A HIGH-RESOLUTION ATLAS OF URANIUM–NEON IN THE H BAND

The growing interest in high-precision astrophysical spectroscopy (such as the detection of low-mass exoplanets) requires the measurement of radial velocities (RVs) at the meter per second (m s⁻¹) level. This corresponds to a Doppler shift in the astrophysical spectrum of 3 parts in 10⁹. Such precision is being achieved now in the optical with iodine cells or hollow-cathode lamps (HCLs) with stable heavy elements such as thorium (²³²Th) and a fill gas (usually argon; Marcy & Butler 1992; Mayor et al. 2009). In the near-infrared (NIR), sub 10 m s⁻¹ precision calibration sources such as telluric CO₂ lines (Figueira et al. 2010) and gas cells (Mahadevan & Ge 2009; Bean et al. 2010) are limited to narrow spectral regions, and thorium HCLs have few bright, calibrated lines compared with the visible. We propose that uranium–neon (U/Ne) HCLs are a more suitable calibration emission source in the NIR than Th/Ar (Redman et al. 2011; Ramsey et al. 2008).

The ideal calibration source would be a picket fence of bright emission lines or deep absorption lines covering a wide wavelength coverage and exhibiting a uniform density and uniform intensities throughout the observed spectral region. Such a calibration source should not be too dense because blended lines are difficult—if not impossible—to use for precise dispersion solutions. The calibration source should also be repeatable over long temporal baselines and should be relatively inexpensive and easy to operate.

Laser-frequency combs (LFCs) provide a picket fence of lines and are an ideal calibration source. Even with their high cost and technical complexity, they have quickly been adopted by the astrophysical community, which is anxious for the development of such a high-precision source. Steinmetz et al. (2008) used the Vacuum Tower Telescope (VTT) solar telescope to demonstrate a comb with sunlight. Efforts are being made to develop and test such astro-combs by many groups in the optical for the High-Accuracy Radial velocity Planet Searcher (HARPS), the Trans-Atlantic Exoplanet Survey (TrES), and the High-Accuracy Radial velocity Planet Searcher North (HARPS-N) spectrographs (Li et al. 2008; Szentgyorgyi et al. 2008; Wilken et al. 2010), as well as future European Southern Observatory (ESO) facilities such as the Echelle Spectrograph for Rocky Exoplanet and Stable Spectroscopic Observations (ESPRESSO; Pepe et al. 2010) and the Cosmic Dynamics and Exo-earth experiment (CODEX; Pasquini et al. 2005). Our own efforts have focused on developing NIR LFCs (Quinlan et al. 2010) to enable the detection of terrestrial-mass planets in the habitable zones of M dwarfs (the smallest and most populous stars in the galaxy) when coupled to the next generation of NIR precision RV spectrographs like the Habitable Zone Planet Finder (HZPF) that are now being built (Mahadevan et al. 2010).

In the meantime, HCLs remain the next best solution for the high-precision calibration of fiber-fed spectrographs in the NIR. Thorium is an ideal atomic emission source as it has only one natural isotope, zero nuclear spin, a long half-life, and an abundance of lines. Uranium shares all of these traits except the first: there are three naturally occurring isotopes of uranium, but the abundance is dominated by ²³⁸U, which makes up 99.275% of the element, followed by 0.720% of ²³⁵U and 0.005% of ²³⁴U.⁸ Uranium also has approximately twice as many emission lines in the NIR (intrinsically). Since only a fraction of the thorium lines in the NIR have been published (Kerber et al. 2008), a more complete uranium line list represents a significant gain in the number of measured calibration lines. Palmer et al. (1980) published an atlas of uranium for the optical, and Engleman (2003) commented on the significantly higher density of bright lines in the emission spectra of uranium compared with thorium. There are now three uranium line lists: Palmer et al. (1980), which covers the optical up through 910 nm, Redman et al. (2011, hereafter R11), which features uranium lines between 850 nm and 4000 nm, and Conway et al. (1984), which covers the NIR between 1800 nm and 5500 nm. We refer the interested reader to the work of Kerber et al. (2007) for a review of the development of these lamps and to Kerber et al. (2008) for a detailed NIR thorium line list.

