JOEL M.G.

Measurement of the Hg spectrum using a

Michelson interferometer
Joel Martı́nez González

Abstract—Using a Michelson interferometer, all five persistent

emission wavelengths in the spectrum of neutral mercury (Hg
I) between 400 and 600 nm were measured as 404.69 ± 0.07,
435.89 ± 0.04, 546.09 ± 0.02, 576.96 ± 0.04 and 579.07 ± 0.03
nm. Multiple corrections were used, but the use of a calibration
wavelength was especially vital in the process.

I. I NTRODUCTION
S elements are heated or excited, electrons jump between
A energy states, emitting in the process photons with the
energy corresponding to the energy difference between these
states. Therefore, according to Ei = hfi , specific frequencies
are emitted for each element, depending on the energy levels.
The aim of this experiment was to determine the most
prevalent visible emitted wavelengths for neutral mercury
(Hg I) inside the range 400-600 nm.
Fig. 1: Diagram of a basic Michelson interferometer. The light
To achieve this, a so-called Michelson interferometer from the source is split into two perpendicular beams by the
was employed as an effective spectrometer. A Michelson beam splitter through a half-mirror. The two beams travel are
interferometer (Fig. 1) is an optical instrument that splits a reflected in the mirrors M1 and M2, traveling a distance of
beam of light and then recombines to form an interference 2d1 and 2d1 respectively before interfering as they recombine.
pattern, which can be studied as a function of the path Finally, the recombined beam is intercepted by the detector.
difference between the split beams. Studying the interference pattern as a function of the path
difference 2|d2 −d1 | allows for the recovery of the wavelength
If the source light consisted of a single wavelength λ0 , distribution of the source light.
and the intensity was split evenly between beams, solving
for the recombined beam’s intensity as a function of the path
difference ∆ = 2|d2 − d1 | gives an intensity at the detector of If the interferogram is symmetric (ϕ = 0), then (3) would be
I(∆) = I0 (1 + cos (2π∆σ0 + ϕ)) (1) the (shifted) cosine form of the Fourier transform. However,
for some general ϕ(σ), to relate the spectrum to the interfero-
, where σ0 is the wavenumber 1/λ0 (λ and σ provide essen- gram, the complex part of a Fourier transform must be taken
tially the same information, so they will generally be used in- into account to recover the complete spectral information[1]
terchangeably), and ϕ is a phase that allows the interferogram
Z ∞
to be asymmetric. This phase generally depends on σ and is
mainly due to irregularities in the beam splitter’s thickness[1]. S(σ) = N · abs( I(∆)exp(−2πiσ∆)d∆) (4)
−∞
The spectrum (i.e., the wavenumber distribution) of a light
wave with a single wavenumber σ0 , which is the case for (1), , where N is now some proportionality constant. This directly
will evidently be a delta function I0 δ(σ − σ0 ). However, the connects the experimental data I(∆) with the desired result,
emission spectrum corresponding to a wavelength σ0 experi- the spectrum S(σ), so that S(σ) ∝ abs(F {I(∆}).
mentally resembles a sharp Gaussian-like shape centered on σi
and receives the name of spectral line. Defining the spectrum
of each line as Si (σ), a relation between the interferogram and II. E XPERIMENTAL METHOD
the lines that make the spectrum can be found by extending
(1) to the case where the intensity distribution is not a delta The introduction lays the foundation for what seems to
function: be a pretty straightforward workflow: set a mercury lamp as
X the source of the interferometer, build an interferogram by
S(σ) = Si (σ) (2) varying ∆, and obtain its Fourier transform. However, there are
Zi ∞ multiple practical details that will require a more sophfisticated
I(∆) = S(σ)(1 + cos (2π∆σ + ϕ(σ))dσ (3) approach, as the description of the experimental method will
0 show.
JOEL M.G. 2

µsteps and meters must be measured. It is for this purpose

that a calibration wavelength was used, which is a reference
sharp line propagated through the same optical path as the
source of interest.

