Theoretical study of magnetism,

structure and chemical order in

transition-metal alloy clusters
von
Herrn Junais Habeeb Mokkath
aus Kerela, India
Dissertation zur Erlangung des akademischen Grades
eines Doktors der Naturwissenschaften (Dr. rer. nat.)
vorgelegt dem Fachbereich Mathematik und Naturwissenschaften
der Universitat Kassel
Betreuer:
Prof. Dr. G. M. Pastor
Disputation am 15. Februar 2012
Theoretical study of magnetism,
structure and chemical order in
transition-metal alloy clusters
von
Herrn Junais Habeeb Mokkath
born in Kerela, India
A Thesis submitted to the Department of Mathematics and Natural
Sciences in partial fulllment of the requirements for the degree
Doctor of Natural Sciences (Dr. rer. nat.)
Theoretical physics institute
University of Kassel
supervised by
Prof. Dr. G. M. Pastor
Examination on 15
th
February 2012
To my parents, jessi and reyan
Who teacheth by the pen, Teacheth man that which he
knew not.
Holy Quran.
.
Abstract
Research on transition-metal nanoalloy clusters composed of a few atoms is fascinating by their
unusual properties due to the interplay among the structure, chemical order and magnetism. Such
nanoalloy clusters, can be used to construct nanometer devices for technological applications by ma-
nipulating their remarkable magnetic, chemical and optical properties. Determining the nanoscopic
features exhibited by the magnetic alloy clusters signies the need for a systematic global and lo-
cal exploration of their potential-energy surface in order to identify all the relevant energetically
low-lying magnetic isomers.
In this thesis the sampling of the potential-energy surface has been performed by employing
the state-of-the-art spin-polarized density-functional theory in combination with graph theory and
the basin-hopping global optimization techniques. This combination is vital for a quantitative
analysis of the quantum mechanical energetics.
The rst approach, i.e., spin-polarized density-functional theory together with the graph theory
method, is applied to study the Fe
m
Rh
n
and Co
m
Pd
n
clusters having N = m + n 8 atoms.
We carried out a thorough and systematic sampling of the potential-energy surface by taking
into account all possible initial cluster topologies, all dierent distributions of the two kinds of
atoms within the cluster, the entire concentration range between the pure limits, and dierent
initial magnetic congurations such as ferro- and anti-ferromagnetic coupling. The remarkable
magnetic properties shown by FeRh and CoPd nanoclusters are attributed to the extremely reduced
coordination number together with the charge transfer from 3d to 4d elements.
The second approach, i.e., spin-polarized density-functional theory together with the basin-
hopping method is applied to study the small Fe
6
, Fe
3
Rh
3
and Rh
6
and the larger Fe
13
, Fe
6
Rh
7
and Rh
13
clusters as illustrative benchmark systems. This method is able to identify the true
ground-state structures of Fe
6
and Fe
3
Rh
3
which were not obtained by using the rst approach.
However, both approaches predict a similar cluster for the ground-state of Rh
6
. Moreover, the
computational time taken by this approach is found to be signicantly lower than the rst approach.
The ground-state structure of Fe
13
cluster is found to be an icosahedral structure, whereas Rh
13
and Fe
6
Rh
7
isomers relax into cage-like and layered-like structures, respectively. All the clusters
display a remarkable variety of structural and magnetic behaviors. It is observed that the isomers
having similar shape with small distortion with respect to each other can exhibit quite dierent
magnetic moments. This has been interpreted as a probable artifact of spin-rotational symmetry
breaking introduced by the spin-polarized GGA. The possibility of combining the spin-polarized
density-functional theory with some other global optimization techniques such as minima-hopping
method could be the next step in this direction. This combination is expected to be an ideal
sampling approach having the advantage of avoiding eciently the search over irrelevant regions
of the potential energy surface.
3
Abstrakt

Ubergangsmetall-Nano-Gemisch-Cluster bestehend aus einigen wenigen Atomen stellen we-

gen ihren ungewohnlichen Eigenschaften ein faszinierendes Forschungsgebiet dar. Ihre Eigen-
schaften sind zur uckzuf uhren auf das Wechselspiel zwischen Struktur, chemischer Ordnung und
Magnetismus. Solche Nano-Gemisch-Cluster knnen durch Manipulation ihrer bemerkenswerten
magnetischen, chemischen und optischen Eigenschaften benutzt werden, um Nanometer-Bauteile
f ur technologische Anwendungen zu konstruieren. Die Bestimmung der nanoskopischen Besonder-
heiten dieser magnetischen Gemisch-Cluster kennzeichnet die Notwendigkeit einer systematischen
globalen und lokalen Erforschung ihrer Potentialenergie-Oberache, um alle relevanten energetisch
niedrig gelegenen magnetischen Isomere zu identizieren.
In dieser These ist das Abtasten der Potentialenergie-Oberache mittels modernster spinpo-
larisierter Dichtefunktionaltheorie in Kombination mit sowohl Graphentheorie als auch globalen
basin-hopping-Optimierungstechniken durchgef uhrt worden. Diese Kombination ist entscheidend
fr eine quantitative Analyse der quantenmechanischen Energetik.
Der erste Ansatz, d.h. spinpolarisierte Dichtefunktionaltheorie zusammen mit der Graphen-
theorie-Methode, ist angewandt worden, um Fe
m
Rh
n
und Co
m
Pd
n
Cluster mit N = m + n 8
Atomen zu untersuchen. Die Potentialenergie-Oberache wurde vollstandig und systematisch
abgetastet, indem alle moglichen Anfangs-Cluster-Topologien, alle verschiedenen Verteilungen der
zwei Atomsorten innerhalb des Clusters, der gesamte Konzentrationsbereich zwischen den Gren-
zen monoatomarer Cluster, und verschiedene magnetische Anfangskongurationen wie ferro- und
antiferromagnetische Kopplung ber ucksichtigt wurden. Die von FeRh und CoPd aufgezeigten
bemerkenswerten magnetischen Eigenschaften sind auf die extrem reduzierte Koordinationszahl
zusammen mit dem Ladungstransfer von 3d zu 4d Elementen zur uckgef uhrt worden.
Die zweite Herangehensweise, d.h. spinpolarisierte Dichtefunktionaltheorie zusammen mit der
basin-hopping-Methode, ist gewahlt worden, um die kleinen Fe
6
, Fe
3
Rh
3
and Rh
6
und die groeren
Fe
13
, Fe
6
Rh
7
und Rh
13
Cluster als anschauliche Bezugswertsysteme zu studieren. Diese Methode
ist in der Lage, die wahren Grundzustandsstrukturen von Fe
6
und Fe
3
Rh
3
zu identizieren, die
mittels des ersten Ansatzes nicht gefunden worden waren. F ur den Grundzustand von Rh
6
jedoch
sagen beide Ansatze eine ahnliche Clusterstruktur voraus. Dar uber hinaus ist herausgefunden
worden, dass die f ur die zweite Herangehensweise benotigte Computerzeit bedeutend geringer ist
als fr die erste. F ur den Grundzustand des Fe
13
Clusters ist eine Ikosaederstruktur gefunden
worden, wohingegen die Rh
13
und Fe
6
Rh
7
Isomere zu kagartigen bzw. schichtartigen Struk-
turen relaxieren. Alle Cluster zeigen eine bemerkenswerte Vielfalt an strukturellem und magnetis-
chem Verhalten. Es ist beobachtet worden, dass Isomere mit ahnlicher Form aber geringer gegen-
seitiger Verzerrung ziemlich unterschiedliche magnetische Momente aufweisen. Dieses ist als ein
wahrscheinliches Artefakt der Spin-Rotationssymmetrie-Brechung durch die spinpolarisierte GGA
interpretiert worden. Die Moglichkeit, die spinpolarisierte Dichtefunktionaltheorie mit anderen
globalen Optimierungstechniken wie zum Beispiel der Minima-hopping-Methode zu kombinieren,
konnte der nachste Schritt in dieser Richtung sein. Es wird erwartet, dass diese Kombination
einen idealen Abtastansatz darstellt, der den Vorteil, dass das Suchen uber irrelevante Regionen
der Potentialenergie-Oberche ezient vermieden wird, aufweist.
4
Acknowledgments
A journey is easier when we travel together. This thesis is the result of four years of
work whereby I have been accompanied and supported by many people. It is a pleasant
aspect that I now have the opportunity to express my gratitude for all of them.
The rst person I would like to sincerely thank is my supervisor, Prof. Dr. G. M.
Pastor for his guidance, enthusiastic and integral view on research. He gave me enough
freedom in thinking, choosing my own problems and collaborating with other colleagues
which helped me to bring out my best. Many thanks to Prof. Dr. Martin E. Garcia
for reviewing this thesis. I would like to express my sincere gratitude to Prof. Dr. J.
Dorantes Davilla for his friendly relationship. I gratefully acknowledge Dr. Luis Daz for
an active collaboration on basin-hopping method. He gave me motivation to learn python
programing language. I thank also Dr. J. L. Chavez who guided my rst steps through
VASP. I thank also Dr. Pedro Ruiz Diaz, who helped me whenever I needed his help
in Linux and VASP. Special thanks to Waldemar Tows, who made my life in Germany
wonderful. Special thanks to Matthieu Sauban`ere for reading and correcting this thesis.
Thanks also to the secretary Katarina Schmidt and Andrea Wecker for their great help
solving any situation. I thank all my friends for standing beside me all the time.
5
Contents
1 Introduction 9
1.1 Magnetism in the bulk phase . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Magnetism in reduced dimensionality . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Magnetic characterization methods . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 Stern-Gerlach (SG) experiment . . . . . . . . . . . . . . . . . . . . . 15
1.3.2 X-ray magnetic circular dichroism . . . . . . . . . . . . . . . . . . . 16
1.4 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Background on electronic structure theory 19
2.1 The quantum many-body problem . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.1 Hartree-Fock Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1.2 The Thomas-Fermi Theory . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 Density Functional theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 The Hohenberg-Kohn theorems . . . . . . . . . . . . . . . . . . . . . 23
2.2.2 The Kohn-Sham theory . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Kohn-Sham equations . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Exchange-correlation energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.1 Exchange and correlational functionals . . . . . . . . . . . . . . . . . 25
2.4 Projector augmented wave method . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.1 The PAW transformation operator . . . . . . . . . . . . . . . . . . . 28
2.4.2 Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5 Magnetic anisotropy energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3 Exploring the ground-state energy surface of nanoclusters 34
3.1 Local Optimization methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.1.1 Steepest Descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.2 Conjugate Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.3 The Broyden Fletcher Goldfarb Shanno method . . . . . . . . . . . 36
3.2 Global Optimization schemes . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.1 Graph theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.2 Simulated annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.3 Basin-hopping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.4 Genetic algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4 Interplay of structure, magnetism and chemical order in small FeRh
clusters 42
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Ab-initio relaxation of clusters . . . . . . . . . . . . . . . . . . . . . . . . . 44
6
Contents
4.3 Structure and magnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3.1 FeRh dimers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3.2 FeRh trimers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3.3 FeRh tetramers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.4 FeRh pentamers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3.5 FeRh hexamers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3.6 Exploring heptamers and octamers . . . . . . . . . . . . . . . . . . 57
4.4 Trends as a function of size and composition . . . . . . . . . . . . . . . . . . 62
4.4.1 Binding energy and magnetic moments . . . . . . . . . . . . . . . . . 62
4.4.2 Relative stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.4.3 Electronic structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5 Structure and magnetism of small CoPd clusters 71
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.2 Computational details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.3.1 Dimers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.3.2 Trimers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.3.3 Tetramers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.3.4 Pentamers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3.5 Hexamers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.3.6 Heptamers and Octamers . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4 Size and composition trends . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4.1 Binding energy and magnetic moments . . . . . . . . . . . . . . . . . 89
5.4.2 Relative stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6 First principles spin-polarized basin-hopping method 93
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Computational aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.1 Basin-Hopping method . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.2 Measuring the sampling eciency . . . . . . . . . . . . . . . . . . . 96
6.3 Results for small homogeneous clusters . . . . . . . . . . . . . . . . . . . . . 98
6.3.1 Dominant isomers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.3.2 Magnetic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.4 Results for heterogeneous small clusters . . . . . . . . . . . . . . . . . . . . 104
6.4.1 Dominant isomers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.4.2 Magnetic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.5 Results for larger clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.6 Summary and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7 Composition dependent orbital magnetism 113
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.2 Computational aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.3.1 Size eects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7
Contents
7.3.2 Composition eects on the magnetic properties . . . . . . . . . . . . 117
7.3.3 Angular dependence of the MAE . . . . . . . . . . . . . . . . . . . . 120
7.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8 Summary and outlook 122
Bibliography 125
List of gures 136
List of tables 139
Abbreviations and symbols 140
List of publications 142
8
1
Introduction
Nanoclusters composed of small number of atoms show special structural, chemical, mag-
netic, and optical properties. Their properties dier qualitatively and quantitatively from
those of atomic and the bulk limits [1]. These special features can be placed under a
common label of quantum size eects (QSE). These are due to the dramatic change in
the electronic structure of a system when its size reduced to few atoms, thus replacing the
quasi-continuous density of states by a discrete energy level spectrum. Nanoclusters have
large surface to volume ratio. This leads to the eective reduction of the mean coordi-
nation number for the atoms on the cluster surface. Therefore, electronic structure of a
surface atom and an interior atom within the nanocluster will be quite dierent. This is in
fact the main reason for the remarkable features shown by nanoclusters. One of the most
perplexing questions in the eld of magnetic nanoclusters is how their magnetic properties
change in the non-scalable, reduced dimensionality regime as compared to the atomic and
bulk limits. For example, it is found that nanoclusters composed of 3d transition metal
(TM) elements such as Fe
N
, Co
N
and Ni
N
show spin moments, orbital moments, and
magnetic anisotropy energies (MAEs) that are enhanced with respect to the correspond-
ing crystalline solids. Moreover, 4d TM nanoclusters composed of elements like Rh and
Pd are known to be magnetic despite being non-magnetic in the bulk [2, 410].
The possibilities of optimizing the cluster magnetic behavior by simply tuning the sys-
tem size in the case of pure clusters have been rather disappointing, particularly concern-
ing the magnetic anisotropy energy, which remains relatively small despite being orders
of magnitude larger than in solids due to the rather weak spin-orbit (SO) coupling in the
3d atoms [2]. Even though magnetic nanoclusters possess sizable magnetic moments per
atom, their blocking temperatures are found to be very small. From a technological per-
spective the most important question is: how to realize magnetic objects which are nano
and ferro simultaneously. In other words, how we can enhance the blocking temperature
of a nanocluster in order to utilize them as recording or storage media even at room tem-
perature applications. The most appropriate answer found so far to the above challenge
seems to be alloying ferromagnetic 3d elements with highly polarizable 4d or 5d elements.
9
This so called nanoalloying is expected to materialize the long time goal of realizing the
materials having large magnetic moments and suciently large MAEs per atom.
Although the potential advantages of nanoalloying can be grasped straightforwardly,
the problem involves a number of serious practical challenges for both theory and experi-
ment. Dierent growth or synthesis conditions can lead to dierent chemical orders, which
can be governed not just by energetic reasons but also by kinetic processes. For instance,
one may have to deal with segregated clusters having either a 4d core and a 3d outer shell
or vice versa. Post-synthesis manipulations can induce dierent degrees of intermixing,
including for example surface diusion or disordered alloys. Moreover, the inter atomic
distances are also expected to depend strongly on size and composition. Typical TM-
cluster bond-lengths are in fact 1020% smaller than in the corresponding bulk crystals.
Taking into account that itinerant magnetism is most sensitive to the local and chemical
environments of the atoms [3, 1618], it is clear that controlling the distribution of the
elements within the cluster is crucial for understanding magnetic nanoalloys. Systematic
theoretical studies of binary-metal clusters are hindered by the diversity of geometrical
conformations, ordered and disorder arrangements, as well as segregation tendencies that
have to be taken into account.
The above diversity of magnetic nanoclusters would make the sampling of the potential
energy surface (PES) nearly an impossible task due to the exponentially increasing number
of local minima. This exponential growth is analogous to the famous Levinthals paradox,
according to which a protein would never reach its native state within the lifetime of the
universe if it would have to go through all local PES minima completely randomly [19, 20].
At the early stages of cluster research the most widely used approach in order to circumvent
this problem was to consider educated guesses of cluster geometries. Such educated guesses
maneuvered by chemical intuition (for example: FCC, BCC, HCP, etc.) is a rst trial to
explore the magnetic behavior. However, this approach is not suitable for nanoclusters due
to the diversity of structural isomers, for example, due to strong Jahn-Teller distortions
or to the coexistence of isomers and dierent spin congurations in the case of magnetic
nanoclusters. This signals the need for a systematic, ecient, and unbiased sampling of
the local and global congurational space of magnetic nanoclusters.
Concerning the solution of the many-electron problem underlying the cluster behav-
ior we intent to employ the state-of-the-art DFT in the Kohn-Sham (KS) formalism [11].
DFT has proven its suitability and popularity as the most powerful method in the elec-
tronic structure theory. It has demonstrated to be a good compromise between accuracy
and computational time. It scales favorably with the system size O(N
3
) compared with
Hartree-Fock (HF) and high-level correlated ab initio methods. For example, the second
order perturbation theory by Mller and Plesset (MP2) [12] scales as O(N
5
). DFT has
a further advantage over HF since it is able to treat electron correlation eects. From
the implementation point of view, DFT is well suited for modern parallel computing and
linear-scaling techniques. Although the available computer resources are commonly one of
the main considerations when choosing one particular model over the others, the dominant
factor still hinges on the physical adequacy of the chosen model for the specic property
one is interested in. The spin-polarized version of DFT [13, 14] is a powerful tool to study
the properties displayed by the magnetic materials and to bring forth results which are
having the desired physical accuracy.
This thesis is devoted to study the unique properties exhibited by the TM magnetic
nanoclusters. The main goals are the following: i) The magnetic properties of FeRh and
10
Chapter 1. Introduction
CoPd nanoclusters have been investigated as a function of size, structure, and chemical
order. ii) We developed a novel method which combines spin-polarized DFT with the
basin-hopping (BH) global optimization method. iii) This approach has been successfully
employed to study the properties of pure and binary magnetic nanoclusters. iv) The spin
moments, orbital moments and magnetic anisotropy energy of fcc-like FeRh clusters having
N = 13 and 19 atoms have been determined as a function of composition. A remarkable
non-monotonous dependence of the MAE is observed as a function of Fe content, i.e., upon
going from pure Fe to pure Rh. This leads to an important increase of the MAE, which
reaches about 300% at the optimal Fe concentration.
In the rest of this chapter, a brief description on the magnetic nanoalloy clusters is
given and it is followed by a comparative study on the magnetic properties exhibited
by the materials in the bulk phase and in the reduced dimensionality such as nanowires,
nanoclusters, etc. The comparison is based on the spin moments, orbital moments, and the
MAE. The main purpose of this chapter is to review the concepts and the formalism that
will be useful for further discussion. In addition, we review two of the most widely used
magnetic characterization methods, namely, the Stern-Gerlach (SG) cluster experiment
and X-ray magnetic circular dichroism (XMCD).
1.1 Magnetism in the bulk phase
Magnetism in thermal equilibrium (not in steady states) is a quantum mechanical eect
which cannot be explained by classical statistical mechanics. Theory shows that the par-
tition function of a classical system is always independent of the vector potential eld

A [15]. Microscopically, the magnetism of solids originates nearly exclusively from elec-
trons. Nuclear moments contribute negligibly to the magnetization, but can be important,
for example, in resonance imaging. In magnetic materials, the atomic moments interact
with each other to form a variety of magnetically ordered structures. A bulk phase is
said to be ferromagnetic (FM) if all the atomic moments mutually align parallel. It is
antiferromagnetic (AF) when the atomic moments of two sub-lattices of opposite direc-
tion that cancel out each other. More complex non-collinear magnetic orders can also be
formed depending on the interactions between the atomic magnetic moments [22]. At high
temperatures the regular magnetic order is destroyed by spin uctuations and the bulk
becomes paramagnetic [23].
One of the key parameters to quantify the magnetism in the bulk is the mean saturation
magnetization per atom (MSM). The MSM as a function of the electron lling of the d
bands in TMs and their alloys can be represented by the renowned Slater-Pauling (SP)
curve (see Fig. 1.1). The most remarkable feature of this plot is its triangular shape
showing that most alloy species follow the two equal and opposite slopes. On the right
side of this plot, Co and Ni based alloys such as CoNi, NiCu, NiZn, and Co rich Fe-Co alloys
are located, while the BCC Fe based alloys such as FeV, FeCr, and Fe rich FeCo alloys are
on the left side. One can see from SP curve that as the total electron number decreases
from 28.5, where the majority and minority d-states are both lled, electron vacancies
appear only in the minority states. Therefore, magnetic moment which is dened as the
dierence in the number of majority and minority electrons, increases with decreasing
number of d-electrons at a rate of 1
B
(Bohr magneton) per electron. This leads to
a straight line on the right side with a slope of -45% degrees. In the case of FeNi, the
11
1.2. Magnetism in reduced dimensionality
Figure 1.1: Mean saturation magnetization for ferromagnetic alloys as a function of total
electron number (after Ref. [25]).
saturation moment drops sharply as the electron number decreases and approaches the
phase boundary. At the peak of this curve, located at 35% Ni in Fe, a very low thermal
expansion coecient is measured at room temperature. This alloy is called Invar, and
its peculiar thermal eect is called Invar eect. The name Invar comes from the word
invariable, referring to its lack of expansion or contraction with temperature changes [24].
1.2 Magnetism in reduced dimensionality
This section presents a compact description concerning some of the unusual magnetic
properties exhibited by materials in reduced dimensionality with a special focus on nan-
oclusters. In order to study the remarkable eects shown by the magnetic objects in the
reduced dimensionality, we are going to analyze qualitatively the dierences in electronic
structure, and the magnetic moment, in the case of Cobalt, when proceeding from one
single atom to the bulk phase (see Fig. 1.2).
Let us consider a Co atom having seven electrons in the 3d band and two electrons in
the 4s band. In order to satisfy the Hunds rst rule, the spin magnetic moment should
be 3
B
(see Fig. 1.2). Now we gradually add Co atoms one after the other to the rst Co
atom. This is called bottom-up approach and this have been utilized to model many novel
functional devices. The one by one addition of Co atoms will induce a tendency for atoms
to come closer by creating bonds with each other. As the atoms combine to form solid, sd
hybridization occurs and the 3d electrons participate in bonding. In fact, the localization
of d states near to the atomic core is the key reason for the magnetism exhibited by the
transition-metal atoms. Each Co atom provides seven electrons in the 3d band. This
is localized picture (Heisenberg model) and is based on the assumption that s-electrons
provide the metallic bonding and d-electrons generate the magnetic moments. This ap-
proach can successfully explain the Curie-Weiss susceptibilities and the existence of spin
waves in transition metals. However, it is not really successfully to explain the saturation
12
Chapter 1. Introduction
Figure 1.2: Illustration of the transition from atomic to bulk behavior in Co.
magnetization nor the magnetic moments deduced from Curie-Weiss constant [15].
In the case of bulk Co, the magnetic moment per atom amounts to 1.70
B
, which is
much lower than the magnetic moment of an isolated Co atom. In fact, the appearance of
non-integer value is due to the partial delocalization of the 3d electrons. That makes the
magnetic moments not entirely localized near to the atomic core. This kind of magnetism
can be explained by the Stoner model (itinerant electron model) [15, 23, 26]. This model
starts from Bloch-like electrons which are spread out through the system (localized in
reciprocal k space) and the electrons are assumed to have itinerant character, forming
broad sp bands and narrow d bands. The width of the band is inversely proportional to
the degree of localization of the atomic states, and increases almost continuously when
adding one atom after the other until it reaches the bulk value. This model can successfully
explain the fractional number of magnetic moments in transition metals, however, it gives
poor description of their thermal properties.
The models mentioned above are apparently insucient to study the unique magnetic
properties exhibited by the nanoclusters, since in clusters a large number of atoms are
on their surface. Consequently, there are signicant dierences in the electronic structure
of an atom on the surface compared to an atom inside the cluster. The reduced coor-
dination number on the cluster surface will reduce the eective hybridization of orbitals
that give rise to a prominent contribution to the total magnetic moment. We can restate
the above sentence as reducing the coordination number of a nanostructure can lead to
enhanced magnetic moments. This behavior can be quantied by the Stoner criterion [37],
which states that ferromagnetism occurs only when the product of the density of states
at the Fermi level D(E
F
) and the material dependent exchange integral I is larger than
1: D(E
F
) I > 1. In the bulk only Fe, Co and Ni fulll this condition, due to their high
d-density of states (DOS) at the Fermi level. A narrow d band width W
d
implies a larger
number of states at the Fermi level: W
d
1/D(E
F
). Tight binding calculations have
derived a simple relationship between W
d
and the local environment in nanostructures:
13
1.2. Magnetism in reduced dimensionality
Figure 1.3: Experimentally determined size dependence of the magnetic moment per atom
of Rh
N
clusters in the size range N = 934. Notice that the Rh
15
, Rh
16
and Rh
19
exhibit
particularly large magnetic moment. These results have been obtained by Cox et al. [51].
W
d
= 2

Z h
d
, where Z is the number of nearest neighbors and h
d
is an average hopping
matrix element. h
d
depends on the overlap of nearest neighbors d orbitals and is therefore
element specic. It is then clear that, by reducing the eective coordination number, the d
band width is reduced, with a subsequent increase of D(E
F
) (for more information about
this concept see Ref [4]). In particular, TM elements that are non-magnetic in bulk such
as Rh, and Pd etc might become magnetic in the reduced dimensions, for example, in
nanowires, nanoclusters, thin lms etc. [3850]
The rst magnetic measurements of 3d and 4d nanoclusters as a function of the number
of constituent atoms were performed in the beginning of the 1990s by the groups of W. de
Heer [30, 31] and L. Bloomeld [3234, 51]. Both groups determined the average magnetic
moments experimentally by using a SG magnet. In Ref. [51] magnetic moments of Rh
clusters were investigated, as a function of cluster size (see Fig. 1.3). It has been found
that, the average magnetic moment per atom increased signicantly by reducing the cluster
size. In addition, the magnetic moment exhibits an oscillating behavior, which is probably
due to ferromagnetic-like or antiferomagnetic-like coupling among atomic shells inside the
cluster.
First theoretical calculations on the magnetic properties of Rh clusters were performed
by Galicia [52] by using a molecular orbital approach. He computed a magnetic moment
of 1
B
per atom for Rh
13
cluster with bulk-like structure. Later, Reddy, Khanna, and
Dunlap [53] predicted the magnetic moment for Rh 13-atom clusters with icosahedral and
cubo-octahedral symmetry. They found 1.60
B
per atom for icosahedral Rh
13
and 1.46
for fcc Rh
13
. The rst theoretical calculations on the electronic structure and magnetic
properties were performed on small Fe and Ni clusters in the 1980s by Lee et al. [29].
They found narrowing of the d bands and consequent enhancement in spin polarization
compared to the bulk in the small Fe clusters. Shortly later, G. M. Pastor et al. [4]
extended these calculations by taking dierent geometric structures in the size range from
2 to about 50 atoms per cluster. This study shown that Fe clusters having less than 10
14
Chapter 1. Introduction
atoms present magnetic moments of about 3
B
with decreasing values for larger clusters.
These theoretical prediction where subsequently veried by the SG experiment of W. de
Heer et al. [6].
1.3 Magnetic characterization methods
A concise description of two important magnetic characterization methods used to study
the behavior of nanostructures are presented in the following. The rst one concerns the
working principle of a SG device, while the second one presents the XMCD technique.
1.3.1 Stern-Gerlach (SG) experiment
Stern-Gerlach experiment [54] have been used to study the magnetic moment of clusters
and established a very fundamental empirical fact: The electron has an intrinsic magnetic
moment
S
which adopt quantized values
B
= e/2mc along any axis,
B
is the Bohr
magneton. A schematic diagram of the experiment is shown in the Fig. 1.4.
Figure 1.4: Schematic diagram of the Stern-Gerlach cluster beam experiment. The cluster
beam is collimated before traveling down the length of a magnet with a strongly inho-
mogeneous eld along the z. (a) If the spatial components of angular momentum are
quantized, the beam will be split into distinct beams along z, resulting in separate spots
at the rear detection plane. (b) If angular momentum is classically distributed in space,
the beam will be evenly spread in z.
A compact description about the working principle is as follows. A beam of clusters
produced in a cluster source passes through an inhomogeneous magnetic eld along the z
direction. Due to the space quantization, the magnetic moment of clusters in the beam
will have quantized z components and will experience a quantized deecting force in the
z direction. After the magnet, the beam hits a screen so that the deected distribution of
clusters could be directly observed. After inspection of the detector screen, and in contrast
to the atomic experiment and initial expectations, a broadening of the cluster beam prole
and a shift of the distribution towards the direction of the increasing magnetic eld has
been observed.
The eective magnetic moment per atom
eff
can then be derived by assuming a
longer superparamagnetic relaxation, namely,
15
1.3. Magnetic characterization methods

eff
=
L
(
NB
k
B
T
)
=
[
coth
(
NB
k
B
T
)

k
B
T
NB
]
,

1.3.1
where N is the number of atoms in the cluster, k
B
is the Boltzmanns constant, T is the
temperature of the cluster ensemble, and B is the external magnetic eld.
1.3.2 X-ray magnetic circular dichroism
The XMCD technique has been experimentally developed by Sch utz et al. [55]. By using
XMCD one can probe the magnitude of both orbital and spin magnetic moments of specic
elements within a sample. Consequently, this versatile and powerful technique has become
well-known in recent years. XMCD can sample specic elements of FM ordered materials,
but it is unresponsive to AF order. This is because the signal is proportional to the average
magnetic moment. Therefore the signals from parallel and antiparallel orientated spins
cancel out. A schematic diagram of the density of states is shown in Fig. 1.5, depicting
the spin-dependent absorption as a single-electron two-step process.
Figure 1.5: Schematic representation of L edge X-ray absorption of a) a nonmagnetic
metal and b) a magnetic metal using right circularly polarized (RCP) and left circularly
polarized (LCP) light. The asymmetry in the spin up and spin down states give rise to
XMCD signal (after Ref. [27]).
The 2p core states are split by the spin-orbit interaction into the j = 3/2 and j = 1/2
levels. Right circularly polarized (RCP) photons excites preferentially spin-up electrons
at the p
3/2
(L
3
) edge because the orbital and spin angular momenta are parallel j = l +s,
while left circularly polarized (LCP) photons excites preferentially spin-down electrons at
the p
1/2
(L
2
) edge because the orbital and spin angular momenta are antiparallel j = l s.
16
Chapter 1. Introduction
Figure 1.6: Upper part: photo-absorption spectra of 7.5 nm FeCo particles deposited on
a Ni (111) lm grown on W (110). The spectra are taken in remanence with circularly
polarized radiation from the 2p core levels. The Fe and Co spectra have been enlarged by
a factor 30. Lower part: the corresponding XMCD data given by the intensity dierences
(after Ref. [59]).
These dierences are the reason behind the L
2,3
XMCD spectrum. The L
3
and L
2
peaks
are opposite in sign. The sum of the white line intensities, denoted I
L3
and I
L2
respectively,
is directly proportional to the number of d holes.
The orbital and spin moments are related to the absorption spectra by

L
=
B
L
z
= 2n
h

L
3
+L
2

dE

L
3
+L
2

+
+
0
+

dE
,

1.3.2
and

S
= 2
B
(
S
z
+
7
2
T
z

)
=
3
2
n
h

L
3

dE 2

L
2

dE

L
3
+L
2

+
+
0
+

dE
.

1.3.3
The method measures the quantities L
z
and S
z
+ 7T
z
per valence band hole n
h
,
where L
z
and S
z
are the expectation values of the z components of the orbital and
spin angular momenta of the atoms and T
z
is the expectation value of the magnetic
quadrupole operator. As an example, the XMCD spectra of FeCo particles deposited on
a Ni (111) lm grown on W (110) are shown in Fig. 1.6
17
1.4. Thesis outline
1.4 Thesis outline
As discussed in the previous sections, the 3d-4d and 3d-5d magnetic nanoalloy clusters
(e.g., FeRh, CoPd etc.) exhibit special properties and oer many interesting features for
future magnetic nanometer device applications. In this thesis, we have investigated the
magnetic properties of these clusters by employing the state-of-the-art density functional
theory (DFT) method. We have used dierent optimization methods such as graph theory
and basin hopping in order to study the complex energy landscape of these clusters. Some
remarkable results have been observed from these studies. For instance, in the FeRh dimer
we found an important transfer of d electrons from Fe 3d to Rh 4d orbital, which allows
the Fe atom to develop very large spin moment, due to the larger number of d holes.
The rest of the thesis is organized as follows. Chapters 2 provides the background
information on theoretical and computational procedures we used in this thesis. For the
sake of clarity we begin with Hatree-Fock wave function method. Then we focus on
DFT in which framework all the calculations reported in this thesis have been performed.
Dierent approximations to the exchange and correlation functional are discussed in more
detail: the local density approximation (LDA), the generalized gradient approximation
(GGA). Chapter 3 is mainly dedicated to review the various local and global optimization
methods used in present-day calculations. Among the local optimization methods we nd
the steepest descent (SD) and the conjugate gradient (CG) schemes, while the global
optimization methods include the graph theory, simulated annealing (SA), basin-hopping
(BH), and genetic algorithms (GA). Chapter 4 and 5 presents the main results we obtained
for FeRh and CoPd nanoalloy clusters. As already discussed these studies were motivated
by the peculiar magnetic properties exhibited by the pure clusters of Fe, Co, Rh, and Pd
and the possibilities of tailoring them through alloying. Indeed, a number of interesting
new eects are revealed by these calculations. Chapter 6 presents the results of our
combined spin-polarized DFT and BH approach on some representative pure and alloy
magnetic clusters. Chapter 7 analyzes the spin moments, orbital moments and the MAE
of the FeRh clusters having 13 and 19 atoms. In this study we focus on the eect of
size, and the composition on the magnetic properties. Finally, Chapter 8 summarizes our
conclusions and outlines some relevant directions for future work in this eld.
18
2
Background on electronic structure theory
2.1 The quantum many-body problem
Atoms, clusters and solids are composed of mutually interacting electrons and nuclei. The
precise determination of the electronic structure is a very dicult task due to the number
of electrons involved, which renders an analytic solution impossible for systems with more
than one electron. Obviously, the complexity grows dramatically with increasing number
of electrons.
The time-independent many-body Schrodinger equation can be written as

H(r
1
, r
2
, ...r
N
) = E(r
1
, r
2
, ...r
N
),

2.1.1
where

H is the Hamiltonian, (r
1
, r
2
, ...r
N
) is the many-body wavefunction and E is the
total energy of the system [69]. The Hamiltonian in atomic units can be written as

H =
1
2m
z
i
M

i=1

2
R
i

1
2
N

i=1

2
r
i
+
M

i
M

j>i
Z
i
Z
j
|R
i
R
j
|

i=1
M

j=1
Z
i
|r
i
R
j
|
+
N

i=1
N

j>i
1
|r
i
r
j
|
,

2.1.2
where M and N are total number of nuclei and electrons in the system, m
z
, Z and R are
the mass, charge and position of the nuclei, and r
i
represents the position of the electron
i. The rst two terms in Eq. 2.1.2 are the kinetic energy of the nuclei and the electrons,
respectively. The remaining three terms are Coulomb energies, representing the ion-ion re-
pulsion, the ion-electron attraction and the electron-electron repulsion, respectively. The
Schrodinger Eq. 2.1.1 with the Hamiltonian given above is impossible to be solved ana-
lytically except for selected single-electron problems. Therefore, it is unavoidable to make
approximations for both the Hamiltonian and the many-body wavefunction .
19
2.1. The quantum many-body problem
The rst major simplication is introduced by the Born-Oppenheimer (BO) approxi-
mation, which allows to decouple the electronic and ionic degrees of freedom due to the
large dierence in mass between the electrons m
e
and the ions M
I
(around 10
3
) [70].
In the BO approximation the ions behave like classical particles. In other words, ions
are regarded as if they were at rest at some xed positions, while the electrons are mov-
ing within a xed external potential due to the nuclei. With the BO approximation the
Hamiltonian can be simplied to a electronic model. In the non-relativistic approximation
it is given by

H =
1
2
N

i=1

2
r
i

N

i=1
M

j=1
Z
j
|r
i
R
j
|
+
N

i=1
N

j>i
1
|r
i
r
j
|
.

2.1.3
Despite this assumption the electronic Schrodinger equation, remains extremely com-
plicated to solve for most of the realistic situations. The next sections give a short review of
the most famous eective eld approaches, namely Hartree-Fock (HF) theory and density
functional theory (DFT).
2.1.1 Hartree-Fock Theory
Fock formulated an approximation [74] in order to deal with the problem associated with
the Hatree approximation by writing the wavefunction as a fully antisymmetrised product
of single-orbital states:

HF
=
1

N!

1
(r
1
s
1
)
2
(r
1
s
1
)
N
(r
1
s
1
)

1
(r
2
s
2
)
2
(r
2
s
2
)
N
(r
2
s
2
)
.
.
.
.
.
.
.
.
.

