Magneto-optical Kramers-Kronig analysis

Julien Levallois, Ievgeniia Nedoliuk, Iris Crassee, Alexey B. Kuzmenko1

1
Department of Quantum Matter Physics, University of Geneva, CH-1211 Geneva 4, Switzerland∗
(Dated: March 4, 2015)
We describe a simple magneto-optical experiment and introduce a magneto-optical Kramers-
Kronig analysis (MOKKA) that together allow extracting the complex dielectric function for left-
and right-handed circular polarizations in a broad range of frequencies without actually generating
circularly polarized light. The experiment consists of measuring reflectivity and Kerr rotation, or
alternatively transmission and Faraday rotation, at normal incidence using only standard broadband
polarizers without retarders or quarter-wave plates. In a common case, where the magneto-optical
rotation is small (below ∼ 0.2 rad), a fast measurement protocol can be realized, where the polarizers
arXiv:1503.01030v1 [cond-mat.str-el] 3 Mar 2015

are fixed at 45◦ with respect to each other. Apart from the time-effectiveness, the advantage of this
protocol is that it can be implemented at ultra-high magnetic fields and in other situations, where an
in-situ polarizer rotation is difficult. Overall, the proposed technique can be regarded as a magneto-
optical generalization of the conventional Kramers-Kronig analysis of reflectivity on bulk samples
and the Kramers-Kronig constrained variational analysis of more complex types of spectral data.
We demonstrate the application of this method to the textbook semimetals bismuth and graphite
and also use it to obtain handedness-resolved magneto-absorption spectra of graphene on SiC.

PACS numbers: 78.20.-e,78.20.Ci,78.20.Ls,78.40.Kc,78.67.Wj

I. INTRODUCTION instead of one (the field B is along the z axis). If the

light polarization is confined in the xy plane, which is
Applying a magnetic field to materials influences pro- the case for normal-incidence experiments in the Faraday
foundly their optical properties by the Zeeman splitting, geometry, where the field is parallel to the propagation of
formation of Landau levels (LLs), modification of the op- light), then only the two components, xx and xy , play a
tical selection rules and sometimes inducing phase tran- role. This allows us to restrict our discussion to the 2 × 2
sitions. The strength of magneto-optical (MO) effects tensor (we omit the B argument for brevity):
depends on many factors, such as the density and mo-  
bility of the charge carriers, the value of magnetic mo- xx (ω) xy (ω)
ˆ(ω) = , (2)
ments and their interaction, the band structure and the −xy (ω) xx (ω)
spin-orbit coupling. Magnetic field breaks the time rever-
sal symmetry (unless it is already broken spontaneously) which has the eigenvalues ± (ω) = xx (ω) ± ixy (ω) cor-
and triggers non-reciprocal optical phenomena, such as responding to the right- and the left-handed circular po-
the Faraday1 and the Kerr2 rotation. The MO rotation, larizations. It is notable that this reduction is valid not
which was historically the first compelling evidence of the only for fully isotropic but also for uniaxial media with
electromagnetic nature of light, plays nowadays a crucial the optical axis along the z direction, which increases
role in telecommunications, data storage, laser technol- dramatically the number of materials to which our dis-
ogy and material characterization. cussion applies.
It is well-known that magneto-resistivity and Hall-
The goal of a complete linear-response optical-
effect measurements are in general more informative than
spectroscopy experiment in zero field is to determine the
the DC zero-field transport. Likewise, the magneto-
real and the imaginary parts of the dielectric function.
optical response carries more ample information about
While certain techniques, such as spectroscopic ellipsom-
a physical system than zero-field optics. It is also more
etry or time-domain spectroscopy, can provide experi-
complicated from the experimental and theoretical point
mental access to both of them, other methods rely on
of view. While the linear electromagnetic response of an
the Kramers-Kronig (KK) relations4–7 due to a limited
isotropic material without field is described by a scalar
set of measured optical quantities. A widespread tech-
function, such as the complex optical conductivity σ(ω)
nique of this kind is reflectance spectroscopy, where one
or the dielectric function (ω) = 1 + 4πiσ(ω)/ω, its MO
measures the absolute reflectivity R(ω) in a broad range
counterpart is essentially a 3 × 3 tensor. The Onsager
of frequencies and determines the phase θ(ω) of the com-
reciprocity relations3 impose on it the following struc-
plex reflectivity coefficient8 :
ture:
 
xx (ω, B) xy (ω, B) 0
p
p iθ(ω) 1− (ω)
ˆ(ω, B) = −xy (ω, B) xx (ω, B) 0  (1) r(ω) = R(ω)e = p . (3)
1 + (ω)
0 0 zz (ω, B)

which involves three complex functions, xx , xy and zz , Jahoda found9 that the KK relation can be applied to

