Journal of Alloys and Compounds 442 (2007) 231–234

Y3Fe5O12 yttrium iron garnet and lost magnetic moment

(computing of spin density)
A. Bouguerra a , G. Fillion a , E.K. Hlil b , P. Wolfers b,∗
a Laboratoire Louis Néel-CNRS, Grenoble, France
b Laboratoire de Cristallographie, CNRS, Grenoble, France
Received 15 July 2006; received in revised form 13 October 2006; accepted 16 October 2006
Available online 6 February 2007

Abstract
The yttrium iron garnet (YIG) has been intensively studied in the past as a convenient ferrimagnetic insulator for resonance devices and magnetic
memories. The YIG was considered as the model of ionic compound with the ion Fe3+ and 5 ␮B on each of two iron sites with a resulting
magnetization of 5 ␮B . Contrarily, all neutron diffraction studies lead to a localized moment close to 4 ␮B . The present work is a tentative to
understand these noticeable discrepancies by using the electronic structure computing.
© 2007 Elsevier B.V. All rights reserved.

Keywords: Oxide materials; Electronic band structure; Spin density

1. Introduction 2. Electronic structure calculation details

The yttrium iron garnet compound (YIG) was intensively
studied in the past as a convenient ferromagnetic compound for
magnetic memories (magnetic cores or bubbles). Traditionally,
the tetrahedral and octahedral sites of the iron atoms were known
as the state Fe3+ and typical ionic oxide. All the performed mag-
netic measurements seem to be coherent with a pure ionic model
with 5 ␮B of magnetic moment by iron atoms in the two sites
at low temperature [1,2]. Unfortunately, all neutron diffraction
studies (powder or single crystal) lead to a magnetic moment
close to 4 ␮B at low temperature [3,4]. Our aim in the present
paper is to understand this apparent discrepancy between the
magnetization measurements and the neutron diffraction experi-
ments. The basic idea was to find an unlocalized spin distribution
of spin to compensate the lowered localized magnetic moment
of 4 ␮B (affected to each iron atom) and to check if this spin
distribution was in agreement with the neutron diffraction data.
As the polarized neutron diffraction experiments are not easy
due to the large secondary extinction [3,4], we have undertaken
to determine the spin density in the YIG unit cell by electronic
structure computing.
∗ Corresponding author. Tel.: +33 76 88 74 05; fax: +33 76 88 10 38.
E-mail address: pierre.wolfers@grenoble.cnrs.fr (P. Wolfers).
We used the full linearized potential augmented plane wave
(FLAPW) method [5,6] which performs DFT calculations
using the general gradient approximation (GGA) where the
Kohn–Sham equation and energy functional are evaluated
self-consistently. For this method, the space is divided into the
interstitial region and the non-overlapping muffin tin spheres
centered on the atomic sites. The employed basis function inside
each atomic sphere is a linear expansion of the radial solution of
a spherically potential multiplied by spherical harmonics. In the
interstitial region, the wave function is taken as an expansion
of plane waves and no shape approximation for the potential
is introduced in this region which is consistent with the full
potential method convenient to non-metallic compounds. The
core electrons are described by atomic wave functions which
are solved fully relativistically using the current spherical
function base and the valence electrons are treated with the spin
polarized potential in our case.
The crystallographic structure is defined in the space group
Ia-3d (#230) with four independent atoms: yttrium, oxygen and
the two iron atoms successively at 24c, 96h, 16a (octahedral)
and 24d (tetrahedral) sites. The YIG material is considered to
be in the collinear magnetic state where the magnetic moments
of octahedral and tetrahedral sites are opposite.

0925-8388/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.jallcom.2006.10.178
232 A. Bouguerra et al. / Journal of Alloys and Compounds 442 (2007) 231–234

Fig. 1. Total (a) and partials YIG density of states (for spin up and down) showing the not electric conductor character of YIG ((b) Feo , (c) Fet , (d) O, (e) Y). The
Fermi energy is located for energy = 0.

3. Electronic structure calculation results of Y3 Fe5 O12
Take into account the electrical insulator character of the YIG
as confirmed by the total density of state (Fig. 1), we expect that
the spin density is localized in some well defined place in the
crystallographic structure not previously fixed by a particular
model of wave function basis. The spin density is deduced from
the difference between the spin up and spin down electronic
density by using the FLAPW package that build 3D maps after
convergence of the self-consistent process.
Table 1 presents the resulting magnetic moments as deduced
from the differences between the spin up and spin down popu-
lations of electron in each atomic sphere and in the interstitial
region. These results clearly reveal the existence of a noticeable
spin density out of iron atomic. The contribution of the intersti-
tial spin density must be just as expected because the integration
of (spin) moments on the whole of unit cell is very close to the
low temperature experimental value of 5 ␮B /f.u. [1].
The 3D spin density (in the primitive unit cell) is built and
displayed by using the Xcrysden program [7] (Figs. 2 and 3)
 A. Bouguerra et al. / Journal of Alloys and Compounds 442 (2007) 231–234 233

Table 1
Computed and measured magnetic moments

Sites RASA (Å) FLAPW (␮B ) Neutron diffraction (␮B )

