Estimating star formation rates from infrared and radio luminosities: The origin of the radio-infrared correlation

I have assembled a diverse sample of galaxies from the literature with far-ultraviolet (FUV), optical, infrared (IR), and radio luminosities to explore the calibration of radio-derived and IR-derived star formation (SF) rates and the origin of the radio-IR correlation. By comparing the 8-1000 mu;m IR, which samples dust-reprocessed starlight, with direct stellar FUV emission, I show that the IR traces most of the SF in luminous ~L* galaxies but traces only a small fraction of the SF in faint ~0.01L* galaxies. If radio emission were a perfect SF rate indicator, this effect would cause easily detectable curvature in the radio-IR correlation. Yet, the radio-IR correlation is nearly linear. This implies that the radio flux from low-luminosity galaxies is substantially suppressed, compared to brighter galaxies. This is naturally interpreted in terms of a decreasing efficiency of nonthermal radio emission in faint galaxies. Thus, the linearity of the radio-IR correlation is a conspiracy: both indicators underestimate the SF rate at low luminosities. SF rate calibrations that take into account this effect are presented, along with estimates of the random and systematic error associated with their use