Chiron Data Reductions
September 2017: This software now works on both fiber-mode and slicer-mode spectra.

The standard Chiron data reductions are optimized for the standard Chiron observing program - radial velocity studies of bright stars searching for exoplanets. Radial velocity stability is crucial. For these purposes the reduced data products more than suffice.

Chiron is fiber-fed: there is no sky subtraction capability (except after the fact). The standard reduction does not seem to subtract the instrumental background, which is not a problem for short integrations in bright stars. The flat field is flattened prior to division into the data, retaining the blaze function.

Our needs are different. Novae fade rapidly, and the spectra are often dominated by broad emission lines. We need to be able to subtract the instrumental background (our exposures run up to 30 minutes). We need to be able to flatten the spectra, because when the width of the line is comparable to the free spectral range it can be difficult to determine where the continuum is.

We also discovered that the algorithm employed to reject cosmic rays has the unfortunate side effect of rejecting strong emission lines. This is unacceptable. So I wrote a new reduction scheme in IDL. The salient features include:

Full details and documentation are available here.

IDL procedures and master calibration files are available in these tar files.

Figures from the full writeup follow.

A. Motivation for the new reduction scheme.

Order 10 of achi150728.1127.fits. The target is the nova V5668 Sgr . The emission line is Fe II λ4921. The narrow dropouts are an artifact of the default data reduction process.

What you've been missing. Orders 124-138 in a slicer-mode spectrum of η Car. The extracted data have been flux-calibrated, and the orders have been spliced. The strong broad lines are Hδ and Hγ. Narrow blueshifted wind absorption is visible in both lines. The narrow emission lines are mostly Fe III.

B. Technical Details/Quality Control.

The normalized ratio of the boxcar-extracted to the Gaussian-extracted flat for order 108. The green trace is the scaled width of the Gaussian used to extract the flat. They track well, with the larger ratio (less flux in the Gaussian extraction) when the fit is broader. This occurs when the trace is centered between two pixels. I suspect this is attributable to the non-Gaussianity of the PSF. In any event, the effect seems to divide out.

The ratio of the boxcar-extracted to the Gaussian-extracted spectra, after flat division, for order 108 of chi131014.1143 (μ Col). The RMS scatter of the ratio is 0.6%. The median is 0.999; no normalization has been applied. The structure visible in the flat ratio is greatly reduced. The SNR of the spectrum, from counting statistics alone, peaks about 900, and exceeds 500 between pixels 250 and 850. Within this region the RMS is about 0.4%, comparable to the 0.3% expected from counting statistics alone.

A particularly boring part of the spectrum of the O9.5V star μ Col, a spectrophotometric standard. Note that division by the actual flat extraction yields a very flat spectrum.

C. Wavelength Solution.

The Na D lines in N Oph 2015 on 150728. There are few narrow lines in novae - these are interstellar. The achi spectrum is in aqua. The wavelength solutions are in good agreement.

A comparison of the slicer mode spectrum of σ Ori using my reductions (black) and the spectrum from the achi file (blue) processed using the Yale reduction. This is a portion of order 42 (order 27 in the achi file) containing the Na D1 line. The difference in slope is attributable to fact that I divide by the extracted flat, while the achi spectrum is not flattened. A heliocentric correction is applied to both spectra; fluxes are normalized to the median in the plotted region. The spectra match well; there is a wavelength offset corresponding to 0.4 km/s between the spectra that could arise from many sources.

D. Quality Control: Cosmic Rays.

These panels illustrate the effects, both good and bad, of cosmic ray filtering single images. The lower panel shows how cosmic ray rejection is supposed to work. The black trace is the boxcar-extracted spectrum with cosmic ray detection turned off; the green trace is the same extraction with nominal CR rejection. A narrow positive deviation, seen in black. The upper panel shows what happens when a real narrow feature is mis-identified. The default extraction produced the green trace. Note the absence of the strong interstellar Na D1 line. The magenta trace is the extraction run with CR rejection and without spectral cleaning. Points flagged as bad are set to zero. The flagged points are interpolated over (the green trace). Eschewing CR rejection results in the black spectrum, and a normal-looking pair of Na D lines. This is a fiber-mode spectrum of V339~Del shortly after maximum.

