Mon. Not. R. Astron. Soc.

345, 16–38 (2003)

Absorption and reflection infrared spectra of MgO

and other diatomic compounds

A. M. Hofmeister,1  E. Keppel1 and A. K. Speck2

1 Department of Earth and Planetary Sciences, Washington University, St Louis, MO 63130, USA
2 Department of Physics, University of Missouri, Columbia, MO 65211, USA

Accepted 2003 June 9. Received 2002 December 19; in original form 2002 March 19

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

ABSTRACT
Oxide and sulphide minerals are expected to occur in diverse astronomical environments. How-
ever, optical constants for such minerals are either lacking or poorly characterized. Minimizing
errors in laboratory data, while extrapolating over wide frequency ranges, is the focus of this
report. We present reflection and absorption spectra of single-crystal MgO from about ∼100 to
18 000 cm−1 (∼100 to 0.5 µm), and derive emissivity, dielectric and optical functions (n and
k) using classical dispersion analysis and supplementary data to ensure that the reflectivity
values are correct at the low- and high-frequency limits. Absorbance spectra of thin films of
oxides (MgO, CaO, FeO and ZnO) and sulphides (MgS, CaS and FeS) are in good agreement
with available reflectivity measurements, and provide information on the various effects of
chemical composition, structure and optical depth. The greatest mismatch occurs for MgO,
connected with this compound having the broadest peak in reflectance. The ferrous compounds
(FeO and FeS) have relatively weak infrared features and may be difficult to detect in astro-
nomical environments. Previous optical data based on transmission spectra of dispersions have
underestimated the strength of the main infrared features because this approach includes spec-
tral artefacts that arise from the presence of opaque particulates, or from non-uniform optical
depth. We show that areal coverage, not grain size, is the key factor in altering absorption
spectra from the intrinsic values, and discuss how to account for ‘light leakage’ in interpreting
astronomical data. Previous reflectivity data on polycrystals differ from intrinsic values be-
cause of the presence of additional, internal reflections, creating errors in the derived optical
functions. We use classical dispersion analysis and supplemental data from optical microscopy
to provide correct n- and k-values for FeO from the far-infrared to the visible, which can then
be used in radiative transfer models. Thin-film absorption data are also affected by internal
reflections in the transparent regions: we show how to recognize these features and how to
obtain the absorption coefficient, n, and k from thin-film infrared data on CaO, CaS and MgS
using the damped harmonic oscillator model.
Key words: line: identification – methods: laboratory – techniques: spectroscopic – infrared:
general.

(wavelengths λ < 25 µm), the so-called mid-IR region. However,

1 INTRODUCTION
modes in crystalline solids from about ∼50 to 400 cm−1 (the far-IR
Dust particles occur in a wide range of astronomical environments region, ∼ 25
λ
200 µm) show a greater variety of intensities,
(e.g. circumstellar shells around evolved stars, pre-planetary sys- widths and positions, and are thus better suited for identification.
tems around young stars, and the interstellar medium). Identifica- Only recently have direct astronomical observations at comparable
tion is made through comparison with infrared (IR) laboratory data. frequencies become available, through the Infrared Space Obser-
Mostly, the high-frequency lattice modes have been studied, because vatory (ISO: Kessler 1996). Hence the existing mineralogical data
of the ease and availability of laboratory equipment, coupled with are insufficient for the simple task of identifying substances in in-
collection of observational data also at frequencies above 400 cm−1 terstellar and circumstellar dust. In addition, absorption spectra are
usually obtained from a powdered sample dispersed in a medium.
 E-mail: hofmeist@levee.wustl.edu However, variations in particle thickness and areal coverage within a


C 2003 RAS
 IR spectra of MgO compounds 17
dispersion provide spectral artefacts (Horak & Vitek 1978; Hofmeis- The methods used here are not new, as each component exists in
ter 1995). To address these problems, we are compiling a data base various scientific literatures. However, to our knowledge the cross-
of IR spectra of minerals and other inorganic solids that may ex- checks have not been used in astronomical studies, and have not
ist in space. Both reflection and absorption techniques are utilized, been taken full advantage of elsewhere. For the most part, in miner-
as advantages and disadvantages exist for each approach, and be- alogical studies large crystals are used in reflectivity measurements
cause the optical functions obtained from reflectivity are used as to avoid potential problems, and the focus has largely been on fre-
inputs to radiative transfer models in astronomy, whereas absorp- quencies of the peaks, which are less prone to artefacts than peak
tion measurements are directly compared with spectral patterns for profiles, band strengths and widths. Because all peak parameters are
the purpose of identification and for estimates of concentration. important in astronomical applications of the data, however, multi-
This paper is the first of a series, and therefore focuses on the ple measurements and cross-checks are necessary.
methodology. Our plan is to present the data base in segments, each
of which covers structurally and/or chemically related minerals. Di- 2 E X P E R I M E N TA L M E T H O D S
atomic oxides and sulphides are examined here, with the focus being
Three single crystals of synthetic, pure MgO were purchased from
on MgO. This choice is motivated by astronomy, as well as prag-
Atomergenic Chemetals Corporation.1 The large faces are perpen-

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

matics. Foremost, ISO has provided unprecedented high-resolution
dicular to a crystallographic axis and 10 × 10 mm2 . Thicknesses
mid- and far-IR spectra of a variety of astronomical environments
were initially 0.5, 1.0 and 1.0 mm. Spectra were collected on freshly
(see e.g. Waters et al. 2000), and references therein). One of the ma-
polished or cleaved surfaces to minimize surface hydration. In some
jor new results from ISO data was the discovery of absorption fea-
cases, however, the surface hydrated rapidly.
tures indicative of crystalline silicates in the spectra of evolved stars.
Powdered MgO was prepared by dehydrating pure Mg(OH)2 at
However, reliable optical constants for such solid-state species are
600 K in air for 2 d. A sample of Mg18 O was prepared by reacting Mg
lacking. Numerous other unidentified dust features have also been
metal with enriched water in a sealed Pt tube at ∼1100 K and 0.6 GPa
observed, e.g. the ‘30-µm’ feature in evolved star spectra (e.g. Hony,
with a gas-media apparatus. This starting product was dehydrated
Waters & Tielens 2001) or the ‘23-µm’ feature in the spectra of
and recrystallized at 1700 K for 16 h. Pure (better than 99.5 per cent)
Herbig Ae/Be stars (Malfait, Bogaert & Waelkens 1998). These
CaO, FeO, MgS, CaS and FeS were purchased from Alfa/Aesar.2
quality astronomical spectra exemplify the need for new laboratory
The vials, packed in argon, were opened and immediately measured
data with which to compare. Secondly, the structural and chemi-
in the lowest frequency region where their fundamental modes ex-
cal simplicity of diatomic compounds makes these an ideal starting
ist. Samples were subsequently stored in an evacuated desiccator.
point for a such a data base. The IR spectra are simple as a conse-
Higher frequency data (if needed) were collected months to a year
quence, and spectral artefacts are relatively easy to recognize. How-
later. CaO showed evidence of hydration, which was almost entirely
ever, artefacts are also easily engendered as a result of the extreme
removed by heating in N2 at 800 K for 4 d. An additional sample of
variation in optical functions between frequency regions where the
FeS is as synthesized and described by Lauretta, Kremser & Fegley
intense, broad bands are active and where they are not. Thus, in com-
(1996). Naturally occurring magnetite (Fe3 O4 ) from Magnet Cove,
paring this class of substances with astronomical data, one must be
Arkansas is essentially chemically pure. Natural pyrrhotite from Eli
aware how laboratory data are affected by the particulars of the
mine, Vershire, Vermont and pyrite, locality unknown, were also
experiment.
studied. These minerals tend to have few impurities (Arnold 1967;
Our general approach is to obtain reflectivity data from suitably
Deer et al. 1992).
large crystals of good optical quality from the far-IR to the visi-
IR reflectivity spectra of MgO were acquired at near-normal
ble. A Kramers–Kronig analysis of the reflectivity data provides
incidence using a Spectratech Fourier transform IR spectrome-
the optical and dielectric functions as well as the absorption coef-
ter (FTIR)3 microscope and using an evacuated Bomem DA 3.02
ficients. Classical dispersion analysis, which is also known as the
Fourier transform spectrometer4 at 1 cm−1 resolution. The accuracy
damped harmonic oscillator model (Spitzer et al. 1962), additional
of the instrument is ∼0.01 cm−1 . An Si-bolometer and a coated
single-crystal absorption data for the transparent regions, i.e. fre-
mylar beamsplitter were used for the far-IR range from ∼50 to 650
quencies above 1000 cm−1 (λ < 10 µm), and previous, independent
cm−1 (∼200–15 µm). An HgCdTe detector with a KBr beamsplit-
measurements of the low-frequency dielectric constant ( 0 ) and the
ter covered the mid-IR from ∼450 to 4000 cm−1 (∼23–2.5 µm). A
high-frequency index of refraction (n =  1/2 ∞ ) are used to verify CaF2 beamsplitter and an InSb detector were used for the near-IR
the absolute values of the reflectivity, and to extrapolate the data
from ∼1800 to 8500 cm−1 (∼6–1.2 µm). An SiC source was used
into the visible region. The Lydanne–Sachs–Teller relationship pro-
for these IR regions. A quartz source, an Si avalanche detector and
vides a cross-check between IR peak positions and dielectric data.
a quartz beamsplitter probed the visible region of ∼8500 to 20 000
Classical dispersion analysis can be used to obtain n and k from ab-
cm−1 (∼1.2–0.45 µm). A gold mirror was used as the reference for
sorption data if the thickness is known. Emissivity is obtained from
the far- and mid-IR, and silver for the near-IR and visible regions.
the optical functions using Kirchoff’s law. For the lattice modes be-
The number of scans ranged from 800 to 2000.
low 1000 cm−1 , we obtain absorption spectra from films of various
Optically thin (<1 µm) films were made by crushing particles
thicknesses formed by compression in a diamond anvil cell. This
in a diamond anvil cell (DAC). The films were thinned until the
technique provides spectra consistent with Kramers–Kronig analy-
absorbance was near or below unity, or until the profile of the main
sis of reflectivity data of complex silicates (McAloon & Hofmeister
peak was unchanged by further thinning. Thicker films were used
1993; Hofmeister 1995), thereby allowing expansion of the data
to probe the less absorbing regions. Spectra were acquired using
base to substances for which sufficiently large and clear crystals for
the DAC as a holder in a beam condenser using one of the source,
reflection measurements are not available. Films with a thickness of
6 µm have been made using a spacer. We focus on the far-IR, owing 1 Brighton, MA, USA.
to the paucity of laboratory data, on recognizing and compensating 2 Ward Hill, MA, USA.
for spectral artefacts, and on problems and limitations in comparing 3 Spectra-Tech Inc., Stamford, CT, USA.
observational with laboratory data. 4 Bomem Inc., Quebec, Canada.


C 2003 RAS, MNRAS 345, 16–38
18 A. M. Hofmeister, E. Keppel and A. K. Speck
detector and beamsplitter combinations as above. Very thick films Imeas
were made by placing a small amount of powder in the centre of a
25 cm diameter disc of BaF2 or of KBr. A tin gasket with a thickness
of 6 µm (NSG Precision Cells, Inc., Farmingdale, NY, USA) was I0 I0R
placed at the edge of the disc. The sample was compressed to the
same thickness as the gasket by a second BaF2 (or KBr) disc using
I0R(1-R)2(1-Ω)2
a commercial holder (Spectratech, Inc.). To attain the same thick-
ness for sample and gasket we made sure that the powder did not I0(1-R) I0R(1-R)(1-Ω)2
extend to the edge of the disc, but that the area covered by powder d
occluded the entire IR beam, and was without visible cracks and was
even in appearance upon examination under a binocular microscope. I0(1-R)(1- Ω) I0R(1-R)(1- Ω)
These measurements provide useful information on the overtones
and O–H modes at frequencies above ∼500 cm−1 . Use of BaF2 lim-
its transmission to above 690 cm−1 (<14.5 µm). Absorption spectra I0(1-R)(1- Ω)(1-R) = Imeas

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

were also collected from single crystals of MgO placed on an aper-
ture directly in the beam, or in a Spectratech IR microscope. For
the latter approach, the image was masked. To avoid errors and non- Figure 1. Schematic of light loss in a dielectric material showing the most
linear response as transmission becomes small, thicknesses yielding intense ray paths at near-normal incidence (the angle is exaggerated). The
absorbance below unity were sought for the frequency region of in- incident light with intensity I 0 is partially reflected and transmitted at the
terest. The thicknesses of the films made without a spacer can be front surface. The transmitted beam is partially absorbed, to an extent deter-
mined by thickness, d, as described by equation (6). At the back surface, the
determined by comparing peak heights of features that are com-
beam attenuated to I 0 (1 − R)(1 −
) is again partially reflected and trans-
mon and trustworthy in the various spectra, but such features do not mitted. For the case of very weakly absorbing and thin sample, and if the
always exist, and slopes and shoulders sometimes need to be used. angle of incidence allows the beam to reach the detector, then I sum
2R I 0
Procedural details regarding reflectivity measurements or use of in a reflection experiment.
the DAC were given previously (e.g. Hofmeister 1997; Hofmeister
& Mao 2001). to go as frequency to the fourth power (Wooten 1972), was used as a
cross-check. This approach provides the trend expected for damped
harmonic oscillators, but can create artefacts if errors exist in the
3 DATA A N A LY S I S
measured high-frequency reflectivity. Another cross-check exists
In any reflectivity study, some uncertainty exists in the absolute value against n ∞ , measured in the visible as 1.736 by minimizing the re-
of the reflectivity because the reference mirror and sample may not lief in liquids of known n in an optical microscope (Deer et al. 1992).
be precisely aligned in the same position in the beam or may differ Obtaining accurate values of the optical functions at high frequency
slightly in the surface polish, or the sample may provide additional requires that the sample be large enough to avoid back reflections
secondary reflections from fractures, grain boundaries, or the back for the transparent spectral regions. Our sample of MgO has back
surface. The reflectivity in the far-IR region is cross-checked as reflections (Fig. 1) above ∼1200 cm−1 (< 8.25 µm). For this case,
follows against independent measurements of the low-frequency classical dispersion analysis and measured absorption spectra from
dielectric constant  0 . The reflectivity is crystal fragments are used to constrain the high-frequency results,
discussed below.
(n − n med )2 + k 2
R= , (1) IR modes are split by the interaction with light into transverse
(n + n med )2 + k 2 optic (TO) and longitudinal optic (LO) components. The ith TO
where n and k are the real and imaginary components of the complex frequency (ν i ) occurs at maxima in  2 (ν) and the LO frequency is
index of refraction (the optical functions) and n med is the index of determined from minima in the imaginary part of 1/. The damp-
refraction of the medium (n med = unity for air or vacuum). The ing coefficient  i = 2πFWHMi , where FWHMi is the full width
complex dielectric function  =  2 + i 1 is related to the complex at half-maximum of each peak in  2 (ν) in frequency ν units. The
optical function (n + ik) through oscillator strength f i = 2FWHMi σ max /ν i2 where the conductivity
σ (ν) = ν 2 (ν)/2 [see Spitzer et al. (1962) or Wooten (1972)]. The
1 (ν) = n 2 − k 2 , 2 (ν) = 2nk. (2)
absorption coefficient A is calculated from
For dielectric materials at frequencies near 0, k is negligible, so
1/2 1/2 2πν2 (ν)
R 0 = ( 0 − 1)2 /( 0 + 1)2 . The most accurate type of electrical A(ν) = = 4πνk(ν). (3)
n(ν)
measurements provides  0 = 9.83 for MgO (Fontanella, Andeen
& Schuele 1974), which is within the 1 per cent uncertainty of re- The reflectance was fitted using classical dispersion analysis (e.g.
cent measurements of 9.9 derived from a less accurate technique Spitzer et al. 1962). This method treats the vibrations as damped har-
(Subramanian et al. 1989). Both values are in excellent agreement monic oscillators, and as a result the peaks in  2 and Im(1/) have
with the raw far-IR data. We therefore scaled the mid-IR spectra Lorentzian shapes. A perfect harmonic oscillation would produce a
to match and performed a Kramers–Kronig analysis on the merged delta function shape in the dielectric functions, and reflectivity that is
file (Spitzer et al. 1962; Andermann, Caron & Dons 1965). The re- shaped like a square wave. The widths of the peaks in  2 , k and in the
flectivity above and below the range of measured frequencies was spectral measurements are due to phonon–phonon scattering, which
assumed to be a constant in the presented results. This approach is damps the vibrations. This method represents IR spectra well be-
used because the presence of reflections from the back surfaces (see cause the classical and quantum mechanical analyses of damped har-
below) induces some errors in the near-IR reflectivity, and assum- monic oscillations provide essentially the same formulae. For further
ing constant R creates the fewest artefacts for this case. Wooten’s discussions, see Mitra (1969) and Hofmeister (2001). We use a three-
approximation, wherein the high-frequency reflectivity is assumed parameter description of the peaks, wherein the ith LO frequency


