Using the Sloan Digital Sky Survey (SDSS), we probe the halo mass distribution by studying the velocities of satellites orbiting isolated galaxies. In a subsample that covers 2500 sq. degrees on the sky, we detect about 3000 satellites with absolute blue magnitudes going down to M_B = -14; most of the satellites have M_B=-16 to -18, comparable to the magnitudes of M32 and the Magellanic Clouds. After a careful, model-independent removal of interlopers, we find that the line-of-sight velocity dispersion of satellites declines with distance to the primary. For an L* galaxy the r.m.s. line-of-sight velocity changes from ~120 km/s at 20 kpc to ~60 km/s at 350 kpc. This decline agrees remarkably well with theoretical expectations, as all modern cosmological models predict that the density of dark matter in the peripheral parts of galaxies declines as rho_DM propto r^{-3}. Thus, for the first time we find direct observational evidence of the density decline predicted by cosmological models; we also note that this result contradicts alternative theories of gravity such as MOND. We also find that the velocity dispersion of satellites within 100 kpc scales with the absolute magnitude of the central galaxy as sigma propto L^{0.3}; this is very close to the Tully-Fisher relation for normal spiral galaxies.

We propose a new explanation for the origin of angular momentum in galaxies and their dark halos, in which the halos obtain their spin through the cumulative acquisition of angular momentum from satellite accretion. In our model, the build-up of angular momentum is a random walk process associated with the mass assembly history of the halo's major progenitor. We assume no correlation between the angular momenta of accreted objects. Using the extended Press-Schechter approximation, we calculate the growth of mass, angular momentum, and spin parameter $\lambda$ for many halos. Our random walk model reproduces the key features of the angular momentum of halos found in N-body simulations: a lognormal distribution in $\lambda$ with an average of $<\lambda> \approx 0.04$, independent of mass and redshift. The evolution of the spin parameter in individual halos in this model is quite different from the steady increase with time of angular momentum in the tidal torque picture. We find both in N-body simulations and in our random walk model that the value of $\lambda$ changes significantly with time for a halo's major progenitor. It typically has a sharp increase due to major mergers, and a steady decline during periods of gradual accretion of small satellites. The model predicts that on average the $\lambda$ of halos which had major mergers after redshift $z=2$ should be substantially larger than the $\lambda$ of those which did not. Perhaps surprisingly, this suggests that halos that host late-forming elliptical galaxies should rotate faster than halos of spiral galaxies.