Bright lines from the fill gases of HCLs are useful for moderate-precision wavelength calibration but are a hindrance for sensitive high-resolution instruments that require a very high precision calibration and minimal scattered light. Whaling et al. (2002) have demonstrated that pressure changes shift the measured wavenumbers of transitions from the 4f, 5g, and 6g levels of Ar i. As such, these lines are not suitable for precision RV measurements below tens of m s⁻¹ (Lovis & Pepe 2007). In the NIR, bright argon lines hamper the analysis of the fainter actinide lines that are more stable and suitable for wavelength calibration. Neon, on the other hand, has relatively few lines beyond 900 nm and thus produces less scattered light and pixel saturation—see Figure 1 of R11 for a comparison between Th/Ar, U/Ar, and U/Ne. These factors make U/Ne a potentially useful calibration source for precise NIR spectroscopy.

Spectral atlases are essential to the initial wavelength calibration of grating spectra. They provide both approximate relative intensity and wavelength information that the human brain can quickly compare to an observed spectrum in a fraction of time that would otherwise require extensive computer programming and processing time. Atlases are useful both for gross wavelength calibration of an unknown spectral region (e.g., when a spectrograph is commissioned) and for the identification of specific spectral regions (e.g., when a telescope user obtains a spectrum at a particular grating setting). Once the observed spectral region has been identified, the user can refer to the original line list for specific line information and measure the line centroids to extract the dispersion solution.

We have created an atlas of U/Ne between 1454 nm and 1638 nm using a simultaneous LFC (Murphy et al. 2007; Osterman et al. 2007; Braje et al. 2008) with a second calibration fiber to provide a first-order estimate of the wavelength calibration of the hollow-cathode spectrum. We then refined the dispersion solution using the uranium line list of R11. We observed this spectrum at a current (14 mA) and resolution (50,000) that are more typical of astrophysical spectrographs than the high-current (75–300 mA), high-resolution Kitt Peak Fourier transform spectrometer (FTS) (Brault 1976) spectra that were used to measure the uranium wavelengths in R11. The format of this atlas makes it suitable for day-to-day use by astronomers both for visual inspection and for wavelength calibration, and its layout is comparable to Th/Ar atlases that are in routine use at telescopes around the world.⁹

A number of stable fiber-fed instruments that are now being planned or are in the design stages could benefit from the atlas published here. The Sloan Digital Sky Survey III (SDSS-III) Apache Point Observatory Galactic Evolution Experiment (APOGEE) galactic kinematics survey is currently commissioning a 300-fiber R ≈ 22,500 H-band fiber-fed spectrograph covering the spectral region between 1530 nm and 1680 nm (Wilson et al. 2010). New high-resolution NIR spectrographs like the Immersion Grating Echelle Spectrograph (iSHELL; Tokunaga et al. 2008; already under construction) can benefit from better calibration sources. The next generation of high-resolution NIR spectrographs built specifically for precision RV surveys, such as the HZPF (Mahadevan et al. 2010), the Calar Alto high-resolution search for M dwarfs with Exo-earths with Near-infrared and optical Échelle Spectrographs (CARMENES; Quirrenbach et al. 2010), and SpectroPolarimetre InfraROUge (SPIRou; J. Donati 2012, in preparation), are now in their planning stage, and the U/Ne calibration source we present here may help them achieve their desired precision.

We describe the spectrometer and our H-band frequency comb in Section 2. In Section 3 we briefly describe how we obtained our spectra. In Section 4 we explain how we reduced and wavelength-calibrated the LFC and U/Ne data. In Section 5 we discuss some of the limitations of the atlas and compare it to the precision of the frequency comb. This atlas is available as supplementary material in the online journal as a series of annotated figures and as a FITS file.

