High-Resolution Observations of the Infrared Spectrum of Neutral Neon

We have observed the spectrum of neutral neon (Ne I) emitted by a microwave-excited electrodeless discharge lamp with the National Institute of Standards and Technology 2 m Fourier transform spectrometer. The spectra cover the regions 6929 Å to 11 000 Å with a resolution of 0.01 cm−1 and 11 000 Å to 47 589 Å with a resolution of 0.007 cm−1. We present a line list that includes more than 650 classified lines and provides an accurate and comprehensive description of the infrared spectrum. The response of the Fourier transform spectrometer was determined by using a radiometrically calibrated tungsten strip lamp, providing relative intensities that for moderate to strong lines are accurate to approximately 10 % over the entire range of the observations. The identities of many lines that were previously multiply classified are unambiguously resolved.


Introduction
Neon discharges are widely used in scientific, technical, and commercial applications. Despite the fact that Ne is frequently used as a buffer gas in sources for laboratory spectroscopy, no comprehensive description of its spectrum in the extraphotographic infrared region has appeared in the literature. Photographic spectra recorded with large grating spectrographs in the near infrared region have reasonably high resolution but extend to only about 12 000 Å [1]. At longer wavelengths, spectra obtained with infrared scanning instruments have resolution and accuracy that are very low by current standards [2,3].
The most complete line list for Ne at wavelengths longer than 11 000 Å was given by Humphreys [4] based on spectra recorded with a scanning 1 m grating spectrometer. The relative intensities in this list are experimental values from Humphreys's grating observations, but the wavelengths are calculated from the Ne energy levels of Kaufman and Minnhagen [5]. Because of the low resolution and accuracy of the experimental spectra, many lines are multiply classified. In some cases as many as five classfications and calculated wavelengths are associated with a single feature in the observed spectrum. For such a case Ref. [4] provides no information about the relative intensities of these possible transitions in a high resolution spectrum, nor does it provide any basis for estimating the effective wavelength of the unresolved transitions at low resolution.
Measurements for 118 infrared lines of neon were reported by Chang et al. [6] based on hollow cathode spectra from the archives of the Fourier transform spectrometer of the National Solar Observatory at Kitt Peak National Observatory. These are the most accurate published measurements for neon in the infrared. Unfortunately, the work of Chang et al. has little value as a description of the spectrum since it gives no intensities and includes only selected lines.
In order to provide a comprehensive high-resolution description of the infrared spectrum of Ne, we made new observations with the National Institute of Standards and Technology (NIST) 2 m Fourier transform spectrometer (FTS). This is one component of a broader program of observations and compilations for the noble gases currently in progress at NIST. A more detailed description of this program and of our experimental observations for the infrared spectra of the noble gases has been given in Ref. [7]. We have also produced a complete compilation of transitions and energy levels for Ne I [8], which includes many data from the present work.