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 19
is a function of the TO peak parameters: ν i ,  i and f i .The equa- (Shankland, Nitsen & Duba 1979). For small R, the last term is close
tions for m oscillators using the circular frequency (ω = 2πν) are to log(1 − 2R). This formula assumes parallel, smooth surfaces in a
  non-converging beam. If the reflectance is unknown, then the mea-


m
f j ω2j ω2j − ω2
1 = ∞ +  2 , (4) sured absorbance is obtained relative to a baseline. This approach
j=1 ω2j − ω2 +  2j ω2 is valid for weak peaks, and is necessary if the surface is rough or
if additional internal reflections are present (e.g. from fractures or
where  ∞ is obtained from measurements of n 2 in the visible, and grain boundaries) or when the beam converges strongly.


m
f j ω2j  j ω Parallel surfaces can create interference fringes if the spacing (b)
2 =  2 . (5) is related to an integer multiple of the wavelength. The relationship is
j=1 ω2j − ω2 +  2j ω2
1
b= , (11)
This approach allows extrapolation of the data to frequencies above 2nν
and below the range of measurements, and aids in deciphering arte- where ν is the fringe spacing. In this study, the spacing is probably
facts from overtone-combination bands which are present as weak the gap between the diamonds or between the KBr or BaF2 discs.
shoulders or structure on the main peaks. Generally, the Kramers–

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

Fringes can also be correlated with the size of the aperture (e.g. the
Kronig parameters are considered as the rough description of the 500-µm diameter of the diamond tip or of any masking aperture).
IR spectrum of a solid, and the classical dispersion parameters as
the preferred representation (Wooten 1972). We used the Kramers–
Kronig parameters as the starting point for the initial fit to the di- 4 S Y M M E T RY A N A LY S I S
electric functions, and made the final fit to the raw reflectivity data.
Thin film data were also fitted. The number of fundamental IR modes expected for a compound
Wave propagation is described by E = E 0 exp[i(q · r − ωt)], depends on its crystallographic structure. The number, type and po-
where E is the electric field strength, q is the wavevector, r is the larizations of the modes can be predicted from applying group the-
space coordinate and t is time. Physical properties such as the di- ory to the symmetry elements for the atoms in the structure (e.g.
electric functions and optical functions are derived from reflectivity Cotton 1971). Most of the diatomic compounds examined here
data using this form and Maxwell’s equations (e.g. Wooten 1972). (MgO, CaO, FeO, MgS and CaS) crystallize in the cubic NaCl
This representation of wave propagation requires exponential atten- (rocksalt or B1) structure. The rocksalt structure, with both types
uation of intensity (I), which is proportional to E, with thickness. of atom in octahedral sites, has few expected modes. One strong IR
Consequently, the true absorbance (a) and absorption coefficients band is predicted from symmetry (Fateley et al. 1972) which con-
(A) are related to light transmission through sists of a doubly degenerate TO component and one LO component.
Transverse acoustic (TA) and longitudinal acoustic (LA) modes also
a = Ad = − ln[(Iϒ /I0 ) = − ln[1 − Iabs /I0 ], (6) exist, but symmetry considerations suggest that these should not be
detected by IR spectroscopy. However, IR spectra of compounds
where I ϒ is the transmitted intensity, I 0 is the incident intensity and
with the rocksalt structure often have weak IR modes superimposed
d is the sample thickness (Brewster 1992).
on the one strong IR band (Kachare, Andermann & Brantley 1972).
From Kirchoff’s law, emissivity (ξ = I emit /I 0 ) equals absorptivity
Such resonances are generally attributed to acoustic modes (Mitra
(
= I abs /I 0 ). From equations (3) and (6), ξ depends on frequency
1969). Resonances or combination/overtone modes cannot be pre-
and thickness:
dicted by symmetry analysis because these types of vibrations are
ξ (ν, d) = 1 − exp[−d A(ν)]. (7) due to interactions among the fundamental modes (Mitra 1969).
The presence of phonon interactions is implicit in the measurable
The emission spectrum is
widths of the peaks, whereas non-interacting vibrational modes
(T, ν) = ξ (T, ν)Ibb (T, ν), (8) would be manifest as delta functions in the dielectric constants
(e.g. Hofmeister 2001).
where I bb is the blackbody curve of Planck. The temperature depen-
FeO is the nominal composition for wüstite, but this mineral is
dence of ξ is derived from measurements of A at T. The intensity,
neither purely ferrous (Fe2+ ), nor perfectly cubic. The structure and
width and position of the peaks depend on T and  is truncated
physical properties of wüstite are reviewed by Hazen & Jeanloz
at a high frequency when I bb becomes negligible: this truncation
(1984). In summary, the real mineral contains a small portion of
frequency depends strongly on T.
ferric (Fe3+ ) iron, and a sufficient number of octahedra cation va-
Comparing the ideal absorbance calculated from Kramers–
cancies () to balance charge, i.e. 2Fe3+ +  = 3Fe2+ . However,
Kronig or classical dispersion analysis of reflectivity data to mea-
the situation is even more complicated because Fe3+ ions not only
sured absorption spectra requires two adjustments. In most spectro-
occupy the octahedral sites expected for the rocksalt structure, but
scopic, chemical and mineralogical literature, the following conven-
also occur in tetrahedral interstices that are normally not filled in
tion is used: absorbance (a scm ) and the absorption coefficient (Ascm )
the rocksalt structure. Therefore the fraction of octahedral vacan-
are defined by common, not natural logarithms:
cies is larger than the chemical formula (Fe1−x x O) implies. Thus
ascm = Ascm d = − log(Imeas /I0 ). (9) the vacancy content is usually not explicitly stated, and composi-
tions are indicated as Fey O. The distribution of vacancies and of
This convention is pervasive to the extent that equation (9) is in-
the cations is not random, but occurs as defect clusters, leading to
corporated in the software for commercial IR spectrometers and in
variable amounts of short- and long-range order within the basically
spectral analysis packages. From Fig. 1, the measured amount of
cubic structure. Also, synthesis tends to produce compositions near
light transmitted (I meas ) is affected by reflections at the front and
Fe0.93 O, by exsolution either of Fe metal or of magnetite (Fe3 O4 ).
back interfaces. Combining Fig. 1 with equations (6) and (9) gives
The above complications can potentially alter the IR spectrum from
Ad that intrinsic to pure FeO which has the rocksalt structure, and which
ascm = − 2 log(1 − R) (10)
2.3026 could form in a highly reducing environment possible in space.


C 2003 RAS, MNRAS 345, 16–38
20 A. M. Hofmeister, E. Keppel and A. K. Speck
ZnO has the hexagonal wurtzite structure, in which both atoms λ, µm
occupy tetrahedral sites. This structure is identical with the simplest 50 25 20 16.7 14.2 12.5 10
2H polytype of hexagonal α-SiC (Wyckoff 1963). From symmetry a
analysis (Fateley et al. 1972), one band is expected in each of the po- MgO 5
0.8

Absorption coefficient, 1/µm

larizations along and perpendicular to the c-axis, such that the latter
mode is degenerate. These modes are also Raman-active. Symme- 4
R
try predicts two additional acoustic, two Raman and two inactive

Reflectivity
0.6
modes. Although two IR bands exist for this structure, it is unlikely
that these can be distinguished for ZnO by thin-film or dispersive 3
methods, as these methods revealed only one strong band in the IR 0.4
spectrum of α-SiC (e.g. Speck, Hofmeister & Barlow 1999). The 2
underlying reasons for the equivalence of the IR modes from the A
two polarizations are that (1) the two cation–anion bond lengths in 0.2
the wurtzite structure are similar, and (2) the motion of the atoms 1

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

involved in producing the TO mode of E
c is related to the motion
producing the LO mode of E⊥c, and vice versa. Raman data on 0
100 0
wurtzite-type compounds (Arguello, Rousseau & Porto 1969) are
0
consistent with this analysis. Furthermore, the stacking disorder that
* b
forms the additional hexagonal or rhombohedral polytypes of SiC 80 -1
results in additional weak modes being superimposed on the strong

Dielectric Functions
modes in the IR spectra (Choyke & Patrick 1962). Such a stacking -2
disorder has not been reported for ZnO, but exists for ZnS. 60 ε2 im(1/ε)
FeS is the nominal composition for a family of structurally and -3
chemically complex minerals (see e.g. Vaughan & Craig 1978; Deer
et al. 1992; Gaines et al. 1997). As occurs in wüstite, some of the 40
-4
Fe is ferric, and charge balance is attained through cation vacancies.
Chemical compositions are reported as Fe1−x S to indicate the min- 20
-5
imum degree of defects. Compositions near Fe0.94 S (ferric troilite)
crystallize in the NiAs parent structure (Deer et al. 1992) which has -6
anions in hexagonal close-packing, whereas the cations are octahe- 0 * -7
drally coordinated. Symmetry analysis of NiAs gives mode types
that are identical to those found in the wurzite structure, suggest-
8 MgO c
ing only one IR mode would be observed for such ferric troilites.
The troilite structure, which is formed with lower ferric contents, x
< 0.06, is described as a superlattice of the NiAs structure which
Optical Functions

6
contains 12 FeS molecules in the crystallographic unit cell (Wyck-
n
off 1963). Using symmetry analysis, we predict two IR modes for
troilite when E is polarized along the c-axis and six IR modes perpen-
dicular to c (acoustic modes are not included). Because the modes 4
polarized perpendicular to c are doubly degenerate, these are more k
intense, and six bands should be seen in unpolarized measurements.
However, the relationship of the troilite structure to the NiAs parent 2
structure through multiple stackings of the small NiAs unit cell par-
allels the polytypism of SiC, suggesting that many modes are weak. *
It is possible that troilite could have fewer IR bands, or one strong 0
200 400 600 800 1000
peak and several weak ones. Wavenumbers, cm
-1

To complicate matters, other structures exist for Fe1−x S. Stacking

disorder analogous to that of hexagonal SiC exists as x increases in Figure 2. Reflectivity at normal incidence of MgO and derived functions.
the series Fe1−x S, such that the various polytypes are associated with Asterisks mark weak peaks. (a) Raw reflectivity (thick solid line, left y-axis)
specific values of x. Pyrrhotite, with x below or near 0.17, is com- was joined at 600 cm−1 by dividing the mid-IR by 1.125. Data below 125
cm−1 are approximated by constant R 0 = 0.266 from the dielectric measure-
mon in terrestrial environments, whereas troilite is almost always
ments (Fontanella et al. 1974). For the Kramers–Kronig analysis, R was as-
meteoritic in origin. Samples with x near 0 can occur metastably in
sumed to be constant at and above that at 2690 cm−1 . Reflectivity calculated
either a cubic form (isostructural with β-SiC) or a tetragonal form from the classical dispersion fitting parameters in Table 2 is shown as open
(mackinawite, isostructural with PbO). circles. Absorption coefficients (right y-axis) are determined from Kramers–
Kronig (light solid line) and classical dispersion (dotted line, almost hidden
by solid line). (b) Dielectric functions determined from Kramers–Kronig
5 R E S U LT S
(solid lines) and classical dispersion (dotted and dashed lines). Heavier lines
indicate  2 . One main peak is seen in  2 , but the shape indicates that it is a
5.1 Single-crystal reflection spectra of MgO composite of three unresolved components. (c) Optical functions determined
A strong, broad band dominates the reflectivity spectrum of MgO from Kramers–Kronig (solid lines) and classical dispersion (broken lines).
Heavier lines indicate n. Although the mode at 650 cm−1 is prominent in
(Fig. 2a), as expected from symmetry analysis. Superimposed on
reflectivity and n, it is subtle in A, k and  2 .
the intense band is a depression near 650 cm−1 . This feature, in


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 21
Table 1. Kramers–Kronig analysis for MgO using  = 9.83 (Fontanella λ, µm
et al. 1974) to constrain the data below 125 cm−1 . Frequencies and FWHM
are in cm−1 . The weakest peaks appear as shoulders and reliable parameters 40 25 20 16.7
3 0
could not be derived. Instead these were determined by classical dispersion
– see Table 2. The 646 and 286 cm−1 positions match frequencies observed
at the Brillouin zone edge (Peckham et al. 1967; Sangster et al. 1970). The 2.5
*
-0.01
mode assigned as transverse acoustic (TA) may be an artefact. **

Dielectric Functions
2 ε2 im(1/ε)
Assignment ν TO FWHM f ν LO
-0.02
TA 286 weak 290
1.5
LA 408, 430*
TO 415.3 33.8 8.05 739.0 -0.03
2TA ∼580 weak ∼580 1
TA+LA 646 ∼50 0.023 644

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

∗ These two poorly resolved peaks are seen in the expanded view of Fig. 3. -0.04
0.5 *
0 -0.05
violation of symmetry analysis, has been previously observed in 300 400 500 600 700
-1
MgO and isostructural LiF (Piriou 1964; Andermann & Duesler Wavenumbers, cm
1970). Although our spectrum repeats the essential features of the
earlier measurements, our profile differs near 400 cm−1 , and in Figure 3. Expanded view of the dielectric functions in Fig. 2(b), using the
same symbols. Asterisks mark the TA mode at 286 cm−1 and two poorly
the absolute values. The present reflectivity values appear to be
resolved, weak modes revealed in Im(1/) from 400 to 450 cm−1 . A very
correct, as these agree with independent dielectric measurements weak mode is suggested at 585 cm−1 by the mismatch of the Kramers–
(Fontanella et al. 1974). The previous, more squared profiles were Kronig and classical dispersion analyses, but may be an artefact, as this is
duplicated in our instrument by slightly mis-aligning the sample or close to the merger of the far- and mid-IR segments. The weak mode near
the spectrometer (not shown). These differences may also be related 650 cm−1 has been previously detected, e.g. Kachare et al. (1972).
to the lower, and frequency-dependent, resolution used in the pre-
vious studies involving scanning instruments. A detailed analysis
will be given in a companion paper that focuses on peak widths, The LO component at 737 cm−1 arises from a singlet, as it is re-
overtones, and the effect of high temperature (Hofmeister, in prepa- produced by one Lorentzian oscillator. The LO positions of the two
ration). The purpose of the current paper is to describe the major IR weak components (Table 1) are revealed by expanding the field of
bands, which could potentially be present in astronomical observa- view (cf. Figs 2b to 3) because these are located on a flat region
tional data, and to provide various optical functions. of the dielectric function. Similarly, a weak mode at 283 cm−1 ap-
Parameters from Kramers–Kronig analysis (Table 1) should be pears to be resolved in  2 and Im(1/) because this spectral region
accurate because the resulting dielectric functions are classical in is relatively flat. We assign this mode as a TA component, but it is
appearance: nearly Lorentzian shapes exist for the imaginary part of possible that this is an artefact arising in a relatively noisy section
the dielectric function,  2 , and the imaginary part of the inverse of of the spectrum.
the dielectric function, Im(1/) (Fig. 2b), and a derivative shape is In contrast, the TO components of the threefold multiplet could
seen for the real part of the dielectric function,  1 (not shown). The not be completely resolved because these are shoulders on the
derived absorption coefficient (Fig. 2a) is asymmetric with more steeply rising main peak: expanding the axis smears the features. The
intensity towards the LO position, as expected (see Wooten 1972). spectra show weak manifestations of these extra modes: a shoulder
The overall appearance of all functions obtained from Kramers– occurs in reflectivity from about 385 to 410 cm−1 (Fig. 2a) and ex-
Kronig analysis (Fig. 2) suggests that the functions and parameters cess intensity is seen in  2 on the high-frequency side compared with
are reliable. Using Wooten’s approximation for the high-frequency a single-oscillator fit (not shown). Similarly, an additional very weak
extrapolation gives more asymmetry to the peak than that shown in feature is seen near 590 cm−1 . The visibility of the weaker features
Fig. 2, which was obtained by assuming that R is constant for fre- strongly depends on their position in the spectrum, relative to that
quencies above the range measured here. Appearance of asymmetry of the strong mode, and the same feature can be resolved to varying
in the derived dielectric peak, regardless of the approximation used degrees in the raw data and functions derived by Kramers–Kronig
to extrapolate the results to high frequency, points to the main IR analysis. This observation is clarified by comparing the expression
peak being a multiplet. More problems arise with Wooten’s approx- of the 650 cm−1 peak with that of the 285 cm−1 peak in Figs 2 and 3.
imation because the measured R in the transparent regions is higher The weak peaks listed in Table 1 were derived by examining both the
than the true value, because of back reflections (discussed below). raw data and the derived functions. Their existence is confirmed by
Back reflections do not occur in the absorbing parts of the spectrum classical dispersion analysis, in that the measured reflectivity could
because the strong attenuation of the light does not allow measur- not be fitted with fewer than three peaks near 400 cm−1 (Table 2).
able intensity to reach the back side of the sample or the internal The fit to the region near 590 cm−1 could be further improved by the
cracks. addition of another oscillator. However, because this is near the join
The details and weak features that appear in the functions (Figs 2b of the far-IR and mid-IR frequency regions, and because the existing
and c, and Fig. 3) confirm that the main band is composed of over- fit is reasonable (Figs 2 and 3), we did not complicate the analysis
lapping peaks. For example, in Im(1/) near 400 cm−1 , a doublet with this additional peak. For similar reasons, the peak at 285 cm−1 ,
of LO peaks is seen in the expanded view of Fig. 3. This suggests which could be the TA mode, was not include in the fitting.
that the main TO feature is composed of three poorly resolved com- Considerably weaker features (barely above the noise level) exist
ponents because a strong LO band occurs at 737 cm−1 (Fig. 2b). at 380, 440 and 505 cm−1 . These features coincide with absorption