1
(r
N
s
N
)
2
(r
N
s
N
)
N
(r
N
s
N
)

2.1.4

HF
=
1

N!
det [
1
(r
1
s
1
)
2
(r
2
s
2
)
N
(r
N
s
N
)]

2.1.5
This ansatz fullls the antisymmetry condition and known as a Slater determinant [75].
The Eq. 2.1.4 has the desired full antisymmetry property, since interchanging the position
of two electrons is equivalent to interchanging the corresponding columns in the determi-
nant, which changes its sign. The orbitals are subject to the orthonormality conditions

i
(r)
j
(r)dr =
i
|
j
=
ij
,

2.1.6
The HF energy can be evaluated by taking the expectation value of the Hamiltonian
with the above Slater determinant. This yields:
20
Chapter 2. Background on electronic structure theory
E
HF
=
HF
|

H|
HF

=
N

1
2
|
i
(r)|
2

j=1
Z
j
|r R
j
|
|
i
(r)|
2

d
3
r
+
1
2
N

i,j,i=j

|
i
(r)|
2
1
|r r

|
|
i
(r

)|
2
d
3
rd
3
r

1
2
N

i,j,i=j

s
zi
s
zj

i
(r)

j
(r

)
1
|r r

|

j
(r)
i
(r

) d
3
r d
3
r

The rst term is the kinetic energy of the electrons, the second one is the electron-ion
interaction and the third one is the interaction between electrons. The last term arises
from the antisymmetric nature of the HF wavefunction. It vanishes when s
i
= s
j
. It is
known as the exchange energy E.
Consider a spin-up electron at position r
i
. Then, according to the Paulis exclusion
principle, all the other spin-up electrons will be eectively moved away from the position
r
i
. On the other hand, the spin-down electrons will not be repelled by the Pauli principle,
or more generally by the antisymmetry of the wavefunction, since they have a dierent
spin quantum number. In fact there is a sort of enhancement of the probability of nding
two electrons on the same point r
i
when they form a singlet spin state (antiparallel spins).
Thus spin-up electron is encircled by a space which has been depleted of other spin-up
electrons. Similarly, for a spin-down electron there is a space depleted of other spin-down
electrons. This region is called the exchange hole.
In fact, the electronic motion in realistic systems are correlated beyond the predic-
tions of the HF formalism. The interaction energy neglected by HF is known as the
correlation energy E
c
[76].
E
c
= E
0
E
HF
,

2.1.7
where E
0
is the exact ground state energy [77]. Taking into account the variational prin-
ciple it is clear that E
c
< 0.
The electronic structure calculations which neglect electron correlation are found to
produce signicant deviations from experimental results. One way to bring in the cor-
relation eects is to construct linear combinations of Slater determinants corresponding
to excited-state congurations. These methods are conjointly called post Hartree-Fock
methods, e.g., the conguration interaction (CI), the coupled cluster (CC) and Mller-
Plesset (MP) methods. These have allowed one to include electron correlations into the
multi-electron wave function. However, these implementations require a prohibitively huge
computer time. In practice they do not allow to study systems having more than 10 elec-
trons [78].
Another kind of hole (a dynamic electron density depletion) is present in the neighbor-
hood of any electron, due to the electronic correlation beyond HF. This electron-depleted
region that surrounds each electron is known as the Coulomb hole or correlation hole.
The HF approach is an improvement over the Hartree theory. It is often applied as
a rst approximation to oxides and crystals of small organic molecules, which have small
number of localized electrons. For transition metals, it is not appropriate, in particular
because of the high electron density. The HF method cannot be used for metals, as it
21
2.2. Density Functional theory
neglects the collective Coulomb screening in a completely delocalized electron system. By
the way, one can show that HF DOS is always zero precisely at the Fermi energy. Thus,
the HF approximation can never describe metallic behavior.
2.1.2 The Thomas-Fermi Theory
Thomas and Fermi (TF) introduced theory [79, 80], which can be considered as the rst
DFT, since it employs electron density n(r), rather than the many-electron wavefunction,
as the fundamental unknown of the many-body problem. The total energy functional can
be written as
E
TF
[n(r)] = C
F

n(r)
5/3
d
3
r +

n(r) v
ext
(r) d
3
r +
1
2

n(r)n(r

)
|r r

|
d
3
r d
3
r

2.1.8
The rst term is the kinetic energy of the non-interacting electrons in a homogeneous
electron gas with density n. The second term is the electrostatic interaction energy between
nuclei and electrons, where v
ext
(r) is the Coulomb potential
v
ext
(r) =
M

j=1
Z
j
|r R
j
|
.

2.1.9
generated by the nuclei. Finally, the third term is the classical Coulomb repulsion between
electrons, known as the Hartree energy. The TF model is simple and appealing, but it
should not be used when quantitative predictions on realistic systems are desired. Its main
failure comes from the incorrect approximation of the kinetic energy. Another problem
is the over-simplied description of the electron-electron interactions, which are treated
classically and hence do not take account of quantum phenomena such as the exchange
interaction [81, 82]. In fact TF theory has been rapidly abandoned since one can shows
that it predicts that the energy of isolated atoms is always lower than the energy of
any molecule. Thus the TF theory cannot predict any molecular bonding. Still, the
idea of replacing the wavefunction (r
1
s
1
, r
N
s
N
) by the density of electrons n(r) is a
remarkable approach, which later on provided a major breakthrough.
2.2 Density Functional theory
Finding the approximate solution by taking advantage of modern computational power
seems not surprising nowadays. Diracs famous statement concerning the equations gov-
erning chemistry are still authoritative today, if one takes for granted that the Schrodinger
equation can be directly numerically solved. DFT appears as an alternative method to
the theory of electronic structure, in which the electron density n(r), rather than the
many-body wave function , plays the key role [84]. This idea is proven to be excellent in
all respects, since it represents an exact ground-state quantum theory which is alternative
to the wavefunction approach. Walter Kohn has been indeed awarded the Nobel Prize
in Chemistry in 1998 for his remarkable work in this eld, shared with John Pople, who
contributed with the quantum chemistry approach. Nevertheless, although in principle
exact, the theory involves an universal functional F
HK
[n], which is in general not known
and which must be therefore approximated.
22
Chapter 2. Background on electronic structure theory
DFT deals with systems of identical particles [85], basically fermions, providing a sim-
ple method for describing the eects of exchange and correlation in an inhomogeneous
electron gas. The minimum of the total energy as a functional of n(r) is the ground-state
energy of the system, and the density that yields this minimum value is in principle the
exact ground-state density n(r). Kohn and Sham [11] showed how to replace the many-
electron problem by an exactly equivalent set of self-consistent one-electron equations.
Furthermore, they showed that all other ground-state properties of the system (e.g. cohe-
sive energy, lattice constant, etc.) are functionals of the ground-state electron density [87].
They derived eigenvalue equations from the variational approach which take a simple form
and which are analogous to the HF equation. These so-called Kohn-Sham equations are
actually simpler to solve than HF ones.
DFT has had a huge impact on realistic calculations of the properties of molecules
and solids, and its applications to diverse systems will continue to grow. There are state-
of-the-art applications of DFT in wide variety of areas, including magnetism, catalysis,
surface science, nanomaterials, biomaterials and geophysics [93]. DFT has been applied
to degenerate ground states [88], spin-polarized ground states [84, 88], quantum Hall eect
[94], etc. However, in some special cases DFT predicts wrong results. For example,
DFT has limited accuracy in the calculation of excited states. A particularly famous
example is the underestimation of band gaps in semiconductors and insulating materials
[87]. Among the failures of DFT, with the functionals known nowadays, one may mention
the larger binding energies in LDA, inaccurate results associated with weak van der Waals
forces, the Kohn-Sham potential decays exponentially for large distances instead of 1/r,
and strongly correlated solids such as NiO and FeO are predicted as metals and not as
antiferromagnetic insulators [97].
2.2.1 The Hohenberg-Kohn theorems
The formal footing of the DFT is provided by two theorems demonstrated by Hohenberg
and Kohn in 1964.
Theorem I: The ground-state density n(r) of a many-body quantum system in some
external potential v
ext
(r) determines this potential uniquely.
This means that n(r) determines N and v
ext
(r), and hence all properties of the ground-
state, for example, the kinetic energy T[n(r)], the potential energy V [n(r)] =

n(r)v
ext
(r)
and the electron-electron interaction energy U
ee
[n(r)]. The total energy E[n(r)] [99] is
given by
E[n(r)] = V [n(r)] +T[n] +U
ee
[n]

2.2.1
We can group together all functionals which do not involve v
ext
(r) as
E[n(r)] = V [n] +F
HK
[n] =

n(r) v
ext
[n] d
3
r +F
HK
[n],

2.2.2
where F
HK
is known as the Hohenberg-Kohn functional.
Theorem II: For any trial density n(r) it holds E
0
E[n(r)], where E
0
is the ground-
state energy of the system.
In other words, the minimum value of the total-energy functional E[n(r)] is the ground-
state energy of the system, and the density which yields this minimum value is the exact
ground-state density of the many-body system [99].
23
2.2. Density Functional theory
2.2.2 The Kohn-Sham theory
Although providing the theoretical foundations of DFT, the HK work of 1964 does not
unravel an explicit way to solve the many-body problem. One year after, Kohn and Sham
(KS) developed this important scheme in practice. The KS formulation [11] is based on
mapping the full interacting electronic system onto a ctitious non-interacting system,
so that the complex many-body problem can be transformed into a set of self-consistent
single-particle equations.
2.2.3 Kohn-Sham equations
The TF idea of obtaining the ground-state kinetic energy from a non-interacting system is
also the starting point for the KS scheme. In order to assess the kinetic energy of N non-
interacting particles given only their density distribution n(r), the corresponding potential
v
s
[r] yielding the exact ground-state density in the absence of interactions is introduced.
The single particle Schrodinger equation reads
(

1
2

2
+v
s
(r)
)

i
(r) =
i

i
(r)

2.2.3
So that
n(r) =
N

i=1
|
i
(r)|
2
.

2.2.4
Since the potential v
s
(r) is a functional of the density, then Eqs. 2.2.3 and 2.2.4 have
to be solved self-consistently.
The KS total-energy functional for a set of occupied electronic states
i
can be written
as
E[n(r)] =

i
d
3
r +

V
ion
(r) n(r) d
3
r +
1
2

n(r)n(r

)
|r r

|
d
3
r d
3
r

+E
xc
[n(r)] +E
ion
(R
I
),
where E
ion
stands for the Coulomb energy associated with interactions between the nuclei
(or ions) at the positions R
I
and V
ion
is the electron-ion energy, and E
xc
[n(r)] is the
exchange and correlation (XC) energy.
Strictly speaking only the minimum of the KS energy functional has physical meaning,
corresponding to the ground-state energy of the system of electrons with the ions at
positions R
I
[100]. It is essential to nd out the set of wave functions
i
that minimize
the KS energy functional. These are given by the self-consistent KS equations [11], which
take the form
{

1
2

2
+V
ion
(r) +V
H
(r) +V
XC
(r)
}

i
(r) =
i

i
(r).

2.2.5
As before,
i
denotes the wave function of electronic state i and
i
the corresponding KS
eigenvalue. The terms within the brackets can be regarded as an eective single-particle
Hamiltonian, where V
H
is the Hartree potential given by
24
Chapter 2. Background on electronic structure theory
V
H
(r) =

n(r

)
|r r

|
d
3
r,

2.2.6
and the exchange and correlation potential
V
XC
(r) =
E
xc
[n(r

)]
n(r)

2.2.7
is formally given by the functional derivative of the exchange and correlation energy with
respect to the density.
The KS equations must be solved self-consistently so that the occupied electronic states
render a charge density that produces the electronic potential that is used to construct the
equations. The sum of the single-particle KS eigenvalues does not give the total electronic
energy because of some kind of double counting of the eects of the electron-electron
interaction in the Hartree energy and in the exchange-correlation energy [101].
2.3 Exchange-correlation energy
The exchange and correlation (XC) potential V
XC
is a functional derivative of the ex-
change and correlation energy with respect to the local density [see Eq. 2.2.7]. For a
homogeneous electron gas, this will only depend on the value of the electron density. For
a non-homogeneous system, the value of the exchange correlation potential at the point
r depends not only on the value of the density at r, but also on its variation close to
r, where

close

is a microscopic distance of comparable magnitude as the local Fermi

wavelength or the TF screening length [78]. It can therefore be written as an expansion
over the gradients to arbitrary order of the density:
V
xc
[n(r)] = V
xc
[n(r), n(r), (n(r)), ....].

2.3.1
The exact form of the energy functional is, so far, unknown, as the inclusion of density
gradients makes the solution of the DFT equations dicult. The simplest way to obtain
this contribution is to assume that the exchange and correlation energy leads to an XC
potential at point r
i
that depends only on the value of the density n(r) at r, i.e., not on
its gradient. This is known as the local density approximation (LDA).
2.3.1 Exchange and correlational functionals
Some of the main XC functionals used nowadays are the following:
The LDA which takes into account only the local aspects of the density.
The generalized gradient approximation (GGA) in which the dependence on the
gradients of the density is further added.
The meta-GGA (MGGA), which including the dependence on the kinetic energy
density.
The hybrid functionals, in which exchange HF-like contributions are added to the
dependence on the occupied orbitals. An example of this kind is the exact exchange
approach.
The fully non-local functionals, in which more complex dependence on the unoccu-
pied orbitals are incorporated.
25
2.3. Exchange-correlation energy
The local density approximation
LDA is a simplest approximation to quantify the exchange and correlation energy [11].
In this approximation the XC energy of an electronic system is constructed by assuming
that the XC energy per electron at a point r in the electron gas
xc
[n(r)] is equal to the
XC energy per electron in a homogeneous electron gas having the same (homogeneous)
density n(r). Thus,
E
LDA
xc
[n(r)]

xc
[n(r)]n(r)dr

2.3.2
and
E
xc
[n(r)]
n(r)
=
[n(r)
xc
(r)]
n(r)

2.3.3
with

xc
(r) =
hom
xc
[n(r)].

2.3.4
Numerical correlated calculations allow one to know
xc
(r) to a very high accuracy. Phys-
ically, one expects LDA to become exact when the length scale over which n(r) varies is
very large [84].
By construction LDA is exact for a uniform electron gas. In other words, it should be
a good approximation for slowly varying densities. Surprisingly, the LDA has proven to
yield very good results in many applications, even for atomic systems where the hypoth-
esis slowly varying density evidently violated [100]. This is due to the fact that the LDA
shows many formal features, such as the sum rule for the exchange and correlation hole.
However, the LDA also has several important failures. For example, the LDA systemat-
ically underestimates the band gap in semiconductors. In the case of Ge the calculated
band gap is even negative, which erroneously indicates that Ge should be metal, and not
a semiconductor [100].
The generalized gradient approximation
The next level of approximation beyond LDA is given by a number of non-local approaches.
They are usually termed as generalized gradient approximations (GGAs). They take the
form
E
GGA
XC
=

f(n(r, |n(r)|))dr,

2.3.5
in which f(n(r, |n(r)|)) is a suitably chosen function of n(r) and n(r) [84].
Even though GGAs should render a systematic improvements over the LDA, it remains
a main problem that the gradients in real materials can be so large that the expansion
breaks down. By imposing conditions on the correct exchange hole given by the gradient
expansion, Perdew [89] proposed a model which leaves only a 1% error in exchange energy.
This model has also been further simplied [90] to
E
GGA
XC
= C
1

dr
3
n
4/3
F
x
(s),

2.3.6
with
26
Chapter 2. Background on electronic structure theory
s =
|n(r)|
2k
F
n
,
k
F
= C
2
n
1/3
.
and
F
x
(s) = (1 + 1.29s
2
+ 14s
4
+ 0.2s
6
)
1/15
,

2.3.7
C
1
and C
2
are constants. Note that the LDA corresponds to F
x
(s) = 1. However, several
other forms for F
x
(s) have been suggested [91].
The most widely used XC functionals have been proposed by Becke (B88) [115], Perdew
and Wang (PW91), and Perdew, Burke, and Enzerhof (PBE) [116]. Finally, a useful
collection of explicit expressions for some GGAs can be found in the appendix of Ref. [92].
2.4 Projector augmented wave method
In order to compute the electronic structure of a system using the DFT method, the Kohn-
Sham equations have to be solved in some ecient numerical way. One of the central
issues concerning eciency is the rather dierent behavior of the electronic wavefunction
at dierent distances from the atomic nuclei. i.e., the dierent behavior of the outer
valance electrons and the inner core states. Indeed, the atomic wavefunctions should be
mutually orthogonal. In order to comply with this condition, the valence wavefunctions
oscillate rapidly in the core region, making it very challenging to describe them precisely
without using a very large basis set. The core electrons are located very near to the
atomic nucleus and therefore do not contribute to the chemical bond formation and to
magnetism. It is clear that focusing on the valence states signicantly reduces the actual
computational time.
A widely used method to handle the core electrons is to introduce pseudopotentials by
which nuclei and core electrons are replaced. This ideally smooth potential is constructed
in order to reproduce the correct eect on the remaining valence electrons. The pseudopo-
tentials are computed and tabulated once for each element. The Kohn-Sham equations
then apply only to the valence electrons. Consequently the computational time signi-
cantly reduced. However, this simplication has mainly two major disadvantages. First,
all information about the true wavefunction close to the atomic nuclei is lost, rendering
it impossible to determine any property which depends on the core region (e.g, electric
eld gradients, hyperne parameters, etc). A second major limitation is that there is no
systematic way to construct reliable transferable pseudopotentials.
Another commonly used scheme is the all-electron method, in which all the information
about the wavefunction is available. This method is normally connected to the frozen-
core approximation, in which the core orbitals are computed and tabulated once and kept
xed. This is justied by the fact that the core states do not participate in the formation
of chemical bonds. One of the most important of such schemes is the augmented plane
wave method (APW) introduced by Slater in 1937. In this scheme the space is partitioned
in two regions: a sphere one around each atom in which the wavefunction is expanded
onto a local basis that is able to reproduce the fast variations, and an interstitial region
in which a plane wave basis is used. The single particle wavefunctions are of matched
at the surface of the sphere according to the usual continuity conditions. A recently
27
2.4. Projector augmented wave method
more widely spread scheme is the projector augmented wave (PAW) method developed
by P. Blochl [170] as an extension to both the APW and pseudopotential methods, and
which in fact can be retrieved by well-dened approximations [106]. The PAW method
unies all electron and pseudopotential approaches. In the following section we briey
present this formalism, since all the calculations contained in this thesis have been within
its framework as implemented in the Vienna ab-initio simulation package (VASP) by G.
Kresse and D. Joubert [168].
2.4.1 The PAW transformation operator
The dierent shape or behavior of the wavefunctions in dierent regions points towards
the need for a proper partitioning of the space around the nuclei. The PAW method takes
into account this and separates the wavefunction in two parts: a partial wave expansion
within an atom-centered sphere, and an envelope function outside. The two parts are then
matched smoothly at the sphere edge.
We seek a linear unitary transformation

T which maps some computationally con-
venient pseudo or smooth auxiliary wavefunctions |

to the physically relevant true or

all-electron wavefunctions |:
|
n
=

T |

n
,

2.4.1
where n is a quantum-state label, including the band index, spin and k-vector index. The
ground-state pseudo-wavefunction is then obtained by solving the Kohn-Sham equations
in the transformed Hilbert space

H

T |

n
=
n

T

T |

2.4.2
Since the true wavefunctions are smooth enough at a denite distance from the core, we
anticipate that the transformation is just the unity operator beyond the augmentation
cut-o and a sum of atom-centered contributions inside:

T = 1 +

T
a
,

2.4.3
where a is an atom index and

T
a
=

T
a
(r R
a
) = 0 for |r R
a
| > r
a
c
. The cut-o radii
r
a
c
is such that there is no overlap between augmentation spheres.
Within the augmentation region
a
, we expand the pseudo wavefunction into partial
waves

a
as
|

n
=

ia
c
a
ni
|

a
i
.

2.4.4
Similarly, for the all-electron counterpart we have
|
n
=

ia
c
a
ni
|
a
i

2.4.5
within
a
. By applying Eq. (2.4.1) we obtain
|
a
i
= (1 +

T
a
) |

a
i


T
a
|

a
i
= |
a
i
|

a
i
,

2.4.6
28
Chapter 2. Background on electronic structure theory
for all a and i, which fully determines the transformation

T in terms of the partial waves.
Hence we can express the true wavefunction as
|
n
= |

ia
c
a
ni
|

a
i
+

ia
c
a
ni
|
a
i

2.4.7
with the expansion coecients to be determined.
Since

T is linear, the coecients must be linear functionals of the pseudo wavefunction
|

n
, i.e., scalar products
c
a
ni
= p
a
i
|

n
P
a
ni
,

2.4.8
where p
a
i
are some xed functions. They are known as projector functions. The notation
P
a
ni
is consistent with existing literature.
Since there is zero overlap between dierent augmentation spheres, the one-center
expansion of a real pseudo wavefunction

i
p
a
i
|

n
|

a
i
has to be identical to |

a
i
inside
the augmentation sphere. This is equivalent to fullling the completeness relation

i
|

a
i
p
a
i
| = 1

2.4.9
within
a
which in turn implies that

p
a
i
1
|

a
i
2
=
i
1
,i
2

2.4.10
within
a
. In other words the pseudo functions and the partial waves are mutually
orthonormal within the augmentation sphere.
Finally, by inserting Eq. 2.4.8 into Eq. 2.4.7 we obtain a closed form for the transfor-
mation operator

T =

i
(|
a
i
|

a
i
) p
a
i
| ,

2.4.11
which permits us to obtain the true, all-electron, Kohn-Sham wavefunction
n
(r) = r|
n

as

n
(r) =

n
(r) +

i
(
a
i
(r)

a
i
(r)) p
a
i
|

n
.

2.4.12
It is usually convenient to bring in the one center expansions

a
n
(r) =

a
i
(r) p
a
i
|

a
n
(r) =

a
i
(r) p
a
i
|

2.4.13
Then, the true wavefunction can be written as

n
(r) =

n
(r) +

a
(
a
n
(r)

a
n
(r),

2.4.14
which split the extended-space and the atom-centered contributions. This is frequently
employed to obtain compact expression for various quantities in PAW. The rst term can
be evaluated on an extended grid, or on a soft basis set, while the last two terms are
evaluated on ne radial grids.
In summary, the three ingredients that determine the PAW transformation are:
29
2.5. Magnetic anisotropy energy
The partial waves
a
i
, which are obtained as solutions of the Schrodinger equation for
the isolated atom and utilized as an atomic basis for the all-electron wavefunctions
within the augmentation sphere.
The smooth pseudo partial waves

a
i
, which match with the corresponding true
partial waves outside the augmentation sphere but are smooth continuations inside
the spheres. These are employed as atomic basis-sets for the pseudo wavefunctions.
The smooth projector pseudo functions p
a
i
, one for each partial wave, which satisfy
the condition p
a
i
1
|

a
i
2
=
i
1
,i
2
inside each augmentation sphere.
2.4.2 Approximations
Up to this point the PAW method may be considered as an exact implementation of the
DFT. In order to make it a practical scheme, the following three approximations are re-
quired.
i ) Frozen Core
The frozen-core approximation assumes that the core states are localized in the
augmentation spheres and that the core states are not modied by the formation of
chemical bonds. The core Kohn-Sham states are thus chosen to be exactly the core
states of the isolated atoms:
|
c
n
= |
a,core

2.4.15
Notice that, in contrast to Eq. 2.4.5, no projector functions need to be specied for
the core states.
ii ) Finite basis set
The extended pseudo contribution

n
in Eq. 2.4.14 is measured outside the augmen-
tation spheres by using an appropriate basis set or on a real-space grid. In both
cases the lack of completeness in the basis, or equivalently the nite grid-spacing,
will introduce some controllable error.
iii ) Finite number of partial waves and projectors
The number of partial waves and projector functions is obviously nite. This means
that that the completeness condition we have assumed is not strictly fullled. How-
ever the approximation can be controlled by increasing the number of partial waves
and projectors, so that they form a satisfactory complete space for the expansion of
the wavefunctions within the augmentation spheres.
More detailed information about the PAW method may be found in Ref. [170].
2.5 Magnetic anisotropy energy
Magnetic anisotropy can be dened as the change of the total energy of a magnetic system
as a function of the orientation of the magnetization

M with respect to the crystalline axis.

30
Chapter 2. Background on electronic structure theory
Figure 2.1: Denition of direction cosine. Taken from Ref. [25].
This means that there are some unique directions in space in which a magnetic material
is easy or dicult to magnetize. The former is called easy axis for

M, whereas the latter
is called hard axis for

M. The dierence in the total energy between easy and hard axis
magnetization is known as magnetic anisotropy energy (MAE).
The main sources of the magnetic anisotropies are following:
Magneto-crystalline anisotropy: The magnetization is oriented along specic crys-
talline axes.
Shape anisotropy: The magnetization is aected by the macroscopic shape of the
nanoparticle.
Induced magnetic anisotropy: Specic magnetization directions can be stabilized by
tempering the sample in an external magnetic eld.
Stress anisotropy or magnetostriction: Leading to a spontaneous deformation.
Surface and interface anisotropy: Surfaces and interfaces often exhibit dierent mag-
netic properties compared to the bulk due to their asymmetric environment.
As an example, we review the magnetic anisotropy exhibited by the cubic crystals. In
cubic crystals, the magneto-crystalline anisotropy energy is given by a series expansion in
terms of the angles between the direction of magnetization (
1
,
2
,
3
) and the axes of the
cube (see Fig. 2.1). In a spherical coordinate system the direction cosines are given by

1
= sincos,
2
= sinsin, and
3
= cos,

2.5.1
with and being the polar and azimuthal angles, respectively.
The energy per unit volume E
MCA
can be expanded phenomenologically as a power
series of the direction cosines
1
,
2
and
3
:
E
cubic
MCA
= K
0
+K
1
(
2
1

2
2
+
2
1

2
3
+
2
2

2
3
) +K
2

2
1

2
2

2
3
+...

2.5.2
The corresponding phenomenological parameters are obtained by inserting the direc-
tion cosine into Eq. (2.5.2):
31
2.5. Magnetic anisotropy energy
E
100
= K
0
,

2.5.3
E
110
= K
0
+
1
4
K
1
,

2.5.4
and
E
111
= K
0
+
1
3
K
1
+
1
27
K
2
.

2.5.5
The K
n
are known as the anisotropy constants. The MAE in the cubic 3d transition
metals is of the order of a few eV/atom. Typical values of K
2
for bulk 3d metals
are 4.02 eV/atom for Fe, 45 eV/atom for Co, and -0.3 eV/atom for Ni [117]. On
the other hand, nanostructured materials such as monolayers, chains, clusters, or even
single magnetic atoms deposited on non-magnetic surfaces, the MAE is of the order of
meV/atom, and it can be as high as 20 meV/atom as in the case of Co/Pt(111) [50].
One of the major microscopic sources of MAE is the spin-orbit (SO) interaction. The
SO term is a relativistic correction to the Hamiltonian and is given by

H
SO
= S.L
which couples the spin operator S with the lattice, represented by the orbital momentum
operator L. The spin operator

S is proportional to the Pauli matrices and is the spin-
orbit coupling constant.

H
SO
depends on both and the magnitude of the L. is energy
dependent, whereas L strongly depends on the local environment of the atom. For example,
in bulk materials the orbital moment for 3d electrons is quenched, due to the crystal eld
interaction.
The other main contributions to the MAE is the dipolar interaction which is related to
the shape of the object under consideration. The shape anisotropy is found in materials
having non-spherical form. It has a classical macroscopic analogue: since dipoles sponta-
neously align such that the north-pole of the rst one is pointing to the south pole of the
second one, to generate a chain. Thus, in case of a chain of atoms the shape anisotropy
would lead to an easy axis along the chain axis. In the case of a thin lm, the shape
anisotropy gives an in-plane easy axis.
According to the Brunos [118] model based on the second-order perturbation theory,
the MAE is due to spin-orbit interaction (no shape anisotropy) and is proportional to the
anisotropy of the orbital moment. Thus, magneto-crystalline anisotropy can be derived as
E =

4
B
L

2.5.6
where is the spin-orbit parameter and L is the dierence in orbital moments between
the easy and hard axis.
Chapter 7 is devoted to the determination of MAE in nanoclusters. In fact, tight-
binding method including spin-orbit coupling [2], have proved to be suitable to provide
the correct trends for the MAE. Although, these methods can provide the correct trends,
they do not yield accurate values for MAE. The accurate computation of MAE can be
fullled by employing rst principles methods, since they provide both the accurate values
and systematic trends, even though, they are computationally very expensive to perform.
In order to determine the MAE we followed some specic steps. Firstly, we conned the
rotation of the magnetization vector

M with respect to the cluster frame. This allows
to restrict the calculations into a 2D coordinate system dened by the plane, where the
polar angle is taken as the angle between the magnetization and the easy axis direction,
32
Chapter 2. Background on electronic structure theory
(i) 13 atom cluster (ii) 19 atom cluster
Figure 2.2: The fcc clusters with N = 13 and 19 atoms showing the zx plane in which the
magnetization angle is varied.
usually chosen as the z axis. Moreover, has been increased in the steps of 10 degrees
until

M completes one full rotation around the cluster.
Fig. 2.2 illustrates our procedure to determine the MAE. For = 0,

M is parallel to
the z axis. When is gradually increased,

M passes through the middle of a bond and
also through the center of a triangular facet. It points along a bond axis for = /2.
In the calculations reported here we employed the so called magnetic force theorem [227].
This implies that we estimated the total energy dierence by the corresponding dierence
of the sum of the KS eigenvalues.
33
3
Exploring the ground-state energy surface of
nanoclusters
The potential energy surface (PES) represents the energy of a molecule or cluster as a
function of its geometry. For a molecule having N atoms, there are 3N independent
translational degrees of freedom corresponding to each atom minus the 3 translational
degrees of freedom of the center of the mass and 3 rotational degrees of freedom around
it. Thus, the PES has in principle 3N 6 dimensions. The concept of PES has been
employed in a wide variety of phenomena, from atomic clusters, over protein folding to
glasses.
Finding the global minimum of the PES is one of the most dicult and important
goals in a number of branches of physics, chemistry, and biology. For a periodic system,
the global minimum renders the crystalline ground state structure of a solid, while, for
a non-periodic systems such as a nanocluster it determines the ground-state geometry.
The determination of the relevant isomers in the neighborhood to the ground state is also
crucial, since the former might be observed in an experiment due to the thermodynamic
eects.
A short description of some of the most widely used techniques to explore PES and to
optimize the energy from both local and global perspectives is given below.
3.1 Local Optimization methods
Local optimization (LO) methods are employed to locate the local minima of a function.
In any LO scheme, only the information about a function from the close proximity of the
initial conguration or starting point is used in order to update the conguration and the
function value. The optimization procedure advances to whatever nearby local minima
by following the surface downhill in some way. Eventually, the LO techniques are called
deterministic schemes. At each iteration step, local information such as the energy and
its gradients or forces are collected in order to obtain the next structure until iteratively
34
Chapter 3. Exploring the ground-state energy surface of nanoclusters
Figure 3.1: Schematic picture of the SD scheme (black arrows) and a conjugate gradient
(red arrows). After Ref. [141].
the forces disappear and a local minimum has been achieved.
3.1.1 Steepest Descent
The steepest descent (SD) is the simplest of all LO methods. It strictly seeks downhill to
locate the local minimum [144]. In each iteration step, the atoms are displaced according
to the forces acting on them
R
,k+1
= R
,k
+
k
F

({R
,k
})

3.1.1
where
k
is a technical parameter to manipulate. This method has some major disadvan-
tages. If
k
is too small, many iteration steps are required, the convergence speed is rather
slow, and the process can in principle go on forever. Setting a larger
k
can enhance the
convergence speed. However, the system might begin to swing around the local minimum.
3.1.2 Conjugate Gradient
The CG method is introduced by Hestenes and Stiefel in 1951 [120123,144]. This method
employs the successive line minimizations along a search direction G
,k
:

k
= arg min

E(R
,k
+
k
G
,k
({R
,k
}),

3.1.2
where arg min

E represents the argument which minimizes the energy E. The atomic

coordinates are then correspondingly updated to
R
,k+1
= R
,k
+
k
G
,k
.

3.1.3
In the rst step one has to begin along the atomic forces as rst search direction, i.e.,
G
,k
= F

(R
,0
). The main dierence between the SD and CG methods lies in the choice
of the following search direction from current point. The SD scheme gives a criss-crossed
pattern, since every new line step only takes local information into account. Instead the
CG scheme does not strictly seek the PES downhill but along a search direction that is
somewhat tilted or conjugate to the preceding search directions. Which is accomplished
by including a portion of the preceding search direction to the atomic forces:
G
,k
= F
,k
+
k
G
,k1
.

3.1.4
35
3.2. Global Optimization schemes
(see Fig. 3.1). If the system is close to a local minima then the CG methods in principle
render the same results. However, if the system is away from the local minima, the search
directions may become unreasonable and it is convenient often to start with a few SD
steps to bring the system near to the local minimum. One can show that for quadratic
form the CG method converges in at most N line searches.
3.1.3 The Broyden Fletcher Goldfarb Shanno method
The Broyden-Fletcher-Goldfarb-Shanno (BFGS) method is sometimes called quasi-Newton
method. Dierent methods under this category dier in their updating of the inverse of
the Hessian matrix H

. Instead of directly taking the inverse of H

, an iterative up-
dating technique is used. The new search direction G
,i
can be obtained by solving the
Newton equation

G
,i
= F

({R
,i
})

3.1.5
The next structure is then obtained by performing a line minimization as in the CG
scheme Eq. 3.1.2. If the PES were quadratic and the Hessian would be known accurately,
the local minimum would be identied within one line search. In fact, the computation
of the Hessian matrix in each iteration step can be very expensive, so that it is instead
successively approximated in each iteration step, therewith being a quasi-Newton scheme.
Since more information of the PES is taken into account, the BFGS method can be more
ecient than SD or even CG schemes. The best performance is achieved when the system is
near to the local minimum, i.e., where the harmonic approximation is justied. Therefore,
as in the case of the CG scheme, if a LO is aimed it is important to begin with a few
SD steps to bring the system close to the local minimum. Otherwise the BFGS approach
remains very ecient but the system might converge to an attractor basin that is dierent
from the one corresponding to the initial optimization.
3.2 Global Optimization schemes
Global optimization (GO) has long been an intensive research topic in many elds of
science. Some of the important applications of GO are the design of integrated circuits
such as microprocessors [129], the prediction of protein structures [126, 127], and the ab-
initio computation of nano-size atomic structures [125,133]. In fact GO is a non polynomial
(NP) hard problem, i.e., non-deterministic and non-polynomial-time hard [124]. In other
words there is no GO algorithm available which guarantees to locate the global minimum
within a time that goes as some power of the number of variables.
GO schemes can be classied according to the way in which new structures are gener-
ated from the initial structure or previous structures. A technique called trial move can
be used for this purpose. Every trail move generates a new structure by deforming or
distorting the initial structure in some way. The newly generated structures are accepted
or rejected according to some acceptance criterion. The total energy of the system can be
taken as acceptance criterion since it is the fundamental quantity to optimize. Dierent
kinds of GO schemes are available at present. The most important ones will be outlined
in the following.
36
Chapter 3. Exploring the ground-state energy surface of nanoclusters
3.2.1 Graph theory
The starting cluster geometries used for the calculations reported in chapters 4 and 5 are
generated from graph theory. Here we provide only the important aspects of this method.
For more detailed information the reader refer to the Refs. [17, 172, 176].
The topographical bonding in the clusters can be represented by the adjacency matrix
A and it is important to verify that it can be represented by a true structure in D 3
dimensions. The adjacency matrix can be associated to an undirected simple graph g. The
ensemble of topographical structures for cluster size N is a subset of the set G of undirected
simple graphs having N vertices. A graph is acceptable as a cluster structure, only if a set
of atomic coordinates

R
i
with i = 1, . . . , N exists, such that the interatomic distances R
ij
satisfy the conditions R
ij
= R
0
if the sites i and j are connected in the graph (i.e., if the
adjacency matrix element A
ij
= 1) and R
ij
> R
0
otherwise (i.e., if A
ij
= 0). Here R
0
refers
to the nearest neighbor (NN) distance, which at this stage can be regarded as the unit
of length, assuming for simplicity that it is the same for all clusters. The complete set of
topographical structures can be generated by computing all the unidirected simple graphs
having N vertices and disregarding equivalent (isomorphic) graphs that correspond to a
permutation of atomic indexes without changing the connectivity. The simplest test for
graph isomorphism consist in the comparison of the standard representation of adjacency
matrices calculated by including atomic connectivities up to third neighbors [172]. A
further numerically more expensive test is the comparison of eigenvalue spectra of the
adjacency matrices. Once the topologies are obtained the next important step is the
determination of 3D atomic coordinates for each structure, which are consistent with
the interatomic distances and connectivity information. This is known as the distance-
geometry problem which belongs to the class of non-polynomial hard computations.
The adjacency matrices for two similar cluster structures are shown in Fig. 3.2 . The
only dierence between these structures is that the atoms 1 and 6 are connected in the clus-
ter structure on the bottom. The connectivity matrices for the corresponding structures,
given by S
1
and S
2
, are of course dierent.
One easily nds that for N 4 all graphs are possible cluster structures. These are
tetrahedron, rhombus, square, star, triangular racket and linear chain shown in Fig. 3.3
[17]. However, for N 5 there are graphs, i.e., topologies, which cannot be realized in
practice. For instance, it is not possible to have ve atoms being NNs from each other
in a three dimensional space. Consequently, for N 5 there are less real structures than
mathematical graphs. The total number of graphs (structures) is 21 (20), 112 (104), and
853 (647) for N = 5, 6 and 7, respectively [17].
As discussed earlier, in order to perform reliable structure calculations for clusters one
needs to consider a large, most possibly complete and unbiased set of initial structures. In
the case of mixed clusters, a thorough geometry optimization must include not only the
representative cluster geometries or topologies, but also all relevant chemical orders. Thus
all distributions of the dissimilar atoms for any given size and composition must also be
taken into account. The dierent distributions of atoms for representative compositions
are illustrated in the Fig. 3.4, where a square pyramid structure is considered with one
and two impurity atoms.
37
3.2. Global Optimization schemes
S
1
=