Experiment
The spectrum was excited in a commercial sealed electrodeless discharge lamp filled for this work with Ne at a pressure of 200 Pa (1.5 Torr). The lamp design, illustrated in Fig. 1, is derived from that used by Wilkinson and Tanaka [9]. The lamp is equipped with a cemented magnesium fluoride window that permits viewing along the axis of the discharge. The double wall in the area of the window protects the epoxy seal from the discharge as demonstrated by Bass [10]. The lamp contains a barium getter in a side arm to trap impurities released from the walls during operation. Initial attempts to power the lamp using an Evenson cavity [11] were unsuccessful because we could not maintain stable tuning of the cavity over the several hours required for data acquisition. This resulted in large discontinuous variations in light output that were unacceptable for the FTS. We therefore turned to a cuptype antenna directed toward the side of the lamp near its midpoint to couple power to the discharge. This produced very stable excitation.
Light from the electrodeless lamp was directed to the entrance aperture of the FTS through a path purged with dry air to avoid infrared absorption by atmospheric water vapor. The purged path incorporates a remotely-actuated rotating mirror and a concave mirror as shown in Fig. 2, permitting light from either the Ne lamp or a radiometric standard source to be imaged to the entrance aperture of the FTS. The radiometric standard source, a tungsten strip lamp with sapphire window, was used for calibration of the instrumental response.
Spectra were recorded with different combinations of beamsplitter, detector, and filter in three overlapping regions, covering the range from 7000 Å to 50 000 Å. The combinations used and resolution for each of these regions are summarized in Table 1. The instrumental resolution was 0.01 cm -1 for the shortest wavelength region, extending to 11 000 Å, and 0.007 cm -1 for the longer wavelength regions. This was sufficient to fully resolve all of the lines. Optical filters were used to eliminate the visible spectrum and scattered light from the He/Ne reference laser. Two detectors at the complementary outputs of the FTS were operated differentially to acquire the interferogram. Sixteen scans of the interferogram were coadded, representing a data acquisition time of about one hour for each spectrum. Six Ne  spectra were recorded, two for each of the three spectral regions. Before or after each observation of the Ne lamp, a spectrum of the tungsten strip lamp was recorded. Representative spectra of this standard lamp for the three spectral regions are shown in Fig. 3. These calibration spectra were later used to correct the intensities in the Ne spectra for the response of the FTS, filters, and detectors.
The interferograms were transformed and the spectra were phase corrected and measured by using the interactive program Xgremlin, an X-windows implementation of the FTS analysis program GREMLIN [12] with graphical user interface and significantly enhanced capabilities that was developed by Griesmann at NIST. A brief description of Xgremlin is given in [13].
Ne has only small isotope shifts between the dominant 20 Ne, which accounts for 91 % of the natural abundance, and 22 Ne which makes up virtually all of the balance. Neither of these isotopes has magnetic hyperfine structure (hfs). Consequently, the observed lines in our Ne spectra were generally narrow and symmetric as illustrated by the lines in Fig. 4. Lines of this type were measured by fitting the observed profile with a Voigt function. A few lines had sufficiently large isotope shifts to display significant asymmetry. These lines were measured by interactively marking upper and lower wave-number limits that enclose the line and calculating the center of gravity of the spectrum between those limits. For all lines the integrated intensity under the line profile was determined.

Calibration and Data Processing
Spectra recorded with an FTS are linear in wave number to very high precision, but to obtain absolute accuracy of better than a few parts in 10 6 it is necessary to correct the scale by a multiplicative constant derived from accurately known internal standard lines. For this purpose we used wave numbers calculated from optimized level values of the 2p 5 3s, 3p, 3d, and 4s configurations. Lines of these transition arrays have been recommended as secondary standards by the International Astronomical Union (IAU) [14]. Sixty-one lines of the 3s-3p, 3p-3d, and 3p-4s transition arrays were used for the calibration. Lines of the 3d-4p and 4s-4p transition arrays were not used because they gave results that were systematically inconsistent with the other calibration lines.
For each spectrum the lines were initially measured with respect to the uncorrected wave number scale. A correction factor was then determined by taking the unweighted average of individual correction factors calculated from each of the standard lines. Not all standard lines appeared in all spectra. For spectra covering the range 7000 Å to 11 000 Å about 40 standards were used and for the longer wavelength regions about 27 standards were used. In all cases the average correction was in the range 4 to 6 parts in 10 7 and the standard deviation of the individual values was about 1 part in 10 7 .
The final wave number for each line was calculated from the corrected values as the unweighted average of the individual measurements. For most lines there were two to four observations. For a few lines there was only a single observation; for others there were as many as six. The uncertainty for each line was calculated as the quadrature sum of three terms: the calibration uncertainty as measured by the standard deviation of the individual line correction factors, the standard deviation of the multiple measurements of the line, and the estimated precision with which the line position could be measured in the spectra. For most lines of moderate or greater intensity the uncertainty is dominated by the first two terms. The third term is calculated as the line width divided by twice the signal-to-noise ratio at the line center [15]. It is included to insure that weak or broad lines are not assigned an unreasonably low uncertainty because of an accidentally high degree of agreement between a small number of measurements.
For each Ne spectrum the radiometric response of the combination of FTS, filters, and detectors was determined by recording the spectrum of the standard tungsten strip lamp with the same filters, detectors, and FTS observing parameters. This spectrum was compared to the previously calibrated output of the strip lamp to generate an instrumental response curve that was used to adjust the integrated intensities of the spectral lines to a uniform linear dependence on the number of photons detected. With this choice for the calibration, the ratios of intensities of lines with a common upper level give directly the branching ratios of the various decay paths.
After calibrated intensities had been obtained for each spectrum, lines of moderate intensity in the overlapping spectral regions were used to determine scaling factors that were applied to place all of the spectra on a common intensity scale. The intensities from the multiple observations of each line were then averaged. For the two spectra in the 13 000 Å to 50 000 Å region, intensities of the 11 lines between 9218 cm -1 and 9465 cm -1 were omitted from the intensity average because they differed systematically from the intensities of the same lines measured in the two shorter wavelength regions. These lines were at the extreme end of the long wavelength region where the instrumental response was very low. Finally, the entire set of average intensities was scaled to obtain values on a linear scale from 1 to 100 000.
In order to estimate the uncertainty of the intensities, we examined the ratio of the standard deviation to the average value for each of the approximately 600 lines for which more than one measurement was made. From this analysis we observed that the relative uncertainty of the intensities is about 10 % independent of intensity for lines with intensity of 100 or greater, 15 % for lines with intensities 10 to 100, and 25 % for lines weaker than 10. The analysis of uncertainties is summarized in Table 2, where we present the percentage of lines for which the relative standard deviation lies within 1, 2, and 3 times the stated uncertainty for each decade in the intensity. Based on this distribution, the stated uncertainties represent approximately a 90 % level of confidence. The statistics for the 10 000 to 100 000 intensity range reflect the fact that the few lines with intensities greater than 50 000, especially those that are self-reversed, are somewhat less reproducible than weaker lines.

