnetism. This term is expressed by down, respectively. As we take account of the spin-orbit
coupling, these functions are not eigenfunctions of the
D. LS, &&S,j, (1.1) —
spin component S, but mixtures of and +
spin
where D is a constant vector. In n-FesOs, D is parallel states. The one electron Hamiltonian is written in terms
to the trigonal axis. This theory, however, is a phe- of annihilation and creation operators as follows:
nomenological one and does not clarify how this
interaction arises and how D can be calculated.
E=Q Q s„(R)Ln„t*(R)n„t(R)+n„g*(R)n„g (R) j
We erst note that the expression (1.1) is the anti-
symmetric part of the most general expression for Q Q (b„ „(R'—R)fn t~(R')rr t(R)
+BWB' n„n'
bilin ear spin-spin interaction. Only the symmetric
interactions have been familiar to us so far. In insulators + „., *(R') „,(R) j+C„„(R'—R))~„,,*(R') „,(R)
the known spin-spin interactions are isotropic super- (R' —
—n„g*(R')n„g (R)]+C„„— R)n„.t*(R')n„g (R)
exchange interaction, magnetic dipole-dipole interaction
and anisotropic superexchange interaction of pseudo- +C„.„+(R' R)n„g*(R')n„t (—
R)), (2.1)
dipolar type. Among them the magnetic dipole-dipole where
interaction can make a contribution to the antisym-
metric coupling in certain special cases because of the b ~ (R' —R)+C „'(R'—R)
orbital moments. Its magnitude, however, is generally
very small and it vanishes in the above mentioned P„ t*(r R')HgP„t (r R— —
R)—
)dr,
trigonal crystals. The anisotropic superexchange inter-
action studied so far4 is the so-called pseudodipolar b (R' —R) —C '(R'
interaction and is symmetric. However, the previous
theories do not seem to be general.
We have developed a general theory of anisotropic
P„.g*(r R')Hgg„g(r R—
)dr, 2)—
(2.
superexchange interaction with the use of the new
formalism recently developed by Anderson. ' We take
account of the spin-orbit coupling in the mechanisms C„„(R' R) = P„ t*(r —
R')HgP„g (r R)dr—
,
—
of superexchange while the previous theories do not
explicitly consider the mechanism of superexchange.
We could show that when the crystal symmetry is C„.„+(R'—R) = P„.g*(r R')H, y„t(r R)dr—
,
—
suf6ciently low, the largest term of the anisotropic
' I. Dzialoshinski, J. Phys. Chem. Solids 4, 241 (1958). = p'/2rN+
4
J. H. Van Vleck, J. phys. radium 12, 262 (1951); T, Moriya Hr V (r)
and K. Yosida, Progr. Theoret. Phys. (Kyoto) 9, 663 (1953);
T. Xagamiya, K. Yosida, and R. Kubo, Advances in Physics,
+ (A/2m'c') S.PgradV(r) g pj,
edited by N. F. Mott (Taylor and Francis, Ltd. , London, 1955), C„.„+(R'—R) = C„„*(R'—R) &ic„„"(R'—R),
Vol. 4, p. 1.
5 P. W.
Anderson, Phys. Rev. 115„2 (1959). s T, Moray@, Phys. Rev. Letters 4, 228 (1960),
 ANiso TROpiC sUPERExCHANGE INTERACTioN
n„t (R) and n„t*(R) are the annihilation and the creation g isthe gyromagnetic ratio and Ag is its deviation from
operators of the electrons in the state f„t(r R—), etc. the value for a free electron. Therefore the orders of
The most important contribution to the superexchange magnitudes of D and 1' are estimated by
comes from the conlguration mixing of the polar states
due to the transfer terms of (2.1). By the second order
D (hg/W) J, F (Ag/g)'J. (2.6)
perturbation calculation we get the usual isotropic The contribution of the next importance is the direct
superexchange from the 6' terms, the interaction of the exchange interaction. This has the same form as (2.3)
form (1.1) from the bC terms and pseudodipolar and the coeKcients of the 6rst two terms Jgg" and
interaction from the C' terms. We have for the inter- I'gg ' are given as follows:
action between the spins at R and R'
(2.7)
Eg, g &'&