0 1 1 1 1 0
1 0 1 0 1 1
1 1 0 1 0 1
1 0 1 0 1 1
1 1 0 1 0 1
0 1 1 1 1 0

=
S
2
=

0 1 1 1 1 1
1 0 1 0 1 1
1 1 0 1 0 1
1 0 1 0 1 1
1 1 0 1 0 1
1 1 1 1 1 0

=
Figure 3.2: Clusters 1 and 2 with their corresponding adjacency matrices S
1
and S
2
,
respectively. The clusters are similar except that the atoms 1 and 6 are connected in the
cluster 2.
tetrahedron rhombus square star triangular
racket
linear chain
Figure 3.3: Structures for the cluster size N = 4.
=
=
Figure 3.4: The illustration of the dierent chemical orders (also known as homotops) of
single- and double-impurity clusters with N = 5 atoms having square pyramid topology.
38
Chapter 3. Exploring the ground-state energy surface of nanoclusters
3.2.2 Simulated annealing
The simulated annealing (SA) method is one of the most commonly used GO schemes.
It was rst proposed in 1983 by Kirkpatrick [129, 130] and is based on the Metropolis
Monte-Carlo (MC) scheme, by including a variable temperature to simulate the annealing
procedure of a physical system. The working principle of a typical SA method is the
following. Starting from a random structure with a total energy E, a new structure is
generated by randomly displacing the atoms, leading to a change of the total energy E.
If the energy has decreased, i.e., E < 0, the new structure is accepted and used as starting
point for the following iteration. If the energy has increased, however, the structure is not
rejected unconditionally. It is accepted with a probability of P(E) exp(E/k
B
T).
According to the acceptance criterion of the Metropolis algorithm one generates a canonical
ensemble of atomic congurations at a given temperature T.
At strictly zero temperature, only structures that lower the energy would be accepted.
At rst sight this might seem a good choice since one intends to push the system towards
the ground state. However, this would be very inecient, since the system is most likely to
get stuck to a particular local minimum and never reach the true global minimum. A nite
temperature assists the system to jump out of a local minimum by following controlled
uphill steps. Once the system arrives at a high temperature, an annealing schedule can
be employed. This procedure successively cools down the system, and facilitates reach-
ing thermal equilibrium. Coupled to the temperature are the random displacements of
the individual atoms R

which usually adopt a Gaussian distribution in the classical

simulated annealing scheme [131]:
p(R

) exp
(
(R

)
2
/T
)

3.2.1
With decreasing temperature, the step width is hereby reduced,
T
T
0
log(1 +t)
.

3.2.2
thus ideally bringing the system towards the optimal ground state geometry.
3.2.3 Basin-hopping
The calculations reported in chapter 7 are performed by using the BH method [133135].
Rather than sampling the original PES E{R
i
}, the method explores the transformed PES

E{R
i
} given by

E{R} = min E{R},

3.2.3
where min indicates a local structural relaxation.
In this scheme the energy associated to a specic point R
i
on PES is that of the local
minimum obtained starting at this point R
i
and following a local optimization. This is
performed, for example with the CG or BFGS methods. Thus, the BH scheme maps the
real PES onto a set of interpenetrating staircases with plateaus corresponding to each
basins of attraction. A schematic view of the BH scheme is shown in Fig. 3.5.
The BH method has been derived in order to get rid of the unfavorable transition states
or high barriers on the PES. Thus, it allows for fast and ecient interbasin transitions.
39
3.2. Global Optimization schemes
Figure 3.5: Illustration of principle of the basin-hopping method. Depicted is a model
energy surface together with its transformed landscape. The green arrow indicates a trial
move performed on a local minimum, being followed by a local structural relaxation (red
arrows). See Ref. [222].
In contrast to the original SA scheme, Markov steps that initially result in an increase
of energy are accepted much more probably, since the subsequent LO allows the system
to relax into the corresponding local minimum which has usually a much lower energy.
Although, the shapes of the original and the transformed PES are dierent, the nature of
the local minima is not modied, since the Hamiltonian itself is not changed. A comparison
between molecular-dynamics and BH generated minima is thus straightforward.
In 1999, Wales et al. located global minima of Lennard-Jones (LJ) clusters having up
to 110 atoms by using BH scheme. The LJ
38
is a well-known example of a double-funnel
PES [137], consisting of an icosahedral funnel with a large free energy (large conguration
energy of the corresponding attraction barrier) and a fcc funnel with a comparably smaller
free energy but which actually leads to the global minimum. The original PES renders only
a small overlap in the canonical occupation probabilities with respect to the temperature,
so that there is a very high possibility that the system gets trapped in the wrong funnel.
Transforming the PES like in the BH scheme, however, results in a broadening of the
overlap region, which considerably helps for the system to jump out of the wrong funnel
[137,138]. In fact, the BH scheme not only facilitates the interbasin transitions by removing
the often high barriers but also renders the transition between dierent funnels easier
because of the barrier less stochastic-dynamics.
3.2.4 Genetic algorithms
The basic idea behind genetic algorithms (GA) is based on the Darwinian theory of natural
selection [139, 140]. New cluster structures are generated from two candidate structures
selected from a population of cluster structures, namely the parents, which are then mated
to create a child.
One important condition for a good mating process is that the child should preserve
40
Chapter 3. Exploring the ground-state energy surface of nanoclusters
some of the structural properties of the parents. A common procedure for that is to chop
the parent structures by a plane that is randomly aligned, and to cross the resulting halfs
followed by a relaxation to generate a child (see Fig. 3.6). In this way structural motifs
from dierent points on the PES are then joined together to form the child rather than
performing a pure local search like in the BH scheme.
Figure 3.6: Mating between two parent structures generating a child. After the mating,
the new child is locally relaxed (after Ref. [141]).
One interesting question here is what kind of features a child should have in order to
substitute a parent from the population. Deaven and Ho [139] formulated two arguments.
The rst one is that in order to substitute a parent, a child should have lower energy.
The second argument is that a child that is expected to replace a parent must be dierent
from all the other members of the population, since this will facilitates the population to
remain diverse.
41
4
Interplay of structure, magnetism and chemical
order in small FeRh clusters
The present chapter is devoted to the study of the structural, electronic and magnetic
properties of small Fe
m
Rh
n
clusters having N = m + n 8 atoms in the framework of a
generalized-gradient approximation to density-functional theory. The correlation between
structure, chemical order, and magnetic behavior is analyzed as a function of size and
composition. For N = m + n 6 a thorough sampling of all cluster topologies has been
performed, while for N = 7 and 8 only a few representative topologies are considered. In
all cases the entire concentration range is systematically investigated. All the clusters show
ferromagnetic-like order in the optimized structures. As a result, the average magnetic
moment per atom
N
increases monotonously with Fe content, in an almost linear way
over a wide range of concentrations. A remarkable enhancement of the local Fe moments
beyond 3
B
is observed as result of Rh doping. This is a consequence of the increase in the
number of Fe d holes, due to charge transfer from Fe to Rh, combined with the extremely
reduced local coordination. The Rh local moments, which are important already in the
pure clusters (N 8) are not signicantly enhanced by Fe doping. However, the overall
stability of magnetism, as measured by the energy gained upon spin polarization, increases
when Rh is replaced by Fe. The composition dependence of the electronic structure and
the inuence of spin-orbit interactions on the cluster stability are discussed.
4.1 Introduction
Alloying elements with complementary qualities in order to tailor their physical behavior
for specic technological purposes has been a major route in material development since
the antiquity. Cluster research is no exception to this trend. After decades of system-
atic studies of the size and structural dependence of the most wide variety of properties
of monoelement particles, the interest has actually been moving progressively over the
past years towards investigations on nite-size binary alloys [150]. The magnetism of
42
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
transition-metal (TM) clusters opens numerous possibilities and challenges in this con-
text [1, 151164]. For example, one would like to understand how to modify the magnetic
characteristics of clusters, in particular the saturation magnetization and the magnetic
anisotropy energy (MAE), as it has been done in solids. This would indeed allow one
to design new nanostructured materials from a microscopic perspective. Nevertheless, it
also true that controlling composition, system size, and magnetic behavior sets serious
diculties for both experiment and theory.
Pure TM clusters such as Fe
N
, Co
N
and Ni
N
show spin moments, orbital moments, and
MAEs that are enhanced with respect to the corresponding periodic solids [2, 49]. Still,
the possibilities of optimizing the cluster magnetic behavior by simply tuning the system
size have been rather disappointing, particularly concerning the MAE, which remains
relatively small despite being orders of magnitude larger than in solids [2] due to the
weakness of the spin-orbit (SO) coupling in the 3d atoms. This is one of the motivations
for alloying 3d TMs with 4d and 5d elements which, being heavier, are subject to stronger
SO interactions. In this context it is useful to recall that large nanoparticles and three
dimensional solids of these elements are non-magnetic. However, at very small sizes the 4d
and 5d clusters often develop a nite spontaneous low-temperature magnetization, due to
the reduction of local coordination and the resulting d-band narrowing [3, 52, 53, 165, 230].
The rst experimental observation of this important nite-size eect has been made by
Cox et al. by performing Stern-Gerlach-deection measurements on Rh
N
clusters. In this
work the average magnetic moments per atom
N
= 0.150.80
B
have been experimentally
determined for N 3050 atoms [230]. In view of these contrasting features one expects
that 3d-4d and 3d-5d alloy clusters should show very interesting structural, electronic and
magnetic behaviors.
The purpose of this chapter is the investigation of the ground-state properties of the
small FeRh clusters in the framework of Hohenberg-Kohn-Shams density functional the-
ory [11]. Besides the general interest of the problem from the perspective of 3d-4d nano-
magnetism, these clusters are particularly appealing because of the remarkable phase dia-
gram of FeRh bulk alloys (see Fig. 4.1) [166]. In the case of Fe
50
Rh
50
the magnetic order
at normal pressure and low temperatures is antiferromagnetic (AF). As the temperature
increases this so-called

phase undergoes a rst order transition to a ferromagnetic (FM)

state, the

phase, which is accompanied by a change in lattice parameter. The corre-

sponding transition temperature T

c
increases rapidly with increasing external pressure
P, eventually displacing the FM

phase completely for P 7 GPa (T

c
290K for
Fe
50
Rh
50
at normal pressure). Moreover, T

c
decreases very rapidly with decreasing Rh
content (see Fig. 4.1). At low pressures the FM

phase undergoes a FM to paramag-

netic (PM) transition at (T
C
670K) [166]. In addition, the properties of -FeRh bulk
alloys have been the subject of rst principles and model theoretical investigations [167].
These show that the relative stability of the FM and AF solutions depends strongly on
the interatomic distances. Such remarkable condensed-matter eects enhance the appeal
of small FeRh particles as specic example of 3d-4d nanoscale alloy. Investigations of
their magnetic properties as a function of size, composition, and structure are therefore
of fundamental importance.
The remainder of this chapter is organized as follows. In section. 4.2 the main details
of the computational procedure are presented. The results of our calculations for FeRh
clusters having N 8 atoms are reported in sections 4.3 and 4.4. First, we focus on the
interplay between structure, chemical order and magnetism in the most stable geometries
43
4.2. Ab-initio relaxation of clusters
Figure 4.1: The phase diagram of FeRh bulk alloy (after Ref. [166]).
for dierent cluster sizes. Second, we analyze the concentration dependence of the cohesive
energy, the local and average magnetic moments, the spin-polarized electronic structure,
relative stability and magnetic stabilization energy. Finally, we conclude in section. 4.5
with a summary of the main trends and a short outlook to future extensions.
4.2 Ab-initio relaxation of clusters
The calculations reported in this chapter have been performed in the framework of Hohenberg-
Kohn-Shams density functional theory, [11] as implemented in the Vienna ab initio sim-
ulation package (VASP) [168]. The exchange and correlation energy is described by us-
ing both the spin-polarized local density approximation (LDA) and Perdew and Wangs
generalized-gradient approximation (GGA) [224]. As discussed in the section. 2.2 the
VASP solves the spin-polarized Kohn-Sham equations in an augmented plane-wave basis
set, taking into account the core electrons within the projector augmented wave (PAW)
method [170]. See section. 2.4 where the PAW method has extensively been described.
This ecient frozen-core all-electron approach allows to incorporate the proper nodes of
the Kohn-Sham orbitals in the core region and the resulting eects on the electronic struc-
ture, total energy and interatomic forces. The 4s and 3d orbitals of Fe, and the 5s and 4d
orbitals of Rh are treated as valence states. The wave functions are expanded in a plane
wave basis set with the kinetic-energy cut-o E
max
= 268 eV. In order to improve the
convergence of the solution of the self-consistent KS equations the discrete energy levels
are broadened by using a Gaussian smearing = 0.02 eV. A number of tests have been
performed in order to assess the numerical accuracy of the calculations. Increasing the
cut-o energy E
max
= 268 eV and supercell size a = 12

A to E
max
= 500 eV and a =
22

A in Rh
4
increases the computation time by a factor 47. This yields total energy
dierences of 1.75 meV and 0.25 meV, respectively. In the above calculations the changes
44
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
in average bond length (bond angle) amounts to 10
3

A (10
4
degrees). These dierence
are not signicant for our physical conclusions. In fact, typical isomerization energies in
these clusters are an order of magnitude larger, i.e., of the order of 1030 meV. We also
found that the total energy is nearly independent of the choice of the smearing parame-
ter , provided it is not too large ( 0.05 eV). Values from = 0.01 to 0.1 eV have
been checked. Therefore, we judge that our set of standard parameters (E
max
= 268 eV,
supercell size a from 10 to 22

A, and = 0.02 eV) oers a suciently good accuracy
at a reasonable computational costs. The PAW sphere radii for Fe and Rh are 1.302

A
and 1.402

A, respectively. A simple cubic supercell is considered with the usual periodic
boundary conditions. The linear size of the cell is a = 1022

A, so that any pair of images
of the clusters are well separated and the interaction between them is negligible. Since we
are interested in nite systems, the reciprocal space summations are restricted to the
point.
For smaller clusters (N 6) all the possible structures have been taken as starting
points of our structural relaxations. See section. 3.2.1 where the cluster generation method
for the present calculations has been described in detail. Out of this large number of
dierent initial congurations the unconstrained relaxations using VASP lead to only
a few geometries, which can be regarded as stable or metastable isomers. For larger
clusters (N = 7 and 8) we do not aim at performing a full global optimization. Our
purpose here is to explore the interplay between magnetism and chemical order as a
function of composition for a few topologies that are representative of open and close-
packed structures. Taking into account our results for smaller sizes, and the available
information on the structure of pure Fe
N
, Rh
N
, Co
N
, and Pd
N
clusters, we have restricted
the set of starting topologies for the unconstrained relaxation of heptamers and octamers to
the following: bicapped trigonal bipyramid, capped octahedra, and pentagonal bipyramid
for N = 7, and tricapped trigonal bipyramid, bicapped octahedra, capped pentagonal
bipyramid and cube for N = 8. Although, the choice of topologies for N = 7 and 8
is quite restricted, it includes compact as well as more open structures. Therefore, it is
expected to shed light on the dependence of the magnetic properties on the chemical order
and composition.
The dependence on concentration is investigated systematically for each topology of
Fe
m
Rh
n
by varying m and for each size N = m + n 8, including the pure Fe
N
and
Rh
N
limits. Moreover, we take into account all possible non-equivalent distributions of
the m Fe and n Rh atoms within the cluster. In this way, any a priori assumption on the
chemical order is avoided. Obviously, such an exhaustive combinatorial search increasingly
complicates the computational task as we increase the cluster size, and as we move away
from pure clusters towards alloys with equal concentrations. Finally, in order to perform
the actual density-functional calculations we set for simplicity all NN distances in the
starting cluster geometry equal to the Fe bulk value [173] R
0
= 2.48

A. Subsequently, a
fully unconstrained geometry optimization is performed from rst principles by using the
VASP [168]. The atomic positions are fully relaxed by means of conjugate gradient or
quasi-Newtonian methods, without imposing any symmetry constraints, until all the force
components are smaller than a threshold of 5 meV/

A. The convergence criteria are set

to 10
5
eV/

A for the energy gradient and 5 10

4

A for the atomic displacements [174].

The same procedure applies to all considered clusters regardless of composition, chemical
order, or total magnetic moment. Notice that the diversity of geometrical structures and
atomic arrangements often yields many local minima on the ground-state energy surface,
45
4.3. Structure and magnetism
which complicates signicantly the location of the lowest-energy conguration.
Lattice structure and magnetic behavior are intimately related in TMs, particularly in
weak ferromagnets such as Fe and its alloys. On the one side, the optimum structure and
chemical order depend on the actual magnetic state of the cluster as given by the average
magnetic moment per atom
N
and the magnetic order. On the other side, the magnetic
behavior is known to be dierent for dierent structures and concentrations. Therefore,
in order to rigorously determine the ground-state magnetic properties of FeRh clusters,
we have varied systematically the value of the total spin polarization of the cluster S
z
by
performing xed spin-moment (FSM) calculations in the whole physically relevant range.
Let us recall that S
z
= (

)/2 where

) represents the number of electrons in the

majority (minority) states. In practice we start from the non-magnetic state (S
min
z
= 0)
and increase S
z
until the local spin moments are fully saturated, i.e., until the Fe mo-
ments in the PAW sphere reach
Fe
4
B
and the Rh moments
Rh
2.5
B
(typically,
S
max
z
3N/2). The above described global geometry optimizations are performed inde-
pendently for each value of S
z
. These FSM study provides a wealth of information on
the isomerization energies, the spin-excitation energies, and their interplay. These are
particularly interesting for a subtle magnetic alloy such as FeRh. In the present chapter
we focus on the ground-state properties by determining for each considered Fe
m
Rh
n
the
most stable structural and magnetic conguration corresponding to energy minimum as a
function of S
z
and of the atomic positions [171].
Once the optimization with respect to structural and magnetic degrees of freedom is
achieved, we derive the binding energy per atomE
B
= [mE(Fe)+nE(Rh)E(Fe
m
Rh
n
)]/N
in the usual way by referring the total energy E to the corresponding energy of m Fe and
n Rh isolated atoms. Moreover, for each stationary point of the total energy surface (i.e.,
for each relaxed structure having a nearly vanishing |

E| we determine the vibrational

frequencies from the diagonalization of the dynamical matrix. The latter is calculated
from nite dierences of the analytic gradients of the total energy. In this way we can
rule out saddle points to which the local optimization procedure happens to converge on
some occasions. Only congurations which correspond to true minima are discussed in
the following. Finally, a number of electronic and magnetic properties for example, the
magnetic energy E
m
= E(S
z
= 0) E(S
z
), the local magnetic moments
i
integrated
within the Wigner-Seitz (WS) or Bader atomic cells of atom i, [175, 176] and the spin
polarized density of electronic states (DOS)

() are derived from the self-consistent

spin-polarized density and Kohn-Sham spectrum.
4.3 Structure and magnetism
In this section we discuss the ground-state structure, chemical order, binding energy, and
magnetic moments of Fe
m
Rh
n
clusters having N = m+n 8 atoms. The main emphasis
is here on understanding how the various electronic, structural and magnetic properties
depend on the chemical composition of the alloy. First, each cluster size N is analyzed
separately, since a strong dependence on N is expected in the small size, non-scalable
regime. Comparisons between the various N are stressed by means of cross-references
between dierent subsections. In addition the main trends as a function of size and
concentration are summarized in section. 4.4.
46
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
Table 4.1: Structural, electronic and magnetic properties of FeRh dimers. Results are
given for the binding energy E
B
(in eV), the magnetic stabilization energy E
m
= E(S
z
=
0) E(S
z
) (in eV), the average interatomic distance d

(in

A) between atoms and
(, = Fe or Rh), the average spin moment per atom
N
= 2S
z
/N (in
B
), the local spin
moment

(in
B
) at the Fe or Rh atoms, and the vibrational frequency
0
(in cm
1
).
Cluster Struct. E
B
E
m
d

N

Fe

Rh

0
Fe
2
1.35 0.77 1.98 3.00 2.82 288
FeRh 1.95 0.24 2.07 2.50 3.34 1.33 359
Rh
2
1.65 0.00 2.21 2.00 1.83 224
4.3.1 FeRh dimers
Despite being the simplest possible systems, dimers allow to infer very useful trends on the
relative strength, charge transfers and magnetic order in the various types of bonds which
are found in FeRh alloy clusters. The results summarized in Table 4.1 show that the FeRh
bond yields the highest cohesive energy, followed by the Rh
2
bond, the Fe
2
bond being the
weakest. The particular strength of the heterogeneous bond is conrmed by the fact that
the corresponding vibrational frequency is the highest. The bond length, however, follows
the trend of the atomic radius which, being larger for Rh, gives d
RhRh
> d
FeRh
> d
FeFe
.
Quantitatively, the binding energy per atom E
GGA
B
= 1.35 eV obtained for Fe
2
within
the GGA is smaller than the LDA result E
LDA
B
= 2.25 eV [178] although it still remains
larger than the experimental value E
expt
B
= 0.65 eV reported in Ref. [180]. The calculated
vibrational frequency
0
(Fe
2
) = 288 cm
1
is consistent with previous experimental results
[
0
(Fe
2
) = 299.6 cm
1
from Ref. [180] and
0
(Fe
2
) = 300 15 cm
1
from Ref. [181]]. Our
result for E
B
and
N
of Rh
2
coincide with previous GGA calculations by B. V. Reddy et
al. [182]. These are however larger than the experimental values E
expt
B
(Rh
2
) = 1.46 eV
derived from Knudsen diusion [185], E
expt
B
(Rh
2
) = 0.700.15 eV derived from resonance
Raman in Ar matrices [186] and E
expt
B
(Rh
2
) = 1.203 eV derived from the resonant two-
photon ionization [187]. The calculated vibrational frequency of
0
(Rh
2
)
GGA
= 224 cm
1
should be compared with the experimental value
0
(Rh
2
)
expt
= 283.9 cm
1
reported in
Ref. [186].
The stability of magnetism, as measured by the dierence in the total energy E
m
of
the non-magnetic (S
z
= 0) and optimal magnetic solutions, is largest for Fe
2
and smallest
for Rh
2
. The same trend holds for the average magnetic moment per atom which decreases
linearly from
2
= 3
B
to 2
B
as one goes from Fe
2
, to FeRh, to Rh
2
. These average
magnetic moments per atom correspond to a full polarization of all d electrons in the WS
spheres:
d
7 for Fe and
d
8 for Rh, where
d
stands for the number of valence
d electrons of the corresponding atom. The local magnetic moments

( Fe and
Rh), are obtained by integrating the spin density within the PAW spheres which have the
radius r
PAW
(Fe) = 1.3

A for Fe and r
PAW
(Rh) = 1.4

A for Rh. In the pure dimers, the
local moments
Fe
= 2.82
B
and
Rh
= 1.83
B
are close to the respective total moment
per atom
2
= 3
B
and 2
B
, which indicates that the spin-density m(r) = n

(r) n

(r)
is quite localized around the atoms. Actually, the dierences between

and
N
give
47
4.3. Structure and magnetism
a measure of the small spill-o eect in m(r). Taking this into account, the results for

in the FeRh dimer seem quite remarkable. Here the Fe local moment is signicantly
enhanced with respect to the Fe
2
or Fe-atom value, while the Rh moment is reduced by
a similar amount (
Fe
= 0.52
B
and
Rh
= -0.50
B
, see Table 4.1). This is mainly
the consequence of a transfer of d electrons from Fe to Rh, which allows the Fe atom to
develop a larger spin moment, due to the larger number of d holes. This occurs at the
expense of the moment at the Rh atom, which has less d holes to polarize. An integration
of the electronic density in the Bader cells [175] shows that 0.33 electrons are transferred
from the Fe to the Rh atom in FeRh. This behavior is qualitatively in agreement with the
higher Pauling electronegativity of the Rh atom (
Fe
= 1.83 and
Rh
= 2.28) [188].
4.3.2 FeRh trimers
Having established the properties of dimers, we turn now to the trimers. The results for
trimers are summarized in Table 4.2 and Fig. 4.2. We considered both triangle and linear
shapes as the initial structures for the optimization. As expected, the lowest energy isomers
are found to be triangles for all compositions. According to our calculations the ground
state of Rh
3
is an equilateral triangle (D
3h
) with E
B
= 2.31 eV, bond length d = 2.37

A
and average magnetic moment
3
= 1
B
. The local magnetic moments

= 0.93
B
in
the WS cells align parallel to each other and are almost as large as
3
. These results are
consistent with those reported in previous GGA studies of Rh
3
(E
B
= 2.35 eV, d = 2.45

A
and
3
= 1
B
) [182]. A single Fe substitution yields an isosceles FeRh
2
with an elongated
base composed of the two Rh atoms. We notice that the bond-length d
RhRh
= 2.57

A is
larger than in Rh
3
. The linear isomer of the form Rh-Fe-Rh, i.e., with only FeRh bonds,
lies 0.4 eV above the optimal structure. It is the only true local minimum among the
linear FeRh trimers. The other linear structures (Rh-Rh-Fe, Fe-Rh-Fe, and Fe-Fe-Rh) are
all found to be saddle points connecting triangular minima of the potential energy surface
(PES). Further Fe substitution yields a isosceles Fe
2
Rh in which the FeFe bond is the
shortest. One observes, as in the dimers, that the interatomic distances follow the trends
in the atomic radii. Finally, for Fe
3
, the calculated lowest-energy structure is a Jahn-Teller
distorted isosceles triangle with two longer bonds (d
12
= d
13
= 2.30

A) and a shorter one
(d
23
= 2.07

A). The calculated average magnetic moment of Fe
3
is
3
= 3.33
B
. These
results coincide with previous GGA studies [177] predicting d
12
= d
13
= 2.33

A and
d
23
= 2.09

A. In contrast, LDA calculations [178] yield an equilateral Fe
3
with average
magnetic moment
3
= 2.66
B
and d = 2.10

A. By using the spin-polarized LDA, we also
obtain an equilateral triangle similar to the one reported in Ref. [178]. In contrast, in
the GGA one nds that the equilateral triangle (D
3h
) is unstable with respect to a Jahn-
Teller distortion. The isosceles shape of Fe
3
can therefore be interpreted as a consequence
of exchange and correlation eects. Moreover, we have analyzed the GGA Kohn-Sham
spectrum in the equilateral structure and found a high degeneracy at the Fermi energy,
which is consistent with the interpretation that the distortion is triggered by a Jahn-Teller
eect.
Concerning the composition dependence of E
B
one observes a non-monotonous be-
havior as for N = 2, which indicates that the FeRh bonds are the strongest. The lowest
vibrational frequency follows a similar trend, despite the larger mass of Rh. Notice that
FeRh
2
is somewhat more stable than Fe
2
Rh, since the bonds between Rh atoms are in
general stronger than between Fe atoms. Finally, one may also notice that the energy
48
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
Table 4.2: Structural, electronic and magnetic properties of FeRh trimers. Results are
given for the binding energy per atom E
B
(in eV), the magnetic stabilization energy per
atom E
m
= [E(S
z
=0) E(S
z
)]/N (in eV), the average interatomic distance d

(in

A)
ordered from top to bottom as d
FeFe
, d
FeRh
and d
RhRh
, the average spin moment per atom

N
= 2S
z
/N (in
B
) , the local spin moment

(in
B
) at the Fe or Rh atoms. The
asterisks refer to the rst exited isomers. See also Fig. 4.2.
Cluster E
B
E
m
d

N

Fe

Rh
Fe
3
1.80 0.69 2.22 3.33 2.99
Fe
2
Rh 2.24 0.32 2.25 3.00 3.35 1.21
2.35
Fe
2
Rh* 1.64 0.48 2.13 3.00 3.30 1.35
2.29
FeRh
2
2.45 0.05 2.21 2.00 3.27 1.18
2.57
FeRh
2
* 1.99 0.49 2.20 2.00 3.30 1.17
Rh
3
2.31 0.02 2.37 1.00 0.93
gain E
m
associated to magnetism only plays a quantitative role in the relative stability
of triangular and linear FeRh
2
. E
m
is actually larger for the linear chain than for the
triangle. Therefore, the later remains the most stable structure even in the non-magnetic
case, although with somewhat dierent bond lengths.
The average magnetic moment per atom
3
amounts to 1
B
for Rh
3
. In the alloys
it increases monotonously with Fe doping, reaching
3
= 10/3
B
for Fe
3
. The local
magnetic moments

always show a FM-like coupling. They are all identical in Rh

3
,
which is consistent with the C
3
point-group symmetry. In the pure clusters

is always
close to
3
. This indicates that the spin polarization is dominated by electrons occupying
localized states and that spill-o contributions are not important. For example, in the
case of Fe
3
, one nds
1
= 3.23
B
and
2
=
3
= 2.87
B
, the latter corresponding to the
pair of atoms forming the shorter bond. On the other side, the average local moments

Rh
= 0.93
B
in Rh
3
should be compared with (Rh
3
) = 1
B
. As soon as FeRh bonds are
present, for mixed compositions, the local Fe moments are enhanced beyond 3
B
. This
is mainly due to a charge transfer from Fe to Rh, leading to an increase in the number
of Fe d holes as already observed in the dimer. Quantitatively, the local
Fe
and
Rh
in
mixed trimers are similar, though somewhat smaller than the corresponding values in the
FeRh dimer. Notice, moreover, the enhancement of the Rh local moments in Fe
2
Rh and
FeRh
2
as compared to pure Rh
3
. This reects the importance of the proximity of Fe on
the magnetic behavior of the Rh atoms.
4.3.3 FeRh tetramers
The most stable FeRh tetramers are all tetrahedra and the rst low-lying isomers are
rhombi (see Fig. 4.2). The distribution of the atoms within the optimal topology does
not play a role since all sites are equivalent in a tetrahedron. In the case of Rh
4
we
49
4.3. Structure and magnetism
Fe
3
Fe
2
Rh Fe
2
Rh* FeRh
2
FeRh
2
* Rh
3
Fe
4
Fe
3
Rh Fe
3
Rh* Fe
2
Rh
2
Fe
2
Rh
2
* FeRh
3
FeRh
3
*
Rh
4
Figure 4.2: The optimal structure for the ground-state and lowest energy isomers (indi-
cated by an asterisk) of trimer and tetramer FeRh clusters.
obtain a nonmagnetic undistorted tetrahedron having E
B
= 2.75 eV and bond length
d = 2.45

A. The closest isomer is found to be a bent rhombus with an average bond length
d = 2.35

A. Similar results have been obtained in previous studies on Rh clusters [183].
Notice, however, that Bae et al. [190] have obtained a bend rhombus as the ground-state
structure for Rh
4
also by using VASP. This discrepancy is likely to be a consequence of
the dierent choice of the pseudopotential and cuto energy E
max
. In our calculations we
considered the PAW method and E
max
= 268 eV, while in Ref. [190] one used ultrasoft
pseudopotentials and E
max
= 205.5 eV.
The binding energy of the alloys shows a characteristic non-monotonous dependence
on concentration, which was also found in smaller clusters. In fact Fe
2
Rh
2
and FeRh
3
are the most stable tetramers with E
B
= 2.74 eV and E
B
= 2.76 eV, respectively. This
conrms that the FeRh bonds are somewhat stronger than others. It is worth noting that
these trends are not altered qualitatively if magnetism is neglected, i.e., if one considers
E
B
for S
z
= 0. In addition, it is interesting to follow how E
B
changes from Rh
4
to Fe
4
.
The stability of the clusters can be qualitatively related to the number of homogeneous
and heterogeneous bonds by counting them for each of the clusters shown in Fig. 4.2. For
instance, FeRh
3
, which is the most stable composition, has 3 FeRh and 3 RhRh bonds.
Replacing a Rh by an Fe to obtain Fe
2
Rh
2
implies replacing 2 RhRh bonds by a stronger
FeRh and a weaker FeFe bond. Therefore, E
B
does not change signicantly. The fact that
E
B
depends weekly on composition for Rh rich tetramers shows that FeRh and RhRh
bonds are comparably strong in these clusters.
Concerning the magnetic moments one observes a approximately linear dependence of

N
as a function of Fe content. In general, the substitution of a Rh by and Fe atom results
in an increase of the total moment 2S
z
by 3 or 4
B
, or equivalently,
N
= (0.751)
B
(see Table 4.3). The magnetic order is always FM-like. In the alloys the local moments

Fe
show the above mentioned enhancement, which is due to a Fe-to-Rh d-electron charge
transfer that increases the number of d holes and allows for the development of
Fe
3.2
3.4
B
. In addition, the presence of Fe in Fe
m
Rh
n
enhances the Rh local moments as
compared to pure Rh
4
.
50
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
Table 4.3: Structural, electronic and magnetic properties of FeRh tetramers as in Table
4.2. See also Fig. 4.2.
Cluster E
B
E
m
d

N

Fe

Rh
Fe
4
2.21 0.35 2.28 3.50 3.08
Fe
3
Rh 2.49 0.58 2.34 3.00 3.18 1.03
2.40
Fe
3
Rh* 2.43 0.90 2.44 2.75 3.08 1.07
2.22
Fe
2
Rh
2
2.74 0.37 2.52 2.50 3.39 1.03
2.31
2.72
Fe
2
Rh
2
* 2.67 0.49 2.28 2.50 3.33 1.20
2.57
FeRh
3
2.76 0.21 2.30 1.75 3.25 1.12
2.60
FeRh
3
* 2.61 0.47 2.34 1.75 3.34 1.08
Rh
4
2.75 0.00 2.45 0.00 0.00
4.3.4 FeRh pentamers
In Table 4.4 and Fig. 4.3 the results for FeRh pentamers are summarized. Although all
possible cluster topologies (20 structures) were considered as starting geometries for each
composition, only the most highly coordinated trigonal bipyramid (TBP) and the square
pyramid (SP) are found to be most stable geometries. The low coordinated structures
transform into compact structures after the relaxation. Except for Rh
5
, which optimal
structure is a SP, all the other FeRh pentamers have the TBP as ground-state geometry.
The trend in the composition dependence of the binding energy E
B
of pentamers conrms
the behavior we started to observe for N = 4. Indeed, in the Fe-rich limit E
B
increases
rapidly with increasing Rh content, as the weakest FeFe bonds are replaced by FeRh bonds.
Later on, near 50% concentration and in the Rh-rich limit, the composition dependence is
weak since FeRh and RhRh bonds are comparably strong (see Table 4.4). In particular for
Rh-rich compositions, replacing Fe by Rh atoms no longer results in weaker binding. In
other words, FeRh bonds are no longer primarily preferred. This is possibly a consequence
of the increasing coordination number, which enhances the role of electron delocalization
and band formation, thus favoring the larger Rh hybridizations.
The calculated optimal structure of Rh
5
, a square pyramid, coincides qualitatively with
previous DFT calculations [182]. Nevertheless, we obtain a binding energy that is 0.07 eV
per atom lower than in Ref. [182]. Substituting one Rh atom by Fe yields FeRh
4
and
changes the optimal cluster topology to the more compact TBP. The SP remains a local
minimum of the ground-state energy surface, which is only 3 meV per atom less stable than
the optimal TBP geometry. The average magnetic moment (FeRh
4
) = 1.2
B
is enhanced
with respect to Rh
5
due to the contribution of a large Fe local moment
Fe
= 3.31
B
.
Notice that the Rh moments are no longer enhanced as in the smaller FeRh
N1
but
51
4.3. Structure and magnetism
Table 4.4: Structural, electronic and magnetic properties of FeRh pentamers as in Table
4.2. See also Fig. 4.2.
Cluster E
B
E
m
d

