08.11.2014 Views

Diploma - Max Planck Institute for Solid State Research

Diploma - Max Planck Institute for Solid State Research

Diploma - Max Planck Institute for Solid State Research

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

– Photoemission on EuRh 2 Si 2 –<br />

disentanglement of surface and bulk<br />

structures<br />

<strong>Diploma</strong>rbeit<br />

zur Erlangung des akademischen Grades<br />

Diplom-Physiker<br />

vorgelegt von<br />

Marc Höppner<br />

geboren am 16.11.1986 in Großröhrsdorf<br />

Institut für Festkörperphysik<br />

Fachrichtung Physik<br />

Fakultät Mathematik und Naturwissenschaften<br />

Technische Universität Dresden<br />

2011


The Wannier function depicted on the cover page is<br />

mainly based on Europium 4f z(x 2 −y 2 ) and Rhodium<br />

4d z 2 illustrating a likely hybrid orbital in EuRh 2 Si 2 .<br />

It is the Fourier trans<strong>for</strong>med Bloch band with Europium<br />

4f z(x 2 −y 2 ) as its major contribution. Fractions<br />

based on sites in neighbouring unit cells have<br />

been omitted.<br />

1. Gutachter: Prof. Dr. C. Laubschat<br />

2. Gutachter: Dr. Helge Rosner<br />

Datum des Einreichens der Arbeit: 24. November 2011


Abstract<br />

A thorough knowledge of correlated electron systems is indispensable to describe materials’<br />

properties like superconductivity or heavy-fermion behaviour. Owing to that,<br />

the interaction of localized europium 4f electrons with itinerant rhodium 4d valence<br />

electrons in EuRh 2 Si 2 is discussed in this thesis. The results of angular resolved photoemission<br />

on Si and Eu terminated surfaces are compared to calculated band structures<br />

based on density functional theory. In doing so, the emphasize is laid on the<br />

treatment of the Eu 4f electrons in the calculation. The photoemission spectra of both<br />

surface terminations are reproduced by means of a simple hybridization model. Using<br />

a projection-based method to create Wannier functions, the interaction of the Eu 4f<br />

electrons with the valence band can basically be related to Rh 4d electrons. Moreover,<br />

<strong>for</strong> the first time a quasi-linear band originated at the surface is described, which could<br />

manifest similar properties like that in graphene [Varykhalov et. al., Phys. Rev. Lett.<br />

101:157601 (2008)] or Bi 2 Se 3 [Xia et. al., Nature Physics 06, 398-402 (2009)]. By the<br />

interplay of the Dirac-like band and the 4f states various new material properties are<br />

conceivable.<br />

Kurzfassung<br />

Um Materialeigenschaften wie Supraleitung und Schwere-Fermion Verhalten erklären<br />

zu können, ist ein grundlegendes Verständnis korrelierter elektronischer Systeme unerlässlich.<br />

In dieser Arbeit wird die Wechselwirkung der lokalisierten Europium 4f Elektronen<br />

mit den itineranten Rhodium 4d Valenzbandelektronen in EuRh 2 Si 2 untersucht.<br />

Dabei werden winkelaufgelöste Photoemissionsmessungen an Si- als auch Euterminierten<br />

Oberflächen mit Bandstrukturen verglichen, welche auf Dichtefunktionalrechnungen<br />

basieren. Hierbei wird insbesondere auf die theoretische Beschreibung<br />

der lokalisierter Elektronen und ihrer Wechselwirkung mit dem Valenzband eingegangen.<br />

Für beide Oberflächenterminierungen können die Photoemissionsspektren mit Hilfe<br />

eines einfachen Hybridisierungsmodells zufriedenstellend reproduziert werden. Durch<br />

Projektion auf Wannierorbitale kann die Wechselwirkung der Eu 4f Elektronen mit dem<br />

Valenzband im Wesentlichen den Rh 4d Elektronen zugeordnet werden. Zudem wird erstmals<br />

ein quasi-lineares Oberflächenband beschrieben, welches ähnliche Eigenschaften<br />

aufweisen könnte wie der Dirac Cone in Graphen [Varykhalov et. al., Phys. Rev. Lett.<br />

101:157601 (2008)] oder Bi 2 Se 3 [Xia et. al., Nature Physics 06, 398-402 (2009)]. Die<br />

Wechselwirkung zwischen diesem und den 4f Zuständen könnte neue, faszinierende Materialeigenschaften<br />

ermöglichen.


v<br />

Contents<br />

1 Introduction 1<br />

2 Theoretical foundation 3<br />

2.1 Band Structure Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />

2.1.1 Density Functional Theory . . . . . . . . . . . . . . . . . . . . . 5<br />

2.1.2 Exchange-correlation Functionals . . . . . . . . . . . . . . . . . . 7<br />

2.1.3 Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8<br />

2.2 Photoemission Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11<br />

2.2.1 Three-step model . . . . . . . . . . . . . . . . . . . . . . . . . . . 12<br />

2.2.2 One-step model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14<br />

2.2.3 General remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 15<br />

3 Experimental foundations 17<br />

3.1 Photoemission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17<br />

3.2 General setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18<br />

3.3 BESSYII: 1 3 ARPES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20<br />

3.4 SLS: SIS-HRPES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20<br />

4 EuRh 2 Si 2 – semi-localized electrons 23<br />

4.1 Overview – properties and classification . . . . . . . . . . . . . . . . . . 23<br />

4.1.1 Brillouin zone and computational setups . . . . . . . . . . . . . . 24<br />

4.1.2 Treatment of strongly localized electrons beyond L(S)DA . . . . 27<br />

4.1.3 Cleavage behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . 29<br />

4.1.4 Surface and bulk band structure . . . . . . . . . . . . . . . . . . 33<br />

4.2 Hybridization: localized versus itinerant states . . . . . . . . . . . . . . 41<br />

4.2.1 Symmetry considerations . . . . . . . . . . . . . . . . . . . . . . 41<br />

4.2.2 Estimation of the hybridization strength . . . . . . . . . . . . . . 43<br />

4.2.3 The hybridization model . . . . . . . . . . . . . . . . . . . . . . . 48<br />

4.3 Perspective: quasi-linear dispersion in a 4f compound . . . . . . . . . . . 53<br />

5 Summary 57<br />

Bibliography 59<br />

List of Figures 65<br />

List of Tables 67


vii<br />

Nomenclature<br />

AF<br />

APW<br />

ARPES<br />

BO<br />

BZ<br />

CI<br />

DFT<br />

FPLO<br />

GGA<br />

GS<br />

HF<br />

HK1<br />

HK2<br />

LDA<br />

LEED<br />

LMTO-ASA<br />

antiferromagnetic<br />

augmented plane waves<br />

angular resolved photoemission spectroscopy<br />

Born-Oppenheimer approximation<br />

Brillouin Zone<br />

configuration interaction<br />

density functional theory<br />

Full Potential Local Orbital code developed at IFW Dresden<br />

generalized gradient approximation<br />

ground state<br />

Heavy Fermion<br />

first theorem of Hohenberg and Kohn<br />

second theorem of Hohenberg and Kohn<br />

local density approximation<br />

low electron energy diffraction<br />

Linear Muffin Tin Orbital – Atomic Sphere Approximation Code mainly<br />

developed/mainted by A. Perlov & A. Yaresko<br />

LSDA+U [AL] local spin density approximation plus correlation correction in the atomic<br />

limit<br />

PE<br />

PES<br />

SE<br />

SPG<br />

SS<br />

UHV<br />

VB<br />

WF<br />

photoemission<br />

photoemission spectroscopy<br />

Schrödinger equation<br />

space group<br />

surface state<br />

ultra-high vacuum<br />

valence band<br />

Wannier function


1<br />

1 Introduction<br />

Materials research is one of the basic topics of science being the foundation <strong>for</strong> modern<br />

and innovative applications ranging from lightweight materials <strong>for</strong> automotive engineering<br />

over superconductors <strong>for</strong> dissipationless current transport up to low-dimensional<br />

systems <strong>for</strong> new nanodevices and computational systems. Thereby electronic properties<br />

play a crucial role and the interplay between different microscopic phenomena (e.g.<br />

electron-phonon interaction) strongly determines the macroscopic properties. The interaction<br />

of localized electrons in a bath of itinerant ones belongs to the same class and<br />

is addressed in this diploma thesis.<br />

Typical examples of localized electrons are the 4f states of rare earth (RE) elements,<br />

which are energetically in the range of the valence electrons, but spatially localized near<br />

the nuclei. In a solid, the overlap of neighbouring atoms’ 4f orbitals can be neglected and<br />

hence they almost do not contribute to chemical bonding conserving in this way many of<br />

their atomic properties, among them particularly their magnetic moments. Depending<br />

on symmetry, interactions with itinerant valence states are possible that may lead to<br />

a spin-polarization of conduction states and thereby to an indirect coupling of neighbouring<br />

4f moments and, thus, to magnetic ordering (Ruderman-Kittel-Kasuya-Yosida<br />

interaction). With increasing interaction strength conduction electrons’ polarization<br />

may lead to a screening of the magnetic moments known as Kondo effect [1], and a periodic<br />

arrangement of Kondo impurities causes heavy-fermion behavior [2] characterized<br />

by an increase of the effective mass of the charge carriers up to a factor of thousand.<br />

In the recent past rare earth transition metal silicides and pnictides have attracted<br />

considerable interest due to their exotic electronic properties ranging from different<br />

types of magnetic order [3, 4] to heavy-fermions’ properties [5] and even superconductivity<br />

[6]. At so-called quantum critical points of the ternary phase diagrams the<br />

individual phases may be degenerate with each other, and in certain regions coexistence<br />

of competing interactions like magnetic order and superconductivity may be found, that<br />

lead to instabilities of the electronic properties being interesting <strong>for</strong> applications in spintronics.<br />

To get an insight into the electronic properties of these compounds, angle-resolved<br />

photoelectron spectroscopy (ARPES) which provides an image of band structures and<br />

Fermi surfaces is the method of choice. The localized 4f states are hereby reflected<br />

by broad final-state multiplets and interactions with the valence bands by additional<br />

energy splittings and dispersions. From a theoretical point of view the description of<br />

these data and the proper treatment of the 4f states is a challenging subject. Due to the


2 1 Introduction<br />

conflicting limits of itinerant and localized electrons, a feasible, general solution is still<br />

missing and there<strong>for</strong>e the dominating energy scales <strong>for</strong> each material’s type determine<br />

the applicable approximations. Since photoelectron spectroscopy represents a surface<br />

sensitive technique, surface preparation is a very delicate question. The present thesis<br />

deals with ternary silicides of the type RE Rh 2 Si 2 which are available in <strong>for</strong>m of rather<br />

large single crystalls, that reveal a layered structure and may, thus, easily be cleaved<br />

under ulta-high vacuum conditions leading to clean and structurally ordered surfaces.<br />

While YbRh 2 Si 2 is a well-known heavy-fermion system which is close to a quantum<br />

critical point and has been studied extensively in the recent past [7–9], EuRh 2 Si 2 is<br />

a stable divalent antiferromagnet with almost unknown properties. The intention of<br />

this diploma thesis is to present a detailed analysis of PE results of EuRh 2 Si 2 considering<br />

both, surface and bulk, contributions which is mandatory because of the surface<br />

sensitivity [10]. In addition, the interplay between the 4f electrons and the itinerant<br />

conduction electrons is explored with the attempt of disentangling ground state’s from<br />

excited state’s properties. A similar analysis has been prepared <strong>for</strong> YbRh 2 Si 2 indicating<br />

that the <strong>for</strong>mer are at least to some extent accessible by photoemission [11]. In order<br />

to study the interaction of the two electron species, the PE spectra are simulated using<br />

ab initio density functional theory (DFT) calculations and a simple hybridization model<br />

whereat the main challenge is the accurate treatment of the localized 4f electrons within<br />

these methods.<br />

The present thesis is organized as follows: at first, the key concepts of DFT and the<br />

primarily-used methods are presented followed by a sketch of the PE process and a<br />

description of the utilized experimental stations with their special properties. Subsequently,<br />

EuRh 2 Si 2 is briefly introduced including an overview of the crystal structure,<br />

the band structure, possible cleavage planes and different computational setups. The<br />

analysis <strong>for</strong> different surface configurations and a disentanglement of surface and bulk<br />

contributions followed by a detailed discussion of the coupling as well as the simulation<br />

of the PE spectra is presented afterwards. Finally, an outlook regarding peculiar surface<br />

states and a summary are given.


3<br />

2 Theoretical foundation<br />

2.1 Band Structure Theory<br />

Matter consists of atoms assembled of a nucleus and electrons whereat the latter are<br />

responsible <strong>for</strong> bonding between atoms governing the state of aggregation. To understand<br />

the manifold properties of solids one thus has to find a solution of a manybody<br />

Hamiltonian (in this case: only Coulomb interaction, disregarding relativistic effects)<br />

H = T e + V ee + V ne + T n + V nn (2.1)<br />

in which T e (T n ) is the kinetic energy of the electrons (nuclei) and V ee plus V nn are the<br />

electron and nuclei Coulomb repulsion terms, respectively. Both subsystems are coupled<br />

via the Coulomb interaction operator V ne . Solving the time-independent Schrödinger<br />

equation <strong>for</strong> this Hamiltonian is unfeasible (besides some simple or highly symmetric<br />

molecules). Hence a simplification is needed <strong>for</strong> complex molecules and especially <strong>for</strong><br />

solids.<br />

Note: all further <strong>for</strong>mulae in this section are presented in atomic units ( = m e =<br />

e = 4πɛ 0 = 1). An overview of used variables is given in tab. 2.1.<br />

Assuming that the momenta of ( the nuclei ) and electrons are in the same order of<br />

magnitude, their kinetic energies ∼ p2<br />

2m<br />

should differ in at least three orders. Theresymbol<br />

explanation<br />

A operator A in Dirac notation<br />

A s/m spatial / momentum representation of A<br />

e.g. SOMETHING missing<br />

W operator representation of the Born-Oppenheimer surface<br />

A n,e,ne A depends on nuclei, electrons or both of them<br />

N e/n number of electrons / nuclei<br />

r tuple {r 1 , . . . , r Ne } of the electrons’ spatial coordinates<br />

R tuple {R 1 , . . . , R Nn } of the nuclei spatial coordinates<br />

µ variational parameter<br />

B, s Bravais vector, basis vector of the unit cell respectively<br />

Table 2.1: signs and symbols <strong>for</strong> chapter 2.1


4 2 Theoretical foundation<br />

<strong>for</strong>e the electronic (represented by ψ(r, R)) and the nuclear part (represented by ψ(R))<br />

decouple using a product ansatz Ψ (r, R) = φ (R) ψ (r, R):<br />

H s e<br />

{ }} {<br />

(Te<br />

s + Vee s + Vne) s ψ (r, R) = E e (R) · ψ (r, R) (2.2)<br />

(T s<br />

n + W s (R)) φ (R) = E ne · φ (R) (2.3)<br />

with<br />

W s (R) = E e (R) + V nn (R)<br />

This procedure is called Born-Oppenheimer approximation (BO) [12]. Equation 2.2 is<br />

referred to as electronic Schrödinger equation (SE) because the coordinates R of the<br />

nuclei can be regarded as parameters since the spatial representation (of the operators)<br />

does not contain a differential operator with respect to them. Eq. 2.2 can only be solved<br />

approximately (e.g. <strong>for</strong> small molecules / compounds: Configuration Interaction; <strong>for</strong><br />

molecules as well as solids: Hartree-Fock, Thomas Fermi theory, density functional<br />

theory (DFT); all mentioned algorithms are solved iteratively) despite the analytical<br />

results <strong>for</strong> H and H + 2 . The solution of eq. 2.3 depends on the knowledge of the highlydimensional<br />

Born-Oppenheimer surface and can just be solved <strong>for</strong> rather small or vastlysymmetric<br />

molecules (cf. C 60 ).<br />

Assuming that there is no interest in vibrational or rotational modes of the nuclei (i.e.<br />

phonons in crystals), their positions can be locked and eq. 2.3 could be neglected. In<br />

principle the statement is even stronger since the time frame <strong>for</strong> the motion of electrons<br />

is considerably smaller than the one <strong>for</strong> the nuclei.<br />

The solution of eq. 2.2 allows to deduce fundamental electronic and optical properties<br />

of crystalline solids (e.g. if the solid is an insulator, a semiconductor or a metal)<br />

by analyzing the dispersion of the eigenvalues (the so-called band structure) in the<br />

reciprocal space (also called momentum space, k-space). The Wigner-Seitz cell (smallest<br />

unit cell of the compound) trans<strong>for</strong>med into the momentum space is called first Brillouin<br />

Zone (BZ). The periodicity (originated in translational invariance of real space) is an<br />

intrisic property of the momentum space introducing two different models: the reduced<br />

and the extended BZ scheme. The latter is the reduction to the first BZ disregarding<br />

the absolute value of the momentum (being defined as recurrent with period 2π a , a being<br />

the lattice constant) which is – in most cases – sufficient. The <strong>for</strong>mer scheme is more<br />

suitable <strong>for</strong> evaluating photoemission spectra because in this case one needs momentum<br />

conservation.<br />

Besides the a<strong>for</strong>e mentioned iterative solutions (like DFT), there exists a second class<br />

of methods <strong>for</strong> calculating the band structure of solids employing either pertubation<br />

theory or the superposition of atomic solutions (e.g. k · p perturbation theory or tight<br />

binding).


2.1 Band Structure Theory 5<br />

Figure 2.1: DFT scheme – the GS density can be calculated iteratively with eq. 2.9 and 2.10<br />

because the Hamiltonian, especially the effective potential, is solely determined by the density<br />

of the previous iteration step (HK1)<br />

The approach of density functional theory to solve eq 2.2 will be discussed in detail<br />

below using explicitly the symmetry of crystals. Furthermore, there are also schemes<br />

<strong>for</strong> disordered materials[13].<br />

2.1.1 Density Functional Theory<br />

Hohenberg and Kohn [14] demonstrated that the external potential Ṽext ≡ 〈 ψ | V ne | ψ 〉<br />

is solely defined by the ground state (GS) density ρ (HK1) as well as that Ṽext is determined<br />

by an universial functional F HK [ρ] which does not depend on the external<br />

potential V ext (HK2). Considering the expectation value of H e with a product wavefunction<br />

ansatz <strong>for</strong> non-interacting electrons eq. 2.2 trans<strong>for</strong>ms into<br />

E e [ρ] =<br />

(<br />

)<br />

T e + V ee + Ṽext [ρ] (2.4)<br />

Thereby one loses the exact solution of the electronic manybody problem due to the<br />

approximation of a product wavefunction.<br />

The variation of 2.4 under the constraint of charge conservation ∫ ∞<br />

−∞ ρ d3 r = N e yields<br />

µ = δE e [ρ]<br />

δρ (r) = ṽ ext (r) + δF HK [ρ]<br />

δρ (r)<br />

with<br />

F HK [ρ] = T e [ρ] + V ee [ρ]<br />

(2.5)<br />

which is regarded as the basic equation in DFT. Since the exact solution of H e is<br />

approximated in eq. 2.5, one tried to incorporate the manybody phenomena into the<br />

Hamiltonian H e . In particular, Kohn and Sham have shown that there always exists<br />

a system of non-interacting electrons which has the same density as the system of<br />

interacting electrons [15]. They constructed F HK as a sum of a non-interacting electron


6 2 Theoretical foundation<br />

system and an exchange-correlation energy functional which represents the difference<br />

between the real system and the approximation of a non-interacting system<br />

F HK [ρ] = T e [ρ] + V ee [ρ] + E xc [ρ] (2.6)<br />

with<br />

E xc [ρ] = T [ρ] − T e [ρ] + V [ρ] − V ee [ρ] (2.7)<br />

leading to a rescaled effective potential v eff in eq. 2.5:<br />

v eff [ρ (r)] = ṽ ext (r) + v xc [ρ (r)] (2.8)<br />

Processing this scheme one approaches the following set of equations (in spatial representation):<br />

(<br />

− 1 2<br />

3∑<br />

i<br />

∂ 2<br />

∂r 2 i<br />

+ v eff [ρ (r)]<br />

)<br />

ψ k (r) = ɛ k ψ k (r) (2.9)<br />

with ρ (r) =<br />

N occupied<br />

∑<br />

k<br />

|ψ k (r)| 2 (2.10)<br />

The first equation is not equivalent to a one particle Schrödinger equation since the<br />

latter is a linear differential equation (linear in the algebraic sense, cf. linear operator)<br />

whereat eq. 2.9 is non-linear because the potential depends on the density which<br />

infact results from the wave functions. There<strong>for</strong>e the density has to be computed selfconsistently<br />

by choosing a start density ρ 0 (e.g. a spatial homogeneous one or using<br />

ρ n from the previous run) and evaluating the potential v eff at the given density to get<br />

eq. 2.9 with the eigenvalues ɛk 0 and wave functions ψ k 0 (r) as its solution. Subsequently<br />