4K 300 K (sphere) [2] 300 K (cube) [2]

Y 1.11 0.007 – – –
O 0.77 0.078 – – –
Feo 1.03 −3.308 −4.53 −3.78 −3.75
Fet 1.03 3.146 4.21 3.62 3.70
Interstitial region (␮B /f.u.) 1.219 – – –
Magnetization (␮B /f.u.) 4.998 3.57 3.30 3.60

In this table Feo and Fet denote, respectively, the octahedral and tetrahedral iron sites. The DFT deduced magnetic moments in the atomic spheres are given
in FLAPW column. Our typical neutron powder diffraction study at 4 K (ILL/D1B instrument) are displayed in the 4 K column. The 300 K values are taken
from reference [2] were the magnetic moment integration was a sphere or a cube. For this data the corresponding observed magnetization is 3.63 ␮B (at
300 K).

and our program map to diag which allows creating simulated

neutron diffraction patterns by fast Fourier transformation with
addition of nuclear contributions from a traditional structure fac-
tor computing. This last way enables us to evaluate the effect of
the found unlocalized spin density on the neutron diffraction
pattern and also gives some keys on a future experimental deter-
mination by neutron diffraction on a small single crystal to avoid
a too large secondary extinction.
By the examination of the 3D density map display
(Figs. 2 and 3) the following results can be deduced: the spin
density takes the form of the dual polyhedral of the coordination
ones with rounded edges and vertex (Fig. 2). The octahe-
dral iron atom presents a spin distribution in form of rounded
cube and the tetrahedral iron a rounded tetrahedral distribution.
These spin distributions can be explained as the 3d electrons
avoid the oxygen atoms. The oxygen spin distribution is very
anisotropic and presents a net magnetic quadrupolar feature
(Fig. 3).

Fig. 3. Perspective view displaying the spin density in the plane Feo –O–Fet . We
can see form of spin density on the oxygen atoms that exhibits a quadrupolar
spin distribution with the positive (in red) and negative (in blue) parts (yellow
arrows). (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of the article.)

4. Discussion

Our resulting density of state (Fig. 1) shows the electric iso-

Fig. 2. Perspective view showing the unlocalized spin density on Fe sites. The
dual coordination polyhedron form of the spin densities on the octahedral (dual
cube) and tetrahedral (dual tetrahedron) is clearly evidenced.
lator character of the YIG. The 3d character of the band just
above the narrow gap displayed by the total DOS is the sum
of contribution of the empty 3d band as well displayed by the
DOS of two iron sites (Fig. 1b and c). The d electrons cannot
give significant conductivity owing to their high effective mass.
So, the narrow gap (∼0.25 eV wide) does not flag a traditional
semiconductor.
A second point is the weakness of iron magnetic moments.
This feature can be enforced by the anisotropic distribution of
the spin around the iron atom, especially for the tetrahedral site.
This anisotropic spin distribution should be take into account
by an anisotropic form factor when we perform the magnetic
234 A. Bouguerra et al. / Journal of Alloys and Compounds 442 (2007) 231–234
structure fitting from neutron diffraction data, and this effect
can be augmented by the particular oxygen spin distribution.
We know that the used method is not ideal because a com-
mon feature of the DFT computing is to exaggerate the effect of
coulomb interaction despite the effect of exchange-correlation
term. The main effect in the present study is to do lower the
localized magnetic moment. A possibility is to use the LDA + U
processing that can be better for a magnetic isolating compound
but, with the spin-orbit coupling (in the same energy range) this
kind of computing would be heaviest in terms of time computing
owing to the large number of atoms in the garnet structure (80
atoms in the primitive unit cell).
For the present time, we consider the actual results as glob-
ally correct except for the localized moments that are decreased
at the benefice of the spin density in the interstitial region. Per-
haps, a new accurate study with a large number of diffraction
lines (>2000) would allow to get a very sharp experimental spin
density where small details could be compared to the DFT com-
puting spin density map. This study would allow to evaluate the
possible accuracy of the DFT spin density.
References
[1] S. Geller, H.J. Williams, R.C. Sherwood, J.P. Remeika, G.P. Espinosa, Phys.
Rev. 131 (1963) 1080.
[2] L. Néel, R. Pauthenet, B. Dreyfus, in: C.J. Gorter (Ed.), Progress in Low
Temperature Physics, vol. IV, Interscience Publish. Inc., NY, 1964, p. 344.
[3] M. Bonnet, A. Delapalme, F. Tchéou, H.S. Fuess, Proceedings of the Interna-
tional Congress on Magnetism, Moscow, 1973, vol. IV, 1974, pp. 251–256.
[4] M. Bonnet, A. Delapalme, H. Fuess, M. Thomas, Acta Crystallogr. B31
(1975) 2233.
[5] P. Blaha, K. Schwarz, P. Sorantin, S.B. Trickey, Comput. Phys. Commun.
59 (1990) 399.
[6] P. Blaha, K. Schwarz, J. Luitz, WIEN97, Vienna University of Technology,
1997 (improved and updated Unix version of ref. [5]).
[7] A. Kokalj, Comput. Mater. Sci. 28 (2003) 155–168.