2.1. The Fiber-fed NIR Pathfinder Spectrograph

The Pathfinder testbed used for this work is similar to the instrument we used as a testbed to demonstrate 7–10 m s⁻¹ NIR RV precision on sunlight in Ramsey et al. (2008). Details of our current setup and improvements are described in Ramsey et al. (2010), but we mention key design elements here for completeness. Pathfinder is a fiber-fed warm-pupil cross-dispersed echelle spectrograph that yields a spectral resolution λ/Δλ ≈ 50,000. The input fiber coupling is designed to match the f/4.2 output of the fiber from the Hobby–Eberly Telescope (HET) fiber instrument feed. A pair of commercial NIR achromats optimized for the 0.75–1.55 μm region image the beam from the HET and calibration fibers onto a 100 μm slit. The diverging beam is folded by a small gold mirror on the way to the collimator, which in turn directs the light to a 715 blaze angle (R3) echelle grating. While this grating is undersized in the dispersion direction for the 100 mm Pathfinder beam, it is twice as efficient as that used in Ramsey et al. (2008). This R3 echelle operates in-plane at an off-Littrow angle of ≈5° to achieve high dispersion. This mode of operation (no γ angle tilt of the echelle) ensures that there are no slit curvature effects. A gold-coated 150 l mm⁻¹ grating blazed at 546 provides cross dispersion, enabling our 1024 × 1024 HAWAII-1K array (Hodapp et al. 1996, 2004) to simultaneously image seven to eight echelle orders in the Y band and three to four orders in the H band. We use the same camera system as described in detail in Ramsey et al. (2008), which uses a parabolic mirror with a weak coma-correcting lens near the entrance of the NIR dewar. The uncooled Pathfinder testbed is made possible by liquid-nitrogen-cooled (78 K) thermal blocking filters that suppress the out-of-band thermal radiation to a high degree (S. Mahadevan et al. 2012, in preparation).

In its H-band configuration, Pathfinder has three optical fibers (one stellar fiber leading to the focal plane of the HET and two calibration fibers that lead to our calibration bench) that can simultaneously be illuminated. These fibers were polished and assembled on-site at the HET and then aligned with a 100 μm slit as shown in Figure 1. The alignment of the three fibers with the slit is not perfect, and this causes a constant offset in the wavelength calibration from the three fibers. By feeding LFC light through all three fibers simultaneously, the dispersion solution of each fiber can be independently measured to obtain the offset between the different fibers (see Section 5.1).

Zoom In Zoom Out Reset image size

2.2. The H-band Frequency Comb as an NIR Calibration Source

The National Institute of Standards and Technology (NIST) and University of Colorado at Boulder (CU) H-band LFC is built around a 250 MHz passively mode-locked erbium fiber laser. The laser is frequency-stabilized by locking both the repetition rate f_rep (direct detection) and the carrier-envelope offset frequency f_ceo (f − 2f optical heterodyne; Jones et al. 2000) to an Rb clock that is steered on the long term by the global positioning system (GPS). In this configuration, the passively mode-locked laser generates modes at frequencies of

$$\begin{equation} f_n = n \times f_{\rm rep} + f_{\rm ceo}, \end{equation} \tag{ 1 }$$

centered at 1550 nm with a bandwidth of about 70 nm. The accuracy of the mode frequencies is limited by the accuracy of the GPS-stabilized Rb clock and is better than 1 part in 10¹¹ (Quinlan et al. 2010). For use with Pathfinder, the mode spacing is increased using Fabry–Pérot filter cavities. Two identical cavities are employed with 25 GHz free spectral range and finesse of about 2000; one is placed immediately after the mode-locked laser and the other is placed after a series of optical amplifiers (see Figure 2). The lengths of the cavities are actively locked by the use of piezoelectric actuators to the LFC for maximum transmission of one set of 25 GHz spaced comb modes. After filtering and amplification, the 25 GHz comb is broadened in a highly nonlinear fiber and then transmitted via a single-mode fiber to an integrating sphere, where a fraction of this light enters a 300 μm fiber that leads to the spectrograph. After filtering, the comb equation for the 25 GHz comb is

$$\begin{equation} f_n = \left(n_0 + 100 n\right) \times f_{\rm rep} + f_{\rm ceo}, \end{equation} \tag{ 2 }$$

where n₀ is the index of one comb mode transmitted through the filter cavity, measured with a wavemeter. For these experiments, n₀ was 782,971, f_rep was 250.1132048 MHz, and f_ceo was 70.11320482 MHz. Further details of the comb and our tests with Pathfinder are presented in G. G. Ycas et al. (2012, in preparation).

