Jour

nal
Name

Finite size effects in local properties of magnetic

nanoparticles†
Jacques R. Eone,∗a Olivier M. Bengone,a‡ and Christine Goyhenexa

Using first principle calculations, the local electronic structure is characterized in cobalt magnetic
nanoparticles containing from 13 to 405 atoms (or 1-2nm of diameter). Within both a shell and
a local fine analyses, these calculations enabled to unravel the effects of local structural environ-
ment including atomic coordination, structural deformations, finite size effects and shape change
of the nanoparticles. In particular, in icosahedral nanoparticles, the evolution of the local spin
magnetic moment in the nanoparticles is found to decrease in the particles from one shell to an-
other, going from the surface to the core. Charge transfers between shells are obeying a general
rule of variation where donors sites are identified to be exclusively the subsurface atoms and more
unexpected vertex sites from the surface. The variations of the magnetic moment, however, are
dictated by the deformations in the nanoparticles coupled to complex hybridization effects, prop-
agating from the surface, giving rise to singular characteristics like a strong decrease of the spin
moment in the core of the nanoparticles and particular electronic distribution near the surface.
Larger nanoparticles, in the form of a truncated octahedron, show a more uniform distribution
of the local magnetic moment in the core but keep the memory of some characteristics of the
lower sized icosahedral particles like the low intersite charhe transfer or the reversal from anti-
ferromagnetic coupling to ferromagnetic of the electronic sp states. These specificities should be
related to further properties not only magnetism but also reactivity following the conditions of use
of magnetic nanoparticles.

1 Introduction of a transition metal element is indeed drastically sensitive to

The research on metallic nanoparticles is in constant progress
both experimentally and theoretically 1–4 . Thanks to this contin-
uous increase of knowledge, many potential applications have
been envisaged or even realized in various fields including catal-
ysis, magnetism and optics. In the case of computer materials
science, a common aim is still to improve the predictability of the
calculations and to extend their transferability toward realistic
physico-chemical environments of the considered nanoparticles.
The variety of physico-chemical properties of nanoparticles
are indeed strongly related to finite size effects which makes
strongly varying their local atomic environment as a function
of size. The main involved characteristics are the presence of
inequivalent sites and the variation of their number in terms
of atomic coordination, the atomic bonds deformations and the
shape of the particles. These statements take all their sense
when considering magnetic nanoparticles. The magnetic state
a a detailed study at the electronic structure level 3,15 . Conversely,
Institut de Physique et Chimie des Matériaux de Strasbourg, Université de Strasbourg,
CNRS UMR 7504, 23 rue du Lœss, BP 43, F-67034 Strasbourg, Cedex 2, France; E-mail:
christine.goyhenex@ipcms.unistra.fr
small variations of the details in local electronic structure, and
therefore to changes in the close atomic environment.
Advances in electronic structure descriptions together with the
increasing computer accessible power has enabled the upscaling
of the standard ab initio codes, mainly implementing the Density
Functional Theory (DFT), toward extended sizes of systems 3,5–8 .
Most interestingly, the latter fact has opened the route to new
possibilities to explore or to revisit the electronic structure, and
the properties related to it, of increasingly larger nanoparticles.
Magnetic clusters in a size range going from a few atoms to a few
tens, for the most numerous, were intensively investigated 9–14 .
As magnetic nanoparticles, in the range size of one hundred
to a few hundreds, have been much less studied within DFT
calculations, one still can expect to highlight some specificities in
the properties at a scale closer to experimental systems, relatively
to the very small clusters. In addition, these few studies are
more focused on total energy calculations and scarcely propose
the studies at higher sizes could even generalize and extend to
higher size some results obtained at the lower sizes. One main

