Chapter VI: Frequency Modulation Spectroscopy of YbOH
6.3 Modeling FM lineshapes
0.79) I determined that a modulation frequency in the range 0.5 β² ππ/Ξ β² 0.79 is ideal for our spectroscopic purposes. Specifically, we operate at a modulation frequency of ππ/Ξ β 0.56, indicated by the vertical dashed line in Fig. 6.3 and 6.4. The specific value of ππ/Ξ was determined partially due to the fact that we already had a 2πΓ50.3 MHz EOM in the laboratory. A 2πΓ50.3 MHz modulation frequency falls in the optimal range for FM spectroscopy inside our buffer gas cell (the measured in cell linewidth isΞβΌ2πΓ90 MHz).
a function π(π₯) is given by
π»(π) (π¦) = 1 π
π
β« β
ββ
π(π₯) π¦βπ₯
ππ₯ . (6.14)
Therefore, given the absorption lineshape,πΏ(πβ²) the dispersion line shape is given by the negative Hilbert transform of the absorption lineshape
π(π) =βπ»(πΏ) (π). (6.15) If we make the substitution π’ = (πβ² β ππ π π )/(β
2π) where π = Ξ/2.355 then π π’ =1/(β
2π)π πβ²,πΏ(πβ²)= π(π’) = π΄ exp(βπ’2) and π(π) = 1
π π
β« β
ββ
π(π’) π’β πββππ ππ
2π
π π’
= 1 π
π
β« β
ββ
βπ΄Γexp(βπ’2) π‘βπ’
π π’
=βπ΄ π»[exp(βπ’2)] (π‘),
(6.16)
whereπ‘ = (πβππ π π /(
β
2π). The Hilbert transform of π(π’) =exp(βπ’2) is known and is related to the Dawson function [155, 156]
π»[exp(βπ’2)] (π‘) = 2
β π
πΉ(π‘), (6.17)
whereπΉ(π‘) is the Dawson function. Therefore, the dispersion lineshape is given by π(π) =βπ΄
β2 π
πΉ
(πβππ π π )
β
2(Ξ/2.355)
!
. (6.18)
The dispersion lineshape given in Eq. 6.18 is convenient for numerical modeling since the Dawson function is a built in function in several programming languages4. Simulated Gaussian absorption and dispersion lineshapes (Eq. 6.12 and 6.18) are shown in the top left and right panels of Fig. 6.5, respectively.
Ultimately, we want to model a true absorption spectrum which will contain an arbitrary superposition of Gaussian lineshapes
πΏπ‘ ππ‘(π) =βοΈ
π
πΏπ(π) =βοΈ
π
π΄π exp
β (πβππ)2 2(Ξπ/2.355)
, (6.19)
4The Dawson function is a built in special function in python, which is the language used here to model the FM lineshapes.
where π denotes the absorption lineshape due to the ith transition, and ππ is the resonance frequency of the ith transition. Therefore, the total dispersion lineshape due to the combination of multiple transitions is given by
ππ‘ ππ‘(π)= 1 π
π
β« β
ββ
πΏπ‘ ππ‘(πβ²) πβ²βπ
π π
=βοΈ
π
π΄π
β2 π
πΉ
(πβππ)
β
2(Ξπ/2.355)
! .
(6.20)
The simulated total Gaussian absorption and dispersion lineshapes for five closely spaced and blended transitions are shown in the top left and right panels of Fig. 6.6, respectively.
With Eq. 6.19 and 6.20, any arbitrary absorption and dispersion line shape can be modeled. Any arbitrary FM lineshape can be modeled by using Eq. 6.19 and 6.20 in Eq. 6.7, 6.8, 6.6, and 6.11. However, Eq. 6.7 and 6.8 account for the attenuation of the carrier due to strong absorption as well as the effects of the sidebands of all orders. While modeling the FM lineshapes using Eq. 6.7 and 6.8 would provide the most accurate models, including the aforementioned effects is excessive and could complicate the fitting by providing too large of an optimization space.
We are specifically utilizing FM spectroscopy to measure weak absorption lines so operating in the weak absorption limit, discussed in Section 6.1, is justified.
Therefore, in our FM linshape model we assume exp[β2πΏ(π)] β 1. Additionally, Eq. 6.9 and 6.10 indicate that at lower modulation depth π΄πΉ π(π)/π·πΉ π(π) β π and therefore, operating at higher modulation depths will result in larger FM signals.
