Chapter 9

Theoretical Concepts of
Spin-Orbit Splitting
9.1

Free-electron model

In order to understand the basic origin of spin-orbit coupling at the surface

of a crystal, it is a natural starting point to consider a free-electron model.
The model describes a 2D electron gas (2D-EG) with spin-orbit interaction,
applicable to surface states that are localized in the near surface plane [83].
The model Hamiltonian will be written as
H=

p2
+ HR ,
2m

(9.1)

with the Rashba Hamiltonian

HR = (~ez ~k) ~ .

(9.2)

Solutions of the Schrodinger equation with the Hamiltonian of Eq. 9.1 can
be found analytically. We choose a Cartesian coordinate system with the zaxis perpendicular to the 2D plane of electron motion. It can be shown (see
Appendix C) that the free-electron parabolic dispersion will now be split in two
parabola, shifted in k direction with energies
E1,2 =

~2 k 2
2m

~k

(9.3)

The associated wave functions have the form

~

1 (r) = eik~r (| i + ieik | i)

~

f or E1 =

~2 k2
2m

+ ~k,

~2 k2

(9.4)

(9.5)
~k.
2m
The spin functions | i, | i denote spin-up and spin-down electron states,
respectively, with respect to the z direction; k is the angle between ~k and xaxis, according to geometry illustrated in Fig. 9.1. The linear combination of
2 (r) = eik~r (ieik | i + | i)

85

f or E2 =

86

CHAPTER 9. THEORETICAL CONCEPTS

Figure 9.1: Geometry considered in the free-electron model.

spin-up and spin-down states in the eigenfunctions 9.4 and 9.5 describe a spin
polarization within the xy-plane (see Appendix C), which is oriented perpendicular to the direction of electron motion. This means that, the electrons with
the same k vector and opposite spins will have different energies.
The results derived from this model are illustrated in Fig. 9.2. In the lower
part, the Rashba spin-orbit split energy dispersion of a (free-electron like) surface state is presented. The two concentric circles in the upper part represent
the corresponding Fermi surfaces, with arrows indicating the in-plane spin orientations that are always perpendicular to the electron momentum.
The most important results of the model are: (1) it correctly describes
the nature of the splitting and, (2) it shows that Rashba type spin-orbit interaction will orient the spins of propagating electrons. However, the size of
the splitting, estimated by using the surface potential gradient, is much too
small compared with experimental values. The surface potential gradient can
be roughly approximated by (V )z /F , with the work function and
the Fermi wavelength F [84] . Applied to the case of Au(111) (F = 5
A and
= 4.3 eV), this leads to an estimated value of E = 106 eV, which is several
orders of magnitude smaller than the experimentally observed splitting. This
deficit is inherent in the free electron model, which does not account for regions
of steep nuclear potential gradients near atomic cores.
Shockley-type surface states penetrate as deep as several atomic layers into
the bulk [85]. It has been shown by G. Bihlmayer [70] for the example of the
gold surface state, that regions near atomics cores have also to be taken into
account when describing the Rashba splitting of a surface state. The potential
gradients close to the nucleus of an atom, leading to atomic spin-orbit splitting,
are orders of magnitude larger than the gradient of surface barrier, and its
consideration leads to an increase of the Rashba splitting.
In order to obtain a better understanding of the atomic contributions to
spin-orbit interaction at the surface, a tight-binding model shall be considered
next.

9.2. TIGHT-BINDING MODEL

87

Figure 9.2: Energy dispersion of free-electron state with Rashba splitting bottom part together with the corresponding Fermi surfaces (upper part). Arrows
indicate spin orientations of electrons at the Fermi surface.

9.2

Tight-binding model

The free-electron model is built upon the assumption that electrons propagate
almost freely and that their wave functions can be approximated by plane waves.
As an opposite starting point, one can use localized atomic orbitals as a basis
set to perform band-structure calculations; this is the tight binding (TB) or
linear combination of atomic orbitals (LCAO) approach.
This approach was developed in molecular physics and has been extended
to the description of electronic states in solids. As a starting point, we consider
two identical but separate atoms with their (atomic) wave function for a single
valence electron. As we bring them together, the interaction between the two
valence charges will lead to the formation of new orbitals, the so-called bonding
and antibonding orbitals. The energy of the bonding orbital is lowered by an
amount that is determined by the interaction Hamiltonian, and the energy of
the antibonding orbital is raised by the same amount. The matrix elements of
the interaction Hamiltonian between atomic orbitals are usually referred to as
overlap parameters. They have a simple physical interpretation as interaction
strength between electrons on adjacent atoms. The concept of bonding and
antibonding orbitals is easily extendable to crystals, if we assume that atomic
orbitals of each atom in the crystal overlap with whose of its nearest neighbors
only. This is a reasonable approximation for many solids. As a result of the
overlap (during crystal formation) bonding and antibonding orbitals broaden
forming valence and conductions bands, respectively. Of course, the crystal

88

CHAPTER 9. THEORETICAL CONCEPTS

structure strongly affects the overlap between the atomic orbitals.