E. Wavelength Solution.

The drift in pixels over the course of a night of the Th-Ar spectra in fiber mode. The drift correlates well with various instrumental temperatures. While the drift is systematic, and is accounted for in the extraction software, it is also insignificant for many purposes. Chiron is stable.

The flux-calibrated slicer spectrum of μ Col (image chi131008.1140).

A comparison of the boxcar extraction (black), Gaussian extraction (dark blue) and the Yale extraction (aqua) for the Hα region of Nova Sco 2015 (image chi150728.1124). The Yale extraction is scaled so the medians match. Note that the central peak of the Hα line is missing in the Yale extractions. It has apparently been flagged as a cosmic ray or some other type of defect. The top of the line is not saturated.

A comparison of the boxcar extraction (black) with the Yale extraction (aqua) for the sodium D line region of Nova Sco 2015 (image chi150729.1124). The Yale extraction is scaled so the medians match. Note that the line strengths are much weaker in the Yale reduction, while the S/N is higher. This suggests that the background has not been subtracted in the Yale reductions. The emission line is He I λ5876. Multiple velocity components are visible in the Na D1 and D2 lines. The low velocity components are galactic foreground; the blue-shifted absorption lines may be ejecta from the nova.

This is the same data as shown in the figure above, except that the data (black) are not flattened, and the background has not been subtracted. This provides a reasonably good match to the Yale reductions (magenta).

Order 67, showing strong emission from the λ8498A component of the Ca II IR triplet. The line is not evident in the Yale reductions. The apparent strengths of the λ8446A O I line are also very different.

Order 66, showing strong emission from the λ8542A component of the Ca II IR triplet. As in the previous figure, the line is not evident in the Yale reductions.

A region dominated by emission lines (mostly Fe III) in η Car. In the lower plot, the black trace is from the new reductions while the magenta trace is the Yale extraction after flux calibration and order splicing. The upper plot shows the ratio of the two. The dotted vertical line shows an order splice. The upper plot shows 3 artifacts. There are wavelength offsets in the 5261A and 5275A lines, but the senses are different, suggesting that one wavelength scale is stretched relative to the other. The positive excursion in the ratio at 5274A is a cosmic ray or hot pixel not flagged in my reduction scheme (it is not designed to work well in highly non-linear regions). The 1% step at the order splice at 5279A is in the new reduction scheme.

A region on the blue side of η Car's broad Na~I emission line. The deep absorption line (presumably due to discrete Na I absorption at -200 km/s in the homunculus) is deeper in my reductions, probably because of the global background subtraction. The narrow peaks between absorption lines (especially near 5881A) are sharper in my extractions

The NaD line region in η Car. The low velocity Na D lines are under-subtracted in the Yale extractions. My reductions return sharper features on the blue sides of the Na D absorption lines. For low contrast lines (e.g., in the 5896-5905A region and between the Na D lines), the two schemes agree to better than 1%.

H. Pushing the Limits

The H-delta λ4100A line in N Sgr 2015b. The other line is probably Fe II λ4128A (multiplet 27). This is as blue as Chiron goes. All 1028 points are plotted; nothing is masked. There is not much sensitivity at the ends of the order.

The Hα region of V745 Sco on 150603. This is the sum of two 1800 second integrations. The R mag at the time of observation was about 16.2. The dotted red vertical lines mark the spliced regions. No background has been subtracted; the data are unsmoothed. The default trims have been expanded by 10 pixels. The high points are due to low counts at the edge of the detector rather than to cosmic rays. The narrow emission is Hα from one of the two stars in the system; the source of the broad emission near λ6585A is not currently known. No sky has been subtracted; much of the continuum and the λ6563A absorption may be from the night sky.

Finally - the end!

Slicer-mode spectra of nova Sct 2017 on 10 nights between 26 July and 7 August 2017, taken during the recommissioning of the spectrograph. Fluxes are normalized to the 6470-6507A continuum. After an initial dip from days 7961 to 7964, the line grew steadily stronger and broader. The fuzz at 6456 and 6517A is the emerging broad and flat-topped Fe II emission. The 6613A absorption is a diffuse interstellar band.