Experimentally, we observe this trend with larger signal sizes for larger modulation depths. In order to maximize our FM signal sizes we operate at a modulation depth of π = 0.84. The modulation depth is fixed to this value for all modeled FM lineshapes shown in this chapter. At this modulation depth we find that the amplitude of the the first-order sidebands is β 20% of the carrier amplitude and the amplitude of second-order sidebands are β 1% of the carrier amplitude. In order to include the effects of the second order sidebands we model theπ΄πΉ π(π)and π·πΉ π(π)lineshapes to second order (the n=0,1 terms in Eq. 6.7 and 6.8). Therefore,
in the small absorption limit the FM absorption lineshape is given by [153]
π΄2ππ
πΉ π(π) =π½0(π)π½1(π) [πΏ(πβππ) βπΏ(π+ππ)]
+π½1(π)π½2(π) [πΏ(πβ2ππ) βπΏ(π+2ππ) +πΏ(πβππ) βπΏ(π+ππ)], (6.21) and the FM dispersion lineshape is given by
π·2ππ
πΉ π(π) =π½0(π)π½1(π) [π(πβππ) +π(π+ππ) β2π(π)]
+π½1(π)π½2(π) [π(πβ2ππ) +π(π+2ππ) βπ(πβππ) βπ(π+ππ)]. (6.22) Given a set of transitions (each with transition frequencyππ, widthΞπ, and amplitude π΄π) and a phase angle, π, the πΌπΉ π(π)andππΉ π(π)lineshapes are calculated using Eq. 6.6 and 6.11 respectively, where π΄πΉ π(π) is given by Eq. 6.21, π·πΉ π(π) is given by Eq. 6.22, πΏ(π) is given by Eq. 6.19, and π(π) is given by Eq. 6.20.
The simulated I and Q FM lineshapes at various values of π between π = 0 and π = π for a single transition with a Gaussian absorption lineshape are shown in Fig. 6.5. A modulation frequency ofππ/Ξ =0.56, the same modulation frequency used in our FM setup, was used when modeling the lineshapes in Fig. 6.5. The simulated I and Q FM lineshapes at various values of π between π = 0 and π = π for five closely spaced transitions are shown in Fig. 6.6. Here, the FWHM of all transitions was set to 90 MHz and a modulation frequency of 50.3 MHz was used (ππ/Ξ = 0.56). In Fig. 6.6 several of the transitions are blended to illustrate the resulting FM lineshapes when blended lines are present. In all cases, both isolated transitions in Fig. 6.5 and blended transitions in Fig. 6.6, at a phase angle ofπ =0 πΌπΉ π(π) (π =0) = π΄πΉ π(π) andππΉ π(π) (π =0) =βπ·πΉ π(π). As the phase angle increases the I and Q signals become mixtures of π΄πΉ π(π) andπ·πΉ π(π) up toπ = π/2 whereπΌπΉ π(π) (π = π/2) = π·πΉ π(π) andππΉ π(π) (π = π/2) = π΄πΉ π(π). For phase angles betweenπ/2 andπthe I and Q signals are again a mixture ofπ΄πΉ π(π) and π·πΉ π(π). At π = π πΌπΉ π(π) (π = π) = βπΌπΉ π(π) (π = 0) = βπ΄πΉ π(π) and ππΉ π(π) (π = π) =βππΉ π(π) (π =0) = π·πΉ π(π). For phase angles ofπ < π < 2π πΌπΉ π(π) (π) =βπΌπΉ π(π) (πβ²=πβπ) andππΉ π(π) (π) =βππΉ π(π) (πβ² =πβπ). In all casesπΌπΉ π(π) (π+2π) =πΌπΉ π(π) (π)andππΉ π(π) (π+2π) =ππΉ π(π) (π). Ultimately, we want to perform a simultaneous fit of our modeled I and Q FM linshapes to the measured I and Q FM data. This is accomplished with a non-linear
least squares optimization5which takes the FM data and initial guesses for the phase angle and the parameters of each transition present (the transition frequency, width, and amplitude of each transition) as inputs. The optimization works to minimize the set of residuals provided to it. To accomplish the simultaneous fit the following residual function was used
π (π) =
βοΈ
[πΌπππ π(π) βπΌπππ‘ π(π)]2+ [ππππ π(π) βππππ‘ π(π)]2, (6.23) whereπΌπππ π(π)(ππππ π(π)) is the calculated value ofπΌπΉ π(ππΉ π) at the frequencyπ and πΌπππ‘ π(π) (ππππ‘ π(π)) is the measured value of πΌπΉ π (ππΉ π) at the frequencyπ. The sum of the squares of the individual I and Q residuals as opposed to just the sum of the I and Q residuals was used to prevent the residual from taking on inaccurately small values due to a cancellation resulting from the I and Q residuals being opposite in sign. In the fit, the phase angle π and the lineshape parametersππ, Ξπ, and π΄π are floated. Any arbitrary number of transitions can be fit by the algorithm. An example of a fit of the model to FM data is shown later in this chapter in Fig. 6.11.
