Add batch_0602 typeI: 67 tasks (binaries via LFS; excl. data_raw symlinks & cache cruft)

.gitattributes

@@ -58,3 +58,221 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+*.pdf filter=lfs diff=lfs merge=lfs -text
+*.csv filter=lfs diff=lfs merge=lfs -text
+baseball_pythagorean_winpct_lahman__win_pct/data_raw filter=lfs diff=lfs merge=lfs -text
+binary_pulsar_post_keplerian_atnf__PBDOT/data_raw filter=lfs diff=lfs merge=lfs -text
+binary_pulsar_post_keplerian_atnf__PBDOT/reference/peters_1963.txt filter=lfs diff=lfs merge=lfs -text
+bird_flight_speed_alerstam__Ue/data_raw filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/data_raw filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/data_raw filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/prep_data.py filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/data_raw filter=lfs diff=lfs merge=lfs -text
+bns_postmerger_waveform_core__f2/data_raw filter=lfs diff=lfs merge=lfs -text
+budyko_partitioning_mach__et_over_p/data_raw filter=lfs diff=lfs merge=lfs -text
+budyko_partitioning_mach__et_over_p/reference/summary_formula_yang_2008.md filter=lfs diff=lfs merge=lfs -text
+catchment_mean_runoff_camels__q_mean/data_raw filter=lfs diff=lfs merge=lfs -text
+cepheid_period_luminosity__M_W/data_raw filter=lfs diff=lfs merge=lfs -text
+cloud_cover_parameterization__cloud_cover/data_raw filter=lfs diff=lfs merge=lfs -text
+commuting_gravity_lodes__flow_Tij/data_raw filter=lfs diff=lfs merge=lfs -text
+dna_melting_temperature_khandelwal__Tm/data_raw filter=lfs diff=lfs merge=lfs -text
+drosophila_isolation_by_distance_dest__M_stat/data_raw filter=lfs diff=lfs merge=lfs -text
+eclipsing_binary_mass_luminosity_debcat__log_L_Lsun/data_raw filter=lfs diff=lfs merge=lfs -text
+exfor_neutron_capture_resonance_gold__sigma_E/data_raw filter=lfs diff=lfs merge=lfs -text
+exoplanet_mass_radius__R/data_raw filter=lfs diff=lfs merge=lfs -text
+galaxy_mass_metallicity_sdss__oh/data_raw filter=lfs diff=lfs merge=lfs -text
+galaxy_stellar_mass_sdss__log_M_star/data_raw filter=lfs diff=lfs merge=lfs -text
+geomagnetic_storm_dst__dDst_dt/data_raw filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__CDE/data_raw filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__EC/data_raw filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__ME/data_raw filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__NOE/data_raw filter=lfs diff=lfs merge=lfs -text
+gravity_wgs84_somigliana__g0/data_raw filter=lfs diff=lfs merge=lfs -text
+ground_motion_ita18__log10_pga/data_raw filter=lfs diff=lfs merge=lfs -text
+gutenberg_richter_b_value_usgs__gr_freq_mag/data_raw filter=lfs diff=lfs merge=lfs -text
+hacks_law_river_length_hydrosheds__main_length_km/data_raw filter=lfs diff=lfs merge=lfs -text
+hea_hardness_lattice_distortion_couzinie__HV/data_raw filter=lfs diff=lfs merge=lfs -text
+her_hor_pt_acid_kucernak__current_density/data_raw filter=lfs diff=lfs merge=lfs -text
+island_species_richness_terceira__S_terceira/data_raw filter=lfs diff=lfs merge=lfs -text
+island_species_richness_terceira__S_terceira/reference/cardoso_2020.txt filter=lfs diff=lfs merge=lfs -text
+lake_thermocline_depth_pilla__z_t/data_raw filter=lfs diff=lfs merge=lfs -text
+life_expectancy_preston__e0_vs_gdppc/data_raw filter=lfs diff=lfs merge=lfs -text
+liion_capacity_fade_nasa_pcoe__capacity_Ah/data_raw filter=lfs diff=lfs merge=lfs -text
+liquid_dispersion_glycols__refractive_index/data_raw filter=lfs diff=lfs merge=lfs -text
+lpbf_meltpool_rosenthal_inconel__depth/data_raw filter=lfs diff=lfs merge=lfs -text
+magnet_3c90_core_loss__log_P_vol/data_raw filter=lfs diff=lfs merge=lfs -text
+mars_crater_size_frequency__N_cum/data_raw filter=lfs diff=lfs merge=lfs -text
+materials_project_elastic_birch_murnaghan__K_VRH/data_raw filter=lfs diff=lfs merge=lfs -text
+mauna_loa_co2_keeling_curve_noaa__co2_ppm/data_raw filter=lfs diff=lfs merge=lfs -text
+metabolic_scaling_anage__log_bmr_W/data_raw filter=lfs diff=lfs merge=lfs -text
+mincer_earnings_acs_pums__log_wage/data_raw filter=lfs diff=lfs merge=lfs -text
+neo_size_frequency_distribution__N_cum_H/data_raw filter=lfs diff=lfs merge=lfs -text
+nuclear_binding_energy_ame2020__BE_per_A/data_raw filter=lfs diff=lfs merge=lfs -text
+ocean_mesoscale_eddy_closure__Sx/data_raw filter=lfs diff=lfs merge=lfs -text
+parker_spiral_imf__B_phi/data_raw filter=lfs diff=lfs merge=lfs -text
+pfas_soil_sorption__log_kd/data_raw filter=lfs diff=lfs merge=lfs -text
+pp_total_cross_section_donnachie_landshoff__sigma_tot/data_raw filter=lfs diff=lfs merge=lfs -text
+proton_em_form_factor__GE_over_GD/data_raw filter=lfs diff=lfs merge=lfs -text
+proton_em_form_factor__GE_over_GD/prep_data.py filter=lfs diff=lfs merge=lfs -text
+protoplanetary_disk_mmflux__F_mm/data_raw filter=lfs diff=lfs merge=lfs -text
+pv_module_temperature_efficiency_pvpmc__P_max/data_raw filter=lfs diff=lfs merge=lfs -text
+qg_turbulence_subgrid_pyqg__Sq/data_raw filter=lfs diff=lfs merge=lfs -text
+red_giant_asteroseismology__delta_nu/data_raw filter=lfs diff=lfs merge=lfs -text
+red_giant_asteroseismology__nu_max/data_raw filter=lfs diff=lfs merge=lfs -text
+rock_core_permeability_norweg__log_k/data_raw filter=lfs diff=lfs merge=lfs -text
+running_endurance_iaaf__velocity/data_raw filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/data_raw filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/summary_formula+dataset_kormendy_2013.md filter=lfs diff=lfs merge=lfs -text
+sne_ia_distance_modulus_pantheonplus__mu/data_raw filter=lfs diff=lfs merge=lfs -text
+solar_flat_plate_collector_graz_efficiency__eta_thermal/data_raw filter=lfs diff=lfs merge=lfs -text
+spirometry_nhanes__FEV1_L/data_raw filter=lfs diff=lfs merge=lfs -text
+spirometry_nhanes__FVC_L/data_raw filter=lfs diff=lfs merge=lfs -text
+trade_gravity_cepii__log_tradeflow/data_raw filter=lfs diff=lfs merge=lfs -text
+traffic_flow_density_ngsim__q_flow/data_raw filter=lfs diff=lfs merge=lfs -text
+tree_biomass_allometry_chave__agb/data_raw filter=lfs diff=lfs merge=lfs -text
+ttauri_accretion__L_acc/data_raw filter=lfs diff=lfs merge=lfs -text
+volcanic_column_mer_ivespa__H_top/data_raw filter=lfs diff=lfs merge=lfs -text
+wind_turbine_power_curve_engie__power_kW/data_raw filter=lfs diff=lfs merge=lfs -text
+geomagnetic_storm_dst__dDst_dt/reference/omni_lro_documentation.html filter=lfs diff=lfs merge=lfs -text
+baseball_pythagorean_winpct_lahman__win_pct/reference/miller_2007.txt filter=lfs diff=lfs merge=lfs -text
+binary_pulsar_post_keplerian_atnf__PBDOT/reference/manchester_2005.txt filter=lfs diff=lfs merge=lfs -text
+bird_flight_speed_alerstam__Ue/reference/alerstam_2007.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/bernuzzi_2016.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/camilletti_2022.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/camilletti_2024.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/coughlin_2018.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/darc_2025.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/dietrich_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/hotokezaka_2013.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/kruger_2020.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/lund_2025.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/nedora_2020.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/radice_2018_a.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/radice_2018_b.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/sekiguchi_2016.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/siegel_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mdisk/reference/villar_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/bernuzzi_2016.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/camilletti_2022.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/camilletti_2024.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/coughlin_2018.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/darc_2025.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/dietrich_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/hotokezaka_2013.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/kruger_2020.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/lund_2025.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/nedora_2020.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/radice_2018_a.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/radice_2018_b.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/sekiguchi_2016.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/siegel_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__Mej/reference/villar_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/bernuzzi_2016.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/camilletti_2022.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/camilletti_2024.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/coughlin_2018.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/darc_2025.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/dietrich_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/hotokezaka_2013.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/kruger_2020.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/lund_2025.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/nedora_2020.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/radice_2018_a.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/radice_2018_b.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/sekiguchi_2016.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/siegel_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_merger_disk_ejecta__vej/reference/villar_2017.txt filter=lfs diff=lfs merge=lfs -text
+bns_postmerger_waveform_core__f2/reference/gonzalez_2023.txt filter=lfs diff=lfs merge=lfs -text
+bns_postmerger_waveform_core__f2/reference/hotokezaka_2013.txt filter=lfs diff=lfs merge=lfs -text
+bns_postmerger_waveform_core__f2/reference/vretinaris_2020.txt filter=lfs diff=lfs merge=lfs -text
+budyko_partitioning_mach__et_over_p/reference/yang_2008.txt filter=lfs diff=lfs merge=lfs -text
+cepheid_period_luminosity__M_W/reference/ripepi_2019.txt filter=lfs diff=lfs merge=lfs -text
+cloud_cover_parameterization__cloud_cover/reference/grundner_2024.txt filter=lfs diff=lfs merge=lfs -text
+cloud_cover_parameterization__cloud_cover/reference/grundner_2025.txt filter=lfs diff=lfs merge=lfs -text
+commuting_gravity_lodes__flow_Tij/reference/lenormand2016_systematic.txt filter=lfs diff=lfs merge=lfs -text
+commuting_gravity_lodes__flow_Tij/reference/yang2014_pwo_WRONG.txt.bak filter=lfs diff=lfs merge=lfs -text
+dna_melting_temperature_khandelwal__Tm/reference/khandelwal_2010.txt filter=lfs diff=lfs merge=lfs -text
+exoplanet_mass_radius__R/reference/chen_2016.txt filter=lfs diff=lfs merge=lfs -text
+galaxy_mass_metallicity_sdss__oh/reference/andrews_2013.txt filter=lfs diff=lfs merge=lfs -text
+galaxy_mass_metallicity_sdss__oh/reference/curti_2020.txt filter=lfs diff=lfs merge=lfs -text
+galaxy_mass_metallicity_sdss__oh/reference/mannucci_2010.txt filter=lfs diff=lfs merge=lfs -text
+galaxy_mass_metallicity_sdss__oh/reference/tremonti_2004.txt filter=lfs diff=lfs merge=lfs -text
+galaxy_stellar_mass_sdss__log_M_star/reference/bell_2003.txt filter=lfs diff=lfs merge=lfs -text
+galaxy_stellar_mass_sdss__log_M_star/reference/taylor_2011.txt filter=lfs diff=lfs merge=lfs -text
+geomagnetic_storm_dst__dDst_dt/reference/king_2005.txt filter=lfs diff=lfs merge=lfs -text
+geomagnetic_storm_dst__dDst_dt/reference/markidis_2025.txt filter=lfs diff=lfs merge=lfs -text
+geomagnetic_storm_dst__dDst_dt/reference/obrien_2000.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__CDE/reference/meinshausen_2020.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__CDE/reference/pan_2019.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__CDE/reference/vazquez_2022.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__CDE/reference/york_2003.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__EC/reference/meinshausen_2020.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__EC/reference/pan_2019.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__EC/reference/vazquez_2022.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__EC/reference/york_2003.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__ME/reference/meinshausen_2020.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__ME/reference/pan_2019.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__ME/reference/vazquez_2022.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__ME/reference/york_2003.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__NOE/reference/meinshausen_2020.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__NOE/reference/pan_2019.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__NOE/reference/vazquez_2022.txt filter=lfs diff=lfs merge=lfs -text
+ghg_emission_drivers__NOE/reference/york_2003.txt filter=lfs diff=lfs merge=lfs -text
+gutenberg_richter_b_value_usgs__gr_freq_mag/reference/hainzl_2008.txt filter=lfs diff=lfs merge=lfs -text
+gutenberg_richter_b_value_usgs__gr_freq_mag/reference/helmstetter_2003.txt filter=lfs diff=lfs merge=lfs -text
+gutenberg_richter_b_value_usgs__gr_freq_mag/reference/kagan_2010.txt filter=lfs diff=lfs merge=lfs -text
+hea_hardness_lattice_distortion_couzinie__HV/reference/couzinie_dib_2018.txt filter=lfs diff=lfs merge=lfs -text
+hea_hardness_lattice_distortion_couzinie__HV/reference/gorsse_dib_2018.txt filter=lfs diff=lfs merge=lfs -text
+hea_hardness_lattice_distortion_couzinie__HV/reference/maresca_curtin_2020.txt filter=lfs diff=lfs merge=lfs -text
+hea_hardness_lattice_distortion_couzinie__HV/reference/temesi_2023_hardness_bonding.txt filter=lfs diff=lfs merge=lfs -text
+hea_hardness_lattice_distortion_couzinie__HV/reference/temesi_2024_cohesion_hardness.txt filter=lfs diff=lfs merge=lfs -text
+life_expectancy_preston__e0_vs_gdppc/reference/deaton_2003.txt filter=lfs diff=lfs merge=lfs -text
+life_expectancy_preston__e0_vs_gdppc/reference/preston_1975.txt filter=lfs diff=lfs merge=lfs -text
+liion_capacity_fade_nasa_pcoe__capacity_Ah/reference/he_2011.txt filter=lfs diff=lfs merge=lfs -text
+liion_capacity_fade_nasa_pcoe__capacity_Ah/reference/saha_goebel_2009.txt filter=lfs diff=lfs merge=lfs -text
+materials_project_elastic_birch_murnaghan__K_VRH/reference/dejong_2015_scidata.txt filter=lfs diff=lfs merge=lfs -text
+metabolic_scaling_anage__log_bmr_W/reference/de_magalhaes2024_anage.txt filter=lfs diff=lfs merge=lfs -text
+metabolic_scaling_anage__log_bmr_W/reference/demagalhaes_2024.txt filter=lfs diff=lfs merge=lfs -text
+metabolic_scaling_anage__log_bmr_W/reference/west2002.txt filter=lfs diff=lfs merge=lfs -text
+metabolic_scaling_anage__log_bmr_W/reference/west_2004.txt filter=lfs diff=lfs merge=lfs -text
+nuclear_binding_energy_ame2020__BE_per_A/reference/goriely_2013.txt filter=lfs diff=lfs merge=lfs -text
+nuclear_binding_energy_ame2020__BE_per_A/reference/moller_1995.txt filter=lfs diff=lfs merge=lfs -text
+nuclear_binding_energy_ame2020__BE_per_A/reference/wang_2014.txt filter=lfs diff=lfs merge=lfs -text
+ocean_mesoscale_eddy_closure__Sx/reference/perezhogin_2023.txt filter=lfs diff=lfs merge=lfs -text
+ocean_mesoscale_eddy_closure__Sx/reference/tipping_2001.txt filter=lfs diff=lfs merge=lfs -text
+ocean_mesoscale_eddy_closure__Sx/reference/zanna_bolton_2020.txt filter=lfs diff=lfs merge=lfs -text
+pp_total_cross_section_donnachie_landshoff__sigma_tot/reference/block_halzen_2005.txt filter=lfs diff=lfs merge=lfs -text
+proton_em_form_factor__GE_over_GD/reference/arrington_2007.txt filter=lfs diff=lfs merge=lfs -text
+proton_em_form_factor__GE_over_GD/reference/bradford_2006.txt filter=lfs diff=lfs merge=lfs -text
+proton_em_form_factor__GE_over_GD/reference/ye_2018.txt filter=lfs diff=lfs merge=lfs -text
+protoplanetary_disk_mmflux__F_mm/reference/dalessio_2006.txt filter=lfs diff=lfs merge=lfs -text
+qg_turbulence_subgrid_pyqg__Sq/reference/zanna_bolton_2020.txt filter=lfs diff=lfs merge=lfs -text
+red_giant_asteroseismology__delta_nu/reference/Belkacem2011.txt filter=lfs diff=lfs merge=lfs -text
+red_giant_asteroseismology__delta_nu/reference/KjeldsenBedding1995.txt filter=lfs diff=lfs merge=lfs -text
+red_giant_asteroseismology__nu_max/reference/Belkacem2011.txt filter=lfs diff=lfs merge=lfs -text
+red_giant_asteroseismology__nu_max/reference/KjeldsenBedding1995.txt filter=lfs diff=lfs merge=lfs -text
+rock_core_permeability_norweg__log_k/reference/cabalar_akbulut_2016_perm_formulae.txt filter=lfs diff=lfs merge=lfs -text
+running_endurance_iaaf__velocity/reference/emig2020_running.txt filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/bentz_2015.txt filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/davis_2023.txt filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/jin_2023.txt filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/kormendy_2013.txt filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/mcconnell_2013.txt filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/sahu_2019.txt filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/tremaine_2002.txt filter=lfs diff=lfs merge=lfs -text
+smbh_mass_sigma_relation__log_M_BH/reference/vandenbosch_2016.txt filter=lfs diff=lfs merge=lfs -text
+sne_ia_distance_modulus_pantheonplus__mu/reference/chevallier_2001.txt filter=lfs diff=lfs merge=lfs -text
+sne_ia_distance_modulus_pantheonplus__mu/reference/riess_2022.txt filter=lfs diff=lfs merge=lfs -text
+solar_flat_plate_collector_graz_efficiency__eta_thermal/reference/tschopp_2023.md filter=lfs diff=lfs merge=lfs -text
+trade_gravity_cepii__log_tradeflow/reference/anderson_vanwincoop_2003.txt filter=lfs diff=lfs merge=lfs -text
+trade_gravity_cepii__log_tradeflow/reference/gaulier_zignago_2010.txt filter=lfs diff=lfs merge=lfs -text
+trade_gravity_cepii__log_tradeflow/reference/head_mayer_2014.txt filter=lfs diff=lfs merge=lfs -text
+tree_biomass_allometry_chave__agb/reference/chave_2014.txt filter=lfs diff=lfs merge=lfs -text
+ttauri_accretion__L_acc/reference/alcala_2014.txt filter=lfs diff=lfs merge=lfs -text
+ttauri_accretion__L_acc/reference/manara_2012.txt filter=lfs diff=lfs merge=lfs -text
+ttauri_accretion__L_acc/reference/natta_2006.txt filter=lfs diff=lfs merge=lfs -text
+wind_turbine_power_curve_engie__power_kW/reference/burton_2011.txt filter=lfs diff=lfs merge=lfs -text
+wind_turbine_power_curve_engie__power_kW/reference/carrillo_2013.txt filter=lfs diff=lfs merge=lfs -text
+wind_turbine_power_curve_engie__power_kW/reference/kragh_hansen_2014.txt filter=lfs diff=lfs merge=lfs -text
+wind_turbine_power_curve_engie__power_kW/reference/lydia_2014.txt filter=lfs diff=lfs merge=lfs -text
+wind_turbine_power_curve_engie__power_kW/reference/sohoni_2016.txt filter=lfs diff=lfs merge=lfs -text

baseball_pythagorean_winpct_lahman__win_pct/data/test.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d974472e98b7386716cb942788233a1cb3789f9cae81719eb2531984db61646
+size 15962

baseball_pythagorean_winpct_lahman__win_pct/data/train.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5718100d60c7ff2e1a0ed725df485e7bb5ad1410f26f7fdaf3c90491916e008
+size 34267

baseball_pythagorean_winpct_lahman__win_pct/formulas/__init__.py

@@ -0,0 +1,16 @@
+"""Reference-baseline formula bank for baseball_pythagorean_winpct_lahman__win_pct.
+
+Type I task — each module exposes the v2 contract:
+    USED_INPUTS, LAW_CONSTANTS, OTHER_CONSTANTS, LOCAL_FITTABLE (= {}),
+    predict(X, **law_constants)
+No fit() — Type I has no per-cluster parameters.
+
+Two reference baselines, both implementing the Pythagorean won-loss formula
+with different gamma exponents:
+  james_1980   — gamma = 2.0 (original Bill James 1980 canonical value)
+  miller_2007  — gamma = 1.79 (Miller 2007 least-squares mean, AL 2004)
+"""
+from importlib import import_module
+
+_STEMS = ("james_1980", "miller_2007")
+REGISTRY = {stem: import_module(f".{stem}", __name__) for stem in _STEMS}

baseball_pythagorean_winpct_lahman__win_pct/formulas/james_1980.py

@@ -0,0 +1,68 @@
+"""Pythagorean won-loss formula in the original Bill James (1980) formulation
+with the canonical exponent gamma = 2.
+
+James, B. (1980). The Bill James Baseball Abstract. Self-published.
+(Citation-only reference; no PDF mirrored. The canonical mathematical
+derivation is Miller 2007.)
+
+Formula
+-------
+Bill James' original "Pythagorean" formula (named for the sum-of-squares form):
+
+    win_pct = R^2 / (R^2 + RA^2)
+
+where R is season runs scored and RA is season runs allowed.
+This is the beta = 0 special case of Miller 2007 Eq. (1.2) with gamma = 2.
+
+The formula is cited and reproduced in Miller (2007), abstract and PDF p. 1
+(Introduction, first paragraph): "Initially in baseball the exponent gamma
+was taken to be 2 (which led to the name)."
+
+LAW_CONSTANTS
+-------------
+- gamma = 2: the canonical Bill James exponent (1980 Baseball Abstract),
+  cited in Miller (2007) PDF p. 1 Introduction paragraph 1.
+
+OTHER_CONSTANTS
+---------------
+None. The formula is dimensionless.
+
+Type designation: Type I. The exponent gamma = 2 is the universal constant
+for this baseline. LOCAL_FITTABLE = {}.
+
+Column mapping (paper -> CSV):
+  RS (runs scored per game / season) -> R  (column 1, season total)
+  RA (runs allowed per game / season) -> RA (column 2, season total)
+  Scale-invariant (as with all Pythagorean formulas; see miller_2007.py).
+
+Caveats:
+- The gamma = 2 constant is systematically too large for modern
+  scoring environments; the empirical optimum is near 1.82 (Miller).
+  This module ships the historical canonical value, not a refitted one.
+  It is expected to produce slightly higher RMSE than miller_2007.py.
+"""
+
+import numpy as np
+
+USED_INPUTS  = ["R", "RA"]
+PAPER_REF    = "summary_formula_miller_2007.md"
+EQUATION_LOC = "Bill James (1980); cited in Miller 2007 PDF p. 1 Introduction"
+
+LAW_CONSTANTS = {
+    "gamma": 2.0,    # James (1980) canonical exponent; cited in Miller 2007 PDF p. 1
+}
+OTHER_CONSTANTS = {}   # dimensionless formula
+LOCAL_FITTABLE  = {}   # Type I
+
+
+def predict(X: np.ndarray, gamma: float) -> np.ndarray:
+    """Pythagorean win percentage under the original Bill James gamma = 2.
+
+    X: (n, 2) — columns R (runs scored), RA (runs allowed).
+    Returns array of shape (n,) — win_pct in (0, 1).
+    """
+    R  = np.asarray(X[:, 0], dtype=float)
+    RA = np.asarray(X[:, 1], dtype=float)
+    Rg  = np.power(R,  gamma)
+    RAg = np.power(RA, gamma)
+    return Rg / (Rg + RAg)

baseball_pythagorean_winpct_lahman__win_pct/formulas/miller_2007.py

@@ -0,0 +1,78 @@
+"""Pythagorean won-loss formula from Miller (2007), using Miller's empirical
+best-fit gamma.
+
+Miller, S. J. (2007). A derivation of the Pythagorean won-loss formula in
+baseball. Chance, 20(1):40-48. American Statistical Association.
+Open-access mirror: https://web.williams.edu/Mathematics/sjmiller/public_html/
+399/handouts/PythagWonLoss_Paper.pdf
+
+Formula
+-------
+For season-total runs R (scored) and RA (runs allowed), with the Weibull
+location parameter beta negligible relative to season totals:
+
+    win_pct = R^gamma / (R^gamma + RA^gamma)          (simplified Eq. 1.2)
+
+Full formula with beta (Eq. 1.2, PDF p. 1; Theorem 2.2 / Eq. 2.6, PDF p. 4):
+
+    win_pct = (R - beta)^gamma / ((R - beta)^gamma + (RA - beta)^gamma)
+
+For season totals, |beta| ~ 0.5 * G (runs per game) is negligible relative
+to R, RA ~ 500-800, so this module sets beta = 0.
+
+LAW_CONSTANTS
+-------------
+- gamma = 1.79: mean best-fit exponent from the method of least squares
+  applied to the 14 American League teams of the 2004 baseball season
+  (Miller 2007, abstract and PDF p. 1 / p. 2 first paragraph).
+  Standard deviation 0.09. The maximum-likelihood estimate is 1.74 (std 0.06).
+  This module uses the least-squares mean 1.79 as the primary LAW constant.
+
+OTHER_CONSTANTS
+---------------
+None. The formula is dimensionless; no universal physics constants are needed.
+
+Type designation: Type I. The exponent gamma is a universal league-wide
+constant; no per-team or per-season fitting is performed in Miller's primary
+analysis. LOCAL_FITTABLE = {}.
+
+Column mapping (paper -> CSV):
+  RS (paper runs scored per game) -> R  (column 1, season total)
+  RA (paper runs allowed per game) -> RA (column 2, season total)
+  The formula is scale-invariant: R^gamma/(R^gamma+RA^gamma) =
+  (R/G)^gamma / ((R/G)^gamma + (RA/G)^gamma), so season totals and
+  per-game averages give the same result.
+
+Caveats:
+- Miller fits beta = -0.5 runs/game for the per-game discrete correction;
+  at season-total scale (R ~ 600, RA ~ 600, G ~ 162) the correction is
+  |beta_season| = 0.5 * 162 ~ 81, which is ~13% of R -- not entirely
+  negligible. This module ships gamma frozen at the paper's value and sets
+  beta = 0 (season-total approximation), consistent with the canonical
+  Bill James formulation. The sister module james_1980.py uses gamma = 2.
+"""
+
+import numpy as np
+
+USED_INPUTS  = ["R", "RA"]
+PAPER_REF    = "summary_formula_miller_2007.md"
+EQUATION_LOC = "Eq. (1.2) PDF p. 1; Theorem 2.2 / Eq. (2.6) PDF p. 4"
+
+LAW_CONSTANTS = {
+    "gamma": 1.79,   # least-squares mean over 14 AL 2004 teams; Miller 2007 abstract + PDF p. 1-2
+}
+OTHER_CONSTANTS = {}   # dimensionless formula; no external physics constants needed
+LOCAL_FITTABLE  = {}   # Type I — no per-cluster parameters
+
+
+def predict(X: np.ndarray, gamma: float) -> np.ndarray:
+    """Pythagorean win percentage under Miller 2007's best-fit gamma.
+
+    X: (n, 2) — columns R (runs scored), RA (runs allowed).
+    Returns array of shape (n,) — win_pct in (0, 1).
+    """
+    R  = np.asarray(X[:, 0], dtype=float)
+    RA = np.asarray(X[:, 1], dtype=float)
+    Rg  = np.power(R,  gamma)
+    RAg = np.power(RA, gamma)
+    return Rg / (Rg + RAg)

baseball_pythagorean_winpct_lahman__win_pct/formulas/reference_metrics.json

@@ -0,0 +1,61 @@
+{
+  "baselines": {
+    "james_1980": {
+      "equation_loc": "Bill James (1980); cited in Miller 2007 PDF p. 1 Introduction",
+      "error": null,
+      "failed": false,
+      "kind": "reference",
+      "law_constants": {
+        "gamma": 2.0
+      },
+      "local_fittable": [],
+      "metrics": {
+        "log_mae": 0.018073578706717868,
+        "mae": 0.02035253889196662,
+        "mape": 0.0416209828963854,
+        "mdae": 0.016923316811408906,
+        "mse": 0.0006665903205117677,
+        "n_finite": 938,
+        "r2": 0.8723463499046444,
+        "rmse": 0.02581841049545397,
+        "smape": 0.041594077969012565
+      },
+      "other_constants": {},
+      "paper_ref": "summary_formula_miller_2007.md"
+    },
+    "miller_2007": {
+      "equation_loc": "Eq. (1.2) PDF p. 1; Theorem 2.2 / Eq. (2.6) PDF p. 4",
+      "error": null,
+      "failed": false,
+      "kind": "reference",
+      "law_constants": {
+        "gamma": 1.79
+      },
+      "local_fittable": [],
+      "metrics": {
+        "log_mae": 0.0179324763805722,
+        "mae": 0.02012816862599297,
+        "mape": 0.041496949291881115,
+        "mdae": 0.01727645471286568,
+        "mse": 0.000643510981005811,
+        "n_finite": 938,
+        "r2": 0.8767660989454998,
+        "rmse": 0.025367518227170174,
+        "smape": 0.04127015407815537
+      },
+      "other_constants": {},
+      "paper_ref": "summary_formula_miller_2007.md"
+    }
+  },
+  "derived_caps": {
+    "fit_timeout_seconds": null,
+    "max_init_size_per_param": 1,
+    "max_law_constants": 1,
+    "max_local_params": 0
+  },
+  "metric_declared": "rmse",
+  "n_test_rows": 938,
+  "reference_baseline_id": null,
+  "task": "baseball_pythagorean_winpct_lahman__win_pct",
+  "type": "typeI"
+}

baseball_pythagorean_winpct_lahman__win_pct/metadata.yaml

@@ -0,0 +1,119 @@
+task_id: baseball_pythagorean_winpct_lahman__win_pct
+domain: social_science / sports_analytics
+license: CC-BY-SA-3.0
+
+# Type I — each franchise-season is an independent row; the Pythagorean
+# formula uses a single universal exponent (no per-team or per-season refit).
+type: typeI
+has_group_id: false
+
+context: >-
+  This task is drawn from Major League Baseball, where each team plays a full season of games and scores and allows runs.
+
+target:
+  name: win_pct
+  symbol: "W%"
+  unit: ""
+  description: Season win percentage — fraction of decided games won, W / (W + L).
+  range: [0.12, 0.90]
+
+inputs:
+  - name: R
+    symbol: RS
+    unit: runs
+    description: Season total runs scored by the team.
+    range: [219, 1220]
+  - name: RA
+    symbol: RA
+    unit: runs
+    description: Season total runs allowed by the team.
+    range: [209, 1252]
+
+data_files:
+  train: data/train.csv   # seasons 1872-1989 (historical era)
+  test:  data/test.csv    # seasons 1990-2021 (modern expansion era, OOD probe)
+
+n_train: 2012
+n_test:  938
+
+# Optional audit note (not shown to SR system):
+# Temporal split: train = yearID <= 1989, test = yearID >= 1990.
+# The test set spans the modern high-scoring era (steroids era peak ~5.4 R/G
+# in 1996-2000 and pitcher-dominant era ~4.0 R/G in the 2010s) — the formula
+# claims universality across eras, which this OOD split tests.
+# The training set includes low-scoring deadball era seasons (~3.5 R/G in
+# the 1900s-1910s) plus moderate mid-century play.
+
+# Audit history:
+#   C12 2026-05-26: removed priors gamma_james=2.0 and gamma_miller_ls=1.79 (numerical leak
+#     with james_1980/miller_2007 LAW['gamma']; pure empirical regression exponents, no
+#     upstream physical constant). Replaced with sports-analytics distractors below.
+#   H1  2026-05-26: removed prior gamma_miller_ml=1.74 (semantic leak — same physical quantity
+#     as LAW gamma, just a different point estimate of the Pythagorean exponent).
+#   2026-05-28: removed davenport_2014 baseline (unsourced LAW a=1.50; σ-of-log10 form gives
+#     effective γ≈0.65 vs Davenport-Woolner 1999 published γ≈1.83 in direct-Pythagorean form;
+#     no PDF on disk; half-integrated wiring).
+# Final priors: 3 distractors, 0 candidates.
+priors:
+  - name: games_per_season_modern
+    value: 162.0
+    unit: games
+    description: Number of regular-season games per team in the modern MLB schedule (since 1961 AL / 1962 NL). Not consumed by the Pythagorean formula, which is scale-invariant in R and RA.
+    source: Historical MLB schedule; not used directly in the Pythagorean formula (formula is scale-invariant in R, RA).
+    _role: distractor
+  - name: innings_per_game
+    value: 9.0
+    unit: innings
+    description: Standard number of innings in a baseball game. A structural game-rule constant unrelated to the run-ratio power law.
+    source: Official Baseball Rules, Rule 9.03(a); MLB Official Rules (any edition).
+    _role: distractor
+  - name: batting_avg_lg_typical
+    value: 0.260
+    unit: ""
+    description: Approximate long-run MLB league batting average, used in offensive-efficiency models (sabermetrics). Not a parameter of the Pythagorean run-ratio formula.
+    source: Thorn & Palmer (1984), The Hidden Game of Baseball, p. 64; historical MLB statistics (Retrosheet).
+    _role: distractor
+
+references:
+  - id: james_1980
+    label: "James (1980) — original Pythagorean formula, gamma = 2"
+    formula_file: formulas/james_1980.py
+    reference_pdf: reference/miller_2007.pdf
+    n_law_constants: 1
+    n_other_constants: 0
+    n_local_params: 0
+    measured:
+      rmse: 0.0258
+      r2: 0.8723
+  - id: miller_2007
+    label: "Miller (2007) — Pythagorean formula, gamma = 1.79 (least-squares)"
+    formula_file: formulas/miller_2007.py
+    reference_pdf: reference/miller_2007.pdf
+    n_law_constants: 1
+    n_other_constants: 0
+    n_local_params: 0
+    measured:
+      rmse: 0.0254
+      r2: 0.8768
+
+caps:
+  max_law_constants:        1
+  max_local_params:         0
+  max_init_size_per_param:  1
+  fit_timeout_seconds:      null
+
+metric: rmse
+# === Validity rubric ===
+# Frozen list — the scoring-time judge (eval_consistency.py) scores a
+# submission by how many of THESE criteria it satisfies:
+# validity_score = (criteria satisfied) / len(validity_rubrics).
+# De-leaked context + comprehensive multi-baseline extraction, 2026-06-01.
+validity_rubrics:
+  - "base expected winning percentage on the relative run-scoring balance between runs scored and runs allowed"
+  - "predicted win_pct stays within [0, 1]"
+  - "predicted win_pct does not decrease when R increases with RA fixed"
+  - "predicted win_pct does not increase when RA increases with R fixed"
+  - "predicted win_pct is above 0.5 when R exceeds RA and below 0.5 when R is less than RA"
+  - "swapping R and RA maps the prediction to its complement about 0.5"
+
+best_baseline: 0.0253675    # miller_2007 — rmse (lower is better); regenerated 2026-05-30

baseball_pythagorean_winpct_lahman__win_pct/prep_data.py

@@ -0,0 +1,234 @@
+"""prep_data.py — baseball_pythagorean_winpct_lahman__win_pct
+
+=== SOURCE ===
+data_raw/Teams.csv: 2985 raw rows (header + 2984 data rows, covering 1871-2021)
+produced directly from the Lahman Baseball Database GitHub mirror
+(https://github.com/cbwinslow/baseballdatabank).
+sha256(data_raw/Teams.csv): 827bef5af12158eeb3d18a4a3f54ff846e395965b47c09b2564ad7bb29c5e545
+License: CC-BY-SA-3.0 (Lahman Baseball Database).
+
+=== CANONICAL COLUMN PROJECTION ===
+Column 0  win_pct  : float, dimensionless — win fraction W/(W+L) (SR target)
+Column 1  R        : int → float, runs — season total runs scored by the team
+Column 2  RA       : int → float, runs — season total runs allowed by the team
+
+Dropped from working CSV (not written to released files):
+  'W'          — tautology leak: win_pct = W/(W+L) by definition (data_spec §9.4)
+  'L'          — same leak (same reason)
+  'yearID'     — auxiliary bookkeeping; not a declared SR input
+  'lgID'       — categorical; not a declared SR input
+  'teamID'     — string id; not a declared SR input
+  'G'          — games played; auxiliary (not in any baseline's USED_INPUTS)
+  All other Lahman columns (HR, BB, ERA, …) — not relevant to Pythagorean formula
+
+=== TYPE I vs TYPE II DECISION ===
+Verdict: TYPE I (group_id column NOT emitted).
+
+Criteria checked:
+(a) Row-per-identifier ratio: one row per franchise-season; rows across
+    different team-seasons are fully interchangeable in the formula. ✗ Type II.
+(b) LOCAL_FITTABLE in formulas/: both formulas/james_1980.py and
+    formulas/miller_2007.py have LOCAL_FITTABLE = {} — no per-cluster
+    fitted parameters. ✗ Type II.
+(c) Summary declarations: summary_formula_miller_2007.md states gamma is
+    a universal league-wide constant, not a per-team or per-cluster
+    parameter; Miller (2007) fits one gamma across all 14 AL 2004 teams
+    collectively. ✗ Type II.
+(d) Natural cluster identifier: team-season rows have a 'teamID' cluster,
+    but the Pythagorean formula has no per-team refit — gamma is global
+    to the league. ✗ Type II.
+
+All four Type II criteria fail → TYPE I.
+
+=== TRAIN / TEST SPLIT ===
+Tier 1 (dataset-side authority):
+  Inspected: no summary_dataset_*.md present (Miller 2007 is the sole
+  reference, and it is a formula paper; the Lahman database has no
+  prescribed train/test split).
+  → EMPTY.
+
+Tier 2 (formula-side authority):
+  Inspected: summary_formula_miller_2007.md.
+  Finding: Miller (2007) uses only the 14 American League teams of the
+  2004 season for calibration. No prescribed train/test split for a
+  multi-season benchmark. → EMPTY.
+
+Tier 3 (LLM-proposed OOD split — temporal):
+  Chosen axis: yearID (season year), deterministic cutoff.
+  Rule: train = yearID < 1990 (seasons 1872–1989);
+        test  = yearID >= 1990 (seasons 1990–2021).
+
+  Justification:
+  - Temporal OOD is the natural axis for a team-season dataset spanning
+    150 years: the run environment has changed substantially (deadball
+    era ~3.5 R/G in the 1900s-1910s; live-ball era surge post-1920;
+    steroids era peak ~5.4 R/G in 1996-2000; pitcher-dominant era
+    ~4.0 R/G in the 2010s). Training on pre-1990 data and testing on
+    1990-2021 probes the formula's extrapolation into the modern
+    high-scoring era and subsequent correction.
+  - The Pythagorean formula is claimed to be universal (Miller 2007
+    derives it from assumptions about the run distribution, not from
+    era-specific fitting). A temporal split exercises whether that
+    universality holds.
+  - The split produces a train set (1872-1989) larger than the test
+    set (1990-2021), consistent with data_spec §8 check on n_train >
+    n_test.
+  - Cutoff at 1990 is deterministic and physically motivated: 1990 is
+    the start of the modern expansion era (Florida Marlins, Colorado
+    Rockies in 1993; wildcard era in 1994; significant park construction
+    affecting scoring). Any post-1900 cutoff would be defensible; 1990
+    cleanly separates the large historical record from the modern era.
+  - OOD claim: test set probes the Pythagorean formula in the modern
+    scoring environment (R/G ~ 4.2-5.5, with steroids-era peaks) from
+    a train set dominated by low-scoring historical eras and moderate
+    mid-century play.
+
+Split counts:
+  train: 2012 rows  (yearID 1872-1989)
+  test:  938 rows   (yearID 1990-2021)
+"""
+
+from __future__ import annotations
+
+import hashlib
+import sys
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+HERE = Path(__file__).resolve().parent
+RAW = HERE / "data_raw" / "Teams.csv"
+OUT = HERE / "data"
+
+# ---------------------------------------------------------------------------
+# Constants
+# ---------------------------------------------------------------------------
+EXPECTED_SHA256 = "827bef5af12158eeb3d18a4a3f54ff846e395965b47c09b2564ad7bb29c5e545"
+MIN_W_PLUS_L    = 50        # minimum decided games; excludes ultra-short seasons
+TRAIN_YEAR_MAX  = 1989      # inclusive; train = yearID <= TRAIN_YEAR_MAX
+TEST_YEAR_MIN   = 1990      # inclusive; test  = yearID >= TEST_YEAR_MIN
+OUT_COLS        = ["win_pct", "R", "RA"]
+
+
+# ---------------------------------------------------------------------------
+# Load and verify
+# ---------------------------------------------------------------------------
+def load_raw(fp: Path) -> pd.DataFrame:
+    if not fp.is_file():
+        raise FileNotFoundError(f"Raw data not found: {fp}\nRun data_raw/download.sh first.")
+    # Hash check (soft warning)
+    sha = hashlib.sha256(fp.read_bytes()).hexdigest()
+    if sha != EXPECTED_SHA256:
+        print(f"  WARNING: sha256 mismatch on {fp.name}\n"
+              f"    expected: {EXPECTED_SHA256}\n"
+              f"    got:      {sha}", file=sys.stderr)
+    return pd.read_csv(fp)
+
+
+# ---------------------------------------------------------------------------
+# Assemble working DataFrame
+# ---------------------------------------------------------------------------
+def assemble(raw: pd.DataFrame) -> pd.DataFrame:
+    required = ["yearID", "lgID", "teamID", "W", "L", "R", "RA", "G"]
+    for c in required:
+        if c not in raw.columns:
+            raise ValueError(f"Required column '{c}' missing from Teams.csv")
+
+    df = raw[required].copy()
+    df = df.dropna(subset=["W", "L", "R", "RA", "G"])
+
+    # Cast to int; drop non-positive R/RA and negative W/L
+    for c in ("W", "L", "R", "RA", "G"):
+        df[c] = df[c].astype(int)
+    df = df[(df["W"] >= 0) & (df["L"] >= 0)]
+    df = df[(df["R"] > 0) & (df["RA"] > 0)]
+
+    # Drop ultra-short franchise-seasons (folded teams, pre-1876 NA seasons)
+    df = df[(df["W"] + df["L"]) >= MIN_W_PLUS_L].copy()
+
+    # Compute SR target
+    df["win_pct"] = df["W"] / (df["W"] + df["L"])
+
+    return df.reset_index(drop=True)
+
+
+# ---------------------------------------------------------------------------
+# Sanity checks
+# ---------------------------------------------------------------------------
+def sanity_checks(df: pd.DataFrame) -> None:
+    assert (df["win_pct"] > 0).all(), "win_pct must be > 0"
+    assert (df["win_pct"] < 1).all(), "win_pct must be < 1"
+    assert (df["R"] >= 100).all(), f"unexpectedly small R: {df['R'].min()}"
+    assert (df["R"] <= 1500).all(), f"unexpectedly large R: {df['R'].max()}"
+    assert (df["RA"] >= 100).all(), f"unexpectedly small RA: {df['RA'].min()}"
+    assert (df["RA"] <= 1500).all(), f"unexpectedly large RA: {df['RA'].max()}"
+
+    # Cross-check: Bill James gamma=2 median |residual| should be < 0.04
+    pred = df["R"] ** 2 / (df["R"] ** 2 + df["RA"] ** 2)
+    mae = (pred - df["win_pct"]).abs().median()
+    print(f"  sanity: gamma=2 Pythagorean median |residual| = {mae:.4f}")
+    assert mae < 0.04, f"Pythagorean gamma=2 fit unexpectedly poor (median |err|={mae:.4f})"
+
+
+# ---------------------------------------------------------------------------
+# Temporal split (deterministic, no random seed)
+# ---------------------------------------------------------------------------
+def split(df: pd.DataFrame) -> dict[str, pd.DataFrame]:
+    """Deterministic temporal OOD split.
+
+    train: yearID <= 1989 (historical era, 1872-1989)
+    test:  yearID >= 1990 (modern era, 1990-2021)
+
+    See the docstring TRAIN/TEST SPLIT section for full justification.
+    The split is deterministic: no randomness, reproducible by year cutoff.
+    """
+    train = df[df["yearID"] <= TRAIN_YEAR_MAX].copy()
+    test  = df[df["yearID"] >= TEST_YEAR_MIN].copy()
+    assert len(train) > len(test), f"n_train={len(train)} must exceed n_test={len(test)}"
+    assert len(train) + len(test) == len(df), "train + test must equal total rows"
+    return {"train": train, "test": test}
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+def main() -> int:
+    raw = load_raw(RAW)
+    df = assemble(raw)
+    print(f"assembled {len(df)} franchise-seasons after filtering "
+          f"(W+L >= {MIN_W_PLUS_L}; "
+          f"year range {df['yearID'].min()}-{df['yearID'].max()}).")
+    sanity_checks(df)
+
+    parts = split(df)
+    OUT.mkdir(parents=True, exist_ok=True)
+
+    for split_name in ("train", "test"):
+        sub = parts[split_name][OUT_COLS].copy()
+        out_fp = OUT / f"{split_name}.csv"
+        sub.to_csv(out_fp, index=False, lineterminator="\n",
+                   float_format="%.6f")
+        print(f"wrote {out_fp}: {len(sub)} rows  "
+              f"(win_pct {sub['win_pct'].min():.4f}..{sub['win_pct'].max():.4f})")
+
+    # Verify columns on disk
+    for split_name in ("train", "test"):
+        fp = OUT / f"{split_name}.csv"
+        df2 = pd.read_csv(fp)
+        assert list(df2.columns) == OUT_COLS, \
+            f"{split_name}: expected {OUT_COLS}, got {list(df2.columns)}"
+    print("column-order check: OK")
+
+    # Print md5sums
+    import hashlib as hl
+    for split_name in ("train", "test"):
+        fp = OUT / f"{split_name}.csv"
+        h = hl.md5(fp.read_bytes()).hexdigest()
+        print(f"  md5({split_name}.csv) = {h}")
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

baseball_pythagorean_winpct_lahman__win_pct/reference/README.md

@@ -0,0 +1,21 @@
+# reference — baseball_pythagorean_winpct_lahman__win_pct
+
+| File | Reference | Note |
+|------|-----------|------|
+| `miller_2007.pdf` | Steven J. Miller (2007). "A Derivation of the Pythagorean Won-Loss Formula in Baseball." *Chance* 20(1):40–48. American Statistical Association. | **Primary baseline.** Provides the Weibull-distribution derivation of the Pythagorean formula; Eq. (1.2) PDF p. 1 and Theorem 2.2 / Eq. (2.6) PDF p. 4. Empirical gamma = 1.79 (least-squares). |
+| `miller_2007.txt` | Text extract of `miller_2007.pdf` via `pdf_to_text.py`. | Auto-generated; used for coefficient grepping. |
+| `miller_2007.bib` | BibTeX entry for Miller (2007). | `refkey = miller2007pythag`. |
+| `refs.bib` | Combined bibliography for all references cited in this task (Miller 2007, James 1980 citation, Davenport-Woolner 1999 citation, Lahman DB). | Retained from SRC for completeness; per-paper `.bib` files are the authoritative splits. |
+| `summary_formula_miller_2007.md` | Page-level summary of Miller (2007) for this benchmark. | Records Eq. (1.2) provenance, gamma values, and column mapping. |
+
+---
+
+**DOIs / URLs:**
+
+- Miller 2007 open-access author mirror:
+  https://web.williams.edu/Mathematics/sjmiller/public_html/399/handouts/PythagWonLoss_Paper.pdf
+- James (1980) Baseball Abstract: citation-only; no DOI or PDF mirrored.
+- Davenport-Woolner (1999) Baseball Prospectus:
+  https://www.baseballprospectus.com/news/article/342/ — citation-only.
+- Lahman Database: https://sabr.org/lahman-database/ ;
+  https://github.com/cbwinslow/baseballdatabank

baseball_pythagorean_winpct_lahman__win_pct/reference/miller_2007.bib

@@ -0,0 +1,12 @@
+@article{miller2007pythag,
+  author       = {Steven J. Miller},
+  title        = {A Derivation of the {Pythagorean} Won-Loss Formula in Baseball},
+  journal      = {Chance},
+  volume       = {20},
+  number       = {1},
+  pages        = {40--48},
+  year         = {2007},
+  publisher    = {American Statistical Association},
+  url          = {https://web.williams.edu/Mathematics/sjmiller/public_html/399/handouts/PythagWonLoss_Paper.pdf},
+  note         = {Open-access author mirror (Williams College). Provides the Weibull-distribution derivation of Bill James' Pythagorean Won-Loss formula: when runs scored and runs allowed per game are independent random variables drawn from Weibull distributions with the same (beta, gamma) but different alpha, the team's expected winning percentage is (RS-beta)^gamma / ((RS-beta)^gamma + (RA-beta)^gamma) (Theorem 2.2, p. 4). Equations 2.1-2.3 give the Weibull density and the mean integrals leading to this result; Eq. (1.2) in the Introduction is the formula's algebraic form. For aggregate season-total runs the Weibull location parameter beta is small relative to RS, RA, and the formula collapses to the classical W% = R^gamma / (R^gamma + RA^gamma) with gamma about 1.74-1.82 (least-squares / max-likelihood, Miller PDF p. 1-2) -- in the vicinity of the canonical Bill James gamma = 2.}
+}

baseball_pythagorean_winpct_lahman__win_pct/reference/miller_2007.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8ecb3567060de9be00754733972a6f0a785ec26e65d72d40dd3c365f9d5298f
+size 324923

baseball_pythagorean_winpct_lahman__win_pct/reference/miller_2007.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1965ee2fb8feb84fb558c5c05686cea9e0101a02ae699c83be19db749f8a7c28
+size 40048

baseball_pythagorean_winpct_lahman__win_pct/reference/refs.bib

@@ -0,0 +1,46 @@
+% Reference bibliography for the baseball Pythagorean win-percentage SR task.
+%
+% Single OA primary reference is Miller 2007 (Chance), whose author-mirror
+% PDF at Williams College is bundled in this directory as miller_2007.pdf.
+% The James (1980) "Baseball Abstract" and Davenport-Woolner (1999)
+% "Revisiting the Pythagorean Theorem" entries are citation-only (no PDF
+% is mirrored); they are the historical origins of the formula and the
+% gamma=1.83 refinement respectively.
+
+@article{miller2007pythag,
+  author       = {Steven J. Miller},
+  title        = {A Derivation of the {Pythagorean} Won-Loss Formula in Baseball},
+  journal      = {Chance},
+  volume       = {20},
+  number       = {1},
+  pages        = {40--48},
+  year         = {2007},
+  publisher    = {American Statistical Association},
+  url          = {https://web.williams.edu/Mathematics/sjmiller/public_html/399/handouts/PythagWonLoss_Paper.pdf},
+  note         = {Open-access author mirror (Williams College).  Provides the Weibull-distribution derivation of Bill James' Pythagorean Won-Loss formula: when runs scored and runs allowed per game are independent random variables drawn from Weibull distributions with the same (beta, gamma) but different alpha, the team's expected winning percentage is (RS-beta)^gamma / ((RS-beta)^gamma + (RA-beta)^gamma) (Theorem 2.2, p. 4).  Equations 2.1-2.3 give the Weibull density and the mean integrals leading to this result; Eq. (1.2) in the Introduction is the formula's algebraic form.  For aggregate season-total runs the Weibull location parameter beta is small relative to RS, RA, and the formula collapses to the classical W% = R^gamma / (R^gamma + RA^gamma) with gamma about 1.74-1.82 (least-squares / max-likelihood, Miller PDF p. 1-2) -- in the vicinity of the canonical Bill James gamma = 2.}
+}
+
+@misc{james1980abstract,
+  author       = {Bill James},
+  title        = {The {Bill James} Baseball Abstract},
+  year         = {1980},
+  publisher    = {Self-published},
+  note         = {Historical origin of the Pythagorean Won-Loss formula in baseball, with the original exponent gamma = 2.  Citation-only; no PDF is mirrored.  Bill James reformulated and refined the formula across many subsequent Baseball Abstracts; the canonical mathematical derivation is Miller 2007.}
+}
+
+@misc{davenport1999pythag,
+  author       = {Clay Davenport and Keith Woolner},
+  title        = {Revisiting the {Pythagorean Theorem}: Putting {Bill James'} Pythagorean Theorem to the Test},
+  howpublished = {Baseball Prospectus (online)},
+  year         = {1999},
+  url          = {https://www.baseballprospectus.com/news/article/342/revisiting-the-pythagorean-theorem-putting-bill-james-pythagorean-theorem-to-the-test/},
+  note         = {Refines the exponent to gamma about 1.83 by fitting MLB team-season records.  Citation-only; no PDF is mirrored.  This is the source of the modern saber-metrics "Pythagenport" exponent that improves on Bill James' gamma = 2 in modern high-scoring eras.}
+}
+
+@misc{lahman2025db,
+  author       = {Sean Lahman and {Chadwick Baseball Bureau}},
+  title        = {Lahman Baseball Database ({baseballdatabank}), {Teams.csv}},
+  year         = {2025},
+  howpublished = {GitHub mirror: \url{https://github.com/cbwinslow/baseballdatabank}; SABR landing page: \url{https://sabr.org/lahman-database/}},
+  note         = {The canonical historical Major League Baseball record (1871-2021 in this mirror; SABR's official 2025 release goes through the 2024 season).  Distributed under the Creative Commons Attribution-ShareAlike 3.0 Unported License (CC-BY-SA-3.0).  Derivative files (data/train.csv, data/test.csv) inherit the same license.}
+}

baseball_pythagorean_winpct_lahman__win_pct/reference/summary_formula_miller_2007.md