Typeset by REVTEX
 2
p
the function f (ω) = ln r(ω) = ln R(ω) + iθ(ω): many Muller-matrix elements17 . While allowing a num-
ber of independent quantities to be measured separately,
Z ∞ p
1 ln R(x) this experiment unavoidably mixes the xy and z compo-
θ(ω) = − ℘ dx + π, (4) nents of the 3 × 3 MO tensor due to the finite angle of
π −∞ x−ω
incidence and thus adds extra complexity to the analysis.
which can be used to restore the unknown complex phase To our knowledge, a model-independent extraction of the
by a direct integration of the measured reflectivity prop- complex functions + (ω) and − (ω) by this method has
erly extrapolated outside the experimental range. Due to not been demonstrated yet.
the parity relation R(−ω) = R(ω) this formula reduces In this paper we propose a rather simple and fast
to pathway towards a complete broadband magneto-optical
Z ∞ spectroscopy of materials with the MO conductivity de-
ω ln R(x) scribed by Equation (2). It involves the measurement of
θ(ω) = − ℘ dx + π. (5)
π 0 x2 − ω 2 the reflectivity (or transmission) and the Kerr (or Fara-
day) rotation at a normal, or near-normal angle of in-
The knowledge of the phase together with the reflectivity
cidence using two conventional polarizers without the
itself allows, via the Fresnel equations or other boundary-
need to measure other quantities, such as the elliptic-
conditions, the determination of other optical functions,
ity. The treatment of this set of data relies on a general-
including the dielectric function and the optical conduc-
ized magneto-optical KK analysis (MOKKA), which we
tivity.
introduce and mathematically justify in this paper.
In magneto-optics, a complete experiment should al-
To emphasize the basic ideas of the method, we shall
low determination of the real and the imaginary parts
restrict our consideration to materials, where the mag-
for both xx (ω) and xy (ω) or equivalently for both
netic and magneto-electric susceptibility are negligible as
± (ω) = 1+4πiσ± (ω)/ω. Obviously, the number of quan-
compared to the dielectric function, which is valid in a
tities to be measured should be doubled as compared to
vast majority of non-magnetic compounds. In the dis-
the case of B = 0. At a first glance, one should simply
cussion, we shall however speculate on a possible inclu-
work with circular polarized light and apply the zero-field
sion of these electromagnetic terms that play an essential
techniques to each handedness separately. Although con-
role in magnetic materials18 , multiferroics19 and topo-
ceptually appealing, this strategy is difficult to realize in
logical insulators20 . For the sake of simplicity we shall
broadband spectroscopy as it is experimentally challeng-
also assume that the Fresnel equations are satisfied, even
ing to generate circular polarized light in a broad range
though the essential part of MOKKA holds even in the
of frequencies. Standard waveplates, typically made of
non-local limit, where more general electromagnetic re-
quartz, operate at selected wavelengths. There are re-
lations should be used.
ports of multilayer waveplates that cover a continuous
range of frequencies, which is however still somewhat The rest of the paper is organized as follows. Section
limited10,11 . Many waveplates must therefore be com- II is devoted to the generalization of the conventional
bined to cover a reasonably broad spectrum of circular Kramers-Kronig reflectivity formalism to the case, where
polarized light, which is especially hard in the range of the magnetic field is set perpendicular to the sample
the phonon absorption in quartz. An alternative solution surface. In section III, the variational KK-constrained
using elliptically polarized light was also demonstrated12 analysis21 is extended to MO experiments. Section IV
that allows reducing the number of waveplates but sig- describes the experimental technique. Three application
nificantly complicates the analysis. A broadband phase- examples are presented in section V. The results are sum-
shifting element (retarder) using an internal reflection in marized in Section VI.
silicon and other materials was fabricated13,14 . Working
with it is however associated with extra light absorption,
the need of a delicate alignment and the penalty (in the II. DIRECT KRAMERS-KRONIG ANALYSIS
case of Si) of not seeing visible light after the retarder. OF MAGNETO-REFLECTIVITY AND KERR
ROTATION
An alternative broadband solution without generating
circular light is based on polarimetry, where one can mea-
sure the MO rotation and ellipticity, or related quanti- We start by considering the magneto-reflectivity and
ties. A rather sensitive polarimetric method is based on a Kerr rotation spectra of a bulk sample at normal inci-
photoelastic modulation (PEM), where one can measure dence. The complex magneto-reflectivity tensor has the
these quantities using a lock-in demodulation at the main same symmetry as the MO tensor (2):
PEM frequency and at its second harmonic15,16 . How-  
ever, in order to achieve an experiment that is complete in rxx (ω) rxy (ω)
r̂(ω) = (6)
the sense indicated above, one has to measure addition- −rxy (ω) rxx (ω)
ally the complex reflectivity or transmission. Further-
more, the calibration procedure for the PEM-based mea- Its eigenvalues are given accordingly by:
surement is quite involved. Another relevant technique is p
a spectroscopic magneto-ellipsometry with an access to r± (ω) = rxx (ω) ± irxy (ω) = R± (ω)eiθ± (ω) . (7)
 3

In the local limit they are related to the momentum- The sought phases can now be obtained by applying the
independent dielectric function via the Fresnel equations: KK transformation to R+ (ω) and R− (ω):
p
1 − ± (ω) Z ∞ p
r± (ω) = . (8) 1 ln R± (x)
p θ± (ω) = − ℘ dx + π, (15)
1 + ± (ω) π −∞ x−ω
The magneto-reflectivity R(ω), unlike the zero-field re- which can be reduced, using the parity relation
flectivity, depends on the polarization state of the inci- R± (−ω) = R∓ (ω), to:
dent light even if the sample is optically isotropic. For a
circular polarization, it takes one of the values R± (ω) de- θ± (ω) = θ̃(ω) ∓ θK (ω), (16)
pending on the handedness. For the linear polarization,
which can be represented as the sum of the two oppo- where
sitely handed circular waves with equal amplitudes, it is Z ∞
given by the average value: θ+ (ω) + θ− (ω) ω ln R̃(x)
θ̃(ω) = =− ℘ dx+π (17)
2 π 0 x2 − ω 2
R+ (ω) + R− (ω)
R(ω) = . (9)
2 and
For a general (elliptical) polarization, the reflectivity
p
p 2 c(ω)
is an unequally weighted superposition of R− (ω) and R̃(ω) = R+ (ω)R− (ω) = R(ω). (18)
1 + c(ω)
R+ (ω). Notably, many spectrometers, especially of the
Fourier-transform type, generate light with an ill-defined Equations (13)-(18) constitute a self contained recipe
and frequency-dependent polarization. This imposes to extract the complex MO reflectivity for each hand-
placing an extra polarizing element in a magneto-optical edness and Equation (8) allows computing the complex
experiment in order to have a well determined polariza- dielectric functions ± (ω). An important implication of
tion. Below we shall consider the case where the incident this result is that spectra of R(ω) and θK (ω) measured
light is linearly polarized and use Equation (9). in a broad spectral range and supplemented with proper
It is obvious that the standard KK method based on extrapolations, bear sufficient information for complete
Equation (5) does not apply since the reflectivity phases MO spectroscopy.
θ+ (ω) and θ− (ω) are different. Moreover, it does not use It is worth noting that the described algorithm relies
the essential information coming from the Kerr angle: on the non-trivial assumption that the functions ln r± (ω)
1

r− (ω)

θ− (ω) − θ+ (ω) are analytical, i.e. they do not exhibit poles or other
θK (ω) = arg = . (10) singularities in the upper complex semiplane =ω > 0.
2 r+ (ω) 2 While this holds for normal reflectance from a bulk sam-
Here we prove mathematically that the complex re- ple, the analyticity can break down, for example, for a
flectivities r± (ω) can be model-independently extracted partially transparent sample. In this case, the variational
from R(ω) and θK (ω) if these function are known on the KK constrained analysis21 should instead be applied. In
entire real-frequency axis. We first introduce an auxiliary the next section we describe the application of the varia-
complex function: tion approach to a much broader set of magneto-optical
  experiments.
r− (ω) 1
f (ω) = ln = ln c(ω) + 2iθK (ω), (11)
r+ (ω) 2
III. VARIATIONAL KRAMERS-KRONIG
where ANALYSIS OF MAGNETO-OPTICAL SPECTRA
R− (ω)
c(ω) = (12) First we remind the basic idea of the KK constrained
R+ (ω)
variational analysis in zero field21 . The measured spectra
is the circular-dichroism ratio. As the imaginary part such as reflectivity, transmission or any combination of
of f (ω) is the Kerr angle known from experiment, them are initially fitted with a KK consistent and sum-
the circular dichroism can be computed via the KK rule satisfying model dielectric function. Most typically,
transformation26 : this is a sum of a limited number of Drude-Lorentz oscil-
 Z ∞  lators:
8ω θK (x)
c(ω) = exp ℘ dx . (13)
π 0 x2 − ω 2
2
X ωp,k
Combining this result with Equation (9) we can extract mod (ω) = ∞ + 2 − ω 2 − iγ ω , (19)
separately the reflectivity for each handedness: ω0,k k
k

2R(ω) 2c(ω)R(ω) where each Lorentzian is parametrized by the oscillator

R+ (ω) = , R− (ω) = . (14)
1 + c(ω) 1 + c(ω) frequency ω0,k (equal to zero for a Drude component),
 4

the linewidth γk and the plasma frequency ωp,k . The with the following structure:
parameter ∞ is the contribution from the high-energy
N −1 h
optical transitions. Once a reasonable fit is achieved, i
Ai± 4 4∗
X
these parameters are fixed, and a so-called variational var± (ω) = i (ω) + A i∓ i (−ω) (26)
dielectric function (VDF) is added to the model: i=2

that satisfies the physical relation ± (−ω) = ∗∓ (ω).