ρ 1 is calculated utilizing eq. 2.10. After such an iteration the convergence is checked<br />

either by comparing the densities ρ n and ρ n+1 or the total energies whereat the convergence<br />

criteria define the computional ef<strong>for</strong>t (cf. flow chart in fig. 2.1). In addition, to<br />

determine the effective potential an exchange-correlation functional (see ch. 2.1.2) has<br />

to be chosen. For a more detailed (and mathematically-emphasized) review I refer to<br />

[16].<br />

Since here only crystalline solids are regarded, the successive paragraphs are restricted<br />

to them. Being described as periodically continuous in all three spatial dimensions, the<br />

external potential (originated by the fixed nuclei) must have the same periodicity. The<br />

eigenvectors of such a periodic Hamiltonian are Bloch states<br />

| kn 〉 = ∑ Bsµ<br />

c kn<br />

sµ | Bsµ 〉e ik(B+s) (2.11)


2.1 Band Structure Theory 7<br />

respecting the crystal symmetry (B is a Bravais vector and s a vector of the basis<br />

pointing to a Wyckoff position). Thereby the problem is confined to a primitive unit<br />

cell with periodic boundary conditions because the Hamiltonian commutes with the<br />

translation operators of the lattice. Furthermore the amount of unique sites is reduced<br />

by point symmetry. Whether the Bloch Ansatz is a result of the self-consistency cycle<br />

or used <strong>for</strong> the definition of the basis set depends on the chosen scheme (cf. 2.1.3),<br />

nevertheless its appearance in the self-consistent eigenstates arises from the translational<br />

symmetry of the crystal.<br />

2.1.2 Exchange-correlation Functionals<br />

Since there is no general scheme to obtain a universal functional E xc which achieves<br />

highly-accurate results <strong>for</strong> all input configurations it is necessary to choose a wellbalanced<br />

approximation. In principle, the exchange-correlation functional can be expressed<br />

as<br />

E xc [ρ] = 1 2<br />

∫<br />

∫<br />

d 3 r 1 ρ (r 1 )<br />

d 3 r 2<br />

1<br />

|r 1 − r 2 | ɛ xc [ρ, ∇ρ, . . .] (r 1 , r 2 )<br />

where the exchange-correlation density ɛ xc depends on both spatial coordinates. Most<br />

functionals used correspond to one of these three classes [17, p. 479–481]:<br />

(a) Local Density Approximation (LDA) type functionals are most widely-used<br />

since they are simple, fast and yield good results <strong>for</strong> systems whose electrons are<br />

itinerant.<br />

The exchange correlation density (whose expectation value is the exchange<br />

correlation energy) depends only on the density at the same spatial position<br />

as the density evaluated <strong>for</strong> the expectation value. Thus it is a local correction.<br />

(b) Generalized Gradient Approximation (GGA) are based on the LDA with<br />

higher expansion terms (∇ρ . . .), there<strong>for</strong>e in general the lattice constants and total<br />

energy are typically better than obtained by LDA [? ]. But in general it is not possible<br />

to decide whether LDA or GGA is more sufficient because by the construction<br />

principle the absolute value of the remainder depends on the compound [18, 19].<br />

(c) Hybrid Functionals are constructed by empirical fits between HF (exact exchange),<br />

exchange as well as correlation energies of LDA and GGA [20]. Due to the<br />

portion of exact exchange the description of band gaps in semiconductors is better<br />

than in LDA/GGA but the computational ef<strong>for</strong>t exceeds that of the others.<br />

For all calculations presented in this diploma thesis an LDA-type (spin-dependent: LSDA)<br />

functional [21] was used although it has been proven to be not accurate <strong>for</strong> rather localized<br />

electrons (d or f electrons). The main drawback is, that the assumption of a<br />

slowly-varying charge density is not fulfilled anymore.<br />

If one would try to calculate


8 2 Theoretical foundation<br />

such a system anyway (<strong>for</strong> example as a zeroth order approximation), one has to challenge<br />

additionally the convergence instability of the Fermi level determination process,<br />

since the dispersion of the localized states is weak and there<strong>for</strong>e a small rearrangement<br />

of the Fermi level changes the occupation number strongly. A possible solution is<br />

to modify the functional so that it contains some correction to the correlation energy<br />

(e.g. L(S)DA+U). To circumvent this issue the localized electrons have been treated as<br />

“core” electrons 1 (so-called open core approximation) neglecting the overlap from different<br />

sites. This approximation is legitimate, because their magnitude of localization<br />

is comparable to orbitals treated as “core” electrons, but their single-particle energy is<br />

considerably higher. Moreover, in ch. 4.1 it is shown, that the Fermi level obtained<br />

by this method is comparable to L(S)DA+U results with respect to the valence band<br />

structure.<br />

2.1.3 Codes<br />

There are a lot of DFT codes available with various approximations depending on the<br />

implemented basis set (a few implementations of the respecting methods are given at<br />

the end of each block). In general, a distinction[17, p. 233–235] can be drawn between<br />

(a) Plane wave methods present the most general way <strong>for</strong> solving differential equations.<br />

It is easy to implement them <strong>for</strong> computation and since being the solution of<br />

the Schrödinger equation with constant potential, they are an effective basis <strong>for</strong> the<br />

nearly-free electron model (covering the crystal potential as a small perturbation)<br />

there<strong>for</strong>e one gets a valuable insight to the bandstructure of sp-metals and semiconductors.<br />

The disadvantage is enclosed in the potential representation because plane<br />

waves demand a smooth potential whereas the Coulomb potential has a singularity.<br />

Hence, those methods are often accompanied by pseudopotentials (smoothed<br />

potentials, nucleus and core electrons are combined) or grids.<br />

(e.g. Abinit [http://www.abinit.org], VASP [http://cms.mpi.univie.ac.at/vasp], Quantum-<br />

Expresso [http://www.quantum-espresso.org], CPMD [http://www.cpmd.org], ...)<br />

(b) Choosing localized (atomic-like) orbitals as a basis set respects automatically<br />

the symmetry in the vicinity of the atomic sites. As basis functions of the Bloch<br />

states are usually selected Gaussians, or numerically adjusted atomic-like orbitals<br />

(demands Bloch Theorem, cf. (2.11), (2.12)). The advantage of being able to use<br />

the bare Coulomb potential (superposition of the atomic Coulomb potentials) is<br />

gaining high accuracy <strong>for</strong> heavier elements as well as having a smaller basis set<br />

compared to plane wave methods. Since atomic orbitals from different sites are not<br />

1 Further on, this approximation will be called “open core approximation”, in literature also frozen core<br />

or quasi-core, because we deal with not fully-occupied “core” electrons. Since it is not a stable noble<br />

gas configuration, the occupation number should in principle be determined variationally. Given<br />

that the overlap of the localized orbitals is small, one can set a fixed occupation number as an initial<br />

parameter according to the experiment


2.1 Band Structure Theory 9<br />

orthogonal, all multicenter-integrals of the basis set have to be computed. There are<br />

also various approximate, non-DFT solutions possible e.g. tight-binding[22] which<br />

are mainly used to estimate parameters of model Hamiltonians (e.g.<br />

model, Hubbard model) <strong>for</strong> comparison with real compounds.<br />

Anderson<br />

(e.g. FPLO [http://www.fplo.de], Gaussian [http://www.gaussian.com], Siesta [http://www.<br />

icmab.es/siesta/], Crystal [http://www.cse.scitech.ac.uk/cmg/CRYSTAL/], ...)<br />

(c) Atomic sphere methods are the natural approach to adopt the basis set to<br />

the given problem dividing the arrangement of atoms into atomic sphere-like (centered<br />

around the sites) and interstitial parts. The potential in the <strong>for</strong>mer is similar<br />

to the atomic potential, whereat in the latter case it is smooth suggesting<br />

an augmented basis set consisting of localized functions with boundary conditions<br />

satisfying smoothly varying functions in the interstitial region (so-called APWs -<br />

Augmented Plane Waves). Adversely, this results in non-linear equations 2 which<br />

solutions are demanding. There<strong>for</strong>e one introduced a linerization[23] around fixed<br />

energy values (eg. LAPW, ...) receiving the most accurate method today.<br />

(e.g. fleur [http://www.flapw.de], Wien2k [http://www.wien2k.at/], elk/exciting [http://exciting.<br />

source<strong>for</strong>ge.net], Stuttgart LMTO [http://www.fkf.mpg.de/andersen/docs/manual.html], ...)<br />

The following codes have already been used successfully in our group <strong>for</strong> LDA calculations<br />

of Heavy Fermion (HF) and mixed-valent compounds in the past (and were<br />

used <strong>for</strong> all calculations in this diploma thesis) but this does not imply that they are<br />

the most suitable ones.<br />

1. Full Potential Local Orbital code (FPLO) [24]<br />

FPLO uses a nonorthogonal local-orbital basis set | Bsµ 〉 (cf. 2.11) whose orbitals<br />

are the solution of a Schrödinger Equation with a spherically-averaged crystal potential<br />

and a limiting potential part v lim =<br />

(<br />

r<br />

r 0<br />

) 4.<br />

The latter ensures a minimized<br />

basis set 3 since otherwise the amount of atomic-like basis functions needed <strong>for</strong><br />

a sufficient expansion of weakly-bound extended states would be severely larger.<br />

Another option is a basis set extension by plane waves but the complexity and thus<br />

the additional computational ef<strong>for</strong>t is in no relation to the gained accuracy. The<br />

basis orbitals <strong>for</strong> which the differences between the real crystal and the sphericalaveraged<br />

potential is not perceptible, are called core orbitals – and their overlap<br />

is defined as zero. All remaining orbitals are treated as valence orbitals. Nevertheless,<br />

the overlap between core orbitals and valence orbitals from different sites<br />

is regarded. Due to the distinction between core and valence orbitals (or even<br />

2 this does not influence the basic linearity of the SE, but due to the energy-dependence of the basis<br />

functions one is not able to solve the equations <strong>for</strong> all eigenenergies at one time<br />

3 basis functions in the vicinity of a weak potential influence get compressed compared to the same<br />

eigenfunctions in an unmodified potential, there<strong>for</strong>e the compressed ones are more suitable to describe<br />

weakly bound / unbound states


10 2 Theoretical foundation<br />

Figure 2.2: muffin tin with pastries: the potential in LMTO is similar to a muffin tin devided<br />

into two parts – a spherical symmetric Coulomb-like potential around the atomic sites and a<br />

constant interstitial region. In difference to the general case of atomic sphere methods which<br />

are based on fragmented potentials as well, the potential is not continuously differentiable, but<br />

the basis functions can be defined in a more convenient manner.<br />

core, semi-core and valence orbitals) the matrix equation 2.12 gets simplyfied –<br />

whereat this classification is solely artificial governed by the required accuracy.<br />

Inserting the Bloch ansatz into eq. 2.9 projected onto a Kohn-Sham orbitals yields<br />

the secular equation<br />

⎛<br />

⎞<br />

∑<br />

⎜<br />

⎝〈 0s ′ µ ′ | H | Bsµ 〉 − 〈 0s ′ µ ′ ⎟<br />

| Bsµ 〉ɛ<br />

} {{ } } {{ kn ⎠ c kn<br />

}<br />

Bsµ<br />

(1)<br />

(2)<br />

sµ e ik(B+s−s′ ) !<br />

= 0 (2.12)<br />

with the Hamiltonian matrix (1) and the overlap matrix (2). This equation is now<br />

solved iteratively.<br />

All calculations based on this method are labelled as (fplo 9.07.41, parameters<br />

and approximations used).<br />

2. Linear Muffin Tin Orbital – Atomic Sphere Approximation code (LMTO-<br />

ASA) [17, p. 331–333, p. 355–363]<br />

The Muffin Tin Orbital [25] approach is a special case of an APW method employing<br />

a spherically-symmetrized potential <strong>for</strong> the atomic part and a “flat” one<br />

in the interstitial region 4 (see fig. 2.2) with a smart choice <strong>for</strong> the basis functions<br />

– using surface spherical harmonics multiplied by<br />

a) the radial solution plus a term proportional to the spherical Bessel function<br />

(regular at the origin) inside the sphere and<br />

4 comparable to the Korringa-Kohn-Rostoker[26, 27] method


2.2 Photoemission Process 11<br />

Figure 2.3: mean free inelastic scattering path of electrons in solids; only weak material<br />

dependence [28, p. 8]<br />

b) the Neumann function (regular at r → ∞) in the interstitial region.<br />

Those basis functions and their derivatives are by generation smooth at the<br />

sphere’s boundary. Furthermore, assuming that only closed packaged structures<br />

(corresponding to tightly bound compounds) will be regarded, the atomic spheres<br />

can be extended until the interstitial part vanishes completely (ASA). The major<br />

drawback of the LMTO code is its dependence on ASA which does not yield good<br />

approximations <strong>for</strong> anisotropic or “open” structures not to mention slab calculations.<br />

In principle, one has to take care of each setup by manually adjusting<br />

the spherical overlap, eventually even introducing “empty spheres” (an additional<br />

atomic sphere without a nuclei potential but with local basis functions), to obtain<br />

valid configurations.<br />

All calculations will be labelled as (lmto 5.01.1, parameters and approximations<br />

used).<br />

Since the applied approximations and the basis sets are different <strong>for</strong> LMTO and FPLO,<br />

their spherical contributions (l, m projection) can differ severely although the converged<br />

charge densities are approximately the same. Due to the fact that LMTO is based on<br />

ASA and a spherically-averaged potential the results are in principle less accurate than<br />

the ones obtained by FPLO, thus the latter has been used <strong>for</strong> most of the calculations.<br />

Nevertheless, to estimate the hybridization strength it was necessary to extract the<br />

coefficients from LMTO (see ch. 4.2).<br />

2.2 Photoemission Process<br />

Photoemission spectroscopy (PES) has been one of the first techniques which allowed<br />

to study the quantized nature of electrons inside solids, but due to the small electron<br />

escape depth (see fig. 2.3) the interpretation of the spectra was difficult. The interest<br />

in electron-based spectroscopy has risen in the 1970s (especially <strong>for</strong> solids), because


12 2 Theoretical foundation<br />

Figure 2.4: photoemission models – left: three-step model, dividing the photoemission process<br />

into (1) excitation, (2) transport and (3) transmission to the vacuum; right: one-step model,<br />

quantum mechanical description of the excitation process by calculating the transition propability<br />

between the bound state and the unbound free state whose tail decays into the solid [28,<br />

p. 245]<br />

routine methods to obtain ultra-high vacuum (UHV) have been developed to establish<br />

the precondition <strong>for</strong> analyzing clean surfaces [28, p. 8] which enabled the community<br />

<strong>for</strong> the first time to distinguish surface and bulk originated spectral features. Since the<br />

discovery of the high-T c compounds and the development of devices capable of angular<br />

resolved photoemission spectroscopy (ARPES) , the interest in PES is regrowing.<br />

Photoemission (PE) is basically a “photon in – electron out” process granting access<br />

to the electronic structure of solids. Measuring the emission angle and the energy of the<br />

electron one is able to analyze the manybody transition. Initial states will be marked<br />

by the index i, final states correspondingly by f. In the following the two major models<br />

are sketched (a detailed description is given in [28]).<br />

2.2.1 Three-step model<br />

The originally single-step quantum mechanical PE process is devided into three steps,<br />

which will be discussed below. Nevertheless, this approximation is purely phenomenological<br />

and has been described in detail in [29, 30].<br />

(i) Optical excitation<br />

The incoming photon (excited electron) is characterized by the energy E ph (E e )<br />

and the momentum p ph (p e ), respectively. Neglecting the photon’s momentum<br />

(since p ph ≪ p e ≈ 100p ph , this is only valid <strong>for</strong> ω ≪ 500 eV) and respecting<br />

momentum conservation allows only “vertical” transitions – where the electron’s<br />

momentum changes by plus / minus a reciprocal lattice vector. In principle, one<br />

should deal with the extended zone scheme because elsewise one is not able to


2.2 Photoemission Process 13<br />

Figure 2.5: (a) general scheme of the three step model, (1) the photo excitation of the electron<br />

inside solid, (2) transport to the surface and (3) the transmittion to the vacuum [28, p. 12]; (b)<br />

sketch of the third step: penetration through the surface, only the parallel component of k is<br />

conserved [28, p. 249]<br />

distinguish between the wave vector of the crystal states k f and the momentum<br />

of the excited electron K f = k i + G inside the solid.<br />

(ii) Transport to the surface<br />

After the excitation (having overcome the atomic potential) the electron travels<br />

arbitrarily through the solid whereat the scattering is dominated by electronelectron<br />

interaction. The electronic inelastic mean free path reads<br />

λ (E, k) = τ d E<br />

d k<br />

and is approximately 3-5 Å <strong>for</strong> an energy range of 30-150 eV. Hence, one has to<br />

include inelastic scattering processes <strong>for</strong> an appropriate description but they will<br />

be neglected here. For further in<strong>for</strong>mation I refer to [30].<br />

(iii) Transmission to the vacuum<br />

All electrons <strong>for</strong> which the component of the kinetic energy perpendicular to the<br />

surface is large enough to overcome the surface potential, will transmit to the<br />

vacuum:<br />

2<br />

2m K ⊥ 2 ≥ (E F − E 0 ) + Φ (2.13)<br />

whereat E F is the Fermi level, E 0 the binding energy of the electron state and the<br />

work function is denoted by Φ. The transmission through the surface conserves<br />

the parallel momentum (cf. fig. 2.5b) and using the energy dispersion of the free<br />

electron yields<br />

K ‖ =<br />

( 2m<br />

2 E kin<br />

) 1/2<br />

sin ϑ out =<br />

( 2m<br />

2 E f − E F<br />

) 1/2<br />

sin ϑ in (2.14)


14 2 Theoretical foundation<br />

<br />

<br />

<br />

Figure 2.6: different quantum states: upper row – final states, lower row – initial states; (a)<br />

bulk Bloch wave weakly damped, (b) strongly damped Bloch wave, (c) “surface” / gap state,<br />

(d) bulk Bloch wave and (e) surface state inside the bulk band gap [28, p. 273]<br />

which resembles Snell’s law (refraction of the wave vector). Choosing the reference<br />

frame outside, the maximum angle equals ϑ out, max = 90 ◦ . Since E kin = E f − E F +<br />

Φ, the angle ϑ in < ϑ out and thus all electrons “inside” this cone (ϑ < ϑ in,max ) will<br />

transmit to the vacuum (cf. [28, p. 248]). Evidently, all inelastically scattered<br />

electrons – depending on the number of scatter events – will have different escape<br />

cones.<br />

A reduced type of this model has been used <strong>for</strong> evaluating the experimental photoemission<br />

spectra and transfering the results to reciprocal space.<br />

2.2.2 One-step model<br />

The decomposition of the PE process into several parts neglects important interference<br />

effects between different emission channels (e.g. bulk and surface emission) and simplyfies<br />

the transmission probability severely [28, p. 280]. Hence describing it properly as a<br />

transition between two quantum mechanical states will respect the wave / particle duality.<br />

Using Fermi’s Golden rule and an approximation <strong>for</strong> the interaction Hamiltonian<br />

H int w fi = 2π <br />

∣<br />

∣〈 f | H int | i 〉 ∣ ∣ 2 δ (E f − E i − ω) (2.15)<br />

with e.g. H int = 1 (A · p + p · A) (2.16)<br />

2mc<br />

as well as a reasonable composition of initial | i 〉 and final states 〈 f | yields generally the<br />

“correct” spectra. For H int the interaction between an electron (momentum operator p)<br />

and a photon (field operator A) can be assumed. In a manybody description the transition<br />

operator (e.g. f emission: t f (ω) (fψ + + ψ + f); f + (f) is the creation (annihilation)<br />

operator <strong>for</strong> an f electron and ψ the corresponding photon operator, t f (ω) represents the<br />

weight <strong>for</strong> this emission channel) may be used <strong>for</strong> specific emission channels. In addition,<br />

spectral broadening can be dealt with special representations <strong>for</strong> the δ-distribution


2.2 Photoemission Process 15<br />

in eq. 2.16. The best compromise between complexity and feasible computational ef<strong>for</strong>t<br />

<strong>for</strong> initial and final states has been found with inverse Low Electron Energy Diffraction<br />

(LEED) states. Since in LEED a valid description <strong>for</strong> initial, transmissible and<br />

reflected electron states (which respect the low mean free path of electrons as well as<br />

the vacuum / solid transition by a potential step) has been developed, the left task is<br />

to reverse the transmitted (incoming) LEED beam and to add the photon receiving the<br />

initial and final states <strong>for</strong> photoemission.<br />

2.2.3 General remarks<br />

As it has been sketched, PE spectra reflect the transition between two eigenstates and<br />

hence differ fundamentally from ground state Kohn-Sham eigenenergies [31]. At least<br />

<strong>for</strong> Hatree-Fock theory Koopmans stated, that the energy of the highest occupied orbital<br />

is approximately the ionization energy (with negative sign) [32]. A more detailed review<br />

is given in [31]. However, <strong>for</strong> a qualitative description of valence band photoemission<br />

of metals one can use DFT results hence the final state almost resembles the initial<br />

state because the photoexcitation hole is delocalized – which means well-screened. In<br />

contrast, core level or localized electron photoemission depend strongly on the relaxation<br />

of the final state.<br />

Besides, since the inelastic mean free path of electrons is in the order of the lattice<br />

constant, the k ⊥ component is not conserved. Thus, one has to partially integrate<br />

over k ⊥ <strong>for</strong> initial states. The <strong>for</strong>mer and the finite lifetime of final states cause the<br />

emergence of projected band structure in photoemission.


