ISIS image slicer

• Performance summary
• Introduction
• Spectral resolution
• Throughput predictions
• Throughput measurements
• Image slicer performance and recommendations
• Observing with the image slicer
• Notes for the data reduction

1. The image slicer is designed to reduce light-loss at the spectrograph entrance slit, and to give the highest possible spectral resolution.
2. We recommend using slicer B because it is more efficient than slicer A by a factor of ~ 1.4-2 over most of the wavelength range.
3. For seeing of about 1 arcsec, image slicer B is as efficient as the long slit with width set to match the seeing. Therefore we recommend using the image slicer in such conditions if optimum spectral resolution is desired.
4. For seeing of about 1 arcsec, the image slicer B is a factor of ~ 1.9-3.3 more efficient than the long slit set to project to 2 pixels (of course giving similar spectral resolution).
5. For seeing of about 1.5 arcsec, the numbers in the previous two points are a factor of ~ 1.5-2.0 lower, thus the actual seeing should be taken into account.
6. With the image slicer one can in principle reach a spectral resolution of 13500 for the R1200R grating at 7000Å, and 20500 for the H2400B grating at 4500Å. However, in practice charge diffusion especially in the blue arm degrades the delivered spectral resolution for wavelengths shorter than ~6000Å by up to a few tens of per cent.
7. With the image slicer each spectrum is collected over more pixels on the CCD than when using the long slit, and this can impact signal-to-noise for faint targets due to the increased CCD readout noise. To slice all 5 slices for image slicer B, one has to add ~ 50 pixels in each row.
8. In order to subtract the sky background properly, sky exposures should be taken at least once during the night, or for each spectrograph configuration used during observations.

The concept of the image slicer as a means of reducing loss of light at the entrance slit of a stellar spectrograph was introduced by Bowen (ApJ 88, 113, 1938). The image slicer rearranges the stellar image into a series of strips of width equal to that of the slit, and arranges them end to end along the slit, thus allowing for higher spectral resolution without slit losses.


The ISIS image slicer viewed from above, where one can see two circular apertures labelled A and B, and the mini-slit of width 0.45 arcsec between them.


The Bowen image slicer for ISIS was built by Tom Gregory for one of the ISIS multi-slit positions, and is used in place of the long slit when deployed. A photo of the image slicer is shown above, where one can see two circular apertures labelled A and B, and the mini-slit of 100 µm (~ 0.45 arcsec) width between them. The apertures A and B each have a diameter of 500 µm (~ 2.26 arcsec), and each has one image slicer below. Two metal plates around apertures A and B are covered with reflective material, so that when acquiring a target the acquisition camera can see the reflected light.

The details of the optical arrangement are shown in the diagram below, taken from and discussed in Pierce (PASP 77, 216, 1965). Once the star is acquired in aperture A or B, the light encompassed by the 2.26 arcsec circular aperture travels down towards the small mirror (M in the picture below) and is reflected to the slicer (S in the picture below), from where it is reflected once more to the slit below, producing six and five slices on the detector for slicers A and B respectively. The slit below slicers A and B also has a width of 0.45 arcsec and is just a continuation of the mini-slit, so physically it is one slit below and between apertures A and B.

The details of the optical arrangement of the image slicer. Once the star is acquired, the light encompassed by the circular aperture travels down towards the small mirror M and is reflected to the slicer S, from where it is reflected once more to the slit below.


The image slicer A has a broad-band, aluminium coated surface to cover the spectral range from 310 nm to the near infrared. The image slicer B has a wide-band dielectric multi-layer coating to enhance the throughput within the region of 390 to 850 nm.


Spectra from image slicer B (top) and A (bottom) taken on the night of December 25 2012 with the R1200R grating and the Red+ detector in the red arm of ISIS. The sky lines can be clearly seen. Also, it can be seen that slicer B has 5 slices, whereas slicer A has 6 slices.
The image-slicer mini slit, width 0.45-arcsec, is strictly speaking only fully sampled by the ISIS detectors for the R158B, R300B and R600B gratings. The tables below list the expected spectral resolutions, with the spectrograph optimally focused and with the image slicer deployed, at wavelengths of 4500Å in the blue arm and 7000Å in the red arm. A two-pixel resolution element is assumed for those gratings where the image-slicer slit is undersampled (projects to less than two pixels). For comparison, the expected spectral resolutions with the long-slit deployed and set to a width of 1.0-arcsec are also listed for each grating.