An already present line in its spectrum was used: the

green line. The value for the wavelength of the green mercury
line is 546.0735 nm according to NIST[4]. However, this is
the wavelength in a vacuum: the wavelength measured will
effectively be λ = λ0 /n, where n is the refractive index
of air. An estimation for the n was obtained following the
equation derived by Edlén (1966)[5] with the modifications
made to account for factors like air humidity by Ciddor
(1996)[6]. The following environmental values were chosen
as an estimation for the lab conditions and their variability:
Fig. 2: Diagram of the experimental setup used: 1) Hg lamp. P = 1 ± 0.1 kPa, T = 15 ± 10, U = 80 ± 10 %,
2) 25 mm plano-convex lens to collimate the light. 3) 50:50 c = 500 ± 200 ppm, for the pressure, temperature, humidity,
dielectric beam splitter of the Michelson interferometer. 4) and CO2 concentration in the air, respectively. The uncertainty
Mirror with adjustable tilt. 5) Movable mirror, varying d2 in ranges were chosen so that they were reasonably ensured to
steps of ∼0.0376 nm through a rotating motor. 6) Secondary contain the real conditions during the experiment. This gives
50:50 dielectric beam splitter. 7) Green filter used to isolate n = 1.000243 ± 0.000019, and so the resulting calibration
the green line. 8,9) Photodiode detectors. This configuration wavelength is λ = 545.941 ± 0.011. The raw spectrum
allows the calibration of the interferogram with respect to the obtained when using this correction gives values for air, so
green Hg line. the wavelengths will need to be converted to vacuum again.
The effects and usefulness of this vacuum-air correction
(which will be referenced as VAC) are discussed in the next
A. Data collection section.
The basic Michelson interferometer shown in Fig. 1 was
extended by adding a second beam splitter at the output, and A mean value for the nm/µstep can be found by averaging
sending one of the split beams through a green filter before the number of µsteps between zero-crossing points of the
each beam is intercepted by its own detector, as shown in green line interferogram and equating that to λ/2. However,
Fig. 2. The data recorded for the filtered light will act as a the motor presented some inconsistencies in its movement,
calibration reference, which is further discussed in the next so this value varied significantly across the interferogram.
subsection. Therefore, the position data was converted to meters through
a systematic correction to each half-wavelength by requiring
An interferogram was recorded around ∆ = 0 (namely that each distance between consecutive crossing points (which
the null point), so as to provide the Fourier transform with were interpolated between data points) in the interferogram
information regarding the width of the interferogram to better of the green line corresponds to λ/2, and then correcting the
estimate the reciprocal spectrum width. This requires an position scale. Dividing the position data {X} in sections
interferogram long enough to show its curvature. The null {Xi }, where i corresponds to the crossing points Xi and
point also constitutes the point of biggest contrast between the Xi+1 that enclose the section, the wavelength calibration
fringes in the interferogram. In our particular interferometer, algorithm will be
the beam splitter used was a dielectric half-mirror that added
λ ({Xi } − Xi )
a π phase, so the null point corresponded to a dark central {Xi }correct = Xicorrect + (5)
2 (Xi+1 − Xi )
fringe of destructive interference. It is worth pointing out
λ
that an appropriate sampling rate was chosen to record the Xicorrect = correct
+ Xi−1 (6)
interferogram, which must be at least double the highest 2
frequency to measure (here, this was ensured by aiming for a X0correct := 0 (7)
higher order of magnitude) due to the Nyquist theorem[2].
Another technical aspect to be considered is the fact that since
the interferogram is made out of discrete points, a discrete
B. Pre-processing: data correction Fourier transform (or DFT) is required. In particular, the fast
The raw data was then processed with a Butterworth Fourier transform (or FFT) algorithm was used, which is a
filter[3] to reduce the noise in the interferogram and remove fast implementation of a DFT. These algorithms require evenly
the intensity offset. The motor controlling d2 measured its spaced points as inputs, which is not necessarily the case
estimated distance moved in µsteps, which is a unit that for the data. Therefore, the interferograms were interpolated
corresponds to a 1/(200 · 256) fraction of a revolution of through a cubic spline and a new set of evenly spaced data
the stepper motor. This means that the conversion between points was resampled.
JOEL M.G. 3

C. Post-processing: obtaining results and estimating errors

The final step is concerned with the spectrum obtained
from the FFT, and determining the mean wavelengths of each
peak by fitting the lines to some profile. Instead of using
Gaussians, the more general Voigt profiles[7] were used,
so that each line S(σi ) is modeled as the convolution of a
Gaussian and a Lorentzian, both centered at σi . This way,
the two main mechanisms that broaden the lines away from
their ideal δ(σ − σi ) shape can be modelled[8]. These are
quantum uncertainty (also known as natural or Lorentzian
broadening) and Doppler broadening (also known as Gaussian
broadening). This fit will provide us with the final estimation
for the mean wavelengths.