17<br />

3 Experimental foundations<br />

3.1 Photoemission<br />

Having already regarded the principles of photoemission (see ch. 2.2) this paragraph<br />

concentrates on experimental issues. At least, one has to note the following effects and<br />

differences:<br />

• To obtain angle-resolved spectra high-quality monocrystalline samples are needed<br />

because the translational invariance is a requirement <strong>for</strong> momentum conservation<br />

in photoemission.<br />

• Since the incident photon beam has a finite spot size, we have spatial as well<br />

as temporal integration <strong>for</strong> the outgoing particle wave functions in addition to<br />

the intrinsic interference contribution of different PE channels. Furthermore, the<br />

sample’s surface is not homogeneous (breaking exactly at one well defined layer)<br />

and exhibits terraces and single atoms adsorbed at the surface. Additionally, the<br />

sample can have different oriented domains and there<strong>for</strong>e one usually integrates<br />

(spatially) over a non-homogeneous surface region (we selected samples whith<br />

large single-domain regions <strong>for</strong> PE).<br />

• Having a semi-infinite solid, we face in principle three different states (cf. fig. 2.6)<br />

which are crucial <strong>for</strong> photoemission: bulk states, surface resonances and surfaces<br />

states (the last two will be subsumed as surface states). Given that the<br />

Figure 3.1: manipulator including sample holder inside the preparation chamber: left: raw<br />

sample with lever stick; right: cleaved sample


18 3 Experimental foundations<br />

mean free inelastic scattering path of electrons in solids follows a general curve<br />

(cf. fig. 2.3) yields <strong>for</strong> the used energy range [45. . .140] eV approximately [4. . .8] Å,<br />

thus PE is rather surface sensitive. Hence, to obtain clean surfaces with few adsorbed<br />

atoms / molecules, we prepare them in-situ by cleaving (a stick which<br />

has been glued previously onto the sample is used as lever to split the sample,<br />

see fig. 3.1). The difference between surface and bulk states can experimentally<br />

be distinguished by choosing different surfaces or by quenching of surface emission<br />

involving the deposition of adlayers. In ab-initio calculations, one can use<br />

the contribution of the first layer to the band structure of a supercell (slab) as<br />

an estimate, where the translational invariance is broken parallel to the surface<br />

normal. Nevertheless, also the top layer contributes to bulk states, and there<strong>for</strong>e<br />

one should generally use the states, the charge density of which is localized<br />

at the surface. As the DFT calculations deal solely with itinerant electrons, the<br />

PE of rather localized 4f electrons has to be incooperated in another way. The<br />

insignificant overlap in all three spatial dimensions of the respective states allows<br />

to use calculated atomic PE spectra as a first approximation. The 4f emission<br />

of Eu adlayers is shifted towards higher binding energies due to the asymmetry<br />

in the potential at the surface. Since the lower coordination number effects<br />

the distribution of valence electrons, the binding energy of europiums’ 4f orbitals<br />

changes and there<strong>for</strong>e the shift of the surface emission is comparable to that of a<br />

surface core level shift (predictable by a Born-Haber process [33]).<br />

• As a coarse approximation atomic cross sections <strong>for</strong> photoemission will be used<br />

to analyze the corresponding spectra. Basically it has been shown, that they are<br />

different in solids (e.g. revealing several Fano resonances [34]) caused by deviations<br />

in the shape of the wave function, particularly due to different boundary conditions<br />

in the solid and the free atom. However, the calculation is non-trivial and in zeroth<br />

order the atomic cross sections are a suitable estimate.<br />

3.2 General setup<br />

The photoemission spectra were taken at two experimental setups located at different<br />

synchrotron sources each having its unique beamline layout and endstation characteristic.<br />

Their specifics will be discussed below.<br />

Generally, a synchroton radiation source consists of a booster unit composed of diverting<br />

coils and accelerating parts, which take charged particles to high energies (typically<br />

several GeV). The variation of the magnetic field according to the particles’ speed keeps<br />

them on a closed orbit. Afterwards they are induced into a storage ring where the energy<br />

loss (proportional to the amount of synchrotron radiation) is compensated by linear accelerators<br />

mounted into the ring so the particles remain at an orbit with constant speed.<br />

Since the radiation at high energies peaks strongly in the <strong>for</strong>ward direction in contrast


3.2 General setup 19<br />

Figure 3.2: characteristics of the dipole radiation of a charged particle [35, p. 25]<br />

to a classical dipole [35], synchrotrons are the preferred sources <strong>for</strong> a high photon flux<br />

(see fig. 3.2). In the case of linear accelerators, the radiation peaks in <strong>for</strong>ward direction<br />

as well, but the flux is significantly lower and one has to separate the photons from the<br />

charged particles since their directions are equal.<br />

The Swiss Light Source (SLS) as well as the Berliner Elektronenspeicherring-Gesellschaft<br />

für Synchrotronstrahlung m.b.H. II (BESSYII) – where all experiments have been per<strong>for</strong>med<br />

– use electrons. Whereas the <strong>for</strong>mer supports a continues current by top-up<br />

injection – which means that electron losses 1 are compensated by periodic reinjection<br />

into the storage ring – the latter does not provide an uninterrupted beam. The endstations<br />

are composed of connected vacuum chambers separating special purpose environments<br />

(a general sketch is given in fig. 3.3a). Being separated by different valves<br />

ensures well-defined vacuum conditions <strong>for</strong> the different steps of sample preparation and<br />

measurement. Usually there is:<br />

1. a small (fast-)entry lock designed <strong>for</strong> sample exchange evacuated by at least<br />

one pre pump (e.g. rotary vane pump) in addition with a turbomolecular pump,<br />

so that a fast transfer is garantueed.<br />

2. a preparation chamber, which is used e.g. <strong>for</strong> cleaving, pre-cooling, heating<br />

(annealing), vacuum deposition, Low Electron Energy Diffraction (LEED) and<br />

other preliminary sample modification or analysis. Since the volume of the preparation<br />

/ analyzing chamber is larger than that of the fast-entry lock, additional<br />

pumps <strong>for</strong> a higher throughput are needed to obtain ultra high vacuum condition.<br />

3. an analyzing chamber usually separated to preserve the prepared / cleaned<br />

sample structure by providing a stable UHV environment during the measurement<br />

since due to deposition or degasing the vacuum condition of the preparation<br />

1 even though using ultra high vacuum (in the range of ≈ 10 −10 mbar) the collission probability<br />

with remaining molecules is so high, that there is a bisection of the number of electrons after<br />

approximately 10 hours (cf. BESSYII, current loss per shift)


20 3 Experimental foundations<br />

chamber varies in time. For this purpose, especially getter pumps are used because<br />

the final pressure of mechanical pumps like turbo pumps is limited by their<br />

reverse flow. In laboratories, usually the radiation from electrical discharge lamps<br />

(e.g. He) is used, and there<strong>for</strong>e, in that case, turbo pumps are preferred to getter<br />

pumps which are not able to bind noble gases. Commonly, there are at least<br />

three ports: a transfer valve to the preparation chamber, the beamline and an<br />

analyzer port apart from windows, ports <strong>for</strong> ion gauges, pumps and other.<br />

Below, the different designs and experimental conditions of the two endstations are<br />

illustrated.<br />

3.3 BESSYII: 1 3 ARPES<br />

The 1 3 ARPES setup (see fig. 3.3b) was constructed to achieve ≤1 meV energy resolution<br />

of the beamline as well as of the analyzer at a sample temperature of ≤1 K. Since<br />

the latter is restricted by the cooling agent as well as by the thermal isolation and<br />

the thermal contact of the sample holder to the cooling reservoir, it was necessary<br />

to build a cooling shield combined with a cryostat which is connected to the sample<br />

holder. The cooling process is nested using liquid nitrogen <strong>for</strong> the outer shield, filled<br />

with liquid helium inside which surrounds the closed 3 He cooling cycle. A compromise<br />

between cooling efficiency and sample orientations’ degrees of freedom was found by<br />

allowing only x, y, z and incident angle movements limiting the remaining rotations<br />

and hence minimizing the radius of the cryostat. It has one entry (beamline) and one<br />

exit (analyzer) slit which are usually open and a sample entrance which is locked by<br />

a mechanical-driven flap gate. Principally it is possible to obtain (projected) constant<br />

energy surfaces, however this is limited by the precision of the manipulator. Since the<br />

size of a homogeneous sample region (meaning same kind of surface atoms, little surface<br />

defects and same crystal domain) is very small (in the range of 100 µm 2 ), the accuracy of<br />

the manipulator should be as good as possible. Un<strong>for</strong>tunately, having adapters to steer<br />

the sample holder inside which have a rather large play when changing the screwing<br />

direction does not allow to stay at the same sample region. The analyzer slit has been<br />

mounted vertically.<br />

3.4 SLS: SIS-HRPES<br />

This endstation (see fig. 3.3c) is able to achieve around 10 K without the necessity of<br />

shielding. Thus one has the advantage of an high precision manipulator with no angular<br />

restriction which is fully computer-controlled and hence each position can be approached<br />

several times without losing the absolute sample position. There<strong>for</strong>e, most of the BZ<br />

maps and high-symmetry cuts of the BZ have been obtained here. In contrast to


3.4 SLS: SIS-HRPES 21<br />

1 3 ARPES the analyzer slit is mounted horizontally. Accordingly vertical and horizontal<br />

polarized spectra are not directly comparable <strong>for</strong> both endstations.


Figure 3.3: (a) general setup of a photoemission endstation at synchrotron facilities, (b) on the left-hand side: the 1 3 ARPES, (c) on the right-hand<br />

side: the SIS ARPES endstation<br />

22 3 Experimental foundations


23<br />

4 EuRh 2 Si 2 – semi-localized electrons<br />

4.1 Overview – properties and classification<br />

The ternary compound EuRh 2 Si 2 crystallizes in the tretragonal body-centered ThCr 2 Si 2<br />

structure. The respective lattice parameters and Wyckoff positions are given in tab. 4.1<br />

(experimental and calculated relaxed 1 parameters).<br />

Surprisingly, the differences between<br />

the computionally relaxed parameters and the experimental ones are rather small<br />

(an indication of the well-chosen basis set in FPLO and keeping the 4f occupation fixed<br />

a suitable approximation – at least <strong>for</strong> total energy). The crystal structure is depicted<br />

in fig. 4.5a, whereas the tretagonal unit cell is bordered by dotted lines. It is evident,<br />

that this is a layered structure with respect to the [001] direction.<br />

The main transport properties and the phase diagram of EuRh 2 Si 2 have already<br />

been explored [3], hence only a short review is given here. Since similar ternary compounds<br />

show <strong>for</strong> example Heavy-Fermion behaviour (YbRh 2 Si 2 in [5]), superconductivity<br />

(CeRh 2 Si 2 in [6]), mixed-valent (EuPd 2 Si 2 in [37, 38]) or spin-density wave behaviour<br />

(EuFe 2 As 2 in [4]) one expects also interesting electronic properties and magnetic behaviour<br />

in EuRh 2 Si 2 . Belonging to the stable divalent europium materials it reveals<br />

an antiferromagnetic (AF) ordered phase of the localized 4f 7 moments below 25 K, the<br />

exact configuration of which is unknown.<br />

It is supposed to be a ferromagnetic coupling<br />

in the Eu plane and an AF order between the respective layers [3]. The energy<br />

gain of magnetic order is small and thus yet a small pertubation induces a different<br />

magnetic order or partially non-magnetic contribution to the ground state. There<strong>for</strong>e<br />

1 In the case of the lattice constants the computational relaxation has been per<strong>for</strong>med by a self-written<br />

implementation of the gradient method minimizing the total energy. Plotting the energy functional<br />

<strong>for</strong> a deviation of 15% from the experimental lattice parameters shows a smooth energy surface.<br />

The Wyckoff positions were <strong>for</strong>ce-minimized afterwards by the implemented procedure in FPLO. In<br />

both cases, the total error of the computation has been smaller than the experimental one and the<br />

<strong>for</strong>mer has been rounded to the accuracy of the latter.<br />

Table 4.1: (a) lattice constants and (b) Wyckoff positions: experimental [36] and calculated<br />

relaxed parameters (fplo 9.07.41, LDA, 4f 7 unpolarized open core)<br />

(a) lattice constants<br />

exp. [Å] relaxed [Å]<br />

a x 4.107 4.089<br />

a y 4.107 4.089<br />

a z 10.25 10.244<br />

(b) Wyckoff positions<br />

exp.<br />

relaxed<br />

x y z x y z<br />

Eu 0 0 0 0 0 0<br />

Rh 0 0.5 0.25 0 0.5 0.25<br />

Si 0 0 0.375 0 0 0.375


24 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.1: unit cells <strong>for</strong> different space group symmetry, coloured<br />

atoms depict positions defined by the described symmetry aspects;<br />

(a) bulk; left: Wyckoff positions, right: sites; (b) semi-bulk;<br />

left: Wyckoff positions, right: sites (c) slab configurations; left: Si<br />

terminated, right: Eu terminated (a Wyckoff position is a set of points,<br />

which is invariant with respect to the symmetry operations defined by<br />

the space group; a site is a set of points, which is invariant with respect<br />

to the point symmetry (without elements of the translational group) of<br />

the space group)<br />

spectroscopic studies and experiments based on the de-Haas-van-Alphen effect (e.g. to<br />

determine the resonant orbits of the Fermi surface) will probably not reveal signatures<br />

of the magnetic transition at 25 K.<br />

4.1.1 Brillouin zone and computational setups<br />

Each crystal has an intrinsic, highest symmetry group (space group (SPG): union of<br />

Bravais lattice and point symmetry). Depending on the surface sensitivity of the measurement,<br />

several symmetry operations are not allowed anymore (symmetry breaking).<br />

Since the Bravais lattice determines the BZ, one has to take care in comparing different<br />

setups. Here three miscellaneous symmetry configurations will be discussed:<br />

1. bulk (SPG 139 – I4/mmm) is the highest symmetry group <strong>for</strong> the ThCr 2 Si 2<br />

structure. The body-centered symmetry in real space is reflected by a truncated<br />

octahedral BZ which is de<strong>for</strong>med in z-direction whereat auxiliary the corresponding<br />

quadrangles perpendicular to the z-axis are scaled in comparison to the first<br />

BZ of a fcc crystal. This represents the unique three-dimensional domain <strong>for</strong> the<br />

band structure.<br />

2. semi-bulk (SPG 123 – P 4/mmm) means, that the body-centered symmetry is<br />

disregarded which doubles the number of sites per unit cell and causes backfolding<br />

of bands (with zero bandgaps, in principle).<br />

The BZ is simple tetragonal<br />

and there<strong>for</strong>e one has to map points from SPG 123 and 139 accordingly (high<br />

symmetry points, besides Γ are not identical).<br />

3. Additionally to the semi-bulk configuration, the surface (SPG 123) is described<br />

by a stretching along z-direction and empty space (vacuum) added to the unit cell


4.1 Overview – properties and classification 25<br />

Figure 4.2: overview of the Brillouin Zone: (left) k z = 0 section of the BZ – the green color<br />

represents the BZ of the bodycentered structure (SPG 139), the red color the simple tetragonal<br />

cell (SPG 123); neglecting body-centered basis symmetry, bands get backfolded; respective<br />

directions are marked by the blue and yellow lines (right) 3D representation of the BZ of SPG<br />

139 (grey) and 123 (red); the section of the left hand side is depicted correspondingly<br />

(either at the center or at the boundary), which has to be large enough, in order<br />

that the overlap of the wavefunctions of adjoined atoms is negligible – resulting<br />

in a model, quasi 2D material with no k z dispersion. The amount of layers to<br />

construct this configuration is a compromise between computational time needed<br />

(because the number of atoms per unit cell rises) and the in<strong>for</strong>mation about the<br />

surface / bulk relation which is intended to be obtained.<br />

Since only the Bravais lattice accounts <strong>for</strong> the BZ, one can furthermore reduce the<br />

needed amount of in<strong>for</strong>mation (and thereby the computational time) by using the point<br />

symmetry which results in the irredicuble wedge of the BZ used finally <strong>for</strong> the calculations.<br />

Setups <strong>for</strong> the a<strong>for</strong>e mentioned configurations and a note on the difference<br />

between lattice symmetry and point symmetry are depicted in fig. 4.1.<br />

In principle one can distinguish (dependent on the level of localization) between bulk<br />

states (delocalized), surface resonances (increased probability at the surface) and surface<br />

states (localized at the surface corresponding to two-dimensional states). Using a unit<br />

cell m-times repeated in z-direction results in respective backfolding of bands inside the<br />

BZ (with “reduced” k z dispersion), because it decreases with the same factor the unit<br />

cell increases. Extending the procedure to m → ∞ maps all dispersion with respect to<br />

k z onto the k x × k y plane resulting in a quasi two-dimensional representation of the<br />

bulk band structure, the so-called projected bulk band structure. This process can be<br />

imagined as reflecting the band structure subsequently along a mirror plane – which is<br />

shown <strong>for</strong> the transition from SPG 139 to SPG 123 in fig 4.3c. It should be noted, that<br />

the projected band structure does not depend on a surface or an exponential decay of<br />

the wavefunction into the vacuum because mathematically it contains the same amount<br />

of in<strong>for</strong>mation as the bulk band structure. However, in an ideal bulk the smallest


26 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.3: (a) comparative band structure plot of high-symmetry cuts of SPG 123 (red)<br />

and SPG 139 (green) in the BZ; the corresponding paths have been sketched in (b)+(c);<br />

in (c) additionally the backfolding is sketched: the band structure of the upper green plane is<br />

reflected along the red plane (BZ border of SPG 123) onto the one of the lower green plane<br />

(see the arrow); the band structure along Γ-Z-Γ indicates the reflecting character of the red<br />

plane; since SPG 123 has a cuboid BZ, the bisection can be infinitely repeated resulting in the<br />

mentioned projected band structure<br />

translational invariant quantity – the unit cell – is superior and backfolding does not<br />

occur since the corresponding components of the potential’s Fourier trans<strong>for</strong>mation<br />

vanish. But in PE projected band structure occurs, because the initial state is the sum<br />

of all states in the region of the surface (can have different k z ) and the indeterminacy of<br />

k z in the final state due to finite penetrating depth. There<strong>for</strong>e, the surface configuration<br />

should consist of projected bulk band structure due to the change in periodicity as<br />

well as of surface states / resonances depending on the respective boundary conditions.<br />

Both parts are important to compare the surface / bulk sensitivity of the measurements.<br />

Speer et. al. [39] made a detailed analysis exemplarily <strong>for</strong> silver on the topic of emerging<br />

band structure in photoemission. Since semi-bulk and surface have the same space<br />

group, the projection onto the k x × k y plane <strong>for</strong> the surface BZ will be denoted by a<br />

bar over the respecting high-symmetry points.<br />

Below, the relationship between the BZs of SPG 123 / 139 depicted in fig. 4.2 will be<br />

discussed. A cut at k z = 0 is illustrated denoting the boundary of the BZs inplane as<br />

solid and structures of lower and higher parallel layers by dotted lines. Hence the stacking<br />

of the bulk BZ in the x/y-direction is shifted by (0, 0, ±1/2) <strong>for</strong> the next-neighbour<br />

BZ, the Z-point is the mid-distance point of the Γ-Γ path <strong>for</strong> each spatial direction (x,<br />

y, z, -x, -y, -z). Neglecting body-centered symmetry halves the BZ (cf. the 3D image in<br />

fig. 4.2) and each Z-point is mapped onto Γ. These backfolding from SPG 139 to 123<br />

will be exemplarily shown <strong>for</strong> two high-symmetry directions: the diagonal path Z-X-Γ<br />

(blue) gets mirrored with respect to X reassembling the Γ-M-Γ path in SPG 123 whereat<br />

the folding orientation is given by the arrows. Correspondingly, Z-X maps onto Γ-X.