(ω) = mod (ω) + var (ω). (20) One can see that now there are two adjustable param-
eters for every mesh frequency: Ai+ = var+ (ωi ) and
The VDF is defined on a dense mesh of the experimental Ai− = var− (ωi ). By refitting the data we obtain model-
frequency points ωi (i = 1, .., N ), which usually coincide independently the complex valued + (ω) and − (ω) and
with the experimental datapoints or two times less dense thus achieve the goal of MOKKA.
(i.e. every second datapoint). It is constructed as a sum
of elementary KK-consistent functions 4 i (ω):
IV. EXPERIMENTAL PROCEDURE

N −1 h i In this section we shall describe an experiment

Ai  4 4∗
X
var (ω) = i (ω) + i (−ω) , (21) to measure broadband magneto-reflectivity (magneto-
i=2 transmission) and the MO rotation spectra that can be
where the coefficients Ai = var (ωi ) are regarded as free treated by MOKKA. Although the measurement proto-
cols described below were optimized for Fourier trans-
parameters. The imaginary part of 4 i (ω) has a triangu- form infrared (FTIR) spectroscopy, they can also be ap-
lar shape:
plied, with some modifications, to other types of broad-

ω−ωi−1 band spectrometers. We assume that the experimental
 ωi −ωi−1 , ωi−1 < ω ≤ ωi

setup allows the inversion of the magnetic field direction.
=4i (ω) =
ωi+1 −ω
ω −ω , ωi < ω < ωi+1 (22) Hereafter until the end of the Section we shall refer to the
 i+10 i , otherwise

reflectance and the Kerr rotation, although exactly the
same experimental procedure applies to the transmission
while its real part is obtained by the KK transformation: and the Faraday rotation.
As it was emphasized in Section II, in magneto-optics
=4
Z ∞ 
1 i (x) 1 g(ω − ωi−1 )
<4i (ω) = ℘ dx = the polarization state of the incident light should be well
π −∞ x − ω π ωi − ωi−1 defined even if the sample is isotropic in zero field. Since

(ωi+1 − ωi−1 )g(ω − ωi ) g(ω − ωi+1 ) the spectrometer output may be uncontrollably polar-
− + , (23) ized, a polarizer should be put before the sample. Ad-
(ωi − ωi−1 )(ωi+1 − ωi ) ωi+1 − ωi
ditionally, the optical path between the sample and the
where g(x) = x ln |x|. The second term to Equation (21) detector may also introduce some depolarization effects,
serves to satisfy the physical condition (−ω) = ∗ (ω). in which case placing a second polarizer (often called an-
The entire collection of data is finally refitted by adjust- alyzer) after the sample is also required. Therefore we
ing all Ai values independently. As a rule, the fit quality shall consider a setup, where both polarizers are present,
is nearly perfect since the number of parameters is equal and which allows measuring the reflectivity and the Kerr
or comparable to the number of datapoints. Since the rotation within the same experiment. This configura-
variational method does not imply the analyticity of any tion is similar to standard two-polarizer ellipsometry22 ,
other response function than (ω) it applies to virtually except that here we consider normal incidence, while el-
any type of optical experiments. lipsometry is typically done at large illumination angles.
It appears that this approach can also be applied to We shall assume that the polarizer is fixed and the ana-
magneto-optics. The Drude-Lorentz model is modified lyzer is rotated, although the same result is valid for the
in the magnetic field as follows: case where the analyzer is fixed and the polarizer turns.
2
Since the sample is isotropic, we can set the polarizer an-
X ωp,k gle to zero without loss of generality. As it is shown in
mod,± (ω) = ∞± + 2 − ω 2 − iγ ω ∓ ω , (24)
ω0,k k c,k ω Appendix A, the intensity measured by the detector as a
k
function of the analyzer angle A is given by:
where a cyclotron frequency ωc,k is introduced for each
Lorentz term. At first one has to fit the entire set of MO IA (ω, B) = CA (ω, B)
data as close as possible using Equation (24) or any other R(ω, B) + R̃(ω, B) cos[2A − 2θK (ω, B)]
KK-consistent and sum-rule compliant function. Then × , (27)
2
the model parameters should be fixed and a magneto-
VDF should be added: where the functions R(ω, B), R̃(ω, B) and θK (ω, B) are
defined in Section II and the prefactor CA (ω, B) is deter-
± (ω) = mod± (ω) + var± (ω) (25) mined by all elements in the optical path from the source
 5

