Combining computer models to account for mass loss in stellar evolution

Abstract

AbstractIntricate computer models can be used to describe complex physical processes in astronomy such as the evolution of stars. Like a sampling distribution, these models typically predict observed quantities as a function of a number of unknown parameters. Including them as components of a statistical model, however, leads to significant modeling, inferential, and computational challenges. In this article, we tackle these challenges in the study of the mass loss that stars experience as they age. We have developed a new Bayesian technique for inferring the so‐called initial–final mass relation (IFMR), the relationship between the initial mass of a Sun‐like star and its final mass as a white dwarf. Our model incorporates several separate computer models for various phases of stellar evolution. We bridge these computer models with a parameterized IFMR in order to embed them into a statistical model. This strategy allows us to apply the full force of powerful statistical tools to build, fit, check, and improve the statistical models and their computer model components. In contrast to traditional techniques for inferring the IFMR, which tend to be quite ad hoc, we can estimate the uncertainty in our fit and ensure that our model components are internally coherent. We analyze data from three star clusters: NGC 2477, the Hyades, and M35 (NGC 2168). The results from NGC 2477 and M35 suggest different conclusions about the IFMR in the mid‐ to high‐mass range, raising questions for further astronomical work. We also compare the results from two different models for the primary hydrogen‐burning stage of stellar evolution. We show through simulations that misspecification at this stage of modeling can sometimes have a severe effect on inferred white dwarf masses. Nonetheless, when working with observed data, our inferences are not particularly sensitive to the choice of model for this stage of evolution. © 2013 Wiley Periodicals, Inc. Statistical Analysis and Data Mining 6: 34–52, 2013