.
= Jg, a &'&(S(R) .S(R'))
+D . LS(R)XS(R')]
&&

+S(R) Fg.a &'&

S(R') (2.3)
where the scalar, vector and tensor quantities: Jzz (",
Dg, g &2~, and F~a &'& are given in the case of one electron
1 . (R')
J.. (NN ~m') ~. (2.8)
per ion as follows:
~' e„(R')—e„(R') )
Jg, g = 2 ib„(R—R') i'/U,
&'&
(2.4a) The tensor coeScient I'z& ' of the last term has just
the same form as that which has been discussed by the
Dg, s, &" = (4i/U) Lb„„(R— R') C ~ (R. —
'R) previous authors4 though they regarded to include J
—C (R —R')b„„(R'—R)], (2.4b) superexchange implicitly. The exchange integral is
rgg &'&= (R — R')C„(R' —R) given by
(4/U)LC
+C (R'~
— R) C. (R —R') .
—(C„„.(R —R') C„(R'—R))1]. (2.4c)
J~g (ee'mm') = '

J J
rp„*(r~ —R) p„*(r2—R')

The expression for J» (2)

has been obtained by Ander-
X q&„(~2—R) y (r~ —R')dr&dr2, (2.9)
son. ' We assumed here that the ground state of an ion where «„(r— R), etc. , are the orthogonal orbital wave
is nondegenerate except for being a Kramers' doublet. functions obtained without the spin-orbit coupling.
e and e' represent the ground states of the ions at E, We see here again that the ratios of Dgg"" and I'gg""
and R', respectively, and U is the energy required to to Jzz '* are approximately given by (2.6). Therefore,
transfer an electron from an ion to its nearest neighbor, these terms are generally smaller than the corresponding
thus making a polar state. second order superexchange terms as was shown by
When the orbital level separation in an ion is larger Anderson in the case of isotropic superexchange.
than the spin-orbit coupling, we can calculate the The third order perturbation which include transfer
transfer integrals b and C by a perturbation method. terms twice and the intra-atomic exchange interaction
The b terms are the ordinary transfer integrals without gives the same form of coupling as (2.3). The isotropic
spin-orbit coupling and the C terms are given up to 6rst term and the anisotropic second and third terms
the first order in the spin-orbit coupling as follows: come from O'Jo, bCJO, and C'Jo terms, respectively,
Jo being the intra-atomic exchange energy. All the
1„„*(R')
C„„(R'—R) = —-i Q b (R' —R) three terms are generally Jo/U times smaller than the
2 ( m ~, (R') —e„.(R') corresponding second order terms.
We shall here show only the coe%cient of the second
1„„(R)
b„(R' —R) ~+c(R' —R), (2.5) term. The first term is given in reference 5.
e (R)-~„(R) ) Day. &'&= —(~/U')(Q„. J„(R')Lb (R — R')
with
A
X C „(R'—R) —C„„(R—R') b„„(R'—R)]
c(R' —R) = —
I y„*(r—R') +Q„J„„(R)fb „(R'—R) C„„(R—R')
4m'c' ~ —C„„(R'—R) b„„.(R —R')] }, (2.10)
X Lgrad V(r) X p] « „(r—R) dr, (2.6)
where J„(R) represents an intra-atomic exchange
where n, n' represent the ground orbital states and integral in the ion at R, etc. We have considered above
m, m' the excited sta, tes, 1 „(R) the matrix element of the most important three contributions to the super-
the orbital angular momentum of the ion at R, X the exchange interaction. As was seen, all the terms linear
spin-orbit coupling constant, and q „(r— R) the ground in the spin-orbit coupling have the form (1.1) which is
orbital wave function of the ion at R. antisymmetric for the interchange of two spins and the
From (2.5) we see that the magnitude of the C terms of second order in the spin-orbit coupling have
terms is roughly (Ag/g) times that of the b terms where the pseudodipolar form which is symmetric for the two
 TORU MORI YA

biquadratic forms of spin coupling (for S&1) and

couplings which are linear in one spin and of third
order in the other (for S& ~), in addition to the minor
corrections to the bilinear terms. A representative term
of the fourth order perturbation which contributes to
the isotropic biquadratic coupling may be of the form:

Q U-'Xb. (E—E')b„(R' —E)b„(E—R')