Results
Results of this work are presented in Table 3. Virtually all lines previously reported in this spectral region have been observed. Only lines that could be reliably classified as transitions between Ne I levels are reported in the table. Impurity lines of Ar I, Xe I, O I, C I, and Hg I were identified and removed. Also removed were 23 weak lines that we were unable to identify. The strongest of these had intensity 57. All others were much weaker.
The intensities reported in the first column of Table  3 are on a scale linear in photon number as described above. Lines that were used as internal standards for calibration of the spectra are indicated by an S following the intensity. An asterisk in this position denotes a line that is multiply classified. In almost all cases multiply-classified lines are associated with the small intervals between pair-coupled levels in 5p 5 nf and 5p 5 ng configurations. These transitions are unresolvable in observation of the emission spectrum because their Doppler widths exceed the level separation. The many unresolved blends of transitions to 5p 5 nd levels in Ref. [4] have all been resolved in this work.
The observed wavelengths and their uncertainties are given in the second and third columns of Table 3. Wavelengths shorter than 20 000 Å are reported in standard air; wavelengths longer than 20 000 Å are vacuum values. The observed vacuum wave numbers were converted to standard air wavelengths by using the index of refraction of air as calculated from the three term formula of Peck and Reeder [16]. Uncertainties are reported at the one standard deviation level representing a 68 % confidence interval. The wavelengths have been rounded so that the uncertainty in the least significant digit does not exceed 20. Corresponding values of the vacuum wave number and its uncertainty are given in columns four and five.
The classification for each line is given in the remaining columns of Table 3. The configuration, term, and J value for the lower level is given first followed by the same information for the upper level. All level designations are given in the J 1 l coupling notation. The line identifications were initially made based on the Ne I level values reported by Kaufman and Minnhagen [5] and by Chang et al. [6]. A few identifications were later revised based on calculated transition probabilities and re-optimized level values determined in our comprehensive compilation of Ne I wavelengths and energy levels [8]. The classifications in Table 3 are fully consistent with Ref. [8].

Discussion
We have made new observations of the Ne I spectrum in the infrared region with high resolution and signal-to-noise ratio. Our results provide the first comprehensive experimental description of the spectrum in the extraphotographic infrared. Approximately 650 lines have been classified as transitions among previously reported Ne I levels. Most of the multiply-classified lines in the work of Humphreys [4] have now been resolved as described more fully in Ref. [7]. The lines shown in Fig. 4, for example, were reported as a single line with five classifications in Ref. [4].               Letters or symbols in the comment column have the following meanings: a -asymmetric. r -self reversed. * -multiply classified. S -used as an internal standard for absolute calibration.
Our newly measured wavelengths for all transitions to levels of the 5p 5 4p configuration in the 20 000 Å to 50 000 Å region are systematically shifted with respect to their calculated values based on the levels of Kaufman and Minnhagen [5] as we have previously described in Ref. [7]. The 5p 5 4p level values in Ref. [5] are consistent with values recommended by the IAU [14] as suitable for calculation of secondary wavelength standards. Our measurements indicate that the levels of this configuration must be shifted upward by approximately 0.0054 cm -1 . A reoptimization of the 5p 5 4p levels based on all available data from the ultraviolet to the infrared has been made and confirms this shift [8].