Selection of the Space Interferometry Mission astrometric grid

We investigate the choice of stellar population for use as the astrometric grid for the Space Interferometry Mission (SIM). SIM depends on the astrometric stability of about 2000 stars, the so-called grid, against which the science measures are referenced. Low-metallicity and thus relatively high luminosity K giants are shown to be the population of choice, when available. The alternative, nearby G dwarfs, is shown to be susceptible to unmodeled motions induced by gas-giant planetary companions, should there be a significant population of such companions. Radial velocity filtering is quite efficient in selecting grid members from the K giants with yields exceeding 50% if filtering at 30 m s(-1) (1 sigma) is available. However, if the binary fraction of the G dwarfs approaches 100% as some studies suggest, the yield of stable systems would be in the range of 15% at best (with 10 m s(-1) filtering). Use of the initial SIM measurement as a final filter is shown not to be critical in either case, although it could improve the yield of stable grid members. For a grid composed of weak-lined K giants, the residual contamination by large unmodeled motions will amount to about 3% (and rises to about 10% if a 60 m s(-1) radial velocity criterion is used). The selective introduction of quadratic terms in the proper motion solutions during the postmission phase of data reduction can reduce contamination to a remarkable 1% or better in either case. Analytic estimates based on circular orbits are developed that show how these results come about.