Building a Photometric Hubble Diagram from DES-SN Data R. Kessler University of Chicago SN ClassificaCon Workshop April 28 26
The Dark Energy Survey Survey project using 4 complementary techniques: I. Cluster Counts II. Weak Lensing III. Large-scale Structure IV. Supernovae Two multiband imaging surveys: 5 deg 2 grizy to 24th mag 3 deg 2 repeat griz (SNe) DECam on the Blanco New 3 deg 2 FOV camera on the Blanco 4m telescope Survey 23-28 (525 nights) Facility instrument for astronomy community (DES 3% time) Largest optical etendue (FoV x mirror-area) 2 www.darkenergysurvey.org www.darkenergydetectives.org
DES-SN Overview Visit each field once per week in griz 8 shallow fields (24 deg 2 ), depth 23.5 2 deep fields (6 deg 2 ), depth 24.5 Finished three `5-month seasons 3
Source of Photometry (Y+Y2+Y3) Forced Photometry (PSF-fi_ed fluxes) from Difference-Imaging Pipeline used to discover transients Subtracted images from A.Becker s hotpants CalibraCon tweaked few percent based on overlap of PS supercal (Scolnic et al 25) More refined photometry (Scene-Modeling) in progress, but not ready yet. 4
Source of Redshies Host Galaxy Spectroscopic redshies, mostly from OzDES using AAT (Yuan 25) Matching algorithm (DLR) from Gupta et al, 26 (arxiv:64.638) 5
SimulaCons ClassificaCon and HD Fikng make extensive use of SNANA simulacons. Simulate SNIa with SALT2 model (α SIM =.4, β SIM =3.2; c,x populacon from SK6) and intrinsic sca_er models that include chromacc variacons. Simulate CC based on 42 templates (from photometric challenge, 2). CC Rate, Ibc/II raco, and LF from Li 2. 6
Pre-Analysis PreparaCon: Analyze Fake SNIa Overlaid on Images Are photometry errors correct? à Yes, aeer correccng for surface brightness (see Figs 9- in arxiv:57.537) Does fast SNANA sim reproduce fake SNe? à yes, aeer scaling simulated error vs. PSF (see Sim-validaCon slides to follow) 7
Sim-ValidaCon: Overview SN Ia Fakes DiffImg Pipeline (weeks) SNANA SimulaCon (minutes) Analysis & fikng Compare DiffImg Fakes to SNANA sim 8
Empirical Flux-Error Adjustment for SNANA sim.5 Shallow Fields, Y+Y2.5 err(forcephoto) / err(snana Sim).4.3.2..9.8.5.4.3.2..9.8 2 3 4 5 PSF-g PSF-i g 2 3 4 5 i.4.3.2..9.8.5.4.3.2..9.8 2 3 4 5 PSF-r PSF-z r z 2 3 4 5 Scale simulated flux-errors by black curves at lee. Purple-dash curve is prediccon based on floacng background in PSF fit. forcephoto-error excess at small-psf is not understood. PSF units: sigma, pixels [ =.6 PSF(fwhm,arcsec) ] 9
Empirical Flux-Error Adjustment for SNANA sim.5 Shallow Fields, Y+Y2.5.5 Deep Fields, Y+Y2.5.4.4.4.4.3.3.3.3 err(forcephoto) / err(snana Sim).2..9.8.5.4.3.2..9.8 2 3 4 5 PSF-g PSF-i g 2 3 4 5 i.2..9.8.5.4.3.2..9.8 2 3 4 5 PSF-r PSF-z r z 2 3 4 5.2..9.8.5.4.3.2..9.8 2 3 4 PSF-g PSF-i g 2 3 4 i.2..9.8.5.4.3.2..9.8 2 3 4 PSF-r PSF-z r z 2 3 4 PSF units: sigma, pixels [ =.6 PSF(fwhm,arcsec) ]
