The modspec mounted on the MDM 2.4 meter telescope. The blue dewar contains the "Templeton" detector, which is mounted here on the spectrograph's "Long port".
Introduction
The MDM Modular Spectrograph (called modspec) is a near-copy of a spectrograph originally designed for the Las Campanas 2.5-m telescope. It was built in the early 1990s, mostly under the direction of Paul Schechter and Peter Mack.
There is a manual that was written when the spectrograph was new. The manual has much useful detail, but unfortunately does not include practical information about spectral resolution, coverage, and so on, that are needed for planning observations. There is also a list of available grating and grism information available here.
I am hoping that the present document will remedy the shortcomings of the original manual and create a wider user base for this instrument.
Modspec has mostly been used to produce intermediate-resolution spectra that give good radial velocities.
Modspec has no electrically-driven mechanisms -- the usual modus operandi is to lock down a single spectrograph configuration for a whole run. Getting everything into good adjustment takes a little care, but once it's there, it's very solid.
Excellent acquisition and guiding. Modspec is similar to CCDS in that it has a camera that looks at light reflected off the slit jaws, so you can be positive that your target is in the slit. The slit-viewing Andor cameras have very high QE, excellent cooling systems, and can deliver up to 8 full frames per second, which makes it easy to center bright objects. By integrating one can see almost any target that's bright enough for spectroscopy and drop it into the slit. The Andor cameras are much more capable than the SBIG cameras used with CCDS. An entirely separate autoguiding system holds the telescope in place once you have the target in the slit. Target acquisition with modspec is much faster and more positive than with OSMOS (which cannot have a slit viewer because of its design).
Blue limitations. Modspec's collimator is a lens, and it uses commercial camera lenses to project the spectrum onto the detector. Because of the lenses, modspec's throughput and resolution degrade short of 4800 A or so, though it still gives information down to at least 4300. If you need good blue response, CCDS may be your best bet, but in the red and mid-visible modspec can be very good.
Spectral range limitations. When modspec was designed, an 800x800 TI chip was considered pretty big, so its unobstructed spectral range is fairly small, and large detectors vignette pretty badly toward the edges. There is still information toward the ends, but with degraded signal-to-noise. Despite this, the number of resolution elements along the spectrum is typically significantly greater than with CCDS.
Configurability. Modspec can be configured over a rather wide range of dispersions and so on -- the manual gives more detail -- but it has almost always been used as a standard grating spectrograph using the "long" port. See Commonly Used Setups below.
Works on either telescope. The modspec is fairly heavy, but the 1.3m can easily accomodate it. The acquisition and guiding system on the 1.3m is nearly identical to that at the 2.4m.
Throughput. Jules Halpern of Columbia measured the throughput some years ago by opening up the slit and observing a standard. With the 600 line grating and the Echelle chip on the long port, he found a peak efficiency (telescope + spectrograph + detector) of around 20 per cent. I measured a lower figure on the 1.3m in 2013 November -- around 8 per cent, including the atmosphere -- but did not open the slit jaws (and I'm also not sure I trust my calculations at this point).
Odd continua. For reasons I have never been able to explain, many spectra from modpsec have oddly-shaped continua. These effects do tend to average out over many exposures, but if spectrophotometry is critical to your science, this is not the best instrument to use -- its strong suit is excellent wavelength stability, not spectrophotometry.
Detector options. At present, only the older "facility" CCDs can be used on
modspec -- these are the ones controlled by the
Owl program.
Briefly, they are:
Although modspec has a fairly wide range of configurations, only a few setups have been commonly used. All are on the long port.
Exploring other setups. There is an old C-language program, modset, which generates numbers (dispersions, coverage, etc) for various setups with the gratings. Here is the source code. To compile the code on a linux system, this should work:
gcc modset.c -lm -o modsetOnce it's compiled, type
./modsetand answer the questions. Note that the scale along the slit is correct for the 2.4m. Both the 1.3m and 2.4m are f7.5, so it's simple to scale to the other telescope.
The modspec is more flexible than the commonly-used setups would suggest.
In order to optimize your setup, plan to arrive at the observatory by early afternoon at the latest.
While the adjustments on modspec are not motor-driven, they do have indicators attached, and a previous observer may have used a setup that suits your purposes. This simplifies matters a great deal. The old manual has a list of setups with indicator readings, which you can at least use as a starting point.
