An Upgrade of the ARC Echelle Spectrograph
12.13.01
Donald G. York
First DRAFT
Comments welcome

A. Background

The echelle spectrograph for the ARC 3.5 meter telescope (ARCES) 
was installed in December 1998 and commissioned in the first 
quarter of 1999. Since that time, it has been used by scientists from 
Chicago, Washington, Hopkins and Colorado, that I know of.

The spectrograph provides data at 8 km/sec resolution from 3600 A 
to beyond 10,500 A. The resolution is limited by the pixel size (25 
microns). The throughput is the same as the KPNO echelle, the 
agreed upon performance standard for the commissioning activity. 
The spectra are essentially untrailed, though for some purposes, it is 
useful to move the star along the long dimension of the 1.6x3.2 
arcsec slit. Other slits can be installed, by hand. The spectrograph 
has no moving parts. Images can be autoguided, autotrailed, or hand 
guided, using Remark, the remote interface. The current guider is 
adequate down to 15-16th magnitude V, depending on sky 
brightness, and is limited by the dark current of the CCD, which is 
essentially uncooled. The instrument is permanently mounted and 
easy to operate remotely. The instrument change, from the DIS to 
the echelle, for instance, takes less than 5 minutes.

The spectrograph is very stable. We have a master flat field built up 
from many exposures with an unfiltered quartz lamp in the red and 
a filtered quartz lamp in the blue, which works well from night to 
night. In the first half of the night, wavecals must be recorded at 
least every two hours, but after mid-night, less frequent wavecals 
are sufficient. Over the course of a night, the wavelength shift, 
uncalibrated, is up to a tenth of a pixel, but the orders may move as 
much as 1 pixel perpendicular to the dispersion in the first half of 
the night.

The data require more-than-standard use of IRAF, because of the 
pixel sampling (24 micron) and the closely spaced orders. Using one 
dimensional flat fielding, we routinely are able to achieve S/N of 
3000. The S/N improves as expected for a photon limited detector 
in the high signal regime: that is, spectra can be noiselessly co-
added, providing the signal per pixel is above 200 electrons. The 
orders can be co-added in the overlap regions of the orders to 
produce a blazeless spectrum over the entire spectral range. Beyond 
about 8500 A, the reductions are not so good because of the 
internal reflections in the SITe 2048x2048 CCD and because of 
residual fringing (the illumination of the spectrograph by the flat 
field lamp and the star is slightly different.) The resulting spectra 
show only the wavelength dependent sensitivity of the optics and 
CCD, with little or no echelle ripple.

Operationally, exposures of 30 minutes are ideal for minimizing 
radiation events. For this exposure time, the readout overhead (3 
minutes) is negligible and the data reduction problems are 
minimized. For many problems, longer exposures are possible, of 
course. The effective limiting magnitude is 15-16, depending on 
conditions.

The main limitation of the system is the readout noise of the CCD: 7 
electrons. 

With a suitably low noise detector of similar quantum efficiency to 
the present detector, the spectrograph can reach 19.5 magnitude 
(flat spectrum, V band) in a long night of exposure. The detected 
flux per pixel at 6000 A is 1,000,000 photons per 6 minutes per 
resolution element for a star of V magnitude 6. (_/__ is constant.) 
This is equivalent to 2800 detected photons per second per 
resolution element of 8 km/sec. For a star of 16th V magnitude, the 
flux is 0.28 cts/sec per resolution element, or 1000 photons per 
hour, equivalent to at S/N of 30/1. To go fainter, one can integrate 
longer. To achieve S/N=20 at magnitude 19.5 would require 10 
hours of integration. 

In practice, achieving this performance requires making reasonably 
long exposures to get above the read noise before radiation events 
overcome the spectrum, then co-adding multiple exposures. 

One can also bin on-chip, to go fainter in less time, at whatever 
resolution is desired.