Surface states at a crystal surface form a two-dimensional manifold of energy
eigenstates. For the present approach, we take Wannier functions to form a basis
set. Surface states are localized states that decay exponentially into the bulk.
To represent the essential physics of spin-orbit interaction, we use p-orbitals
as an example, keeping the model simple and transparent, and to reveal the
important ingredients of the Rashba effect. For sp-derived surface states, like
those at the Au(111) surface, this is sufficient, since s-states do not contribute
to spin-orbit interaction (zero angular momentum). For Gd and Tb, however,
it is not fully correct to consider p-states, since the surface state in this case
has dz 2 character (derived from the bulk d-band). Construction of the model
with d-orbitals included can be done in full analogy to the case with p-orbitals
as a basic set. However, including d-states will increase the number of basis
functions and will probably conceal the main idea of this consideration, i.e. to
obtain a transparent and clear physical picture of the problem. Instead, we will
rather compare with results of density-functional-theory (DFT) calculations,
see below.
The Gd(0001) and Tb(0001) surfaces considered in this work have a sixfold symmetry at the surface. Accordingly, surface states are described in the
model with sheet of hexagonally arranged atoms using the three atomic orbital
functions px , py , pz , where the in-plane orbitals px and py are coupled in the
usual way by the help of the directional cosine.
Atomic orbitals can be expressed as the product of a radial wave function
and spherical harmonics Ylm (, ), with the atomic nucleus chosen as the origin.
In case of two atoms, it is convenient to choose coordinate axes such that the
z-axis is parallel to the vector d~ connecting the atoms, and the azimuthal angles
are the same. The interaction Hamiltonian will have cylindrical symmetry with
respect to d~ and therefore cannot depend on . Symmetry consideration permits
to gradually reduce the number of overlap parameters, which have to be found
by sorting out zero and equals integrals. For a crystal lattice, it is convenient
to choose crystallographic axes as coordinate axes, and the spherical harmonics
Ylm (, ) of the atomic orbitals are defined with respect to a fixed coordinate
system. In calculating the overlap parameter for any pair of neighboring atoms,
one expands the spherical harmonics defined with respect to d~ in terms of
Ylm (, ). Figure 9.3 illustrates this expansion for p-orbitals.
If one considers the surface layer as an isolated system, px and py would not
couple to the pz orbital. Inversion symmetry z z would lead to an equally
large overlap of a px , py lobe with the positive and negative pz lobe, resulting in
a zero total overlap. At the surface, this symmetry is obviously broken due to
the surface potential, V (z), which breaks the symmetry of the electron density
distribution of surface states with respect to the surface plane. Only the px , py
orbitals will retain the symmetry of their atomic counterparts. This effect can
be included by actually coupling px , py , and pz states.
As was found by Petersen and Hedegard [83], the simplest tight-binding
Hamiltonian that allows us to discuss the influence of SOC (on the 2D electronic
structure) has the form

9.2. TIGHT-BINDING MODEL

89

Figure 9.3: Projection of the overlap parameter between py orbitals onto direc~ From
tions along the vector d~ joining the two atoms and perpendicular to d.
Ref. [86].

H0 =

t (Ri Rj ) |p (Ri ), i hp (Rj ), | ,

(9.6)

where t (Ri Rj ) is the overlap matrix element given by

w cos2 (ij ) + sin2 (ij )

(w + ) cos(ij ) sin(ij )

w sin2 (ij ) cos2 (ij )

t (Ri Rj ) =
cos(ij )

sin(ij )

(, ) = (x, x)
(, ) = (x, y) = (y, x)
(, ) = (y, y)
(, ) = (x, z) = (z, x)
(, ) = (y, z) = (z, y)
(, ) = (z, z)

for Ri and Rj pointing towards nearest-neighbor lattice sites only; all other
matrix elements are zero. ij is the angle between the vector (Ri Rj ) and
the x-axis. Parameter w denotes the Vpp and is a shorthand for Vpp . Vpp
and Vpp are visualized in Fig. 9.3; their physical interpretation is given by
the matrix elements of the interaction Hamiltonian (9.6) that have been calculated [87, 88] for different lattices and can be found in the literature. The
parameter =hpz (R)| V |pn (R + x)i (n = x, y) is essentially a measure of the
surface potential gradient; it has a similar meaning as V
z in the free electron
model discussed above. The case = 0 corresponds to a 2D model with inversion symmetry.
By forming Bloch waves with overlap matrix elements written above and
by diagonalizing the resulting matrix, we can solve the problem given by the
Hamiltonian (9.6). Without taking into account the electron spin, the resulting
matrix will be only 3 3. From here, with /w = 0.1 and /w = 0.3, we
obtain the band structure shown in Fig. 9.4. The band with a minimum at the
point is primarily of pz character. The true surface states correspond to the
free-electron-like part of this band around the point. Also, there are strongly
dispersive bands with minima at the M-point and along the -K direction; they
have primarily px , py character. The gaps appear due to the broken symmetry
with respect to reflection z z, which is controlled by the parameter; the