FIG. 1. Crystal structure of SfS
mfm
o.-Fe203. The open circles represent
0 ions and the shadowed ones
Fe'+ ions. &&b„.„(Z' —Z) P P n„.'(R)n„. .(R')n„...*(R')
&&n„.. (R)n„),*(R)n g(R')n„), *(R')n„g (R). (2.11)
When the ground ionic state can be expressed by the
numbers of electrons in the basic states, the first sum
is over the singly occupied Kramers' doublets. In
general, we should take the expectation value of (2. 11)
in the ground ionic state. The order of magnitude of
the isotropic biquadratic coupling may be (b/U)' times
spins. The orders of magnitude of these anisotropic the usual bilinear superexchange interaction. Contri-
couplings are generally given in relation to the isotropic butions to the anisotropic coupling can be obtained by
coupling by (2.6). replacing one, two or three b's in (2. 11) by C's with the
corresponding modifications of the spin sures of the
The crystal symmetry is of particular importance
annihilation and creation operators. From the order of
for the antisymmetric coupling (1.1). In crystals of
magnitude relation: C (Ag/g) b, we may estimate
high symmetry this coupling vanishes though it is
their magnitudes.
important in crystals of low symmetry, particularly in
Though we will not discuss any further these higher
relation to weak ferromagnetism. Discussion of the
order perturbations, we already see how to provide
symmetry properties will be given in the following
the mechanisms and to estimate the orders of magnitude
section.
of various forms of higher order anisotropic super-
Extension of the theory to the case of many-electron
exchange couplings.
ions is straightforward only when the ground. ionic
state has a single configuration. In this case we need CRYSTAL SYMMETRY AND THE ANTISYMMETRIC
only to express the spin operators of the electrons by SPIN COUPLING
the equivalent operators of the ionic spins with some
modifications in the coe%cients. However, in the first In the preceding section a general theory of calcu-
transition group ions the term separation is much larger lating the anisotropic superexchange interaction was
than the spin-orbit coupling and the ground states developed. In an actual crystal, some components of
taking account of the spin-orbit coupling cannot the symmetric and antisymmetric coupling tensors
generally be expressed by a single configuration. More- vanish because of the crystal symmetry. Here we
over, in crystals of low symmetry in which the anti- discuss the antisymmetric coupling (1.1) from the
symmetric coupling (1.1) is of particular importance, crystal symmetry point of view.
even the purely orbital states cannot generally be The coupling between two ions in the crystal is
represented by a single configuration. The calculation, considered first. The two ions 1 and 2 are located at
therefore, is very complicated even for the isotropic the points A and 8, respectively, and the point bisecting
coupling and we would content ourselves with the
the straight line AB is denoted by C. The following
expectation that also in this case the order estimation rules are obtained easily.
of (2.6) may be correct. 1. %'hen a center of inversion is located, at C,
Ke shall now further discuss the contribution from
the higher order perturbations qualitatively. For the D=o.
one electron ions there are no other types of interaction 2. When a mirror plane perpendicular to AB passes
between two ions 'than those given by (2.3). So the through C,
higher order perturbation calculation gives only minor
corrections to the contributions treated above. On the D ~~
mirror plane or D J AB.
other hand, for the ions with many electrons the 3. When there is a mirror plane including A and 8,
couplings of more than fourth order in the spin vari-
ables come out from higher order perturbations. For D J mirror plane.
example, from the fourth order perturbation we get 7 This was first pointed out by P. %. Anderson.
 A N I SOT ROP I C S U P E RE X C HA N GE IN I E RA C T I 0 N

4. When a two-fold rotation axis perpendicular to AB

passes through C,
D J two-fold axis. t t
5. When there is an n;fold axis (m& 2) along AB,

