Bayesian Reasoning in Eternal Inflation: A Solution to the Measure Problem

Probabilities in eternal inflation are traditionally defined as limiting frequency distributions, but a unique and unambiguous probability measure remains elusive. In this paper, we present a different approach, based on Bayesian reasoning. Our starting point is the master equation governing vacuum dynamics, which describes a random walk on the network of vacua. Our probabilities require two pieces of prior information, both pertaining to initial conditions: a prior density $\rho(t)$ for the time of nucleation, and a prior probability $p_\alpha$ for the ancestral vacuum. For ancestral vacua, we advocate the uniform prior as a conservative choice, though our conclusions are fairly insensitive to this choice. For the time of nucleation, we argue that a uniform prior is consistent with the time-translational invariance of the master equation and represents the minimally-informative choice. The resulting predictive probabilities coincide with Bousso's "holographic" prior probabilities and are closely related to Garriga and Vilenkin's "comoving" probabilities. Despite making the least informative priors, these probabilities are surprisingly predictive. They favor vacua whose surrounding landscape topography is that of a deep funnel, akin to the folding funnels of naturally-occurring proteins. They predict that we exist during the approach to near-equilibrium, much earlier than the mixing time for the landscape. We also consider a volume-weighted $\rho(t)$, which amounts to weighing vacua by physical volume. The predictive probabilities in this case coincide with the GSVW measure. The Bayesian framework allows us to compare the plausibility of the uniform-time and volume-weighted hypotheses to explain our data by computing the Bayesian evidence for each. We argue, under general and plausible assumptions, that posterior odds overwhelmingly favor the uniform-time hypothesis.