@@ -0,0 +1,149 @@
+# Miller 2007 -- A Derivation of the Pythagorean Won-Loss Formula in Baseball
+
+**File:** `reference/miller_2007.pdf` (325 kB, 3 pages excerpt covering the
+derivation section -- the published *Chance* version is 9 pages; the
+Williams College open mirror serves the technical core).
+**Venue:** Steven J. Miller, *Chance* 20(1):40-48 (2007), American
+Statistical Association.
+**Open-access URL:** https://web.williams.edu/Mathematics/sjmiller/public_html/399/handouts/PythagWonLoss_Paper.pdf
+**License of the paper:** author pre-print, hosted by the Williams
+College Mathematics Department.  Freely served.
+
+This summary is the per-paper companion to `metadata.yaml`.  It records
+the page-level provenance of every coefficient and equation the
+benchmark uses.
+
+## Cover-page verification (`pdftotext reference/miller_2007.pdf - | head`)
+
+```
+A DERIVATION OF THE PYTHAGOREAN WON-LOSS FORMULA IN
+BASEBALL
+STEVEN J. MILLER
+Abstract.  It has been noted that in many professional sports leagues
+a good predictor of a team's end of season won-loss percentage is
+Bill James' Pythagorean Formula RS^gamma / (RS^gamma + RA^gamma),
+where RS_obs (resp. RA_obs) is the observed average number of runs
+scored (allowed) per game and gamma is a constant for the league;
+for baseball the best agreement is when gamma is about 1.82.
+```
+
+Title, author, and venue match `refs.bib` entry `miller2007pythag`.
+
+## The formula used by this benchmark
+
+Source: PDF page 1, Eq. (1.2):
+
+```
+                                       (RS - beta)^gamma
+Won-Loss Percentage(RS, RA, beta, gamma) = -----------------------------------
+                                       (RS - beta)^gamma + (RA - beta)^gamma
+```
+
+For full-season run totals (R, RA in the hundreds, beta of order
+-0.5 runs per game per Miller's choice), beta is negligible relative
+to (R, RA) and the formula collapses to the classical Bill James form:
+
+```
+                        R^gamma
+       win_pct = -----------------
+                R^gamma + RA^gamma
+```
+
+This is the closed form used by `formulas/james_1980.py`
+(with gamma = 2) and `formulas/miller_2007.py` (with gamma = 1.79,
+Miller's least-squares-best mean from 14 AL 2004 teams; also
+bracketed by his max-likelihood-best gamma = 1.74 +/- 0.06).
+The exponent gamma is the **structural universal constant invariant**
+in this Type I task.
+
+## Derivation -- Eq. (2.1), (2.2), (2.3) of Miller 2007
+
+The derivation lives in Section 2.  Starting from the three-parameter
+Weibull density (PDF Eq. (1.1)):
+
+* **Eq. (2.1)** (PDF p. 4, line 200): the substitution
+  u = ((x - beta) / alpha)^gamma yields the indefinite integral for
+  the Weibull mean integral
+  E[X] = integral over [beta, infinity) of x * f(x; alpha, beta, gamma) dx.
+* **Eq. (2.2)** (PDF p. 4, line 221): the closed-form Weibull mean
+  mu_{alpha, beta, gamma} = beta + alpha * Gamma(1 + 1/gamma); and
+  variance sigma^2_{alpha, beta, gamma} = alpha^2 * [Gamma(1 + 2/gamma)
+  - Gamma(1 + 1/gamma)^2].  Both are stated formally as Lemma 2.1
+  on PDF p. 4 line 316.
+* **Eq. (2.3)** (PDF p. 4, line 256): the probability integral
+  P(X > Y) for two independent Weibulls with parameters
+  (alpha_RS, beta, gamma) and (alpha_RA, beta, gamma) reduces (via
+  the same substitution) to
+  alpha_RS^gamma / (alpha_RS^gamma + alpha_RA^gamma).
+
+**Theorem 2.2 (Pythagorean Won-Loss Formula; PDF p. 4 line 338-341):**
+
+> "Let the runs scored and runs allowed per game be two independent
+> random variables drawn from Weibull distributions with parameters
+> (alpha_RS, beta, gamma) and (alpha_RA, beta, gamma) [...] If
+> gamma > 0 then
+> [P(team wins)] = (RS - beta)^gamma / ((RS - beta)^gamma +
+> (RA - beta)^gamma)."
+
+Substituting alpha_RS = RS - beta and alpha_RA = RA - beta recovers
+Eq. (1.2) above.
+
+## Numerical anchor: best-fit gamma values
+
+PDF p. 2 first paragraph and abstract:
+
+* Bill James' canonical exponent: gamma = 2 (the source of the
+  formula's name; "Pythagorean" because of the sum-of-squares form).
+* Miller's empirically best exponent (aggregate MLB season-end
+  records): gamma about 1.82 (Miller 2007 abstract, PDF p. 1).
+* Miller's best-fit gamma from the 14 2004 American-League teams:
+  - method of least squares: mean gamma = 1.79, std 0.09;
+  - method of maximum likelihood: mean gamma = 1.74, std 0.06.
+
+All three values agree within +/- 0.1; the structural invariant is the
+*form* `R^gamma / (R^gamma + RA^gamma)`, not the precise numerical
+value of gamma (which depends only weakly on the run environment).
+
+## Mapping to this benchmark
+
+| benchmark symbol | source-paper symbol              | meaning                                |
+|------------------|----------------------------------|----------------------------------------|
+| `win_pct`        | RS^gamma / (RS^gamma + RA^gamma) | team-season win percentage W / (W + L) |
+| `R`              | RS                               | runs scored in the season              |
+| `RA`             | RA                               | runs allowed in the season             |
+| `gamma`          | gamma                            | universal Bill James exponent (= 2; empirically best ~1.82 per Miller 2007 p. 1) |
+
+The benchmark uses season totals for R, RA, W, L (one row per
+franchise-season).  Miller's derivation uses per-game means; the
+formula is invariant under per-game-mean = season-total / G, so
+applying it to season totals is mathematically identical to applying
+it to per-game means once G cancels.
+
+## What is NOT done by this task
+
+* The Pythagenpat refinement `gamma = ((R+RA)/G)^0.287` (Smyth-Patriot
+  2003) is the subject of a possible sister task
+  (`pythagenpat_winpct_lahman__win_pct`) and is intentionally NOT
+  included here -- this is the strict Type I task with the universal
+  scalar gamma.  See the formula module `formulas/miller_2007.py` for
+  implementation details.
+* The Weibull location parameter beta is set to zero in this
+  benchmark.  Miller's beta = -0.5 runs per game is an order of
+  magnitude smaller than R, RA in season totals and contributes
+  approximately 1.2% to the predicted win_pct -- below the per-row
+  natural noise of a 162-game season.
+
+## Citation
+
+```
+@article{miller2007pythag,
+  author    = {Steven J. Miller},
+  title     = {A Derivation of the Pythagorean Won-Loss Formula in Baseball},
+  journal   = {Chance},
+  volume    = {20},
+  number    = {1},
+  pages     = {40--48},
+  year      = {2007},
+  publisher = {American Statistical Association}
+}
+```

binary_pulsar_post_keplerian_atnf__PBDOT/data/report.md

@@ -0,0 +1,195 @@
+# Data Construction Report — binary_pulsar_post_keplerian_atnf__PBDOT
+
+## Source
+
+| Field | Value |
+|-------|-------|
+| File | `data_raw/atnf_binary_pulsar_pk.csv` |
+| Produced by | `data_raw/parse_psrcat.py` from ATNF Pulsar Catalogue v2.7.0 |
+| Dataset paper | Manchester et al. (2005), *AJ* 129, 1993 — DOI 10.1086/428488 |
+| Data rows | 91 |
+| sha256 | `39d4efe1057276295f6e6ba858115e1cb38f3e41388d48f835a9c9aef829c18d` |
+
+The raw CSV contains binary pulsars with a measured orbital period derivative (PBDOT) extracted
+from the ATNF Pulsar Catalogue v2.7.0.  All 91 rows have a non-NaN PBDOT value; other columns
+(m1, m2, omdot, sini, gamma) are NaN for the majority of rows where the corresponding PK
+parameter has not been measured.
+
+---
+
+## Schema
+
+### Canonical columns (released CSVs)
+
+| # | Column | Raw name | Unit | Notes |
+|---|--------|----------|------|-------|
+| 0 | `PBDOT` | `PBDOT_obs` | s/s (dimensionless) | Target; observed orbital period derivative |
+| 1 | `Pb` | `Pb_d` | days | Keplerian orbital period |
+| 2 | `e` | `e` | — | Orbital eccentricity |
+| 3 | `m1` | `Mp_Msun` | M☉ | Pulsar (neutron star) mass; NaN ~80 % of rows |
+| 4 | `m2` | `Mc_Msun` | M☉ | Companion mass; NaN ~55 % of rows |
+| 5 | `omdot` | `omdot_deg_yr` | deg yr⁻¹ | Periastron advance rate; NaN where unmeasured |
+| 6 | `sini` | `sini` | — | Sin(orbital inclination); NaN where unmeasured |
+| 7 | `gamma` | `gamma_s` | s | Gravitational redshift + time dilation PK param; NaN where unmeasured |
+
+### Dropped columns
+
+- **`PSRJ`** — pulsar J-name identifier.  Host-only provenance column; not a physical input
+  to any formula in the reference set.  Dropping prevents SR methods from learning a name-based
+  memorisation path.
+- **`binary_model`** — TEMPO timing-model flag (DD, ELL1, BT, DDH, ...).  Metadata about the
+  timing analysis method used to derive the orbital parameters, not a physical property of the
+  binary system.  No reference formula in `reference/summary_*.md` uses `binary_model` as an
+  input variable; the Peters / Hulse-Taylor formula requires only Pb, e, m1, m2 (and optionally
+  omdot, gamma for the Damour 1991 observables-only form).
+
+### Format conventions
+
+- `PBDOT`, `gamma`: scientific notation with 6 decimal places (`.6e`) — values span > 10
+  orders of magnitude.
+- All other numeric columns: plain float with 6 decimal places (`.6f`).
+- Missing values: empty string (CSV blank field).
+- Line terminator: `\n` (Unix).
+
+---
+
+## Type I / Type II Decision
+
+**Verdict: Type I** — no `group_id` column emitted.
+
+Evidence:
+
+1. **One row per pulsar**: each binary pulsar contributes exactly one PBDOT measurement to the
+   dataset.  There is no repeated-observation structure within a single system.
+2. **No within-object clustering**: the ATNF catalogue is a literature compilation; rows are
+   independent astrophysical objects, not time-series or multi-visit observations of the same
+   object.
+3. **Explicit statement in dataset summary**: `summary_dataset_manchester_2005.md §5` states:
+   "Because each row is an independent pulsar (no natural within-object clustering), this is a
+   Setting 1 / Type I dataset."
+4. **No LOCAL_FITTABLE declarations**: no `formulas/` directory exists; therefore no per-cluster
+   fitted parameters are declared.
+
+---
+
+## Split Decision
+
+**Tier chosen: Tier 3 (LLM-proposed OOD split)**
+
+### Tier 1 — dataset-side authority
+Papers inspected:
+- `summary_dataset_manchester_2005.md` §4: *"The ATNF catalogue does not itself define a
+  train/test split. It is a literature-compilation database, not a designed experiment."*
+- `summary_formula_dataset_weisberg_2016.md` §5: *"There is no explicit train/test split within
+  this paper; the full TOA set is the calibration set."*
+- `summary_formula_damour_1991.md` §5: single-system analytical paper; no multi-system split.
+- `summary_formula+dataset_kramer_2006.md` §5: single-system timing paper (J0737-3039A/B only);
+  no multi-system split.
+
+**Finding: no dataset-side split. Tier 1 empty.**
+
+### Tier 2 — SR-method-side authority
+Papers inspected:
+- `summary_formula_peters_1963.md` §5: purely theoretical derivation; no observational dataset.
+- `summary_formula_will_2014.md` §5: review paper; split protocol deferred to ATNF catalogue.
+- `summary_formula_damour_1991.md` §5 (also checked under Tier 2): single system, forward model.
+- `summary_formula+dataset_kramer_2006.md` §5 (also checked under Tier 2): single system.
+
+**Finding: no SR-method-side split. Tier 2 empty.**
+
+### Tier 3 — LLM-proposed OOD split
+
+**OOD axis**: orbital eccentricity (e).
+
+**Rationale**: The structurally critical element of the PBDOT formula is the eccentricity
+enhancement factor (Peters 1963, Eq. 17; Weisberg 2016, Eq. 22):
+
+```
+f(e) = (1 + 73/24·e² + 37/96·e⁴) / (1 − e²)^(7/2)
+```
+
+For near-circular systems (e ≈ 0), f(e) ≈ 1 and PBDOT reduces to the circular-orbit limit.
+For eccentric DNS systems (e up to ~0.91 in this dataset), f(e) exceeds 2000 — a factor of
+2000× enhancement in GW emission rate.  Training exclusively on near-circular systems (where
+f(e) ∈ [1.000, 1.004]) and testing on eccentric systems (where f(e) ∈ [1.015, ~2500]) forces
+the SR method to extrapolate the exact polynomial coefficients (73/24, 37/96) and denominator
+exponent (7/2) that characterise the Peters formula.  These invariant structural constants are
+identified explicitly in all five reference formula summaries.
+
+**Split rule**: `e < 0.05` → train; `e ≥ 0.05` → test.
+
+**Threshold choice**: e = 0.05 chosen to capture the natural bimodal eccentricity structure of
+the ATNF sample (~71 near-circular recycled MSPs vs ~20 eccentric DNS/relativistic systems)
+while maintaining a ~79 %/21 % train/test split.
+
+**Seed**: none required (deterministic threshold split).
+
+**OOD claim**: The test set probes extrapolation of f(e) from the near-circular regime (train:
+e ∈ [0.000, 0.027], f(e) ∈ [1.000, 1.004]) to the eccentric relativistic regime (test: e ∈
+[0.071, 0.905], f(e) ∈ [1.015, ~2500]).
+
+### Counts
+
+| Split | Rows | Eccentricity range | f(e) range |
+|-------|------|--------------------|------------|
+| Train | 71 | [0.000, 0.027] | [1.000, 1.004] |
+| Test  | 20 | [0.071, 0.905] | [1.015, ~2500] |
+
+Test systems include PSR B1913+16 (J1915+1606, the Hulse-Taylor pulsar, e = 0.617), PSR
+J0737-3039A/B (the double pulsar, e = 0.088), and PSR J1757-1854 (e = 0.606), covering the
+canonical GR test cases for the Peters formula.
+
+---
+
+## Smoke Results
+
+### Determinism check
+
+Two consecutive `python3 prep_data.py` runs produce byte-identical outputs:
+
+| File | md5 |
+|------|-----|
+| `data/train.csv` | `755a022e02c7820cfa197dbedfd34ec3` |
+| `data/test.csv`  | `74a4855f56546742bf4bac12de917879` |
+
+Determinism verified ✓
+
+### Formula smoke test
+
+No `formulas/` directory exists for this task at time of data preparation.  Formula smoke test
+(step 2 of section C) is therefore not applicable; it will be run when `formulas/` is authored
+in a subsequent task.
+
+### Class-coverage check
+
+No categorical inputs in the schema — check passes trivially. ✓
+
+---
+
+## Caveats
+
+1. **PBDOT_obs vs PBDOT_intrinsic**: The target column `PBDOT` is the *observed* orbital period
+   derivative (`PBDOT_obs` in the raw ATNF catalogue), which includes kinematic contributions
+   (Shklovskii proper-motion effect, Galactic differential acceleration) in addition to the
+   intrinsic GR quadrupole radiation term.  The Peters formula predicts the intrinsic GR term;
+   for most systems the kinematic correction is small (< 1 % for well-characterised systems like
+   B1913+16, per Damour & Taylor 1991), but for globular-cluster pulsars (J0024-7204*,
+   J1748-2446*, J1824-2452*, etc.) the PBDOT may be dominated by the cluster's gravitational
+   potential acceleration rather than GW energy loss.  SR methods will observe systematic
+   residuals on these systems.  No explicit filtering of globular-cluster pulsars was applied;
+   they are included as legitimate data points to keep the dataset self-contained.
+
+2. **High NaN rates**: m1, m2, omdot, sini, and gamma are NaN for the majority of rows.  The
+   Peters formula requires m1 and m2; SR methods using these inputs will be limited to the
+   ~20–30 rows with measured masses.  The baseline in `baselines.py` handles this by imputing
+   1.4 M☉ for missing masses.
+
+3. **Train rows with non-zero eccentricity**: The highest eccentricity in train is e = 0.027
+   (J1748-2446au and J1618-3921 at e ≈ 0.026–0.027).  At this eccentricity, f(e) ≈ 1.004, which
+   is essentially indistinguishable from the circular-orbit limit.  The eccentricity enhancement
+   is therefore effectively invisible in the training data.
+
+4. **Legacy files in data/**: The directory also contains `filtered.csv`, `ood_dns.csv`,
+   `val.csv`, and `split.json` from a prior pipeline version.  These are not part of the
+   released benchmark (spec §5 ships only `train.csv` and `test.csv`).  `val.csv` in particular
+   conflicts with spec §5 (only train + test ship); it should be removed by a future maintainer.

binary_pulsar_post_keplerian_atnf__PBDOT/data/test.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1a76b6b880840c0712777339fe5fd09dc933812a272bb85149dd35fe543832c
+size 331

binary_pulsar_post_keplerian_atnf__PBDOT/data/train.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c10ec03a33a8396b91e5b9bd1a561ab84a2b029e592d66f3d9fa290321140f7
+size 2253

binary_pulsar_post_keplerian_atnf__PBDOT/formulas/__init__.py

@@ -0,0 +1,50 @@
+"""Reference-baseline formula bank for binary_pulsar_post_keplerian_atnf__PBDOT.
+
+Type I task -- each module exposes the v2 contract:
+    USED_INPUTS, LAW_CONSTANTS, OTHER_CONSTANTS, LOCAL_FITTABLE (= {}),
+    predict(X, **params)
+No fit() -- Type I has no per-cluster parameters.
+
+Wave-17 single live baseline (2026-05-27, user-directed rescue):
+
+  peters_1963 -- chirp-mass form of the Peters & Mathews 1963 GR quadrupole
+        orbital-decay law using only (Pb, e) inputs and a canonical
+        double-neutron-star (DNS) chirp mass M_c = 1.20 M_sun:
+            PBDOT = -(192 pi / 5) * (2 pi * M_c * T_sun * f_b)^(5/3) * F(e),
+            F(e) = (1 + 73/24 e^2 + 37/96 e^4) / (1 - e^2)^(7/2).
+        Peters & Mathews 1963 Eq. 16/17 (PDF p. 437); Will 2014 Eq. 108
+        (PDF p. 69); Kramer 2006 PDF p. 9 (T_sun); Antoniadis 2013 /
+        Bagchi 2013 (canonical M_c).
+
+Why only one baseline (SB tag accepted)
+----------------------------------------
+The Wave-17 schema (PBDOT, Pb, e) admits ONE physical formula: the GR
+quadrupole orbital decay law. Every "alternative" GR derivation (Damour &
+Taylor 1991 Eq. 5.4 PK-observables form, Will 2014 Eq. 108 chirp-mass form,
+Weisberg 2016 Eq. 22 explicit-mass form) is the same Peters & Mathews
+quadrupole law in a different parameterisation -- they are algebraic
+re-arrangements that produce numerically identical predictions on the same
+(Pb, e) inputs once the chirp mass is fixed. A second baseline drawn from
+these papers would be padding ("ladder padding" per Wave-17 user feedback).
+
+Genuinely-distinct alternative formulas would have to be either
+  - non-GR (e.g. Brans-Dicke dipole radiation, Damour 1992 Phys. Rev. D 45),
+    requiring scalar-tensor coupling parameters that are not measured for
+    individual systems, OR
+  - higher-PN corrections (Will 2014 Eq. 108 explicit "+ X_P^GR/c^2"
+    contains coefficients of order 2e-5 relative to the leading term; below
+    measurement noise for all but the cleanest DNS systems).
+
+Wave-13 / Wave-16 history (deleted in Wave-17):
+  damour_1991 -- observables-only via (Pb, e, omdot, gamma) [Eq. 5.4];
+                 deleted because omdot/gamma columns were 74.6 %/98.6 % NaN
+                 in train and the formula required them.
+  will_2014   -- chirp-mass form via (Pb, e, omdot, gamma) deriving masses
+                 from PK equations; deleted because its predictions were
+                 numerically identical to damour_1991 to 6 digits
+                 ("ladder padding").
+"""
+from importlib import import_module
+
+_STEMS = ("peters_1963",)
+REGISTRY = {stem: import_module(f".{stem}", __name__) for stem in _STEMS}

binary_pulsar_post_keplerian_atnf__PBDOT/formulas/peters_1963.py

@@ -0,0 +1,137 @@
+"""GR orbital-period decay of a compact binary -- Peters & Mathews 1963.
+
+Peters & Mathews (1963), "Gravitational Radiation from Point Masses in a
+Keplerian Orbit", Phys. Rev. 131:435 (DOI 10.1103/PhysRev.131.435).
+PDF: reference/peters_1963.pdf.
+
+The paper's principal result is Eq. (16), PDF p. 437, for the orbit-averaged
+gravitational-wave power radiated by a Keplerian binary of arbitrary
+eccentricity:
+
+    <P> = (32/5) * (G^4 / c^5) * m1^2 m2^2 (m1+m2) / [a^5 (1-e^2)^(7/2)]
+          * (1 + (73/24) e^2 + (37/96) e^4)
+
+The corresponding eccentricity enhancement factor (Eq. 17, PDF p. 437) is
+
+    f(e) = (1 + (73/24) e^2 + (37/96) e^4) / (1 - e^2)^(7/2).
+
+The orbital-period derivative PBDOT (the benchmark target) follows from
+energy balance dE/dt = -<P> combined with the Keplerian energy
+E = -G m1 m2 / (2a) and Kepler's third law P_b^2 = 4 pi^2 a^3 / [G (m1+m2)]:
+
+    PBDOT = -(192 pi / 5) * (G / c^3)^(5/3)
+            * (P_b / 2pi)^(-5/3) * m1 m2 / (m1 + m2)^(1/3) * f(e).
+
+In solar / geometric units (T_sun = G M_sun / c^3 = 4.925490947 us,
+Kramer 2006 PDF p. 9) this rearranges to the chirp-mass form (Will 2014
+Eq. 108, PDF p. 69):
+
+    PBDOT = -(192 pi / 5) * (2 pi * M_c * T_sun * f_b)^(5/3) * F(e)
+
+    where  M_c = eta^(3/5) * (m1+m2) = (m1 m2)^(3/5) / (m1+m2)^(1/5)
+    is the chirp mass [M_sun],  f_b = 1 / P_b [Hz],  F(e) = f(e).
+
+The benchmark task ships only (Pb, e) as inputs. Individual component masses
+are not measured for most ATNF binaries (Mp_Msun is NaN for 91/92 rows in
+psrcat v2.8.0), so the chirp mass is supplied here as a single canonical
+value M_c = 1.20 M_sun -- the population mean for galactic
+double-neutron-star (DNS) binaries reported by Antoniadis et al. (2013)
+ApJ 778 and Bagchi (2013) MNRAS 428 (~1.20 M_sun, sigma ~0.07 M_sun).
+This value is the **canonical DNS chirp mass** used throughout the literature
+for population-level GR-decay estimates when per-system masses are not
+available.
+
+Symbol map (paper -> released CSV columns):
+    P_b  <-  Pb  (days; converted to seconds inside predict)
+    e    <-  e
+
+LAW_CONSTANTS / OTHER_CONSTANTS classification (v2 contract):
+
+  LAW_CONSTANTS  (paper-published, frozen, NOT refit):
+      None -- every numerical coefficient (192 pi / 5, 73/24, 37/96, 7/2,
+              5/3) is a PN structural constant of GR (Peters 1963 Eq. 16-17;
+              Will 2014 Eq. 88, 108).
+
+  OTHER_CONSTANTS  (universal physics constants / unit conversions):
+      T_sun       = 4.925490947e-6  s    G M_sun / c^3     (Kramer 2006 PDF p. 9)
+      M_c         = 1.20            M_sun   canonical DNS chirp mass
+                  (Antoniadis 2013 ApJ 778; Bagchi 2013 MNRAS 428)
+      DAY_TO_SEC  = 86400.0         s/d    days -> seconds unit conversion
+                  (Pb is catalogued in days; 1 d = 86400 s exactly, SI)
+
+  LOCAL_FITTABLE: {} -- Type I, no per-system parameters.
+
+Runnability
+-----------
+The formula is defined for every test row (Pb > 0, 0 <= e < 1 satisfied by
+data construction). It uses **only** the inputs that are NaN-free in both
+train and test partitions (Wave-17 column drop). The canonical M_c = 1.20
+M_sun is a population-level approximation;
+individual systems deviate by a factor of <2 (J0737-3039A/B M_c ~ 1.13;
+B1913+16 M_c ~ 1.23). At the per-pulsar level the prediction can be off
+by a factor ~2 -- correct to the GR-power scaling exponent (5/3) but with
+a constant offset reflecting the canonical-vs-actual chirp-mass mismatch.
+"""
+
+import numpy as np
+
+USED_INPUTS = ["Pb", "e"]
+PAPER_REF = "summary_formula_peters_1963.md"
+EQUATION_LOC = ("Peters & Mathews 1963 Eq. 16 (PDF p. 437) + Eq. 17 (PDF p. 437); "
+                "PBDOT derived form -- Will 2014 Eq. 108 (PDF p. 69) "
+                "and Kramer 2006 PDF p. 9 (T_sun definition); "
+                "canonical M_c from Antoniadis 2013 ApJ 778 / Bagchi 2013 MNRAS 428")
+
+# === LAW_CONSTANTS -- paper-published, frozen ===
+LAW_CONSTANTS = {}     # all numerical coefficients are PN structural constants
+
+# === OTHER_CONSTANTS -- universal physics factors / unit conversions / canonical means ===
+OTHER_CONSTANTS = {
+    "T_sun":      4.925490947e-6,   # s; solar mass in geometric units G M_sun / c^3
+    "M_c":        1.20,             # M_sun; canonical DNS chirp mass
+                                    # (Antoniadis+2013, Bagchi 2013)
+    "DAY_TO_SEC": 86400.0,          # s/d; days -> seconds (1 d = 86400 s, SI exact)
+}
+
+LOCAL_FITTABLE = {}    # Type I -- no per-cluster parameters
+
+
+def predict(X: np.ndarray) -> np.ndarray:
+    """Predict GR PBDOT via Peters 1963 / Will 2014 Eq. 108, canonical M_c.
+
+    Type I, LAW_CONSTANTS = {} -- the harness calls predict(X, **{}) with no
+    kwargs. All given constants (T_sun, the canonical chirp mass M_c, and the
+    days->seconds conversion) are read from OTHER_CONSTANTS (gold style);
+    every other coefficient (192 pi / 5, 2 pi, 73/24, 37/96, 7/2, 5/3) is a
+    structural rational of the published GR law, kept inline.
+
+    Parameters
+    ----------
+    X : np.ndarray, shape (n, 2)
+        Columns in USED_INPUTS order: Pb [d], e [dimensionless].
+
+    Returns
+    -------
+    np.ndarray, shape (n,)
+        Predicted PBDOT [dimensionless, s/s]. Always negative for bound
+        orbits (energy is radiated away).
+    """
+    T_sun = OTHER_CONSTANTS["T_sun"]
+    M_c = OTHER_CONSTANTS["M_c"]
+    day_to_sec = OTHER_CONSTANTS["DAY_TO_SEC"]
+
+    Pb_d = np.asarray(X[:, 0], dtype=float)
+    e = np.asarray(X[:, 1], dtype=float)
+
+    Pb_s = Pb_d * day_to_sec
+    f_b = 1.0 / Pb_s            # orbital frequency [Hz]
+
+    e2 = e * e
+    e4 = e2 * e2
+    # f(e) = (1 + 73/24 e^2 + 37/96 e^4) / (1 - e^2)^(7/2)   (Eq. 17)
+    F_e = (1.0 + (73.0 / 24.0) * e2 + (37.0 / 96.0) * e4) / (1.0 - e2) ** 3.5
+
+    # PBDOT = -(192 pi / 5) * (2 pi M_c T_sun f_b)^(5/3) * F(e)   (Will 2014 Eq. 108)
+    return -(192.0 * np.pi / 5.0) \
+           * (2.0 * np.pi * float(M_c) * float(T_sun) * f_b) ** (5.0 / 3.0) \
+           * F_e

binary_pulsar_post_keplerian_atnf__PBDOT/formulas/peters_1963_new.py

@@ -0,0 +1,137 @@
+"""GR orbital-period decay of a compact binary -- Peters & Mathews 1963.
+
+Peters & Mathews (1963), "Gravitational Radiation from Point Masses in a
+Keplerian Orbit", Phys. Rev. 131:435 (DOI 10.1103/PhysRev.131.435).
+PDF: reference/peters_1963.pdf.
+
+The paper's principal result is Eq. (16), PDF p. 437, for the orbit-averaged
+gravitational-wave power radiated by a Keplerian binary of arbitrary
+eccentricity:
+
+    <P> = (32/5) * (G^4 / c^5) * m1^2 m2^2 (m1+m2) / [a^5 (1-e^2)^(7/2)]
+          * (1 + (73/24) e^2 + (37/96) e^4)
+
+The corresponding eccentricity enhancement factor (Eq. 17, PDF p. 437) is
+
+    f(e) = (1 + (73/24) e^2 + (37/96) e^4) / (1 - e^2)^(7/2).
+
+The orbital-period derivative PBDOT (the benchmark target) follows from
+energy balance dE/dt = -<P> combined with the Keplerian energy
+E = -G m1 m2 / (2a) and Kepler's third law P_b^2 = 4 pi^2 a^3 / [G (m1+m2)]:
+
+    PBDOT = -(192 pi / 5) * (G / c^3)^(5/3)
+            * (P_b / 2pi)^(-5/3) * m1 m2 / (m1 + m2)^(1/3) * f(e).
+
+In solar / geometric units (T_sun = G M_sun / c^3 = 4.925490947 us,
+Kramer 2006 PDF p. 9) this rearranges to the chirp-mass form (Will 2014
+Eq. 108, PDF p. 69):
+
+    PBDOT = -(192 pi / 5) * (2 pi * M_c * T_sun * f_b)^(5/3) * F(e)
+
+    where  M_c = eta^(3/5) * (m1+m2) = (m1 m2)^(3/5) / (m1+m2)^(1/5)
+    is the chirp mass [M_sun],  f_b = 1 / P_b [Hz],  F(e) = f(e).
+
+The benchmark task ships only (Pb, e) as inputs. Individual component masses
+are not measured for most ATNF binaries (Mp_Msun is NaN for 91/92 rows in
+psrcat v2.8.0), so the chirp mass is supplied here as a single canonical
+value M_c = 1.20 M_sun -- the population mean for galactic
+double-neutron-star (DNS) binaries reported by Antoniadis et al. (2013)
+ApJ 778 and Bagchi (2013) MNRAS 428 (~1.20 M_sun, sigma ~0.07 M_sun).
+This value is the **canonical DNS chirp mass** used throughout the literature
+for population-level GR-decay estimates when per-system masses are not
+available.
+
+Symbol map (paper -> released CSV columns):
+    P_b  <-  Pb  (days; converted to seconds inside predict)
+    e    <-  e
+
+LAW_CONSTANTS / OTHER_CONSTANTS classification (v2 contract):
+
+  LAW_CONSTANTS  (paper-published, frozen, NOT refit):
+      None -- every numerical coefficient (192 pi / 5, 73/24, 37/96, 7/2,
+              5/3) is a PN structural constant of GR (Peters 1963 Eq. 16-17;
+              Will 2014 Eq. 88, 108).
+
+  OTHER_CONSTANTS  (universal physics constants / unit conversions):
+      T_sun       = 4.925490947e-6  s    G M_sun / c^3     (Kramer 2006 PDF p. 9)
+      M_c         = 1.20            M_sun   canonical DNS chirp mass
+                  (Antoniadis 2013 ApJ 778; Bagchi 2013 MNRAS 428)
+      DAY_TO_SEC  = 86400.0         s/d    days -> seconds unit conversion
+                  (Pb is catalogued in days; 1 d = 86400 s exactly, SI)
+
+  LOCAL_FITTABLE: {} -- Type I, no per-system parameters.
+
+Runnability
+-----------
+The formula is defined for every test row (Pb > 0, 0 <= e < 1 satisfied by
+data construction). It uses **only** the inputs that are NaN-free in both
+train and test partitions (Wave-17 column drop). The canonical M_c = 1.20
+M_sun is a population-level approximation;
+individual systems deviate by a factor of <2 (J0737-3039A/B M_c ~ 1.13;
+B1913+16 M_c ~ 1.23). At the per-pulsar level the prediction can be off
+by a factor ~2 -- correct to the GR-power scaling exponent (5/3) but with
+a constant offset reflecting the canonical-vs-actual chirp-mass mismatch.
+"""
+
+import numpy as np
+
+USED_INPUTS = ["Pb", "e"]
+PAPER_REF = "summary_formula_peters_1963.md"
+EQUATION_LOC = ("Peters & Mathews 1963 Eq. 16 (PDF p. 437) + Eq. 17 (PDF p. 437); "
+                "PBDOT derived form -- Will 2014 Eq. 108 (PDF p. 69) "
+                "and Kramer 2006 PDF p. 9 (T_sun definition); "
+                "canonical M_c from Antoniadis 2013 ApJ 778 / Bagchi 2013 MNRAS 428")
+
+# === LAW_CONSTANTS -- paper-published, frozen ===
+LAW_CONSTANTS = {}     # all numerical coefficients are PN structural constants
+
+# === OTHER_CONSTANTS -- universal physics factors / unit conversions / canonical means ===
+OTHER_CONSTANTS = {
+    "T_sun":      4.925490947e-6,   # s; solar mass in geometric units G M_sun / c^3
+    "M_c":        1.20,             # M_sun; canonical DNS chirp mass
+                                    # (Antoniadis+2013, Bagchi 2013)
+    "DAY_TO_SEC": 86400.0,          # s/d; days -> seconds (1 d = 86400 s, SI exact)
+}
+
+LOCAL_FITTABLE = {}    # Type I -- no per-cluster parameters
+
+
+def predict(X: np.ndarray) -> np.ndarray:
+    """Predict GR PBDOT via Peters 1963 / Will 2014 Eq. 108, canonical M_c.
+
+    Type I, LAW_CONSTANTS = {} -- the harness calls predict(X, **{}) with no
+    kwargs. All given constants (T_sun, the canonical chirp mass M_c, and the
+    days->seconds conversion) are read from OTHER_CONSTANTS (gold style);
+    every other coefficient (192 pi / 5, 2 pi, 73/24, 37/96, 7/2, 5/3) is a
+    structural rational of the published GR law, kept inline.
+
+    Parameters
+    ----------
+    X : np.ndarray, shape (n, 2)
+        Columns in USED_INPUTS order: Pb [d], e [dimensionless].
+
+    Returns
+    -------
+    np.ndarray, shape (n,)
+        Predicted PBDOT [dimensionless, s/s]. Always negative for bound
+        orbits (energy is radiated away).
+    """
+    T_sun = OTHER_CONSTANTS["T_sun"]
+    M_c = OTHER_CONSTANTS["M_c"]
+    day_to_sec = OTHER_CONSTANTS["DAY_TO_SEC"]
+
+    Pb_d = np.asarray(X[:, 0], dtype=float)
+    e = np.asarray(X[:, 1], dtype=float)
+
+    Pb_s = Pb_d * day_to_sec
+    f_b = 1.0 / Pb_s            # orbital frequency [Hz]
+
+    e2 = e * e
+    e4 = e2 * e2
+    # f(e) = (1 + 73/24 e^2 + 37/96 e^4) / (1 - e^2)^(7/2)   (Eq. 17)
+    F_e = (1.0 + (73.0 / 24.0) * e2 + (37.0 / 96.0) * e4) / (1.0 - e2) ** 3.5
+
+    # PBDOT = -(192 pi / 5) * (2 pi M_c T_sun f_b)^(5/3) * F(e)   (Will 2014 Eq. 108)
+    return -(192.0 * np.pi / 5.0) \
+           * (2.0 * np.pi * float(M_c) * float(T_sun) * f_b) ** (5.0 / 3.0) \
+           * F_e

binary_pulsar_post_keplerian_atnf__PBDOT/formulas/reference_metrics.json

@@ -0,0 +1,40 @@
+{
+  "baselines": {
+    "peters_1963": {
+      "equation_loc": "Peters & Mathews 1963 Eq. 16 (PDF p. 437) + Eq. 17 (PDF p. 437); PBDOT derived form -- Will 2014 Eq. 108 (PDF p. 69) and Kramer 2006 PDF p. 9 (T_sun definition); canonical M_c from Antoniadis 2013 ApJ 778 / Bagchi 2013 MNRAS 428",
+      "error": null,
+      "failed": false,
+      "kind": "reference",
+      "law_constants": {},
+      "local_fittable": [],
+      "metrics": {
+        "log_mae": null,
+        "mae": 9.683126914531676e-14,
+        "mape": 0.13348956381077243,
+        "mdae": 7.643692387886755e-14,
+        "mse": 1.481736628537652e-26,
+        "n_finite": 10,
+        "r2": 0.994398063852058,
+        "rmse": 1.2172660467365594e-13,
+        "smape": 0.1172416817741102
+      },
+      "other_constants": {
+        "DAY_TO_SEC": 86400.0,
+        "M_c": 1.2,
+        "T_sun": 4.925490947e-06
+      },
+      "paper_ref": "summary_formula_peters_1963.md"
+    }
+  },
+  "derived_caps": {
+    "fit_timeout_seconds": null,
+    "max_init_size_per_param": 1,
+    "max_law_constants": 0,
+    "max_local_params": 0
+  },
+  "metric_declared": "rmse",
+  "n_test_rows": 10,
+  "reference_baseline_id": "peters_1963",
+  "task": "binary_pulsar_post_keplerian_atnf__PBDOT",
+  "type": "typeI"
+}