C 2003 RAS, MNRAS 345, 16–38
22 A. M. Hofmeister, E. Keppel and A. K. Speck
Table 2. Classical dispersion fitting parameters using ∞ = n 2 = 3.0176 λ, µm
10 5 4 3 2.5
(Fontanella et al. 1974). For MgO, the fit was made to the reflectivity data. 0.1 0.2
Frequencies and FWHM are in cm−1 .
a
R
Phase Assign. ν TO FWHM f ν LO ε0 0.08

Absorption coefficient, 1/µm

0.1
MgO TO1 403 18 1.50 407 9.39
TO2 416 20 3.90 738
LA* 429 18 0.95 426 0.06

Reflectivity
TA+TO* 650 75 0.025 645 0
CaO TO 298 32 9 570 12.4 A
MgS TO 240 31 11 435 15.8 0.04
CaS TO 232 22 9.5 417 14.6
∗ Resonances involving edges of the first Brillouin zone. The TO, LO -0.1
and LA branches cross and resonate within the Brillouin zone so the 0.02

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

assignments are intended to indicate the dominant character of modes. For
CaO and CaS, n =1.84 and 2.14, respectively (Gaines et al. 1997). For
MgS, n is 2.19 (Boswarva 1970). Although resonances for these substances 0 -0.2
1000 2000 3000 4000
occur within the main IR mode, these are neglected in fitting the thin-film Wavenumbers, cm
-1

data. λ, µm
5 2 1.25 1 0.8 0.625
0.1 0.4
peaks in the coated mylar beamsplitter, and appear to be artefacts of
less than perfect alignment of the spectrometer and its accessories, b
R
or of the sample with respect to the reference. (Slight motions of 0.08 H O
2
the thin, mylar beamsplitter are also a likely cause.) The features 0.3
are resolved because of the high signal-to-noise ratio. We there- calc. R

SCM absorbance
fore omitted these probable artefacts from the classical dispersion 0.06
Reflectivity

analysis. Weak peaks exist near 850 cm−1 , but were not included
0.2
in the analysis, because this transparent spectral region is affected
by a secondary reflection (as sketched in Fig. 1) and because the 0.04 ascm
absorption measurements provide better constraints (see below).
Roughly 300 synthetic spectra with differing numbers of modes 0.1
and mode parameters were calculated to probe the limitations and 0.02
trade-offs in fitting, particularly regarding the threefold multiplet.
There is little leeway in the fitting because the steep reflectivity
edge at high frequency and the gradual slope in reflectivity at low fre- 0 0
5000 10000 15000
quency provide opposing constraints (Fig. 2a). This steep slope con- Wavenumbers, cm
-1

strains the FWHM as 18 to 20 cm−1 for all components of the multi-

plet, the maximum reflectivity constrains the oscillator strength f i of Figure 4. Single-crystal IR data of MgO at high frequency. The y-axes
the strongest peak, and the gradual slope from 300 to 400 cm−1 con- are expanded from those of Fig. 2. (a) Mid-IR. Grey line = raw reflectivity
strains f i of the weaker peaks (Table 2). The asymmetry of the main from a cleaved surface. An internal crack at about 0.5-mm depth provided
peak requires that the component near 405 cm−1 be more intense a back reflection. Black line = raw reflectivity from a polished surface. The
thickness is 1 mm, providing a back reflection with some contribution from
than the high-frequency component at 429 cm−1 . The position of
absorption. Plus signs = raw reflectivity from a 1 mm thick crystal that
the main LO band also constrains f i of the main TO peak. the manufacturer had polished on one side. Grey = raw reflectivity from
The existence of the threefold multiplet is corroborated by the another crystal. Circles = reflectivity calculated from classical dispersion
positions and numbers of IR overtone and combination bands (Table 2). Heavy, jagged black line = absorption coefficient (right y-axis)
(Hofmeister, in preparation; see Section 5.2) and of second-order derived from Kramers–Kronig. Short dashes = absorption coefficient de-
Raman peaks (Manson, der Ohe & Chodos 1971). Although the Ra- rived from classical dispersion analysis. (b) Near-IR and visible regions.
man spectra are adequately described in terms of combinations of Black line = measured absorbance for 0.98-mm thickness, doubly pol-
the IR and acoustic fundamentals, the IR overtones are too numer- ished surface. Surface hydration occurs as both OH and H2 O species. Grey
ous (see next section) to have originated from only one strong IR lines = reflectivity from t = 1 mm, singly polished crystal. The mid-IR,
fundamental. The multiplet exists because the resonances occur be- near-IR and visible data were acquired in that order, and over the period of
3 days the sample hydrated significantly. The source of the large amount of
tween the TO, LO and LA modes at various places in the Brillouin
noise for the visible detector is unclear. Square = reflectivity calculated from
zone (which is the reciprocal of the primitive cell, and is used to the measured index of refraction in yellow light (Fontanella et al. 1974). Be-
describe scattering of X-rays and particles from the crystal lattice). cause of the problems with back reflections for the three small, transparent
As shown by inelastic neutron scattering (Peckham, Saunderson & samples, we did not try to improve the signal-to-noise ratio in the visible,
Sharpe 1967; Sangster, Peckham & Saunderson 1970), for a large nor did we try to repolish to remove the water, and other crystals that were
portion of the Brillouin zone, the LA mode is situated between the thinner but with less hydration had similar problems.
TO and LO modes. The TO and LO modes are described as compo-
nents, but in actuality these are the same vibration. IR and Raman 2001). The IR modes are mixed through phonon–phonon scatter-
methods excite modes at the centre of the Brillouin zone only; how- ing. A detailed discussion of the known dependence of vibrational
ever, these modes interact with all vibrational modes in the crystal frequencies on position in the Brillouin zone, the IR resonances,
to various degrees, as indicated by finite peak widths (Hofmeister and the overtone-combination bands in MgO is in preparation


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 23
(Hofmeister, in preparation). Similar behaviour was seen in the IR λ, µm
spectra of Mg2 SiO4 and Fe2 SiO4 with the spinel structure, which 20 10 5 4 3.33 2.5
also have simple IR spectra (Hofmeister & Mao 2001). For the cur-
a
rent paper, it suffices that the measured reflectivity is reproduced MgO
with a threefold multiplet (Fig. 2a), and the dielectric and optical 2
HO
2
functions from the Kramers–Kronig and classical dispersion analy-
ses (Figs 2b, c and 3) are in excellent agreement over the fundamental
region. 1.5
Reflectivity data from 1000 to 4000 cm−1 (10–2.5 µm) contain

SCM Absorbance
some additional weak and broad peaks (Fig. 4). Their poor resolu- d =0.006 mm
tion, along with intensities <1/100th of the fundamentals, and broad 1
between KBr disks

widths, suggests an origin as overtones. The spectrum above ∼1200

cm−1 is affected by a back reflection, and thus the peak parameters HO
2
are not trustworthy, nor are the values of R (note expanded y-axis

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

0.5
in Fig. 4). The back reflection exists to various degrees in all mea- d = 0.28 mm
surements, regardless of whether a cleaved or polished sample was 0.10 mm in microscope
used, or whether the sample was doubly or singly polished. The back 0.10 mm
reflection should double the light received, less a contribution for 0
500 1000 1500 2000 2500 3000 3500 4000
transmissivity departing from unity (Fig. 1). Reflectivity in excess -1
Wavenumbers, cm
of the classical dispersion calculation is seen for the cleaved crys- λ, µm
100 50 25 20
tal and the three polished crystals of different thicknesses (Fig. 4).
2.5
The differences between the spectra are attributed to some surface
hydration on the two polished samples, and its greater thickness
b
for the polished sample in Fig. 4(a), leading to more absorption.
Multiple reflections existed for the cleaved crystal because of the 2

internal cracks developed during cleaving. The intensities of the ab-

sorption features are exaggerated because of the path, and corrupted
SCM Absrobance
wedge
because both reflectivity and absorption are superimposed. Fig. 4 1.5 15 µm
MgO
shows, however, that even the exaggerated overtones at ν > 1200 crystal
cm−1 have sufficiently low intensity that these are unlikely to ap- d = 100 µm I I
pear in astronomical observations for dust of µm sizes. Owing to 1 I
the excess reflectivity and exaggerated absorptions, the absorption
coefficient from the Kramers–Kronig analysis first goes negative, MgO
and then steeply rises with increasing ν (Fig. 4a). The problems 0.5
film
exist into the visible region (Fig. 4b). The absolute values are gen-
erally larger than that predicted from independent measurements of calc. from refl.
n, consistent with back reflections. The one exception is the visible 0
range. The lowered R-values in the visible may be connected with 100 200 300 400 500 600
-1
surface hydration, which increased over the course of the measure- Wavenumberc, cm
ments, as this effectively reduces the polish. It is difficult to obtain
Figure 5. Single-crystal absorption spectra of MgO and comparison with
accurate visible spectra with a Fourier transform instrument, and films. (a) High-frequency spectral range. The doubly polished samples (solid
the low R may be due to baseline drift. Because of the combina- lines) were placed on an aperture. The large baseline absorbance of the
tion of reflection and absorption existing at normal incidence for thicker sample is due to an imperfect surface polish. Peaks near 2900 cm−1
such weakly reflecting and absorbing samples (Fig. 1), the spectra are organic residues from polishing. The thinner piece was also viewed in
of MgO with thickness
1 mm are not reliable in the transparent a microscope (dotted line). Despite the fact that the area viewed was at the
region. For extrapolation to high frequency, the classical dispersion centre of the sample and was about half of the total sample area, stray light
results should be used for A, n and k above ∼1000 cm−1 . From and scattered light were received by the detector, providing both incorrect
Fig. 4, the calculated reflectivity gradually increases with ν above relative intensities and spurious peaks. For high absorbance (a scm > 2.5),
about 2000 cm−1 to R = 0.0724 for yellow light (∼18 000 cm−1 ). the values cannot be trusted, even for samples sitting on an aperture because
of detector limitations. Both types of errors underestimate the true values of
This extrapolation should be valid until the metal–oxygen charge
a. Dashed line = spectrum of a 6-µm film. The rise towards high ν is due
transfer bands in the ultraviolet are encountered (see e.g. Blazey to scattering for this incompressible substance. Peaks due to sorbed water
1977). are labelled. (b) Far-IR spectral range. Heavy line = spectra from a 0.10-
mm crystal on an aperture. Because a room-temperature detector (deuterated
triglycerine sulphate) was used, resolution was limited to 4 cm−1 . Dotted
5.2 Single-crystal and thick-film absorption spectra of MgO
line = cleaved wedge of about 15-µm thickness. Thin line = spectra from
The prominent fundamental bands in the IR spectra are accompa- a wedge with t = 15 µm, corrected for surface reflections (equation 10).
nied by overtones and combinations occurring at higher frequency. Grey line = ascm (equation 9) obtained from the Kramers–Kronig analysis.
Absorption spectra were obtained to constrain better the optical Dashed line = thin film of MgO with t near 1 µm. A shoulder at 285 cm−1
functions where the crystals are largely transparent, and to try to is assigned to the TA peak.
determine the thicknesses of the films from which the IR funda-
mentals were studied, which could not be accurately measured in a
binocular microscope.


C 2003 RAS, MNRAS 345, 16–38
24 A. M. Hofmeister, E. Keppel and A. K. Speck
Direct observation of overtones and combinations features at ν > 5 100
1000 cm−1 requires thicknesses of ∼0.1 to 0.3 mm (Fig. 5a). Thicker
(0.5 to 1 mm) crystals were studied at frequencies up to 8500 cm−1 , total a
and indicated sorbed H2 O, but were otherwise featureless (Fig. 4b). 4 coverage 80
MgO is hygroscopic and polished surfaces become dull (foggy) after
%T

% Transmission
exposure to air over hours to days. Fine powders also accumulate

Absorbance
surface hydration on a similar time-scale, as indicated by features 3 60
near 1500 and 3400 cm−1 in the thick-film spectrum (Fig. 5a). At
low frequencies, the thick-film spectrum compares reasonably well
with the single-crystal results, considering the differences in optical 2 40
path. Towards high ν, the absorbance of the film strongly increases
(Fig. 5a). This behaviour is due to scattering. Because MgO is hard
and incompressible, the grain boundaries were not eliminated by 1 20
A
pressing between KBr discs.

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

The overtone features are weak but are in good agreement with
0 0
absorption coefficients calculated from Kramers–Kronig analysis 1 100
for the sample with the least back reflections. The thinnest crystal
has a feature at 850 cm−1 with an absorbance of 0.0168 µm−1 , which 50 % b
is within the uncertainty of a noisy feature (Fig. 2a) at 847 cm−1 0.8 coverage 80
with A = 0.0155 µm−1 . The 6-µm film has a stronger absorbance %T
of 0.035 µm−1 but this value is affected by interference fringes.