4.1 Overview – properties and classification 27<br />

A short comparison of the measured high-symmetry directions of bulk and semi-bulk<br />

is given in fig. 4.3. It is evident that increasing the translational period in z-direction<br />

causes a projection from the k z -dispersion onto the Γ-X cut revealing, <strong>for</strong> example, a<br />

bunch of very steep bands at the Γ-point. This relation will be regarded in chapter 4.3.<br />

4.1.2 Treatment of strongly localized electrons beyond L(S)DA<br />

Since already a few calculations have been presented to demonstrate the connection of<br />

the different SPGs, it will be explained in more detail why the argumentation already<br />

given in ch. 2.1.2 holds true.<br />

The published literature emphasizes, that in principle<br />

there is no general solution to overcome the limits of L(S)DA. An approach often implemented<br />

is joining L(S)DA and a model Hamiltonian (e.g. Hubbard model) to treat the<br />

correlation in a self-consistent scheme [40–42]. The majority of them depends on additional,<br />

empirical parameters resulting from the correlation model which vastly influence<br />

the result. The major drawback of those methods is, that they are not as easily comparable<br />

to each other as full-potential L(S)DA/GGA results are because the obtained<br />

solutions depend strongly on the implemented Hamiltonian as well as on the limit in<br />

which they are solved. There exist complicated modifications of the exchange correlation<br />

functional to include a self-consistent dynamical mean field solution of the Hubbard<br />

model (LDA+DMFT) [40], but since those schemes are not as stable as L(S)DA/GGA,<br />

an analytical obtained implementation of the Hubbard model has been used. The latter<br />

is called L(S)DA+U and the implemented functionals (atomic limit [AL], around mean<br />

field [AMF]) 2 in FPLO [42] are used in comparison to the open core calculations. The<br />

parameters have been chosen as U = 8 eV and J = 1 eV (Slater parameters / input<br />

parameters <strong>for</strong> FPLO: F 0 = 8 eV, F 2 = 11.92 eV, F 4 = 7.96 eV, F 6 = 5.89 eV from<br />

which U and J are computed) in accordance to [45, 46]. In that only the magnitude<br />

of U and J define the minimum, the solution persists stable under variation (10% of<br />

deviation from U, J). The outcome of the LDA+U [AL] calculation is an occopation<br />

of 6.7 being roughly 7.0 which has been utilized in the open core calculation (remembering<br />

the divalent limit: [Xe] 6s 2−x 5d x 4f 7 ). Under the premise that their Fermi levels<br />

are equal, a comparison between the 4f 7 open core (SPG 139 does not allow AF order,<br />

2 The LDA contains already all electron-electron interactions in a mean-field way (by definition as<br />

exchange correlation functional). Since the Hubbard model incoporates the exact Coulomb term,<br />

one cannot simply add both together because this would lead to a double counting term (a term<br />

present in LDA as well as in the Hubbard model). There<strong>for</strong>e one has to substract the mean-field part<br />

from either the LDA result or the Hubbard model – whereat the <strong>for</strong>mer is unfavourable because it<br />

already contains spatial variations of the potential which we want to keep, the latter method is often<br />

implemented. This can be achieved in two limits: (a) adding the non-mean field part of the Hubbard<br />

Hamiltonian resulting in the around mean field limit (E LSDA+U [AMF] = E LSDA + H int − 〈H int〉)<br />

or (b) adding the difference of the Hubbard Hamiltonian and its atomic limit (E LSDA+U [AL] =<br />

E LSDA + H int − E atomic limit ). The <strong>for</strong>mer works well <strong>for</strong> almost uniquely populated orbitals (e.g.<br />

in the case of Yb) whereas the latter should be taken <strong>for</strong> rather localized d / f electrons because it<br />

separates occupied and unoccupied levels emphasizing behaviour of one-particle excitations. For a<br />

detailed review see [43, 44].


28 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.4: comparison between an LDA+U (fplo 9.07.41, LDA+U [AL], U = 8 eV, J = 1 eV)<br />

and an open core calculation (fplo 9.07.41, LDA, 4f 7 unpolarized open core); the bandwidth<br />

of the 4f states is small, and since the overlap with neighbouring orbitals is insignificant, the<br />

VB structures are comparable; the band structure after the first iteration process with fixed<br />

occupation matrix <strong>for</strong> the correlated orbital is depicted in red, the fully-converged LDA+U<br />

band structure in blue; (1) away from the 4f levels the VB structures <strong>for</strong> all calculations are<br />

comparable; (2a) level crossing; (2b) avoided level crossing (hybridization); (3) the Fermi surface<br />

depends strongly on the applied functional, there<strong>for</strong>e an exhaustive analysis should be done to<br />

recover the correct one; most promising at the moment is a sufficient solution of a model<br />

Hamiltonian (e.g. Anderson model) based upon an L(S)DA calculation<br />

there<strong>for</strong>e an unpolarized configuration has been used legitimated by the rather unstable<br />

AF configuration) and the calculation with correlated orbitals is depicted in fig. 4.4<br />

showing that in principle the valence band (VB) structure (neglecting the hybridization<br />

of the valence band with the shallow 4f bands) of both is comparable. Assuming<br />

that the hybridization between the localized 4f electrons and the VB is small, and that<br />

the dispersion of the <strong>for</strong>mer is negligible, one can use the open core calculation as a<br />

first approximation to the VB structure applying afterwards a hybridization model (e.g.<br />

Anderson model) to include the interaction between localized and itinerant electrons<br />

again. In principle, the L(S)DA+U results can be interpreted as a correction to the<br />

single-particle self-energy yielding the spectral function (which corresponds to the PE<br />

spectrum), but it is, nonetheless, only a vast approximation of the PE process [42]. In<br />

addition, the convergence cycle of the L(S)DA+U calculation is cumbersome because<br />

the energy manifold depends on the path taken during the convergence (more precise:<br />

the proportion between the L(S)DA and the occupation matrix iteration procedure). To<br />

obtain a configuration close to divalent, the occupation has been fixed to 7.0 converging


4.1 Overview – properties and classification 29<br />

the charge density a first time. Afterwards, the self-consistency cycle of the occupation<br />

matrix has been permitted and the final result obtained. Returning to fig. 4.4, the<br />

characteristica of this iterative process will be discussed. Fixing the occupation matrix<br />

of the correlated orbital avoids shifting the orbital energies during the iterative process<br />

due to the model Hamiltonian. Hence the eigenenergies of all 4f orbitals are located near<br />

the Fermi level (red dotted representation). Allowing the fully self-consistent treatment<br />

shifts the occupied (unoccupied) 4f orbitals below (above) the Fermi level, respectively<br />

(blue dashed representation). Having 7 localized orbitals (spin degenerate) with an<br />

occupation of 6.7, at least 3 should lie below the Fermi level (since their dispersion is<br />

negligible). One can see, that <strong>for</strong> the self-consistent LDA+U calculation one 4f orbital<br />

is 3 eV, one 2 eV below the Fermi level and two are around the Fermi level (being<br />

partially above). Regarding regions marked by 1, it is apparent that the open core<br />

approximation is suitable <strong>for</strong> the VB in an energy range not allocated by 4f orbitals.<br />

In difference comparing 2a to 2b (being in the range of localized states), the symmetry<br />

of the VB determines the hybridization strength, and thus the deviation from the open<br />

core calculation. There<strong>for</strong>e it is necessary to apply a hybridization model (see ch. 4.2)<br />

to rearrange the band structure properly in the open core approximation (e.g. regarding<br />

the Fermi surface of the compound).<br />

4.1.3 Cleavage behaviour<br />

As already mentioned in ch. 3, the surface is prepared in-situ by sample splitting, hence<br />

the structure has several rupture lines (since it is layered). The lever stick has been<br />

oriented in the [001] direction, so that we were able to measure the [001] surface BZ in<br />

both setups (because k x , k y are equivalent and there<strong>for</strong>e the orientation of the entrance<br />

slit does only matter <strong>for</strong> the type of polarization). To estimate the possible cleavage<br />

plane and thus the atom type of the surface layer (termination) two different approaches<br />

have been used:<br />

(a) comparison of bond strength<br />

The bond strength between the layers can be estimated by the “charge transfer”<br />

inside the solid compared to the atomic allocation because electrons mediate bonds<br />

which implies that the iso-charge density is a valid measure <strong>for</strong> their distribution.<br />

There<strong>for</strong>e the sum of the atomic charge densities (<strong>for</strong> each site) has been substracted<br />

from the final (converged) charge density, so positive (negative) remaining charge<br />

density corresponds to an inflow (outflow) of electrons. If one compares Fig. 4.5b<br />

to fig. 4.5c it becomes evident, that the charge density is redistributed from the<br />

Eu layer and the region between Eu and Si to the composition of Si-Rh-Si, mostly<br />

between the atoms Rh-Si / Si-Rh, exactly in bonding direction. Thus the structural<br />

integrity of the latter is larger than that of the Eu-Si bond and one expects either Si<br />

or Eu terminated surfaces. Auxiliary, one can discuss the bonding type evaluating


30 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.5: (a) charge isosurface and extended lattice configuration of EuRh 2 Si 2 : an<br />

(arbitrarily-selected) isosurface of the converged charge density is plotted with respect to the<br />

lattice denoting the bonding between the Rh and Si layers; (b) + (c): difference between the<br />

converged total charge density and the “atomic” charge densities to estimate the charge transfer<br />

inside the unit cell, positive (negative) isocharge depicts inflow (outflow) of electrons with<br />

respect to the start configuration


4.1 Overview – properties and classification 31<br />

Figure 4.6: cleavage behaviour: (a) interlayer distances as a function of doubled c-lattice<br />

constant; increasing strain pulls apart mainly Eu and Si-Rh-Si structures, whereas the distances<br />

inside the Si-Rh-Si compound remain approximately constant; if the strain is too strong<br />

(a c ≈ 28 Å), the crystal breaks into two parts revealing one Eu and one Si terminated surface<br />

(the maroon marked graph depicts the maximum distance of next-neighbour Eu-Si layers); the<br />

inset magnifies the small region of elastic de<strong>for</strong>mation; (fplo 9.07.41, LDA, 4f 7 unpolarized<br />

open core | <strong>for</strong>ce minimization) (b) a slab geometry <strong>for</strong> a z < 28 Å; (c) a slab <strong>for</strong> a z > 28 Å,<br />

the calculations shown are marked in (a)<br />

several isocharge surfaces. Recognizing that the charge density between the Eu<br />

layer and the Si-Rh-Si part is localized around the compounds, not in between,<br />

points to ionic character. Contrariwise, the localization of charge between two<br />

neighbouring atoms <strong>for</strong> Si-Rh / Si-Rh is an aspect <strong>for</strong> rather covalent bonds. Both<br />

aspects are supported by the total charge density distribution in fig. 4.5a, because<br />

the iso-charge level shown reveals no connection between the Eu layer and the Si-<br />

Rh-Si part. Whether the covalent or the ionic bond is stronger, cannot be answered<br />

generally, but since the covalent one is directional in contrast to the undirectional<br />

ionic bond, we can expect that shear strain will <strong>for</strong>ce to break up the ionic bonds<br />

first and hence the cleave will reveal Si and Eu termination. However, one should be<br />

careful, because this is a suggestive interpretation of (differences in) charge density.<br />

In addition, using only the corresponding Wyckoff positions (e.g. only Eu, leaving<br />

all other positions empty) <strong>for</strong> the atomic charge densities neglects the possible<br />

<strong>for</strong>mation of bonds during the convergence (e.g. metallic bonding <strong>for</strong> Eu, since the<br />

distance between two atoms is not small enough). In principle, one could use the<br />

start density be<strong>for</strong>e the iteration process, but due to the special compressed orbital<br />

basis set, the configuration is as well not equal to the atomic configuration. Further<br />

investigations should be undertaken to clarify this issue. All in all, this method<br />

is computationally inexpensive since only the converged charge density has to be<br />

extracted in real space.


32 4 EuRh 2 Si 2 – semi-localized electrons<br />

(b) first-principle <strong>for</strong>ce and energy minimization [47]<br />

Stretching the z-axis of a supercell (1 × 1 × n, n ∈ N ) and per<strong>for</strong>ming <strong>for</strong>ce<br />

(relaxation) minimization <strong>for</strong> each configuration yields different interlayer distances<br />

<strong>for</strong> Eu-Si and Si-Rh because of their different bond strength. For a z > a z, crit.<br />

either the Eu-Si or Rh-Si bond will break up and two surfaces will emerge (which<br />

corresponds to a slit in the crystal). The resulting interlayer distances as functions of<br />

the z-axis lattice constant a z are depicted in fig. 4.6. Obviously the bond strength of<br />

Eu-Si is smaller than that of Rh-Si, because the distance of the <strong>for</strong>mer is increasing<br />

more intensely than that of the latter with growing strain. For a z being larger than<br />

the critial lattice constant a z, crit. ≈ 28 Å, the evolution of two surfaces – either<br />

terminated by Eu or Si – can be observed (cf. fig. 4.6c). Besides, one can note that<br />

the distance of the topmost layer (labelled as surface in fig. 4.6a) is smaller than<br />

the corresponding interlayer distance in the bulk. This relaxation due to bonding<br />

asymmetry sometimes severely influences the available surface features.<br />

It has to be mentioned, that this method presumes a large amount of processing<br />

power, because be<strong>for</strong>e each <strong>for</strong>ce optimization step a self-consistency calculation has<br />

to be per<strong>for</strong>med and the process can diverge if the energy self-consistency cycle does<br />

not converge sufficiently (usually the first self-consistency cycles in a stretched cell<br />

do not achieve the same convergence criteria in the limits of iteration and deviation<br />

in charge density in comparison to the bulk calculation, but if the density at least<br />

converges linearly, the next <strong>for</strong>ce minimization mostly does not diverge). For this<br />

calculations a moderate k-mesh of 8 × 8 × 6 (a compromise between accuracy and<br />

computing time) as well as SPG 99 has been used, because the latter does not<br />

reflect the mirror symmetry of the z-axis compared to SPG 123. All atom positions<br />

were allowed to be varied, there<strong>for</strong>e the slit position is totally arbitrary (it depends<br />

additionally on the convergence algorithm). Because our processing resources were<br />

limited, the <strong>for</strong>ce minimization was stopped after 4 weeks having reached an overall<br />

minimum <strong>for</strong>ce F min ≈ 0.1 eV/Å. Although the minimization has not been finished,<br />

the results are representative. Moreover it should be noted, that the Eu 4f–Rh 4d<br />

interaction (because it is rather weak) has been completely neglected due to the<br />

open core approximation.<br />

This is in coincidence wtih experimental observations: upon cleaving along the Eu-<br />

Si plane, the Eu atoms stick either on one or the other side of the cleaved crystal.<br />

Regarding cohesive energies, the <strong>for</strong>mation of smooth surfaces is energetically favoured.<br />

There<strong>for</strong>e, Eu atoms are expected to <strong>for</strong>m large islands and hence the cleaved surface<br />

constists of regions terminated either by Eu or Si atoms. Thus one should search an<br />

area with mainly one kind of surface atom to measure plain terminated surfaces. The<br />

corresponding spectral signatures will be discussed below.


4.1 Overview – properties and classification 33<br />

Figure 4.7: (a) Atomic cross section of Eu 4f, Rh 4d, Si 3s and Si 3p. A discussion on the<br />

validity of the their usage has been given in ch. 3.1; (b) Wide range overview <strong>for</strong> Eu 4d-4f<br />

resonance measured at hν = 142 eV with vertical polarization. The pure divalent character of<br />

Eu in EuRh 2 Si 2 is in accordance with Mössbauer experiments in [48] (SLS-SIS, T ≈ 15 K)<br />

4.1.4 Surface and bulk band structure<br />

In general, the PE spectrum of EuRh 2 Si 2 can be regarded in a first approximation as a<br />

superposition of a valence band PE spectrum (states with significant dispersion) and a<br />

spectrum of atomic transitions. The latter can be identified as lines because they do not<br />

reveal an angular / k dependency because the overlap of their atomic-like wavefunctions<br />

from different sites is negligible. Furthermore, we have to deal with the short mean<br />

free inelastic scattering path of the photo electrons (see ch. 3.1), thus it is necessary to<br />

distinguish between electronic states localized at the surface of the compound and states<br />

belonging to the bulk since both have a comparable share in the spectra. There<strong>for</strong>e in<br />

the following part the major contribution of specific spectral structures will be discussed<br />

with respect to the studied [001]-surface. To identify surface and bulk states one can<br />

per<strong>for</strong>m theoretical calculations (see bulk, projected bulk and surface configuration in<br />

ch. 4.1.1) to compare the band structure to the PE spectrum. The calculations <strong>for</strong><br />

the itinerant states were per<strong>for</strong>med mainly with FPLO whereat the atomic transitions<br />

<strong>for</strong> the localized 4f states were taken from configuration interaction based calculations<br />

in [49]. Experimentally it would be possible to verify the origin of a state by surface<br />

deposition of a noble metal, e.g. Ag (a surface state changes in binding energy whereat<br />

a bulk state does not vary severely), or by adjusting the energy of the photons probing<br />

different layers k x × k y since the incident photon energy determines k z . This is just<br />

an indication because also bulk states can have a small or negligible k z -dispersion.<br />

Both have not been per<strong>for</strong>med, because the atomic cross section in the range of the<br />

available photon energies already varies strongly. Hence PE on EuRh 2 Si 2 is sensitive<br />

to valence states (mainly Rh 4d) <strong>for</strong> hν = [40 − 55] eV whereas <strong>for</strong> hν = [120 − 140] eV


34 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.8: spectral overview measured at hν = 120 eV with linear vertical polarization along<br />

X − Γ − X; marked, white-shaded regions correspond to the integration limits <strong>for</strong> multiplet<br />

analysis; (a) Si terminated surface: the hybridization between the surface states and the Eu<br />

4f bulk is rather large indicated by the blur of the 4f emission lines. (b) Eu terminated<br />

surface: well-resolved bulk PE multiplet and surface Eu 2+ 4f emission at 1.5 eV binding energy.<br />

The trivalent PE component is missing (see fig. 4.7b). (c) The PE signal in (a)+(b) is related<br />

to the specifically-marked Eu layers. (SLS-SIS, T ≈ 15 K)<br />

Eu 4f emission dominates (cf. fig. 4.7). As the valence band and the 4f electrons <strong>for</strong>m<br />

hybride states, both emission channels are not separable. Generally, also the amount<br />

of electrons as a function of their origin (depth of the layer) should be included to<br />

get an approximation <strong>for</strong> the main contribution at a fixed photon energy. In that the<br />

experimental identification was not accessible, this paragraph is primarily predicated on<br />

theoretical modelling. The atomic PE signal of Eu 4f has been determined by Gerken<br />

et al. [49] within the sudden approximation and a basis set of linear combinations of<br />

Slater determinants (configuration interaction). The occupation of the ground state<br />

has been fixed to 4f 7 and the main contribution (> 97%) is a 8 S 7/2 configuration (each<br />

atomic orbital is occupied by one electron and all spins are equally aligned). The 4f 6<br />

final states (electron removal states, mainly 7 F 0...6 ) have been determined variationally<br />

and corresponding to Fermi’s Golden rule, the transition probability was computed<br />

whereupon a lower boundary omits states with intensities considerably smaller than 1%<br />

of the maximum intensity. The result is depicted on the right hand side in fig. 4.9, the<br />

topmost panel.<br />

The characteristic intensity distribution (lowest / highest intensity <strong>for</strong> the final state<br />

configuration dominated by 7 F 0 / 7 F 6 ) can be understood in a first approximation<br />

using Hund’s rules and no mixed states. The initial and final states could be regarded<br />

as the atomic levels mentioned be<strong>for</strong>e. Each final state J is represented by m J different<br />

microstates depending on the orientation of J with respect to the quantisation axis, and<br />

there<strong>for</strong>e the statistical weight <strong>for</strong> each final state 7 F J is (2J + 1). Since the excitation


4.1 Overview – properties and classification 35<br />

Figure 4.9: characterisation of the 4f final state multiplet: (left) the average shift of the<br />

multiplet lines comparing Eu and Si terminated surfaces is 33 meV, the Fermi level of both<br />

terminations has been aligned (cf. inset, see text). The splitting of the Eu surface component<br />

is related to the hybridization with the valence band, see ch. 4.2. (right) comparison between<br />

the relative position and intensity of the multiplet components with respect to the lowest energy<br />

level E 0 (E = E i − E 0 , i = 0 . . . 6), <strong>for</strong> orientation grey lines have been plotted as a guide <strong>for</strong><br />

the position of the calculated atomic transitions<br />

energy hν ≈ [40 − 150] eV is much larger than the energy splitting <strong>for</strong> the final states<br />

(E J=6 − E J=0 ≈ 0.6 eV, see experiment), one can assume that all possible microstates<br />

are equally occupied because of their “degeneracy” in energy. In that a combination of<br />

L, S and J determines the total energy in absence of electromagnetic fields [50], the<br />

(2J + 1)-degeneracy <strong>for</strong> each J is reflected in intensity resulting in the highest one <strong>for</strong><br />

the 8 S 7/2 → 7 F 6 transition. This is in accordance with the results obtained by Gerken<br />

including additionally weight differences due to non-degeneracy and a mixed ground<br />

state. A review on calculating many-body atomic states and transitions is given by<br />

H. Friedrich [50].<br />

Since in [49] only the partially occupied 4f shell has been used, the first emission line<br />

is located directly at the Fermi level. Comparing that to the experimentally obtained<br />

spectra (see fig. 4.8), one notices a shift of approximately 0.15 eV to higher binding<br />

energies which is probably related to many-body effects: the quasi-core hole due to<br />

4f emission cannot be “screened” totally by the remaining 4f electrons and thus the<br />

potential <strong>for</strong> the valence electrons changes resulting in an energy shift <strong>for</strong> all final states,<br />

because the average unscreened charge is the same <strong>for</strong> all configurations. It has already<br />

been shown that 4f emission at the Fermi level is prevalent if the final state resembles<br />

the ground state at least to some part [51–53]. But neither the 4f 8 admixture to the