For isolated lines, we find that the linecenters extracted from the fit exactly match our zero crossing measurements and have equivalent or smaller errors.
5The lmfit Python package [157] is used for the non-linear least squares fitting.
-4 -2 0 2 4
( )
-4 -2 0 2 4
( )
-4 -2 0 2 4
-5 0
5 I
FM( ) = 0
-4 -2 0 2 4
-5 0
5 Q
FM( ) = 0
-4 -2 0 2 4
-5 0
5 I
FM( ) = /8
-4 -2 0 2 4
-5 0
5 Q
FM( ) = /8
-4 -2 0 2 4
-5 0
5 I
FM( ) = /4
-4 -2 0 2 4
-5 0
5 Q
FM( ) = /4
-4 -2 0 2 4
-5 0
5 I
FM( ) = /2
-4 -2 0 2 4
-5 0
5 Q
FM( ) = /2
-4 -2 0 2 4
-5 0
5 I
FM( ) = 3 /4
-4 -2 0 2 4
-5 0
5 Q
FM( ) = 3 /4
-4 -2 0 2 4
-5 0
5 I
FM( ) =
-4 -2 0 2 4
-5 0
5 Q
FM( ) =
Frequency ( res )/
FM signal Amplitude (arb.)
Figure 6.5: Simulated in-phase,πΌπΉ π(π), and in-quadrature,ππΉ π(π), FM signals for an isolated spectral feature with a Gaussian absorption, πΏ(π), and dispersion, π(π), lineshape. The in-phase and in-quadrature signals are shown for various phase angles,π, ranging fromπ =0 toπ =π. For phase angles> π πΌπΉ π(π) (π =π+π)=
βπΌπΉ π(π) (π =π) andππΉ π(π) (π =π+π) =βππΉ π(π) (π =π)for 0 β€ π β€ π.
0.00 0.02 0.04 0.06
( )
0.00 0.02 0.04 0.06
( )
0.00 0.02 0.04 0.06 -5 0 5 I
FM( ) = 0
0.00 0.02 0.04 0.06 -5 0 5 Q
FM( ) = 0
0.00 0.02 0.04 0.06 -5 0 5 I
FM( ) = /8
0.00 0.02 0.04 0.06 -5 0 5 Q
FM( ) = /8
0.00 0.02 0.04 0.06 -5 0 5 I
FM( ) = /4
0.00 0.02 0.04 0.06 -5 0 5 Q
FM( ) = /4
0.00 0.02 0.04 0.06 -5 0 5 I
FM( ) = /2
0.00 0.02 0.04 0.06 -5 0 5 Q
FM( ) = /2
0.00 0.02 0.04 0.06 -5 0 5 I
FM( ) = 3 /4
0.00 0.02 0.04 0.06 -5 0 5 Q
FM( ) = 3 /4
0.00 0.02 0.04 0.06 -5 0 5 I
FM( ) =
0.00 0.02 0.04 0.06 -5 0 5 Q
FM( ) =
Frequency (cm 1 )
FM signal Amplitude (arbitrary)
Figure 6.6: Simulated in-phase,πΌπΉ π(π), and in-quadrature,ππΉ π(π), FM signals for five closely spaced and blended spectral features. All five spectral features were modeled with a Gaussian absorption,πΏ(π), and dispersion, π(π), lineshape. The in-phase and in-quadrature signals are shown for various phase angles, π, ranging fromπ =0 toπ =π. For phase angles> π πΌπΉ π(π) (π =π+π) =βπΌπΉ π(π) (π =π) and ππΉ π(π) (π = π + π) = βππΉ π(π) (π = π) for 0 β€ π β€ π. A modulation frequency of 50.3 MHz and a FWHM for all transitions of 90 MHz were used in the simulations. The absolute value of the frequency of the x-axis is arbitrary.