Order-sorter? If you're working at wavelengths long enough so that second-order light could be a problem, insert an order-sorter filter. You can pretty much assume that very little light passes shortward of maybe 3800 Angtroms (though I have not tested this myself!) so this only becomes necessary longward of, perhaps, 7600 A or so. A GG495 suppresses everything short of 4950 A very effectively; there's one in the filter wheel at the top of the instrument.
Subarray readout. The CCDs are normally mounted with columns parallel to the dispersion, so the spectrum appears "up and down". All the detectors have more columns than the slit illuminates, so you will want to set up a subarray in Owl to read a limited range of columns and possibly rows. I've also found that a spectrum put in the middle of the guide field falls a bit to the side, so I tend to center the subarray a bit toward higher columns. On my last run I used these settings with Echelle:
Tilt plate. The detector dewar is mounted on an adjustable plate that can be moved up or tilted slightly to align the detector with the focal surface (which will not in general be exactly planar). The position of the tilt plate is shown on three dial indicators, labeled PL1, PL2, and TL. The "P" indicators are for "piston", and "T" is for "tilt". There are several screws used to adjust the position of the tilt plate by acting as "feet". Once you have good adjustment, the plate is clamped into position with four rosette knobs.
The tilt plate is mechanically very cumbersome -- it's difficult to move one adjustment without affecting the others, and it's easy to get into a position where the bolts are blocking each other. You just have to do the best you can. For guidance, see sections 3.3 and 3.4 of the original manual.
If you have PL1, PL2, and TL readings from an existing setup, tell the staff and they'll wrassle the plate into position and lock it. But if you don't, prepare to spend some time getting the values; again, see the original manual. Briefly, you set the collimator to its nominal value, and then scan through PL values until the spectrum is in pretty good focus in the middle; then adjust TL to get it in focus across the whole range. Finally, you tune the focus by adjusting the collimator (see below).
Detector rotation. Unfortunately, the only way to align the detector rows with the slit is by trial-and-error. Have a staff member loosen the attachments that hold the dewar, rotate it slightly, and check the location of a comparison line on each end of the slit. Do not attempt this on your own as you could potentially drop the CCD dewar off the spectrograph!!
Loosening the dewar mounting changes the focus, so don't try focusing until it's clamped down again.
Grating tilt. This is straightforward; just release the grating brake using the rosette knob, rotate, and retighten the brake. The tilt angle reads in units of degrees; it should be close to that predicted by modset, but it isn't exact. To reproduce an exact setting -- e.g., putting a certain comparison line on the same pixel of the chip as last time -- use the "grating adjustment factor" and pixel dispersion from the "modset" program to compute a correction, and iterate the tilt. With care you can put a line within a couple of pixels of the desired location.
Slit adjustment. The slit is "biparting", which means that both jaws move. You open or close the slit by turning a knob on the slit assembly. Don't try to close the slit all the way, since you may damage the delicate mechanism. A fine scale ('analog indicator' in the picture) reads the fractional turn, and an odometer-like device shows how many turns the knob has made. At the 2.4m at f7.5, a 1-arcsecond slit corresponds to about 0.96 turn (i.e. 96 counter units).
This sounds easy enough, but because of misuse, the turn counter does not click over when the fine scale goes through zero. And because you can't really close the slit to find the zero, to avoid (further) damage to the slit jaws, you need to do something else to find the zero point.
Here's a procedure for finding the reading at which the slit is closed. Take exposures of a flat lamp with different slit opening values, recording both the "odometer" reading and the dial reading separately. Note where in the dial cycle the "odometer" turns over, and use some consistent recipe to form a single number from the two readings. Measure the amount of light in the test exposures, taking care to account for the bias (the values in the bias strip); IRAF's "imstat" will do this. Finally, make a plot of the average counts (minus bias) as a function of your consistently-determined slit width number. This should form a nice straight line, so you can estimate the slit zero point by seeing where the line crosses the horizontal axis. You can now set the slit width confidently by simply adding the amount of counter units you want to the zero point you've determined.
This is a bit of a rigamarole, but it works well. It's good to do this test once in a while, in case the slit has been disturbed by people honking down on it.
