Classification of Double Neutron Stars Using the Extreme Deconvolution Method

Abstract The double neutron star (DNS) systems are pivotal compact binary systems for testing gravitational waves and general relativity—evidenced by radio telescope observations of their orbital decay and LIGO-Virgo-KAGRA detections of their coalescences. Thus, a systematic classification study of DNS systems can profoundly advance our understanding of their origins and evolutionary pathways. Based on a Gaussian mixture model, we applied the Extreme Deconvolution Method to the classification of an updated sample of 33 pairs of DNSs. Bayesian Information Criterion analysis reveals that these systems are best modeled by a three-Gaussian distribution in the orbital period–eccentricity plane, partitioning them into three distinct populations. Type I DNSs (5 systems): Characterized by low eccentricity and orbital periods shorter than 1 day, they are likely to merge within the Hubble time. Type II DNSs (13 systems): Exhibiting significantly diversified parameter distributions, these systems show longer orbital periods and higher eccentricities compared to Type I. Type III DNSs (14 systems): These systems exhibit much longer orbital periods and various eccentricities. DNS orbital period–eccentricity correlations infer distinct formation channels: Type I systems are likely associated with ultra-stripped supernovae, and Type II systems may arise from electron capture supernovae, while Type III systems likely originate from iron-core collapse supernovae. We conclude that three unconventional DNSs (PSR J0509+3801, B1534+12, and J1757-1854) in which the non-recycled NSs are heavier than the recycled pulsars should belong to Type II DNSs. The non-recycled NSs in these systems possibly formed through electron capture supernovae.