36 4 EuRh 2 Si 2 – semi-localized electrons<br />

ground state is reasonable (enabling 4f 8 →4f 7 transitions) – following an arguement of<br />

Hund’s rules that half-filled shells are energetically favourable – nor the hybridization<br />

with the valence band at the Fermi level (admitting 4f 7 →4f 6 →VB −1 4f 7 transitions)<br />

seems to be substantial in EuRh 2 Si 2 (cf. ch. 4.2).<br />

Atomic-like surface emission<br />

The presence of Eu atoms at the surface is identified by a shift of the 4f emission by<br />

approx. 1 eV towards higher binding energies (see fig. 4.9), which reflects the altered<br />

bonding properties with respect to bulk Eu and can be determined by a Born-Haber<br />

process because the 4f level is rather localized and there<strong>for</strong>e the PE spectrum behaves<br />

similar to that of a corelevel. The shift is comparable to other divalent Eu compounds<br />

(cf. EuPd x in [54], EuPd 2 Si 2 in [38]), since it is mainly determined by the coordination<br />

number at the surface. As PE is rather surface sensitive, in principle solely the<br />

emission of the first (labelled surface) and second (labelled subsurface) Eu layer have<br />

a reasonable contribution to the spectrum (cf. fig. 2.3). Hence, they will be used as<br />

an approximation <strong>for</strong> the surface and bulk signal of Eu, respectively. In addition, the<br />

spectrum of the surface state seems to deviate from the seven final states of the bulk<br />

signal. There are two possible explanations <strong>for</strong> that behaviour: on the one hand, the<br />

potential cannot be regarded as spherically symmetric on the surface (lower coordination<br />

number) anymore, and there<strong>for</strong>e the final states can differ severely. On the other<br />

hand, the energy broadening increases linearly with binding energy because lifetimes of<br />

final states decrease due to stronger relaxation. It will be shown later, that the splitting<br />

of the Eu surface emission is probably related with hybridization (see ch. 4.2.3).<br />

Examining the effect of surface termination on 4f bulk emission, only a part of the<br />

angle-resolved spectrum (integrated spectrum between |8 ◦ − 9 ◦ | in order to reduce the<br />

impact of hybridization, marked in fig. 4.8) has been chosen <strong>for</strong> evaluation of the multiplet<br />

maxima. After the integration, the Fermi level of both spectra were aligned (cf. inset<br />

in fig. 4.9). Although the 4f multiplet is <strong>for</strong>mally attributed to the bulk, the positions<br />

of the multiplet lines in fig. 4.9 seem to depend on the surface termination. Thereby,<br />

the position of the 4f 6 subsurface multiplet <strong>for</strong> a Si terminated surface is shifted by<br />

about 33 meV towards higher binding energies as compared to the respective emission<br />

from a Eu terminated surface. On the one hand, it may be related with the charge<br />

transfer from the Eu surface layer to the outermost subsurface layers, that is missing<br />

<strong>for</strong> a Si terminated surface. On the other hand, comparing the binding energies of the<br />

multiplet lines relative to the highest level in binding energy (cf. in fig. 4.9, schemata on<br />

the right-hand side) suggests that the 4f state of the Eu subsurface atom hybridizes at<br />

the selected emission angle partly with the VB states. There<strong>for</strong>e a detailed discussion<br />

on bulk / surface hybrid states will be given in ch. 4.2.


4.1 Overview – properties and classification 37<br />

Figure 4.10: projected Fermi surface onto the surface BZ [001] <strong>for</strong> EuRh 2 Si 2 . The measurement<br />

(hν = 53 eV, linear vertical polarization, T ≈ 15 K) is shown in grey shades. The red<br />

marked areas represent the k z -projected bulk band structure (see text <strong>for</strong> details), green (blue)<br />

lines correspond to surface states <strong>for</strong> Si (Eu) termination.<br />

Valence band surface emission<br />

For itinerant electrons, one can characterize surface bands (SB), which are energetically<br />

shifted bulk bands [55], and surface states (SS), which are located inside a bulk band<br />

gap [56]. The <strong>for</strong>mer arise due to the asymmetry of charge density (missing bonds<br />

at the surface); the latter are confined states at the surface between the <strong>for</strong>bidden<br />

region in the bulk (band gap) and the surface potential. Representatives of both are<br />

depicted <strong>for</strong> Si as well as <strong>for</strong> Eu terminated surfaces in fig. 4.11 (high symmetry cut<br />

along Γ − X − M − Γ) and in fig. 4.10 (“Fermi surface”) based on Slab calculations 3<br />

with at least 11 Eu layers. Comparing the SB and SS <strong>for</strong> both, it is evident that the<br />

surface contribution of Si terminated surfaces is dominated by a SS which has the shape<br />

of a star located around the M-point in the surface BZ (see fig. 4.10, labelled as 1a in<br />

fig. 4.11). In case of Eu terminated surfaces a similar feature in the calculation is absent.<br />

For both terminations, there evolves a SB at the Γ-point with rather linear dispersion<br />

(2a in fig. 4.11), which will be discussed in chapter 4.3.<br />

In figure 4.10, an experimental energy surface taken at 53 eV and linear vertical<br />

polarization in the vicinity of the Fermi level (left) is compared to the calculated SS and<br />

SB of the Si (center; green) and the Eu terminated surface (right; blue) superimposed by<br />

3 To separate the projected band structure from SS and SB, an analysis of the eigenfunction <strong>for</strong> each<br />

ɛ(k) is made. If the sum of all basis coefficients dedicated to the surface is 5 times higher than<br />

the sum of all other coefficients of similar structures devided by the number of bulk layers, the<br />

respective Kohn-Sham eigenfunction is labelled as surface-originated. A linear interpolation <strong>for</strong> the<br />

ratio of 1 to 5 times has been used to obtain a smooth transition from bulk to surface. For the sake<br />

of simplicity, the coefficients of the 4 topmost layers have been interpreted as possibly surface-based:<br />

Eu-Si-Rh-Si (Si-Rh-Si-Eu) <strong>for</strong> Eu (Si) terminated surface, respectively.


38 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.11: a superposition of bulk and surface states as expected <strong>for</strong> different surface terminations;<br />

the k z -projected bulk band structure is depicted in shades of maroon (small k z -dispersion: dark maroon;<br />

large k z − dispersion: light maroon) overlayed by the bulk (grey lines) and surface states<br />

(coloured points) of the respective slab calculation; in (a) the Si terminated and in (b) the Eu<br />

terminated surface is depicted; <strong>for</strong> comparison some SS and SB are labelled, whereat in the case of<br />

the latter the dashed lines point to bulk and the continous lines to corresponding surface features;<br />

1a/b in (a) mark the star-like surface state (cf. fig.4.10), 2 marks the conical SB around the Γ [in (a)<br />

as well as in (b)]


4.1 Overview – properties and classification 39<br />

(a) FS: 33 (b) FS: 34 (c) FS: 35<br />

Figure 4.12: Fermi sheets (FS) of EuRh 2 Si 2 ; outer face is depicted in red, the inner face in<br />

blue. They are probably not comparable to the “true” bulk Fermi sheets, if the 4f are in the<br />

range of the Fermi level. The numbering of the eigenvalues (sorted) is arbitrary originated in the<br />

distinction between valence and core orbitals (fplo 9.07.41, LDA, 4f 7 unpolarized open core).<br />

the calculated 4 projected bulk band structure (red). The border of the BZ is marked by<br />

white-dashed lines. Regarding the bulk Fermi surface (cf. fig. 4.12), one recognizes that<br />

the isosurface projected along the [001] direction consists mainly of Fermi sheet 34 and<br />

35 representing the connected square-like structure around Γ with a gap at the M-point.<br />

Apparently, the intensities of the experimental bulk emuissions seem to be inverse to<br />

the calculated ones <strong>for</strong> the first BZ, but similar to the calculation in the second BZ.<br />

This points to selection rules (best seen at the BZ border in the measurement), which<br />

probably can be simulated by means of a sophisticated PE model. As already mentioned<br />

be<strong>for</strong>e, inside the bulk band gap around the M-point resides an electron-like SS at<br />

the Si terminated surface. In the measurement, there seem to be two nested states,<br />

whereas the calculation reproduces only one. But regarding the structures labelled<br />

1a and 1b in fig. 4.11 one recognices that the second surface state (1b) is above the<br />

Fermi level in the calculation, hence it does not (severely) contribute to the shown<br />

isoenergy surface. The deviation can probably be explained by a surface relaxation,<br />

because the distances of the topmost layers are usually smaller than the corresponding<br />

bulk intervals (cf. fig. 4.6). Moreover, there is a lobe-like SB oriented from the Γ-point<br />

towards the M-point at Si terminated surfaces which motivates the observed intensityvariation<br />

around Γ in the first BZ. The nodal point of the surface state 2 in fig. 4.11 is<br />

above the Fermi level <strong>for</strong> Si termination (a), but below <strong>for</strong> Eu termination (b), which is in<br />

good agreement with the measurement depicted in fig. 4.8 despite of that the calculated<br />

SS seems to be weaker at Eu terminated surfaces. This has a more technical than<br />

physical reason since fixing the 4f occupation and treating these orbitals as open core,<br />

one reduces the freedom to generate an asymmetry in charge density at the surface. For<br />

4 The projection onto the surface BZ has been obtained by a superposition of isosurfaces perpendicular<br />

to the z-axis of the BZ. A Gaussian was used <strong>for</strong> energy broadening, which has been chosen around<br />

15 meV being in the order of magnitude of the integration interval chosen <strong>for</strong> the measured spectrum.<br />

The two major Fermi sheets which contribute to the projected bulk band structure, are depicted<br />

in fig. 4.12, FS 34 and FS 35.


40 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.13: similar slab configurations<br />

compared to fig. 4.11 substituting<br />

europium by strontium. Besides<br />

the additional weight <strong>for</strong> the Sr terminated<br />

setup, there are no substantial<br />

differences to the calculations<br />

based on europium (fplo 9.07.41,<br />

LDA)<br />

(a) Si terminated surface configuration.<br />

The surface states 1a and 1b<br />

at the M-point and the linear dispersive<br />

state around Γ are identical<br />

to the ones of the open core calculations.<br />

(b) Sr terminated surface configuration.<br />

In difference to the calculation<br />

with fixed 4f basis orbitals,<br />

two surface states with quasi linear<br />

dispersion emerge around Γ labelled<br />

as 2a and 2b. One could speculate,<br />

whether this are the bands which are<br />

distinguishable in the measured spectrum.<br />

Si termination this is not crucial since the first Eu layer is 5 Å below the topmost layer<br />

and a SSs / SBs usually reside in a range of ±5 Å perpendicular to the surface. But<br />

<strong>for</strong> an Eu layer on top, the corresponding charge density can only partially evolve. To<br />

illustrate this issue, the calculation has been repeated replacing Eu by Sr yielding the<br />

iso-structural compound SrRh 2 Si 2 . Strontium is a good substitute <strong>for</strong> europium having<br />

the same valency and a similar atomic radius. Given that none of the valence orbitals<br />

in Sr are restricted, the freedom to choose a charge relocation at the surface is larger<br />

than <strong>for</strong> Eu and reproduces the linear SS <strong>for</strong> both, Si as well as Sr terminated surfaces<br />

(cf. fig. 4.13).<br />

In summary, it has been shown that EuRh 2 Si 2 reveals remarkable differences between<br />

the surface and bulk-derived band structures as well as a non-negligible k z -<br />

dispersion which demands a three-dimensional description comparable to other 122-<br />

compounds [57]. Explicitly, the main SS and SB have been determined <strong>for</strong> the [001]-<br />

surface and compared to PE data. It is evident that the description of two noninteracting<br />

electron systems (an itinerant – VB and a localized one) is not sufficient<br />

to understand the PE spectra, and there<strong>for</strong>e probably the ground state and low-energy<br />

excitations as well, which govern macroscopic properties like heat transport or conductivity.<br />

Thus, the following section will trace a route to what extent both calculations<br />

can be merged.


4.2 Hybridization: localized versus itinerant states 41<br />

Figure 4.14: the orbital contributions to the band structure <strong>for</strong> the direction (0,0,0)-<br />

(0,0.580·π/a x ,0) are shown illustrating the hybridization of the 4f levels in the LDA+U scheme.<br />

The linear combinations of spherical harmonics with even m l [blue] do not mix in contrast to<br />

the pairs (4f y(3x 2 −y 2 ), 4f yz 2) [red] and (4f xz 2, 4f x(x 2 −3y 2 )) [green] with odd m l (fplo 9.07.41,<br />

LDA+U [AL], U = 8 eV, J = 1 eV). Red and green mark similar symmtries. The related linear<br />

combinations of the complex spherical harmonics Y m l<br />

l<br />

are given.<br />

4.2 Hybridization: localized versus itinerant states<br />

To understand the phenomenology of the interactions between localized and itinerant<br />

electrons, one needs on the one hand in<strong>for</strong>mation on the hybridization strength and on<br />

the other hand knowledge of the symmetry of coupling orbitals. We have already seen<br />

in the LSDA+U calculation, that level crossings (allowed and avoided) are reproduced<br />

in this scheme (cf. fig. 4.4) <strong>for</strong> both subsystems.<br />

4.2.1 Symmetry considerations<br />

Assuming that the chosen L(S)DA+U [AL] functional is more suitable than the pure<br />

L(S)DA approximation, one can try to extract the symmetry from the <strong>for</strong>mer by a basis<br />

trans<strong>for</strong>mation. To emphasize bonds in molecules, usually hybrid orbitals, so-called<br />

molecular orbitals, are constructed. A similar trans<strong>for</strong>mation exists <strong>for</strong> Bloch bands<br />

– the construction of Wannier Functions [58–60]. In principle, they are an orthogonal<br />

basis set localized in real space and based on the Fourier trans<strong>for</strong>m of Bloch bands.<br />

Since there is a gauge freedom of the Bloch phase, the definition of Wannier orbitals<br />

is not unique. To elimenate the degree of freedom, one can choose maximally localized<br />

Wannier functions (WFs) or another fixed construction principle [60]. Here, the


42 4 EuRh 2 Si 2 – semi-localized electrons<br />

–<br />

–<br />

WF FPLO orbital used energy contribution by the nextneighbour<br />

contribution by<br />

<strong>for</strong> projection window<br />

[eV]<br />

Rh sites the next-neighbour<br />

Si sites<br />

(a) 4f x(x 2 −3y 2 )−4f xz 2 [-0.2, 5.5] Rh x : 4d z 2, 4d xz , 4d x2 −y 2<br />

∼ 4f x 3<br />

Rh y : 4d xy , 4d xz<br />

(b) 4f y(3x2 −y 2 )+4f yz 2 [-0.2, 5.5] Rh x : 4d xy , 4d xz<br />

∼ 4f y 3<br />

Rh y : 4d z 2, 4d xz , 4d x 2 −y 2<br />

(c) 4f x(x 2 −3y 2 )+4f xz 2 [-0.2, 5.5] Rh x : 4d z 2, 4d xz , 4d x 2 −y 2 3p x , 3p y , 3p z<br />

∼ 4f x(z 2 −y 2 )<br />

Rh y : –<br />

(d) 4f y(3x 2 −y 2 )−4f yz 2 [-0.2, 5.5] Rh x : –<br />

3p x , 3p y , 3p z<br />

∼ 4f y(z2 −x 2 )<br />

Rh y : 4d z 2, 4d xz , 4d x2 −y 2<br />

(e) 4f xyz [4.0, 5.0] – 3s, 3p x , 3p y , 3p z<br />

(f) 4f z 3 [-1.7, -2.3] Rh x : 4d z 2, 4d x 2 −y 2, 4d xz –<br />

Rh y : 4d z 2, 4d x 2 −y 2, 4d yz<br />

(g) 4f z(x 2 −y 2 ) [-2.5, 3.5] Rh x : 4d z 2, 4d x2 −y 2, 4d xz<br />

Rh y : 4d z 2, 4d x2 −y 2, 4d yz<br />

–<br />

Table 4.2: Symmetry in<strong>for</strong>mation of the obtained WF localized at the Eu<br />

site. The second and third column define the projection operator U k nµ <strong>for</strong> the<br />

respective WF. The positions “off x” and “off y” are relative to the Eu site at<br />

which the WF is localized. Contributions to a WF which were smaller than<br />

10% of the maximum one have been neglected. The positions <strong>for</strong> Rh x and<br />

Rh y are sketched on the right hand side.<br />

scheme implemented in FPLO is used resulting in highly localized WFs [57, 61], whose<br />

construction will be described shortly. The n’th Bloch band in momentum space<br />

Ψ k n ≡ Ψ n (k) = ∑ Bsµ<br />

c kn<br />

sµ Φ Bsµ e ik(B+s) (4.1)<br />

whereat Φ Bsµ denotes the local orbital basis set, gets trans<strong>for</strong>med to the WF of type µ<br />

localized at R:<br />

W Rµ =<br />

V<br />

(2π)<br />

∫BZ<br />

3 d 3 k e ∑ −ikR Ψ k nUnµ k (4.2)<br />

n<br />

(V denotes the volume of unit cell in real space)<br />

The related projection operator Unµ k has to be chosen in such a manner, that the resulting<br />

WF is localized and has the symmetry we intended to get. As the atomic<br />

basis orbitals are already localized, we can define a test function χ (an arbitrary linear<br />

combination of the basis orbitals respecting the space group’s symmetry) and evaluate<br />

the projection of the Kohn-Sham eigenfunctions onto χ. This measure defines to what<br />

extent the Kohn-Sham function resembles χ which determines the composition of eigenfunctions<br />

Ψ k n. There<strong>for</strong>e Unµ k acts as a selector <strong>for</strong> the Kohn-Sham functions used to<br />

create the WF. Furthermore, one can define an energy window to separate bonding and<br />

anti-bonding orbital configurations. To demonstrate this procedure, one example will<br />

be given: if we only choose a sole basis orbital to define the projector, and do not limit


4.2 Hybridization: localized versus itinerant states 43<br />

the energy window, we will get a WF which yields exactly this particular atomic orbital,<br />

because <strong>for</strong> each k-point at least one Kohn-Sham function has a major contribution by<br />

this orbital.<br />

To create a (minimal) WF basis set, one should define molecular orbitals, in this<br />

case mainly hybrids of Eu 4f and Rh 4d as it will be shown later, and check if the WF<br />

Hamiltonian reproduces the band structure inside the chosen energy window. Because<br />

this is a tough task, and we are only interested in the in<strong>for</strong>mation of hybridizing orbitals,<br />

a different approach has been applied. Defining a projector <strong>for</strong> each Eu 4f orbital and<br />

restricting the energy to this particular band of the band structure should define a WF<br />

which consists of the 4f orbital and its symmetry related counterpart at other sites (due<br />

to “hybridization”). This rough procedure is solely suitable, because the dispersion of the<br />

4f dominated bands is very small and all 4f levels can be separated in energy. Internally,<br />

FPLO uses a real representation <strong>for</strong> the complex spherical harmonics of 4f orbitals (the<br />

“general set”). Since the 4f levels are not decoupled (cf fig. 4.14), we use a different<br />

superposition (similar to the “cubic set”) to create a representation so that all orbitals<br />

are well separated by energy and / or symmetry.<br />

In fig. 4.15 the 4f orbitals with respect to the applied projector are depicted, the<br />

corresponding parameters are given in tab. 4.2. In that the energy windows are chosen<br />

arbitrarily, the amount of hybridization cannot be compared between the WFs. Nevertheless,<br />

comparing the WFs in fig. 4.15 the Eu 4f orbitals seem to hybridize primarily<br />

with Rh 4d states despite of the ones in (c) and (e) although silicon is their nearestneighbour.<br />

If the occupation of the 4f orbitals in LSDA+U [AL] reflects the ground<br />

state occupancy, one could restrict the analysis to partially- and entirely-filled orbitals.<br />

But since this scheme does not include spin-orbit coupling and respective excited final<br />

states have mixed occupations of all 4f orbitals, this is not suitable. Regarding tab. 4.2,<br />

one can at least conclude that the Rh 4d z 2 orbital possibly match some contribution to<br />

a hybrid orbital because of its orientation towards the Eu layers and its occurance in<br />

all WFs.<br />

4.2.2 Estimation of the hybridization strength<br />

Depending on the symmetry analysis, one has two possible options to get a first approximation<br />

<strong>for</strong> the interaction strength. If one cannot relate the 4f orbitals with special<br />

linear combinations of Rhodium 4d orbitals by e.g. a basis trans<strong>for</strong>mation or group theory,<br />

then solely a quantative estimation within muffin-tin methods remains. Otherwise<br />

the basis coefficients of the Rh 4d orbitals can be used directly as an estimate <strong>for</strong> the<br />

coupling strength. The first option will be discussed below.<br />

Remembering that in LMTO-ASA no interstitial regions exist, the redistribution of<br />

charge density during the convergence process compared to the initial guess of a superposition<br />

of atomic solutions causes non-vanishing contributions of initially unoccupied<br />

orbitals in the considered sphere as well as in the surrounding ones. Since in our cal-


44 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.15: Wannier functions generated by FPLO and rendered with blender [62] using the projectors defined in tab. 4.2. The major contribution<br />

(and there<strong>for</strong>e identical to the projection orbital) at the Eu site is (a): 4f x 3, (b): Eu 4f y 3, (c): 4f x(z 2 −y 2 ) , (d): 4f y(z 2 −x 2 ) , (e): Eu 4fxyz, (f): 4f z 3 and<br />

(g): 4f z(x 2 −y 2 ) . The sections depicted represent very low iso-energy levels because otherwise the symmetry at neighbouring sites would not be apparent.<br />

There<strong>for</strong>e the 4f orbitals are rather large compared to their size in which an electron can be found with 90% probability. Tails of the WFs originated at<br />

sites not belonging to the unit cell have been omitted.