N

Fe

Rh
Fe
5
2.51 1.00 2.41 3.20 2.93
Fe
4
Rh 2.76 1.01 2.30 3.00 3.09 1.06
2.47
Fe
4
Rh* 2.67 0.75 2.39 3.00 3.02 1.03
2.18
Fe
3
Rh
2
2.96 0.92 2.37 2.40 3.13 0.75
2.39
Fe
3
Rh
2
* 2.85 0.66 2.39 2.40 3.05 0.97
2.40
2.71
Fe
2
Rh
3
3.06 0.55 2.35 2.20 3.36 1.08
2.71
Fe
2
Rh
3
* 3.03 0.53 2.36 2.20 3.28 1.12
2.61
2.73
FeRh
4
3.01 0.33 2.39 1.20 3.31 0.57
2.51
FeRh
4
* 3.00 0.25 2.36 0.80 3.20 0.15
2.52
Rh
5
3.03 0.70 2.48 1.00 0.95
signicantly reduced:
Rh
= 0.62
B
for the apex atoms and
Rh
= 0.52
B
for the Rh
atoms sharing a triangle with the Fe. This is of course related to the fact that the ground-
state S
z
is relatively low. The eect is even stronger in the case of the SP isomer of FeRh
4
.
Here we nd two Rh moments
Rh
= 0.43
B
that couple parallel to the Fe moment, one
very small Rh local moment
Rh
= 0.05
B
, and an antiparallel moment
Rh
= 0.48
B
.
This explains the reduced average total moment
5
= 0.8
B
and the very small average
Rh moment
Rh
= 0.15
B
found in SP isomer of FeRh
4
. The present example illustrates
the subtle competition between cluster structure and magnetism in 3d-4d nanoalloys.
Further increase in the Fe content does not change the topology of the optimal struc-
ture. Moreover, we start to see that for nearly equal concentrations of Fe and Rh (i.e.,
Fe
2
Rh
3
and Fe
3
Rh
2
) the low-lying isomers are the result of changes on the chemical order,
i.e., changes in the distribution of the Fe and Rh atoms within the cluster, rather than the
result of changes in the cluster topology. The most stable conguration corresponds to the
case where the 3 Rh atoms (in Fe
2
Rh
3
) or the 3 Fe atoms (in Fe
3
Rh
2
) are all NNs of each
other (see Fig. 4.3). This is understandable from a single-particle perspective, since the
band energy is lower when orbitals having nearly the same energy levels are hybridized.
In addition, the most stable congurations maximize rst the number of FeRh NN pairs,
followed by the number of RhRh pairs. The optimal Fe
3
Rh
2
structure has 6 FeRh and 3
52
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
FeFe NN pairs, while the rst isomer has 5 FeRh, 3 FeFe, and 1 RhRh NN pairs. The
optimal Fe
2
Rh
3
structure has 6 FeRh and 3 RhRh NN pairs, while the rst isomer has 6
FeRh, 2 RhRh, and 1 FeFe NN pairs. Finally, in the Fe-rich limit, for example in Fe
4
Rh,
the lowest-energy structure remains a TBP but the closest isomer corresponds to the SP,
which has a dierent topology, rather than a dierent position of the Rh atom in the TBP.
The trends in the magnetic properties are dominated by the Fe content. As for smaller
clusters the average magnetic moment per atom
N
increases monotonously with increas-
ing Fe concentration. This holds for all optimal structures and in most of the rst excited
isomers. In fact the latter show in general the same
N
as the optimal structure. The
only exception is FeRh
4
, which is also the only case where an antiparallel alignment of Rh
local moments is found. In all other investigated cases the magnetic order was found to
be FM-like. The local Fe moments show the usual enhancement with respect to pure Fe
N
,
due to an increase in the number of Fe d-holes. This eect is stronger for Rh-rich clusters,
since the larger the number of Rh atoms is, the stronger is the FeRh charge transfer (see
Fig. 4.3). In contrast, the substitution of Rh by Fe does not always enhances the Rh
local moments, as we observed systematically for smaller sizes. Finally, it is interesting to
observe that the dierent chemical orders found in the low lying isomers of Fe
2
Rh
3
and
Fe
3
Rh
2
correspond to dierent local magnetic moments. The environment dependence
of

follows in general the well-known trend of higher spin polarization at the lowest
coordinated sites.
4.3.5 FeRh hexamers
In Table 4.5 and Fig. 4.3 the results for FeRh hexameters are summarized. For each com-
position all possible cluster topologies (63 dierent graphs [17,172]) and all non-equivalent
distributions of Fe and Rh atoms were taken into account as initial guess for the ab initio
optimization of the cluster geometry. As in previous cases, all relevant values of the total
magnetic moment 2S
z
are scanned. Despite the diversity of starting topologies most low
coordinated structures relax into compact ones in the course of the unconstrained relax-
ations. In the end, the square bipyramid (SBP), in general somehow slightly distorted,
yields the lowest-energy regardless of composition. Other more open structures appears,
however, as the rst exited isomers (see Fig. 4.3).
The binding energy per atom E
B
shows a similar composition dependence as for pen-
tamers. For Fe-rich clusters E
B
increases steadily with increasing Rh content, by about
0.2 eV each time a Rh replaces an Fe (see Table 4.5). Qualitatively, this conrms that the
bonding between Fe and Rh is stronger than between Fe atoms. However, for nearly equal
concentrations and in the Rh-rich clusters (Fe
m
Rh
6m
with m 2) E
B
becomes almost
independent of m. This seems to be the result of a compensation of bonding and magnetic
contributions. In fact, on the one side, the magnetic energy E
m
continues to decrease
with increasing Rh content, by about 0.10.2 eV per Rh substitution, even for high Rh
content. And on the other side, this is compensated by an increase of the bonding energy
with increasing number of Rh atoms.
In the case of Rh
6
an octahedron with an average moment
6
= 1
B
and average bond
length d = 2.54

A yields the lowest energy. The rst isomer, a trigonal biprism (TBP),
lies only 28 meV above the optimum, showing a somewhat shorter average bond length
d = 2.46

A and a higher average moment
6
= 1.67
B
. These results are consistent with
previous DFT calculations [189]. A single Fe substitution enhances the average moment to
53
4.3. Structure and magnetism
Fe
5
Fe
4
Rh Fe
4
Rh* Fe
3
Rh
2
Fe
3
Rh
2
*
Fe
2
Rh
3
Fe
2
Rh
3
* FeRh
4
FeRh
4
* Rh
5
Fe
6
Fe
5
Rh Fe
5
Rh* Fe
4
Rh
2
Fe
4
Rh
2
*
Fe
3
Rh
3
Fe
3
Rh
3
* Fe
2
Rh
4
Fe
2
Rh
4
* FeRh
5
FeRh
5
* Rh
6
Figure 4.3: Lowest energy isomers of FeRh pentamers and hexamers as in Fig. 4.2.
54
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
Table 4.5: Structural, electronic and magnetic properties of FeRh hexamers as in Table
4.2. See also Fig. 4.3.
Cluster E
B
E
m
d

N

Fe

Rh
Fe
6
2.74 1.03 2.38 3.33 2.95
Fe
5
Rh 2.94 1.07 2.38 3.16 3.10 1.20
2.46
Fe
5
Rh* 2.92 1.29 2.44 3.16 3.13 1.23
2.35
Fe
4
Rh
2
3.14 0.85 2.38 3.00 3.28 1.22
2.45
Fe
4
Rh
2
* 3.00 0.71 2.42 3.00 3.24 1.39
2.48
2.62
Fe
3
Rh
3
3.21 0.56 2.62 2.50 3.32 1.14
2.38
2.61
Fe
3
Rh
3
* 3.18 0.77 2.56 2.33 3.22 1.04
2.36
2.63
Fe
2
Rh
4
3.26 0.38 2.46 2.33 3.38 1.39
2.51
Fe
2
Rh
4
* 3.15 0.48 2.76 2.33 3.32 1.30
2.39
2.59
FeRh
5
3.24 0.28 2.44 1.83 3.37 1.33
2.54
FeRh
5
* 3.20 0.34 2.37 1.16 3.20 0.54
2.37
Rh
6
3.20 0.19 2.54 1.00 0.91

6
= 1.83
B
but does not change the topology of the optimal FeRh
5
. The RhRh distances
remain essentially unchanged and the FeRh distances are somewhat shorter. The impor-
tant increase in the ground-state spin polarization (5
B
in all) is not only due to the larger
Fe moment (
Fe
= 3.37
B
in the PAW sphere) but also results from the enhancement of
the local Rh moments (
Rh
= 1.33
B
, see Table 4.5). The rst isomer of FeRh
5
corre-
sponds to a distorted trigonal prism with signicantly contracted FeRh and RhRh bond
lengths. In Fe
2
Rh
4
the Fe atoms occupy the opposite apex positions of the octahedron.
In this way each Fe is four-fold coordinated with all Rh atoms. Heterogeneous bonds are
favored over FeFe ones. The local Fe moments in Fe
2
Rh
4
are the largest among all hex-
ameters:
Fe
= 3.38
B
, slightly beyond the value found in FeRh
5
. This corresponds to a
large number of d holes. In addition, particularly important spin polarizations are induced
at the neighboring Rh atoms (
Rh
= 1.39
B
). The rst isomer of Fe
2
Rh
4
corresponds to
55
4.3. Structure and magnetism
Fe
6
= 0.010
Fe
5
Rh
= 0.017
Fe
4
Rh
2
= 0.02
Fe
3
Rh
3
= 0.04
Fe
2
Rh
4
= 0.006
FeRh
5
= 0.007
Rh
6
= 0.009
Figure 4.4: Constant magnetization density plots

(r)

(r) = and local moments

(in
B
) for the ground-state structures of FeRh hexamers. The value of the constant
magnetization density is given in
B
/

A
3
.
a capped trigonal bipyramid (CTBP) having a short FeFe NN bond. This structure lies
only 0.11 eV higher in energy and has the same total moment as the optimal geometry.
Replacing a further Fe atom yields Fe
3
Rh
3
, whose optimal structure is an octahedron.
Here we nd two isosceles open Fe
3
and Rh
3
triangles that form a /2 angle with respect
to each other (see Fig. 4.3). Out of the 12 NN pairs in the Fe
3
Rh
3
octahedron, 8 are
FeRh and only 4 are homogeneous (2 FeFe and 2 RhRh). The local Fe magnetic moments
are similar to the other clusters but the Rh moments are somewhat smaller in average
(
Rh
= 1.14
B
). The rst excited isomer of Fe
3
Rh
3
is a CTBP that lies 25 meV per atom
above the ground state. The lowest-energy structure found for Fe
4
Rh
2
is an octahedron,
while a distorted CTBP is an isomer lying 0.14 eV per atom above. In the former the Rh
atoms are far apart occupying the apical positions, whereas in the latter they are NNs.
The situation is thus similar to what we nd in Fe
3
Rh
2
. For low Rh or Fe concentrations
the atoms are distributed in order to favor the FeRh bonds rather than homogeneous NN
pairs between the atoms in the minority. The octahedron and a distorted CTBP remain
the two most stable structures as one further reduces the Rh content (see Table 4.5 for
Fe
5
Rh and Fe
6
).
Concerning the magnetic properties one observes qualitatively similar trends as in the
smaller clusters. The average magnetic moment per atom
N
increases monotonously with
Fe content. Accordingly, the energy gain E
m
associated to magnetism also increases with
the number of Fe atoms. There are in general very little dierences in
N
between the
56
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
optimal structure and the rst low-lying isomer. The largest part of the spin polarization
(about 90%) can be traced back to the local d magnetic moments with the PAW sphere of
the atoms. As expected, the s and p spin polarizations are almost negligible in comparison
to the d-orbital contributions. A signicant increase of the Fe moments is observed upon
Rh doping, which result from the larger number of available Fe d holes and the low coor-
dination number. Moreover, the Rh moments in Fe
m
Rh
n
are stabilized by the proximity
of the Fe atoms. In the alloy hexameters the values of
Rh
are larger than in the pure
Rh
6
. However, this is not a general trend, since the magnetic moments in small Rh
n
are
often quite important due to the extremely reduced coordination numbers. Fig. 4.4 shows
plots of constant magnetization density

(r)

(r) = for the optimal structure of the

dierent hexameters. Only the positive values of the magnetization density are consid-
ered since all local moment are ferromagnetically aligned. In addition the local moment

i
for Fe and Rh atoms are indicated. Notice the strongly environment dependent.
4.3.6 Exploring heptamers and octamers
For Fe
m
Rh
n
clusters having m+n = N 7 we did not attempt to perform a systematic
sampling of initial topologies for further unconstrained structural relaxation, as was done
for the smaller sizes. Only a few compact and open starting structures are considered.
For N = 7, this includes the bicapped trigonal bipyramid (BCTBP), capped octahedron
(CO) and pentagonal bipyramid (PBP), while for N = 8, they are the tricapped trigonal
bipyramid (TCTBP), bicapped octahedron (BCO), capped pentagonal bipyramid (CPBP)
and cube (C). This choice is motivated by previous results for pure clusters and by the
trend to compact geometries observed for smaller sizes N 6. Although far from exhaus-
tive, the considered geometries allow to explore various relevant growth patterns with a
reasonable computational eort. Certainly, a more complete study would be necessary in
order to draw denitive conclusions about the optimal topologies. For each composition,
all possible distributions of the Fe and Rh atoms within the cluster, as well as all rele-
vant values of the total magnetization S
z
are taken into account (from the non-magnetic
state to saturation). Therefore, the trends on the interplay between chemical order and
magnetic behavior remain rigorous within the framework of the sampled topologies.
The results for N = 7 are summarized in Table 4.6 and Fig. 4.5. As in smaller Fe
m
Rh
n
the binding energy per atom increases rst with increasing Rh content and becomes es-
sentially independent of composition in the Rh-rich limit (5 n 7). For pure Rh
7
the
PBP is the most stable structure among the considered starting geometries. This result is
consistent with some earlier DFT studies [183]. However it contrasts with the calculations
by Wang et al., [189] who used the GGA functional of Ref. [224] (PW91) and obtained a
capped octahedron, and with the calculations of Bae et al., [190] who found a prism plus
an atom on a square face. According to our results, these structures are, respectively 20
and 4 meV per atom higher in energy than the PBP. In the case of a prism plus an atom
on the square face, the energy dierence with the ground state seems too small to be able
to draw denitive conclusions.
FeRh heptamers with high Rh concentrations also favor a PBP topology. In FeRh
6
the Fe atom occupies an apex site, while in Fe
2
Rh
5
and Fe
3
Rh
4
the Fe atoms belong to
the pentagonal ring. Notice that the distances between the two apex atoms in Rh
7
and
between the Fe and Rh apex atoms in FeRh
6
are relatively short. In Fe
2
Rh
5
and Fe
3
Rh
4
the Fe atoms are as far as possible from each other and the distance between the Rh apex
57
4.3. Structure and magnetism
Table 4.6: Structural, electronic and magnetic properties of FeRh heptamers as obtained
from a restricted representative sampling of cluster topologies. See Fig 4.5.
Cluster E
B
E
m
d

N

Fe

Rh
Fe
7
2.95 0.84 2.47 3.14 2.88
Fe
6
Rh 3.11 0.78 2.51 3.00 2.98 1.17
2.43
Fe
6
Rh* 3.09 1.08 2.40 3.00 3.00 1.06
2.34
Fe
5
Rh
2
3.25 0.73 2.45 2.86 3.12 1.19
2.46
Fe
5
Rh
2
* 3.23 1.00 2.50 2.86 3.14 1.19
2.41
Fe
4
Rh
3
3.36 0.62 2.42 2.71 3.27 1.26
2.55
2.71
Fe
4
Rh
3
* 3.33 0.94 2.41 2.71 3.26 1.26
2.46
2.57
Fe
3
Rh
4
3.38 0.57 2.48 2.28 3.25 1.16
2.37
2.67
Fe
3
Rh
4
* 3.36 0.67 2.62 2.28 3.29 1.13
2.50
2.60
Fe
2
Rh
5
3.41 0.45 2.40 2.14 3.34 1.33
2.62
Fe
2
Rh
5
* 3.38 0.56 2.37 2.14 3.23 1.39
2.42
2.55
FeRh
6
3.37 0.28 2.52 1.71 3.20 1.29
2.57
FeRh
6
* 3.25 0.37 2.40 1.76 3.211 1.24
2.59
Rh
7
3.33 0.22 2.61 1.86 1.62
atoms is larger. This is consistent with the previously discussed trend to favor the stronger
FeRh bonds. For example, the energy involved in changing the position of the Fe atom in
FeRh
6
from the apex (7 FeRh bonds) to the pentagonal ring (5 FeRh bonds) is 0.016 eV
per atom.
As the Fe content increases, the topology of Fe
m
Rh
n
changes. In fact Fe
4
Rh
3
cor-
responds to a CO, while for m 5 the conguration yielding the lowest energy can be
regarded as a strongly distorted PBP (see Fig. 4.5). Already in Fe
4
Rh
3
, but also in Fe
5
Rh
2
,
58
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
Fe
7
Fe
6
Rh Fe
6
Rh* Fe
5
Rh
2
Fe
5
Rh
2
*
Fe
4
Rh
3
Fe
4
Rh
3
* Fe
3
Rh
4
Fe
3
Rh
4
* Fe
2
Rh
5
Fe
2
Rh
5
* FeRh
6
FeRh
6
* Rh
7
Fe
8
Fe
7
Rh Fe
7
Rh* Fe
6
Rh
2
Fe
6
Rh
2
* Fe
5
Rh
3
Fe
5
Rh
3
* Fe
4
Rh
4
Fe
4
Rh
4
* Fe
3
Rh
5
Fe
3
Rh
5
*
Fe
2
Rh
6
Fe
2
Rh
6
* FeRh
7
FeRh
7
* Rh
8
Figure 4.5: Lowest energy isomers of FeRh heptamers and octamers.
59
4.3. Structure and magnetism
one observes a tendency of the Fe atoms to group in sub-clusters, bringing the Rh atoms
to outer positions, so that the number of FeRh bonds is largest. Concerning the shape of
the Fe-rich heptamers, one observes important deformations of the pentagonal bipyramid
(D
5h
symmetry) which are similar to the distortions found in pure Fe
7
[191, 192]. While
the precise origin of the symmetry lowering is dicult to establish in the alloys, it is
reasonable to expect that it is similar to the case of pure Fe
7
. According to Ref. [192],
the deformations found in Fe
7
are due to the presence of degenerate electronic states in
the undistorted PBP structure. In order to verify this hypothesis we have analyzed the
Kohn-Sham spectrum of the symmetric structure (D
5h
symmetry) and found that it is
highly degenerate at
F
. In contrast the spectrum of the distorted structure has a band
gap about 0.4 eV at
F
. This suggests that the distortions in the Fe-rich clusters can be
interpreted as a Jahn-Teller eect.
In Table 4.7 and Fig. 4.5 results for FeRh octamers are reported. The general trends
concerning the composition dependence of the binding energy, chemical order, as well as
the average and local magnetic moments are very similar to smaller clusters. The most
stable structure that we obtain for Fe
8
is a BCO having E
B
= 3.03 eV, an average magnetic
moment
8
= 3
B
, and a relatively short average bond-length d = 2.42

A (see Table 4.7).
A similar structure is also found in previous spin-polarized LDA calculations, where E
B
=
4.12 eV and
8
= 3
B
were obtained [193]. We have repeated these calculation for the BCO
structure with our computational parameters and atomic reference energies and found
E
B
= 3.51 eV. The discrepancies between LDA and GGA results reect the importance
of exchange and correlation to the binding energy. In the other extreme, for pure Rh
8
,
the structure that we nd with the considered starting topologies is a regular cube having
E
B
= 3.59 eV, an average magnetic moment
8
= 1.5
B
, and all bond lengths equal to
2.40

A. These results are in good agreement with previous calculations by Bae et al. [196].
It is interesting to observe that the substitution of a single Rh atom by Fe in FeRh
7
results in a compact topology, which is more stable than the relatively open (relaxed)
cube-like structures derived from pure Rh
8
. The same trend holds for higher Fe content
(i.e., Fe
m
Rh
8m
with m 1). The dominant structure for non-vanishing Fe content is
a BCO with slight distortions. Only for Fe
5
Rh
3
we nd a dierent topology, namely, a
distorted CPBP. The typical isomerization energies between the BCO and the TCTBP
are E
iso
= 1030 meV per atom. The average magnetic moments in the lowest lying
isomers are either the same or very similar.
The magnetic properties of heptamers and octamers follow qualitatively the behavior
observed in smaller clusters. In most cases the average magnetic moment per atom
N
and the magnetic energy E
m
increase with Fe concentration. The only exception is the
pure Rh heptamer, for which
7
is somewhat larger than in FeRh
6
. This is not due to
AF-like coupling between Fe impurity moment and the remaining Rh atoms but rather
to a reduction of the Rh local moments in FeRh
6
(
Rh
i
1.611.63
B
in Rh
7
, while

Rh
i
1.251.30
B
in FeRh
6
). Remarkably, the Rh local moments in Rh
7
are the largest
among all the heptamers. They amount to 87% of the total moment, which stresses the
importance of the local d-electron contributions. Also in Rh
8
one nds quite large local
moments, which are actually larger than the Rh moments in most Fe doped clusters. This
shows that for these sizes the Fe atoms do not necessarily increase the Rh moments by
simple proximity eects (see Tables 4.6 and 4.7). Nevertheless, a dierent behavior is
expected for larger N, where pure Rh clusters are no longer magnetic on their own. The
local Fe moments are strongly enhanced with respect to pure Fe
N
(
Fe
2.8
B
in Fe
7
60
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
Table 4.7: Structural, electronic and magnetic properties of FeRh octamers as obtained
from a restricted representative sampling of cluster topologies (see text).
Cluster E
B
E
m
d

N

Fe

Rh
Fe
8
3.03 0.76 2.42 3.00 2.77
Fe
7
Rh 3.19 0.73 2.43 2.87 2.88 1.04
2.47
Fe
7
Rh* 3.16 1.08 2.48 2.88 2.86 1.11
2.40
Fe
6
Rh
2
3.32 0.66 2.43 2.75 3.18 1.04
2.47
Fe
6
Rh
2
* 3.29 0.54 2.45 2.75 3.22 1.10
2.52
Fe
5
Rh
3
3.42 0.69 2.43 2.63 3.15 1.10
2.57
Fe
5
Rh
3
* 3.41 0.84 2.44 2.63 3.13 1.15
2.51
2.72
Fe
4
Rh
4
3.47 0.53 2.45 2.50 3.20 1.25
2.42
2.70
Fe
4
Rh
4
* 3.45 0.84 2.51 2.50 3.24 1.23
2.48
2.60
Fe
3
Rh
5
3.54 0.42 2.77 2.37 3.36 1.34
2.39
2.65
Fe
3
Rh
5
* 3.50 0.45 2.77 2.40 3.36 1.34
2.45
2.56
Fe
2
Rh
6
3.53 0.32 2.70 2.00 3.25 1.30
2.59
2.57
Fe
2
Rh
6
* 3.51 0.47 2.40 2.00 3.28 1.25
2.59
FeRh
7
3.49 0.23 2.45 1.62 3.27 1.18
2.57
FeRh
7
* 3.47 0.24 2.48 1.37 3.30 1.04
2.45
Rh
8
3.59 0.09 2.40 1.50 1.33
or Fe
8
) reaching values up to 3.36
B
, particularly when the Fe atoms are in a Rh rich
environment. As in the smaller clusters, this is a consequence of a charge transfer from
the Fe to the Rh atoms, which increases the number of polarizable Fe d-holes. Notice that
some kind of interaction between the Fe atoms seems to favor this eect, since the largest

Fe
are found for clusters having 2 or 3 Fe atoms rather than for the single Fe impurity.
Large Fe moments are also found in bulk FeRh alloys [161, 194].
61
4.4. Trends as a function of size and composition
To conclude this section it is interesting to compare the cluster results with available
experiments and calculations for macroscopic alloys [161, 194, 195]. Band structure calcu-
lations for the periodic Fe
0.5
Rh
0.5
alloy having a CsCl structure yield an antiferromagnetic
(AF) ground state, which is more stable than the ferromagnetic solution [195]. This is
qualitatively in agreement with experiments showing AF order when the Rh concentration
is above or equal to 50% [194]. In contrast our results for small clusters show a FM-like
order for all Rh concentrations, even for the pure Rh clusters. This is a consequence of
the reduction of local coordination number and the associated eective d-band narrowing,
which renders the Stoner criterion far easier to satisfy, and which tends to stabilize the
high-spin states with respect to the low-spin AF states. In fact, even in the bulk cal-
culations on FeRh, the energies of the AF and FM states are not very dierent, and a
coexistence of both solutions is found over a wide range of volumes [195]. Moreover, exper-
iment shows an AF to FM transition with increasing temperature, which is accompanied
by an enhanced thermal expansion [194]. Recent ab initio calculations have revealed the
importance of competing FM and AF exchange interactions in stoichiometric -FeRh [161].
Moreover, neutron diraction experiments [194] on Fe
1x
Rh
x
for 0.35 < x < 0.5 and cal-
culations [161] for x = 0.5 show that the Fe moments
Fe
are signicantly enhanced with
respect to
Fe
in pure -Fe, particularly in the FM state where it reaches values of about
3.2
B
[161, 194]. These bulk results are remarkably similar to the trends found in Fe
m
Rh
n
clusters over a wide range of compositions. As in the clusters, the induced Rh moments

Rh
play an important role in the stability of the FM phase. Bulk experiments [194] on
Fe
1x
Rh
x
yield
Rh
1
B
for 0.35 < x < 0.5 which is comparable to, though somewhat
smaller than the present cluster results.
4.4 Trends as a function of size and composition
The main purpose of this section is to focus on the dependence of the electronic and
magnetic properties of Fe
m
Rh
n
clusters as a function of size and composition for N =
m+n 8.
4.4.1 Binding energy and magnetic moments
In Fig. 4.6 the binding energy per atom E
B
is given as a function of the number of Fe
atoms m. Besides the expected monotonic increase of E
B
with increasing N, an interesting
concentration dependence is observed. For very small sizes (N 4) E
B
is maximal for
m = 1 or 2, despite the fact that E
B
is always larger for pure Rh than pure Fe clusters.
This indicates that in these cases the bonding resulting from FeRh pairs is stronger than
RhRh bonds. Only for m N 1, when the number of weaker FeFe bonds dominates,
one observes that E
B
decreases with increasing m. For larger sizes (N 5) the strength
of RhRh and FeRh bonds becomes very similar, so that the maximum in E
B
is replaced
by a range of Fe concentrations x = m/N 0.5 where E
B
depends very weakly on m.
In Fig. 4.7 the average magnetic moments
N
of Fe
m
Rh
n
are shown as a function
of m for N 8. As already discussed in previous sections,
N
increases monotonously,
with the number of Fe atoms. This is an expected consequence of the larger Fe local
moments and the underlying FM-like magnetic order. The average slope of the curves
tends to increase with decreasing N, since the change in concentration per Fe substitution
is more important the smaller the size is. The typical increase in
N
per Fe substitution is
62
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
0 1 2 3 4 5 6 7 8
m
1.0
1.5
2.0
2.5
3.0
3.5
B
i
n
d
i
n
g

E
n
e
r
g
y

[
e
V
/
a
t
o
m
]
Fe
m
Rh
n
N = 2
3
4
5
6
7
8
Figure 4.6: Binding energy per atom E
B
of Fe
m
Rh
n
clusters as a function of the number
of Fe atoms. The lines connecting the points for each N = m+n are a guide to the eye.
about (1/N)
B
per Fe substitution. Notice, moreover, the enhancement of the magnetic
moments of the pure clusters in particular for Fe
N
(m = N), which go well beyond 3
B
,
the value corresponding to a saturated d-band in the d
7
s
1
conguration. In contrast, the
moments of pure Rh
N
are far from saturated except for N = 2 and 7 (see Fig. 4.7 for
m = 0). In this context it is important to recall that a thorough global optimization, for
example, by considering a large number of initial topologies, could aect the quantitative
values of the magnetic moments for N = 7 and 8.
The local magnetic moments in the PAW sphere of the Fe and Rh atoms provide
further insight on the interplay between 3d and 4d magnetism in Fe
m
Rh
n
. In Fig. 4.8

Fe
and
Rh
are shown as a function of m for N = 68. The Fe moments are essentially
given by the saturated d-orbital contribution. For pure Fe clusters the actual values of

Fe
within the PAW sphere are somewhat lower than 3
B
due to a partial spill-o of the
spin-polarized density. Notice that the Fe moments increase as we replace Fe by Rh atoms
showing some weak oscillations as a function of m. The increase is rather weak for a single
Rh impurity in Fe
N1
Rh but becomes stronger reaching a more or less constant value as
soon as the cluster contains 2 or more Rh (m N 2, see Fig. 4.8). This eect can
be traced back to a d electron charge transfer from Fe to Rh which, together with the
extremely low coordination number, which yields a full polarization of the larger number
of Fe d holes. On the other side the Rh moments are not saturated and therefore are
more sensitive to size, structure and composition. The values of
Rh
are in the range of
11.5
B
showing some oscillations as a function of m. No systematic enhancement of
Rh
with increasing Fe content is observed. This behavior could be related to charge transfers
eects leading to changes in the number of Rh d electrons as a function of m.
The magnetic stabilization energy of FeRh clusters is shown in Fig. 4.9. One observes
that the pure Fe clusters show highest E
m
with a largest value of 1 eV for Fe
6
, while
most of the pure Rh clusters show small E
m
values. For smaller cluster sizes (N 5)
63
4.4. Trends as a function of size and composition
0 1 2 3 4 5 6 7 8
m
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5

]
Fe
m
Rh
n
N = 2
3
4
5
6
7 8
Figure 4.7: Total magnetic moment per atom
N
of Fe
m
Rh
n
clusters as a function of
number of Fe atoms. The symbols corresponding to each size are the same as in Fig. 4.6.
The lines connecting the points for each N = m+n are a guide to the eye.
0 1 2 3 4 5 6 7 8
1.0
1.5
2.0
2.5
3.0
3.5

]
N = 6
N = 7
N = 8
Fe
Rh
m
Figure 4.8: Local magnetic moment

at the Fe and Rh atoms as a function of the

number Fe atoms m.
the variation in E
m
as a function of the number of Fe atoms m display a somewhat non-
monotonous behavior. This is due to the fact that the change in E
m
is more signicant
the smaller the cluster is. The similar behavior are been already found in the binding
energy per atom and average magnetic moment (see Figs. 4.6 and 4.7).
It is interesting to analyze the role played by magnetism in dening the cluster structure
64
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
0 1 2 3 4 5 6 7 8
m
0
0.5
1

E
m

[
e
V
]
Fe
m
Rh
n
N = 2
3
4
5
6
7
8
Figure 4.9: The magnetic stabilization energy E
m
= E(S
z
= 0) E(S
z
) of Fe
m
Rh
n
clusters as a function of number of Fe atoms. The symbols corresponding to each size are
the same as in Fig. 4.6. The lines connecting the points for each N = m + n are a guide
to the eye.
by comparing magnetic and non-magnetic calculations. For the smallest FeRh clusters
(N = 3 and 4) the magnetic energy E
m
= E(S
z
=0) E(S
z
) gained upon magnetization
is higher in the rst excited isomer than in the most stable structure. This implies that the
contribution of magnetism to the structural stability is not crucial, since the non-magnetic
calculations yield the same ordering, at least concerning the two best structures. This
suggests that for the smallest sizes the kinetic or bonding energy dominates the structural
stability, which also explains that the two most stable isomers have dierent topologies.
The situation changes for large clusters. For N 5 one nds a number of FeRh clusters for
which the optimal structure is actually stabilized by magnetism. For example, in Fe
4
Rh,
Fe
3
Rh
2
, and FeRh
4
the energy ordering of the two most stable isomers would be reversed
if magnetism were neglected. It should be noted that in these cases the structures dier
only in the chemical order, not in the topology which is a TBP. In the FeRh hexameters
the energy dierences between the low-lying isomers are more important and only in one
case, Fe
4
Rh
2
, magnetism appears to be crucial for stabilizing the actual optimal structure.
A similar strong interplay between structure, chemical order and magnetism is expected
for larger FeRh clusters.
4.4.2 Relative stability
In order to measure the relative stability, we plot in Fig. 4.10 the second energy dierence

2
E(N, m) = E(N, m + 1) + E(N, m1) 2E(N, m)

4.4.1
as a function of the number of Fe atoms m for dierent cluster sizes N. One observes that
some clusters show higher stability than their adjacent clusters for the dierent composi-
65
4.4. Trends as a function of size and composition
tions with the same cluster size N and also among the dierent N. Among the octamers,
FeRh
7
and Fe
3
Rh
5
clusters display highest and lowest stability, respectively. This can be
justied by analyzing Figs. 4.5 and 4.11 for the number of dierent bonds and dierent
average bond length, respectively. For the optimal FeRh
7
cluster, there are 6 FeRh bonds
(with average bond length of 2.45

A) and 12 RhRh bonds (with average bond length of

2.57

A). Notice that this cluster has no FeFe bond which we have shown to be weakest
among the dimers (see Table. 4.1). In the case of Fe
3
Rh
5
cluster, there are 11 FeRh
bonds (with average bond length of 2.39

A), 6 RhRh bonds (with average bond length of

2.65

A) and 3 FeFe bonds (with average bond length of 2.77

A). It can be argued that the

absence of FeFe bond provides higher relative stability to the FeRh
7
cluster compared to
the Fe
3
Rh
5
cluster where 3 FeFe bonds are present.
1 2 3 4
m
-0.15
-0.10
-0.05
0.00
0.05
0.10

2
E
(
5
)

[
e
V
]
1 2 3 4 5
m
-0.15
-0.10
-0.05
0.00

2
E
(
6
)

[
e
V
]
1 2 3 4 5 6
m
-0.10
-0.08
-0.06
-0.04
-0.02
0.00
0.02

2
E
(
7
)

[
e
V
]
1 2 3 4 5 6 7
m
-0.10
-0.05
0.00
0.05
0.10
0.15

2
E
(
8
)

[
e
V
]
Figure 4.10: Relative stability of Fe
m
Rh
n
clusters as a function of composition for dierent
N = m+n,
2
E(N, m) = E(N, m+1) +E(N, m1) 2E(N, m) is given as a function of
number of Fe atoms. The symbols corresponding to each size are the same as in Fig. 4.6.
The lines connecting the points for each N = m+n are a guide to the eye.
66
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
1 2 3 4
m
2.3
2.4
2.5
2.6
2.7
2.8
d

(

)
1 2 3 4 5
m
2.3
2.4
2.5
2.6
2.7
2.8
d

(

)
1 2 3 4 5 6
m
2.3
2.4
2.5
2.6
2.7
2.8
d

(

)
1 2 3 4 5 6 7
m
2.3
2.4
2.5
2.6
2.7
2.8
d

(

)
N = 5 N = 6
N = 7 N = 8
Figure 4.11: Bond lengths for FeFe (cross), FeRh (circle) and RhRh (square) pairs as a
function of number of Fe atoms in Fe
m
Rh
n
clusters having N = 58 atoms. The lines
connecting the points for each N = m+n are a guide to the eye.
4.4.3 Electronic structure
In the previous sections the structure and spin moments of FeRh alloy clusters have been
discussed as a function of 3d/4d concentration. Although these properties are intimately
related to the size and composition dependence of electronic structure, it is in general
very dicult to achieve a physical transparent correlation between global and microscopic
behaviors. Nevertheless, it is very interesting to analyze, at least for some representative
examples, how the electronic structure depends on the composition of magnetic nanoalloys.
Fig. 4.12 shows the spin-polarized d-electron density of states (DOS) of representative
FeRh octamers having the relaxed structures illustrated in Fig. 4.5. Results for pure Fe
8
and Rh
8
are also shown for the sake of comparison. The spin-polarized DOS can be given
by
N

() =
1
N

k
(
k
)

4.4.2
represents the number of electronic states per unit energy and per atom having spin and
energy . The eigenenergies
k
are the eigenvalues of the Kohn-Sham equations.
In all the clusters, the dominant peaks in the relevant energy range near
F
correspond
either to the Fe-3d or to the Rh-4d states. The valence spectrum is largely dominated by
67
4.4. Trends as a function of size and composition
these d-electron contributions. In fact the total DOS and the d-projected DOS are dicult
to tell apart.
-50
0
50
-50
0
50
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
-5 -4 -3 -2 -1 0 1 2

F
[eV]
-50
0
50
-5 -4 -3 -2 -1 0 1 2

F
[eV]
(a)
(b)
(c)
(d)
(e) (f)
Rh
8
Fe
2
Rh
6
Rh
8
Fe
4
Rh
4
Fe
6
Rh
2
Fe
8
Cube Bicapped octahedron
Figure 4.12: Electronic density of states (DOS) of FeRh octamers. Results are given
for the total (solid), the Fe-projected (dotted), and the Rh-projected (dashed) d-electron
DOS. Positive (negative) values correspond to majority (minority) spin. A Lorentzian
width = 0.02 eV has been used to broaden the discrete energy levels. The considered
structures are the optimal ones illustrated in Fig. 4.5.
First of all, let us consider the DOS of the pure clusters. Our results for Rh
8
with
a cube structure are similar to those of previous studies [196]. They show the dominant
d-electron contribution near
F
, with the characteristic ferromagnetic exchange splitting
between the minority and majority spin states. In Fig. 4.12 we also included the DOS for
Rh
8
with a BCO structure, since it allows us to illustrate the dierences in the electronic
structure of compact and open geometries. Moreover, the DOS of pure Rh
8
with BCO
structure is very useful in order to demonstrate the dependence of DOS on Fe content, since
the structures of Fe
m
Rh
8m
with m 1 are similar to the BCO. Both Fe
8
and Rh
8
show
relatively narrow d-bands which dominate the single-particle energy spectrum in the range
5eV
F
3eV. The spin polarization of the DOS clearly reects the ferromagnetic
68
Chapter 4. Interplay of structure, magnetism and chemical order in small FeRh clusters
order in the cluster. Putting aside the exchange splitting, the peak structure in the up and
down DOS

() are comparable. There are even qualitative similarities between the two
elements. However, looking in more detail, one observes that the eective d-band width
in Fe
8
(about 4 eV) is somewhat smaller than in Rh
8
(about 5 eV). Moreover, in Rh
8
the DOS at
F
is non-vanishing for both spin directions and the nite-size gaps are very
small (see Fig. 4.12). In contrast, the majority d-DOS is fully occupied in Fe
8
, with the
highest majority state lying about 0.5 eV below
F
. In addition there is an appreciable
gap (about 0.1 eV) in the corresponding minority spectrum. These qualitative dierences
are of course consistent with the fact that Fe
8
is a strong ferromagnet with saturated
moments, while Rh
8
should be regarded as a weak unsaturated ferromagnet.
The trends as a function of concentration reect the crossover between the previous
contrasting behaviors. For low Fe concentration (e.g., Fe
2
Rh
6
) we still nd states with
both spin directions close to
F
. The magnetic moments are not saturated, although the
Fermi energy tends to approach the top of the majority band. Moreover, the majority-spin
states close to
F
have dominantly Rh character. Small Fe doping does not reduce the
d-band width signicantly. Notice the rather important change in the shape of the DOS in
Fe
2
Rh
6
as compared to the DOS in Rh
8
. This is a consequence of the change in topology
from cubic to bicapped octahedron.
For equal concentrations (Fe
4
Rh
4
) the rst signs of d-band narrowing and enhanced
exchange splitting start to become apparent. The spin-up states (majority band) which
in Fe
2
Rh
6
contribute to the DOS at
F
now move to lower energies (0.3 eV below
F
) so
that the majority band is saturated. Only spin-down (minority) states are found around

F
, although there is a signicant gap in

() (see Fig. 4.12). In the majority band Rh

dominates over Fe at the higher energies (closer to
F
), while Fe dominates in the bottom
of the band. In the minority band the participation of Rh (Fe) is stronger (weaker) below

F
and weaker (stronger) above
F
. This is consistent with the fact that the Rh local
moments are smaller than the Fe moments.
Finally, in the Fe rich limit (e.g., Fe
6
Rh
2
), the majority-band width becomes as narrow
as in Fe
8
, while the minority band is still comparable to Rh
8
. The exchange splitting is
large, the majority band saturated and only minority states are found close to
F
. As
in Fe
8
,