Fake/Sim Overlays of Fi_ed Params 6 Shallow Fields, Y 6 6 4 4 4 4 2 4 2.5 SALT2c 2 3 2..2 err(c) 4 2 2 2 Fake SN à 2DiffImg 3 err(pkmjd) SNANA sim 75 5 5 25-2 2 PKMJD PKMJD true 4 3 2-5 5 SALT2x 2 25 SALT2mB 6 4 2 2 err(x)..2.3 err(mb).4.2 -.2 -.4.5.5. z phot -.25.25 err(z phot ) ZPHOT ZPHOT true.5 -.5 -.2.4.6.8.2.4.6.8 mean c vs z phot mean x vs z phot
Done with pre-analysis of fake SN Now start real analysis on data 2
Photometric SelecCon Box cuts: * 3 bands with obs having SNR>5 * at least one epoch with Trest < -2 * at least one epoch with Trest > + * x < 4 * FitProb >.5 (from SALT2 LC fit) 25 2 5 5 Nearest Neighbor (Sako 24) P Ia >.5 with >σ significance..2.3.4.5.6.7.8.9 P fit = Fit Prob 3
Nearest Neighbor (NN): 3D Space of Redshie, Stretch, Color DATA Fit color. Fit stretch 4
Nearest Neighbor (NN): 3D Space of Redshie, Stretch, Color DATA Sim(Ia+CC) Fit color. Fit color P Ia =8/2 True SNIa True SN CC Fit stretch Fit stretch NN Training finds opcmal radius of 3D sphere by maximizing Eff x Purity. 5
NN Performance on SimulaCon: Aeer box cuts: SNIa Purity 9% Aeer P NN,Ia > ½: SNIa Purity à 98% with 2% SNIa loss 6
Data/Sim Comparisons Box cuts + P NN,Ia > ½ Sim zhost tuned to match DES data (Eff zhost not known yet) No other sim tuning (except flux-errors to match fakes) µ(α=.4, β=3.2) 46 45 44 43 42 4 4 39 38 37 36 DATA, NN(z,s,c).2.4.6.8 zspec zspec µ(α=.4, β=3.2) 46 45 44 43 42 4 4 39 38 37 36 SIM, NN(z,s,c).2.4.6.8 zspec 7
Data Sim Data/Sim Comparisons 8 6 4 2 5 tuned.2.4.6.8 zhd 75 5 25 2 4 6 SNRMAX 5 5 5.5..5.2 err c 75 5 25.5.5 2 err x 5 5 5 5 -.5.5 SALT2-c -4-2 2 4 SALT2-x.5..5.2 err mb 2 3 err PKMJD 5 2 5 2 5 5 8 2 22 24 26 SALT2-mB.5.6.7.8.9 NN-P Ia 2 4 6 SNRMAX3.2.4.6.8 P fit 8
.5..5.2.5.5 2 c B 6 4 4 3 2 2.5.2 4 8 4 Data Sim.5..5.2 8 6 3 6 4 err c err x Data/Sim 2 Comparisons.5.5 2 err mb 75 5 err x (24.5 5 mag) 2 2 2 25 5 tuned.5.2.2.4.6.8 22 4 6-3 zhd SNRMAX err err c 5 2 4 PKMJD 6.2.4 5.6.8 SNRMAX3 Shallow Fields (23.5 mag) P fit 2 4 3 2 2 3 err PKMJD DEEP Fields 5 5 5 5.5..5.2 75 5 25.5.5 2 err x -.5.5 SALT2-c -4-2 2 4 SALT2-x.5..5.2 err mb 2 3 err PKMJD AX3 5 4 6 5 2.2.4.6.8 P fit 5 5 2 8 2 22 24 26 SALT2-mB.5.6.7.8.9 NN-P Ia 2 4 6 SNRMAX3.2.4.6.8 P fit 9
Data/Sim Comparisons NN(z,s,c) 2 DATA SIM (Ia+CC) SIM (CC-only) - -3-2 - 2 3 µ SN µ fit (SALT2mu fit) Sim CC/Tot =.6%. Data contaminacon is clearly few Cmes more. 2
When all else Fails, Tune the CC Sim 2 DATA SIM (Ia+CC) SIM (CC-only) Untuned: CC/Tot=.6% CC sim - 2 DATA SIM (Ia+CC) SIM (CC-only) CC.4 mag Brighter: CC/Tot=2.3% -3-2 - 2 3 µ SN µ fit (SALT2mu fit) - 2 - DATA SIM (Ia+CC) SIM (CC-only) -3-2 - 2 3 µ SN µ fit (SALT2mu fit) Double CC: CC/Tot=2.6% -3-2 - 2 3 µ SN µ fit (SALT2mu fit) Can tune CC to match μ-resid tail, but cannot fix shoulder problem. See Jones talk for tests with 9bg, Iax.... 2
Fikng the Hubble Diagram: CorrecCng for Biases and CC ContaminaCon 46 45 44 43 42 4 4 39 38 37 DATA, NN(z,s,c) z µ(α=.4, β=3.2) SIM, NN(z,s,c) Average Bias Corrected 36.2.4.6.8 zspec Distances zspec 22