B. Description of Upgrade 

To achieve the limiting performance of the spectrograph requires a 
CCD with lower noise and with less area affected by radiation events 
(cosmic rays, or gamma rays from local sources.) We propose to 
install a 4096x2048 CCD with 15 micron pixels and a read noise of 
about 3 electons per pixel. Studies of the line profiles of the current 
echelle suggest that a resolution of 5.6 km/sec will be achieved 
(optics limited). Since 95% of radiation events are single pixel 
events, only 1/3 as much area will be affected by cosmic rays, so 
longer exposures can be made. For a flux of 40 electrons per hour 
(19.5 magnitude), one would get 80 electrons per resolution 
element. With a read noise of 3 electrons per pixel and 4 pixels per 
resolution element (2 in each direction), the S/N would be 7.4 per 
pixel per two hour exposure, or 17 in 5, 2 hour exposures. For an 
electron noise of 5 electrons, the S/N for the same configuration of 
exposures would be 13.. For an electron noise of 2, the S/N would 
be 19. For a resolution of 10 km/sec (4 pixels), 5, 2 hr. exposures 
would yield a S/N of 24 with a 3 electron read noise per pixel. 

The CCD would be oriented with the edge of the short dimension 
roughly parallel with the echelle orders. The blazeless spectral 
coverage would be 3600 A to 9400 A, limited on the short 
wavelength side by the falling transmission of the cross dispersive 
prisms and on the long side by the falling CCD efficiency and the 
chip geometry. The CCD of choice, determined by Shu-i Wang, Dale 
Sanford and Connie Rockosi after extensive study, is the Marconi 
CCD44-82. The thinned device is on a number of telescopes and 
regularly delivers 2-5 electron read noise. The cost is about $100K. 
A possible alternative is a Lincoln labs, thick chip of the type Chris 
Stubbs plans to put in the mosaic imager at APO, but too little is 
known at present to allow a comparison with the Marconi device. 
The current SITe CCDs are larger, but have a new architecture that 
significantly increases the cost of the electronics for the upgrade, 
and the devices deliver more than 5 electron read noise (with little 
hope of a lower value.) Note that for this many exposures, the 
radiation events can be removed by applying a median filter to the 
spectra.

C. Science Programs

a. Quasar Absorption Lines
( collaborative work between U. Chicago, Colorado, Princeton)

The SDSS will discover tens of thousands of QSO absorption line 
systems in QSOs brighter than 19th magnitude. With such a large 
sample, one can hope to address several key questions of QSO 
absorption line research. One of the most important is the issue of 
what percentage of the systems are intergalactic as opposed to being 
circum QSO material, ejected at very high speeds. From the large 
SDSS sample, it will be possible to pick absorption line systems that 
are far enough away in apparent redshift [z(QSO)-z(QSOALS)>0.5] 
to allow a systematic study of these two possible classes of system 
with the echelle. Current studies indicate that there will be profile 
differences and probably abundance differences that will allow 
discernment of the systems. It is important not only to determine 
the percentage of systems that might be circum QSO but to find a 
way to determine if a particular system is circum QSO or 
intergalactic. The former question will affect many statistical 
inferences drawn from the QSO absorbers, including spatial 
correlation functions and large scale structure, association of 
QSOALS with galaxies and evolution of the elements in łnormal˛ 
galaxies.

Once a set of truly intergalactic absorbers can be picked out, the 
large SDSS sample will enable numerous classical QSOALS studies, 
especially those involving evolution of the elements. These studies 
will aim to observe, say, 100 QSO absorption line systems to track 
the evolution of Si, Fe, Mn and Zn over cosmic time (10 absorber 
systems, far from the QSO redshifts, every 0.4 in z from z=1 to 4. 
Such programs would require a large commitment of telescope 
time.)

b. Gamma Ray Bursts and the Intergalactic Medium
( interests at JHU and Chicago)

Gamma ray bursts are now understood to be at great distances and 
intergalactic absorption lines have been detected in their spectra. 
Early detection of the bursts is now possible with the HETE satellite 
and, for bursts with optical counterparts, with the ROTSE III 
telescope at Los Alamos. The brightest such event in the optical was 
9th magnitude.

