3.1. Are we misinterpreting the data?
After the original supernova results (Riess et al. 1998, Perlmutter et al. 1999) were announced in 1998, cosmologists converted rather quickly from skepticism about universal acceleration to a tentative acceptance, which has grown substantially stronger with time. The primary reason for this sudden conversion has been the convergence of several complementary lines of evidence in favor of a concordance model; foremost among the relevant observations are the anisotropy spectrum of the cosmic microwave background (Spergel et al. 2003) and the power spectrum of large-scale structure (Verde et al. 2002), but a number of other methods have yielded consistent answers.
Nevertheless, it remains conceivable that we have dramatically misinterpreted the data, and the apparent agreement of an = 0.7, M = 0.3 cosmology with a variety of observations is masking the true situation. For example, the supernova observations rely on the nature of Type Ia supernovae as "standardizable candles," an empirical fact about low-redshift supernovae which could somehow fail at high redshifts (although numerous consistency checks have confirmed basic similarities between SNe at all redshifts). Given the many other observations, this failure would not be enough to invalidate our belief in an accelerating universe; however, we could further imagine that these other methods are conspiring to point to the wrong conclusion. This point of view has been taken by Blanchard et al. (2003), who argue that a flat matter-dominated (M = 1) universe remains consistent with the data. To maintain this idea, it is necessary to discard the supernova results, to imagine that the Hubble constant is approximately 46 km/sec/Mpc (in contrast to the Key Project determination of 70 ± 7 km/sec/Mpc, Freedman et al. 2001), to interpret data on clusters and large-scale structure in a way consistent with M = 1, to relax the conventional assumption that the power spectrum of density fluctuations can be modeled as a single power law, and to introduce some source beyond ordinary cold dark matter (such as massive neutrinos) to suppress power on small scales. To most workers in the field this conspiracy of effects seems (even) more unlikely than an accelerating universe.
A yet more drastic route is to imagine that our interpretation of the observations has been skewed by the usual assumption of an isotropic universe. It has been argued (Linde, Linde & Mezhlumian 1995) that some versions of the anthropic principle in an eternally inflating universe lead to a prediction that most galaxies on a spacelike hypersurface are actually at the center of spherically symmetric domains with radially-dependent density distributions; such a configuration could skew the distance-redshift relation at large distances even without dark energy. This picture relies heavily on a choice of measure in determining what "most" galaxies are like, an issue for which there is no obvious correct choice.
Finally, we may imagine that there is indeed new physics involved in making distant supernovae dimmer than we would expect, but it is the propagation of light which is being altered rather than the expansion of the universe. Such a scenario has been worked out by Csaki, Kaloper & Terning (2002), who suggested that photons passing through intergalactic magnetic fields may be converting into light axion-like particles, resulting in a diminuition of the total flux from supernovae. In addition to introducing a new field with significantly constrained mass and coupling, this model still requires a form of dark energy to reconcile the flatness of the universe with the low observed matter density (although the required energy source could decay away more rapidly than ordinary dark energy).
The lengths to which it seems necessary to go in order to avoid concluding that the universe is accelerating is a strong argument in favor of the concordance model.