() shows a clear gap at

F
(see Fig. 4.12). However, the Rh contribution to
the minority states below
F
remains above average despite the relative small Rh content.
The Fe contribution largely dominates the unoccupied minority-spin DOS, in agreement
with the larger local Fe moments.
4.5 Summary
The structural, electronic and magnetic properties of small Fe
m
Rh
n
clusters having N =
m+n 8 atoms have been investigated systematically in the framework of a generalized
gradient approximation to density-functional theory. For very small sizes (N 4 atoms)
the binding energy E
B
shows a non-monotonous dependence on concentration, which
implies that the FeRh bonds are stronger than the homogeneous ones. However, for larger
sizes the FeRh and RhRh bond strengths become comparable, so that E
B
depends weakly
on concentration for high Rh content.
The magnetic order of the clusters having the most stable structures is found to be
FM-like. Moreover, the average magnetic moment per atom
N
increases monotonously,
69
4.5. Summary
which is almost linear over a wide range of concentration with Fe content. Consequently,
the energy gain E
m
associated to magnetism also increases with the number of Fe atoms.
The largest part of the spin polarization (about 90%) can be traced back to the local d
magnetic moments within the PAW sphere of the atoms. The s and p spin polarizations
are almost negligible in general. A remarkable enhancement of the local Fe moments is
observed as result of Rh doping. This is a consequence of the increase in the number of Fe
d holes, due to charge transfer from Fe to Rh, combined with the extremely reduced local
coordination. The Rh local moments are important already in the pure clusters (N 8).
Therefore, they are not signicantly enhanced by Fe doping. However, the overall stability
of magnetism, as measured by the total energy gained by the onset of spin polarization,
is found to increase with increasing Fe content.
A further interesting aspect, particularly for future studies is the role of spin-orbit (SO)
interactions on the magnetism of nanoalloys. We have performed some representative
calculations by taking into account spin-orbit coupling (SOC) in order to explore their
eect on the ground-state structure, chemical order and spin moments. For example
in Fe
6
, Fe
3
Rh
3
and Rh
6
we nd that the changes in the ground-state energy resulting
from SO interaction are typically of the order of 0.2 eV for the whole cluster. This is
often comparable to or larger than the energy dierences between the low-lying isomers.
However, the SO energies are very similar for dierent structures, so that the ground-state
structures remain essentially the same as in the scalar relativistic (SR) calculations. The
changes in the bond lengths and in the spin moments resulting from SOC are also very
small (e.g., |
SOC

SR
| 0.01
B
and |d
SOC
ij
d
SR
ij
| 0.001

A in Fe
3
Rh
3
). As a result,
the conclusions drawn from our SR calculations on the relative stability and local spin
moments seem to be unaected by the spin-orbit contributions. The eect of spin orbit
interaction are discussed in more detial in chapter 7.
70
5
Structure and magnetism of small CoPd clusters
In the present chapter we discuss the electronic and magnetic properties of Co
m
Pd
n
clus-
ters having N = m+n 8. Our study shown that the optimized cluster structures have
a clear tendency to maximize the number of nearest-neighbor CoCo pairs. The magnetic
order is found to be ferromagnetic-like for all the ground-state topologies. However, AF-
like order has also been found in some rst exited isomers. The binding energies and
average magnetic moments show an approximately linear increase with Co concentration.
This is consistent with the trends found in the solid where E
B
is larger in Co than in Pd
[E
B
(Co-bulk) = 4.39 eV/atom and E
B
(Pd-bulk) = 3.89 eV/atom]. The maximal local
spin polarization for Co and Pd atoms are found in the equiatomic or nearly equiatomic
compositions (e.g., Co
3
Pd
3
, Co
3
Pd
4
and Co
4
Pd
4
). A signicant enhancement of the Co
local moments is found as a result of Pd doping. This is a consequence of the increase in
the number of Co d holes, due to Co to Pd charge transfer, combined with the extremely
reduced local coordination numbers.
5.1 Introduction
The central goal in the eld of nanoalloys is to explore, understand, and characterize the
rich variety of alloy properties at the nanoscale as a function of size and composition [150].
By using dierent experimental and theoretical tools, now it is known that combining
two metallic elements to form nanoalloys leads to an even higher degree of novelty and
complexity [1, 2, 47, 65, 151, 157159, 164]. The elements considered in the present study,
Co and Pd, have very interesting properties in the low dimensional non-scalable regime. Co
clusters show enhanced magnetic moments compared to the bulk [65, 158, 159], whereas
Pd clusters display a nite magnetic moment despite being non-magnetic in the solid
[10, 183, 205, 211]. The magnetic properties of free standing Co
N
clusters were investigated
via Stern-Gerlach molecular beam deection experiment by Bloomeld et al. in the size
range N = 20200 (see Fig. 5.1 as well as Ref. [5]) and by de Heer et al. for N = 30300
Ref. [6]. These studies shown that in the temperature range of 77-300 K, the Co
N
clusters
71
5.1. Introduction
Figure 5.1: The average magnetic moment per atom of Co
N
clusters as a function of the
cluster size N (after Ref. [5]).
show high-eld deections which are characteristic of supermagnetic behavior. There
are also many theoretical results available for Co clusters. Castro et al. [200] performed
all-electron density-functional calculations using both the local density and generalized
gradient approximations. However, the size of the clusters was limited to only up to ve
atoms. Later on, Lopez et al. [202] studied Co
N
clusters (4 N 60), by performing
geometry optimization by using an evolutive algorithm based on a many-body Gupta
potential [203]. The magnetic properties of the cluster structures were studied by a spd
tight-binding method.
The magnetism of small Pd clusters is still a subject of debate. Photoemission exper-
iments [205] predicted a Ni-like spin arrangement in Pd
N
clusters having N 6 and a
Pt-like non-magnetic behavior for N 15. DC susceptibility measurements [206] found
magnetic moments of 0.230.19
B
/atom in huge Pd clusters with diameters in the range
of 5070

A. Recent experiment [204] by using gas-evaporative method in a high purity
Ar gas atmosphere has revealed a magnetic moment of 0.750.31
B
/atom for ne Pd
particles. A large number of theoretical studies have been devoted to the study of the Pd
clusters. The DFT calculations by M. Moseler et al. [10] have shown that both neutral
and anionic Pd
N
clusters having 2 N 7 and N = 13 atoms are magnetic. Moreover,
there are some theoretical studies on CoPd lms. J. Dorantes-Davila et al. [209] carried
out tight-binding calculations by depositing Co 1D chains on Pd(110) surface. They found
that small magnetic moments are induced (
Pd
= 0.21 0.29)
B
for Pd atoms nearest
neighbor of Co. In view of these interesting behaviors one expects that CoPd clusters
should show very interesting structural, electronic and magnetic properties.
The remainder of this chapter is organized as follows. In section. 5.2 the main details
of the theoretical background and computational procedure are presented. The results
of our calculation for CoPd clusters having N 8 atoms are reported in Secs. 5.3 and
5.4. Here, we focus on the interplay between structure, chemical order and magnetism in
72
Chapter 5. Structure and magnetism of small CoPd clusters
the most stable geometries for dierent cluster sizes. We also analyze the concentration
dependence of the cohesive energy, the local and average magnetic moments, the relative
stability and the magnetic stabilization energy. Finally, we conclude in section. 5.5 with
a summary of main trends.
5.2 Computational details
The structural, electronic and magnetic properties of small Co
m
Pd
n
clusters having N =
m + n 8 atoms has been studied using DFT and the VASP [11, 168] as in the chapter
4. Electronic exchange and correlation are described in the generalized gradient approx-
imation, by means of the functional proposed by Perdew and Wang (PW91) [170]. As
discussed in Sec. 2.2 the VASP solves the spin-polarized Kohn-Sham equations in an aug-
mented plane-wave basis set, taking into account the core electrons within the projector
augmented wave (PAW) method [170]. The PAW method has extensively been described
in section. 2.4. The 4s and 3d orbitals of Co, and the 5s and 4d orbitals of Pd are treated as
valence states. The PAW sphere radii for Co and Pd are 1.302

A and 1.434

A, respectively.
The validity of the present choice of computational parameters has been veried. A
number of tests have been performed in order to assess the numerical accuracy of the
calculations. For example, increasing the cuto and supercell size E
max
= 268 eV, a =
12

A to 500 eV, a = 22

A in Co
4
yields a total energy dierences of only 1.0 meV/atom and
0.2 meV/atom, respectively. This implies an increase in the computation time by a factors
of 37. In these calculations above calculations we obtain changes in the average bond
length (bond angle) of only 10
3

A (10
4
degrees). Such dierence are not signicant
for our physical conclusions. In fact typical isomerization energies in CoPd clusters are
an order of magnitude larger, of the order of 1050 meV/atom. We also found that the
total energy is essentially independent of the choice of the smearing parameter values ,
provided it is suciently small ( 0.05 eV). Values of between 0.01 to 0.1 eV have
been explicitly tested. We conclude that set of standard parameters E
max
= 268 eV,
supercell size a from 10 to 22

A and smearing parameter 0.02 eV oer a good accuracy at
a reasonable computational costs.
The rest of the computational procedure employed in this chapter is analogous to the
one used in the chapter 4 in the case of FeRh. This concerns in particular the thorough
sampling of cluster topologies, total magnetic moments S
z
, and all possible chemical orders.
Further methodological details may be found in Sec. 4.2.
5.3 Results and discussion
In this section we discuss the structure, chemical order, binding energy and magnetic
moments of Co
m
Pd
n
clusters having N = m + n 8 atoms. Here the most important
electronic and magnetic properties are analyzed for the dierent cluster sizes N and for
all compositions. The general trends as a function of size and concentration are discussed
in Sec. 5.4.
73
5.3. Results and discussion
Table 5.1: Structural, electronic and magnetic properties of CoPd dimers. Results are
given for the binding energy E
B
(in eV/atom), point group symmetry (PGS) of the struc-
ture, the magnetic stabilization energy E
m
= E(S
z
= 0) E(S
z
) (in eV/atom), the
average interatomic distance d

(in

A) between atoms and (, = Co or Pd), the
average spin moment per atom
N
= 2S
z
/N (in
B
), and the local spin moment

at
the Co or Pd atoms.
Cluster PGS E
B
E
m
d

N

Co

Pd
Co
2
D
h
1.53 0.67 1.96 2.0 1.93
CoPd D
h
1.19 0.57 2.23 1.5 2.42 0.50
Pd
2
D
h
0.64 0.31 2.48 1.0 0.60
5.3.1 Dimers
The result for dimers are summarized in Table 5.1. Quantitatively, one observes that
the CoCo dimer yields the highest cohesive energy, followed by the CoPd dimer and the
PdPd dimer. This is qualitatively consistent with the known cohesive energies in the
corresponding solids. Moreover, the bond length follows the trend in the atomic radii,
namely d
PdPd
> d
CoPd
> d
CoCo
. Let us consider the homogeneous cases rst. Our
calculations yield E
B
= 1.53 eV, d
CoCo
= 1.96

A, and
2
= 2
B
for Co
2
. These results are
similar to those reported in the previous DFT studies by S. Datta et al. [198] using the
GGA. However, the collision-induced dissociation experiments [199] reported a somewhat
lower E
expt
B
(Co
2
) = 1.32 eV. In the case of Pd
2
, our results are E
B
= 0.64 eV, d
PdPd
=
2.48

A and
2
= 1
B
. Notice the bond length contractions with respect to the bulk values
(d
bulk
CoCo
= 2.51

A and d
bulk
PdPd
= 2.80

A). These results are in good agreement with earlier
DFT studies [10, 183, 207]. In contrast, Barretau et al. [212] found a non-magnetic ground
state for Pd
2
by using a spd tight-binding model. The binding energy calculated in the
GGA compares reasonably well with available experimental results which scatter between
0.73 and 1.13 eV/atom [214]. High-level quantum chemistry calculations [215] also predict
a triplet ground- state, having d
PdPd
= 2.48

A, and a somewhat smaller E
B
= 0.43 eV.
Finally, the calculations based on hybrid functionals [213] yield E
B
= 0.48 eV/atom and
d
PdPd
= 2.53

A. Concerning the local magnetic moments in the WS cell, one observes
contrasting behaviors in the Co and Pd dimers. In Co
2
, the largest part of the spin
magnetization is localized close to the atom:
WS
Co
= 1.93
B
in the WS cell while
2
=
2
B
. In contrast, in the case of Pd
2
, a signicant part of the spin polarization is found
in the spill-o region, beyond WS radius r
WS
Pd
= 1.434

A, since
WS
Pd
= 0.6
B
while
2
=
1
B
.
For the CoPd dimer we found E
B
= 1.19 eV, d
CoPd
= 2.23

A, and
2
= 1.5
B
. Notice
that the local Co moment
Co
= 2.42
B
is signicantly enhanced by, 0.49
B
, while
Pd
is
reduced by 0.1
B
as compared to the respective pure dimers (see Table 5.1). A signicant
charge transfer from Co to Pd can be inferred from this. Indeed, an integration of the
electronic density in the Bader cells [175] shows that 0.11 electrons are transferred from the
Co to the Pd atom. This behavior is qualitatively in agreement with the higher Pauling
electronegativity of the Pd atom (
Co
= 1.88 and
Pd
= 2.20).
74
Chapter 5. Structure and magnetism of small CoPd clusters
Table 5.2: Structural, electronic and magnetic properties of CoPd trimers as in Table 5.1.
The average interatomic distance d

(in

A) ordered from top to bottom as d
CoCo
, d
CoPd
and d
PdPd
. The most stable structures are illustrated in Fig 5.3.
Cluster PGS E
B
E
m
d

N

Co

Pd
Co
3
C
2v
1.87 0.57 2.16 2.00 1.94
Co
2
Pd C
2v
1.84 0.47 2.01 1.33 1.95 0.06
2.44
Co
2
Pd* D

1.66 0.61 2.03 2.00 2.32 0.12

2.28
CoPd
2
C
2v
1.70 0.19 2.31 1.00 2.25 0.33
2.62
CoPd
2
* D

1.41 0.34 2.23 1.44 2.46 0.64

Pd
3
D
3h
1.25 0.10 2.52 0.67 0.63
5.3.2 Trimers
Table 5.2, Fig. 5.2 and Fig. 5.3 summarize all the important results obtained for trimers.
The lowest energy isomers are found to be the triangle for all the compositions. For Co
3
,
the ground-state structure is an isosceles triangle with E
B
= 1.87 eV, d
CoCo
= 2.16

A, and

2
= 2
B
. The highest occupied molecular orbital (HOMO) is degenerate, and partially
occupied (see Fig. 5.2). These results are in agreement with the all-electron (AE) DFT
calculations by M. Castro et al. [200]. The linear Co
3
is found to be 0.43 eV/atom less
stable than the ground-state. The lowest energy state for Pd
3
is an equilateral triangle with

3
= 0.67
B
. This result is similar to the previous DFT study reported by T. Futschek et
al. [183]. The DOS of Pd
3
shows that the HOMO is degenerate and partially occupied.
Therefore a Jahn-Teller distortion could have been expected. However, our calculation do
not display such an instability. In fact, Valerio and Toulhaot [213] have shown that Pd
3
is stable against Jahn-Teller distortions at the GGA level, but unstable when a hybrid
functional combining DFT and exact exchange is used or if the calculations are performed
at the Hartree-Fock plus conguration interaction (HF+CI) level. For Co
2
Pd, the lowest
energy state is an isosceles triangle with E
B
= 1.84 eV, and
3
= 1.33
B
. The local
moments are coupled ferromagnetically. A signicant enhancement in the local moments
could be expected as in the CoPd dimer. Surprisingly, there is no such an enhancement in
the
Co
, which is essentially the same as in Co
3
. Moreover, the Pd has nearly vanishing
moment. The lack of enhancement of
Co
is likely to be related to the rather contracted
CoCo bond length d
CoCo
= 2.01

A (see Table 5.2). We found that the linear Pd-Co-
Co (Co-Pd-Co) structures are 0.18 (0.48) eV/atom less stable than the the ground-state
structure. The lowest energy state of CoPd
2
is also an isosceles triangle (C
2v
) with E
B
=
1.7 eV and
3
= 1
B
. Unlike the ground-state Co
2
Pd, the CoPd
2
cluster shows a large
enhancement in the
Co
= 2.25
B
. This is due to the combined eect of larger CoCo
interatomic distance (d
CoCo
= 2.31

A) and to the signicant charge transfer from Co to

Pd. While,
Pd
in CoPd
2
is rather reduced compared to Pd
3
local moment.
75
5.3. Results and discussion
-4 -2 0 2 4 6
-40
-20
0
20
40
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
-4 -2 0 2 4 6
-40
-20
0
20
40
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
-4 -2 0 2 4 6

F
[eV]
-40
-20
0
20
40
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
-4 -2 0 2 4 6

F
[eV]
-80
-40
0
40
80
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
Co
3
Co
2
Pd
1
Co
1
Pd
2
Pd
3
Figure 5.2: Electronic density of states (DOS) of CoPd trimers. Results are given for the
total DOS. Positive (negative) values correspond to majority (minority) spin. A Lorentzian
width = 0.02 eV has been used to broaden the discrete energy levels.
Co
3
Co
2
Pd Co
2
Pd* CoPd
2
CoPd
2
*
Pd
3
Co
4
Co
3
Pd Co
3
Pd* Co
2
Pd
2
Co
2
Pd
2
* CoPd
3
CoPd
3
* Pd
4
Figure 5.3: Lowest energy isomers of CoPd trimers and tetramers. The asteriks indicate
rst exited isomers.
76
Chapter 5. Structure and magnetism of small CoPd clusters
Table 5.3: Structural, electronic and magnetic properties of CoPd tetramers as in Ta-
ble 5.2.
Cluster PGS E
B
E
m
d

N

Co

Pd
Co
4
S
4
2.37 0.72 2.43 2.50 2.19
Co
3
Pd D
2h
2.18 0.52 2.18 1.75 2.10 0.32
2.40
Co
3
Pd* S
4
2.13 0.53 2.45 1.75 2.06 0.26
2.48
Co
2
Pd
2
D
2h
1.874 0.44 2.39 1.50 2.30 0.43
2.62
Co
2
Pd
2
* S
4
1.87 0.18 2.24 -0.01 -0.02 0.01
2.47
2.63
CoPd
3
S
4
1.94 0.41 2.37 0.75 2.18 0.24
2.71
CoPd
3
* D
2h
1.83 0.42 2.36 0.75 2.29 0.24
2.63
Pd
4
T
d
1.67 0.09 2.61 0.50 0.47
5.3.3 Tetramers
Table 5.3 and Fig. 5.3 summarize the relevant geometric and magnetic information ob-
tained for the ground-state tetramers. The lowest energy structures are tetrahedra for m
= 0, 1 and 4, and rhombi for m = 2 and 3. Co
4
has a Jahn-Teller distorted ground-state
with
4
= 2.5
B
. In this cluster there are two kinds of bond lengths: two pairs have equal
short bond lengths d = 2.14

A, while the third pair has a much larger d = 2.72

A. The
shorter bond lengths should enhance the binding since the localized 3d electrons can also
take part in the bonding. The rst excited isomer is a rhombi, which is 0.11 eV less stable
than the ground-state structure. Our results for Co
4
are similar to the previous DFT
calculations by S. Datta et al. [198]. So far, there are, to our knowledge, no experimental
result available on the neutral tetramers. However, Yoshida et al. [201] predicted a tetra-
hedron with a bond length of 2.25 0.2

A as the optimal structure for Co

4
anion. The
most stable structure obtained for Pd
4
is a tetrahedron with d = 2.61

A and
4
= 0.50
B
.
This result is in good agreement with previous DFT studies [10, 183, 207]. The HOMO
(minority spin) is degenerate and partially occupied by only one electron (see Fig. 5.4).
The corresponding DOS is similar to the one reported by V. Kumar et al. [207].
Let us move further to Co
3
Pd. The optimal structure is a rhombus and the rst exited
isomer is a tetrahedron lying 41 meV/atom above the ground-state. The rhombus is also
found to be the optimal structure for Co
2
Pd
2
. This case is an ideal example to study the
inuence of chemical order on the binding energy and magnetism. Notice that E
B
has
been reduced in the ground-state of Co
2
Pd
2
as compared to the ground-state of Co
3
Pd
and CoPd
3
. This is probably due to the presence of a Pd dimer, which is weakest among
the dimers (see Fig. 5.3). The local moments
Co
and
Pd
show somewhat enhanced
77
5.3. Results and discussion
-4 -2 0 2 4 6
-60
-40
-20
0
20
40
60
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
-4 -2 0 2 4 6
-60
-40
-20
0
20
40
60
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
-4 -2 0 2 4 6

F
[eV]
-60
-40
-20
0
20
40
60
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
-4 -2 0 2 4 6

F
[eV]
-60
-40
-20
0
20
40
60
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
-4 -2 0 2 4 6
-60
-40
-20
0
20
40
60
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
Co
4
Co
3
Pd
1
Co
2
Pd
2
Co
1
Pd
3
Pd
4
Figure 5.4: Electronic density of states (DOS) of CoPd tetramers.
values 2.30
B
and 0.43
B
, respectively. This is because Co atoms are farther away and
weakly bonded. Consequently, Co d states will be more localized and provide a signicant
enhancement to the resulting local spin polarization. The rst isomer is a tetrahedron and
which is only 4 meV/atom less stable. Here the magnetic order is found to be AF-like. The
local moments are distributed as
Co1
= -1.73
B
,
Co2
= 1.71
B
,
Pd1
= 0.03
B
,
Pd2
= -0.02
B
. The small but still signicant energy dierence of 4 meV/atom between FM-
and AF-like magnetic order pointing towards a strong competition between the dierent
magnetic isomers to become the ground-state structure. In the rich Pd limit (CoPd
3
), the
optimal structure is a distorted tetrahedron and the rst excited isomer is a rhombus. One
observes that ground-state and rst excited isomer have the same
4
= 0.75
B
and
Pd
= 0.24
B
, but the latter shows 0.11
B
enhancement for the
Co
compared to the former
(see Table 5.3).
5.3.4 Pentamers
The results are tabulated in Table 5.4, and the surface of constant magnetization den-
sity is shown in Fig. 5.5. The optimal structures are trigonal bipyramids (TBP) for all
composition expect for Co
4
Pd, where a square pyramid (SP) is obtained. For Co
5
, the
ground-state structure is a trigonal bipyramid (D
3h
) having E
B
= 2.65 eV,
5
= 2.60

B
and d = 2.41

A. In this structure there are two types of bond-lengths: all the sides of
upper and lower triangular pyramids have the same small length d = 2.18

A, while the NN
78
Chapter 5. Structure and magnetism of small CoPd clusters
Table 5.4: Structural, electronic and magnetic properties of CoPd pentamers as in 5.2.
Cluster PGS E
B
E
m
d

N

Co

Pd
Co
5
D
3h
2.65 0.61 2.41 2.60 2.24
Co
4
Pd C
4v
2.50 0.56 2.19 2.00 2.12 0.37
2.48
Co
4
Pd* D
3h
2.47 0.47 2.32 1.60 1.91 0.18
2.36
Co
3
Pd
2
D
3h
2.35 0.42 2.24 1.40 2.05 0.25
2.48
2.73
Co
3
Pd
2
* C
4v
2.32 0.47 2.14 1.78 2.25 0.48
2.51
2.57
Co
2
Pd
3
D
3h
2.19 0.30 2.10 1.20 2.16 0.31
2.47
2.70
Co
2
Pd
3
* D
3h
2.14 0.39 2.45 1.20 2.25 0.31
2.67
CoPd
4
D
3h
2.05 0.20 2.43 0.60 2.16 0.19
2.76
CoPd
4
* C
4v
2.03 0.20 2.43 0.60 2.14 0.19
2.62
Pd
5
D
3h
1.87 0.05 2.65 0.40 0.38
distances d = 2.65

A along the triangle shared by the two triangular pyramids are much
larger. Our results coincide with the previous DFT calculations by S. Datta et al. [198].
The optimal Pd
5
is a TBP with E
B
= 1.87 eV, and
5
= 0.40
B
. The length of the
bonds shared by both triangular pyramids along the triangle is 2.66

A, whereas the rest
of bonds are slightly shorter by 0.03

A. This result is in agreement with previous DFT
results [10, 183]. The rst isomer is a SP structure, which is 5.9 meV less stable.
The optimal Co
4
Pd is a SP and the average moment
5
is reduced by 0.60
B
compared
to the pure Co
5
. The Pd atom is located on the most coordinated cap position with
Pd
= 0.37
B
. The rst excited isomer is a TBP, which is 27 meV less stable. In the optimal
Co
3
Pd
2
, Co atoms form an isosceles triangle on the triangular face of TBP. SP is found to
be the rst excited isomer, which is 31 meV less stable than the ground-state structure.
As one goes from Co
5
to Co
3
Pd
2
over Co
4
Pd,
5
is reduced by about 0.60
B
per Pd atom
doping. However, only a small reduction of about 0.20
B
in the
5
is found as one goes
from Co
3
Pd
2
to Co
2
Pd
3
and CoPd
4
to Pd
5
(see Table 5.4). The binding energy per atom
shows an approximately linear decrease as a function of the number of Pd atoms, since
more the Pd bonds are, weaker the cluster. The surface of magnetization density shown
in Fig. 5.5 for Pd
5
coincides with previous DFT results by T. Futschek et al. [183].
79
5.3. Results and discussion
Co
5
= 0.02
Co
4
Pd
= 0.015
Co
3
Pd
2
= 0.006
Co
2
Pd
3
= 0.007
CoPd
4
= 0.009
Pd
5
= 0.008
Figure 5.5: Constant magnetization density plots

(r)

(r) = and local moments

(in
B
) for the ground-state structures of CoPd pentamers. The value of the constant
magnetization density is given in
B
/

A
3
.
5.3.5 Hexamers
The results for hexamers are summarized in Table 5.5 and Figs. 5.6, 5.7 and 5.8. The
optimal structures for Co
6
, CoPd
5
and Pd
6
clusters is an octahedra and capped-TBP
(CTBP) are found for the rest of the CoPd hexamers.
The optimal Co
6
has E
B
= 3.03 eV and
6
= 2.33
B
. This result is in good agreement
with those reported previous DFT calculations by S. Datta et al. [198]. The pentagonal
byramid (PBP) lies 1.7 eV higher in energy. However, photoelectron spectroscopic studies
[201] predicted a PBP with bond length 2.750.1

A to be the most probable structure for
the Co

6
anion cluster. On the other extreme, a perfect octahedron (O
h
) with E
B
= 1.95
eV,
6
= 0.33
B
and d = 2.65

A is found optimal for Pd
6
. The length of the bonds in
the middle plane is d = 2.60

A, while we nd d = 2.69

A for the bonds connecting the

middle plane and the apex positions. A non-magnetic octahedron with the length of the
bond forming the middle plane d = 2.65

A and the length of the bonds involving the cap

position being d = 2.67

A is nearly degenerate and is found to be only 3 meV less stable. It

80
Chapter 5. Structure and magnetism of small CoPd clusters
Co
5
Co
4
Pd Co
4
Pd* Co
3
Pd
2
Co
3
Pd
2
* Co
2
Pd
3
Co
2
Pd
3
* CoPd
4
CoPd
4
* Pd
5
Co
6
Co
5
Pd Co
5
Pd* Co
4
Pd
2
Co
4
Pd
2
*
Co
3
Pd
3
Co
3
Pd
3
* Co
2
Pd
4
Co
2
Pd
4
* CoPd
5
CoPd
5
* Pd
6
Figure 5.6: Lowest energy isomers of CoPd pentamers and hexamers. The asteriks indicate
rst exited isomers.
is interesting to observe that the cluster structures having a similar shape display dierent
total magnetic moments. This is probably an artifact of broken spin-rotational symmetry.
The optimal CTBP structure is 46 meV less stable than the ground-state. These results
coincide with those reported previous DFT calculations by T. Futschek et al. [183] and
Kumar et al. [207]. The optimal Co
5
Pd is a CTBP structure. The Co atoms form a
TBP structure by pushing Pd atom to occupy the outer position. The average magnetic
moment
6
is reduced by 0.50
B
compared to the pure Co
6
. The rst exited isomer is
a capped square pyramid (CSP), with Co atoms forming the square pyramid structure
and Pd atom to occupy the cap position. The optimal Co
4
Pd
2
is a CTBP with Co atoms
sub-clustered to form a tetrahedron, and Pd atoms are capped without forming a dimer.
81
5.3. Results and discussion
Table 5.5: Structural, electronic and magnetic properties of CoPd hexamers as in Ta-
ble 5.2.
Cluster E
B
E
m
d

N

Co

Pd
Co
6
3.03 0.77 2.27 2.33 2.08
Co
5
Pd 2.73 0.55 2.30 1.83 1.99 0.31
2.46
Co
5
Pd* 2.73 0.57 2.27 1.83 2.00 0.25
2.49
Co
4
Pd
2
2.63 0.47 2.30 1.67 2.12 0.31
2.47
Co
4
Pd
2
* 2.56 0.46 2.27 1.67 2.07 0.45
2.45
2.62
Co
3
Pd
3
2.44 0.38 2.24 1.50 2.24 0.43
2.48
2.70
Co
3
Pd
3
* 2.42 0.36 2.24 1.16 2.05 0.12
2.50
2.60
Co
2
Pd
4
2.31 0.27 2.22 1.00 2.18 0.32
2.49
2.68
Co
2
Pd
4
* 2.26 0.22 2.08 1.00 2.23 0.196
2.50
2.68
CoPd
5
2.15 0.15 2.43 0.50 2.13 0.13
2.70
CoPd
5
* 2.13 0.14 2.46 0.50 2.24 0.19
2.71
2.71
Pd
6
1.95 0.10 2.65 0.32 0.32
A CSP structure with a PdPd bond is the rst exited isomer which is 67 meV less stable
in energy. For the equiatomic composition (i.e., Co
3
Pd
3
) a CTBP structure is found to
be optimal. Here Co atoms are grouped together to form a triangle (see Fig. 5.6). The
segregation of similar atoms in a cluster is important since orbitals with same symmetry
can hybridize, so that cluster can reduce its total energy. The rst excited isomer is a
double tetrahedron with Co atoms making a triangle. This is 15 meV/atom less stable. In
the Pd rich Co
2
Pd
4
cluster, a CTBP structure with a relatively shorter CoCo bond d
CoCo
= 2.22

A is the optimal. The rst excited isomer is an octahedron, which is 47 meV less
stable. Finally, the optimal CoPd
5
is an octahedron and the rst exited isomer is a CTBP
structure that is 20 meV less stable.
82
Chapter 5. Structure and magnetism of small CoPd clusters
-50
0
50
100
-4 -2 0 2 4
-50
0
50
100
-4 -2 0 2 4
D
e
n
s
i
t
y

o
f

s
t
a
t
e
s

[
e
V
-
1
]
Co
6
Co
4
Pd
2
Co
2
Pd
4
Pd
6

F
[eV]
F
[eV]
Total
Co
Pd
(a)
(b)
(c)
(d)
Figure 5.7: Electronic density of states (DOS) of hexamers. Results are given for the total
(solid black) DOS, Co-projected (narrow red) and Pd-projected (thick blue) d-electron
DOS. Positive (negative) values correspond to majority (minority) spin.
The binding energy per atom shows a similar trend as in the smaller clusters. In the
case of average magnetic moment per atom
N
an interesting behavior has been observed.
The average magnetic moment
6
exhibits a similar eect found in the case of pentamers,
which depends on the Co content level:
6
shows a reduction of 0.5
B
as one goes from
Co
5
Pd to Co
2
Pd
4
over CoPd
5
, while this reduction is only around 0.16
B
as one moves
from Co
4
Pd
2
to Pd
6
over Co
3
Pd
3
.
In the previous sections, the total DOS has been plotted to show the variation of the
HOMO-LUMO gap with concentration and size. It is also very interesting to analyze the
site projected local density of states (LDOS), at least for some representative examples.
To this aim, we plot in Fig. 5.7 the spin polarized d-electron LDOS of representative CoPd
hexameters having the relaxed structures illustrated in Fig 5.6. The DOS for pure Co
6
and
Pd
6
is also shown for the sake of comparison. In all the clusters, the dominant peaks in the
relevant energy range (i.e., the occupied valence orbitals and the unoccupied ones near
F
)
correspond either to the Co-3d or to the Pd-4d states. It is found that the contributions
from s and p states in the DOS is negligible when compared to the d band contribution.
The Co
6
cluster shows a relatively narrow d-band which dominates the single-particle
energy spectrum in the range 4.3 eV
F
1.15 eV. The majority d-DOS is
fully occupied with highest majority state lying about 1.62 eV below
F
. In addition
83
5.3. Results and discussion
Co
6
= 0.020
Co
5
Pd
= 0.010
Co
4
Pd
2
= 0.007
Co
3
Pd
3
= 0.006
Co
2
Pd
4
= 0.001
CoPd
5
= 0.003
Figure 5.8: Constant magnetization density plots

(r)

(r) = and local moments

(in
B
) for the ground-state structures of CoPd hexamers. The value of the constant
magnetization density is given in
B
/

A
3
.
there is an appreciable gap (about 0.26 eV) in the corresponding minority spectrum. The
spin polarized DOS clearly reects the FM-like order in the Co
6
cluster. For low Pd
concentration (e.g., Co
4
Pd
2
), the magnetic moments are not saturated. Only spin down
(minority) states are found around
F
. The Fermi level
F
lies on the top of the minority
band (see Fig. 5.7). In the majority band Pd dominates over Co at the higher energies
(near to
F
) while Co dominates in the bottom of the band. In the minority band the
participation of Pd (Co) is stronger (weaker) below
F
and weaker (stronger) above
F
,
which is consistent with the fact that the Pd local moments are smaller than the Co
moments.
Finally for the Pd rich limit (e.g., Co
2
Pd
4
), the majority states are almost saturated
and only minority states are found close to
F
. The Co contribution is signicant just
below and above
F
. The Co contribution largely dominates the unoccupied minority-
spin DOS, in agreement with the larger local Co moments. The DOS of Co
6
is similar to
previous DFT calculation by Kumar et al. [207].
84
Chapter 5. Structure and magnetism of small CoPd clusters
Co
7
Co
6
Pd Co
6
Pd* Co
5
Pd
2
Co
5
Pd
2
*
Co
4
Pd
3
Co
4
Pd
3
* Co
3
Pd
4
Co
3
Pd
4
* Co
2
Pd
5
Co
2
Pd
5
* CoPd
6
CoPd
6
* Pd
7
Co
8
Co
7
Pd Co
7
Pd* Co
6
Pd
2
Co
6
Pd
2
* Co
5
Pd
3
Co
5
Pd
3
* Co
4
Pd
4
Co
4
Pd
4
* Co
3
Pd
5
Co
3
Pd
5
*
Co
2
Pd
6
Co
2
Pd
6
* CoPd
7
CoPd
7
* Pd
8
Figure 5.9: Lowest energy isomers of CoPd heptamers and octamers. Note that only very
few topologies have been considered as starting congurations of geometry relaxation.
85
5.4. Size and composition trends
5.3.6 Heptamers and Octamers
For Co
m
Pd
n
clusters having m+n = N 7 we have employed the same initial structures
considered in the previous chapter.
The results for N = 7 are summarized in Table 5.6. As in smaller Co
m
Pd
n
clusters
the binding energy per atom increases approximately linearly with increasing Co content.
For pure Pd
7
the PBP is the most stable of the considered starting geometries. The rst
isomer is found to be a capped octahedron (CO). Our result are in good agreement those
of T. Futschek et al. [183], and M. Moseler et al. [10].
Heptamers with high Pd concentrations also favor a PBP topology except CoPd
6
. In
the CoPd
6
ground-state the Co atom is located on the square ring of the octahedron,
while in the Co
2
Pd
5
ground-state the Co atoms are along the axis of the PBP. The results
for Co
2
Pd
5
(e.g., structure, local moments and chemical order) coincide with the previous
DFT and tight binding calculations [218]. As the Co content increases, the topology of
Co
m
Pd
n
ground-state geometry changes. In fact, the ground-state structures of Co
4
Pd
3
and Co
5
Pd
2
corresponds to a CO (see Fig 5.9). However, the optimal structure for Co
5
Pd
2
obtained in Ref. [218] is a PBP with the Pd atoms occupying the two apex sites. For the
pure Co
7
the PBP is the most stable of the considered starting geometries. This result
is similar with the one reported DFT work by S. Datta [198]. However, the experimental
magnetic moment (2.360.25) [65] is higher than the present value.
Table 5.7 and Fig 5.9 summarizes the results for CoPd octamers. The general trends
concerning the composition dependence of the binding energy, chemical order, as well
as the average and local magnetic moments are similar to smaller clusters. The most
stable structure that we obtain for Co
8
is a BCO having
8
= 2
B
, and d = 2.31

A
(see Fig 5.9). The experimentally measured magnetic moment, 2.510.15
B
/atom [65]
is higher than the present value. In all the mixed octamers, there is a clear tendency for
Co atoms to group together. The most stable structure that we obtain for Pd
8
is a BCO
having E
B
= 2.07 eV,
8
= 0.25
B
and d = 2.68

A. This result is similar with earlier
reported work using DFT [183]. The magnetic properties of heptamers and octamers
follow the qualitative behavior established in smaller clusters. In all the cases the average
magnetic moment per atom
N
and the magnetic stabilization energy E
m
increase with
Co concentration.
To conclude this section it is interesting to compare the present results with available
experiments and theoretical results for macroscopic alloys [217]. The dilute solutions of
Co in Pd have been investigated by R. M. Bozorth et al. [217] shown that Co and Pd local
moments are coupled ferromagnetically. They predicted a magnetic moment of 1.7
B
for
the Co atom and a moment 0.6
B
for the NN Pd atoms whereas more distant Pd atoms
are found nearly unpolarized. Our results for small CoPd clusters also predict a FM-like
order at least in the ground-state clusters. Indeed, the local moments of Co and Pd are
signicantly enhanced upon alloying. This is due to the combined eect of reduction of
local coordination number and the charge transfer from Co to Pd atom.
5.4 Size and composition trends
The main purpose of this section is to focus on the dependence of the electronic and
magnetic properties of Co
m
Pd
n
clusters as a function of size and composition. In the
86
Chapter 5. Structure and magnetism of small CoPd clusters
Table 5.6: Structural, electronic and magnetic properties of CoPd heptamers as obtained
from a restricted representative sampling of cluster topologies.
Cluster E
B
E
m
d

