2012ApJ...758...21C

All globular clusters (GCs) studied to date show evidence for internal (star-to-star) variation in their light-element abundances (including Li, C, N, O, F, Na, Mg, Al, and probably He). These variations have been interpreted as evidence for multiple star formation episodes within GCs, with secondary episodes fueled, at least in part, by the ejecta of asymptotic giant branch (AGB) stars from a first generation of stars. A major puzzle emerging from this otherwise plausible scenario is that the fraction of stars associated with the second episode of star formation is observed to be much larger than expected for a standard initial mass function. The present work investigates this tension by modeling the observed anti-correlation between [Na/Fe] and [O/Fe] for 20 Galactic GCs. If the abundance pattern of the retained AGB ejecta does not depend on GC mass at fixed [Fe/H], then a strong correlation is found between the fraction of current GC stellar mass composed of pure AGB ejecta, f_p, and GC mass. This fraction varies from 0.20 at low masses (10^4.5 M_☉) to 0.45 at high masses (10^6.5 M_☉). The fraction of mass associated with pure AGB ejecta is directly related to the total mass of the cluster at birth; the ratio between the initial and present mass in stars can therefore be derived. Assuming a star formation efficiency of 50%, the observed Na-O anti-correlations imply that GCs were at least 10-20 times more massive at birth, a conclusion that is in qualitative agreement with previous work. These factors are lower limits because any mass-loss mechanism that removes first- and second-generation stars equally will leave f_punchanged. The mass dependence of f_p probably arises because lower mass GCs are unable to retain all of the AGB ejecta from the first stellar generation. Recent observations of elemental abundances in intermediate-age Large Magellanic Cloud clusters are re-interpreted and shown to be consistent with this basic scenario. The small scatter in f_p at fixed GC mass argues strongly that the process responsible for the large mass loss is internal to GCs. A satisfactory explanation of these trends is currently lacking.