% Transmission
Absorbance
Absorption spectra were acquired at low frequencies (Fig. 5b) to 0.6 60
probe the possible feature at 290 cm−1 in the reflectance (Fig. 2).
The feature is of comparable strength to the noise for the thick-
nesses examined (Fig. 5b), whereas for thicker samples the steep 0.4 40
rise in absorbance as frequency increased towards the main peak
obscures this feature (not shown). The feature is too weak to be
0.2 A 20
seen in the thinner films (Fig. 5b). Although it appears that this
TA mode is real as all data are consistent (Table 1), it is not suf-
ficiently resolved over the noise level of the spectra to be used to
determine film thickness accurately. It can serve as a rough check on 0 0
200 300 400 500 600 700 800 900
Kramers–Kronig analysis. From Fig. 5(b), the true absorption coef- -1
Wavenumbers, cm
ficient (peak minus sloping baseline) is ∼ 0.12 × 2.3/0.1 mm = 3
mm−1 . The calculated absorption coefficient of ∼15 mm−1 (Fig. 2a) Figure 6. Schematic indicating how spectral artefacts are created from light
is larger, but may be enhanced (i.e. roughly doubled in strength) by leakage. For each section, the circle (top) illustrates conditions of sampling:
a back reflection, as were the weak modes at high frequency (Fig. 4). heavy dashed line = the transmission spectrum of MgO with t = 1 µm;
To a good extent, these values differ simply as a result of the low light dashed line = the corresponding absorbance. Heavy solid line = trans-
signal-to-noise ratio and the weakness of the feature. We also com- mission for a sample that is opaque near 430 cm−1 ; light solid line = the
pared the measured absorbances in the far-IR with values calculated corresponding SCM absorbance. (a) Ideal case where the aperture is covered
entirely by the sample at a uniform thickness. Low values of transmission
from the Kramers–Kronig relationships, after subtracting the con-
drastically affect the absorbance. (b) Case of uniform thickness, but par-
tribution for the reflections (equation 10). The wedge absorbs more tial coverage. Light leakage alters the spectrum from that intrinsic to the
light than expected between 100 and 330 cm−1 (Fig. 5b). Probably material. Note that absorbance is about the same regardless of whether the
this is due to internal scattering, as the comparison for the thick crys- sample partially transmits near the peak or whether the sample is opaque at
tal showed a larger discrepancy. Once the absorbance is moderate frequencies near the peak.
(>0.4), calculations and measurements are in good agreement, and
for strong absorbances (>1.5) the reflectivity correction has little
impact (Fig. 5b). Absorption measurements of even single crystals response becomes non-linear. For this reason, commercial software
are therefore an approximation of the ideal absorption coefficient, often truncates absorbance (a SCM ) at values of 2, 3 or 4, at the dis-
with some problems occurring in the more transparent regions. cretion of the manufacturer. In our experience, absorbance near 1
The spectra collected from samples lying on an aperture show an minimizes artefacts of this nature. This observation is consistent
increase in the intensity of the peak near 1000 cm−1 that is propor- with use of a > 2 to describe optically thick conditions (Siegel &
tional to thickness, as expected from the connection of peak area Howell 1972).
or height with concentration (Beer’s law). However, the intensity of A different type of non-linear response is seen in the spectrum
the peak near 850 cm−1 does not increase as much as the difference obtained from a crystal using an IR microscope. In this case, light
in thickness would suggest. Note also the high degree of noise for leaking around the edge of the sample reduces the absorbance
measured absorbance that is greater than 2. The loss of proportion- from the actual values (Fig. 5a). As shown schematically in Fig. 6,
ality is a problem of sampling. Detectors are not noise-free, and, the leakage of light reduces absorbance of the strong peaks more
as the transmission decreases, it will lie within the uncertainty of than the weak peaks (or shoulder), creating either rounded or
‘zero’ before the true absorbance value is infinite (see Fig. 6a: a flattened profiles. Placing apertures at the image points before
small amount of noise in the transmission spectrum for the mini- and after the sample minimizes the light leakage, but not all IR
mum near 425 cm−1 strongly alters the maximum value of a). Thus, microscope systems use the double-aperturing technique. Nei-
at some low value of transmission (high absorbance), the system ther does the second aperture completely remove sampling artefacts


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 25
λ, µm (Refner, Hornlein & Hellgeth 1997). When light leakage is present,
50 25 16.7 12.5 10 8.3 peaks are seen at frequencies where the sample is opaque [as shown
in Fig. 5(a) when the same sample covers the aperture]. At fre-
1 a 6
quencies above 800 cm−1 , where the sample transmits to a greater
M degree, peaks are seen at the same positions as the sample of equal
5 thickness that was masked by a physical aperture, but the relative in-
0.8
tensities of the peaks, their components and shoulders are incorrect
SCM Absorbance

MgO
(Fig. 5a). For strong absorbers, the spectral profile is a predomi-
4
nantly determined by the amount of light leakage, as indicated by
0.6 M FI Fig. 6. Note that opaque regions of a sample with light leakage (i.e.
3 a dispersion) provide finite absorbance, and that infinite absorbance
ZnO cannot be distinguished from a = 4 at 50 per cent coverage. These
0.4
18
Mg O conclusions are irrespective of the type of material sampled. Thus,
I F 2
I for dispersions, the fraction of areal coverage is a primary factor in

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

I
determining the appearance of the spectra.
0.2
1
F
LO 5.3 Thin-film spectra of diatomic compounds
xtl
0 0
1.6
5.3.1 Oxides
b
1.4 Absorption spectra obtained from optically thin films of the sim-
CaO ple oxides consist of a broad, asymmetric peak with superimposed
1.2 structure (Fig. 7). For absorbances below 2, the spectra of MgO
SCM Absorbance

F and CaO essentially scale as the sample is thinned, in accordance

1 with Beer’s law. Multiple spectra were not acquired for ZnO. For
the FeO sample, very low absorbance (the maximum below 0.2) is
0.8 F
needed to obtain peak profiles that do not change upon further thin-
I ning. Because of this, and because of superimposed scattering (the
0.6 rise to the visible, Fig. 7b), the width measured from the absorption
spectra is large for FeO (Table 3). The width for FeO is difficult to
0.4 F determine as it depends on the baseline. The CaO and ZnO samples
have smaller widths than that of MgO, and are more symmetric.
0.2 Their longitudinal acoustic frequencies lie below that of the main
1.5 µm band (Table 3), whereas in MgO the TO and LA frequencies are
0 nearly equal. In FeO, the sum of the TA and LA frequencies is
1
c 382 cm−1 which is close to the main peak position for the thicker
samples (Fig. 7b). The combination of scattering and a multicompo-
nent peak probably contributes to the dependence of spectral profile
0.8
Fe O on thickness.
SCM Absorbance

3 4
I The peak for Mg18 O is shifted to 469 cm−1 and is broader than that
I
of MgO (Fig. 7a). The large breadth is probably due to the enriched
0.6
sample containing a significant proportion of the normal isotope
16
O. The normal sample would have levels below 0.2 atom per cent
I of the heavy isotope (Criss 1999). Isotopic exchange should affect
0.4 the mass without altering the lattice constant, producing a decrease
I
in frequency. Instead, an increase in ν is observed, and attributed to
FeO changes in the resonances between the IR and acoustic modes.
0.2 I The superimposed structure on the high-frequency side of the
M
peaks for MgO and CaO is mostly interference fringes (Fig. 7a).
The fringes are related to the spacing between the diamond faces,
0 which is larger than the sample thickness, and no simple method
200 400 600 800 1000 1200
Wavenumber, cm-1 exists to remove them without affecting the spectra. The spacing of
the fringes is broken by some weak combination peaks, and the LO
Figure 7. Thin-film IR spectra of X–O compounds. (a) Data on MgO and modes. For MgO, its acoustic combination band at 645 cm−1 and its
ZnO, showing procedures to merge spectral segments. Thick line = Mg18 O; strongest IR overtone near 850 cm−1 are also present. The strength
long-dashed line = ZnO; all other lines are MgO. M indicates merge point. of the 850 cm−1 band suggests a 2.3-µm film from comparison
F indicates interference fringes. Short lines indicate peaks in addition to
with the single-crystal data, whereas t = 1 µm is suggested from
the fringes (e.g. the overtones in MgO near 860 cm−1 ). The left-hand axis
is for the thick film of MgO (grey line); a 6-µm film (dotted line), and
data from a crystal scaled to t = 6 µm (thin line near x-axis marked ‘xtl’). high n is due to index of refraction mismatch. Arrows indicate approximate
Comparison of the overtone intensity indicates that the thin film of MgO has extrapolation between segments and to high ν. (c) Natural Fe3 O4 (light line)
t near 1 µm thick. (b) Effect of thickness on CaO. Heavy dotted line = film and FeO (dotted and thick lines). Scattering causes the rising baseline to
between KBr discs, obtained at t = 6 µm but scaled to t = 1.5 µm. The rise to high frequency. Vertical lines show peak positions.


C 2003 RAS, MNRAS 345, 16–38
26 A. M. Hofmeister, E. Keppel and A. K. Speck
Table 3. Thin-film absorption spectra parameters (in cm−1 ) and related information. Width is twice the FWHM on the low-frequency side
of the main peak. For comparison, acoustic frequencies, reduced atomic masses and bond lengths are included. Cation–anion distances and
acoustic (Debye) frequencies are from compilations of structural (Wyckoff 1963; Gaines et al. 1997) and elastic data (e.g. Bass 1995; Knittle
1995). Acoustic frequencies of the sulphides were estimated (except for ZnS) because the shear moduli are not available. For MgS and CaS,
resonances in the IR were used to estimate some acoustic frequencies.

MgO CaO FeO ZnO MgS CaS FeS ZnS

ν TA 271 180 116 130 225* ∼130 ∼100 108

ν LA 409 295 266 275 282* 270¶ ∼200 207
ν shoulder 230§
ν main 423 318 363§ 420 255 242 256§ 274#
width 110 60 ∼130 80 60 30 12
ν TA+LA 650 ∼500† 390 437 ∼400 294§
ν LO 560 570 590 390 380 352#
ν LO –ν TO

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

325 270 200 170 145 150 80
µ (amu) 9.6 11.4 12.4 12.9 13.8 17.8 20.4 21.5
X–A (Å) 2.10 2.40 2.15 1.98 2.58 2.84 2.5
2.35
∗ Present in thicker samples of MgS, whereas the main band at 255 cm−1 is no longer resolved. †Also seen by Jacobson & Nixon (1968) and
Galtier et al. (1972). § The main peak in FeO shifts to 390 cm−1 for thicker samples. ¶ Inferred from a resonance. § Troilite has several weak
peaks, but no strong main peak.
Trolite has six Fe–S distances at 2.359, 2.379, 2.416, 2.504, 2.565 and 2.721 Å (Evans 1970). The distances
for the other hexagonal phases (ZnO and ZnS) are averages, with a spread in bond lengths of ∼0.02 Å. #Raman data on ZnS (Arguello et al.
1970) are from averaging the polarizations of the hexagonal polymorph.

comparison with the thick film. The agreement is good, consider- surements. (A quantitative comparison will be made in Section 6.3.)
ing that the fringes interfere with determination of peak strengths. A thicker film (∼2 µm) is also shown. In this case, the rapid rise
For the thinnest FeO film, a shoulder appears at 570 cm−1 . From ends with a shoulder at the LA resonance (Fig. 7a). This data better
comparison with previous reflectivity data, this is the LO mode. Sim- constrains the join to the thinnest film.
ilarly, for ZnO, a shoulder exists at 580 cm−1 , which is between the Applying the same procedure to CaO indicates a rapid rise at 590
LO positions determined from Raman spectroscopy (Arguello et al. cm−1 , confirming the above assignment to the LO mode (Table 3).
(1969). For MgO, the LO position is not revealed in the thin-film The thicker film of CaO appears to be 1.5 µm, as this thickness
spectrum; instead, the resonance peak at 650 cm−1 breaks the spac- provides the best joint between the thin-film data and the spectrum
ing of the fringes. A similar shoulder is seen in the CaO spectrum, from t = 6 µm. The rise to high ν is scattering, and the intrinsic
which is assigned as the LO component, because sums involving the spectrum is approximated by a gradual decrease in absorbance above
acoustic modes do not match this position (Table 3). The resulting 750 cm−1 .
splitting between the LO and TO modes systematically decreases for
the oxides (Table 3). This marker combines the effects of oscillator
5.3.2 Sulphides
strength and damping coefficient.
Joining the two thin-film spectra results in an asymmetric peak for Far-IR absorption spectra of the cubic sulphides MgS and CaS also
all of the oxides (ZnO data are not shown for clarity). The asymme- have one main peak, with some structure (Fig. 8). Mid-IR spectra
try increases from ZnO to CaO to MgO. For FeO, strong scattering revealed no additional features and are not shown. Their peak po-
makes it difficult to obtain data except for the thinnest films. For sitions are very similar, but occur near 250 cm−1 rather than near
these low thicknesses, the LO mode becomes less important (the 400 cm−1 for the oxides, consistent with the larger mass of the sul-
ideal is approached) and the peak appears symmetric. Magnetite phur anion, and with the longer bond distances (Table 3). In addition,
exhibits the same sort of scattering baseline as FeO (Fig. 7c). From the peaks are much narrower than those of the oxides. Structure ex-
comparing Fig. 7(a) with Figs 2(a) and 5(a), the thin-film spectra ists on the main band of MgS, which is attributed to a resonance
for MgO are clearly delivering excess intensity at high frequency, with the acoustic modes. The positions of the shoulders are in rea-
which provides the asymmetry. This compound is highly reflecting sonably good agreement with estimates of the acoustic frequencies
over a wide frequency range (Fig. 2a), which creates an artificially (Table 3). The TO and LO positions are also compatible with fre-
high apparent absorbance, in accord with equation (10). Using equa- quencies observed for α-ZnS and CdS using Raman spectroscopy
tion (10) requires reflectivity data, and it is desirable to obtain the (Arguello et al. 1969). The peak widths and the LO–TO splittings
intrinsic absorption independently. We devised the following pro- are smaller than those of the oxides and decrease as the reduced
cedure. For MgO, appearance of the overtone at 860 cm−1 indicates mass increases (Table 3). The narrower widths and low LO–TO
a film thickness near 1 µm. The single-crystal absorption spectrum splittings for MgS and CaS are consistent with the reflectivity being
(Fig. 5a) was therefore scaled to t = 1 µm. Although the absorbance high over the narrow range where the sample mostly absorbs, and
values should have been adjusted for reflectance (equation 10), the with the measured absorbance reasonably representing the intrinsic
scaling reduces the values to near zero, making the correction unim- absorbance for the thin films.
portant. The single-crystal spectrum is truncated at 745 cm−1 , be- FeS has an extremely weak IR spectrum, consisting of a few
cause of the rapid rise in absorbance associated with the LO mode. narrow, weak peaks superimposed on a scattering curve. The same
The spectrum for the 6 µm thick film, scaled to t = 1 µm, rapidly type of baseline was seen for FeO, and it also has a weak peak
rises just below this frequency. This trend, extrapolated to meet the compared with the other oxides, but for FeO a strong IR mode still
thin-film measurement, represents the closest approximation to the exists, consistent with the wüstite structure being essentially that of
intrinsic absorption spectrum obtainable without reflectivity mea- rocksalt. FeS appears to lack a strong IR mode, which is attributed to


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 27
λ, µm this mineral occupying a much different structure, related to that
100 50 33.3 25 20 16.7 of NiAs. The correlation of peak positions with those of the di-
3.2
atomic sulphides listed in Table 3 is given for convenience. Our
I I a
2.8
data differ somewhat from the optical functions derived from pre-
SCM Absorbance

vious reflectivity measurements (Henning & Mutschke 1997) – see

2.4
Section 6.4.
TA LO

2
6 DISCUSSION
MgS
1.6
LA 6.1 Useful relationships between spectral parameters and
physical properties
1.2 TA + TO
For a simple harmonic oscillator
0.8
I 

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

ν= F/µ, F ∝ X −3/2 , (12)
0.4 where µ is the reduced mass and F is a force constant. The above
power-law relationship of F with bond distance (X) is phenomeno-
0
2.8 logical (Batsonov & Derbeneva 1969). In accordance with equation
b (11), the main peak position does not simply depend on atomic
masses, but instead is determined by a combination of the mass
SCM Absorbance

2.4
and the bond distance. For example, the low frequency of CaO
CaS is related to its large bond length and intermediate reduced mass
2 (Table 3, Fig. 7). Although equation (11) is not precisely adhered
to, the trend for the oxides (Fig. 9) is consistent with the above phe-
1.6 nomenological model, and suggests that the measured frequencies
are reasonably correct.
1.2 The Lydanne–Sachs–Teller (LST) relationship relates the high-
and low-frequency limits of the real part of the dielectric function
LA LO to the IR frequencies:
0.8
2TO  2 m 
 2
0 νLO νLOi
= = , (13)
0.4 ∞ νTO νTOi
i=1

0 where the product pertains to multiple band substances. The LST

1.6 relationship is derived by comparing the damped harmonic oscillator
equations in the low- and high-frequency limits (Burns 1990). This
1.4 relationship holds for each polarization separately. For the diatomics
SCM Absorbance

(except FeS), one band dominates, so the simpler formula is used

1.2
and FeS is omitted from consideration. The TO positions were taken
as 10 to 15 cm−1 (in round numbers) below the peak positions from
the thin-film spectra, because this relationship was seen for MgO
1
(cf. Tables 1 to 3). For non-opaque minerals,  ∞ is very close to
n 2 (equation 2). For the diatomic compounds examined here and
0.8 previously (Table 4) for which data on all four parameters exist, good
adherence to equation (12) exists, considering that  0 is generally
0.6 uncertain by 10 per cent.
FeS The good agreement in Table 4 also allows estimation of  0 if
0.4
both the index of refraction and IR data exist. FeO has a high di-
electric constant (Table 4), regardless of whether parameters from
c
the thin-film measurements (Table 3) or classical dispersion analysis
0.2
100 200 300 400 500 600 of reflectivity data (Table 5, discussed further below) are used. The
Wavenumber, cm
-1 value from Kramers–Kronig analysis [from Henning & Mutschke
(1997), listed in Table 5] does not fit the LST relationship, indicating
Figure 8. Thin-film IR spectra of diatomic sulphides. (a) Effect of thick- that R is incorrect at either low or high frequency (see Section 6.2).
ness on MgS far-IR spectra. Thin, vertical lines connect features among the Classical dispersion analysis of the thin-film data was also used to
spectra. The main band in MgS is inferred to be a multiplet. Some of the obtain  0 (Table 2, see Section 6.6). The results are consistent with
differences are artificially enhanced by low instrumental throughput. (b)
Effect of thickness on CaS far-IR spectra. These spectra mostly scale with
thickness, although the shoulders are exaggerated somewhat. (c) FeS. Thin removed. The absorbance of FeS is high because the films are relatively
black lines = far-IR data on varying thicknesses of commercial FeS. Heavy thick: otherwise the weak peaks could not be observed. The weak peaks
black line = FeS synthesized by Lauretta et al. (1996). Grey lines = mid-IR below 180 cm−1 are artefacts of low throughput and small amounts of water
data on commercial samples. Spectral artefacts near 385 and 440 cm−1 were vapour remaining in the instrument.


