No evidence for phantom crossing: local goodness-of-fit improvements do not persist under global Bayesian model comparison

Abstract

Recent cosmological data have been interpreted as indicating deviations from $\Lambda$CDM within the standard $w_0w_a$ parametrization, including hints of phantom crossing and dynamical dark energy. However, such inferences can be parametrization-dependent and need not imply a statistically robust detection. We test these claims by comparing $\Lambda$CDM, $w_0w_a$, and thawing quintessence models, using the Deviance Information Criterion (DIC) and the Bayesian evidence $\ln \mathcal{Z}$. We find that $w_0w_a$ can provide a slightly improved local fit; however, this improvement is confined to a limited region of the parameter space. The global Bayesian evidence does not support it once the full prior volume is accounted for. In particular, cases with $\Delta{\rm DIC}<0$ but $\Delta \ln \mathcal{Z}<0$ indicate that these improvements are not statistically significant. We show that all models are statistically indistinguishable, and that there is no statistically consistent evidence across different datasets for either dynamical dark energy or phantom crossing.