to the detector apart from the sample. It may depend on B. Fast protocol
the analyzer angle due to depolarization effects in the de-
tection channel and on the field because of the changes in The measurement scheme described above is rigorous
detector sensitivity and all mechanical movements that but quite slow as it involves measuring the raw intensity
it may induce. spectra at dozens of analyzer angles for the sample and
If we replace the sample by a reference with R = R̃ = 1 the reference. Fortunately, in many situations one can
and θK = 0 then the detector intensity will be given by use an alternative protocol, which requires much fewer
the Malus law: spectra to be measured. This method is accurate if the
rotation angle is small (for example θK < 0.2 rad), so
IA,ref (ω, B) = CA (ω, B) cos2 A. (28) that sin θK ≈ θK and cos θK ≈ 1. Since for small rota-
tions the dichroism is also weak, |c − 1|  1, the ratio
Thus the reference measurement allows us to determine R̃(ω)/R(ω) = 1 + O((c − 1)2 ) and hence we can ignore
CA (ω, B).23 . Since the choice of the best reference proce- the small difference between R(ω) and R̃(ω).
dure is setup- and sample-dependent, we do not discuss The fast method hinges on the fact that the intensity
this issue here24 . is most sensitive to the MO rotation when the polarizers
Below we discuss two measurement protocols, both are oriented at +45◦ or −45◦ with respect to each other
based on Equation (27) but adapted for different experi- and is almost insensitive to it for A = 0, i.e. when they
mental situations. are parallel. Within the approximations made above,
Equation (27) gives the following detector intensity for
these analyzer positions:
1 ± 2θK (ω, B)
A. Complete protocol I±45 (ω, B) ≈ C±45 (ω, B)R(ω, B) , (31)
2
I0 (ω, B) ≈ C0 (ω, B)R(ω, B). (32)
As it is common in polarimetric experiments, the most
rigorous approach involves measuring the raw spectra and for the reference:
IA (ω, B) for a large number of analyzer angles spread C±45 (ω, B)
I±45,ref (ω, B) ≈ , (33)
equidistantly between 0◦ and 360◦ . The same should 2
be done for the reference, in order to obtain the prefac- I0,ref (ω, B) ≈ C0 (ω, B). (34)
tor CA (ω, B) as discussed above. For each frequency and An obvious way to determine the reflectivity is to use
each field value one can fit the function F (A) = IA /CA by the measurement in the parallel polarizer configuration:
the formula α + β cos(2A − 2Am ) with the fitting param-
eters α, β and Am . The reflectivity can be determined I0 (ω, B)
R(ω, B) ≈ , (35)
by the relation R(ω, B) = 2α(ω, B). The parameter Am I0,ref (ω, B)
formally corresponds to the Kerr angle. However, using which has the additional advantage that the optical signal
the relation θK (ω, B) = Am (ω, B) may result in a con- is at maximum.
siderable systematic error due to polarizer imperfections. In order to obtain the magneto-optical rotation, we
This problem is easily removed by subtracting the zero- first we assume that CA (ω, +B) = CA (ω, −B). Then we
field value: θK (ω, B) = Am (ω, B) − Am (ω, 0). have:
Our experience shows that measuring at the two po- I±45 (ω, +B)
larities of the magnetic field improves both the system- ρ±45 (ω, B) = ≈ 1 ± 4θK (ω, B), (36)
I±45 (ω, −B)
atic accuracy and the signal-to-noise ratio. Based on the
parity relations R(ω, −B) = R(ω, B) and θK (ω, −B) = where again we used that θK (ω, −B) = −θK (ω, B). This
−θK (ω, B) the following equations can be derived: suggests that the rotation can be obtained based on the
measurement for only one of the angles +45◦ or −45◦ ,
R(ω, B) = α(ω, B) + α(ω, −B), (29) without reference:
Am (ω, B) − Am (ω, −B) ρ±45 (ω, B) − 1
θK (ω, B) = . (30) θK (ω, B) ≈ ± . (37)
2 4
In this way the rotation can be measured even in exper-
that we actually recommend to use. p imental setups, where in situ rotation of the analyzer is
The value of β/α = R̃(ω)/R(ω) = 2 c(ω)/(1 + c(ω)) not possible as one can simply place the sample between
can in principle be utilized to determine the circular two fixed polarizers mounted at 45◦ with respect to each
dichroism ratio c(ω). However, this only works well if other.
the dicroism, and therefore the ellipticity of the reflected If CA (ω, +B) is not equal to CA (ω, −B) then the pro-
light, is strong enough, which is often not the case25 . cedure should be modified. The most obvious modifica-
Fortunately, the implementation of MOKKA does not tion is to take the same measurement with a reference:
require c(ω) since it is computed via a KK transforma- I±45,ref (ω, +B)
ρ±45,ref (ω, B) = (38)
tion of θK (ω). I±45,ref (ω, −B)
 6

1.0
and to use the corrected formula: (a) (b)
  0.8 holes
1 ρ±45 (ω, B) 0T z

Re σ (10 Ω cm )
4

-1
Reflectivity
θK (ω, B) ≈ ± −1 . (39) 0.6
4 ρ±45,ref (ω, B)

-1
electrons

3
0.4 2
If the analyzer can be rotated then a more accurate
0.2
result can be obtained by combining measurements at 0T
+45◦ and −45◦ . One can use the formula: 0.0 0
0 200 400 600 0 100 200 300
-1 -1
Wavenumber (cm ) Wavenumber (cm )
ρ+45 (ω, B) − ρ−45 (ω, B)
θK (ω, B) ≈ (40) 1.0
8 (c) (d)
0.10

Re σ+/- (10 Ω cm )
0.8

-1
+
if CA (ω, +B) = CA (ω, −B) or

Kerr angle (rad)

7T 4

Reflectivity

-1
0.6
0.05 -

3
 
1 ρ+45 (ω, B) ρ−45 (ω, B)
θK (ω, B) ≈ − (41) 0.4 2
8 ρ+45,ref (ω, B) ρ−45,ref (ω, B) 0.00
0.2
7T
otherwise. Equations (40) and (41) are less sensitive 0.0 -0.05 0
0 200 400 600 0 100 200 300
to polarizer imperfections than Equations (37) and (39). -1 -1
Wavenumber (cm )
Wavenumber (cm )
Additionally, one can show that Equation (40) provides
a more accurate result than Equation (37) when a weak
FIG. 1. Far-infrared optical and magneto-optical infrared
field asymmetry of CA (ω, B) is present. This allows a spectra of bismuth at 10 K for the polarization perpendicular
reference-free measurement of the MO rotation without to the trigonal (z) axis. Panels (a) and (b) show the zero-
imposing stringent conditions on the stability of the op- field reflectivity and optical conductivity respectively. Panel
tical elements with respect to the field. (c) presents magneto-reflectivity (in blue) and Kerr rotation
(in red) at 7 T. (d) The real (absorptive) part of the optical
conductivity for the two circular polarizations obtained using
V. APPLICATION EXAMPLES MOKKA. The inset of panel (b) depicts the electron and hole
Fermi pockets (adapted From Ref.38 ).
A. Bismuth

Bismuth is a canonical semimetal with a low car- bolometric detector were used. A crystal of 99.9999%
rier density (∼1017 cm−3 ), a long mean-free path and a pure Bi had a surface of 3 × 5 mm2 and a thickness of 1
small cyclotron mass. These properties, combined with mm. Free standing wire-grid gold polarizers were used.
a strong spin-orbit interaction, give rise to spectacular The magnetic field was applied along the trigonal axis
magneto-optical effects27–31 . While some classical phe- and therefore almost parallel to the propagation of light.
nomena, such as quantum oscillations32,33 , were first dis- The noise in the reflectivity spectra was less that 0.5 %.
covered in bismuth, a number of new phenomena were The accuracy of the Kerr angle was better than 1 mrad.
recently found in it and in related compounds, such as a The zero-field reflectivity, shown in Figure 1a, features
transition to a 3D topological insulating state upon Sb a sharp plasma edge at around 150 cm−1 due to highly
substitution34 , a valley-ferromagnetic state36 , a valley- mobile charge carriers and it is in good agreement with
nematic Fermi liquid state37 and avoided Lifshitz-type previous works39,40 . The optical conductivity obtained
semimetal-semiconductor transition35 . Therefore this by the usual KK analysis, is presented in Figure 1b. It
material makes an excellent case for applying MOKKA. has a low-energy Drude peak and a linearly growing con-
Bismuth has a rhombohedral (or trigonal) crystal lat- tribution from low-lying interband transitions.
tice and therefore it is optically anisotropic. However, it The magneto-reflectivity and the Kerr angle spectra at
is optically isotropic within the plane, which is perpen- 7 T are shown in Figure 1c. One can note that the plasma
dicular to the trigonal axis and therefore the described edge is significantly blue-shifted by the field, which is
technique applies when the propagation of light and the a magnetoplasma effect41 . The reflectivity also shows a
magnetic field are along the trigonal axis. Fortunately, strong decrease below 100 cm−1 , which can be attributed
the material cleaves easily (in liquid nitrogen) in the same to the enhancement of the penetration depth and there-
plane, making it possible to prepare a high-quality atom- fore an enhanced optical absorption below the cyclotron
ically flat surface of the needed orientation. frequency. The Kerr rotation has a strong spectral struc-
Infrared magneto-reflectivity and Kerr angle spectra at ture clearly associated with the plasma edge at 200 cm−1
nearly normal incidence (' 8◦ ) were measured between and a weaker feature at 100 cm−1 , close to the frequency
20 and 650 cm−1 with the resolution of 1 cm−1 at 10 below which the reflectivity starts to decrease. Remark-
K in a split-coil superconducting magnet attached to an ably, the Kerr angle is essentially zero above 250 cm−1 .
FTIR spectrometer. Both protocols described in Section Figure 1d presents the optical conductivity σ± (ω) for
IV were used for comparison and they gave almost iden- the left- and right-handed circular polarized light ob-
tical results. A Hg light source and a liquid He cooled tained by the variational magneto-optical KK analysis,
 7

