Master Thesis Defence by Morten Raahauge Bastholm

Title: Machine learning approach to calibrate Type-Ia supernovae luminosity relation using their host galaxy properties

This project presents a novel approach to calibrate the peak luminosities of Type Ia 
supernova for use in cosmological analysis and estimation of the Hubble constant H0 . This is done with the goal to possibly minimize the intrinsic scatter present in the current calibrations of Type Ia and to shed some light on possible solutions to the Hubble tension. The analysis performed tries to take into consideration the reddening RV from local environment of the supernova by using the global reddening RV of the host galaxy as a tracer. The light curves of the supernovae are fitted with the SNooPy light curve fitter using the host reddening RV to correct the color of the light curves. This eliminates the need for the global extinction R from the classical Tripp-calibration of Type Ia supernovae in the supernova calibration and replaces it with a local reddening correction intrinsic to each supernova.
For supernovae without available reddening information of the host galaxy, a random forest model is build to estimate RV using stellar mass, specific star formation rate, and total metallicity as traced by [MgFe] of the host galaxy. A cosmological sample consisting of 81 supernovae is build with observations from the CfA supernova Archive survey, the Carnegie Supernova Project survey, the Foundation Supernova Survey, and the Young Supernovae Experiment survey. A calibrating sample of 17 supernovae from the Pantheon+ sample with host reddening estimated using the random forest model is used to anchor the cosmological supernovae in the final analysis. From the final analysis using a Bayesian approach the best-fit estimated value of H0 =(73.96 ± 1.47) km s−1 Mpc−1 using a second order polynomial correction for the absolute magnitude of the supernovae. This model also gave the smallest intrinsic scatters of the two samples with σcosmo (0.26 ± 0.02) mag and σcalib = (0.08 ± 0.03) mag. The estimate of the Hubble constant is in good correspondence with other estimates based on Type Ia supernovae observations. The intrinsic scatter from corrections of the supernovae samples are however not reduced compared to similar methods.


  • Christa Gall, University of Copenhagen, Niels Bohr Institute

  • Luca Izzo - Co supervisor, University of Copenhagen & INAF


  • Hans Kjeldsen -Aarhus University