Zoom In Zoom Out Reset image size

The simultaneous frequency comb and U/Ne spectrum for this atlas were collected as part of an on-sky commissioning run of the Pathfinder spectrograph at the HET. Light from the frequency comb was fed into the primary calibration fiber, while U/Ne light was fed simultaneously through the secondary calibration fiber (the HET stellar fiber was not needed during this experiment). We ran a Photron¹⁰ U/Ne HCL (part number P863, serial number HGL0170) continuously at 14 mA. The Pathfinder HAWAII 1K detector can only cover about 20% of the observable spectral range, so the grating settings had to be adjusted incrementally throughout the experiment. At each grating setting, we obtained 10 30 s U/Ne and frequency comb images and five 30 s flat-field images from a quartz tungsten halogen bulb. All exposures were of the same length so that we could take exposures continuously. Thirty second dark frames were taken throughout the observing run and used to background subtract the exposures. The flat-field exposures were taken with neutral density filters in front of each fiber, which had to be removed during the U/Ne and comb exposures. During these exchanges, a shutter in front of the fiber output in the pathfinder experiment was closed to minimize the effects of persistence in our images. The frequency comb was monitored continuously to ensure the stability of n₀, f_rep, and f_ceo.

An annotated map of U/Ne atlas, created by overlapping adjacent spectral images, can be seen in Figure 3. In each order, the frequency comb spectrum is above the corresponding U/Ne spectrum. Wavelength increases to the left and down. The index of every tenth frequency comb line has been marked, and the frequency of each comb line can be found by using Equation (2), which is also printed on the image for easy reference. The blackbody radiation of the HCL is visible as a faint continuum throughout the U/Ne spectrum; this continuum is not present in the frequency comb spectrum since the comb is not a thermal source. Several bright neon lines have been marked throughout the map.

Zoom In Zoom Out Reset image size

4.1. Data Reduction

4.2. Wavelength Calibration

4.2.1. Wavelength Calibration of the Frequency Comb Spectrum

Wavelength identification of the frequency comb spectrum was conducted by manually identifying the comb line indices (n) specific to a given grating setting and fitting the comb peaks with Gaussians using MPFIT routines (Markwardt 2009) in Interactive Data Language (IDL). Unlike a hollow-cathode emission spectrum, frequency comb lines are evenly spaced and identical in appearance. Because n₀, f_rep, and f_ceo are known, we can determine n by feeding a single fiber with both comb light and a calibration with a well-known spectrum. In our case, since we know the fibers are separated by less than a pixel in dispersion (≈1 GHz), while the comb lines are separated by about 25 pixels (≈25 GHz), we used bright (but not saturated) uranium and neon lines from the adjacent fiber to determine the index of the nearest comb lines and thereby determine the index (and frequency) for all lines in that order. Note that the determination of n requires much less precision (by two orders of magnitude) than our later comparison between the FTS uranium wavelengths and the comb-based uranium wavelengths.

The LFC dispersion solutions were found by fitting the comb line wavelengths as a function of modeled Gaussian centroids with a fourth-order polynomial, ignoring outliers that fell more than three standard deviations from the mean residual to avoid the influence of coincident cosmic rays or stray light from bright lines in adjacent orders. An example comb spectrum can be seen in panel (a) of Figure 4. The residuals of this dispersion solution can be seen in the same figure as diamonds; the scale for the residuals is on the right-hand axis.

Zoom In Zoom Out Reset image size

In the absence of systematic sources of noise, the rms of the frequency comb dispersion solution residuals should decrease as the square root of the signal-to-noise-ratio (S/N) of the lines that were used to solve the dispersion solution. However, we found that the frequency comb rms residuals plateaued at about 0.0004 nm. We hypothesize that this noise floor is induced by a combination of modal noise and detector inhomogeneities.