90

CHAPTER 9. THEORETICAL CONCEPTS

Figure 9.4: Tight-binding-model band structure for surface states in a twodimensional hexagonal lattice, derived from p-orbitals without spin-orbit interaction. The true surface states belong to the free-electron-like band around the
point with primarily pz character; from Ref. [83].
gaps disappear when = 0, and symmetry is restored. Only the free-electronlike band around the point resides inside a gap of the projected bulk band
structure in a real crystal and therefore can be identified as a true surface state.
To introduce spin-orbit interaction into our model, we take the Hamiltonian
+
(L + L + + Lz z ),
(9.7)
2
where L are angular momentum step-up/step-down operators and are the
same for the spin, is the atomic spin-orbit constant. Now, when taking
into account the spin parts of the atomic wave functions used as basis set
{|px , i , |px , i , |py , i , |py , i , |pz , i , |pz , i}, the resulting matrix becomes


 0 i 0
0 0 1 

 i 0
0
0 0 1 


0 0
0 1 i 0 
.
(9.8)
Hsoc = 
2  0 0 1 0 i 0 
 0 0 i i 0 0 


 1 i
0
0 0 0 
Hsoc = L S =

Inspection of the Hsoc matrix directly reveals that, for spin-orbit interaction
to become effective, it is necessary to have px and py states in the model.
Otherwise, as can be seen from the 2 2 block of zeros in the lover right corner
of the matrix that corresponds to pz states, no spin-orbit interaction will be
present.

9.2. TIGHT-BINDING MODEL

91

Figure 9.5: Band structure with spin-orbit splitting for surface states within the
tight-binding model constructed from p-orbitals in a two-dimensional hexagonal
lattice. The true surface states belong to the free-electron-like band around the
point. The splitting is determined by the atomic spin-orbit coupling strength,
.
By including Hsoc in the Hamiltonian (9.6), one can derive the spin-orbitsplit band structure. The result of the calculation with = 0.2w is presented
in Fig. 9.5 .
The idea of this section is not only to demonstrate that the free-electron
model and the tight-binding model can provide some insight into the role of spinorbit interaction in the surface electronic structure, but also to demonstrate the
important contribution of atomic spin-orbit coupling to the Rashba splitting of
surface states. For this purpose, we will limit ourselves to the subspace of
the Hilbert space that is spanned by pz -like states. A simple truncation of
the Hamiltonian to this subspace will give no spin-orbit interaction, since this
requires breaking parity which is only achieved by an admixture of px,y states,
i.e. by a coupling between pz and px,y orbitals. To account for these bands,
these virtual transitions to them should be included to second order in the
coupling. This is possible as long as the pz and the px , py bands are well
separated in energy.
From a mathematical point of view, we follow the work by Petersen and
Hedegard [83]. One writes the resolvent ( H)1 , which is the Laplace transformation of the time evolution operator, and consequently includes all necessary information. Since we consider only the subspace spanned by pz and we are
interested in the time evolution of the low-energy bands, our projected resolvent is P ( H)1 P , where P is the projection operator onto the pz subspace.
From linear algebra one obtains [83]:

92

CHAPTER 9. THEORETICAL CONCEPTS

1
1
,
P =
1
H
QHP
P HP P HQ QHQ

where Q projects onto the complement of P , i.e. Q = 1 P . Therefore the

effective Hamiltonian for the pz subspace takes the form:
Hef f = P HP + P HQ

1
QHP .
QHQ

For and QHQ in the denominator, we make the following approximation.

Since we are interested in k points close to the point, 0z (k = 0) and
QHQ Q0x,y (k = 0) will be a good approximation, where 0i (k)... denotes the
unperturbed band structure.
This procedure, performed to second order in , , and k, yields the following
expression for the subspace spanned by the pz orbitals:


6i(kx iky )/w
6 + ( 23 + 9 2 /w)k2
.
Hef f =
6i(kx + iky )/w
6 + ( 32 + 9 2 /w)k2
We can now directly identify the contributions to the Hamiltonian that describe
the system with spin-orbit interaction. The diagonal terms are similar to the
free-electron-model case, with the effective mass determined by . The offdiagonal term corresponds to the Rashba spin-orbit term, and the parameter
R is equal to 6/w. This explicitly demonstrates that the Rashba spin-orbit
interaction at the surface depends on the atomic spin-orbit parameter (), as
well as on the potential gradient at the surface, described by the parameter .
When setting to zero, which corresponds to a switching-on of the inversion
symmetry, the Rashba splitting will be removed. Also, the disappearing of the
atomic spin-orbit interaction will lead to a vanishing of the Rashba splitting.
With this consideration, we have achieved a connection between the freeelectron model and the tight-binding model.
Finally, we like to mention that atomic spin-orbit interaction is responsible
for a pronounced Z-dependence of the Rashba surface-state splitting. From the
present tight-binding model calculations (Ref. [83]) it can be understood why
the Rashba splitting should be of the same order of magnitude as the atomic
splitting, and not much smaller as was estimated earlier using the free-electron
approach.