J
our
nal
Name,
[yea
r][
,vol
.
], 1–9 | 1
aim of this article is to address this question of scaling magnetic
properties in nanoparticles.
In this context, we will undertake the case of cobalt systems
which are archetypal examples of systems in which magnetism
plays a major role, both from a fundamental point of view and
for their wide range of technical applications. Nanoparticles of
magnetic metals receive indeed much attention because of their
applications in fields like recording media, magneto-optics, catal-
ysis and biomedical potential applications. Rationalizing the link
between the structure and the local properties, is of prime impor-
tance to enable the design of nanoparticles with targeted proper-
ties.
In the case of magnetic systems, one delicate point is to un-
ravel the effects of the local atomic environment by attributing
the change in local properties to charge transfer and/or LDOS
deformation the latter being mainly due to atomic coordination
variations and finally to relate these changes to local distance de-
formations which are enhanced in nanoparticles relatively to ex-
tended surfaces and large particles. We will consider fully relaxed
magnetic particles but include also the results of local properties
obtained with the initial configurations in order to be able to re-
late the results to specific distance deformations.
2 Methods for the electronic structure
2.1 The SIESTA method
SIESTA 6 is a widely used method to perform self-consistent Kohn-
Sham DFT electronic structure calculations on molecules and
solids. Its efficiency is tightly related to the use of localized
basis sets. It can be used with a wide variety of exchange-
correlation functionals (including GGA-PBE used in the present
work). Practical implementation of DFT calculations with SIESTA
requires some particular choices including exchange correlation
functional, basis set, and pseudopotentials which are used to
describe the effects of the core (i.e. non-valence) electrons of
an atom and its nucleus. In our case, the latter are norm-
conserving, improved Troullier-Martins pseudopotentials 16 with
core corrections, factorized in the Kleinman-Bylander form 17 in-
cluding scalar-relativistic effects. Valence states are described us-
ing double-ζ polarized (DZP) basis sets. For cobalt, a DZP basis
was used with two different radial functions to represent the 4s
and 3d orbitals and one polarization 4p shell. The reference cal-
culations are performed on surface systems whithin the supercell
scheme using a slab of about twelve layers, each plan containing
four atoms. The calculations with nanoparticles are performed
like for isolated molecules, i.e. a cluster of atoms is placed in a
vacuum of sufficiently large spans in all directions in order to ef-
fectively isolate it from its images introduced by the application of
the periodic conditions inherent to DFT procedure of calculation.
In the present work, a cubic supercell of fixed dimensions 10 nm
was used which resulted in at least 10 nm of spacing between the
periodic cluster images in each direction. Structural optimization
was performed within the conjugate gradient method.
The local charge determination is performed within a Mulliken
analysis 18 Since a localized basis is used it is possible to simply
2| J
our
nal
Name,
[yea
r][
,vol
.
],
1–9
project the charge on the basis functions associated with given
atom and obtain the orbital charge directly. In practice, the cor-
responding matrix elements of density and overlap matrices are
multiplied to form the population matrix in such way, that the
electronic density from overlap matrix is partitioned equally be-
tween atoms. Is worth noticing that the accuracy of the obtained
Mulliken populations is very sensitive to the choice and optimiza-
tion of the basis.However we have previously shown it to be ac-
curate enough for metallic (Co,Pt) based systems, further Bader
analysis giving the same results in this case 19 .
2.2 The tight-binding linear muffin-tin orbital (TB-LMTO)
approach
For the sake of comparison, we also performed TB-LMTO calcula-
tions to benchmark the Siesta results for the surfaces. The tight-
binding linear muffin-tin orbital method 20 , within the atomic
sphere approximation (ASA) has been used. TB-LMTO-ASA re-
sults have been obtained using a surface Green’s function tech-
nique. 21 This approach allows to simulate on an ab initio level
the open boundaries of a surface with a true semi-infinite stack-
ing of layers, from the substrate to the vacuum. This is realized
by using a combination of delocalized basis functions in recipro-
cal space within the two-dimensional plane of the surface (xy),
and tight-binding real space basis functions in the direction per-
pendicular to the surface.
The electronic structure was determined self-consistently
within the local spin density approximation, with Vosko-Wilk-
Nusair parameterization for the exchange-correlation poten-
tial. 20
To model the surfaces, we used ten layers of metals (Co or
Fe), with four layers of empty spheres to simulate the vacuum on
top of the metal surfaces. for the TB-LMTO method we have not
considered structural relaxation.
The TB-LMTO-ASA method is particularly accurate and effi-
cient for the closed-packed structures considered here.
3 Results: Reference study of close-packed
extended surfaces of cobalt
As we will focus more on icosahedral nanoparticles presenting
facets of (111) orientation only, we start, in this reference study,
by considering the extended close packed Co(111) surface. We
think it important to perform such preliminary work in the aim
to ensure the correctness of the analyses, as well of the local
magnetic moments as of the charge analysis, knowing that these
quantities are very sensitive to the method of calculation and the
one for analyzing the charge. We have used the SIESTA-DFT and
the TB-LMTO to extract the charge and spin magnetic moment
per site. The results for local magnetic moments and of total
charge per site are presented in Figure 1 where S-N (N=0,1,...)
denote the atoms of the plans starting from the surface one
(N=0).
Both unrelaxed and relaxed slabs are considered with SIESTA,
while TB-LMTO is performed on unrelaxed slabs only. The
LMTO results are consistent with previous studies on surfaces 22 .
Then all the results are consistent with each other whatever FCC - Co(111)

the method which gives confidence in the analyses performed 1.75

SIESTA (with relaxation)

here and their transferability to the studies of nanoparticles.

