Does DESI DR2 challenge $\Lambda$CDM paradigm ?

TL;DR

DESI DR2 data shows some tension with the $ ext{Lambda}$CDM model at certain redshifts, but overall does not decisively challenge it; the apparent deviations are mainly driven by specific galaxy samples and model degeneracies.

Contribution

This study provides a detailed analysis of DESI DR2 data, examining its consistency with $ ext{Lambda}$CDM and exploring implications for dynamical dark energy models.

Findings

DR2 shows improved agreement with $ ext{Lambda}$CDM compared to DR1.

Tensions at specific redshifts are driven mainly by LRG1 and LRG2 samples.

Model comparison indicates weak evidence against $ ext{Lambda}$CDM when certain samples are removed.

Abstract

Although debate on DESI DR1 systematics remains, DESI DR2 is consistent with DR1 and strengthens its trends. In our analysis, the LRG1 point at $z_{\mathrm{eff}}=0.510$ and the LRG3+ELG1 point at $z_{\mathrm{eff}}=0.934$ are in tension with the $\Lambda$CDM-anchored $\Omega_m$ inferred from Planck and SNe Ia (Pantheon$^{+}$, Union3, DES-SN5YR): for LRG1 the tensions are $2.42\sigma$, $1.91\sigma$, $2.19\sigma$, and $2.99\sigma$; for LRG3+ELG1 they are $2.60\sigma$, $2.24\sigma$, $2.51\sigma$, and $2.96\sigma$. Across redshift bins DR2 shows improved agreement relative to DR1, with the $\Omega_m$ tension dropping from $2.20\sigma$ to $1.84\sigma$. Nevertheless, DR2 alone is not decisive against $\Lambda$CDM, and the apparent deviation is driven mainly by LRG1 and LRG2. In a $\omega_0\omega_a$CDM fit using all tracers we find a posterior mean with $w_0>-1$, consistent with dynamical dark energy and nominally challenging $\Lambda$CDM. Removing LRG1 and/or LRG2 restores $\Lambda$CDM concordance ($\omega_0\to-1$); moreover, $\omega_0^{\mathrm{(LRG2)}}>w_0^{\mathrm{(LRG1)}}$, indicating that LRG2 drives the trend more strongly. Model selection via the natural-log Bayes factor $\ln\mathrm{BF}\equiv\ln(Z_{\Lambda\mathrm{CDM}}/Z_{\omega_0\omega_a\mathrm{CDM}})$ yields weak evidence for $\Lambda$CDM when LRG1, LRG2, or both are removed, and is inconclusive for the full sample. Hence the data do not require the extra $\omega_a$ freedom, and the apparent $\omega_0>-1$ preference should be interpreted cautiously as a reflection of the $\omega_0$$\omega_a$ degeneracy with limited per-tracer information.