Fro. 2. Symmetry e1e- 4

For example, for an Fe'+ —Fe'+ pair in 0;-Fe203
ments of CuC12. 2H20
crystal.
oriented along the three-fold axis, 0 is parallel to the C

trigonal axis when the two ions are (1 or 4) and (2 or 3)

in Fig. 1 and D is zero for the other pairs. For the
rutile type iron group difluorides, D is not zero for the
pairs of corner and body-center ions. D for the nearest c== c
neighbor interaction is given by the following table.
The Positions of the Ions Direction Magnitude
~1 shown in Fig. 2. The principal axes of the crystalline
000 and 1
2) [1 10] ~D
electric field around the corner- and the body-center
000 and 1
2)
1
27
1
2 [1 10] ~D
sites are different. Only the b axis is a common principal
000 and 1
2)
1
27
1
2 [1 10] &D
000 and 1
2)
1
27
1
2 [1 10] &D axis. We denote it as the s axis. The two other principal
axes at the corner site and those at the base-center
When the spins are ordered and form sublattices we are sites are written as $&r1& and bri&, respectively. Writing
interested in the coupling between the sublattice the a- and the c axes as y and x, respectively, we have
magnetizations. In the case of iron group diQuorides, a relation:
we see from the above table that there is no anti-
(q= x cos0+y sin0, $q
—g cos0 —y sin0,
symmetric coupling of the form (1.1) between the two ——x sin0+y cos0,
sublattice magnetizations. In n-Fe203, this coupling is p& p& —x sin8+y cos0,
not zero and D for the coupling between sublattice where the angle 0 is coristant but not known so far.
magnetizations is parallel to the trigonal axis. This The five localized 3d orbital states of a Cu'+ ion are
was first shown by Dzialoshinski. ' nondegenerate in this crystalline field and have the
Symmetry consideration on the transfer integrals is following forms:
also useful in theoretical calculation of the anisotropic
exchange. This will be discussed in the following section. Pt pp Cp(3s —y )+Cp(P —yiP)
Pp pp Cp'(3s' —y')+Cp'(P —rj),
MAGNETIC ANISOTROPY IN THE ANTIFERRO-
MAGNET CuC12 2H20
6" kn (4.1)
4 ~ 'QS)
The magnetic anisotropy energy of CuC12 2H20 A" &4
which determines the spin orientation has been dis-
cussed by Yosida and the present author. ' However, where the constants Cp, C2, Cp, C2 are in principle
they neglected the antisymmetric coupling of the form determined by the crystalline potential. We shall denote
(1.1). This interaction may be of particular importance the transfer integral between the eth state of a corner
in understanding the antiferromagnetic spin arrange- ion and the mth state of its neighboring base-center
ment in this crystal as was briefly discussed in the ion by b„.
reference 4. We shall show in this section how the Then the following relations are obtained from
coeKcient D of the anisotropic exchange interaction symmetry considerations.
can be calculated in CuCl2. 2820 as a typical example b13= —b31, b23= —b3~, b45 b54 )
of the preceding section. b15= —b51, b25= —b52, b34 = —b43, (4.2)
A Cu'+ ion has only one hole in the 3d shell. The
b14 b41) b24 b42) b35 b53 ~
crystal structure is orthorhombic and two Cu'+ ions
are located at the corner and the base-center sites in a All the nonvanishing matrix elements of 1 are listed
unit cell. The symmetry elements of the crystal are below.

i—
(3 1) = i2C p,
i. — (3 l, 2)= —i2Cp', (41l, 15) = i, —
(1 /p 5) = iv3Cp i—
Cp, (2 —
l( 5) = iv3Cp' i—
Cp', —
(314 4) =i, (4.3)
(1 l„ 4) = iVBCp —iCp,
(3 l„ 5)=
 TORU MOR I YA

cells oriented in the c direction. A four sublattice

model may be reasonable. In this case, D for the
interaction between the corner and the base-center
sublattice magnetizations vanishes. However, non-
vanishing D„ is very important in determining the spin
arrangement. The spin arrangement which is accepted
at present is the ferromagnetic layers parallel to the
ab plane with the spins pointing in the +a and — a
directions alternatingly. When D„ is not zero this
arrangement cannot be stable. We propose instead the
FIG. 3. Proposed spin arrangement in antiferromagnetic spin arrangement shown in Fig. 3. The c component
CuCl2 2H20. The spins are in the ac plane. The black and the of a spin is expected to be (Ag/g) times, i.e. , about
open circles represent the ions at the positions 0, 0, 0 and —, ~, 0, ',
respectively. 10/o of, the u component. We expect that this spin
arrangement may give a more consistent understanding
Using (4.2) and (4.3) we can calculate D from (2.4b)
of the proton resonance data both above and below the
and (2.5). Though we do not know which of the five Neel temperature' and the neutron diffraction data"
states (4.1) the ground state is, the following formulas for the positions of the protons. Neutron diGraction
are valid, regardless of the ground state: experiment below the Neel temperature is highly
desirable.
(i*)
D (&) = D~(2) = b„„, VfEAK FERROMAGNETISM OF SOME
U ~ ~m
—~n b TRIGONAL CRYSTALS