N

Co

Pd
Co
7
3.04 0.68 2.28 2.14 1.96
Co
6
Pd 3.05 0.60 2.29 2.00 2.08 0.28
2.48
Co
6
Pd* 2.94 0.61 2.30 2.00 2.06 0.362
2.54
Co
5
Pd
2
2.88 0.51 2.28 1.86 2.15 0.43
2.48
Co
5
Pd
2
* 2.83 0.50 2.28 1.86 2.16 0.42
2.49
2.68
Co
4
Pd
3
2.71 0.45 2.26 1.43 2.08 0.37
2.49
2.73
Co
4
Pd
3
* 2.70 0.43 2.39 1.71 2.25 0.50
2.46
Co
3
Pd
4
2.50 0.30 2.30 1.29 2.16 0.43
2.50
2.72
Co
3
Pd
4
* 2.48 0.31 2.34 1.29 2.21 0.40
2.48
2.70
Co
2
Pd
5
2.39 0.26 2.22 1.14 2.24 0.50
2.52
2.66
Co
2
Pd
5
* 2.37 0.23 2.26 0.86 2.18 0.27
2.46
2.70
CoPd
6
2.21 0.14 2.42 0.71 2.31 0.36
2.70
CoPd
6
* 2.19 0.13 2.48 0.43 2.25 0.15
2.73
Pd
7
1.99 0.04 2.69 0.29 0.28
87
5.4. Size and composition trends
Table 5.7: Structural, electronic and magnetic properties of CoPd octamers as obtained
from a restricted representative sampling of cluster topologies.
Cluster E
B
E
m
d

N

Co

Pd
Co
8
3.17 0.70 2.31 2.00 1.88
Co
7
Pd 3.11 0.57 2.30 1.88 1.95 0.27
2.51
Co
7
Pd* 3.08 0.56 2.32 1.75 1.91 0.30
2.48
Co
6
Pd
2
3.05 0.53 2.31 1.75 2.06 0.29
2.49
Co
6
Pd
2
* 3.00 0.53 2.30 1.75 2.03 0.32
2.53
Co
5
Pd
3
2.93 0.49 2.27 1.63 2.13 0.42
2.48
Co
5
Pd
3
* 2.91 0.47 2.33 1.63 2.13 0.41
2.49
Co
4
Pd
4
2.76 0.41 2.36 1.50 2.23 0.49
2.46
Co
4
Pd
4
* 2.75 0.41 2.31 1.50 2.23 0.47
2.47
2.73
Co
3
Pd
5
2.62 0.28 2.29 1.13 2.19 0.36
2.48
2.73
Co
3
Pd
5
* 2.59 0.29 2.28 1.13 2.19 0.26
2.47
2.75
Co
2
Pd
6
2.47 0.21 2.34 1.00 2.26 0.43
2.45
2.71
Co
2
Pd
6
* 2.46 0.20 2.34 1.00 2.26 0.43
2.44
2.72
CoPd
7
2.27 0.12 2.42 0.63 2.31 0.34
2.69
CoPd
7
* 2.25 0.12 2.47 0.63 2.33 0.32
2.74
Pd
8
2.07 0.59 2.68 0.25 0.25
88
Chapter 5. Structure and magnetism of small CoPd clusters
following we present and discuss results for the binding energy, average and local spin
moments, relative stability and the magnetic stabilization energy.
5.4.1 Binding energy and magnetic moments
In order to summarize the important trends, we present in Fig. 5.10 the binding energy
per atom E
B
as a function of the number of Co atoms m. For most of the sizes studied,
0 1 2 3 4 5 6 7 8
m
0.5
1
1.5
2
2.5
3
B
i
n
d
i
n
g

E
n
e
r
g
y

[
e
V
/
a
t
o
m
]
Co
m
Pd
n
N = 2
3
4
5
6
7
8
Figure 5.10: Binding energy per atom E
B
of Co
m
Pd
n
clusters as a function of the number
of Co atoms. The lines connecting the points for dierent total number of atoms N = m+n
are a guide to the eye.
the slope of E
B
as a function of m is maximal for single Co doping (m = 1) and decreasing
for m > 1. There is an almost linear increase in the E
B
with increasing Co content (see
Fig 5.10). Notice the small dip for Co
2
Pd
2
in Fig. 5.10. This can be explained as follows:
Co
2
Pd
2
is a 2D rhombus in which two Co atoms are well separated by a Pd dimer. The
presence of weaker Pd dimer is the key reason for the reduced binding energy for Co
2
Pd
2
.
In Fig. 5.12 the average magnetic moments
N
of Co
m
Pd
n
is shown as a function of
m. As already discussed in the previous sections,
N
increases almost linearly with the
Co content. This is expected due to the larger Co local moments and the underlying FM-
like order. Among the pure clusters, Co
5
and Pd
2
show the highest magnetic moments.
Notice, the enhancement of the magnetic moments of the pure clusters in particular for
Co
N
, which go well beyond 2.5
B
(e.g., in Co
4
and Co
5
). In general, the slope of the curves
tend to decrease with increasing N, since the change in concentration per Co substitution
is more important the smaller the size is.
The magnetic stabilization energy E
m
is shown in Fig. 5.11. One observes that E
m
increases approximately linearly with the Co concentration except for CoPd
7
. The pure
Co clusters show the highest E
m
. This is understandable since Co is a FM-3d element.
In contrast, Pd clusters show low E
m
, since Pd is a non-magnetic element.
89
5.4. Size and composition trends
0 2 4 6 8
m
0
0.2
0.4
0.6
0.8

E
m

[
e
V
]
Co
m
Pd
n
N = 2
3
4
5
6
7
8
Figure 5.11: The magnetic stabilization energy E
m
= E(S
z
=0) E(S
z
) of Co Pd
m
Pd
n
clusters as a function of number of Co atoms. The symbols corresponding to each size
N = m + n are the same as in Fig. 5.10. The lines connecting the points for each N are
a guide to the eye.
0 1 2 3 4 5 6 7 8
m
0.5
1.0
1.5
2.0
2.5

]
Co
m
Pd
n
N = 2
3
4
5
6
7
8
Figure 5.12: Total magnetic moment per atom
N
of Co
m
Pd
n
clusters as a function of
number of Co atoms. The lines connecting the points for each N = m+n are a guide to
the eye.
90
Chapter 5. Structure and magnetism of small CoPd clusters
5.4.2 Relative stability
1 2 3 4
m
-0.05
0.00

2
E
(
5
)

[
e
V
]
1 2 3 4 5
m
-0.10
-0.05
0.00
0.05

2
E
(
6
)

[
e
V
]
1 2 3 4 5 6
m
1 2 3 4 5 6 7
m
-0.05
0.00

2
E
(
8
)

[
e
V
]
0.20
0.00
-0.18
0.10

2
E
(
7
)

[
e
V
]
0.03
0.15
0.10
Figure 5.13: The relative stability
2
E(N, m) = E(N, m+1)+E(N, m1)2E(N, m) of
Co
m
Pd
n
clusters as a function of the number of Co atoms. The symbols corresponding to
each size are the same as in Fig. 5.10. The lines connecting the points for each N = m+n
are a guide to the eye.
1 2 3 4
m
2
2.2
2.4
2.6
2.8
d

(

)
1 2 3 4 5
m
2
2.2
2.4
2.6
2.8
d

(

)
1 2 3 4 5 6 7
m
2
2.2
2.4
2.6
2.8
d

(

)
1 2 3 4 5 6 7
m
2
2.2
2.4
2.6
2.8
d

(

)
N = 5
N = 6
N = 7
N = 8
Figure 5.14: Bond lengths for CoCo (cross), Co Pd (circle) and PdPd (square) pairs as
a function of number of Co atoms in Co
m
Pd
n
clusters having N = 58 atoms. The lines
connecting the points for each N = m+n are a guide to the eye.
91
5.5. Summary
Like FeRh clusters, the CoPd clusters also show signicant variation in the relative
stability among the dierent cluster sizes N having dierent compositions. For instance,
for N = 6, Co
5
Pd shows the highest stability while Co
4
Pd
2
has lowest stability. This can
be justied by analyzing the number of dierent bonds and average bond length, as was
done in the Sec. 4.4.2 in the case of FeRh clusters. In Co
5
Pd, there are 9 CoCo bonds
(average bond length 2.30

A) and 3 CoPd (average bond length 2.46

A). While in Co
4
Pd
2
,
there are 5 CoCo bonds (average bond length 2.30

A) and 6 CoPd bonds (average bond
length 2.47

A) present. Let us recall that the CoCo bond is the strongest, as it was shown
in the case of the dimers (see Table. 5.1).
5.5 Summary
The structural, electronic and magnetic properties of small Co
m
Pd
n
clusters having N =
m+n 8 atoms have been investigated systematically in the framework of a generalized
gradient approximation to density-functional theory. Geometry optimization yields both
compact 3D and 2D open topologies. The magnetic order of the clusters is found to be
FM-ike, at least in the most stable structures. The average magnetic moment per atom

N
increases approximately linearly with the increasing Co content. Accordingly, the
energy gain E
m
associated to magnetism also increases with the number of Co atoms.
The maximal spin polarization for Co and Pd atom is found in the equiatomic or nearly
equiatomic compositions (Co
3
Pd
3
, Co
3
Pd
4
and Co
4
Pd
4
). The s and p spin polarizations
are almost negligible in general. A remarkable enhancement of the local Co moments
is observed as result of Pd doping. This is due to the enhancement in the number of
Co d holes, due to Co to Pd charge transfer, combined with the extremely reduced local
coordination.
The spin-orbit coupling (SOC) is not included in the present calculations. Several
authors have discussed the inuence of SOC for a variety of 3d, 4d, and 5d transition-metal
clusters [219221]. These investigations have predicted that for clusters of 3d and 4d TMs
the SOC was found much smaller whereas for 5d TMs like Pt SOC can be strong enough
so that it can change the energetic ordering of structural isomers. We have performed
some representative calculations by taking into account spin-orbit coupling (SOC) in order
to explore their eect on the ground-state structure, chemical order and spin moments.
For example in Co
6
, Co
3
Pd
3
and Pd
6
we nd that the changes in the ground-state energy
resulting from SO interaction are typically of the order of 0.2 eV for the whole cluster. This
is often comparable to or larger than the energy dierences between the low-lying isomers.
However, the SO energies are very similar for dierent structures, so that the ground-state
structures remain essentially the same as in the scalar relativistic (SR) calculations. The
changes in the bond lengths and in the spin moments resulting from SOC are also very
small (e.g., |
SOC

SR
| 0.04
B
and |d
SOC
ij
d
SR
ij
| 0.001

A in Co
4
Pd
4
). As a result,
the conclusions drawn from our SR calculations on the relative stability and local spin
moments seem to be unaected by the spin-orbit contributions.
92
6
First principles spin-polarized basin-hopping
method
6.1 Introduction
The reported results in the chapters 5 and 6 have been obtained by employing those struc-
tures which are derived from graph theory method [17]. However, it should be emphasized
that performing calculations by considering all the possible initial cluster structures, all
the dierent magnetic congurations and all the possible dierent distribution of two kinds
of atoms is nearly impossible (or impracticable) beyond the hexamer clusters (N = 6) due
to the prohibitively huge computational time. This was the main reason for considering
only a small set of initial cluster topologies for cluster sizes N = 7 and 8 in chapters 5 and
6. On the other hand, we would like to elucidate whether the ground-state structures we
identied in the last chapters represent the true ground-states, at least in the case of FeRh
clusters. The overall situation signals the need for an ecient global optimization method,
and by employing this we could x the inherent problems associated with the graph theory
method. For this purpose, we chose and employed the basin-hopping (BH) global opti-
mization method. Then we developed a scheme which combines the BH method with the
spin-polarized DFT code VASP. As was discussed in Sec. 3.2.3, the BH method has already
proven its applicability and suitability to optimize smaller and larger clusters [19, 20, 222].
In this work we applied the proposed scheme to study both small (Fe
6
, Fe
3
Rh
3
and Rh
6
)
and larger (Fe
13
, Fe
6
Rh
7
and Rh
13
) clusters. Having selected some clusters with dierent
sizes and compositions, the next step should be the choice of suitable technical parameters
in order to optimize the performance of the BH calculations. In the past, an investigation
have been reported by combining DFT with the BH scheme [222]. However, that study
was based on the Si and Cu clusters, which are obviously non-magnetic. To the best of our
knowledge, this is the rst time one proposes to combine the spin-polarized DFT with a
BH scheme. Although one easily understand the potential advantages of proposed scheme,
it poses a number of serious computational challenges. We begin our computations on ho-
93
6.1. Introduction
Figure 6.1: Schematic diagram of potential energy surface (PES) of a nanoalloy cluster.
mogeneous clusters and then follow with the heterogeneous ones. In order to implement
the proposed scheme, one should consider dierent initial cluster structures and various
starting magnetic congurations. The dierent magnetic congurations such as FM and
AF-like are introduced by assigning random initial magnetic moment to the BH sampling
runs. Obviously, the proposed scheme would take more computational time in the case
of mixed clusters. This is because one has to consider dierent homotops for a specic
composition. This has been done by exchanging or swapping the dissimilar atoms within a
cluster on-the-y. Adding the complexities mentioned above makes the computation very
expensive. This indeed shows the need for optimizing the eciency of the BH runs.
A widely used approach to measure the sampling eciency of a global optimization
calculation is the mean rst encounter (MFE), which is dened as the number of BH steps
until the algorithm locates the ground-state structure for the rst time. The MFE is found
to be an ideal sampling eciency indicator. In order to obtain useful or meaningful re-
sults, the sampling should be performed by monitoring on-the-y analyzable performance
indicators that allow to modify an ongoing run. However, there is no previous information
which illustrates how to build move parameters that do not depend upon system-specic
intuition in the case of spin-polarized clusters. The inherent stochastic nature of the BH
scheme can be another source of diculties in the proposed scheme. Any investigation
which judges the credibility of BH technical parameters or on-the-y analyzable perfor-
mance indicators must involve an averaging over a fairly large number of BH sampling
runs.
Studies have been reported in the past, which combine the BH scheme with numerically
undemanding model potentials such as Sutton-Chen [223] and Lennard-Jones [19,20]. But
these model potentials are unsuitable to sample the quantum-mechanical PESs of magnetic
TM nanoclusters. In fact, the basic motivation behind this work is to illustrate an ecient
framework for a systematic performance analysis of rst-principles spin-polarized BH sam-
pling runs and investigate the unique magnetic properties exhibited by TM nanoclusters.
This method should considerably reduce the computational cost of the spin-polarized BH
runs required for the averaging process and help to obtain statistically promising perfor-
mance results.
94
Chapter 6. First principles spin-polarized basin-hopping method
6.2 Computational aspects
The calculations reported in this chapter have been performed in the framework of Hohenberg-
Kohn-Shams density functional theory, [11] as implemented in the Vienna ab initio simu-
lation package (VASP) [168]. The detailed information concerning this method has been
explained in the Sec. 4.2.
6.2.1 Basin-Hopping method
In this section, all the relevant parameters that govern the performance of the BH sampling
run are explained. The detailed information regarding this method has been discussed in
the Sec. 3.2.3. Three key points in a BH sampling method deserve special attention: i)
the trial move method to generate new cluster structures, ii) the acceptance criteria, and
iii) the entropy driven dissociations. These are discussed in the following.
The trial move method
The BH trail moves class can be classied into single and collective particle moves. This
is schematically shown in the Fig. 6.2. The studies carried out by R. Gehrke [222] made
a systematic comparison on the performance between these move classes and reported
that they does not produce any signicant dierence in the performance. Therefore, we
employed collective particle moves with uniform distribution of atoms to perform all the
calculations reported in this chapter.
Figure 6.2: Single and collective particle move classes.
In the collective random move all the atoms in the cluster are randomly displaced
within a predened range of distance. In the coordinate space, the position of an atom i
within a cluster is represented by

R
i
= (x
i
, y
i
, z
i
) Randomly displacing an atom means
that changing the current (x
i
, y
i
, z
i
) position into a new position and the corresponding
displacement vector of an atom i can be written as
R
i
= d {x x, y y, z z}.

6.2.1
The move distance d controls the extend of the jumps in the congurational space and
the resulting algorithmic performance. The real advantage of the proposed scheme is that
it can be implemented to study any atomic cluster (for instance, metallic, semiconductor,
etc.) without a prior knowledge about the properties owned by the specic clusters. The
95
6.2. Computational aspects
move distances are calculated from the shortest dimer bond distance a within a specic
cluster. Atoms in the cluster are uniformly displaced using d = 0.10a, 0.15a, 0.30a, etc.
The optimal move distance can then be identied from this procedure.
The acceptance criterion
An essential parameter that must be included in the BH sampling is the acceptance crite-
rion, according to which a newly generated trial structure is accepted and substitutes the
current structure as starting point for the next trail move. Naively, one might think that
in order to minimize the energy of the system, one should accept a new structure when the
energy is lower. However, considering also less stable structures along the Markov chain
in order to able to overcome energy barriers. One criterion can be that new structures
are unconditionally accepted. However, in its classical form, the BH scheme proposed to
accept less stable structures according to a Metropolis rule, exp(E/k
B
T), where k
B
is
the Boltzmann constant and T is an eective temperature. The Metropolis rule facilitates
the BH algorithm to eectively climb the high-energy barrier regions on multiple-funnel
type transformed PESs and thus enables a cluster structure to hop from one local minima
to other more rare but potentially more stable local minima (see Fig. 6.1). However, in
practice our results demonstrate that the acceptance criterion has only a secondary im-
portance. This is because the employed update strategy itself is adequate to sample all
the relevant regions of the underlying PES. This is veried by performing BH sampling
runs with a window acceptance criterion.
Entropy driven dissociation
It is also very important to circumvent the problems of entropy-driven dissociation of the
clusters along the Markov chain while sampling the congurational space systematically
and without any prior considerations is critical. This problem has been xed by discarding
the partly dissociated structures featured by an atom having a nearest-neighbor distance
larger than twice the dimer bond length. Likewise we discarded trail moves that place
atoms at distances of less than 1.5

A from each other.

6.2.2 Measuring the sampling eciency
As already mentioned, the MFE can be used to determine the performance analysis of BH
sampling. The basic aim of the present work is to identify all the relevant isomers for a
specic cluster. Therefore, we moved one step forward by introducing another parameter,
namely, the number of moves N
av
until all the relevant isomers have been found at least
once. Of course, one has to specify what a relevant isomer is. N
av
is found to be an
ideal parameter for the performance analysis of the employed BH moves. However, it
should be stressed that, due to the marginally varying number of geometry steps for
the local relaxation of each trial structure, N
av
is only roughly proportional to the total
computational time taken by the BH run. Due to the stochastic nature of the BH method,
N
av
is only a statistically meaningful quantity after averaging over suciently large number
of runs. Even for the small cluster sizes considered here, this implies having to perform
hundreds of dierent rst-principles spin-polarized BH calculations in order to obtain a
N
av
. This has been done for each BH settings (e.g. move distance) we wish to analyze. The
96
Chapter 6. First principles spin-polarized basin-hopping method
Figure 6.3: Pictorial illustration of successful, unsuccessful and high-energy trial moves in
the BH scheme. The horizontal red line indicates the energy-window acceptance criterion
for BH sampling. After Ref. [222].
outcome of trail moves can be divided into three categories: i) unsuccessful trail moves,
ii) high-energy trail moves, and iii) successful trail moves. These are shown schematically
in Fig. 6.3.
Unsuccessful trail moves
A trial move is considered to be unsuccessful (see Eq. 6.2.2) if the initial cluster, where
the trial move has been started, and the optimized cluster where the optimization ends
up are the same. In this case no new information concerning the isomers is obtained. The
fraction of unsuccessful moves
uns.
up to the n
th
trial move can is given by

uns
=

uns
n
,

6.2.2
where
uns
is the number of unsuccessful moves up to that point.
High-energy trail moves
A Markov step is rejected (see Eq. 6.2.3) if the move generates a new relaxed cluster
having a total energy dierence E that is above the BH energy window imposed by the
acceptance criterion. The fraction of moves rejected in this way is given by

hE
=

hE
n
,

6.2.3
where
hE
is the corresponding number of rejected moves up to that point.
Successful trail moves
A Markov step is considered to be successful (see Eq. 6.2.4) once the move and subsequent
relaxation brings the cluster into local minimum within the energy window that is dierent
from the starting structure. The fraction of successful moves is given by

suc
=
n
s
n
= 1
uns

hE

6.2.4
97
6.3. Results for small homogeneous clusters
6.3 Results for small homogeneous clusters
This section is devoted to discuss the performance and the results of the proposed scheme
in the case of homogeneous small clusters. The main emphasis is here on identifying
the ground-state structures and other low-lying isomers, and discuss their energetic and
magnetic properties.
6.3.1 Dominant isomers
As pointed out in the introduction of this chapter, we have varied the move distance by
starting from d = 0.10a up to a maximum value d = 1.0a, in terms of the smallest bond
length a within the specic cluster. In order to make the discussion compact and more
informative, we provide here only the results for three dierent move distances d = 0.15a,
0.26a and 0.40a. It would be however noted that many other move distances (for instance,
0.10a, 0.20a, 0.60a, etc) have been considered.
{0, 0.00, 20} {1, 0.02, 20} {2, 0.18, 20} {3, 0.20, 20} {4, 1.18, 6}
{5, 1.20, 12} {6, 1.22, 12} {7, 1.24, 14} {8, 1.25, 12} {9, 1.34, 8}
{10, 1.38, 10} {11, 1.41, 14} {12, 1.74, 12}
Figure 6.4: Identied isomers of Fe
6
cluster in the energy range up to 1.74 eV above the
ground-state. The isomers are ordered by increasing energy. In the subcaption the rst
entry refers to the isomer number, the second entry to the energy relative to the ground-
state energy in eV, and third entry to the the cluster magnetic moment in
B
. Bond
lengths are indicated in

A.
98
Chapter 6. First principles spin-polarized basin-hopping method
This choice is justied since the BH algorithm run with very small move distance (say
d = 0.10a) ends up most of the time in the same structure as the starting one. In contrast
BH runs using a too large move distance (say d = 0.60a) mostly produce loosely connected
open cluster structures having a much higher energy than the ground-state structure (see
Fig. 6.6). We present in Fig. 6.7 the binding energy of isomers starting from the ground-
state structure, the probability of nding a given isomer is given by the number of times
that the low-energy isomers were located, and the cluster magnetic moment along the BH
runs obtained for Fe
6
and Rh
6
. It should be emphasized that each spin-polarized BH run
for a particular move distance is composed of several hundred steps and was performed
until the probability plots for all the relevant isomers were converged. We have explicitly
checked convergence of the probability by taking Fe
3
Rh
3
cluster as a representative test
system. The progress towards convergence is schematically shown in Fig. 6.5. The plot
shows the probability over consecutive sampling periods containing 50 calculations each.
One may observes that some isomers show up as dominant. The main reason for the
existence of dominant isomers is that their corresponding basins of attraction on the PES
is large and thus hit by the trial moves many times. In other words, the existence of
dominant isomers is because of the reduced system size and the resulting small number of
low-lying isomers.
50 100 150 200
BH step n
0
0.1
0.2
0.3
0.4
0.5
0.6
P
r
o
b
a
b
i
l
i
t
y
# 1
# 2
# 3
# 4
> # 4
Figure 6.5: Normalized probabilities for the lowest-energy isomers of Fe
3
Rh
3
as shown in
the Fig. 6.10. The probabilities over consecutive sampling periods containing 50 moves
each are shown by using collective-particle moves and uniformly distributed move distances
around the average values d = 0.26a. The probabilities for all isomers higher in energy
than isomer # 4 are bundled into one entry labeled > # 4.
The following analysis provides more information concerning the dominant isomers.
Fig. 6.7 demonstrates the results obtained for Fe
6
and Rh
6
cluster using three dierent
move distances. First we discuss the results for Fe
6
with and without window acceptance
99
6.3. Results for small homogeneous clusters
0 10 20 30 40 50 60 70 80 90 100
-36.0
-35.5
-35.0
-34.5
-34.0
E
n
e
r
g
y

(
e
V
/
c
l
u
s
t
e
r
)
-36.4
-36.2
-36.0
-35.8
-35.6
-35.4
-35.2
d = 0.15a
Number of BH steps (n)
0 10 20 30 40 50 60 70 80 90 100
-36.0
-35.0
-34.0
-33.0
-32.0
-31.0
-30.0
E
n
e
r
g
y

(
e
V
/
c
l
u
s
t
e
r
)
-36.4
-36.2
-36.0
-35.8
-35.6
-35.4
-35.2
Number of BH steps (n)
d = 0.26a
0 10 20 30 40 50 60 70
-36.0
-35.0
-34.0
-33.0
-32.0
E
n
e
r
g
y

(
e
V
/
c
l
u
s
t
e
r
)
-36.4
-36.2
-36
-35.8
-35.6
-35.4
-35.2
Number of BH steps (n)
d = 0.40a
0 10 20 30 40 50 60 70 80
-36
-35
-34
-33
-32
-31
-30
E
n
e
r
g
y

(
e
V
/
c
l
u
s
t
e
r
)
-36.4
-36.2
-36.0
-35.8
-35.6
-35.4
d = 0.60a
Number of BH steps (n)
Figure 6.6: The total energy of Fe
6
as a function of the number of BH steps n for dierent
move distances d = 0.15a, 0.26a, 0.40a and 0.60a.
100
Chapter 6. First principles spin-polarized basin-hopping method
-2.8
-2.7
-2.6
-2.6
-2.5
E
B

[
e
V
/
a
t
o
m
]
d = 0.15a
d = 0.26a
d = 0.40a
0
0.2
0.4
0.6
0.8
P
r
o
b
a
b
i
l
i
t
y
0 2 4 6 8 10 12 14 16 18
Isomer number
8
12
16
20
T
o
t
a
l

m
o
m
e
n
t
Fe
6
NAC
(a)
-3.2
-3.1
-3
E
B
[
e
V
/
a
t
o
m
]
0
0.2
0.4
0.6
0.8
P
r
o
b
a
b
i
l
i
t
y
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Isomer number
0
2
4
6
8
10
T
o
t
a
l

m
o
m
e
n
t
Rh
6
NAC
(c)
-2.7
-2.6
E
B

[
e
V
/
a
t
o
m
]
0
0.2
0.4
0.6
0.8
P
r
o
b
a
b
i
l
i
t
y
0 2 4 6 8
Isomer number
12
16
20
T
o
t
a
l

m
o
m
e
n
t
Fe
6
WAC
(b)
-3.2
-3.1
-3
E
B

[
e
V
/
a
t
o
m
]
0
0.1
0.2
P
r
o
b
a
b
i
l
i
t
y
0 2 4 6 8 10 12 14 16 18 20 22 24 26
Isomer number
0
3
6
9
T
o
t
a
l

m
o
m
e
n
t
Rh
6
WAC
(d)
Figure 6.7: The binding energy per atom (in eV), the probability that trial moves generate
the lowest energy isomers, and the total magnetic moment are plotted as a function of
isomer number for Fe
6
and Rh
6
cluster. NAC refers to the calculations using No Accep-
tance Criterion and WAC refers to With Acceptance Criterion. The obtained isomers are
numbered with decreasing stability, with isomer 0 refers to the ground-state structure.
The gure contains all the isomers found for collective moves with uniform distribution
of move distances d = 0.15a, 0.26a and 0.40a, where a refers to the shortest NN distance
within the cluster. The geometric structures of obtained isomers for Fe
6
and Rh
6
are given
in Fig. 6.4 and in Fig. 6.8, respectively.
101
6.3. Results for small homogeneous clusters
{0,0.000,6} {1,0.020,0} {2,0.023,6} {3,0.054,6} {4,0.162,4} {5,0.179,2}
{6,0.217,10} {7,0.359,10} {8,0.363,6} {9,0.367,6} {10,0.386,8}
{11,0.434,10} {12,0.435,4} {13,0.469,10} {14,0.472,8} {15,0.508,4}
{16,0.545,6} {17,0.555,8} {18,0.567,0} {19,0.664,2}
{20,0.684,8} {21,0.698,0} {22,0.743,4} {23,0.771,0}
{24,0.799,6} {25,0.841,2} {26,0.903,8} {27,0.947,4} {28,1.045,2}
{29,1.119,2} {30,1.390,0}
Figure 6.8: Calculated isomers of Rh
6
cluster in the energy range up to 1.39 eV above the
ground-state energy. The entries below the illustration have the same meaning as in the
Fig. 6.4
102
Chapter 6. First principles spin-polarized basin-hopping method
criterion of 1eV. See plots (a) and (b) in the Fig. 6.7.
One observes that only a small set of isomers which are close to the ground-state
structure (labeled as isomer number 0) are much more often sampled than other isomers.
Notice that, more than 2/3 of all executed moves in the BH runs ended up in the ground-
state isomer, regardless of the actual move distance and the window acceptance criterion
employed. The plots (a) and (b) in the Fig. 6.7 display some dierences too. One can
see that for the same move distances the calculations without an energy window (with an
energy window) generate 9 (4) isomers. All the isomers are having compact geometries
(see Fig. 6.4). The results for Rh
6
[see plots (c) and (d) in the Fig. 6.7] also display the
existence of the dominant isomers as in the case of Fe
6
cluster.
One can see that, some of the Rh
6
isomers relax into the open structures higher up
in energy compared to the ground-state structure. However, the possibility for a move
to end up in an open structure is very small. They are not often identied in the BH
sampling run. Overall, these results show that in the case of small homogeneous clusters
it is sucient to carryout the calculations without using a window acceptance criterion.
As pointed out earlier, without a window acceptance criterion, the system can jump out
of the dominant isomers and is able to visit dierent rare local minima.
6.3.2 Magnetic properties
Fe
6
and particularly Rh
6
clusters display remarkable structural and the magnetic diversity.
In fact, the results for Rh
6
reveal a unique relation between the magnetism and the cluster
structure. It was been shown in chapter 4 that the reduced system dimensionality and
the subsequent d-band narrowing helps the Rh clusters to fulll the Stoner criteria of
ferromagnetism, despite being non-magnetic in the bulk. And even a small distortion
in the cluster structure could produce quite dierent magnetic ordering (see Fig. 6.8).
For instance, the shape of the ground-state structure and the rst isomer of Rh
6
cluster
are octahedra. However, their cluster magnetic moments 6
B
and 0
B
, respectively are
quite dierent. It is also interesting to see that isomers having the same cluster magnetic
moments adopt dierent geometries. See for instance, the 5
th
and 8
th
isomers in the
Fig. 6.4 and ground-state, 2
nd
, and 24
th
isomer in the Fig. 6.8. The rst example of AF-
like coupling of the local moments is found in the 4
th
isomer of Fe
6
clusters, where two
local moments couple antiferromagnetically with the rest of the local moments and the
total cluster moment is reduced to 6
B
(see Fig. 6.4). In the case of Rh
6
clusters, AF-like
coupling is observed in the rst isomer. The associated total cluster moment is zero.
It is interesting to compare the present results for Fe
6
and Rh
6
with those reported
in chapter 4. One can see that the true ground-state structure for Fe
6
cluster is the one
reported in the present chapter, namely a distorted octahedra. The perfect octahedra
reported in chapter 4 (see Fig 4.3) is only a low-lying isomer. The corresponding energy
dierence is 0.04 eV/atom. This shows the signicance of the BH method, since it can
eectively take into account even very small cluster distortions, and thereby identifying
the true ground-state structure. The conclusion drawn in chapter 4 concerning the local
magnetic moments are also aected by the distortions found in the present BH calculation.
103
6.4. Results for heterogeneous small clusters
6.4 Results for heterogeneous small clusters
In this section, results for mixed Fe
3
Rh
3
clusters are discussed. The initial cluster struc-
tures were the same as those considered in the pure clusters. Unlike pure clusters, the
mixed clusters oer many potential complexities as was discussed in chapters 5 and 6. One
of the important conclusions of the analysis of small pure clusters is that the window ac-
ceptance criterion has only a secondary importance. However, we have explicitly checked
the role of window acceptance criterion in the case of mixed clusters too. The results
and discussions are divided according to the specic computational schemes used. Four
cases are considered: i) no acceptance criterion (NAC) and no swapping (NS), ii) with
acceptance criterion (WAC) and no swapping (NS), iii) no acceptance criterion (NAC) and
with swapping (WS), iv) with acceptance criterion (WAC) and with swapping (WS). The
results of the BH runs are analyzed for dierent computation parameters by comparing
their relative performance.
6.4.1 Dominant isomers
Fig. 6.9 shows the existence of dominant isomers in the case of small mixed clusters as was
observed in the analysis of small pure clusters. Therefore, we conclude that the emergence
of dominant isomers is an intrinsic property of the BH scheme, at least for the small cluster
sizes considered here. Now we analyze the obtained results step by step according to the
dierent schemes.
First we compare the performance of the calculations with (Fig. 6.9 (a)) and without
(Fig. 6.9 (b)) window acceptance criterion. The major similarity between these BH runs
is the existence of dominant isomers (for instance, isomer number 3). There are two major
dierences. The rst one concern the number of isomers being sampled. The calculations
without window acceptance criterion are able to identify up to 52 isomers, while the
calculation with window acceptance criterion were able to visit only up to 37 isomers.
This shows that calculation with window acceptance criterion enhance the performance
of the BH sampling by precluding the system to hop into non-relevant regions of the
underlying PES. The second important dierence is that for the smallest move distance d
= 0.15a, the calculation without window acceptance criterion has not been able to identify
the ground-state structure even after 100 Markov steps.
We compare now the relative performance of calculations by introducing the swapping
of dissimilar atoms on the y (see Figs. 6.9 (a) and (c)). The result shows many signicant
dierences with the previous calculations. The rst one is that the calculations with
swapping of dissimilar atoms is able to locate the ground-state structure even for the
smallest move distance d = 0.15a. The second one is the emergence of many dominant
isomers close to the ground-state structure regardless of the move distance. Therefore,
the calculations with swapping of dissimilar atoms on the y increases the algorithmic
performance, since, it enables the system to visit many of the relevant homotops of Fe
3
Rh
3
and reduce the total computing time even for the smallest move distance of 0.15a (see
Fig. 6.9). Concerning the combined eect of introducing a window acceptance criterion
and the swapping of dissimilar atoms (see Fig. 6.9 (a) and (d)), we observe that this
scheme facilitates the overall performance, since the procedure can eciently explore the
underlying PES by eciently hoping among the dierent local minima.
104
Chapter 6. First principles spin-polarized basin-hopping method
-3.2
-3.1
-3
E
B

