See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/243430725

Theoretical Study of Electronic, Magnetic, and Structural Properties of α-Fe2O3

(Hematite)

Article  in  Physical Review B · March 1995

DOI: 10.1103/PhysRevB.51.7441

CITATIONS READS
203 426

3 authors, including:

Roberto Dovesi
Università degli Studi di Torino
381 PUBLICATIONS   15,363 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Statistics and benchmarking View project

REFeO3 thin film synthesis and characterization View project

All content following this page was uploaded by Roberto Dovesi on 26 November 2014.

The user has requested enhancement of the downloaded file.

PHYSICAL REVIEW B VOLUME 51, NUMBER 12 15 MARCH 1995-II

Theoretical study of electronic, magnetic, and structural properties of a-Fe~O, (hematite)

M. Catti and G. Valerio

Dipartimento di Chimica Fisica ed Elettrochimica, Universita di Milano, via Golgi 19, 20133 M&lano, Italy

R. Dovesi
Dipartimento di Chimica Inorganica, Fisica e dei Materiali, Universita di Torino, via Giuria 5, 10125 Torino, Italy
(Received 3 November 1994)
Antiferromagnetic rhombohedral a-Fe&03 has been studied by calculations of the ground-state spin-
polarized wave function and total energy, using the ab initio periodic unrestricted Hartree-Fock ap-
proach. All-electron basis sets of contracted Gaussian-type functions are employed to represent the 0
and Fe atoms (18 and 27 orbitals, respectively); Fe is alternatively described by a large-core pseudopoten-
tial plus 18 valence-shell orbitals. Computations have been performed for both the antiferromagnetic
(AF) and ferromagnetic (FM) structures; the correct relative stability is reproduced, with
hE(AF-FM) = —0. 0027 hartree per formula unit (including a correction for correlation energy). The
dependence of AE(AF-FM) on variations of the Fe-0 bond lengths and Fe-0-Fe' angles involved in su-
perexchange is analyzed, finding that for some configurations the FM structure becomes more stable
than the AF one. The athermal equation of state, equilibrium crystal structure, elastic bulk modulus,
and binding energy have been computed and compared to experimental quantities. An analysis of the
density of electronic states show that the band gap is of p-d rather than d-d type, confirming the charge-
transfer-insulator nature of hematite as inferred from photoelectron spectra. The overall shape of the
valence band is also fully consistent with spectroscopic results. Mulliken electron population data indi-
cate a charge back transfer of 0.29 ~e~ from 0 to the d shell of Fe'+, causing a partial spin pairing with
a magnetic moment of 4. 7p&.

INTRODUCTION C

The erst-row transition-metal oxides are a class of ma-

terials with outstanding electronic and magnetic proper-
ties, which have been the object of intensive experimental
and theoretical research for a long time. ' Most work has
concerned the compounds with 1:1 stoichiometry and
NaCl-type crystal structure. In particular, these have
been very recently investigated by ab initio periodic
unrestricted-Hartree-Fock (UHF) methods, ' which have
proved to be suitable theoretical techniques for crystal-
line systems containing open-shell ions. Other calcula-
tions, based on the density-functional approach in con-
junction with linearized augmented plane-waves (DFT-
LAPW) schemes, have been reported. ' Because of the
high symmetry of the rocksalt structure, these oxides are
relatively simple to study, but do not give information on
more subtle questions such as the inAuence of bond an-
gles on superexchange and magnetic properties. For this
reason, and for the interest of considering oxides of tran-
sition metals in different oxidation states, the case of
rhombohedral a-Fe203 (hematite) seems to be particular-
ly attractive. Iron sesquioxide is an important antiferro-
magnetic material, and also occurs as a natural product
of surface alteration of many iron-containing rock-
forming minerals. Its study is framed into a theoretical FIG. 1. View of the crystal structure of antiferromagnetic a-
investigation of corundum-type and spinel-type oxides, Fe203, with outline of the hexagonal unit cell. Black, gray, and
the first part of which concerned a alumina and magnesi- open circles denote Fe1 (spin-up), Fe2 (spin-down), and 0
um aluminate. atoms, respectively. The angular superexchange interactions in-
Hematite is isostructural with a-A1203 (space group volving a particular Fel atoms, two neighboring oxygens, and
R 3c, two and six formula units in the primitive rhom- the related Fe2 atoms are emphasized.