binary_pulsar_post_keplerian_atnf__PBDOT/metadata.yaml

@@ -0,0 +1,230 @@
+task_id: binary_pulsar_post_keplerian_atnf__PBDOT
+metric: rmse
+domain: astronomy / pulsar_timing
+type: typeI
+has_group_id: false
+
+license: "GPL-3.0-or-later (ATNF psrcat v2.8.0 distribution, CSIRO/ATNF); pulsar timing measurements are numeric facts per Feist 1991; AUDIT_CHECKLIST v0.4 D5 compatible"
+
+context: >-
+  A binary pulsar is a compact system in which a pulsar orbits another neutron star, studied in relativistic astrophysics as a test of gravity.
+
+problem_statement: >-
+  Predict the GR-driven orbital-period derivative (PBDOT, dimensionless
+  s/s, always negative) of compact-DNS binary pulsars catalogued in
+  ATNF psrcat v2.8.0 from (Pb [days], e [dimensionless]) alone. The
+  split places near-circular systems (e < 0.05, predominantly recycled
+  millisecond pulsars in NS-WD orbits whose Pb-decay is dominated by
+  Galactic-acceleration kinematic terms rather than GR quadrupole
+  losses) in train (71 rows) and the eccentric GR-dominated compact
+  DNS systems (e >= 0.05, PBDOT < 0, Pb < 1 d) in test (10 rows, after FM-B8 dedup of the J0737-3039A/B physical-alias double-count).
+  The hard constraints to recover are the GR quadrupole prefactor
+  (192 pi / 5), the eccentricity-enhancement polynomial coefficients
+  (73/24, 37/96, exponent 7/2), and the chirp-mass scaling (5/3 power)
+  of Peters 1963 / Will 2014.
+
+target:
+  name: PBDOT
+  symbol: dot_P_b
+  unit: s/s
+  description: >-
+    Observed time derivative of the binary orbital period (dimensionless
+    rate). Measured directly from multi-year pulse-timing residuals.
+    Negative for GR-dominated binaries (inspiraling DNS systems).
+  # E1 (2026-05-29): bounds the FULL target column (train+test). The near-circular
+  # train PBDOT spans positive (Shklovskii/kinematic) to deep-negative (redback /
+  # cluster contamination); the GR-valid regime is the test subset
+  # [-5.294e-12, -1.366e-13]. See prep_data.py "Honest train-set limitation".
+  range:
+    - -3.5e-9
+    - 6.0e-11
+
+# === Wave-17 column cleanup, 2026-05-27 ===
+# Dropped omdot, gamma -- 74.6 % / 98.6 % NaN in train (most ATNF binaries
+# are recycled near-circular MSPs that lack measurable post-Keplerian
+# precession). The two baselines that consumed them (damour_1991,
+# will_2014) were deleted because they were algebraically equivalent
+# parameterisations of the same GR quadrupole law (numerically identical
+# predictions to 6 digits on the test set). The Wave-17 baseline
+# (peters_1963) uses a canonical DNS chirp mass instead of per-system
+# omdot+gamma-derived masses, eliminating the NaN dependency. Prior
+# Wave-13 already dropped m1, m2, sini for the same reason.
+inputs:
+  - name: Pb
+    symbol: P_b
+    unit: d
+    description: Keplerian orbital period of the binary system.
+    range:
+      - 0.06
+      - 1236.73
+  - name: e
+    symbol: e
+    unit: ""
+    description: Orbital eccentricity, dimensionless and bounded in [0, 1).
+    range:
+      - 0.0
+      - 0.91
+
+data_files:
+  train: data/train.csv   # near-circular MSP-dominated split: e < 0.05, 71 rows
+  test:  data/test.csv    # compact DNS GR-test split: e >= 0.05 AND PBDOT < 0 AND Pb < 1 d, 10 rows (FM-B8 dedup of J0737-3039A/B)
+
+n_train: 71
+n_test: 10
+
+# Wave-17 (2026-05-27, user-directed rescue):
+#   - Refreshed against ATNF psrcat v2.8.0 (released 2026-05-07; +1 binary
+#     with PBDOT measured: J1946+2052, a DNS in a 1.9-h compact orbit).
+#   - Dropped omdot, gamma columns (74.6 % / 98.6 % NaN in train).
+#   - Dropped will_2014 baseline (numerically equivalent to damour_1991).
+#   - Dropped damour_1991 baseline (structurally requires omdot+gamma).
+#   - Added peters_1963 baseline using only (Pb, e) and a canonical DNS
+#     chirp mass M_c = 1.20 M_sun (Antoniadis 2013, Bagchi 2013).
+#   - Added Pb < 1 d test filter to exclude cluster pulsars (J0024-7204H
+#     in 47 Tuc, J1518+0204B in M5) and Be-star binaries (J0045-7319)
+#     whose PBDOT is contaminated by external accelerations and tidal
+#     effects rather than GR quadrupole losses.
+#
+# Post-Wave-17 R^2 (peters_1963 on 10-row test): +0.9944 (was +0.78 on the
+# old 10-row damour_1991 split). RMSE 1.217e-13 s/s.
+#
+# Auditor notes:
+#
+# 1. Eccentricity split + HONEST TRAIN LIMITATION (2026-05-29 audit). Train
+#    (e<0.05, 71 rows) is dominated by recycled MSPs in near-circular orbits whose
+#    observed Pbdot is NOT GR: 41/71 train rows have POSITIVE PBDOT (Shklovskii /
+#    Galactic-acceleration kinematic terms — opposite sign to GR inspiral), and the
+#    negative ones are largely redbacks / transitional MSPs / cluster pulsars
+#    (tidal / mass-loss / cluster-potential, 100-10000x the GR term). So the GR
+#    quadrupole law is NOT learnable by curve-fitting the train fold — the task
+#    tests RECOVERY of the known GR law on the clean eccentric-DNS test set against
+#    the Peters reference. Only ~12 clean GR-test DNS exist worldwide, so a >=30-row
+#    clean-only split is infeasible (FM-B4); the near-circular train is the contrast
+#    regime. This is a physics-imposed limitation, not a fixable split defect.
+#    Test (e>=0.05, PBDOT<0, Pb<1 d, 10 rows) is the compact-DNS regime where the
+#    GR Peters quadrupole loss dominates the timing residual.
+#
+# 2. Compact-DNS Pb<1 d filter. Wave-17 adds Pb<1 d to the test selection
+#    to exclude wide-orbit cluster pulsars and Be-star binaries whose PBDOT
+#    sign is consistent with GR (negative) but whose magnitude is set by
+#    cluster-potential or tidal contamination at a level orders of magnitude
+#    larger than the GR quadrupole contribution. The 4 systems removed by
+#    this filter (J0024-7204H, J0045-7319, J1518+0204B, J2129+1210C if
+#    Pb>1 d -- actually 0.34 d so retained) would each contribute residuals
+#    ~1e-7 against a GR prediction ~1e-15, breaking the metric.
+#
+# 3. Per-row data availability (test split): all 10 rows have Pb, e, PBDOT
+#    NaN-free. The peters_1963 baseline returns 10/10 finite predictions.
+#
+# 4. Sample-size justification. Measuring PBDOT to 1e-12 precision requires
+#    10-30 years of continuous pulsar timing; fewer than 15 such DNS
+#    systems exist worldwide as of 2026. Sample size reflects the physical
+#    rarity of the phenomenon, not a data collection limitation. The
+#    Wave-17 refresh to psrcat v2.8.0 added 1 new system (J1946+2052,
+#    Meyers+2025); future updates will incrementally add ~1-2 systems/year.
+#
+# 5. License framework. ATNF psrcat is distributed under GPL-3.0-or-later
+#    (https://www.atnf.csiro.au/research/pulsar/psrcat/download.html);
+#    copyright CSIRO/ATNF; required citation per psrcat documentation
+#    is the Manchester 2005 reference (see references: block below).
+#    The catalogued pulsar timing measurements are numeric astrophysical
+#    facts per Feist v. Rural Telephone (US Supreme Court 1991);
+#    AUDIT_CHECKLIST v0.4 D5 explicitly covers this case.
+#
+# 6. Why one live baseline (SB tag). Every published GR-quadrupole
+#    PBDOT formula that takes (Pb, e) alone reduces to the same Peters
+#    chirp-mass law once a chirp-mass value is fixed; alternative
+#    parameterisations (Damour 1991 Eq. 5.4 PK-observables form,
+#    Will 2014 Eq. 108 chirp-mass form, Weisberg 2016 Eq. 22
+#    explicit-mass form) produce numerically identical predictions
+#    after the algebraic re-arrangement. Adding a "Will-2014" or
+#    "Weisberg-2016" baseline alongside peters_1963 would be ladder
+#    padding. Genuinely-distinct alternatives (non-GR Brans-Dicke
+#    dipole radiation, higher-PN corrections at order 1e-5) are
+#    either below measurement noise on the test set or require
+#    free parameters not measured for these systems. See WAVE17_ATTEMPT.md.
+
+priors:
+  - name: T_sun
+    value: 4.925490947e-6
+    unit: s
+    description: Solar mass in geometric time units (G M_sun / c^3). Universal physics constant.
+    source: Kramer et al. 2006, PDF p. 9
+    _role: physics_anchor
+  - name: M_c_canonical
+    value: 1.20
+    unit: M_sun
+    description: Canonical chirp mass for galactic DNS binaries (population mean).
+    source: Antoniadis 2013 ApJ 778 / Bagchi 2013 MNRAS 428
+    _role: physics_anchor
+  - name: coeff_73_24
+    value: 3.0416667
+    unit: ""
+    description: GR eccentricity-polynomial coefficient 73/24 from Peters & Mathews 1963 Eq. 17 -- structural constant.
+    source: Peters & Mathews 1963, Eq. 17, PDF p. 437
+    _role: distractor
+  - name: coeff_37_96
+    value: 0.3854167
+    unit: ""
+    description: GR eccentricity-polynomial coefficient 37/96 from Peters & Mathews 1963 Eq. 17 -- structural constant.
+    source: Peters & Mathews 1963, Eq. 17, PDF p. 437
+    _role: distractor
+
+# Reference-baseline bank
+# Wave-17 (2026-05-27): single live baseline (SB tag accepted).
+#   peters_1963 -- chirp-mass form with canonical DNS M_c=1.20 M_sun
+#                  using only (Pb, e). RMSE 1.23e-13, R^2 +0.994 on
+#                  10-row compact-DNS test.
+references:
+  - id: peters_1963
+    label: "Peters & Mathews 1963 Eq. 16/17 (PDF p. 437) -- GR PBDOT via canonical DNS chirp mass (M_c=1.20 M_sun, Antoniadis 2013); chirp-mass form per Will 2014 Eq. 108 (PDF p. 69)"
+    formula_file: formulas/peters_1963.py
+    reference_pdf: reference/peters_1963.pdf
+    n_law_constants: 0
+    n_other_constants: 3   # T_sun, M_c, DAY_TO_SEC (days->seconds conversion)
+    n_local_params: 0
+    measured:
+      rmse: 1.2172660467365594e-13
+      mse: 1.481736628537652e-26
+      mae: 9.683126914531676e-14
+      smape: 0.1172416817741102
+      r2: 0.994398063852058
+      n_finite: 10
+      n_test: 10
+
+caps:
+  max_law_constants: 0
+  max_local_params: 0
+  max_init_size_per_param: 1
+  fit_timeout_seconds: null
+
+reference_baseline_id: peters_1963
+# === Validity rubric ===
+# Frozen list — the scoring-time judge (eval_consistency.py) scores a
+# submission by how many of THESE criteria it satisfies:
+# validity_score = (criteria satisfied) / len(validity_rubrics).
+# De-leaked context + comprehensive multi-baseline extraction, 2026-06-01.
+validity_rubrics:
+  - "includes eccentricity-enhanced gravitational radiation with circular orbits least efficient"
+  - "includes quadrupolar gravitational-wave energy loss from the binary orbit"
+  - "includes the Keplerian orbital-period scaling that makes shorter-period binaries decay faster"
+  - "includes the compact-binary chirp-mass scale in the period-decay normalization"
+  - "predicted PBDOT is strictly negative for every bound binary with Pb > 0 and 0 <= e < 1"
+  - "predicted PBDOT remains defined throughout the physical input domain Pb > 0 and 0 <= e < 1"
+
+best_baseline: 1.21727e-13    # peters_1963 — rmse (lower is better); regenerated 2026-05-30
+
+wave17_notes: >-
+  Wave-17 rescue (2026-05-27). User flagged Wave-16 will_2014 baseline as
+  algebraic re-arrangement of damour_1991 ("ladder padding", numerically
+  identical predictions to 6 digits). User also rejected the omdot/gamma
+  74.6%/98.6% NaN columns as "documented but not fixed". Wave-17 actions:
+  (1) deleted will_2014.py and damour_1991.py from formulas/; (2) dropped
+  omdot, gamma from CSV schema; (3) added peters_1963.py using only (Pb, e)
+  + canonical DNS chirp mass M_c=1.20 M_sun (Antoniadis 2013); (4) refreshed
+  ATNF data to psrcat v2.8.0 (+1 DNS: J1946+2052); (5) added Pb<1 d test
+  filter (drops 4 cluster/Be-star contaminated systems). Result: 0-NaN
+  dataset, 10 GR-clean test rows (11 from the v2.8.0 refresh minus 1 J0737-3039A/B
+  physical-alias dedup per FM-B8), R^2 0.994 (was 0.78), single
+  live baseline (SB tag accepted -- no genuinely-distinct 2nd form exists
+  with [Pb, e]-only inputs). See WAVE17_ATTEMPT.md for full diff.

binary_pulsar_post_keplerian_atnf__PBDOT/prep_data.py

@@ -0,0 +1,294 @@
+#!/usr/bin/env python3
+"""binary_pulsar_post_keplerian_atnf__PBDOT — data preparation script
+============================================================================
+
+Source
+------
+data_raw/atnf_binary_pulsar_pk.csv  (92 data rows + header)
+sha256 : 859e7c9d9d64d863e8792cc100ebf8a6019c042c87d22ad1e76cd9660a169b22
+Produced by data_raw/parse_psrcat.py from ATNF Pulsar Catalogue v2.8.0
+(Manchester et al. 2005, AJ 129, 1993;
+ https://www.atnf.csiro.au/research/pulsar/psrcat/).
+
+Wave-17 (2026-05-27) — schema simplified to (PBDOT, Pb, e) only
+----------------------------------------------------------------
+The previous schema kept omdot, gamma as inputs to support the Damour 1991
+PK-observables-only baseline (formulas/damour_1991.py) and the chirp-mass-form
+re-parameterisation (formulas/will_2014.py). Both are removed in Wave-17
+because (a) the two baselines are numerically equivalent algebraic
+re-arrangements of the same GR quadrupole law (RMSE identical to 6 digits)
+— "ladder padding" — and (b) the omdot / gamma columns were 74.6 % / 98.6 %
+NaN in train because most ATNF binaries are recycled near-circular MSPs that
+do not require those PK parameters to be measurable. Dropping them yields a
+0-NaN dataset with a single, honestly-distinct Peters 1963 reference baseline.
+
+Canonical column projection (Wave-17)
+--------------------------------------
+col 0  PBDOT    observed orbital period derivative (s s^-1, dimensionless)
+                renamed from raw column PBDOT_obs
+col 1  Pb       orbital period (days); renamed from Pb_d
+col 2  e        orbital eccentricity (dimensionless)
+
+Dropped columns (Wave-17)
+-------------------------
+omdot, gamma  -- 74.6%/98.6% NaN in train; the only baselines that consumed
+                 them were the algebraically-equivalent will_2014/damour_1991
+                 forms (deleted in Wave-17).
+m1, m2, sini  -- already dropped in Wave-13 (no active baseline ever used
+                 them; were 65-100 % NaN in train).
+PSRJ          -- pulsar J-name identifier; host-only provenance column,
+                 not a physics formula input.
+binary_model  -- TEMPO timing-model flag (DD, ELL1, BT, DDH, ...); metadata
+                 about the timing analysis method, not a physical property.
+
+Type I / Type II decision
+--------------------------
+TYPE I.  Each row is an independent binary pulsar system contributing exactly
+one PBDOT measurement.  There is no within-system repeated-measurement
+hierarchy.  The dataset is a literature compilation catalogue (Manchester et al.
+2005); each pulsar is a distinct astrophysical object with a single summary
+entry.  summary_dataset_manchester_2005.md SS5 states explicitly: "Because each
+row is an independent pulsar (no natural within-object clustering), this is a
+Setting 1 / Type I dataset."  No formulas/ directory exists (no LOCAL_FITTABLE
+declarations).  -> No group_id column emitted.
+
+Split-tier consultation
+------------------------
+Tier 1 -- dataset-side authority:
+  Inspected: summary_dataset_manchester_2005.md and
+    summary_formula_dataset_weisberg_2016.md.
+  manchester_2005 SS4: "The ATNF catalogue does not itself define a train/test
+    split. It is a literature-compilation database, not a designed experiment."
+  weisberg_2016 SS5: "There is no explicit train/test split within this paper;
+    the full TOA set is the calibration set."
+  damour_1991 SS5: single-system analytical derivation; no multi-system split.
+  kramer_2006 SS5: single-system timing paper (PSR J0737-3039A/B only);
+    no multi-system split.
+  Finding: no dataset-side split defined in any paper. Tier 1 EMPTY.
+
+Tier 2 -- SR-method-side authority:
+  Inspected: summary_formula_damour_1991.md, summary_formula_peters_1963.md,
+    summary_formula_will_2014.md, summary_formula+dataset_kramer_2006.md.
+  All four papers are either single-system or purely theoretical derivations;
+  none define a multi-system train/test partition applicable to the full ATNF
+  dataset.  Finding: no SR-method-side split. Tier 2 EMPTY.
+
+Tier 3 -- LLM-proposed OOD split (implemented in this script):
+  OOD axis chosen: orbital eccentricity (e).
+  Rationale: The field's primary structural challenge in the PBDOT formula is
+  the eccentricity enhancement factor
+      f(e) = (1 + 73/24 * e^2 + 37/96 * e^4) / (1 - e^2)^(7/2)   [Peters 1963 Eq.17]
+  which is the invariant structural element identified in all reference
+  formula summaries.  For near-circular orbits (e ~ 0), f(e) ~ 1 and PBDOT
+  reduces to the circular-orbit limit.  For highly eccentric DNS systems
+  (e up to ~0.68 in this dataset post-Wave-17), f(e) exceeds 20, amplifying
+  the GW power by orders of magnitude.  Training exclusively on near-circular
+  systems forces SR to extrapolate f(e) into the eccentric regime where the
+  exact polynomial coefficients (73/24, 37/96) and denominator exponent (7/2)
+  are the determining structural constants.
+  Split rule: train <- e < 0.05 (near-circular); test <- e >= 0.05 (eccentric).
+  Seed: none (deterministic threshold, no randomness required).
+
+  Wave-17 test-set physical filters:
+   - PBDOT < 0   (Wave-16): drop the kinematic-positive PBDOT systems
+                  whose observed Pb-decay is dominated by Galactic-acceleration
+                  / Shklovskii kinematic terms, NOT by GR quadrupole losses;
+                  any GR-only formula structurally cannot fit them.
+   - Pb < 1 d    (Wave-17): restrict to compact-DNS "clean GR test" systems.
+                  Globular-cluster pulsars with Pb >> 1 d (e.g. J0024-7204H
+                  in 47 Tuc, J1518+0204B in M5) and Be-star companions
+                  (J0045-7319) have PBDOT contaminated by external accelerations
+                  (cluster potential) or tidal/wind effects of orders of
+                  magnitude larger than the GR quadrupole contribution.
+                  The Pb<1d threshold is the standard literature delimiter
+                  for the "compact DNS GR-test" subsample (Tauris+2017,
+                  Antoniadis+2013) and corresponds to characteristic
+                  GW-driven inspiral timescales tau_GW < 10 Gyr.
+
+  Train rows: 71   Test rows: 10 (after FM-B8 dedup of J0737-3039A/B; see step 3.5).
+
+  Honest train-set limitation (documented 2026-05-29 audit)
+  ----------------------------------------------------------
+  The near-circular train split (e < 0.05) is NOT a clean GR-quadrupole signal:
+  41 of its 71 rows have POSITIVE PBDOT (orbital period INCREASING — dominated by
+  Shklovskii / Galactic-acceleration kinematic terms, the opposite sign of GR
+  inspiral), and most of the negative-PBDOT near-circular systems are recycled
+  redbacks / transitional MSPs / cluster pulsars (e.g. J1227-4853 -8.7e-10,
+  J1723-2837 -3.5e-9, J1023+0038 -7.3e-11) whose decay is set by tidal / mass-loss
+  / cluster-potential effects 100-10000x larger than the GR quadrupole term.
+  The GR law is cleanly observable only in compact ECCENTRIC double-neutron-star
+  systems, which are exactly the ~10-12 systems in the test split. Consequently
+  this task tests RECOVERY OF THE KNOWN GR LAW on the clean eccentric DNS test set
+  against the Peters reference; it does NOT (and physically cannot) supply a train
+  fold from which the GR quadrupole law is learnable by curve-fitting. Only ~12
+  clean GR-test DNS exist worldwide, so a >=30-row clean-only split is infeasible
+  (FM-B4 floor); the near-circular train is retained as the contrast regime. This
+  is a known, physics-imposed limitation, not a fixable split defect.
+"""
+
+import csv
+import hashlib
+import sys
+from pathlib import Path
+
+TASK_DIR = Path(__file__).parent
+RAW      = TASK_DIR / "data_raw" / "atnf_binary_pulsar_pk.csv"
+OUT      = TASK_DIR / "data"
+
+# Wave-17: refreshed against ATNF psrcat v2.8.0 (released 2026-05-07).
+# Previous SHA (v2.7.0): 39d4efe1057276295f6e6ba858115e1cb38f3e41388d48f835a9c9aef829c18d
+EXPECTED_SHA256 = "859e7c9d9d64d863e8792cc100ebf8a6019c042c87d22ad1e76cd9660a169b22"
+EXPECTED_ROWS   = 92
+
+# Rename map: raw column name -> canonical name
+RENAME = {
+    "PBDOT_obs":    "PBDOT",
+    "Pb_d":         "Pb",
+    # "e"    -> "e"    (unchanged)
+}
+
+# Final output columns: target first, then inputs.
+# Wave-17 (2026-05-27): omdot, gamma also dropped (74.6 % / 98.6 % NaN in train).
+COLS = ["PBDOT", "Pb", "e"]
+
+# Columns where scientific notation is needed (values span > 4 orders of magnitude)
+SCI_COLS = {"PBDOT"}
+
+# Eccentricity threshold: near-circular -> train; eccentric -> test (Tier-3 OOD)
+E_THRESH = 0.05
+
+# Wave-16: PBDOT < 0 -> drop kinematic-positive systems.
+# Wave-17: Pb < 1 d -> restrict test to compact DNS (drop cluster + Be-star
+#                       systems whose PBDOT is non-GR contaminated).
+PBDOT_GR_FILTER = True
+PB_MAX_DAYS = 1.0
+
+
+def _check_sha256(path: Path) -> None:
+    digest = hashlib.sha256(path.read_bytes()).hexdigest()
+    if digest != EXPECTED_SHA256:
+        print(
+            f"ERROR: sha256 mismatch for {path}\n"
+            f"  expected : {EXPECTED_SHA256}\n"
+            f"  got      : {digest}",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+
+def _fmt(col: str, val: str) -> str:
+    """Format one CSV cell. Missing values emit literal 'nan' (harness convention)."""
+    if val == "":
+        return "nan"
+    v = float(val)
+    if col in SCI_COLS:
+        return f"{v:.6e}"
+    return f"{v:.6f}"
+
+
+def _write_csv(path: Path, rows) -> None:
+    with open(path, "w", newline="") as fh:
+        fh.write(",".join(COLS) + "\n")
+        for row in rows:
+            fh.write(",".join(_fmt(c, row.get(c, "")) for c in COLS) + "\n")
+
+
+def main() -> None:
+    # 1. Verify source file integrity
+    _check_sha256(RAW)
+
+    # 2. Load source CSV
+    with open(RAW, newline="") as fh:
+        raw_rows = list(csv.DictReader(fh))
+
+    if len(raw_rows) != EXPECTED_ROWS:
+        print(
+            f"ERROR: expected {EXPECTED_ROWS} data rows, got {len(raw_rows)}",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+    # 3. Rename and project to canonical columns only
+    projected = []
+    for raw_row in raw_rows:
+        row = {}
+        for raw_col, val in raw_row.items():
+            canon = RENAME.get(raw_col, raw_col)
+            if canon in COLS:
+                row[canon] = val
+        projected.append(row)
+
+    # 3.5 De-duplicate physical-alias rows (FM-B8). The ATNF catalogue lists
+    #     J0737-3039A and J0737-3039B as separate pulsars, but they are the two
+    #     members of the SAME double-neutron-star binary and share ONE orbital-decay
+    #     measurement (identical Pb=0.102252 d, e=0.087777, PBDOT=-1.252e-12).
+    #     Keeping both double-counts one physical system. Dedup on the released
+    #     (PBDOT, Pb, e) tuple, keeping the first occurrence (J0737-3039A); this
+    #     removes exactly 1 row (J0737-3039B). Back-ported 2026-05-29: a GM manual
+    #     `drop_duplicates()` had been applied to data/test.csv but NOT to this
+    #     recipe, so re-running prep_data regenerated the duplicate (FM-B9a drift);
+    #     the dedup is now in the pipeline and is deterministic/reproducible.
+    seen, deduped, n_dup = set(), [], 0
+    for row in projected:
+        key = (row.get("PBDOT", ""), row.get("Pb", ""), row.get("e", ""))
+        if key in seen:
+            n_dup += 1
+            continue
+        seen.add(key)
+        deduped.append(row)
+    projected = deduped
+
+    # 4. Tier-3 eccentricity split (deterministic threshold; no randomness).
+    train = [r for r in projected if float(r["e"]) < E_THRESH]
+    test_full = [r for r in projected if float(r["e"]) >= E_THRESH]
+
+    # Wave-16: PBDOT < 0 filter (GR-dominated only)
+    # Wave-17: Pb < 1 d filter (compact DNS only — drop cluster + Be-star contam.)
+    test = []
+    for r in test_full:
+        if r.get("PBDOT", "") == "" or r.get("Pb", "") == "":
+            continue
+        if PBDOT_GR_FILTER and float(r["PBDOT"]) >= 0.0:
+            continue
+        if float(r["Pb"]) >= PB_MAX_DAYS:
+            continue
+        test.append(r)
+
+    # 5. Write output CSVs
+    OUT.mkdir(exist_ok=True)
+    _write_csv(OUT / "train.csv", train)
+    _write_csv(OUT / "test.csv",  test)
+
+    # 6. Summary printout
+    print(f"physical-alias dedup (FM-B8): removed {n_dup} duplicate row(s) [J0737-3039A/B]")
+    print(f"data/train.csv  : {len(train)} rows x {len(COLS)} cols  [e < {E_THRESH}]")
+    print(f"data/test.csv   : {len(test)}  rows x {len(COLS)} cols  "
+          f"[e >= {E_THRESH}, PBDOT < 0, Pb < {PB_MAX_DAYS} d]")
+
+    # 7. Type I assertion: no group_id must appear in the released columns
+    assert "group_id" not in COLS, "group_id must not be in COLS for a Type I task"
+    print("Type I check    : no group_id column -- OK")
+
+    # 8. No categorical inputs in schema -> coverage check trivially passes
+    print("Coverage check  : no categorical inputs in schema -- OK")
+
+    # 9. Eccentricity ranges for audit
+    e_train = [float(r["e"]) for r in train]
+    e_test  = [float(r["e"]) for r in test]
+    print(f"e range (train) : [{min(e_train):.6f}, {max(e_train):.6f}]")
+    print(f"e range (test)  : [{min(e_test):.6f}, {max(e_test):.6f}]")
+
+    # 10. Wave-17 NaN-free assertion: train and test CSVs must have 0 NaN
+    import csv as _csv
+    for split_path in (OUT / "train.csv", OUT / "test.csv"):
+        with open(split_path) as fh:
+            for line in fh.readlines()[1:]:  # skip header
+                if "nan" in line.lower():
+                    print(f"ERROR: NaN found in {split_path.name}: {line.strip()}",
+                          file=sys.stderr)
+                    sys.exit(1)
+    print("NaN-free check  : train.csv and test.csv contain no NaN -- OK")
+
+
+if __name__ == "__main__":
+    main()

binary_pulsar_post_keplerian_atnf__PBDOT/reference/README.md

@@ -0,0 +1,33 @@
+# Reference PDFs — binary_pulsar_post_keplerian_atnf__PBDOT
+
+## Index
+
+| File stem | Citation | Role |
+|-----------|----------|------|
+| `peters_1963` | Peters & Mathews 1963, Phys. Rev. 131:435 | Primary formula: gravitational-wave power loss and orbital period derivative for eccentric binary (Peters formula, Eq. 17) |
+| `damour_1991` | Damour & Taylor 1991, ApJ 366:501 | Observable-only GR PBDOT formula (Eq. 5.4) — basis for baseline_a; no mass inputs needed |
+| `will_2014` | Will 2014, Living Rev. Relativ. 17:4 | GR tests of gravity review; chirp-mass PBDOT form (Eq. 108) — basis for baseline_c |
+| `weisberg_2016` | Weisberg & Huang 2016, ApJ 829:55 | Hulse-Taylor timing; explicit (m1, m2) PBDOT formula (Eq. 22) — basis for baseline_b |
+| `manchester_2005` | Manchester et al. 2005, AJ 129:1993 | ATNF Pulsar Catalogue description paper; primary data source |
+| `kramer_2006` | Kramer et al. 2006, Science 314:97 | Double-pulsar timing; validates PBDOT at the 0.1% level; supporting evidence for baseline_a |
+| `hulse_1975` | Hulse & Taylor 1975, ApJ 195:L51 | Discovery of the first binary pulsar (PSR B1913+16); historical context |
+
+## How to find equation locations
+
+- `summary_formula_peters_1963.md` — Peters 1963 eccentricity enhancement f(e), Eq. 17
+- `summary_formula_damour_1991.md` — Damour & Taylor 1991 observable-only form, Eqs. 5.2d, 5.4–5.6
+- `summary_formula_will_2014.md` — Will 2014 chirp-mass form, Eqs. 88, 108
+- `summary_formula_dataset_weisberg_2016.md` — Weisberg & Huang 2016 explicit-mass form, Eq. 22
+- `summary_dataset_manchester_2005.md` — ATNF catalogue schema, Tables 1–2
+- `summary_formula+dataset_kramer_2006.md` — Double-pulsar timing; PBDOT measurement context
+- `summary_supporting_hulse_1975.md` — Discovery context; no formula sourced from this paper
+
+## Data source and license
+
+Data: ATNF Pulsar Catalogue v2.7.0, distributed via psrcat tarball from
+https://www.atnf.csiro.au/research/pulsar/psrcat/download.html
+
+License: GNU General Public License v3.0 or later (GPL-3.0-or-later).
+The psrcat software and database are Copyright CSIRO/ATNF; freely
+redistributable per the GPL. Required citation: Manchester et al. (2005),
+AJ 129, 1993 (DOI 10.1086/428488).

binary_pulsar_post_keplerian_atnf__PBDOT/reference/damour_1991.bib

@@ -0,0 +1,10 @@
+@ARTICLE{DamourTaylor1991,
+  author = {{Damour}, T. and {Taylor}, J. H.},
+  title = {{On the orbital period change of the binary pulsar PSR 1913+16}},
+  journal = {ApJ},
+  year = 1991,
+  volume = {366},
+  pages = {501--511},
+  doi = {10.1086/169585},
+  note = {Galactic acceleration + Shklovskii kinematic corrections to PBDOT_obs.},
+}

binary_pulsar_post_keplerian_atnf__PBDOT/reference/damour_1991.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f9d89ab0e4985f11d4c5bf0b8856917cad4ab08e144c90a6afa2c50018b2341
+size 1183249