C 2003 RAS, MNRAS 345, 16–38
28 A. M. Hofmeister, E. Keppel and A. K. Speck
Table 4. Dielectric constants and frequencies (in cm−1 ). Values calculated
NiO MgO from the LST relation are in parentheses and boldface. A value near unity in
400
last column indicates adherence to the LST relation (equation 11). Dielectric
ZnO data are from compilations (Arguello et al. 1969; Boswarva 1970; Cook &
FeO
CoO Jaffe 1979; Subramanian et al. 1989; Gaines et al. 1997). The  0 -values have
uncertainties generally near ±10 per cent, whereas n and  ∞ are precise. For
FeO the other compounds, data are unavailable on  0 , but  ∞ = 4.796 for MnO,
-1

FeS
Peak position, cm

300 5.617 for NiO and 6.200 for CdO, all of which crystallize in the rocksalt
CaO structure (Gaines et al. 1997). FeS is opaque, so data on n are lacking.
ZnS MgS
CaS 0 TO 2
CdS 0 ∞ ν TO ν LO ∞ ( LO )
SrO BeO⊥c 6.87 2.95 724* 1098* 1.013
200 BeO
c 7.66 3.00 680* 1083* 1.007
MgO 9.83 3.018 409 737 1.003
300 ± 5† 578 ± 2†

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

CaO 11.95 3.378 0.953
FeO (14.4) 5.38 325# 534#
CoO 12.9 5.3 349‡ 550‡ 0.98
BaO
NiO 11.9 5.61 401‡ 580‡ 1.01
100 ZnO⊥c 8.7 4.052 413§ 591§ 1.049
PbS
ZnO
c 11.3 4.117 380§ 579§ 1.182
0.08 0.1 0.12 0.14 0.16 0.18 SrO 13.1 3.28 233 ± 2† 484 ± 4† 0.93
3/2 -1/2 -3/4 BaO (37) 3.57 132† 425†
1/Square root (µx ), amu A
β−ZnS 8.37 5.617 0.90

Figure 9. Dependence of the main IR peak position on reduced mass and ZnS⊥c 8.58 5.55 274§ 352§ 0.937
interatomic bond distance (equation 12). Dots = frequency determined from ZnS
c 8.00 5.65 274§ 352§ 0.856
thin-film absorption data. Triangles = FeS from thin-film data. Squares = CdS⊥c 8.7 5.29 243§ 307§ 1.030
frequency from single-crystal studies. The line is a least-squares fit to the CdS
c 9 5.36 234§ 305§ 1.052
single-crystal measurements of the oxides (Tables 3 and 4). The bonding in PbS (150) 16.8@ 71@ 212@
the sulphides has a metallic character and falls on a different curve. The light ∗ Loh (1968). †Jacobson & Nixon (1968); Galtier et al. (1972) also
Mg cation falls off the trends for both oxides and sulphides. analyse BaO data using classical dispersion analysis. ‡Gielisse et al.
(1965). #From classical dispersion analysis of the previous IR reflectivity
data by Henning & Mutschke (1997). § Raman data of Arguello et al.
the reflectivity measurements for CaO and MgO (Table 4), sug-
(1969).
Assumes that LO and TO values of the hexagonal polymorph
gesting that the values for CaS and MgS obtained from fitting the are valid for the cubic phase. @Classical dispersion analysis of Geick (1964).
thin-film data are reasonably accurate.
For FeS, data on the index of refraction are not available because
this mineral is opaque in the visible. Two constraints exist. First, the Table 5. Peak parameters for the main mode in FeO.
low-frequency dielectric constant does not vary much among the
sulphides (Table 4), suggesting that  0 of FeS is 8.5 to 9. Compari- ν TO FWHM f ν LO 0 Analysis
son of the oxides with the sulphides with the same cation suggests 325 71 10.9 525 ∼23 Kramers–Kronig*
that  ∞ of FeS is roughly 6 to 7. Roughly, the index of refraction of 325 71 10.9 563 16.3 initial fit
FeS is about 2.6. Secondly, we use existing reflectivity values ob- 325 71 9.0 534 14.38 final fit
tained by comparison with standards in a optical microscope. Such ∗ Kramers–Kronig analysis and measurements from Henning &
reflectivity values in the visible are available from the literature on
Mutschke (1997). The raw data were fitted using classical dispersion
ore minerals (mostly opaques, meaning that 30-µm thicknesses are with  ∞ = 5.38 from microscope observations (Bowen & Schairer
impenetrable to visible light). For wavelengths near 590 nm, the 1935) as an input parameter:  0 is a result. Frequencies and FWHM
compilation of Peckett (1992) provides R = 18.4 per cent for FeO, are in cm−1 .
11 per cent for α-ZnO, 17 per cent for cubic ZnS, 22 per cent for
MnS, 44 per cent for PbS, 40 per cent for FeS (troilite), 50 per cent
for NiS and 40 per cent for FeS2 . The hexagonal minerals are bire- in Table 4, which suggests that either k is significant (∼2) or  0 for
flective and the values reported are the average. Some of these values troilite is greater than 9.
√
can be checked against the index of refraction data ( ∞ in Table 4), Reflectivity measurements on polycrystalline FeS (Henning &
which indicates that R vis = 15.8 per cent for FeO, 11.4 per cent for Mutschke 1997) give R = 53.5 per cent at 1000 cm−1 , which is
ZnO and 16.5 per cent for ZnS. The match is excellent for ZnO inconsistent with microscope observations of R = 40 per cent in the
and ZnS. The discrepancy for FeO is probably due to the variable visible. The excess is attributed to back reflections from the small
structural chemistry of this mineral (Hazen & Jeanloz 1984), partic- crystals; see Section 6.4.
ularly the amount of magnetite exsolved in the wüstites specifically
examined for these two different measurements. Because the same
6.2 Re-analysis of FeO data
type of behaviour occurs for troilite, we infer that its high-frequency
reflectivity should also be within a few per cent of the microscope The wüstite sample from which IR reflectivity was obtained was
determinations of 40 per cent. Because troilite is an opaque, k is synthesized by quenching (Henning & Mutschke 1997) and is poly-
non-zero. For this case, equation (1) gives an upper limit of n = crystalline (Mutschke, private communication). Grain sizes were
4.4 for FeS in the visible (with k = 0). This value is significantly not reported. At ambient conditions R of FeO has a minimum near
larger than that estimated from trends with chemistry and structure 125 cm−1 and rises to lower frequencies (Fig. 10a). A reflectivity


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 29
λ, µm spectrum of single-crystal Fe0.93 O obtained previously (Bowen,
400 50 25 16.7 12.5 10 Adler & Auker 1975) resembles Fig. 10(a), but additional weak
0.8
bands are seen at 300 and 569 cm−1 , the high-frequency side of the
FeO a peaks is shifted to longer wavelength, the maximum peak reflectiv-
Fe Mg O ity is about 10 per cent higher, and the minimum occurs at 270 cm−1 .
0.6 0.56 0.44
Bowen et al. (1975) did not perform a quantitative analysis, so an

Absorption coefficient, 1/µm

H&M (1997)
2 in-depth comparison cannot be made. Kramers–Kronig analysis by
Henning & Mutschke (1997) and previously (Henning et al. 1995),
0.4 R calc. from which assumes that R = 1 at ν = 0 (Begemann et al. 1995) and
Reflectivity

K-K parameters
that R is constant at frequencies above the data, leads to a value of
n in the IR (Fig. 10b) that is larger than the n = 2.32 determined
1 for the visible region from optical microscopy (Bowen & Schairer
0.2
calc. R 1935). The high-frequency reflectivity is not correct because strong
final fit electronic transitions contribute negligibly to k in the visible, as im-

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

A plied by equation (3) (also see Fig. 2). Only for opaque minerals
0 A
final fit
such as FeS is k affected. Thus R of wüstite is determined by n at
high frequency through equation (1). The excess reflectivity at high
0
50 frequency in the measurements is attributed to reflections from the
0
back surface of the topmost layer of crystals in this polycrystalline
b sample as schematically shown in Fig. 1. As shown above for MgO,
40 -0.2 even thicknesses of 1 mm show back reflections in the mid-IR re-
gion. Sizes of this order are impossible to obtain through quenching,
Im(1/ε) and thus back reflections are unavoidable in the polycrystalline sam-
Dielectric function

30 -0.4 ples studied by the Henning group. Consequently, the dielectric and
H&M (1997)
optical functions obtained by Kramers–Kronig analysis (Henning
et al. 1995; Henning & Mutschke 1997) err at high frequency.
-0.6
20 The upturn in R at frequencies below 125 cm−1 is a spectral
artefact for similar reasons, leading to n and k being suspect at
-0.8 all frequencies. Work in progress suggests that single crystals with
10 compositions between MgO and Mg0.44 Fe0.56 O have constant re-
ε
2 flectivity from 50 to 200 cm−1 . Henning and co-workers extrapo-
final fit -1
late their trend to R = 1 at ν = 0 cm−1 , assuming that FeO is a
0 free carrier. This assumption is contradicted by previous work. It is
true that the electrical conductivity of single-crystal wüstite is high,
6 c 10 ohm−1 cm−1 at room temperature, but neither a Hall voltage nor
intrinsic conduction has been observed (Bowen et al. 1975). FeO
5 is thought to be a partially compensated semiconductor where the
compensating defects are oxygen vacancies or trivalent impurities
Optical functions

4 (Bowen et al. 1975). Thus the electronic properties depend on the

n non-stoichiometry of the particular sample. The wüstite prepared
final fit
k by Henning et al. (1995) has 10 per cent Fe3+ from wet chemical
3
nvis analysis. The widths of the X-ray diffraction (XRD) peaks were
H&M (1997) sufficiently large that cell dimensions could not be accurately deter-
2 mined, suggesting a high degree of lattice disorder (Henning et al.
1995). In addition, in polycrystalline wüstites, Fe3 O4 is commonly
1 co-precipitated or exsolved (Hazen & Jeanloz 1984). Magnetite ex-
solution is expected for compositions similar to the sample examined
0
by Henning & Mutschke (1997) (i.e. x > 0.07), and occurs rapidly
0 200 400 600 800 1000 and at scales too fine for detection by X-ray diffraction (Hazen &
Wavenumber, cm-1 Jeanloz 1984). The variable structure and chemistry for individual
samples and the differences in R between the previous studies cast
Figure 10. Re-analysis of FeO reflectivity data at quasi-normal incidence
doubt on the rise in R towards ν = 0 being intrinsic behaviour, but
at room temperature and derived functions. (a) Raw reflectivity (thick solid
line, left y-axis) from Henning & Mutschke (1997). For their Kramers–
Kronig analysis, R was assumed to be constant above the measurements. (dotted line). (b) Dielectric functions determined from Kramers–Kronig
Reflectivity calculated from the classical dispersion fitting parameters in (solid lines) and classical dispersion (broken lines). Heavier lines indicate
Table 5 is shown as open circles. R calculated from a classical dispersion 2 . One main peak is seen in 2 ; no components are indicated, within the un-
analysis using their Kramers–Kronig parameters is shown as a dot–dashed certainty of the data. (c) Optical functions determined from Kramers–Kronig
line. Classical dispersion analysis gives much lower R at high ν in order to (solid lines) and classical dispersion (broken lines). Heavier lines indicate
match n in the visible. The raw data apparently have a secondary reflection. n. The filled square is n from optical microscopy. Problems in measured R
The rise in R as ν approaches zero is probably also due to back reflections. are apparent from comparing n and k from Kramers–Kronig and classical
Absorption coefficients (right y-axis) are determined from Kramers–Kronig dispersion analysis at low and high frequency, as well as the value of n in
of Henning & Mutschke (light solid line) and classical dispersion analysis the visible (square).