0163-1829/95/51(12)/7441(10)/$06. 00 1995 The American Physical Society

 M. CATTI, G. VALERIQ, AND R. DOVESI 51

bohedral, and in the conventional hexagonal unit cells, interaction (CI) cluster calculations. ' Some details of the
respectively); its structure was refined from x-ray interpretation of the density of electron energy states are
di6'raction data at ambient conditions and also at high sti11 not completely understood, however. Other simpler
pressure. ' Hexagonal close-packed (001) layers of 0 cluster-type quantum-mechanical
' ' studies of the MS-Xa
atoms are present, with Fe atoms 611ing — ', of the octahe- type have been performed, but no periodic calcula-
dral holes in between (Fig. 1). The stable phase turns out tions have appeared yet, to our knowledge.
to be antiferromagnetic (AF) below the Neel temperature The aim of the present study is to apply the UHF com-
T&=953 K. In the corresponding magnetic structure putational approach to the e-Fez03 crystalline system,
(space group R 3 ), alternating (001) layers of spin-up which should be related to MnO (Ref. 3), because Fe + is
(Fel) and spin-down (Fe2) iron atoms are observed. 9 Out a d ion isoelectronic with Mn +. On the other hand, its
of the four Fe atoms in the primitive cel1, two are in- more complex crystal structure is expected to introduce
dependent (Fel and Fe2), and the other ones are cen- important di6'erences both in the magnetic and in the
trosymmetrically related to them. Two di6'erent electronic behavior with respect to MnO. Further, the
modifications of this antiferromagnetic structure are ob- structural changes of hematite with pressure and its
served: below the Morin temperature (TM =250 K) the athermal equation of state are investigated here, in order
Fe electron spins are aligned with the [001] hexagonal to compare them with the results found for isostructural
axis, while above TM, they lie on the (001) planes. ' A corundum.
weak ferromagnetic moment is present at TM & T & T&, CGMPUTATIONAI,
due to a slight canting of the spins. " A hypothetical fer- Ab initio periodic I.CAO (linear combination of atomic
romagnetic (FM) structure would correspond to space orbitals) Hartree-Fock methods, as implemented in the
group R 3c with four Fe1-type (spin-up) symmetry-related cRYSTAL92 code, ' were used, adding the supplementary
iron atoms in the primitive cell. option for unrestricted treatment of the spin-dependent
The electronic structure of a-Fe2O3 has been thorough- part of the wave function in open-shell systems. ' The
ly investigated recently by photoelectron spectrosco- latter feature was essential to obtain spin-polarized eigen-
py' ' and techniques, ' and functions of the Fock Hamiltonian, taking into account
by inverse-photoemission
the results have been interpreted as due to a charge- the ls 2s 2p 3s 3p 3d electronic configuration of isolat-
transfer insulator on the basis of con6guration- ed Fe + ions. The radial factors of atomic orbitals are

TABLE I. Exponents (bohr ) and contraction coefticients of the (individually normalized) Gaussian
functions adopted for iron in the present study. y[+z] stands for y X 10+—'. Durand-Barthelat pseudo-
potential ' (PP) and all-electron (AE) basis sets are reported.
Fe (PP) Fe (AE)
Shell CoeKcients Coe ancients
type Exponents Exponents
3. 154[+ 5] 2.270[ — 4]
4. 569[+ 4] 1.900[ —3]
9.677[+ 3] 1.110[—2]
2.521[+3] 5.010[ — 2]
7.597[+ 2] 1.705[ —1]
2.630[+2] 3.692[ —1]
1.028[+ 2] 4.033 [ —1]
4.297[+ 1] 1.434[ —1]
7.983[+2] —5.200[ —3] 8.500[ — 3]
1.912[+2] —6.800[ —2] 6.080[ —2]
6.369[+ 1] —1.314[ —1] 2. 114[ —1]
2.536[+ 1] 2.517[ —1] 3.944[ —1]
1.073[+ 1] 6.433[ —1] 3.980[ —1 ]
3.764 2. 825[ —1] 2.251[ —1]
4.814[+ 1] 1.220[ —2] —2. 150[ —2]
1.746[+ 1] —2.278[ —1] —8.500[ —2]
6.997 —8.801[ —1] 2.010[ —1 ]
3.079 9.755[ —1] 1.302
1.486 3.365[ —1] 1.0 1.314 1.0 1.0
1.132 —5.327[ —1] 1.0
Ssp 4. 500[ —1] 1.0 1.0 5. 532[ —1] 1.0
38 4.081[ + 1] 2.574[ —2] 3.048[+ 1] 5.830[ —2]
1.219[+ 1] 1.431[ —1] 8.692 2.591[ —1]
4. 126 3.535[ —1] 3.101 5. 162[ —1]
1.526 4. 846[ —1] 1.171 5.656[ —1]
5. 117[—1] 1.0 4.298[ —1] 1.000
51 THEORETICAL STUDY OF ELECTRONIC, MAGNETIC, AND. . .