4.2 Hybridization: localized versus itinerant states 45<br />

culation the localized 4f electrons are treated as core orbitals, we introduce additional,<br />

higher-lying 5f-orbitals to the basis set of the Eu sphere, because they only differ in the<br />

radial wave function in comparison to the 4f orbitals. Bonds 5 of 4f-like electrons with<br />

the itinerant VB should be apparent by additional 5f weight inside the Eu sphere after<br />

the self-consistency process. Doubtless, this weight depends particularly on the chosen<br />

ratio of the atomic spheres and there<strong>for</strong>e the result can only be regarded in a qualitative<br />

way. Thus different compounds are only comparable if their spherical ratios are similar.<br />

But since the ratios are used to adopt the calculated band structure to experiment, it<br />

is a firm task.<br />

The weight’s distribution of the Eu 5f is depicted in fig. 4.16a <strong>for</strong> Si and Eu terminated<br />

surfaces considering surface and subsurface emission. The respective Eu spheres from<br />

which the 5f weight has been taken are highlighted in the insets. The surface originated<br />

states S1 (Si) / S3 (Eu) are shifted to higher binding energies <strong>for</strong> both configurations,<br />

but the bulk projected bands are similar to that of the FPLO calculations. Consequently<br />

it is not possible to use the coefficients <strong>for</strong> the surface band structure directly which<br />

could be related with the states shown in the PE spectrum. A comparison between the<br />

Rh 4d characters of FPLO and the distribution of the 5f weight in LMTO is used as a<br />

starting point <strong>for</strong> the modelling of the coupling strength assuming that a large portion<br />

of 5f weight initially stems from Rh 4d which in fact is supported by the WF analysis.<br />

To rate the hybridization you thus use largly the distribution of the 5f weight of LMTO<br />

and shift the corresponding surface bands / states according to the results obtained by<br />

FPLO. This will be discussed seperately <strong>for</strong> both configurations:<br />

(a) The 5f weight distribution <strong>for</strong> the Si terminated surface is dominated by two<br />

triangularly-shaped areas A1 and A2 (see fig. 4.16a).<br />

These are, if we compare<br />

them to the FPLO calculation, based on the Rh 4d yz surface and bulk component<br />

<strong>for</strong> silicon terminated surfaces (cf. fig. 4.16b). Since the number of bands in the<br />

projected band structure of a slab calculation depends on the size of the slab, an<br />

analytical approximation (given below) is used. For A1 a superposition of parabola<br />

and <strong>for</strong> A2 a linear combination of cosine is applied. To emphasize the origin as<br />

projected band structure, calculations with variable number of bands are made (cf.<br />

fig. 4.19a-c). Moreover, the surface state S1 is shifted according to the FPLO calculation<br />

(mainly Rh 4d xy ), so the nodal point appears 0.2 eV above the Fermi level.<br />

Additionally, the weight of the projected bulk band structure below the surface<br />

state S1 is taken into account (cf. Rh 4d xy ), because the 5f weight of LMTO in<br />

this region is shared between several eigenenergy values <strong>for</strong> one k-point. The resulting<br />

distribution of the coupling parameter V ij (k) in Γ-X direction is presented<br />

in fig. 4.17a.<br />

used dispersions (k ‖ in [π/a x ] and ɛ(k ‖ ) in [eV]):<br />

5 bonds illustrate overlapping wave functions of different sites, which are signatures of “charge shifts”<br />

inside solids compared to spherical symmetric atomic configurations


46 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.16: (a) 5f weight of the Eu layers in LMTO <strong>for</strong> a slab calculation which contribute to<br />

the respective surface termination. In contrast to the FPLO calculation presented in fig. 4.11, no<br />

distinction between surface / bulk Kohn-Sham functions has been made, which is reflected by a<br />

mixture of “surface” and “bulk” components especially <strong>for</strong> the Si terminated surface. The leading<br />

contributions have been marked <strong>for</strong> discussion (lmto 5.01.1, LDA, 4f 7 unpolarized open core)<br />

(b) surface contribution of Rhodium 4d states in the Si as well as in the Eu terminated slab<br />

calculation. The surface state with quasi linear dispersion has mainly 4d xy character and the<br />

nodal point is above (below) the Fermi level <strong>for</strong> Si (Eu) termination. Due to the partially-fixed<br />

basis set, the linear surface state is only indicated <strong>for</strong> 4d xy states. Below, the bulk contribution<br />

of the Si terminated slab is shown (identical to that of Eu termination) (fplo 9.07.41, LDA, 4f 7<br />

unpolarized open core)


4.2 Hybridization: localized versus itinerant states 47<br />

Figure 4.17: coupling strength depending<br />

on the surface termination<br />

(a) hybridization strength <strong>for</strong> the<br />

Si terminated surface as a superposition<br />

of surface and projected bulk<br />

contributions<br />

(b) weight distribution of the hybridization<br />

<strong>for</strong> Eu terminated surface<br />

which is mainly based on surface<br />

contributions<br />

white S1 : ɛ(k ‖ ) = −5.8 · k ‖ + 0.1<br />

white S1 projected: ɛ(k ‖ ) = −5.8 · k ‖ − 0.5 · i/30, i ∈ [0, 29]<br />

white A1 : ɛ(k ‖ ) = 23 · (k ‖ − 0.25) 2 − 0.75 + i/50, i ∈ [0, 49]<br />

white A2 : ɛ(k ‖ ) = 0.03 · cos(30 · k ‖ ) − 0.6 · i/26 − 0.15, i ∈ [0, 25]<br />

(b) Since the surface and subsurface emission are originated in different slab layers <strong>for</strong><br />

Eu termination, one has to evaluate both positions. The two distributions depicted<br />

in fig. 4.16a are rather similar, thus in both cases S2 and S3 are dominating<br />

(see fig. 4.16a). Both are shifted in comparison to the Rh 4d xy and Rh 4d yz surface<br />

contribution, so their nodal points remain below the Fermi level. Due to the similarity<br />

the 5f weights were not chosen differently <strong>for</strong> surface and subsurface emission.<br />

The corresponding distribution is depicted in fig. 4.17 b.<br />

used dispersions (k ‖ in [π/a x ] and ɛ(k ‖ ) in [eV]):<br />

whiteS3 : ɛ(k ‖ ) = −5.8 · k ‖ − 0.3<br />

whiteS2 part1 : ɛ(k ‖ ) = −30 · k‖ 2 ⎧<br />

− 0.6<br />

⎪⎨<br />

−5.8 · k ‖ − 0.1 ‖k ‖ ‖ < 0.19<br />

whiteS2 part2 : ɛ(k ‖ ) = −5.8 · k ‖ + 2.5 · (k ‖ − 0.19) 2 − 0.2 0.19 < ‖k ‖ ‖ < 0.38<br />

⎪⎩<br />

−1.6 ‖k ‖ ‖ > 0.38<br />

In the subsurface layer, additional weight of projected band structure can be seen,<br />

which seem to have minor influence (in comparison to the measured spectrum, see<br />

fig. 4.8). At a first glance it is not obvious, that in case of Eu termination no<br />

effect of the projected band structure is observed. It may be argued, that since<br />

the mean free scattering path of electrons is in the same range as the depth of<br />

the subsurface layer <strong>for</strong> Eu termination, the influence of the bulk band structure is<br />

much weaker in comparison to the Si termination. Perhaps it can be evaluated at<br />

higher photon energies increasing the mean free path if the instrumental resolution<br />

admits to resolve the final state multiplet structure.<br />

In the following, the motivated coefficients V ik (k) will be used to introduce an interaction<br />

between the two electronic subsystems. Due to the analytical expressions, not<br />

all details have been taken into account.


48 4 EuRh 2 Si 2 – semi-localized electrons<br />

4.2.3 The hybridization model<br />

To this end, a simple hybridization model 6 (already demonstrated in [67, 68]) will<br />

be applied. Therewith band gaps and hybrid states can be assessed treating the hybridization<br />

as a small pertubation to the pre-calculated atomic final state PE spectrum.<br />

Furthermore the transfer of spectral weight to the Fermi level can be estimated. The<br />

corresponding Hamiltonian<br />

H =<br />

N∑<br />

ɛ i d † i d ∑N f<br />

i + ˜ɛ i f † i f i<br />

i<br />

} {{ }<br />

(1)<br />

i<br />

} {{ }<br />

(2)<br />

N,N<br />

∑ f<br />

+<br />

i,j<br />

V ij<br />

(f † j d i + d † i f j<br />

)<br />

} {{ }<br />

(3)<br />

N∑<br />

f ,N f<br />

+<br />

i≠j<br />

C ij<br />

(f † j f i + f † i f j<br />

)<br />

} {{ }<br />

(4)<br />

(4.3)<br />

consists of two basic count terms, one <strong>for</strong> itinerant states (1) and one <strong>for</strong> localized<br />

states (2), and a hopping term (3) describing the probability <strong>for</strong> mixing 7 . In addition,<br />

there is a fourth term (4) <strong>for</strong> symmetry relations between the localized states. Regarding<br />

them as possibly energy degenerate states and using fermionic operators, contradicts<br />

itself and does not allow to determine any quantitative measure <strong>for</strong> occupation. Strictly<br />

speaking, knowing that the final states of PE are many-body states, the usage of a<br />

fermionic operator <strong>for</strong> each of them is not justified. Nevertheless knowing this peculiarity<br />

and neglecting further many-body effects, you gain qualitative corrections to the<br />

eigenenergies as long as the coupling strength V ij is small enough. The k-dependence<br />

of H is solely introduced by the parameters C ij , V ij . There<strong>for</strong>e, the Hamiltonian can<br />

be evaluated in the following one-particle pseudo basis set<br />

| d j , k 〉 = d † jk<br />

| 0 〉 . . . itinerant valence state<br />

| f j , k 〉 = f † jk<br />

| 0 〉 . . . localized state<br />

whereat | 0 〉 denotes the free vacuum state. Since many-body effects are already covered<br />

by LDA (generally: exchange-correlation functional in DFT) and corresponding interactions<br />

are included in Gerken’s et. al. [49] calculation of the Eu PE spectrum as well,<br />

6 In general, the task would have been to solve the Anderson model [63]. It involves two sorts of<br />

electrons: itinerant and localized ones. Furthermore, it includes interaction between them (hybridization)<br />

and adds auxiliary Coulomb repulsion between the localized electrons if they occupy<br />

the same orbital at one site, the “correlation” depends on the degeneracy of the orbital. Since the<br />

Eu impurities are translational invariant, one should include this periodicity, which is not possible<br />

without severe approximations [64–66]. There<strong>for</strong>e the extracted hybridization model has been used.<br />

A motivation on how to renormalize the periodic Anderson model to obtain a hybridization model<br />

is given in [44].<br />

7 This model is not limited to a coupling between localized and itinerant states. In principle, there is<br />

a term proportional to the number operator <strong>for</strong> each species and one describing the mixing of them,<br />

whereat the dispersion of the states can be arbitrary, because ɛ/˜ɛ as well as V ij are k-dependent.<br />

Nevertheless, there exists no intrinsic mixing between different k-dependent variables (in contrast<br />

to the Anderson model).


4.2 Hybridization: localized versus itinerant states 49<br />

justifies the choice of the basis set. Due to this simplification, the matrix representation<br />

can be directly written as:<br />

⎛<br />

⎞<br />

ɛ 1 0 . . . V 11 . . . V Nf 1<br />

. 0 .. . . .. .<br />

. ɛ Nd V 1Nd . . . V Nf N d<br />

V 11 . . . V 1Nd ˜ɛ 1 C 12 . . . C 1Nf<br />

(4.4)<br />

. . C .. . .. ⎜ . .. 21 .<br />

.<br />

.<br />

⎝<br />

. .. . .. ⎟<br />

CNf −1N f<br />

⎠<br />

V Nf 1 . . . V Nf N d<br />

C Nf 1 . . . C Nf N f −1 ˜ɛ Nf<br />

This matrix is k-dependent and so the solution yields a rearrangement of the eigenenergies<br />

<strong>for</strong> each k-point corresponding to the hybridization matrix elements V ij , which<br />

are chosen proportional to the overlap of the hybridizing states. The number of valence<br />

states N d corresponds to the number of eigenenergie values in the respective energy<br />

range of the band structure calculation (itinerant electron system). In contrast to that,<br />

the number of localized levels N f and their mixing C ij depends on the interaction of<br />

different levels in photoemission, or more precise, on their origin. As mentioned be<strong>for</strong>e,<br />

the m J -degeneracy can be lifted by an electro-magnetic field whereat the field strength<br />

is proportional to the splitting (cf. ch. 4.1.4). The different symmetry considerations<br />

shall be illustrated exemplarily by the atomic PE spectra of Yb and Eu.<br />

Ytterbium [68–71] exhibits as well as europium [49, 67, 72, 73] a complex PE multiplet<br />

structure which has been extensively studied. Since Rare-Earth elements are metals,<br />

they act mainly as electron donors in compounds. This charge redistribution with<br />

respect to the simple superposition of free atoms’ charges causes an increasing electric<br />

field due to the charge separation which lifts the m J -degeneracy. This is known as<br />

crystal electric field splitting. In EuRh 2 Si 2 , this effect cannot be resolved because the<br />

intensity of 4f emission at the Fermi level is rather weak. Additionally, the resolution<br />

gets worse with increasing binding energy due to finite lifetime effects of final states in<br />

photoemission. On the contrary, in YbRh 2 Si 2 the 4f 7/2 multiplet component is located<br />

in the range of the Fermi level and exhibits a resolvable splitting into four different<br />

levels [74]. Since these are mixed representations which lift the 4f shell degeneracy,<br />

they are partially coupled to each other and accordingly the model should respect their<br />

symmetry. In this case N f equals four having two representation ( Γ i t6 and Γi t7 ) with<br />

two elements each. To adopt the symmetry relations, each representation is coupled<br />

only once to the VB and a repulsive “<strong>for</strong>ce” is added between elements of the same<br />

representation because they are <strong>for</strong>bidden to be degenerate (Pauli principle). This<br />

has already been presented by Vyalikh et. al. [68] (thereafter called model 1 “coupled<br />

localized levels”). Contrariwise, the coupling between states with different J is negligible<br />

and thus a superposition of the hybridized spectra of each final state configuration with


50 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.18: comparison of different hybridization<br />

models: on the left-hand side the input is<br />

displayed followed by the eigenvalue redistribution,<br />

and the photoemission spectrum on the<br />

right-hand side (left: f emission, right: VB emission).<br />

As initial point, two localized levels<br />

(ef = [−0.15, −1.15] eV) and a parabolic ) hole-<br />

(−10 ∗ k<br />

‖ 2 + 0.2<br />

like band (ɛ(k) =<br />

eV) are used.<br />

Since atomic PE emission can have different transition<br />

probabilities, the spectral weight of the localized<br />

levels was chosen as 1 : 2 to demonstrate<br />

this influence and a broadening proportional to<br />

σ(ɛ) = 100/15 · (ɛ/5 + 1) eV was applied (ɛ in<br />

units of binding energy).<br />

(a) superposition of localized levels: the hybridization model has been solved two times, <strong>for</strong> each localized level once, the eigenvalue distributions of<br />

which are shown next to the initial configuration. The superposition of both represents the final eigenvalue dispersion. The hybridization parameter<br />

has been chosen constant as Vij = 0.15 eV.<br />

(b) coupled localized levels: in difference to the <strong>for</strong>mer model, hybridization gaps evolve <strong>for</strong> both levels and due to the coupling between both also<br />

their absolute distance gets larger <strong>for</strong> regions where the hybridization with the VB can be neglected. Vij is the same as in (a) with Cij = 0.2 eV


4.2 Hybridization: localized versus itinerant states 51<br />

the VB yields a more appropriate model [67] (denoted by model 2 “superposition of<br />

localized levels”).<br />

To illustrate this distinction, both models are presented in fig. 4.18. Exemplarily, two<br />

localized levels differing by a factor of two in spectral intensity and a hole-like parabolic<br />

VB with its apex located 0.2 eV above the Fermi level are taken. On the left-hand side<br />

in (a) as well as in (b) the initial configuration is depicted followed by the rearranged<br />

eigenenergy distribution. A sketch of both components in model 2 the sum of which<br />

represents the spectrum, clarifies the degeneracy of the spectral eigenvalues in regions<br />

of the VB separated from the localized levels. The main disparity between both is the<br />

occurance of a band gap in the case of coupled localized levels, which is not present in<br />

case of superposition. Furthermore, due to the constant symmetry parameter C ij also<br />

the localized states in model 1 will be pushed apart if no valence band approaches. A<br />

feasible correction would be to make it proportional to the distance between the valence<br />

band and the second localized level or solving the model self-consistently re-adjusting<br />

the input parameters of the localized levels so the solution fits in the limit of ɛ i → ∞<br />

to the original PE spectrum.<br />

In the following part, the motivated model 2 <strong>for</strong> europium (hence C ij (k) = 0 and<br />

N f = 1) will be evaluated <strong>for</strong> Si and Eu terminated surfaces. As input <strong>for</strong> the subsurface<br />

PE levels the spectrum calculated by Gerken et. al shifted by 0.15 eV is used. In the<br />

case of a Eu terminated surface, the surface 4f emission is chosen proportional to the<br />

subsurface emission adopted in intensity (6x) and energy position (shifted by 1 eV to<br />

higher binding energies) to the experiment. The distribution of V ij (k), the evolution<br />

of which already has been sketched, is given in fig. 4.17. To verify the coupling to the<br />

projected band structure <strong>for</strong> the Si terminated surface, different numbers of bands have<br />

been taken to simulate the transition from a single band to a continuously-projected<br />

band structure. On the one hand, this effect displaces spectral weight to the edges of the<br />

region where the VB is located and on the other hand distributes the residual weight in<br />

the coupled area yielding an almost homogeneous intensity distribution. In fig. 4.19, this<br />

process is indicated <strong>for</strong> the triangularly shaped area A1 next to the linear dispersive state<br />

at Γ. The final spectra <strong>for</strong> Si and Eu terminated surfaces in Γ−X direction are depicted<br />

in fig. 4.20. The main spectral features are well-reproduced <strong>for</strong> both terminations.<br />

Besides, the measured spectrum <strong>for</strong> Si termination evidences additional interaction with<br />

the projected band structure especially in the region marked by 2. Furthermore, there<br />

has to be a some part of the band structure which causes the accumulation of spectral<br />

weight at the tip of the state S1 (1) not evidenced in the calculated spectrum. Because<br />

only f emission is taken into account in the simulation, the weak VB like contribution (3)<br />

is missing. For the Eu terminated surface, the shift of surface 4f emission with respect<br />

to the subsurface emission has been estimated imprecisely, as well as its width. The<br />

latter is probably related to a different multiplet structure, because the potential is no<br />

longer rotationally symmetric as it can be regarded in bulk as a first approximation.


52 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.19: simulated spectra<br />

based on the distribution of V ij (k)<br />

derived in ch. 4.2.2;<br />

(a) to (c) the evolving triangularlyshaped<br />

area A1 by increasing the<br />

number of projected bands from<br />

1 to 50 bands, respectively<br />

(d) hybridization of the linear surface<br />

state around Γ with the final<br />

state multiplet, whereat the apex<br />

remains above the Fermi level<br />

(e) projected bulk band structure<br />

in A2 which hybridizes with the<br />

4f states to <strong>for</strong>m a shaded area below<br />

the surface state S1<br />

(f) final spectrum containing all<br />

parts of the band structure which<br />

contribute to the hybrid states<br />

Figure 4.20: measured dispersion <strong>for</strong> X−Γ−M and modelled spectra <strong>for</strong> Eu and Si terminated<br />

surfaces. The colormaps are only vaguely adopted since two different programs have been used<br />

to create the figures. (a) show the simulated f emission and (b) the measured spectrum (hν =<br />

120 eV, linear vertical polarization, T ≈ 10 K) dominated due to the cross section by f character<br />

<strong>for</strong> Si terminated surfaces; (c) + (d) depict the respective results <strong>for</strong> the Eu terminated surface.<br />

Differences between measurement and simulation are marked by numbers and discussed in the<br />

text.