Echelle spectra can, in principle, be obtained of the optical burst if 
the events can be detected and located fast enough (a few hours, 
depending on the brightness.) Spectra of these objects would help 
answer at questions about the intergalactic medium, as discussed 
above for QSO absorption lines. They may be visible to very high 
redshifts. They are particularly important for addressing the issue of 
the origin of QSO absorption lines because the circum QSO lines are 
not an issue.

APO is ideally suited for such opportunities because the echelle can 
be made available in 5 minutes, once the schedule observer logs off. 

c. Abundances in Faint Stars (George Wallerstein)
(George sent me these notes, which I include with minor editing.)

During the next few years I will be an active user of the 3.5-m 
echelle. I will soon start a program to derive the composition of 
distant anti-center cepheids to establish the galactic abundance 
gradient of metals, s-process and r-process species. 

In addition a future program on abundances in globular clusters will 
start when I hire my next postdoc, probably in September. This 
program needs to reach to V=15 or 16 which can only be done with 
a CCD with a very low readout noise. Working with me and 
Guillermo Gonzalez, now at Iowa State, Chris Laws will be obtaining 
spectra of numerous stars with planets. His work to date shows that  
the echelle at McDonald with a resolution 60,000 provides 
significantly better data than does ours at 40,000. The main 
difference is the resolution. 

A Brazilian astronomer, Luciana Pompeia, has applied for a Brazilian 
postdoc position to work here on the abundances of Cu and Zn in 
carbon and other types of stars to trace the effects of small neutron 
exposures. The highest resolution is important for that program 
because of the paucity of Cu and Zn lines and the blending of those 
rather weak features with lines of other elements.

Lew Hobbs and Doug Duncan can add more here.

d. Distances to High Velocity Clouds in the Galaxy

There is a growing controversy as to the location of the 21cm high 
velocity clouds (HVCs). As was suggested by Oort in his early work, 
there are a number of possibilities, none of which have been ruled 
out in the 35 years since their discovery. They may be primoridal, 
intra-group clouds that did not form galaxies. They may be material 
associated with the halo of the Milky Way. They may be gas ejected 
from the plane of the Milky Way. Recent studies of abundances with 
FUSE/HST show that the element abundances are consistent with the 
second choice. But such measurements have been made for few 
clouds.

The most obvious way to determine the origins of the clouds is to 
determine their distances. A difficult, but straight forward way to do 
this is to observe stars that are at successively higher distances, in 
the direction of each HVC, until the interstellar absorption lines of 
Na I and/or Ca II show up in the spectra. The stars best suited for 
such studies are the RR Lyr stars, which have a high enough density 
on the sky to offer several options per cloud, in distance steps of 10 
kpc, say. The objects are now easy to find (SDSS, ROTSE III). While 
the optical interstellar lines have been detected in a number of 
clouds, the phenomenology of the Na I lines and Ca II lines is not 
well determined. Thus, a preliminary study must be done using the 
spectra of bright AGNs toward HVCs to establish the frequency of 
occurrence of the interstellar absorption lines. Once this is known, 
and with a large enough sample of HVCs, the distances can be 
determined. The spectra of 30-40 AGNs will be needed to begin the 
study. This many are known to be behind HVCs that are brigher than 
16th magnitude, so, with the upgraded instrument, an observing 
program of 6-8 nights will be adequate for this preliminary study. 
The actual spectra of RR Lyr stars are more difficult to obtain, 
especially at the greater distances. The V magnitudes (18) and the 
deep stellar Ca II lines place these objects at the limit of detection of 
the upgraded ARCES.

e. Distances to Interstellar Clouds (FAME)

The specific location of an individual interstellar cloud is important 
for several reasons. For instance, the chemistry of interstellar clouds 
depends, through cosmic ray ionization of molecular hydrogen, on 
the location of a cloud in relation to the cosmic ray flux that peaks 
in O associations. We may not understand the chemistry in several 
cases until this issue of the location of the cloud compared to 
cosmic ray (and X-ray, for that matter) sources is sorted out, 
particularly for cases with anomalous observations, such as the 
excess 6Li in the gas toward omicron Per. Another example is the 
possibility of accretion of cloud material onto stars as their 
trajectories cross, a phenomenon that can affect surface 
abundances of stars as well as the configuration of the astrospheres 
(analogous to the heliosphere) around stars. Knowing that a cloud is 
near a star represents an important step in understanding the details 
of the cloud and the atmospheres of planets around the star itself.