expressed as linear combinations of Gaussian-type func- plexity of syrnrnetry lowering in the AF case. We believe
tions of the electron-nucleus distance. The 0 atom was that the features of the equilibrium structure and its
represented on an all-electron basis by 18 atomic orbitals changes with pressure should not be affected by the type
(Gaussian basis set of type 8-411G'), according to the of magnetic configuration, owing to the small energy
scheme previously used for a-A1203; the Gaussian ex- difference between the FM and AF cases. On the other
ponents and coefficients for the inner shells 1s, 2sp, and hand, computations were performed with both the AE
3sp are therein reported. Iron was dealt with at an all- and PP basis sets.
electron (AE) level, according to a 8-6-411G' contraction The unit-cell volume was set to ten values ranging from
with two d-type (41G) shells. However, owing to limita- —10 to +8% of the experimental result, and the total
tions of the available computational resources, for some energy was calculated keeping the c/a ratio and the z(Fe)
calculations it was necessary to devise a less expensive, and x(O) fractional coordinates fixed at their measured
slightly less satisfactory scheme. This was done by em- values. These three structural variables were relaxed for
ploying the Durand-Barthelat large-core pseudopoten- some of the points ( — 8, — 4, 0, +4, + g%%uo) with the PP
tial, ' supplemented by two sp-type (21G) and two d-type basis set, and their least-energy values turned out to vary
(41G) shells, to represent iron. The corresponding basis linearly with the volume change. Therefore, with the AE
set, keeping the all-electron representation for oxygen, basis set the relaxation was carried out for the — 8 and
will be denoted as PP (pseudopotential), while the AE +4% points only, interpolating linearly for the other
symbol is employed for the all-electron description of values. In both cases, relaxation energy corrections vs
both the Fe and 0 atoms. Use of the PP scheme will be volume were derived, finally obtaining numerical curves
made in the study of the inAuence of structural changes for E( V), c/a ( V), zF, ( V), and xo( V). The whole pro-
on crystal properties. Further, assessing the overall per- cedure is described in more detail in previous work on
formance of the PP with respect to AE basis set is useful, MgCO3 and n-A1203, which have the same space
because in future studies of more complicated crystal group. ' The E( V) curve was interpolated by the Mur-
structures, possibly containing second- or third-row tran- naghan equation of state, ' containing the zero-pressure
sition elements, use of the less expensive PP scheme could energy (Eo), volume ( Vo ), elastic bulk modulus (Ko ), and
be a necessary choice. its pressure derivative (ICO) as fitting parameters. By
At first, the parameters of the AE and PP basis sets diA'erentiation, the pressure p= dE/dV=—p(V) is ob-
were optimized by minimizing the energy of the isolated tained. Thus, all the above dependences of structural
Fe + ion. Then the exponents of the 5sp and 4d shells of variables on volume were transformed into pressure
0
Fe and those of the 4sp and 3d shells of were refined by dependences. The minimum of the Murnaghan curve
searching for the minimum total crystal energy of a- E(V) gives the equilibrium (zero-pressure) values of all
Fez03 at the experimental structural configuration. The such variables, which are reported in Table II and com-
optimized parameters of both basis sets of Fe are report- pared with the corresponding experimental values. The
ed in Table I; the 4sp and 3d exponents obtained for oxy- latter ones are taken from x-ray measurements at room
gen are 0.210 and 0.35 bohr (AE), and 0.200 and 0.35 temperature and could not be extrapolated to T=0 K
bohr (PP), respectively. Apart from the presence of a d because of lacking thermal expansion data in the litera-
orbital shell, and small differences for the two outer shells ture.
(3' and4') exponents, the 0 basis set is analogous to In all cases, a better agreement between calculated and
that used in studies of the rocksalt-type transition-metal measured data is observed for the AE with respect to PP
oxides. ' Also, the AE basis of iron interpolates well be- results. This proves that the larger computational effort
tween those of Mn and Ni given in the same studies. of the AE basis is definitely worthwhile, in order to simu-
The level of numerical approximation in evaluating the late structural properties. As usual, with Hartree-Fock
Coulomb and exchange series appearing in the self- results for systems including fourth-row or heavier
consistent-field (SCF) equations for periodic systems is atoms, the least-energy unit-cell edges and volume are
controlled by five tolerances. ' These are related to esti- overestimated; this effect is larger for the a than c edge,
mates of overlap or penetration for integrals of Gaussian resulting in a slight decrease of the c/a ratio with respect
functions on different centers, which define cutoff limits to its experimental value. The volume error (+3.6%) is
for series summation. The values used in the present cal- significantly worse than for the isostructural a-AlzO3
culations are 10, 10, 10, 10, and 10 ', which containing the light Al atoms ( — 0. 2% ), but surprisingly
correspond to severe computational conditions. The re- performs better than for MnO (+5.4%). The most
ciprocal space was sampled according to a regular sublat- significant interatomic distances and angles (Fig. 1) are
tice defined by six points in the irreducible Bri11ouin zone. also reported in Table II, and will be particularly useful
below for the discussion of magnetic properties. Here we
need only remark that in the computed structure, the
EQUILIBRIUM AND HIGH-PRESSURE STRUCTURE
shortest Fe-O distance lengthens, but the longest does
not, so as to reduce their difference with respect to exper-
Although the AF configuration of hematite is the imental values.
stable one at low temperature, as confirmed also by this The other parameters of the fitting to the Murnaghan
study (cf. the section on magnetism below), all calcula- equation of state also appear in Table II, except for the
tions concerning structural effects were carried out in the minimum energy Eo (Table III). The experimental value
FM configuration, just to avoid the computational com- of the bulk elastic modulus Ko therein reported comes
 M. CATTI, G. VALERIO, AND R. DOVESI 51

TABLE II. Calculated (least-eneryyi and experimental equilibrium values (see the text for refer-
ences) of hexagonal unit-cell edges
0
(A), volume per formula unit (A ), atomic fractional coordinates,
selected interatomic distances (A) and angles (deg), elastic bulk modulus (CsPa), and its pressure deriva-
tive for o;-Fe203. Linear compressibilities (TPa ) are reported, too. Percentage errors are indicated;
PP and AE mean results obtained with pseudopotential and all-electron basis sets, respectively.
PP exp.

ao 5. 144(+ 2. 2% ) S.112(+1.5%) 5.035

co 13.893(+ 1. 1% ) 13.820(+0. 5%) 13.747
co /ao 2.7O1( —1. 1%) 2.703( —1.0% ) 2.730
Vo 53.o6(+ s. 5% ) 52. 13(+3. 6% ) 50.30

x(O) 0.3003 0.3020 0.3056

z(Fe) 0.3539 0.3541 0.3553

Fe-0 2.005 1.988 1.945 X 3

2. 114 2. 110 2. 113X 3
Fe1-Fe2 2.887 2.877 2.895
3.444 3.421 3.361 X 3
3.766 3.744 3.701X 3
Fe1-0-Fe2 86. 1 86.0 86.5
118.4 118.8 119.6
132.2 132.0 131.6X2

Ko 243(+6%) 261(+ 13% ) 230

Ko 3.3( —6% ) 3. 1( —11%) 3.5

d (a /ao)/dp —0.97 —0.90 —1.27

d (c/co)/dp —1.37 —1.29 —1.61
d (cao /aco ) /dp —0.41 —0.40 —0.34

from a room-temperature x-ray diffraction (synchrotron 1. 04—

radiation) static compression measurement up to 50
GPa, and should be, thus, underestimated by about
—3% with respect to the value extrapolated to the
athermal limit. Therefore, the real error of computed 1. 00—
data should be accordingly reduced, so as to be similar,
for the AE case, to the figure usually observed also for
lighter substances. A smaller error is observed with the 0. 96—
PP basis set, which is, however, due to the larger comput-
ed unit-cell volume, producing a reduced elastic stiffness,
with respect to the AE case. It should also be noticed
that a more accurate K0 experimental value would be 0. 92—
given by knowledge of all elastic constants, but unfor-