1.0
where a Drude-Lorentz model with only a few terms was (a) 20 z (b)
used as a starting point. Applying the direct integration

Re σ (10 Ω cm )
-1
Reflectivity
procedure described in Section II provided essentially the 0T
0.9

-1
same conductivity curves. The most prominent spectral

3
10
structure in both polarizations is a absorption band at
80-110 cm−1 , which consists of several peaks. The spec- 0.8 0T
tral shape of this absorption is quite different for the two 0
light helicities. The Fermi surface in bismuth is highly 0 200 400
-1
600 0 200 400
-1
600
Wavenumber (cm ) Wavenumber (cm )
anisotropic. It features one hole pocket and three elec-
1.0 0.10
tron pockets with a Dirac-like band dispersion as shown (c) 20 (d)
in the inset of Figure 1b. The presence of multiple cy-

Re σ+/- (10 Ω cm )
7T

-1
+ -

Kerr rotation (rad)

Reflectivity
clotron energies is obviously related to this complex band

-1
0.9
structure. -

3
0.05
10
7T
The physical interpretation of this data is beyond the
scope of this paper and will be presented elsewhere42 . 0.8

Here we only note that MOKKA allows us to obtain the 0.00 0

true shape of the MO conductivity and to determine the 0 200 400 600 0 200 400 600
-1 -1
Wavenumber (cm ) Wavenumber (cm )
absolute spectral weight of the cyclotron resonance peaks
for each handedness separately in a model-independent
FIG. 2. Far-infrared optical and magneto-optical spectra
fashion.
of graphite at 10 K. Panels (a) and (b) show the zero-field
reflectivity and optical conductivity respectively. Panel (c)
presents magneto-reflectivity (in blue) and Kerr rotation (in
B. Graphite red) at 7 T. (d) The real (absorptive) part of the optical
conductivity for the two circular polarizations obtained using
Graphite is another textbook semimetal, where several MOKKA. The dashed curves in Panels (c) and (d) correspond
to the tight-binding Hamiltonian fits49 . The inset depicts the
physical phenomena were discovered, and which is addi-
Fermi surface (electrons pockets in blue, hole pockets in red),
tionally of a tremendous practical importance. In the last adapted From Ref.50 ).
decade it again attracted much interest in the research
community because of the exfoliation of graphene43 .
Graphite is composed of weakly coupled honeycomb car-
bon layers with predominantly Bernal stacking and is
therefore extremely anisotropic. The tiny Fermi surface
consists of trigonally warped electron and hole pockets
stretched along the K-H corner of the hexagonal Brillouin these data in fields from 0 to 7 T using a Slonczewski-
zone, as shown in the inset of Figure 2b. The very pe- Weiss-McClure Hamiltonian with Landau-level quanti-
culiar optical and magneto-optical properties of graphite zation and the Kubo formula as described in Ref. 49.
were widely studied44–49 . Recently, we measured infrared The fits at 7 T are shown as dashed lines. Although
magneto-reflectivity and Kerr rotation spectra and fitted very good, they show some deviations from the experi-
them using a tight-binding model Hamiltonian49 . Here ment and therefore do not provide the precise shape of
we shall demonstrate a model-independent extraction of the optical conductivity. Here we use these physically
handedness-resolved magneto-optical conductivity from meaningful fits, instead of the phenomenological Drude-
the same data. Lorentz model, with fixed parameters as a starting point
Figures 2a and 2b present the far-infrared region of the and apply a magneto-VDF as discussed in Section III.
zero-field reflectivity and the KK-derived optical conduc- This allows us to achieve a nearly perfect fit (not shown)
tivity of highly-ordered pyrolytic graphite (HOPG) at 10 and calculate the optical conductivity for the two circu-
K48 . The reflectivity is growing monotonously towards lar polarizations shown in Figure 2d. One can see that
unity at zero frequency and the optical conductivity is the true optical conductivity shows much sharper peaks
essentially constant, apart from the lowest frequencies, than the model. This is likely due to the fact that the
where a small Drude peak is observed. Above 200 cm−1 Kubo formula used in Ref. 49 was assuming a frequency-
the conductivity is close to σ0 = e2 /4~ per layer (≈ 1820 independent scattering rate, thus strongly overestimating
Ω−1 cm−1 ), which is the theoretically expected conduc- the optical width of low-energy LL transitions.
tivity of monolayer graphene due to optical transitions
between conical valence and conduction bands48 .
Figure 2c shows the experimental results at 7 T. The In contrast to the case of bismuth, the two conduc-
reflectivity modifies dramatically in field showing intense tivity curves of graphite are markedly different. This is
spectral structures owing to the formation of Landau lev- explained by non-trivial magneto-optical selection rules
els. The Kerr rotation also shows a strong and rather in graphite that are dominated by the trigonal warping,
nontrivial spectral dependence. We successfully modeled as discussed in details in Ref.49.
 8

C. Multilayer graphene on SiC 4.0 1.0

* (a) (b)
4T
3.5 3.5 T *
0.8
As a third example, we show an extraction of 3T

Faraday rotation θF (rad)

3.0
handedness-resolved magneto-optical conductivity of an 2.5 T
0.6
ultrathin film on a MO-inactive substrate by applying 2T

T(B)/T(0)
2.5
MOKKA to a combination of magneto-transmission and 1.5 T
0.4
Faraday rotation. Figures 3a and 3b present far-infrared 2.0 1T
spectra of T (B)/T (0) and θF of multilayer graphene 0.75 T
0.2
grown on the C-face of 6H-SiC51 for several fields be- 1.5
0.5 T
tween 0.25 and 4 T at 5 K. The Fabry-Perot oscillations 0.25 T
1.0 0.0
in the substrate were intentionally suppressed by reduc-
ing the spectral resolution to 5 cm−1 . 0.5
0 200 400 600 0 200 400 600
The two-dimensional (sheet) optical conductivities Wavenumber (cm-1)
Wavenumber (cm-1)
σ± (ω) of graphene (Figures 3c and 3d)) were calculated
model-independently using the variational MOKKA in- En 2 En 2
+ 1 (c) - 1 (d)
troduced in section III and the Fresnel formulas for the 300
B 0 300
B 0
case of incoherent internal substrate reflections as de- -1 -1
-2 -2
tailed in Appendix B.