binary_pulsar_post_keplerian_atnf__PBDOT/reference/damour_1991.txt

@@ -0,0 +1,1455 @@
+
+=== PAGE 1 ===
+
+1991ApJ. . .366. .50ID
+
+The Astrophysical Journal, 366:501-511,1991 January 10
+© 1991. The American Astronomical Society. All rights reserved. Printed in U.S.A.
+
+ON THE ORBITAL PERIOD CHANGE OF THE BINARY PULSAR PSR 1913 + 16
+Thibault Damour
+Institut des Hautes Etudes Scientifiques; and DARC—Observatoire de Paris
+AND
+J. H. Taylor
+Joseph Henry Laboratories; and Physics Department, Princeton University
+Received 1990 April 10 ; accepted 1990 June 29
+ABSTRACT
+We reexamine the theoretical significance of the raw observational parameter called the “rate of orbital
+period change,” PJbs, of the binary pulsar PSR 1913 + 16. We show that the current precision (~0.8%) on the
+determination of PJbs makes it necessary to take explicitly into account the effects of the galactic accelerations
+of the pulsar and the Sun, and that of the proper motion of the pulsar. Several other possible contributions to
+P£bs are (re)examined and found negligible. As the value of the galactic contribution to Pb/Pb depends explicitly on the distance to the pulsar, say d, we have been led to reexamine the determination of d from dispersion
+measurements. We find that recent progress in H i absorption measurements in the first galactic longitude
+quadrant allows one to constrain the mean electron density along the line of sight to PSR 1913 + 16 (/ = 50°)
+to the range fic(50°) = ([6.39 + 0.93][R0/l kpc])-1 cm-3, where R0 is the galactocentric distance of the Sun.
+This value for ñe is smaller than the “standard” one and leads to a galactic-reduced distance to PSR
+1913 + 16, d/R0 = 1.08 + 0.16. After subtraction of the galactic effects, the latest experimental results yield a
+0.8% confirmation of the general relativistic prediction: P£bs~ßal/P?R = 1.0081 + 0.0022(galactic)
++ 0.0076(observational). Alternatively, this result yields an upper bound to the rate of change of Newton’s
+gravitational constant, G/G = (1.10 ± 1.07) x 10“11 yr_1, which, in the long term, may be limited to the
++ 3 x 10“12 yr-1 level of precision because of the uncertainties in the values of the galactic constants, R0
+and v0.
+Subject headings: gravitation — pulsars — relativity — stars: binaries — stars: stellar dynamics
+I. INTRODUCTION
+The binary pulsar PSR 1913 + 16 has held up to, and even
+surpassed, the expectations raised by its discovery in 1974
+(Hulse, and Taylor 1975) in providing us with a new laboratory
+for relativistic gravity. The regular acquisition of timing data
+with steadily improving precision has motivated several stages
+of corresponding improvements in the theoretical description
+of this system (for a recent review see Taylor and Weisberg
+1989). A partial list of these successive theoretical improvements is: the inclusion of the relativistic time dilation of the
+pulsar clock (Blandford and Teukolsky 1976); the addition of
+timing corrections linked with gravitational retardation of the
+pulsar signals and with 0(v2/c2) relativistic orbital effects
+(Epstein 1977, 1979; Haugan 1985; Damour and Deruelle
+1985, 1986); the timing effects due to aberration (Smarr and
+Blandford 1976; Damour and Deruelle 1986); the calculation
+of higher order relativistic contributions to the periastron
+advance—both spin-orbit (Barker and O’Connell 1975) and
+0(v*/c4) orbital effects (Damour and Schäfer 1988). Each time,
+these refinements have been motivated by the necessity to
+match the theoretical accuracy of the description of the system
+with the current observational precision. The aim of the
+present work is to refine further the link between the raw
+observational parameter traditionally called the “rate of
+orbital period decay,” PJbs, and the corresponding theoretical
+quantity, Pj,heor, defined as the contribution to the intrinsic
+orbital period decay due to a particular relativistic theory of
+gravity (i.e., P^*™ is the orbital decay that would be observed
+in the center-of-mass system of an isolated binary system
+whose internal dynamics are ruled entirely by some theory of
+
+gravity). This refinement is required by the present observational precision of P£bs, which has now reached a level ~0.8%
+(Taylor and Weisberg 1990).
+The theoretical importance of the Pb measurement has long
+been recognized as a direct confirmation that the gravitational
+interaction propagates with velocity c (which is the immediate
+cause of the appearance of a damping force in the binary pulsar
+system) and thereby as a test of the existence of gravitational
+radiation and of its quadrupolar nature (Wagoner 1975a;
+Eardley 1975; Will and Eardley 1977; Weisberg and Taylor
+1981; Will 1981; Damour 1983a, b). More recently, it has also
+been recognized as a new way to constrain the time variability
+of the Newtonian gravitational constant (Damour, Gibbons
+and Taylor 1988). In view of this theoretical importance it is
+essential to have control of all the effects that could contribute
+an apparent orbital period change, i.e., that would contribute
+to Pf35, while not being part of Pf™ (as defined above). A
+(probably partial) list of the effects which have been considered
+in the literature is: (1) acceleration of the center of mass of the
+binary system with respect to the solar system due to the gravitational field of the Galaxy (Brumberg et al 1975; Wagoner
+1975h; Shapiro and Terzian 1976; Will 1976, 1981); (2) mass
+loss from the pulsar or its companion (Smarr and Blandford
+1976; Will 1981); (3) tidal dissipation effects in the case the
+companion is a helium star or a rapidly rotating white dwarf
+(Smarr and Blandford 1976; Will 1981); (4) acceleration by a
+third body (Smarr and Blandford 1976; Will 1981); (5) dynamical friction caused by an enveloping gas cloud (Baroni et al.
+1979, 1980); (6) perturbation by external ultra-low-frequency
+gravitational waves (Bertotti, Carr, and Rees 1983).
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 2 ===
+
+1991ApJ. . .366. .50ID
+
+502
+
+DAMOUR AND TAYLOR
+
+In this work we shall reexamine some of the effects listed
+above, notably (1) and (2), and add more effects to this list.
+Our main conclusion will be that, contrary to the conclusions of Shapiro and Terzian (1976) and Will (1976, 1981),
+the galactic effects are not contributing at the negligible level
++3 x 10 13 yr 1 (i.e., ~10-4 times less than
+hs
+Pl /Pb ^ —2.76 x 10-9 yr-1 and of opposite sign), but rather
+at the level ~ -2 x 10"11 yr"1 (i.e, the 0.7% fractional level,
+which is comparable to the present observational 1 a accuracy,
+and of the same sign). All the other effects that we shall consider appear to be negligible (or very implausible). In order to
+compare conveniently the relevance of various effects, we shall
+systematically consider their contributions to Plhs/Pb9 and use
+as a basic unit the quantity
+
+Vol. 366
+
+d = \xt — x0\ the distance between the pulsar and the Sun.
+The (angular) galactic coordinates of PSR 1913 +16 are
+/ = 49?97 - 50?0 ,
+(2.3a)
+b = 2?12 - 2?1 .
+(2.3b)
+The nonzero value of b means that PSR 1913 + 16 is slightly
+above the galactic plane, by an amount
+
+(100 lems-1)2
+= 1.0810 a .
+(1.3)
+c(kpc)
+The link between the “observational” parameter Plhs/Plhs
+(defined by fitting the raw data to the eqs. [1], and [7]-[14] of
+Taylor and Weisberg 1989), and the corresponding
+“ theoretical ” parameter can be written as
+
+z = d sin h ^ 37 x (d/1 kpc) pc .
+(2.4)
+Therefore the #i10 • a1 term in equation (2.2) contains a contribution ^ — sin h Kz/c, where Kz denotes the vertical component of the galactic acceleration at the pulsar location. For
+d ^ 8.3 kpc (see below), which implies z ^ 307 pc, one has
+Kz ^ 3.0 x (1.65 x 10~9 cm s-2) (Fig. 7 in Bahcall 1984), and
+hence a contribution to (Pb/Pb)gil1 approximately equal to
+— 1.6 x 10“3a, with an uncertainty of perhaps 30%. In fact
+this value is a slight underestimate because the pulsar is
+located (see below) at a galactic radius 12% smaller than the
+Sun, so that Kz, for a given z, is slightly bigger than at the solar
+circle. Because the Sun is located within about 20 pc of the
+plane of symmetry of the mass distribution in its neighborhood, the corresponding term for the solar acceleration in the
+z-direction is negligible.
+Keeping in mind the small “vertical’ contribution, to be
+added in the end, we shall henceforth concentrate on the
+“planar” contribution obtained by assuming that the pulsar
+and Sun are located within the galactic plane and acted upon
+only by the radial component of the galactic acceleration.
+From equation (2.2) this planar contribution can be written in
+the following simple explicit form
+
+The various contributions to equation (1.4) will be discussed in
+the following order: galactic accelerations in § II, various other
+acceleration effects in § III, “mass losses” (and various other
+dissipation effects) in § IV. The theoretical contribution will be
+discussed in § V. Finally, § VI will contain a summary of the
+various contributions, a comparison with the observational
+data, and a discussion of the significance of the results.
+
+- »o cos / vi C0Sl +
+(2.5)
+cR(\
+cRt
+7dwhere v0 is the equilibrium galactic circular velocity at the
+Sun’s position (often denoted ©0), R0 the distance from the
+galactic center to the Sun, vt and
+the corresponding quantities at the pulsar location, and À the angle between the Sun
+and the galactic center as seen from the pulsar (see Fig. 1). Note
+that the circular velocities v0 and
+have entered only as a
+means of expressing the galactic radial accelerations a0 =
+Vo/Ro’ ai = Vi/Ru and that we are not assuming that the
+pulsar is (as the Sun is to a good approximation) moving in a
+circular galactic orbit. Simple triangle relations give
+
+a = 10“18 s"1 = 3.15576 x 10“11 yr"1 ,
+
+(1.1)
+
+which corresponds roughly to the one percent level of what is
+observed. The following conversion coefficients between some
+characteristic accelerations (divided by the velocity of light, c)
+and the unit a will be useful to remember :
+^.4.6493 x 10-««,
+
+(U)
+
+II. GALACTIC EFFECTS
+The “galactic” term in equation (1.4) comes directly from
+the time derivative of the (first-order) Doppler factor,
+1 + VjJc = Plhs/Plb\ connecting the observed binary period,
+P£bs, to the intrinsic one, Pj,nt. Here
+vR = ni0-(v1 -v0)
+
+(2.1)
+
+denotes the radial velocity of the center of mass of the pulsar
+xlf relative to the solar system barycenter x0, with n10 =
+(xi — xo)Z l*i — *o I denoting the unit vector directed from the
+solar system, index 0, toward the binary pulsar, index 1. Differentiating equation (2.1) leads to
+— a0) +
+
+(2.2)
+
+where a0 and ¿jq are the acceleration of the solar system and
+the center of mass of the binary system in the Galaxy, vT =
+l^io x (^i — t>o)l is the transverse relative velocity, and
+
+2
+(d/R0) - cos /
+x cos l + f£i\
+, (2.8)
+Vo/ 1 - 2(d/R0) cos l + (d/R0)2
+where // = vT/d is the angular motion of the pulsar as seen from
+the solar system, i.e., its proper motion.
+Up to now we have made no assumption concerning the
+rotation curve of the Galaxy, i.e., the dependence of r on R. We
+shall assume henceforth a simple linear rotation curve,
+—
+
+K
+
+*0
+
+)’
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+(2.9)
+
+=== PAGE 3 ===
+
+1991ApJ. . .366. .50ID
+
+ORBITAL PERIOD CHANGE OF PSR 1913 + 16
+
+No. 2, 1991
+Sun
+
+503
+
+Harris 1989, R0 = 7.5 + 0.9 kpc, which is nicely consistent
+with eq. [2.14]). Concerning v0 we are aware of no improvements since the review of Kerr and Lynden-Bell (1986), who
+recommend the value
+v0 = 222 + 20 km s"1 .
+
+(2.15)
+
+For the fractional slope parameter, a recent study of the rotation curve of the Galaxy (Fich, Blitz, and Stark 1989) leads to
+the extremely small value b0 = —0.003 as the best fit (for a
+linear rotation curve) for v0 == 220 km s-1 and R0 = 8.0 kpc,
+with b0 ranging between + 0.03 if R0 ranges between 7-9 kpc.
+In other words, we can safely take
+b0 = 0.00 ± 0.03 .
+
+(2.16)
+
+For the pulsar-dependent quantities, we note the Galactic
+longitude /, is known with good precision (eq. [2.3a]), and the
+proper motion has now been measured (Taylor and Weisberg
+1989,1990):
+//a = cos Sdcc/dt = —3.21 + 0.35 mas yr-1 , (2.17a)
+Fig. 1.—Schematic diagram showing the relative positions of the Sun and
+PSR 1913 + 16, projected onto the galactic plane. The pulsar lies at Galactic
+longitude / and distance d from the Sun; the two objects have galactocentric
+distances R0 and Rx.
+depending only on the new parameter, the fractional velocity
+gradient (“ slope parameter ”) at the Sun’s radius :
+/ R dv_\
+_A+B
+(2.10)
+\v dRjR=Ro ~ A — B ’
+where connection has been made with the traditional Oort
+parameters A and B. As the pulsars we shall consider will never
+be at galactic radii R much larger or much smaller than R0, the
+use of a linear rotation curve is both numerically valid and
+analytically convenient. In particular it allows us to express
+analytically the galactic effect (2.8) in terms of observable
+quantities. We will evaluate the function
+(PJPbr' = g(R0, v0, b0, ô, n, l),
+
+(2.11)
+
+defined by
+(v 0, v0,...)g(R
+X < cos / +
+
+v
+°
+cRo
+
+js
+[1 + b0(l - Vsin2 i ++)]:
+sin2 l + ß+ tr
+
+Rob
+
+(2.12)
+
+where we have introduced the short-hand notations
+
+fi0 — dô/dt = + 1.04 + 0.42 mas yr-1 .
+
+Knowing that they have a normalized covariance of +0.3147,
+we get for the total proper motion
+t* = (mI + Vô)112 = 3-43 + 0.32 mas yr-1 .
+(2.18)
+(In eqs. [2.17]-[2.18], and only there in this paper, a and Ô
+represent right ascension and declination.)
+In remains to estimate the value of the reduced distance to
+PSR 1913 + 16, ô = d/R0. The only useful distance indicator
+available is the dispersion measure (Taylor and Weisberg
+1982),
+[Z)A/] 1913 + 16 = 168.77 + 0.01 cm-3 pc .
+(2.19)
+Dispersion measure is the integral over the line of sight of the
+electron density, DM = ¡ ne ds, so one needs a model of the
+galactic electron density to convert DM into an estimate of the
+distance d = ¡ds. Reciprocally, the models of electron density,
+ne, are calibrated by means of independently measured pulsar
+distances : mainly through the measurement of the shape of the
+21 cm absorption line of broad-band pulsar signals by cool
+neutral-hydrogen clouds located between the Earth and the
+pulsar. The net observational results of H i absorption measurements consist of radial velocities, which can be converted
+into distance estimates only by assuming some galactic rotation model. Until quite recently, these conversions have usually
+relied on the outdated Schmidt (1965) model of the Galaxy.
+This model used, in particular, a galactic radius R0 = 10 kpc
+and a solar velocity v0 = 250 km s_1. The link between the
+radial velocity, vR, of a cloud (assumed to be in circular galactic
+motion) and its distance, d, is
+
+ô = d/R0 (“reduced distance” to the pulsar), (2.13a)
+ß = (d/R0)-cos l.
+
+R0 = 7.7 ± 0.7 kpc .
+
+— i?0 ) sin /,
+
+(2.13h)
+
+We need now the values of the various observable quantities,
+& I appearing as arguments in the galactic acceleration function g. The best current estimate of the galactic
+radius R0, as deduced from a weighted average of several
+methods employing independent calibrations, is (Reid 1989)
+(2.14)
+
+(We are aware of only one more recent determination, not
+taken into account by Reid 1989; it is the one of Racine and
+
+(2.17b)
+
+(2.20)
+
+where v is the circular velocity at the galactic radius of the
+cloud, R = (Ro + d2 — 2R0d cos
+It is useful to remark
+that, if we assume a general linear rotation curve, equation
+(2.9), then equation (2.20) can be solved analytically for d/R0.
+We find
+ô = — = cos l ±
+Ro
+
+(1 + b0)v0 sin .F-
+
+sin
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+(2.21)
+
+=== PAGE 4 ===
+
+1991ApJ. . .366. .50ID
+
+504
+
+DAMOUR AND TAYLOR
+
+where the plus (minus) sign corresponds to a velocity feature
+beyond (before) the “ tangent point,” d = R0 cos /.
+We have considered here the longitude range 0 < / < 90°,
+for which the radial velocity first increases with d from zero up
+to a positive maximum, t$ax = (1 + b0)v0(l - sin 0, when d =
+R0 cos /, and then decreases down to negative velocities (the
+zero crossing being attained for d = 2R0 cos l) (see e.g.,
+Mihalas and Binney 1981). It is to be noted, from equation
+(2.21), that if one converts H i radial velocity measurements
+into reduced distance estimates, ô = d/R09 the conversion
+depends only on the parameter (1 + b0)v0, which has changed
+less in going from the Schmidt (1965) model to recent galactic
+models (e.g., Fich et al 1989) than R0 (whose best recent estimates are smaller by -25% than the “old” IAU value of 10
+kpc). Moreover, even the dependence on (1 + b0)v0 can be
+eliminated if the maximum radial velocity is clearly present in
+the absorption or emission profiles and is used to express ô in
+terms of
+Vr
+(1 — sin /) vR
+(1 + b0)v0 sin /
+sin / v'g™ '
+
+(2.22)
+
+From the foregoing comments, it would seem possible to
+obtain a reasonable estimate of the reduced distance to PSR
+1913 + 16 simply by dividing by 10 kpc its “standard”
+dispersion-measure distance of 5.2 kpc, as listed by Manchester
+and Taylor (1981), using the best available (R0 = 10 kpc based)
+electron-density model (Lyne, Manchester, and Taylor 1985).
+However, the latter authors recognize that their model has
+primarily a statistical validity, and may be in error by as much
+as a factor 2 when estimating individual pulsar distances. And
+indeed, it has been suggested by Abies and Manchester (1976)
+that the mean electron density <ne> = J neds/¡ ds could vary
+
+Vol. 366
+
+by more than a factor of 2 depending on the longitude and
+distance ranges considered, and in particular that <ne> seems
+to have a lower value in the interarm region between the Sagittarius and the Perseus arm. The line of sight to PSR 1913 + 16,
+along / = 50?0, happens to lie mainly in this interarm region, as
+it is nearly tangent to the (inner) Sagittarius arm, and cuts the
+(outer) Perseus arm only at a distance d - 1.2R0 (see e.g.,
+Georgelin and Georgelin 1976). This argument is a qualitative
+indication that (d/R0)1913 + 16 might be appreciably greater
+than 5.2/10 = 0.52.
+Fortunately, recent observational progress in the H i calibration of the pulsar distance scale (Heiles et al 1983; Weisberg et al 1987; Frail 1989) makes it possible to confirm this
+indication and render it quantitative. We have proceeded in
+two ways. First, starting from the 20 pulsars in Frail’s list with
+DM > 40 cm-3 pc, Galactic longitudes 35° < / < 70°, and distance estimates (sometimes only lower limits) from H i absorption or supernova remnant associations, we plotted the ratios
+of their standard (Manchester and Taylor 1981; Lyne, Manchester, and Taylor 1985) dispersion-measure estimated distances to their distances computed by Frail (1989) using a
+galactic model with R0 = 8.5 kpc and v0 = 220 km s“1 (Fich et
+al 1989). We take into account the fact that for three of the
+pulsars listed by Frail (PSRs 1907 + 10, 1914+13 and
+1915 +13) the upper limits quoted by him are only “ tentative ”
+limits (as clearly stated by Weisberg et al 1987, who quote
+them in parentheses) and we have replaced them by the secure
+upper limits recomputed [with (1 + b0)v0 = 220 km s“1] from
+the negative radial velocity emission features quoted by Weisberg et al (1987, p. 308). The resulting “recalibration” graph,
+with error bars, is repoduced in Figure 2. The distance ratios
+are fit very well by a simple curve that is constant at 0.55 from
+/ = 70° to / = 47°, rises smoothly to 1.35 at / = 41°, and
+
+Fig. 2.—Ratios of the distances estimated from the galactic electron-density model of Lyne, Manchester, and Taylor (1985) to those listed by Frail (1989) based on
+H i absorption measurements or supernova assocations {filled
+circles). Thick vertical bars denote the nominal range of distances allowed by the absorption
+measurements; thin bars allow for an additional
+±7 km s"1 of radial velocity dispersion in the absorbing clouds. The pulsar names are listed above each bar
+3
+followed by dispersion measures in units of cm pc.
+’
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 5 ===
+
+1991ApJ. . .366. .50ID
+
+No. 2, 1991
+
+ORBITAL PERIOD CHANGE OF PSR 1913 + 16
+
+remains constant at 1.35 for / < 41°. At / = 50° the curve is at
+0.55, which means that at that longitude the 8.5 kpc-based
+distances listed by Frail (1989) are ^1.82 times larger than
+those given by the 10 kpc-based model of Lyne et al (1985).
+This reasoning yields an estimate of d/R0 = 1.82 x (10/8.5)
+x 0.52= 1.11 for PSR 1913 + 16. A rough estimate of the
+uncertainty in this determination, obtained from Figure 2 by
+inspection, is ± 0.20 (based primarly on the distance limits for
+PSRs 1915 + 13 and 1930 + 22).
+This first estimate having confirmed clearly that the mean
+electron density along the line of sight, ñe = J neds/¡ ds, is longitude dependent [ñe = ñe(/)], we then used a second, more
+direct, method to estimate ñe(50°). We selected the longitude
+range 420-58° centered around / = 50°. This range contains
+nine pulsars with H i-estimated distances, four being below
+50°, and five above 50°. In conformity with our discussion
+above we use a galactic model with v0 = 220 km s“1, h0 = 0
+(flat velocity curve), and R0 = 7.7 kpc. This model differs from
+the one used by Frail (1989) only in the value of R0, and
+equation (2.21) shows that H i-estimated distances scales as R0
+for fixed (1 + b0)v0. For each pulsar we can then compute,
+independently of R0, the ratio between the galactic reduced
+distance d/R0 and the dispersion measure, DM = f neds =
+he d, which gives access to
+1 _d/R0
+ne R0 DM ’
+
+(2.23)
+
+For a small enough longitude range, and for comparable
+values of DM, the values of l/ñeR0 deduced from different
+pulsars should be all consistent. (We are here neglecting the
+influence of the galactic latitude, b, which was taken into
+account by Lyne et al. 1985, to concentrate on the influence of
+the longitude. As PSR 1913 + 16 turns out to have a larger
+
+505
+
+value of DM | sin h | than the pulsars that will be most effective
+in determining l/neR0 around / = 50°, this neglecting of the
+influence of b can only lead to slightly underestimating the
+distance to PSR 1913 +16.) As shown in Figure 3, we find that
+the allowed ranges of l/ñeR0 computed for all nine pulsars
+with 42° < l < 58° are consistent. From their intersection we
+determine a “best” range for the value of l/neR0 around 50°.
+The least upper bound is (6.93 + 0.38) cm3 kpc-1, which
+comes from a firm upper limit to the distance of PSR 1913 +10
+(Frail’s upper limit plus his one sigma uncertainty), and the
+greatest lower bound is (5.79 — 0.33) cm3 kpc-1, from a firm
+lower limit to the distance to PSR 1930 + 22 (Frail’s lower limit
+minus one sigma). Our procedure therefore yields
+—= (6.39 ± 0.93) cm3 kpc-1 .
+(2.24)
+|Ko
+A nice check on this value comes from the result of Heiles et
+al. (1983), who find that the detailed H i observations of PSR
+1937 + 21 (/ = 57?5) and its immediate neighbors on the sky are
+of a precise enough nature to determine (d/R0)i937 + 2i as being
+0.5 (rounded down from cos 57?5 = 0.537). This corresponds to
+l/neR0 = 7.13 cm3 pc-1. However, it is difficult to estimate
+what error bars should be put on this value. Frail treats it
+essentially as a lower limit, with a much larger upper limit (see
+Figs. 2 and 3). Backer (1989) suggests a conservative estimate
+of (d/Ro)i937 + 2i = 0.5 + 0.2, which translates into l/ñcR0 =
+7.13 + 2.82.
+The result (2.24), together with the dispersion measure of
+PSR 1913 +16, eq. (2.19), yields
+<5i9i3 + i6 = (-¿-)
+=1.08 ±0.16.
+(2.25)
+\^0/ 1913 + 16
+This value is in good accord with our estimate obtained from
+
+o
+Cl
+ro
+
+£
+O
+
+60
+
+55
+
+50
+45
+Galactic longitude (deg)
+Fig. 3.—The ratio of reduced distance to dispersion measure, ô/DM, for nine pulsars in the longitude range 42° < / < 58°. Error bars are defined in the same way
+as in Fig. 2; horizontal lines correspond to the adopted scale factor and uncertainty range given in eq. (2.24). Pulsar names and dispersion measures are listed above
+each bar.
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 6 ===
+
+1991ApJ. . .366. .50ID
+
+506
+
+DAMOUR AND TAYLOR
+
+Figure 2; it is more localized to / « 50°, and is based on pulsars
+whose dispersion measures are comparable to that of PSR
+1913 + 16, and we shall adopt it in the following. As a consequence of eq. (2.25), and of the value (2.14) for R0, we get
+^i9i3 + i6 = 8-3 + 1.4 kpc .
+(2.26)
+so that when we shall need d (and not only <5) for some rough
+estimates we shall use d ~ 8 kpc (to replace the usual 5 kpc
+nominal value used up to now). Another consequence of equation (2.25) is that the galactocentric radius of the pulsar (see eq.
+[2.6]) is R1 ^ 0.88Ro, so that the pulsar is not far from the
+solar circle (but at a height z ^ 307 above the galactic plane).
+We have now in hand, to the best of our knowledge, estimates of all the observable quantities entering into the apparent orbital decay rate induced by galactic effects, namely R0
+(eq. [2.14]), i;0 (eq. [2.15]), b0 (eq. [2.16]), Ô (eq. [2.25]), // (eq.
+[2.18]), and / (eq. [2.3a]). Inserting their values into eq. (2.12),
+we get for the planar galactic effect
+p \ gal(plan)
+-0.596 + 0.188 a,
+(2.27)
+where a = 10“18 s“1 is the unit introduced in equation (1.1).
+The 1 a uncertainty in equation (2.27) comes from the addition
+in quadrature of the errors on R0, v0, b0, ô, and p. It is interesting to note that although Ô is the most poorly known
+parameter (~15% fractional uncertainty), it contributes only
+negligibly to the total error \_{dg/dô)oô = 0.037a before adding
+in quadrature]. The proper motion uncertainty (9.3%) contributes also negligibly to the total error
+= 0.044a]. In
+fact the uncertainty in equation (2.27) comes almost entirely
+from poor knowledge of the galactic constants R0 K^^o)
+Gr0 = 0.098a] and v0 [_{dgldv0)avo = 0.150a].
+Finally, we must add to equation (2.27) the “ vertical ” galactic effect, estimated above as being
+p \ gal(vert)
+yj
+= -sin by~ -0.0061 ± 0.0018 a . (2.28)
+It should be noted that the basic reason why the present
+redetermination of galactic effects gives a much bigger result
+(and a different sign) than previous estimates (Shapiro and
+Terzian 1976; Will 1976, 1981) is that these previous estimates
+had assumed that the pulsar was (like the Sun) corotating with
+the Galaxy—an assumption which leads to a delicate cancellation between the negative acceleration effects and the positive
+proper motion ones in equation (2.5) (see e.g., Table 1 in
+Shapiro and Terzian 1976). We now know from the direct
+timing measurement of the pulsar proper motion that this
+assumption was invalid, and that the positive proper motion
+contribution reduces, but does not cancel, the negative acceleration ones. Our use of an upward-recalibrated pulsar distance
+(together with a current, lower, estimate of R0) has only the
+effect of amplifying the negative outcome of this noncancellation.
+III. OTHER ACCELERATION EFFECTS
+The general formula (2.2) of the previous section shows any
+influence causing a relative acceleration of the center of mass of
+the binary system with respect to the solar system barycenter
+will contribute a term
+(p \ accel J
+=-/.10-(¿«1-á«0).
+(3.1)
+
+Vol. 366
+
+As we have already taken into account the average acceleration imposed by all the masses in the Galaxy, the additional
+accelerations ôa0, öa^ entering into equation (3.1) are the fluctuations in the galactic acceleration due to a local clustering of
+accelerating centers near the Sun or near the pulsar. This
+problem has been studied in detail in the literature (Holtsmark
+1919; Verweij 1936; Chandrasekhar 1941 ; Chandrasekhar and
+von Neumann 1942), with a result for the probability distribution of the fluctuating acceleration, ôa, given by the
+“ Holtsmark ” distribution :
+K ut* f,
+J
+V
+
+A
+J
+
+1 H(ß)d3a
+4* ß2 Q3h’
+
+(3.2)
+
+where
+H(ß) = 71
+"^ Í e~ixlß)3/2x sin xdx ,
+P Jo
+
+(3.3a)
+(3.3b)
+
+Qh =
+
+(3.3c)
+
+,
+
+(3.3d)
+
+L <M> J
+
+The averages in eq. (3.3d) are taken over the mass spectrum of
+the attracting centers, and p = n<M> denotes the total mass
+density. It should be noted that for /? -► oo (high fluctuations of
+the perturbing acceleration), H(ß) ~ (l5/S)(2/n)112 ß'5'2, which
+implies that the variance of ôa, <(áa)2>, is infinite. This disquieting remark motivates our present discussion of the probable effects of fluctuating gravitational accelerations.
+Since we have a full knowledge of the statistical distribution
+of ôa we can define, instead of the infinite standard deviation,
+an error bar for ôa by considering some suitable confidence
+level. More precisely, we see from equation (3.1) that we are
+interested not in the modulus of ôa but in the value of the
+projection of ôa along some fixed (nonfluctuating) direction,
+«10 =
+say. A direct calculation from the (isotropic) distribution (3.2) for ôa, shows that the integrated probability distribution of ôaz = e: ■ öa is
+Probability (|<5az| > z0QH) = £ dß^l - jjwß). (3.4)
+As a compromise between quoting, as is usual, a “1 <r” error
+bar, and taking into account the significant tail of the distribution, we choose to define an error bar on 3az as being one-half
+the I öaz I which has a 4.55% probability of being exceeded (i.e.,
+our “1 cr” equals half the usual “2 cr”, or 95.45% confidence
+level). Performing the integral (3.4) leads then to
+I àaz l-i <t” = 6.17GM1/3/?2/3 ,
+
+(3.5a)
+
+i.e., in practical units.
+• ôa
+
+= 0.29 x 10~2|
+
+\ 2/3
+
+According to Mihalas and Binney (1981, p. 224) the total
+mass density around the solar galactic radius due to stars and
+white dwarfs is -0.065 M0 pc-3 with an M ~ 1 M0. This
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 7 ===
+
+1991ApJ. . .366. .50ID
+
+No. 2, 1991
+
+ORBITAL PERIOD CHANGE OF PSR 1913 + 16
+
+yields for the P/P contribution of the fluctuating acceleration
+due to stars on the pulsar (which is located at Rl ^ 0.88 R0)
+p \ * on 1
+-M
+= +4.7 x 10-4 a ,
+(3.6)
+,bj 1 a
+which is in fact an upper bound because the pulsar is at a
+height z « 300 pc above the galactic plane, where p(z) < p(0). A
+priori the same “1 a” upper limit applies to the —/i10 • ôa0/c
+fluctuating acceleration on the solar system barycenter.
+However, in the case of the Sun, we know rather well the actual
+distribution of nearby stars (e.g., Allen 1973, ch. 12) and we
+could, if necessary, compute and subtract the corresponding
+P/P effect. To do this in detail is unnecessary because the effect
+is negligible, as attested by the fact that the nearest massive
+stellar system (a Cen) of total mass MA + MB ^ 2 M0, located
+1.3 pc away, exerts a gravitational pull GM/cd2 ^ 5.5 x 10-4a,
+which happens to be exerted nearly orthogonally (94?2) to the
+#i10 direction of interest so _5
+that its contribution to P/P is only
+—#i10 • Sa0/c ^ +4.0 x 10 a.
+On the other hand, the analytical expressions (3.5) suggest
+that one should worry about the gravitational effect of giant
+molecular clouds which, according to Scoville and Sanders
+(1986), have masses up to 4 x 106 M0 and contribute, near the
+solar circle, a mean density of 0.033 M0 pc~3. This necessitates
+doing another statistical analysis, because the space distribution of giant molecular clouds is essentially two dimensional,
+they being confined to the galactic disk with a half-width at
+half-maximum ~75 pc. Using, for simplicity, only the “nearest
+neighbor approximation ” (Chandrasekhar 1941) for the fluctuating acceleration, one finds that a plane distribution of
+attracting centers exerts on a point within the plane a fluctuating acceleration with an isotropic distribution given by
+/
+1 j \ 1
+d2a
+Probability \ ôa = a + - da ] = - Ga —r3
+(3.7)
+V
+-2)2
+|«|
+where a is the total surface mass density of the attracting
+centers. Defining as above a “1 a” = ^(95.45% confidence
+level) we then deduce from (3.7) the following expected “1 <j ”
+value for the projection of ôa along some fixed direction within
+the plane:
+2 — dim
+”io ' Sa
+a.
+(3.8)
+C
+1 O'”
+According to Dame et al. (1987), the contribution of the mass
+density due to molecular clouds is <r = 1.3 M0 pc 2 in the
+vicinity of the Sun. Assuming this estimate to be valid anywhere close to the solar galactic circle, we expect an effect
+GMC on 1
+= ±6.6 x lO“3 a ,
+(3.9)
+“1 O'”
+which is an upper bound because the pulsar is ~ 300 pc above
+the nearly planar distribution of clouds, which cuts off* the
+useful “ nearest neighbor ” tail of the fluctuating attraction. As
+in the case of stars discussed above, the same “1 a ” limit
+applies also to the —n10* ôa0/c fluctuating acceleration on the
+solar system barycenter (without the diminishing influence of a
+nonzero height above the galactic plane). However, in this case
+we know in some detail the actual distribution of nearby giant
+molecular clouds (Dame et al. 1987), and we could estimate the
+corresponding P/P effect. In fact, there is a string of close
+massive clouds (Aql Rift, Vul Rift, Cyg Rift, Cyg 0B1) which
+
+507
+
+pull the Sun roughly toward / = 50°, thereby contributing an
+effect
+~ —3 x 10 3 a ,
+
+(3.10)
+
+consistent with the estimated “ 1 cr ” effect of equation (3.9).
+Another type of acceleration effects one must keep in mind is
+the fact that the orbital motion of PSR 1913 + 16 is observed
+from the moving Earth, and not from the solar system barycenter, and that the motion of the Earth around the solar
+system barycenter is known with less precision than the
+motion of the Earth around the Sun. For instance, a small
+change in the masses of the outer planets will affect the ephemeris of the Earth and thereby modify the “observed” Pb (as
+deduced from fitting the data to a timing formula which uses
+some Earth ephemeris). Since GM0/c(AU)2 = 1.98 x 107a this
+effect is quite sensitive to any small error in the masses of the
+outer planets. One test for this type of effect has been carried
+out by Fairhead (1989). Any unmodeled acceleration of the
+Earth affects equally the orbital period and the period of a
+pulsar itself, Pp/Pp = Pb/Pb, and Fairhead has studied the
+change of apparent Pp, for PSR 1937 + 21, when using the
+same atomic time scale but two different ephemerides (namely
+PEP 740R from the Center for Astrophysics and DE 200 from
+the Jet Propulsion Laboratory). He found ôPp (PEP 740R-DE
+200) = 1.9 x 10“ 23, which (with Pp = 1.5578 ms) yields
+yj
+
+= +1.2 x 10-2 a ,
+
+(3.11)
+
+an effect twice bigger than the maximum expected gravitational perturbation by giant molecular clouds, equation (3.9).
+Finally, let us note for completeness that some authors
+(Smarr and Blandford 1976; Will 1981) have pointed out that if
+PSR 1913 + 16 were part of a triple system, the interpretation
+of Pb would be affected because of the term n10 • ôajc.
+However, our subsequent understanding (even admittedly
+rough) of the evolution and formation of binary pulsars makes
+quite implausible the survival of such a triple system, and we
+shall therefore not pursue this possibility here.
+IV. MASS LOSS FROM THE SYSTEM
+We shall reconsider the effect on P£bs of mass loss by the
+pulsar or its companion because its treatment in the literature
+contains an error of a factor 4 (Smarr and Blandford 1976; Will
+1981), and because no upper bound for the probable contribution of the companion has been given. If either star loses massenergy at a rate ma (a = 1 for the pulsar, and 2 for the
+companion) so that there is no linear momentum loss in the
+instantaneous rest frame of the star, the (intrinsic) orbital
+period will change at a rate given by
+Pb\m
+2m1
+2m2
+(4.1)
+m
+PbJ
+i + w2 mi -\- m2
+This result was first derived by Jeans (1924, 1925) and can be
+also recovered from the effect of a time variation of G, Pb/Pb =
+— 2G/G (see e.g., Damour, Gibbons, and Taylor 1988) since the
+gravitational coupling constant ruling the relative orbital
+motion is the product G(mx + m2). (Probably due to an oversight, Smarr and Blandford [1976] and Will [1981] show
+factors of y, rather than 2, in the numerators of equation [4.1]).
+If the pulsar and its companion lose energy in the form of fields
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 8 ===
+
+1991ApJ. . .366. .50ID
+
+508
+
+DAMOUR AND TAYLOR
+
+and relativistic particles, the corresponding mass-energy loss
+can be identified with the loss or rotational kinetic energy,
+ÈT - mac2, where £'°* = \IaQ.2a. Here Ia denotes the moment
+of inertia of body a (a = 1 or 2) and
+is given in terms of the
+corresponding spin period by na = 2n/Pa. Hence
+2ma
+4n2 h 2 Pg
+ml + m2
+mac Pi '
+
+(4.2)
+
+At the increase of the spin period (Pi = 59.03 ms) of PSR
+1913 +16 is directly observed,
+= 8.626 x 10"18 (Taylor and
+Weisberg 1989), we can directly estimate the contribution of
+mass loss from the pulsar itself [assuming Ix = (1.0 ± 0.3)
+x 1045 g cm2 and using the timing-determined values of the
+masses,
+= 1.442 M0, m2 = 1.386 M0; Taylor and Weisberg 1989]:
+(P \titl
+
+^ +(6.6 ± 2.0) x 104 a .
+
+(4.3)
+
+No direct observations give access to the rotational energy
+loss of the unseen companion. However, it is important to
+place an upper limit on P2/Pl because Smarr and Blandford
+(1976) have pointed out that if the companion were a Crab-like
+active pulsar the contribution (Pt/P^2 could be comparable
+to (minus) the observed orbital period decay. Fortunately, it is
+possible to place a fairly rigorous limit on P2 in the following
+indirect way. First, the spin period decay of pulsars is statistically well represented by a simple law of the form (Lyne,
+Manchester, and Taylor 1985),
+PP = Tf exp (~2t/tB),
+6
+
+where tB = 9.1 x 10 yr and where In
+variable with mean value
+
+(4.4)
+
+is a Gaussian statistical
+
+Ini* = \n(ABf) = In(0.55 x 10“15s),
+
+(4.5a)
+
+and standard deviation
+= 2<TlnB. = 1.38 .
+(4.5b)
+Note that Lyne et al. (1985) express their result in terms of a
+statistical distribution of “initial surface magnetic field,” Pj.
+However, phenomenologically speaking, it is simpler to think
+directly in terms of the statistical distribution of t* = ABf
+(which has the advantage of incorporating the variations of the
+factor A oc R6/I). The solution of equation (4.4) is
+p2(t) = P2(0) + tBTll - exp (-2t/tBy\ .
+
+(4.6)
+
+2
+
+If we consider the case in which P (0)/2ti tB/2, [i.e.,
+3 x 103 yr i 4.55 x 106 yr for P(0) ~ 10 ms, and ~ Tf;
+we shall see below that this condition holds], the solution (4.6)
+can be approximated by P2(t) = 2t, í, so that
+P
+
+1
+
+As the probability distribution of t, is assumed known, we
+can, as above, define an effective “la” level for P/P3 oc 1 as
+being one-half the value corresponding to a tí(2(T) such that
+Probability (t^1 > t^) = 95.45%. One finds lnTii2<r) = Inr,- 1.70alnt., i.e., ii(2ff) = e~2 34r6'ci = 0.53 x 10~1¿ s, and
+thereby, using equations (4.2) and (4.7),
+
+Vol. 366
+
+The estimate (4.8) depends on the age of the companion as a
+rotating neutron star, i.e., the time of formation of the binary
+system PSR 1913-1-16 (through the supernova event that
+formed the younger companion). We can get an estimate of the
+age t by noticing that the angular measurements (2.3) and (2.17)
+imply that PSR 1913 + 16 is above the Galactic plane (b =
+2? 12) and moving away from it with angular velocity
+jub = f> = 3.38 + 0.31 mas yr-1 ,
+
+(4.9)
+
+(the longitude component of its proper motion happens to be
+small, = cos b t = —0.61, and does not enter the present
+argument). As the corresponding linear velocity vb = A.lAfib(d/
+1 kpc) ~ 130 km s_ 1 is large compared to the random vertical
+velocities caused by the attraction of the galactic disk, and is
+directed away from it, it is clear that we are seeing the direct
+runaway of the binary pulsar, out of the galactic plane, after its
+having received a “kick” during the last supernova. As the
+scale height of progenitors of systems like PSR 1913 + 16 is
+small compared to its present height (z ~ 300 pc from eq.
+[2.4]), a direct estimate of the age of the system is simply
+z tan b ^ ^
+,
+i = 7 = —g— = 2.26 x 106 yr .
+
+(4.10)
+
+Assuming I2 ^ 1045 g cm3 in equation (4.8), we finally obtain
+(p \m2
+^ ^+7.3xl0-3a,
+(4.11)
+more than 10 times bigger than the directly constrained pulsar
+mass-loss effect (4.3). Note that the corresponding “1 a ”
+energy loss is -£r2ot - 1.8 x 1034 ergs s'1. This upper limit is
+appreciably smaller than the observational upper limit on the
+averaged isotropic X-ray luminosity obtained by Davidsen et
+al. (1975), Lx < 1.1 x 1035 ergs s'1 (scaled up to the present
+distance estimate d ~ 8 kpc).
+Let us note that we disregard here the possibility of mass
+accretion (ma > 0), and of a corresponding dynamical friction
+caused by PSR 1913 + 16 being embedded in a diffuse gas
+cloud. This possibility, considered by Baroni et al. (1979,1980),
+appears very implausible because of the formation of the
+system ~2 x 106 yr ago and its subsequent runaway to
+z ~ 300 pc above the galactic plane. Moreover, a gas cloud is
+strongly constrained by the observational upper limits on any
+orbital phase-dependent dispersion variations (Taylor and
+Weisberg 1982).
+Finally, to complete this section dealing with “dissipative”
+effects happening in the binary system, let us remark that the
+various tidal dissipation effects considered by Smarr and
+Blandford (1976) and Will (1981) are significant only if the
+companion is a (stripped) helium star or a rapidly rotating
+white dwarf. The possibility of such companions (not to speak
+of the contrived nature of the dissipation mechanisms which
+have to be evoked to get a sizable effect) is stronly constrained
+both theoretically and observationally. See the discussion in
+§ Ylb of Taylor and Weisberg (1989), which concludes that the
+companion is almost certainly a neutron star.
+V. THEORETICAL ORBITAL PERIOD CHANGE
+In § I we defined the “ theoretical ” contribution to Pb as the
+orbital period change that would be observed in the center-ofmass system of an isolated binary system whose internal
+dynamics are ruled entirely by some relativistic theory of
+gravity. In this work we shall consider only the class of gravita-
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 9 ===
+
+19 91ApJ. . .366.
+
+No. 2, 1991
+
+ORBITAL PERIOD CHANGE OF PSR 1913 + 16
+
+tion theories smoothly connected with Einstein’s theory, i.e.,
+theories that contain some parameters, say e (which represents
+one or several parameters), such that their predictions differ
+from the general relativistic ones by terms ~ e + a2 €2 + • ••
+that tend smoothly to zero when € -► 0. Examples of such theories are the Jordan-Brans-Dicke one (Jordan 1959; Brans and
+Dicke 1961; with € = a>-1), the nonsymmetric gravitational
+theory [Einstein 1925, 1954; Moffat 1979, 1989; with € =
+(/e>/p>/n>/c)]> and a recently developed model (Damour,
+Gibbons, and Gundlach 1990) which contains two independent small parameters, a combination of which allows one to
+introduce a time variation of Newton’s constant, G, independently of its locally influenced space variation (so that we can
+take e = G/G). In a pure general relativistic description the
+theoretical value of Pb can be written as
+.?R
+
+P
+
+v(GMn)5/3
+
+^7/2 [1 + XGR-, >
+
+(5 1}
+
+where
+M = m1 + m2 ,
+
+(5.2a)
+
+2
+
+v = m^JM ,
+
+(5.2b)
+
+« = 2n/Pb ,
+
+(5.2c)
+
+e
+
+P4Í T) = 1 + 24
+
++ 96
+
+>
+
+(5.2d)
+
+and where eT denotes a suitably defined relativistic eccentricity
+(Damour and Deruelle 1986). The term X$R denotes higher
+order relativistic corrections [starting at order (rorb/c)2]. The
+lowest order term in equation (5.1) (when XpR = 0) was first
+heuristically written down by Esposito and Harrison (1975)
+and Wagoner (1975a), using previous results of Peters and
+Mathews (1963). It has been more rigorously derived by a
+direct dynamical calculation of secular orbital effects in relativistic binary systems (Damour 1983a, b). The higher order
+fractional correction term XpR is of the form
++
+
+(5.3)
+
+where the explicit expression of X2R has been recently
+(heuristically) obtained by Blanchet and Schäfer (1989). The
+numerical value of X%R/c2 in the case of PSR 1913 +16 is only
++ 2.15 x 10~5. Such a fractional correction is negligible both
+compared to the observational precision on Pb (now around
+0.8%), and compared to the sum of the additional contributions discussed above (see Table 1), and we shall henceforth
+neglect it.
+Besides the observationally well-determined parameters Pb
+and eT, the theoretical result (5.1) contains the two a priori
+unknown masses of the pulsar and its companion mum2. Fortunately there are two other observational parameters, the rate
+of periastron advance, cb, and the time dilation parameter, y
+(Blandford and Teukolsky 1976), whose measurement allow
+one, within each theory of gravity, to determine m1 and m2, or
+equivalently M = ml + m2 and x2 = m2/M (and therefore v =
+x1x2 with xl = mJM = 1 — x2).
+In general relativity the resulting expression for the theoretically predicted Pb in terms of the observables Pb, eT, <b, and y is
+=
+
+“ 157!/c5/2[(1 + 4r,)l'2 ~ 1 ~ ^
+
+’
+
+(5 4)
+
+-
+
+509
+
+where we have introduced the following combinations of
+observed parameters :
+á>P
+k = ——b (fractional periastron advance per orbit), (5.5)
+
+^
+
+3ny
+kej{\ — 6t) ’
+
+(5.6)
+
+and where we have neglected the higher order relativistic
+effects which would bring only fractional corrections of order
+1(T5.
+We should emphasize at this point that we have included in
+our definition of the “ theoretical ” contribution to Pb the fact it
+corresponds to what could be observed in the center-of-mass
+system of the binary pulsar. In real life, all the observables of
+the system (Pfe, P&, cb, y,...) are measured in the solar barycenter system. This introduces Doppler factors between the
+various Earth-fitted parameters (like Pb, <b, y,...) and their
+“intrinsic” counterparts. However, as shown in detail by
+Damour and Deruelle (1986), these Doppler factors (when they
+are constant) amount to an overall unobservable change of
+units of length, time and mass which drops out when performing tests of gravitational theoreies [e.g., the test we
+are considering here, P£bs 1P^01 (observed parameters) + • • •). Another way to state this result is that the
+Doppler factors do not affect the dimensionless parameters,
+like eT, k, r¡, Pb, so that the relation (5.4) is still valid in a
+“boosted frame.” Note however that this result holds only for
+constant Doppler factors. On the other hand, a slowly changing Doppler factor, D = D0 + D0 i + • • -, is observable by its
+causing secular drifts in the various parameters with dimensions, like Pb. This is indeed the way we have obtained
+(iyp&)gal in § II above. [A more rigorous proof based on the
+formulas given in Damour and Deruelle 1986 modifies equation (2.2) only by some power of the Doppler factor, D0 =
+1 + Vr/c + 0(v2/c2), which introduces unknown but negligible
+fractional corrections ~t;R/c < 10“3 in the right-hand side of
+equation (2.2)].
+According to the most recent experimental data (Taylor and
+Weisberg 1990) the values of the observables entering equation
+(5.4) are Pb = 27906.9808968(16) s, eT = 0.6171312(8),
+œ = 4.226605(30)
+degrees
+per
+Julian
+year,
+and
+y = 4.303(6) x 10“3 s, where the numbers in parentheses indicate the observational uncertainties on the corresponding last
+digits. These values, together with the knowledge of the correlations between them (see Taylor and Weisberg 1989) give for
+the (lowest order) general relativistic prediction, eq. (5.4)
+ppR -2.402576(69) x 10“12 .
+(5.7)
+When generalizing equation (5.4) to more general theories
+(within the e-smooth class considered above), one must be
+careful to note that, at order O(e), there will be three distinct
+e-dependent new contributions : one from a direct modification
+of the basic “theoretical” Pb, and two, indirectly, from modifications of the œ and y theoretical links to mx and m2. These
+modifications have been worked out explicitly in two of the
+examples of theoretics quoted above (respectively by Will and
+Zaglauer 1989 and by Moffat 1989). Concerning the effect of a
+time variation of G, we shall adopt here the phenomenological
+approach of Shapiro et al (1971), taken up by Damour,
+Gibbons, and Taylor (1988), in which one assumes that the
+parameter G appearing in the usual orbital equations of
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 10 ===
+
+1991ApJ. . .366. .50ID
+
+510
+
+DAMOUR AND TAYLOR
+
+motion is replaced by G + Ót. This leads then only to a direct
+XP type term and adds to the right-hand side of equation (5.4)
+a simple term (Damour, Gibbons, and Taylor 1988):
+Ó
+(Phf = -2-Pb.
+
+(5.8)
+
+VI. SUMMARY AND DISCUSSION
+Table 1 presents a comprehensive summary of the observations, biases, theoretical values, and uncertainties that we are
+discussing. At the top of the table we quote the latest experimental value for the observed parameter (Ph/Pb)ohs (Taylor and
+Weisberg 1990):
+/p \obs
+lyj = -87.39 + 0.65 a ,
+(6.1)
+where a = 10~18 s-1 and the quoted uncertainty is intended to
+represent aid confidence interval including estimates of both
+random and systematic errors. In the following lines we list the
+various “additional” terms that contribute to the observed
+orbital period change and do not come purely from relativistic
+gravitational effects internal to the binary pulsar system. Their
+sum,
+
+Vol. 366
+
+bs
+
+on PJ . This result is in excellent accord with the “ theoretical ”
+prediction of Einstein’s theory, namely, from eq. (5.7):
+gr
+= -86.0923 ± 0.0025 a .
+(6.5)
+The ratio
+j^obs-add
+¿GR = 1.0081 + 0.0022(gal) ± 0.0076(obs),
+(6.6)
+Pb
+shows that the binary pulsar test of the existence and Einsteinian nature of gravitational radiation has now reached a precision better than 1%. If we recall that the values of the masses
+that have been used in the theoretical estimate, PpR, are the
+fully general relativistic “Schwarzschild masses” of the two
+gravitationally condensed objects (see e.g., Damour 1983a)
+which include large (~15%) self-gravity contributions, we see
+clearly that the 0.8% agreement (6.6) includes a verification of
+the strong gravitational-field effects of Einstein’s theory.
+Let us now consider this Pb test as a way to quantify, within
+the above considered class of theories continuously connected
+to general relativity, how far Nature allows us to deviate from
+Einstein’s theory. In particular, if we consider deviations due to
+a time varying gravitational constant, we get from equations
+(5.9), (6.5) and (6.6)
+G
+1 pjbs-add ppR-i
+=
+G ~2l Pb "Pj
+= [1.10 ± 0.30(gal) ± 1.03(obs)] x 10"11 yr"1 . (6.7)
+
+In fact, Table 1 shows that, at the present levels of precision,
+only the galactic effects (discussed in § II) contribute significantly to equation (6.3).
+Subtracting equation (6.3) from the experimental value leads
+to
+obs-add
+I
+= - 86.79 ± 0.19(gal) ± 0.65(obs) a , (6.4)
+where we are indicating separately the 1 cr uncertainties
+coming from galactic effects and from the obsevational errors
+TABLE l
+Contributions to the Observed {Pb/Pb)
+Contribution Uncertainty
+Quantity
+Observed
+Galactic acceleration, planar ...
+Galactic acceleration, vertical ..
+Stars near pulsar
+Stars near Sun
+GMCs near pulsar
+GMCs near Sun
+Ephemeris uncertanties
+Pulsar mass loss
+Companion mass loss
+Total correction
+Observed—correction
+Theoretical (General relativity)
+
+(a = 10-18s-1)
+-87.39
+-0.596
+-0.0061
++ 0.00004
+-0.003
++ 0.00066
++ 0.0037
+-0.60
+-86.79
+-86.0923
+
+±0.65
+±0.188
+±0.0018
+±0.00047
+±0.00001
+±0.0066
+±0.001
+±0.012
+±0.00020
+±0.0037
+±0.19
+±0.68
+±0.0025
+
+The addition in quadrature of galactic and observational
+uncertainties gives a limit on the time variation of G, G/G =
+(1.10 ± 1.07) x 10-11 yr-1, which is comparable to those
+obtained from radar ranging data in the solar system, notably
+from the Viking lander data (Hellings et al. 1983; Shapiro
+1990). As the binary pulsar timing data continue to be acquired
+in the future, the observational uncertainty will continue to
+decrease at least as fast as the inverse 2.5 power of the total
+time span, and we can anticipate that within 5-10 yr it will be
+the galactic effects that will limit the precision of determination
+of G/G at the level ±3 x 10-12 yr-1. The possibility to reach
+such a level is interesting, on the one hand because “ asteroid
+noise” presently limits the precision of the determination of
+G/G in the solar system at the level ±5 or 6 x 10“12 yr“1
+(Anderson et al. 1989), and on the other hand because a newly
+developed field theory model of the time variation of G
+(Damour, Gibbons, and Gundlach 1990) finds ±6.6 x 10“12
+yr “1 as the maximum possible value for ó/G.
+Alternatively, if we assume that the standard general relativistic framework (with G = 0) is valid we see that, in a few
+years, the measurement of Pgbs can be turned into a measurement of the galactic contribution (2.12) and thereby furnish a
+new handle on the galactic constants R0 and v0 (especially v0,
+which presently contributes the biggest uncertainty). Such a
+“ pulsar timing ” measurement of v0 would be free from many
+of the astrophysical uncertainties that have plagued its other
+determinations. Note that it is fortunate (and unexpected) that
+the 15% fractional uncertainty on the reduced distance (that
+will be hard to decrease) contributes almost negligibly to the
+uncertainty in (Pb/P&)8al (namely ± 0.037a before adding errors
+in quadrature).
+Finally, let us remark that when using the difference
+pobs _ padd _ pGR to est¿mate
+time variation of G, we are
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 11 ===
+
+1991ApJ. . .366. .50ID
+
+No. 2, 1991
+
+ORBITAL PERIOD CHANGE OF PSR 1913 + 16
+
+assuming that we can neglect the effect of ultra-low-frequency
+gravitational waves. Bertotti, Carr, and Rees (1983) have
+emphasized that such waves with periods intermediate
+between the span of the timing observations and the lighttravel time to the pulsar bring an extra contribution to P£bs. In
+the case of a stochastic background of gravitational radiation,
+the difference (6.7) corresponds to a ratio Qg of the integrated
+energy density within the frequency band 10“12 to 10“ 9 Hz to
+the cosmological closure density, of the order of 0.02 x (100
+km s"1 Mpc-1/^o)2- In the stated frequency range there are
+no other direct limits on the stochastic gravitational radiation
+background. However, there are much better limits on
+Qg(f) = fdQg/df both for lower frequencies (/~ 10“15 Hz,
+Rees 1983) and for higher frequencies (/ ~ 10“ 8 Hz, Stinebring
+et al. 1990). Many cosmological scenarios producing such stochastic gravitational radiation backgrounds endow them with
+a roughly scale-free spectrum, üg(/) ^ constant, so it can be
+argued that the bracketing strong limits on Qg(f) [most
+notably Q^(4 x 10-9 Hz) < 9 x 10-8, Stinebring et al 1990]
+ensure indirectly that such stochastic backgrounds contribute
+only negligibly to P£bs.
+Note added in manuscript 1990 June 19—Results from the
+
+511
+
+Ph.D. thesis of Frail (1989), used extensively in § II, have now
+been published in the open literature (Frail and Weisberg
+1990). We note that these authors do not accept the line of
+reasoning of Heiles et al. (1983) and Backer (1989), leading to a
+tight upper (as well as lower) limit on the distance to PSR
+1937 + 21. We also note that Frail and Weisberg appear to feel
+more secure than did Weisberg et al. (1987) in assigning
+“tangent point” upper distance limits for PSRs 1914+13 and
+1915 + 13. These considerations serve to emphasize that at
+some level of accuracy, it is an oversimplification to treat the
+interstellar electron density as constant, even over a small
+range of Galactic longitude. Our conclusions in § II and other
+sections of the paper are not affected, however. As we have
+mentioned following equation 2.27, uncertainties in the distance to PSR 1913 + 16 contribute negligibly to the overall
+error budget of the experiment.
+Part of this work has been supported by National Science
+Foundation grant AST88-17826 to Princeton University. We
+thank D. C. Backer, P. Dasgupta, D. Frail, J. Laskar, B. S.
+Sathyaprakash, and J. M. Weisberg for informative communications and other assistance.
+
+REFERENCES
+Abies, J. G., and Manchester, R. N. 1976, Astr. Ap., 50,177.
+Hellings, R. W., et al. 1983, Phys. Rev. Letters, 51,1609.
+Allen, C. W. 1973, Astrophysical Quantities (3rd ed., London: The Athlone
+Holtsmark, J. 1919, Ann. Physik, 58, 576.
+Press).
+Hülse, R. A., and Taylor, J. H. 1975, Ap. J. {Letters), 195, L51.
+Anderson, J. D., et al. 1989, Adv. Space. Res., 9, (9)71.
+Jeans, J. H. 1924, M.N.R.A.S., 84,2.
+Backer, D. C. 1989, private communication.
+. 1925, M.N.R.A.S., 85,912.
+Bahcall, J. N. 1984, Ap. J., 287,926.
+Jordan, P. 1959, Zs. Phys., 157,112.
+Backer, B. M, and O’Connell, R. R. 1975, Ap. J. {Letters), 199, L25.
+Kerr, F. J., and Lynden-Bell, D. 1986, M.N.R.A.S., 221,1023.
+Baroni, L., Callegari, G., Gualdi, C., and Fortini, P. 1979, Lettere al Nuovo
+Lyne, A. G., Manchester, R. N., and Taylor, J. H. 1985, M.N.R.A.S., 213,613.
+Cimento, 25,260.
+Manchester, R. N., and Taylor, J. H. 1981, A.J., 86,1953.
+. 1980, Lettere al Nuovo Cimento, 27, 509.
+Mihalas, D., and Binney, J. 1981, Galactic Astronomy (San Francisco:
+Bertotti, B., Carr, B. J., and Rees, M. J. 1983, M.N.R.A.S., 203,945.
+Freeman).
+Blanchet, L, and Schäfer, G. 1989, M.N.R.A.S., 239,845.
+Moffat, J. W. 1979, Phys. Rev. D19, 3554.
+Blandford, R. D., and Teukolsky, S. A. 1976, Ap. J., 205,580.
+. 1989, University of Toronto, preprint UTPT-89-30.
+Brans, C., and Dicke, R. H. 1961, Phys. Rev., 124,925.
+Peters, P. C., and Mathews, J. 1963, Phys. Rev., 131,435.
+Brumberg, V. A., Zel’dovich, Ya. B., Novikov, I. D., and Shakura, N. I. 1975,
+Racine, R., and Harris, W. E. 1989, A.J., 98,1609.
+Soviet Astr. Letters, 1,2.
+Rees, M. J. 1983, in Gravitational Radiation, ed. N. Deruelle and T. Piran
+Chandrasekhar, S. 1941, Ap. J., 94, 511.
+(Amsterdam. North-Holland) p. 297.
+Chandrasekhar, S., and von Neumann, J. 1942, Ap. J., 95,489.
+Reid, M. J. 1989, in IAU Symposium 136, The Center of the Galaxy, ed. M.
+Dame, T. M., et al. 1987, Ap. J., 322, 706.
+Morris (Dordrecht: Kluwer), p. 37.
+Damour, T. 1983«, in Gravitational Radiation, ed. N. Deruelle and T. Piran
+Schmidt, M. 1965, in Galactic Structure, ed. A. Blaauw and M. Schmidt
+(Amsterdam : North-Holland), p. 59.
+(Chicago: The University of Chicago Press), p. 513.
+. 19836, Phys. Rev. Letters, 51,1019.
+Scoville, N. Z., and Sanders, D. B. 1986, in The Galaxy and the Solar System,
+Damour, T., and Deruelle, N. 1985, Ann. Inst. H. Pioncaré {Phys. Théorique)
+ed. R. Smoluchowski, J. N. Bahcall, and M. S. Matthews (Tucson: The
+43,107.
+University of Arizona Press), p. 69.
+. 1986, Ann. Inst. H. Pioncaré {Phys. Théorique) 44,263.
+Shapiro, I. I. 1990, in General Relativity and Gravitation ed. N. Ashby,
+Damour, T., Gibbons, G. W., and Gundlach, C. 1990, Phys. Rev. Letters, 64,
+(Cambridge: Cambridge University Press), in press.
+123.
+Shapiro, I. L, Smith, W. B., Ash, M. B., Ingalls, R. P., and Pettengill, G. H.
+Damour, T., Gibbons, G. W., and Taylor, J. H. 1988, Phys. Rev. Letters., 61,
+1971, Phys. Rev. Letters, 26,27.
+1151.
+Shapiro, S. L., and Terzian, Y. 1976, Astr. Ap., 52,115.
+Damour, T., and Schäfer, G. 1988, Nuovo Cimento, B101,127.
+Smarr, L. L., and Blandford, R. 1976, Ap. J., 207, 574.
+Davidsen, A., et al. 1975, Ap. J. {Letters), 200, L19.
+Stinebring, D. R., Ryba, M. F., Taylor, J. H., and Romani, R. W. 1990, Phys.
+Eardley, D. M. 1975, Ap. J. {Letters), 196, L59.
+Rev. Letters, submitted.
+Einstein, A. 1925, Preuss. Akad. Wiss. Sitzungberichte, p. 414.
+Taylor, J. H., and Weisberg, J. M. 1982, Ap. J., 235,908.
+1954, The Meaning of Relativity, Appendix to the 5th edition
+. 1989, Ap. J., 345,434.
+(Princeton, NJ : Princeton University Press).
+. 1990, unpublished.
+Epstein, R. 1977, Ap. J., 216,92.
+Verweij, S. 1936, Pub. Astr. Inst. Amsterdam, 5, 3.
+. 1979, Ap. J., 231,644.
+Wagoner, R. V. 1975a, Ap. J. {Letters), 196, L63.
+Esposito, L. W., and Harrison, E. R. 1975, Ap. J. {Letters), 196, LI.
+. 19756, Gravity Essay (Gravity Research Foundation).
+Fairhead, L. 1989, Thèse de Doctorat de l’Observatoire de Paris.
+Weisberg, J. M., Rankin, J. M., and Boriakoff, V. 1987, Astr. Ap., 186,307.
+Fich, M, Blitz, L., and Stark, A. A. 1989, Ap. J., 342,272.
+Weisberg, J. M., and Taylor, J. H. 1981, Gen Rel. Grav., 13,1.
+Frail, D. 1989, Ph.D. thesis, University of Toronto.
+Will, C. M. 1976, Ap. J., 205, 861.
+Frail, D., and Weisberg, J. M. 1990, A.J., in press.
+. 1981, Theory and Experiment in Gravitational Physics (Cambridge:
+Georgelin, Y. M., and Georgelin, Y. P. 1976, Astr. Ap., 49, 57.
+Cambridge University Press).
+Haugan, M. P. 1985, Ap. J., 296,1.
+Will, C. M., and Eardley, D. M. 1977, Ap. J. {Letters), 212, L91.
+Heiles, C., Kulkarni, S. R., Stevens, M. A., Backer, D. C., Davis, M. M., and
+Will, C. M., and Zaglauer, H. W. 1989, Ap. J., 346, 366.
+Goss, W. M. 1983, Ap. J. {Letters), 273, L75.
+T. Damour: Institut des Hautes Etudes Scientifiques, 91440 Bures sur Yvette, France, and D.A.R.C., Observatoire de ParisC.N.R.S., 92195 Meudon Cedex, France
+J. H. Taylor: Physics Department, Princeton University, Princeton, NJ 08544
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System

binary_pulsar_post_keplerian_atnf__PBDOT/reference/hulse_1975.bib

@@ -0,0 +1,10 @@
+@ARTICLE{HulseTaylor1975,
+  author = {{Hulse}, R. A. and {Taylor}, J. H.},
+  title = {{Discovery of a pulsar in a binary system}},
+  journal = {ApJ},
+  year = 1975,
+  volume = {195},
+  pages = {L51--L53},
+  doi = {10.1086/181708},
+  note = {PSR B1913+16 discovery — first binary pulsar; Nobel Prize 1993.},
+}

binary_pulsar_post_keplerian_atnf__PBDOT/reference/hulse_1975.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67b7d475a4718141c2fb6211a64dab2cbce900ae6633f151dddadcdd9e1915ac
+size 290764

