The spiral patterns in computer simulations of isolated disks must be self-excited. However, without an understanding of their origin, they could all be dismissed as arising from some unknown numerical artifact, perhaps related to the small number of particles (Bertin & Lin 1996). If true, such a skeptical viewpoint would require spirals in real galaxies to have a different origin, though they need not necessarily be long-lived; a possible alternative universal mechanism for recurrent transient spirals could be forcing by substructure in the dark matter halos, as reported by Dubinski (these proceedings).
The development of short-lived spiral features has not changed over the years as the numerical quality of simulations has improved. Codes have held up under extensive testing (Sellwood 1983; Inagaki, Nishida & Sellwood 1984; Sellwood & Athanassoula 1986; Earn & Sellwood 1995; Sellwood & Evans 2001), but a successful test in one problem is no guarantee of the code's performance in other problems. I have recently (Sellwood, in preparation) conducted a suite of simulations of a stable Mestel disk model with particle numbers ranging up to N = 5 × 108. Simulations of a linearly stable disk should manifest no structure exceeding that expected from amplified particle noise. However, I find evidence that the linear theory prediction fails whenever the disturbance amplitude exceeds ~ 2% of the undisturbed density. Again this could be an important physical result or another manifestation of the supposed artifact, although the greatest particle number is now only 2 orders of magnitude below the number of stars in a disk.
The most effective way to counter this dismissive viewpoint would be to understand the mechanism for recurrent spiral generation in simulations and to find evidence that the process also works in nature. I (Sellwood 2000) outlined a possible mechanism for recurrent spiral generation in collisionless particle disks, but observational evidence to support it was lacking. However, the detailed phase-space structure of stars in the solar neighborhood, as revealed in the monumental study of local F & G dwarfs by Nordström et al. (2004) has opened the door to empirical tests. Clearly, data from the upcoming GAIA mission will be even more useful.
The central idea of the recurrence mechanism is that scattering of stars at the principal resonances of the spiral pattern (e.g. BTII) changes the distribution function (DF) in such a way as to seed the growth of a new instability in the disk. Each instability is caused by locally steep gradients in the DF, supported vigorously by the response of the surrounding disk (Sellwood & Kahn 1991). The spiral wave grows until non-linear effects become important, and Sellwood & Binney (2002) showed that the onset of horse-shoe orbits at corotation causes the disturbance to begin to disperse. At this moment, all the action stored in the wave is carried away to the Lindblad resonances at the group velocity where stars are scattered to create the conditions for a new instability. While the idea is simply stated, many of the details remain to be worked out.
Support for this idea comes from analysis of the velocity distribution of 13240 nearby F & G dwarf stars in the Geneva-Copenhagen sample (Nordström et al. 2004). The distribution of stars in phase space is very far from smooth, as originally noted by Dehnen (1998), but is broken into various "star streams," which cannot simply be dissolved star clusters (Famaey et al. 2007; Bensby et al. 2007), with no underlying smooth component.
Some of these features have been modeled as scattering by the bar (Kalnajs 1991; Dehnen 2000) or by spirals (Yu & Tremaine 2002) or a combination of both (Quillen & Minchev 2005), but all these studies modeled the velocity-space distribution, not integral-space, and none was self-consistent. The substructure in the DF is too complicated to have a single origin, and such ideas could be responsible for one or more of the features.
I find (in preparation) that the local DF, when plotted as a function of energy and angular momentum, contains at least one clear feature of high statistical significance that appears to have resulted from either scattering or trapping at a resonance. Analysis of the sample in action-angle variables reveals that it is an inner Lindblad resonance that has most recently sculptured the local DF, which is exactly the feature Sellwood (1994) predicted might be observable in such a sample.
If the mechanism invoked to interpret the behavior of simulations does indeed appear to occur in the Milky Way, then the entire picture stands on much firmer ground.