/atom)
1.70 SIESTA (no relaxation)

In summary, the local spin magnetic moment is (as expected) TB-LMTO

B
1.65

m (m
strongly enhanced at the surface layer, due to the lowering of
Bulk
the coordination at the surface. The presence of the surface still 1.60

affects the subsurface plane where the local magnetic moment

9.04
even if it decreases, is still higher than in the next planes where
9.02
the bulk values is rapidly reached, anyway, already from the third

Ne
9.00
plan (S-2). The relaxation (SIESTA-DFT results) affect slightly Bulk

8.98
the results. The same considerations can be inferred from the
8.96
reported values of total charge per site. For unrelaxed structures
1 2 3 4 5
the charge variation does not exceed ±0.01 e and is limited to th
N layer
the two first planes from the surface. It corresponds therefore to
a net charge transfer of 0.01 e from the subsurface atoms to the Fig. 1 Co(111) surface. Upper panel: magnetic moment per site; lower
surface ones. The relaxation tends to increase this charge transfer panel: total charge per site. The layers (abscissa) are numbered from
from 0.01 e/atom to 0.02 e/atom so that the contraction between bulk (N=1) to surface (N=5)
these two plans S and S − 1 favors charge transfer. However,
upon relaxation, the local magnetic moment is decreased both
at the surface (by 0.02 µB ) and at the subsurface (by 0.01 µB )
so that the correlation between interatomic charge transfer and 4.8

magnetic moment variation is not trivial. A closer analysis to up 4.7

Nd (e/atom)
and down population separately shows that the effects are also 4.6

limited to the two first plans from the surface. The relaxations at 4.5

the surface do not change the trends in the spin up - spin down 2.9
distribution. Additionally one can see an anti-ferromagnetic
2.8
coupling between sp and the total distribution which is found in
both used methods and which is preserved at the surface. It is
0.90
worth noticing that, despite the very good agreement on the total 0.85
Nsp (e/atom)

distribution, a noticeable difference is visible between SIESTA 0.80

and LMTO regarding the electronic distribution between sp up 0.75

and down. It however does not affect the general conclusions in 0.70

0.65
particular in the case of the anti-ferromagnetic coupling between
0.60
sp and the total distribution.
0.55

It is worth noticing that all the mentioned variations of both 5.4

Nspd (e/atom)

charges and magnetic moments are very small and the main
5.3
variations in the local magnetic properties are essentially due to up (initial)

down (initial) LMTO - up

the change in atomic coordination. However, in nanoparticles up (relax)
LMTO - down
down (relax)
where deformations are more important, the role of relaxation 3.7

could take a bigger part in the modification of local magnetic

properties. Moreover changes of charge distributions at the 1 2 3 4 5
surface may alter, if not the magnetic properties, the surface th
N layer
reactivity to adsorbed gases. Based on these considerations, the
next section is devoted to the nanoparticles and the effects of
finite size and deformations on the local electronic structure and
magnetic properties, that we intend to quantify in the aim of a Fig. 2 Co(111) surface. From top to bottom : charge per orbital d,
better rationalization of magnetic nanostructures in a computer charge per orbitals sp, total charge per site (spd). The layers (abscissa)
material design perspective. are numbered from bulk (N=1) to surface (N=5)

4 Results: Cobalt nanoparticles
In the case of transition metals, in a size range from 1 to 2 ≈
nm, icosahedral shapes are frequently observed in experiments
and we take this archetypal shape to perform our study of the
influence of structural changes onto local electronic structure and
magnetism. Icosahedral nanoparticles are based on the Mackay
icosahedron, a shape deriving from the regular cuboctahedron.
It is obtained therefore starting from a regular cuboctahedron
where the constituting atoms are separated by the interatomic
distances of the bulk Co in an FCC stacking. The intra-layer

J
our
nal
Name,
[yea
r][
,vol
.
], 1–9 | 3
distances are then dilated by about 5% relatively to the inter-
layer distances, which leads to a distortion of the (100) facets
into triangular (111) facets. It is worth noticing that starting in
the calculations with a cuboctahedron leads to its deformation
towards an icosahedron which is more stable within the same
number of atoms. To our knowledge, the question of the accurate
transition to another morphology at higher sizes is not clearly
established. From recent DFT calculations it is clear that up to
309 atoms the icosahedral shape is favored for Co nanoparti-
cles 15 . On the other hand, at much higher size (diameter ≈ 10
nm), the experiments show clearly the Wulff shape (truncated
octahedron or TO in the following) of Co nanoparticles 23 , as
expected for close-packed transition metals. In this context, we
will compare the main results for the icosahedra to the case of a
TO morphology for a higher size (405 atoms, ≈ 2nm of diameter). Fig. 3 Images of icosahedral nanoparticles colored following the local
spin magnetic moment per site. The color scale ranges from the minimal
to the maximal value for each nanoparticle independently. Number of
The relaxed icosahedral nanoparticles are represented in Fig.3 atoms in the nanoparticles, from left to right: 13, 55, 147 and 309.
where the represented atoms are colored following their local
spin magnetic moment using a scale based on the visible spec-
trum so that higher moments correspond to the red color and the 4.5

Ico 13
d1
4.5

Ico 55

nm)
lower ones to the purple color. The distances inside these parti- d2

-1
(10
interatomic distances
cles are analyzed first in order to consider the structural effects
4.0 4.0