C 2003 RAS, MNRAS 345, 16–38
30 A. M. Hofmeister, E. Keppel and A. K. Speck
rather suggest that the behaviour at low ν could depend on the im-
purity content and on the presence of exsolved magnetite or iron a b c
metal (typically of 0.03 < x < 0.07). More likely, the rise towards
low frequency is an instrumental problem in that the pyroxene glass
examined by Henning & Mutschke (1997) has a similar rise in R
towards ν = 0. d<λ
One possibility is inaccuracies in ratioing the sample spectrum
to the reference spectrum when the throughput is low. It is not triv-
ial to obtain precise values of R below 100 cm−1 . If a SiC globar
was used as the source in this frequency region, its intensity is low.
R=16.7% R from Eq.1
Mercury-arc lamps produce more light at low frequency, but their
output is not completely stable, leading to baseline drift. The partic- Figure 11. Ray paths (arrows) in a DAC. The trapezoids are the diamonds,
ular beamsplitter used also has an effect. The broad-range, coated the light grey represents an air gap, and the sample is the stippled grey area.
mylar beamsplitters have little throughput below 50 cm−1 , whereas (a) Empty cell used as the reference. The reflections are caused by a small

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

using the uncoated mylar beamsplitters of various thicknesses re- air gap. Using n = 2.387 for diamond, k = 0 in equation (1) gives 16.7
quires merging multiple segments. In addition to these problems, per cent as the ideal reflectance. (b) Ideal sample sandwiched between the
the rise in R towards low frequency contains some contribution diamonds. (c) More realistic lumpy surface with cracks, irregular thickness,
from back reflections from the small grains because the samples and sections adhering to both diamonds.
are partially transparent in this frequency region. Therefore at low
frequency R should be flat-lying, as observed for MgO (Fig. 2) and
Quantitative comparison of calculated and measured absorption
Mg0.44 Fe0.56 O (Fig. 10a), to represent best the lattice dynamics of
spectra is made through equations (1) and (10). These equations
pure FeO.
presumes that the light reflects upon entering and exiting the sam-
To constrain n and k of FeO, a classical dispersion analysis
ple from a medium (Fig. 1). The situation in the DAC is slightly
was performed. The parameters obtained from the Kramers–Kronig
different (Fig. 11). The empty cell, serving as the reference, has two
analysis (Henning & Mutschke 1997) lead to a poor match to the
air–diamond reflections, whereas the loaded cell could have two
reflectivity profile (Fig. 10a), and thus these values are inaccurate.
sample–diamond reflections. However, the thin film is not bonded
The reflectivity above or below the peak cannot be used to determine
to the diamond, and an air gap could exist between at least one of the
the fitting parameters. As a first approximation, we assume that the
diamonds and the film. The diamond faces are very smooth, whereas
shape of the strong main IR band represents the intrinsic behaviour
the films contain grain boundaries. One approximation would be
of FeO (it is absorbing enough that back reflections are unlikely
that the air–diamond reflections are eliminated when the raw data
over the peak). The maximum reflectivity of 70 per cent is less than
are ratioed to the reference spectrum, leaving air–sample reflections
those of MgO with R = 92 per cent (Fig. 2a), Fe0.56 Mg0.44 O with
that need to be accounted for. Another approximation would be
R = 76 per cent (Fig. 10a), CoO with R = 80 per cent, and NiO with
that the air–diamond and sample–diamond reflections are roughly
R = 90 per cent (Gielisse et al. 1965). The trends seem consistent,
equal, and thus this effect need not be necessarily be accounted
suggesting that the FeO values near the peak are reasonably correct.
for. However, because the reflectivity for MgO is high and variable
We altered the oscillator strength until a reasonable match across the
(∼100 per cent between 400 and 600 cm−1 , ∼ 80 per cent between
main band was obtained (Fig. 10a). The final TO and LO parameters
600 and 740 cm−1 , and drops rapidly to either side) which contrasts
(Table 5) do not differ much from the Kramers–Kronig analysis of
strongly with the constant reflectivity of air–diamond as 17 per cent,
Henning & Mutschke (1997), but the dielectric and optical functions
reflections at the sample surface could affect the absorption experi-
are affected to varying degrees (Figs 10b and c). The calculated R
ment. We therefore calculated spectra allowing for extra reflections,
at ν = 0 is 34 per cent, which is compatible with reflectivity data
both sample–diamond and sample–air, for comparison both with
on Mg0.44 Fe0.56 O (Fig. 10a, details to be published elsewhere). The
the calculated absorptivity and with the thin-film measurements.
fit is also compatible with the LST relationship (Table 4), as is built
The results show that the peak position of the film is very close to
into the damped harmonic oscillator model. The classical disper-
that of the intrinsic absorption spectra (Fig. 12a) but is altered if re-
sion fit should therefore reasonably represent FeO, but may not be
flections are accounted for (Fig. 12b). Reflections both broaden and
exact.
shift the peak, and in particular increase the effective peak height.
Calculated values for the thinner films are most affected by the addi-
6.3 Comparison of measured and calculated tional reflections (Fig. 13). Such behaviour is not observed: instead,
absorption spectra intensities tend to scale upon thinning (Figs 7 and 8), suggesting
that these are essentially absorption spectra.
Qualitatively, the peak positions from thin films of the diatomic com-
The lack of a match suggests that equation (10) does not ade-
pounds are in generally good agreement with available reflectivity
quately describe the situation. It appears that the intrinsic spectrum
and Raman data (Tables 3 and 4). The small difference between
(Figs 12 and 13) dominates, but a mismatch exists. The imperfect
the TO frequencies and the absorption positions (Fig. 9, Tables 3
correspondence (Fig. 12) is attributed to the departure from ideal
and 4) suggests that this method documents the intrinsic behaviour
behaviour in the cell.
(i.e. material properties) of these solids. The absorption peak lies
closer to the TO position of ZnO with E⊥c, as is expected because (i) The coverage of the diamond tip is not perfect, and the slight
of the larger strength of the bands in this polarization because it amount of light leakage will ‘round’ the top of the peak and decrease
is doubly degenerate, whereas bands in E||c are singly degenerate. the overall absorbance (as shown in Fig. 6). The intensity of the
Note that the positions for FeO show the greatest disparity. This shoulders and weaker peaks will be exaggerated compared with the
could be associated with the large amount of scattering or errors in main peak (Hofmeister et al. 2000), and the profile will be flattened
the reflectivity from polycrystalline measurements. and have lower absorbance than an ideal film (Fig. 6).


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 31
λ, µm λ, µm
50 25 20 12.5 100 50 25 20 16.7 12.5

MgO a 0.4 1 µm a
1.6 0.5
calculated absorbance calc. FeO
SCM absrorbance

SCM Absorbance
0.3
1.2
t = 1 µm

0.2
0.8
LO
0.6
µm
LO 0.2 meas.

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

0.1
0.4 thin film
0.5 µm 0.1

0.1 µm
0
0 0.7

b b
2.5 t = 0.5 µm
0.6 0.5 µm
calculated for one
SCM absrorbance

sample-diamond
2 0.5 reflection
calculated: diamond 0.2

SCM Absrobance
air to sample to sample
to air to diamond 0.4 0.1
1.5 0.01 µm

0.3

1
thin film
0.2

0.5 0.1
meas.
no refl.
0
0 100 200 300 400 500 600 700 800
200 300 400 500 600 700 800
-1 Wavenumbers, cm-1
Wavenumber, cm
Figure 13. Comparison of measured absorbance of FeO (solid lines) with
Figure 12. Comparison of measured absorbance of MgO (grey line) with
calculations based on classical dispersion analyses (broken lines) for various
calculations based on Kramers–Kronig analyses (thin lines). Vertical lines
sample thicknesses. Dot-dashed = 1 µm; long dashes = 0.5 µm; short
show peak positions in the measured absorption spectra. (a) Comparison of
dashes = 0.2 µm; dots = 0.1 µm; broken = 0.01 µm. (a) Comparison
measurements with perfect absorption without surface reflections for various
of measurements with perfect absorption without surface reflections; (b)
thicknesses. Long dashes = 1 µm; dot–dashed = 0.6 µm; short dashes =
assuming that one diamond–sample reflection exists.
0.5 µm; dots = 0.2 µm; thin solid line = 0.1 µm. (b) Calculated spectra
for thickness of 0.5 µm, assuming that two additional surface reflections
exist. Short dashes = no reflections; dotted = two reflections arising from (v) Non-normal incidence exaggerates the LO mode (Berreman
the air–sample interfaces; long dashes = two reflections arising from the
1963).
diamond–sample interfaces.
Discerning the microscopic behaviour is difficult because some
of these effects compete: for example, light leakage lowers the ab-
(ii) The incident angle is high inside the diamond, owing both to sorbance from intrinsic values whereas a reflection raises it, and
the use of converging light and to the fact that diamond has a high excitation of the LO mode raises a for the high-frequency side of
index of refraction. Whilst n = 1 for air or vacuum, the dielectric the peak. However, the imperfect correspondence of Fig. 12 is most
constant  0 is 5.7 and n = 2.4099 at 15 237 cm−1 for diamond likely connected with light leakage (compare Figs 6b and 12b) and
(Fontanella et al. 1977), implying that n is close to 2.387 across also with the surface roughness of the sample (which reduces the
the far-IR. Effectively, the diamonds act like lenses, making the reflectance) compared with the smoothness of the diamond and the
relevance of Fig. 1 questionable. The sample, being thin, is also focusing effect. Fig. 11(c) probably best describes the sample, which
analogous to a coating, which is used in the optical industry to touches both diamonds, with some air gaps and some light leakage.
prevent surface reflections. Thus, equation (10) provides excessive reflection. A smaller amount
(iii) The film is not perfectly smooth, so the loss of light may than the ideal, roughly 10 per cent, and a mixture of sample–air
involve scattering rather than the specular reflection effect assumed and sample–diamond reflections would provide a better match. The
in deriving Fig. 12. spectrum is also affected by the light leakage depicted in Fig. 7. The
(iv) The film does not have a uniform thickness, which enhances excess intensity on the high-frequency side is ascribed to the non-
the shoulder. normal incidence. MgO, with its broad peak and strong LO mode,


C 2003 RAS, MNRAS 345, 16–38
32 A. M. Hofmeister, E. Keppel and A. K. Speck
is a worst test case (Fig. 12) with its extreme changes in R across the λ, µm
peak (Fig. 2) and with the reflectivity being high near the LO mode, 50 33.3 25 20 16.7
0.7
1
but the absorbance being moderate, rather than strong. It is also the
hardest of the materials studied, which makes it difficult to obtain FeS
an even film without cracks. The comparison for FeO (Fig. 13) with 0.8

Reflectivity
lower R and less contrast (Fig. 10), and a little softer, is better, but R 0.6

SCM absorbance
the reflectivity data are uncertain. Much better agreement is seen for
CaO which has a narrower peak, and is soft: see below. Excellent 0.6
agreement was seen for garnets (Hofmeister 1995). These complex a
silicates have 13 bands with widths about 1/10th those of the ox- 0.5

ides, and thus the reflectivity is only high where the sample absorbs 0.4
strongly.
The thicknesses created by compression are tenths of a µm to 1 R
vis
0.4
µm, as ascertained by transmission electron microscopy measure- 0.2

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

ments of garnet samples (Hofmeister 1995). The films are too thin to calc. a
t = 0.5 µm
be removed without damage, and so thicknesses cannot be reliably
determined from comparison with a microscope reticule. From re- 0 0.3
200 300 400 500 600
lief observed visually in a binocular microscope, all films prepared -1
Wavenumber, cm
in this study have thicknesses below 1 µm. The lack of precise cor-
respondence of calculated and measured spectra does not allow film Figure 14. Comparison of measured absorbance of FeS (solid black lines)
thickness to be ascertained exactly from comparison. However, the with that calculated for a thickness of 0.5 µm using Kramers–Kronig anal-
above discussion suggests that the rough comparisons in Figs 12 ysis (heavy black line) of reflectivity data (grey line, right y-axis, Henning
and 13 are valid. The thicknesses of the MgO films are estimated to & Mutschke 1997). The thin films are less than 1 µm thick, from relief. The
be 0.5 µm, consistent with visual observations, and a small amount positions of the absorption features agree well, but the calculations overes-
timate the strengths by about four-fold.
of reflection.
Similar relationships exist between the absorption data on FeO
and the classical dispersion calculation (Fig. 13). Specifically, the Our absorption data imply that problems exist in the previous
profiles from a calculation that assumes no exit reflection best match reflectivity data at low frequency as well. Comparing FeO and FeS
the data, and film thicknesses of about 0.15 and 0.4 µm are suggested absorption spectra in the far-IR (Figs 7 and 8) suggests that the re-
for the two films. However, the peak maximum of the film occurs flectivity of these samples should behave the same at low frequency.
at a higher frequency than the calculation, which did not occur for Free-carrier absorption is unlikely to be the cause of the rise in R
MgO. Perhaps the difference in peak positions represents a larger towards low frequency.
amount of reflection for FeO. For FeO, a strong peak is seen above Although the maximum value of R obtained for the main IR peak
the LO position. Assuming that the peak at 564 cm−1 is the LO of FeO appears to be correct (cf. Figs 10 and 13), the same behaviour
mode and that the TO mode occurs at 15 cm−1 below the thin-film is not necessarily true for FeS. For FeO, R ranges from about 5 to
peak gives a reasonable value for  0 from the LST relation (Table 4). 70 per cent, whereas R of FeS ranges from about 40 to 60 per cent.
These values may be just as valid as the classical dispersion analysis, The smaller range in R is consistent with the peaks being weak,
given the obvious problems with the reflectivity data obtained from and further suggests that the reflectivity across the entire spectrum
polycrystalline material. Providing exact values for the peak param- could be artificially enhanced by back reflections. Such problems
eters and for the optical functions of FeO will require single-crystal were seen by Long et al. (1993), who compared optical functions
reflectivity measurements. obtained from reflectivity of single crystals and pressed powders of
calcium carbonate and calcium sulphate. This inference is confirmed
by comparing the data sets (Fig. 14). Peaks in R occurring at 230, 255
and 295 cm−1 for FeS at room temperature (Henning & Mutschke
6.4 Assessment of FeS reflectivity data
1997) correspond well with the peaks from the thin-film measure-
The thin-film measurements show that the peaks of FeS are weak. ments (Table 3). The shoulder in R at 200 cm−1 is not resolved in the
It is difficult to resolve the features in the thin samples, whereas in- thin-film measurements at room temperature. This feature is close
creasing the thickness mainly increases the baseline (owing to high to the level of noise in the reflectivity data, and may not be real. This
overall reflectivity) without enhancing the peaks (Fig. 8). Our data correspondence of reflectivity and absorbance suggests that the re-
are consistent with the essentially featureless spectrum obtained flectivity spectrum of FeS (Fig. 14) is a mixture of absorption and
from a dispersion (Nuth et al. 1985), because in a dispersion, light reflection, as occurred for the weak overtone-combination bands at
leakage can overwhelm such small absorption features. Reflectiv- high frequency of MgO (Fig. 4). Fig. 1 shows the process schemat-
ity measurements in the visible provide R = 40 per cent (Peckett ically. This superposition enhances the features in the reflectivity,
1992), which is much lower than R = 53 per cent at 1000 cm−1 giving overly large intensities in the absorption spectrum calculated
determined by Henning & Mutschke (1997). This discrepancy at from Kramers–Kronig analysis. Fig. 14 shows the absorbance cal-
high ν is clearly due to back reflections, which exist in the trans- culated for t = 0.5 µm, compared with the measurements on two
parent spectral regions of our reflectivity measurements of thick different films with thicknesses of 1–2 µm. Even if the films are as
(1 mm) MgO crystals. Henning & Mutschke (1997) did not report thin as 0.2 µm, the previous calculations still provide excessively
the grain sizes of their FeS sample, but such a synthesis by quench- large absorbance. The positions are in good agreement for the three
ing would provide grain sizes closer to µm than to mm. It is clear that well-resolved bands below 300 cm−1 (>33 µm). In contrast, the
the high-frequency region of FeS has enhanced reflectivity because broad peak extending from 370 to almost 600 cm−1 obtained from
of secondary reflections (Fig. 1). Kramers–Kronig (Henning & Mutschke 1997) is not present in any


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 33
of the six absorbance measurements on two different samples of peaks reported for FeS and pyrrhotite, although it could create a
troilite. In fact, a depression is seen over this frequency range in weak peak. Moreover, absorption strengths obtained from the re-
some of our spectra (Fig. 8c). This band is not evident in the raw flectance data were shown above to be too large, and thus additional
reflectivity data, and we conclude that this feature between 17 and problems exist in the FeS and FeO absorption spectra reported by
30 µm is a spectral artefact, probably induced by back reflections Keller et al. (2002). To probe the source of the discrepancies, we
and the use of constant R as an extrapolation of the data to high collected mid-IR data from the above minerals as well as from pyrite
frequency. (FeS2 ).
The n- and k-values extracted by Kramers–Kronig analysis by The pyrrhotite spectra (Fig. 15), like that of troilite, show no ev-
Henning & Mutschke (1997) clearly err. This inference is corrob- idence of an extremely broad band from about 400 to 600 cm−1 .
orated by the large values of about 6 so obtained for n, whereas Pyrite has a fairly narrow bands at 600 and 1644 cm−1 and over-
equation (1) sets an upper limit of n = 4.4 in the visible (for k = 0) lapping peaks at 1000 and 1200 cm−1 . The pyrite structure contains
from independent observations of R. Lower values of about 2.5 for S2 dimers which should be active in the mid-IR, and probably pro-
n are seen for the compounds FeO, ZnS and CdS, but these com- duce the higher frequency modes. Pyrite peak widths are similar
pounds are more transparent. If the lower n holds also for FeS, then to those of the iron oxides. Our thin-film data on sulphides do not