binary_pulsar_post_keplerian_atnf__PBDOT/reference/hulse_1975.txt

@@ -0,0 +1,325 @@
+
+=== PAGE 1 ===
+
+1975ApJ...195L..51H
+
+The Astrophysical Journal, 195 :L51-L53, 1975 January 15
+© 1975. The American Astronomical Society. All rights reserved. Printed in U.S.A.
+
+DISCOVERY OF A PULSAR IN A BINARY SYSTEM
+R. A. Hulse and J. H. Taylor
+Department of Physics and Astronomy, University of Massachusetts, Amherst
+Received 1974 October 18
+ABSTRACT
+We have detected a pulsar with a pulsation period that varies systematically between 0S058967 and 0S059045 over
+a cycle of 0d3230. Approximately 200 independent observations over 5-minute intervals have yielded a well-sampled
+velocity curve which implies a binary orbit with projected semimajor axis U] sin i = 1.0 Rq, eccentricity e = 0.615, and
+mass function/(w) = 0.13 M0. No eclipses are observed. We infer that the unseen companion is a compact object with
+mass comparable to that of the pulsar. In addition to the obvious potential for determining the masses of the pulsar and
+its companion, this discovery makes feasible a number of studies involving the physics of compact objects, the astrophysics of close binary systems, and special- and general-relativistic effects.
+Subject headings: binaries — black holes — neutron stars — pulsars — relativity
+I. INTRODUCTION
+We wish to report the detection of an unusual pulsar
+discovered during the course of a systematic survey for
+new pulsars being carried out (Hulse and Taylor 1974)
+at the Arecibo Observatory in Puerto Rico. The object
+has a pulsation period of about 59 ms—shorter than
+that of any other known pulsar except the one in the
+Crab Nebula—and periodic changes in the observed
+pulsation rate indicate that the pulsar is a member of a
+binary system with an eccentric orbit of 0d3230 period.
+Thus for the first time it is possible to observe the
+gravitational interactions of a pulsar and another massive object, and additional observations should make it
+possible to determine the masses of the two objects unambiguously.
+II. DISCOVERY OF THE BINARY PULSAR
+The equipment and searching method used in the
+pulsar survey have been described previously (Hulse
+and Taylor 1974). Forty pulsars have now been detected in this work, of which 32 were not previously
+known; the parameters of the 21 most recently discovered will be given in another paper (Hulse and
+Taylor 1975). The 59-ms pulsar, PSR 1913 + 16, was
+first detected in 1974 July. Attempts to measure its
+period to an accuracy of + 1 /¿s were frustrated by apparent changes in period of up to ^80 ¿¿s from day to
+day, and sometimes by as much as 8 ^s over 5 minutes.
+Such behavior is quite uncharacteristic of other pulsars:
+the largest known secular changes of period are of order
+10 ¿¿s per year, and irregular changes of period are many
+orders of magnitude smaller (Manchester and Taylor
+1974). It soon became clear that Doppler shifts resulting from orbital motion of the pulsar could account
+for the observed period changes, and by the end of
+September an accurate velocity curve of this “singleline spectroscopic binary’’ had been obtained (see
+figure 1).
+The parameters of the pulsar are given in table 1. In
+the table, celestial and galactic coordinates are followed
+by Pcm, the “center of mass” pulsar period (corrected
+for the orbital motion of the pulsar and for the motion
+
+of the observer in the solar system) ; an upper limit for
+dPcm/dt, the first derivative of the period; DAI, the
+dispersion measure; J+o, the average flux density at
+430 AIHz; and an upper limit to We, the effective pulse
+width. (The pulses observed at 430 MHz are probably
+significantly broadened by multipath scattering in the
+interstellar medium.)
+The elements of the binary orbit are given in table 2.
+Yi is the semiamplitude of radial velocity variation of
+the pulsar with respect to the center of mass of the
+system; Pb is the period of the binary orbit, corrected
+for the motion of the observatory; e is the eccentricity
+of the orbit; co is the longitude of periastron; T is the
+time of periastron passage; ai sin i is the projected
+semimajor axis of the pulsar orbit, i being the inclination between the orbit and the plane of the sky; and
+f{m) = (Mo sin iy/(Mi + M2)2 is the mass function.
+These quantities were evaluated from the velocity
+measurements shown as filled circles in figure 1. The
+TABLE 1
+Parameters of the Binary Pulsar
++1950.0) = 19h13om13s ± 4s
+0(1950.0) = +16 00'24" ± 60"
+l = 49?9
+b = 2?1
+Pcm = 0S059030 ± 0*000001
+dPcJdt < 1 x 10-12 -3
+DM = 167 ± 5 cm pc
+S430 = 0.006 ± 0.003 Jy
+We < 10 ms
+
+TABLE 2
+Elements of the Orbit
+Ki =199 + 5 km s"1
+Pb = 27908 + 7 s
+e = 0.615 ± 0.010
+co = 179° + Io
+T = JD 2,442,321.433 ± 0.002
+ai sin i = 1.00 + 0.02 R0
+f(m) = 0.13 ± 0.01 M0
+
+LSI
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 2 ===
+
+1975ApJ...195L..51H
+
+HÜLSE AND TAYLOR
+
+L52
+
+Voi. 195
+
+PHASE
+Fig. 1.—Velocity curve for the binary pulsar. Points represent measurements of the pulsar period distributed over parts of 10 different
+orbital periods. The curve corresponds to equations (l)-(4), with parameters from table 2.
+TABLE 3
+Possible Parameters of Binary Pulsar System
+
+velocity curve also shown in the figure was then computed from the elements using the equations (Aitken
+1964)
+Vri = Ah[cos (0 + co) + 6 COS <x>] ,
+tan
+
+1/2
+
+= [(1 + e)/(l “ <0]
+
+tan
+
+(1)
+,
+
+(2)
+
+M = E — e sm E ,
+
+(3)
+
+0 = M/27T = (^ - T)/Ph,
+
+(4)
+
+where Vrl is the radial velocity of the pulsar (the “visible” member of the binary pair); M, E, and 6 are respectively the mean, eccentric, and true anomaly of the
+orbit of the pulsar about the center of mass; 0 is the
+orbital phase; and t is the time.
+The orbital elements given in table 2 were obtained
+from direct measurements of the pulsar period over
+about 200 different 5-minute intervals distributed over
+10 days. The 5-minute intervals are long enough that
+the period can be measured to an accuracy of about 1
+¿¿s, but short enough that the period does not change too
+drastically within the interval.
+III. PHYSICAL PARAMETERS OE THE BINARY PAIR
+The mass of the pulsar is, of course, a quantity of
+great interest, as is the size and mass of the unseen
+companion. The observed mass function permits a wide
+range of values for Mi and M2. However, if we restrict
+attention to values of Mi thought to be reasonable for
+neutron stars, the picture becomes clearer. Table 3 gives
+the required values for Fi, the maximum velocity of the
+pulsar, and M2, the mass of the companion, for assumed
+inclinations i = 90°, 60°, 30°, 20°, and 10°, and pulsar
+masses Mi = 0.3, 1.0, and 1.5 M0. Evidently the mass
+ratio Mi/M2 cannot be very different from unity unless
+
+i
+(degrees)
+
+Fi
+(max)
+
+Mi = 0.3
+
+Mi = 1.0
+
+Mi = 1.5
+
+M2
+
+R2
+
+m2
+
+m2
+
+R2
+
+90
+60
+30
+20
+10
+
+0.001k
+0.0012c
+0.002k
+0.003k
+0.006k
+
+0.4
+0.5
+1.5
+3.8
+26
+
+0
+0.7
+<0.6 0.9
+<1.3 2.2
+<1.9 4.8
+<3.5 27
+
+0
+0.9
+<0.8 1.1
+<1.6 2.6
+<2.1 5.4
+<3.7 28
+
+0
+<0.8
+<1.8
+<2.3
+<3.7
+
+R2
+
+the inclination i is rather small, which seems unlikely
+in view of the large observed radial velocity (^-TCTk).
+Furthermore, the orbit is such that if the inclination
+were close to 90° and the size of the companion were
+large enough, eclipses of the pulsar would occur at
+orbital phase 0 = 0.93. No eclipses are observed,
+which requires the radius of the companion to be less
+than
+^2,max = (öl + Ö2)(l - e2) sin f/tan i
+= Rq(1 + M1/M2)(l - e2)/tan i ,
+
+(5)
+
+where a2 is the semimajor axis of the orbit of the companion about the center of mass and Mi and M2 are the
+masses of the two objects. Comparison of these upper
+limits for R2 with the corresponding values of M2,
+together with the known dependence of radius on mass
+for main-sequence stars (Allen 1973), virtually rules out
+the possibility that the companion is a main-sequence
+star. We conclude that the companion must be a compact object, probably a neutron star or a black hole. A
+white dwarf companion cannot be ruled out, but seems
+unlikely for evolutionary reasons.
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System
+
+=== PAGE 3 ===
+
+1975ApJ...195L..51H
+
+No. 2, 1975
+
+DISCOVERY OF PULSAR IN BINARY SYSTEM
+
+IV. ADDITIONAL OBSERVATIONS
+We cannot at present rule out the possibility that the
+unseen companion is also a radiofrequency pulsar. If
+pulsations from the companion can be found, the system
+will be in effect a ‘‘double-line” spectroscopic binary
+and the mass ratio of the two bodies will be directly
+measurable. This is an exciting possibility, because
+then only the inclination would have to be determined
+in order to solve for the two masses.
+Timing data much more accurate than that already
+available can in principle be obtained by recording the
+absolute time of arrival of the pulses. Observations of
+this sort done on other pulsars yield absolute arrival
+times accurate to ~10-4 s. Measurements of comparable quality are now being acquired for PSR 1913 +
+16, and in due course the data will yield greatly improved accuracies for the celestial coordinates and for
+the orbital elements of the binary system. This in turn
+will allow a number of interesting gravitational and
+relativistic phenomena to be studied. The binary configuration provides a nearly ideal relativity laboratory
+including an accurate clock in a high-speed, eccentric
+orbit and a strong gravitational field. We note, for
+example, that the changes of both i+T2 and GM/âr
+during the orbit are sufficient to cause changes in observed period of several parts in 106. Therefore, both the
+relativistic Doppler shift and the gravitational redshift
+will be easily measurable. Furthermore, the generalrelativistic advance of periastron should amount to
+about 4° per year, which will be detectable in a short
+
+L53
+
+time. The measurements of these effects, not usually
+observable in spectroscopic binaries, would allow the
+orbit inclination and the individual masses to be obtained.
+The star field in the direction of the pulsar is crowded,
+and the observed dispersion measure suggests that PSR
+1913 + 16 is about 5 kpc distant. Probably there are
+some 5 to 10 mag of optical absorption along the line of
+sight, so we should expect the apparent visual magnitude of the pulsar (and its companion) to be some 18 to
+23 mag fainter than the absolute magnitudes. Thus, the
+prospects for optical observations do not seem good
+unless a large fraction of the observed dispersion is the
+result of ionized material close to the pulsar. No changes
+in dispersion measure exceeding ± 20 cm-3 pc have been
+observed over the binary period, so it is clear that at
+most a small fraction of the dispersion can arise from
+electrons within the binary orbit.
+We thank the staff of the Five College Radio Astronomy Observatory for assistance in construction of
+the pulsar search apparatus, and the staff of the
+Arecibo Observatory for assistance with the observations. This work is supported by the National Science
+Foundation under grants GP-37917 and GP-32414X.
+The Arecibo Observatory is part of the National
+Astronomy and Ionosphere Center and is operated by
+Cornell University under contract with the National
+Science Foundation. This paper is contribution number
+196 of the Five College Observatories.
+
+REFERENCES
+Aitken, R. G. 1964, The Binary Stars (New York: Dover), pp. 79,
+Hülse, R. A., and Taylor, J. H. 1974, Ap. J. (Letters), 191, L59.
+158
+. 1975, in preparation.
+’
+Allen, C. W. 1973, Astrophysical Quantities '(3d ed.; London:
+Manchester, R. N., and Taylor, J. H. 1974, Ap. J. (Letters), 191,
+Athlone Press), p. 209.
+L63.
+
+R. A. Hulse and J. H. Taylor: Department of Physics and Astronomv, University of Massachusetts, Amherst,
+MA 01002
+
+© American Astronomical Society • Provided by the NASA Astrophysics Data System

binary_pulsar_post_keplerian_atnf__PBDOT/reference/kramer_2006.bib

@@ -0,0 +1,10 @@
+@ARTICLE{Kramer2006,
+  author = {{Kramer}, M. and {Stairs}, I. H. and {Manchester}, R. N. and {McLaughlin}, M. A. and {Lyne}, A. G. and {Ferdman}, R. D. and {Burgay}, M. and {Lorimer}, D. R. and {Possenti}, A. and {D'Amico}, N. and {Sarkissian}, J. M. and {Hobbs}, G. B. and {Reynolds}, J. E. and {Freire}, P. C. C. and {Camilo}, F.},
+  title = {{Tests of General Relativity from Timing the Double Pulsar}},
+  journal = {Science},
+  year = 2006,
+  volume = {314},
+  pages = {97--102},
+  doi = {10.1126/science.1132305},
+  note = {Double pulsar PSR J0737-3039A/B: 5 PK parameters measured; PBDOT consistent with GR at the 1.4e-3 level.},
+}

binary_pulsar_post_keplerian_atnf__PBDOT/reference/kramer_2006.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47caf499fe46675815edca0aca1e758ac8a2ef5623de000ef9c8add0c23e54c3
+size 561369

binary_pulsar_post_keplerian_atnf__PBDOT/reference/kramer_2006.txt

@@ -0,0 +1,1191 @@
+
+=== PAGE 1 ===
+
+arXiv:astro-ph/0609417v1 14 Sep 2006
+
+Tests of general relativity from timing the double
+pulsar
+M. Kramer,1∗ I.H. Stairs,2 R.N. Manchester,3 M.A. McLaughlin,1,4
+A.G. Lyne,1 R.D. Ferdman,2 M. Burgay,5 D.R. Lorimer,1,4
+A. Possenti,5 N. D’Amico,5,6 J.M. Sarkissian,3 G.B. Hobbs,3
+J.E. Reynolds,3 P.C.C. Freire7 and F. Camilo8
+1
+2
+
+University of Manchester, Jodrell Bank Observatory, Macclesfield, SK11 9DL, UK
+
+Dept. of Physics and Astronomy, University of British Columbia, 6224 Agricultural Road,
+Vancouver, BC V6T 1Z1, Canada
+
+3
+
+Australia Telescope National Facility, CSIRO, P.O. Box 76, Epping NSW 1710, Australia
+4
+Department of Physics, West Virginia University, Morgantown, WV 26505, USA
+5
+INAF - Osservatorio Astronomica di Cagliari, Loc. Poggio dei Pini, Strada 54,
+
+09012 Capoterra, Italy
+6
+Universita’ degli Studi di Cagliari, Dipartimento di Fisica, SP Monserrato-Sestu km 0.7,
+09042 Monserrato (CA), Italy
+7
+NAIC, Arecibo Observatory, HC03 Box 53995, PR 00612, USA
+8
+
+Columbia Astrophysics Laboratory, Columbia University, 550 West 120th Street,
+New York, NY 10027, USA
+
+∗
+
+To whom correspondence should be addressed; E-mail: mkramer@jb.man.ac.uk
+
+The double pulsar system, PSR J0737-3039A/B, is unique in that both neutron
+stars are detectable as radio pulsars. This, combined with significantly higher
+mean orbital velocities and accelerations when compared to other binary pulsars, suggested that the system would become the best available testbed for
+general relativity and alternative theories of gravity in the strong-field regime.
+1
+
+=== PAGE 2 ===
+
+Here we report on precision timing observations taken over the 2.5 years since
+its discovery and present four independent strong-field tests of general relativity. Use of the theory-independent mass ratio of the two stars makes these
+tests uniquely different from earlier studies. By measuring relativistic corrections to the Keplerian discription of the orbital motion, we find that the “postKeplerian” parameter s agrees with the value predicted by Einstein’s theory
+of general relativity within an uncertainty of 0.05%, the most precise test yet
+obtained. We also show that the transverse velocity of the system’s center of
+mass is extremely small. Combined with the system’s location near the Sun,
+this result suggests that future tests of gravitational theories with the double
+pulsar will supersede the best current Solar-system tests. It also implies that
+the second-born pulsar may have formed differently to the usually assumed
+core-collapse of a helium star.
+
+Introduction.
+
+Einstein’s general theory of relativity (GR) has so far passed all experimental
+
+tests with flying colours (1), with the most precise tests achieved in the weak-field gravity
+conditions of the Solar System (2, 3). However, it is conceivable that GR breaks down under
+extreme conditions such as strong gravitational fields where other theories of gravity may apply
+(4). Predictions of gravitational radiation and self-gravitational effects can only be tested using
+massive and compact astronomical objects such as neutron stars and black holes. Studies of
+the double-neutron-star binary systems, PSR B1913+16 and PSR B1534+12, have provided the
+best such tests so far, confirming GR at the 0.2% and 0.7% level, respectively (5, 6) 1 . The
+recently discovered double pulsar system, PSR J0737-3039A/B, has significantly higher mean
+1
+
+Stairs et al. (2002, ref. (6)) find an agreement of their measured values for PSR B1534+12 with GR at the
+0.05% level, but the measurement uncertainty on the most precisely measured parameter in the test, s, is only
+0.7%.
+
+2
+
+=== PAGE 3 ===
+
+orbital velocities and accelerations than either PSR B1913+16 or PSR B1534+12 and is unique
+in that both neutron stars are detectable as radio pulsars (7, 8).
+PSR J0737−3037A/B consists of a 22-ms period pulsar, PSR J0737−3039A (henceforth
+called A), in a 2.4-hr orbit with a younger 2.7-s period pulsar, PSR J0737−3039B (B). Soon
+after the discovery of A (7), it was recognised that the orbit’s orientation, measured as the
+longitude of periastron ω, was changing in tine with a very large rate of ω̇ = dω/dt ∼ 17◦
+yr−1 , which is four times the corresponding value for the Hulse-Taylor binary, PSR B1913+16
+(5). This immediately suggested that the system consists of two neutron stars, a conclusion
+confirmed by the discovery of pulsations from B (8). The pulsed radio emission from B has
+a strong orbital modulation, both in intensity and in pulse shape. It appears as a strong radio
+source only for two intervals, each of about 10-min duration, while its pulsed emission is rather
+weak or even undetectable for most of the remainder of the orbit (8, 9).
+In double-neutron-star systems, especially those having short orbital periods, observed pulse
+arrival times are significantly modified by relativistic effects which can be modelled in a theoryindependent way using the so-called “Post-Keplerian” (PK) parameters (10). These PK parameters are phenomenological corrections and additions to the simple Keplerian description of the
+binary motion, describing for instance a temporal change in period or orientation of the orbit, or
+an additional “Shapiro-delay” that occurs due to the curvature of space-time when pulses pass
+near the massive companion. The PK parameters take different forms in different theories of
+gravity and so their measurement can be used to test these theories (11,1). For point masses with
+negligible spin contributions, GR predicts values for the PK parameters which depend only on
+the two a priori unknown neutron-star masses and the precisely measurable Keplerian parameters. Therefore measurement of three (or more) PK parameters provides one (or more) tests of
+the predictive power of GR. For the double pulsar we can also measure the mass ratio of the
+two stars, R ≡ mA /mB = xB /xA . The ability to measure this quantity provides an important
+3
+
+=== PAGE 4 ===
+
+constraint because in GR and other theories this simple relationship between the masses and
+semi-major axes is valid to at least first post-Newtonian (1PN) or (v/c)2 order (12, 11).
+
+Observations. Timing observations of PSR J0737−3039A/B have been undertaken using the
+64-m Parkes radio telescope in New South Wales, Australia, the 76-m Lovell radio telescope
+at Jodrell Bank Observatory (JBO), UK, and the 100-m Green Bank Telescope (GBT) in West
+Virginia, USA, between 2003 April and 2006 January.
+At Parkes, observations were carried out in bands centred at 680 MHz, 1374 MHz and
+3030 MHz. While timing observations were frequent after the discovery of the system, later
+observations at Parkes were typically conducted every 3-4 weeks, usually covering two full
+orbits per session. Observations at the GBT were conducted at monthly intervals, with each
+session consisting of a 5- to 8-hour track (i.e., 2 to 3 orbits of the double pulsar). Typically, the
+observing frequencies were 820 and 1400 MHz for alternate sessions. Occasionally, we also
+performed observations at 340 MHz, in conjunction with pulse profile studies to be reported
+elsewhere. In addition, we conducted concentrated campaigns of five 8-hour observing sessions,
+all at 820 MHz, in 2005 May and 2005 November. Observations at JBO employed the 76-m
+Lovell telescope. Most data were recorded at 1396 MHz, while some observing sessions were
+carried out at the lower frequency of 610 MHz. The timing data obtained at Jodrell Bank
+represent the most densely sampled dataset but, because of the limited bandwidth, requiring
+longer integration times per timing point. The Parkes dataset is the longest one available and
+hence provides an excellent basis for investigation of secular timing terms.
+The time-series data of all systems were folded modulo the predicted topocentric pulse
+period. The adopted integration times were 30 s for pulsar A (180 s for JBO data) and 300 s for
+pulsar B. For A, these integration times reflect a compromise between producing pulse profiles
+with adequate signal-to-noise ratio and sufficient sampling of the orbit to detect and resolve
+
+4
+
+=== PAGE 5 ===
+
+phenomena that depend on orbital phase, such as the Shapiro delay. The integration time for B
+corresponds to about 108 pulse periods and is a compromise between the need to form a stable
+pulse profile while resolving the systematic changes seen as a function of orbital phase.
+
+Timing measurements.
+
+For each of the final profiles, pulse times-of-arrival (TOAs) were
+
+computed by correlating the observed pulse profiles with synthetic noise-free templates (see
+Fig. 1 in (13), cf. ref. (7)). A total of 131,416 pulse TOAs were measured for A while 507
+TOAs were obtained for B. For A, the same template was used for all observations in a given
+frequency band, but different templates were used for widely separated bands. We note that our
+observations still provide no good evidence for secular evolution of A’s profile (15) despite the
+predictions of geodetic precession. The best timing precision was obtained at 820 MHz with
+GASP backend (see ref. (13) for details of this and other observing systems) on the GBT, with
+typical TOA measurement uncertainties for pulsar A of 18 µs for a 30-s integration.
+For B, because of the orbital and secular dependence of its pulse profile (9), different templates were also used for different orbital phases and different epochs. A matrix of B templates
+was constructed, dividing the data set into 3-month intervals in epoch and 5-minute intervals in
+orbital phase. The results for the 29 orbital phase bins were studied, and it was noticed that,
+while the profile changes dramatically and quickly during the two prominent bright phases, the
+profile shape is simpler and more stable at orbital phases when the pulsar is weak. This apparent stability at some orbital phases cannot be attributed to a low signal-to-noise ratio as secular
+variations in the pulse shape are still evident. Consequently, the orbital phase was divided into
+five groups of different lengths to which the same template (for a given 3-month interval) was
+applied as shown in Fig. 2 of (13). In the final timing analysis, data from the two groups representing the bright phases (IV & V in Fig. 2 of (13)) were excluded to minimize the systematic
+errors caused by the orbital profile changes. Also, because of signal-to-noise and radio inter-
+
+5
+
+=== PAGE 6 ===
+
+ference considerations, only data from Parkes and the GBT BCPM backend were used in the B
+timing analysis.
+All TOAs were transferred to Universal Coordinated Time (UTC) using the Global Positional System (GPS) to measure offsets of station clocks from national standards and Circular T of the BIPM to give offsets from UTC, and then to the nominally uniform Terrestrial
+Time TT(BIPM) timescale. These final TOAs were analysed using the standard software package TEMPO (16), fitting parameters according to the relativistic and theory independent timing
+model of Damour & Deruelle (17, 10). In addition to the DD model, we also applied the “DDShapiro” (DDS) model introduced by Kramer et al. (ref. (18)). The DDS model is a modification
+of the DD model designed for highly inclined orbits. Rather than fitting for the Shapiro parameter s, the model uses the parameter zs ≡ − ln(1 − s) which gives a more reliable determination
+of the uncertainties in zs and hence in s. We quote the final result for the more commonly used
+parameter s and note that its value computed from zs is in good agreement with the value obtained from a direct fit for s within the DD model. Derived pulsar and binary system parameters
+are listed in Table 1.
+In the timing analysis for pulsar B, we used an unweighted fit to avoid biasing the fit toward
+bright orbital phases. Uncertainties in the timing parameters were estimated using Monte Carlo
+simulations of fake data sets for a range of TOA uncertainties, ranging from the minimum estimated TOA error to its maximum observed value of about 4 ms. For B, we also fitted for offsets
+between datasets derived from different templates in the fit since the observed profile changes
+prevent the establishment of a reliable phase relationship between the derived templates. This
+precludes a coherent fit across the whole orbit and hence limits the final timing precision for B.
+It cannot yet be excluded that different parts of B’s magnetosphere are active and responsible
+for the observed emission at different orbital phases.
+In the final fit, we adopted the astrometric parameters and the dispersion measure derived
+6
+
+=== PAGE 7 ===
+
+for A and held these fixed during the fit, since A’s shorter period and more stable profile give
+much better timing precision than is achievable for B. Except for the semi-major axis which
+is only observable as the projection onto the plane-of-the-sky xB = (aB /c) sin i, where i is the
+orbital inclination angle, we also adopted A’s Keplerian parameters (with 180◦ added to ωA ) and
+kept these fixed. We also adopted the PK parameter ω̇ from the A fit since logically this must
+be identical for the two pulsars; this equality therefore does not implicitly make assumptions
+about the validity of any particular theory of gravity (see next section). The same applies for
+Ṗb . In contrast, the PK parameters γ, s and r are asymmetric in the masses and their values
+and interpretations differ for A and B. In practical terms, the relatively low timing precision
+for B does not require the inclusion of γ, s, r or Ṗb in the timing model. We can however
+independently measure ω̇B , obtaining a value of 16.96 ± 0.05 deg yr−1 , consistent with the
+more accurately determined value for A.
+Since the overall precision of our tests of GR is currently limited by our ability to measure
+xB and hence the mass ratio R ≡ mA /mB = xB /xA (see below), we adopted the following
+strategy to obtain the best possible accuracy for this parameter. We used the whole TOA data
+set for B in order to measure B’s spin parameters P and Ṗ , given in Table 1. These parameters
+were then kept fixed for a separate analysis of the concentrated 5-day GBT observing sessions
+at 820 MHz. On the timescale of the long-term profile evolution of B, each 5-day session
+represents a single-epoch experiment and hence requires only a single set of profile templates.
+The value of xB obtained from a fit of this parameter only to the two 5-day sessions is presented
+in Table 1.
+Because of the possible presence of unmodelled intrinsic pulsar timing noise and because
+not all TOA uncertainties are well understood, we adopt the common and conservative pulsartiming practice of reporting twice the parameter uncertainties given by TEMPO as estimates of
+the 1-σ uncertainties. While we believe that our real measurement uncertainties are actually
+7
+
+=== PAGE 8 ===
+
+somewhat smaller than quoted, this practice facilitates the comparison with previous tests of
+GR using pulsars. The timing model also includes timing offsets between the datasets for the
+different instruments represented by the entries in Table 1 in (13). The final weighted rms
+post-fit residual is 54.2µs. In addition to the spin and astrometric parameters, the Keplerian
+parameters of A’s orbit and five PK parameters, we also quote a tentative detection of a timing
+annual parallax which is consistent with the dispersion-derived distance. Further details are
+given in ref. (13).
+
+Tests of general relativity. Previous observations of PSR J0737−3039A/B (7, 8) resulted in
+the measurement of R and four PK parameters: the rate of periastron advance ω̇, the gravitational redshift and time dilation parameter γ, and the Shapiro-delay parameters r and s.
+Compared to these earlier results, the measurement precision for these parameters from PSR
+J0737−3039A/B has increased by up to two orders of magnitude. Also, we have now measured the orbital decay, Ṗb . Its value, measured at the 1.4% level after only 2.5 years of timing,
+corresponds to a shrinkage of the pulsars’ separation at a rate of 7mm per day. Therefore, we
+have measured five PK parameters for the system in total. Together with the mass ratio R, we
+have six different relationships that connect the two unknown masses for A and B with the observations. Solving for the two masses using R and a one PK parameter, we can then use each
+further PK parameter to compare its observed value with that predicted by GR for the given
+two masses, providing four independent tests of GR. Equivalently, one can display these tests
+elegantly in a “mass-mass” diagram (Fig. 1). Measurement of the PK parameters gives curves
+on this diagram that are in general different for different theories of gravity but which should
+intersect in a single point, i.e., at a pair of mass values, if the theory is valid (11).
+As shown in Fig. 1, we find that all measured constraints are consistent with GR. The
+most precisely measured PK parameter currently available is the precession of the longitude
+
+8
+
+=== PAGE 9 ===
+
+of periastron, ω̇. We can combine this with the theory-independent mass ratio R to derive
+the masses given by the intersection region of their curves: mA = 1.3381 ± 0.0007 M⊙
+and mB = 1.2489 ± 0.0007 M⊙ .2 Table 2 lists the resulting four independent tests that are
+currently available. All of them rely on comparison of our measured values of s, r, γ and
+Ṗb with predicted values based on the masses defined by the intersection of the allowed regions for ω̇ and R in the mA –mB plane. The calculation of the predicted values is somewhat complicated by the fact that the orbit is nearly edge-on to the line of sight, so that the
+formal intersection region actually includes parts of the plane disallowed by the Keplerian
+mass functions of both pulsars (see Fig. 1). To derive legitimate predictions for the various
+parameters, we used the following Monte Carlo method. A pair of trial values for ω̇ and
+xB (and hence R and the B mass function) is selected from gaussian distributions based on
+the measured central values and uncertainties. (The uncertainty on xA is very small and is
+neglected in this procedure.) This pair of trial values is used to derive trial masses mA and
+Pb −5/3
+)
+(T⊙ M)2/3 (1 − e2 )−1 , where M = mA + mB and
+mB , using the GR equation ω̇ = 3( 2π
+
+T⊙ ≡ GM⊙ /c3 = 4.925490947 µs, and the mass-ratio equation mA /mB = xB /xA . If this trial
+mass pair falls in either of the two disallowed regions (based on the trial mass function for B)
+it is discarded. This procedure allows for the substantial uncertainty in the B mass function.
+Allowed mass pairs are then used to compute the other PK parameters, assuming GR. This procedure is repeated until large numbers of successful trials have accumulated. Histograms of the
+PK predictions are used to compute the expectation value and 68% confidence ranges for each
+of the parameters. These are the values given in Table 2.
+The Shapiro delay shape illustrated in Fig. 2 gives the most precise test, with sobs /spred =
+2
+
+The true masses will deviate from these values by an unknown, but essentially constant, Doppler factor,
+probably of order 10−3 or less (10). Moreover, what is measured is a product containing Newton’s gravitational
+constant G. The relative uncertainty of G of 1.5×10−4 limits our knowledge of any astronomical mass in kilograms
+but since the product T⊙ = GM⊙ /c3 = 4.925490947µs is known to very high precision, masses can be measured
+precisely in solar units.
+
+9
+
+=== PAGE 10 ===
+
+0.99987 ± 0.00050.3 This is by far the best available test of GR in the strong-field limit, having
+a higher precision than the test based on the observed orbit decay in the PSR B1913+16 system
+with a 30-year data span (19). As for the PSR B1534+12 system (6), the PSR J0737−3039A/B
+Shapiro-delay test is complementary to that of B1913+16 since it is not based on predictions
+relating to emission of gravitational radiation from the system (20). Most importantly, the four
+tests of GR presented here are qualitatively different from all previous tests because they include
+one constraint (R) that is independent of the assumed theory of gravity at the 1PN order. As a
+result, for any theory of gravity, the intersection point is expected to lie on the mass ratio line in
+Fig. 1. GR also passes this additional constraint.
+In estimating the final uncertainty of xB and hence of R, we have considered that geodetic
+precession will lead to changes to the system geometry and hence changes to the aberration of
+the rotating pulsar beam. The effects of aberration on pulsar timing are usually not separately
+measurable but are absorbed into a redefinition of the Keplerian parameters. As a result, the
+int
+observed projected sizes of the semi-major axes, xobs
+A,B , differ from the intrinsic sizes, xA,B by
+
+a factor (1 + ǫA
+A,B ). The quantity ǫA depends for each pulsar A and B on the orbital period,
+the spin frequency, the orientation of the pulsar spin and the system geometry (11). While
+aberration should eventually become detectable in the timing, allowing the determination of a
+further PK parameter, at present it leads to an undetermined deviation of xobs from xint , where
+the latter is the relevant quantity for the mass ratio. The parameter ǫA
+A,B scales with pulse period
+and is therefore expected to be two orders of magnitude smaller for A than for B. However,
+because of the high precision of the A timing parameters, the derived value xobs
+A may already
+be significantly affected by aberration. This has (as yet) no consequences for the mass ratio
+obs
+obs
+R = xobs
+B /xA , as the uncertainty in R is dominated by the much less precise xB . We can
+
+explore the likely aberration corrections to xobs
+B for various possible geometries. Using a range
+3
+
+Note, s has the same relative uncertainty as our determination of the masses.
+
+10
+
+=== PAGE 11 ===
+
+of values given by studies of the double pulsar’s emission properties (21), we estimate ǫA
+A ∼
+−4
+10−6 and ǫA
+B ∼ 10 . The contribution of aberration therefore is at least one order of magnitude
+
+smaller than our current timing precision. In the future this effect may become important,
+possibly limiting the usefulness of R for tests of GR. If the geometry cannot be independently
+determined, we could use the observed deviations of R from the value expected within GR to
+determine ǫA
+B and hence the geometry of B.
+Space motion and inclination of the orbit.
+
+Because the measured uncertainty in Ṗb de-
+
+creases approximately as T −2.5 , where T is the data span, we expect to improve our test of
+the radiative aspect of the system to the 0.1% level or better in about five years’ time. For the
+PSR B1913+16 and PSR B1534+12 systems, the precision of the GR test based on the orbitdecay rate is severely limited both by the uncertainty in the differential acceleration of the Sun
+and the binary system in the Galactic gravitational potential and the uncertainty in pulsar distance (22, 6). For PSR J0737−3039A/B, both of these corrections are very much smaller than
+for these other systems. Based on the measured dispersion measure and a model for the Galactic
+electron distribution (23), PSR J0737−3039A/B is estimated to be about 500 pc from the Earth.
+From the timing data we have measured a marginally significant value for the annual parallax,
+3 ± 2 mas, corresponding to a distance of 200 − 1000 pc (Table 1), which is consistent with the
+dispersion-based distance that was also used for studies of detection rates in gravitational wave
+detectors (7). The observed proper motion of the system (Table 1) and differential acceleration
+in the Galactic potential (24) then imply a kinematic correction to Ṗb at the 0.02% level or less.
+Independent distance estimates also can be expected from measurements of the annual parallax
+by Very Long Baseline Interferometry (VLBI) observations, allowing a secure compensation
+for this already small effect. A measurement of Ṗb at the 0.02% level or better will provide
+stringent tests for alternative theories of gravity. For example, limits on some scalar-tensor
+
+11
+
+=== PAGE 12 ===
+
+theories will surpass the best current Solar-system tests (25).
+In GR, the parameter s can be identified with sin i where i is the inclination angle of the
+◦
+
+.50
+orbit. The value of s given in Table 1 corresponds to i = 88◦ .69+0
+−0◦ .76 . Based on scintillation
+
+observations of both pulsars over the short time interval when A is close to superior conjunction,
+Coles et al. (26) derived a value for |i − 90◦| of 0◦ .29 ± 0◦ .14. This is consistent with our
+measurement only at the 3-σ level. As mentioned above, we used the DDS model to solve for
+the Shapiro delay. Fig. 3 shows the resulting χ2 contours in the zs – mB plane. The value and
+uncertainty range for s quoted in Table 1 correspond to the peak and range of the 68% contour.
+Because of the non-linear relationship between zs and s, the uncertainty distribution in s (and
+hence in i) corresponding to these contours is very asymmetric with a very steep edge on the 90◦
+side. Only close to the 99% confidence limit is the timing result consistent with the scintillationderived value of |i − 90◦ | of 0◦ .29 ± 0◦ .14 (26). We note that the scintillation measurement is
+based on the correlation of the scintillation fluctuations of A and B over the short interval when
+A is close to superior conjunction (i.e., behind B). In contrast, the measurement of i from timing
+measurements depends on the detection of significant structure in the post-fit residuals after a
+portion of the Shapiro delay is absorbed in the fit for xA (27). As shown in Fig. 2, the Shapiro
+delay has a signature that is spread over the whole orbit and hence can be cleanly isolated. We
+also examined the effects on the Shapiro delay of using only low- or high-frequency data, and
+found values of s consistent withing the errors in each case. The scintillation result is based on
+the plasma properties of the interstellar medium and may also be affected by possible refraction
+effects in B’s magnetosphere. We believe that the timing result is much less susceptible to
+systematic errors and is therefore more secure.
+Scintillation observations have also been used to deduce the system transverse velocity.
+Ransom et al. (28) derive a value of 141 ± 8.5 km s−1 while Coles et al. (26) obtain 66 ± 15 km
+s−1 after considering the effect of anisotropy in the scattering screen. Both of these values are
+12
+
+=== PAGE 13 ===
+
+in stark contrast to the value of 10 ± 1 km s−1 (relative to the Solar system barycentre) obtained
+from pulsar timing (Table 1). We note that the scintillation-based velocity depends on a number
+of assumptions about the properties of the effective scattering screen. In contrast, the proper
+motion measurement has a clear and unambiguous timing signature, although the transverse
+velocity itself scales with the pulsar distance. Even allowing that unmodelled effects of Earth
+motion could affect the published scintillation velocities by about 30 km s−1 , the dispersionbased distance would need to be underestimated by a factor of several to make the velocities
+consistent. We believe this is very unlikely, particularly as the tentative detection of a parallax
+gives us some confidence in the dispersion-based distance estimate. Hence, we believe that our
+timing results for both inclination angle and transverse velocity are less susceptible to systematic
+errors and are therefore more secure than those based on scintillation.
+We note that, with the inclination angle being significantly different from 90◦ , gravitational
+lensing effects (29) can be neglected. The implied low space velocity, the comparatively low
+derived mass for B and the low orbit eccentricity are all consistent with the idea that the B pulsar
+may have formed by a mechanism different to the usually assumed core-collapse of a helium
+star (30, 31). A discussion of its progenitor is presented elsewhere (32). We also note that,
+as expected for a double-neutron-star system, there is no evidence for variation in dispersion
+measure as a function of orbital phase.
+
+Future tests.
+
+In contrast to all previous tests of GR, we are now reaching the point with PSR
+
+J0737−3037A where expressions of PK parameters to only 1PN order may not be sufficient
+anymore for a comparison of theoretical predictions with observations. In particular, we have
+measured ω̇ so precisely (i.e., to a relative precision approaching 10−5) that we expect corrections at the 2PN level (12) to be observationally significant within a few years. These corrections
+include contributions expected from spin-orbit coupling (33, 34). A future determination of the
+
+13
+
+=== PAGE 14 ===
+
+system geometry and the measurement of two other PK parameters at a level of precision similar to that for ω̇, would allow us to measure the moment of inertia of a neutron star for the
+first time (12, 35). While this measurement is potentially very difficult, a determination of A’s
+moment of inertia to a precision of only 30% would allow us to distinguish between a large
+number of proposed equations of state for dense matter (36, 37). The double pulsar would then
+not only provide the best tests of theories of gravity in the strong-field regime as presented here
+but would also give insight into the nature of super-dense matter.
+
+References and Notes
+1. C. Will, Living Reviews in Relativity 4, 4 (2001).
+2. B. Bertotti, L. Iess, P. Tortora, Nature 425, 374 (2003).
+3. J. G. Williams, S. G. Turyshev, D. H. Boggs, Phys. Rev. Lett. 93, 261101 (2004).
+4. T. Damour, G. Esposito-Farèse, Phys. Rev. D 58, 1 (1998).
+5. J. H. Taylor, J. M. Weisberg, ApJ 345, 434 (1989).
+6. I. H. Stairs, S. E. Thorsett, J. H. Taylor, A. Wolszczan, ApJ 581, 501 (2002).
+7. M. Burgay, et al., Nature 426, 531 (2003).
+8. A. G. Lyne, et al., Science 303, 1153 (2004).
+9. M. Burgay, et al., ApJ 624, L113 (2005).
+10. T. Damour, N. Deruelle, Ann. Inst. H. Poincaré (Physique Théorique) 44, 263 (1986).
+11. T. Damour, J. H. Taylor, Phys. Rev. D 45, 1840 (1992).
+
+14
+
+=== PAGE 15 ===
+
+12. T. Damour, G. Schäfer, Nuovo Cim. 101, 127 (1988).
+13. Supporting Online Material
+14. M. Kramer, et al., ApJ 526, 957 (1999).
+15. R. N. Manchester, et al., ApJ 621, L49 (2005).
+16. http://www.atnf.csiro.au/research/pulsar/tempo.
+17. T. Damour, N. Deruelle, Ann. Inst. H. Poincaré (Physique Théorique) 43, 107 (1985).
+18. M. Kramer, et al., Annalen der Physik 15, 34 (2006).
+19. J. M. Weisberg, J. H. Taylor, Binary Radio Pulsars, F. Rasio, I. H. Stairs, eds. (Astronomical Society of the Pacific, San Francisco, 2005), pp. 25–31.
+20. J. H. Taylor, A. Wolszczan, T. Damour, J. M. Weisberg, Nature 355, 132 (1992).
+21. M. Lyutikov, MNRAS 362, 1078 (2005).
+22. T. Damour, J. H. Taylor, ApJ 366, 501 (1991).
+23. J. M. Cordes, T. J. W. Lazio, NE2001. I. A New Model for the Galactic Distribution of Free
+Electrons and its Fluctuations (2002). astro-ph/0207156.
+24. K. Kuijken, G. Gilmore, MNRAS 239, 571 (1989).
+25. T. Damour, G. Esposito-Far‘ese, to appear. (2006).
+26. W. A. Coles, M. A. McLaughlin, B. J. Rickett, A. G. Lyne, N. D. R. Bhat, ApJ 623, 392
+(2005).
+27. C. Lange, et al., MNRAS 326, 274 (2001).
+15
+
+=== PAGE 16 ===
+
+28. S. M. Ransom, et al., ApJ 609, L71 (2004).
+29. R. R. Rafikov, D. Lai, Phys. Rev. D 73, 063003 (2006).
+30. E. Pfahl, S. Rappaport, P. Podsiadlowski, H. Spruit, ApJ 574, 364 (2002).
+31. T. Piran, N. J. Shaviv, Phys. Rev. Lett. 94, 051102 (2005).
+32. I. H. Stairs, S. E. Thorsett, R. J. Dewey, M. Kramer, C. McPhee, MNRAS in press (2006).
+33. T. Damour, R. Ruffini, Academie des Sciences Paris Comptes Rendus Ser. Scie. Math. 279,
+971 (1974).
+34. B. M. Barker, R. F. O’Connell, ApJ 199, L25 (1975).
+35. N. Wex, Class. Quantum Grav. 12, 983 (1995).
+36. I. A. Morrison, T. W. Baumgarte, S. L. Shapiro, V. R. Pandharipande, ApJ 617, L135
+(2004).
+37. J. M. Lattimer, B. F. Schutz, ApJ 629, 979 (2005).
+38. E. M. Standish, A&A 336, 381 (1998).
+39. We thank Thibault Damour and Norbert Wex for useful discussions. The Parkes radio telescope is part of the Australia Telescope which is funded by the Commonwealth of Australia
+for operation as a National Facility managed by CSIRO. The National Radio Astronomy
+Observatory is a facility of the U.S. National Science Foundation operated under cooperative agreement by Associated Universities, Inc. GASP is funded by an NSERC RTI-1 grant
+to IHS and by US NSF grants to Donald Backer and David Nice. We thank Paul Demorest,
+Ramachandran and Joeri van Leeuwen for their contributions to GASP hardward and software development. IHS holds an NSERC UFA, and pulsar research at UBC is supported by
+16
+
+=== PAGE 17 ===
+
+an NSERC Discovery Grant. MB, AP and ND’A acknowledge financial support from the
+Italian Ministry of University and Research (MIUR) under the national program Cofin 2003.
+FC is supported by NSF, NASA, and NRAO.
+
+17
+
+=== PAGE 18 ===
+
+with an inset showing an expanded view of the region of principal interest.
+Fig. 1. The tests of general relativity parameter summarized in a graphical form. Constraints on
+the masses of the two stars (A and B) in the PSR J0737−3039A/B binary system. Shaded regions are forbidden by the individual mass functions of A and B since sin i must be ≤ 1. Other
+constraining parameters are shown as pairs of lines, where the separation of the lines indicates
+the measurement uncertainty. For the diagonal pair of lines labelled as R, representing the mass
+ratio derived from the measured semi-major axes of the A and B orbits, the measurement precision is so good that the line separation only becomes apparent in the enlarged inset, showing
+an expanded view of the region of principal interest. The other constraints shown are based
+on the measured post-Keplerian (PK) parameters interpreted within the framework of general
+relativity. The PK parameter ω̇ describes the relativistic precession of the orbit, γ combines
+gravitational redshift and time dilation, while Ṗb represents the measured decrease in orbital
+period due to the emission of gravitational waves. The two PK parameters s and r reflect the
+observed Shapiro delay, describing a delay that is added to the pulse arrival times when propagating through the curved space-time near the companion. The intersection of all line pairs is
+consistent with a single point that corresponds to the masses of A and B. The current uncertainties in the observed parameters determine the size of this intersection area which is marked in
+blue and which reflects the achieved precision of this test of GR and the mass determination for
+A and B.
+Fig. 2. Measurement of a Shapiro delay demonstrating the curvature of space-time. Timing
+residuals (differences between observed and predicted pulse arrival times) are plotted as a function of orbital longitude and illustrate the Shapiro delay for PSR J0737−3039A. (a) Observed
+timing residuals after a fit of all model parameters given in Table 1 except the Shapiro-delay
+terms r and s which were set to zero and not included in the fit. While a portion of the delay
+is absorbed in an adjustment of the Keplerian parameters, a strong peak at 90◦ orbital longitude
+18
+
+=== PAGE 19 ===
+
+remains clearly visible. This is the orbital phase of A’s superior conjunction, i.e. when it is
+positioned behind B as viewed from Earth, so that its pulses experience a delay when moving
+through the curved space-time near B. The clear detection of structure in the residuals over the
+whole orbit confirms the detection of the Shapiro delay, which is isolated in (b) by holding all
+parameters to their best-fit values given in Table 1, except the Shapiro delay terms which were
+set to zero. The line shows the predicted delay at the centre of the data span. In both cases,
+residuals were averaged in 1◦ bins of longitude.
+Fig. 3. Contour plots of the χ2 distribution in the plane of the Shapiro-delay parameter zs ≡
+− ln(1 − s) and the mass of the B pulsar, mB . The contours correspond to 68%, 95% and 99%
+confidence limits.
+
+19
+
+=== PAGE 20 ===
+
+Table 1: Parameters for PSR J0737−3039A (A) and PSR J0737−3039B (B). The values were
+derived from pulse timing observations using the DD (10) and DDS (18) models of the timing
+analysis program TEMPO and the Jet Propulsion Laboratory DE405 planetary ephemeris (38).
+Estimated uncertainties, given in parentheses after the values, refer to the least significant digit
+of the tabulated value and are twice the formal 1-σ values given by TEMPO. The positional
+parameters are in the DE405 reference frame which is close to that of the International Celestial
+Reference System. Pulsar spin frequencies ν ≡ 1/P are in barycentric dynamical time (TDB)
+units at the timing epoch quoted in Modified Julian Days. The five Keplerian binary parameters
+(Pb , e, ω, T0, and x) are derived for pulsar A. The first four of these (with an offset of 180◦
+added to ω) and the position parameters were assumed when fitting for B’s parameters. Five
+post-Keplerian parameters have now been measured. An independent fit of ω̇ for B yielded a
+value (shown in square brackets) that is consistent with the much more precise result for A. The
+value derived for A was adopted in the final analysis (see (13)). The dispersion-based distance
+is based on a model for the interstellar electron density (23).
+
+20
+
+=== PAGE 21 ===
+
+Timing parameter
+Right Ascension α
+Declination δ
+Proper motion in the RA direction (mas yr−1 )
+Proper motion in Declination (mas yr−1 )
+Parallax, π (mas)
+Spin frequency ν (Hz)
+Spin frequency derivative ν̇ (s−2 )
+Timing Epoch (MJD)
+Dispersion measure DM (cm−3 pc)
+Orbital period Pb (day)
+Eccentricity e
+Projected semi-major axis x = (a/c) sin i (s)
+Longitude of periastron ω (deg)
+Epoch of periastron T0 (MJD)
+Advance of periastron ω̇ (deg/yr)
+Gravitational redshift parameter γ (ms)
+Shapiro delay parameter s
+Shapiro delay parameter r (µs)
+Orbital period derivative Ṗb
+Timing data span (MJD)
+Number of time offsets fitted
+RMS timing residual σ (µsec)
+Total proper motion (mas yr−1 )
+Distance d(DM) (pc)
+Distance d(π) (pc)
+Transverse velocity (d = 500 pc) (km s−1 )
+Orbital inclination angle (deg)
+Mass function (M⊙ )
+Mass ratio, R
+Total system mass (M⊙ )
+Neutron star mass (m⊙ )
+
+PSR J0737−3039A
+PSR J0737−3039B
+07h 37m 51s .24927(3)
+−
+−30◦ 39′ 40′′ .7195(5)
+−
+−3.3(4)
+−
+2.6(5)
+−
+3(2)
+−
+44.054069392744(2)
+0.36056035506(1)
+−3.4156(1) × 10−15
+−0.116(1) × 10−15
+53156.0
+53156.0
+48.920(5)
+−
+0.10225156248(5)
+−
+0.0877775(9)
+−
+1.415032(1)
+1.5161(16)
+87.0331(8)
+87.0331 + 180.0
+53155.9074280(2)
+−
+16.89947(68)
+[16.96(5)]
+0.3856(26)
+−
+0.99974(−39, +16)
+−
+6.21(33)
+−
+−1.252(17) × 10−12
+−
+52760 – 53736
+52760 – 53736
+10
+12
+54
+2169
+4.2(4)
+∼ 500
+200 − 1000
+10(1)
+88.69(-76,+50)
+0.29096571(87)
+0.3579(11)
+1.0714(11)
+2.58708(16)
+1.3381(7)
+1.2489(7)
+
+21
+
+=== PAGE 22 ===
+
+Table 2: Four independent tests of GR provided by the double pulsar. The second column lists
+the observed PK parameters obtained by fitting a DDS timing model to the data. The third
+column lists the values expected from general relativity given the masses determined from the
+intersection point of the mass ratio R and the periastron advance ω̇. The last column gives the
+ratio of the observed to expected value for each test. Uncertainties refer to the last quoted digit
+and were determined using Monte Carlo methods.
+PK parameter
+Ṗb
+γ (ms)
+s
+r(µs)
+
+Observed
+1.252(17)
+0.3856(26)
+0.99974(−39,+16)
+6.21(33)
+
+22
+
+GR expectation
+1.24787(13)
+0.38418(22)
+0.99987(−48,+13)
+6.153(26)
+
+Ratio
+1.003(14)
+1.0036(68)
+0.99987(50)
+1.009(55)
+
+=== PAGE 23 ===
+
+Fig. 1
+
+23
+
+=== PAGE 24 ===
+
+Fig. 2
+
+24
+
+=== PAGE 25 ===
+
+Fig. 3
+
+25
+
+=== PAGE 26 ===
+
+Supporting Online Material
+
+1 Observing systems
+The experimental data presented in the main paper are based on pulsar timing observations
+at several frequencies between 320 MHz and 3100 MHz using the Parkes radio telescope in
+Australia, the Lovell radio telescope at Jodrell Bank Observatory, UK, and the Green Bank
+Telescope (GBT) in the USA, between 2003 April and 2006 January. Details of the observing
+systems are summarized in Supporting Table 1.
+At the Parkes 64-m radio telescope observations were carried out using the centre beam of
+the 20-cm multibeam receiver and a coaxial 10cm/50cm receiver. For each of these cryogenically cooled receivers, two orthogonally polarized signals were amplified and down-converted
+to an intermediate frequency. These signals were transferred to band splitters and fed into a
+filterbank system (FB) for each polarization of each feed. The output of each filter was detected
+and summed with its corresponding polarization pair. These summed outputs were high-pass
+filtered and integrated for the sampling interval of 80 µs and then one-bit digitised. While the
+original frequency channels were folded with a reference frequency corresponding to the band
+centre, timing was performed on sub-bands.
+Observations at the GBT utilized two different data acquisition systems. The BerkeleyCaltech Pulsar Machine (BCPM) is a flexible filterbank system (1), with which we collected 4bit summed-polarization data. The Green Bank Astronomical Signal Processor (GASP) carries
+out 8-bit Nyquist-sampling of the incoming dual-polarization signal, after which it performs
+coherent dedispersion in software on a Linux-based cluster for each of several 4-MHz channels
+(2, 3). The data stream is then detected, and the two polarizations are usually flux-calibrated
+before summation using a diode noise source as a reference.
+At Jodrell Bank we used a incoherently dedispersing filterbank system. Its parameters are
+26
+
+=== PAGE 27 ===
+
+summarized in Table 1, while details of the observing system can be found in ref. 4.
+
+2 Dedispersion
+Since the interstellar medium (ISM) is ionized, the propagation speed of radio pulses depends
+on their radio frequency with pulses emitted at a high radio frequencies arriving earlier than
+low-frequency pulses. Unless this effect is accounted for, pulses will be broadened over the
+finite observing bandwidth. Two dedispersion techniques are in use. For “incoherent dedispersion”, the bandwidth is sub-divided into a number of frequency channels which are detected
+and sampled independently. Dispersion smearing is thereby reduced to the smearing across an
+individual filterbank channel. The “coherent dedispersion” technique involves the application
+of an inverse “ISM-filter” to the raw voltage data received from the antenna (5). This technique
+is computationally more intensive but removes the effects of dispersion completely.
+At Parkes and Jodrell Bank we obtained incoherently dedispersed data using the filterbank
+systems listed in Table 1. The resulting profiles were summed across frequency channels with
+appropriate delays to remove the effects of interstellar dispersion. For the wide-bandwidth
+Parkes data, where the original frequency channels were folded with a reference frequency corresponding to the band centre, timing was performed on sub-bands. The number of sub-bands
+was chosen such that the dispersion delay across the sub-bands was significantly smaller than
+the overall timing precision. Analysis of TOA data separately for the different sub-bands properly accounts for the fact that data at different frequencies received at a given time correspond
+to different orbital phases at emission due to the differential dispersion delay (see e.g. (6)).
+At the GBT, the BCPM data were divided in four frequency sub-bands, separately dedispersed, folded and timed. In contrast, each GASP 4-MHz channel was coherently dedispersed
+and folded using the channel centre frequency as a reference. The GASP channels were then
+summed appropriately to give a single TOA for each integration.
+27
+
+=== PAGE 28 ===
+
+3 Pulse Time-of-Arrival analysis
+Pulse times-of-arrival (TOAs) were computed by correlating the observed pulse profiles with
+synthetic noise-free templates (see Figs. 1 and 2; cf. ref. (7)). All datasets obtained at different
+epochs and frequencies with different data acquisition hardware and telescopes were studied
+for possible systematic errors and artificial correlations. Firstly, correlations between successive TOAs were investigated by computing the post-fit root-mean-square (rms) timing residuals
+with averaging of consecutive TOAs, expecting that the rms residual should decrease with the
+square-root of the number of averaged TOAs. Datasets with significant deviations from this
+expected scaling were excluded from the analysis. Secondly, for the GBT observations where
+we recorded data with two different data acquisition systems in parallel, we preferred to use to
+more accurate GASP data and only used BCPM data if no GASP TOAs were available within 2
+minutes of a BCPM TOA. Thirdly, the uncertainties of the TOAs in the remaining datasets were
+studied by inspecting the reduced χ2 achieved in the fit of the timing model. For most datasets
+we applied a small quadrature addition and a scaling factor to the uncertainties to obtain the expected value of χ2red = 1. No adjustments to the TOA uncertainties were needed for the GASP
+data; this is not surprising as the 8-bit sampling provides excellent profile fidelity. Finally, all
+retained datasets were combined in a weighted least-squares fit of the DD and DDS models.
+Following these fits, we verified that the χ2red for each data subset was still close to unity. A
+total of 131,416 arrival times were included in the final analysis of A while 507 TOAs were
+used for B, most at frequencies close to 820 MHz and 1400 MHz. The much smaller number
+of TOAs for B results from several factors: JBO data were not used, the integration time for B
+was a factor of ten larger than for A, the data were summed over the entire observed frequency
+band, only about 20% of the orbit was used and finally, even in the analysed regions, B was
+often too weak to give a significant TOA. Figures 3 and 4 summarise the TOA distributions for
+
+28
+
+=== PAGE 29 ===
+
+the different observatories for pulsars A and B respectively. Finally, we present the covariance
+matrix as computed by TEMPO for the fit of the DDS timing model in Table 2.
+
+References and Notes
+1. D. C. Backer, et al., PASP 109, 61 (1997).
+2. P. Demorest, et al., American Astronomical Society Meeting Abstracts 205, (2004).
+3. R. D. Ferdman, et al., American Astronomical Society Meeting Abstracts 205, (2004).
+4. D. M. Gould, A. G. Lyne, MNRAS 301, 235 (1998).
+5. T. H. Hankins, B. J. Rickett, Methods in Computational Physics Volume 14 — Radio Astronomy (Academic Press, New York, 1975), pp. 55–129.
+6. G. B. Hobbs, R. T. Edwards, R. N. Manchester, MNRAS 369, 655 (2006).
+7. M. Kramer, et al., ApJ 526, 957 (1999).
+
+29
+
+=== PAGE 30 ===
+
+Supporting Table 1. Summary of the observing systems used for timing observations of the
+double pulsar.
+Telescope
+
+Instrument
+
+Parkes
+
+FB
+
+GBT
+
+BCPM
+GASP
+
+Jodrell Bank
+
+FB
+FB
+
+Centre
+freq. (MHz)
+680
+1374
+3030
+820
+1400
+340
+820
+1400
+610
+1396
+
+Gain
+(K/Jy)
+0.66
+0.74
+0.62
+2.0
+2.0
+2.0
+2.0
+2.0
+1.1
+1.1
+
+Tsys
+(K)
+45
+22
+28
+25
+20
+70
+25
+20
+32
+32
+
+Sample
+interval (µs)
+80
+80
+80
+72
+72
+0.25
+0.25
+0.25
+44.4
+44.4
+
+Bandwidth
+(MHz)
+64
+256
+768
+48
+96
+16
+64
+64
+8
+64
+
+Number
+of channels
+128
+512
+256
+96
+96
+4
+16a
+16a
+32
+64
+
+a The number of channels and hence bandwidth that was used varied occasionally within a given session due to the removal of channels
+
+contaminated with radio frequency interference and/or occasional recording disk-space limitations.
+
+30
+
+=== PAGE 31 ===
+
+Supporting Table 2. Covariance matrix as computed by TEMPO for a fit of the DDS timing
+model to the TOAs of A.
+ν
+ν̇
+Dec
+RA
+PMDec
+PMRA
+x
+e
+T0
+Pb
+ω
+ω̇
+γ
+DM
+π
+Ṗb
+zs
+m2
+
+ν
+1.00
+-0.76
+0.16
+0.10
+-0.25
+0.01
+-0.02
+0.00
+-0.54
+0.47
+-0.54
+0.47
+-0.02
+-0.02
+0.05
+0.01
+-0.01
+0.01
+
+ν̇
+
+Dec
+
+RA
+
+PMDec
+
+PMRA
+
+x
+
+e
+
+T0
+
+Pb
+
+ω
+
+ω̇
+
+γ
+
+DM
+
+π
+
+Ṗb
+
+zs
+
+m2
+
+1.00
+-0.31
+-0.08
+0.39
+-0.28
+0.02
+0.00
+0.43
+-0.47
+0.42
+-0.48
+-0.02
+0.02
+-0.02
+-0.02
+0.02
+-0.03
+
+1.00
+0.18
+-0.83
+0.04
+0.01
+0.01
+-0.01
+0.01
+-0.01
+0.01
+0.00
+-0.01
+-0.21
+0.00
+0.01
+-0.01
+
+1.00
+-0.16
+-0.71
+0.00
+-0.01
+0.02
+-0.02
+0.02
+-0.02
+0.01
+0.01
+0.16
+0.03
+-0.01
+0.00
+
+1.00
+0.12
+-0.01
+-0.02
+0.02
+-0.02
+0.02
+-0.02
+0.00
+0.03
+0.20
+0.00
+-0.01
+0.01
+
+1.00
+-0.01
+-0.01
+-0.03
+0.02
+-0.03
+0.02
+0.00
+0.02
+-0.12
+-0.02
+0.00
+0.01
+
+1.00
+0.66
+0.00
+-0.15
+0.06
+-0.15
+0.44
+0.00
+-0.03
+0.05
+0.55
+-0.85
+
+1.00
+0.01
+-0.15
+0.01
+-0.15
+0.01
+0.00
+-0.04
+0.00
+0.44
+-0.75
+
+1.00
+-0.85
+0.99
+-0.85
+-0.03
+0.00
+-0.01
+-0.03
+0.01
+-0.02
+
+1.00
+-0.84
+1.00
+0.02
+0.00
+-0.01
+-0.06
+-0.10
+0.18
+
+1.00
+-0.84
+0.10
+0.00
+0.00
+0.01
+0.01
+-0.02
+
+1.00
+0.03
+0.00
+-0.01
+0.01
+-0.10
+0.18
+
+1.00
+0.00
+0.02
+0.13
+-0.01
+-0.02
+
+1.00
+-0.02
+0.00
+0.00
+0.00
+
+1.00
+-0.01
+-0.01
+0.03
+
+1.00
+0.00
+0.00
+
+1.00
+-0.72
+
+1.00
+
+31
+
+=== PAGE 32 ===
+
+Supporting Figure 1. Pulse profile templates used for TOA determinations for pulsar A.
+Supporting Figure 2. Regions of orbital phase (hatched) used for timing of pulsar B and pulse
+profile templates for these phases derived from and used for the 820 MHz GBT observations
+in May 2005. Each of the template plots covers a range of 60/360 = 0.17 in pulse phase.
+Similar but different templates were used for other frequencies and epochs. While B was clearly
+detectable in these three regions, it is actually brightest in the two cross-hatched regions, but
+because the shape of the profile evolves quickly and dramatically in these regions, they were
+excluded from the timing analysis.
+Supporting Figure 3. Timing residuals obtained for pulsar A for the three telescopes and their
+distribution. The upper panel shows the distribution of observations in frequency.
+Supporting Figure 4. Timing residuals obtained for pulsar B for Parkes and the GBT and their
+distribution. The upper panel shows the distribution of observations in frequency.
+
+32
+
+=== PAGE 33 ===
+
+Fig. 1
+
+33
+
+=== PAGE 34 ===
+
+Fig. 2
+
+34
+
+=== PAGE 35 ===
+
+Fig. 3
+
+35
+
+=== PAGE 36 ===
+
+Fig. 4
+
+36