(relaxation, deformation) and their interplay with the local elec-

2.5 2.5

tronic structure. The local distribution of distances, 1rst and 2nd

shells of neighbors for each atom, is represented in Fig.4 for the 4.5
0 2 4 6 8 10 12 14

4.5
0 5 10 15 20 25 30 35 40 45 50 55

Ico 147 Ico 309

relaxed icosahedral nanoparticles. A general characteristic shown
nm )
-1
(10

by these distributions is the decreasing distance from one layer of

interatomic distances

4.0 4.0

atoms to the other, going from the surface to the center, meaning
that the core of the icosahedron is in the strongest compressive 2.5 2.5

state which is expected for icosahedral nanoparticles 24 . In more 0 20 40 60 80 100 120 140 0 50 100 150 200 250 300

details, when looking closer to the local distances variations some

atom number atom number

other specificities are noticeable. First, there is a dilation of the

second nearest distance implying the vertex atoms of the surface,
which is a consequence of the intralayer dilation. Looking at the
particular case of vertex sites, their first neighbors distances are
contracted relatively to the other surface atoms, but in return
their second nearest neighbors distances are clearly longer. For
the two largest nanoparticles (containing 147 and 309 atoms),
the distribution of the second nearest distances is very inhomoge-
neous with a particular behavior identified along the axis joining
the central atom to the surface vertex atom. Along this direc-
tion, there is a connection between the enhanced contraction of
first nearest neighbors and the enhanced dilation of second near-
est neighbors distances. This configuration is expected to play a
part in the evolution of the electronic structure through variable
hybridizations of the orbitals and charge distributions.
4.1 Shell analysis of the electronic structure in Co icosahe-
dral nanoparticles
The main variations of the magnetic moment and of the distances
(1rst nearest neighbors) are occurring from one shell to another in
the nanoparticles so that the local electronic properties in relation
to the atomic local environment, can be analyzed by considering
values per shell. From Figure 3, it can indeed be clearly seen that,
in relation to the symmetric icosahedral configuration, the spin
Fig. 4 Local distribution of distances in the relaxed icosahedral
nanoparticles containing from 13 to 309 atoms: 1rst and 2nd shells of
atoms for each atom.
magnetic moment presents a rather constant value per shell ex-
cept for the surface where sites with different coordination num-
bers coexist. It increases steadily, starting from a much reduced
magnetic moment at the central site to an enhanced magnetic
moment at the surface sites. This can be related to the variation
of the compressive state which is itself related to the reduction
of the next nearest distances when going towards the center of
the nanoparticles. At the surface, as expected, the more the coor-
dination is reduced the more enhanced is the magnetic moment
which reaches a maximum at vertex sites. Pursuing this analy-
sis per shell, in a more quantitative way, we have plotted both
the average magnetic moment per shell (Figure 5) and the total
charge variation per shell (Figure 6). In each case, the initial (not
relaxed) configuration has also been added.
As already mentioned, for each size, the average magnetic mo-
ment increases in each shell of atoms, from the center to the
surface. At the surface it is clear that the reduced coordination
dominates the enhancement of the magnetic moment leading to
an increased moment from the facet atoms to the least coordi-

4| J
our
nal
Name,
[yea
r][
,vol
.
],
1–9
 2.5 2.0
1.5
1.0
(relax)
0.5
2.0
0.0
Bulk (initial)
-0.5
-1.0 N (relax)
1.5 -1.5
/atom)

Ih Co 13 -2.0
N (initial) Ih - Co
-2.5 13
1.0 -3.0

1 2 3 4 5 1 2 3 4 5

N
B

2.5 2.0
1.5
average magnetic moment per shell (

1.0

total charge variation per shell

0.5
2.0
0.0
-0.5
-1.0
1.5 -1.5
Ih Co 55 -2.0 Ih - Co
-2.5 55
1.0 -3.0

1 2 3 4 5 1 2 3 4 5

2.5 2.0
1.5
1.0
0.5
2.0
0.0
-0.5
-1.0
1.5 -1.5
Ih Co 147 -2.0
Ih - Co
-2.5 147
1.0 -3.0

1 2 3 4 5 1 2 3 4 5

2.5 2.0
1.5
1.0
0.5
2.0 0.0
-0.5
-1.0
1.5 -1.5
-2.0
Ih - Co
Ih Co 309 -2.5 309
-3.0
1.0
1 2 3 4 5 1 2 3 4 5

th th
N shell N shell

Fig. 5 Icosahedra : average magnetic moment (µB /atom) per shell. Fig. 6 Icosahedra : total charge variation per shell.