(lr cosg —
D (2) —D (2)—
U m &m
l„sing)
&n
„
.
b. b. .
The following trigonal crystals are known to be weak
ferromagnets: n-Pe~03, MnCO3, CoCO3, CrF3, and
possibly FeF3. n-Fe203 has a corundum type crystal
D (2) —D (2)-P structure and MnCO3 and CoCO& are of NaNO& type.
The crystal structures of CrF3 and FeF3 have recently
where e
denotes the ground orbital state and m the been determined. " All of these crystals have the same
excited states. There is no reason why any one of the space group D3d' — R3C. The spin superstructure of
transfer integrals should vanish and therefore we see n-Fe203" gives rise to a nonvanishing antisymmetric
from (4.3) and (4.4) that the a and b components of spin coupling of the form (1.1) between the sublattice
D are nonzero regardless of which the ground state magnetizations and D is parallel to the trigonal axis as
may be. D, or D, in (4.4) consists of only one term was first shown by Dzialoshinski. ' For the remaining
when the ground state is other than 1ta and two terms crystals, the carbonates and the triQuorides, non-
when the ground state is f3. Similarly the direct vanishing D parallel to the trigonal axis exists when the
exchange contribution is calculated as follows. magnetic unit cell is just the same as the chemical unit
cell (with two molecular units) as may actually be the
(i*) ~ case.
D;*=4i) Q J(nnmn), When the spins are perpendicular to the trigonal
axis in the ordered state, a net magnetic moment is
(l(cos8 —i, sin8) „ induced by this interaction. The magnitude of this net
D ' =4iX Q —
J(nnmn),
(4.5)
moment at O'K may be estimated by

M D )Agq
(5.1)
where J(nnmn) is defined by (2.9). These expressions 1Vg1iiiS 2J (g )
are also valid regardless of what the ground state may Comparison of the value estimated by (5.1) with the
be. measured moments is shown in Table I. The agreement
Quantitative calculation of the transfer integrals b „ is generally reasonable. In CoCO&, the orbital moment
and the direct exchange integrals J(nnmn) is necessary
to get D. However, we may expect that 6 „and N. J. Poulis and G. E. G. Hardeman, Physica 18, 201 (1952).
J. Itoh, R. Kusaka, Y. Yamagata, R. Kiriyama, and H.
J(nnmn) are of the same order of magnitude as b Ibamoto, Physica 19, 415 (1953).
and J(nnnn), respectively, because of the low symmetry "S. W. Peterson and H. A. Levy, J. Chem. Phys. 26, 220
at the positions of Cu'+. The magnitude of D may be (1957).
"For CrF3 see K. H. Jack and R. Maitland, Proc. Chem. Soc.
about (hg/g) times the isotropic superexchange inter- (London) 232 (1957); K. Knox has also obtained the same result
action. (private communication). For FeF3 see M. A. Hepworth, Acta.
Cryst. 10, 345 (1957).
In the antiferromagnetic CuC12. 2H20, a magnetic ~ C. G. Shull, W. A. Strauser, and E. O. Wollan, Phys. Rev.
unit cell is considered to consist of two chemical unit 83, 333 (1951).
 AN ISOTROPI C SUPE REXCHANGE INTERACTION
may be of particular importance. Though the g values TABIK I. Comparison between fractional magnetic moments
and (hg/g) in various weak ferromagnets.
are not available. we may naturally expect a compara-
tively large D and therefore a large moment, in quali-
tative agreement with the experiment. In Ot-Fe203,
Substance (W&gPI 5') (Ag/g)