TABLE III. Total energies per formula unit (hartree) for o.-
Fe203 ( Eo ) and for individual atoms, and binding energy Ez of
n-Fe203. Hartree-Fock results and HF + electron correlation
0. 84—
corrections are reported for the AE basis set. h% is the per- I

centage error with respect to the experimental binding energy of 10 20 30 40 50

0.9250 hartree.
p (G»)
HF HF + corr
E(Fe) —1262.3679 —1263.5895
E(O) —74.8012 —75.0672
Eo(a-Fe203) —2749.7835 —2753. 1915 FICs. 2. Calculated dependence of the unit-cell constants of
—0.6441 —0.8109 a-Fe&03 divided by their equilibrium values on pressure. Trian-
—30.4 12.3 gles, squares, and circles indicate the a/ao, c/co and V/Vo ra-
tios, respectively.
51 THEORETICAL STUDY OF ELECTRONIC, MAGNETIC, AND. . .

tunately the C, 3 component was not measured; the set respect to results relative to systems with the same struc-
of elastic components necessary to calculate the bulk tures but containing lighter atoms.
modulus is incomplete. Values of Eo obtained in older
static compression studies ' compare well with those re- MAGNETIC PROPERTIES
ported in Table II.
The curves of V/Vo, a/ao, and c/co versus pressure The energy difference between the antiferromagnetic
and the ferromagnetic configurations, hE(AF-FM), was
are shown in Fig. 2. Least-squares linear approximations
of the p dependences of a/ao, c/co, and (c/a)/(co/ao) computed for the experimental crystal structure of hema-
tite (Table IV). Hartree-Pock results uncorrected and
give the slopes reported in Table II, and are therein com-
corrected for the correlation energy are reported. The
pared to values obtained from linear fits of the data of
relative stabilities of the two magnetic structures are
Ref. 8. The behavior of c/a is particularly interesting,
reproduced rightly, and this should be considered as an
because its pressure coefficient turns out to be three times
important achievement of the method of calculation, be-
larger than that of a-alumina. This reproduces closely
cause of the small energy value involved. Use of the
what is observed experimentally in a much smaller pres-
pseudopotential rather than all-electron scheme to
sure range, and shows that hematite is much more
represent the Fe atom seems to overestimate somehow
compressible along the threefold axis than the isostruc-
the b, E(AF-FM) energy. Performing the correction for
tural a-Alz03. The computed compressibilities seem to
correlation energy enhances the stability of the antiferro-
be slightly smaller than the measured ones, for which the
magnetic structure significantly, with respect to the bare
estimated error should be of the order of 10%. Hartree-Fock result. Qn the other hand, most of the sta-
bilization effect of the AF configuration is related to the
BINDING ENERGY superexchange mechanism of antiferromagnetism, ' ac-
cording to which the net spin moments of Fel and Fe2
The results of calculations performed to obtain the atoms are coupled in an antiparallel way through overlap
binding energy of a-Fe203 are summarized in Table III. of the corresponding electron distributions with that of
Atomic energies of isolated Fe and 0 atoms were com- the in-between 0 atom. Further, such a mechanism is
puted by using the same all-electron basis sets used in the most effective when the Fe1-0-Fe2 bond angle ap-
crystal, but supplemented by two more diffuse shells, the proaches 180', and its energy contribution decreases as
exponents of which were optimized. In the case of Fe, this angle shrinks. ' In hematite, the 0 atom is sur-
0. 127 and 0.044 bohr were obtained, and also the ex- rounded by four Fe atoms, two of Fel and two of Fe2
ponents of the 5sp and 4d shells were reoptimized to type (Fig. 1). Out of the six Fe-0-Fe' angles, four are of
0.552 and 0.392 bohr, respectively. The binding ener- Fel-0-Fe2 type. Three of these are wider, 120 and
gy E~ is obtained as the difference between the total crys- 132' ( X 2), with Fe 1-Fe2 distances of 3.361 and
tal energy per formula unit Eo (with optimized structural 3. 701 (X2) A, respectively. The fourth one (86. 5') is
geometry) and the energies of the constituent isolated much smaller, and thus, though corresponding to the
atoms. The experimental value of E& was derived by an shortest Fel-Fe2 contact (2. 900 A, parallel to the three-
appropriate Born-Haber thermochemical cycle, accord- fold c axis), should give little contribution, if any, to su-
ing to the relationship perexchange interactions.
Therefore, from the point of view of superexchange,
Eii(expt) =X, [H, , +(H, —H, )j the structure of hematite is much more complicated than
those usually considered (typically, the rocksalt case),
where metal-oxygen-metal angles are either 90' or 180;
+gH298 (H298 Ho) Eovib
f thus this appears ideally suited to a deeper investigation
The sum is extended to all chemical elements in the for- of the structural features of superexchange. This has
been done by computing the value of b, E(AF-FM) for
mula unit. The formation enthalpy AHf"' of a-Fe203
strained unit-cell geometries, with the PP basis set in or-
sublimation (Fe) and dissociation (Oz) enthalpies H, ;,
der to contain the cost of calculations. Either the edge a
and heating enthalpies H; — H, for all three species or c was changed, keeping all other parameters fixed at
were taken from Ref. 26. The zero-point vibrational en-
their experimental equilibrium values; the energy
ergy of hematite E„.b was estimated by the Debye model,
using an isotropic approximation for the mean acoustic
wave velocity derived from the bulk modulus, which gave
a Debye temperature of 911 K. A value of 0.9250 hartree
was, thus, obtained for the experimental binding energy
of a-Pe&03. TABLE IV. Energy difference per formula unit AE(AM-FM)
(hartree) between the antiferromagnetic and ferromagnetic
As usual, the Hartree-Fock binding energy turns out to
configurations of hematite, computed at the experimental crys-
be underestimated by about 30%. The a posteriori
tal structure, for the PP and AE basis sets.
correction of total energies based on an estimate of the
correlation energy obtained by density-functional HF HF + corr
theory leads to a considerable improvement. The error —0.0027 —0.0034
obtained ( —12%) is comparable to that found for MnO
pp
AE —0.0019 —0.0027
and NiO (Ref. 19), but appears to be slightly larger with
7446 M. CATTI, G. VALERIO, AND R. DOVESI 51