nated vertex one. One other general effect of the relaxation is
to reduce the overall interatomic distances relatively to the initial
configuration, leading to a reduced local spin magnetic moment,
in particular in the core of the two larger nanoparticles which be-
comes in a particular compressive state. As a matter of fact, from
the size of N ≥ 147 atoms, and inside the nanoparticles from the
first subsurface shell, the magnetic moment is decreasing under
the bulk value, so that the curve of the inner part of the particle
is always under the bulk value which is reached or overcome only
from the subsurface. This effect is a particular finite size effect
which leads to a very different behavior in a nanoparticle from
what could be inferred from a rationale purely based on coordi-
nation numbers. It is worth noticing that the central atomic site
is in particular subject to a drop in the magnetic moment which
is even more marked after relaxation. The central atom is in-
deed in a particular state of compression (shortened first nearest
neighbors) and certainly subject also to the perturbation originat-
ing from the surface. This perturbation, owing to its symmetry
produces strong deformations in the local density of states which
will be presented in the last section of the article. As expected its
influence decreases when the size increases but even so, it acts on
a long range distance and the bulk magnetic moment is probably
a limit which is reachable only in the core of very large parti-
cles. This behavior, already mentioned for very small clusters of
a few atoms 25 , is found here to persist in larger nanoparticles
of a few hundreds of atoms. Remaining in the shell analysis, a
first insight in the electronic state is to quantify the total charge
transfers between layers. While in the initial structure the total
charge transfers are very limited, in the relaxed structure a very
specific behavior appears. The subsurface layer acts as a single
donor of electrons transferred to the close layers mainly at the
surface and with a less extent to its close inner shells. Exactly
the same behavior is already mentioned in the literature for mag-
netic Pt nanoparticles 26 so that it can be described as a general
phenomenon. In addition, as visible in Ref. 26 , it does not depend
on the shape, at least considering icosahedra, cuboctahedra and
decahedra in the same size range from 13 to 309 atoms, which
supports its general character. However, reduced to the charge
transferred per atom, the involved values do not exceed a few
hundredths of electron per atom and it is therefore not obvious
how the magnetic properties can be accordingly modified in Co
systems. As these features have not been studied into detail, they
deserve a closer analysis which is done in the next section at the
local electronic level. It is worth mentioning that, preliminarily to
this, the spin local magnetic moment has been also extracted for
each site, the only local particularity being the slightly enhanced
magnetic moment at under-vertex sites which are visibly under
the influence of the surface, as mentioned in section 3, and which
is, here, more marked in the case of vertices relatively to facets.
In addition, at this stage, no other particular correlation could be
found between local magnetic moments and the local structure
apart from the decrease of the local magnetic moment with the

J
our
nal
Name,
[yea
r][
,vol
.
], 1–9 | 5
reduced 1rst next neighbors distances and its enhancing with the
lowering of coordination at the surface.
4.2 Truncated octahedron shaped nanoparticles (TO, Wulff
shape)
Although the shape transition is not well defined in Co, one can
attempt to extend the study to the TO shapes which should take
to be examined in order to detect singularities apart from the
previously identified average behavior per shell. The charges
per site (spd), decomposed in spin up and spin down contri-
butions, are represented in Fig. 8 (the total charge is also shown).
6.0
spd up
10.5

over the icosahedral ones at sufficiently higher size 23 . We take

13
spd down

5.5 10.0

here the case of a TO of 405 atoms which should be enough rep-

Nspd (up,down)
5.0 9.5

Ne (total)
resentative. Anyway, the DFT calculations start to be difficult to 4.5 9.0

N e (total)
4.0 8.5

handle at larger sizes, in particular relaxations can hardly be car-

3.5 8.0

ried out in a reliable way. The relaxed nanoparticles are repre- 3.0 7.5
0 5 10 15 -5 0 5 10 15 20 25 30 35 40 45 50 55 60

sented in Fig.7 where the represented atoms are colored follow- 6.0
13 55 13 55 147
10.5

ing their local spin magnetic moment using a scale based on the 5.5 10.0

Nspd (up,down)
5.0 9.5

visible spectrum so that higher moments correspond to the red

Ne (total)
4.5 9.0

color and the lower ones to the purple color. 4.0 8.5

3.5 8.0

3.0 7.5
0 50 100 150 0 100 200 300

atom number atom number

Fig. 8 Charge per site (spd) decomposed in spin up (majority spin, red)
and spin down (minority spin, blue) contributions for the same
icosahedra as in Fig. 4. The total charge (sum, right y axis) is
represented in green. The x axis indicates the atom number starting
from the central atom.