CrF3, and FeF3y we have neutron diGraction data"" a-Feg03 1.4X10 ' 1X10 '
MnCO3 2 6X10 ' 1X10 '
which show that the spins are perpendicular to the CoCO3 2~6X10 '
L111j axis below the Neel temperature. For MnCOs CrF3 1X10 ~ ~1X10~
neutron data" suggest the spin directions are between FeF3 1X10 3

the 111) axis and the (111) plane while a magnetic

I

susceptibility measurement" seems to show that the

spins are almost perpendicular to the 111j axis.
Various magnetic properties of Q.-Fe203 have been
I
"
the increase of X& near T~. This behavior seems to be
observed in CrF3. For MnCO3 there is some sign of
"
discussed by Dzialoshinski' from phenomenological this behavior but it is not yet clear. Similar behavior
has been expected theoretically'~ and was observed in
grounds with the essential use of an interaction of the
form (1.1) with D as a parameter. We have now
"
NiF2. This behavior of the paramagnetic susceptibility
provided the mechanism for this interaction, thus giving may be one of the distinctive features of weak ferro-
a firmer ground to his theory. We shall here remark magnets.
one property of these weak ferromagnets which has not Our next discussion is on the possible other mecha-
been discussed. The paramagnetic susceptibility of nisms of weak ferromagnetism. Interactions between
these weak ferromagnets shows a very sharp increase two spins have been fully discussed in 2 and the higher
near the Neel temperature. A simple model to show order interactions are more than two orders of magni-
this may be the following Hamiltonian: tude smaller than the bilinear term discussed above.
We shall consider the anisotropy energy of single spin
H=J P (S;,S;)+8 + S, s nature which comes from the spin-orbit coupling under
j
(i. i the crystalline electric field. This interaction does not
.
&

+D Q (S;Q;„—
S;„S, )+gp~H Q S;, (5.2) depend on the superexchange mechanism and therefore
is more important in substances with low Neel temper-
atures. In NiF2 this kind of interaction is responsible
where summations are taken over nearest neighboring for its weak ferromagnetism. "
In the trigonal crystals
pairs. We assume that 8 is positive, i.e., the easy which we are discussing now the fourth order terms of
direction is perpendicular to the s axis. By a molecular trigonal symmetry are the lowest order terms which
field approximation we can easily show that Xii (s may give rise to a net magnetization of the crystals.
component) obeys the Curie-Weiss law fairly well This is possible because there are essentially two kinds
down to TN while X& is given by

Ng'pnsS(S+1) T
3k (T+ Tsr) T
Tp-
T~— (5 3)
of positions for the magnetic ions and the crystalline
fields around them are di6erent. They are written as

"
(1)
E Q S;,S;„(S;„'—3S;,')
with
(2)
—Q S;,Sy„(S;„' 3S;,'), (5. 6)—
(5.4)
JZS(S+1) where the erst and the second summations are taken
~0
over one and the other kinds of ions, respectively. This

where Z is the number of nearest neighbors. As (T~

is small and positive, X& increases very sharply near T&.
Tp)— interaction is e6ective in producing weak ferromagnet-
ism only when the equilibrium spin orientation is
somewhere between the trigonal axis and the plane
A measure of the sharpness may be given by perpendicular to it. This is possible when the total
eGect of the second order and the other fourth order
1pDi'
— -2I ) M anisotropy energies and the dipolar and pseudodipolar
(TN Tp)/TN -I (5 5)
2 J) interactions favor it. In Q.-Fe203, CrF3, FeF3 this does
L
I

(NgpgS)
I

not seem to be the case. ""

Moreover, the estimated
The smaller the net moment below TN the sharper is values of E
for O.-Fe203 and MnCO3 are more than one
"E. O. Wollan, H. R. Child, W. C. Koehler, and M. K. Will- "W. N. Hansen and M. Griiiel, J. Chem. Phys. 30, 913 (1959).
T. Moriya, Phys. Rev. 117, 635 (1960).
sinson, Phys. Rev. 112, 1132 (1958).
"S.
' J. Pickart, Bull. Am. Phys. Soc. 5, 59 (1960). "
17
R. G. Shulman, NMR measurement in NiF2 (to be published);
A. S. Borovik-Romanov, J. Exptl. Theoret. Phys. (U. S.S.R. ) A. H. Cooke, xq of NiF2 (private communication); J. C. Burgiel,
36, 766 (1959) /translation: —
Soviet Phys. JETP 36 (9), 539 V. Jaccarino, and A. L. Schawlow, x of powdered NiF2 and
(1959)j. Ni(IO&)s 2H&O LBull. Am. Phys. Soc. 4, 424 (1959) j.
 T OR U MORI YA