TABLE V. Energy difference per formula unit (hartree) between antiferromagnetic and ferromagnet-
ic configurations of hematite, computed by changing either the0 a or c unit-cell edge or the c/a ratio (at
constant volume), and corresponding interatomic distances (A) and angles (deg) relevant for superex-
change. The PP basis set is used.
HF HF + corr Fe-0 Fe1-0-Fe2
5.034 —0.0027 —0.0034 1.944 2. 114 131.6 119.7 86.5
5.334 —0.0008 —0.0011 2.038 2. 181 130.9 119.0 83.2
5.634 +0.0002 +0.0002 2. 134 2.251 130.3 118.3 80. 1
6.034 +0.0008 +0.0011 2.262 2.346 129.6 117.6 76.3

HF HF + corr F@-0 Fe1-0-Fe2

13.75 —0.0027 —0.0034 1.944 2. 114 131.6 119.7 86.5
15.75 —0.0016 —0.0021 2.000 2.263 133.4 121.5 94.3
18.75 —0.0006 —0.0009 2.095 2.504 136.0 124.4 104. 1
20.75 —0.0000 +0.0001 2. 165 2.673 137.8 126.4 109.7

c/a HF HF + corr Fe-0 Fe1-0-Fe2

3.502 —0.0042 —0.0054 1.895 2.220 135.0 123.3 100.6
3.085 —0.0036 —0.0045 1.914 2. 157 133.2 121.3 93.4
2.731 —0.0027 —0.0034 1.944 2. 114 131.6 119.7 86.5
2.430 —0.0017 —0.0021 1.981 2.087 130.3 118.3 79.8
2. 172 —0.0010 —0.0011 2.024 2.075 129.2 117.2 73.6
1.948 —0.0003 —0.0002 2.072 2.075 128.3 116.3 67.7

difference results, together with interatomic distances and larger than the positive effect due to the decrease of the
angles involved in superexchange, are reported in Table longer distance.
V. Of course expanding the unit cell tends to destabilize By considering the significant contribution of the
the antiferromagnetic with respect to ferromagnetic spin correlation correction to the energy of antiferromagnetic
configuration, thus finally leading to a change of sign of stabilization (Table IV), it turns out that exchange (in-
AE(AF-FM). However, the effect is much stronger for cluding superexchange) effects are not exclusively re-
the a than for the c expansion; to achieve the stability in- sponsible for the type of magnetic ordering observed in
version requires that either of the two edges be this compound; but, the electron correlation plays a not
lengthened by 10 or 50/o, respectively. negligible role as well.
This can be understood mainly by considering that an
expansion of the a edge causes the Fel-0-Fe2 angles ELECTRONIC PROPERTIES
relevant for superexchange to decrease, while the oppo-
site occurs for the c lengthening. Thus, in the latter case %'e refer in this section to electronic properties of
the expansion of Fe-0 bond lengths, leading to a weaker hematite computed (AE basis set) for the antiferromag-
superexchange, is contrasted by the widening of bond an- netic experimental crystal-structure geometry. In this
gles, which favors superexchange. It is then clear that a case, by definition, Fe l atoms have a net excess of e
larger expansion is necessary in the c than in the a case, (spin-up) or p (spin-down) electrons, and Fe2 atoms have
in order to achieve the same effect on hE(AF-FM). A a and p populations exactly exchanged with respect to
second reason is also related to the shorter (1.94 A) Fe-0 0
Fel; the u and p electron populations of atoms are per-
bonds having directions closer to the (001) plane, and fectly balanced. Results concerning the ferromagnetic
longer (2. 11 A) Fe-0 bonds closer to the [001] axis. structure would show only minor differences, referring
Therefore, the expansion of a produces a comparatively mainly to a slight unbalance of the spin-up —spin-down
larger increase of the shorter bond with respect to the electrons on oxygen.
same expansion of c, yielding a stronger effect on In Table VI, the a+p Mulliken population data per
bE(AF-FM). This analysis is confirmed by considering orbital shell of Fe and O atoms is reported, together with
the effect of changing the c/a ratio at constant unit-cell net charges and total a-P values. The corresponding re-
volume on the AF stabilization energy (Table V). The sults obtained in previous work on isoelectronic MnO
main effect is the angular distortion, which, as c/a de- are also shown for comparison; this is made possible by
creases, narrows all Fel-0-Fe2 angles weaking the su- the use of electron basis sets designed very similarly in
perexchange and destabilizing the AF with respect to the the two oxides, as discussed above. The Fe net charge of
FM configuration. The average Fe-0 distance remains 2. 62~e~ appears to be slightly larger than that of 2. 30~e~
approximately constant, but the increase of the shorter obtained for Al in o'-A12O3. Also the bond overlap pop-
bond length has again a negative effect on superexchange ulations (0.012 against 0. 098~e~ for Fe-0 and Al-O, re-
51 THEORETICAL STUDY OF ELECTRONIC, MAGNETIC, AND. . .