binary_pulsar_post_keplerian_atnf__PBDOT/reference/manchester_2005.bib

@@ -0,0 +1,10 @@
+@ARTICLE{Manchester2005,
+  author = {{Manchester}, R. N. and {Hobbs}, G. B. and {Teoh}, A. and {Hobbs}, M.},
+  title = {{The Australia Telescope National Facility Pulsar Catalogue}},
+  journal = {AJ},
+  year = 2005,
+  volume = {129},
+  pages = {1993--2006},
+  doi = {10.1086/428488},
+  note = {Description paper for psrcat v1; current version 2.7.0 (2025) used here. ATNF psrcat is freely redistributable for academic use with citation. Tarball: https://www.atnf.csiro.au/research/pulsar/psrcat/downloads/psrcat_pkg.tar.gz.},
+}

binary_pulsar_post_keplerian_atnf__PBDOT/reference/manchester_2005.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba885f66905706e2907bce6e130ac3f22aa87c938bdf0bff4644eb409e6809da
+size 512901

binary_pulsar_post_keplerian_atnf__PBDOT/reference/manchester_2005.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb63ac852a064c916666f681193c19dad87868ca2c893653ee90d4adbc8f7639
+size 61708

binary_pulsar_post_keplerian_atnf__PBDOT/reference/peters_1963.bib

@@ -0,0 +1,10 @@
+@ARTICLE{PetersMathews1963,
+  author = {{Peters}, P. C. and {Mathews}, J.},
+  title = {{Gravitational Radiation from Point Masses in a Keplerian Orbit}},
+  journal = {Phys. Rev.},
+  year = 1963,
+  volume = {131},
+  pages = {435--440},
+  doi = {10.1103/PhysRev.131.435},
+  note = {Foundational PN derivation of dPb/dt for eccentric binaries.},
+}

binary_pulsar_post_keplerian_atnf__PBDOT/reference/peters_1963.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e59fec4a24852daa3ccdbc5ff2d88b6a04914f2a6ea719c9cd68c0049b4e31a
+size 23566944

binary_pulsar_post_keplerian_atnf__PBDOT/reference/peters_1963.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3705f814016ff6ed5a92226727b8a583401bc622e773a830fd2f38309234fff6
+size 20046

binary_pulsar_post_keplerian_atnf__PBDOT/reference/summary_dataset_manchester_2005.md

@@ -0,0 +1,96 @@
+**Type**: Dataset — primary data source for the `binary_pulsar_post_keplerian_atnf__PBDOT` task; provides observed PBDOT values and all supporting Keplerian/post-Keplerian parameters for binary pulsars.
+**Source**: PDF at `reference/manchester_2005.pdf`
+**arXiv / DOI / journal**: DOI 10.1086/428488 — *The Astronomical Journal*, 129:1993–2006, 2005 April
+**Pages read**: 1,2,3,4,5,10,11,12
+
+---
+
+## 1. Background
+
+Manchester et al. (2005) describe the Australia Telescope National Facility (ATNF) Pulsar Catalogue, a database of all published parameters for 1509 pulsars compiled from the literature by ATNF, Jodrell Bank Observatory, and collaborating groups. The catalogue supersedes the Taylor et al. (1993) catalogue (558 pulsars) and is accessed via the command-line program PSRCAT and a World Wide Web interface. The underlying database is an ASCII keyword–value store built on the system originally developed at the University of Massachusetts (Manchester & Taylor 1972). The catalogue is updated periodically; the description paper covers v1, while the .bib note records that v2.7.0 (2025) is the version actually used in this benchmark. (PDF p. 1)
+
+## 2. What the Dataset Contains
+
+The catalogue stores 67 predefined parameters per pulsar plus expert-mode parameters; all parameters with their keywords and units are listed in Tables 1 and 2 of the paper. For binary pulsars, the parameters directly relevant to the benchmark are:
+
+**Basic Binary System Parameters (Table 1, PDF p. 2)**
+
+| Keyword | Description | Units |
+|---------|-------------|-------|
+| Binary | Binary timing model (BT, ELL1, DD, …) | — |
+| T0 | Epoch of periastron | MJD |
+| PB | Binary period of pulsar | days |
+| A1 | Projected semimajor axis of pulsar orbit, $a_1 \sin i$ | s |
+| OM | Longitude of periastron, $\omega$ | deg |
+| Ecc | Eccentricity, $e$ | — |
+| Tasc | Epoch of ascending node (ELL1 model) | MJD |
+| Eps1 | $e \sin\omega$ (ELL1 model) | — |
+| Eps2 | $e \cos\omega$ (ELL1 model) | — |
+| MinMass | Minimum companion mass ($i = 90°$, $M_\mathrm{NS} = 1.35\,M_\odot$) | $M_\odot$ |
+| MedMass | Median companion mass ($i = 60°$) | $M_\odot$ |
+
+**Expert Binary Parameters (Table 2, PDF p. 4–5)**
+
+| Keyword | Description | Units |
+|---------|-------------|-------|
+| OMDOT | Periastron advance $\dot\omega$ | deg yr⁻¹ |
+| **PBDOT** | **First time derivative of binary period $\dot P_b$** | **dimensionless (s s⁻¹)** |
+| A1DOT | Rate of change of projected semimajor axis | — |
+| ECCDOT | Rate of change of eccentricity | s⁻¹ |
+| GAMMA | Relativistic time dilation term | s |
+| PPNGAMMA | PPN parameter $\gamma$ | — |
+| SINI | Sine of inclination angle $\sin i$ | — |
+| M2 | Companion mass | $M_\odot$ |
+| MTOT | Total system mass | $M_\odot$ |
+| MASSFN | Pulsar mass function | $M_\odot$ |
+| UPRMASS | 90% confidence companion mass upper limit ($i = 26°$) | $M_\odot$ |
+| XPBDOT | Rate of change of orbital period minus GR prediction | — |
+| XOMDOT | Rate of periastron advance minus GR prediction | deg yr⁻¹ |
+| DTHETA | Relativistic deformation of the orbit | — |
+
+The catalogue contains data for **1509 pulsars total** as of the description paper; binary pulsars with a measured PB are tagged `type BINARY` (PDF p. 5). PBDOT is an expert-mode parameter measured only for binary systems with sufficient timing baselines; its value is blank (missing) for most pulsars.
+
+Table 10 (PDF p. 12) illustrates the kind of data stored: globular-cluster binary pulsars with Period, Period Derivative, Binary Period (days), and Median Companion Mass. Table 9 (PDF p. 11) lists high-energy and AXP/SGR pulsars with Period, Age, and Surface B-field.
+
+## 3. Selection / Quality Cuts
+
+- Includes **all spin-powered pulsars** with published information up to the compilation date; excludes accretion-powered X-ray pulsars (very different physical regime) and non-pulsating systems. (PDF p. 1)
+- AXPs and SGRs (magnetars) are included because they are similar to ordinary pulsars in most respects. (PDF p. 1)
+- **All data values have an associated reference key** pointing to a BIBTEX bibliography; the catalogue contains > 360 bibliography entries. (PDF p. 1)
+- PBDOT values are present only for binary systems where pulsar timing has accumulated sufficient baseline; no filtering criterion is stated beyond "published information currently exists."
+- The binary timing model used (BT / ELL1 / DD) is recorded per pulsar: DD (Damour & Deruelle 1986) is used where relativistic effects are important; ELL1 (Wex 2000) for near-circular orbits; BT (Blandford & Teukolsky 1976) most commonly otherwise. (PDF p. 3)
+
+## 4. Train / Test Split — Qualitative
+
+The ATNF catalogue does not itself define a train/test split. It is a literature-compilation database, not a designed experiment. Each pulsar is an independent measurement entry. For benchmark purposes, the split must be defined externally (e.g., by the v0.5 §4 pipeline), stratified on system properties such as binary period, companion type (NS/WD/Main-sequence), or eccentricity class. The paper makes no remark about cross-validation or holdout sets.
+
+## 5. Cluster / Group Structure
+
+The dataset is **not hierarchically clustered** in the usual sense: each pulsar is a distinct astrophysical object observed independently. However, natural sub-groupings exist and may inform split design:
+
+- **By association**: globular-cluster pulsars (Table 10, PDF p. 12) vs. field pulsars — globular-cluster systems may have distinct PBDOT behaviour due to cluster gravitational acceleration (a kinematic bias on $\dot P_b$).
+- **By binary type**: double neutron star (DNS) systems vs. pulsar–white-dwarf (recycled MSP) systems vs. other — DNS systems are the relativistic systems for which PBDOT is dominated by gravitational-wave energy loss.
+- **By timing model**: DD-model pulsars are the relativistic binaries where PBDOT is a meaningful post-Keplerian parameter.
+
+The benchmark task selects binary pulsars with measured PBDOT; each pulsar contributes one row.
+
+## 6. Validation Criteria — Layer 2 Rubric
+
+The paper itself does not derive or discuss the physical formula for PBDOT. It is a catalogue paper; the theoretical formula (Peters 1963 quadrupole radiation) and its GR tests are in the companion papers in this reference set. However, the catalogue defines one directly relevant check:
+
+- **XPBDOT is defined as "rate of change of orbital period minus GR prediction"** (Table 2, PDF p. 4); any model for PBDOT should recover XPBDOT ≈ 0 for well-tested systems consistent with GR.
+- **PBDOT is listed as an expert-mode parameter** (Table 2, PDF p. 4); it is non-trivially measured and only present for a subset of binary pulsars, implying that the benchmark rows are restricted to systems with high-quality timing solutions.
+- **Binary period PB is a basic parameter (Table 1, PDF p. 2)** stored in days; any formula must be consistent with the scale of PB values in the catalogue (millisecond-period binaries to multi-day orbits visible in Table 10, PDF p. 12).
+
+## 7. License / Redistribution Constraints
+
+The .bib note states: *"ATNF psrcat is freely redistributable for academic use with citation."* No NDA, no paywall, no "research use only" restriction. SPDX: no formal identifier given; treat as `CC-BY` equivalent for academic redistribution. The catalogue paper itself is published in AJ (2005), all rights reserved by AAS; however, the **database** (psrcat) is separately distributed as a freely downloadable tarball.
+
+## 8. Notes for Benchmark Integration
+
+- **Target column**: `PBDOT` (keyword in PSRCAT; dimensionless, i.e., s s⁻¹). Typical values for GW-dominated DNS systems are $\sim -10^{-12}$ (negative = orbital decay). Millisecond pulsar–WD systems often have small positive or negative PBDOT dominated by kinematic terms.
+- **Input columns** likely include: `PB` (days), `A1` (s), `Ecc`, `OM` (deg), `M2` ($M_\odot$), `MTOT` ($M_\odot$), `OMDOT` (deg yr⁻¹), `GAMMA` (s), `SINI` — all drawn from the same catalogue.
+- **Unit convention**: PB in days in the catalogue; the Peters formula uses SI or CGS internally — conversion factors must be checked in `baselines.py`.
+- **Globular-cluster bias**: pulsars in globular clusters (keyword `Assoc` contains `GC:`) have PBDOT contaminated by the cluster's gravitational potential acceleration. They should either be excluded from the benchmark dataset or treated as a separate Type II cluster if the benchmark is testing GR predictions.
+- **Setting / Type eligibility**: Because each row is an independent pulsar (no natural within-object clustering), this is a **Setting 1 / Type I** dataset. Setting 2 / Type II would require multiple measurements of the same system under varying conditions, which is not the catalogue structure.
+- **Version note**: The description paper covers psrcat v1 (2005, 1509 pulsars); the benchmark uses v2.7.0 (2025), which contains substantially more pulsars and updated measurements. The schema described in Tables 1 and 2 is stable across versions.

binary_pulsar_post_keplerian_atnf__PBDOT/reference/summary_formula+dataset_kramer_2006.md

@@ -0,0 +1,110 @@
+**Type**: Formula+Dataset — primary observational source for PBDOT in the double pulsar PSR J0737-3039A/B; provides GR prediction framework and all Keplerian/PK system parameters that benchmark rows are built from.
+**Source**: PDF at `reference/kramer_2006.pdf`
+**arXiv / DOI / journal**: arXiv:astro-ph/0609417; DOI 10.1126/science.1132305; *Science* **314**, 97–102 (2006)
+**Pages read**: 1, 9–12, 20–22, 27–31
+
+---
+
+## 1. Background and landscape
+
+The double pulsar PSR J0737-3039A/B (discovered 2003–2004) is unique: both neutron stars are detectable as radio pulsars. Its short 2.4-hour orbital period and edge-on orientation yield five measurable post-Keplerian (PK) parameters plus the theory-independent mass ratio R = m_A/m_B = x_B/x_A, providing four independent tests of GR within a single system. The orbital period derivative Ṗ_b (due to gravitational wave emission) is one of these five PK parameters; it had previously been tested only in PSR B1913+16 (Hulse-Taylor pulsar, ~30-year baseline, 0.2% precision) and PSR B1534+12 (0.7% precision).
+
+## 2. Research gap
+
+Earlier tests of the radiative sector of GR required decades of data and were limited by uncertain kinematic corrections (Galactic potential acceleration, pulsar distance) and by fitting a single-pulsar system. PSR J0737-3039A/B has much smaller kinematic corrections (estimated ≤ 0.02% for Ṗ_b; PDF p. 11) and allows a mass-ratio constraint independent of any gravitational theory, making GR tests qualitatively different from prior work.
+
+## 3. Core formula(s)
+
+### 3a. GR prediction for orbital period derivative Ṗ_b (Peters / Damour–Deruelle formula)
+
+The GR formula for Ṗ_b due to gravitational-wave energy loss is **not written explicitly in this paper**. It is used implicitly via the Damour–Deruelle formalism (refs. 10, 17 in the paper: Damour & Deruelle 1985, 1986) to compute the GR prediction 1.24787(13) × 10^{-12} given in Table 2 (PDF p. 22). The formula referenced therein is the standard Peters quadrupole formula:
+
+$$\dot{P}_b^{\rm GR} = -\frac{192\pi}{5}\,T_\odot^{5/3}\left(\frac{P_b}{2\pi}\right)^{-5/3} \frac{m_A\,m_B}{M^{1/3}}\,f(e)$$
+
+where
+
+$$f(e) = \frac{1 + \tfrac{73}{24}e^2 + \tfrac{37}{96}e^4}{(1-e^2)^{7/2}}, \quad T_\odot \equiv \frac{GM_\odot}{c^3} = 4.925490947\;\mu{\rm s}$$
+
+> **Note:** The constant T⊙ = 4.925490947 μs is stated explicitly on PDF p. 9. The full Ṗ_b formula is not written in this paper; its explicit form appears in the companion references (Peters 1963, Damour & Deruelle 1986, also referenced as the `peters_1963.bib` entry in this task's reference dir).
+
+**Parameter classification for Ṗ_b^GR formula:**
+
+| Symbol | Role | Physical meaning | Units | Typical range / magnitude | Notes |
+|---|---|---|---|---|---|
+| Ṗ_b | **Output / target** | Orbital period time derivative (GR prediction) | dimensionless (s/s) | ~10^{-12} (negative for decaying orbit) | Benchmark SR target |
+| P_b | **Input variable** | Orbital period | days or seconds | 0.1 d (this system); wider range in ATNF catalog | Keplerian parameter |
+| e | **Input variable** | Orbital eccentricity | dimensionless | 0–1; this system e ≈ 0.088 | Keplerian parameter |
+| m_A | **Input variable** | Mass of pulsar A (recycled) | M⊙ | 1.0–1.5 M⊙ | Derived from ω̇ + R |
+| m_B | **Input variable** | Mass of pulsar B (young) | M⊙ | 1.0–1.5 M⊙ | Derived from ω̇ + R |
+| M = m_A + m_B | (derived) | Total system mass | M⊙ | ~2–3 M⊙ | Not independent input |
+| T⊙ = GM⊙/c³ | **Fixed constant** | Solar mass in time units | μs | 4.925490947 μs | Universal constant; stated PDF p. 9 |
+| 192π/5 | **Fixed constant** | Numerical prefactor in quadrupole formula | dimensionless | ~120.5 | From GR quadrupole radiation; structural |
+| 73/24, 37/96 | **Fixed constants** | Eccentricity-enhancement coefficients in f(e) | dimensionless | ~3.04, ~0.385 | Structural constants of the Peters formula; encode GW energy loss at different orbital phases |
+| 7/2 (exponent in f(e)) | **Fixed constant** | Denominator exponent in f(e) | dimensionless | 3.5 | Structural |
+
+### 3b. GR formula for periastron advance ω̇ (explicitly stated in paper)
+
+On PDF p. 9, the paper explicitly gives:
+
+$$\dot\omega = 3\left(\frac{P_b}{2\pi}\right)^{-5/3}(T_\odot M)^{2/3}(1-e^2)^{-1}$$
+
+This formula is used to derive the masses from the timing data. It is not the benchmark target but is needed to understand how m_A and m_B are inferred.
+
+### 3c. Validity domain
+
+All formulas apply in the point-mass, negligible-spin limit. For PSR J0737-3039A/B spin contributions are estimated to be at least one order of magnitude below current precision. The GR Ṗ_b formula is valid for bound elliptical orbits (0 ≤ e < 1); the double pulsar's low eccentricity (e ≈ 0.088) makes the f(e) correction modest (~1.17 relative to circular).
+
+---
+
+## 4. Variable / parameter consolidation table
+
+(System-specific values from Table 1, PDF p. 21)
+
+| Symbol | Role | Physical meaning | Units | Value / range (this system) |
+|---|---|---|---|---|
+| Ṗ_b | Output / target | Orbital period derivative | dimensionless | −1.252(17) × 10^{-12} (observed); GR pred. −1.24787(13) × 10^{-12} |
+| P_b | Input variable | Orbital period | day | 0.10225156248(5) |
+| e | Input variable | Eccentricity | dimensionless | 0.0877775(9) |
+| m_A | Input variable | Pulsar A (recycled) mass | M⊙ | 1.3381(7) |
+| m_B | Input variable | Pulsar B (young) mass | M⊙ | 1.2489(7) |
+| x_A = (a_A/c) sin i | per-system covariate | Projected semi-major axis of A | s | 1.415032(1) |
+| x_B = (a_B/c) sin i | per-system covariate | Projected semi-major axis of B | s | 1.5161(16) |
+| R = m_A/m_B = x_B/x_A | derived | Theory-independent mass ratio | dimensionless | 1.0714(11) |
+| ω̇ | per-system covariate | Advance of periastron | deg/yr | 16.89947(68) |
+| γ | per-system covariate | Gravitational redshift + time dilation PK param | ms | 0.3856(26) |
+| s | per-system covariate | Shapiro delay shape | dimensionless | 0.99974(−39,+16) |
+| r | per-system covariate | Shapiro delay range | μs | 6.21(33) |
+| T⊙ | Fixed constant | GM⊙/c³ | μs | 4.925490947 |
+| i | per-system covariate | Orbital inclination | deg | 88.69(−76,+50) |
+| M = m_A + m_B | derived | Total system mass | M⊙ | 2.58708(16) |
+| d | per-system covariate | Distance (DM-based) | pc | ~500 (DM); 200–1000 (parallax) |
+
+---
+
+## 5. Train / test split — qualitative
+
+This is a single-system observational paper, not a multi-system catalog study. There is no train/test split. The timing data cover MJD 52760–53736 (~2.5 years, 2003 April – 2006 January) using Parkes, GBT, and Jodrell Bank telescopes. A total of 131,416 TOAs for pulsar A and 507 TOAs for B. The mass determination uses ω̇ and R as anchors; GR predictions for the remaining four PK parameters are then compared to observations as independent tests (Table 2).
+
+For the `binary_pulsar_post_keplerian_atnf__PBDOT` benchmark, this paper contributes one row (PSR J0737-3039A/B) with precise measured Ṗ_b and system parameters. The broader training data come from the ATNF pulsar catalog, which aggregates many binary pulsars.
+
+---
+
+## 6. Validation criteria — Layer 2 rubric
+
+- **Ṗ_b must be negative**: Gravitational wave emission removes orbital energy, causing the orbit to shrink and P_b to decrease; Ṗ_b = −1.252(17) × 10^{-12} for this system (PDF p. 8 text and Table 1 p. 21).
+- **Kinematic correction to Ṗ_b is ≤ 0.02% for PSR J0737-3039A/B**: The small transverse velocity (10 ± 1 km/s) and proximity (~500 pc) imply Galactic-potential and Shklovskii corrections are negligible compared to measurement uncertainty; any candidate Ṗ_b formula must be insensitive to these corrections at this level (PDF p. 11, "Space motion and inclination of the orbit").
+- **Ṗ_b^obs / Ṗ_b^GR = 1.003 ± 0.014**: The observed orbital decay rate must match the GR quadrupole-radiation prediction to within ~1.4% after 2.5 years of data; this is the primary benchmark criterion for PBDOT (PDF p. 22, Table 2).
+- **Measurement precision of Ṗ_b scales approximately as T^{-2.5}** with data span T, reflecting the growing lever-arm of orbital phase measurements (PDF p. 11); candidate expressions must reproduce a Ṗ_b that improves in agreement with more data rather than diverging.
+- **All five PK parameters plus R must intersect at a single point in the m_A–m_B plane**: A physically valid Ṗ_b prediction, given the same masses inferred from ω̇ and R, must be consistent with the other PK-parameter constraints (PDF p. 8–9, Fig. 1 description p. 18).
+
+---
+
+## 7. Notes for benchmark integration
+
+- **Benchmark target**: `PBDOT` = Ṗ_b, the orbital period derivative, in units of 10^{-12} (dimensionless, but conventionally quoted as s/s × 10^{12}).
+- **baselines.py mapping**: `f(Pb, e, mA, mB)` → `Ṗb^GR`. Inputs in SI or solar units; output in 10^{-12}. The formula's fixed constants (192π/5, 73/24, 37/96, T⊙) must be embedded.
+- **Sign convention**: Ṗ_b is negative for a decaying orbit. Some catalogs (ATNF) may tabulate |Ṗ_b| — check sign convention in the catalog column.
+- **Units**: ATNF catalog tabulates P_b in days; convert to seconds (× 86400) for SI Peters formula.
+- **Kinematic correction**: For the full ATNF sample, the observed Ṗ_b^obs includes a kinematic (Shklovskii + Galactic) bias. The GR formula predicts Ṗ_b^intrinsic. For well-characterized systems the correction is small but non-negligible; the benchmark target is the intrinsic value or the observed value depending on the ATNF column used — verify which column is ingested.
+- **Setting / Type eligibility**: Single-system paper — contributes to `setting1_typeI` (formula evaluated per pulsar system using that system's own parameters). If multiple systems are combined from the ATNF catalog, `setting2_typeI` applies (different systems sharing the same formula form but different per-system parameters).
+- **invariant_constants**: 192π/5, T⊙ = 4.925490947 μs, 73/24, 37/96 — these are the fixed structural constants of the Ṗ_b formula that must remain identical across all binary pulsar applications.

binary_pulsar_post_keplerian_atnf__PBDOT/reference/summary_formula_damour_1991.md

@@ -0,0 +1,129 @@
+**Type**: Formula — establishes the theoretical link between the raw observed orbital-period derivative $\dot{P}_b^{\rm obs}$ and the intrinsic general-relativistic prediction $\dot{P}_b^{\rm GR}$ for PSR 1913+16 by accounting for galactic acceleration and Shklovskii (proper-motion) kinematic corrections; primary correction-formula source for the `binary_pulsar_post_keplerian_atnf__PBDOT` benchmark target.
+**Source**: PDF at `reference/damour_1991.pdf`
+**arXiv / DOI / journal**: DOI 10.1086/169585; *The Astrophysical Journal*, 366:501–511, 1991 January 10
+**Pages read**: 1, 2, 9, 10, 11
+
+---
+
+## 1. Background and landscape
+
+The binary pulsar PSR 1913+16 (Hulse & Taylor 1975) provides a laboratory for testing general relativity through the secular decay of its orbital period. By 1991 the raw timing parameter $\dot{P}_b^{\rm obs}$ had been measured to ~0.8% precision (Taylor & Weisberg 1990). At that level of precision, the naive comparison $\dot{P}_b^{\rm obs} \approx \dot{P}_b^{\rm GR}$ was no longer valid: kinematic and galactic effects contribute at the same fractional level (~0.7%) and must be subtracted before the residual can be interpreted as a test of gravitational-wave emission. Prior literature (Shapiro & Terzian 1976; Will 1976, 1981) had estimated these corrections to be negligible (~10⁻⁴ × the observed value), but that estimate was based on an incorrect (too short) distance to PSR 1913+16. This paper supersedes that estimate.
+
+## 2. Research gap
+
+Prior analyses used a nominal distance $d \simeq 5$ kpc to PSR 1913+16. Damour & Taylor 1991 show, using updated H i absorption measurements in the first galactic-longitude quadrant, that $d/R_0 = 1.08 \pm 0.16$ (i.e., $d \simeq 8.3$ kpc for $R_0 = 7.7$ kpc), nearly double the standard value. At this revised distance the galactic contribution $(\dot{P}_b/P_b)^{\rm gal}$ is ~0.7% of $(\dot{P}_b/P_b)^{\rm obs}$ — comparable to the observational precision — and cannot be neglected. The paper also corrects a factor-of-4 error in the mass-loss formula of Smarr & Blandford (1976).
+
+## 3. Core formulas
+
+### 3a. Decomposition of the observed orbital-period derivative (eq. 1.4, PDF p. 2)
+
+$$
+\left(\frac{\dot{P}_b}{P_b}\right)^{\!\rm obs}
+= \left(\frac{\dot{P}_b}{P_b}\right)^{\!\rm theor}
++ \left(\frac{\dot{P}_b}{P_b}\right)^{\!\rm gal}
++ \left(\frac{\dot{P}_b}{P_b}\right)^{\!\rm accel}
++ \left(\frac{\dot{P}_b}{P_b}\right)^{\!\dot{m}_1}
++ \left(\frac{\dot{P}_b}{P_b}\right)^{\!\dot{m}_2}
++ \cdots
+$$
+
+This is the master correction identity. After subtracting all non-GR terms the residual should equal $\dot{P}_b^{\rm GR}/P_b$.
+
+### 3b. Galactic (planar) + Shklovskii correction (eq. 2.8, PDF p. 2)
+
+$$
+\left(\frac{\dot{P}_b}{P_b}\right)^{\!\rm gal}
+= -\frac{v_0^2}{cR_0}
+\left\{
+  \cos l + \left(\frac{v_1}{v_0}\right)^{\!2}
+  \frac{(d/R_0) - \cos l}{1 - 2(d/R_0)\cos l + (d/R_0)^2}
+\right\}
++ \mu^2 \frac{d}{c}
+$$
+
+The last term $\mu^2 d/c$ is the **Shklovskii kinematic term** (always positive, since it arises from centripetal acceleration due to the pulsar's transverse velocity). The first brace is the **planar galactic acceleration** term.
+
+### 3c. General-relativistic prediction for $\dot{P}_b$ (eqs. 5.1 and 5.4, PDF p. 9)
+
+**Mass-dependent form (eq. 5.1):**
+$$
+\dot{P}_b^{\rm GR} = -\frac{192\pi}{5c^5}\,\nu(GM n)^{5/3}\,
+\frac{P_4(e_T)}{(1-e_T^2)^{7/2}}\,[1 + X_P^{\rm GR}]
+$$
+
+where $M \equiv m_1+m_2$, $\nu \equiv m_1 m_2/M^2$, $n \equiv 2\pi/P_b$, and the higher-order correction $X_P^{\rm GR}/c^2 \approx +2.15\times10^{-5}$ is negligible at 0.8% precision.
+
+**Observables-only form (eq. 5.4, PDF p. 9):**
+$$
+\dot{P}_b^{\rm GR} = -\frac{64\pi}{15\sqrt{3}}\,k^{5/2}
+\bigl[(1+4\eta)^{1/2} - 1 - \eta\bigr]
+\frac{P_4(e_T)}{1-e_T^2}
+$$
+
+with auxiliary combinations (eqs. 5.5–5.6, PDF p. 9):
+$$
+k \equiv \frac{\dot{\omega}\,P_b}{2\pi}
+\quad\text{(fractional periastron advance per orbit)},
+$$
+$$
+\eta \equiv \frac{3n\gamma}{k\,e_T(1-e_T^2)},
+$$
+$$
+P_4(e_T) \equiv 1 + \tfrac{73}{24}\,e_T^2 + \tfrac{37}{96}\,e_T^4 .
+$$
+
+**Parameter classification:**
+
+| Symbol | Role | Physical meaning | Units | Typical range / magnitude | Notes |
+|---|---|---|---|---|---|
+| $\dot{P}_b^{\rm GR}$ (= $\dot{P}_b^{\rm theor}$) | **Output / target** | Intrinsic GR orbital period derivative | dimensionless (s s⁻¹) | ~ −2.4 × 10⁻¹² for PSR 1913+16 | benchmark target PBDOT |
+| $\dot{P}_b^{\rm obs}$ | **Input variable** | Raw timing-measured orbital period derivative | dimensionless | ~ −87.4 × 10⁻¹⁸ s⁻¹ × $P_b$ | from timing fit |
+| $P_b$ | **Input variable** | Orbital period | s | ~8 h (PSR 1913+16: 27906.98 s) | well-measured timing param |
+| $e_T$ | **Input variable** | Relativistic (timing) eccentricity | dimensionless | 0–1; PSR 1913+16: 0.617 | Damour–Deruelle parameterization |
+| $\dot{\omega}$ | **Input variable** | Rate of periastron advance | deg yr⁻¹ | PSR 1913+16: 4.2266 deg yr⁻¹ | post-Keplerian timing param |
+| $\gamma$ | **Input variable** | Einstein time-dilation + gravitational redshift parameter | s | PSR 1913+16: 4.303 × 10⁻³ s | post-Keplerian timing param |
+| $m_1, m_2$ | **Per-system fit values** | Pulsar and companion masses | $M_\odot$ | ~1.4 $M_\odot$ each for NS–NS | derived from $\dot{\omega}$, $\gamma$ within GR |
+| $v_0$ | **Per-cluster observed covariate** | Galactic circular velocity at Sun | km s⁻¹ | ~220 km s⁻¹ ($\Theta_0$) | from galactic model; principal source of galactic uncertainty |
+| $R_0$ | **Per-cluster observed covariate** | Galactocentric distance of Sun | kpc | 7–8.5 kpc | uncertain at ~5% level |
+| $l$ | **Per-pulsar observed covariate** | Galactic longitude of pulsar | deg | PSR 1913+16: $l = 49.97°$ | from pulsar position |
+| $d$ | **Per-pulsar observed covariate** | Distance to pulsar | kpc | PSR 1913+16: $d \simeq 8.3$ kpc | from DM + H i; dominant uncertainty in galactic correction |
+| $v_1$ | **Per-pulsar observed covariate** | Galactic circular velocity at pulsar location | km s⁻¹ | ~220 km s⁻¹; enters via rotation-curve model | inferred from galactic model + $R_1$ |
+| $\mu$ | **Per-pulsar observed covariate** | Proper motion (angular) = $v_T/d$ | rad s⁻¹ | ~2 mas yr⁻¹ for PSR 1913+16 | drives Shklovskii term |
+| $c$ | **Fixed constant** | Speed of light | cm s⁻¹ | 2.998 × 10¹⁰ | universal |
+| $G$ | **Fixed constant** | Newton's gravitational constant | cgs | 6.674 × 10⁻⁸ | assumed constant in GR baseline |
+| $\frac{73}{24}, \frac{37}{96}$ | **Fixed constants** | Structural eccentricity-enhancement coefficients in $P_4(e_T)$ | dimensionless | — | from Peters & Mathews (1963); invariant across all binary pulsars |
+| $\frac{192\pi}{5}$, $\frac{64\pi}{15\sqrt{3}}$ | **Fixed constants** | GR numerical prefactors | dimensionless | — | intrinsic to GR quadrupole formula |
+
+**Validity domain:** derived for relativistic compact-object binaries (NS–NS, NS–WD) where the post-Newtonian expansion applies. The higher-order term $X_P^{\rm GR} \sim 2\times10^{-5}$ is negligible at 0.8% precision. The galactic correction formula assumes a nearly circular galactic orbit for the Sun and uses a linear rotation-curve model (eq. 2.9) for the pulsar.
+
+## 4. Variable / parameter consolidation table
+
+(Consolidated into §3 above for compactness; all symbols listed there with role, units, and magnitude.)
+
+## 5. Train / test split — qualitative
+
+This is a single-system analytical derivation, not a statistical fit to a dataset. There is no train/test split in the conventional sense. PSR 1913+16 is the **only** system analysed in detail. The observational inputs ($P_b$, $e_T$, $\dot{\omega}$, $\gamma$) come from Taylor & Weisberg (1989, 1990) timing solutions. The galactic distance $d$ is estimated from H i absorption (Frail 1989) combined with the DM + electron-density model of the first galactic quadrant. The paper presents a forward model (correction formula → residual → GR test), not a regression.
+
+For the ATNF benchmark: the formula is applied per-pulsar across the ATNF catalog; each binary pulsar constitutes a single data point; no cross-validation is described. The split protocol is entirely at the benchmark curation level, not within this paper.
+
+## 6. Validation criteria — Layer 2 rubric
+
+- The galactic correction $(\dot{P}_b/P_b)^{\rm gal}$ for PSR 1913+16 evaluates to $-0.596 \pm 0.188\,\alpha$ (Table 1, PDF p. 10; $\alpha = 10^{-18}$ s⁻¹), which is ~0.7% of the observed value $-87.39\,\alpha$; any candidate formula whose galactic correction deviates from this by more than a few per cent is inconsistent with the paper. (PDF p. 10, Table 1)
+
+- The Shklovskii term $\mu^2 d/c$ is **always positive** (centripetal acceleration is always directed away from the observer's line of sight); a formula that predicts a negative Shklovskii contribution violates the physics. (PDF p. 2, eq. 2.8)
+
+- After subtracting all non-GR corrections, the residual ratio $\dot{P}_b^{\rm obs-add}/\dot{P}_b^{\rm GR}$ should be consistent with unity; the paper establishes $1.0081 \pm 0.0022(\rm gal) \pm 0.0076(\rm obs)$ (eq. 6.6, PDF p. 10) as the validated benchmark value for PSR 1913+16, so any formula predicting the intrinsic PBDOT should recover a corrected-to-GR ratio within ~1%.
+
+- The mass-loss contribution to $\dot{P}_b/P_b$ is proportional to $-2\dot{m}_a/(m_1+m_2)$ per component (eq. 4.1, PDF p. 7, correcting the factor-of-4 error in Smarr & Blandford 1976); for PSR 1913+16 this evaluates to $+3.7 \times 10^{-6}\,\alpha$ (companion) and $+6.6 \times 10^{-8}\,\alpha$ (pulsar spin-down), always **positive** (mass loss lengthens the orbit). (PDF p. 10, Table 1)
+
+- The total non-GR additive correction $({\dot{P}_b}/{P_b})^{\rm add} = -0.60 \pm 0.19\,\alpha$ (eq. 6.3, PDF p. 10) is **negative** and dominated by the planar galactic acceleration; the vertical galactic term contributes only $-0.006\,\alpha$ and all stellar/GMC perturbations are $< 0.01\,\alpha$ — a formula that elevates these minor terms to significance violates the paper's hierarchy. (PDF p. 10, Table 1)
+
+## 7. Notes for benchmark integration
+
+- **Benchmark target**: `PBDOT` = $\dot{P}_b^{\rm GR}$ (intrinsic GR orbital period derivative), dimensionless (or in s s⁻¹).
+- **Baselines.py mapping**: implement eq. (5.4) as `PBDOT_GR(Pb, eT, omegadot, gamma)` using the $k$–$\eta$ auxiliary definitions (eqs. 5.5–5.6). The prefactor coefficients $64\pi/(15\sqrt{3})$ and $P_4$ coefficients $73/24$, $37/96$ are **invariant fixed constants** that must appear verbatim in any recovered formula.
+- **Setting / Type eligibility**: `setting1_typeI` (each pulsar is an independent system; no per-cluster covariate hierarchy beyond the galactic model constants). Potentially `setting2_typeI` if multiple pulsars are grouped by galactic environment (shared $v_0$, $R_0$), though the paper itself treats each pulsar individually.
+- **`invariant_constants`** (v0.5 §3.3 field): `73/24`, `37/96` (coefficients in $P_4(e_T)$); `192π/5`, `64π/(15√3)` (GR quadrupole prefactors); factor `2` in the mass-loss formula (eq. 4.1) — corrected from the erroneous `1/2` of Smarr & Blandford 1976.
+- **Sign convention**: $\dot{P}_b^{\rm GR}$ is negative for energy-losing (inspiralling) binaries. The ATNF catalog reports PBDOT as a signed dimensionless quantity (s s⁻¹); the galactic-corrected value should be more negative than the raw observed PBDOT.
+- **Unit convention**: the paper uses $\alpha \equiv 10^{-18}$ s⁻¹ as a convenience unit for $\dot{P}_b/P_b$; dimensional PBDOT (s s⁻¹) = $(\dot{P}_b/P_b) \times P_b$.
+- **Distance uncertainty is the limiting factor** for the galactic correction: the 15% uncertainty in $d$ propagates to a $\pm 0.19\,\alpha$ uncertainty in the total correction (dominated by the planar galactic term), which in turn limits the GR test to ~0.22% in galactic uncertainty alone. Any benchmark row for which $d$ is poorly constrained will carry large PBDOT correction uncertainties.

