2000ApJ...540.1041S

Interior layers of stars that have been exposed by surface mass loss reveal aspects of their chemical and convective histories that are otherwise inaccessible to observation. It must be significant that the surface hydrogen abundances of luminous blue variables (LBVs) show a remarkable uniformity, specifically X_surf=0.3-0.4, while those of hydrogen-poor Wolf-Rayet (WN) stars fall, almost without exception, below these values, ranging down to X_surf=0. According to our stellar model calculations, most LBVs are post-red-supergiant objects in a late blue phase of dynamical instability, and most hydrogen-poor WN stars are their immediate descendants. If this is so, stellar models constructed with the Schwarzschild (temperature-gradient) criterion for convection account well for the observed hydrogen abundances, whereas models built with the Ledoux (density-gradient) criterion fail. At the brightest luminosities, the observed hydrogen abundances of LBVs are too large to be explained by any of our highly evolved stellar models, but these LBVs may occupy transient blue loops that exist during an earlier phase of dynamical instability when the star first becomes a yellow supergiant. Independent evidence concerning the criterion for convection, which is based mostly on traditional color distributions of less massive supergiants on the Hertzsprung-Russell diagram, tends to favor the Ledoux criterion. It is quite possible that the true criterion for convection changes over from something like the Ledoux criterion to something like the Schwarzschild criterion as the stellar mass increases.