Re σ+ (units of σ0)

Re σ- (units of σ0)
The transmission features several absorption dips
200 200
(marked by arrows for B=0.25 T) corresponding to opti-
cal transitions between Landau levels, in agreement with
52,53
previous measurements
√ . The transition energies are
proportional to B and match almost perfectly the ex- 100 100
pected values in monolayer graphene, where the LL en-
ergies are given by the formula:
0→1 -1→2 -1→0 -2→1
0 0
q 0 200 400 600 0 200 400 600
En = sign(n) 2~evF2 |nB|. (42) Wavenumber (cm-1) Wavenumber (cm-1)

FIG. 3. Far infrared magneto-transmission (a) and Faraday

Here vF ≈ 106 m/s is the Fermi velocity and n = rotation (b) of multilayer graphene on the Si-face of 6H-SiC
0, ±1, ±2, .. is the LL quantum number. It was argued52 at 5 K at several values of perpendicular magnetic field be-
that the monolayer-like dependence is due to a twisted tween 0.25 T and 4 T. Arrows indicate the LL transitions
stacking of the layers which decouples them from each from neutral layers, the asterisk shows the Drude contribution
other55 much more efficiently than in the Bernal-stacked from highly doped layers. Panels (c) and (d) show the two-
graphite. The absolute spectral weight of the LL absorp- dimensional optical conductivity normalized by σ0 = e2 /4~
tion is in fact much smaller than the expected value ac- for circular light of each handedness obtained using varia-
tional MOKKA. The corresponding Landau level transitions
cording to the number of layer (about 20 in this sample)
are sketched in the insets. In all graphs, the curves are shifted
for reasons that are not fully understood at the moment. vertically for clarity, except at 0.25 T. The zero levels for the
In addition to quantum Landau level transitions, there is Faraday rotation curves are shown by the dashed lines.
a structure at low energies (shown by the asterisk for 4
T) stemming from the cyclotron resonance of free carriers
in strongly doped layers close to the substrate.
VI. SUMMARY
While the LL transitions −n → n + 1 and −n − 1 → n
have the same energy, they are excited by circular polar-
izations of light56 as sketched in the insets of Figures 3c The goal of this paper is to introduce a technically sim-
and 3d). In a simple magneto-transmission experiment ple solution for a complete-tensor broadband magneto-
they unavoidably overlap. The advantage of MOKKA optical spectroscopy. The pathway that we propose is to
over is that it allows an unambiguous separation of transi- perform an ellipsometry-like measurement at normal in-
tion with different handedness. In the presented case, one cidence with two polarizers and to supplement it with a
can clearly see that the shape of the peaks with the cor- properly generalized Kramers-Kronig analysis. We have
responding energy is quite different in σ+ (ω) and σ− (ω). proven mathematically that in the case of a bulk sam-
For example, the peaks in σ+ (ω) are sharper. One can ple a combination of broadband magneto-reflectivity and
also notice that at low fields (0.25 and 0.5 T) the inten- Kerr angle that can be measured using this simple tech-
sity of the 0 → 1 peak is significantly larger than that nique, allow a unique and model-independent extraction
of the −1 → 0 peak. A discussion of this electron-hole of handedness resolved dielectric function. Furthermore,
asymmetry was given in Ref.54 . we extended the variational Kramers-Kronig analysis21
 9

to magneto-optical spectra, which can be applied in a APPENDIX A

variety of experimental situations, including transmis-
sion and Faraday rotation measurements on thin films Here we use the Jones-matrix formalism to derive
deposited on a substrate as well as reflectivity on semi- Equation (27). For simplicity, we shall refer to the re-
transparent samples. This method can be also regarded flection experiment, although exactly the same deriva-
as an optical-Hall type technique since it provides access tion applies to transmission. Below we consider the case
both to the longitudinal and the Hall components of the where the first polarizer is fixed and the second polarizer
optical conductivity. (analyzer) is rotating. However, the same result applies
As the examples of bismuth, graphite and epitaxial when the first polarizer is rotated and the second one
graphene on SiC have clearly shown, handedness resolved is fixed. The transformation of the polarization of light
MO spectroscopy is especially useful for the systems, along the optical path is sketched in Figure 4.
where both electrons and holes are present. In contrast
to the often used approach, where the measured spectra Ey Ey
(including Kerr or Faraday angles) are only compared
to model curves49,57–59 , our technique allows extracting P̂ Ŝ θK
source Ex
the true shape of the MO dielectric function and opti- polarizer
Ex
sample
cal conductivity. This is important, where such a shape
is difficult to reproduce theoretically, for example, due Ey
to a frequency-dependent scattering rate. Although all A Â
spectra presented in this paper were measured in the far- detector analyzer
infrared region, the same technique can be obviously ap- Ex
plied to any other optical range.
Further extensions of this method to more complicated
situations can be envisaged. First, in the case, where FIG. 4. Transformations of the polarization of light in the
the magnetic and magneto-electric susceptibility play a two-polarizer optical setup used for the magneto-optical ex-
role one would need to increase the number of indepen- periment proposed in this paper.
dent measurements. For example, measuring magneto-
transmission and Faraday rotation for several films of Without loss of generality, we assume that the polar-
different thickness and properly modifying the Fresnel izer is aligned along the x axis, so that its Jones matrix
equations may allow an independent extraction of ± (ω) is
and µ± (ω). Second, if non-local effects are essential  
then one should introduce the momentum dependence 1 0
P̂ = tP (43)
to the dielectric function and replace the Fresnel equa- 0 0
tions with more general electromagnetic relations. Re-
markably, even in this case the complex reflectivity for where tP is the polarization transmission for the wanted
each handedness can be extracted using the formalism polarization. This results in the following Jones vector
presented in Section II, which does not rely on the Fres- of the radiation incident to the sample:
nel equations. Third, if a sample is anisotropic in zero
field and the MO tensors ˆ(ω) and r̂(ω) no longer sat-  
~ in = P̂ E
~ source = 1
isfy the symmetry of Equations (2) and (6) the measure- E t E , (44)
0 P 0
ment technique has to be modified to involve rotating of
both polarizers. As a final remark, the time-domain po- where E0 is the x component of the amplitude of light
larimetric spectroscopy59–61 offers further opportunities leaving the spectrometer. Using the Jones matrix of the
when combined with the method described in this pa- sample
per, as it allows measuring more independent quantities  
simultaneously. rxx rxy
Ŝ = (45)
−rxy rxx

