Wolfgang Kerzendorf

• Position:

  Student, supervised by Brian Schmidt

• The title of your research proposal:

  Exploring the physics of Type Ia supernovae through spectral analysis

• A brief description of the nature of the computations to be performed (resources required, interactive/batch job, i/o requirements, single processor/parallel jobs, scalability):

  Type Ia supernovae are one of the most luminous events in the universe. They are important in astronomy as a cosmological distance indicator. Type Ia supernovae are the physical endpoint of some white dwarfs in a binary system. The nucleosynthesis in the explosion has an important impact in the abundance pattern of the interstellar medium. Embarrassingly, we know little about them. Their progenitor scenario as well as the details of the explosion mechanism remain a mystery.

  In recent years there has been a dramatic increase in surveys targeting these events. There are now many well sampled light curves which are mainly used for cosmological distance measurements. The number of spectroscopically well sampled Type Ia explosions is also rapidly increasing. I will construct an auutomatic analysis tool for Type Ia supernova spectra. This tool will produce matching synthetic spectra to exisiting observed ones. For the radiative transfer code I use a code developed by Mazzali et al. (2008). It is well tested and reproduces supernova spectra well. It produces a supernova spectrum for a given luminosity, photospheric velocity, time since explosion and several elemental abundances. Tweaking these parameters to produce a spectrum that fits an observed spectrum is a complex, as of yet, manual task.

  An automation will also provide us with error analysis, as well as assuring reproducibility. To summarize the problem: Given an observed spectrum I will produce a fitting synthetic spectrum tweaking 10 parameters with different numerical ranges. The calculation for one set of parameters on a state of the art laptop computer is roughly 1 minute per parameter set. The search space is large and complex.

  Sampling the whole parameter space is not feasible due to cpu time and storage space requirements. As an individual parameter calculation takes a minute the algorithm should search the parameter space in parallel. Additionally it needs to be very robust against local extrema. Genetic Algorithms fulfill all of these requirements. Currently we are working with 150 indivdual parameter sets for each generation. A scheduler distributes each parameter set to a cloud-like computing environment (COALA contributes 32 processors). This enables us to decrease the average time for each spectrum from 1 minute to about 1 second on average. The project is still in the testing phase, but has already delivered promising results.