Regarding the uncertainties, an estimation for the random

error can be obtained from the uncertainties in the parameters
of the Voigt fits (in particular, in the mean wavelength). Fig. 4: Spectrograms of the Hg emissions in the range 400-
The precision of the calibration wavelength was taken at 600 nm, showing effects of the 0.0376 nm/µstep average
the last decimal place in the non-VAC case. The uncertainty approximation and the wavelength correction. The corrected
in the refractive index for air n affects both the calibration spectrum shows four clear lines at ∼435, ∼546, ∼577, and
wavelength and the spectrum once it is converted back to ∼579 nm, as well as a smaller line at ∼404 nm. Each
vacuum: all of these uncertainties are propagated in the final correlates with existent persistent lines, whereas the spectrum
results. without the wavelength calibration is noisy and shifted. With-
out the correction, accurately identifying the lines would not
III. R ESULTS AND DISCUSSION be possible.
Neither the interferograms (Fig. 3) of the Green line nor the
complete spectrum present symmetry, supporting the hypoth-
esis that there is in fact some wavelength-dependent phase
ϕ(σ). The source of this phase is not clear, but the two The position correction does not have a visible effect
main possibilities are the beam splitter (due to thickness on the interferograms; however, it does greatly affect the
imperfections[1]) and the mirrors (due to dispersion[9]). The spectrum after taking the FFT. Shown in Fig. 4, the spectrum
motor’s acceleration was discarded as a source of asymmetry (in a vacuum) with the wavelength calibration shows five
by taking interferograms in both directions and allowing the clear peaks in our target range between 400 and 600 nm,
motor to gain speed before taking data. whereas the spectrum obtained with a fixed average relation
of 0.0376 nm/µstep is noisy and shifted (away from the
reference values that is). In the corrected spectrum, the
lines found are the green line (∼546 nm), the so-called
yellow doublet at ∼577 nm, and the blue-green peak on
the ∼435 nm mark, as well as a small violet line (better
depicted in Fig. 5) at ∼404 nm. The violet line, though
dim, presents a profile very distinct from the noise found
in the region, with a relative scale at least 20 times bigger
than the tallest peaks deemed noise, so it is interpreted as such.

It is clear how without the correction, the interferogram

would provide almost no information about the spectrum, and
lines like the violet one would be completely invisible. The
five lines found correlate with the five persistent lines (i.e.,
stable lines which are found in a wide range of conditions) of
Hg I, though the definitive accuracy of the spectrum cannot
yet be determined.
Setting five Voigt profiles at the initial positions of each
Fig. 3: Recorded interferograms after position correction using peak, the whole spectrum is fitted as a function of each
the calibration wavelength, together with their cubic interpo- profile’s σ0 , and Lorentzian and Gaussian widths (Fig. 5). The
lation. The subplots show: a) interferogram for the green line mean wavelengths from the fit can be obtained for both the
b) interferogram for the complete spectrum. Their asymmetry calibration wavelength assuming vacuum, and the vacuum-air
hints at some wavelength-dependent phase ϕ(σ). correction (Table I).
JOEL M.G. 4

present in that case (1). Either knowing the exact spectrum

from the green line a priori (to then either compare and adjust
the complete interferogram or account for the variance in the
error) or using a calibration light with a sharper spectrum
would increase the accuracy of the experiment.

