Plate Parallelism Procedure

If the finesse of your instrumental profile is less than about 33 -- assuming you are not looking at a blend -- the plate parallelism needs to be tweaked. Watch out for blends since there are more blends than not in typical TTF-style spectra. The finesse is simply the separation between two emission lines spaced by a free-spectral-range (i.e. the same line at two different orders) divided by the FWHM of the lines.

Here we give a rapid procedure for establishing plate parallelism.

The procedure can be explained rather succinctly. We segment the pupil with quadrant masks and obtain shuffle arcs through each of the quadrants. If the arc spectrum has the same phase through each quadrant, then the plates are deemed parallel.

With the current control system, the parallelism check is a little more cumbersome but fun all the same. The user obtains a spectrum through each pupil quadrant and compares the relative phases. Three quadrants are enough in principle, but I tend to take four quadrant arcs for subtle reasons to do with bending modes (Ennos 1960).


It is assumed that your support astronomer has already dialled up the optimal CS100 settings for rough balancing of the plates. Now you can tweak this analogue setting using the software variables (X_s,Y_s) under setup / change_cs100 in the SMS window. Step around a list of settings like (0,0), (0,30), (30,0), (30,30), (-30,0) and so on looking for which pair give you the best phase match between the four quadrants.

The procedure is to obtain arc spectra for all four settings of the quadrant mask in the usual way. This means either getting a sausage cube or an 80-shuffle image through all four settings of the quadrant masks in the pupil wheel (PUPIL=4,5,6,7). Simply observe through each of the pupil positions in turn over the same Z-steps, all else remaining unchanged.

Sausage cube approach.

This is the recommended method. A rapid approach to paralleiism is to pick out a single bright spectral line and do a sausage cube through the line. If you keep to 10 steps, you end up with a single sausage cube with 40 little images for the 4 quadrants, so that you can fit all four peaks in one go in the usual way. You need now simply compare the fit centroids (modulo 10).

You will need the big hole in the aperture wheel, the TTF in the beam, and any blocking filter that gives you nice bright lines. Don't forget to insert the small CCD window mitll_on_off.

In the example below, we observed the Ne lamp with the BTTF (Zc=0, Zf=0) obtaining 20 little images  for each quadrant. The stacked sausage has 80 stacked images revealing 4 lines which should be evenly spaced and be identical in character. (The uneven illumination of the lamps means the lines are not all of equal strength.)

The first example shows the plates are not very parallel. Note the lines have different widths and are not very symmetric. Here are the lorentzian fit parameters from splot.  Note that the centroids range from 5.05 to 7.46 (modulo 20). Since the stepping is dz = 6, this range is bigger than the bandpass which should be about 10 (= 6 x lfwhm).

    center        cont           flux         eqw      core     gfwhm   lfwhm
    7.45986  170.737     5616.25    -32.89   2733.95        0.     1.308
  26.35128  159.3639   9070.13    -56.91   2434.59        0.     2.372
  45.05762  148.1022   3315.03    -22.38   870.689        0.     2.424
  66.48528  135.2022   3619.64    -26.77   1172.68        0.     1.965

Here is a much better solution for parallelism for the same set up. Note that the centroid spread is now an order of magnitude smaller, the line widths are the same, and the lines are more symmetric and more lorentzian.

    center        cont           flux         eqw      core     gfwhm   lfwhm

    8.57212  161.445     5911.42    -36.62   2389.34        0.      1.575
  28.52686  152.7101   8804.69    -57.66   3375.19        0.      1.661
  48.45477  143.9869   3232.04    -22.45   1339.54        0.      1.536
  68.57748  135.1785   3618.91    -26.77   1406.8          0.      1.638


80-shuffle slit approach.

In the shuffle arc below, the spatial direction runs horizontally, and the dispersion vertically. What are seen are copper argon lines at high resolution. A fairly good parallel setting is shown in (a), compared to a terrible setting in (b). By tweaking (X_s,Y_s), you should be able to achieve (c). Once again, do not be put off by the uneven illumination clearly seen in (a) and (c). This arises from poor diffusion of the calibration lamps in the chimney.

You will need the big slit in the aperture wheel, the TTF in the beam, and any blocking filter that gives you nice bright lines. Don't forget to insert the CCD window MITLL_SHUFFLE80. Now obtain a shuffle arc for each quadrant in the usual way.

Which direction do I shift in X_s and/or Y_s if the spectra are not in phase? Check that the quadrants are in the correct slots as illustrated here. The X axis is defined by NE and SW, the Y axis is perpendicular to this. If the NE/SW quadrant looks off, try adjusting X in steps of 10. Positive X pushes the plates further apart which will shift the spectrum redwards. How to equate precisely the correction to the movement is presently difficult because our chimney lamps are so poorly diffused.