4.3 Perspective: quasi-linear dispersion in a 4f compound 53<br />

Figure 4.21: dispersion of the linear surface state S1 around Γ measured<br />

<strong>for</strong> several cuts parallel to Γ−X towards the X-point. The different<br />

cuts are sketched in (c). In (a) are the results <strong>for</strong> Si terminated surface<br />

and in (b) <strong>for</strong> Eu terminated surface depicted. (SLS-SIS)<br />

There<strong>for</strong>e using atomic excitation spectra is only a zeroth-order approach. In principle, a<br />

model <strong>for</strong> atomic 4f emission with additional, non-spherical potential could address both<br />

shortcomings. Furthermore, the additional spectral weight at Γ cannot be explained (4).<br />

Nevertheless, the splitting of the surface 4f emission evident in fig. 4.9 can be explained<br />

in the framework of hybridization, because the surface state S2 seems to displace the<br />

spectral weight to higher and lower binding energies with a simultaneously emerging<br />

hybridization gap (5).<br />

As presented, the PE spectra show mainly surface states in hybridization with the<br />

4f multiplet of europium. Since the momentum perpendicular to the surface is not<br />

conserved, shaded bands (part of the projected band structure) emerge additionally. To<br />

disentangle the bulk band structure successfully, a less surface sensitive method has to<br />

be chosen. On the other hand, PE is a good choice to describe surface and edge states.<br />

This is especially important in a new class of solids, the surface states of which have<br />

macroscopically-different properties – e.g. the bulk is an insulator whereat the surface<br />

is a metal. A short outline will be given in the next chapter.<br />

4.3 Perspective: quasi-linear dispersion in a 4f compound<br />

Dimensionality and thereby confined electronic states are of recent interest in current<br />

research [75–78]. Especially after certain topological insulators have been predicted [76]<br />

and experimentally explored [77], the surface of which is stable metallic, a new field of<br />

research has emerged <strong>for</strong> spintronic and magnetoelectric devices. The topological classification<br />

scheme [76, 79] allows to determine whether the surface states are expected to<br />

be stable under small pertubation (e.g. disorder). It is expected, that due to supressed<br />

back-scattering in the case of strong topological insulators [75], dissipationless transport


54 4 EuRh 2 Si 2 – semi-localized electrons<br />

Figure 4.22: dispersion in X − Γ direction <strong>for</strong> different chosen k z . For k z = [0.0 − 0.285] · π/a x ,<br />

there exists a band gap at Γ which closes at k z = 0.285 · π/a x being exactly the k-point where<br />

a three-fold degenerate eigenenergy value near the Fermi level occurs in the Γ − Z dispersion<br />

(cf. fig. 4.3). When the bands cross, a “Dirac Cone” seems to emerge. Going further to Z, the<br />

band gap vanishes.<br />

occurs at the surface. Similar states have been observed also in ironpnictides the groundstate<br />

of which is metallic, hence it is an ongoing discussion whether the argumentation<br />

made <strong>for</strong> insulators is transferable to intermetallics [80–82]. Many experimentally-based<br />

publications claim, that there is a connection between the Dirac-like dispersion (microscopic<br />

property) and transport (macroscopic property). The major difficulty in their<br />

argumentations is the lacking knowledge of competing effects (e.g. impurity scattering,<br />

magnetic order, correlation between different electronic subsystems) of such complex<br />

systems. There<strong>for</strong>e a short reasoning is given here, why it is worth to investigate the<br />

origin of the linear dispersive state around Γ further and what investigations could be<br />

done.<br />

In fig. 4.21 the dispersion of the linear surface states <strong>for</strong> several cuts parallel to<br />

k x are shown. It emphasizes a fourth-fold symmetry and there are evidences from<br />

observed projected Fermi surfaces (not shown), that the quasi-Dirac cone is rather<br />

de<strong>for</strong>med and that a section of the dispersion in the k x × k y plane can be regarded as a<br />

superposition of two ellipsoids. The Fermi velocity amounts to (3.0±1.0) eV Å≈ 10 −3 ·c<br />

[(2.5 ± 1.0) eV Å≈ 10 −3 · c] (c being the speed of light) <strong>for</strong> Si [Eu] termination which<br />

is three times smaller than in graphene [83]. In the latter transport is dominated by<br />

the Dirac cone but <strong>for</strong> intermetallics, since there are several bands intersecting the<br />

Fermi level, it is not obvious and deserves a careful study.<br />

As it has been demonstrated be<strong>for</strong>e (cf. ch. 4.1.4), the quasi-linear states seem to<br />

be of surface origin. To explain their possible evolution, the bulk band structure is regarded<br />

again. It reveals a three-fold degenerate eigenenergy value in going from Γ to Z<br />

(cf. fig. 4.3) near the Fermi level, which evidences a rather steep slope in the Γ − M direction<br />

(in bulk <strong>for</strong> example: Z−Γ 3 ). To examine this part further, different paths parallel<br />

to the k x × k y plane are depicted in fig. 4.22 demonstrating the k z dispersion. For<br />

k z = [0.0 − 0.285] · π/a x , there exists a band gap at Γ which closes at k z = 0.285 · π/a x .<br />

Arriving at that plane, the degeneracy seems to <strong>for</strong>ce a linear dispersive behaviour<br />

in the vicinity of k 1 = (0, 0, 0.285) · π/a x . Since the projected band structure <strong>for</strong> the


4.3 Perspective: quasi-linear dispersion in a 4f compound 55<br />

[0, 0, 1]-surface is the weighted sum of the dispersion from all k x × k y planes, also this<br />

particular linear dispersion should contribute to it. Thus the linear surface states <strong>for</strong><br />

both terminations in the vicinity of Γ can perhaps be regarded as surface bands, being<br />

similar to a special k x × k y plane but shifted in energy due to unsaturated bonds at<br />

the surface related to the bulk. The linearity of the surface states has not been studied<br />

extensively within DFT / LDA so far, but it appears to have a strong dependence on<br />

the regarded number of slab layers which determines the complexity of the projected<br />

band structure in the calculation. There<strong>for</strong>e a proper analysis <strong>for</strong> the semi-infinite bulk<br />

has to be done. Additionally, the Berry curvature seems to play a crucial role [84, 85]<br />

<strong>for</strong> such phenomena, and thus it should provide an insight on whether respective states<br />

are protected against small pertubations and to what extent qualitatively a correction<br />

to the Fermi velocity should occur. Due to competing processes in transport (scattering<br />

on magnetically-ordered Eu 4f, possible Dirac cone, disorder) it is not yet clear<br />

without ambiguity, if macroscopic traces of the surface states can be measured. Hence<br />

to simplify the task, one could try to substitute europium by strontium removing the<br />

local 4f impurity.


57<br />

5 Summary<br />

In this diploma thesis angle-resolved photoemission spectroscopy and density functional<br />

theory based calculations have been used to study the ternary rare-earth compound<br />

EuRh 2 Si 2 . Thereby the emphasis was laid on many-body interactions exploring the<br />

interplay of two opposing limits <strong>for</strong> the description of electrons: itinerant character vs.<br />

localization. Symmetry dependencies of photoemission as depicted in fig. 5.1 have been<br />

neglected.<br />

The crystal cleaves solely along the Eu–Si planes which has been demonstrated theoretically<br />

by <strong>for</strong>ce-minimization. The amount of europium atoms at the surface of the<br />

sample is not controllable, there<strong>for</strong>e one has to search <strong>for</strong> regions at the surface which<br />

are predominantly terminated by Si or Eu atoms. The kind of termination may be<br />

identified by the surface component of the Eu 4f emission which is shifted by approximately<br />

1 eV to higher binding energies whereat it is missing <strong>for</strong> Si termination. On<br />

the other hand, a star-like surface state centered around the M-point of the quadratic<br />

surface BZ located in a gap of the projected Fermi surface manifests Si termination in<br />

agreement with results of slab calculations. For the first time, a linear dispersive Diraclike<br />

state around the Γ-point has been observed <strong>for</strong> a rare earth compound. It seems<br />

to be of surface origin and has its nodal point close above (below) the Fermi energy at<br />

Si (Eu) terminated surfaces. This observation is in accordance with the surface calculations<br />

(4f treated as core orbitals) indicating that the pertubation of the eigenvalues due<br />

Figure 5.1: PE spectra<br />

of EuRh 2 Si 2 with<br />

Si termination <strong>for</strong> different<br />

polarisations<br />

at hν = 100 eV and<br />

T ≈ 2 K, aligned roughly<br />

in X − Γ − X direction<br />

(1 3 ARPES (BESSYII))<br />

(a) has been measured<br />

with horizontal polarisation<br />

and (b) with linear<br />

vertical polarisation. In<br />

comparison to SLS-SIS<br />

the polarization effect<br />

seems to be swapped<br />

(cf. fig. 4.8a)


58 5 Summary<br />

to the f-d interaction is small. Whether this Dirac-like state demonstrates changes in<br />

the macroscopic properties and whether the interplay of correlated electrons with that<br />

state effects transport features could not be proven.<br />

The observed spectra have been reproduced by a simple hybridization model using<br />

the calculated band structures and atomic Eu 4f PE spectra. In doing so, a Wannier<br />

function analysis of the Eu 4f orbitals in a LSDA+U calculation revealed that they<br />

mainly hybridize with surrounding 4d orbitals of Rhodium. Based on that, the hybridization<br />

matrix elements have been chosen proportional to the overlap of Eu 4f and<br />

Rh 4d orbitals. The calculated spectra are in accordance with the observed ones. The<br />

contribution of subsurface emissions to the PE spectra depends on the surface termination.<br />

Hereby, the bulk sensitivity is apparently larger in the case of Si termination,<br />

what may possibly be related to the small mean-free inelastic scattering path. In contrast<br />

to YbRh 2 Si 2 [68], hybridization gaps are not evident (though they exist <strong>for</strong> each<br />

final state) since the difference in binding energy of the final states is an order of magnitude<br />

smaller in EuRh 2 Si 2 and there<strong>for</strong>e the superposition of final state spectra covers<br />

them. The absence of 4f emission at the Fermi level can probably associated to either<br />

<strong>for</strong>bidden hybridization in the ground state (which is unlikely) or a finite difference<br />

(above thermal excitations) between the Fermi energy and the 4f one particle energy in<br />

the ground state.<br />

A detailed evaluation in the framework of the Anderson model is missing which could<br />

shed light on the issue of the many-body ground state whose structure is not accessible<br />

by experiment.


59<br />

Bibliography<br />

[1] Jun Kondo. Resistance Minimum in Dilute Magnetic Alloys. Progress of Theoretical<br />

Physics, 32(1):37–49, 1964. doi:10.1143/PTP.32.37.<br />

[2] F. Steglich, J. Aarts, C. D. Bredl, W. Lieke, D. Meschede, W. Franz, and H. Schäfer.<br />

Superconductivity in the Presence of Strong Pauli Paramagnetism: CeCu 2 Si 2 . Phys. Rev.<br />

Lett., 43:1892–1896, Dec 1979. doi:10.1103/PhysRevLett.43.1892.<br />

[3] Z Hossain, O Trovarelli, C Geibel, and F Steglich. Complex magnetic order in EuRh 2 Si 2 .<br />

Journal of Alloys and Compounds, 323-324:396 – 399, 2001. Proceedings of the 4th International<br />

Conference on f-Elements. doi:DOI:10.1016/S0925-8388(01)01095-7.<br />

[4] A. Błachowski, K. Ruebenbauer, J. Żukrowski, K. Rogacki, Z. Bukowski, and J. Karpinski.<br />

Shape of spin density wave versus temperature in AFe 2 As 2 (A = Ca, Ba, Eu): A Mössbauer<br />

study. Phys. Rev. B, 83(13):134410, Apr 2011. doi:10.1103/PhysRevB.83.134410.<br />

[5] O. Trovarelli, C. Geibel, S. Mederle, C. Langhammer, F. M. Grosche, P. Gegenwart,<br />

M. Lang, G. Sparn, and F. Steglich. YbRh 2 Si 2 : Pronounced Non-Fermi-Liquid Effects<br />

above a Low-Lying Magnetic Phase Transition. Phys. Rev. Lett., 85(3):626–629, Jul 2000.<br />

doi:10.1103/PhysRevLett.85.626.<br />

[6] R. Movshovich, T. Graf, D. Mandrus, J. D. Thompson, J. L. Smith, and Z. Fisk. Superconductivity<br />

in heavy-fermion CeRh 2 Si 2 . Phys. Rev. B, 53(13):8241–8244, Apr 1996.<br />

doi:10.1103/PhysRevB.53.8241.<br />

[7] D. V. Vyalikh, S. Danzenbächer, A. N. Yaresko, M. Holder, Yu. Kucherenko, C. Laubschat,<br />

C. Krellner, Z. Hossain, C. Geibel, M. Shi, L. Patthey, and S. L. Molodtsov. Photoemission<br />

Insight into Heavy-Fermion Behavior in YbRh 2 Si 2 . Phys. Rev. Lett., 100:056402, Feb 2008.<br />

doi:10.1103/PhysRevLett.100.056402.<br />

[8] S.L. Molodtsov, S. Danzenbächer, Yu. Kucherenko, C. Laubschat, D.V. Vyalikh, Z. Hossain,<br />

C. Geibel, X.J. Zhou, W.L. Yang, N. Mannella, Z. Hussain, Z.-X. Shen, M. Shi,<br />

and L. Patthey. Hybridization of 4f states in heavy-fermion compounds YbRh2Si2 and<br />

YbIr2Si2. Journal of Magnetism and Magnetic Materials, 310(2, Part 1):443 – 445, 2007.<br />

Proceedings of the 17th International Conference on Magnetism<br />

The International Conference on Magnetism. doi:<br />

10.1016/j.jmmm.2006.10.501.<br />

[9] P. Gegenwart, T. Westerkamp, C. Krellner, Y. Tokiwa, S. Paschen, C. Geibel, F. Steglich,<br />

E. Abrahams, and Q. Si. Multiple energy scales at a quantum critical point. SCIENCE,<br />

315(5814):969–971, FEB 16 2007. doi:{10.1126/science.1136020}.<br />

[10] D. V. Vyalikh, S. Danzenbächer, Yu. Kucherenko, C. Krellner, C. Geibel, C. Laubschat,<br />

M. Shi, L. Patthey, R. Follath, and S. L. Molodtsov. Tuning the Hybridization at the<br />

Surface of a Heavy-Fermion System. Phys. Rev. Lett., 103(13):137601, Sep 2009. doi:<br />

10.1103/PhysRevLett.103.137601.<br />

[11] S. Danzenbächer, D. V. Vyalikh, and K. Kummer. Insight into the f-derived Fermi surface<br />

of the heavy-fermion compound YbRh 2 Si 2 . unpublished, 2011.<br />

[12] <strong>Max</strong> Born and J. Robert Oppenheimer. Zur Quantentheorie der Molekeln. Annalen der<br />

Physik, 84:457––484, 1927.


60 Bibliography<br />

[13] B. Kramer. A Pseudopotential Approach <strong>for</strong> the Green’s Function of Electrons in<br />

Amorphous <strong>Solid</strong>s. physica status solidi (b), 41(2):649–658, 1970. doi:10.1002/pssb.<br />

19700410220.<br />

[14] P. Hohenberg and W. Kohn. Inhomogeneous Electron Gas. Phys. Rev., 136(3B):B864–<br />

B871, Nov 1964. doi:10.1103/PhysRev.136.B864.<br />

[15] W. Kohn and L. J. Sham. Self-Consistent Equations Including Exchange and Correlation<br />

Effects. Phys. Rev., 140(4A):A1133–A1138, Nov 1965. doi:10.1103/PhysRev.140.A1133.<br />

[16] Helmut Eschrig. The Fundamentals of Density Functional Theory. Edition am Gutenbergplatz<br />

Leipzig, 2003.<br />

[17] Richard M. Martin. Electronic Structure – Basic Theory and Practical Methods. Cambridge<br />

University Press, 2004. Available from: http://electronicstructure.org/.<br />

[18] G.D. Barrera, D. Colognesi, P.C.H. Mitchell, and A.J. Ramirez-Cuesta. LDA or GGA? A<br />

combined experimental inelastic neutron scattering and ab initio lattice dynamics study<br />

of alkali metal hydrides. Chemical Physics, 317(2-3):119 – 129, 2005. doi:10.1016/j.<br />

chemphys.2005.04.027.<br />

[19] L.A. Palomino-Rojas, M. López-Fuentes, Gregorio H. Cocoletzi, Gabriel Murrieta, Romeo<br />

de Coss, and Noboru Takeuchi. Density functional study of the structural properties of<br />

silver halides: LDA vs GGA calculations. <strong>Solid</strong> <strong>State</strong> Sciences, 10(9):1228 – 1235, 2008.<br />

doi:10.1016/j.solidstatesciences.2007.11.022.<br />

[20] Axel D. Becke. Density-functional thermochemistry. III. The role of exact exchange. Journal<br />

of Chemical Physics, 98:5648–5652, 1993. doi:10.1063/1.464913.<br />

[21] John P. Perdew and Yue Wang. Accurate and simple analytic representation of the<br />

electron-gas correlation energy. Phys. Rev. B, 45(23):13244–13249, Jun 1992. doi:<br />

10.1103/PhysRevB.45.13244.<br />

[22] Felix Bloch. Über die Quantenmechanik der Elektronen in Kristallgittern. Zeitschrift für<br />

Physik A Hadrons and Nuclei, 52:555–600, 1929. 10.1007/BF01339455. doi:10.1007/<br />

BF01339455.<br />

[23] O. Krogh Andersen. Linear methods in band theory. Phys. Rev. B, 12(8):3060–3083, Oct<br />

1975. doi:10.1103/PhysRevB.12.3060.<br />

[24] Klaus Koepernik and Helmut Eschrig. Full-potential nonorthogonal local-orbital minimumbasis<br />

band-structure scheme. Phys. Rev. B, 59(3):1743–1757, Jan 1999. doi:10.1103/<br />

PhysRevB.59.1743.<br />

[25] O. K. Andersen and R. V. Kasowski. Electronic <strong>State</strong>s as Linear Combinations of Muffin-<br />

Tin Orbitals. Phys. Rev. B, 4(4):1064–1069, Aug 1971. doi:10.1103/PhysRevB.4.1064.<br />

[26] J. Korringa. On the calculation of the energy of a Bloch wave in a metal. Physica, 13(6-<br />

7):392 – 400, 1947. doi:DOI:10.1016/0031-8914(47)90013-X.<br />

[27] W. Kohn and N. Rostoker. Solution of the Schrödinger Equation in Periodic Lattices<br />

with an Application to Metallic Lithium. Phys. Rev., 94(5):1111–1120, Jun 1954. doi:<br />

10.1103/PhysRev.94.1111.<br />

[28] Stefan Hüfner. Photoelectron Spectroscopy, volume 82 of Springer Series in <strong>Solid</strong>-<strong>State</strong><br />

Sciences. Springer-Verlag, 1995.<br />

[29] C. N. Berglund and W. E. Spicer. Photoemission Studies of Copper and Silver: Experiment.<br />

Phys. Rev., 136(4A):A1044–A1064, Nov 1964. doi:10.1103/PhysRev.136.A1044.