Using the same technique as described for RR Lyr stars and HVCs, 
the location of interstellar clouds can be determined, using stars in 
the galactic disk. Since the most interesting clouds are of order 1-
10pc in size, the distances to the stars must be known to a parallax 
of 0.1 milliarcsecond or better. The FAME satellite will obtain 
parallaxes with this precision. The interstellar lines can be observed 
in those stars to delimit the cloud distances (by watching lines 
appear (as one increases the distance of the stars) and disappear (as 
one moves laterally away from a cloud with the sampling stars.) 
Such studies have been done for a few clouds, but the number of 
stars available with adequate interstellar spectra and precisely 
known distances is so small as to leave most of the cases ambiguous. 
FAME will make such measurements to well below 12th visual 
magnitude. The higher resolution of the upgraded ARCES is 
important to separate the groups of interstellar components. The 
study will involve 15 minute exposures on thousands of stars. (York 
is a FAME co-Investigator.)

f. Diffuse Interstellar Bands
( Collaborative program between U. Chicago, JHU, Univ. of 
Colorado)

The echelle has been extremely useful in studying the origins of the 
diffuse interstellar bands, a set of 400, diffuse, interstellar 
absorption lines between 4000 A and 10,000 A that have never been 
identified, despite being first seen over 100 years ago, and being 
understood to be interstellar for 80 years. With very high signal to 
noise spectra of reddened stars, we are able to establish strong 
correlations between some lines and the absence of correlations in 
others. Dozens of new lines have been found. Specific candidate 
large molecules have been ruled out. And, we have shown that the 
strongest of the bands appear to arise in material in the cloud 
surfaces as opposed to in the molecular parts of the clouds, where 
the molecular fraction of H2 is high. We hope to establish energy 
level diagrams for the likely molecules (by looking for correlations 
strong enough to indicate a common molecule and by looking for 
fixed spacings that might indicate molecular structure. Based on our 
ongoing program, both of these steps seem realizable.

The 8 km/sec spectra are not adequate to isolate the bands to the 
particular cloud components on a given sight line. The higher 
resolution of the upgraded ARCES will be needed to determine the 
velocities and shapes of some of the weaker bands. In the case of C3, 
now detected by with ARCES in many stars, the band structure 
cannot be resolved without the higher resolving power. Even 
incipient resolution can lead to determinations of temperature and 
other features of the clouds in which the molecule exists. It is 
anticipated that the same will be true for the diffuse bands.

The relevant stars are reddened, so much so that conventional UV 
studies cannot be done with either FUSE or HST. The optical lines of 
K I, Na I and Ca II thus provide the only high resolution identification 
of the interstellar component structure needed to analyze the 
diffuse band profiles. Once we finish our current diffuse band 
survey, dozens of stars will be used for follow up at high resolution. 
The total exposure times needed to reach the high signal to noise 
required are as high as 20 hours. per star.


D. Cost

The team that assembled and mounted the echelle is still together. 
Careful estimates based on the known steps needed have been made. 
The replacement of the CCD itself, with associated fixturing to 
accommodate the different packaging and cooling aspects of the 
CCD, is estimated to cost $350K (450 days of labor, 37 days of 
machining, $40K worth of materials.) The CCD recommended 
Marconi CCD is $100K. Additional work that should be done while 
the structure is apart (including a likely doubling of the throughput 
by AR coating the refractive optics, realigning some optical 
components to improve the resolution and adding a new guider) 
would cost an additional$70K, if we could find the funds to do it. 

I am recommending that ARC provide the CCD and that I propose to 
NSF for the CCD replacement activity. I will talk to the APO 
management about the additional items to see if they can be 
covered internally.