Radial Partition of Predictive Information in Bulge-Dominated SPARC Galaxies

The SPARC galaxies with a separately modeled bulge occupy a markedly different baryonic-velocity regime from the rest of the SPARC population, raising the question of whether their rotation-curve discrepancies contain a distinct distribution of predictive information [1]. Using all 175 SPARC galaxies and 3,391 radial measurements, we compare the full complement of 32 galaxies with nonzero modeled bulge contributions with 143 galaxies without a separate bulge term. Across these unmatched populations, cross-validation prediction fails strongly, but most of that failure is associated with the different dynamical ranges occupied by the two groups, especially in maximum baryonic rotation speed. Restricting the comparison to common Vbar,max-Rmax support lowers the bulge-trained prediction error from 123.7 to 34.5 km/s, although those two scores use different target galaxy sets. In a like-for-like check on the 26 non-bulge galaxies that fall inside the shared support region (the other 117 lie outside it), restricting the bulge training set to common support lowers RMSE from 51.50 to 34.07 km/s, a 33.8% reduction. Within common support, each galaxy’s measured radial range is divided into four equal quarters (0–25%, 25–50%, 50–75%, and 75–100% of Rmax). The largest difference occurs in the second quarter, while the outermost quarter is nearly equal, although the uncertainty remains broad. Common support contains 22 bulge galaxies, so an equal-size check pairs those 22 with 22 of the 26 common-support non-bulge galaxies. Apparent steep inner residual structures are visually prominent but contribute little to transferable prediction. The firm result is therefore strong support-controlled transferability; the suggested radial partition remains a question for further testing.