Fiber modal noise (Baudrand & Walker 2001) is an inevitable result of the finite number of modes propagating in the fiber and leads to additional noise in all fiber-fed spectrographs. The modal noise in NIR fiber-fed instruments is exacerbated (compared to the visible) because the longer NIR wavelengths lead to fewer modes in the optical fiber. In spite of mitigation efforts, such as putting the LFC light through the integrating sphere and fiber agitation, we discovered that the primary and secondary calibration fibers were still impacted by modal noise. While the effects of this were mitigated for on-sky observations by averaging together many cumulative exposures, the modal noise does affect the U/Ne and LFC calibration experiment at the level of a few tens of m s⁻¹ and sets a floor on the achievable precision. Better agitation and scrambling mechanisms are expected to mitigate this problem.

The uncertainties of the frequency comb lines are the sum in quadrature of two components: the statistical uncertainty of the line (as measured by the width, W_{fcomb, i}, and S/N, S/N_{fcomb, i}, of the model Gaussian), and the uncertainty due to modal noise and detector inhomogeneities:

$$\begin{equation} \delta _{{\rm fcomb}, i} = \sqrt{\left(\frac{W_{{\rm fcomb}, i}}{{\rm S/N}_{{\rm fcomb}, i}}\right)^2 + (0.000 4 \,{\rm nm})^2}. \end{equation} \tag{ 3 }$$

Technically, we should also include the uncertainty in the frequency comb wavelength, but this is several orders of magnitude smaller than the other uncertainties and therefore negligible for our purposes.

4.2.2. Wavelength Calibration of the Uranium–Neon Spectrum

The measurement of frequency comb centroids and the calculation of the dispersion solution from their corresponding frequencies are simple tasks compared with the wavelength calibration of a hollow-cathode spectrum because the latter spectrum contains blended lines. The dispersion solution of U/Ne was first estimated by offsetting the frequency comb dispersion solution of the same grating order by 0.005 nm, the approximate separation between the fibers. This separation varies across the order slightly and is different for each order and for each grating setting, so the dispersion solutions had to be solved using the U/Ne lines alone. In an effort to derive the most accurate and precise dispersion solutions possible, we solved our U/Ne dispersion solutions in a consistent and automated manner. The following process is applicable to the measurement of any hollow-cathode dispersion solution.

Each U/Ne spectrum was modeled as the sum of Gaussians, one for each line in the uranium and neon line lists of R11 and Saloman & Sansonetti (2004). For the uranium lines, we used the relative intensities of R11 to guide our initial guesses. We also added a constant offset to the model for the continuum. This model spectrum was iteratively modified to fit the observed spectrum using MPFIT (Markwardt 2009). We then restricted the dispersion solution to a careful selection of lines. First, we neglected to use any neon lines for our dispersion solutions. Some of the gas lines of argon are susceptible to pressure shifts (Whaling et al. 2002); while it is not known whether or not any of the neon energy levels are susceptible to these shifts, we thought it prudent to ignore these lines from the beginning. Second, we restricted ourselves to those lines with an S/N above 100. We lowered this threshold if fewer than 10 lines per order had an S/N above 100, but the threshold was usually above 40. Third, we used the modeled Gaussian goodness of fit to guide our selection. Lines that did not appear in our observed spectra had modeled widths of zero, and lines that were poorly modeled exhibited relatively large or noisy residuals. In an effort to avoid the influence of heavily blended and poorly modeled lines, we calculated the residuals as the difference between the observed spectrum and the modeled line (neglecting the other modeled lines). If either the mean difference or standard deviation in the difference of the three closest pixels was larger than 5% of the line peak, the line was not used to solve the dispersion solution. This threshold was increased in the event that there were fewer than 10 usable lines, but it was usually below 8%. The uranium lines that met these criteria accounted for approximately 20% of the uranium lines from R11 in this spectral range and are marked throughout the atlas with an asterisk ("*").

An example uranium spectrum and model (in gray; red in the online version) are shown in panel (b) of Figure 4. Using the above-mentioned selection of uranium lines, we solved the dispersion solutions by fitting a fourth-order polynomial to the uranium wavelengths as a function of the modeled line centroids. The residuals to the dispersion solution are shown as diamonds. The uncertainties of the dispersion solution residuals are the sum in quadrature of three components: the statistical uncertainty of the model line, the uncertainty due to modal noise and detector inhomogeneities, and the uncertainty of the wavelength of the line, as measured in R11:

$$\begin{equation} \delta _{{\rm U/Ne}, i} = \sqrt{\left(\frac{W_{{\rm U/Ne}, i}}{{\rm S/N}_{{\rm U/Ne}, i}}\right)^2 + (0.000 4 \,{\rm nm})^2 + \delta _{{\rm R}11_i}^2}, \end{equation} \tag{ 4 }$$

where W_{U/Ne, i} is the width of the model line and S/N_{U/Ne, i} is the S/N of the model line. The residuals from the fit are shown in panel (c) of Figure 4.