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

k is substantial in the visible, not near 0 as derived by Henning & provide evidence of extraordinarily broad bands for these phases in
Mutschke (1997). the mid-IR (Fig. 15) or in the far-IR (Fig. 8c).
Given the uncertainty in the thickness of the thin film of FeS, and Our wüstite (FeO) spectrum has a shoulder at 560 cm−1 on the
the likelihood of back reflections pervasively altering the reflectance main band at 363 cm−1 (Figs 7c and 13), whereas the FeO spectrum
data, we cannot extract the optical functions using either Kramers– of Keller et al. (2002) consists of a single peak at ∼550 cm−1 on
Kronig or classical dispersion analyses. Reflectivity data on larger a flat lying baseline. Our spectra were collected from a thickness
crystals are needed. of about 0.2 µm. Keller et al. (2002) do not provide thickness or
absorbance values, but state that the sulphides have t < 0.2 µm. The
spectra of FeO should thus be equivalent, and we conclude that their
6.5 Comparison with previous absorption spectra of iron main, far-IR peak of FeO was removed during baseline subtraction
sulphides and oxides by Keller et al. (2002).
Keller et al. (2002) provide mid-IR absorption spectra of the di- For magnetite, four IR bands are observed in the far-IR (Fig. 7c),
atomics examined here, pyrrhotite (Fe1−x S) and magnetite (Fe3 O4 ). as expected from symmetry analysis (Chopelas & Hofmeister 1991;
Their data are at odds with the thin-film far-IR spectra, and give Hofmeister & Mao 2001). The pattern of two strong and two weak
overly large absorption strengths compared with those extracted bands is typical for the IR spectra of minerals with the spinel struc-
from reflectance data (Henning & Mutschke 1997). This difference ture. The highest frequency band occurs at 570 cm−1 , and is similar
in absorption strengths can be attributed to Keller et al. (2002) not in position and width to the LO mode of FeO (Fig. 7c). (Probably
applying the correction of Shankland et al. (1979). This correction the similar peak positions in FeO and Fe3 O4 reflect similar atomic
(equation 10) is particularly important for FeS, as R is large. How- motions and bonding.) This band and the other moderately strong
ever, omitting the correction is unlikely to create the pronounced band in magnetite at 353 cm−1 are seen in the spectra of Keller et al.
(2002), although the 353 cm−1 band is lower in intensity. It is likely
that their spectral profile of the 353 cm−1 band was altered because
λ, µm it occurs at the limit of detection. The two weak bands at 421 and
20 12.5 10 8 6
1.8 479 cm−1 were not detected by Keller et al. (2002). This comparison
corroborates the assumption that problems exist below 500 cm−1 in
1.6 FeS the absorption spectra of Keller et al. (2002). Because differences
exist between all four minerals in the same frequency range, the
pyrrhotite
1.4 mismatch with our sulphide spectra is not due to slightly differ-
ent samples being examined (e.g. from differences in Fe3+ content,
SCM Asorbance

1.2 vacancies, disorder or polytypism), but is related to experimental

technique.
1 We ascribe the extraordinarily broad features reported in FeS
and pyrrhotite to spectral artefacts. Baseline subtraction is probably
0.8 the major cause. Another contributing factor may be use of carbon
films to support their samples. Carbon is both highly reflective and
0.6 absorbent, like FeS. The surface reflections (equations 1 and 10)
pyrite would thus depend on frequency and may contribute to spectral
0.4 artefacts. A third consideration is that KBr beamsplitters have low
throughput below 500 cm−1 , which can create problems near the end
0.2 of the spectral range. Fourthly, it is highly likely that the detector was
600 800 1000 1200 1400 1600 saturated and responding in a non-linear fashion to intensity input.
Wavenumbers, cm-1 Although a small aperture of 32 × 32 µm2 was used by Keller
et al. (2002), the synchrotron beam is intense. Mid-IR detectors are
Figure 15. Comparison of mid-IR data from thin films of various iron
sulphides. Dotted line = pyrite. Solid black = natural pyrrhotite. Grey = prone to saturation, which can occur even if only a small spot on
troilite. Pyrite peaks are indicated by ticks. All other peaks and sways appear the detector element is flooded with light. A low overall throughput
to be interference fringes. Noise is seen near the detector limit. The three does not guarantee that detector response is linear over the whole
spectra of pyrrhotite (and the two curves for troilite) represent successive frequency range. Saturation was avoided in the present study by
thinnings of the film. use of wire mesh screens to reduce throughput or by restricting


C 2003 RAS, MNRAS 345, 16–38
34 A. M. Hofmeister, E. Keppel and A. K. Speck
the aperture near the source. The phase correction spectra were λ, µm
examined in all cases to ensure that saturation had indeed been 100 50 25 20 12.5
8
avoided. Because the throughput of our relatively dim globar was
too much, even when a DAC was used which reduces throughput Calculated functions
7
by a factor of 1000, one might reasonably conclude that a bright CaO
synchrotron source requires special measures to avoid saturation. 6
Because no peak exists for iron sulphides in the vicinity of 25 µm,

Optical Functions
identification of this material in protoplanetary discs is predicated 5
on a match with the weak features of troilite near 250 cm−1 (Figs 8c n
and 14). Of the objects examined by Keller et al. (2002), only M2-43 4
has features in this region, but these are at or barely above the noise
level. The spectral match is unconvincing. 3 nvis
MgS
6.6 Estimated optical constants for CaO, CaS and MgS 2 CaS

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

In principle, absorption data from a known thickness, corrected for
1 CaO
reflections (equation 10), can be fitted using a classical dispersion k
analysis (equations 4 and 5). The other input is the index of refraction
0
in the visible (n 2vis =  ∞ ). An additional constraint could be provided 100 200 300 400 500 600 700 800 900
by values of  0 , but such data seem unavailable for MgS or CaS. Wavenumbers, cm
-1

Data for  0 of CaO are available (Table 4), along with reflectivity
Figure 17. Calculated optical functions for CaS (heavy black lines), MgS
measurements (Jacobson & Nixon 1968; Galtier, Montaner & Vidal
(grey lines) and CaO (thin black lines). Solid lines = real part of the complex
1972). Because we found these sources after classical dispersion
index of refraction. Broken lines = imaginary part. Measurements of n in
analysis had been carried out, and because the oscillator strengths the visible are shown by arrows.
were not reported, we fit the thin-film data and compare the resulting
parameters with the previous reports.
For CaO, the film thickness was ascertained by comparison with
inferred position for the LO component. Relevant parameters (given
thick-film data (Fig. 7b). For CaS and MgS, thicknesses are esti-
in Table 2) must be considered as estimates for MgS and CaS. How-
mated as about 1 and 0.75 µm, respectively, by comparison with
ever, the results for CaO should be fairly accurate because thickness
results for the oxides (Figs 7, 12 and 13). These are rough numbers.
could be constrained through comparison with spectra obtained from
The fit is made by scaling the spectrum to correspond to t = 1 µm
a t = 6 µm film. The parameters are not affected by changing the
and to provide the true absorbance. This allows direct comparison
thickness by 20 per cent, which is roughly the uncertainty in t. The
with the calculated absorption coefficient.
resulting frequencies for CaO (Table 2) are within a few cm−1 of
Good fits were obtained for all substances (Fig. 16). The results
those obtained from reflectivity of single crystals (Table 4). The
are consistent, in that fitting the peak resulted in structure near the
uncertainty given in Table 4 is derived from the range of reported
frequencies (Jacobson & Nixon 1968; Galtier et al. 1972): our re-
λ, µm sults for the LO mode are slightly high, and as a result  0 is high,
100 50 25 20 cf. Tables 2–4. If we had altered the damped harmonic oscillator
parameters to fit the low-frequency dielectric measurements of Ja-
2.5 cobson & Nixon (1968), better agreement would have been obtained
Absorbance

CaS for the LO mode. As is, our results are within 1.2 per cent.
Absorption Coefficient, 1/µm

232,22,9.5 Results for MgS and CaS are less accurate, as thickness is uncer-
2 CaO 2
tain and because the spectrum consists of two overlapping bands.
298,32,9 The spectrum for MgS appears to have a baseline problem at the
1.5 high frequency, probably due to high reflectance. Even with these
MgS problems, for frequencies above about 1200 cm−1 , the results for n
240,31,11 and k (Fig. 17) are independent of the peak parameters, but instead
1 1 are determined by n vis . The results suggest that if the thin-film thick-
ness for MgS could be determined from interference fringes or if
the reflectance at high frequency could be minimized by collecting
0.5 data from a thicker film, then fitting the absorption spectra would
set a limit on n in the visible, and better constrain k.
0 0 The peaks of CaO, CaS and MgS are narrower than those of MgO
100 200 300 400 500 600 and FeO, resulting in a better match of ideal absorption coefficients
-1
Wavenumbers, cm with the measurements. The better match is largely due to the reflec-
tivity and absorptivity being high over the same frequency interval.
Figure 16. Classical dispersion analysis of thin-film absorption data. The
raw data were scaled to represent a 1 µm thick film, and converted to true
An additional factor is that the CaO and the sulphides are soft, which
absorbance units (solid lines). The fits to the data are shown as broken lines. is conducive to preparing thin, even films. The existence of multiple,
Heavy black lines = CaS; grey = MgS; thin black = CaO. We fitted the overlapping peaks contributes some uncertainty to the fits, but is of
low-frequency slope, and assumed that some rounding at the peak occurred, less importance. We conclude that materials such as silicates with
and that the spectra near the LO mode are overly high because of partial relatively narrow peaks, and which are frequently softer than MgO,
reflectance (for CaO) or were disturbed by interference fringes. would provide absorption spectra closer to intrinsic values.


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 35

6.7 Effect of temperature λ, µm

100 50 25 20 12.5 10
MgO reflectivity has been obtained from 8 to 1950 K (Jasperse et al. 1
2 5
1964) and analysed using the classical dispersion model (Kachare MgO
et al. 1972). However, the data at room temperature vary slightly
1 µm
from the present results (Fig. 2), which is probably due to alignment 0.8
or resolution problems with the dispersive technology used at that
time. The main IR band was assumed to consist of one strong peak t = 100 µm
0.5
and one shoulder, whereas our analysis indicates that four poorly re- 0.6 t = 10 µm

Emissivity
50
solved peaks are present (Tables 1 and 2). Neither were the previous
data adequately fitted by two oscillators. An additional problem is
that high resolution is needed to measure the steep profiles found at 0.4 0.2
cryogenic temperatures (Bowey et al. 2001). Undersampling rounds
the peaks. Therefore the temperature response of MgO deduced by

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

Kachare et al. (1972) has some degree of associated uncertainty. The
0.2
measurements show that, for both peaks, the frequencies decrease 10
as temperature increases (dν/dT = −0.0186 cm−1 K−1 ), whereas 0.1 µm
the damping coefficients and oscillator strengths increase. All pa-
0
rameters depend linearly on T. The product of  and f causes the 200 400 600 800 1000 1200
peaks in k (and also the absorption peaks) to become less intense as Wavenumbers, cm
-1

temperature increases. This deduction is consistent with the trend

in reflectivity wherein the main peak becomes shorter and broader Figure 18. Emissivity of MgO at room temperature for various grain sizes.
as T increases. Such behaviour is expected. As T increases, more Black lines = calculations based on classical dispersion analysis (light solid
line = thickness of 0.1 µm; light dots = t of 0.2 µm; light dashes = 0.5
states are populated and the number of phonon–phonon scattering
µm; heavy solid = 1 µm; heavy dots = 2 µm; heavy dashes = 5 µm; long
events increases, leading to larger widths. The number of Mg–O
dashes = 10 µm). Grey lines = calculations from measured absorbance of
dipoles producing the main IR band should be unaffected as it is a 100-µm crystal (dashes = t of 100 µm; dots = 50 µm). For grain sizes
the ground state, and thus the area under the peak should remain the above about 20 µm, the overtone bands contribute to the emissivity. For
same. The broader peaks must therefore be shorter. If the population grain sizes of about 2 to 10 µm, the emissivity is roughly unity over the peak
of the main band does increase, as is possible at very low tempera- and minimal elsewhere. For grain sizes 1 µm or below, the emission peak
tures, then the band area and height could increase as temperature strongly resembles the absorption profiles.
increases.
The reflectivity data of FeO and FeS with temperature (Henning & 6.8 Calculation of emissivity and emissions
Mutschke 1997) may be problematic as these involve measurements
of polycrystals. Specifically, the raw data indicate that the maximum Equation (7) provides emissivity. From Fig. 2, ideal absorptions for
R of FeO decreases as T increases, consistent with the behaviour of MgO obtained from Kramers–Kronig and classical dispersion anal-
MgO, but the FeO peaks become narrower, and the TO position is yses are nearly identical. Because back reflections may be present
constant (in both raw R data and derived functions). Frequency be- near 1000 cm−1 , the synthetic spectra are used to compute ξ . The
ing independent of T is inconsistent with finite thermal expansivity, emissivity mimics absorption spectra at low grain sizes (Fig. 18).
particularly in that the IR band of diatomic solids involves stretch- For larger grains, emissivity is pinned at unity for the absorbing
ing and compression of the cation–anion bond. Moreover, wüstite region, and extends into the overtone combination region for the
and MgO have nearly identical values of the thermal expansivity largest grains (submillimetre sizes). Higher temperatures would
(e.g. Fei 1995) and thus their frequencies should shift at about the cause additional broadening, whereas cryogenic features would be
same rate. Errors in the absolute value of R due to secondary reflec- narrower than those presented here. The peak shifts expected are
tions may cause these inconsistencies. Undersampling due to low relatively small.
resolution is a contributing factor, and this effect probably explains Emission spectra are a convolution of the emissivity and Planck’s
the small degree of change in R for the pyroxene sample and the blackbody curve. If the grains are only MgO, then the emission spec-
glasses examined in the same study (Henning & Mutschke 1997). tra for temperatures above 200 K are quite similar to the emissivity
It is particularly revealing that the main band decreases in both in- with the intensity being highly temperature-dependent (Fig. 19). Be-
tensity and breadth as T increases. Although the intensity decrease low 200 K, the higher frequency side of the emissions is truncated by
is compatible with the behaviour of MgO (Jasperse et al. 1964), the decrease in blackbody radiation: only the TO mode is manifest.
the width decrease is not. Moreover, simultaneous decreases in in- The peaks shown in Fig. 19 are broader than expected at cryogenic
tensity and width require that the vibrational energy states become temperatures, because room-temperature spectra were used in the
decreasingly populated at T increases, which is incompatible with calculation. However, a distribution of temperatures among the dust
statistical thermodynamics. We did not try to use classical disper- grains would also cause peak broadening, and thus the results of
sion to obtain optical functions from the reflectivity measurements Figs 18 and 19 should be relevant.
at low temperature because data on n in the visible at cryogenic Grain sizes larger than 1 µm would provide broader emission
temperatures may be needed to constrain the results. Single-crystal spectra at 298 K and above, whereas below 200 K the blackbody
reflectivity data on FeO and FeS at temperature, and probably mea- function would dominate as shown in Fig. 19. Smaller grain sizes
surements of n(T) or of R(T ) in the visible, are needed for accu- would provide emission spectra that more closely resemble absorp-
rate determination of the optical functions at temperature of these tion spectra from samples of the same grain size. Another compli-
materials. cating factor is the possibility of temperature gradients within the


C 2003 RAS, MNRAS 345, 16–38
36 A. M. Hofmeister, E. Keppel and A. K. Speck

λ, µm from flat, parallel interfaces (from equations 1 and 10) do not fit
100 50 25 20 12.5 10 the data and thus the amount of reflection is lower, consistent with
8
1.2 10 the non-normal incidence and some surface roughness of the films.
MgO However, this effect is most pronounced in the transparent regions
and near the LO mode, where the sample reflects, but is weakly
1 10
8
t = 1 µm absorbant. The mismatch of ideal and measured absorption wors-
Emissions, erg/sec/cm3

298 K ens as the LO–TO splitting increases, as the frequency interval of

7
8 10 high reflectivity with low absorbance increases concomitantly. As
the LO–TO splitting depends on the band strength, the Si–O stretch-
6 10
7 ing band at 10 µm and the bending mode near 20–25 µm could be
affected by these problems, but the low-frequency, weak modes of
silicates should not.
7
4 10 Spectra collected from dispersions have similar problems to the
200 K
thin-film data. This method is useful as the conditions in space in-