TABLE VI. Mulliken population data of electron-charge distribution in atomic-orbital shells (a+P)
and net atomic spins (a — 0
P) for Fe and in a-Fez03 (AF, this work) and Mn and 0
in MnO (AF„Ref.
3), with all-electron basis sets.
a-Fe&03 Mno
Fe 0 Mn 0
1s 2.000 1.997 2.000 1.997
2$p 8.085 2.678 8.082 2.640
3$p 2. 348 2.540 2. 164 2.626
4sp 4. 315 2.494 4.291 2.601
5$p 1.340 1.486
3d 4.356 0.036 4.502
4d 0.939 0.611
d (a+P) 5.295 5. 113
Total (a+P) 23.382 9.745 23. 136 9.864

z {net charge) + 2.618 —1.745 +0.864 —1.864

d (a — p) 4.726 0.0
Total (a — P) 4.741 0.0 4.924 0.0

exp. magnetic 4.9' 4. 79, 4. 58

moment (p~ )
'See Ref. 30.
bSee Ref. 3.

spectively) support a slightly more ionic character of of 2p(O) nonbonding type with a minor
gy, is essentially
bonding in e-Fe203 with respect to e-A1203. Most of the presence of 3d~(Fel) [or 3d (Fe2)] states due to charge
charge back transfer from 0 to Fe + goes into the d transfer from 0 to Fe + (cf. the discussion of electron
orbitals (0. 30~e~ ), giving a population of 5.01 spin-up and populations above). A conduction band of 3d~(Fe1) char-
0.29 spin-down electrons. This effect of partial spin pair- acter, with a small contribution from 2p(O) levels giving
ing in the d shell, due to charge transfer, is stronger than it some Fe-0 antibonding nature, is finally observed
for isoelectronic Mn + in MnO; in that case 0. 11~e~ are
transferred, so that 5. 02a and 0. 10@ electrons populate
the d orbitals. The computed spin magnetic moment of
4. 74pz compares favorably with the experimental re- total
sult of 4. 9p~. Here a slight underestimate of measured
data is observed, while the deviation has the opposite sign
for the magnetic moment of MnO. By considering the
detailed atomic-shell populations, the outer anionic and s, p(O)
cationic shells appear to be more populated in MnO than
in a-Fe203, showing a slight shrinkage of the electron
cloud in hematite related to the higher nuclear charge of
Fe + with respect to Mn +. It can be noticed that the Fe2
Ms-Xa calculations on the Fe06 cluster showed a much
more covalent Fe-0 bonding, with Fe net charges of 1.30
(Ref. 17) and 1. 83~e~ (Ref. 16), and a larger charge
transfer from 0 to Fe + (0. 75 e~, Ref. 16), with Fe1
respect to the present results for periodic, crystalline cx-
Fe203.
In Fig. 3 the density of electronic states (DOS) of anti-
ferromagnetic a-Fe203 is shown. By considering the pro- -.75 -.25 .25

jections over atomic contributions therein outlined, we EF

can interpret the sequence of peaks appearing from low
to high energy as follows. The first one is a 2s(O) non- FIG. 3. DOS (density of electronic states, energy in hartrees)
bonding band, followed by a narrow band of mainly of antiferromagnetic a-Fe203 relative to spin-up electrons, with
3d (Fe1) [or 3d~(Fe2)] character, with some admixture projections onto atomic-orbital contributions. The spin-down
with 2p(O) states and thus a limited Fe-O bonding nature. diagram would exchange the d(Fe1) and d(Fe2) projections,
The upper valence band, lying just below the Fermi ener- leaving the rest unvaried.
 M. CATTI, G. VAI. ERIO, AND R. DOVESI

above EI;. Some important features should be stressed. DOS, and thus to be responsible of the (small) covalency
First, the band gap is of p-d instead of d-d type: thus o.- observed in a-Fez03.
Fez03 is shown by Hartree-Fock calculations to be a The electron-charge difference between crystal and ion-
charge transfer rather than Mott-Hubbard insulator, just ic superposition is shown for the plane (010) in Fig. 4(a).
as was the case for MnO and NiO. This confirms the re-
sults of experimental photoemission spectroscopy on
hematite, interpreted on the basis of configuration-
interaction calculations on the FeO6 cluster. ' The pho-
toelectron spectrum of o.'-Fe2O3 shows a three-peak struc- I
J
I

I
J
J
ture extending 18 eV below the Fermi level, with maxima
r

at approximately — 3. 1, —6. 2, and —12. 9 eV. ' The first

J
) a
f

/ ( A, /
J

two peaks are assigned to 3d I. photoemission final /

/ (

states, which result from L (0 ligand)~d (Fe + metal)

P
1

charge transfer transforming the original metal d holes /

J

(d states) into ligand p holes. The region in the lower

energy range ( — 9 to —18 eV), including the third peak,
I

final states (d holes). Since 3d "L in-

/
is ascribed to 3d r (
stead of 3d" ' final states are located near the Fermi lev-
l I

(
J
el in the photoemission spectra, ferric oxide is classified /

as a charge transfer rather than Mott-Hubbard insula- /

I

tor. ' There is a surprising agreement between this I

I /
scheme and the Hartree-Fock density of states shown in
Fig. 3, concerning not only qualitatively the relative posi-
tions of the 3d(Fe ) and 2p(O) valence bands, but also
the quantitative energy values of density maxima. This
would indicate that in the present case the Koopman's 1

theorem holds successfully, so that effects of relaxation of / 'I

the electron configuration after the ionization process I

would play a minor role, at least to a first approximation.