binary_pulsar_post_keplerian_atnf__PBDOT/reference/summary_formula_dataset_weisberg_2016.md

@@ -0,0 +1,133 @@
+**Type**: Formula+Dataset — primary source for the GR-predicted orbital period derivative (PBDOT) formula and the PSR B1913+16 timing measurements that anchor the `binary_pulsar_post_keplerian_atnf__PBDOT` benchmark target.
+**Source**: PDF at `reference/weisberg_2016.pdf`
+**arXiv / DOI / journal**: DOI 10.3847/0004-637X/829/1/55 — *The Astrophysical Journal*, 829:55 (10pp), 2016 September 20. Confirmed against `.bib` and PDF title page.
+**Pages read**: 1,4,5,6,7,8,9
+
+---
+
+## 1. Background and Landscape
+
+PSR B1913+16 (Hulse & Taylor 1975) was the first binary pulsar discovered, consisting of two neutron stars in a highly eccentric, tight orbit. It remains the prototypical laboratory for strong-field relativistic gravitation tests. This paper (hereafter WH16) presents analyses of 9257 times-of-arrival (TOAs) accumulated over 35 years (1981–2012) at Arecibo Observatory, extending the earlier Weisberg, Nice & Taylor (2010, WNT) dataset with post-2006 data and improved analysis procedures. The core test is the comparison of the observed (kinematically corrected) orbital period derivative to the general relativistic prediction from gravitational wave damping.
+
+## 2. Research Gap
+
+WNT (2010) reported a ~1.8σ discrepancy between measured and GR-predicted PBDOT. WH16 resolves this by (a) incorporating post-2006 data that more than double the baseline and (b) adopting improved galactic parameters from Reid et al. (2014), which substantially revise the galactic acceleration correction. WH16 also measures for the first time in this system the Shapiro gravitational propagation delay parameters and the relativistic orbital shape correction δθ.
+
+---
+
+## 3. Core Formulas
+
+### 3a. GR Prediction for Orbital Period Derivative (Eq. 22, PDF p. 8)
+
+$$\dot{P}_b^{\rm GR} = -\frac{192\pi}{5\,c^5}\,G^{5/3}\left(\frac{P_b}{2\pi}\right)^{-5/3}
+\left(1 + \frac{73}{24}e^2 + \frac{37}{96}e^4\right)(1-e^2)^{-7/2}
+\,m_1\,m_2\,(m_1+m_2)^{-1/3}$$
+
+Equivalently, in solar units (using $T_\odot = G M_\odot / c^3$):
+
+$$\dot{P}_b^{\rm GR} = -\frac{192\pi}{5}\,T_\odot^{5/3}\left(\frac{P_b}{2\pi}\right)^{-5/3}
+\left(1 + \frac{73}{24}e^2 + \frac{37}{96}e^4\right)(1-e^2)^{-7/2}
+\times\frac{m_1}{M_\odot}\frac{m_2}{M_\odot}\left(\frac{m_1+m_2}{M_\odot}\right)^{-1/3}$$
+
+**Origin**: Peters & Mathews (1963), as implemented in the GR timing framework of Damour & Deruelle (1986) and Damour & Taylor (1992). WH16 is the measurement paper; the formula itself is standard GR.
+
+**Parameter classification:**
+
+| Symbol | Role | Notes |
+|--------|------|-------|
+| 192π/5 | Fixed constant (universal) | Structural GR coefficient |
+| 73/24 | Fixed constant (universal) | Eccentricity enhancement coefficient |
+| 37/96 | Fixed constant (universal) | Eccentricity enhancement coefficient |
+| 7/2 in $(1-e^2)^{-7/2}$ | Fixed constant (universal) | GR exponent |
+| $T_\odot = G M_\odot/c^3 \approx 4.9255\times10^{-6}$ s | Fixed constant (universal) | Solar mass in time units |
+| $P_b$ | Input variable | Orbital period, observed per binary system |
+| $e$ | Input variable | Orbital eccentricity, observed per binary system |
+| $m_1$ | Per-cluster covariate | Pulsar mass, derived from ⟨ω̇⟩ and γ (post-Keplerian observables) — read once per system, not refit |
+| $m_2$ | Per-cluster covariate | Companion mass, derived from ⟨ω̇⟩ and γ — read once per system |
+| $\dot{P}_b^{\rm GR}$ | Output variable / target | GR-predicted orbital period derivative (PBDOT) |
+
+**Validity domain**: Applies to any relativistic binary system with $0 \le e < 1$ in the weak-emission approximation (Peters & Mathews 1963). For PSR B1913+16: $P_b \approx 0.323$ d, $e \approx 0.617$, $m_1 \approx 1.438\,M_\odot$, $m_2 \approx 1.390\,M_\odot$.
+
+### 3b. Galactic Kinematic Correction (Eq. 15, PDF p. 4)
+
+$$\dot{P}_b^{\rm intr} = \dot{P}_b^{\rm obs} - \dot{P}_b^{\rm gal}$$
+
+The galactic term $\dot{P}_b^{\rm gal}$ arises from relative line-of-sight accelerations of the solar and binary systems (Damour & Taylor 1991). For B1913+16: $\dot{P}_b^{\rm gal} = -(0.025 \pm 0.004)\times10^{-12}$ (PDF p. 5).
+
+### 3c. Post-Keplerian Mass Equations (Eqs. 16–17, PDF p. 5–6)
+
+Masses $m_1$, $m_2$ are determined by jointly solving:
+
+$$\langle\dot{\omega}\rangle = 3\,T_\odot^{2/3}\left(\frac{P_b}{2\pi}\right)^{-5/3}(1-e^2)^{-1}
+\left(\frac{m_1+m_2}{M_\odot}\right)^{2/3}$$
+
+$$\gamma = T_\odot^{2/3}\,e\left(\frac{P_b}{2\pi}\right)^{1/3}
+\frac{m_{2;\langle\dot\omega\rangle}}{M_\odot}\left(\frac{m_{1;\langle\dot\omega\rangle}+2m_{2;\langle\dot\omega\rangle}}{m_{1;\langle\dot\omega\rangle}+m_{2;\langle\dot\omega\rangle}}\right)^{4/3}$$
+
+---
+
+## 4. Variable / Parameter Consolidation Table
+
+| Symbol | Role | Physical meaning | Units | Typical range / magnitude | Notes |
+|--------|------|-----------------|-------|--------------------------|-------|
+| $\dot{P}_b^{\rm GR}$ | Output (target) | GR-predicted orbital period derivative | dimensionless (s/s) | $\sim -2.4\times10^{-12}$ for B1913+16 | Negative (orbit shrinks) |
+| $P_b$ | Input variable | Orbital period | days (or seconds) | ~0.1–100 d for relativistic binary pulsars | B1913+16: 0.32299745 d |
+| $e$ | Input variable | Orbital eccentricity | dimensionless | 0–1 (B1913+16: 0.617) | Higher $e$ enhances GW emission |
+| $m_1$ | Per-cluster covariate | Pulsar (neutron star) mass | $M_\odot$ | ~1.2–2.0 $M_\odot$ | B1913+16: $1.438 \pm 0.001\,M_\odot$ (PDF p. 6) |
+| $m_2$ | Per-cluster covariate | Companion (neutron star) mass | $M_\odot$ | ~1.2–2.0 $M_\odot$ | B1913+16: $1.390 \pm 0.001\,M_\odot$ (PDF p. 6) |
+| $T_\odot$ | Fixed constant | $G M_\odot / c^3$ | s | $\approx 4.9255\times10^{-6}$ s | Converts solar units |
+| 192π/5 | Fixed constant | GR leading coefficient | — | 120.531… | From Peters & Mathews 1963 |
+| 73/24 | Fixed constant | Eccentricity polynomial coefficient | — | 3.0417 | Inside $f(e)$ numerator |
+| 37/96 | Fixed constant | Eccentricity polynomial coefficient | — | 0.3854 | Inside $f(e)$ numerator |
+| $\dot{P}_b^{\rm gal}$ | Per-cluster covariate | Galactic kinematic correction | dimensionless | $\sim\!-0.025\times10^{-12}$ for B1913+16 | Requires distance, proper motion, galactic parameters |
+| $\dot{P}_b^{\rm obs}$ | Per-cluster covariate | Observed orbital period derivative (TEMPO fit) | dimensionless | $-2.423(1)\times10^{-12}$ for B1913+16 (Table 2, PDF p. 4) | Raw TEMPO output |
+
+---
+
+## 5. Train / Test Split — Qualitative
+
+This is a single-system timing paper, not a multi-system dataset paper. The 9257 TOAs for PSR B1913+16 span 1981–2012, all from Arecibo Observatory near 1400 MHz. WH16 uses TEMPO to fit the full TOA set simultaneously for all orbital and spin parameters. There is no explicit train/test split within this paper; the full TOA set is the calibration set. The derived orbital parameters (Table 2) serve as covariates in the benchmark. The paper does not define a held-out test set — the "test" is the comparison of derived $\dot{P}_b^{\rm intr}$ to the GR prediction (Eq. 22).
+
+For benchmark use: the ATNF catalog provides multi-system data; each binary pulsar system is one cluster/row. WH16 provides parameter values and the formula; it does not itself define a split across systems.
+
+---
+
+## 6. Validation Criteria — Layer 2 Rubric
+
+- **Sign constraint**: $\dot{P}_b^{\rm GR} < 0$ everywhere in the supported domain ($0 \le e < 1$, $m_1,m_2 > 0$) — the orbit must shrink as energy is radiated away via gravitational waves (PDF p. 8, Eq. 22, and Section 5.1 discussion).
+- **Eccentricity enhancement**: The factor $(1 + \frac{73}{24}e^2 + \frac{37}{96}e^4)(1-e^2)^{-7/2}$ must exceed 1 for any $e > 0$, diverging as $e \to 1$ — higher eccentricity strongly enhances gravitational wave emission (PDF p. 8, Eq. 22 structure; Peters 1964 as cited).
+- **GR agreement to $\lesssim0.2\%$**: For PSR B1913+16, the ratio $\dot{P}_b^{\rm intr}/\dot{P}_b^{\rm GR} = 0.9983 \pm 0.0016$ (PDF p. 8, Eq. 23) — a recovered formula should reproduce the GR value to within observational uncertainties.
+- **Galactic correction is small but non-negligible**: $|\dot{P}_b^{\rm gal}/\dot{P}_b^{\rm obs}| \approx 1\%$ for B1913+16; neglecting it produces a systematic bias (PDF p. 4–5, Eq. 15 and surrounding text).
+- **Limit $e \to 0$**: Formula reduces to the circular-orbit Peters expression $\dot{P}_b^{\rm GR}|_{e=0} = -(192\pi/5)\,T_\odot^{5/3}(P_b/2\pi)^{-5/3} m_1 m_2 (m_1+m_2)^{-1/3}$ (PDF p. 8, Eq. 22 — the eccentricity factor evaluates to 1 at $e=0$).
+
+---
+
+## 7. Dataset Content (for PSR B1913+16)
+
+| Field | Value | Source |
+|-------|-------|--------|
+| System | PSR B1913+16 | PDF p. 1 |
+| N TOAs | 9257 | PDF p. 1, Abstract |
+| Observing epoch | 1981–2012 | PDF p. 1, Section 2 |
+| Telescope | Arecibo Observatory | PDF p. 1, Section 2 |
+| Frequency | ~1400 MHz | PDF p. 1, Section 2 |
+| $P_b$ | 0.322997448918(3) d | Table 2, PDF p. 4 |
+| $e$ | 0.6171340(4) | Table 2, PDF p. 4 |
+| $\dot{P}_b^{\rm obs}$ | $-2.423(1)\times10^{-12}$ | Table 2, PDF p. 4 |
+| $\dot{P}_b^{\rm GR}$ | $(-2.40263 \pm 0.00005)\times10^{-12}$ | PDF p. 8, Eq. 22 evaluated |
+| $m_1$ | $1.438 \pm 0.001\,M_\odot$ | PDF p. 6 |
+| $m_2$ | $1.390 \pm 0.001\,M_\odot$ | PDF p. 6 |
+
+TEMPO input files and TOAs are published with the paper as a `.tar.gz` package (PDF p. 1). License: AAS/IOP open-access article; data files published with the article — **SPDX: CC-BY-4.0** (AAS standard for 2016 ApJ articles).
+
+---
+
+## 8. Notes for Benchmark Integration
+
+- **Target**: `PBDOT` = $\dot{P}_b^{\rm GR}$ as computed from Eq. (22) using inputs $(P_b, e, m_1, m_2)$.
+- **`baselines.py` mapping**: `f(X)` where `X[:,0]=Pb_days`, `X[:,1]=e`, `X[:,2]=m1_solar`, `X[:,3]=m2_solar`; output is $\dot{P}_b^{\rm GR}$ (dimensionless, negative).
+- **Unit convention**: $P_b$ in seconds when computing with SI $G$, $c$; or use the $T_\odot$ solar-unit form with $P_b$ in seconds. The ATNF catalog typically lists $P_b$ in days — convert to seconds by multiplying by 86400.
+- **Sign**: Output is always negative.
+- **Clipping**: Formula is singular at $e=1$; benchmark data should exclude $e \ge 1$ (not physically realizable for bound orbits).
+- **Setting/Type eligibility**: Each binary pulsar system is one row; inputs $(P_b, e)$ are observed Keplerian parameters, masses $(m_1, m_2)$ are derived per-system post-Keplerian covariates. This supports **`setting1_typeI`** (single-system regression) and **`setting2_typeI`** (cross-system regression with all four inputs observed per system). The masses are per-cluster covariates (not refit from raw data each time), making Type II (hold out masses) potentially applicable, but the paper does not define a secondary/holdout structure — flag for v0.5 §4 split-decision pipeline.
+- **Galactic correction**: The benchmark likely uses $\dot{P}_b^{\rm intr}$ (not $\dot{P}_b^{\rm obs}$) as the label, having already applied Eq. (15). Confirm with data pipeline.

binary_pulsar_post_keplerian_atnf__PBDOT/reference/summary_formula_peters_1963.md

@@ -0,0 +1,99 @@
+**Type**: Formula — foundational derivation of the gravitational-wave power radiated by an eccentric Keplerian binary; directly supplies the eccentricity enhancement factor $f(e)$ whose fixed coefficients ($73/24$, $37/96$) and denominator structure $(1-e^2)^{7/2}$ appear verbatim in the PBDOT formula for the `binary_pulsar_post_keplerian_atnf__PBDOT` benchmark target.
+**Source**: PDF at `reference/peters_1963.pdf`
+**arXiv / DOI / journal**: DOI 10.1103/PhysRev.131.435; *Physical Review*, Vol. 131, No. 1, pp. 435–440, 1 July 1963
+**Pages read**: 1,2,3,4,5
+
+---
+
+## 1. Background and landscape
+
+Peters & Mathews (1963) is the foundational calculation of gravitational radiation emitted by two point masses orbiting each other in a Keplerian ellipse under their mutual gravitational attraction. The work exploits the analogy between linearised general relativity and classical electromagnetism: just as accelerating charges radiate electromagnetic waves, accelerating masses radiate gravitational waves. Two independent methods are developed — an inertia-tensor method and a multipole expansion — and both yield identical results, providing mutual verification (§§I–III, PDF pp. 435–437). The paper covers arbitrary eccentricity but restricts to orbital velocities small compared with $c$ (post-Newtonian approximation). Results include the total radiated power (averaged and instantaneous), its angular distribution, and its polarisation properties.
+
+Prior to this paper, estimates of gravitational radiation existed for special cases (circular orbits), but no closed-form result for arbitrary eccentricity with the full Keplerian orbit kinematics had been published in this form.
+
+## 2. Research gap
+
+Circular-orbit estimates of gravitational-wave power existed in the literature but could not capture the strong eccentricity enhancement relevant to real astrophysical binaries. Highly eccentric orbits radiate orders of magnitude more power per unit time than circular orbits with the same semimajor axis (by factor $f(0.9)\sim 10^3$, Fig. 2, PDF p. 437). The paper fills this gap with an analytic, closed-form result valid for any eccentricity $0 \le e < 1$.
+
+## 3. Core formulas
+
+### 3a. Instantaneous power radiated (Eq. 15, PDF p. 436)
+
+$$
+P = \frac{8}{15}\frac{G^4 m_1^2 m_2^2(m_1+m_2)}{c^5 a^5(1-e^2)^5}
+(1+e\cos\psi)^4
+\bigl[12(1+e\cos\psi)^2 + e^2\sin^2\psi\bigr]
+$$
+
+where $\psi$ is the true anomaly (orbital phase from periastron).
+
+### 3b. Average power radiated over one orbital period (Eq. 16, PDF p. 437)
+
+$$
+\langle P \rangle = \frac{32}{5}\frac{G^4 m_1^2 m_2^2(m_1+m_2)}{c^5\,a^5(1-e^2)^{7/2}}
+\!\left(1+\frac{73}{24}e^2+\frac{37}{96}e^4\right)
+$$
+
+This is the principal result of the paper. The eccentricity polynomial in the numerator and the $(1-e^2)^{7/2}$ denominator together define the enhancement factor $f(e)$ (Eq. 17).
+
+### 3c. Eccentricity enhancement factor (Eq. 17, PDF p. 437)
+
+$$
+f(e) = \frac{1+\tfrac{73}{24}e^2+\tfrac{37}{96}e^4}{(1-e^2)^{7/2}}
+$$
+
+This function appears verbatim as $P_4(e_T)/(1-e_T^2)^{7/2}$ in the PBDOT formula (see Damour & Taylor 1991, eqs. 5.1 and 5.4). The fixed coefficients $73/24$ and $37/96$ are structural constants of the GR quadrupole formula, not fit parameters.
+
+**Parameter classification:**
+
+| Symbol | Role | Physical meaning | Units | Typical range / magnitude | Notes |
+|---|---|---|---|---|---|
+| $\langle P \rangle$ | **Output / target** | Orbit-averaged gravitational wave power | erg s⁻¹ (cgs) | depends on system; enters PBDOT via $dE/dt = -\langle P\rangle$ | intermediate target; PBDOT is derived from this |
+| $m_1$ | **Per-system fit value** | Mass of body 1 (e.g. pulsar) | g (cgs) / $M_\odot$ | ~1–2 $M_\odot$ for NS binaries | refit per binary system |
+| $m_2$ | **Per-system fit value** | Mass of body 2 (e.g. companion) | g (cgs) / $M_\odot$ | ~0.1–2 $M_\odot$ | refit per binary system |
+| $a$ | **Per-system fit value** | Semimajor axis of relative orbit | cm | $10^{10}$–$10^{12}$ cm for binary pulsars | related to $P_b$ via Kepler's third law |
+| $e$ | **Input variable** | Orbital eccentricity | dimensionless | $0 \le e < 1$ | measured from pulsar timing |
+| $\psi$ | **Input variable** | True anomaly (orbital phase) | rad | $[0, 2\pi)$ | enters only instantaneous Eq. (15); averages out in Eq. (16) |
+| $G$ | **Fixed constant** | Gravitational constant | cgs: $6.67\times10^{-8}$ dyn cm² g⁻² | — | universal |
+| $c$ | **Fixed constant** | Speed of light | $3\times10^{10}$ cm s⁻¹ | — | universal |
+| $\tfrac{32}{5}$ | **Fixed constant** | GR quadrupole prefactor | dimensionless | — | structural coefficient of GR; invariant |
+| $\tfrac{73}{24}$ | **Fixed constant** | Eccentricity-polynomial coefficient (order $e^2$) | dimensionless | — | invariant across all binaries; appears in PBDOT enhancement |
+| $\tfrac{37}{96}$ | **Fixed constant** | Eccentricity-polynomial coefficient (order $e^4$) | dimensionless | — | invariant across all binaries; appears in PBDOT enhancement |
+| $(1-e^2)^{7/2}$ | **Fixed structural form** | Denominator of enhancement factor | dimensionless | — | exact exponent $7/2$ is structural; must not be altered |
+| $f(e)$ | **Derived fixed function** | Enhancement factor over circular-orbit power | dimensionless | $f(0)=1$; $f(0.6)\sim10$; $f(0.8)\sim10^2$; $f(0.9)\sim10^3$ | rises steeply with $e$; see Fig. 2, PDF p. 437 |
+
+**Validity domain:** post-Newtonian regime ($v/c \ll 1$); arbitrary eccentricity $0 \le e < 1$; point-mass approximation (no finite-size, tidal, or spin-precession effects). The paper states this explicitly in the abstract (PDF p. 435): "assume orbital velocities are small."
+
+## 4. Variable / parameter consolidation table
+
+Consolidated into §3 above. Key point: $m_1$, $m_2$, $a$ are per-system values (refit each binary); $e$ is the observationally measured eccentricity; $G$, $c$, and the numerical coefficients $32/5$, $73/24$, $37/96$, and the exponent $7/2$ are universal fixed constants.
+
+## 5. Train / test split — qualitative
+
+This is a purely analytic theoretical derivation. No observational dataset is fitted or held out. The result is exact within the post-Newtonian, point-mass approximation; there is no calibration split. Verification is provided by the two independent methods (inertia tensor vs. multipole expansion) yielding identical answers, and by the self-consistency check that the sum over harmonics $\sum_{n=1}^{\infty} g(n,e) = f(e)$ proven in the Appendix (PDF pp. 439–440, Eq. A1–A3). For the ATNF benchmark, each binary pulsar constitutes a single data point; split protocol is at the benchmark curation level, not within this paper.
+
+## 6. Validation criteria — Layer 2 rubric
+
+- **Circular-orbit limit:** At $e=0$, $f(0)=1$ exactly, so $\langle P \rangle$ reduces to the circular-orbit power $\frac{32}{5}\frac{G^4 m_1^2 m_2^2(m_1+m_2)}{c^5 a^5}$; any candidate formula that does not satisfy $f(0)=1$ violates the derivation. (PDF p. 437, text following Eq. 16; "the average power equals the power radiated from a circular orbit of equal semimajor axis … times an enhancement factor")
+
+- **Monotonic increase with eccentricity:** $f(e)$ is a strictly increasing function of $e$ for $e \in [0,1)$; specifically $f(0.6)\sim 10$, $f(0.8)\sim 10^2$, $f(0.9)\sim 10^3$ (Fig. 2, PDF p. 437 and text: "the power radiated is a steeply rising function of the eccentricity"). A formula predicting lower power at higher eccentricity (for fixed $a$, $m_1$, $m_2$) is unphysical.
+
+- **Harmonic completeness:** The total power Eq. (16) must equal the sum over all harmonics $\sum_{n=1}^{\infty} P(n)$ where $P(n)$ is given by Eq. (19); i.e. $\sum_{n=1}^{\infty} g(n,e) = f(e)$. This is a self-consistency constraint verified in the Appendix (PDF pp. 439–440, Eqs. A1–A3). A recovered symbolic formula for $\langle P \rangle$ that differs from this sum violates energy conservation within the quadrupole approximation.
+
+- **Divergence at $e\to 1$:** $\langle P \rangle \to \infty$ as $e \to 1^-$ (the $(1-e^2)^{7/2}$ denominator vanishes), reflecting infinite power in the parabolic-orbit limit; any formula with a finite limit at $e=1$ is structurally wrong. (PDF p. 437, Fig. 2 shows $f(e) \to \infty$; text states "a steeply rising function of the eccentricity $e$")
+
+- **Positive-definite power:** $\langle P \rangle > 0$ for all $m_1, m_2 > 0$ and $0 \le e < 1$; the energy is always radiated away (never absorbed). The numerator $1 + \frac{73}{24}e^2 + \frac{37}{96}e^4 > 0$ for all real $e$. (PDF p. 437, Eq. 16 — all terms non-negative)
+
+## 7. Notes for benchmark integration
+
+- **Benchmark target PBDOT**: this paper does not write $\dot{P}_b$ explicitly. The PBDOT formula is obtained via energy balance $dE/dt = -\langle P \rangle$ combined with the Keplerian energy $E = -Gm_1m_2/(2a)$ and Kepler's third law $P_b^2 = 4\pi^2 a^3/[G(m_1+m_2)]$, yielding $\dot{P}_b = -(192\pi/5)(G/c^3)^{5/3}(P_b/2\pi)^{-5/3} m_1 m_2/(m_1+m_2)^{1/3} \times f(e)$. The $f(e)$ function is Eq. (17) of this paper. This derivation step is completed in Peters (1964), Phys. Rev. 136, B1224.
+
+- **`invariant_constants`** (v0.5 §3.3): `73/24`, `37/96` (numerator polynomial coefficients in $f(e)$); `7/2` (exponent in denominator $(1-e^2)^{7/2}$); `32/5` (overall GR prefactor). These must appear verbatim in any correctly recovered PBDOT formula.
+
+- **baselines.py mapping**: implement $f(e)$ as `f_e = (1 + 73/24*e**2 + 37/96*e**4) / (1 - e**2)**(7/2)` and use it to scale the circular-orbit power or feed directly into the PBDOT expression.
+
+- **Setting / Type eligibility**: `setting1_typeI` (per-pulsar; each binary system is independent); coefficients $73/24$, $37/96$, $7/2$ are the invariant Type II constants if the benchmark groups multiple pulsars per cluster.
+
+- **Sign convention**: $\langle P \rangle > 0$ (energy emitted); the orbital energy $E < 0$ decreases with time, so $a$ shrinks and $P_b$ decreases ($\dot{P}_b < 0$). The PBDOT benchmark target is a negative number for inspiralling systems.
+
+- **Page 6 of the PDF** (journal p. 440 bottom onward) is a completely separate paper ("Quasiparticles and the Born Series" by Weinberg, same journal issue) and is not part of this Peters & Mathews paper.

binary_pulsar_post_keplerian_atnf__PBDOT/reference/summary_formula_will_2014.md

@@ -0,0 +1,73 @@
+**Type**: Formula + Supporting — canonical GR formula source for the post-Keplerian orbital period derivative `PBDOT` (Ṗ_b); contextual review of binary-pulsar tests of general relativity.
+**Source**: PDF at `reference/will_2014.pdf`
+**arXiv / DOI / journal**: doi:10.12942/lrr-2014-4 — *Living Reviews in Relativity* 17 (2014), 4 — confirmed against `will_2014.bib` and PDF title page.
+**Pages read**: 1, 62, 68, 69, 70, 71
+
+---
+
+## 1. Background and landscape
+
+Will (2014) is a comprehensive living review of experimental tests of general relativity, updated from the 2006 edition. Section 6 ("Stellar System Tests of Gravitational Theory") covers binary-pulsar tests. Section 6.1 focuses on the Hulse–Taylor pulsar B1913+16, which provided the first empirical evidence for gravitational-wave energy loss through the secular decay of the orbital period. The paper situates the orbital-period-derivative formula within the post-Newtonian (PN) hierarchy: it is a 2.5PN (order ε^{5/2}) correction to the orbital dynamics. The ATNF online pulsar catalogue (reference [28] in the paper) is flagged as the authoritative per-system parameter source.
+
+## 2. Research gap
+
+Prior to the Hulse–Taylor discovery, the quadrupole formula for gravitational radiation had not been verified against a dynamical, self-gravitating system. The paper contextualises Ṗ_b as the critical observable that closed this gap, and examines whether alternative gravity theories with dipole radiation terms can equally explain the measured orbital decay.
+
+## 3. Core formula
+
+### Eq. (108), PDF p. 69 — GR post-Keplerian orbital period derivative
+
+$$\dot{P}_b = -\frac{192\pi}{5}(2\pi\mathcal{M}f_b)^{5/3} F(e)$$
+
+### Eq. (88), PDF p. 62 — eccentricity enhancement factor
+
+$$F(e) = (1 - e^2)^{-7/2}\!\left(1 + \frac{73}{24}e^2 + \frac{37}{96}e^4\right)$$
+
+### Chirp mass definition (PDF p. 62, text after eq. 88)
+
+$$\mathcal{M} = \eta^{3/5}\,m, \qquad \eta = \frac{m_1 m_2}{m^2}, \qquad m = m_1 + m_2$$
+
+### Parameter classification
+
+| Symbol | Role | Physical meaning | Units | Typical range / magnitude | Notes |
+|---|---|---|---|---|---|
+| $\dot{P}_b$ | **Output / target** | Rate of change of orbital period (PBDOT benchmark target) | dimensionless (s s⁻¹) | ~−10⁻¹² for tight NS binaries | Negative: period shrinks due to GW energy loss |
+| $P_b$ | **Input variable** | Orbital period | days | 0.1–few days for relativistic NS binaries | Keplerian parameter; $f_b = 1/P_b$ |
+| $e$ | **Input variable** | Orbital eccentricity | dimensionless | 0–0.9; B1913+16 has e = 0.617 | Keplerian parameter |
+| $m_1$ | **Input variable** | Pulsar (primary) mass | $M_\odot$ | ~1.2–1.5 $M_\odot$ for NS | Per-system, inferred from PK constraints |
+| $m_2$ | **Input variable** | Companion mass | $M_\odot$ | ~1.2–1.5 $M_\odot$ for NS companions | Per-system, inferred from PK constraints |
+| $\mathcal{M}$ | **Derived** | Chirp mass $= \eta^{3/5}m$ | $M_\odot$ | ~1.2 $M_\odot$ for B1913+16 | Not an independent parameter; computed from $m_1, m_2$ |
+| $f_b$ | **Derived** | Orbital frequency $= 1/P_b$ | Hz | ~3.5 × 10⁻⁵ Hz for B1913+16 | Derived from $P_b$ |
+| $\frac{192\pi}{5}$ | **Fixed constant** | Quadrupole-formula prefactor | — | 120.52… | Universal GR structural constant; never refit |
+| $\frac{73}{24}$ | **Fixed constant** | Eccentricity coefficient in $F(e)$, $e^2$ term | — | ≈ 3.042 | Structural coefficient in Peters–Mathews formula |
+| $\frac{37}{96}$ | **Fixed constant** | Eccentricity coefficient in $F(e)$, $e^4$ term | — | ≈ 0.385 | Structural coefficient in Peters–Mathews formula |
+| $\frac{7}{2}$ | **Fixed constant** | Exponent in $(1-e^2)^{-7/2}$ factor of $F(e)$ | — | 3.5 | Structural exponent |
+| $\frac{5}{3}$ | **Fixed constant** | PN exponent on $(2\pi\mathcal{M}f_b)$ | — | 1.667 | Marks 2.5PN order |
+
+### Validity domain
+
+The formula applies to compact binary systems (neutron stars or black holes) in the weak-field exterior regime, at orbital separations large compared to the bodies' radii, with tidal effects negligible. It covers the full eccentricity range $0 \le e < 1$. It excludes dipole radiation (relevant only when bodies have unequal internal gravitational binding energies under non-GR theories).
+
+## 4. Variable / parameter consolidation table
+
+*(See §3 table above — all symbols covered there.)*
+
+## 5. Train / test split — qualitative
+
+Will (2014) is a review paper; it does not define a train/test split for the benchmark. The system parameters used to illustrate the formula (Table 7, PDF p. 68) are drawn from Weisberg, Nice & Taylor (2010) [ref 409 in the paper] based on B1913+16 timing data through 2006. For the `binary_pulsar_post_keplerian_atnf__PBDOT` benchmark the rows come from the ATNF pulsar catalogue (ref [28]), not from this paper directly. Split design follows the v0.5 §4 pipeline applied to the ATNF catalogue.
+
+## 6. Validation criteria — Layer 2 rubric
+
+- **Ṗ_b must be strictly negative** for gravitationally bound binaries emitting quadrupole radiation: the leading minus sign in eq. (108) is structural (PDF p. 69, eq. 108; p. 62, eq. 88 context).
+- **F(e) → 1 in the circular-orbit limit (e → 0)**: $(1-0)^{-7/2}(1 + 0 + 0) = 1$, so eq. (108) reduces to the circular-orbit Peters formula; the eccentricity enhancement is monotonically increasing with |e| (PDF p. 62, eq. 88).
+- **GR quadrupole formula agrees with observation to better than 0.5%**: the corrected ratio $\dot{P}_b^{\rm corr}/\dot{P}_b^{\rm GR} = 0.997 \pm 0.002$ for B1913+16 (PDF p. 70–71, eq. 109); candidate formulae should respect this calibration benchmark.
+- **Ṗ_b is an O(ε^{5/2}) = 2.5PN correction** and therefore orders-of-magnitude smaller than the orbital period itself; $|\dot{P}_b| \sim 10^{-12}$ for the Hulse–Taylor pulsar with $P_b \approx 0.323$ days (PDF p. 68, Table 7; p. 69 text).
+- **The formula includes only quadrupole radiation** and explicitly excludes tidal dissipation and higher-multipole terms; any systematic residual at the ~0.5% level in tight NS binaries is attributable to these omitted effects, not formula error (PDF p. 69, text below eq. 108).
+
+## 7. Notes for benchmark integration
+
+- **baselines.py mapping**: implement as `PBDOT_will2014(X)` where `X` columns are `[m1_Msun, m2_Msun, Pb_days, e]` (output-column-0 convention). Compute $f_b = 1/(P_b \times 86400)$ Hz, $m = m_1 + m_2$ in solar-mass geometric units ($G=c=1$, $M_\odot = 4.925 \times 10^{-6}$ s), $\eta = m_1 m_2/m^2$, $\mathcal{M} = \eta^{3/5} m$, then apply eq. (108).
+- **Units**: Ṗ_b is dimensionless (s/s); in ATNF catalogue it is often tabulated as $\times 10^{-12}$. Confirm sign convention (negative = decaying orbit).
+- **Setting / Type eligibility**: each row corresponds to one pulsar system (one "cluster") with its own $(m_1, m_2, P_b, e)$; there are no free refit parameters — all inputs are observational covariates. This is therefore `setting1_typeI` (formula is universal; no per-cluster fit parameters) and `setting2_typeI` (multi-system generalisation; cross-cluster constants $\frac{192\pi}{5}$, $\frac{73}{24}$, $\frac{37}{96}$ must remain invariant).
+- **Invariant constants** (v0.5 `invariant_constants` field): `192π/5`, `73/24`, `37/96`, exponent `5/3`, exponent `7/2`.
+- **License**: CC BY-NC 3.0 Germany (confirmed from PDF footer, p. 1). This covers the paper text. The ATNF catalogue data licence is governed by ATNF separately.

binary_pulsar_post_keplerian_atnf__PBDOT/reference/summary_supporting_hulse_1975.md

@@ -0,0 +1,40 @@
+**Type**: Supporting — discovery paper for PSR B1913+16; establishes the binary system whose orbital decay (PBDOT) is the benchmark target; contains no PBDOT formula or measurement.
+**Source**: PDF at `reference/hulse_1975.pdf`
+**arXiv / DOI / journal**: DOI 10.1086/181708 — *The Astrophysical Journal*, 195, L51–L53, 1975 January 15. Confirmed against `hulse_1975.bib` and PDF title page.
+**Pages read**: 1-3
+
+---
+
+## Summary
+
+Hulse & Taylor (1975) announce the discovery of PSR 1913+16 (later renamed PSR B1913+16), the first pulsar found in a binary system, detected during a systematic pulsar survey at Arecibo. The 59-ms pulsar shows periodic Doppler-shift variations over a 0.3230-day orbit, establishing it as a "single-line spectroscopic binary." Approximately 200 independent 5-minute observations over 10 days yielded a well-sampled velocity curve (Fig. 1) fitted by the four standard spectroscopic-binary equations (eqs. 1–4, p. 2).
+
+**Orbital elements from Table 2 (p. 1):**
+- Radial velocity semi-amplitude: K₁ = 199 ± 5 km s⁻¹
+- Binary orbital period: P_b = 27908 ± 7 s (≈ 0.3230 days)
+- Eccentricity: e = 0.615 ± 0.010
+- Longitude of periastron: ω = 179° ± 1°
+- Time of periastron passage: T = JD 2,442,321.433 ± 0.002
+- Projected semimajor axis: a₁ sin i = 1.00 ± 0.02 R☉
+- Mass function: f(m) = 0.13 ± 0.01 M☉
+
+**Pulsar parameters from Table 1 (p. 1):**
+- Center-of-mass pulsar period: P_cm = 0.059030 ± 0.000001 s
+- Spin period derivative (upper limit): dP_cm/dt < 1 × 10⁻¹²
+- Dispersion measure: DM = 167 ± 5 cm⁻³ pc
+- Flux density at 430 MHz: S₄₃₀ = 0.006 ± 0.003 Jy
+- Effective pulse width: W_e < 10 ms
+
+The companion mass is constrained by Table 3 (p. 2) for a grid of inclinations and pulsar masses M₁ = 0.3, 1.0, 1.5 M☉; eclipse non-detection sets an upper limit on companion radius R₂ via eq. (5). The companion is inferred to be a compact object (neutron star or black hole; white dwarf not excluded but disfavored on evolutionary grounds).
+
+Section IV (p. 3) predicts that future high-accuracy timing will yield: (a) general-relativistic advance of periastron ~4° per year; (b) measurable relativistic Doppler shift and gravitational redshift (changes in observed period of several parts in 10⁶ during orbit); (c) eventual measurement of orbital period decay due to gravitational-wave energy loss (PBDOT). These predictions frame the scientific motivation for the PBDOT benchmark target but are not realized in this paper.
+
+## Why it appears in this source group
+
+This is the foundational discovery paper for PSR B1913+16; every subsequent post-Keplerian measurement (including PBDOT) references this work for the system identification and initial orbital elements.
+
+## Notes
+
+- **No PBDOT formula or measurement appears in this paper.** The paper only reports an upper limit on the *spin* period derivative dP_cm/dt, not the orbital period derivative. PBDOT formulas and observational fits are in Peters (1963), Taylor & Weisberg (1982), and subsequent timing papers.
+- The orbital elements in Table 2 (P_b, e, ω, a₁ sin i, f(m)) feed directly into later PBDOT calculations — they are reproduced in several of the other reference papers.
+- No Layer-2 rubric criteria are extracted (Supporting paper; the paper makes no quantitative claims about PBDOT).

