The DESI DR1/DR2 evidence for dynamical dark energy is biased by low-redshift supernovae

Recently, a $3\sim 4\sigma$ preference for dynamical dark energy has been reported by the Dark Energy Spectroscopic Instrument (DESI) collaboration, which has inspired hot debates on new physics or systematics. In this paper, we reveal that this preference is significantly biased by an external low-redshift supernova (low-$z$ SN) sample, which was combined with the Dark Energy Survey SN program (DES-SN) in their Year-Five data release (DESY5). Using the intercept in the SN magnitude-distance relation as a diagnostic for systematics, we find not only large dispersions but also a large discrepancy in the low-$z$ SN sample when compared to the high-$z$ DES-SN sample within the single DESY5 compilation, in contrast to the uniform behavior found in the PantheonPlus data. Correcting for this low-$z$ systematics with or without including the cosmic microwave background data can largely reduce the preference for dynamical DE to be less than $2\sigma$. Therefore, the DESI preference for dynamical DE is biased by some unknown systematics in the low-$z$ SN sample.