Focus. Fine focus is controlled by the collimator setting, which is read out on a dial on the face of the spectrograph. Ideally the tilt plate should put the focus near the nominal value of 0.568, because then the collimator focal point corresponds to the slit. However, given how painful it is to adjust the tilt plate, this may be off a ways -- I've worked with collimator at, say, 0.640 without obvious problems (but then again I'm doing stellar work, not long-slit for which optical distortions might be more important).
Once your tilt, slit, and so on are set correctly, you can measure the focus by taking comparison lamps at different collimator settings. A change of 30 units (i.e., 30 thousandths of an inch) in the collimator setting produces a perceptible change. I like to back away by 90-120 thousandths from the rough focus, and then step through to a similar offset on the other side. That way you can be sure you've bracketed the correct focus. IRAF has a specfocus task for processing the focus-test data.
Telescope focus. The Andor camera should be focused on the slit jaws, which is also where you want to focus stars. So when stars near the slit (which is tilted, remember) are in focus in the Andor image, the telescope should be in focus.
I have sometimes run into a condition where this isn't the case -- stars in focus in the Andor are out of focus in the spectrograph (even producing "donut slices" along the slit). A sure way to fix this is to get a brightish star in the slit, take a series of exposures at different telescope foci and select the one that gives the narrowest profile for the star along the slit, and then focus the Andor so the star is in focus. Now the Andor and the telescope are at the same focus, and the Andor can be used as a proxy for the telescope focus as it should be.
That last step - focusing the Andor - is a bother, since it involves removing a cover that's held by 6 screws, and then fiddling with the lens barrel. Be sure not to close the f-stop on the lens when you're doing this!
Hopefully, you shouldn't have to do this, since over the last year or so I think I've run this issue to ground. The staff just "dials in" the correct focus on the Andor's lens.
Flat fields. The MIS flat lamps work well until one gets down to the blue end, where they get very weak -- if you're observing there, you'll want to take a great many flats to build up signal-to-noise.
The slit jaws are irregular enough that one sees vertical (i.e., along the dispersion) features in the flat. These appear to divide out well.
If you're using modspec at the 2.4m, you should be aware that the prism used to divert light into RETROCAM does not illuminate the slit uniformly. -- it's too narrow to do so. If you divide MIS flats into your data, you'll create an artificial gradient along the spatial direction. To correct for this, take a few exposures of bright twilight sky and derive a smooth correction to the MIS flat's "tilt", assuming the twilight sky is uniform.
Comparison lamps. Here you're limited to the lamps in the MIS.
There is no easy way to script a double exposure in Owl, but you can simply add two frames together to create a single comparison.
Using the night sky spectrum to track flexure. If precise velocities are important -- and if they're not, why use modspec? -- you must be aware that the spectrograph flexes as it moves around the sky. The classic way to account for this is to take comparison lamps with every new target, but this adds quite a bit of overhead. I have had great success with the 600 line/mm visible setup using the 5577 airglow line as a fiducial marker, and simply shifting the whole wavelength solution to force its velocity to zero. As a check, I take a section of spectrum that contains a complex of night-sky OH bands and measure its apparent velocity by cross-correlating it against a master sky. With the 600 line/mm grating, this test yields mean velocities well below 10 km/s, and dispersions typically smaller than that, provided the twilight is not too bright or the exposures extremely short.
To implement this, then, my procedure is to take comparison lamp spectra at the positions of the first few targets, and then dispense with comparison lamps once the twilight has fallen below three magnitudes, according to JSkyCalc.
Finding targets. The field of view of the slit viewer is small, and you're dealing with two different camera systems -- one for the slit viewer, and one for the autoguider. For general guiding, refer to the autoguiding and acquisition manual. For instructions on how to use the Andor slit viewer, refer to the Andor manual. This last also contains information about how to take time series photometry with the Andor; it's also easy to preserve an image of the slit field of view for your records.
Quick-look reduction. I've written a quick-look reduction tool for Modspec and Mark III data (i.e., the spectrographs that take Owl-driven detectors). This lets you see a sky-subtracted, optimally-extracted, and even wavelength-calibrated version of the spectrum you just took, once you've done just a bit of setup.
All the lines from 5852 longward are Neon. Typical residuals to a 5-th order polynomial fit ("legendre 6" in IRAF) are around 0.03 A.