The extracted U/Ne H-band atlas is available online; an example of this atlas is presented in Figure 5. The wavelengths and uncertainties of the uranium and neon lines are taken from R11 and Saloman & Sansonetti (2004), respectively. Lines of an unknown origin (marked with a "?") are also from R11 and are most likely uranium lines with unknown upper energy levels. The flux has been normalized to the brightest feature in the spectrum, the (saturated) neon line at 1523.48765 nm. An electronic version of this atlas and the wavelength-calibrated spectra used to create it are available as supplementary material in the online journal. This atlas provides astronomers with a useful visual reference for deciding on which spectral region to use for calibration, as well as a means of confirming the observed spectral region.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

View figure set (62 panels)

Tarball of figure set images

There are many features of note throughout the atlas, and we highlight a few of them here.

• 1.  
  There is a uranium line at 1511.05233 nm that falls within 0.006 nm of a known neon line. The uranium line is based on its existence in U/Ar spectra. Based on the relative photon fluxes from R11 and the fluxes of the nearby uranium lines, both lines contribute an approximately equal amount of flux to the feature.
• 2.  
  Some of the many blended lines are readily apparent throughout the spectrum and confirmed by both the FTS spectra and the Pathfinder spectrum (e.g., the pair of uranium lines at 1634.1 nm).
• 3.  
  There are several unidentified lines throughout the spectrum, such as the line near ≈1555.1 nm. These lines are present neither in the uranium line list of R11 nor in the neon line list of Saloman & Sansonetti (2004). As such, they are not marked in the atlas but are apparent with a visual inspection. Others include the lines at 1578.6 nm and 1480.6 nm. These might be unknown uranium or neon lines, but they may also be due to contamination in our lamp. No known lines of any element were found at these locations in the Atomic Spectral Database (Ralchenko et al. 2011).
• 4.  
  Sharp artificial features sometimes appear in the spectrum when bright lines near the edge of the detector shine scattered light onto nearby grating settings. These illumination patterns produce discontinuities in the spectral averaging algorithm; an example can be seen near the neon line at 1493.38863 nm.

5.1. Comparison between the Wavelength Calibrations

A comparison between the precision of the LFC and U/Ne hollow-cathode dispersion solutions is shown in the top two panels of Figure 6. The standard deviation of the LFC residuals was 0.0008 nm, while the standard deviation of the U/Ne hollow-cathode residuals was 0.0012 nm. The former was limited by modal noise and detector imperfections, while the latter was limited by blended lines in the spectrum, which offset the modeled centroids in a systematic fashion.

Zoom In Zoom Out Reset image size

The limitation induced by modal noise was largely unexpected as we were agitating the output end of the fiber, which is known to reduce modal noise in the visible and very NIR (Baudrand & Walker 2001). We suspect that this modal noise was caused primarily by the way in which we fed light into the fibers: the optical alignment of the calibration system was done by eye (e.g., in the optical), and it is possible that different indices of refraction from the beam splitter caused a slight offset in the position of the H-band light relative to the optical. By injecting fewer modes into the fiber, we diminished the number of modes at the output. We anticipate that we can reduce modal noise in the H band in future Pathfinder experiments by aligning our fibers in the NIR and using a double scrambler.

The uranium HCL is dense with emission lines, especially in the NIR. This density is a double-edged sword as it provides many potential calibration lines throughout the NIR, but also a large number of blends, many of which cannot be used for precise RV measurements because they obfuscate the true dispersion solution. The precision of a given line depends on the precise nature of the blend—two equally bright lines will induce more of a shift in their measured line centers than will a blend where one line is significantly weaker than the other. This is a significant advantage that a frequency comb has over hollow-cathode emission lamp sources, since blends in the former are completely avoidable if the comb lines are optimally filtered.