Fig. 7 Nanoparticle with the shape of a truncated octahedron (TO, Wulff
shape) containing 405 atoms. The color scale ranges from the minimal
to the maximal value of spin magnetic moment (µmin = 1.57µB and
µmax = 1.96µB .
The higher spin magnetic moment is 1.96µB at vertex sites and
the lower one is 1.56µB at the center of the nanoparticle. Some os-
cillations are visible in the magnetic moment between the shells
from the center to the second layer under the surface, however
their amplitude is in fact very small (≤ 0.04 µB ) and a mean mag-
netic moment of 1.59 µB is obtained in the core (excluding sub-
surface and surface shells) which is a very close to the bulk value
(1.61 µB ).
4.3 Local charge analysis
For a detailed and comprehensive local analysis, we will consider
the charge per site and orbital (up, down and total respectively)
and finally some local densities of states (LDOS). The latter will
be used to better clarify the relative importance of pure charge
transfer and deformations of the LDOS on the modification of the
local magnetism.
As it has been previously underlined that the subsurface
layer, in the nanoparticles, present a global charge loss dis-
tributed mainly to the surface, and to a less extent to the
underlying layers, the local charge distributions indeed deserve
On the total charge curves, one indeed sees that the subsur-
face shell presents a lack of charge, mainly a minority spin charge
loss (spin down), and the other shells a slight excess of charge
equally distributed between minority and majority spin (spin up)
populations. One single exception is at the vertex sites which
also undergo a charge loss. This effect appears already from the
smallest icosahedron with at least 3 shells (55 atoms). The loss is
rather undergone by the minority spin bands since there is clearly
an asymmetry between the increase of the majority spin popula-
tion and the larger decrease of the minority spin one. It therefore
contributes, in addition to the coordination effect, to the increase
of the magnetic moment while probably in a lesser extent. The
other noteworthy fact is that the redistribution between majority
and minority populations is very important at the center, in par-
ticular for the icosahedron of 147 atoms. On the total population,
at this site, an excess of 0.04 e appears for the icosahedron of
147 atoms and of 0.02 e for the icosahedron of 309 atoms. These
value are too small to relate them to a significant change in the
spin local magnetic moment which is more likely due to the de-
formations of the LDOS in response to local structural changes
and the persistent effect due to the finite size of the particles. The
latter is logically more pronounced for the smaller nanoparticle of
147 atoms than for the larger one of 309 atoms. However, it is in-
teresting to have a closer look to the charge redistribution effects
by also separating d(up, down) and sp(up, down) contribution per
site. The results are gathered in Fig. 9 and 10. On the overall,
the charge variation per site, are the most important for sp pop-
ulations. At the subsurface sites, the charge variation, obtained
by comparison with bulk, reaches on the average −0.16e/atom
for the cluster of 13 atoms (limit case where the subsurface site

6| J
our
nal
Name,
[yea
r][
,vol
.
],
1–9
is also the center) but it decreases progressively when the size of
the nanoparticles increases (it is about −0.04 e/atom for the larger 5.0 5.0
13

4.8 4.8

size of 309 atoms). For the d population, the charge variations are 4.6
Bulk up

4.6

Nd (up, down)
lower, from −0.02 to +0.01 e/atom, the latter value being reached 4.4 d up
d down
4.4

3.4 3.4

for the size of 55 atoms and does not change afterwards when 3.2 3.2

3.0 Bulk down 3.0

the size increases. Putting sp and d together, the maximal charge 2.8

2.6
2.8

2.6

loss per atom at the subsurface of the nanoparticles is therefore 5.0

0 5 10 15

5.0
0 10 20 30 40 50 60

13 55 13 55 147

evolving rather rapidly, as a function of the size, towards the limit 4.8 4.8

value observed at the surface which is −0.02 e/atom. Then, it

4.6 4.6

Nd (up, down)
4.4 4.4

is worth noticing that some deviations to the mean value, per 3.4 3.4

3.2 3.2

atom and per shell, are occurring at particular sites like the sites 3.0

2.8
3.0

2.8

situated along the axis joining the vertices to the center. It is in- 2.6

0 50 100 150
2.6

0 100 200 300

deed the place where the distance deformations are the most im- atom number atom number

portant which favors certainly some charge transfers. However,

again, the involved quantities are very small apart at the least co-
ordinated vertex sites themselves, where a change occurs since,
there, the antiferromagnetic coupling, observed in extended sys-
tems (bulk or surfaces) between sp and and the total electronic
distribution, becomes strongly ferromagnetic with a total reversal
between the minority spin and majority spin electronic fillings.
As the contribution to the magnetic moment becomes now posi-
Fig. 9 Icosahedra : d charge per site decomposed in spin up (majority
spin, red) and spin down (minority spin, blue) contributions for the same
icosahedra as in Fig. 4. The corresponding charge in bulk is
represented by an horizontal line: red for spin up and blue for spin down.
The x axis indicates the atom number starting from the central atom.
1.2 1.0

sp up
1 13
sp up
tive, this effect also contributes to enhance the magnetic moment 1.0
sp down
0.9
sp down

and this, moreover, exclusively at these vertex sites. The effect,

sp (up, down)
0.8 Bulk down 0.8

at least in the size range of this study, seems independent of the 0.6
Bulk up

0.7

sp N
size of the nanoparticles, with a value of 0.12 ≤ ∆Nup−down ≤ 0.3, 0.4 0.6

sp
the tendency being a decreasing of ∆Nup−down when the size in- 0.2
0 2 4 6 8 10 12 14
0.5
0 10 20 30 40 50 60

creases. Finally, we go back to the case of the drop of the magnetic 1.0
13 55
sp up
1.0
13 55 147

