Introduction

The CS (shortened C) configuration was first suggested by Robert Braun (1993) as a means of filling in the central hole in the uv-coverage of C configuration. The goal was to reduce the short-spacing problem and allow single-configuration imaging for the many projects (primarily extragalactic HI experiments) which require relatively high resolution observations of extended sources. Holdaway's (1994) noiseless simulations showed that Braun's CS configuration led to significantly better imaging than C configuration alone for a complicated source (Cas A, scaled to different sizes), particularly for relatively short observing times ( $\pm2\rm\,hours$) and relatively large sources ( $\stackrel{\textstyle >}{\sim}8\rm\,arcmin$ at 1.4 GHz). This led to a practical test of the new configuration, and Rupen's (1997) observations of the HI in a large, face-on spiral galaxy showed that CS configuration was indeed extremely useful in imaging extended, complex emission, in this case of fairly low signal-to-noise ratio (SNR). Encouraged by these results, NRAO asked that observers proposing for C configuration in 1997 indicate their preference for either C or the new CS; roughly a third of the proposals specifically requested CS, while only a handful expressed a preference for C.¹ In the end roughly half the time was spent in C and half in the CS configuration. So far no complaints have surfaced concerning the new arrangement of antennas, and a number of projects which would formerly have required dual-configuration imaging have been carried out successfully with CS configuration alone.

Given the obvious benefits of this new configuration, the next question is whether we ought to abandon C configuration entirely. The practical advantages of this are, first, that a couple antenna moves would be avoided; and second, that there would be no need to schedule an additional configuration. Since moving from C to CS requires moving only two antennas a relatively short distance, that is at worst a minor inconvenience. Scheduling presents more difficulties, since one has to weigh the relative benefits of C and CS for each project, decide how much total time to spend in each configuration, ensure that the full 24 hours are used efficiently each day, etc. But is CS always better than C configuration? Would any projects be seriously hurt by replacing C configuration entirely by CS?

These are very difficult questions to answer. CS is C configuration with two antennas moved from midway along the east and west arms into the center (Figure ), effectively trading intermediate for short baselines (e.g., Figure ). The advantages of adding short baselines are obvious and easy to quantify - the array becomes more sensitive to large-scale structures, and deconvolution algorithms can more readily extrapolate inwards to find the total (zero spacing) flux. Both Holdaway (1994) and Rupen (1997) found basically what they expected, that CS does a much better job for large sources, and seems to have only positive effects in the two extremes they tested, of noiseless and low SNR imaging. However, it must be true that the missing intermediate spacings make it more difficult to recover intermediate-scale structures; and it must also be true that, at some stage and for some sources, throwing away those intermediate baselines causes problems too severe for existing deconvolution routines to repair.

The purpose of this memorandum is to examine quantitatively the possible deleterious effects of shifting from C to CS configuration. I have taken a `worst case' approach, comparing the two configurations for imaging problems for which CS' additional short spacings probably don't matter, while its missing intermediate spacings may lead to real problems:

• Spectral index mapping: The four standard VLA configurations are scaled versions of each other, in principle allowing the observation of a source with identical uv-coverage at similarly-scaled frequencies. This is presumably the best of all possible worlds for spectral index mapping, and trading C for CS configuration would lose this pleasing symmetry. However, this perfect match is not really very important, and is seldom achieved even now in practice. First, the images are always deconvolved before the comparison is made, and nonlinearities in that process far outweigh small differences in the uv-coverage, especially since source never look precisely the same at different wavelengths (which is of course why we observe them!). Second, there are always differences between the actual uv-coverage obtained in the two configurations: one seldom observes over the identical LST range, and one or more antennas are often out (or coming and going) during one or both observations. Precisely identical coverage is therefore neither terribly important, nor in practice possible to achieve.
• Snapshot imaging: Long observations using CS configuration will tend to reduce the holes in uv-coverage caused by the missing intermediate baselines, as will data taken at low elevations where projection effects increase the density of the sampling of the uv-plane. Snapshots at high elevation are therefore likely to be among the observations most severely affected by a switch from C to CS configuration.
• Very detailed mapping: At the other extreme, one might worry that very high SNR imaging using long integrations would suffer significantly, since certain parts of the uv-plane would simply not be covered by CS where they were by C configuration: some holes at intermediate baselines would never be filled in, even by a fairly long synthesis. For very large sources this effect might be masked by the benefits of CS' short spacings (although high SNR imaging would in that case likely involve D configuration anyhow), but for small or moderately extended objects this might be a significant problem.
• Multi-configuration studies: A similar situation would obtain when combining data from several configurations. In particular, while combining A with C configuration gives excellent uv-coverage over the entire range of baselines, CS would leave fairly large gaps, especially since one would normally observe for a much shorter time in the smaller configuration. One might have to throw in B configuration - for a much longer time than C - to fill in the intermediate-spacing gaps.

All of these potential problems are exacerbated by the fact that one frequently loses several antennas to maintenance or instrument failure. Currently the VLA has a ``three-antenna'' rule, where technicians can be called out at any time only if more than three antennas are down (out of the array). Since this does not count the occasional loss of an additional antenna for single-dish VLBI, one can lose up to four antennas from the array without strong measures being taken; with two more antennas moved in from C to CS configuration,² one might lose up to six antennas from the standard C configuration, and the resulting imaging problems might confidently be expected to be severe. In the next sections I compare the uv-coverage and (through simulations) the imaging quality of C and CS configurations, both with all antennas operational and with two missing (one of which is chosen to highlight any problems with CS configuration).