We combine the CfA Redshift Survey (CfA2) and the Southern Sky Redshift Survey (SSRS2) to estimate the pairwise velocity dispersion of galaxies sigma(12) on a scale of similar to 1 h(-1) Mpc. Both surveys are complete to an apparent magnitude limit B(0)=15.5. Our sample includes 12 812 galaxies distributed in a volume 1.8x10(6)h(-3) Mpc(3). We conclude: (1) The pairwise velocity dispersion of galaxies in the combined CfA2+SSRS2 redshift survey is sigma(12)=540+/-180 kms(-1). Both the estimate and the variance of sigma(12) significantly exceed the canonical values sigma(12)=340+/-40 measured by Davis and Peebles (1983) using CfA1. (2) We derive the uncertainty in sigma(12) from the variation among subsamples with volumes on the order of 7x10(5)h(-3) Mpc(3). This variation is nearly an order of magnitude larger than the formal error, 36 km s(-1) derived using least-squares fits to the CfA2+SSRS2 correlation function. This variation among samples is consistent with the conclusions of Mo ef al. (1993) for a number of smaller surveys and with the analysis of CfA1 by Zurek et al. (1994). (3) When we remove Abell clusters with R greater than or equal to 1 from our sample, the pairwise velocity dispersion of the remaining galaxies drops to 295+/-99 km s(-1). (4) The dominant source of variance in sigma(12) is the shot noise contributed by dense virialized systems. Because sigma(12) is pair weighted, the statistic is sensitive to the few richest systems in the volume. This sensitivity has two consequences. First, sigma(12) is biased low in small volumes, where the number of clusters is small. Second, we can estimate the variance in sigma(12) as a function of survey volume from the distribution of cluster and group velocity dispersions n(sigma). For either a COBE-normalized CDM universe or for the observed distribution of Abell cluster velocity dispersions, the volume required for sigma(12) to converge (delta sigma(12)/sigma(12)<0.1 is similar to 5x10(6)h(-3) Mpc(3), larger than the volume of CfA2+SSRS2. (5) The distribution of pairwise velocities is consistent with an isotropic exponential with velocity dispersion independent of scale. The inferred single-galaxy velocity distribution function is incompatible with an isotropic exponential. Thus the observed kinematics of galaxies differ from those measured in the hydrodynamic simulations of Cen & Ostriker (1993). On the other hand, our observations appear to be consistent with the velocity distribution function measured by Zurek et al. (1994) using collisionless simulations with much higher resolution than Cen & Ostriker (1993). The large dynamic range in the simulations by Zurek ef al. (1994) affords a more accurate treatment of galaxy interactions, which dramatically alter the dynamical evolution the galaxy distribution (Couchman & Carlberg 1999; Zurek et al. 1994). The agreement between the observed velocity distribution and the one predicted by these high-resolution simulations may be another clue that mergers play an important role in the evolution of galaxies and large-scale structure. (C) 1995 American Astronomical Society.