sp down sp up
moment at the center of the nanoparticle, after finding that the 0.9 0.9
sp down
Nsp (up,down)

charge transfers are too small at this place to affect noticeably 0.8 0.8

the local magnetic moment, so that it remains to take a closer 0.7 0.7

look at the deformation of the LDOS consequently to local struc- 0.6 0.6

tural changes. The LDOS at the center of the two largest relaxed
0.5 0.5
0 50 100 150 0 100 200 300

atom number atom number

nanoparticles (147 and 309 atoms) are represented in the graph

of Fig. 11 together with the LDOS in the pure bulk cobalt system.
It can be seen that the LDOS undergo strong deformation in the Fig. 10 Icosahedra : sp charge per site decomposed in spin up
center of the nanoparticles and are rather different from the bulk (majority spin, red) and spin down (minority spin, blue) contributions for
the same icosahedra as in Fig. 4. The corresponding charge in bulk is
one. As found previously, the charge transfer to the central site
represented by an horizontal line: red for spin up and blue for spin down.
being very small (≤ 0.04 e), only the LDOS deformations should The x axis indicates the atom number starting from the central atom.
be at the origin of the drop in the magnetic moment at the center
of the considered nanoparticles.
tems (bulk or surfaces) between sp and and the total electronic
distribution, becomes strongly ferromagnetic with a total reversal
4.4 Truncated octahedron shaped nanoparticles (TO, Wulff
between the minority and majority spin electronic fillings.
shape)

It is now interesting to check how these behaviors scale with 5 Conclusion

larger sizes and a different shape, taking again the case of a TO
of 405 atoms. On the overall, charge transfers are never exceed-
ing 0.02 e confirming the low impact of charge transfers in the
magnetic properties of ferromagnetic nanoparticles. Then, the
characteristics of the subsurface layer acting as a single donor of
electrons transferred to the close layers is not found anymore.
The only two sites presenting a charge loss are eventually the
vertices and their underlying neighbors in the subsurface with a
charge loss reaching 0.06 e for under-vertex sites. Moreover, as for
the smaller nanoparticles, at the single least coordinated vertex
sites, the antiferromagnetic coupling, observed in extended sys-
Within both a shell and a local fine analyses of SIESTA-DFT
calculations, we have unraveled the effects of local structural
environment including atomic coordination, structural deforma-
tions and finite size effects in cobalt ferromagnetic nanoparticles.
A shell analysis of the quantities obtained from SIESTA-DFT
calculations has been performed. Concerning the charge analy-
sis, a particular effect of donor-acceptor behavior between the
sublayer and its close environment with in particular the surface
layer is observed in ferromagnetic Co icosahedra. It can be
considered as a general effect, since it has already been obtained