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

2 10
7
50 K
volve distributions of particles. However, laboratory spectra and
100 K
(x 10000) (x 10)
astronomical data differ in several important ways. Foremost, the
processing differs. In laboratory experiments light is forced through
0
200 400 600 800 1000 some sort of aperture, and the raw data are ratioed to a baseline
-1
Wavenumbers, cm involving this aperture, whereas astronomical data are processed by
subtraction of a baseline. Thus the artefacts induced by light leak-
Figure 19. Emission spectra of MgO. Dots = temperatures of 298 K; dashes
= 200 K; light solid line = 100 K, but intensity is scaled by ×10; Heavy age shown in Fig. 6 (which is equivalent to mismatching the aper-
solid line = 50 K, scaled by ×104 . The second peak at 50 K is due to the tures, or ratioing to an inappropriate baseline) will not as strongly
finite reflectivity at low frequency, coupled with the steep climb for Planck’s affect observational data, but will affect laboratory measurements
blackbody function. The calculation presumes that the emissions are solely from dispersions more than measurements from thin films. For well-
due to MgO grains. dispersed dust in space with various grain sizes, the largest, opaque
particles will simply reduce the overall throughput in the observa-
sample. For very large samples, the colder outside layers absorb the tional data, and the signal received is either that which misses the
essentially blackbody emissions from the interior, producing spec- grains or light that is partially absorbed by the smaller grains. The
tra in emission that look like absorption [cf. emission spectra of features will essentially scale, although rounding of the strongest
Christensen et al. (2000) to those of Bates (1978)]. The sizes used peaks will be seen because each dust particle provides a range of
by Christensen et al. (2000) are large, 0.75 mm, and their temper- thickness. Because large grains ( µm) in dispersions can induce
ature gradients are small. Although these conditions may not be artefacts, comparisons with astronomical data should be made us-
encountered in space, it is possible that the effect occurs in smaller ing dispersions with fine grain sizes. The lack of parallel surfaces
grains at the colder temperatures in space, especially if one side of in space also means that secondary reflections are weaker than that
dust grains is heated and large thermal gradients exist. predicted by the equations here. Therefore the thin-film data with
If the grains are a mixture of phases, the emission spectra could be minimal reflections are appropriate. This can be met by merging
altered significantly from the predicted curves of Fig. 19. Probably spectra from films of varying thicknesses (i.e. obtaining data in the
the greatest effect would be seen for mixtures with graphite or FeS, transparent regions from thicker films). Dust peaks in observational
which are good greybodies, because these have roughly constant data will occur at the same positions as the ideal absorption spectra,
reflectance over a wide frequency range. For this case, emission and the higher frequency side of the peaks will be stronger than the
peaks of MgO would be seen superimposed on a blackbody curve ideal. The angle of incidence is essentially normal, so the LO modes
from the opaques. Mixtures of oxides and silicates would provide should not be as prominent as in the thin film. The peak absorbance
convoluted absorption spectra. The signature of the small grains should roughly indicate the optical depth of the dust.
could be identified from the TO positions, whereas for large grains Dense dust clouds with large particles can have the same spectral
the spectral profiles would be difficult to distinguish. artefacts as seen for the thicker films, and as schematically indicated
in Fig. 6. Interpretation of the features based on comparison is less
robust in the optically thick regime. For the oxides, with one strong
6.9 Comparing laboratory spectra with astronomical spectra
peak, special conditions of fairly uniform and fine grain sizes are
Absorption coefficients calculated from reflectivity data represent needed for a definitive identification. The spectral signatures of MgS
the ideal case where the sample is uniformly thick and covers the and CaS, with their relatively narrow peaks, should be easier to
field of view, the direction of propagation is perpendicular to the recognize in astronomical data.
surface, and no surface reflections exist. In the laboratory measure- Emission peaks should not be as prey to artefacts, and the cal-
ments of absorption spectra from thin films, these conditions are culations based on ideal absorption coefficients should provide the
approximately met, and the spectra are affected to varying degrees. appropriate comparison. As discussed above, the temperature of the
Comparing thin-film measurements with ideal absorbance shows object determines which vibrations are excited, and affects the width
that peak positions are reproduced, but that the profiles are flatter and intensity of the peaks.
than those calculated, and that LO modes are present. Flatter peak
tops are connected with light leakage from cracks in the film, incom-
7 S U M M A RY A N D C O N C L U S I O N S
plete coverage, and/or variable film thickness. Given the consistent
results upon thinning, the flattening seen in the present study is only Interpretation of observed astronomical spectra is made through
partially caused by such spectral artefacts, and surface reflections comparison with laboratory data. The correctness of an interpre-
are present to some degree. Relatively strong reflections expected tation is predicated on the observational and laboratory data being


C 2003 RAS, MNRAS 345, 16–38
 IR spectra of MgO compounds 37
collected under similar conditions. We cannot control the conditions ing and providing the Mg18 O. Comments by anonymous review
in space (temperature, dust density, grain size, phase and chemistry), substantially improved the manuscript.
and thus it is imperative that we understand the effect such variables
have on laboratory measurements in order to gauge the relevance of
the laboratory spectra to the astronomical environment. This paper REFERENCES
not only presents new optical data on oxides and sulphides, but also
Andermann G., Duesler E., 1970, J. Opt. Soc. Am., 60, 53
provides methods to ascertain the accuracy of such measurements.
Andermann G., Caron A., Dows D. A., 1965, J. Opt. Soc. Am., 55, 1210
Analysing IR spectra in detail reveals the limitations inherent in Arguello C. A., Rousseau D. L., Porto S. P. S., 1969, Phys. Rev., 181,
laboratory data. In particular, the absolute values of n and k and the 1351
positions of absorption features can be compromised by sampling Arnold R. G., 1967, Canad. Mineral., 9, 31
techniques. Understanding these issues allows astronomers to be Bass J. D., 1995, in Ahrens T. J., ed., Mineral Physics and Crystallogra-
able to judge whether imperfectly fitted observed spectra are due phy: A Handbook of Physical Constants. American Geophysical Union,
to variance of the conditions in the laboratory from those remotely Washington, DC, p. 45
probed, or to the presence of different phases. Bates J. B., 1978, Fourier Transform IR Spectrosc., 1, 99

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

Specifically, new reflectivity data on MgO at room temperature Batsonov S. S., Derbeneva S. S., 1969, J. Struct. Chem. (USSR), 10, 510
are cross-checked against existing independent dielectric measure- Begemann B., Henning T., Mutschke H., Dorschner J., 1995, Planet. Space
Sci., 43, 1257
ments at low frequency and measurements of the index of refraction
Berreman D. W., 1963, Phys. Rev., 130, 2193
in the visible. Secondary reflections were found to exist even for Blazey K. W., 1977, J. Phys. Chem. Solids, 38, 671
very large thickness (∼1 mm) over frequency regions where the Boswarva L. M., 1970, Phys. Rev. B, 1, 1698
sample essentially transmits, and for the ideal conditions of a nearly Bowen N. L., Schairer J. F., 1935, Am. J. Sci., 29, 151
perfect polish on only one side. The Kramers–Kronig analysis of the Bowen H. K., Adler D., Auker B. H., 1975, J. Solid State Chem., 12, 355
raw data was therefore used as the starting point for fitting the spec- Bowey J. E., Lee C., Hofmeister A. M., Ade P. A. R., Barlow M. J., 2001,
trum to a series of damped harmonic oscillators. Such a classical MNRAS, 325, 886
dispersion analysis provides the optical functions, dielectric func- Brewster M. Q., 1992, Thermal Radiative Transfer and Properties. Wiley,
tions, absorption coefficients and emissivity from the far-IR to the New York
visible. Above about 25 000 cm−1 , the tail of strong, metal–oxygen Burns G., 1990, Solid State Physics. Academic Press, San Diego
Chopelas A., Hofmeister A. M., 1991, Phys. Chem. Minerals, 18, 279
charge transfer bands affects the values of k, and the classical dis-
Choyke W. J., Patrick L., 1962, Phys. Rev., 127, 1868
persion results are no longer valid. The same approach was applied Christensen P. R., Bandfield J. L., Hamilton V. E., Howard D. A., Lane M.
to previous reflectivity data on FeO (Henning & Mutschke 1997). D., Piatek J. L., Ruff S. W., Stefanov W. L., 2000, J. Geophys. Res., 105,
Because of the small grain size, spurious reflections affect the results 9735
at low as well as high frequencies. We presume that the region of Cook W., Jaffe H., 1979, in Hellwege K. H., ed., Landolt-Bornstein Group
strong absorbance is not affected, but single-crystal measurements III, Vol. 11. Springer Verlag, Berlin, p. 287
are needed to confirm this. For FeS, with its weaker peaks, R from Cotton F. A., 1971, Chemical Applications of Group Theory. Wiley-
polycrystals appears to be corrupted at all frequencies, so that the Interscience, New York
functions extracted from the available measurements have substan- Criss R. E., 1999, Principles of Stable Isotope Distribution. Oxford Univ.
tial errors, particularly in n and k at high frequency, and in overly Press, Oxford
Deer W. A., Howie R. A., Zussman J., 1992, Introduction to the Rock-
large absorption band strengths. Inference of FeS in protoplanetary
Forming Minerals. Wiley, New York
objects based on these R-values and problematic absorption data Evans H. T., Jr, 1970, Sci, 167, 621
(Keller et al. 2002) is not well founded. Fateley W. G., Dollish F. R., McDevitt N. T., Bentley F. F., 1972, Infrared
The diatomic solids studied here, with their one strong IR band, and Raman Selection Rules for Molecular and Lattice Vibrations: the
represent an extreme of behaviour because the high reflectivity near Correlation Method. Wiley-Interscience, New York
the absorption peak contrasts strongly with the low baseline reflec- Fei Y., 1995, in Ahrens T. J., ed., Mineral Physics and Crystallography: A
tivity at high and low frequency. Reasonable agreement was ob- Handbook of Physical Constants. American Geophysical Union, Wash-
tained between the thin-film measurements and calculated spectra ington, DC, p. 29
for MgO and FeO which have broad bands, and good fits could be ob- Fontanella J., Andeen C., Schuele D., 1974, J. Appl. Phys., 45, 2852
tained to the absorption data of the compounds with narrower peaks Fontanella J., Johnston R. L., Colwell J. H., Andeen C., 1977, Appl. Optics,
16, 2949
(CaO, CaS and MgS), suggesting that ideal spectra will represent
Gaines R. V., Skinner H. C. W., Foord E. E., Mason B., Rosenzwieg A.,
dust features of polyatomic compounds in astronomical environ- 1997, Dana’s New Mineralogy. John Wiley and Sons, New York
ments, e.g. complex oxides and silicates. Only the strongest peaks Galtier M., Montaner A., Vidal G., 1972, J. Phys. Chem. Solids, 33, 2295
(such as the Si–O stretching bands near 10 µm) may be altered from Geick R., 1964, Phys. Lett., 10, 51
the ideal by light leakage and surface reflections. None the less, un- Gielisse P. J., Plendi J. N., Mansur L. C., Marshall R., Mitra S. S., Mykola-
der certain conditions, even the broad bands of simple oxides and jewycz R., Smakula A., 1965, J. Appl. Phys., 36, 2446
sulphides can be distinguished. Fine grain sizes, well dispersed in Hazen R. M., Jeanloz R., 1984, Rev. Geophys. Space Phys., 22, 673
an optically thin cloud, should provide spectra closely resembling Henning T., Mutschke H., 1997, A&A, 327, 743
the intrinsic absorption or emissions of the phase. Henning T., Begemann B., Mutschke H., Dorschner J., 1995, A&AS, 112,
143
Hofmeister A. M., 1995, in Humecki H. J., ed., A practical guide to infrared
AC K N OW L E D G M E N T S microspectroscopy. Marcel Dekker Inc., New York, p. 377
Hofmeister A. M., 1997, Phys. Chem. Mineral., 24, 535
This work is supported by NSF-AST-9805924. We thank N. Johnson Hofmeister A. M., 2001, Am. Mineral., 86, 1188
(Washington University) for providing synthetic powders of MgO Hofmeister A. M., Mao H. K., 2001, Am. Mineral., 86, 622
and FeS, and for dehydrating the samples. We thank T. C. Hoering, Hofmeister A. M., Cynn H., Burnley P. C., Meade C., 1999, Am. Mineral.,
I. Kushiro and B. Mysen (Geophysical Laboratory) for synthesiz- 84, 454


C 2003 RAS, MNRAS 345, 16–38
38 A. M. Hofmeister, E. Keppel and A. K. Speck
Hofmeister A. M., Keppel E., Bowey J. E., Speck A. K., 2000, in Kessler Peckham G., Saunderson D. H., Sharpe R. I., 1967, Br. J. Appl. Phys., 18,
M. F., Salama A., eds, ESA SP-456, ISO Beyond the Peaks: The 2nd 473
ISO Workshop on Analytical Spectroscopy. ESA, Noordwijk, p. 343 Piriou B., 1964, Comptes Rendes Acad. Sci. Paris, 259, 1052
Hony S., Waters L. B. F. M., Tielens A. G. M., 2001, A&A, 378, L41 Refner J. A., Hornlein R. W., Hellgeth J. W., 1997, Micros. Microanalysis,
Horak M., Vitek A., 1978, Interpretation and Processing of Vibrational Spec- 3, 867
tra. John Wiley and Sons, New York Sangster M. J. L., Peckham G., Saunderson D. H., 1970, J. Phys. C, Solid
Jacobson J. L., Nixon E. R., 1968, J. Phys. Chem. Solids, 29, 967 State Phys., 3, 1026
Jasperse J. R., Karan A., Plendl J. N., Mitra S. S., 1966, Phys. Rev., 146, Shankland T. J., Nitsen U., Duba A. G., 1979, J. Geophys. Res., 84,
526 1603
Kachare A., Andermann G., Brantley L. R., 1972, J. Phys. Chem. Solids, Siegel R., Howell J. R., 1972, Thermal Radiation Heat Transfer. McGraw-
33, 467 Hill, New York
Keller L. P. et al., 2002, Nat, 417, 148 Speck A. K., Hofmeister A. M., Barlow M. J., 1999, ApJ, 513, L87
Kessler M. F. et al., 1996, A&A, 315, L27 Spitzer W. G., Miller R. C., Kleinman D. A., Howarth L. W., 1962, Phys.
Knittle E., 1995, in Ahrens T. J., ed., Mineral Physics and Crystallogra- Rev., 126, 1710
phy: A Handbook of Physical Constants. American Geophysical Union, Subramanian M. A., Shannon R. D., Chai B. H. T., Abraham M. M.,

Downloaded from https://academic.oup.com/mnras/article/345/1/16/984419 by guest on 05 October 2021

Washington, DC, p. 98 Wintersgill M. C., 1989, Phys. Chem. Minerals, 16, 741
Lauretta D. S., Kremser D. T., Fegley B., 1996, Icarus, 122, 288 Vaughan D. J., Craig J. R., 1978, Mineral Chemistry of Metal Sulphides.
Loh E., 1968, Phys. Rev., 166, 673 Cambridge Univ. Press, Cambridge
Long L. L., Querry M. R., Bell R. J., Alexander R. W., 1993, Infrared Phys., Waters L. B. F. M., 2000, in Salama A., Kessler M. F., Leech K., Schulz B.,
34, 191 eds, ESA SP-456, ISO Beyond the Peaks: The 2nd ISO Workshop on
McAloon B. P., Hofmeister A. M., 1993, Am. Mineral., 78, 957 Analytical Spectroscopy. ESA, Noordwijk, p. 39
Malfait K., Bogaert E., Waelkens C., 1998, A&A, 331, 211 Wooten F., 1972, Optical Properties of Solids. Academic Press Inc., San
Manson N. B., der Ohe W. V., Chodos S. L., 1971, Phys. Rev. B, 3, 1968 Diego
Mitra S. S., 1969, in Nudelman S., Mitra S. S., eds, Optical Properties of Wyckoff R. W. G., 1963, Crystal Structures, Vol. 1. John Wiley and Sons,
Solids. Plenum Press, New York, p. 333 New York
Nuth J. A., Mosley S. H., Silverberg R. F., Goebel J. H., Moore W. J., 1985,
ApJ, 290, L41
Peckett A., 1992, The Colours of Opaque Minerals. Van Nostrand Reinhold,
New York This paper has been typeset from a TEX/LATEX file prepared by the author.


C 2003 RAS, MNRAS 345, 16–38