Bibliography 61<br />

[30] C. N. Berglund and W. E. Spicer. Photoemission Studies of Copper and Silver: Theory.<br />

Phys. Rev., 136(4A):A1030–A1044, Nov 1964. doi:10.1103/PhysRev.136.A1030.<br />

[31] Stephan Kümmel and Leeor Kronik. Orbital-dependent density functionals: Theory and<br />

applications. Rev. Mod. Phys., 80(1):3–60, Jan 2008. doi:10.1103/RevModPhys.80.3.<br />

[32] T. Koopmans. Über die Zuordnung von Wellenfunktionen und Eigenwerten zu den<br />

einzelnen Elektronen eines Atoms. Physica, 1(1-6):104 – 113, 1934. doi:10.1016/<br />

S0031-8914(34)90011-2.<br />

[33] Börje Johansson and Nils Mårtensson. Core-level binding-energy shifts <strong>for</strong> the metallic<br />

elements. Phys. Rev. B, 21(10):4427–4457, May 1980. doi:10.1103/PhysRevB.21.4427.<br />

[34] S. L. Molodtsov, S. V. Halilov, V. D. P. Servedio, W. Schneider, S. Danzenbächer, J. J.<br />

Hinarejos, Manuel Richter, and C. Laubschat. Cooper Minima in the Photoemission Spectra<br />

of <strong>Solid</strong>s. Phys. Rev. Lett., 85(19):4184–4187, Nov 2000. doi:10.1103/PhysRevLett.<br />

85.4184.<br />

[35] D. H. Tomboulian and P. L. Hartman. Spectral and Angular Distribution of Ultraviolet<br />

Radiation from the 300-Mev Cornell Synchrotron. Phys. Rev., 102(6):1423–1447, Jun 1956.<br />

doi:10.1103/PhysRev.102.1423.<br />

[36] Israel Felner and Israel Nowik. Itinerant and local magnetism, superconductivity and<br />

mixed valency phenomena in RM 2 Si 2 , (R = rare earth, M = Rh, Ru). Journal of Physics<br />

and Chemistry of <strong>Solid</strong>s, 45(4):419 – 426, 1984. doi:10.1016/0022-3697(84)90149-5.<br />

[37] E V Sampathkumaran, L C Gupta, R Vijayaraghavan, K V Gopalakrishnan, R G Pillay,<br />

and H G Devare. A new and unique Eu-based mixed valence system: EuPd 2 Si 2 . Journal<br />

of Physics C: <strong>Solid</strong> <strong>State</strong> Physics, 14(9):L237, 1981. doi:10.1088/0022-3719/14/9/006.<br />

[38] N. Mårtensson, B. Reihl, W. D. Schneider, V. Murgai, L. C. Gupta, and R. D. Parks. Highly<br />

resolved surface shifts in a mixed-valent system: EuPd 2 Si 2 . Phys. Rev. B, 25(2):1446–1448,<br />

Jan 1982. doi:10.1103/PhysRevB.25.1446.<br />

[39] N. J. Speer, M. K. Brinkley, Y. Liu, C. M. Wei, T. Miller, and T.-C. Chiang. Surface<br />

vs. bulk electronic structure of silver determined by photoemission. EPL (Europhysics<br />

Letters), 88(6):67004, 2009. doi:10.1209/0295-5075/88/67004.<br />

[40] K. Held, I. A. Nekrasov, N. Blümer, V. I. Anisimov, and D. Vollhardt. Realistic Modeling<br />

of Strongly Correlated Electron Systems: An Introduction to The LDA+DMFT Approach.<br />

International Journal of Modern Physics B: Condensed Matter Physics; Statistical Physics;<br />

Applied Physics, 15(19/20):2611, 2001. doi:10.1142/S0217979201006495.<br />

[41] S. Biermann, F. Aryasetiawan, and A. Georges. First-Principles Approach to the Electronic<br />

Structure of Strongly Correlated Systems: Combining the GW Approximation<br />

and Dynamical Mean-Field Theory. Phys. Rev. Lett., 90:086402, Feb 2003. doi:<br />

10.1103/PhysRevLett.90.086402.<br />

[42] H. Eschrig, K. Koepernik, and I. Chaplygin. Density functional application to strongly<br />

correlated electron systems. Journal of <strong>Solid</strong> <strong>State</strong> Chemistry, 176(2):482 – 495, 2003.<br />

Special issue on The Impact of Theoretical Methods on <strong>Solid</strong>-<strong>State</strong> Chemistry. doi:10.<br />

1016/S0022-4596(03)00274-3.<br />

[43] M. T. Czyżyk and G. A. Sawatzky. Local-density functional and on-site correlations: The<br />

electronic structure of La 2 CuO 4 and LaCuO 3 . Phys. Rev. B, 49:14211–14228, May 1994.<br />

doi:10.1103/PhysRevB.49.14211.


62 Bibliography<br />

[44] T. M. Rice and K. Ueda. Gutzwiller Variational Approximation to the Heavy-Fermion<br />

Ground <strong>State</strong> of the Periodic Anderson Model. Phys. Rev. Lett., 55:995–998, Aug 1985.<br />

doi:10.1103/PhysRevLett.55.995.<br />

[45] R. Gumeniuk, M. Schmitt, C. Loison, W. Carrillo-Cabrera, U. Burkhardt, G. Auffermann,<br />

M. Schmidt, W. Schnelle, C. Geibel, A. Leithe-Jasper, and H. Rosner. Boron induced<br />

change of the Eu valence state in EuPd 3 B x (0 ≤ x ≤ 0.53) : A theoretical and experimental<br />

study. Phys. Rev. B, 82(23):235113, Dec 2010. doi:10.1103/PhysRevB.82.235113.<br />

[46] M. D. Johannes and W. E. Pickett. Magnetic coupling between nonmagnetic ions: Eu3+<br />

in EuN and EuP. Phys. Rev. B, 72(19):195116, Nov 2005. doi:10.1103/PhysRevB.72.<br />

195116.<br />

[47] Helmut Eschrig, Alexander Lankau, and Klaus Koepernik. Calculated cleavage behavior<br />

and surface states of LaOFeAs. Phys. Rev. B, 81(15):155447, Apr 2010. doi:10.1103/<br />

PhysRevB.81.155447.<br />

[48] B. Chevalier, J. M. D. Coey, B. Lloret, and J. Etourneau. EuIr 2 Si 2 : a new intermediate<br />

valence compound. Journal of Physics C: <strong>Solid</strong> <strong>State</strong> Physics, 19(23):4521, 1986. doi:<br />

10.1088/0022-3719/19/23/015.<br />

[49] F Gerken, A S Flodström, J Barth, L I Johansson, and C Kunz. Surface Core Level Shifts<br />

of the Lanthanide Metals Ce 58 -Lu 71 : A Comprehensive Experimental Study. Physica<br />

Scripta, 32(1):43, 1985. doi:10.1088/0031-8949/32/1/006.<br />

[50] Harald Friedrich. Theoretische Atomphysik. Springer-Verlag, Berlin | Heidelberg | New<br />

York, 1994, 2. Auflage.<br />

[51] O. Gunnarsson and K. Schönhammer. Electron spectroscopies <strong>for</strong> Ce compounds in the<br />

impurity model. Phys. Rev. B, 28(8):4315–4341, Oct 1983. doi:10.1103/PhysRevB.28.<br />

4315.<br />

[52] R. Hayn, Yu. Kucherenko, J.J. Hinarejos, S.L. Molodtsov, and C. Laubschat. Simple<br />

numerical procedure <strong>for</strong> the spectral function of 4f photoexcitations. Physical Review B,<br />

64:115106, 2001. doi:10.1103/PhysRevB.64.115106.<br />

[53] J.-M. Imer and E. Wuilloud. A simple model calculation <strong>for</strong> XPS, BIS and EELS 4fexcitations<br />

in Ce and La Compounds. The European Physical Journal B - Condensed<br />

Matter and Complex Systems, 66:153–169, 1987.<br />

[54] En-Jin Cho, Se-Jung Oh, S. Suga, T. Suzuki, and T. Kasuya. Electronic structure study<br />

of Eu intermetallic compounds by photoelectron spectroscopy. Journal of Electron Spectroscopy<br />

and Related Phenomena, 77(2):173 – 181, 1996. doi:10.1016/0368-2048(95)<br />

02495-6.<br />

[55] Ig. Tamm. Über eine mögliche Art der Elektronenbindung an Kristalloberflächen.<br />

Zeitschrift für Physik, 76:849–850, 1932. 10.1007/BF01341581. doi:10.1007/BF01341581.<br />

[56] William Shockley. On the Surface <strong>State</strong>s Associated with a Periodic Potential. Phys. Rev.,<br />

56:317–323, Aug 1939. doi:10.1103/PhysRev.56.317.<br />

[57] Helmut Eschrig and Klaus Koepernik. Tight-binding models <strong>for</strong> the iron-based superconductors.<br />

Phys. Rev. B, 80:104503, Sep 2009. doi:10.1103/PhysRevB.80.104503.<br />

[58] Gregory H. Wannier. The Structure of Electronic Excitation Levels in Insulating Crystals.<br />

Phys. Rev., 52:191–197, Aug 1937. doi:10.1103/PhysRev.52.191.<br />

[59] N Marzari, I. Souza, and D. Vanderbilt. An introduction to maximally-localized Wannier<br />

functions. Psi-K Scientific Highlight of the Month, 57:129–168, 2003. Available from:<br />

http://wannier.org/papers/MSVpsik.pdf.


Bibliography 63<br />

[60] Nicola Marzari and David Vanderbilt. <strong>Max</strong>imally localized generalized Wannier functions<br />

<strong>for</strong> composite energy bands. Phys. Rev. B, 56:12847–12865, Nov 1997. doi:10.1103/<br />

PhysRevB.56.12847.<br />

[61] Klaus Koepernik. Wannier functions with FPLO, February 2010. Available from: http:<br />

//www.fplo.de/download/wan_user.pdf.<br />

[62] open source 3D content creation suite. Available from: http://www.blender.org.<br />

[63] P. W. Anderson. Localized Magnetic <strong>State</strong>s in Metals. Phys. Rev., 124:41–53, Oct 1961.<br />

doi:10.1103/PhysRev.124.41.<br />

[64] M. Potthoff, M. Aichhorn, and C. Dahnken. Variational Cluster Approach to Correlated<br />

Electron Systems in Low Dimensions. Phys. Rev. Lett., 91:206402, Nov 2003. doi:10.<br />

1103/PhysRevLett.91.206402.<br />

[65] Gabriel Kotliar and Dieter Vollhardt. Strongly Correlated Materials: Insights from Dynamical<br />

Mean-Field Theory. Physics Today, 57(3):53–59, 2004. doi:10.1063/1.1712502.<br />

[66] V. Moskalenko, L. Dohotaru, and R. Citro. Diagram theory <strong>for</strong> the periodic anderson<br />

model: Stationarity of the thermodynamic potential. Theoretical and Mathematical<br />

Physics, 162:366–382, 2010. 10.1007/s11232-010-0029-z. doi:10.1007/<br />

s11232-010-0029-z.<br />

[67] S. Danzenbächer, D. V. Vyalikh, Yu. Kucherenko, A. Kade, C. Laubschat, N. Caroca-<br />

Canales, C. Krellner, C. Geibel, A. V. Fedorov, D. S. Dessau, R. Follath, W. Eberhardt,<br />

and S. L. Molodtsov. Hybridization Phenomena in Nearly-Half-Filled f-Shell Electron<br />

Systems: Photoemission Study of EuNi 2 P 2 . Phys. Rev. Lett., 102(2):026403, Jan 2009.<br />

doi:10.1103/PhysRevLett.102.026403.<br />

[68] D. V. Vyalikh, S. Danzenbächer, Yu. Kucherenko, K. Kummer, C. Krellner, C. Geibel,<br />

M. G. Holder, T. K. Kim, C. Laubschat, M. Shi, L. Patthey, R. Follath, and S. L.<br />

Molodtsov. k Dependence of the Crystal-Field Splittings of 4f <strong>State</strong>s in Rare-Earth<br />

Systems. Phys. Rev. Lett., 105(23):237601, Dec 2010. doi:10.1103/PhysRevLett.105.<br />

237601.<br />

[69] Wolf-Dieter Schneider, Clemens Laubschat, and Bruno Reihl. Temperature-dependent<br />

microstructure of evaporated Yb surfaces. Phys. Rev. B, 27:6538–6541, May 1983. doi:<br />

10.1103/PhysRevB.27.6538.<br />

[70] L. I. Johansson, J. W. Allen, I. Lindau, M. H. Hecht, and S. B. M. Hagström. Photoemission<br />

from Yb: Valence-change-induced Fano resonance. Phys. Rev. B, 21:1408–1411, Feb 1980.<br />

doi:10.1103/PhysRevB.21.1408.<br />

[71] I. Abbati, L. Braicovich, C. Carbone, J. Nogami, I. Lindau, I. Iandelli, G. Olcese, and<br />

A. Palenzona. Photoemission studies of mixed valence in Yb 3 Si 3 , YbSi and Yb 5 Si 3 :<br />

Equivalent versus inequivalent Yb sites. <strong>Solid</strong> <strong>State</strong> Communications, 62(1):35 – 39, 1987.<br />

doi:10.1016/0038-1098(87)90079-2.<br />

[72] W. D. Schneider, C. Laubschat, G. Kalkowski, J. Haase, and A. Puschmann. Surface<br />

effects in Eu intermetallics: A resonant photoemission study. Phys. Rev. B, 28:2017–2022,<br />

Aug 1983. doi:10.1103/PhysRevB.28.2017.<br />

[73] W. A. Henle, M. G. Ramsey, F. P. Netzer, and K. Horn. Reversible Eu 2+ ↔ Eu 3+<br />

transitions at Eu-Si interfaces. 58(15):1605–1607, 1991. doi:doi:10.1063/1.105139.<br />

[74] A.M. Leushin and V.A. Ivanshin. Crystalline electric fields and the ground state of<br />

YbRh2Si2 and YbIr2Si2. Physica B: Condensed Matter, 403(5-9):1265 – 1267, 2008.<br />

Proceedings of the International Conference on Strongly Correlated Electron Systems.<br />

doi:10.1016/j.physb.2007.10.122.


64 Bibliography<br />

[75] Hari C. Manoharan. Topological insulators: A romance with many dimensions. Nature<br />

Nanotechnology, 5:477–479, 2010. doi:10.1038/nnano.2010.138.<br />

[76] Liang Fu and C. L. Kane. Topological insulators with inversion symmetry. Physical Review<br />

B, 76:045302, 2007. doi:10.1103/PhysRevB.76.045302.<br />

[77] Y. Xia, D. Qian, D. Hsieh, L. Wray, A. Pal, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J.<br />

Cava, and M. Z. Hasan. Observation of a large-gap topological-insulator class with a single<br />

Dirac cone on the surface. Nature Physics, 06:398–402, 2009. doi:10.1038/nphys1274.<br />

[78] A. Grüneis, C Attaccalite, A. Rubio, D. V. Vyalikh, S. L. Molodtsov, J. Fink, R. Follath,<br />

W. Eberhardt, B. Büchner, and T. Pichler. Angle-resolved photoemission study of graphite<br />

interlaction compound KC 8 : A key to graphene. Phys. Rev. B, 80:075431, 2009. doi:<br />

10.1103/PhysRevB.80.075431.<br />

[79] M. Z. Hasan and C. L. Kane. Colloquium: Topological insulators. Rev. Mod. Phys.,<br />

82:3045–3067, 2010. doi:10.1103/RevModPhys.82.3045.<br />

[80] Khuong K. Huynh, Yoichi Tanabe, and Katsumi Tanigaki. Both Electron and Hole Dirac<br />

Cone <strong>State</strong>s in Ba(FeAs) 2<br />

Confirmed by Magnetoresistance. Phys. Rev. Lett., 106:217004,<br />

May 2011. doi:10.1103/PhysRevLett.106.217004.<br />

[81] Takao Morinari, Eiji Kaneshita, and Takami Tohyama. Topological and Transport Properties<br />

of Dirac Fermions in an Antiferromagnetic Metallic Phase of Iron-Based Superconductors.<br />

Phys. Rev. Lett., 105(3):037203, Jul 2010. doi:10.1103/PhysRevLett.105.037203.<br />

[82] N. Harrison and S. E. Sebastian. Dirac nodal pockets in the antiferromagnetic parent<br />

phase of FeAs superconductors. Phys. Rev. B, 80:224512, 2009. doi:10.1103/PhysRevB.<br />

80.224512.<br />

[83] A. Varykhalov, J. Sánchez-Barriga, A. M. Shikin, C. Biswas, E. Vescovo, A. Rybkin,<br />

D. Marchenko, and O. Rader. Electronic and Magnetic Properties of Quasifreestanding<br />

Graphene on Ni. Phys. Rev. Lett., 101:157601, Oct 2008. doi:10.1103/PhysRevLett.<br />

101.157601.<br />

[84] Di Xiao, Ming-Che Chang, and Qian Niu. Berry phase effects on electronic properties.<br />

Rev. Mod. Phys., 82:1959–2007, Jul 2010. doi:10.1103/RevModPhys.82.1959.<br />

[85] Yugui Yao, Leonard Kleinman, A. H. MacDonald, Jairo Sinova, T. Jungwirth, Dingsheng<br />

Wang, Enge Wang, and Qian Niu. First Principles Calculation of Anomalous<br />

Hall Conductivity in Ferromagnetic bcc Fe. Phys. Rev. Lett., 92:037204, Jan 2004.<br />

doi:10.1103/PhysRevLett.92.037204.


List of Figures 65<br />

List of Figures<br />

2.1 DFT self-consistency scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5<br />

2.2 muffin tin with pastries: the construction of the potential in LMTO . . . . . . . 10<br />

2.3 mean free inelastic scattering path of electrons . . . . . . . . . . . . . . . . . . . 11<br />

2.4 photoemission: one- and three-step model . . . . . . . . . . . . . . . . . . . . . . 12<br />

2.5 the three step model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13<br />

2.6 initial and final states in photoemission . . . . . . . . . . . . . . . . . . . . . . . 14<br />

3.1 manipulator including sample holder inside the preparation chamber: left: raw<br />

sample with lever stick; right: cleaved sample . . . . . . . . . . . . . . . . . . . . 17<br />

3.2 characteristics of the dipole radiation of a charged particle [35, p. 25] . . . . . . . 19<br />

3.3 (a) general setup of a photoemission endstation at synchrotron facilities, (b) on<br />

the left-hand side: the 1 3 ARPES, (c) on the right-hand side: the SIS ARPES<br />

endstation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22<br />

4.1 overview: real space symmetry and slab configurations . . . . . . . . . . . . . . . 24<br />

4.2 overview of the BZ structure: differences between space group 123 – 139 and<br />

their projections onto the surface BZ . . . . . . . . . . . . . . . . . . . . . . . . . 25<br />

4.3 band structure folding: comparison between SPG 123 and SPG 139 . . . . . . . . 26<br />

4.4 comparison: L(S)DA+U vs. open core approximation . . . . . . . . . . . . . . . 28<br />

4.5 charge isosurfaces and lattice configuration of EuRh 2 Si 2 illustrating the charge<br />

transfer inside the unit cell as well as stronger and weaker bonds . . . . . . . . . 30<br />

4.6 calculated cleavage behaviour: interlayer distances and slab representatives . . . 31<br />

4.7 atomic cross section <strong>for</strong> photoemission and an overview spectrum . . . . . . . . . 33<br />

4.8 spectral overview <strong>for</strong> different surface termination measured at hν = 120 eV . . . 34<br />

4.9 characterisation of the 4f final state multiplet – comparison between atomic calculation<br />

and different surface terminations . . . . . . . . . . . . . . . . . . . . . . 35<br />

4.10 projected Fermi surface: experiment and theory . . . . . . . . . . . . . . . . . . . 37<br />

4.11 band structure of EuRh 2 Si 2 along Γ − X − M − Γ with characterization of bulk<br />

and surface originated states (calculation) . . . . . . . . . . . . . . . . . . . . . . 38<br />

4.12 theoretical bulk Fermi surfaces (4f open core) . . . . . . . . . . . . . . . . . . . . 39<br />

4.13 band structure of SrRh 2 Si 2 along Γ − X − M − Γ with characterization of bulk<br />

and surface originated states (calculation) . . . . . . . . . . . . . . . . . . . . . . 40<br />

4.14 symmetry and dispersion of the FPLO 4f basis set (LDA+U) demonstrated in a<br />

particular direction in the BZ <strong>for</strong> EuRh 2 Si 2 . . . . . . . . . . . . . . . . . . . . . 41<br />

4.15 Wannier representation based on Eu 4f orbitals . . . . . . . . . . . . . . . . . . . 44<br />

4.16 hybridization strength: distribution of the 5f weight in LMTO <strong>for</strong> both terminations<br />

and associated states from FPLO calculations . . . . . . . . . . . . . . . . . 46<br />

4.17 analytically-modelled coupling strength <strong>for</strong> both terminations . . . . . . . . . . . 47<br />

4.18 comparison of two different hybridization models: superposition of localized levels<br />

and coupled localized levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50<br />

4.19 evolution of the PE spectrum <strong>for</strong> Si termination taking into account different<br />

contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52<br />

4.20 simulated spectra based on the distribution of V ij (k) in comparison to the experiment<br />

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52<br />

4.21 linear dispersion in the vicinity of the Γ-point <strong>for</strong> Si / Eu termination . . . . . . 53<br />

4.22 k z dispersion of EuRh 2 Si 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54<br />

5.1 polarization dependent PE spectra of EuRh 2 Si 2 . . . . . . . . . . . . . . . . . . . 57


List of Tables 67<br />

List of Tables<br />

2.1 signs and symbols <strong>for</strong> chapter 2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />

4.1 experimental and calculated relaxed lattices constants / Wyckoff positions . . . . 23<br />

4.2 composition of the WFs created by projection of a particular molecular orbital<br />

and energy range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42


Acknowledgements<br />

I am deeply grateful to Prof. C. Laubschat <strong>for</strong> the possibility to do my diploma thesis<br />

in his research group. Particulary, I want to thank <strong>for</strong> the granted freedom to discover<br />

the aspects of photoemission myself and having the chance to freely shape my work in<br />

conjunction with invaluable discussions which motivated me to go further. Moreover,<br />

I thank every member of the group <strong>for</strong> their support, especially Steffen Danzenbächer<br />

and Denis Vyalikh <strong>for</strong> their valuable discussions and cheer-ups in times of struggling<br />

problems.<br />

I am grateful <strong>for</strong> the love, patience and balance which my girlfriend gave me. Her<br />

support strengthens me to proceed my tasks. I apologize <strong>for</strong> my grumbling, nights spent<br />

<strong>for</strong> work and late arrivals at evenings.<br />

Additionally, I thank Yuri Kucherenko, Klaus Koepernik and Helge Rosner <strong>for</strong> valuable<br />

hints on theoretical fundamentals and their patience in our discussions. Without<br />

their theoretical support and knowledge, I would not have been able to write this thesis.


Erklärung<br />

Hiermit versichere ich, dass diese Arbeit selbständig und ohne andere als die angegebenen<br />

Hilfsmittel angefertigt worden ist.<br />

Dresden, 24. November 2011<br />

Marc Höppner

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!