Fikng the Hubble Diagram: Framework From Hlozek 2 L = L(Ia) + L(CC) μ μ model 4 3 2 - -2-3 -4 with NN.25.5.75 redshift 23
Fikng the Hubble Diagram: Framework From Hlozek 2 L = L(Ia) + L(CC) Fix cosmology params (w,ω M ) and fit for μ-offset in redshie bins (along with α,β): SALTmu program in SNANA, Marriner 2. Can then fit mu-offset vs. redshie with any cosmo program 24
Fikng the Hubble Diagram: CorrecQng for Biases and CC ContaminaCon Avg bias correccons from simulacon using color & stretch populacons from SK6: CorrecCons binned in 3D map of { z, x, c } 25
Fikng the Hubble Diagram: CorrecQng for Biases and CC ContaminaCon.6.5 c <.5 and x <.4.6.5 c <..6.5 x <.4 δ mb = m B m B,sim.4.3.2. -. αδ x = α(x x,sim ).4.3.2. -. < x < 2 βδ c = β(c c sim ).4.3.2. -. c >.2 c <. -.2 -.3 -.2 -.3 3 < x < 2 -.2 -.3 c <. -.4.2.4.6.8.2 redshift -.4.2.4.6.8.2 redshift -.4.2.4.6.8.2 redshift 26
Fikng the Hubble Diagram: CorrecCng for Biases : Sim SNIa only fit µ offset fit µ offset.8.6.4.2 -.2 -.4 -.6 -.8.8.6.4.2 -.2 -.4 -.6 -.8 a) chi2 only 2.9σ slope.2.4.6.8 redshift c) chi2 + 2ln(σ) 7.σ slope.2.4.6.8 redshift SNIa only (no CC) fit µ offset fit µ offset.8.6.4.2 -.2 -.4 -.6 -.8.8.6.4.2 -.2 -.4 -.6 -.8 b) chi2 + biascor 5.9σ slope.2.4.6.8 redshift d) chi2 + 2ln(σ) + biascor.σ slope.2.4.6.8 redshift Unbiased results require both ln(σ) term + biascor (leaving out either is bad): α fit /α SIM =.977(6) β fit /β SIM =.(8) 27
Fikng the Hubble Diagram: CorrecQng for Biases and CC ContaminaQon fit µ offset.8.6.4.2 -.2 -.4 -.6 -.8 Fit Ia + double-cc Sim c) ln(pa+pcc) 7.4σ slope redshift.2.4.6.8 redshift fit µ offset.8.6.4.2 -.2 -.4 -.6 -.8 redshift d) ln(pa+pcc) + biascor.σ slope.2.4.6.8 redshift Unbiased results Require biascor: α fit /α SIM =.986(2) β fit /β SIM =.999() S CC =.82() 28
Fikng the Hubble Diagram: CorrecQng for Biases and CC ContaminaQon fit µ offset.5..5 -.5 -. -.5 d) ln(pa+pcc) + biascor.5σ slope.2.4.6.8 redshift DES Data: Blind Analysis For data, SALT2mu program adds crazy μ-offset in each redshie bin unless explicitly asked to unblind. Fi_ed α, β, σ int are shown to compare with spec samples. 29
Final Caveat: Bias CorrecCon is Similar to NN Cut NN cut removes events where sim CC are majority in local {z,x,c} region Bias CorrecQon removes events where there are no sim-ia in local {z,x,c} bin (since correccon cannot be made) Requirement ater box cuts True CC/Tot raqo for sim with double CC None. Bias CorrecCon (no NN).39 NN (no bias correccon).26 NN + Bias correccon.24 3
Conclusions Overall data/sim agreement is good. Fake SN à photometry pipeline is important for tuning & validacng SNANA simulacon. Simulated Hubble Residuals underescmate high-side (faint) shoulder: Wrong LF? Wrong rate? Missing templates? New HD fikng method to simultaneously account for biases and CC contaminacon. 3