ISIS red arm spectral resolution at 7000Å (Red+)
Slit configuration	R158R	R316R	R600R	R1200R
Image slicer	1934	3763	7143	13462
Long slit 1 arcsec	909	1844	3857	9288

ISIS blue arm spectral resolution at 4500 Å (EEV12)
Slit configuration	R158B	R300B	R600B	R1200B	H2400B
Image slicer	1296	2442	5000	9783	20455
Long slit 1 arcsec	576	1099	2225	5283	12682

An effect which is known to degrade spectral resolution for wavelengths <6000Å, and therefore will most strongly affect the resolution performance of the imager slicer in the blue arm, is charge diffusion in the detector. As the wavelength of incident light decreases, absorption of photons occurs closer to the surface of the CCD where the pixel potential wells are less-well defined. Therefore photoelectrons generated there have a greater tendency to diffuse into the wells of neighbouring pixels, degrading resolution in both the spectral and spatial directions.

This effect is enhanced as wavelength decreases, and has a proportionally larger effect on the spectral resolution delivered with the smallest slits. It has been measured on the EEV12 detector to broaden lines for 2-pixel and 4-pixel long-slit projected widths by 20% (to 2.4 pixels) and by 10% (to 4.4 pixels) respectively, at a wavelength of 4000Å. Therefore, the expected spectral resolution at 4000Å wth the R600B grating and image slicer deployed would be reduced from 4444 to 3700 by charge diffusion.

Measurements of the FWHM of unresolved lines in CuAr and CuNe calibration lamp spectra taken with carefully focused R158R and R158B gratings are consistent with the expected spectral resolutions.

To estimate the throughput of the image slicer we calculate expected light losses caused by the following factors. Firstly, both slicers have a window which protects the aperture from dust, condensation etc. This window is BK-7 glass with an MgF anti-reflection coating. Light loss is about 2% when the window is clean. Secondly, both slicers have a double reflection of the beam before it reaches the slit, and estimated reflection losses in the two reflections are 4% and 26% for slicers B and A, respectively. Lastly, both slicers suffer from slit losses due to the de-focus of the beam, and this is the greatest source of loss. Every millimeter that the beam travels along the slicer mirrors produces a de-focus of ~91µm (the slit width is only 100µm). Taking into account the physical size of each slicer, the slit losses due to the de-focus of the light beam are estimated as 40% and 44% for slicers B and A, respectively.

Finally, we estimate the total light losses to be 46% and 72% for slicers B and A, respectively, assuming the entrance window protecting the apertures is clean. Therefore, the total throughput is estimated to be 54% and 28% for the slicers B and A, respectively.

The data from three commissioning nights have been analysed to assess the throughput of the image slicer. Each slicer image was compared with several long slit exposures:
1) long slit with an 8 arcsec slit width to measure the throughput of the slicer versus total throughput of ISIS. Here we assume that the 8 arcsec slit width encompasses all of the light from the point source, taking into account that seeing was always <2 arcsec throughout all three nights.
2) long slit with 0.4 arcsec slit width to measure the throughput of the slicer versus throughput of the long slit image with a similar spectral resolution.
3) long slit with a slit width matching the seeing, to measure at which seeing it is beneficial to use the image slicer rather than the long slit.

As an example we show the results from the night of 14 October 2013, when we used the R158B grating with a central wavelength of 4500Å in the blue arm, and the R158R grating with a central wavelength of 7500Å and a GG495 blocking filter in the red arm, and the 5300 dichroic. The extracted spectra for slicer B, slicer A, long slit of 8 arcsec, long slit of 0.4 arcsec and long slit matched to the seeing can be seen at the following links for the blue and red arms. The corresponding ratios of the extracted spectra can be seen at the following links for the blue and red arms.

1. Slicer B is more efficient than slicer A by a factor of ~ 1.4-2 over most of the wavelength range; below ~ 3550 and above ~ 9000 Å, slicer A becomes more efficient than slicer B.
2. For seeing of about 1 arcsec, image slicer B (A) is a factor of ~0.47 (0.3) as efficient as a wide long slit of slit width 8 arcsec. This agrees quite well with the total expected throughput as estimated earlier.
3. For seeing of about 1 arcsec, image slicer B (A) is a factor of ~0.95 (0.6) as efficient as a long slit of slit width matching the seeing.
4. For seeing of about 1 arcsec, image slicer B (A) is a factor of ~3.3 (2) as efficient as a long slit of slit width 0.4 arcsec projecting to 2 pixels, giving similar spectral resolution.
5. For seeing of about 1.5 arcsec, the numbers in the previous three points are a factor of ~1.5-2.0 lower.