[
e
V
/
a
t
o
m
]
d = 0.15a
d = 0.26a
d = 0.40a
0
0.2
0.4
0.6
0.8
P
r
o
b
a
b
i
l
i
t
y
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52
Isomer number
0
5
10
15
20
T
o
t
a
l

m
o
m
e
n
t
Fe
3
Rh
3
NAC_NS
(a)
-3.2
-3.1
-3
E
B

[
e
V
/
a
t
o
m
]
0
0.2
0.4
0.6
0.8
P
r
o
b
a
b
i
l
i
t
y
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
Isomer number
0
5
10
15
20
T
o
t
a
l

m
o
m
e
n
t
Fe
3
Rh
3
WAC_NS
(b)
-3.2
-3.1
-3
E
B

[
e
V
/
a
t
o
m
]
0
0.1
0.2
0.3
P
r
o
b
a
b
i
l
i
t
y
0 2 4 6 8 10 1214 16 1820 22 24 26 2830 32 34 36 3840 42 44
Isomer number
0
5
10
15
20
T
o
t
a
l

m
o
m
e
n
t
Fe
3
Rh
3
NAC_WS
(c)
-3.2
-3.1
-3
E
B

[
e
V
/
a
t
o
m
]
0
0.1
0.2
0.3
0.4
P
r
o
b
a
b
i
l
i
t
y
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Isomer number
5
10
15
T
o
t
a
l

m
o
m
e
n
t
Fe
3
Rh
3
WAC_WS
(d)
Figure 6.9: Binding energy per atom (in eV), the probability with which trial moves
generate the lowest energy isomers, and the total magnetic moment as a function of isomer
number for Fe
3
Rh
3
. The isomers are numbered with decreasing stability, with the isomer
0 referring to the ground-state structure. The gure contains all the isomers found for
collective moves with uniform distribution of move distances d = 0.15a, 0.26a and 0.40a,
where a refers to the shortest bond distance within the cluster. The corresponding isomer
structures are illustrated in Fig. 6.10.
105
6.4. Results for heterogeneous small clusters
6.4.2 Magnetic properties
The magnetism of the mixed clusters display remarkable diversity since not only the cluster
structure but also the dierent distribution of dissimilar atoms within the cluster matters.
Fig. 6.10 shows the obtained isomers of Fe
3
Rh
3
cluster. The ground-state structure is
found to be an octahedra, which coincide with the one reported in chapter 4 by using
graph theory method in which two isosceles open Fe
3
and Rh
3
triangles form a /2 angle
{0,0.00,15} {1,0.06,17} {2,0.15,15} {3,0.21,15} {4,0.31,13}
{5,0.33,5} {6,0.35,13} {7,0.39,17} {8,0.41,13} {9,0.43,13}
{10,0.44,15} {11,0.46,15} {12,0.49,15} {13,0.50,15} {14,0.57,11}
{15,0.58,15} {16,0.59,15} {17,0.62,15} {18,0.63,5}
{19,0.64,5} {20,0.65,15} {21,0.67,15}
Figure 6.10: The optimal structures of Fe
3
Rh
3
cluster. The entries below each structures
have the same meaning as in the Fig. 6.4
106
Chapter 6. First principles spin-polarized basin-hopping method
with respect to each other (see Fig 6.10). The total cluster magnetic moment is 15
B
.
The isomers display the magnetic diversity with respect to the cluster shape and dier-
ent arrangements of Fe and Rh atoms. In some cases an even small cluster distortion or a
rather small contraction or elongation in the bond lengths produce entirely dierent clus-
ter magnetic moments. For instance, the ground-state, rst and 5
th
isomers in Fig. 6.10
have the same shape and having the same distribution of Fe and Rh atoms, although we
observe small dierences among them. Interestingly, their cluster magnetic moments are
found to be 15
B
, 17
B
, and 5
B
, respectively. In the case of rst isomer having =
17
B
it is clear that this should correspond to a dierent magnetic state as the ground
state, which has a smaller = 15
B
. However, in the case of the isomer having =
5
B
we cannot exclude the possibility that this is an artifact of broken spin-rotational
symmetry. Notice that the segregation of similar atoms starts to be observed already for
the 7
th
isomer, which is 0.39 eV less stable than the ground-state structure. The cluster
magnetic moment exhibits more diversity while they do segregation with a similar shapes.
See for instance, the isomers 7(17
B
), 9(13
B
), 11(15
B
), and 14(15
B
).
The bottom line is that an ecient BH algorithm should sample all the relevant parts
of the PES. One way to improve the eciency of BH sampling is to reduce the number
of times that the BH algorithm gets stuck in the same dominant isomers, so that the
probability of jumping into the rare minima increases. From the perspective of an unbiased
sampling ansatz, with purely stochastic moves, it seems inevitable that the algorithm will
often revisit the same dominant isomers, at least for the limited isomer number of the
small cluster sizes studied here. The overall performance is then obviously dictated by the
way the algorithm can deal with these dominant isomers, e.g. how eciently it can hop
out of them.
6.5 Results for larger clusters
This section presents results for somewhat larger clusters: Fe
13
, Rh
13
and Fe
6
Rh
7
. The
computational procedure is the same as the one used for N = 6 atoms. For a cluster
having 13 atoms, a complete systematic analysis as was done for the smaller clusters
would be computationally too expensive, since the number of dierent isomers increase
exponentially with the cluster size. Indeed there will be more isomers above the energy
window of interest, the probability to jump in high-energy regions of the congurational
space is much more important than in the case of the smaller clusters. Therefore, rather
than varying the move distance from d = 0.10a to 1.0a, as was done in the case of small
clusters, we choose the move distance of d = 0.30a which was the most appropriate for
both pure and mixed smaller clusters. In fact, the move distance d = 0.30a is a good
choice, based on our analysis of smaller clusters. The study on small clusters reveal that
move distance should not be neither too small (d < 0.15a) nor too large (d > 0.60a).
Moreover, in the case of small mixed clusters we found that a small move distance (e.g.,
d = 0.15a) combined with the interchange of the dissimilar atoms on the y has able
to sample the relevant region on the PES and thus locate isomers having structural and
magnetic diversity. Concerning the window acceptance criterion, we set that to 1.5 eV.
From the analysis we found that this value is optimal. Since, on the one hand by using this
value BH algorithm is able visit many of the low-lying isomers near to the ground-state
and on the other hand BH algorithm is able to jump into rare local minima higher in
107
6.5. Results for larger clusters
energy than ground-state. The diversity of the initial starting cluster structures is also
an essential prerequisite for a successful BH global optimization. Therefore, we chose the
dierent structures such as hcp, icosahedra and fcc. The chemical order of the mixed
cluster during the calculations has been changed on the y.
The calculations reveal the remarkable structural and magnetic diversity among the
isomers of Fe
13
, Rh
13
and Fe
6
Rh
7
. In the case of Fe
13
most of the isomers relax to highly
symmetric structures. The magnetism displayed by the cluster structures having similar
shapes is analogous to what has been observed in the small clusters. In fact, dierent
solutions of the Kohn-Shams equations corresponding to the dierent values of the total
spin moment may be found which yield very similar structures after relaxation. For
example, we nd icosahedral isomers having small distortions with respect to each other,
which correspond to dierent magnetic moments. For instance, the structures labeled as
0 (44
B
), 1 (42
B
), 2 (40
B
), and 3 (38
B
), see Fig. 6.11. This is surely an artifact
of the broken spin rotational symmetry of the spin-polarized GGA approximation, only
the ground state is relevant. Although most of the isomers relax into the icosahedral
shape, some of the isomers optimize into dierent shapes. For instance, we obtained the
isomer 4 as a distorted hcp structure with a cluster magnetic moment of 40
B
that is 0.64
eV above the the ground-state structure. We also found a cage-like structure (isomer 5)
with a cluster magnetic moment of 40
B
that is 0.75 eV less stable than the ground-state
structure. The isomer number 6 deserve special attention since it presents an AF-like
coupling among the local magnetic moments. Fe atom located on one of the pentagonal
rings (-3.11
B
) couples anti-ferromagnetically to the rest of the local moments. Taking
into account that the structure is similar to the ground-state, it is likely that this situation
is a artifact of broken spin-symmetry.
The next step is to compare the present results for Fe
13
with those reported in previous
studies [at least the results of the ground-state and the rst isomer]. The DFT studies
reported in Ref. [229] predicted that the ground-state structure is an icosahedra with a
cluster magnetic moment of 44
B
, which coincides with our results. The rst isomer
in Ref. [229] is also an icosahedra with S
z
= 34
B
. We also found the rst isomer is an
icosahedra, but with a cluster magnetic moment of 40
B
. In fact, for S
z
= 40
B
we should
have obtained the same energy and the structures for S
z
= 44
B
, i.e., the ground-state.
Unlike Fe
13
, Rh
13
does not relax into highly symmetric structures but into a cage-
like and layered structures (see Fig. 6.12). The ground-state is a cage-like structure with
a total cluster magnetic moment S
z
= 13
B
in which all local moments are coupled
ferromagnetically. The DFT studies reported in Ref. [190] also yield a cage-like ground-
state structure but with a magnetic moment S
z
= 17
B
. We actually nd that the
structure given in the Ref. [190] as the optimal one is our rst isomer, which lies 0.09 eV
above of ground-state. The isomer number 1 shows the highest total moment S
z
= 17
B
with all the local moments above 1
B
and coupled ferromagnetically. The isomer number
9 is the rst AF-like coupled cluster and having a total cluster moment S
z
= 1
B
. In
fact, 5 out of 13 local moments coupled anti-ferromagnetically with the rest of the local
moments. Finally, we comment on the results for Fe
6
Rh
7
cluster. In this case most of the
isomers adopt the cage-like and layered structures as in Rh
13
. There are rich structural
and magnetic diversity due to the presence of two kind of atoms. In fact, there are 20
homotops in an energy interval of 0.61 eV. It is interesting to analyze the diversity in the
structures having same cluster magnetic moment. Out of 19 isomers 13 have a cluster
magnetic moment S
z
= 27
B
. They display dierent shapes and the dierent distribution
108
Chapter 6. First principles spin-polarized basin-hopping method
{0,0.00,44} {1,0.09,42} {2,0.18,40} {3,0.34,38}
{4,0.64,40} {5,0.75,40} {6,0.91,34} {7,1.37,28}
{8,1.51,36} {9,1.56,8} {10,2.28,24} {11,2.42,2}
{12,2.53,16} {13,2.99,30} {14,3.08,0} {15,3.13,4}
Figure 6.11: Obtained isomers of Fe
13
cluster in the energy range up to 3.13 eV above the
ground-state energy. The entries below the structures start with isomer number followed
by energy of the isomer relative to the ground-state, and last entry is the total cluster
moment.
109
6.5. Results for larger clusters
{0,0.00,13} {1,0.09,17} {2,0.13,5} {3,0.13,13}
{4,0.16,11} {5,0.16,15} {6,0.21,13} {7,0.22,15.6}
{8,0.23,15} {9,0.26,1} {10,0.27,15} {11,0.28,11}
{12,0.29,13} {13,0.33,15} {14,0.35,13} {15,0.37,17}
{16,0.40,9} {17,0.41,15} {18,0.42,7} {19,0.43,5}
Figure 6.12: Obtained isomers of Rh
13
cluster in the energy range up to 0.43 eV above
the ground-state energy.
110
Chapter 6. First principles spin-polarized basin-hopping method
{0,0.00,27} {1,0.024,27} {2,0.03,27} {3,0.09,27}
{4,0.17,27} {5,0.19,27} {6,0.23,27} {7,0.24,15}
{8,0.27,25} {9,0.34,17} {10,0.36,27} {11,0.38,17}
{12,0.41,17} {13,0.42,27} {14,0.44,27} {15,0.49,17}
{16,0.53,27} {17,0.56,27} {18,0.57,27} {19,0.61,17}
Figure 6.13: Low-lying isomers of Fe
6
Rh
7
in the energy range up to 0.61 eV above the
ground-state energy.
111
6.6. Summary and conclusion
of Fe and Rh atoms. Isomer number 7 is the rst one showing AF-like order. Here one
Fe and two Rh local moments are anti-parallel to the rest of the local moments. To our
knowledge, there are no previously reported results concerning the Fe
6
Rh
7
clusters.
6.6 Summary and conclusion
In this work, the overall performance of a rst-principles spin-polarized BH global opti-
mizations sampling was quantitatively analyzed in the case of pure and mixed TM nan-
oclusters. We obtained new results, some of which are unexpected. We have discussed
extensively the technicalities used for choosing the ideal parameters for the optimizations.
In order to elaborate on the question which isomers are relevant, long basin-hopping runs
under dierent move settings, have been performed, revealing dominant isomers or big
attractors as an intrinsic feature of the PES. These are characterized by having being
sampled far more often than the others. Since the system spends most of the time hop-
ping between the dominant isomers, they govern the overall performance of the algorithm.
The performance analysis of the trial move is thereby separated from the acceptance cri-
terion by using a window acceptance scheme, according to which isomers that lie within
the energy window containing the dominant isomers are unconditionally accepted. This
strategy is supported by the fact that for the small pure cluster sizes the simple stochastic
move schemes enable ecient jumps between all parts of the conguration space. The
sampling in connection with simple stochastic moves turns out to be quite robust with
respect to the analyzed parameters. The signicantly inecient settings to be avoided
correspond either to too small move distances, for which the system relaxes essentially
always back to the original minimum, or to too large distances, which tend to bring the
system into the uninteresting high-energy regions of the congurational space. In the case
of mixed clusters, we employed four dierent computational schemes. The ground-state
structure of Fe
13
cluster is found to be an icosahedral structure, whereas Rh
13
and Fe
6
Rh
7
isomers relax into cage-like and layered-like structures, respectively. One of our original
contributions has been to introduce swap or exchange of the positions of the dissimilar
atoms on the y. We have shown that the method of swapping the positions of the dissim-
ilar atoms on-the-y can identify the relevant isomers, even with a small move distance
d = 0.15a in the case of Fe
6
Rh
7
cluster, thus greatly enhancing the reliability of our cal-
culations with signicantly reduced computer time. Both the pure and mixed clusters
display a remarkable variety of structural and magnetic behaviors. We have found that
isomers having similar shapes with small distortion with respect to each other can exhibit
quite dierent magnetic moments. This has been interpreted as a probable artifact of
spin-rotational symmetry introduced by the spin-polarized LDA or GGA.
112
7
Composition dependent orbital magnetism
7.1 Introduction
This chapter is devoted to study the orbital magnetism and MAE in the FeRh clusters as
a function of size and composition. Atomic magnetism is characterized by Hunds rules,
which predicts maximum orbital angular moment L compatible with maximum spin multi-
plicity S, while in TM solids electron delocalization and subsequent band formation result
in an almost complete quenching of L. A similar kind of quenching or reduction is also
observed in the MAE. In fact, MAE in the cubic 3d transition metals is the order of only
a few eV/atom. The purpose of the present investigations of the orbital magnetism and
MAE in the alloy TM clusters is to reveal novel composition and size-dependent features
that are important both from a fundamental standpoint and in view of applications of the
cluster-based magnetic nanometer devices.
7.2 Computational aspects
The calculations reported in this work have been performed in the framework of Hohenberg-
Kohn-Shams density functional theory, [11] as implemented in the Vienna ab initio sim-
ulation package (VASP) [168]. See Sec. 2.2 where the DFT method has been described in
some detail. The wave functions are expanded in a plane wave basis set with the kinetic
energy cut-o E
max
= 700 eV. A simple cubic supercell a is considered having a linear
size of 20

A for 13 atom clusters and 25

A for 19 atom clusters, with the usual periodic
boundary conditions. In this way, pairs of images of the clusters are well separated and
the interaction between them can be neglected.
Magnetic anisotropy and orbital moments are relativistic eects depending on the
strength of the spin-orbit coupling. Spin-orbit coupling has been implemented in VASP
by Kresse and Lebacq [225]. The MAE is dened as the variation of the total energy
of a system as a function of magnetization direction (

M) with respect to the crystalline

113
7.3. Results and Discussion
frame. See Sec. 2.5 where the theory of the MAE has been discussed. In this study, we
considered an octahedral fcc structure for 13 atom clusters and a spherical fcc structure
for 19 atom clusters, which have a highly symmetric shape. Computing the MAE of highly
symmetric structures oer many potential computational advantages to the rst-principles
calculations.
The MAE is composed of mainly two contributions: one is originating from the spin-
orbit coupling and the other one from the magneto-static dipolar interactions (shape
anisotropy). The shape anisotropy is zero in cubic solids, usually small even in anisotropic
solids but often relevant in ultrathin magnetic lms [228] where it can trigger a magnetic
reorientation transition. The minute value of MAE necessitates a very high accuracy of
the numerical calculations since any minute error can accumulate and lead to faulty results
when the resulting energy dierences are so small. In order to achieve correct results the
spin charge density must be extremely well converged and consistent with the symmetry
of the underlying system. A distortion of the symmetry might occur due to numerical
inaccuracies. These can in general be controlled for example by the choice of the energy
cuto and the number of k points in the case of periodic systems. Moreover, the accuracy
can be reduced by an inappropriate choice of the Fourier grid and by perturbations due
to the set up of model system itself. It is however possible to quantify very small changes
in the equilibrium structure due to the changes in the magnetization direction. This can
be the results of electrostatic interactions between the images of the periodic supercell. If
MAE needs to be calculated for a relaxed structure, the structural relaxation has to be
pursued to suciently small values of residual forces on atoms, not exceeding 10

8 eV/

A
in our case. The MAE itself is calculated by keeping the structure xed.
7.3 Results and Discussion
In this section the main important results that I obtained will be presented. In order to
arrange all the data in a compact way, the results are summarized separately.
7.3.1 Size eects
In order to systematically analyze and compare the orbital moments and MAE of two
dierent cluster sizes (13 and 19) having the same symmetry (fcc), we preserved the initial
cluster geometry all along the relaxation process. Thus, the structural transformation are
avoided but elongation or contraction of bonds have been taken into account. However,
it should be emphasized that the small Fe clusters (for instance, N = 13) with unique
symmetry have a tendency to a symmetry-lowering deformation, which is due to the high
degeneracy of d-states at the Fermi level
F
[229].
The reduction of local coordination number and the resulting d-band narrowing in
the nanostructures (such as nanowires, nanoclusters, etc.) composed of TM elements has
interesting consequence for the relative size of the orbital (
L
) and spin (
S
) magnetic
moments. A larger relative enhancement is anticipated for
L
(and MAE), which is in fact
more sensitive to changes in the local coordination because of its dependence on the crystal
eld [26, 230]. We start our analysis by presenting the results from the previous studies
on the magnetic properties of Fe and Rh in the bulk and in the reduced dimensionality.
The
S
and
L
of Fe bulk is 2.25
B
and 0.06
B
, respectively. Whereas, Fe dimer shows
much more enhanced moments:
S
= 2.92
B
and
L
= 0.16
B
[231]. Rh is non-magnetic
114
Chapter 7. Composition dependent orbital magnetism
(i) Fe
11
Rh
2
(ii) Fe
9
Rh
4
(iii) Fe
8
Rh
5
(iv) Fe
6
Rh
7
(v) Fe
4
Rh
9
(vi) Fe
3
Rh
10
(vii) Fe
2
Rh
11
Figure 7.1: The distribution of Fe and Rh atoms in the FeRh clusters having N = 13
atoms. The green (blue) color refers to Fe (Rh) atom.
in the bulk. However, small Rh clusters show nite spin polarization [51]. Interestingly,
the Rh dimer shows enhanced moments
S
= 1.93
B
and
L
= 0.91
B
[231].
The results for
S
and
L
in pure Fe and Rh clusters shall be compared with those of
extreme limits mentioned in the last paragraph (see Tables. 7.1 and 7.2, and Fig. 7.3). We
observe a remarkable enhancement in magnetic moments for both Fe and Rh pure clusters
as compared to their bulk moments. One can see that the magnetic enhancements are
more appealing in pure Rh
N
. For instance,
S
= 1.461
B
and
L
= 0.336
B
for Rh
13
,
while,
S
= 0.578
B
and
L
= 0.075
B
for Rh
19
. In the case of pure Fe
N
, we nd that
Fe
19
shows enhanced magnetic moments compared to Fe
13
. This trend is opposite to the
case of pure Rh
N
since Rh
13
exhibits enhanced magnetic moments compared to Rh
19
(see
Fig. 7.3). It should be stressed that the above conclusions are based on our optimal fcc
structures. However, we cannot completely exclude the possibility of identifying cluster
structures that are more stable for the clusters sizes and compositions considered here.
Fig. 7.4 demonstrates that the relative change in the MAE due to the size change
is more important for Rh clusters compared to Fe clusters. In the past, several authors
pointed out that one can relate the MAE with the anisotropy of other spin-orbit induced
properties. The most prominent example for this is the relation of the MAE and the
anisotropy of the orbital angular momentum, that was considered by Bruno [118] on the
basis of second-order perturbation theory. According to Brunos model the MAE can be
derived as
E =

4
B
L

7.3.1
where is the spin-orbit parameter and L are the dierence in orbital moment
between the easy and the hard axis. Eq. 7.3.1 shows that MAE depends on the spin-orbit
115
7.3. Results and Discussion
(i) Fe
17
Rh
2
(ii) Fe
15
Rh
4
(iii) Fe
13
Rh
6
(iv) Fe
10
Rh
9
(v) Fe
9
Rh
10
(vi) Fe
6
Rh
13
(vii) Fe
4
Rh
15
(viii) Fe
2
Rh
17
Figure 7.2: The distribution of Fe and Rh atoms in the FeRh clusters having N = 19
atoms. The green (blue) color refers to Fe (Rh) atom.
116
Chapter 7. Composition dependent orbital magnetism
Table 7.1: Electronic and magnetic properties of FeRh clusters having N = 13 atoms.
Results are given for the average spin moment per atom
S
= 2S
z
/N (in
B
), average
orbital moment per atom
L
(in
B
), the ratio between orbital and spin moment

L

,
magnetic anisotropy energy (MAE) in meV per cluster, the easy axis direction (), and
is the dierence in total energy per atom between calculations with and without SOI (in
meV).
Cluster x
Fe

S

L

S
Fe

L
Fe

MAE

S
Rh

L
Rh
Fe
13
1.00 2.769 0.056 2.597 0.023 2.73 x 6.6
Fe
11
Rh
2
0.84 2.769 0.072 2.856 0.062 0.028 2.21 z 11.1
0.835 0.066
Fe
9
Rh
4
0.69 2.615 0.097 3.031 0.082 0.041 7.80 z 16.3
0.899 0.130
Fe
8
Rh
5
0.61 2.231 0.096 2.966 0.101 0.045 5.07 xz 19.4
0.780 0.106
Fe
6
Rh
7
0.53 2.231 0.094 3.209 0.091 0.046 7.93 z 25.2
1.002 0.105
Fe
4
Rh
9
0.31 2.077 0.103 3.355 0.065 0.055 14.95 xz 29.7
1.216 0.119
Fe
3
Rh
10
0.23 1.692 0.113 2.959 0.068 0.073 6.24 x 34.0
1.125 0.127
Fe
2
Rh
11
0.15 1.615 0.161 3.146 0.102 0.109 28.6 xz 39.8
1.166 0.171
Rh
13
0.00 1.461 0.336 1.304 0.257 15.6 xz 45.1
parameter and the magnitude of the orbital moment. The obtained values of magnetic
moments and MAE can be now related by using the Eq. 7.3.1. Rh, as a 4d element, is
heavier than Fe. We may expect that Rh can display larger orbital moments and spin-orbit
energy compared to Fe provided that it is magnetic. As we shall see the calculations show
that this is not always the case.
7.3.2 Composition eects on the magnetic properties
This section presents the magnetic properties as a function of the composition by increas-
ing the Fe content level systematically. Computing the magnetic properties by taking
into account all possible homotops may be possible for a cluster having small number of
atoms (for instance N 6). However this procedure is impracticable for the cluster sizes
considered here. Therefore, in order to reduce the computing time, we placed the atoms
117
7.3. Results and Discussion
0 0.2 0.4 0.6 0.8 1
0.5
1
1.5
2
2.5
3
S
p
i
n

m
o
m
e
n
t

[

/
a
t
o
m
]
13
19
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
O
r
b
i
t
a
l

m
o
m
e
n
t

[

/
a
t
o
m
]
Fe concentration (m)
(a)
(b)
Figure 7.3: Spin and orbital moments in N = 13 and 19 FeRh clusters as a function of Fe
concentration.
0 0.2 0.4 0.6 0.8 1
Fe concentration (m)
0
5
10
15
20
25
30
M
A
E

(
m
e
V
/
c
l
u
s
t
e
r
)
13
19
Figure 7.4: Magnetic anisotropy energy in FeRh clusters having N = 13 and 19 atoms as
a function of Fe concentration.
118
Chapter 7. Composition dependent orbital magnetism
Table 7.2: Electronic and magnetic properties of FeRh clusters having N = 19 atoms as
in Table 7.1.
Cluster x
Fe

S

L

S
Fe

L
Fe

MAE

S
Rh

L
Rh
Fe
19
1.00 2.947 0.101 2.797 0.038 0.064 xz 6.3
Fe
17
Rh
2
0.89 2.736 0.056 2.816 0.062 0.022 4.18 z 9.4
0.642 0.006
Fe
15
Rh
4
0.78 2.485 0.066 2.931 0.071 0.027 0.251 z 12.6
0.814 0.052
Fe
13
Rh
6
0.68 2.316 0.091 2.924 0.104 0.041 2.85 xz 15.7
0.713 0.062
Fe
10
Rh
9
0.52 2.263 0.067 3.164 0.065 0.032 7.22 z 21.9
0.911 0.069
Fe
9
Rh
10
0.47 1.578 0.038 2.897 0.075 0.025 5.89 x 24.6
0.244 0.004
Fe
6
Rh
13
0.31 1.631 0.078 3.168 0.109 0.050 0.235 z 30.9
0.833 0.064
Fe
4
Rh
15
0.21 0.377 0.023 1.587 0.090 0.063 3.23 x 35.8
0.036 0.005
Fe
2
Rh
17
0.10 0.772 0.023 2.908 0.121 0.031 24.89 z 37.9
0.494 0.012
Rh
19
0.00 0.578 0.075 0.568 0.132 6.27 xz 42.6
of the same kind on the symmetric positions of the cluster, specially for low Fe or Rh
concentration (see Figs. 7.1 and 7.2).
For the 13 atom cluster,
S
increases almost linearly with the Fe content level, whereas

L
decreases. The decrease in the
L
curve (0.336
B
0.161
B
) is found for the lowest
Fe concentration (Rh
13
Fe
2
Rh
11
). In contrast in the 19 atom clusters, an oscillatory
behavior in the
S
and
L
are observed. The plot for
S
in the Fig. 7.3 shows a dip
around Fe
4
Rh
15
composition. This reduction in the spin moment is partly due to the
antiferromagnetic-like coupling among the Fe and Rh moments and also due to the small
average moment contribution from the Fe atoms. The Fe spin moments are distributed
as
1
= 2.962,
2
= 2.972,
3
= 0.075,
4
= 0.341
B
. The strong oscillations in the
L
values for 19 atoms clusters shows that orbital moments are more sensitive to the local
environment within the cluster.
We found that the
S
contribution to the total magnetic moment is fairly insensitive
to the direction of magnetization, while there is a large anisotropy in the orbital moments
depending on the magnetization direction. In the pure clusters, the easy axis is parallel to a
direction which is having largest orbital moment. While for some specic composition (for
119
7.3. Results and Discussion
instance, Fe
2
Rh
17
), it is observed that the cluster axis having the largest L
z
are not easy
axis. A similar eect has been found in FeCo clusters having uni-axial symmetry [235].
Fig. 7.4 presents the variation of MAE as a function of composition. One can observe
that the MAE displays a strong oscillation for both sizes and the maxima of MAE is for
low but non-zero Fe concentration. It eventually approaches very small values for the pure
Fe clusters. The quantitative analysis of Fig. 7.4 even reveals some remarkable results.
A remarkable non-monotonous dependence of the MAE is observed as a function of Fe
content, i.e., upon going from pure Fe to pure Rh. This leads to an important increase of
the MAE, which reaches about 300% at the optimal Fe concentration. This oers multiple
possibilities of tailoring the magneto-anisotropic behavior in nanoalloys [197].
7.3.3 Angular dependence of the MAE
-0.06
-0.04
-0.02
0
0
2
4
0
0.1
0.2
-2
-1
0
1
0
2
4
6
-6
-4
-2
0
0
0.1
0.2
-3
-2
-1
0
0 30 60 90 120 150 180
0
10
20
0 30 60 90 120 150 180
-6
-4
-2
0
E
n
e
r
g
y

d
i
f
f
e
r
e
n
c
e

(
m
e
V
/
c
l
u
s
t
e
r
)