order of magnitude smaller than D. We may fairly APPENDIX

safely conclude that this interaction is not of primary We shall show here an alternative derivation of the
importance in the weak ferromagnetism of O. -Fe20~, interaction of the form (1.1). We start from the localized
MnCO3, CrF3, and FeF3. The sixth order anisotropy electron orbital states and consider the spin-orbit
energy is probably negligibly small as compared with D. coupling as a perturbation. The unperturbed states of
the ions 1 and 2 are specified by (n, m, ) and
CONCLUDING REMARKS (e', m', . ), respectively, n, and n' being their ground
states. The second order perturbation energy bilinear
It shouM be emphasized here that now a firm ground with respect to the spin-orbit coupling and the exchange
has been given to the mechanism of weak ferromagnet- interaction is given as follows:
ism first proposed by Dzialoshinski. It is no longer a
phenomenological model but is a real effect established ((n! Xlr Sr! m)2J(mn'e'n)Sr Sz
theoretically. Though there may be some impurity
&n &m
effects (chemical or structural), this mechanism always
exists as an intrinsic property of the crystal. 2J(nn'e'm) (Sr Sz)(m! Xlr S&!n) ~

At present we have many weak ferromagnets and we

&n &m
know two types of interactions which can be the origin
of weak ferromagnetism. One is a coupling between the
( (n'! Xl, S z! m) 2 J( m' nnn') (S z S,)
spins such as the antisymmetric part of the anisotropic +&
m'
I

&n' &m'
superexchange interaction, and the other is the single
spin anisotropy energy. In o.-Fe20&, etc. , the first 2J(n' nnm)(S Srz) (m'! Xlz S !zn') q
mechanism and in NiF~ the second one are the origins (A 1)
of their weak ferromagnetism. There may be substances &n' &m'

in which both of these types of interactions are im-

where J's are the exchange integrals given by (2.9) in
portant. Generally speaking, there is a tendency that
the text. When the orbital state is nondegenerate, the
when the Neel temperature is high the first type of
matrix elements of 1 are purely imaginary and we can
interaction is more important and when the Neel rewrite (A. 1) as follows:
temperature is low, the second mechanism is more
important. This is because the first is approximately J (nn'e'm)
proportional to the exchange interaction while the Er, z 2X Q (e!1&!m) . [Sr, (S&. S&)]
second is independent of the exchange interaction.
When the single spin anisotropy energy is comparable J(ne'm'n)
+2K P (n'!Iz!m') [Sz, (Sr S,)]
with the exchange energy, as may be possible for some m ~nr —~ml
substances with low Neel temperature, or when (Ag/g)
is not small, the magnetic moment may not necessarily J(nn'e'm) (n! 1, ! m)
be small. This may be the case in ludlamite where
=2z I 2
E m &n &m
T~=20'K and the magnetic moment is 0.8 p~ per ion
of divalent iron. " J (ee'm'e) (n'! lz! m') ) ' X S,
The antisymmetric spin coupling treated here is m' &n' &m' )I [Sr ]. (A. 2)
important riot only as an origin of weak ferromagnetism
but also in determining the spin arrangement in This is equivalent to (2.8). If we replace J(nn'm'n)
antiferromagnets as was illustrated in the case of by the effective superexchange integral: b„b„„/U,
CuC12 2H20. etc. , we get the expression equivalent to (2.4b) and
(2 5).
ACKNOWLEDGMENT Though this derivation is more elementary, the one
given in the text is superior for the systematic study
The writer is very much indebted to Dr. P. W. of the anisotropic superexchange interaction.
Anderson for his continuous interest and kind advice Note added in proof. It has been shown by Professor —
on this work. K. W. H. Stevens [Revs. Modern Phys. 25, 166 (1953) )
' R. M. Bozorth and V. Kramer, Collogue International de
that there can exist anisotropic terms of the exchange
zzzagzzetzszae, Grezzoble, 1959 PJ. phys. radium 20, 393 (1959)g. interaction which are linear in the spin-orbit coupling.