Acknowledgments we obtain the Jones vector of the reflected light:

 
We thank M. Tran, N. Ubrig, D. van der Marel, J.-M. E ~ in = rxx tP E0 ,
~ ref = Ŝ E (46)
−rxy
Poumirol and A.Akrap (University of Geneva) for valu-
able discussions. We are grateful to C. Uher (University which in general corresponds to an elliptical polarization.
of Michigan) and T. Seyller (University of Chemnitz) for The Jones matrix of the analyzer set at the angle A with
providing us high-quality samples of bismuth and epi- respect to the x axis is:
taxial graphene on SiC. This work was supported by
cos2 A
 
the Swiss National Science Foundation and EU Graphene cos A sin A
 = tA , (47)
Flagship (contract no. CNECT-ICT-604391). cos A sin A sin2 A
 10

where tA is the analyzer transmission. After the analyzer

the following polarization state is achieved:
−2
Z 0 σ±
~ out = ÂE
E ~ ref T±0 = 1 + , (54)
  ns + 1
cos A 1

ns + 1 + Z 0 σ+

= (rxx cos A − rxy sin A) tA tP E0 , (48) θF0 = arg , (55)
sin A 2 ns + 1 + Z 0 σ−
The signal at the detector is finally given by
where Z0 is the vacuum impedance. In this situation the
2
2 effect of the substrate as compared to a free standing
~ out
IA (ω) = ζ E = CA |rxx cos A − rxy sin A| (49) film is to reduce the effective value of the film conductiv-
ity by (ns + 1)/2 due to a screening of the electric field
where ζ is the cumulated transmission coefficient of all experienced by the film.
optical elements between the analyzer and the detector If multiple reflections contribute significantly to the
and CA = ζ|tA tP E0 |2 . Note that ζ and CA may depend signal then the effect of the substrate is more complex
on A if some of these elements are polarization sensitive. and is dependent on the degree of their mutual phase
In the following, we take into account that coherence. Decoherence can be caused by wedging of
the substrate, a distribution of the angles of incidence,
R+ e−iθK + R− eiθK i(θ+ +θK )
p p
rxx = e (50) a finite resolution of the spectrometer and other factors.
2 p While it is in general quite hard to reproduce this effect
R+ e−iθK − R− eiθK i(θ+ +θK )
p
quantitatively, the analytical result can be easily found
rxy = e (51)
2i in the two limiting cases. If the internal reflections are
fully coherent then
as it follows from Equation (7). Substituting this into
Equation (49) yields the dependence of the detected sig-
nal on the analyzer angle: 1 − p2
2
T±c = T±0 × , (56)
CA p 1 − pq±
( R+ e−iθK + R− eiθK ) cos A
p
IA =
4
 
c 0 1 1 − pq+
2 θF = θF + arg , (57)
+i( R+ e−iθK − R− eiθK ) sin A 2 1 − pq−
p p
(52)

After simple algebra this reduces to: where

ns − 1 n ω o
R + R̃ cos(2A − 2θK ) p= exp i ns ds , (58)
IA = CA . (53) ns + 1 c
2 ns − 1 − Z 0 σ± n ω o
This result has a straightforward geometrical interpre- q± = exp i ns ds . (59)
ns + 1 + Z0 σ± c
tation. The detector signal is minimized when the ana-
lyzer is orthogonal to the main axis of the ellipse drawn and the Fabry-Perot effect is present both in the trans-
by the electric-field vector of light reflected from the sam- mission and the Faraday rotation62 . In the opposite case
ple (Figure 4). On the other hand, the orientation of this of the full incoherence63 one obtains:
axis is given precisely by the Kerr angle.
1 − |p|4
T±i = T±0 × , (60)
APPENDIX B 1 − |pq± |2
1 ∗
θFi = θF0 + arg 1 − |p|2 q−


q+ . (61)
For the reference purpose, here we present equations 2
needed to calculate the magneto-transmission T = (T+ +
T− )/2 and the Faraday rotation θF of a MO-active film Interestingly, at selected frequencies the Fabry-Perot
with the two-dimensional optical conductivity σ± de- effect can increase the transmission and the Faraday rota-
posited on a MO-inactive substrate with the complex tion simultaneously as compared to the incoherent case,
refractive index ns and thickness ds . The result de- which can be important for MO applications62 . On the
pends strongly on whether or not the multiple internal other hand, applying MOKKA is easier if the Fabry-
reflections inside the substrate are taken into account. Perot oscillations are fully suppressed.
If they are negligible, which is, for example, the case
for a strongly absorbing substrate, then the transmission
normalized to the bare identical substrate and Faraday REFERENCES
rotation are given by:
 11