IV. S UMMARY AND CONCLUSION

By slightly modifying a basic Michelson interferometer

so that the green line of the Hg spectrum was isolated,
two interferograms were recorded. The position scale was
corrected according to the calibration interferogram of the
green line (both considering and ignoring the vacuum-air
variations), which resulted in a great accuracy increase in
Fig. 5: Hg spectrum with the fit for five superposed Voigt the spectrograms. Fitting Voigt profiles to the lines in the
profiles. The violet line is expanded, showing a clear line FFT of the complete Hg interferogram, the mean emission
profile. Voigt profiles account for the two main sources of wavelengths were obtained for the five persistent emission
line broadening, and so deviations are assumed as noise/error. lines between 400 and 600 nm; these values were all at a
maximum distance of 0.06 nm from the reference values.
It could be argued that the vacuum-air correction proved
VAC No VAC Reference value
404.69 ± 0.07 nm 404.62 ± 0.06 nm 404.6563 nm useful, though the uncertainty in the environmental conditions
435.89 ± 0.04 nm 435.797 ± 0.005 nm 435.8328 nm prevented a definitive conclusion.
546.09 ± 0.02 nm 546.058 ± 0.002 nm 546.0735 nm
576.96 ± 0.04 nm 576.92 ± 0.02 nm 576.9598 nm The Michelson interferometer has proven to be a very
579.07 ± 0.03 nm 579.03 ± 0.01 nm 579.0663 nm
precise optical system with the right corrections, allowing us
TABLE I: Mean wavelengths (in a vacuum) of the five to get reasonably close to research-grade accuracy in the five
identified peaks for the data with and without the vacuum- lines measured.
air correction (VAC), as well as the reference values from the
NIST database[4]. All the corresponding lines lie within a 0.06
nm range.
R EFERENCES
As expected, using the VAC implies a bigger uncertainty,
since it is introduced in the vacuum-air conversions through [1] A. Thorne and M. Howells, “Interferometric spectrome-
the refractive index n = 1.000243 ± 0.000019. All of the lines ters,” Jun. 1997.
lie within 0.06 nm from the reference values, so both of the [2] A. V. Oppenheim, A. S. Willsky, and S. H. Nawab,
approaches have proven to be accurate methods. Comparing Signals and Systems. Prentice Hall, 1996.
the two approaches, the smallest two wavelengths were better [3] P. Virtanen, R. Gommers, T. E. Oliphant, et al., “SciPy
estimated (as of the reference values) by the non-VAC lines, 1.0: Fundamental Algorithms for Scientific Computing
while the VAC approach was more accurate for the other in Python,” Nature Methods, vol. 17, pp. 261–272, 2020.
three. Except for the ∼435 nm, all of the reference values are DOI : 10.1038/s41592-019-0686-2.
inside the VAC uncertainty range, while the exact opposite [4] A. Kramida, Y. Ralchenko, J. Reader, and N. A. Team,
applies to the non-VAC lines. Overall, the VAC values seem “Nist atomic spectra database,” NIST Physics Laboratory,
to offer only slightly better results, and truly assessing its 2021, https://physics.nist.gov/asd.
effectiveness would require controlling for the environmental [5] B. Edlén, “The refractive index of air,” Metrologia, vol. 2,
conditions, reducing the VAC uncertainty, since it is also no. 2, pp. 71–80, 1966.
possible that the added complexity to the process negatively [6] P. E. Ciddor, “Refractive index of air: New equations for
affected the results. the visible and near infrared,” Applied optics, vol. 35,
no. 9, pp. 1566–1573, 1996.
The main potential improvements in the experiment [7] W. Voigt, “Das Linearenprofil,” Annalen der Physik,
focus on the position correction based on the green Hg vol. 340, pp. 573–587, 1913. DOI: 10 . 1002 / andp .
line. By equating each distance between crossing points in 19133401005.
the interferogram to λ/2, it is implicitly assumed that the [8] W. Demtröder, Molecular Physics: Theoretical Princi-
experimental green line only presents one wavelength (i.e., ples and Experimental Methods. Wiley-VCH, 2008.
SG (σ) = I0 δ(σ −σG )), which is only an approximation, since [9] B. E. A. Saleh and M. C. Teich, Fundamentals of
its interferogram (Fig. 3) does not resemble the form it should Photonics. Wiley, 1991.
JOEL M.G. 5

A PPENDIX
For the two previous experiments, some feedback was
provided. An effort has been made to correct all of those things
found inadequate (especially those from the Radioactivity
experiment regarding the report), and strengthen those which
were praised.

A. Waves & Propagation

For the first experiment, the lab book was found to be
well-organized with clear methods and results but lacked a
deeper understanding of the physical concepts. The impor-
tance of the dispersion relation and its implications on the
phase/group velocity and distortions seen in the square pulse
was not addressed, which is a crucial aspect of the experiment.
However, the presentation was praised for its clarity, concise
slides, excellent delivery and pace, and a good understanding
of data analysis and physics comprehension.

B. Radioactivity
For the second experiment, the assessor found that the
group had a good understanding of the physics being
investigated and managed to complete most of the main
tasks. However, evidence of the dead time estimate was not
seen in the work, and the group worked quite slowly despite
having an extra person. The lab book was found to be clear
and well-formatted, with insightful discussions around tasks,
but the key findings from each task could have been made
clearer, and slightly longer introductions to each task would
have been an improvement. Nevertheless, all the feedback
provided was incorporated into the lab book.

The abstract and the introduction sections of the report

could benefit from a spellcheck and clearer numerical results,
while the diagram in one of the figures needed annotating
with more details, and the caption lacked any takeaway
message. In general, figure captions should contain all the
information necessary to understand the figures. Given the
length constraint, too much of the report was spent on the
theory rather than the experiment itself. A lack of exploration
of the experimental system was also found. The description
of the measurement method and strategy to minimize the
uncertainty and general practical aspects of the experiment
was lacking. The conclusion drawn from the data analysis
was found to be incorrect due to misinterpretation of the data.
The use of the passive is preferred over the plural of the first
person.