J
our
nal
Name,
[yea
r][
,vol
.
], 1–9 | 7

Fig. 11 Spin up and down contributions to the local (projected) densities
of states at the central site of the relaxed icosahedra of 147 (green
curve) and 309 (red curve) atoms. The energies refer to the Fermi
energy EF . The blue curve corresponds to the bulk material reference.)
in weakly magnetic nanoparticles of Pt 26 . However, in a more
local site analysis, it can be seen that there is actually a very
weak charge transfer between atoms (sites and orbitals). Then
a particular magnetic behavior is observed exclusively at vertex
sites with the reversal from an antiferromagnetic coupling to a
ferromagnetic one for sp electrons. The particular compressive
state present in icosahedral nanoparticles leads to a redistribution
between up and down electronic populations leading to a drop
of the magnetic moment. Since the electron transfer are rather
weak, the drop of the local magnetic moment is the consequence
of important deformations of the local densities of states through
hybridizations due to the structural deformations.
In the attempt to extend the rationale on the scaling of mag-
netic properties with both the increasing size of nanoparticles
and shape transition, we gave an overview of the same type
of analysis in the case of a TO nanoparticle containing 405
atoms. In summary, some characteristics are well preserved
like : the weak charge transfer between atoms in the particles,
the maximal charge loss at vertices and under-vertices sites
and finally the reversal from an antiferromagnetic coupling to
a ferromagnetic one for sp electrons of vertex sites exclusively.
The latter effect leads to an increased local magnetic moment
at these sites. Then conversely to icosahedral nanoparticles, the
particular donor-acceptor shell behavior observed in icoshedra
between subsurface and surface shells is, in TO nanoparticles,
limited to the under-vertex and vertex sites. The value of the
local spin magnetic moment, in the core, show oscillations of
weak amplitude (< 0.04 µB ) so that it can be rather considered as
uniform with a mean value very close to the bulk one.
Connecting, in this way, the different contributions between
structural changes and electronic structure in nanoparticles
enables to extend the transferability of DFT calculations toward
realistic physico-chemical environments of the considered
nanoparticles and therefore to contribute to a better choice of
systems with targeted properties either magnetic or catalytic.
8| J
our
nal
Name,
[yea
r][
,vol
.
],
1–9
In the aim of a further multi-scale approach, the important
result on the local neutrality per site and orbitals (no charge
transfer) is fundamental for the implementation of self-consistent
semi-empirical optimization algorithms of the structure for larger
size of particles, inaccessible with first principle calculations.
Acknowledgments
This work was performed thanks to the french Institut du
Développement et des Ressources en Informatique Scientifique
through a grant of computer time (Project No. 100525).
References
1 J. Bansmann, S. H. Baker, C. Binns, J. A. Blackman, J. P.
Bucher, J. Dorantes-Dávila, V. Dupuis, L. Favre, D. Kechrakos,
A. Kleibert, K. H. Meiwes-Broer, G. M. Pastor, A. Perez,
O. Toulemonde, K. N. Trohidou, J. Tuaillon and Y. Xie, Sur-
face Science Reports, 2005, 56, 89–275.
2 Metal Nanoparticles for Catalysis: Advances and Applications,
ed. F. Tao, Royal Society of Chemistry, 2014.
3 P. Entel and M. E. Gruner, Journal of Physics Condensed Matter,
2009, 21, year.
4 D. Yohan and B. D. Chithrani, Journal of Biomedical Nanotech-
nology, 2014, 10, 2371–2392.
5 J. V. J. Hutter, M. Ianuzzi, F. Schiffman, Comput. Mol. Sci.,
2014, 4, 15.
6 J. M. Soler, E. Artacho, J. D. Gale, A. García, J. Junquera,
P. Ordejón and D. Sánchez-Portal, Journal of Physics: Con-
densed Matter, 2002.
7 C. J. Cramer and D. G. Truhlar, Physical Chemistry Chemical
Physics, 2009, 11, 10757.
8 M. F. M. E. Gruner, G. Rollmann, P. Entel, Phys. Rev. Lett.,
2008, 100, 087203.
9 F. Aguilera-Granja, J. M. Montejano-Carrizales, J. Guevara
and A. M. Llois, Physica Status Solidi (B) Basic Research, 2000,
220, 455–460.
10 S. Dennler, J. Morillo and G. M. Pastor, Surface Science, 2003,
532-535, 334–340.
11 S. Dennler, J. L. Ricardo-Chavez, J. Morillo and G. M. Pastor,
European Physical Journal D, 2003, 24, 237–240.
12 J. L. Rodríguez-López, F. Aguilera-Granja, K. Michaelian and
A. Vega, Physical Review B - Condensed Matter and Materials
Physics, 2003, 67, 1–9.
13 S. Datta, M. Kabir, S. Ganguly, B. Sanyal, T. Saha-Dasgupta
and A. Mookerjee, Physical Review B - Condensed Matter and
Materials Physics, 2007, 76, 1–11.
14 A. Langenberg, K. Hirsch, A. Ławicki, V. Zamudio-Bayer,
M. Niemeyer, P. Chmiela, B. Langbehn, A. Terasaki, B. V. Is-
sendorff and J. T. Lau, Physical Review B - Condensed Matter
and Materials Physics, 2014, 90, 1–14.
15 J. Souto-Casares, M. Sakurai and J. R. Chelikowsky, Physical
Review B, 2016, 93, 1–6.
16 J. L. M. N. Troullier, Phys. Rev. B, 1991, 43, 1993–2001.
17 L. K. D. M. Bylander, Phys. Rev. Lett., 1982, 48, 1425–1428.
18 R. S. Mulliken, J. Chem. Phys., 1955, 23, 1833.
19 L. Zosiak, C. Goyhenex, R. Kozubski and G. Tréglia, Journal of
Physics Condensed Matter, 2015, 27, year.
20 M. N. S. H. Vosko, L. Wilk, Can. J. Phys., 1980, 58, 1200.
21 P. W. Ilja Turek, Vaclav drchal, Josef Kurdovsky, Mojmir Sob,
Electronic Structure of Disordered Alloys, Surfaces and Inter-
faces, Kluwer Academic Publishers, Dordrecht, 1997.
22 M. Alden, S. Mirbt, H. L. Skriver, N. M. Rosengaard and B. Jo-
hansson, Surface magnetism in iron, cobalt, and nickel, 1992.
23 A. Kleibert, A. Balan, R. Yanes, P. M. Derlet, C. A. F. Vaz,
M. Timm, A. F. Rodríguez, A. Béché, J. Verbeeck, R. S. Dhaka,
M. Radovic, U. Nowak and F. Nolting, Phys. Rev. B, 2017, 95,
195404.
24 D. Bochicchio and R. Ferrando, 2013, 165435, 1–13.
25 K. H. B. G. M. Pastor, J. Dorantes-Davila, Physica B, 1988, 149,
22–25.
26 C. Di Paola, R. D’Agosta and F. Baletto, Nano Letters, 2016,
16, 2885–2889.

J
our
nal
Name,
[yea
r][
,vol
.
], 1–9 | 9