∗ 31
Alexey.Kuzmenko@unige.ch M. P. Vecchi, J. R. Pereira, and M. S. Dresselhaus, Phys.
1
M. Faraday, Phil. Trans. R. Soc. 136, 104 (1846). Rev. B 14, 298 (1976).
2 32
J. Kerr, Phil. Mag. 3, 321 (1877). L. Shubnikov and W.Y. de Haas, Proc. Acad. Sci. Amster-
3
L. Onsager, Phys. Rev. 37, 405 (1931). dam, 33 130 (1930).
4 33
H. A. Kramers, Nature 117, 775 (1926). W.Y. de Haas and P.M. van Alphen, Proc. Acad. Sci. Am-
5
R. de Kronig, J. Opt. Soc. Am. 12, 547 (1926). sterdam 33 1106 (1930).
6 34
L. D. Landau and E. M. Lifshitz, Electrodynamics of Con- D. Hsieh, D. Qian, L. Wray, Y. Xia, Y. S. Hor, R. J. Cava,
tinuous Media, Oxford, Pergamon Press (1975). and M. Z. Hasan, Nature 452, 970 (2007).
7 35
M. Dressel and G. Grüner, Electrodynamics of Solids, N. P. Armitage, R. Tediosi, F. Lévy, E. Giannini, L. Forro
Cambridge University Press (2002). and D. van der Marel, Phys. Rev. Lett. 104, 237401 (2010).
8 36
While the first equality is universal, the second one is valid L. Li, J. G. Checkelsky, Y. S. Hor, C. Uher, A. F. Hebard,
only in the local limit, which is however widely used in R. J. Cava, and N. P. Ong, Science 321, 547 (2008).
37
material optics. Z. Zhu, A. Collaudin, B. Fauqué, W. Kang, and K. Behnia,
9
F. C. Jahoda, Phys. Rev. 107, 1261 (1957). Nature Physics 8, 89 (2011).
10 38
J.-B. Masson and G. Gallot, Optics Lett. 31, 265 (2006). Z. Zhu, B. Fauqué, Y. Fuseya, and K. Behnia, Phys. Rev.
11
A. K. Kaveev et al, Appl. Optics 52, B60 (2013). B 84, 115137 (2011).
12 39
K. Karrai, E. Choi, F. Dunmore, S. Liu, X. Ying, Qi Li, T. R. Tediosi, N. P. Armitage, E. Giannini, and D. van der
Venkatesan, H. D. Drew, Qi Li, and D. B. Fenner, Phys. Marel, Phys. Rev. Lett. 99, 016406 (2007).
40
Rev. Lett. 69, 355 (1992). A. D. LaForge, A. Frenzel, B. C. Pursley, T. Lin, X. Liu,
13
T. D. Kang, E. Standard, G. L. Carr, T. Zhou, M. J. Shi, and D. N. Basov, Phys. Rev. B 81, 125120 (2010).
41
Kotelyanskii, and A. A. Sirenko, Thin Solid Films 519, E.D. Palik, J.K. Furdyna, Rep. Prog. Phys. 33, 1193
2698 (2011). (1970).
14 42
X. Xi, R. J. Smith, T. N. Stanislavchuk, A. A. Sirenko, S. J. Levallois et al., in preparation.
43
N. Gilbert, J. J. Tu, G. L. Carr, Infrared Phys. Tech. 67, K.S. Novoselov, A. K.Geim, S. V. Morozov, D. Jiang, Y.
436 (2014). Zhang, S.V. Dubonos, I.V. Grigorieva, and A.A. Firsov,
15
K. Sato, Jpn. J. Appl. Phys. 20, 2403 (1981). Science 306, 666 (2004).
16 44
S. Yamaguchi, Y. Okimoto, K. Ishibashi, and Y. Tokura, J. K. Galt, W. A. Yager, and H. W. Dail, Jr., Phys. Rev.
Phys. Rev. B 58, 6862 (1998). 103, 1586 (1956).
17 45
P.Kühne, C. M. Herzinger, M. Schubert, J.A. Woollam, E. A. Taft and H. R. Philipp, Phys. Rev. 138, A197 (1965).
46
and T. Hofmann, Rev. Sci. Instrum. 85, 071301 (2014). W. W. Toy, M. S. Dresselhaus and G. Dresselhaus, Phys.
18
A.J. Sievers and M. Tinkham, Phys. Rev. 129, 1566 Rev. B 15, 4077 (1977).
47
(1963). Z. Q. Li, S.-W. Tsai, W. J. Padilla, S. V. Dordevic, K. S.
19
A. Pimenov, A. A. Mukhin, V. Yu. Ivanov, V. D. Travkin, Burch, Y. J. Wang, and D. N. Basov, Phys. Rev. B 74,
A. M. Balbashov, and A. Loidl, Nature Physics 2, 97 195404 (2006).
48
(2006). A. B. Kuzmenko, E. van Heumen, F. Carbone, and D. van
20
X.-L. Qi, T.L. Hughes and S.-C. Zhang, Phys. Rev. B 78, der Marel, Phys. Rev. Lett. 100, 117401 (2008).
49
195424 (2008). J. Levallois, M. Tran, and A. B. Kuzmenko, Solid State
21
A.B. Kuzmenko, Rev. Sci. Instrum. 76, 083108 (2005). Commun. 152, 1294 (2012).
22 50
R.M.A. Azzam and N. M. Bashara, Ellipsometry and po- J. Schneider, PhD thesis, LNCMI Grenoble (2010).
51
larized light, North-Holland (1977). C. Berger, Z. Song, T. Li, X. Li, A. Y. Ogbazghi, R. Feng,
23
While CA (ω, B) can be determined as IA,ref (ω, B)/ cos2 A, Z. Dai, A. N. Marchenkov, E. H. Conrad, P. N. First, and
a better way to do obtain it is by fitting Iref (A) while W. A. de Heer, J. Phys. Chem. B, 108, 19912 (2004).
52
assuming a reasonable dependence of C(A), for example, M. L. Sadowski, G. Martinez, M. Potemski, C. Berger, and
C(A) = C1 + C2 cos A. W. A. de Heer, Phys. Rev. Lett. 97, 266405 (2006).
24 53
For a small or an irregularly shaped sample an evaporation I. Crassee, J. Levallois, A. L.Walter, M. Ostler, A. Bost-
of a thin metallic film on the surface may be required to wick, E. Rotenberg, T. Seyller, D. van der Marel and A.
obtain the correct absolute reflection. B. Kuzmenko, Nature Physics 7, 48 (2011).
25 54
This problem is analogous to the poor sensitivity of two- I. Crassee, J. Levallois, D. van der Marel, A. L. Walter,
polarizer ellipsometry to a weak absorption resulting in a T. Seyller, and A. B. Kuzmenko, Phys. Rev. B 84, 035103
small ellipticity of reflected light. In ellipsometry, this can (2011).
be solved by using retarders22 . Here we avoid this for the 55
J. M. B. Lopes dos Santos, N. M. R. Peres, and A. H.
reasons mentioned in the introduction. Castro Neto Phys. Rev. Lett. 99, 256802 (2007).
26 56
D. Y. Smith, J. Opt. Soc. Am. 66, 454 (1976). M. Orlita and M. Potemski, Semicond. Sci. Technol. 25,
27
W. S. Boyle, A. D. Brailsford, and J. K. Galt, Phys. Rev. 063001 (2010) .
57
109, 1396 (1958). F. Salghetti-Drioli, P. Wachter and L. Degiorgi, Solid State
28
B. Lax, J. G. Mavroides, H. J. Zeiger, and R. J. Keyes, Commun. 109, 773 (1999).
58
Phys. Rev. Lett. 5, 241 (1960). P.M. Oppeneer, Handbook on Magnetic Materials ed. by
29
J. C. Burgiel and L. C. Hebel, Phys. Rev. 140, A925 K.H.J. Buschow, Elsevier, Amsterdam 13 229 (2001).
59
(1965). J. N. Hancock, J. L. M. van Mechelen, A. B. Kuzmenko,
30
H. R. Verdún and H. D. Drew, Phys. Rev. B 14, 1370 Dirk van der Marel, C. Brne, E. G. Novik, G. V. Astakhov,
(1976). H. Buhmann, and L. Molenkamp Phys. Rev. Lett. 107,
 12

136803 (2011). (2012).

60 62
J.L.M. van Mechelen, D. van der Marel, I. Crassee and T. N. Ubrig, I. Crassee, J. Levallois, I. O. Nedoliuk, F.
Kolodiazhnyi, Phys. Rev. Lett. 106 217601 (2011) . Fromm, M. Kaiser, T. Seyller, and A. B. Kuzmenko, Opt.
61
M. Neshat, and N.P. Armitage, Opt. Express 20, 29063 Express 21, 24736 (2013).
63
I. Crassee, PhD thesis, University of Geneva (2013).