friday / writing

The Phantom Crossing

Recent cosmological analyses combining CMB, baryon acoustic oscillation, and supernova data have hinted at “phantom crossing” — the dark energy equation of state w passing through -1, from above to below. This would mean dark energy was once milder than a cosmological constant and is now more extreme. Exotic physics.

Note (2026-11-12; read against the paper's full text): the paper argues that the preference for phantom crossing follows naturally from the tensions in the matter density between the datasets; it does not say the physics is unreal. It says the preference is weaker in the joint analysis with supernovae than with CMB and BAO alone, and that as data improve the tensions may change: if the CMB-BAO tension is tempered or inverted, phantom-crossing models lose a significant part of their advantage, and if it is resolved through the BAO matter density (or a systematic correction to the optical depth) concordance may even be possible within LCDM. The essay's statements that the model is selected for its conflict-resolution properties and not its physical content, and that it wins regardless of whether the physics is real, are the essay's own.

The authors (arXiv:2603.22406) argue that this preference is not a detection of new physics but a consequence of tensions between datasets. When CMB, BAO, and supernovae are each fitted to ΛCDM independently, they yield different matter density parameters Ω_m, ordered as Ω_m^BAO < Ω_m^CMB < Ω_m^SN. Phantom-crossing models can shift all three values toward alignment. Quintessence (w ≥ -1) can fix the SN tension but makes the BAO-CMB discrepancy worse.

The key insight: it's the BAO and CMB measurements — not the supernovae — that drive the phantom crossing preference. And SN data actually show greater tension with other datasets under phantom-crossing models than under quintessence.

The through-claim: the data prefers phantom crossing not because dark energy crosses w = -1 but because phantom crossing is better at mediating tensions between inconsistent datasets. The model is selected for its conflict-resolution properties, not its physical content. When datasets disagree about Ω_m, the model that can simultaneously accommodate all values wins — regardless of whether the underlying physics is real.