binary_pulsar_post_keplerian_atnf__PBDOT/reference/weisberg_2016.bib

@@ -0,0 +1,10 @@
+@ARTICLE{WeisbergHuang2016,
+  author = {{Weisberg}, J. M. and {Huang}, Y.},
+  title = {{Relativistic Measurements from Timing the Binary Pulsar PSR B1913+16}},
+  journal = {ApJ},
+  year = 2016,
+  volume = {829},
+  pages = {55},
+  doi = {10.3847/0004-637X/829/1/55},
+  note = {Updated PSR B1913+16 timing: PBDOT_obs vs GR prediction agreement to 0.13% after Damour-Taylor galactic corrections.},
+}

binary_pulsar_post_keplerian_atnf__PBDOT/reference/weisberg_2016.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d15c9e4eda175a815f8624dc60866be7effb9ee0a5ced22563d09fde5cedbca8
+size 706005

binary_pulsar_post_keplerian_atnf__PBDOT/reference/weisberg_2016.txt

@@ -0,0 +1,1369 @@
+
+=== PAGE 1 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+doi:10.3847/0004-637X/829/1/55
+
+© 2016. The American Astronomical Society. All rights reserved.
+
+RELATIVISTIC MEASUREMENTS FROM TIMING THE BINARY PULSAR PSR B1913+16
+J. M. Weisberg and Y. Huang
+Department of Physics and Astronomy, Carleton College, Northfield, MN 55057, USA; jweisber@carleton.edu
+Received 2016 January 19; revised 2016 April 20; accepted 2016 June 1; published 2016 September 21
+
+ABSTRACT
+We present relativistic analyses of 9257 measurements of times-of-arrival from the first binary pulsar, PSR
+B1913+16, acquired over the last 35 years. The determination of the “Keplerian” orbital elements plus two
+relativistic terms completely characterizes the binary system, aside from an unknown rotation about the line of
+sight,leading to a determination of the masses of the pulsar and its companion: 1.438±0.001 M☉ and
+1.390±0.001 M☉, respectively. In addition, the complete system characterization allows forthe creation of
+relativistic gravitation test by comparing measured and predicted sizes of various relativistic phenomena. We find
+that the ratio of the observed orbital period decrease caused by gravitational wave damping (corrected by a
+kinematic term) to the general relativistic predictionis 0.9983±0.0016,thereby confirms the existence and
+strength of gravitational radiation as predicted by general relativity. For the first time in this system, we have also
+successfully measured the two parameters characterizing the Shapiro gravitational propagation delay, and found
+that their values are consistent with general relativistic predictions. For the first time in any system, we have also
+measured the relativistic shape correction to the elliptical orbit, δθ, although its intrinsic value is obscured by
+currently unquantified pulsar emission beam aberration. We have also marginally measured the time derivative of
+the projected semimajor axis, which, when improved in combination with beam aberration modeling from geodetic
+precession observations, should ultimately constrain the pulsar’s moment of inertia.
+Key words: binaries: close – gravitation – gravitational waves – pulsars: individual (PSR B1913+16)
+Supporting material: .tar.gz files
+1. INTRODUCTION
+
+the various observing systems and the number of TOAs from
+each through epoch 2006 are tabulated in WNT; WAPP
+spectrometer observations since then have added another 1652
+TOAs to the total, each acquired by three WAPPs deployed
+simultaneously at approximately contiguous 100 MHz bands
+near 1400 MHz.
+Geodetic precession of the pulsar spin axis has induced pulse
+profile changes (Weisberg et al. 1989; Kramer 1998; Weisberg
+& Taylor 2002; Clifton & Weisberg 2008) that have lately
+grown increasingly larger, presumably as our line of sight
+approaches the edge of the pulsar beam. Nevertheless, for
+purposes of uniformitywe use only a single profile template
+while finding TOAs for all WAPP data. This procedure induces
+time offsets into our TOA dataset between different sessions
+and frequencies, which have grown to a level where they
+should be compensated for. To do so, we adopted the following
+process. First, we formed a pulse profile at each frequency band
+for each two-week session. Each resulting “session-band”
+standard profile has a much greater signal-to-noise ratio than
+does a single five-minute integration, while still being short
+enough to avoid the secular changes we are trying to measure.
+Next, we measured the offset of the midpoint of this sessionband standard profile with respect to the grand standard profile.
+(The midpoint is assumed to correspond to a fixed longitude on
+the pulsar regardless of profile shifts).
+Then we fitted out a “primary” linear model of the profile
+offsets as a function of time at each band. In this fashion, we
+provided an empirically determined, model independent, firstorder TOA correction that accounts for the effects of profile
+changes, thereby minimizing the long-term effects of profile
+shifts that might be mistaken as the signature of other
+phenomena. We next fitted the timing model to all such
+“primarily offsetted” TOAs and chose the resulting dispersion
+measure as our nominal value. Finally, secondary offsets were
+
+Pulsar B1913+16 was the first binary pulsar discovered
+(Hulse & Taylor 1975). The system consists of two neutron
+stars (one is an observed pulsar) orbiting in a very tight, highly
+eccentric orbit, and it remains one of the best for studying
+relativistic gravitation (Weisberg & Taylor 1981; Taylor &
+Weisberg 1982, 1989; Weisberg et al. 2010, hereafter WNT).
+In this paperwe update WNT with the addition of post-2006
+data and with further relativistic timing analyses. The addition
+of significant quantities of data acquired with modern dataacquisition devices has enabled us to measure several
+additional relativistic phenomena for the first time in this
+system, while also refining previously measured ones. Among
+the parameters newly measured with various degrees of
+accuracy are the Shapiro gravitational propagation delay, a
+relativistic correction to the elliptical orbital shape, and the time
+derivative of projected pulsar semimajor axis. All of the the
+data used in this study are published with this paper in a .tar.gz
+package and in two online repositories, while our analysis
+software is published on sourceforge.
+We describe the observations used in this work in Section 2,
+while Section 3 delineates the scope and methods of our
+relativistic analyses of these data. The results of our fits to the
+data are explained in Section 4, and their applications for tests of
+relativistic gravitation are discussed in Section 5. We conclude in
+Section 6 by summarizing our work and placing it in the context
+of results from other relativistic binary pulsar systems.
+2. DATA
+The data for our analyses consists of 9257 pulse times-ofarrival (TOAs) derived from five-minute integrations of the
+pulsar signal at frequencies near 1400 MHz measured at
+Arecibo Observatory from 1981 to 2012. The parameters of
+1
+
+=== PAGE 2 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+determined through single session fits, where the dispersion
+measure was fixed at its nominal value and the residual offset of
+each band was fitted for and then removed. This process ensures
+that the infinite-frequency TOAs calculated from each band and
+session are self-consistently de-dispersed1 and offsetted.
+To verify that the above profile variation correction process
+does not contaminate our parameter measurements, we also
+employed an alternate approach, fitting for an offset for each
+band in each session simultaneously with all other parameters
+(Demorest et al. 2013). This procedure yields parameters that
+agree with our method to within 1σ for all parameters,
+suggesting that our measurements are robust with respect to the
+methods used to remove profile-shift-induced timing offsets.
+
+h3, in the alternate FW parametrization; (ii) the orbital elliptical
+shape correction parameter d qobs (to our knowledge never
+previously fitted for in any binary system), which appears
+in DT92ʼs full expression for Equation (1)ʼs Roemer term3; and
+(iii) x˙ obs and e˙obs, the observed time derivatives of x and e. All of
+the new parameters are discussed in greater detail in Sections 3.1–
+3.3, while the fit results for both old and new parameters are
+described in Section 4, and relativistic tests resulting from these
+measurements are discussed in Section 5. A set of TEMPO input
+files containing input parameters and the TOAs is available in a
+.tar.gz package published with this paper, in a persistent
+repository http://dx.doi.org/10.5281/zenodo.54764,and with
+the article preprint http://arxiv.org/e-print/1606.02744v1.
+
+3. RELATIVISTIC ANALYSIS OF TOAS
+
+3.1. Fitting for the Shapiro Gravitational Propagation Delay
+via Two Different Parametrizations
+
+Using an augmented version of the TEMPO software
+program, we fitted the relativistic timing model of Damour &
+Deruelle (1989, hereafter DD),or, in certain cases, the DD
+model augmented by the Freire & Wex (2010, hereafter FW)
+Shapiro parametrization (see Section 3.1),2 to our TOAs. In
+these models, the pulsar signal encounters several distinct types
+of delays on its journey from the orbiting pulsar to the solar
+system barycenter, such that the infinite-frequency pulse arrival
+time at the solar system barycenter, tssbc, is given by
+tssbc = D-1 [T + DRoemer (T ) + DEinstein (T )
++ DShapiro (T ) + DAberration (T )] ,
+
+Until this work, we were unable to explicitly measure the
+two Shapiro gravitational propagation delay terms that
+characterize Equation (1)ʼs ΔShapiro(T), because of their
+relatively small timing signature and their covariance. The
+two terms are identified as s and r in DD,while FW recently
+developed an alternate parametrization of the phenomenon
+wherein their two fitted parameters, ς and h3, are orthonormal.
+Our software implementation of their parametrization for higheccentricity pulsars is available in our augmented version of
+TEMPO (see footnote 2).
+The measurements of either pair of Shapiro parameters, (s, r)
+or (V , h3), can be utilized in either of two different manners, as
+described below.
+First, if general relativity is assumed to be the correct theory
+of gravitation, then either pair of Shapiro measurables can be
+utilized as independent constraints on the orbital inclination
+and binary companion mass. We summarize the theory here,
+and then apply it in Section 4.2.
+In the DD formulation, s and r translate directly into sin i and
+m2 (the companion mass), respectively
+
+(1 )
+
+where each delay is a function of the pulsar proper time of
+pulse emission, T, and whose details depend on a number of
+physical parameters (The Doppler factor, D, accounts for the
+relative motion of the solar system and binary system
+barycenters) The various terms in Equation (1) are detailed in
+DD and Damour & Taylor (1992, hereafter DT92), and we will
+comment further on the last two terms of Equation (1) in the
+following two sections.
+Among the fitted parameters, we determined improved
+values of the pulsar spin and orbital parameters that were
+published in WNT, plus a number of new ones. For the first
+time, we have successfully fitted for the Shapiro (1964)
+gravitational propagation delay while also placing constraints
+on two additional ones: a relativistic correction to the quasielliptical shape of the orbit, and the shrinkage rate of its
+projected semimajor axis, as described in further detail below.
+The improved, previously fitted parameters include the
+pulsar spin frequency and derivative(s) f, f˙ ,... ; five
+“Keplerian” orbital elements defined as projected pulsar
+semimajor axis x º a1 sin i where i is the orbital inclination;
+orbital period,Pb;eccentricity, e;reference epoch, T0;the
+reference epoch’s longitude of periastron, ω0; relativistic
+“post-Keplerian” parameters defined as themean rate of
+periastron advance áw˙ ñ; gravitational redshift-time dilation term
+γ;and orbital period derivative P˙b .
+The newly fitted post-Keplerian parameters include the
+following: (i) two Shapiro delay terms called shape (s) and range
+(r) in the DD parametrization, or two (different) quantities, ς and
+
+sin i = s ,
+
+(2 )
+
+⎛r ⎞
+⎛ c3 ⎞
+m 2 = ⎜ ⎟ r = ⎜ ⎟ M☉,
+⎝G ⎠
+⎝ T☉ ⎠
+
+(3 )
+
+and
+
+with c the speed of light, G as the Newtonian gravitational
+constant, T☉=G M☉/c3=4.925,490,947 μs.
+The alternate FW parametrization of the Shapiro delay gives
+sin i =
+
+2V
+,
+V2 + 1
+
+(4 )
+
+while m2 is a combination of the two measurables (h3, ς)
+⎛ h3 V 3 ⎞
+⎛ c 3 ⎞ h3
+m2 = ⎜
+⎟ 3 =⎜
+⎟ M☉.
+⎝2 G ⎠ V
+⎝ T☉ ⎠
+
+(5 )
+
+Alternatively, each measured parameter of the Shapiro pair
+can be considered to be an independent test of relativistic
+gravitation. We apply this procedure in Section 5.2.
+
+1
+
+This procedure also absorbs TOA variations induced by DM fluctuations at
+the levels and timescales expected from studies of millisecond pulsars (You
+et al. 2007).
+2
+See http://sourceforge.net/projects/TEMPO/for our augmented version
+of TEMPO, which contains our fitting routine for the FW Shapiro parameters in
+high-eccentricity binaries.
+
+3
+
+Before it can be utilized for tests of relativity, the d qobs parameter must be
+corrected for a comparable aberration term which is currently undeterminable
+(see Section 4.4).
+
+2
+
+=== PAGE 3 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+value d qintr
+
+3.2. Determination of the Relativistic Orbital Shape Correction
+δθ in the Presence of the Aberration Delay
+
+quantity x. The corrected
+additional test of gravitation theory.
+
+In order to successfully measure the intrinsic value of δθ,
+which nominally quantifies a relativistic correction to the shape
+of the approximately elliptical orbit in Equation (1)ʼs Roemer
+delay expression, one must compensate the observed value for
+the influence of a phenomenon that comparably affects TOAs,
+namely the orbital-phase dependent aberration of the pulsar
+beam as described by DD and DT92. Those authors provide a
+prescription for calculating and eliminating the confounding
+aberration signature from the observed value of δθif the
+aberration geometry is known. In principle, the necessary
+information can be gleaned from studies of profile changes
+resulting from geodetic precession of the pulsar spin axis
+(Weisberg & Taylor 2002; Clifton & Weisberg 2008). In this
+section, we summarize the theoretical expressions required to
+quantify δθ and aberration, and we will apply our observations
+to these results in Section 4.4.
+The time delay ΔAberration in Equation (1),resulting from
+aberration of the rotating pulsar beam,is dependent on the
+time-variable transverse component of the pulsar’s orbital
+velocity. DD and DT92 parametrize the instantaneous delay via
+the aberration parameters A(t) and B(t)
+DAberration = A (t ){sin (w + A e (u)) + e sin w}
++ B (t ){cos (w + A e (u)) + e cos w},
+
+3.3. Other Parameters Affected by the Aberration Delay
+In addition to affecting δθ, DT92 show that aberration also
+affects the observed x and e values. However, the fractional
+corrections to x and e are tiny. More interesting is the effect of
+geodetic spin axis precession on the time-derivatives of these
+parameters, because DT92 show that they are potentially
+measurable. The precessional motion will cause the aberration
+geometry to change, resulting in secular changes to A(t) and
+hence to òA on precession timescales
+d A 1 dA (t )
+f -1
+1
+=
+=dt
+x dt
+Pb sin i (1 - e2)1 2
+´
+
+⎛ e˙ ⎞obs
+⎛ e˙ ⎞GW
+d
+⎜ ⎟
+= A +⎜ ⎟ ,
+⎝e⎠
+⎝e⎠
+dt
+
+(6 )
+
+f -1
+x
+sin h
+,
+2
+1
+2
+Pb sin i (1 - e )
+sin l
+
+(7 )
+
+B (t ) = -
+
+f -1
+x
+cos i cos h
+,
+2
+1
+2
+Pb sin i (1 - e )
+sin l
+
+(8 )
+
+⎛ a˙ ⎞GW ⎛
+⎛ x˙ ⎞obs d A
+di ⎞SO
+⎜ ⎟
+=
++ ⎜ 1⎟
++ ⎜cot i ⎟
+⎝x⎠
+⎝
+⎝ a1 ⎠
+dt
+dt ⎠
+- m cot i sin (Qm - W) -
+
+A (t )
+.
+x
+
+D˙
+,
+D
+
+(13)
+
+where “SO” refers to spin–orbit coupling. (See Lorimer &
+Kramer 2004) for an expression that includes additional terms
+needed for some other binary pulsars.)
+The quantity di/dt in the third term of Equation (13),
+resulting from pulsar spin–orbit coupling, is developed here,
+from expressions in DT92:
+di
+G⎧
+3 m2 ⎫
+S1
+⎬ 3
+= 2 ⎨2 +
+sin l cos h
+⎩
+⎭
+dt c
+2 m1 aR (1 - e2)3 2
+1 ⎧
+3 m 2 ⎫ I1 (2pf )(2p Pb )2
+⎬
+= 2 ⎨2 +
+sin l cos h ,
+c ⎩
+2 m1 ⎭ (m1 + m 2)(1 - e2)3 2
+(14)
+
+where m1 is the pulsar mass, aR is the semimajor axis of the
+relative orbit, S1 = I1 (2pf ) is the magnitude of the pulsar spin
+angular momentum, and I1 is its moment of inertia. The fourth
+term of Equation (13) results from the changing projection of
+the line of sight onto the orbital plane due to proper motion,
+with μ and Θμ,respectively, the amplitude and position angle
+of proper motion, and Ω the position angle of the line of nodes
+(Kopeikin 1996). The final term of Equation (13), involving
+changes in the Doppler factor D of Equation (1), is caused by
+the relative line of sight galactic accelerations of the solar
+system and the binary system.
+
+(9 )
+
+where the small parameter òA is defined as
+
+A º
+
+(12)
+
+where “GW” designates effects due to gravitational waves,
+while the observed, normalized time derivative of x stems from
+a combination of five terms:
+
+with λ and η the geodetically precessing polar angles of the
+pulsar spin axis with respect to the line of sight and line of
+nodes, respectively. (DD and DT92 suggested the substitution
+of a single fixed parameter, A0, for the two parameters A(t) and
+B(t)because observations at the time suggested that the spin
+and orbital angular momenta are aligned. However, subsequent
+observations of pulse profile changes have shown that this is
+not the case.)
+While the above equations provide a complete description of
+the calculation of ΔAberration at any proper emission time, DD
+and DT92 also provide an alternate approach that focuses on
+aberration parameters that change slowly (on precession
+timescales) as a result of spin axis precession. This procedure,
+detailed below, is more closely tailored to parameters
+determinable from TOA analyses.
+DT92 show that aberration will bias the observed value of
+the relativistic orbital shape parameter, d qobs, with respect to its
+intrinsic value,d qintr
+d qobs = d qintr -  A,
+
+W1geodetic
+(sin i cos l sin 2h + cos i sin l cos h ) ,
+sin2 l
+(11)
+
+where 2p W1geodetic is the geodetic precession period of the
+pulsar spin axis. The observed, normalized time derivative of e
+results from the sum of two phenomena
+
+where Ae(u) is a true-anomaly like quantity, and A(t) and B(t)
+are dependent on the precessing spin-axis geometry
+A (t ) = -
+
+could then serve as an
+
+(10)
+
+Hence the observational bias can be removed, given
+measurements of the aberration parameter A(t) and the Keplerian
+3
+
+=== PAGE 4 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+Table 1
+Astrometric and Spin Parameters
+
+Table 2
+Orbital Parameters
+
+Parameter
+
+Valuea
+
+Parameter
+
+Valuea
+
+t0 (MJD)b
+α (J2000)
+δ (J2000)
+μα (mas yr−1)
+μδ (mas yr−1)
+f (s−1)
+f˙ (s−2)
+
+52984.0
+19 h15m 27.s 99942 (3)
+16°06′27 3868(5)
+−1.23(4)
+−0.83(4)
+16.940537785677(3)
+−2.4733(1)×10−15
+
+T0 (MJD)
+x º a1 sin i (s)
+e
+Pb (d)
+ω0 (deg)
+á ˙wñ (deg yr−1)
+γ (ms)
+obs
+P˙b
+
+52144.90097849(3)
+2.341776(2)
+0.6171340(4)
+0.322997448918(3)
+292.54450(8)
+4.226585(4)
+0.004307(4)
+−2.423(1)×10−12
+4.0(25)×10−6
+−0.014(9)×10−12
+0.0006(7)× 10−12
+
+Glitch Parameters
+Glitch epoch (MJD)
+Δ f (s−1)
+Df˙ (s−2)
+
+d qobs
+x˙ obs
+e˙ obs (s−1)
+
+52777(2)
+5.49(3)×10−10
+−2.7(1)×10−18
+
+Shapiro Gravitational Propagation Delay Parameters
+Damour & Deruelle (1986) Parametrization
+s
+r (μs)
+Freire & Wex (2010) Parametrization
+ς
+h3
+
+Notes.
+a
+Figures in parentheses represent formal TEMPO standard errors in the last
+quoted digit, except for the glitch parameters. The stated uncertainty in glitch
+epoch results from empirically varying the glitch epoch until Δχ2 corresponds
+to the 68% confidence level; the quoted uncertainties in the other glitch
+parameters were derived from their variations as the glitch epoch was varied
+over the chosen range.
+b
+This quantity is the epoch of the next six measurements tabulated here.
+
++0.10
+0.680.06
++2.7
+9.6-3.5
+
+0.38(4)
+0.6(1)×10−6
+
+Note.
+a
+Figures in parentheses represent formal TEMPO standard errors in the last
+quoted digit. The DD Shapiro parameters s and r, which are highly covariant, and
+their uncertainties were refined through a process illustrated in Figure 2.
+
+The above equations demonstrate that measurements of ė or
+ẋ , along with experimental or theoretical determinations of
+some of the other quantities appearing therein, can usefully
+constrain others.
+
+similar to the one tabulated here are now known. (See the
+online Jodrell Bank Pulsar Glitch Catalog4; Espinoza
+et al. 2011). Note that, as with Weisberg et al. (2010),
+10higher-order spin derivatives were also fitted for in order to
+eliminate the effects of timing noise. Their values are not
+shown in the Table as they do not correspond to meaningful
+physical parameters.
+Table 2 displays the results of our fit to orbital parameters,including the eight final entrieswhich are fitted here for
+the first time in this system. Note that the first two of these eight
+new parameters, namely d qobs and x˙ obs, are measured at the
+marginal 1.5σ level, while the third, e˙obs, is only an upper limit.
+All others in this Table, including the new Shapiro terms, are
+measured with high confidence. In the next sections, we discuss
+important orbital measurables, including corrections that must
+be made to some of the observed quantities in order to
+determine their intrinsic values.
+
+4. RESULTS OF THE FITS
+We fitted the parameters discussed above to the full set of
+TOAsusing the TEMPO softwareas modified by us (see
+footnote 2). See Tables 1 and 2 for our results and their
+estimated uncertainties. The uncertainties quoted therein
+represent the standard errors from the TEMPO fit (except as
+noted). This convention differs from our previous practice,
+wherein many uncertainties were instead estimated from
+fitted parameter variations across multiple reasonable fits.
+While the old procedure facilitated the incorporation of some
+systematic uncertainties into the error budget,the more stable
+recent instrumental configurations appear to minimize such
+effects.
+Some of the fitted parameters shifted by several σ with
+respect to the values reported in Weisberg et al. (2010). The
+shifts can all be attributed to the new incorporation of a
+frequency and time offset for each WAPP observing session,
+and center frequency in order to account for geodetic
+precession-induced profile changes (see Section 2), and to
+our new procedure of fitting for rather than freezing at zerothe
+parameter ẋ . The latter procedure also led to a significantly
+larger uncertainty in the fitted value of γ and in quantities
+derived therefrom.
+The astrometric and spin solutions are listed in Table 1.
+These are quite similar to those given in Weisberg et al. (2010),
+except that our longer post-glitch baseline made it clear that the
+previously discovered glitch at MJD ≈ 52770 is better modeled
+with the addition of a change in spin frequency derivative, Df˙ .
+There remains only one known glitch that has a significantly
+smaller value of Δ f/f (in globular cluster millisecond PSR
+B1821-24;Mandal et al. 2009), although several of magnitude
+
+4.1. The Observed and Intrinsic Orbital Period Derivative
+obs
+
+The observed orbital period derivative, P˙b , must be
+gal
+corrected by a term, P˙b , resulting from the relative galactic
+accelerations of the solar system and the binary system
+(Damour & Taylor 1991), in order to yield the intrinsic
+intr
+derivative, P˙b
+intr
+obs
+gal
+P˙b = P˙b - P˙b .
+
+(15)
+
+Using galactic parameters of R0=8.34±0.16 kpc and
+Θ0=240±8 km s−1 from Reid et al. (2014), a pulsar distance
+estimate from Weisberg et al. (2008), and the pulsar proper
+4
+
+4
+
+http://www.jb.man.ac.uk/pulsar/glitches.html
+
+=== PAGE 5 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+Figure 1. Shapiro gravitational propagation delay variation around the orbit at three epochs. The curve represents the expected delay based on a general relativistic
+calculation, while the points and their error bars result from combining all residuals to a special fit (see text) near the given epoch into one of 20 orbital time bins. Time
+is reckoned with respect to TConj, the epoch of the pulsar’s superior conjunction with the companion. Each curve peaks at that epochwhen the pulsar’s earthbound
+signals plunge most deeply into the companion’s gravitational well. The amplitude and shape of the curves evolvedue to relativistic precession of the orbital ellipse,
+as quantified by the advancing longitude of periastron ω.
+gal
+motion from Table 1, we find that P˙b = -(0.025  0.004)
+obs
+´ 10-12 . Inserting P˙b from Table 2 into Equation (15), we
+intr
+calculate that P˙b = -(2.398  0.004) ´ 10-12 . The uncergal
+tainty in this result is dominated by the error in P˙b , which in
+turn is set principally by the pulsar distance uncertainty. A
+VLBA parallax campaign on the pulsarcurrently in progress
+will hopefully improve these uncertainties.
+
+of a pair of Shapiro measurables in the PSR B1913+16 system.
+(The two DD parameters, s and r, had been jointly constrained
+by Taylor & Weisberg (1989). For the ensuing decade,
+unfavorable orbital geometry rendered its amplitude unmeasurably small,while the last decade has seen both improving orbital
+geometry and advances in observing instrumentation.) We
+account for the significant nonlinear covariance of the s and r
+parameters byestimating their values and uncertainties in Table 2
+using a process delineated by Splaver et al. (2002). The
+procedure is illustrated graphically in Figure 2, which also shows
+the best-fitting FW Shapiro parameters. (Tighter constraints on
+the inclination and companion mass can be derived indirectly
+from other measurements. See Section 4.3.)
+
+4.2. First Successful Measurement of the Shapiro Gravitational
+Propagation Delay Parameters in PSR B1913+16
+Due to relativistic precession of the elliptical orbit, the Shapiro
+delay has recently grown to an amplitude of ∼35 μs around the
+orbit, rendering it relatively easy tomeasure. Figure 1 illustrates
+the enhancement of the Shapiro delay signal around the orbit
+over the last dozen years, during which time the WAPP receivers
+also came into use, thereby increasing our observing bandwidth
+ten-fold. The curves in Figure 1 illustrate a general relativistic
+calculation of the expected Shapiro delay variation around the
+orbit, while the data points are residuals ofa TEMPO fit freezing
+all parameters at their best-fit values, except for the Shapiro
+parameter r (corresponding in General Relativity to the
+companion mass). The latter quantity was artificially set to zero
+to simulate the absence of the Shapiro delay in the fit. The
+pattern of residuals systematically matches thetheoretical
+expectations for the Shapiro delay.
+We have now successfully determined the two Shapiro terms
+in both the DD and FW parametrizations (see Table 2). While
+the Shapiro delay has been observed in several other binary
+pulsar systems, these results mark the first successful detection
+
+4.3. Best Determination of Component Masses and Orbital
+Inclination
+The measurement of the first seven quantities in Table 2
+enables the precise general relativistic determination of the
+component masses and orbital inclination. Specifically, our
+measurements of áw˙ ñ and γ, along with the Keplerian elements,
+leave only the two unknowns, m1; áw˙ , g and m 2; áw˙ , g ,in the
+following two general relativistic equations:
+áw˙ ñ = 3 G 2 3 c-2 (Pb 2p )-5 3
+´ (1 - e2)-1 (m1; á ˙wñ, g + m 2; á ˙wñ, g )2 3
+= 3T☉2 3 (Pb 2p )-5 3 (1 - e2)-1
+⎛ m1; á ˙wñ, g + m 2; á ˙wñ, g ⎞2 3
+´⎜
+⎟ ,
+⎠
+⎝
+M☉
+
+5
+
+(16)
+
+=== PAGE 6 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+Figure 2. (a) Measured constraints on ∣cos i∣ and m2 resulting from TEMPO fits for two different parametrizations of the Shapiro gravitational propagation delay within
+the context of general relativity. The Damour & Deruelle (1986) s and r parameters map directly onto the displayed ∣cos i∣ [=+ 1 - s 2 ] and m 2 [=(r T☉) M☉] axes,
+respectively; the black contours show joint 1, 2, and 3s (Dc 2 = 2.3, 6.2, 11.8) confidence limits on those quantitiesderived from a set of TEMPO fits to a large grid
+of (fixed) (∣cos i∣, m 2 ). The alternate Freire & Wex (2010) best-fit parameter constraints and their ±1σ limits are shown in green. Their fitted parameter, ς, transforms
+directly into the displayed ∣cos i∣ axis (see Equation (4)), whereas their h3 parameter does not map uniquely onto either of the axes (see Equation (5)). The marginal
+distributions in (b) and (c) result from collapsing the resulting two-dimensional DD probability distribution onto the ∣cos i∣ and m2 axes,respectively, in which the
+mean (solid black) and the 1σ bounds (gray region) are displayed, yielding ∣cos i∣ = 0.73 (+0.05, -0.11) and m 2 = 1.95 (+0.55, -0.71) M☉ (68.3% confidence).
+
+much more precise than the m2 and sin i or ∣cos i∣ values
+determined directly from the Shapiro propagation delay
+measurements of Section 4.2.
+
+and
+g = G 2 3 c-2 e (Pb 2p )1 3 m 2; á ˙wñ, g
+´ (m1; á ˙wñ, g + 2m 2; á ˙wñ, g )(m1; á ˙wñ, g + m 2; á ˙wñ, g )-4 3
+m 2; á ˙wñ, g
+= T☉2 3 e (Pb 2p )1 3
+M☉
+⎛ m1; á ˙wñ, g + 2m 2; á ˙wñ, g ⎞ ⎛ (m1; á ˙wñ, g + m 2; á ˙wñ, g ⎞-4 3
+´⎜
+.
+⎟
+⎟⎜
+⎠
+⎠⎝
+⎝
+M☉
+M☉
+(17)
+
+4.4. Toward the First Published Measurement of the
+Relativistic Orbital Shape Correction, δθ, in any System
+We have successfully measured the apparent post-Keplerian
+orbital shape correction term, d qobs (see Table 2). As noted in
+Section 3.2, this observed value must be corrected for a
+comparable aberration signal, òA (see Equation (9)). Geodetic
+spin-precession modeling of this system should in principle
+determine the necessary aberration parameters by specifying
+the spin axis orientation (specifically, its polar angles η and λ)
+over time (see Equation (7)). However, we find that the
+currently available pulse shape variation fits (Kramer 1998;
+Weisberg & Taylor 2002; Clifton & Weisberg 2008) yield
+inconsistent solutions for these parameters. Consequently,
+although we have successfully measured d qobs, it is not yet
+
+Simultaneously solving for the two component masses, we find
+that m1; áw˙ , g = 1.438  0.001 M☉ and m 2; áw˙ , g = 1.390 
+0.001 M☉. These values agree with WNT within 2σ, while
+our precision is poorer due to a less-precisely determined γ (see
+Section 4). Furthermore, Newtonian physics then yields an
+additional quantity from the derived masses, and our x and Pb
+measurements, (sin i )áw˙ , g = 0.7327  0.0004 (or, equivalently,
+∣cos i∣áw˙ , g = 0.6806  0.0004). These values are currently
+6
+
+=== PAGE 7 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+possible to determine d qintr nor to use its measured value as an
+
+unique measurement of d A dt . Unfortunately, our fitted value
+of (e˙ e)obs is currently not significantly different from zero,
+although its upper limit is in the 10−15 s−1 range (see Table 2)
+expected for d A dt .
+
+additional test of relativistic gravitation.
+Despite our current inability to measure d qintr , we can
+determine its expected value, d qGR , via a general relativistic
+calculation (DD; DT92)
+
+4.5.2. Constraints from x˙ obs
+
+G 2 3 ⎛ P ⎞-2 3
+d qGR = 2 ⎜ b ⎟
+(m1 + m 2)-4 3
+
+The second term in Equation (13) delineates the gravitational
+wave-induced orbital shrinkage rate, which can be evaluated
+from measurables via
+
+⎝ 2p ⎠
+7 2
+´
+m1 + 6m1m 2 + 2m 22
+2
+-4 3
+⎛ P ⎞-2 3 ⎛ m1 + m 2 ⎞
+= T☉2 3 ⎜ b ⎟
+⎟
+⎜
+⎝ 2p ⎠
+⎝ M☉ ⎠
+c
+
+{
+
+}
+
+GW
+⎛ a˙1 ⎞GW
+2 ⎛ P˙ ⎞
+= ⎜ b⎟
+= - 5.7 ´ 10-17 s-1.
+⎜ ⎟
+⎝ a1 ⎠
+3 ⎝ Pb ⎠
+
+⎫
+⎧ 7 ⎛ m ⎞2
+⎛ m ⎞2 ⎪
+⎛m m ⎞
+⎬,
+⎨ ⎜ 1 ⎟ + 6⎜ 1 2 ⎟ + 2⎜ 2 ⎟ ⎪
+´⎪
+⎝ M☉ ⎠ ⎭
+⎝ M☉ ⎠
+⎩ 2 ⎝ M☉ ⎠
+⎪
+
+The third (spin–orbit) term of Equation (13) varies
+approximately sinusoidally on the geodetic precession timescale with an amplitude of ~3 ´ 10-15 s-1. Details await a
+robust determination of geodetic precession parameters (see
+Equation (14)).
+The fourth (Kopeikin 1996) term of Equation (13) has a
+maximum amplitude5 of ∼2.3×10−16 s−1; while the fifth and
+final term is
+
+(18)
+
+yielding d qGR = (6.187  0.001) ´ 10-6 . Equation (9) can then
+be inverted to give the aberration parameter  A = (2.2  2.5) ´
+10-6 . This timing-derived value of òA will provide a modest
+consistency check on future geodetic precession modeling.
+
+-
+
+4.5. Implications of Fits for the First Time-derivatives of e
+and x
+
+4.5.1. Constraints from e˙obs
+
+The second and final term of Equation (12) involves the time
+evolution of e induced by gravitational wave (GW) emission.
+For the Equation (12) second term, Peters (1964) shows that
+this term,
+⎛ e˙ ⎞GW
+304 G 3
+⎜ ⎟
+=m (m1 + m 2 )2
+⎝e⎠
+15 c 5aR4
+
+304 5 3
+T☉
+15
+
+(
+
+m1 + m2
+M☉
+
+)
+
+The determination of seven particular independent quantities
+suffices to fully determine the dynamics of a binary system within
+the context of a particular theory of gravitation. For example, the
+most accurate determination of component masses and orbital
+inclination in Section 4.3 and of d qGR in Section 4.4 depend upon
+subsets of the first seven measurements listed in Table 2.
+Consequently, any additional measurement would constitute
+an independent test of relativistic gravitation under strong-field
+conditions. In the following two sections, we delineate
+relativistic gravitational tests via measurements of gravitational
+radiation emission and of the Shapiro gravitational propagation
+delay, respectively.
+
+⎞-8 3
+
+⎛ Pb
+⎜
+⎟
+2p ⎠
+
+1 3⎝
+
+⎛
+121 2⎞⎟
+e
+´ (1 - e2)-5 2 ⎜1 +
+⎝
+304 ⎠
+= -2.9 ´ 10-17 s-1,
+
+(21)
+
+5. TESTS OF RELATIVISTIC GRAVITATION
+
+⎛
+121 2⎞⎟
+e
+´ (1 - e2)-5 2 ⎜1 +
+⎝
+304 ⎠
+=-
+
+P˙b,gal
+D˙
+=+
+= - 1 ´ 10-18 s-1
+D
+Pb
+
+(see Section 4.1 for details on the calculation of P˙b,gal ).
+In summary, the first (aberration) and third (spin–orbit) terms
+dominate Equation (13), so all others may be ignored.
+However, neither of these two terms is currently accurately
+determinable. We do have a marginally significant measurement of x˙ obs (see Table 2). Consequently, if either of the two
+terms becomes well-determined in the future, along with an
+improved value of x˙ obs, then the other term will also become
+accessible. For example, there are two possible paths toward
+determiningthe aberrational term: first, additional geodetic
+precession observations and modeling should better constrain
+d A dt; and second, additional observations could better
+determine e˙obs, which, as noted in Section 4.5.1, would then
+be equivalent to a measurement of d A dt . At this pointthe
+spin–orbit term would be calculable, leading to an exciting
+measurement of the pulsar’s moment of inertia, which has
+important implications for neutron star equations of state
+(Lattimer & Schutz 2005). With the measurement precision of
+ẋ and ė improving with time,t, as t -3 2 , another decade or so
+of observations is required. Unfortunately, geodetic spin axis
+precession may cause the pulsar to disappear before that time.
+
+As noted above, Equations (12) and (13) demonstrate that
+the successful measurement of e˙obs and x˙ obs would lead to
+constraints on other quantities of interest. It is therefore useful
+to further investigate the various terms composing these
+equations.
+The first (aberration) term of both equations, d A dt , was
+defined in Equations (10) and (11). However, as discussed in
+Section 4.4 for òA, additional progress in understanding the
+pulsar’s spin axis orientation is needed before d A dt can be
+confidently determined. Nevertheless, our current geodetic
+precession modeling suggests that its value is in the
+∼10−15 s−1 range and varies with spin–precessional phase.
+
+m1 m2
+M☉ M☉
+
+(20)
+
+(19)
+
+with μasthe reduced mass. This term is negligible compared
+to the expected value of d A dt , except at fortuitous
+precessional phases where the latter can drop to zero.
+Consequently, the d A dt term dominates Equation (12), so
+that a successful measurement of (e˙ e)obs would provide a
+
+5
+The exact value depends on the unknown alignment on the sky of the line of
+nodes.
+
+7
+
+=== PAGE 8 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+5.1. Gravitational Radiation Emission and P˙b
+
+next section, that probe different aspects of strong-field
+gravitationare also useful for constraining viable alternatives
+to general relativity.
+
+Gravitational radiation emission should cause the orbit to
+GR
+decay as orbital energy is radiated away. The quantity P˙b is
+the resulting orbital period derivative expected from a general
+relativistic calculation of this phenomenon (Peters & Mathews
+1963):
+
+Each of the two newly measured Shapiro parameters
+represents another independent test of relativistic gravitation.
+As with the P˙b test of Section 5.1, the Shapiro tests also require
+the complementary measurement of áw˙ ñ and γ in order to make a
+testable prediction for the value of the Shapiro parameters. In
+this case, unlike the P˙b test, all of the post-Keplerian quantities
+probe quasi-static phenomena in strong fields. While Shapiro
+parameters have already been measured in several other binary
+systems, it is especially useful to constrain theories via systems
+such as this one and PSRs B1534+12, and J0737-3039Awhere
+at least three “excess” post-Keplerian parameters beyond áw˙ ñ and
+γ (one gravitational radiation parameter and two Shapiro
+quantities) are measurable. Although the precision of the binary
+pulsar Shapiro parameter measurements is well below their
+measurement precision in the weak solar gravitational field, it is
+this simultaneous determination of several parameters in strongfield conditions in each of these binary pulsar systems that leads
+to the important constraints on relativistic theories of gravitation.
+In the DD formulation of the Shapiro delay within general
+relativity, the Shapiro measurables s and r map directly onto
+sin i and m2, respectively. Hence we can test general relativity
+by comparing the Shapiro determination of sin i (ºs ) with that
+determined from áw˙ ñ and γ (called sin iáw˙ , g ; see Section 4.3):
+
+-5 3
+
+192 p G ⎛⎜ Pb ⎞⎟
+GR
+P˙b = ⎝ 2p ⎠
+5 c5
+⎛
+73 2
+37 4⎟⎞
+´ ⎜1 +
+e +
+e (1 - e2)-7 2
+⎝
+24
+96 ⎠
+´ m1 m 2 (m1 + m 2)-1 3
+192 p 5 3 ⎛⎜ Pb ⎞⎟-5 3
+=T☉
+⎝ 2p ⎠
+5
+⎛
+73 2
+37 4⎟⎞
+´ ⎜1 +
+e +
+e (1 - e2)-7 2
+⎝
+24
+96 ⎠
+⎛ m ⎞ ⎛ m ⎞ ⎛ m + m 2 ⎞-1 3
+´ ⎜ 1 ⎟⎜ 2 ⎟⎜ 1
+.
+⎟
+⎝ M☉ ⎠ ⎝ M ☉ ⎠ ⎝ M☉ ⎠
+5 3
+
+5.2. Shapiro Gravitational Propagation Delay
+
+(22)
+
+Inserting our measured and derived values and their
+GR
+uncertainties into Equation (22)6, we find that P˙b =
+12
+(-2.40263  0.00005) ´ 10 . To verify our estimate of
+GR
+the error in P˙b
+that was derived via propagation of
+uncertainty,we also employed a Monte-Carlo method with
+Cholesky decomposition of the covariance matrix. In this
+fashion, we simulated the joint normal distribution of measured
+parameters {g , w˙ , Pb, e}, and then constructed a histogram of
+GR
+1,000,000 derived P˙b and inferred the uncertainty therefrom.
+Consequently, we find that
+
++0.10
+0.68s
+0.06
++0.14
+=
+= 0.930.07 .
+sin iá ˙wñ, g
+0.7327  0.0004
+
+Similarly, we can test general relativity by comparing the
+Shapiro determination of m 2 (ºr M☉ T☉) with that determined
+from áw˙ ñ and γ
+
+intr
+P˙b
+( - 2.398  0.004) ´ 10-12
+=
+GR
+( - 2.40263  0.00005) ´ 10-12
+P˙b
+
+= 0.9983  0.0016.
+
+(24)
+
++0.55
+(1.95r M☉ T☉
+0.71 ) M☉
++0.40
+=
+= 1.400.51 .
+m 2; á ˙wñ, g
+(1.390  0.001) M☉
+
+(23)
+
+This result demonstrates that the system is losing energy to
+gravitational radiation within ∼1σ of the rate predicted by
+general relativity (see also Figure 3 and the red curve in
+Figure 4). The above number represents a significant improvement over the value determined by WNT, 0.997±
+0.002,which represented a 1.8σ discrepancy between our
+measurements and general relativity. Interestingly, the new
+galactic parameters of Reid et al. (2014) are the principal
+gal
+reason for the improvement (via a change in P˙b ), while our
+measured values themselves changed little.
+In addition to confirming general relativistic radiation
+damping at this level, our result rules out large parameter
+spaces in plausible scalar–tensor theories of gravity. In recent
+years, however, other pulsars in neutron-star–white-dwarf
+binary systems have overtaken PSR B1913+16 in constraining
+these alternatives (Freire et al. 2012).
+DD92 point out that this test is a “mixed” strong-field probe
+obs
+in that it involves a combination of radiative effects (via P˙b )
+and quasi-static phenomena (through áw˙ ñ and γ, whose values
+GR
+are needed in order to make a prediction of the expected P˙b ).
+Consequently, additional tests, such as those described in the
+
+(25)
+
+The consistency (albeit with a rather low level of precision) of
+these Shapiro determinations of sin i and m2 with those measured
+via the other post-Keplerian terms, and hence their confirmation
+of general relativity, is also graphically depicted in Figure 4.
+The Shapiro terms have also been measured in several other
+binary pulsar systems with higher precision, and have also been
+shown to be in agreement with general relativity.
+6. CONCLUSIONS
+We report here on the measurements and relativistic analyses
+of 9309 TOAs in over 30 years of high-quality Arecibo data on
+binary pulsar PSR B1913+16. We fitted for a number of
+previously unmeasurable parameters for the first time in this
+system (and in one case, for the first time anywhere), which
+enabled us to significantly advance our relativistic analyses of
+the system. We provide our newest measurents or derivations
+of all relevant physical quantities of the binary system, with the
+exception of Ω, the position angle of the line of nodes. We
+rigorously ascertained the uncertainties in the fitted and derived
+parameters. Having fully characterized the system, we
+proceeded to use it in several tests of general relativity in
+strong-field conditions.
+We have measured a gravitational radiation-induced orbital
+period decrease whose rate agrees with the general relativistic
+
+6
+
+This value may also be calculated directly from the first seven orbital
+measurables in Table 2 alone and without the use of derived quantities such as
+the masses (DT91).
+
+8
+
+=== PAGE 9 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+Figure 3. Orbital decay of PSR B1913+16 as a function of time. The curve represents the orbital phase shift expected from gravitational wave emission according to
+General Relativity. The points, with error bars too small to show, represent our measurements.
+
+expectation to within ∼1σ, which is closer than found
+by WNT, largely as a result of an improved galactic correction
+resulting from more accurate galactic parameters (Reid
+et al. 2014).
+Similar orbital decay tests have now been performed with
+several other binary pulsars (see Table 3 for published
+intr
+GR
+measurements of P˙b P˙b ). The orbital decays of PSRs
+J0348+0432, J0737-3039, J1141-4565, J1738+0333, J1906
++0746, B1913+16, and B2127+11C all exhibit agreement
+between observation and general relativity to within (or very
+close to) the authors’ stated uncertainties. PSR B1913+16
+currently has the most precise determination, and interferometric parallax measurementscurrently in progresswill hopefully further tighten the precision.
+Among the other two, PSR B1534+12 and PSR J17562251, various systematic effects such as an incorrect distance in
+the galactic acceleration correction may explain the small
+observed discrepancies, although it is possible that an
+incompleteness of general relativity or some unknown physical
+effect is responsible. See the work of Ferdman et al. (2014) for
+an especially thorough description of the most significant
+deviation of the orbital decay rate from the general relativistic
+prediction, found in PSR J1756-2251.
+Our new (for this system) measurements of Shapiro gravitational propagation delay parameters represent two additional
+tests of relativistic gravitation, and are fully consistent with
+general relativity, although their relative precision is currently far
+
+Figure 4. Constraints on the masses of the pulsar and the companion as a
+function of five post-Keplerian measurables, within the context of General
+Relativity. The width of each curve represents ±1σ error bounds. The mutual
+near-intersection of all curves illustrates the agreement of our observations with
+general relativity in the strong-field conditions at the binary system.
+
+9
+
+=== PAGE 10 ===
+
+The Astrophysical Journal, 829:55 (10pp), 2016 September 20
+
+Weisberg & Huang
+
+Table 3
+Comparison of Gravitational Radiation-induced Orbital Decay with GR
+Prediction in Binary Pulsars
+PSR
+J0348+0432
+J0737-3039
+J1141-6545
+B1534+12
+J1738+0333
+J1756-2251
+J1906+0746
+B1913+16
+B2127+11C
+
+intr
+GR
+P˙b P˙b
+
+1.05±0.18
+1.003±0.014
+1.04±0.06
+0.91±0.06
+0.94±0.13
+1.08±0.03
+1.01±0.05a
+0.9983±0.0016
+1.00±0.03
+
+Much of this experiment was pioneered by J.H. Taylor, to
+whom we owe our deepest thanks. D.J. Nice assisted with
+observing and analyses, and A.A. Chael assisted in developing
+the FW analysis package. The authors gratefully acknowledge
+financial support from the US National Science Foundation.
+The Arecibo Observatory is operated by SRI International
+under a cooperative agreement with the National Science
+Foundation (AST-1100968), and in alliance with Ana G.
+Mendez-Universidad Metropolitana, and the Universities Space
+Research Association.
+Facility: Arecibo.
+
+References
+Antoniadis et al. (2013)
+Kramer et al. (2006)
+Bhat et al. (2008)
+Stairs et al. (2002)
+Freire et al. (2012)
+Ferdman et al. (2014)
+van Leeuwen et al. (2015)
+This work
+Jacoby et al. (2006)
+
+REFERENCES
+
+Note.
+a
+Assumes negligible proper motion.
+
+Antoniadis, J., Freire, P. C. C., Wex, N., et al. 2013, Sci, 340, 448
+Bhat, N. D. R., Bailes, M., & Verbiest, J. P. W. 2008, PhRvD, 77, 124017
+Clifton, T., & Weisberg, J. M. 2008, ApJ, 679, 687
+Damour, T., & Deruelle, N. 1986, AnIHP, 44, 263 (DD)
+Damour, T., & Taylor, J. H. 1991, ApJ, 366, 501 (DT91)
+Damour, T., & Taylor, J. H. 1992, PhRvD, 45, 1840 (DT92)
+Demorest, P. B., Ferdman, R. D., Gonzalez, M. E., et al. 2013, ApJ, 762, 94
+Espinoza, C. M., Lyne, A. G., Stappers, B. W., & Kramer, M. 2011, MNRAS,
+414, 1679
+Ferdman, R. D., Stairs, I. H., Kramer, M., et al. 2014, MNRAS, 443, 2183
+Freire, P. C. C., & Wex, N. 2010, MNRAS, 409, 199 (FW)
+Freire, P. C. C., Wex, N., Esposito-Farèse, G., et al. 2012, MNRAS,
+423, 3328
+Hulse, R. A., & Taylor, J. H. 1975, ApJL, 195, L51
+Jacoby, B. A., Cameron, P. B., Jenet, F. A., et al. 2006, ApJL, 644, L113
+Kopeikin, S. M. 1996, ApJL, 467, L93
+Kramer, M. 1998, ApJ, 509, 856
+Kramer, M., Stairs, I. H., Manchester, R. N., et al. 2006, Sci, 314, 97
+Lattimer, J. M., & Schutz, B. F. 2005, ApJ, 629, 979
+Lorimer, D. R., & Kramer, M. 2004, Handbook of Pulsar Astronomy, Vol. 4
+(Cambridge: Cambridge Univ. Press), 229
+Mandal, R. D. R., Konar, S., Dey, M., & Dey, J. 2009, MNRAS, 399, 822
+Peters, P. C. 1964, PhRv, 136, 1224
+Peters, P. C., & Mathews, J. 1963, PhRv, 131, 435
+Reid, M. J., Menten, K. M., Brunthaler, A., et al. 2014, ApJ, 783, 130
+Shapiro, I. I. 1964, PhRvL, 13, 789
+Splaver, E. M., Nice, D. J., Arzoumanian, Z., et al. 2002, ApJ, 581, 509
+Stairs, I. H., Thorsett, S. E., Taylor, J. H., & Wolszczan, A. 2002, ApJ,
+581, 501
+Taylor, J. H., & Weisberg, J. M. 1982, ApJ, 253, 908
+Taylor, J. H., & Weisberg, J. M. 1989, ApJ, 345, 434
+van Leeuwen, J., Kasian, L., Stairs, I. H., et al. 2015, ApJ, 798, 118
+Weisberg, J. M., Nice, D. J., & Taylor, J. H. 2010, ApJ, 722, 1030 (WNT)
+Weisberg, J. M., Romani, R. W., & Taylor, J. H. 1989, ApJ, 347, 1030
+Weisberg, J. M., Stanimirović, S., Xilouris, K., et al. 2008, ApJ, 674, 286
+Weisberg, J. M., & Taylor, J. H. 1981, GReGr, 13, 1
+Weisberg, J. M., & Taylor, J. H. 2002, ApJ, 576, 942
+You, X. P., Hobbs, G., Coles, W. A., et al. 2007, MNRAS, 378, 493
+
+lower than the orbital decay test. This binary now joins several
+other systems, including PSRs J0737-3039A, B1534+12, and
+J1756-2251, witheach providing at least three independent tests
+of relativistic, strong-field gravitation.
+We have also marginally measured the orbital shape
+parameter δθ for the first time anywhere, but its intrinsic value
+is corrupted by a comparable, undetermined aberration delay.
+Future geodetic spin–orbit precession measurements should
+lead to an accurate characterization of the aberration and then
+an additional relativistic gravitational test via the comparison of
+the aberration-corrected d qintr with d qGR .
+In addition, we fitted for the time derivative of orbital
+eccentricity,e, and the projected semimajor axis of the pulsar
+orbit, x, and we achieved an upper limit on the former and a
+marginal detection of the latter. We discussed and quantified
+the various physical phenomena that can contribute to these
+parameters. Unless the pulsar disappears in the next few years
+due to geodetic spin axis precession, future timing observations
+should better define these quantities, allowing for a determination of the pulsar’s moment of inertia, I1.
+We have placed online (see footnote 2) a subroutine and
+modifications to the TEMPO TOA fitting software, which
+codes the Freire & Wex (2010) parametrization of the Shapiro
+delay for high-eccentricity binary pulsars such as the PSR
+B1913+16 system. The TEMPO input files and TOAs upon
+which these analyses are based are available in a .tar.gz
+package published with this paper and in a persistent repository
+http://dx.doi.org/10.5281/zenodo.54764 and with the article
+preprint http://arxiv.org/e-print/1606.02744v1.
+
+10