Further, contrary to arguments in favor of a 4s(Fe + ) na-
ture of the lower conduction band, ' inverse- I I

photoemission spectra of hematite' have shown that the (b) I

J I r
empty states just above Ez have a 3d(Fe + ) character,
thus fully confirming the DOS of Fig. 3 also above the ! r
/
i
Fermi level. J
J
J'
/
I
Coming now to details, and comparing with the DOS's
g
I
(
/ I /
of rocksalt type MnO and NiO, we observe that (i) the I( +
3d (Fe + ) band is very narrow (less than 3 eV) with t2
I 1
JI

and e states practically superposed, so as to give a nearly /

vanishing crystal-field splitting, and (ii) the 3d (Fe +)
/
and 2p(O ) bands are separated by an energy gap. On
t
the other hand, the DOS's of MnO and NiO show a wide (
J

3d band (over 8 eV) overlapping largely with the 2p(O) I

/
/'
one, and including separated peaks contributed by t 2g / r
J
and eg states. These features would support a more local-
//
r- —-i ~
ized character of the metal d states in hematite than in /
'II,
+
I

MnO and NiO, probably because of the different charac- g(, ~ J/

J
Jp /
ters of the corundum and rocksalt structures: in the l

former, the metal coordination octahedra are much less (

I

condensed in space than in the latter. Further, although

all three oxides are charge-transfer insulators, a-Fe2O3 / I
S
shows this character more strongly, while MnO and NiO I
l /

seem a bit more inclined towards the Mott-Hubbard I

I
1

regime. It can be finally remarked that the charge

1

I
I

transfer from 2p(O ) to 3d~(Fe +) states, estimated as FIG. 4. Difference (crystal minus ionic superposition)
0. 29~e~ by the Mulliken population results, lies between charge-density maps on the (010) hexagonal plane for a-Fe203
the values of 0.21 calculated by the CI cluster model' (a) and a-A1203 (b, from data of Ref. 6). Three (dots) and two 0
and that of 0. 4~e~ suggested by a Mossbauer study of Fe (stars) or Al (crosses)0 —
atoms are shown. Isodensity curves are
3
hyperfine fields in Fe2 Rh 03. Such a charge transfer separated by 0. 04e A . Continuous, dashed-dotted, and
is proved to give rise to a partial Fe-O o. bonding charac- dashed lines indicate positive, zero, and negative values, respec-
ter of the upper valence band by the above analysis of the tively.
51 THEORETICAL STUDY OF ELECTRONIC, MAGNETIC, AND . . ~ 7449

No metallic bonding density appears between adjacent Fe work on hematite; in our opinion, the interpretations
atoms along the [001] direction, contrary to what was re- given to these experimental results should be taken cau-
ported in some x-ray diffraction studies. ' Such a tiously, because the same authors claim to have observed
feature is not reproduced even by previous theoretical a similar metal-metal bond also in a-A12O3, and this is
in contrast with other careful experimental studies
L /
and also with physicochemical intuition. Thus, the Al-Al
bonding in corundum and Fe-Fe in hematite may be ar-
/'-
tifacts related to an inadequate account of extinction
/'
effects and/or absolute scale factor in the diffraction stud-
ies. An interesting aspect should be noticed in Fig. 4(a):
J the two lobes of highly concentrated charge difference
around each Fe + ion, pointing towards its two neighbor-
I
/ /Ilgwu
I (~(Q ing oxygens. These peaks account for the charge transfer
(f]
2p(O )~3d~(Fe +), and their orientation corresponds
~ ////
g////
rsvp/~// to the small Fe-0 o-type bonding due to d(Fe)/p(O) band
superposition as apparent from the DOS results (Fig. 3).
I

( In Fig. 4(b), the corresponding picture of electron-density

~)
~ '5
I

1 7
difference for a-A1203 (Ref. 6) is shown, and the spherical
j distribution surrounding Al + ions contrasts sharply with
J
I
t /
that of Fe +. Figure 5(a) and 5(b) show the spin densities
S
(a-P) of hematite for the antiferromagnetic and fer-
/'
( romagnetic structures, respectively. In the AF case, the
/
/
symmetrical exchange of a and P 3d-type electrons in the
/ Fel and Fe2 atoms appears clearly. In the FM structure,
on the other hand, Fel and Fe2 have the same spins but a
net spin polarization density is induced on the oxygen
atoms, corresponding to a population of 0. 11 a-P elec-
I

trons derived from the Mulliken analysis.

CONCLUSIONS
The electronic, magnetic, and structural behavior of
corundumlike rhombohedral a-Fe203 has been simulated
correctly by ab initio periodic unrestricted Hartree-Fock
calculations. This includes its nature of charge-transfer
insulator due to a p-d band gap, and the overall features
/ of the valence band in the density of states, compared to
/ photoelectron spectra. A large charge transfer from 0
I
I to the 3d metal ion states, leading to a smaller magnetic
( momentum, is observed with respect to the similar case
of MnO. By comparing the relative energies of the anti-
)
ferromagnetic and ferromagnetic structures, the former
turns out to be more stable, consistently with experiment.
Further, it has been possible to study in detail the effects
0 of structural deformations on the relative stabilities of the
r
two magnetic structures, throwing light onto the impor-
tance of bond angles in superexchange, and finding in
some instances a stabilization of the ferromagnetic
configuration. Finally, the evolution of the crystal struc-
ture with pressure has been determined up to nearly 50
GPa, deriving the compressibilities of unit-cell edges and
g /
of bond lengths. In particular, the c/a ratio proves to
/ decrease with pressure much more rapidly than calculat-
/ ed previously in the case of isostructural a-Alz03.