The recommendation is to use image slicer B rather than slicer A. Also, it seems that using the image slicer brings benefits only if the highest spectral resolutions possible with ISIS are needed. The seeing must be also reasonably good; it appears that with seeing of 1.5 arcsec and slicer B deployed, one is already losing more than half of the light collected by a long slit matching the seeing, and about 75% of the light collected by a long slit opened to 8 arcsec. With seeing of 1.5 arcsec this is a higher light loss than expected by about 25 % (throughput models for 2 arcsec circular aperture and light centered in the middle provided by Marc Balcells). This probably means that the star was not well centered on the circular aperture, or that there is image motion in the aperture. We know that the WHT suffers from a "Cassegrain wobble", seen as oscillations with an amplitude of 0.6 to 1.0 arcsec and a quasi-period of about 35 to 400s. These oscillations appear to be always present, only changing slightly in amplitude and period with time during the night. They do not occur in any particular direction, and auto-guiding does not eliminate them.

One thing to bear in mind when planning to use the image slicer is that each spectrum is collected over a larger number of pixels on the CCD than when using a long slit, and this can limit signal-to-noise for faint targets due to the increased CCD readout noise. Each slicer has a fixed format for displaying individual slices on the CCD, and the width of each slice is not seeing dependent, unlike the case of long slit spectra where the width of the spectrum in the spatial direction is influenced by seeing. Thus, for example, in order to slice all 5 slices for image slicer B, one has to add ~ 50 pixels in each row. In comparison, a typical long slit exposure in 1 arcsec seeing spreads over ~5 pixels. For faint targets this can compromise the delivered signal-to-noise of the image slicer observations.
The image slicer is mounted permanently in one of the slit holder positions. The ISIS Mimic can be used to interchange between the image slicer and the long slit, a procedure which takes less than one minute.

The long slit is tilted at an angle of 7.5 deg to allow viewing of the slit by the acquisition camera. The two plates of the image slicer are tilted by the same amount to permit viewing of the two apertures A and B.

In the acquisition image, the observing support assistant should mark the reference apertures A and B, and then acquire. Light in a calibration unit can be used in order to see labels A and B on the slicer, or you can take an exposure to distinguish between the slicer B (which has 5 slices) and slicer A (which has 6 slices).

The focus of the spectrograph using the image slicer is similar to that when using the normal long slit, thus it is possible to interchange between the image slicer and the long slit without a need to refocus the spectrograph.

During the commissioning nights, there was no need to refocus the slit viewing TV camera used for the acquisition between viewing the image slicer and long slit of ISIS.

The telescope focus for the image slicer and the long slit of ISIS is also very similar and the same value can be used for both.

Dekker clear position is used for both image slicer and long slit observations.

To calibrate properly slicer observations, it is recommended to take internal tungsten lamp flat-fields several times during the night, or at each target position, due to the instrumental flexure.

At least once during the night sky a background exposure free of stellar contamination should be taken with the image slicer. It can be used to determine ratio of sky background in each slice versus sky background in the mini-slit region (see next section for more details).

Wavelength calibration images should be taken independently for image slicer data, as there is a shift in spectral direction compared to the long slit data.

The image slicer is designed to be non-polarising, i.e. <3% according to Tom Gregory in 2010. However, the slicer polarisation has not been measured so far at the telescope.
The image slicer data can be reduced in a standard way, and the only speciality about image slicer data is the sky background removal. As image slicer spectra do not have any proper sky background which can be subtracted from the stellar spectra, the sky background has to be fitted using a mini-slit region between the two apertures A and B. The mini-slit region exaggerates the sky background due to the higher efficiency of the bare slit versus individual slices of the image slicer. For commissioning purposes, sky background removal using a mini-slit region was chosen as a good enough approximation. However, the correct sky background removal should include the amount of sky in each slice of the slicer. The ratio of the sky background in each slice versus that in the mini-slit can be determined from sky exposures taken at least once during the night, or for each spectrograph configuration used during observations. Then, for each spectrum, the sky background in the mini-slit should be corrected by ratios determined from the sky exposures to get proper sky background for each slice and wavelength.

To give an idea about the difference in sky background between each slice and the mini-slit region, we did a comparison for the night of October 14 2013 for the slicer B in the red arm. We subtracted the sky background in each slice and in the mini-slit region, and the final ratios can be seen here. We label as slice 1 the left-most slice in the slicer B as shown on the picture above, and slice 5 the right-most slice. We did not remove cosmic rays and the final ratios have been smoothed over 10 points.