Data-driven approach to Type Ia supernovae: variable selection on the peak luminosity and clustering in visual analytics

摘要

TypeIasupernovae(SNIa)haveanalmostuniformpeakluminosity,sothattheyareusedas"standardcandle"toestimatedistancestogalaxiesincosmology.Inthisarticle,weintroduceourtworecentworksonSNIabasedondata-drivenapproach.ThediversityinthepeakluminosityofSNIacanbereducedbycorrectionsinseveralvariables.Thecoloranddecayratehavebeenusedastheexplanatoryvariablesofthepeakluminosityinpaststudies.However,itisproposedthattheirspectraldatacouldgiveabettermodelofthepeakluminosity.Weusecross-validationinordertocontrolthegeneralizationerrorandaLASSO-typeestimatorinordertochoosethesetofvariables.Using78samplesand276candidatesofvariables,weconfirmthatthepeakluminositydependsonthecoloranddecayrate.Ouranalysisdoesnotsupportaddinganyothervariablesinordertohaveabettergeneralizationerror.Ontheotherhand,thisanalysisisbasedontheassumptionthatSNIaoriginateinasinglepopulation,whileitisnottrivial.Indeed,severalsub-typespossiblyhavingdifferentnaturehavebeenproposed.Weusedavisualanalyticstoolfortheasymmetricbiclusteringmethodtofindbothagoodsetofvariablesandsamplesatthesametime.Using14variablesand132samples,wefoundthatSNIacanbedividedintotwocategoriesbytheexpansionvelocityofejecta.Thoseexamplesdemonstratethatthedata-drivenapproachisusefulforhigh-dimensionallarge-volumedatawhichbecomescommoninmodernastronomy...