(degrees) (degrees)
Fe
19
Fe
17
Rh
2
Fe
15
Rh
4
Fe
13
Rh
6
Fe
9
Rh
10
Fe
10
Rh
9
Fe
6
Rh
13
Fe
4
Rh
15
Rh
19
Fe
2
Rh
17
Figure 7.5: The energy dierence E = E()E(0) in meV/cluster vs of FeRh 19 atom
clusters. The magnetization direction varies in the zx plane in steps of 10 degrees.
120
Chapter 7. Composition dependent orbital magnetism
Fig. 7.5 shows the angle () dependent energy dierence E() of the optimized 19
atom clusters. Here refers to the polar angle and azimuthal angle () has been xed
to zero. The gure reects the underlying symmetry of the clusters. We can classify the
clusters according to the number of easy axis they possess. Fe
19
and Rh
19
clusters have two
easy axis and three hard axis within the zx plane, which are symmetrically equivalent. In
the case of pure clusters, the easy axis is obtained if the magnetization is directed parallel
to a line from the central atom to an atom on the middle of a facet. And the hard axis
is obtained if the magnetization is directed parallel to a line from the central atom to an
atom on the corner. The binary clusters showing a single minimum in the zx plane are
Fe
4
Rh
15
, and Fe
9
Rh
10
, while Fe
2
Rh
17
, Fe
10
Rh
9
, Fe
15
Rh
4
, and Fe
17
Rh
2
show 2 minimum
and Fe
6
Rh
13
shows 3 minimum. However, the optimal structure found for Fe
13
Rh
6
cluster
is somehow distorted and does not display high symmetry properties.
7.4 Summary
The spin moments, orbital moments and magnetic anisotropy energy (MAE) of fcc-like
clusters having N = 13 and 19 atoms have been determined as a function of composition.
A remarkable non-monotonous dependence of the MAE is observed as a function of Fe
content, i.e., upon going from pure Fe to pure Rh. This leads to an important increase of
the MAE, which reaches about 300% at the optimal Fe concentration. This oers multiple
possibilities of tailoring the magneto-anisotropic behavior in nanoalloys.
121
8
Summary and outlook
The principal motivation of this thesis was to enrich the fundamental understanding of
the structural and magnetic properties exhibited by the TM nanoalloy clusters in view of
applications in cluster-based magnetic nanometer devices. The results presented in this
thesis open the way to several near future investigations and developments in this eld.
Specic conclusions concerning each chapter have already been mentioned at the end of
the corresponding chapters. For this reason the present nal chapter is quite compact. In
the following, I briey discuss how the thesis has been progressed.
Chapter 1 through 3 are devoted to present the essential background material for the
calculations performed in this thesis, while chapters 4 through 7 demonstrate the results.
In the rst part, i.e., in chapters 4 and 5, we combined the state-of-the-art Hohenberg-
Kohn Shams DFT with a global optimization technique based on a graph theory method.
This method has been used to perform a thorough and systematic study on the interplay
between cluster structure, magnetism and the chemical order in the Fe
m
Rh
n
and Co
m
Pd
n
nanoclusters having N = m + n 8 atoms. For N = m + n 6 a thorough sampling
of all cluster topologies has been performed. We would like to emphasis that this kind
of study is rather unique. For N = 7 and 8 only a few representative topologies were
considered including both open and compact structures. Choosing a small set of repre-
sentative topologies for N = 7 and 8 is justied by the fact that the number of isomers
(number of local minima on the PES) increases exponentially with the cluster size N.
Indeed, the computational complexity increases even more rapidly in the case of binary
clusters, since one has to take into account all possible homotops. For all the clusters
the entire concentration range is systematically investigated, and the dierent initial mag-
netic congurations such as ferro- and anti-ferromagnetic coupling are considered. The
results in the case of FeRh clusters are the following: an increase of the average magnetic
moment (
N
) and magnetic stabilization energy (E
m
) with increasing Fe concentration,
the presence of small dierences in the average magnetic moment (
N
) between low-lying
isomers, the dominant role of the d-electron spin polarization within the PAW spheres,
the enhancement of the Fe moments upon Rh doping, and a general tendency to max-
122
Chapter 8. Summary and outlook
imize the number of mixed bonds. We have adopted a similar approach in the case of
Co
m
Pd
n
clusters having N = m + n 8 atoms. The main results in this case are the
following: The optimized cluster structures have a tendency to maximize the number of
nearest-neighbor CoCo pairs. An increase of
N
and E
m
is observed with increasing Co
concentration. The magnetic order is ferromagnetic-like (FM) for all ground-state struc-
tures. However, an antiferromagnetic-like (AF) order has been obtained in some of the
rst exited isomers. The maximal local spin polarization for Co and Pd atoms are found
in the equiatomic compositions (Co
2
Pd
2
, Co
3
Pd
3
and Co
4
Pd
4
). We found that taking
into account spin-orbit (SO) interactions in FeRh and CoPd clusters does not alter the
ground-state structures found by using the scalar relativistic (SR) calculations. FeRh and
CoPd clusters are expected to develop a variety of further interesting behaviors, which
still remain to be explored. For instance, larger FeRh cluster should show a more complex
dependence of the magnetic order as a function of concentration. In particular for large
Rh content one should observe a transition from FM-like to AF-like order with increas-
ing cluster size, in agreement with the AF phase found in solids for more than 50% Rh
concentration. Moreover, the metamagnetic transition observed in bulk FeRh alloys also
puts forward the possibility of similar interesting phenomena in nanoalloys as a function
of temperature.
In chapter 6 we have developed a DFT based spin-polarized basing-hopping algorithm.
This methodological approach has been applied to study the structural and magnetic prop-
erties of pure and alloy TM nanoclusters. This method is found to be very impressive.
For instance, in the case of pure clusters, we obtained several Jahn-Teller distorted mag-
netic isomers of the same basic structural motif which, being similar, would have been
most likely missed by using the graph or topographical scheme employed in the rst part
of this thesis. A similar situation has also been encountered in the case of mixed clus-
ters, where we identied several Jahn-Teller distorted magnetic isomers having the same
composition and a similar distributions of the two kinds of atoms. We have discussed
extensively the technicalities used for choosing the ideal move parameters for the opti-
mizations. Moreover, we have implemented a window acceptance criterion. Moreover, in
the case of mixed clusters we swap or exchange the positions of the dissimilar atoms on
the y. We noticed that this method greatly enhances the performance of our calcula-
tions by signicantly reducing the CPU time. For the small clusters (e.g. Fe
6
, Rh
6
and
Fe
3
Rh
3
) the main result is the presence of dominant (or frequently visited) isomers. We
found that this is an intrinsic feature of the basin-hopping method. This is interpreted
as a consequence of the reduced system size and the resulting small number of low-lying
isomers. The dominant isomers govern the overall computational demand of the sampling
and are therefore the relevant isomers for the performance analysis. In the case of larger
clusters (e.g. Fe
13
, Fe
6
Rh
7
and Rh
13
) we applied a similar computational procedure as
the one used in the case of small clusters. In fact, both pure and mixed clusters display
remarkable structural and magnetic diversity. We found that isomers having similar shape
but with a small distortion among each other can exhibit often quite dierent magnetic
moments. This has been interpreted as a probable artifact of the spin-rotational symmetry
breaking introduced by the spin-polarized LDA or GGA. In the case of Fe
6
Rh
7
cluster, an
implementation consisting of small move distances of about 0.15a (where a is the length
of the shortest bond) in combination with swapping of Fe and Rh atoms, could identify
all the relevant isomers including the ground-state. The ground-state structure of Fe
13
cluster is found to be an icosahedral structure, whereas Rh
13
and Fe
6
Rh
7
isomers relax
123
into cage-like and layered-like structures, respectively. The possibility of combining the
spin-polarized density-functional theory with some other global optimization techniques
such as minima-hopping method could be the next step in this direction. This combina-
tion is expected to be an ideal sampling approach by having the advantage of eciently
avoiding search over irrelevant regions of the potential energy surface.
In chapter 7 we investigated the composition dependence of orbital magnetism and
magnetic anisotropy energy in FeRh nanoclusters. A remarkable non-monotonous depen-
dence of the MAE is observed as a function of Fe content, i.e., upon going from pure Fe to
pure Rh. This leads to an important increase of the MAE, which reaches about 3 times the
value for pure clusters at the optimal Fe concentration. This oers multiple possibilities
of tailoring the magneto-anisotropic behavior in nanoalloys. In conclusion, it is our hope
to see the experimental verication of the reported results in this thesis.
124
Bibliography
[1] J. Bansmann, S. Baker, C. Binns, J. Blackman, J.-P. Buecher, J. Dorantes-Davila,
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, Surf. Sci. Rep. 56,
189 (2005).
[2] G. M. Pastor, J. Dorantes-Davila, S. Pick, and H. Dreysse, Phys. Rev. Lett. 75, 326
(1995).
[3] J. Dorantes-Davila, H. Dreysse, and G. M. Pastor, Phys. Rev. B 46, 10432 (1992).
[4] G. M. Pastor, J. Dorantes-Davila, and K. H. Bennemann, Physica B 149, 22 (1988);
Phys. Rev. B 40, 7642 (1989).
[5] J. P. Bucher, D. C. Douglass, and L. A. Bloomeld Phys. Rev. Lett. 66, 3052 (1991);
D. C. Douglass, J. P. Bucher, and L. A. Bloomeld, Phys. Rev. B 45, 6341 (1992); D.
C. Douglass, A. J. Cox, J. P. Bucher, and L. A. Bloomeld, ibid. 47, 12874 (1993).
[6] I. M. L. Billas, J. A. Becker, A. Chatelain, and W. A. de Heer, Phys. Rev. Lett. 71,
4067 (1993); I. M. L. Billas, A. Chatelain, and W. A. de Heer, Science 265, 1682
(1994).
[7] S. E. Apsel, J. W. Emmert, J. Deng, and L. A. Bloomeld, Phys. Rev. Lett. 76, 1441
(1996).
[8] M. B. Knickelbein, Phys. Rev. Lett. 86, 5255 (2001).
[9] G. Nicolas, J. Dorantes-Davila, and G.M. Pastor, Phys. Rev. B 74, 014415 (2006).
[10] M. Moseler, H. Hakkinen, R. N. Barnett, and U. Landman, Phys. Rev. Lett. 86, 2545
(2001).
[11] P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964); W. Kohn and L. J. Sham,
Phys. Rev. 140, A1133 (1965).
[12] C. Moller and M. S. Plesset, Phys. Rev. 93, 618 (1934).
[13] U. V. Barth and L. Hedin, I, J. Phys. C 5, 1629 (1972).
[14] M. M. Pant and A. K. Rajagopal, Solid State Commun. 10, 1157 (1972).
[15] J. Kubler, Theory of itinereant electron magnetism, Oxford Science Publication, re-
vised edition, (2009).
125
Bibliography
[16] M. Mu noz-Navia, J. Dorantes-Davila, D. Zitoun, C. Amiens, B. Chaudret, M.-J.
Casanove, P. Lecante, N. Jaouen, A. Rogalev, M. Respaud, and G. M. Pastor, Faraday
Discuss. 138, 181 (2008); M. Mu noz-Navia, J. Dorantes-Davila, M. Respaud, and G.
M. Pastor, Eur. J. Phys. D 52, 171 (2009).
[17] G. M. Pastor, R. Hirsch, and B. M uhlschlegel, Phys. Rev. Lett. 72, 3879 (1994);
Phys. Rev. B 53, 10382 (1996).
[18] R. Garibay-Alonso, J. Dorantes-Davila, and G. M. Pastor, Phys. Rev. B 79, 134401
(2009).
[19] D. J. Wales, M. A. Miller, and T. R. Walsh, Nature 394, 758 (1998).
[20] J. P. K. Doye and D. J. Wales, J. Chem. Phys. 105, 8428 (1996).
[21] J. H. Van Vleck, Quantum mechanics - the key to understanding magnetism, Rev.
Mod. Phys. vol. 50, no. 2, p. 191, (1978).
[22] M. Saubanere, M. Tanveer, P. Ruiz-Daz, and G. M. Pastor, Status Solidi B. 247,
2610-2620 (2010).
[23] C. Kittel, Introduction to solid state physics, New York: Wiley, 7th ed., (1996).
[24] R. J. Davis, Alloying: understanding the basics, ASM International (2001).
[25] M. Getzla, Fundamentals of magnetism. Berlin: Springer.
[26] S. Chikazumi, Physics of ferromagnetism, Oxford Science Publication, 2th ed.,
(2009).
[27] J. Stohr, Journal of Magnetism and Magnetic Materials, 200, 470-497, (1999).
[28] B.T. Thole, P. Carra, F. Sette, G. van der Laan, Phys. Rev. Lett. 68, 1943, (1992).
[29] K. Lee, J. Callaway, K. Kwong, R. Tang, A. Ziegler, Phys. Rev. B 31 1796, (1985).
[30] W.A. de Heer, P. Milani, A. Chatelain, Phys. Rev. Lett. 65, 488 (1990).
[31] I.M.L. Billas, J.A. Becker, A. Chatelain, W.A. de Heer, Phys. Rev. Lett. 71, 4067
(1993).
[32] J.P. Bucher, D.C. Douglass, L.A. Bloomeld, Phys. Rev. Lett. 66, 3052 (1991).
[33] D.C. Douglass, J.P. Bucher, L.A. Bloomeld, Phys. Rev. Lett. 68, 1774 (1992).
[34] D.C. Douglass, A.J. Cox, J.P. Bucher, L.A. Bloomeld, Phys. Rev. B 47, 12874 (1993).
[35] I.M.L. Billas, A. Chatelain, W.A. de Heer, Science 265, 1682, (1994).
[36] A.J. Cox, J.G. Louderback, L.A. Bloomeld, Phys. Rev. B 49, 12295 (1994).
[37] E. C. Stoner, Proc. R. Soc. Lond. A 154, 656 (1936).
[38] R. Pfandzelter, G. Steierl, C. Rau, Phys. Rev. Lett. 74, 3467 (1995).
126
Bibliography
[39] K. Wildberger, V.S. Stepanyuk, P. Lang, R. Zeller, P.H. Dederichs, Phys. Rev. Lett.
75, 509, (1995).
[40] C. Binns, Surf. Sci. Rep. 44 (2001).
[41] R.A. Guirado-Lopez, J. Dorantes-Davila, G.M. Pastor, Phys. Rev. Lett. 90, 226402
(2003).
[42] P. Gambardella, S. Rusponi, M. Veronese, S. S. Dhesi, C. Grazioli, A. Dallmeyer, I.
Cabria, R. Zeller, P. H. Dederichs, K. Kern, C. Carbone, H. Brune, Science 300, 1130
(2003).
[43] S. Rusponi, T. Cren, N. Weiss, M. Epple, P. Buluschek, L. Claude and H. Brune,
Nature Materials 2, 546 (2003); W. Kuch, i bid. 2, 505 (2003).
[44] T. Jamneala, V. Madhavan, and M. F. Crommie, Phys. Rev. Lett. 87, 256804 (2001).
[45] V. Madhavan, T. Jamneala, K. Nagaoka, W. Chen, Je-Luen Li, S. G. Louie, and M.
F. Crommie, Phys. Rev. B 66, 212411 (2003).
[46] F. Silly, M. Pivetta, M. Ternes, F. Patthey, J. P. Pelz, and W.-D. Schneider, Phys.
Rev. Lett. 92, 16101 (2004).
[47] P. Gambardella, A. Dellmeyer, K. Maiti, M. C. Malagoli, W. Eberhardt, K. Kern,
and C. Carbone, Nature (London) 416, 301 (2002).
[48] V. S. Stepanyuk, L. Niebergall, R. C. Longo, W. Herget and P. Bruno, Phys. Rev. B
70, 75414 (2004).
[49] K. W. Edmonds, C. Binns, S. H. Baker, S. C. Thornton, C. Norris, J. B. Goedkoop,
M. Finazzi, and N. B. Brookes, Phys. Rev. B 60, 472 (1999).
[50] R. Felix-Medina, J. Dorantes-Davila, and G. M. Pastor, Phys. Rev. B. 67, 094430
(2003).
[51] A.J Cox, Phys. Rev. Lett. 86, 5597 (2001).
[52] R. Galicia, Rev. Mex. Fis. 32, 51 (1985).
[53] B. V. Reddy, S. N. Khanna, and B. I. Dunlap, Phys. Rev. Lett. 70, 3323 (1993).
[54] O. Stern, Z. Phys. 7, 249 (1921).
[55] G. Schutz, W. Wagner, W. Wilhelm, P. Kienle, Phys. Rev. Lett. 58, 737 (1987).
[56] D.P. Pappas, A.P. Popov, A.N. Anisimov, B.V. Reddy, S.N. Khanna, Phys. Rev. Lett.
76, 4332 (1996).
[57] D. Gerion, A. Hirt, I.M.L. Billas, A. Chatelain, W.A. de Heer, Phys. Rev. B 62, 7491,
(2000).
[58] D. Gerion, A. Hirt, A. Chatelain, Phys. Rev. Lett. 83, 532 (1999).
[59] J. Bansmann, A. Kleibert, Appl. Phys. A 80 (2005).
127
Bibliography
[60] D.J. Sellmyer, M. Yu, R.D. Kirby, Nanostruct. Mater. 12, 1021 (1999).
[61] S. Sun, C.B. Murray, D. Weller, L. Folks, A. Moser, Science 287, 1989 (2000).
[62] S. Stoyanov, Y. Huang, Y. Zhang, V. Skumryev, G.C. Hadjipanayis, D. Weller, J.
Appl. Phys. 93, 7190 (2003).
[63] T. Burkert, L. Nordstrom, O. Eriksson, O. Heinonen, Phys. Rev. Lett. 93, 027203
(2004).
[64] M.B. Knickelbein, Chem. Phys. Lett. 353, 221 (2002).
[65] M.B. Knickelbein, J. Chem. Phys. 125, 044308 (2006).
[66] J. T. Lau, A. Flisch, R. Nietubyc, M. Reif, and W. Wurth, Phys. Rev. Lett. 89,
57201 (2002).
[67] S. Blundell, Magnetism in Condensed Matter, Oxford University Press, (2001).
[68] S. Rohart, F. Tournus, and V. Dupuis, arXiv:1105.6292v1
[69] E. Schrdinger, Ann. Physik 79, 361, (1926).
[70] M. Born and R. Oppenheimer, Ann. Phys. 84, 457, (1927).
[71] N. Balzs, Phys. Rev. 156, 42, (1967).
[72] E. Lieb and B. Simon, Phys. Rev. Lett. 31, 681, 1973.
[73] E. Teller, Rev. Mod. Phys. 34, 627, (1962).
[74] V. Fock, Z. Phys. 61, 126, (1930).
[75] J. Slater, Phys. Rev. 81, 385, (1951).
[76] A. L. Fetter and J. D. Walecka, 1971, Quantum Theory of Many- Particle Systems,
McGraw-Hill, New York.
[77] V. Nemoshkalenco and V. Antonov, Computational Methods in Solid State Physics,
OPA, (1998).
[78] W. Kohn, Rev. Mod. Phys. 71, 1253 (1999).
[79] E. Fermi, Z. Phys. 61, (1928).
[80] L. Thomas, Proc. Cambridge Philos. Soc. 23, 542, (1927).
[81] R. O. Jones and O. Gunnarsson, Rev. Mod. Phys. 61, 689, (1989).
[82] P. Rushton, Towards a Non-Local Density Functional, Description of Exchange
and Correlation, Dissertation Departments of Chemistry and Physics, University of
Durham (2002).
[83] N. Argaman and G. Makov, Amer. J. Phys. 68, 69, (2000).
128
Bibliography
[84] W. Kohn, A. Becke and Parr R, J. Phys. Chem. 100, 12974, (1996).
[85] H. Eschrig, Lectures notes in Physics 642, 7, (2004).
[86] P. A. M. Dirac, Proc. R. Soc. London, 123 (792): 714-733, 1929.
[87] http://newton.ex.ac.uk/research/qsystems/people/jenkins/mbody/mbody3.html
[88] W. Kohn, 1995, Density Functional Theory, Chapter 21, Euroconference on Computer
Simulation in Condensed Matter Physics and Chemistry 49, Italian Physical Society.
[89] J. Perdew, Phys. Rev. Lett 55, 1665, (1985).
[90] J. Perdew, Phys. Rev. B 33, 8800, (1986).
[91] A. D. Becke, J. Chem. Phys.84, 4524, 1986.
[92] C. Filippi, C. J. Umrigar, and M. Taut, J. Chem. Phys.100, 1290, 1994.
[93] J. Hafner, C. Wolverton and G. Ceder, MRS Bulletin 31, 659.
[94] O. Heinonen, M. Lubin and Johnson M, Phys. Rev. Lett. 75, 4110, (1995).
[95] S. K. Ghosh, Bull. Mater. Sci. 26, 3, (2003).
[96] K. Capelle, arXiv:cond-mat/0211443v5, 2003.
[97] A. Gross, Theoretical Surface Science, A microscopic Perspevtive, Springer, (2002).
[98] p. Hohenberg and W. Kohn, Phys. Rev. B 136, 864, (1964).
[99] A. V. Krasheninnikov, Computational Methods for Material Science, Lecture notes,
University of Helsinki (2000).
[100] M. Payne, M. Teter, D. Allan, T. Arias and J. Joannopoulos, Rev. Mod. Phys. 64,
1045 (1992).
[101] J. Perdew, R. Parr, M. Levy, and J. Balduz, Phys. Rev. Lett. 49, 1691, (1982).
[102] J. Perdew and A. Zunger, Phys. Rev. B 23, 5048, (1981).
[103] S. Vosko, L. Wilk, and M. Nusair, Can. J. Phys. 58, 1200, (1980).
[104] O. Gunnarsson, and R. O. Jones J. Chem. Phys. 72, 5357 (1980).
[105] O. Gunnarsson and B. I. Lundqvist, Phys. Rev. B 13, 4274 (1976).
[106] P. Blochl, Bull. Mater. Sci. 26, p.33 (2003).
[107] O.K. Andersen, Phys. Rev. B 12, 3060 (1975).
[108] J. C. Slater and G. F. Koster, Phys. Rev. 94, 1498 (1954).
[109] P.M. Marcus, Int. J. Quantum. Chem. 1S, 567 (1967).
129
Bibliography
[110] J. Korringa, Physica 13, 392 (1947).
[111] W. Kohn and N. Rostocker, Phys. Rev. 94, 111 (1954).
[112] D.R. Hamann, M. Schlter and C. Chiang, Phys. Rev. Lett. 43, 1494 (1979).
[113] D. Vanderbilt, Phys. Rev. B 41, 7892 (1985).
[114] P. E. Blchl, Phys. Rev. B 50, 17953 (1994).
[115] A. D. Becke, Phys. Rev. A, 38, 3098-3100 (1988).
[116] K. Burke, J. P. Perdew, and M. Ernzerho, Journal of Chem. Phys,109, 3760-3771
(1998).
[117] S. Kaya, Sci. Reports Tohoku Univ. 17, 639 (1928).
[118] P. Bruno, J. Phys. F. Met. Phys. 18, 1291 (1988).
[119] F. Jensen, Introduction to Computational Chemistry, (Wiley, Chichester, 1999).
[120] D. F. Shanno, Math. Oper. Res. 3, 244 (1978).
[121] M. J. D. Powell, Nonconvex minimization calculations and the conjugate gradient
method, in Lecture Notes in Mathematics, Vol. 1066, p. 122141, Springer, Berlin,
(1984).
[122] R. Fletcher and C. M. Reeves, Comp. J. 7, 149 (1964).
[123] M. R. Hestenes and E. L. Stiefel, J. Res. Nat. Bur. Stand. 49, 409 (1952).
[124] L. T. Wille and J. Vennik, J. Phys. A 18, L419 (1985).
[125] L. Zhan, J. Z. Y. Chen, W.-K. Liu, and S. K. Lai, J. Chem. Phys. 122, 244707
(2005).
[126] D. Baker, Nature 405, 39 (2000).
[127] D. Baker and A. Sali, Science 294, 93 (2001).
[128] W. K. Press, S. A. Teukolsky, W. T. Vetterlin, and B. T. Flannery, Numerical
Recipes, 3rd edition ed. (Cambridge University Press, Cambridge, 2007).
[129] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, Science 220, 671 (1983).
[130] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, J.
Chem. Phys. 21, 1087 (1953).
[131] Y. Xiang, D. Y. Sun, W. Fan, and X. G. Gong, Phys. Lett. A 233, 216 (1997).
[132] Y. Xiang, D. Y. Sun, and X. G. Gong, J. Phys. Chem. A 104, 2746 (2000).
[133] D. J. Wales and H. A. Scheraga, Science 285, 1368 (1999).
[134] Z. Li and H. A. Scheraga, Proc. Natl. Acad. Sci. U.S.A. 84, 6611 (1987).
130
Bibliography
[135] D. J. Wales, J. P. K. Doye, M. A. Miller, P. N. Mortenson, and T. R. Walsh, Adv.
Chem. Phys. 115, 1 (2000).
[136] J. Pillardy and L. Piela, J. Phys. Chem. 99, 11805 (1995).
[137] J. P. K. Doye, M. A. Miller, and D. J. Wales, J. Chem. Phys. 110, 6896 (1999).
[138] J. P. K. Doye and D. J. Wales, Phys. Rev. Lett. 80, 1357 (1998).
[139] D. M. Deaven and K. M. Ho, Phys. Rev. Lett. 75, 288 (1995).
[140] D. M. Deaven, N. Tit, J. R. Morris, and K. M. Ho, Chem. Phys. Lett. 256, 195,
(1996).
[141] First-Principles Basin-Hopping for the Structure Determination of Atomic Clusters,
Ralf Gehrke, PhD thesis (2008).
[142] S. Goedecker, J. Chem. Phys. 120, 9911 (2004).
[143] S. Goedecker, W. V. Hellmann, and T. Lenosky, Phys. Rev. Lett. 95, 055501, (2005).
[144] F. Jensen, Introduction to Computational Chemistry (Wiley, Chichester, 1999).
[145] R. P. Bell, Proc. R. Soc. London, Ser.A 154, 414 (1936).
[146] M. G. Evans and M. Polanyi, Trans. Faraday Soc. 32, 1333 (1936).
[147] D. J. Wales and J. P. K. Doye, J. Phys. Chem. A 101, 5111 (1997).
[148] U. H. E. Hansmann and L. T. Wille, Phys. Rev. Lett. 88, 068105 (2002).
[149] L. Zhan, J. Z. Y. Chen, and W.-K. Liu, Phys. Rev. E 73, 015701 (2006).
[150] See, for instance, Faraday Discuss. 138 (2008).
[151] S. Dennler, J. L. Ricardo-Chavez, J. Morillo, and G. M. Pastor, Eur. Phys. J. D 24,
237 (2003).
[152] I. Efremenko, and M. Sheintuch, Chem. Phy. Lett. 401, 232-240 (2005).
[153] S. Ganguly, M. Kabir, S. Datta, B. Sanyal, and A. Mookerjee, Phys. Rev. B. 78,
014402 (2008).
[154] P. Entel, and M. E. Gruner, J. Phys. Condens. Matter 21, 064228 (2009).
[155] A. N. Andriotis, G. Mpourmpakis, G. E. Froudakis, and M. Menon, J. Chem. Phys.
120, 11901 (2004).
[156] G. Rollmann, S. Sahoo, A. Hucht, and P. Entel, Phys. Rev. B 78, 134404 (2008).
[157] C. Antoniak, J. Lindner, M. Spasova, D. Sudfeld, M. Acet, M. Farle, K. Fauth, U.
Wiedwald, H.-G. Boyen, P. Ziemann, F. Wilhelm, A. Rogalev, and S. Sun, Phys. Rev.
Lett. 97, 117201 (2006).
131
Bibliography
[158] S. Yin, R. Moro, X. Xu, and W. A. de Heer, Phys. Rev. Lett. 98, 113401 (2007).
[159] M. B. Knickelbein, Phys. Rev. B 75, 014401 (2007).
[160] R. M. Wang, O. Dmitrieva, M. Farle, G. Dumpich, H. Q. Ye, H. Poppa, R. Kilaas,
and C. Kisielowski, Phys. Rev. Lett. 100, 017205 (2008).
[161] M. E. Gruner, G. Rollmann, P. Entel, and M. Farle Phys. Rev. Lett. 100, 087203
(2008).
[162] Yan Sun, Min Zhang, and Rene Fournier, Phys. Rev. B 77, 075435 (2008).
[163] Yan Sun, Rene Fournier, and Min Zhang, Phys. Rev. A 79, 043202 (2009).
[164] F. Tournus, A. Tamion, N. Blanc, A. Hannour, L. Bardotti, B. Prevel, P. Ohresser,
E. Bonet, T. Epicier, and V. Dupuis, Phys. Rev. B 77, 144411 (2008).
[165] P. Villase nor-Gonzalez, J. Dorantes-Davila, H. Dreysse, and G. M. Pastor, Phys.
Rev. B 55, 15084 (1997).
[166] T. B. Massalski, J. L. Murray, L. H. Bennett, and H. Baker, Binary Alloy Phase
Diagrams, Vol. 1, (American Society for Metals, Metals Park, OH, 1986).
[167] M. E. Gruner, E. Homann, and P. Entel, Phys. Rev. B 67, 064415 (2003).
[168] G. Kresse and J. Furthm uller, Phys. Rev. B 54, 11169 (1996); G. Kresse and D.
Joubert, Phys. Rev. 59, 1758 (1999).
[169] J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, D. J.
Singh, and Carlos Fiolhais, Phys. Rev. B 46, 6671 (1992).
[170] P. E. Blochl, Phys. Rev. B, 50, 17953 (1994).
[171] Results for lower lying isomers and excited magnetic congurations will be reported
elsewhere.
[172] Y. Wang, T. F. George, D. M. Lindsay, and A. C. Beri, J. Chem. Phys. 86, 3493
(1987).
[173] The precise choice of the NN distances is not very important, since this concerns
merely the conguration for starting the unconstrained structural relaxation.
[174] G. Kresse and J urgen Furthm uller, VASP The Guide,
http://cms.mpi.univie.ac.at/vasp.
[175] R.F.W. Bader, Atoms in Molecules, A Quantum Theory (Oxford University Press,
Oxford, 1990).
[176] J.L. Ricardo-Chavez, PhD Thesis, Universite Paul Sabatier, Toulouse, France,
(2007).
[177] G. L. Gutsev, S. N. Khanna, and P. Jena, Phys. Rev. B 62, 1604 (2000); S. Chretien
and D. R. Salahub, Phys. Rev. B 66, 155425 (2002); G. Rollmann and P. Entel,
Computing Letters (CoLe), Vol. 1, No. 4, 288-296 (2004).
132
Bibliography
[178] M. Castro, and D. R. Salahub, Phys. Rev. B 49, 11842 (1994); L. Wang, Q Ge,
Chem. Phys. Lett. 366, 368-376 (2002); C. H. Chien, E. Blaisten-Barojas, and M.
R. Pederson, Phys. Rev. A 58, 2196 (1998); J. L. Chen, C. S. Wang, K. A. Jackson,
and M. R. Pederson, Phys. Rev. B 44, 6558 (1991). D. M. Cox, D. J. Trevor, R. L.
Whetten, E. A. Rohlng, and A. Kaldor, Phys. Rev. B 32, 7290 (1985).
[179] M. Castro, C. Jamorski, and D. R. Salahub, Chem. Phys. Lett. 271, 133 (1997); S.
E. Weber and P. Jena, Chem. Phys. Lett. 281, 401 (1997); B. V. Reddy, S. K. Nayak,
S. N. Khanna, B. K. Rao, and P. Jena, J. Phys. Chem. A 102, 1748 (1998).
[180] M. Moskovits and D. P. Dilella, J. Chem. Phys. 73, 4917 (1980).
[181] D. G. Leopold and W. C. Linenerger, J. Chem. Phys. 85, 51 (1986).
[182] B. V. Reddy, S. K. Nayak, S. N. Khanna, B. K.Rao, and P. Jena, Phys. Rev. B 59,
5214 (1999).
[183] T. Futschek, M. Marsman, and J. Hafner, J. Phys C. M. 17, 5927 (2005).
[184] B. Delley, J. Chem. Phys. 92, 508, (1990); 113, 7756 (2000).
[185] D. L. Cocke and K. A. Gingerich, J. Chem. Phys. 60, 1958 (1974).
[186] H. Wang, H. Haouari, R. Craig, Y. Lui, J. R. Lombardi, and D. M. Lindsay, J.
Chem. Phys., 106, 2101 (1997).
[187] J. D. Langenberg and M. D. Morse, J. Chem. Phys., 108, 2331 (1998).
[188] L. Pauling, J. Am. Chem. Soc. 54, 3570 (1932).
[189] L. Wang and Q. Ge, Chem. Phys. Lett. 366, 368-376 (2002).
[190] Y.-C. Bae, V. Kumar, H. Osanai, and Y. Kawazoe, Phys. Rev. B 70, 195413 (2004).
[191] P. Ballone and R. O. Jones, Chem. Phys. Lett. 233, 632 (1995).
[192] G. Rollmann, P. Entel, and S. Sahoo, Comput. Mater. Sci. 35, 275 (2005).
[193] O. Dieguez, M. M. G. Alemany, C. Rey, P. Ordejon, and L. J. Gallego, Phys. Rev.
B 63, 205407 (2001).
[194] G. Shirane and R. Nathans, Phys. Rev., 134, A1547-A1553 (1964).
[195] V. L. Moruzzi and P. M. Marcus, Phys. Rev. B 46, 2864 (1992).
[196] Y.-C. Bae, V. Kumar, H. Osanai, and Y. Kawazoe, Phys. Rev. B 72, 125427 (2005).
[197] M. Mu noz-Navia, J. Dorantes-Davila, D. Zitoun, C. Amiens, N. Jaouen, A. Rogalev,
M. Respaud, and G. M. Pastor, Appl. Phys. Lett. 95, 233107 (2009).
[198] S. Datta, M. Kabir, S. Ganguly, B. Sanyal, T. Saha-Dasgupta, and A. Mookerjee1,
Phys. Rev. B 76, 014429 (2007).
133
Bibliography
[199] D. A. Hales, C.-X Su, L. Lian, and P.B Armentrout, J. Chem. Phys.100, 1049
(1994).
[200] M. Castro, C. Jamorski and D.R. Salahub, Chem. Phys. Lett. 271 133 (1997).
[201] H. Yoshida, A. Terasaki, K. Kobayashi, M. Tsukada, and T. Kondow, J. Chem.
Phys.102, 5960 (1995).
[202] J. L. Rodriguez-Lopez, F. Aguilera-Granja, K. Michaelian, and A. Vega, Phys. Rev.
B. 67, 174413 (2003).
[203] R. P. Gupta, Phys. Rev. B 23, 6265 (1983).
[204] T. Shinohara, and T. Sato, Phys. Rev. Lett. 91, 197201 (2003).
[205] G. Gentefor, and W. Eberhardt, Phys. Rev. Lett. 76, 4975 (1996).
[206] T. Taniyamaand, E. Ohta, and T. Sato, Europhys. Lett. 38, 195 (1997).
[207] V. Kumar, and Y. Kawazoe, Phys. Rev. B 66, 144413 (2002).
[208] D. Dai and K. Balasubramanian, J. Chem. Phys. 103, 648 (1995).
[209] J. Dorantes-Davila, and G.M. Pastor, Phys. Rev. Lett. 81, 208 (1998).
[210] Y. Wu, J. Stohr, B. D. Hermsmeier, M. G. Samant, and D. Weller, Phys. Rev. Lett.
69, 2307-2310 (1992).
[211] J. Dorantes-Davila, H. Dreysse, and G.M. Pastor, Phys. Rev. Lett. 91, 197206
(2003).
[212] C. Barreteau, R. G. Lopez, D. Spanjaard, M. C. Desjonqeres, and A. M. Oles, Phys.
Rev. B 61 7781 (2000).
[213] G. Valerio, and H. Toulhoat, J. Phys. Chem. 100, 10827 (1996).
[214] S. S. Lin, B. Strauss, and A. Kant, J. Chem. Phys. 51, 2282 (1969).
[215] K. Balasubramanian, J. Chem. Phys. 91, 307 (1989).
[216] H. A. Jahn and E. Teller, Proc. R. Soc. A 161, 220, 1937.
[217] R. M. Bozorth, P. A. Wol, D. D. Davis, V. B. Compton, and J. H. Wernick, Phys.
Rev. 122, 1157-1160 (1961).
[218] F. Aguilera-Granja, A. Vega, J. Rogan, X. Andrade, and G. Garca, Phys. Rev. B
74, 224405 (2006).
[219] M. N Huda, M. N. Niranjan, B. R Sahu, and L. Kleinman, Phys. Rev. A 73, 053201
(2006).
[220] L. Fernandez-Seivane, and J. Ferrer, Phys. Rev. Lett. 99, 183401 (2007).
[221] P. Blonski, S. Dennler, and J. Hafner, J. Chekm. Phys. 134 034107 (2011).
134
Bibliography
[222] R. Gehrke and K. Reuter, Phys. Rev. B 79, 085412 (2009).
[223] J. P. K. Doye and D. J. Wales, New J. Chem.B 22, 733-744, 1998.
[224] J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, D. J.
Singh, and Carlos Fiolhais, Phys. Rev. B 46, 6671 (1992).
[225] G. Kresse, and O. Lebacq, VASP manual, http://cms.mpi.univie.ac.at/vasp/
[226] J. Jellinek, and E. B. Krissinel, Theory of Atomic and Molecular Clusters, Springer,
Berlin, 1999.
[227] G. H. O. Daalderop, P. J. Kelly, and M. F. H. Schuurmans, Phys. Rev. B 41, 11919
(1990).
[228] Y. Guo, W. M. Temmermann, and H. Ebert, J. Phys. Condens.Matter 3, 8205
(1991).
[229] S. Sahoo, A. Hucht, M. E. Gruner, G. Rollmann, P. Entel, A. Postnikov, J. Ferrer,
L. Fernandez-Seivane, M. Richter, D. Fritsch, S. Sil Phys. Rev. B 82, 054418 (2010).
[230] A. J. Cox, J. G. Louderback, and L. A. Bloomeld, Phys. Rev. Lett. 71, 923 (1993);
A. J. Cox, J. G. Louderback, S. E. Apsel, and L. A. Bloomeld, Phys. Rev. B 49,
12295 (1994).
[231] P. Boski and J. Hafner, Phys. Rev. B 79, 224418 (2009).
[232] D. Hobbs, G. Kresse, and J. Hafner, Phys. Rev. B 62, 11556, (2000).
[233] M. Marsman and J. Hafner, Phys. Rev. B 66, 224409, (2002).
[234] G. van der Laan, J. Phys. Condens. Matter 10, 3239 (1998).
[235] A. N. Andriotis and M. Menon, Phys. Rev. Lett. 93, 026402 (2004)
135
List of Figures
1.1 Mean saturation magnetization for ferromagnetic alloys . . . . . . . . . . . 12
1.2 Illustration of the transition from atomic to bulk behavior in Co. . . . . . 13
1.3 Experimentally determined size dependence of the magnetic moment per
atom of Rh
N
clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Schematic diagram of the Stern-Gerlach cluster beam experiment . . . . . . 15
1.5 Schematic representation of L edge X-ray absorption spectra . . . . . . . . 16
1.6 Photo-absorption spectra of 7.5 nm FeCo particles deposited on a Ni (111)
lm grown on W (110) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1 Denition of direction cosine . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 The fcc clusters with N = 13 and 19 atoms showing the zx plane in which
the magnetization angle is varied. . . . . . . . . . . . . . . . . . . . . . . 33
3.1 Schematic picture of the steepest descent scheme . . . . . . . . . . . . . . . 35
3.2 Two clusters with their corresponding adjacency matrices S
1
and S
2
. . . . 38
3.3 Structures for the cluster size N = 4. . . . . . . . . . . . . . . . . . . . . . 38
3.4 The illustration of the dierent chemical orders (also known as homotops)
of single- and double-impurity clusters with N = 5 atoms having square
pyramid topology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5 Illustration of principle of the basin-hopping method . . . . . . . . . . . . . 40
3.6 Mating between two parent structures generating a child . . . . . . . . . . . 41
4.1 The phase diagram of FeRh bulk alloy . . . . . . . . . . . . . . . . . . . . . 44
4.2 The optimal structure for the ground-state and lowest energy isomers (in-
dicated by an asterisk) of trimer and tetramer FeRh clusters. . . . . . . . . 50
4.3 Lowest energy isomers of FeRh pentamers and hexamers as in Fig. 4.2. . . 54
4.4 Constant magnetization density plots

(r)

(r) = and local moments

(in
B
) for the ground-state structures of FeRh hexamers. The value of the
constant magnetization density is given in
B
/

A
3
. . . . . . . . . . . . . . 56
4.5 Lowest energy isomers of FeRh heptamers and octamers. . . . . . . . . . . 59
4.6 Binding energy per atom E
B
of Fe
m
Rh
n
clusters as a function of the number
of Fe atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.7 Total magnetic moment per atom
N
of Fe
m
Rh
n
clusters as a function of
number of Fe atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.8 Local magnetic moment

at the Fe and Rh atoms as a function of the

number Fe atoms m. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.9 The magnetic stabilization energy of Fe
m
Rh
n
clusters as a function of num-
ber of Fe atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
136
List of Figures
4.10 Relative stability of Fe
m
Rh
n
clusters as a function of composition. . . . . . 66
4.11 Bond lengths for FeFe (cross), FeRh (circle) and RhRh (square) pairs as a
function of number of Fe atoms in Fe
m
Rh
n
clusters having N = 58 atoms. 67
4.12 Electronic density of states (DOS) of FeRh octamers . . . . . . . . . . . . . 68
5.1 The average magnetic moment per atom of Co
N
clusters as a function of
the cluster size N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2 Electronic density of states (DOS) of CoPd trimers . . . . . . . . . . . . . . 76
5.3 Lowest energy isomers of CoPd trimers and tetramers. The asteriks indicate
rst exited isomers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.4 Electronic density of states (DOS) of CoPd tetramers. . . . . . . . . . . . 78
5.5 Constant magnetization density plots

(r)

(r) = and local moments

(in
B
) for the ground-state structures of CoPd hexameters. . . . . . . . . . 80
5.6 Lowest energy isomers of CoPd pentamers and hexamers. The asteriks
indicate rst exited isomers. . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.7 Electronic density of states (DOS) of hexamers . . . . . . . . . . . . . . . . 83
5.8 Constant magnetization density plots

(r)

(r) = and local moments

(in
B
) for the ground-state structures of CoPd hexamers. The value of the
constant magnetization density is given in
B
/

A
3
. . . . . . . . . . . . . . 84
5.9 Lowest energy isomers of CoPd heptamers and octamers. Note that only
very few topologies have been considered as starting congurations of ge-
ometry relaxation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.10 Binding energy per atom E
B
of Co
m
Pd
n
clusters as a function of the number
of Co atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.11 Magnetic stabilization energy of Co Pd
m
Pd
n
clusters as a function of num-
ber of Co atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.12 Total magnetic moment per atom
N
of Co
m
Pd
n
clusters as a function of
number of Co atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.13 The relative stability of Co
m
Pd
n
clusters as a function of the number of Co
atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.14 Bond lengths for CoCo (cross), CoPd (circle) and PdPd (square) pairs as a
function of number of Co atoms in Co
m
Pd
n
clusters . . . . . . . . . . . . . 91
6.3 Illustration of successful, unsuccessful and high-energy trial moves in the
BH scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.4 Identied isomers of Fe
6
cluster in the energy range up to 1.74 eV above
the ground-state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.5 Normalized probabilities for the lowest-energy isomers of Fe
3
Rh
3
. . . . . . 99
6.6 The total energy of Fe
6
as a function of the number of BH steps n . . . . . 100
6.7 The binding energy per atom (in eV), the probability that trial moves gen-
erate the lowest energy isomers, and the total magnetic moment are plotted
as a function of isomer number for Fe
6
and Rh
6
cluster . . . . . . . . . . . . 101
6.8 Calculated isomers of Rh
6
cluster in the energy range up to 1.39 eV above
the ground-state energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.9 Binding energy per atom (in eV), the probability with which trial moves
generate the lowest energy isomers, and the total magnetic moment as a
function of isomer number for Fe
3
Rh
3
. . . . . . . . . . . . . . . . . . . . . . 105
137
List of Figures
6.10 The optimal structures of Fe
3
Rh
3
cluster. The entries below each structures
have the same meaning as in the Fig. 6.4 . . . . . . . . . . . . . . . . . . . 106
6.11 Obtained isomers of Fe
13
cluster in the energy range up to 3.13 eV above
the ground-state energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.12 Obtained isomers of Rh
13
cluster in the energy range up to 0.43 eV above
the ground-state energy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.13 Low-lying isomers of Fe
6
Rh
7
in the energy range up to 0.61 eV above the
ground-state energy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1 The distribution of Fe and Rh atoms in the FeRh clusters having N = 13
atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.2 The distribution of Fe and Rh atoms in the FeRh clusters having N = 19
atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.3 Spin and orbital moments in N = 13 and 19 FeRh clusters as a function of
Fe concentration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.4 Magnetic anisotropy energy in FeRh clusters having N = 13 and 19 atoms
as a function of Fe concentration. . . . . . . . . . . . . . . . . . . . . . . . 118
7.5 The energy dierence E = E() E(0) in meV/cluster vs of FeRh 19
atom clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
138
List of Tables
4.1 Structural, electronic and magnetic properties of FeRh dimers . . . . . . . . 47
4.2 Structural, electronic and magnetic properties of FeRh trimers . . . . . . . 49
4.3 Structural, electronic and magnetic properties of FeRh tetramers . . . . . . 51
4.4 Structural, electronic and magnetic properties of FeRh pentamers . . . . . . 52
4.5 Structural, electronic and magnetic properties of FeRh hexamers . . . . . . 55
4.6 Structural, electronic and magnetic properties of FeRh heptamers . . . . . . 58
4.7 Structural, electronic and magnetic properties of FeRh octamers . . . . . . 61
5.1 Structural, electronic and magnetic properties of CoPd dimers . . . . . . . 74
5.2 Structural, electronic and magnetic properties of CoPd trimers . . . . . . . 75
5.3 Structural, electronic and magnetic properties of CoPd tetramers . . . . . . 77
5.4 Structural, electronic and magnetic properties of CoPd pentamers . . . . . 79
5.5 Structural, electronic and magnetic properties of CoPd hexamers . . . . . . 82
5.6 Structural, electronic and magnetic properties of CoPd heptamers . . . . . 87
5.7 Structural, electronic and magnetic properties of CoPd octamers . . . . . . 88
7.1 Electronic and magnetic properties of FeRh clusters having N = 13 atoms . 117
7.2 Electronic and magnetic properties of FeRh clusters having N = 19 atoms . 119
139
Abbreviations and symbols

B
Bohr Magneton

A Amstrong
APW Augmented plane wave
BCC Body centered cubic
BCO Bicapped octahedra
BFGS Broyden Fletcher Goldfarb Shanno
BH Basin Hopping
BO Born Oppenheimer
CC Coupled Cluster
CG Conjugate gradient
CI Conguration Interaction
CO Capped octahedra
CSP Capped square pyramid
DFT Density functional theory
DOS Density of states
FCC Face centered cubic
GA Genetic Algorithm
GGA Generalized gradient approximation
GO Global optimization
HCP Hexagonal close packed
HF Hartree-Fock
HOMO Highest occupied molecular orbital
KS Kohn-Sham
140
LDA Local density approximation
LDOS Local density of states
LO Local optimization
LUMO Lowest unoccupied molecular orbital
MAE Magnetic anisotropy energy
MC Monte Carlo
MFE Mean rst encounter
MP Mller Plesset
N
m
Number of BH moves
PAW Projector augmented wave
PBP Pentagonal bipyramid
QSE Quantum size eects
SA Simulated annealing
SD Steepest descent
SG Stern-Gerlach
SOC Spin orbit coupling
SP Slater Pauling
SP Square pyramid
TF Thomas and Fermi
TM Transition metal
XC exchange-correlation
XMCD X-ray magnetic circular dichroism
141
Veroentlichungen - List of publications
First-principles study of structural, magnetic and electronic properties of small Fe-
Rh alloy clusters, Junais Habeeb Mokkath and G. M. Pastor, Phys. Rev. B 85,
054407 (2012) and arXiv:1201.5971 Materials Science (cond-mat.mtrl-sci)
First-principles study of magnetism, structure and chemical order in small FeRh
alloy clusters, Junais Habeeb Mokkath and G. M. Pastor, arXiv:1110.2669 Atomic
and Molecular Clusters (physics.atm-clus)
Magnetism, structure, and chemical order in small CoPd alloy clusters, Junais
Habeeb Mokkath, and G. M. Pastor. (To be submitted to JPCC)
First principles spin-polarized basin-hopping method, L. Diaz, Junais Habeeb Mokkath,
and G. M. Pastor. (To be sumbitted to PRB)
In addition, results of this research were presented at the following conferences:
15th WIEN2k-WORKSHOP international conference (March 26-29, 2008), Vienna
University of Technology, Austria.
Symposium on Size Selected Clusters (March 8-13, 2009),
Brand, Austria (poster).
DPG Spring Meeting, (March 21-26, 2010),
Regensburg, Germany (poster and talk).
Seminar talk at the Department of Physics of University of Kassel
(April 15, 2009).
Symposium on Nanoscale alloys: from experiment and theory
to quantitative modelling (September 13-16, 2010),
CECAM-HQ-EPFL, Lausanne, Switzerland (poster and talk).
Symposium on Size Selected Clusters (March 20-25, 2011),
Davos, Switzerland (poster).
Cluster treen 2011(November 25-30, 2011)
Burg Rothenfels, Wuerzburg, Germany (talk).
142