ACKNOWLEDGMENTS
W'e thank V. R. Saunders and M. D. Towler for pro-
FIG. 5. Difference spin-density (spin-up spin-down) minus viding the basis set of the isolated Fe + ion. The work
maps on the (010) hexagonal plane for antiferromagnetic (a) and was supported by the Human Capital and Mobility Pro-
ferromagnetic
0 —
(b) a-Fe203. Isodensity curves are separated by gramme of the European Union under Contract No.
3
0. 20e A . Continuous, dashed-dotted, and dashed lines indi- CHRX-CT93-0155, and by the Ministero Universita e
cate positive, zero, and negative values, respectively. Ricerca Scientifica e Tecnologica, Roma.
 M. CATTI, G. VALERIG, AND R. DOVESI

P. A. Cox, The Transition Metal Oxides (Oxford University Initio Molecular Orbital Theory (Wiley, New York, 1986).
Press, Oxford, 1992). 2 P. Durand and J. C. Barthelat, Theor. Chim. Acta 38, 283
W. C. Mackrodt, N. M. Harrison, V. R. Saunders, N. L. Allan, (1975).
M. D. Towler, E. Apra, and R. Dovesi, Philos. Mag. A 68, 2M. Catti, A. Pavese, R. Dovesi, and V. R. Saunders, Phys.
653 (1993). Rev. B 47, 9189 (1993).
M. D. Towler, N. L. Allan, N. H. Harrison, V. R. Saunders, W. 2~F. D. Murnaghan, Proc. Natl. Acad. Sci. U. S.A. 30, 244
C. Mackrodt, and E. Apra, Phys. Rev. B 50, 5041 (1994). (1944).
4P. I. Sorantin and K. Schwarz, Inorg. Chem. 31, 567 (1992). J. Staun Olsen, C. S. G. Cousins, L. Gerward, H. Jhans, and
5P. Dufek, K. Schwarz, and P. Blaha, Phys. Rev. B 48, 12672 B. J. Sheldon, Phys. Scr. 43, 327 (1991)~

(1993). R. C. Liebermann and S. K. Banerjee, J. Geophys. Res. 76,

M. Catti, G. Valerio, R. Dovesi, and M. Causa, Phys. Rev. B 2735 (1971).
49, 14 179 (1994). 26Handbook of Chemistry and Physics, edited by R. C. Weast
7Y. Sato and S. Akimoto, J. Appl. Phys. 50, 5285 (1979). (Chemical Rubber, Boca Raton, 1987) ~

L. W. Finger and R. M. Hazen, J. Appl. Phys. 51, 5362 (1980). J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M.
C. N. R. Rao and J. Gopalakrishnan, New Directions in Solid R. Pederson, D. J. Singh, and C. Fiolhais, Phys. Rev. B 46,
State Chemistry (Cambridge University Press, Cambridge, 6671(1992).
England, 1986). L. J. de Jongh and R. Block, Physica B+C 798, 568 (1975).
IOT. G. Worlton, R. B. Bennion, and R. M. Brugger, Phys. Lett. 2~J. H. van Vleck, J. Chem. Phys. 9, 85 (1941).
24A, 653 (1967). OE. Kren, P. Szabo, and G. Konczos, Phys. Lett. 19, 103 (1965).
I A. Tasaki and S. Iida, J. ~IJ. Zaanen, G. A. Sawatzky, and J. W. Allen, Phys. Rev. Lett.
Phys. Soc. Jpn. 18, 1148 (1963).
A. Fujimori, M. Saeki, N. Kimizuka, M. Taniguchi, and S. 55, 418 (1985).
S g, Phy . R . B 34, 7318 (1986). J. M. D. Coeyand G. A. Sawatzky, J. Phys. C 4, 2386 (1971).
R. J. Lad and V. E. Henrich, Phys. Rev. B 39, 13 478 (1989). V. G. Tsirel'son, M. Yu. Antipin, V. A. Strel'tsov, R. P.
I4G. Draeger, W. Czolbe, and J. A. Leiro, Phys. Rev. B 45, 8283 Qzerov, and Yu. T. Struchkov, Dokl. Akad. Nauk SSSR 298,
(1992). 1137 (1988) [Sov. Phys. Dokl. 33, 89 (1988)].
F. Ciccacci, L. Braicovich, E. Puppin, and E. Vescovo, Phys. ~E. N. Maslen, V. A. Strel'tsov, N. R. Streltsova, and N. Ish-
Rev. B 44, 10444 (1991). izawa, Acta Crystallogr. Sec. B 50, 435 (1994)~

S. Nagel, J. Phys. Chem. Solids 46, 905 (1985). ~5E. N. Maslen, V. A. Streltsov, N. R. Streltsova, N. Ishizawa,
D. M. Sherman, Phys. Chem. Miner. 12, 161 (1985). and Y. Satow, Acta Crystallogr. Sec. B 49, 973 (1993).
R. Dovesi, V. R. Saunders, and C. Roetti, cRYsTAL92. User's A. S. Brown, M. A. Spackman, and R. J. Hill, Acta Crystal-
Manual, Gruppo di Chimica Teorica (Universita di Torino, logr. Sec. A 49, 513 (1993).
Torino, 1992). A. Kirfel and E. Eichhorn, Acta Crystallogr. Sec. A 46, 271
I E.
Apra, Ph. D. thesis, University of Torino, 1993. (1990).
W. H. Hehre, L. Random, P. R. Schleyer, and J. A. Pople, Ab

View publication stats