Unlike measuring the precision of a wavelength calibration source, measuring its accuracy requires a comparison to wavelengths to an independent standard. This is particularly important in the case of a hollow-cathode spectrum, where the observer does not always know if there are unknown, weak lines introducing systematic errors into the dispersion solution. This is undoubtedly the reason that the thorium–argon lamps on HARPS failed to reveal the underlying detector stitching that was so readily seen when the detector was illuminated by an LFC (Wilken et al. 2010).

In an effort to estimate the accuracy of our U/Ne dispersion solutions, we measured the wavelength difference between the calibration fibers by feeding them simultaneously with frequency comb light. The dispersion solution differences between these fibers for each of the three different orders are shown in Figure 7 (solid curves). Also plotted in this figure is the difference between the frequency comb and the U/Ne dispersion solution at the same grating setting (dashed curves). In almost all cases, the U/Ne dispersion solution matches the frequency comb dispersion solution to within the uncertainties of the uranium lines. Unfortunately, we could not perform this analysis at every grating angle for the lack of time, but the bottom panel of Figure 6 shows the difference between U/Ne and frequency comb dispersion solutions throughout the H band. The differences in the three orders of Figure 7 are highlighted in gray (green in the online version). Across the H band, the dispersion solutions exhibit approximately the expected fiber separation.

Zoom In Zoom Out Reset image size

It is possible that one could achieve better precisions with a U/Ne hollow-cathode source through a more careful selection of emission lines, but this experiment provides a unique opportunity to assess both the accuracy and precision of the chosen lines, which would not be possible without the LFC (or another highly accurate and highly precise fiducial). Experiments at lower precision should use a more conservative selection of lines, since blending will be more frequent, while experiments at a higher precision should be able to use a larger selection of uranium lines.

Uranium HCLs have a significantly larger number of usable bright lines than thorium HCLs in the NIR Y, J, and H bands. This high density of lines enables precise wavelength calibration for the vast majority of NIR spectroscopic applications. This atlas of U/Ne significantly reduces the barriers to using such lamps in modern NIR spectrographs by providing an easy-to-use visual reference. In addition, the identification of the least-blended uranium lines enables the user to avoid regions and lines that are unsuitable. This work also serves to illustrate the immense promise and advantage of LFCs for precision astrophysics. While frequency combs offer the best solution for ultra-precise wavelength calibration, their cost and complexity currently limit their applications to astronomical instruments that require an extremely high level of wavelength calibration precision over multi-year baselines. Relatively inexpensive HCLs are the most cost-effective approach for the majority of astronomical spectrographs.

We acknowledge support from NASA through the NAI and Origins grant NNX09AB34G, and the NSF grants AST-0906034, AST-1006676, AST-0907732, and ASF-1126413. This research was partly performed while S. Redman held a National Research Council Research Associateship Award at NIST, and while F. Quinlan was supported by a National Research Council NAS postdoctoral fellowship. We thank M. Hirano of Sumitomo Electric Industries for use of the nonlinear fiber in the frequency comb. This work was partially supported by funding from the Center for Exoplanets and Habitable Worlds, which in turn is supported by the Pennsylvania State University, the Eberly College of Science, and the Pennsylvania Space Grant Consortium. The Hobby–Eberly Telescope (HET) is a joint project of the University of Texas at Austin, the Pennsylvania State University, Stanford University, Ludwig-Maximilians-Universität München, and Georg-August-Universität Göttingen. The HET is named in honor of its principal benefactors, William P. Hobby and Robert E. Eberly.

We thank the HET resident astronomers (John Caldwell, Steve Odewahn, Sergey Rostopchin, and Matthew Shetrone) and telescope operators (Frank Deglman, Vicki Riley, Eusebio "Chevo" Terrazas, and Amy Westfall) for their expertise and support. We also thank the HET engineers and staff who provided us with the necessary assistance during the day: Edmundo Balderrama, Randy Bryant, George Damm, James Fowler, Herman Kriel, Leo Lavender, Jerry Martin, Debbie Murphy, Robert Poenisch, Logan Schoolcraft, and Michael Ward.